Index: stable/7/sys/dev/alc/if_alc.c =================================================================== --- stable/7/sys/dev/alc/if_alc.c (revision 242221) +++ stable/7/sys/dev/alc/if_alc.c (revision 242222) @@ -1,3779 +1,3779 @@ /*- * Copyright (c) 2009, Pyun YongHyeon * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* Driver for Atheros AR813x/AR815x PCIe Ethernet. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* "device miibus" required. See GENERIC if you get errors here. */ #include "miibus_if.h" #undef ALC_USE_CUSTOM_CSUM #ifdef ALC_USE_CUSTOM_CSUM #define ALC_CSUM_FEATURES (CSUM_TCP | CSUM_UDP) #else #define ALC_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP) #endif MODULE_DEPEND(alc, pci, 1, 1, 1); MODULE_DEPEND(alc, ether, 1, 1, 1); MODULE_DEPEND(alc, miibus, 1, 1, 1); /* Tunables. */ static int msi_disable = 0; static int msix_disable = 0; TUNABLE_INT("hw.alc.msi_disable", &msi_disable); TUNABLE_INT("hw.alc.msix_disable", &msix_disable); /* * Devices supported by this driver. */ static struct alc_ident alc_ident_table[] = { { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8131, 9 * 1024, "Atheros AR8131 PCIe Gigabit Ethernet" }, { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8132, 9 * 1024, "Atheros AR8132 PCIe Fast Ethernet" }, { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8151, 6 * 1024, "Atheros AR8151 v1.0 PCIe Gigabit Ethernet" }, { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8151_V2, 6 * 1024, "Atheros AR8151 v2.0 PCIe Gigabit Ethernet" }, { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8152_B, 6 * 1024, "Atheros AR8152 v1.1 PCIe Fast Ethernet" }, { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR8152_B2, 6 * 1024, "Atheros AR8152 v2.0 PCIe Fast Ethernet" }, { 0, 0, 0, NULL} }; static void alc_aspm(struct alc_softc *, int); static int alc_attach(device_t); static int alc_check_boundary(struct alc_softc *); static int alc_detach(device_t); static void alc_disable_l0s_l1(struct alc_softc *); static int alc_dma_alloc(struct alc_softc *); static void alc_dma_free(struct alc_softc *); static void alc_dmamap_cb(void *, bus_dma_segment_t *, int, int); static int alc_encap(struct alc_softc *, struct mbuf **); static struct alc_ident * alc_find_ident(device_t); #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * alc_fixup_rx(struct ifnet *, struct mbuf *); #endif static void alc_get_macaddr(struct alc_softc *); static void alc_init(void *); static void alc_init_cmb(struct alc_softc *); static void alc_init_locked(struct alc_softc *); static void alc_init_rr_ring(struct alc_softc *); static int alc_init_rx_ring(struct alc_softc *); static void alc_init_smb(struct alc_softc *); static void alc_init_tx_ring(struct alc_softc *); static void alc_int_task(void *, int); static int alc_intr(void *); static int alc_ioctl(struct ifnet *, u_long, caddr_t); static void alc_mac_config(struct alc_softc *); static int alc_miibus_readreg(device_t, int, int); static void alc_miibus_statchg(device_t); static int alc_miibus_writereg(device_t, int, int, int); static int alc_mediachange(struct ifnet *); static void alc_mediastatus(struct ifnet *, struct ifmediareq *); static int alc_newbuf(struct alc_softc *, struct alc_rxdesc *); static void alc_phy_down(struct alc_softc *); static void alc_phy_reset(struct alc_softc *); static int alc_probe(device_t); static void alc_reset(struct alc_softc *); static int alc_resume(device_t); static void alc_rxeof(struct alc_softc *, struct rx_rdesc *); static int alc_rxintr(struct alc_softc *, int); static void alc_rxfilter(struct alc_softc *); static void alc_rxvlan(struct alc_softc *); static void alc_setlinkspeed(struct alc_softc *); static void alc_setwol(struct alc_softc *); static int alc_shutdown(device_t); static void alc_start(struct ifnet *); static void alc_start_locked(struct ifnet *); static void alc_start_queue(struct alc_softc *); static void alc_stats_clear(struct alc_softc *); static void alc_stats_update(struct alc_softc *); static void alc_stop(struct alc_softc *); static void alc_stop_mac(struct alc_softc *); static void alc_stop_queue(struct alc_softc *); static int alc_suspend(device_t); static void alc_sysctl_node(struct alc_softc *); static void alc_tick(void *); static void alc_txeof(struct alc_softc *); static void alc_watchdog(struct alc_softc *); static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int); static int sysctl_hw_alc_proc_limit(SYSCTL_HANDLER_ARGS); static int sysctl_hw_alc_int_mod(SYSCTL_HANDLER_ARGS); static device_method_t alc_methods[] = { /* Device interface. */ DEVMETHOD(device_probe, alc_probe), DEVMETHOD(device_attach, alc_attach), DEVMETHOD(device_detach, alc_detach), DEVMETHOD(device_shutdown, alc_shutdown), DEVMETHOD(device_suspend, alc_suspend), DEVMETHOD(device_resume, alc_resume), /* MII interface. */ DEVMETHOD(miibus_readreg, alc_miibus_readreg), DEVMETHOD(miibus_writereg, alc_miibus_writereg), DEVMETHOD(miibus_statchg, alc_miibus_statchg), { NULL, NULL } }; static driver_t alc_driver = { "alc", alc_methods, sizeof(struct alc_softc) }; static devclass_t alc_devclass; DRIVER_MODULE(alc, pci, alc_driver, alc_devclass, 0, 0); DRIVER_MODULE(miibus, alc, miibus_driver, miibus_devclass, 0, 0); static struct resource_spec alc_res_spec_mem[] = { { SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE }, { -1, 0, 0 } }; static struct resource_spec alc_irq_spec_legacy[] = { { SYS_RES_IRQ, 0, RF_ACTIVE | RF_SHAREABLE }, { -1, 0, 0 } }; static struct resource_spec alc_irq_spec_msi[] = { { SYS_RES_IRQ, 1, RF_ACTIVE }, { -1, 0, 0 } }; static struct resource_spec alc_irq_spec_msix[] = { { SYS_RES_IRQ, 1, RF_ACTIVE }, { -1, 0, 0 } }; static uint32_t alc_dma_burst[] = { 128, 256, 512, 1024, 2048, 4096, 0 }; static int alc_miibus_readreg(device_t dev, int phy, int reg) { struct alc_softc *sc; uint32_t v; int i; sc = device_get_softc(dev); /* * For AR8132 fast ethernet controller, do not report 1000baseT * capability to mii(4). Even though AR8132 uses the same * model/revision number of F1 gigabit PHY, the PHY has no * ability to establish 1000baseT link. */ if ((sc->alc_flags & ALC_FLAG_FASTETHER) != 0 && reg == MII_EXTSR) return (0); CSR_WRITE_4(sc, ALC_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ | MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg)); for (i = ALC_PHY_TIMEOUT; i > 0; i--) { DELAY(5); v = CSR_READ_4(sc, ALC_MDIO); if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0) break; } if (i == 0) { device_printf(sc->alc_dev, "phy read timeout : %d\n", reg); return (0); } return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT); } static int alc_miibus_writereg(device_t dev, int phy, int reg, int val) { struct alc_softc *sc; uint32_t v; int i; sc = device_get_softc(dev); CSR_WRITE_4(sc, ALC_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE | (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT | MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg)); for (i = ALC_PHY_TIMEOUT; i > 0; i--) { DELAY(5); v = CSR_READ_4(sc, ALC_MDIO); if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0) break; } if (i == 0) device_printf(sc->alc_dev, "phy write timeout : %d\n", reg); return (0); } static void alc_miibus_statchg(device_t dev) { struct alc_softc *sc; struct mii_data *mii; struct ifnet *ifp; uint32_t reg; sc = device_get_softc(dev); mii = device_get_softc(sc->alc_miibus); ifp = sc->alc_ifp; if (mii == NULL || ifp == NULL || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; sc->alc_flags &= ~ALC_FLAG_LINK; if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: sc->alc_flags |= ALC_FLAG_LINK; break; case IFM_1000_T: if ((sc->alc_flags & ALC_FLAG_FASTETHER) == 0) sc->alc_flags |= ALC_FLAG_LINK; break; default: break; } } alc_stop_queue(sc); /* Stop Rx/Tx MACs. */ alc_stop_mac(sc); /* Program MACs with resolved speed/duplex/flow-control. */ if ((sc->alc_flags & ALC_FLAG_LINK) != 0) { alc_start_queue(sc); alc_mac_config(sc); /* Re-enable Tx/Rx MACs. */ reg = CSR_READ_4(sc, ALC_MAC_CFG); reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB; CSR_WRITE_4(sc, ALC_MAC_CFG, reg); alc_aspm(sc, IFM_SUBTYPE(mii->mii_media_active)); } } static void alc_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr) { struct alc_softc *sc; struct mii_data *mii; sc = ifp->if_softc; ALC_LOCK(sc); if ((ifp->if_flags & IFF_UP) == 0) { ALC_UNLOCK(sc); return; } mii = device_get_softc(sc->alc_miibus); mii_pollstat(mii); ifmr->ifm_status = mii->mii_media_status; ifmr->ifm_active = mii->mii_media_active; ALC_UNLOCK(sc); } static int alc_mediachange(struct ifnet *ifp) { struct alc_softc *sc; struct mii_data *mii; struct mii_softc *miisc; int error; sc = ifp->if_softc; ALC_LOCK(sc); mii = device_get_softc(sc->alc_miibus); LIST_FOREACH(miisc, &mii->mii_phys, mii_list) mii_phy_reset(miisc); error = mii_mediachg(mii); ALC_UNLOCK(sc); return (error); } static struct alc_ident * alc_find_ident(device_t dev) { struct alc_ident *ident; uint16_t vendor, devid; vendor = pci_get_vendor(dev); devid = pci_get_device(dev); for (ident = alc_ident_table; ident->name != NULL; ident++) { if (vendor == ident->vendorid && devid == ident->deviceid) return (ident); } return (NULL); } static int alc_probe(device_t dev) { struct alc_ident *ident; ident = alc_find_ident(dev); if (ident != NULL) { device_set_desc(dev, ident->name); return (BUS_PROBE_DEFAULT); } return (ENXIO); } static void alc_get_macaddr(struct alc_softc *sc) { uint32_t ea[2], opt; uint16_t val; int eeprom, i; eeprom = 0; opt = CSR_READ_4(sc, ALC_OPT_CFG); if ((CSR_READ_4(sc, ALC_MASTER_CFG) & MASTER_OTP_SEL) != 0 && (CSR_READ_4(sc, ALC_TWSI_DEBUG) & TWSI_DEBUG_DEV_EXIST) != 0) { /* * EEPROM found, let TWSI reload EEPROM configuration. * This will set ethernet address of controller. */ eeprom++; switch (sc->alc_ident->deviceid) { case DEVICEID_ATHEROS_AR8131: case DEVICEID_ATHEROS_AR8132: if ((opt & OPT_CFG_CLK_ENB) == 0) { opt |= OPT_CFG_CLK_ENB; CSR_WRITE_4(sc, ALC_OPT_CFG, opt); CSR_READ_4(sc, ALC_OPT_CFG); DELAY(1000); } break; case DEVICEID_ATHEROS_AR8151: case DEVICEID_ATHEROS_AR8151_V2: case DEVICEID_ATHEROS_AR8152_B: case DEVICEID_ATHEROS_AR8152_B2: alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x00); val = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, val & 0xFF7F); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x3B); val = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, val | 0x0008); DELAY(20); break; } CSR_WRITE_4(sc, ALC_LTSSM_ID_CFG, CSR_READ_4(sc, ALC_LTSSM_ID_CFG) & ~LTSSM_ID_WRO_ENB); CSR_WRITE_4(sc, ALC_WOL_CFG, 0); CSR_READ_4(sc, ALC_WOL_CFG); CSR_WRITE_4(sc, ALC_TWSI_CFG, CSR_READ_4(sc, ALC_TWSI_CFG) | TWSI_CFG_SW_LD_START); for (i = 100; i > 0; i--) { DELAY(1000); if ((CSR_READ_4(sc, ALC_TWSI_CFG) & TWSI_CFG_SW_LD_START) == 0) break; } if (i == 0) device_printf(sc->alc_dev, "reloading EEPROM timeout!\n"); } else { if (bootverbose) device_printf(sc->alc_dev, "EEPROM not found!\n"); } if (eeprom != 0) { switch (sc->alc_ident->deviceid) { case DEVICEID_ATHEROS_AR8131: case DEVICEID_ATHEROS_AR8132: if ((opt & OPT_CFG_CLK_ENB) != 0) { opt &= ~OPT_CFG_CLK_ENB; CSR_WRITE_4(sc, ALC_OPT_CFG, opt); CSR_READ_4(sc, ALC_OPT_CFG); DELAY(1000); } break; case DEVICEID_ATHEROS_AR8151: case DEVICEID_ATHEROS_AR8151_V2: case DEVICEID_ATHEROS_AR8152_B: case DEVICEID_ATHEROS_AR8152_B2: alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x00); val = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, val | 0x0080); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x3B); val = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, val & 0xFFF7); DELAY(20); break; } } ea[0] = CSR_READ_4(sc, ALC_PAR0); ea[1] = CSR_READ_4(sc, ALC_PAR1); sc->alc_eaddr[0] = (ea[1] >> 8) & 0xFF; sc->alc_eaddr[1] = (ea[1] >> 0) & 0xFF; sc->alc_eaddr[2] = (ea[0] >> 24) & 0xFF; sc->alc_eaddr[3] = (ea[0] >> 16) & 0xFF; sc->alc_eaddr[4] = (ea[0] >> 8) & 0xFF; sc->alc_eaddr[5] = (ea[0] >> 0) & 0xFF; } static void alc_disable_l0s_l1(struct alc_softc *sc) { uint32_t pmcfg; /* Another magic from vendor. */ pmcfg = CSR_READ_4(sc, ALC_PM_CFG); pmcfg &= ~(PM_CFG_L1_ENTRY_TIMER_MASK | PM_CFG_CLK_SWH_L1 | PM_CFG_ASPM_L0S_ENB | PM_CFG_ASPM_L1_ENB | PM_CFG_MAC_ASPM_CHK | PM_CFG_SERDES_PD_EX_L1); pmcfg |= PM_CFG_SERDES_BUDS_RX_L1_ENB | PM_CFG_SERDES_PLL_L1_ENB | PM_CFG_SERDES_L1_ENB; CSR_WRITE_4(sc, ALC_PM_CFG, pmcfg); } static void alc_phy_reset(struct alc_softc *sc) { uint16_t data; /* Reset magic from Linux. */ CSR_WRITE_2(sc, ALC_GPHY_CFG, GPHY_CFG_SEL_ANA_RESET); CSR_READ_2(sc, ALC_GPHY_CFG); DELAY(10 * 1000); CSR_WRITE_2(sc, ALC_GPHY_CFG, GPHY_CFG_EXT_RESET | GPHY_CFG_SEL_ANA_RESET); CSR_READ_2(sc, ALC_GPHY_CFG); DELAY(10 * 1000); /* DSP fixup, Vendor magic. */ if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B) { alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x000A); data = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data & 0xDFFF); } if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151_V2 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B2) { alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x003B); data = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data & 0xFFF7); DELAY(20 * 1000); } if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151) { alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x0029); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, 0x929D); } if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8131 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8132 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151_V2 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B2) { alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x0029); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, 0xB6DD); } /* Load DSP codes, vendor magic. */ data = ANA_LOOP_SEL_10BT | ANA_EN_MASK_TB | ANA_EN_10BT_IDLE | ((1 << ANA_INTERVAL_SEL_TIMER_SHIFT) & ANA_INTERVAL_SEL_TIMER_MASK); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, MII_ANA_CFG18); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); data = ((2 << ANA_SERDES_CDR_BW_SHIFT) & ANA_SERDES_CDR_BW_MASK) | ANA_SERDES_EN_DEEM | ANA_SERDES_SEL_HSP | ANA_SERDES_EN_PLL | ANA_SERDES_EN_LCKDT; alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, MII_ANA_CFG5); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); data = ((44 << ANA_LONG_CABLE_TH_100_SHIFT) & ANA_LONG_CABLE_TH_100_MASK) | ((33 << ANA_SHORT_CABLE_TH_100_SHIFT) & ANA_SHORT_CABLE_TH_100_SHIFT) | ANA_BP_BAD_LINK_ACCUM | ANA_BP_SMALL_BW; alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, MII_ANA_CFG54); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); data = ((11 << ANA_IECHO_ADJ_3_SHIFT) & ANA_IECHO_ADJ_3_MASK) | ((11 << ANA_IECHO_ADJ_2_SHIFT) & ANA_IECHO_ADJ_2_MASK) | ((8 << ANA_IECHO_ADJ_1_SHIFT) & ANA_IECHO_ADJ_1_MASK) | ((8 << ANA_IECHO_ADJ_0_SHIFT) & ANA_IECHO_ADJ_0_MASK); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, MII_ANA_CFG4); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); data = ((7 & ANA_MANUL_SWICH_ON_SHIFT) & ANA_MANUL_SWICH_ON_MASK) | ANA_RESTART_CAL | ANA_MAN_ENABLE | ANA_SEL_HSP | ANA_EN_HB | ANA_OEN_125M; alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, MII_ANA_CFG0); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); DELAY(1000); /* Disable hibernation. */ alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x0029); data = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); data &= ~0x8000; alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_ADDR, 0x000B); data = alc_miibus_readreg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA); data &= ~0x8000; alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, ALC_MII_DBG_DATA, data); } static void alc_phy_down(struct alc_softc *sc) { switch (sc->alc_ident->deviceid) { case DEVICEID_ATHEROS_AR8151: case DEVICEID_ATHEROS_AR8151_V2: /* * GPHY power down caused more problems on AR8151 v2.0. * When driver is reloaded after GPHY power down, * accesses to PHY/MAC registers hung the system. Only * cold boot recovered from it. I'm not sure whether * AR8151 v1.0 also requires this one though. I don't * have AR8151 v1.0 controller in hand. * The only option left is to isolate the PHY and * initiates power down the PHY which in turn saves * more power when driver is unloaded. */ alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, MII_BMCR, BMCR_ISO | BMCR_PDOWN); break; default: /* Force PHY down. */ CSR_WRITE_2(sc, ALC_GPHY_CFG, GPHY_CFG_EXT_RESET | GPHY_CFG_SEL_ANA_RESET | GPHY_CFG_PHY_IDDQ | GPHY_CFG_PWDOWN_HW); DELAY(1000); break; } } static void alc_aspm(struct alc_softc *sc, int media) { uint32_t pmcfg; uint16_t linkcfg; ALC_LOCK_ASSERT(sc); pmcfg = CSR_READ_4(sc, ALC_PM_CFG); if ((sc->alc_flags & (ALC_FLAG_APS | ALC_FLAG_PCIE)) == (ALC_FLAG_APS | ALC_FLAG_PCIE)) linkcfg = CSR_READ_2(sc, sc->alc_expcap + - PCIR_EXPRESS_LINK_CTL); + PCIER_LINK_CTL); else linkcfg = 0; pmcfg &= ~PM_CFG_SERDES_PD_EX_L1; pmcfg &= ~(PM_CFG_L1_ENTRY_TIMER_MASK | PM_CFG_LCKDET_TIMER_MASK); pmcfg |= PM_CFG_MAC_ASPM_CHK; pmcfg |= (PM_CFG_LCKDET_TIMER_DEFAULT << PM_CFG_LCKDET_TIMER_SHIFT); pmcfg &= ~(PM_CFG_ASPM_L1_ENB | PM_CFG_ASPM_L0S_ENB); if ((sc->alc_flags & ALC_FLAG_APS) != 0) { /* Disable extended sync except AR8152 B v1.0 */ linkcfg &= ~0x80; if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B && sc->alc_rev == ATHEROS_AR8152_B_V10) linkcfg |= 0x80; - CSR_WRITE_2(sc, sc->alc_expcap + PCIR_EXPRESS_LINK_CTL, + CSR_WRITE_2(sc, sc->alc_expcap + PCIER_LINK_CTL, linkcfg); pmcfg &= ~(PM_CFG_EN_BUFS_RX_L0S | PM_CFG_SA_DLY_ENB | PM_CFG_HOTRST); pmcfg |= (PM_CFG_L1_ENTRY_TIMER_DEFAULT << PM_CFG_L1_ENTRY_TIMER_SHIFT); pmcfg &= ~PM_CFG_PM_REQ_TIMER_MASK; pmcfg |= (PM_CFG_PM_REQ_TIMER_DEFAULT << PM_CFG_PM_REQ_TIMER_SHIFT); pmcfg |= PM_CFG_SERDES_PD_EX_L1 | PM_CFG_PCIE_RECV; } if ((sc->alc_flags & ALC_FLAG_LINK) != 0) { if ((sc->alc_flags & ALC_FLAG_L0S) != 0) pmcfg |= PM_CFG_ASPM_L0S_ENB; if ((sc->alc_flags & ALC_FLAG_L1S) != 0) pmcfg |= PM_CFG_ASPM_L1_ENB; if ((sc->alc_flags & ALC_FLAG_APS) != 0) { if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B) pmcfg &= ~PM_CFG_ASPM_L0S_ENB; pmcfg &= ~(PM_CFG_SERDES_L1_ENB | PM_CFG_SERDES_PLL_L1_ENB | PM_CFG_SERDES_BUDS_RX_L1_ENB); pmcfg |= PM_CFG_CLK_SWH_L1; if (media == IFM_100_TX || media == IFM_1000_T) { pmcfg &= ~PM_CFG_L1_ENTRY_TIMER_MASK; switch (sc->alc_ident->deviceid) { case DEVICEID_ATHEROS_AR8152_B: pmcfg |= (7 << PM_CFG_L1_ENTRY_TIMER_SHIFT); break; case DEVICEID_ATHEROS_AR8152_B2: case DEVICEID_ATHEROS_AR8151_V2: pmcfg |= (4 << PM_CFG_L1_ENTRY_TIMER_SHIFT); break; default: pmcfg |= (15 << PM_CFG_L1_ENTRY_TIMER_SHIFT); break; } } } else { pmcfg |= PM_CFG_SERDES_L1_ENB | PM_CFG_SERDES_PLL_L1_ENB | PM_CFG_SERDES_BUDS_RX_L1_ENB; pmcfg &= ~(PM_CFG_CLK_SWH_L1 | PM_CFG_ASPM_L1_ENB | PM_CFG_ASPM_L0S_ENB); } } else { pmcfg &= ~(PM_CFG_SERDES_BUDS_RX_L1_ENB | PM_CFG_SERDES_L1_ENB | PM_CFG_SERDES_PLL_L1_ENB); pmcfg |= PM_CFG_CLK_SWH_L1; if ((sc->alc_flags & ALC_FLAG_L1S) != 0) pmcfg |= PM_CFG_ASPM_L1_ENB; } CSR_WRITE_4(sc, ALC_PM_CFG, pmcfg); } static int alc_attach(device_t dev) { struct alc_softc *sc; struct ifnet *ifp; char *aspm_state[] = { "L0s/L1", "L0s", "L1", "L0s/L1" }; uint16_t burst; int base, error, i, msic, msixc, state; uint32_t cap, ctl, val; error = 0; sc = device_get_softc(dev); sc->alc_dev = dev; mtx_init(&sc->alc_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK, MTX_DEF); callout_init_mtx(&sc->alc_tick_ch, &sc->alc_mtx, 0); TASK_INIT(&sc->alc_int_task, 0, alc_int_task, sc); sc->alc_ident = alc_find_ident(dev); /* Map the device. */ pci_enable_busmaster(dev); sc->alc_res_spec = alc_res_spec_mem; sc->alc_irq_spec = alc_irq_spec_legacy; error = bus_alloc_resources(dev, sc->alc_res_spec, sc->alc_res); if (error != 0) { device_printf(dev, "cannot allocate memory resources.\n"); goto fail; } /* Set PHY address. */ sc->alc_phyaddr = ALC_PHY_ADDR; /* Initialize DMA parameters. */ sc->alc_dma_rd_burst = 0; sc->alc_dma_wr_burst = 0; sc->alc_rcb = DMA_CFG_RCB_64; if (pci_find_extcap(dev, PCIY_EXPRESS, &base) == 0) { sc->alc_flags |= ALC_FLAG_PCIE; sc->alc_expcap = base; - burst = CSR_READ_2(sc, base + PCIR_EXPRESS_DEVICE_CTL); + burst = CSR_READ_2(sc, base + PCIER_DEVICE_CTL); sc->alc_dma_rd_burst = - (burst & PCIM_EXP_CTL_MAX_READ_REQUEST) >> 12; - sc->alc_dma_wr_burst = (burst & PCIM_EXP_CTL_MAX_PAYLOAD) >> 5; + (burst & PCIEM_CTL_MAX_READ_REQUEST) >> 12; + sc->alc_dma_wr_burst = (burst & PCIEM_CTL_MAX_PAYLOAD) >> 5; if (bootverbose) { device_printf(dev, "Read request size : %u bytes.\n", alc_dma_burst[sc->alc_dma_rd_burst]); device_printf(dev, "TLP payload size : %u bytes.\n", alc_dma_burst[sc->alc_dma_wr_burst]); } if (alc_dma_burst[sc->alc_dma_rd_burst] > 1024) sc->alc_dma_rd_burst = 3; if (alc_dma_burst[sc->alc_dma_wr_burst] > 1024) sc->alc_dma_wr_burst = 3; /* Clear data link and flow-control protocol error. */ val = CSR_READ_4(sc, ALC_PEX_UNC_ERR_SEV); val &= ~(PEX_UNC_ERR_SEV_DLP | PEX_UNC_ERR_SEV_FCP); CSR_WRITE_4(sc, ALC_PEX_UNC_ERR_SEV, val); CSR_WRITE_4(sc, ALC_LTSSM_ID_CFG, CSR_READ_4(sc, ALC_LTSSM_ID_CFG) & ~LTSSM_ID_WRO_ENB); CSR_WRITE_4(sc, ALC_PCIE_PHYMISC, CSR_READ_4(sc, ALC_PCIE_PHYMISC) | PCIE_PHYMISC_FORCE_RCV_DET); if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B && pci_get_revid(dev) == ATHEROS_AR8152_B_V10) { val = CSR_READ_4(sc, ALC_PCIE_PHYMISC2); val &= ~(PCIE_PHYMISC2_SERDES_CDR_MASK | PCIE_PHYMISC2_SERDES_TH_MASK); val |= 3 << PCIE_PHYMISC2_SERDES_CDR_SHIFT; val |= 3 << PCIE_PHYMISC2_SERDES_TH_SHIFT; CSR_WRITE_4(sc, ALC_PCIE_PHYMISC2, val); } /* Disable ASPM L0S and L1. */ - cap = CSR_READ_2(sc, base + PCIR_EXPRESS_LINK_CAP); - if ((cap & PCIM_LINK_CAP_ASPM) != 0) { - ctl = CSR_READ_2(sc, base + PCIR_EXPRESS_LINK_CTL); + cap = CSR_READ_2(sc, base + PCIER_LINK_CAP); + if ((cap & PCIEM_LINK_CAP_ASPM) != 0) { + ctl = CSR_READ_2(sc, base + PCIER_LINK_CTL); if ((ctl & 0x08) != 0) sc->alc_rcb = DMA_CFG_RCB_128; if (bootverbose) device_printf(dev, "RCB %u bytes\n", sc->alc_rcb == DMA_CFG_RCB_64 ? 64 : 128); state = ctl & 0x03; if (state & 0x01) sc->alc_flags |= ALC_FLAG_L0S; if (state & 0x02) sc->alc_flags |= ALC_FLAG_L1S; if (bootverbose) device_printf(sc->alc_dev, "ASPM %s %s\n", aspm_state[state], state == 0 ? "disabled" : "enabled"); alc_disable_l0s_l1(sc); } else { if (bootverbose) device_printf(sc->alc_dev, "no ASPM support\n"); } } /* Reset PHY. */ alc_phy_reset(sc); /* Reset the ethernet controller. */ alc_reset(sc); /* * One odd thing is AR8132 uses the same PHY hardware(F1 * gigabit PHY) of AR8131. So atphy(4) of AR8132 reports * the PHY supports 1000Mbps but that's not true. The PHY * used in AR8132 can't establish gigabit link even if it * shows the same PHY model/revision number of AR8131. */ switch (sc->alc_ident->deviceid) { case DEVICEID_ATHEROS_AR8152_B: case DEVICEID_ATHEROS_AR8152_B2: sc->alc_flags |= ALC_FLAG_APS; /* FALLTHROUGH */ case DEVICEID_ATHEROS_AR8132: sc->alc_flags |= ALC_FLAG_FASTETHER; break; case DEVICEID_ATHEROS_AR8151: case DEVICEID_ATHEROS_AR8151_V2: sc->alc_flags |= ALC_FLAG_APS; /* FALLTHROUGH */ default: break; } sc->alc_flags |= ALC_FLAG_ASPM_MON | ALC_FLAG_JUMBO; /* * It seems that AR813x/AR815x has silicon bug for SMB. In * addition, Atheros said that enabling SMB wouldn't improve * performance. However I think it's bad to access lots of * registers to extract MAC statistics. */ sc->alc_flags |= ALC_FLAG_SMB_BUG; /* * Don't use Tx CMB. It is known to have silicon bug. */ sc->alc_flags |= ALC_FLAG_CMB_BUG; sc->alc_rev = pci_get_revid(dev); sc->alc_chip_rev = CSR_READ_4(sc, ALC_MASTER_CFG) >> MASTER_CHIP_REV_SHIFT; if (bootverbose) { device_printf(dev, "PCI device revision : 0x%04x\n", sc->alc_rev); device_printf(dev, "Chip id/revision : 0x%04x\n", sc->alc_chip_rev); } device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", CSR_READ_4(sc, ALC_SRAM_TX_FIFO_LEN) * 8, CSR_READ_4(sc, ALC_SRAM_RX_FIFO_LEN) * 8); /* Allocate IRQ resources. */ msixc = pci_msix_count(dev); msic = pci_msi_count(dev); if (bootverbose) { device_printf(dev, "MSIX count : %d\n", msixc); device_printf(dev, "MSI count : %d\n", msic); } /* Prefer MSIX over MSI. */ if (msix_disable == 0 || msi_disable == 0) { if (msix_disable == 0 && msixc == ALC_MSIX_MESSAGES && pci_alloc_msix(dev, &msixc) == 0) { if (msic == ALC_MSIX_MESSAGES) { device_printf(dev, "Using %d MSIX message(s).\n", msixc); sc->alc_flags |= ALC_FLAG_MSIX; sc->alc_irq_spec = alc_irq_spec_msix; } else pci_release_msi(dev); } if (msi_disable == 0 && (sc->alc_flags & ALC_FLAG_MSIX) == 0 && msic == ALC_MSI_MESSAGES && pci_alloc_msi(dev, &msic) == 0) { if (msic == ALC_MSI_MESSAGES) { device_printf(dev, "Using %d MSI message(s).\n", msic); sc->alc_flags |= ALC_FLAG_MSI; sc->alc_irq_spec = alc_irq_spec_msi; } else pci_release_msi(dev); } } error = bus_alloc_resources(dev, sc->alc_irq_spec, sc->alc_irq); if (error != 0) { device_printf(dev, "cannot allocate IRQ resources.\n"); goto fail; } /* Create device sysctl node. */ alc_sysctl_node(sc); if ((error = alc_dma_alloc(sc) != 0)) goto fail; /* Load station address. */ alc_get_macaddr(sc); ifp = sc->alc_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "cannot allocate ifnet structure.\n"); error = ENXIO; goto fail; } ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = alc_ioctl; ifp->if_start = alc_start; ifp->if_init = alc_init; ifp->if_snd.ifq_drv_maxlen = ALC_TX_RING_CNT - 1; IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen); IFQ_SET_READY(&ifp->if_snd); ifp->if_capabilities = IFCAP_TXCSUM | IFCAP_TSO4; ifp->if_hwassist = ALC_CSUM_FEATURES | CSUM_TSO; if (pci_find_extcap(dev, PCIY_PMG, &base) == 0) { ifp->if_capabilities |= IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST; sc->alc_flags |= ALC_FLAG_PM; sc->alc_pmcap = base; } ifp->if_capenable = ifp->if_capabilities; /* Set up MII bus. */ error = mii_attach(dev, &sc->alc_miibus, ifp, alc_mediachange, alc_mediastatus, BMSR_DEFCAPMASK, sc->alc_phyaddr, MII_OFFSET_ANY, MIIF_DOPAUSE); if (error != 0) { device_printf(dev, "attaching PHYs failed\n"); goto fail; } ether_ifattach(ifp, sc->alc_eaddr); /* VLAN capability setup. */ ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO; ifp->if_capenable = ifp->if_capabilities; /* * XXX * It seems enabling Tx checksum offloading makes more trouble. * Sometimes the controller does not receive any frames when * Tx checksum offloading is enabled. I'm not sure whether this * is a bug in Tx checksum offloading logic or I got broken * sample boards. To safety, don't enable Tx checksum offloading * by default but give chance to users to toggle it if they know * their controllers work without problems. */ ifp->if_capenable &= ~IFCAP_TXCSUM; ifp->if_hwassist &= ~ALC_CSUM_FEATURES; /* Tell the upper layer(s) we support long frames. */ ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); /* Create local taskq. */ sc->alc_tq = taskqueue_create_fast("alc_taskq", M_WAITOK, taskqueue_thread_enqueue, &sc->alc_tq); if (sc->alc_tq == NULL) { device_printf(dev, "could not create taskqueue.\n"); ether_ifdetach(ifp); error = ENXIO; goto fail; } taskqueue_start_threads(&sc->alc_tq, 1, PI_NET, "%s taskq", device_get_nameunit(sc->alc_dev)); if ((sc->alc_flags & ALC_FLAG_MSIX) != 0) msic = ALC_MSIX_MESSAGES; else if ((sc->alc_flags & ALC_FLAG_MSI) != 0) msic = ALC_MSI_MESSAGES; else msic = 1; for (i = 0; i < msic; i++) { error = bus_setup_intr(dev, sc->alc_irq[i], INTR_TYPE_NET | INTR_MPSAFE, alc_intr, NULL, sc, &sc->alc_intrhand[i]); if (error != 0) break; } if (error != 0) { device_printf(dev, "could not set up interrupt handler.\n"); taskqueue_free(sc->alc_tq); sc->alc_tq = NULL; ether_ifdetach(ifp); goto fail; } fail: if (error != 0) alc_detach(dev); return (error); } static int alc_detach(device_t dev) { struct alc_softc *sc; struct ifnet *ifp; int i, msic; sc = device_get_softc(dev); ifp = sc->alc_ifp; if (device_is_attached(dev)) { ether_ifdetach(ifp); ALC_LOCK(sc); alc_stop(sc); ALC_UNLOCK(sc); callout_drain(&sc->alc_tick_ch); taskqueue_drain(sc->alc_tq, &sc->alc_int_task); } if (sc->alc_tq != NULL) { taskqueue_drain(sc->alc_tq, &sc->alc_int_task); taskqueue_free(sc->alc_tq); sc->alc_tq = NULL; } if (sc->alc_miibus != NULL) { device_delete_child(dev, sc->alc_miibus); sc->alc_miibus = NULL; } bus_generic_detach(dev); alc_dma_free(sc); if (ifp != NULL) { if_free(ifp); sc->alc_ifp = NULL; } if ((sc->alc_flags & ALC_FLAG_MSIX) != 0) msic = ALC_MSIX_MESSAGES; else if ((sc->alc_flags & ALC_FLAG_MSI) != 0) msic = ALC_MSI_MESSAGES; else msic = 1; for (i = 0; i < msic; i++) { if (sc->alc_intrhand[i] != NULL) { bus_teardown_intr(dev, sc->alc_irq[i], sc->alc_intrhand[i]); sc->alc_intrhand[i] = NULL; } } if (sc->alc_res[0] != NULL) alc_phy_down(sc); bus_release_resources(dev, sc->alc_irq_spec, sc->alc_irq); if ((sc->alc_flags & (ALC_FLAG_MSI | ALC_FLAG_MSIX)) != 0) pci_release_msi(dev); bus_release_resources(dev, sc->alc_res_spec, sc->alc_res); mtx_destroy(&sc->alc_mtx); return (0); } #define ALC_SYSCTL_STAT_ADD32(c, h, n, p, d) \ SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d) #define ALC_SYSCTL_STAT_ADD64(c, h, n, p, d) \ SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d) static void alc_sysctl_node(struct alc_softc *sc) { struct sysctl_ctx_list *ctx; struct sysctl_oid_list *child, *parent; struct sysctl_oid *tree; struct alc_hw_stats *stats; int error; stats = &sc->alc_stats; ctx = device_get_sysctl_ctx(sc->alc_dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->alc_dev)); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod", CTLTYPE_INT | CTLFLAG_RW, &sc->alc_int_rx_mod, 0, sysctl_hw_alc_int_mod, "I", "alc Rx interrupt moderation"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod", CTLTYPE_INT | CTLFLAG_RW, &sc->alc_int_tx_mod, 0, sysctl_hw_alc_int_mod, "I", "alc Tx interrupt moderation"); /* Pull in device tunables. */ sc->alc_int_rx_mod = ALC_IM_RX_TIMER_DEFAULT; error = resource_int_value(device_get_name(sc->alc_dev), device_get_unit(sc->alc_dev), "int_rx_mod", &sc->alc_int_rx_mod); if (error == 0) { if (sc->alc_int_rx_mod < ALC_IM_TIMER_MIN || sc->alc_int_rx_mod > ALC_IM_TIMER_MAX) { device_printf(sc->alc_dev, "int_rx_mod value out of " "range; using default: %d\n", ALC_IM_RX_TIMER_DEFAULT); sc->alc_int_rx_mod = ALC_IM_RX_TIMER_DEFAULT; } } sc->alc_int_tx_mod = ALC_IM_TX_TIMER_DEFAULT; error = resource_int_value(device_get_name(sc->alc_dev), device_get_unit(sc->alc_dev), "int_tx_mod", &sc->alc_int_tx_mod); if (error == 0) { if (sc->alc_int_tx_mod < ALC_IM_TIMER_MIN || sc->alc_int_tx_mod > ALC_IM_TIMER_MAX) { device_printf(sc->alc_dev, "int_tx_mod value out of " "range; using default: %d\n", ALC_IM_TX_TIMER_DEFAULT); sc->alc_int_tx_mod = ALC_IM_TX_TIMER_DEFAULT; } } SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit", CTLTYPE_INT | CTLFLAG_RW, &sc->alc_process_limit, 0, sysctl_hw_alc_proc_limit, "I", "max number of Rx events to process"); /* Pull in device tunables. */ sc->alc_process_limit = ALC_PROC_DEFAULT; error = resource_int_value(device_get_name(sc->alc_dev), device_get_unit(sc->alc_dev), "process_limit", &sc->alc_process_limit); if (error == 0) { if (sc->alc_process_limit < ALC_PROC_MIN || sc->alc_process_limit > ALC_PROC_MAX) { device_printf(sc->alc_dev, "process_limit value out of range; " "using default: %d\n", ALC_PROC_DEFAULT); sc->alc_process_limit = ALC_PROC_DEFAULT; } } tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD, NULL, "ALC statistics"); parent = SYSCTL_CHILDREN(tree); /* Rx statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD, NULL, "Rx MAC statistics"); child = SYSCTL_CHILDREN(tree); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_frames", &stats->rx_frames, "Good frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames", &stats->rx_bcast_frames, "Good broadcast frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames", &stats->rx_mcast_frames, "Good multicast frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "pause_frames", &stats->rx_pause_frames, "Pause control frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "control_frames", &stats->rx_control_frames, "Control frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "crc_errs", &stats->rx_crcerrs, "CRC errors"); ALC_SYSCTL_STAT_ADD32(ctx, child, "len_errs", &stats->rx_lenerrs, "Frames with length mismatched"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_octets", &stats->rx_bytes, "Good octets"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets", &stats->rx_bcast_bytes, "Good broadcast octets"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets", &stats->rx_mcast_bytes, "Good multicast octets"); ALC_SYSCTL_STAT_ADD32(ctx, child, "runts", &stats->rx_runts, "Too short frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "fragments", &stats->rx_fragments, "Fragmented frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_64", &stats->rx_pkts_64, "64 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127", &stats->rx_pkts_65_127, "65 to 127 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255", &stats->rx_pkts_128_255, "128 to 255 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511", &stats->rx_pkts_256_511, "256 to 511 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023", &stats->rx_pkts_512_1023, "512 to 1023 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518", &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max", &stats->rx_pkts_1519_max, "1519 to max frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs", &stats->rx_pkts_truncated, "Truncated frames due to MTU size"); ALC_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows", &stats->rx_fifo_oflows, "FIFO overflows"); ALC_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs", &stats->rx_rrs_errs, "Return status write-back errors"); ALC_SYSCTL_STAT_ADD32(ctx, child, "align_errs", &stats->rx_alignerrs, "Alignment errors"); ALC_SYSCTL_STAT_ADD32(ctx, child, "filtered", &stats->rx_pkts_filtered, "Frames dropped due to address filtering"); /* Tx statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD, NULL, "Tx MAC statistics"); child = SYSCTL_CHILDREN(tree); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_frames", &stats->tx_frames, "Good frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames", &stats->tx_bcast_frames, "Good broadcast frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames", &stats->tx_mcast_frames, "Good multicast frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "pause_frames", &stats->tx_pause_frames, "Pause control frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "control_frames", &stats->tx_control_frames, "Control frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "excess_defers", &stats->tx_excess_defer, "Frames with excessive derferrals"); ALC_SYSCTL_STAT_ADD32(ctx, child, "defers", &stats->tx_excess_defer, "Frames with derferrals"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_octets", &stats->tx_bytes, "Good octets"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets", &stats->tx_bcast_bytes, "Good broadcast octets"); ALC_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets", &stats->tx_mcast_bytes, "Good multicast octets"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_64", &stats->tx_pkts_64, "64 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127", &stats->tx_pkts_65_127, "65 to 127 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255", &stats->tx_pkts_128_255, "128 to 255 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511", &stats->tx_pkts_256_511, "256 to 511 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023", &stats->tx_pkts_512_1023, "512 to 1023 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518", &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max", &stats->tx_pkts_1519_max, "1519 to max frames"); ALC_SYSCTL_STAT_ADD32(ctx, child, "single_colls", &stats->tx_single_colls, "Single collisions"); ALC_SYSCTL_STAT_ADD32(ctx, child, "multi_colls", &stats->tx_multi_colls, "Multiple collisions"); ALC_SYSCTL_STAT_ADD32(ctx, child, "late_colls", &stats->tx_late_colls, "Late collisions"); ALC_SYSCTL_STAT_ADD32(ctx, child, "excess_colls", &stats->tx_excess_colls, "Excessive collisions"); ALC_SYSCTL_STAT_ADD32(ctx, child, "abort", &stats->tx_abort, "Aborted frames due to Excessive collisions"); ALC_SYSCTL_STAT_ADD32(ctx, child, "underruns", &stats->tx_underrun, "FIFO underruns"); ALC_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns", &stats->tx_desc_underrun, "Descriptor write-back errors"); ALC_SYSCTL_STAT_ADD32(ctx, child, "len_errs", &stats->tx_lenerrs, "Frames with length mismatched"); ALC_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs", &stats->tx_pkts_truncated, "Truncated frames due to MTU size"); } #undef ALC_SYSCTL_STAT_ADD32 #undef ALC_SYSCTL_STAT_ADD64 struct alc_dmamap_arg { bus_addr_t alc_busaddr; }; static void alc_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct alc_dmamap_arg *ctx; if (error != 0) return; KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); ctx = (struct alc_dmamap_arg *)arg; ctx->alc_busaddr = segs[0].ds_addr; } /* * Normal and high Tx descriptors shares single Tx high address. * Four Rx descriptor/return rings and CMB shares the same Rx * high address. */ static int alc_check_boundary(struct alc_softc *sc) { bus_addr_t cmb_end, rx_ring_end, rr_ring_end, tx_ring_end; rx_ring_end = sc->alc_rdata.alc_rx_ring_paddr + ALC_RX_RING_SZ; rr_ring_end = sc->alc_rdata.alc_rr_ring_paddr + ALC_RR_RING_SZ; cmb_end = sc->alc_rdata.alc_cmb_paddr + ALC_CMB_SZ; tx_ring_end = sc->alc_rdata.alc_tx_ring_paddr + ALC_TX_RING_SZ; /* 4GB boundary crossing is not allowed. */ if ((ALC_ADDR_HI(rx_ring_end) != ALC_ADDR_HI(sc->alc_rdata.alc_rx_ring_paddr)) || (ALC_ADDR_HI(rr_ring_end) != ALC_ADDR_HI(sc->alc_rdata.alc_rr_ring_paddr)) || (ALC_ADDR_HI(cmb_end) != ALC_ADDR_HI(sc->alc_rdata.alc_cmb_paddr)) || (ALC_ADDR_HI(tx_ring_end) != ALC_ADDR_HI(sc->alc_rdata.alc_tx_ring_paddr))) return (EFBIG); /* * Make sure Rx return descriptor/Rx descriptor/CMB use * the same high address. */ if ((ALC_ADDR_HI(rx_ring_end) != ALC_ADDR_HI(rr_ring_end)) || (ALC_ADDR_HI(rx_ring_end) != ALC_ADDR_HI(cmb_end))) return (EFBIG); return (0); } static int alc_dma_alloc(struct alc_softc *sc) { struct alc_txdesc *txd; struct alc_rxdesc *rxd; bus_addr_t lowaddr; struct alc_dmamap_arg ctx; int error, i; lowaddr = BUS_SPACE_MAXADDR; again: /* Create parent DMA tag. */ error = bus_dma_tag_create( bus_get_dma_tag(sc->alc_dev), /* parent */ 1, 0, /* alignment, boundary */ lowaddr, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT, /* maxsize */ 0, /* nsegments */ BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_parent_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create parent DMA tag.\n"); goto fail; } /* Create DMA tag for Tx descriptor ring. */ error = bus_dma_tag_create( sc->alc_cdata.alc_parent_tag, /* parent */ ALC_TX_RING_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_TX_RING_SZ, /* maxsize */ 1, /* nsegments */ ALC_TX_RING_SZ, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_tx_ring_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create Tx ring DMA tag.\n"); goto fail; } /* Create DMA tag for Rx free descriptor ring. */ error = bus_dma_tag_create( sc->alc_cdata.alc_parent_tag, /* parent */ ALC_RX_RING_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_RX_RING_SZ, /* maxsize */ 1, /* nsegments */ ALC_RX_RING_SZ, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_rx_ring_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create Rx ring DMA tag.\n"); goto fail; } /* Create DMA tag for Rx return descriptor ring. */ error = bus_dma_tag_create( sc->alc_cdata.alc_parent_tag, /* parent */ ALC_RR_RING_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_RR_RING_SZ, /* maxsize */ 1, /* nsegments */ ALC_RR_RING_SZ, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_rr_ring_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create Rx return ring DMA tag.\n"); goto fail; } /* Create DMA tag for coalescing message block. */ error = bus_dma_tag_create( sc->alc_cdata.alc_parent_tag, /* parent */ ALC_CMB_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_CMB_SZ, /* maxsize */ 1, /* nsegments */ ALC_CMB_SZ, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_cmb_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create CMB DMA tag.\n"); goto fail; } /* Create DMA tag for status message block. */ error = bus_dma_tag_create( sc->alc_cdata.alc_parent_tag, /* parent */ ALC_SMB_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_SMB_SZ, /* maxsize */ 1, /* nsegments */ ALC_SMB_SZ, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_smb_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create SMB DMA tag.\n"); goto fail; } /* Allocate DMA'able memory and load the DMA map for Tx ring. */ error = bus_dmamem_alloc(sc->alc_cdata.alc_tx_ring_tag, (void **)&sc->alc_rdata.alc_tx_ring, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->alc_cdata.alc_tx_ring_map); if (error != 0) { device_printf(sc->alc_dev, "could not allocate DMA'able memory for Tx ring.\n"); goto fail; } ctx.alc_busaddr = 0; error = bus_dmamap_load(sc->alc_cdata.alc_tx_ring_tag, sc->alc_cdata.alc_tx_ring_map, sc->alc_rdata.alc_tx_ring, ALC_TX_RING_SZ, alc_dmamap_cb, &ctx, 0); if (error != 0 || ctx.alc_busaddr == 0) { device_printf(sc->alc_dev, "could not load DMA'able memory for Tx ring.\n"); goto fail; } sc->alc_rdata.alc_tx_ring_paddr = ctx.alc_busaddr; /* Allocate DMA'able memory and load the DMA map for Rx ring. */ error = bus_dmamem_alloc(sc->alc_cdata.alc_rx_ring_tag, (void **)&sc->alc_rdata.alc_rx_ring, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->alc_cdata.alc_rx_ring_map); if (error != 0) { device_printf(sc->alc_dev, "could not allocate DMA'able memory for Rx ring.\n"); goto fail; } ctx.alc_busaddr = 0; error = bus_dmamap_load(sc->alc_cdata.alc_rx_ring_tag, sc->alc_cdata.alc_rx_ring_map, sc->alc_rdata.alc_rx_ring, ALC_RX_RING_SZ, alc_dmamap_cb, &ctx, 0); if (error != 0 || ctx.alc_busaddr == 0) { device_printf(sc->alc_dev, "could not load DMA'able memory for Rx ring.\n"); goto fail; } sc->alc_rdata.alc_rx_ring_paddr = ctx.alc_busaddr; /* Allocate DMA'able memory and load the DMA map for Rx return ring. */ error = bus_dmamem_alloc(sc->alc_cdata.alc_rr_ring_tag, (void **)&sc->alc_rdata.alc_rr_ring, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->alc_cdata.alc_rr_ring_map); if (error != 0) { device_printf(sc->alc_dev, "could not allocate DMA'able memory for Rx return ring.\n"); goto fail; } ctx.alc_busaddr = 0; error = bus_dmamap_load(sc->alc_cdata.alc_rr_ring_tag, sc->alc_cdata.alc_rr_ring_map, sc->alc_rdata.alc_rr_ring, ALC_RR_RING_SZ, alc_dmamap_cb, &ctx, 0); if (error != 0 || ctx.alc_busaddr == 0) { device_printf(sc->alc_dev, "could not load DMA'able memory for Tx ring.\n"); goto fail; } sc->alc_rdata.alc_rr_ring_paddr = ctx.alc_busaddr; /* Allocate DMA'able memory and load the DMA map for CMB. */ error = bus_dmamem_alloc(sc->alc_cdata.alc_cmb_tag, (void **)&sc->alc_rdata.alc_cmb, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->alc_cdata.alc_cmb_map); if (error != 0) { device_printf(sc->alc_dev, "could not allocate DMA'able memory for CMB.\n"); goto fail; } ctx.alc_busaddr = 0; error = bus_dmamap_load(sc->alc_cdata.alc_cmb_tag, sc->alc_cdata.alc_cmb_map, sc->alc_rdata.alc_cmb, ALC_CMB_SZ, alc_dmamap_cb, &ctx, 0); if (error != 0 || ctx.alc_busaddr == 0) { device_printf(sc->alc_dev, "could not load DMA'able memory for CMB.\n"); goto fail; } sc->alc_rdata.alc_cmb_paddr = ctx.alc_busaddr; /* Allocate DMA'able memory and load the DMA map for SMB. */ error = bus_dmamem_alloc(sc->alc_cdata.alc_smb_tag, (void **)&sc->alc_rdata.alc_smb, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->alc_cdata.alc_smb_map); if (error != 0) { device_printf(sc->alc_dev, "could not allocate DMA'able memory for SMB.\n"); goto fail; } ctx.alc_busaddr = 0; error = bus_dmamap_load(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, sc->alc_rdata.alc_smb, ALC_SMB_SZ, alc_dmamap_cb, &ctx, 0); if (error != 0 || ctx.alc_busaddr == 0) { device_printf(sc->alc_dev, "could not load DMA'able memory for CMB.\n"); goto fail; } sc->alc_rdata.alc_smb_paddr = ctx.alc_busaddr; /* Make sure we've not crossed 4GB boundary. */ if (lowaddr != BUS_SPACE_MAXADDR_32BIT && (error = alc_check_boundary(sc)) != 0) { device_printf(sc->alc_dev, "4GB boundary crossed, " "switching to 32bit DMA addressing mode.\n"); alc_dma_free(sc); /* * Limit max allowable DMA address space to 32bit * and try again. */ lowaddr = BUS_SPACE_MAXADDR_32BIT; goto again; } /* * Create Tx buffer parent tag. * AR813x/AR815x allows 64bit DMA addressing of Tx/Rx buffers * so it needs separate parent DMA tag as parent DMA address * space could be restricted to be within 32bit address space * by 4GB boundary crossing. */ error = bus_dma_tag_create( bus_get_dma_tag(sc->alc_dev), /* parent */ 1, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT, /* maxsize */ 0, /* nsegments */ BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_buffer_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create parent buffer DMA tag.\n"); goto fail; } /* Create DMA tag for Tx buffers. */ error = bus_dma_tag_create( sc->alc_cdata.alc_buffer_tag, /* parent */ 1, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ ALC_TSO_MAXSIZE, /* maxsize */ ALC_MAXTXSEGS, /* nsegments */ ALC_TSO_MAXSEGSIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_tx_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create Tx DMA tag.\n"); goto fail; } /* Create DMA tag for Rx buffers. */ error = bus_dma_tag_create( sc->alc_cdata.alc_buffer_tag, /* parent */ ALC_RX_BUF_ALIGN, 0, /* alignment, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ MCLBYTES, /* maxsize */ 1, /* nsegments */ MCLBYTES, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->alc_cdata.alc_rx_tag); if (error != 0) { device_printf(sc->alc_dev, "could not create Rx DMA tag.\n"); goto fail; } /* Create DMA maps for Tx buffers. */ for (i = 0; i < ALC_TX_RING_CNT; i++) { txd = &sc->alc_cdata.alc_txdesc[i]; txd->tx_m = NULL; txd->tx_dmamap = NULL; error = bus_dmamap_create(sc->alc_cdata.alc_tx_tag, 0, &txd->tx_dmamap); if (error != 0) { device_printf(sc->alc_dev, "could not create Tx dmamap.\n"); goto fail; } } /* Create DMA maps for Rx buffers. */ if ((error = bus_dmamap_create(sc->alc_cdata.alc_rx_tag, 0, &sc->alc_cdata.alc_rx_sparemap)) != 0) { device_printf(sc->alc_dev, "could not create spare Rx dmamap.\n"); goto fail; } for (i = 0; i < ALC_RX_RING_CNT; i++) { rxd = &sc->alc_cdata.alc_rxdesc[i]; rxd->rx_m = NULL; rxd->rx_dmamap = NULL; error = bus_dmamap_create(sc->alc_cdata.alc_rx_tag, 0, &rxd->rx_dmamap); if (error != 0) { device_printf(sc->alc_dev, "could not create Rx dmamap.\n"); goto fail; } } fail: return (error); } static void alc_dma_free(struct alc_softc *sc) { struct alc_txdesc *txd; struct alc_rxdesc *rxd; int i; /* Tx buffers. */ if (sc->alc_cdata.alc_tx_tag != NULL) { for (i = 0; i < ALC_TX_RING_CNT; i++) { txd = &sc->alc_cdata.alc_txdesc[i]; if (txd->tx_dmamap != NULL) { bus_dmamap_destroy(sc->alc_cdata.alc_tx_tag, txd->tx_dmamap); txd->tx_dmamap = NULL; } } bus_dma_tag_destroy(sc->alc_cdata.alc_tx_tag); sc->alc_cdata.alc_tx_tag = NULL; } /* Rx buffers */ if (sc->alc_cdata.alc_rx_tag != NULL) { for (i = 0; i < ALC_RX_RING_CNT; i++) { rxd = &sc->alc_cdata.alc_rxdesc[i]; if (rxd->rx_dmamap != NULL) { bus_dmamap_destroy(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap); rxd->rx_dmamap = NULL; } } if (sc->alc_cdata.alc_rx_sparemap != NULL) { bus_dmamap_destroy(sc->alc_cdata.alc_rx_tag, sc->alc_cdata.alc_rx_sparemap); sc->alc_cdata.alc_rx_sparemap = NULL; } bus_dma_tag_destroy(sc->alc_cdata.alc_rx_tag); sc->alc_cdata.alc_rx_tag = NULL; } /* Tx descriptor ring. */ if (sc->alc_cdata.alc_tx_ring_tag != NULL) { if (sc->alc_cdata.alc_tx_ring_map != NULL) bus_dmamap_unload(sc->alc_cdata.alc_tx_ring_tag, sc->alc_cdata.alc_tx_ring_map); if (sc->alc_cdata.alc_tx_ring_map != NULL && sc->alc_rdata.alc_tx_ring != NULL) bus_dmamem_free(sc->alc_cdata.alc_tx_ring_tag, sc->alc_rdata.alc_tx_ring, sc->alc_cdata.alc_tx_ring_map); sc->alc_rdata.alc_tx_ring = NULL; sc->alc_cdata.alc_tx_ring_map = NULL; bus_dma_tag_destroy(sc->alc_cdata.alc_tx_ring_tag); sc->alc_cdata.alc_tx_ring_tag = NULL; } /* Rx ring. */ if (sc->alc_cdata.alc_rx_ring_tag != NULL) { if (sc->alc_cdata.alc_rx_ring_map != NULL) bus_dmamap_unload(sc->alc_cdata.alc_rx_ring_tag, sc->alc_cdata.alc_rx_ring_map); if (sc->alc_cdata.alc_rx_ring_map != NULL && sc->alc_rdata.alc_rx_ring != NULL) bus_dmamem_free(sc->alc_cdata.alc_rx_ring_tag, sc->alc_rdata.alc_rx_ring, sc->alc_cdata.alc_rx_ring_map); sc->alc_rdata.alc_rx_ring = NULL; sc->alc_cdata.alc_rx_ring_map = NULL; bus_dma_tag_destroy(sc->alc_cdata.alc_rx_ring_tag); sc->alc_cdata.alc_rx_ring_tag = NULL; } /* Rx return ring. */ if (sc->alc_cdata.alc_rr_ring_tag != NULL) { if (sc->alc_cdata.alc_rr_ring_map != NULL) bus_dmamap_unload(sc->alc_cdata.alc_rr_ring_tag, sc->alc_cdata.alc_rr_ring_map); if (sc->alc_cdata.alc_rr_ring_map != NULL && sc->alc_rdata.alc_rr_ring != NULL) bus_dmamem_free(sc->alc_cdata.alc_rr_ring_tag, sc->alc_rdata.alc_rr_ring, sc->alc_cdata.alc_rr_ring_map); sc->alc_rdata.alc_rr_ring = NULL; sc->alc_cdata.alc_rr_ring_map = NULL; bus_dma_tag_destroy(sc->alc_cdata.alc_rr_ring_tag); sc->alc_cdata.alc_rr_ring_tag = NULL; } /* CMB block */ if (sc->alc_cdata.alc_cmb_tag != NULL) { if (sc->alc_cdata.alc_cmb_map != NULL) bus_dmamap_unload(sc->alc_cdata.alc_cmb_tag, sc->alc_cdata.alc_cmb_map); if (sc->alc_cdata.alc_cmb_map != NULL && sc->alc_rdata.alc_cmb != NULL) bus_dmamem_free(sc->alc_cdata.alc_cmb_tag, sc->alc_rdata.alc_cmb, sc->alc_cdata.alc_cmb_map); sc->alc_rdata.alc_cmb = NULL; sc->alc_cdata.alc_cmb_map = NULL; bus_dma_tag_destroy(sc->alc_cdata.alc_cmb_tag); sc->alc_cdata.alc_cmb_tag = NULL; } /* SMB block */ if (sc->alc_cdata.alc_smb_tag != NULL) { if (sc->alc_cdata.alc_smb_map != NULL) bus_dmamap_unload(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map); if (sc->alc_cdata.alc_smb_map != NULL && sc->alc_rdata.alc_smb != NULL) bus_dmamem_free(sc->alc_cdata.alc_smb_tag, sc->alc_rdata.alc_smb, sc->alc_cdata.alc_smb_map); sc->alc_rdata.alc_smb = NULL; sc->alc_cdata.alc_smb_map = NULL; bus_dma_tag_destroy(sc->alc_cdata.alc_smb_tag); sc->alc_cdata.alc_smb_tag = NULL; } if (sc->alc_cdata.alc_buffer_tag != NULL) { bus_dma_tag_destroy(sc->alc_cdata.alc_buffer_tag); sc->alc_cdata.alc_buffer_tag = NULL; } if (sc->alc_cdata.alc_parent_tag != NULL) { bus_dma_tag_destroy(sc->alc_cdata.alc_parent_tag); sc->alc_cdata.alc_parent_tag = NULL; } } static int alc_shutdown(device_t dev) { return (alc_suspend(dev)); } /* * Note, this driver resets the link speed to 10/100Mbps by * restarting auto-negotiation in suspend/shutdown phase but we * don't know whether that auto-negotiation would succeed or not * as driver has no control after powering off/suspend operation. * If the renegotiation fail WOL may not work. Running at 1Gbps * will draw more power than 375mA at 3.3V which is specified in * PCI specification and that would result in complete * shutdowning power to ethernet controller. * * TODO * Save current negotiated media speed/duplex/flow-control to * softc and restore the same link again after resuming. PHY * handling such as power down/resetting to 100Mbps may be better * handled in suspend method in phy driver. */ static void alc_setlinkspeed(struct alc_softc *sc) { struct mii_data *mii; int aneg, i; mii = device_get_softc(sc->alc_miibus); mii_pollstat(mii); aneg = 0; if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch IFM_SUBTYPE(mii->mii_media_active) { case IFM_10_T: case IFM_100_TX: return; case IFM_1000_T: aneg++; break; default: break; } } alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, MII_100T2CR, 0); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA); alc_miibus_writereg(sc->alc_dev, sc->alc_phyaddr, MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG); DELAY(1000); if (aneg != 0) { /* * Poll link state until alc(4) get a 10/100Mbps link. */ for (i = 0; i < MII_ANEGTICKS_GIGE; i++) { mii_pollstat(mii); if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE( mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: alc_mac_config(sc); return; default: break; } } ALC_UNLOCK(sc); pause("alclnk", hz); ALC_LOCK(sc); } if (i == MII_ANEGTICKS_GIGE) device_printf(sc->alc_dev, "establishing a link failed, WOL may not work!"); } /* * No link, force MAC to have 100Mbps, full-duplex link. * This is the last resort and may/may not work. */ mii->mii_media_status = IFM_AVALID | IFM_ACTIVE; mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX; alc_mac_config(sc); } static void alc_setwol(struct alc_softc *sc) { struct ifnet *ifp; uint32_t reg, pmcs; uint16_t pmstat; ALC_LOCK_ASSERT(sc); alc_disable_l0s_l1(sc); ifp = sc->alc_ifp; if ((sc->alc_flags & ALC_FLAG_PM) == 0) { /* Disable WOL. */ CSR_WRITE_4(sc, ALC_WOL_CFG, 0); reg = CSR_READ_4(sc, ALC_PCIE_PHYMISC); reg |= PCIE_PHYMISC_FORCE_RCV_DET; CSR_WRITE_4(sc, ALC_PCIE_PHYMISC, reg); /* Force PHY power down. */ alc_phy_down(sc); CSR_WRITE_4(sc, ALC_MASTER_CFG, CSR_READ_4(sc, ALC_MASTER_CFG) | MASTER_CLK_SEL_DIS); return; } if ((ifp->if_capenable & IFCAP_WOL) != 0) { if ((sc->alc_flags & ALC_FLAG_FASTETHER) == 0) alc_setlinkspeed(sc); CSR_WRITE_4(sc, ALC_MASTER_CFG, CSR_READ_4(sc, ALC_MASTER_CFG) & ~MASTER_CLK_SEL_DIS); } pmcs = 0; if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0) pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB; CSR_WRITE_4(sc, ALC_WOL_CFG, pmcs); reg = CSR_READ_4(sc, ALC_MAC_CFG); reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI | MAC_CFG_BCAST); if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0) reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST; if ((ifp->if_capenable & IFCAP_WOL) != 0) reg |= MAC_CFG_RX_ENB; CSR_WRITE_4(sc, ALC_MAC_CFG, reg); reg = CSR_READ_4(sc, ALC_PCIE_PHYMISC); reg |= PCIE_PHYMISC_FORCE_RCV_DET; CSR_WRITE_4(sc, ALC_PCIE_PHYMISC, reg); if ((ifp->if_capenable & IFCAP_WOL) == 0) { /* WOL disabled, PHY power down. */ alc_phy_down(sc); CSR_WRITE_4(sc, ALC_MASTER_CFG, CSR_READ_4(sc, ALC_MASTER_CFG) | MASTER_CLK_SEL_DIS); } /* Request PME. */ pmstat = pci_read_config(sc->alc_dev, sc->alc_pmcap + PCIR_POWER_STATUS, 2); pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE); if ((ifp->if_capenable & IFCAP_WOL) != 0) pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE; pci_write_config(sc->alc_dev, sc->alc_pmcap + PCIR_POWER_STATUS, pmstat, 2); } static int alc_suspend(device_t dev) { struct alc_softc *sc; sc = device_get_softc(dev); ALC_LOCK(sc); alc_stop(sc); alc_setwol(sc); ALC_UNLOCK(sc); return (0); } static int alc_resume(device_t dev) { struct alc_softc *sc; struct ifnet *ifp; uint16_t pmstat; sc = device_get_softc(dev); ALC_LOCK(sc); if ((sc->alc_flags & ALC_FLAG_PM) != 0) { /* Disable PME and clear PME status. */ pmstat = pci_read_config(sc->alc_dev, sc->alc_pmcap + PCIR_POWER_STATUS, 2); if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) { pmstat &= ~PCIM_PSTAT_PMEENABLE; pci_write_config(sc->alc_dev, sc->alc_pmcap + PCIR_POWER_STATUS, pmstat, 2); } } /* Reset PHY. */ alc_phy_reset(sc); ifp = sc->alc_ifp; if ((ifp->if_flags & IFF_UP) != 0) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; alc_init_locked(sc); } ALC_UNLOCK(sc); return (0); } static int alc_encap(struct alc_softc *sc, struct mbuf **m_head) { struct alc_txdesc *txd, *txd_last; struct tx_desc *desc; struct mbuf *m; struct ip *ip; struct tcphdr *tcp; bus_dma_segment_t txsegs[ALC_MAXTXSEGS]; bus_dmamap_t map; uint32_t cflags, hdrlen, ip_off, poff, vtag; int error, idx, nsegs, prod; ALC_LOCK_ASSERT(sc); M_ASSERTPKTHDR((*m_head)); m = *m_head; ip = NULL; tcp = NULL; ip_off = poff = 0; if ((m->m_pkthdr.csum_flags & (ALC_CSUM_FEATURES | CSUM_TSO)) != 0) { /* * AR813x/AR815x requires offset of TCP/UDP header in its * Tx descriptor to perform Tx checksum offloading. TSO * also requires TCP header offset and modification of * IP/TCP header. This kind of operation takes many CPU * cycles on FreeBSD so fast host CPU is required to get * smooth TSO performance. */ struct ether_header *eh; if (M_WRITABLE(m) == 0) { /* Get a writable copy. */ m = m_dup(*m_head, M_DONTWAIT); /* Release original mbufs. */ m_freem(*m_head); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } *m_head = m; } ip_off = sizeof(struct ether_header); m = m_pullup(m, ip_off); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } eh = mtod(m, struct ether_header *); /* * Check if hardware VLAN insertion is off. * Additional check for LLC/SNAP frame? */ if (eh->ether_type == htons(ETHERTYPE_VLAN)) { ip_off = sizeof(struct ether_vlan_header); m = m_pullup(m, ip_off); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } } m = m_pullup(m, ip_off + sizeof(struct ip)); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m, char *) + ip_off); poff = ip_off + (ip->ip_hl << 2); if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) { m = m_pullup(m, poff + sizeof(struct tcphdr)); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } tcp = (struct tcphdr *)(mtod(m, char *) + poff); m = m_pullup(m, poff + (tcp->th_off << 2)); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } /* * Due to strict adherence of Microsoft NDIS * Large Send specification, hardware expects * a pseudo TCP checksum inserted by upper * stack. Unfortunately the pseudo TCP * checksum that NDIS refers to does not include * TCP payload length so driver should recompute * the pseudo checksum here. Hopefully this * wouldn't be much burden on modern CPUs. * * Reset IP checksum and recompute TCP pseudo * checksum as NDIS specification said. */ ip = (struct ip *)(mtod(m, char *) + ip_off); tcp = (struct tcphdr *)(mtod(m, char *) + poff); ip->ip_sum = 0; tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); } *m_head = m; } prod = sc->alc_cdata.alc_tx_prod; txd = &sc->alc_cdata.alc_txdesc[prod]; txd_last = txd; map = txd->tx_dmamap; error = bus_dmamap_load_mbuf_sg(sc->alc_cdata.alc_tx_tag, map, *m_head, txsegs, &nsegs, 0); if (error == EFBIG) { m = m_collapse(*m_head, M_DONTWAIT, ALC_MAXTXSEGS); if (m == NULL) { m_freem(*m_head); *m_head = NULL; return (ENOMEM); } *m_head = m; error = bus_dmamap_load_mbuf_sg(sc->alc_cdata.alc_tx_tag, map, *m_head, txsegs, &nsegs, 0); if (error != 0) { m_freem(*m_head); *m_head = NULL; return (error); } } else if (error != 0) return (error); if (nsegs == 0) { m_freem(*m_head); *m_head = NULL; return (EIO); } /* Check descriptor overrun. */ if (sc->alc_cdata.alc_tx_cnt + nsegs >= ALC_TX_RING_CNT - 3) { bus_dmamap_unload(sc->alc_cdata.alc_tx_tag, map); return (ENOBUFS); } bus_dmamap_sync(sc->alc_cdata.alc_tx_tag, map, BUS_DMASYNC_PREWRITE); m = *m_head; cflags = TD_ETHERNET; vtag = 0; desc = NULL; idx = 0; /* Configure VLAN hardware tag insertion. */ if ((m->m_flags & M_VLANTAG) != 0) { vtag = htons(m->m_pkthdr.ether_vtag); vtag = (vtag << TD_VLAN_SHIFT) & TD_VLAN_MASK; cflags |= TD_INS_VLAN_TAG; } if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) { /* Request TSO and set MSS. */ cflags |= TD_TSO | TD_TSO_DESCV1; cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << TD_MSS_SHIFT) & TD_MSS_MASK; /* Set TCP header offset. */ cflags |= (poff << TD_TCPHDR_OFFSET_SHIFT) & TD_TCPHDR_OFFSET_MASK; /* * AR813x/AR815x requires the first buffer should * only hold IP/TCP header data. Payload should * be handled in other descriptors. */ hdrlen = poff + (tcp->th_off << 2); desc = &sc->alc_rdata.alc_tx_ring[prod]; desc->len = htole32(TX_BYTES(hdrlen | vtag)); desc->flags = htole32(cflags); desc->addr = htole64(txsegs[0].ds_addr); sc->alc_cdata.alc_tx_cnt++; ALC_DESC_INC(prod, ALC_TX_RING_CNT); if (m->m_len - hdrlen > 0) { /* Handle remaining payload of the first fragment. */ desc = &sc->alc_rdata.alc_tx_ring[prod]; desc->len = htole32(TX_BYTES((m->m_len - hdrlen) | vtag)); desc->flags = htole32(cflags); desc->addr = htole64(txsegs[0].ds_addr + hdrlen); sc->alc_cdata.alc_tx_cnt++; ALC_DESC_INC(prod, ALC_TX_RING_CNT); } /* Handle remaining fragments. */ idx = 1; } else if ((m->m_pkthdr.csum_flags & ALC_CSUM_FEATURES) != 0) { /* Configure Tx checksum offload. */ #ifdef ALC_USE_CUSTOM_CSUM cflags |= TD_CUSTOM_CSUM; /* Set checksum start offset. */ cflags |= ((poff >> 1) << TD_PLOAD_OFFSET_SHIFT) & TD_PLOAD_OFFSET_MASK; /* Set checksum insertion position of TCP/UDP. */ cflags |= (((poff + m->m_pkthdr.csum_data) >> 1) << TD_CUSTOM_CSUM_OFFSET_SHIFT) & TD_CUSTOM_CSUM_OFFSET_MASK; #else if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0) cflags |= TD_IPCSUM; if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0) cflags |= TD_TCPCSUM; if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0) cflags |= TD_UDPCSUM; /* Set TCP/UDP header offset. */ cflags |= (poff << TD_L4HDR_OFFSET_SHIFT) & TD_L4HDR_OFFSET_MASK; #endif } for (; idx < nsegs; idx++) { desc = &sc->alc_rdata.alc_tx_ring[prod]; desc->len = htole32(TX_BYTES(txsegs[idx].ds_len) | vtag); desc->flags = htole32(cflags); desc->addr = htole64(txsegs[idx].ds_addr); sc->alc_cdata.alc_tx_cnt++; ALC_DESC_INC(prod, ALC_TX_RING_CNT); } /* Update producer index. */ sc->alc_cdata.alc_tx_prod = prod; /* Finally set EOP on the last descriptor. */ prod = (prod + ALC_TX_RING_CNT - 1) % ALC_TX_RING_CNT; desc = &sc->alc_rdata.alc_tx_ring[prod]; desc->flags |= htole32(TD_EOP); /* Swap dmamap of the first and the last. */ txd = &sc->alc_cdata.alc_txdesc[prod]; map = txd_last->tx_dmamap; txd_last->tx_dmamap = txd->tx_dmamap; txd->tx_dmamap = map; txd->tx_m = m; return (0); } static void alc_start(struct ifnet *ifp) { struct alc_softc *sc; sc = ifp->if_softc; ALC_LOCK(sc); alc_start_locked(ifp); ALC_UNLOCK(sc); } static void alc_start_locked(struct ifnet *ifp) { struct alc_softc *sc; struct mbuf *m_head; int enq; sc = ifp->if_softc; ALC_LOCK_ASSERT(sc); /* Reclaim transmitted frames. */ if (sc->alc_cdata.alc_tx_cnt >= ALC_TX_DESC_HIWAT) alc_txeof(sc); if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING || (sc->alc_flags & ALC_FLAG_LINK) == 0) return; for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) { IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; /* * Pack the data into the transmit ring. If we * don't have room, set the OACTIVE flag and wait * for the NIC to drain the ring. */ if (alc_encap(sc, &m_head)) { if (m_head == NULL) break; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } enq++; /* * If there's a BPF listener, bounce a copy of this frame * to him. */ ETHER_BPF_MTAP(ifp, m_head); } if (enq > 0) { /* Sync descriptors. */ bus_dmamap_sync(sc->alc_cdata.alc_tx_ring_tag, sc->alc_cdata.alc_tx_ring_map, BUS_DMASYNC_PREWRITE); /* Kick. Assume we're using normal Tx priority queue. */ CSR_WRITE_4(sc, ALC_MBOX_TD_PROD_IDX, (sc->alc_cdata.alc_tx_prod << MBOX_TD_PROD_LO_IDX_SHIFT) & MBOX_TD_PROD_LO_IDX_MASK); /* Set a timeout in case the chip goes out to lunch. */ sc->alc_watchdog_timer = ALC_TX_TIMEOUT; } } static void alc_watchdog(struct alc_softc *sc) { struct ifnet *ifp; ALC_LOCK_ASSERT(sc); if (sc->alc_watchdog_timer == 0 || --sc->alc_watchdog_timer) return; ifp = sc->alc_ifp; if ((sc->alc_flags & ALC_FLAG_LINK) == 0) { if_printf(sc->alc_ifp, "watchdog timeout (lost link)\n"); ifp->if_oerrors++; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; alc_init_locked(sc); return; } if_printf(sc->alc_ifp, "watchdog timeout -- resetting\n"); ifp->if_oerrors++; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; alc_init_locked(sc); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) alc_start_locked(ifp); } static int alc_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct alc_softc *sc; struct ifreq *ifr; struct mii_data *mii; int error, mask; sc = ifp->if_softc; ifr = (struct ifreq *)data; error = 0; switch (cmd) { case SIOCSIFMTU: if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > (sc->alc_ident->max_framelen - sizeof(struct ether_vlan_header) - ETHER_CRC_LEN) || ((sc->alc_flags & ALC_FLAG_JUMBO) == 0 && ifr->ifr_mtu > ETHERMTU)) error = EINVAL; else if (ifp->if_mtu != ifr->ifr_mtu) { ALC_LOCK(sc); ifp->if_mtu = ifr->ifr_mtu; /* AR813x/AR815x has 13 bits MSS field. */ if (ifp->if_mtu > ALC_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) { ifp->if_capenable &= ~IFCAP_TSO4; ifp->if_hwassist &= ~CSUM_TSO; VLAN_CAPABILITIES(ifp); } ALC_UNLOCK(sc); } break; case SIOCSIFFLAGS: ALC_LOCK(sc); if ((ifp->if_flags & IFF_UP) != 0) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 && ((ifp->if_flags ^ sc->alc_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) alc_rxfilter(sc); else alc_init_locked(sc); } else if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) alc_stop(sc); sc->alc_if_flags = ifp->if_flags; ALC_UNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: ALC_LOCK(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) alc_rxfilter(sc); ALC_UNLOCK(sc); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: mii = device_get_softc(sc->alc_miibus); error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd); break; case SIOCSIFCAP: ALC_LOCK(sc); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if ((mask & IFCAP_TXCSUM) != 0 && (ifp->if_capabilities & IFCAP_TXCSUM) != 0) { ifp->if_capenable ^= IFCAP_TXCSUM; if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) ifp->if_hwassist |= ALC_CSUM_FEATURES; else ifp->if_hwassist &= ~ALC_CSUM_FEATURES; } if ((mask & IFCAP_TSO4) != 0 && (ifp->if_capabilities & IFCAP_TSO4) != 0) { ifp->if_capenable ^= IFCAP_TSO4; if ((ifp->if_capenable & IFCAP_TSO4) != 0) { /* AR813x/AR815x has 13 bits MSS field. */ if (ifp->if_mtu > ALC_TSO_MTU) { ifp->if_capenable &= ~IFCAP_TSO4; ifp->if_hwassist &= ~CSUM_TSO; } else ifp->if_hwassist |= CSUM_TSO; } else ifp->if_hwassist &= ~CSUM_TSO; } if ((mask & IFCAP_WOL_MCAST) != 0 && (ifp->if_capabilities & IFCAP_WOL_MCAST) != 0) ifp->if_capenable ^= IFCAP_WOL_MCAST; if ((mask & IFCAP_WOL_MAGIC) != 0 && (ifp->if_capabilities & IFCAP_WOL_MAGIC) != 0) ifp->if_capenable ^= IFCAP_WOL_MAGIC; if ((mask & IFCAP_VLAN_HWTAGGING) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; alc_rxvlan(sc); } if ((mask & IFCAP_VLAN_HWCSUM) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWCSUM; if ((mask & IFCAP_VLAN_HWTSO) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) ifp->if_capenable &= ~(IFCAP_VLAN_HWTSO | IFCAP_VLAN_HWCSUM); ALC_UNLOCK(sc); VLAN_CAPABILITIES(ifp); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } static void alc_mac_config(struct alc_softc *sc) { struct mii_data *mii; uint32_t reg; ALC_LOCK_ASSERT(sc); mii = device_get_softc(sc->alc_miibus); reg = CSR_READ_4(sc, ALC_MAC_CFG); reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC | MAC_CFG_SPEED_MASK); if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151_V2 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B2) reg |= MAC_CFG_HASH_ALG_CRC32 | MAC_CFG_SPEED_MODE_SW; /* Reprogram MAC with resolved speed/duplex. */ switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: reg |= MAC_CFG_SPEED_10_100; break; case IFM_1000_T: reg |= MAC_CFG_SPEED_1000; break; } if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) { reg |= MAC_CFG_FULL_DUPLEX; if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0) reg |= MAC_CFG_TX_FC; if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0) reg |= MAC_CFG_RX_FC; } CSR_WRITE_4(sc, ALC_MAC_CFG, reg); } static void alc_stats_clear(struct alc_softc *sc) { struct smb sb, *smb; uint32_t *reg; int i; if ((sc->alc_flags & ALC_FLAG_SMB_BUG) == 0) { bus_dmamap_sync(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); smb = sc->alc_rdata.alc_smb; /* Update done, clear. */ smb->updated = 0; bus_dmamap_sync(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } else { for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) { CSR_READ_4(sc, ALC_RX_MIB_BASE + i); i += sizeof(uint32_t); } /* Read Tx statistics. */ for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) { CSR_READ_4(sc, ALC_TX_MIB_BASE + i); i += sizeof(uint32_t); } } } static void alc_stats_update(struct alc_softc *sc) { struct alc_hw_stats *stat; struct smb sb, *smb; struct ifnet *ifp; uint32_t *reg; int i; ALC_LOCK_ASSERT(sc); ifp = sc->alc_ifp; stat = &sc->alc_stats; if ((sc->alc_flags & ALC_FLAG_SMB_BUG) == 0) { bus_dmamap_sync(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); smb = sc->alc_rdata.alc_smb; if (smb->updated == 0) return; } else { smb = &sb; /* Read Rx statistics. */ for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) { *reg = CSR_READ_4(sc, ALC_RX_MIB_BASE + i); i += sizeof(uint32_t); } /* Read Tx statistics. */ for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) { *reg = CSR_READ_4(sc, ALC_TX_MIB_BASE + i); i += sizeof(uint32_t); } } /* Rx stats. */ stat->rx_frames += smb->rx_frames; stat->rx_bcast_frames += smb->rx_bcast_frames; stat->rx_mcast_frames += smb->rx_mcast_frames; stat->rx_pause_frames += smb->rx_pause_frames; stat->rx_control_frames += smb->rx_control_frames; stat->rx_crcerrs += smb->rx_crcerrs; stat->rx_lenerrs += smb->rx_lenerrs; stat->rx_bytes += smb->rx_bytes; stat->rx_runts += smb->rx_runts; stat->rx_fragments += smb->rx_fragments; stat->rx_pkts_64 += smb->rx_pkts_64; stat->rx_pkts_65_127 += smb->rx_pkts_65_127; stat->rx_pkts_128_255 += smb->rx_pkts_128_255; stat->rx_pkts_256_511 += smb->rx_pkts_256_511; stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023; stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518; stat->rx_pkts_1519_max += smb->rx_pkts_1519_max; stat->rx_pkts_truncated += smb->rx_pkts_truncated; stat->rx_fifo_oflows += smb->rx_fifo_oflows; stat->rx_rrs_errs += smb->rx_rrs_errs; stat->rx_alignerrs += smb->rx_alignerrs; stat->rx_bcast_bytes += smb->rx_bcast_bytes; stat->rx_mcast_bytes += smb->rx_mcast_bytes; stat->rx_pkts_filtered += smb->rx_pkts_filtered; /* Tx stats. */ stat->tx_frames += smb->tx_frames; stat->tx_bcast_frames += smb->tx_bcast_frames; stat->tx_mcast_frames += smb->tx_mcast_frames; stat->tx_pause_frames += smb->tx_pause_frames; stat->tx_excess_defer += smb->tx_excess_defer; stat->tx_control_frames += smb->tx_control_frames; stat->tx_deferred += smb->tx_deferred; stat->tx_bytes += smb->tx_bytes; stat->tx_pkts_64 += smb->tx_pkts_64; stat->tx_pkts_65_127 += smb->tx_pkts_65_127; stat->tx_pkts_128_255 += smb->tx_pkts_128_255; stat->tx_pkts_256_511 += smb->tx_pkts_256_511; stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023; stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518; stat->tx_pkts_1519_max += smb->tx_pkts_1519_max; stat->tx_single_colls += smb->tx_single_colls; stat->tx_multi_colls += smb->tx_multi_colls; stat->tx_late_colls += smb->tx_late_colls; stat->tx_excess_colls += smb->tx_excess_colls; stat->tx_abort += smb->tx_abort; stat->tx_underrun += smb->tx_underrun; stat->tx_desc_underrun += smb->tx_desc_underrun; stat->tx_lenerrs += smb->tx_lenerrs; stat->tx_pkts_truncated += smb->tx_pkts_truncated; stat->tx_bcast_bytes += smb->tx_bcast_bytes; stat->tx_mcast_bytes += smb->tx_mcast_bytes; /* Update counters in ifnet. */ ifp->if_opackets += smb->tx_frames; ifp->if_collisions += smb->tx_single_colls + smb->tx_multi_colls * 2 + smb->tx_late_colls + smb->tx_abort * HDPX_CFG_RETRY_DEFAULT; /* * XXX * tx_pkts_truncated counter looks suspicious. It constantly * increments with no sign of Tx errors. This may indicate * the counter name is not correct one so I've removed the * counter in output errors. */ ifp->if_oerrors += smb->tx_abort + smb->tx_late_colls + smb->tx_underrun; ifp->if_ipackets += smb->rx_frames; ifp->if_ierrors += smb->rx_crcerrs + smb->rx_lenerrs + smb->rx_runts + smb->rx_pkts_truncated + smb->rx_fifo_oflows + smb->rx_rrs_errs + smb->rx_alignerrs; if ((sc->alc_flags & ALC_FLAG_SMB_BUG) == 0) { /* Update done, clear. */ smb->updated = 0; bus_dmamap_sync(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } } static int alc_intr(void *arg) { struct alc_softc *sc; uint32_t status; sc = (struct alc_softc *)arg; status = CSR_READ_4(sc, ALC_INTR_STATUS); if ((status & ALC_INTRS) == 0) return (FILTER_STRAY); /* Disable interrupts. */ CSR_WRITE_4(sc, ALC_INTR_STATUS, INTR_DIS_INT); taskqueue_enqueue(sc->alc_tq, &sc->alc_int_task); return (FILTER_HANDLED); } static void alc_int_task(void *arg, int pending) { struct alc_softc *sc; struct ifnet *ifp; uint32_t status; int more; sc = (struct alc_softc *)arg; ifp = sc->alc_ifp; status = CSR_READ_4(sc, ALC_INTR_STATUS); ALC_LOCK(sc); if (sc->alc_morework != 0) { sc->alc_morework = 0; status |= INTR_RX_PKT; } if ((status & ALC_INTRS) == 0) goto done; /* Acknowledge interrupts but still disable interrupts. */ CSR_WRITE_4(sc, ALC_INTR_STATUS, status | INTR_DIS_INT); more = 0; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if ((status & INTR_RX_PKT) != 0) { more = alc_rxintr(sc, sc->alc_process_limit); if (more == EAGAIN) sc->alc_morework = 1; else if (more == EIO) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; alc_init_locked(sc); ALC_UNLOCK(sc); return; } } if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST | INTR_TXQ_TO_RST)) != 0) { if ((status & INTR_DMA_RD_TO_RST) != 0) device_printf(sc->alc_dev, "DMA read error! -- resetting\n"); if ((status & INTR_DMA_WR_TO_RST) != 0) device_printf(sc->alc_dev, "DMA write error! -- resetting\n"); if ((status & INTR_TXQ_TO_RST) != 0) device_printf(sc->alc_dev, "TxQ reset! -- resetting\n"); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; alc_init_locked(sc); ALC_UNLOCK(sc); return; } if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0 && !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) alc_start_locked(ifp); } if (more == EAGAIN || (CSR_READ_4(sc, ALC_INTR_STATUS) & ALC_INTRS) != 0) { ALC_UNLOCK(sc); taskqueue_enqueue(sc->alc_tq, &sc->alc_int_task); return; } done: if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { /* Re-enable interrupts if we're running. */ CSR_WRITE_4(sc, ALC_INTR_STATUS, 0x7FFFFFFF); } ALC_UNLOCK(sc); } static void alc_txeof(struct alc_softc *sc) { struct ifnet *ifp; struct alc_txdesc *txd; uint32_t cons, prod; int prog; ALC_LOCK_ASSERT(sc); ifp = sc->alc_ifp; if (sc->alc_cdata.alc_tx_cnt == 0) return; bus_dmamap_sync(sc->alc_cdata.alc_tx_ring_tag, sc->alc_cdata.alc_tx_ring_map, BUS_DMASYNC_POSTWRITE); if ((sc->alc_flags & ALC_FLAG_CMB_BUG) == 0) { bus_dmamap_sync(sc->alc_cdata.alc_cmb_tag, sc->alc_cdata.alc_cmb_map, BUS_DMASYNC_POSTREAD); prod = sc->alc_rdata.alc_cmb->cons; } else prod = CSR_READ_4(sc, ALC_MBOX_TD_CONS_IDX); /* Assume we're using normal Tx priority queue. */ prod = (prod & MBOX_TD_CONS_LO_IDX_MASK) >> MBOX_TD_CONS_LO_IDX_SHIFT; cons = sc->alc_cdata.alc_tx_cons; /* * Go through our Tx list and free mbufs for those * frames which have been transmitted. */ for (prog = 0; cons != prod; prog++, ALC_DESC_INC(cons, ALC_TX_RING_CNT)) { if (sc->alc_cdata.alc_tx_cnt <= 0) break; prog++; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; sc->alc_cdata.alc_tx_cnt--; txd = &sc->alc_cdata.alc_txdesc[cons]; if (txd->tx_m != NULL) { /* Reclaim transmitted mbufs. */ bus_dmamap_sync(sc->alc_cdata.alc_tx_tag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->alc_cdata.alc_tx_tag, txd->tx_dmamap); m_freem(txd->tx_m); txd->tx_m = NULL; } } if ((sc->alc_flags & ALC_FLAG_CMB_BUG) == 0) bus_dmamap_sync(sc->alc_cdata.alc_cmb_tag, sc->alc_cdata.alc_cmb_map, BUS_DMASYNC_PREREAD); sc->alc_cdata.alc_tx_cons = cons; /* * Unarm watchdog timer only when there is no pending * frames in Tx queue. */ if (sc->alc_cdata.alc_tx_cnt == 0) sc->alc_watchdog_timer = 0; } static int alc_newbuf(struct alc_softc *sc, struct alc_rxdesc *rxd) { struct mbuf *m; bus_dma_segment_t segs[1]; bus_dmamap_t map; int nsegs; m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = RX_BUF_SIZE_MAX; #ifndef __NO_STRICT_ALIGNMENT m_adj(m, sizeof(uint64_t)); #endif if (bus_dmamap_load_mbuf_sg(sc->alc_cdata.alc_rx_tag, sc->alc_cdata.alc_rx_sparemap, m, segs, &nsegs, 0) != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap); } map = rxd->rx_dmamap; rxd->rx_dmamap = sc->alc_cdata.alc_rx_sparemap; sc->alc_cdata.alc_rx_sparemap = map; bus_dmamap_sync(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_PREREAD); rxd->rx_m = m; rxd->rx_desc->addr = htole64(segs[0].ds_addr); return (0); } static int alc_rxintr(struct alc_softc *sc, int count) { struct ifnet *ifp; struct rx_rdesc *rrd; uint32_t nsegs, status; int rr_cons, prog; bus_dmamap_sync(sc->alc_cdata.alc_rr_ring_tag, sc->alc_cdata.alc_rr_ring_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_sync(sc->alc_cdata.alc_rx_ring_tag, sc->alc_cdata.alc_rx_ring_map, BUS_DMASYNC_POSTWRITE); rr_cons = sc->alc_cdata.alc_rr_cons; ifp = sc->alc_ifp; for (prog = 0; (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0;) { if (count-- <= 0) break; rrd = &sc->alc_rdata.alc_rr_ring[rr_cons]; status = le32toh(rrd->status); if ((status & RRD_VALID) == 0) break; nsegs = RRD_RD_CNT(le32toh(rrd->rdinfo)); if (nsegs == 0) { /* This should not happen! */ device_printf(sc->alc_dev, "unexpected segment count -- resetting\n"); return (EIO); } alc_rxeof(sc, rrd); /* Clear Rx return status. */ rrd->status = 0; ALC_DESC_INC(rr_cons, ALC_RR_RING_CNT); sc->alc_cdata.alc_rx_cons += nsegs; sc->alc_cdata.alc_rx_cons %= ALC_RR_RING_CNT; prog += nsegs; } if (prog > 0) { /* Update the consumer index. */ sc->alc_cdata.alc_rr_cons = rr_cons; /* Sync Rx return descriptors. */ bus_dmamap_sync(sc->alc_cdata.alc_rr_ring_tag, sc->alc_cdata.alc_rr_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* * Sync updated Rx descriptors such that controller see * modified buffer addresses. */ bus_dmamap_sync(sc->alc_cdata.alc_rx_ring_tag, sc->alc_cdata.alc_rx_ring_map, BUS_DMASYNC_PREWRITE); /* * Let controller know availability of new Rx buffers. * Since alc(4) use RXQ_CFG_RD_BURST_DEFAULT descriptors * it may be possible to update ALC_MBOX_RD0_PROD_IDX * only when Rx buffer pre-fetching is required. In * addition we already set ALC_RX_RD_FREE_THRESH to * RX_RD_FREE_THRESH_LO_DEFAULT descriptors. However * it still seems that pre-fetching needs more * experimentation. */ CSR_WRITE_4(sc, ALC_MBOX_RD0_PROD_IDX, sc->alc_cdata.alc_rx_cons); } return (count > 0 ? 0 : EAGAIN); } #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * alc_fixup_rx(struct ifnet *ifp, struct mbuf *m) { struct mbuf *n; int i; uint16_t *src, *dst; src = mtod(m, uint16_t *); dst = src - 3; if (m->m_next == NULL) { for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++) *dst++ = *src++; m->m_data -= 6; return (m); } /* * Append a new mbuf to received mbuf chain and copy ethernet * header from the mbuf chain. This can save lots of CPU * cycles for jumbo frame. */ MGETHDR(n, M_DONTWAIT, MT_DATA); if (n == NULL) { ifp->if_iqdrops++; m_freem(m); return (NULL); } bcopy(m->m_data, n->m_data, ETHER_HDR_LEN); m->m_data += ETHER_HDR_LEN; m->m_len -= ETHER_HDR_LEN; n->m_len = ETHER_HDR_LEN; M_MOVE_PKTHDR(n, m); n->m_next = m; return (n); } #endif /* Receive a frame. */ static void alc_rxeof(struct alc_softc *sc, struct rx_rdesc *rrd) { struct alc_rxdesc *rxd; struct ifnet *ifp; struct mbuf *mp, *m; uint32_t rdinfo, status, vtag; int count, nsegs, rx_cons; ifp = sc->alc_ifp; status = le32toh(rrd->status); rdinfo = le32toh(rrd->rdinfo); rx_cons = RRD_RD_IDX(rdinfo); nsegs = RRD_RD_CNT(rdinfo); sc->alc_cdata.alc_rxlen = RRD_BYTES(status); if ((status & (RRD_ERR_SUM | RRD_ERR_LENGTH)) != 0) { /* * We want to pass the following frames to upper * layer regardless of error status of Rx return * ring. * * o IP/TCP/UDP checksum is bad. * o frame length and protocol specific length * does not match. * * Force network stack compute checksum for * errored frames. */ status |= RRD_TCP_UDPCSUM_NOK | RRD_IPCSUM_NOK; if ((status & (RRD_ERR_CRC | RRD_ERR_ALIGN | RRD_ERR_TRUNC | RRD_ERR_RUNT)) != 0) return; } for (count = 0; count < nsegs; count++, ALC_DESC_INC(rx_cons, ALC_RX_RING_CNT)) { rxd = &sc->alc_cdata.alc_rxdesc[rx_cons]; mp = rxd->rx_m; /* Add a new receive buffer to the ring. */ if (alc_newbuf(sc, rxd) != 0) { ifp->if_iqdrops++; /* Reuse Rx buffers. */ if (sc->alc_cdata.alc_rxhead != NULL) m_freem(sc->alc_cdata.alc_rxhead); break; } /* * Assume we've received a full sized frame. * Actual size is fixed when we encounter the end of * multi-segmented frame. */ mp->m_len = sc->alc_buf_size; /* Chain received mbufs. */ if (sc->alc_cdata.alc_rxhead == NULL) { sc->alc_cdata.alc_rxhead = mp; sc->alc_cdata.alc_rxtail = mp; } else { mp->m_flags &= ~M_PKTHDR; sc->alc_cdata.alc_rxprev_tail = sc->alc_cdata.alc_rxtail; sc->alc_cdata.alc_rxtail->m_next = mp; sc->alc_cdata.alc_rxtail = mp; } if (count == nsegs - 1) { /* Last desc. for this frame. */ m = sc->alc_cdata.alc_rxhead; m->m_flags |= M_PKTHDR; /* * It seems that L1C/L2C controller has no way * to tell hardware to strip CRC bytes. */ m->m_pkthdr.len = sc->alc_cdata.alc_rxlen - ETHER_CRC_LEN; if (nsegs > 1) { /* Set last mbuf size. */ mp->m_len = sc->alc_cdata.alc_rxlen - (nsegs - 1) * sc->alc_buf_size; /* Remove the CRC bytes in chained mbufs. */ if (mp->m_len <= ETHER_CRC_LEN) { sc->alc_cdata.alc_rxtail = sc->alc_cdata.alc_rxprev_tail; sc->alc_cdata.alc_rxtail->m_len -= (ETHER_CRC_LEN - mp->m_len); sc->alc_cdata.alc_rxtail->m_next = NULL; m_freem(mp); } else { mp->m_len -= ETHER_CRC_LEN; } } else m->m_len = m->m_pkthdr.len; m->m_pkthdr.rcvif = ifp; /* * Due to hardware bugs, Rx checksum offloading * was intentionally disabled. */ if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 && (status & RRD_VLAN_TAG) != 0) { vtag = RRD_VLAN(le32toh(rrd->vtag)); m->m_pkthdr.ether_vtag = ntohs(vtag); m->m_flags |= M_VLANTAG; } #ifndef __NO_STRICT_ALIGNMENT m = alc_fixup_rx(ifp, m); if (m != NULL) #endif { /* Pass it on. */ ALC_UNLOCK(sc); (*ifp->if_input)(ifp, m); ALC_LOCK(sc); } } } /* Reset mbuf chains. */ ALC_RXCHAIN_RESET(sc); } static void alc_tick(void *arg) { struct alc_softc *sc; struct mii_data *mii; sc = (struct alc_softc *)arg; ALC_LOCK_ASSERT(sc); mii = device_get_softc(sc->alc_miibus); mii_tick(mii); alc_stats_update(sc); /* * alc(4) does not rely on Tx completion interrupts to reclaim * transferred buffers. Instead Tx completion interrupts are * used to hint for scheduling Tx task. So it's necessary to * release transmitted buffers by kicking Tx completion * handler. This limits the maximum reclamation delay to a hz. */ alc_txeof(sc); alc_watchdog(sc); callout_reset(&sc->alc_tick_ch, hz, alc_tick, sc); } static void alc_reset(struct alc_softc *sc) { uint32_t reg; int i; reg = CSR_READ_4(sc, ALC_MASTER_CFG) & 0xFFFF; reg |= MASTER_OOB_DIS_OFF | MASTER_RESET; CSR_WRITE_4(sc, ALC_MASTER_CFG, reg); for (i = ALC_RESET_TIMEOUT; i > 0; i--) { DELAY(10); if ((CSR_READ_4(sc, ALC_MASTER_CFG) & MASTER_RESET) == 0) break; } if (i == 0) device_printf(sc->alc_dev, "master reset timeout!\n"); for (i = ALC_RESET_TIMEOUT; i > 0; i--) { if ((reg = CSR_READ_4(sc, ALC_IDLE_STATUS)) == 0) break; DELAY(10); } if (i == 0) device_printf(sc->alc_dev, "reset timeout(0x%08x)!\n", reg); } static void alc_init(void *xsc) { struct alc_softc *sc; sc = (struct alc_softc *)xsc; ALC_LOCK(sc); alc_init_locked(sc); ALC_UNLOCK(sc); } static void alc_init_locked(struct alc_softc *sc) { struct ifnet *ifp; struct mii_data *mii; uint8_t eaddr[ETHER_ADDR_LEN]; bus_addr_t paddr; uint32_t reg, rxf_hi, rxf_lo; ALC_LOCK_ASSERT(sc); ifp = sc->alc_ifp; mii = device_get_softc(sc->alc_miibus); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) return; /* * Cancel any pending I/O. */ alc_stop(sc); /* * Reset the chip to a known state. */ alc_reset(sc); /* Initialize Rx descriptors. */ if (alc_init_rx_ring(sc) != 0) { device_printf(sc->alc_dev, "no memory for Rx buffers.\n"); alc_stop(sc); return; } alc_init_rr_ring(sc); alc_init_tx_ring(sc); alc_init_cmb(sc); alc_init_smb(sc); /* Enable all clocks. */ CSR_WRITE_4(sc, ALC_CLK_GATING_CFG, 0); /* Reprogram the station address. */ bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN); CSR_WRITE_4(sc, ALC_PAR0, eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]); CSR_WRITE_4(sc, ALC_PAR1, eaddr[0] << 8 | eaddr[1]); /* * Clear WOL status and disable all WOL feature as WOL * would interfere Rx operation under normal environments. */ CSR_READ_4(sc, ALC_WOL_CFG); CSR_WRITE_4(sc, ALC_WOL_CFG, 0); /* Set Tx descriptor base addresses. */ paddr = sc->alc_rdata.alc_tx_ring_paddr; CSR_WRITE_4(sc, ALC_TX_BASE_ADDR_HI, ALC_ADDR_HI(paddr)); CSR_WRITE_4(sc, ALC_TDL_HEAD_ADDR_LO, ALC_ADDR_LO(paddr)); /* We don't use high priority ring. */ CSR_WRITE_4(sc, ALC_TDH_HEAD_ADDR_LO, 0); /* Set Tx descriptor counter. */ CSR_WRITE_4(sc, ALC_TD_RING_CNT, (ALC_TX_RING_CNT << TD_RING_CNT_SHIFT) & TD_RING_CNT_MASK); /* Set Rx descriptor base addresses. */ paddr = sc->alc_rdata.alc_rx_ring_paddr; CSR_WRITE_4(sc, ALC_RX_BASE_ADDR_HI, ALC_ADDR_HI(paddr)); CSR_WRITE_4(sc, ALC_RD0_HEAD_ADDR_LO, ALC_ADDR_LO(paddr)); /* We use one Rx ring. */ CSR_WRITE_4(sc, ALC_RD1_HEAD_ADDR_LO, 0); CSR_WRITE_4(sc, ALC_RD2_HEAD_ADDR_LO, 0); CSR_WRITE_4(sc, ALC_RD3_HEAD_ADDR_LO, 0); /* Set Rx descriptor counter. */ CSR_WRITE_4(sc, ALC_RD_RING_CNT, (ALC_RX_RING_CNT << RD_RING_CNT_SHIFT) & RD_RING_CNT_MASK); /* * Let hardware split jumbo frames into alc_max_buf_sized chunks. * if it do not fit the buffer size. Rx return descriptor holds * a counter that indicates how many fragments were made by the * hardware. The buffer size should be multiple of 8 bytes. * Since hardware has limit on the size of buffer size, always * use the maximum value. * For strict-alignment architectures make sure to reduce buffer * size by 8 bytes to make room for alignment fixup. */ #ifndef __NO_STRICT_ALIGNMENT sc->alc_buf_size = RX_BUF_SIZE_MAX - sizeof(uint64_t); #else sc->alc_buf_size = RX_BUF_SIZE_MAX; #endif CSR_WRITE_4(sc, ALC_RX_BUF_SIZE, sc->alc_buf_size); paddr = sc->alc_rdata.alc_rr_ring_paddr; /* Set Rx return descriptor base addresses. */ CSR_WRITE_4(sc, ALC_RRD0_HEAD_ADDR_LO, ALC_ADDR_LO(paddr)); /* We use one Rx return ring. */ CSR_WRITE_4(sc, ALC_RRD1_HEAD_ADDR_LO, 0); CSR_WRITE_4(sc, ALC_RRD2_HEAD_ADDR_LO, 0); CSR_WRITE_4(sc, ALC_RRD3_HEAD_ADDR_LO, 0); /* Set Rx return descriptor counter. */ CSR_WRITE_4(sc, ALC_RRD_RING_CNT, (ALC_RR_RING_CNT << RRD_RING_CNT_SHIFT) & RRD_RING_CNT_MASK); paddr = sc->alc_rdata.alc_cmb_paddr; CSR_WRITE_4(sc, ALC_CMB_BASE_ADDR_LO, ALC_ADDR_LO(paddr)); paddr = sc->alc_rdata.alc_smb_paddr; CSR_WRITE_4(sc, ALC_SMB_BASE_ADDR_HI, ALC_ADDR_HI(paddr)); CSR_WRITE_4(sc, ALC_SMB_BASE_ADDR_LO, ALC_ADDR_LO(paddr)); if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B) { /* Reconfigure SRAM - Vendor magic. */ CSR_WRITE_4(sc, ALC_SRAM_RX_FIFO_LEN, 0x000002A0); CSR_WRITE_4(sc, ALC_SRAM_TX_FIFO_LEN, 0x00000100); CSR_WRITE_4(sc, ALC_SRAM_RX_FIFO_ADDR, 0x029F0000); CSR_WRITE_4(sc, ALC_SRAM_RD0_ADDR, 0x02BF02A0); CSR_WRITE_4(sc, ALC_SRAM_TX_FIFO_ADDR, 0x03BF02C0); CSR_WRITE_4(sc, ALC_SRAM_TD_ADDR, 0x03DF03C0); CSR_WRITE_4(sc, ALC_TXF_WATER_MARK, 0x00000000); CSR_WRITE_4(sc, ALC_RD_DMA_CFG, 0x00000000); } /* Tell hardware that we're ready to load DMA blocks. */ CSR_WRITE_4(sc, ALC_DMA_BLOCK, DMA_BLOCK_LOAD); /* Configure interrupt moderation timer. */ reg = ALC_USECS(sc->alc_int_rx_mod) << IM_TIMER_RX_SHIFT; reg |= ALC_USECS(sc->alc_int_tx_mod) << IM_TIMER_TX_SHIFT; CSR_WRITE_4(sc, ALC_IM_TIMER, reg); /* * We don't want to automatic interrupt clear as task queue * for the interrupt should know interrupt status. */ reg = MASTER_SA_TIMER_ENB; if (ALC_USECS(sc->alc_int_rx_mod) != 0) reg |= MASTER_IM_RX_TIMER_ENB; if (ALC_USECS(sc->alc_int_tx_mod) != 0) reg |= MASTER_IM_TX_TIMER_ENB; CSR_WRITE_4(sc, ALC_MASTER_CFG, reg); /* * Disable interrupt re-trigger timer. We don't want automatic * re-triggering of un-ACKed interrupts. */ CSR_WRITE_4(sc, ALC_INTR_RETRIG_TIMER, ALC_USECS(0)); /* Configure CMB. */ if ((sc->alc_flags & ALC_FLAG_CMB_BUG) == 0) { CSR_WRITE_4(sc, ALC_CMB_TD_THRESH, 4); CSR_WRITE_4(sc, ALC_CMB_TX_TIMER, ALC_USECS(5000)); } else CSR_WRITE_4(sc, ALC_CMB_TX_TIMER, ALC_USECS(0)); /* * Hardware can be configured to issue SMB interrupt based * on programmed interval. Since there is a callout that is * invoked for every hz in driver we use that instead of * relying on periodic SMB interrupt. */ CSR_WRITE_4(sc, ALC_SMB_STAT_TIMER, ALC_USECS(0)); /* Clear MAC statistics. */ alc_stats_clear(sc); /* * Always use maximum frame size that controller can support. * Otherwise received frames that has larger frame length * than alc(4) MTU would be silently dropped in hardware. This * would make path-MTU discovery hard as sender wouldn't get * any responses from receiver. alc(4) supports * multi-fragmented frames on Rx path so it has no issue on * assembling fragmented frames. Using maximum frame size also * removes the need to reinitialize hardware when interface * MTU configuration was changed. * * Be conservative in what you do, be liberal in what you * accept from others - RFC 793. */ CSR_WRITE_4(sc, ALC_FRAME_SIZE, sc->alc_ident->max_framelen); /* Disable header split(?) */ CSR_WRITE_4(sc, ALC_HDS_CFG, 0); /* Configure IPG/IFG parameters. */ CSR_WRITE_4(sc, ALC_IPG_IFG_CFG, ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) | ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) | ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) | ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK)); /* Set parameters for half-duplex media. */ CSR_WRITE_4(sc, ALC_HDPX_CFG, ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) & HDPX_CFG_LCOL_MASK) | ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) & HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN | ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) & HDPX_CFG_ABEBT_MASK) | ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) & HDPX_CFG_JAMIPG_MASK)); /* * Set TSO/checksum offload threshold. For frames that is * larger than this threshold, hardware wouldn't do * TSO/checksum offloading. */ CSR_WRITE_4(sc, ALC_TSO_OFFLOAD_THRESH, (sc->alc_ident->max_framelen >> TSO_OFFLOAD_THRESH_UNIT_SHIFT) & TSO_OFFLOAD_THRESH_MASK); /* Configure TxQ. */ reg = (alc_dma_burst[sc->alc_dma_rd_burst] << TXQ_CFG_TX_FIFO_BURST_SHIFT) & TXQ_CFG_TX_FIFO_BURST_MASK; if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B2) reg >>= 1; reg |= (TXQ_CFG_TD_BURST_DEFAULT << TXQ_CFG_TD_BURST_SHIFT) & TXQ_CFG_TD_BURST_MASK; CSR_WRITE_4(sc, ALC_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE); /* Configure Rx free descriptor pre-fetching. */ CSR_WRITE_4(sc, ALC_RX_RD_FREE_THRESH, ((RX_RD_FREE_THRESH_HI_DEFAULT << RX_RD_FREE_THRESH_HI_SHIFT) & RX_RD_FREE_THRESH_HI_MASK) | ((RX_RD_FREE_THRESH_LO_DEFAULT << RX_RD_FREE_THRESH_LO_SHIFT) & RX_RD_FREE_THRESH_LO_MASK)); /* * Configure flow control parameters. * XON : 80% of Rx FIFO * XOFF : 30% of Rx FIFO */ if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8131 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8132) { reg = CSR_READ_4(sc, ALC_SRAM_RX_FIFO_LEN); rxf_hi = (reg * 8) / 10; rxf_lo = (reg * 3) / 10; CSR_WRITE_4(sc, ALC_RX_FIFO_PAUSE_THRESH, ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) & RX_FIFO_PAUSE_THRESH_LO_MASK) | ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) & RX_FIFO_PAUSE_THRESH_HI_MASK)); } if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151_V2) CSR_WRITE_4(sc, ALC_SERDES_LOCK, CSR_READ_4(sc, ALC_SERDES_LOCK) | SERDES_MAC_CLK_SLOWDOWN | SERDES_PHY_CLK_SLOWDOWN); /* Disable RSS until I understand L1C/L2C's RSS logic. */ CSR_WRITE_4(sc, ALC_RSS_IDT_TABLE0, 0); CSR_WRITE_4(sc, ALC_RSS_CPU, 0); /* Configure RxQ. */ reg = (RXQ_CFG_RD_BURST_DEFAULT << RXQ_CFG_RD_BURST_SHIFT) & RXQ_CFG_RD_BURST_MASK; reg |= RXQ_CFG_RSS_MODE_DIS; if ((sc->alc_flags & ALC_FLAG_ASPM_MON) != 0) reg |= RXQ_CFG_ASPM_THROUGHPUT_LIMIT_1M; CSR_WRITE_4(sc, ALC_RXQ_CFG, reg); /* Configure DMA parameters. */ reg = DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI; reg |= sc->alc_rcb; if ((sc->alc_flags & ALC_FLAG_CMB_BUG) == 0) reg |= DMA_CFG_CMB_ENB; if ((sc->alc_flags & ALC_FLAG_SMB_BUG) == 0) reg |= DMA_CFG_SMB_ENB; else reg |= DMA_CFG_SMB_DIS; reg |= (sc->alc_dma_rd_burst & DMA_CFG_RD_BURST_MASK) << DMA_CFG_RD_BURST_SHIFT; reg |= (sc->alc_dma_wr_burst & DMA_CFG_WR_BURST_MASK) << DMA_CFG_WR_BURST_SHIFT; reg |= (DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) & DMA_CFG_RD_DELAY_CNT_MASK; reg |= (DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) & DMA_CFG_WR_DELAY_CNT_MASK; CSR_WRITE_4(sc, ALC_DMA_CFG, reg); /* * Configure Tx/Rx MACs. * - Auto-padding for short frames. * - Enable CRC generation. * Actual reconfiguration of MAC for resolved speed/duplex * is followed after detection of link establishment. * AR813x/AR815x always does checksum computation regardless * of MAC_CFG_RXCSUM_ENB bit. Also the controller is known to * have bug in protocol field in Rx return structure so * these controllers can't handle fragmented frames. Disable * Rx checksum offloading until there is a newer controller * that has sane implementation. */ reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX | ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) & MAC_CFG_PREAMBLE_MASK); if (sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8151_V2 || sc->alc_ident->deviceid == DEVICEID_ATHEROS_AR8152_B2) reg |= MAC_CFG_HASH_ALG_CRC32 | MAC_CFG_SPEED_MODE_SW; if ((sc->alc_flags & ALC_FLAG_FASTETHER) != 0) reg |= MAC_CFG_SPEED_10_100; else reg |= MAC_CFG_SPEED_1000; CSR_WRITE_4(sc, ALC_MAC_CFG, reg); /* Set up the receive filter. */ alc_rxfilter(sc); alc_rxvlan(sc); /* Acknowledge all pending interrupts and clear it. */ CSR_WRITE_4(sc, ALC_INTR_MASK, ALC_INTRS); CSR_WRITE_4(sc, ALC_INTR_STATUS, 0xFFFFFFFF); CSR_WRITE_4(sc, ALC_INTR_STATUS, 0); sc->alc_flags &= ~ALC_FLAG_LINK; /* Switch to the current media. */ mii_mediachg(mii); callout_reset(&sc->alc_tick_ch, hz, alc_tick, sc); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } static void alc_stop(struct alc_softc *sc) { struct ifnet *ifp; struct alc_txdesc *txd; struct alc_rxdesc *rxd; uint32_t reg; int i; ALC_LOCK_ASSERT(sc); /* * Mark the interface down and cancel the watchdog timer. */ ifp = sc->alc_ifp; ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); sc->alc_flags &= ~ALC_FLAG_LINK; callout_stop(&sc->alc_tick_ch); sc->alc_watchdog_timer = 0; alc_stats_update(sc); /* Disable interrupts. */ CSR_WRITE_4(sc, ALC_INTR_MASK, 0); CSR_WRITE_4(sc, ALC_INTR_STATUS, 0xFFFFFFFF); alc_stop_queue(sc); /* Disable DMA. */ reg = CSR_READ_4(sc, ALC_DMA_CFG); reg &= ~(DMA_CFG_CMB_ENB | DMA_CFG_SMB_ENB); reg |= DMA_CFG_SMB_DIS; CSR_WRITE_4(sc, ALC_DMA_CFG, reg); DELAY(1000); /* Stop Rx/Tx MACs. */ alc_stop_mac(sc); /* Disable interrupts which might be touched in taskq handler. */ CSR_WRITE_4(sc, ALC_INTR_STATUS, 0xFFFFFFFF); /* Reclaim Rx buffers that have been processed. */ if (sc->alc_cdata.alc_rxhead != NULL) m_freem(sc->alc_cdata.alc_rxhead); ALC_RXCHAIN_RESET(sc); /* * Free Tx/Rx mbufs still in the queues. */ for (i = 0; i < ALC_RX_RING_CNT; i++) { rxd = &sc->alc_cdata.alc_rxdesc[i]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->alc_cdata.alc_rx_tag, rxd->rx_dmamap); m_freem(rxd->rx_m); rxd->rx_m = NULL; } } for (i = 0; i < ALC_TX_RING_CNT; i++) { txd = &sc->alc_cdata.alc_txdesc[i]; if (txd->tx_m != NULL) { bus_dmamap_sync(sc->alc_cdata.alc_tx_tag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->alc_cdata.alc_tx_tag, txd->tx_dmamap); m_freem(txd->tx_m); txd->tx_m = NULL; } } } static void alc_stop_mac(struct alc_softc *sc) { uint32_t reg; int i; ALC_LOCK_ASSERT(sc); /* Disable Rx/Tx MAC. */ reg = CSR_READ_4(sc, ALC_MAC_CFG); if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) { reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB); CSR_WRITE_4(sc, ALC_MAC_CFG, reg); } for (i = ALC_TIMEOUT; i > 0; i--) { reg = CSR_READ_4(sc, ALC_IDLE_STATUS); if (reg == 0) break; DELAY(10); } if (i == 0) device_printf(sc->alc_dev, "could not disable Rx/Tx MAC(0x%08x)!\n", reg); } static void alc_start_queue(struct alc_softc *sc) { uint32_t qcfg[] = { 0, RXQ_CFG_QUEUE0_ENB, RXQ_CFG_QUEUE0_ENB | RXQ_CFG_QUEUE1_ENB, RXQ_CFG_QUEUE0_ENB | RXQ_CFG_QUEUE1_ENB | RXQ_CFG_QUEUE2_ENB, RXQ_CFG_ENB }; uint32_t cfg; ALC_LOCK_ASSERT(sc); /* Enable RxQ. */ cfg = CSR_READ_4(sc, ALC_RXQ_CFG); cfg &= ~RXQ_CFG_ENB; cfg |= qcfg[1]; CSR_WRITE_4(sc, ALC_RXQ_CFG, cfg); /* Enable TxQ. */ cfg = CSR_READ_4(sc, ALC_TXQ_CFG); cfg |= TXQ_CFG_ENB; CSR_WRITE_4(sc, ALC_TXQ_CFG, cfg); } static void alc_stop_queue(struct alc_softc *sc) { uint32_t reg; int i; ALC_LOCK_ASSERT(sc); /* Disable RxQ. */ reg = CSR_READ_4(sc, ALC_RXQ_CFG); if ((reg & RXQ_CFG_ENB) != 0) { reg &= ~RXQ_CFG_ENB; CSR_WRITE_4(sc, ALC_RXQ_CFG, reg); } /* Disable TxQ. */ reg = CSR_READ_4(sc, ALC_TXQ_CFG); if ((reg & TXQ_CFG_ENB) != 0) { reg &= ~TXQ_CFG_ENB; CSR_WRITE_4(sc, ALC_TXQ_CFG, reg); } for (i = ALC_TIMEOUT; i > 0; i--) { reg = CSR_READ_4(sc, ALC_IDLE_STATUS); if ((reg & (IDLE_STATUS_RXQ | IDLE_STATUS_TXQ)) == 0) break; DELAY(10); } if (i == 0) device_printf(sc->alc_dev, "could not disable RxQ/TxQ (0x%08x)!\n", reg); } static void alc_init_tx_ring(struct alc_softc *sc) { struct alc_ring_data *rd; struct alc_txdesc *txd; int i; ALC_LOCK_ASSERT(sc); sc->alc_cdata.alc_tx_prod = 0; sc->alc_cdata.alc_tx_cons = 0; sc->alc_cdata.alc_tx_cnt = 0; rd = &sc->alc_rdata; bzero(rd->alc_tx_ring, ALC_TX_RING_SZ); for (i = 0; i < ALC_TX_RING_CNT; i++) { txd = &sc->alc_cdata.alc_txdesc[i]; txd->tx_m = NULL; } bus_dmamap_sync(sc->alc_cdata.alc_tx_ring_tag, sc->alc_cdata.alc_tx_ring_map, BUS_DMASYNC_PREWRITE); } static int alc_init_rx_ring(struct alc_softc *sc) { struct alc_ring_data *rd; struct alc_rxdesc *rxd; int i; ALC_LOCK_ASSERT(sc); sc->alc_cdata.alc_rx_cons = ALC_RX_RING_CNT - 1; sc->alc_morework = 0; rd = &sc->alc_rdata; bzero(rd->alc_rx_ring, ALC_RX_RING_SZ); for (i = 0; i < ALC_RX_RING_CNT; i++) { rxd = &sc->alc_cdata.alc_rxdesc[i]; rxd->rx_m = NULL; rxd->rx_desc = &rd->alc_rx_ring[i]; if (alc_newbuf(sc, rxd) != 0) return (ENOBUFS); } /* * Since controller does not update Rx descriptors, driver * does have to read Rx descriptors back so BUS_DMASYNC_PREWRITE * is enough to ensure coherence. */ bus_dmamap_sync(sc->alc_cdata.alc_rx_ring_tag, sc->alc_cdata.alc_rx_ring_map, BUS_DMASYNC_PREWRITE); /* Let controller know availability of new Rx buffers. */ CSR_WRITE_4(sc, ALC_MBOX_RD0_PROD_IDX, sc->alc_cdata.alc_rx_cons); return (0); } static void alc_init_rr_ring(struct alc_softc *sc) { struct alc_ring_data *rd; ALC_LOCK_ASSERT(sc); sc->alc_cdata.alc_rr_cons = 0; ALC_RXCHAIN_RESET(sc); rd = &sc->alc_rdata; bzero(rd->alc_rr_ring, ALC_RR_RING_SZ); bus_dmamap_sync(sc->alc_cdata.alc_rr_ring_tag, sc->alc_cdata.alc_rr_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static void alc_init_cmb(struct alc_softc *sc) { struct alc_ring_data *rd; ALC_LOCK_ASSERT(sc); rd = &sc->alc_rdata; bzero(rd->alc_cmb, ALC_CMB_SZ); bus_dmamap_sync(sc->alc_cdata.alc_cmb_tag, sc->alc_cdata.alc_cmb_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static void alc_init_smb(struct alc_softc *sc) { struct alc_ring_data *rd; ALC_LOCK_ASSERT(sc); rd = &sc->alc_rdata; bzero(rd->alc_smb, ALC_SMB_SZ); bus_dmamap_sync(sc->alc_cdata.alc_smb_tag, sc->alc_cdata.alc_smb_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static void alc_rxvlan(struct alc_softc *sc) { struct ifnet *ifp; uint32_t reg; ALC_LOCK_ASSERT(sc); ifp = sc->alc_ifp; reg = CSR_READ_4(sc, ALC_MAC_CFG); if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) reg |= MAC_CFG_VLAN_TAG_STRIP; else reg &= ~MAC_CFG_VLAN_TAG_STRIP; CSR_WRITE_4(sc, ALC_MAC_CFG, reg); } static void alc_rxfilter(struct alc_softc *sc) { struct ifnet *ifp; struct ifmultiaddr *ifma; uint32_t crc; uint32_t mchash[2]; uint32_t rxcfg; ALC_LOCK_ASSERT(sc); ifp = sc->alc_ifp; bzero(mchash, sizeof(mchash)); rxcfg = CSR_READ_4(sc, ALC_MAC_CFG); rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC); if ((ifp->if_flags & IFF_BROADCAST) != 0) rxcfg |= MAC_CFG_BCAST; if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) { if ((ifp->if_flags & IFF_PROMISC) != 0) rxcfg |= MAC_CFG_PROMISC; if ((ifp->if_flags & IFF_ALLMULTI) != 0) rxcfg |= MAC_CFG_ALLMULTI; mchash[0] = 0xFFFFFFFF; mchash[1] = 0xFFFFFFFF; goto chipit; } IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &sc->alc_ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; crc = ether_crc32_be(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), ETHER_ADDR_LEN); mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f); } IF_ADDR_UNLOCK(ifp); chipit: CSR_WRITE_4(sc, ALC_MAR0, mchash[0]); CSR_WRITE_4(sc, ALC_MAR1, mchash[1]); CSR_WRITE_4(sc, ALC_MAC_CFG, rxcfg); } static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high) { int error, value; if (arg1 == NULL) return (EINVAL); value = *(int *)arg1; error = sysctl_handle_int(oidp, &value, 0, req); if (error || req->newptr == NULL) return (error); if (value < low || value > high) return (EINVAL); *(int *)arg1 = value; return (0); } static int sysctl_hw_alc_proc_limit(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, ALC_PROC_MIN, ALC_PROC_MAX)); } static int sysctl_hw_alc_int_mod(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, ALC_IM_TIMER_MIN, ALC_IM_TIMER_MAX)); } Index: stable/7/sys/dev/bge/if_bge.c =================================================================== --- stable/7/sys/dev/bge/if_bge.c (revision 242221) +++ stable/7/sys/dev/bge/if_bge.c (revision 242222) @@ -1,6219 +1,6219 @@ /*- * Copyright (c) 2001 Wind River Systems * Copyright (c) 1997, 1998, 1999, 2001 * Bill Paul . All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Bill Paul. * 4. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * Broadcom BCM570x family gigabit ethernet driver for FreeBSD. * * The Broadcom BCM5700 is based on technology originally developed by * Alteon Networks as part of the Tigon I and Tigon II gigabit ethernet * MAC chips. The BCM5700, sometimes referred to as the Tigon III, has * two on-board MIPS R4000 CPUs and can have as much as 16MB of external * SSRAM. The BCM5700 supports TCP, UDP and IP checksum offload, jumbo * frames, highly configurable RX filtering, and 16 RX and TX queues * (which, along with RX filter rules, can be used for QOS applications). * Other features, such as TCP segmentation, may be available as part * of value-added firmware updates. Unlike the Tigon I and Tigon II, * firmware images can be stored in hardware and need not be compiled * into the driver. * * The BCM5700 supports the PCI v2.2 and PCI-X v1.0 standards, and will * function in a 32-bit/64-bit 33/66Mhz bus, or a 64-bit/133Mhz bus. * * The BCM5701 is a single-chip solution incorporating both the BCM5700 * MAC and a BCM5401 10/100/1000 PHY. Unlike the BCM5700, the BCM5701 * does not support external SSRAM. * * Broadcom also produces a variation of the BCM5700 under the "Altima" * brand name, which is functionally similar but lacks PCI-X support. * * Without external SSRAM, you can only have at most 4 TX rings, * and the use of the mini RX ring is disabled. This seems to imply * that these features are simply not available on the BCM5701. As a * result, this driver does not implement any support for the mini RX * ring. */ #ifdef HAVE_KERNEL_OPTION_HEADERS #include "opt_device_polling.h" #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "miidevs.h" #include #ifdef __sparc64__ #include #include #include #include #endif #include #include #include #define BGE_CSUM_FEATURES (CSUM_IP | CSUM_TCP) #define ETHER_MIN_NOPAD (ETHER_MIN_LEN - ETHER_CRC_LEN) /* i.e., 60 */ MODULE_DEPEND(bge, pci, 1, 1, 1); MODULE_DEPEND(bge, ether, 1, 1, 1); MODULE_DEPEND(bge, miibus, 1, 1, 1); /* "device miibus" required. See GENERIC if you get errors here. */ #include "miibus_if.h" /* * Various supported device vendors/types and their names. Note: the * spec seems to indicate that the hardware still has Alteon's vendor * ID burned into it, though it will always be overriden by the vendor * ID in the EEPROM. Just to be safe, we cover all possibilities. */ static const struct bge_type { uint16_t bge_vid; uint16_t bge_did; } const bge_devs[] = { { ALTEON_VENDORID, ALTEON_DEVICEID_BCM5700 }, { ALTEON_VENDORID, ALTEON_DEVICEID_BCM5701 }, { ALTIMA_VENDORID, ALTIMA_DEVICE_AC1000 }, { ALTIMA_VENDORID, ALTIMA_DEVICE_AC1002 }, { ALTIMA_VENDORID, ALTIMA_DEVICE_AC9100 }, { APPLE_VENDORID, APPLE_DEVICE_BCM5701 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5700 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5701 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5702 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5702_ALT }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5702X }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5703 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5703_ALT }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5703X }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5704C }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5704S }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5704S_ALT }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5705 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5705F }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5705K }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5705M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5705M_ALT }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5714C }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5714S }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5715 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5715S }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5717 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5718 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5719 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5720 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5721 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5722 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5723 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5750 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5750M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5751 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5751F }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5751M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5752 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5752M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5753 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5753F }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5753M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5754 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5754M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5755 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5755M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5756 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5761 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5761E }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5761S }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5761SE }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5764 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5780 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5780S }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5781 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5782 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5784 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5785F }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5785G }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5786 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5787 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5787F }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5787M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5788 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5789 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5901 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5901A2 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5903M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5906 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM5906M }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57760 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57761 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57765 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57780 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57781 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57785 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57788 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57790 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57791 }, { BCOM_VENDORID, BCOM_DEVICEID_BCM57795 }, { SK_VENDORID, SK_DEVICEID_ALTIMA }, { TC_VENDORID, TC_DEVICEID_3C996 }, { FJTSU_VENDORID, FJTSU_DEVICEID_PW008GE4 }, { FJTSU_VENDORID, FJTSU_DEVICEID_PW008GE5 }, { FJTSU_VENDORID, FJTSU_DEVICEID_PP250450 }, { 0, 0 } }; static const struct bge_vendor { uint16_t v_id; const char *v_name; } const bge_vendors[] = { { ALTEON_VENDORID, "Alteon" }, { ALTIMA_VENDORID, "Altima" }, { APPLE_VENDORID, "Apple" }, { BCOM_VENDORID, "Broadcom" }, { SK_VENDORID, "SysKonnect" }, { TC_VENDORID, "3Com" }, { FJTSU_VENDORID, "Fujitsu" }, { 0, NULL } }; static const struct bge_revision { uint32_t br_chipid; const char *br_name; } const bge_revisions[] = { { BGE_CHIPID_BCM5700_A0, "BCM5700 A0" }, { BGE_CHIPID_BCM5700_A1, "BCM5700 A1" }, { BGE_CHIPID_BCM5700_B0, "BCM5700 B0" }, { BGE_CHIPID_BCM5700_B1, "BCM5700 B1" }, { BGE_CHIPID_BCM5700_B2, "BCM5700 B2" }, { BGE_CHIPID_BCM5700_B3, "BCM5700 B3" }, { BGE_CHIPID_BCM5700_ALTIMA, "BCM5700 Altima" }, { BGE_CHIPID_BCM5700_C0, "BCM5700 C0" }, { BGE_CHIPID_BCM5701_A0, "BCM5701 A0" }, { BGE_CHIPID_BCM5701_B0, "BCM5701 B0" }, { BGE_CHIPID_BCM5701_B2, "BCM5701 B2" }, { BGE_CHIPID_BCM5701_B5, "BCM5701 B5" }, { BGE_CHIPID_BCM5703_A0, "BCM5703 A0" }, { BGE_CHIPID_BCM5703_A1, "BCM5703 A1" }, { BGE_CHIPID_BCM5703_A2, "BCM5703 A2" }, { BGE_CHIPID_BCM5703_A3, "BCM5703 A3" }, { BGE_CHIPID_BCM5703_B0, "BCM5703 B0" }, { BGE_CHIPID_BCM5704_A0, "BCM5704 A0" }, { BGE_CHIPID_BCM5704_A1, "BCM5704 A1" }, { BGE_CHIPID_BCM5704_A2, "BCM5704 A2" }, { BGE_CHIPID_BCM5704_A3, "BCM5704 A3" }, { BGE_CHIPID_BCM5704_B0, "BCM5704 B0" }, { BGE_CHIPID_BCM5705_A0, "BCM5705 A0" }, { BGE_CHIPID_BCM5705_A1, "BCM5705 A1" }, { BGE_CHIPID_BCM5705_A2, "BCM5705 A2" }, { BGE_CHIPID_BCM5705_A3, "BCM5705 A3" }, { BGE_CHIPID_BCM5750_A0, "BCM5750 A0" }, { BGE_CHIPID_BCM5750_A1, "BCM5750 A1" }, { BGE_CHIPID_BCM5750_A3, "BCM5750 A3" }, { BGE_CHIPID_BCM5750_B0, "BCM5750 B0" }, { BGE_CHIPID_BCM5750_B1, "BCM5750 B1" }, { BGE_CHIPID_BCM5750_C0, "BCM5750 C0" }, { BGE_CHIPID_BCM5750_C1, "BCM5750 C1" }, { BGE_CHIPID_BCM5750_C2, "BCM5750 C2" }, { BGE_CHIPID_BCM5714_A0, "BCM5714 A0" }, { BGE_CHIPID_BCM5752_A0, "BCM5752 A0" }, { BGE_CHIPID_BCM5752_A1, "BCM5752 A1" }, { BGE_CHIPID_BCM5752_A2, "BCM5752 A2" }, { BGE_CHIPID_BCM5714_B0, "BCM5714 B0" }, { BGE_CHIPID_BCM5714_B3, "BCM5714 B3" }, { BGE_CHIPID_BCM5715_A0, "BCM5715 A0" }, { BGE_CHIPID_BCM5715_A1, "BCM5715 A1" }, { BGE_CHIPID_BCM5715_A3, "BCM5715 A3" }, { BGE_CHIPID_BCM5717_A0, "BCM5717 A0" }, { BGE_CHIPID_BCM5717_B0, "BCM5717 B0" }, { BGE_CHIPID_BCM5719_A0, "BCM5719 A0" }, { BGE_CHIPID_BCM5720_A0, "BCM5720 A0" }, { BGE_CHIPID_BCM5755_A0, "BCM5755 A0" }, { BGE_CHIPID_BCM5755_A1, "BCM5755 A1" }, { BGE_CHIPID_BCM5755_A2, "BCM5755 A2" }, { BGE_CHIPID_BCM5722_A0, "BCM5722 A0" }, { BGE_CHIPID_BCM5761_A0, "BCM5761 A0" }, { BGE_CHIPID_BCM5761_A1, "BCM5761 A1" }, { BGE_CHIPID_BCM5784_A0, "BCM5784 A0" }, { BGE_CHIPID_BCM5784_A1, "BCM5784 A1" }, /* 5754 and 5787 share the same ASIC ID */ { BGE_CHIPID_BCM5787_A0, "BCM5754/5787 A0" }, { BGE_CHIPID_BCM5787_A1, "BCM5754/5787 A1" }, { BGE_CHIPID_BCM5787_A2, "BCM5754/5787 A2" }, { BGE_CHIPID_BCM5906_A1, "BCM5906 A1" }, { BGE_CHIPID_BCM5906_A2, "BCM5906 A2" }, { BGE_CHIPID_BCM57765_A0, "BCM57765 A0" }, { BGE_CHIPID_BCM57765_B0, "BCM57765 B0" }, { BGE_CHIPID_BCM57780_A0, "BCM57780 A0" }, { BGE_CHIPID_BCM57780_A1, "BCM57780 A1" }, { 0, NULL } }; /* * Some defaults for major revisions, so that newer steppings * that we don't know about have a shot at working. */ static const struct bge_revision const bge_majorrevs[] = { { BGE_ASICREV_BCM5700, "unknown BCM5700" }, { BGE_ASICREV_BCM5701, "unknown BCM5701" }, { BGE_ASICREV_BCM5703, "unknown BCM5703" }, { BGE_ASICREV_BCM5704, "unknown BCM5704" }, { BGE_ASICREV_BCM5705, "unknown BCM5705" }, { BGE_ASICREV_BCM5750, "unknown BCM5750" }, { BGE_ASICREV_BCM5714_A0, "unknown BCM5714" }, { BGE_ASICREV_BCM5752, "unknown BCM5752" }, { BGE_ASICREV_BCM5780, "unknown BCM5780" }, { BGE_ASICREV_BCM5714, "unknown BCM5714" }, { BGE_ASICREV_BCM5755, "unknown BCM5755" }, { BGE_ASICREV_BCM5761, "unknown BCM5761" }, { BGE_ASICREV_BCM5784, "unknown BCM5784" }, { BGE_ASICREV_BCM5785, "unknown BCM5785" }, /* 5754 and 5787 share the same ASIC ID */ { BGE_ASICREV_BCM5787, "unknown BCM5754/5787" }, { BGE_ASICREV_BCM5906, "unknown BCM5906" }, { BGE_ASICREV_BCM57765, "unknown BCM57765" }, { BGE_ASICREV_BCM57780, "unknown BCM57780" }, { BGE_ASICREV_BCM5717, "unknown BCM5717" }, { BGE_ASICREV_BCM5719, "unknown BCM5719" }, { BGE_ASICREV_BCM5720, "unknown BCM5720" }, { 0, NULL } }; #define BGE_IS_JUMBO_CAPABLE(sc) ((sc)->bge_flags & BGE_FLAG_JUMBO) #define BGE_IS_5700_FAMILY(sc) ((sc)->bge_flags & BGE_FLAG_5700_FAMILY) #define BGE_IS_5705_PLUS(sc) ((sc)->bge_flags & BGE_FLAG_5705_PLUS) #define BGE_IS_5714_FAMILY(sc) ((sc)->bge_flags & BGE_FLAG_5714_FAMILY) #define BGE_IS_575X_PLUS(sc) ((sc)->bge_flags & BGE_FLAG_575X_PLUS) #define BGE_IS_5755_PLUS(sc) ((sc)->bge_flags & BGE_FLAG_5755_PLUS) #define BGE_IS_5717_PLUS(sc) ((sc)->bge_flags & BGE_FLAG_5717_PLUS) const struct bge_revision * bge_lookup_rev(uint32_t); const struct bge_vendor * bge_lookup_vendor(uint16_t); typedef int (*bge_eaddr_fcn_t)(struct bge_softc *, uint8_t[]); static int bge_probe(device_t); static int bge_attach(device_t); static int bge_detach(device_t); static int bge_suspend(device_t); static int bge_resume(device_t); static void bge_release_resources(struct bge_softc *); static void bge_dma_map_addr(void *, bus_dma_segment_t *, int, int); static int bge_dma_alloc(struct bge_softc *); static void bge_dma_free(struct bge_softc *); static int bge_dma_ring_alloc(struct bge_softc *, bus_size_t, bus_size_t, bus_dma_tag_t *, uint8_t **, bus_dmamap_t *, bus_addr_t *, const char *); static void bge_devinfo(struct bge_softc *); static int bge_mbox_reorder(struct bge_softc *); static int bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[]); static int bge_get_eaddr_mem(struct bge_softc *, uint8_t[]); static int bge_get_eaddr_nvram(struct bge_softc *, uint8_t[]); static int bge_get_eaddr_eeprom(struct bge_softc *, uint8_t[]); static int bge_get_eaddr(struct bge_softc *, uint8_t[]); static void bge_txeof(struct bge_softc *, uint16_t); static void bge_rxcsum(struct bge_softc *, struct bge_rx_bd *, struct mbuf *); static void bge_rxeof(struct bge_softc *, uint16_t, int); static void bge_asf_driver_up (struct bge_softc *); static void bge_tick(void *); static void bge_stats_clear_regs(struct bge_softc *); static void bge_stats_update(struct bge_softc *); static void bge_stats_update_regs(struct bge_softc *); static struct mbuf *bge_check_short_dma(struct mbuf *); static struct mbuf *bge_setup_tso(struct bge_softc *, struct mbuf *, uint16_t *, uint16_t *); static int bge_encap(struct bge_softc *, struct mbuf **, uint32_t *); static void bge_intr(void *); static int bge_msi_intr(void *); static void bge_intr_task(void *, int); static void bge_start_locked(struct ifnet *); static void bge_start(struct ifnet *); static int bge_ioctl(struct ifnet *, u_long, caddr_t); static void bge_init_locked(struct bge_softc *); static void bge_init(void *); static void bge_stop_block(struct bge_softc *, bus_size_t, uint32_t); static void bge_stop(struct bge_softc *); static void bge_watchdog(struct bge_softc *); static int bge_shutdown(device_t); static int bge_ifmedia_upd_locked(struct ifnet *); static int bge_ifmedia_upd(struct ifnet *); static void bge_ifmedia_sts(struct ifnet *, struct ifmediareq *); static uint8_t bge_nvram_getbyte(struct bge_softc *, int, uint8_t *); static int bge_read_nvram(struct bge_softc *, caddr_t, int, int); static uint8_t bge_eeprom_getbyte(struct bge_softc *, int, uint8_t *); static int bge_read_eeprom(struct bge_softc *, caddr_t, int, int); static void bge_setpromisc(struct bge_softc *); static void bge_setmulti(struct bge_softc *); static void bge_setvlan(struct bge_softc *); static __inline void bge_rxreuse_std(struct bge_softc *, int); static __inline void bge_rxreuse_jumbo(struct bge_softc *, int); static int bge_newbuf_std(struct bge_softc *, int); static int bge_newbuf_jumbo(struct bge_softc *, int); static int bge_init_rx_ring_std(struct bge_softc *); static void bge_free_rx_ring_std(struct bge_softc *); static int bge_init_rx_ring_jumbo(struct bge_softc *); static void bge_free_rx_ring_jumbo(struct bge_softc *); static void bge_free_tx_ring(struct bge_softc *); static int bge_init_tx_ring(struct bge_softc *); static int bge_chipinit(struct bge_softc *); static int bge_blockinit(struct bge_softc *); static uint32_t bge_dma_swap_options(struct bge_softc *); static int bge_has_eaddr(struct bge_softc *); static uint32_t bge_readmem_ind(struct bge_softc *, int); static void bge_writemem_ind(struct bge_softc *, int, int); static void bge_writembx(struct bge_softc *, int, int); #ifdef notdef static uint32_t bge_readreg_ind(struct bge_softc *, int); #endif static void bge_writemem_direct(struct bge_softc *, int, int); static void bge_writereg_ind(struct bge_softc *, int, int); static int bge_miibus_readreg(device_t, int, int); static int bge_miibus_writereg(device_t, int, int, int); static void bge_miibus_statchg(device_t); #ifdef DEVICE_POLLING static void bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count); #endif #define BGE_RESET_START 1 #define BGE_RESET_STOP 2 static void bge_sig_post_reset(struct bge_softc *, int); static void bge_sig_legacy(struct bge_softc *, int); static void bge_sig_pre_reset(struct bge_softc *, int); static void bge_stop_fw(struct bge_softc *); static int bge_reset(struct bge_softc *); static void bge_link_upd(struct bge_softc *); /* * The BGE_REGISTER_DEBUG option is only for low-level debugging. It may * leak information to untrusted users. It is also known to cause alignment * traps on certain architectures. */ #ifdef BGE_REGISTER_DEBUG static int bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS); static int bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS); static int bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS); #endif static void bge_add_sysctls(struct bge_softc *); static void bge_add_sysctl_stats_regs(struct bge_softc *, struct sysctl_ctx_list *, struct sysctl_oid_list *); static void bge_add_sysctl_stats(struct bge_softc *, struct sysctl_ctx_list *, struct sysctl_oid_list *); static int bge_sysctl_stats(SYSCTL_HANDLER_ARGS); static device_method_t bge_methods[] = { /* Device interface */ DEVMETHOD(device_probe, bge_probe), DEVMETHOD(device_attach, bge_attach), DEVMETHOD(device_detach, bge_detach), DEVMETHOD(device_shutdown, bge_shutdown), DEVMETHOD(device_suspend, bge_suspend), DEVMETHOD(device_resume, bge_resume), /* bus interface */ DEVMETHOD(bus_print_child, bus_generic_print_child), DEVMETHOD(bus_driver_added, bus_generic_driver_added), /* MII interface */ DEVMETHOD(miibus_readreg, bge_miibus_readreg), DEVMETHOD(miibus_writereg, bge_miibus_writereg), DEVMETHOD(miibus_statchg, bge_miibus_statchg), { 0, 0 } }; static driver_t bge_driver = { "bge", bge_methods, sizeof(struct bge_softc) }; static devclass_t bge_devclass; DRIVER_MODULE(bge, pci, bge_driver, bge_devclass, 0, 0); DRIVER_MODULE(miibus, bge, miibus_driver, miibus_devclass, 0, 0); static int bge_allow_asf = 0; TUNABLE_INT("hw.bge.allow_asf", &bge_allow_asf); SYSCTL_NODE(_hw, OID_AUTO, bge, CTLFLAG_RD, 0, "BGE driver parameters"); SYSCTL_INT(_hw_bge, OID_AUTO, allow_asf, CTLFLAG_RD, &bge_allow_asf, 0, "Allow ASF mode if available"); #define SPARC64_BLADE_1500_MODEL "SUNW,Sun-Blade-1500" #define SPARC64_BLADE_1500_PATH_BGE "/pci@1f,700000/network@2" #define SPARC64_BLADE_2500_MODEL "SUNW,Sun-Blade-2500" #define SPARC64_BLADE_2500_PATH_BGE "/pci@1c,600000/network@3" #define SPARC64_OFW_SUBVENDOR "subsystem-vendor-id" static int bge_has_eaddr(struct bge_softc *sc) { #ifdef __sparc64__ char buf[sizeof(SPARC64_BLADE_1500_PATH_BGE)]; device_t dev; uint32_t subvendor; dev = sc->bge_dev; /* * The on-board BGEs found in sun4u machines aren't fitted with * an EEPROM which means that we have to obtain the MAC address * via OFW and that some tests will always fail. We distinguish * such BGEs by the subvendor ID, which also has to be obtained * from OFW instead of the PCI configuration space as the latter * indicates Broadcom as the subvendor of the netboot interface. * For early Blade 1500 and 2500 we even have to check the OFW * device path as the subvendor ID always defaults to Broadcom * there. */ if (OF_getprop(ofw_bus_get_node(dev), SPARC64_OFW_SUBVENDOR, &subvendor, sizeof(subvendor)) == sizeof(subvendor) && (subvendor == FJTSU_VENDORID || subvendor == SUN_VENDORID)) return (0); memset(buf, 0, sizeof(buf)); if (OF_package_to_path(ofw_bus_get_node(dev), buf, sizeof(buf)) > 0) { if (strcmp(sparc64_model, SPARC64_BLADE_1500_MODEL) == 0 && strcmp(buf, SPARC64_BLADE_1500_PATH_BGE) == 0) return (0); if (strcmp(sparc64_model, SPARC64_BLADE_2500_MODEL) == 0 && strcmp(buf, SPARC64_BLADE_2500_PATH_BGE) == 0) return (0); } #endif return (1); } static uint32_t bge_readmem_ind(struct bge_softc *sc, int off) { device_t dev; uint32_t val; if (sc->bge_asicrev == BGE_ASICREV_BCM5906 && off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4) return (0); dev = sc->bge_dev; pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4); val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4); pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4); return (val); } static void bge_writemem_ind(struct bge_softc *sc, int off, int val) { device_t dev; if (sc->bge_asicrev == BGE_ASICREV_BCM5906 && off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4) return; dev = sc->bge_dev; pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4); pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4); pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4); } #ifdef notdef static uint32_t bge_readreg_ind(struct bge_softc *sc, int off) { device_t dev; dev = sc->bge_dev; pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4); return (pci_read_config(dev, BGE_PCI_REG_DATA, 4)); } #endif static void bge_writereg_ind(struct bge_softc *sc, int off, int val) { device_t dev; dev = sc->bge_dev; pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4); pci_write_config(dev, BGE_PCI_REG_DATA, val, 4); } static void bge_writemem_direct(struct bge_softc *sc, int off, int val) { CSR_WRITE_4(sc, off, val); } static void bge_writembx(struct bge_softc *sc, int off, int val) { if (sc->bge_asicrev == BGE_ASICREV_BCM5906) off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI; CSR_WRITE_4(sc, off, val); if ((sc->bge_flags & BGE_FLAG_MBOX_REORDER) != 0) CSR_READ_4(sc, off); } /* * Map a single buffer address. */ static void bge_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error) { struct bge_dmamap_arg *ctx; if (error) return; KASSERT(nseg == 1, ("%s: %d segments returned!", __func__, nseg)); ctx = arg; ctx->bge_busaddr = segs->ds_addr; } static uint8_t bge_nvram_getbyte(struct bge_softc *sc, int addr, uint8_t *dest) { uint32_t access, byte = 0; int i; /* Lock. */ CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1); for (i = 0; i < 8000; i++) { if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1) break; DELAY(20); } if (i == 8000) return (1); /* Enable access. */ access = CSR_READ_4(sc, BGE_NVRAM_ACCESS); CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE); CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc); CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD); for (i = 0; i < BGE_TIMEOUT * 10; i++) { DELAY(10); if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) { DELAY(10); break; } } if (i == BGE_TIMEOUT * 10) { if_printf(sc->bge_ifp, "nvram read timed out\n"); return (1); } /* Get result. */ byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA); *dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF; /* Disable access. */ CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access); /* Unlock. */ CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1); CSR_READ_4(sc, BGE_NVRAM_SWARB); return (0); } /* * Read a sequence of bytes from NVRAM. */ static int bge_read_nvram(struct bge_softc *sc, caddr_t dest, int off, int cnt) { int err = 0, i; uint8_t byte = 0; if (sc->bge_asicrev != BGE_ASICREV_BCM5906) return (1); for (i = 0; i < cnt; i++) { err = bge_nvram_getbyte(sc, off + i, &byte); if (err) break; *(dest + i) = byte; } return (err ? 1 : 0); } /* * Read a byte of data stored in the EEPROM at address 'addr.' The * BCM570x supports both the traditional bitbang interface and an * auto access interface for reading the EEPROM. We use the auto * access method. */ static uint8_t bge_eeprom_getbyte(struct bge_softc *sc, int addr, uint8_t *dest) { int i; uint32_t byte = 0; /* * Enable use of auto EEPROM access so we can avoid * having to use the bitbang method. */ BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM); /* Reset the EEPROM, load the clock period. */ CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EEADDR_RESET | BGE_EEHALFCLK(BGE_HALFCLK_384SCL)); DELAY(20); /* Issue the read EEPROM command. */ CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr); /* Wait for completion */ for(i = 0; i < BGE_TIMEOUT * 10; i++) { DELAY(10); if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE) break; } if (i == BGE_TIMEOUT * 10) { device_printf(sc->bge_dev, "EEPROM read timed out\n"); return (1); } /* Get result. */ byte = CSR_READ_4(sc, BGE_EE_DATA); *dest = (byte >> ((addr % 4) * 8)) & 0xFF; return (0); } /* * Read a sequence of bytes from the EEPROM. */ static int bge_read_eeprom(struct bge_softc *sc, caddr_t dest, int off, int cnt) { int i, error = 0; uint8_t byte = 0; for (i = 0; i < cnt; i++) { error = bge_eeprom_getbyte(sc, off + i, &byte); if (error) break; *(dest + i) = byte; } return (error ? 1 : 0); } static int bge_miibus_readreg(device_t dev, int phy, int reg) { struct bge_softc *sc; uint32_t val; int i; sc = device_get_softc(dev); /* Clear the autopoll bit if set, otherwise may trigger PCI errors. */ if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) { CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL); DELAY(80); } CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY | BGE_MIPHY(phy) | BGE_MIREG(reg)); /* Poll for the PHY register access to complete. */ for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); val = CSR_READ_4(sc, BGE_MI_COMM); if ((val & BGE_MICOMM_BUSY) == 0) { DELAY(5); val = CSR_READ_4(sc, BGE_MI_COMM); break; } } if (i == BGE_TIMEOUT) { device_printf(sc->bge_dev, "PHY read timed out (phy %d, reg %d, val 0x%08x)\n", phy, reg, val); val = 0; } /* Restore the autopoll bit if necessary. */ if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) { CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode); DELAY(80); } if (val & BGE_MICOMM_READFAIL) return (0); return (val & 0xFFFF); } static int bge_miibus_writereg(device_t dev, int phy, int reg, int val) { struct bge_softc *sc; int i; sc = device_get_softc(dev); if (sc->bge_asicrev == BGE_ASICREV_BCM5906 && (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL)) return (0); /* Clear the autopoll bit if set, otherwise may trigger PCI errors. */ if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) { CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode & ~BGE_MIMODE_AUTOPOLL); DELAY(80); } CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY | BGE_MIPHY(phy) | BGE_MIREG(reg) | val); for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) { DELAY(5); CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */ break; } } /* Restore the autopoll bit if necessary. */ if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) { CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode); DELAY(80); } if (i == BGE_TIMEOUT) device_printf(sc->bge_dev, "PHY write timed out (phy %d, reg %d, val %d)\n", phy, reg, val); return (0); } static void bge_miibus_statchg(device_t dev) { struct bge_softc *sc; struct mii_data *mii; sc = device_get_softc(dev); mii = device_get_softc(sc->bge_miibus); if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: sc->bge_link = 1; break; case IFM_1000_T: case IFM_1000_SX: case IFM_2500_SX: if (sc->bge_asicrev != BGE_ASICREV_BCM5906) sc->bge_link = 1; else sc->bge_link = 0; break; default: sc->bge_link = 0; break; } } else sc->bge_link = 0; if (sc->bge_link == 0) return; BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE); if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T || IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII); else BGE_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII); if (IFM_OPTIONS(mii->mii_media_active & IFM_FDX) != 0) { BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX); if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0) BGE_SETBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE); else BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE); if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0) BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE); else BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE); } else { BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX); BGE_CLRBIT(sc, BGE_TX_MODE, BGE_TXMODE_FLOWCTL_ENABLE); BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_FLOWCTL_ENABLE); } } /* * Intialize a standard receive ring descriptor. */ static int bge_newbuf_std(struct bge_softc *sc, int i) { struct mbuf *m; struct bge_rx_bd *r; bus_dma_segment_t segs[1]; bus_dmamap_t map; int error, nsegs; if (sc->bge_flags & BGE_FLAG_JUMBO_STD && (sc->bge_ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN))) { m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MJUM9BYTES; } else { m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MCLBYTES; } if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) m_adj(m, ETHER_ALIGN); error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_sparemap, m, segs, &nsegs, 0); if (error != 0) { m_freem(m); return (error); } if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i]); } map = sc->bge_cdata.bge_rx_std_dmamap[i]; sc->bge_cdata.bge_rx_std_dmamap[i] = sc->bge_cdata.bge_rx_std_sparemap; sc->bge_cdata.bge_rx_std_sparemap = map; sc->bge_cdata.bge_rx_std_chain[i] = m; sc->bge_cdata.bge_rx_std_seglen[i] = segs[0].ds_len; r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std]; r->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr); r->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr); r->bge_flags = BGE_RXBDFLAG_END; r->bge_len = segs[0].ds_len; r->bge_idx = i; bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_PREREAD); return (0); } /* * Initialize a jumbo receive ring descriptor. This allocates * a jumbo buffer from the pool managed internally by the driver. */ static int bge_newbuf_jumbo(struct bge_softc *sc, int i) { bus_dma_segment_t segs[BGE_NSEG_JUMBO]; bus_dmamap_t map; struct bge_extrx_bd *r; struct mbuf *m; int error, nsegs; MGETHDR(m, M_DONTWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); m_cljget(m, M_DONTWAIT, MJUM9BYTES); if (!(m->m_flags & M_EXT)) { m_freem(m); return (ENOBUFS); } m->m_len = m->m_pkthdr.len = MJUM9BYTES; if ((sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) == 0) m_adj(m, ETHER_ALIGN); error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_sparemap, m, segs, &nsegs, 0); if (error != 0) { m_freem(m); return (error); } if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i]); } map = sc->bge_cdata.bge_rx_jumbo_dmamap[i]; sc->bge_cdata.bge_rx_jumbo_dmamap[i] = sc->bge_cdata.bge_rx_jumbo_sparemap; sc->bge_cdata.bge_rx_jumbo_sparemap = map; sc->bge_cdata.bge_rx_jumbo_chain[i] = m; sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = 0; sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = 0; sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = 0; sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = 0; /* * Fill in the extended RX buffer descriptor. */ r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo]; r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END; r->bge_idx = i; r->bge_len3 = r->bge_len2 = r->bge_len1 = 0; switch (nsegs) { case 4: r->bge_addr3.bge_addr_lo = BGE_ADDR_LO(segs[3].ds_addr); r->bge_addr3.bge_addr_hi = BGE_ADDR_HI(segs[3].ds_addr); r->bge_len3 = segs[3].ds_len; sc->bge_cdata.bge_rx_jumbo_seglen[i][3] = segs[3].ds_len; case 3: r->bge_addr2.bge_addr_lo = BGE_ADDR_LO(segs[2].ds_addr); r->bge_addr2.bge_addr_hi = BGE_ADDR_HI(segs[2].ds_addr); r->bge_len2 = segs[2].ds_len; sc->bge_cdata.bge_rx_jumbo_seglen[i][2] = segs[2].ds_len; case 2: r->bge_addr1.bge_addr_lo = BGE_ADDR_LO(segs[1].ds_addr); r->bge_addr1.bge_addr_hi = BGE_ADDR_HI(segs[1].ds_addr); r->bge_len1 = segs[1].ds_len; sc->bge_cdata.bge_rx_jumbo_seglen[i][1] = segs[1].ds_len; case 1: r->bge_addr0.bge_addr_lo = BGE_ADDR_LO(segs[0].ds_addr); r->bge_addr0.bge_addr_hi = BGE_ADDR_HI(segs[0].ds_addr); r->bge_len0 = segs[0].ds_len; sc->bge_cdata.bge_rx_jumbo_seglen[i][0] = segs[0].ds_len; break; default: panic("%s: %d segments\n", __func__, nsegs); } bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_PREREAD); return (0); } static int bge_init_rx_ring_std(struct bge_softc *sc) { int error, i; bzero(sc->bge_ldata.bge_rx_std_ring, BGE_STD_RX_RING_SZ); sc->bge_std = 0; for (i = 0; i < BGE_STD_RX_RING_CNT; i++) { if ((error = bge_newbuf_std(sc, i)) != 0) return (error); BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT); } bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE); sc->bge_std = 0; bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, BGE_STD_RX_RING_CNT - 1); return (0); } static void bge_free_rx_ring_std(struct bge_softc *sc) { int i; for (i = 0; i < BGE_STD_RX_RING_CNT; i++) { if (sc->bge_cdata.bge_rx_std_chain[i] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i], BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i]); m_freem(sc->bge_cdata.bge_rx_std_chain[i]); sc->bge_cdata.bge_rx_std_chain[i] = NULL; } bzero((char *)&sc->bge_ldata.bge_rx_std_ring[i], sizeof(struct bge_rx_bd)); } } static int bge_init_rx_ring_jumbo(struct bge_softc *sc) { struct bge_rcb *rcb; int error, i; bzero(sc->bge_ldata.bge_rx_jumbo_ring, BGE_JUMBO_RX_RING_SZ); sc->bge_jumbo = 0; for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) { if ((error = bge_newbuf_jumbo(sc, i)) != 0) return (error); BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT); } bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE); sc->bge_jumbo = 0; /* Enable the jumbo receive producer ring. */ rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb; rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_USE_EXT_RX_BD); CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags); bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, BGE_JUMBO_RX_RING_CNT - 1); return (0); } static void bge_free_rx_ring_jumbo(struct bge_softc *sc) { int i; for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) { if (sc->bge_cdata.bge_rx_jumbo_chain[i] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i], BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i]); m_freem(sc->bge_cdata.bge_rx_jumbo_chain[i]); sc->bge_cdata.bge_rx_jumbo_chain[i] = NULL; } bzero((char *)&sc->bge_ldata.bge_rx_jumbo_ring[i], sizeof(struct bge_extrx_bd)); } } static void bge_free_tx_ring(struct bge_softc *sc) { int i; if (sc->bge_ldata.bge_tx_ring == NULL) return; for (i = 0; i < BGE_TX_RING_CNT; i++) { if (sc->bge_cdata.bge_tx_chain[i] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, sc->bge_cdata.bge_tx_dmamap[i], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, sc->bge_cdata.bge_tx_dmamap[i]); m_freem(sc->bge_cdata.bge_tx_chain[i]); sc->bge_cdata.bge_tx_chain[i] = NULL; } bzero((char *)&sc->bge_ldata.bge_tx_ring[i], sizeof(struct bge_tx_bd)); } } static int bge_init_tx_ring(struct bge_softc *sc) { sc->bge_txcnt = 0; sc->bge_tx_saved_considx = 0; bzero(sc->bge_ldata.bge_tx_ring, BGE_TX_RING_SZ); bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag, sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE); /* Initialize transmit producer index for host-memory send ring. */ sc->bge_tx_prodidx = 0; bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx); /* 5700 b2 errata */ if (sc->bge_chiprev == BGE_CHIPREV_5700_BX) bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bge_tx_prodidx); /* NIC-memory send ring not used; initialize to zero. */ bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0); /* 5700 b2 errata */ if (sc->bge_chiprev == BGE_CHIPREV_5700_BX) bge_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0); return (0); } static void bge_setpromisc(struct bge_softc *sc) { struct ifnet *ifp; BGE_LOCK_ASSERT(sc); ifp = sc->bge_ifp; /* Enable or disable promiscuous mode as needed. */ if (ifp->if_flags & IFF_PROMISC) BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC); else BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC); } static void bge_setmulti(struct bge_softc *sc) { struct ifnet *ifp; struct ifmultiaddr *ifma; uint32_t hashes[4] = { 0, 0, 0, 0 }; int h, i; BGE_LOCK_ASSERT(sc); ifp = sc->bge_ifp; if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) { for (i = 0; i < 4; i++) CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF); return; } /* First, zot all the existing filters. */ for (i = 0; i < 4; i++) CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0); /* Now program new ones. */ IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; h = ether_crc32_le(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), ETHER_ADDR_LEN) & 0x7F; hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F); } IF_ADDR_UNLOCK(ifp); for (i = 0; i < 4; i++) CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]); } static void bge_setvlan(struct bge_softc *sc) { struct ifnet *ifp; BGE_LOCK_ASSERT(sc); ifp = sc->bge_ifp; /* Enable or disable VLAN tag stripping as needed. */ if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) BGE_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG); else BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_KEEP_VLAN_DIAG); } static void bge_sig_pre_reset(struct bge_softc *sc, int type) { /* * Some chips don't like this so only do this if ASF is enabled */ if (sc->bge_asf_mode) bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC); if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) { switch (type) { case BGE_RESET_START: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_START); break; case BGE_RESET_STOP: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_UNLOAD); break; } } } static void bge_sig_post_reset(struct bge_softc *sc, int type) { if (sc->bge_asf_mode & ASF_NEW_HANDSHAKE) { switch (type) { case BGE_RESET_START: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_START_DONE); /* START DONE */ break; case BGE_RESET_STOP: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_UNLOAD_DONE); break; } } } static void bge_sig_legacy(struct bge_softc *sc, int type) { if (sc->bge_asf_mode) { switch (type) { case BGE_RESET_START: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_START); break; case BGE_RESET_STOP: bge_writemem_ind(sc, BGE_SRAM_FW_DRV_STATE_MB, BGE_FW_DRV_STATE_UNLOAD); break; } } } static void bge_stop_fw(struct bge_softc *sc) { int i; if (sc->bge_asf_mode) { bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_PAUSE); CSR_WRITE_4(sc, BGE_RX_CPU_EVENT, CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT); for (i = 0; i < 100; i++ ) { if (!(CSR_READ_4(sc, BGE_RX_CPU_EVENT) & BGE_RX_CPU_DRV_EVENT)) break; DELAY(10); } } } static uint32_t bge_dma_swap_options(struct bge_softc *sc) { uint32_t dma_options; dma_options = BGE_MODECTL_WORDSWAP_NONFRAME | BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA; #if BYTE_ORDER == BIG_ENDIAN dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME; #endif if ((sc)->bge_asicrev == BGE_ASICREV_BCM5720) dma_options |= BGE_MODECTL_BYTESWAP_B2HRX_DATA | BGE_MODECTL_WORDSWAP_B2HRX_DATA | BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE; return (dma_options); } /* * Do endian, PCI and DMA initialization. */ static int bge_chipinit(struct bge_softc *sc) { uint32_t dma_rw_ctl, misc_ctl, mode_ctl; uint16_t val; int i; /* Set endianness before we access any non-PCI registers. */ misc_ctl = BGE_INIT; if (sc->bge_flags & BGE_FLAG_TAGGED_STATUS) misc_ctl |= BGE_PCIMISCCTL_TAGGED_STATUS; pci_write_config(sc->bge_dev, BGE_PCI_MISC_CTL, misc_ctl, 4); /* Clear the MAC control register */ CSR_WRITE_4(sc, BGE_MAC_MODE, 0); /* * Clear the MAC statistics block in the NIC's * internal memory. */ for (i = BGE_STATS_BLOCK; i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t)) BGE_MEMWIN_WRITE(sc, i, 0); for (i = BGE_STATUS_BLOCK; i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t)) BGE_MEMWIN_WRITE(sc, i, 0); if (sc->bge_chiprev == BGE_CHIPREV_5704_BX) { /* * Fix data corruption caused by non-qword write with WB. * Fix master abort in PCI mode. * Fix PCI latency timer. */ val = pci_read_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, 2); val |= (1 << 10) | (1 << 12) | (1 << 13); pci_write_config(sc->bge_dev, BGE_PCI_MSI_DATA + 2, val, 2); } /* * Set up the PCI DMA control register. */ dma_rw_ctl = BGE_PCIDMARWCTL_RD_CMD_SHIFT(6) | BGE_PCIDMARWCTL_WR_CMD_SHIFT(7); if (sc->bge_flags & BGE_FLAG_PCIE) { /* Read watermark not used, 128 bytes for write. */ dma_rw_ctl |= BGE_PCIDMARWCTL_WR_WAT_SHIFT(3); } else if (sc->bge_flags & BGE_FLAG_PCIX) { if (BGE_IS_5714_FAMILY(sc)) { /* 256 bytes for read and write. */ dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(2) | BGE_PCIDMARWCTL_WR_WAT_SHIFT(2); dma_rw_ctl |= (sc->bge_asicrev == BGE_ASICREV_BCM5780) ? BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL : BGE_PCIDMARWCTL_ONEDMA_ATONCE_LOCAL; } else if (sc->bge_asicrev == BGE_ASICREV_BCM5703) { /* * In the BCM5703, the DMA read watermark should * be set to less than or equal to the maximum * memory read byte count of the PCI-X command * register. */ dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(4) | BGE_PCIDMARWCTL_WR_WAT_SHIFT(3); } else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) { /* 1536 bytes for read, 384 bytes for write. */ dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) | BGE_PCIDMARWCTL_WR_WAT_SHIFT(3); } else { /* 384 bytes for read and write. */ dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(3) | BGE_PCIDMARWCTL_WR_WAT_SHIFT(3) | 0x0F; } if (sc->bge_asicrev == BGE_ASICREV_BCM5703 || sc->bge_asicrev == BGE_ASICREV_BCM5704) { uint32_t tmp; /* Set ONE_DMA_AT_ONCE for hardware workaround. */ tmp = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F; if (tmp == 6 || tmp == 7) dma_rw_ctl |= BGE_PCIDMARWCTL_ONEDMA_ATONCE_GLOBAL; /* Set PCI-X DMA write workaround. */ dma_rw_ctl |= BGE_PCIDMARWCTL_ASRT_ALL_BE; } } else { /* Conventional PCI bus: 256 bytes for read and write. */ dma_rw_ctl |= BGE_PCIDMARWCTL_RD_WAT_SHIFT(7) | BGE_PCIDMARWCTL_WR_WAT_SHIFT(7); if (sc->bge_asicrev != BGE_ASICREV_BCM5705 && sc->bge_asicrev != BGE_ASICREV_BCM5750) dma_rw_ctl |= 0x0F; } if (sc->bge_asicrev == BGE_ASICREV_BCM5700 || sc->bge_asicrev == BGE_ASICREV_BCM5701) dma_rw_ctl |= BGE_PCIDMARWCTL_USE_MRM | BGE_PCIDMARWCTL_ASRT_ALL_BE; if (sc->bge_asicrev == BGE_ASICREV_BCM5703 || sc->bge_asicrev == BGE_ASICREV_BCM5704) dma_rw_ctl &= ~BGE_PCIDMARWCTL_MINDMA; if (BGE_IS_5717_PLUS(sc)) { dma_rw_ctl &= ~BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT; if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK; /* * Enable HW workaround for controllers that misinterpret * a status tag update and leave interrupts permanently * disabled. */ if (sc->bge_asicrev != BGE_ASICREV_BCM5717 && sc->bge_asicrev != BGE_ASICREV_BCM57765) dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA; } pci_write_config(sc->bge_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4); /* * Set up general mode register. */ mode_ctl = bge_dma_swap_options(sc) | BGE_MODECTL_MAC_ATTN_INTR | BGE_MODECTL_HOST_SEND_BDS | BGE_MODECTL_TX_NO_PHDR_CSUM; /* * BCM5701 B5 have a bug causing data corruption when using * 64-bit DMA reads, which can be terminated early and then * completed later as 32-bit accesses, in combination with * certain bridges. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5701 && sc->bge_chipid == BGE_CHIPID_BCM5701_B5) mode_ctl |= BGE_MODECTL_FORCE_PCI32; /* * Tell the firmware the driver is running */ if (sc->bge_asf_mode & ASF_STACKUP) mode_ctl |= BGE_MODECTL_STACKUP; CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl); /* * Disable memory write invalidate. Apparently it is not supported * properly by these devices. Also ensure that INTx isn't disabled, * as these chips need it even when using MSI. */ PCI_CLRBIT(sc->bge_dev, BGE_PCI_CMD, PCIM_CMD_INTxDIS | PCIM_CMD_MWIEN, 4); /* Set the timer prescaler (always 66Mhz) */ CSR_WRITE_4(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ); /* XXX: The Linux tg3 driver does this at the start of brgphy_reset. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5906) { DELAY(40); /* XXX */ /* Put PHY into ready state */ BGE_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ); CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */ DELAY(40); } return (0); } static int bge_blockinit(struct bge_softc *sc) { struct bge_rcb *rcb; bus_size_t vrcb; bge_hostaddr taddr; uint32_t dmactl, val; int i, limit; /* * Initialize the memory window pointer register so that * we can access the first 32K of internal NIC RAM. This will * allow us to set up the TX send ring RCBs and the RX return * ring RCBs, plus other things which live in NIC memory. */ CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0); /* Note: the BCM5704 has a smaller mbuf space than other chips. */ if (!(BGE_IS_5705_PLUS(sc))) { /* Configure mbuf memory pool */ CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_BASEADDR, BGE_BUFFPOOL_1); if (sc->bge_asicrev == BGE_ASICREV_BCM5704) CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x10000); else CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_LEN, 0x18000); /* Configure DMA resource pool */ CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_BASEADDR, BGE_DMA_DESCRIPTORS); CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LEN, 0x2000); } /* Configure mbuf pool watermarks */ if (BGE_IS_5717_PLUS(sc)) { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0); if (sc->bge_ifp->if_mtu > ETHERMTU) { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea); } else { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0); } } else if (!BGE_IS_5705_PLUS(sc)) { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x50); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x20); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60); } else if (sc->bge_asicrev == BGE_ASICREV_BCM5906) { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10); } else { CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10); CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60); } /* Configure DMA resource watermarks */ CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5); CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10); /* Enable buffer manager */ val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN; /* * Change the arbitration algorithm of TXMBUF read request to * round-robin instead of priority based for BCM5719. When * TXFIFO is almost empty, RDMA will hold its request until * TXFIFO is not almost empty. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5719) val |= BGE_BMANMODE_NO_TX_UNDERRUN; CSR_WRITE_4(sc, BGE_BMAN_MODE, val); /* Poll for buffer manager start indication */ for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE) break; } if (i == BGE_TIMEOUT) { device_printf(sc->bge_dev, "buffer manager failed to start\n"); return (ENXIO); } /* Enable flow-through queues */ CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF); CSR_WRITE_4(sc, BGE_FTQ_RESET, 0); /* Wait until queue initialization is complete */ for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0) break; } if (i == BGE_TIMEOUT) { device_printf(sc->bge_dev, "flow-through queue init failed\n"); return (ENXIO); } /* * Summary of rings supported by the controller: * * Standard Receive Producer Ring * - This ring is used to feed receive buffers for "standard" * sized frames (typically 1536 bytes) to the controller. * * Jumbo Receive Producer Ring * - This ring is used to feed receive buffers for jumbo sized * frames (i.e. anything bigger than the "standard" frames) * to the controller. * * Mini Receive Producer Ring * - This ring is used to feed receive buffers for "mini" * sized frames to the controller. * - This feature required external memory for the controller * but was never used in a production system. Should always * be disabled. * * Receive Return Ring * - After the controller has placed an incoming frame into a * receive buffer that buffer is moved into a receive return * ring. The driver is then responsible to passing the * buffer up to the stack. Many versions of the controller * support multiple RR rings. * * Send Ring * - This ring is used for outgoing frames. Many versions of * the controller support multiple send rings. */ /* Initialize the standard receive producer ring control block. */ rcb = &sc->bge_ldata.bge_info.bge_std_rx_rcb; rcb->bge_hostaddr.bge_addr_lo = BGE_ADDR_LO(sc->bge_ldata.bge_rx_std_ring_paddr); rcb->bge_hostaddr.bge_addr_hi = BGE_ADDR_HI(sc->bge_ldata.bge_rx_std_ring_paddr); bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREREAD); if (BGE_IS_5717_PLUS(sc)) { /* * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32) * Bits 15-2 : Maximum RX frame size * Bit 1 : 1 = Ring Disabled, 0 = Ring ENabled * Bit 0 : Reserved */ rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, BGE_MAX_FRAMELEN << 2); } else if (BGE_IS_5705_PLUS(sc)) { /* * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32) * Bits 15-2 : Reserved (should be 0) * Bit 1 : 1 = Ring Disabled, 0 = Ring Enabled * Bit 0 : Reserved */ rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0); } else { /* * Ring size is always XXX entries * Bits 31-16: Maximum RX frame size * Bits 15-2 : Reserved (should be 0) * Bit 1 : 1 = Ring Disabled, 0 = Ring Enabled * Bit 0 : Reserved */ rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(BGE_MAX_FRAMELEN, 0); } if (sc->bge_asicrev == BGE_ASICREV_BCM5717 || sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717; else rcb->bge_nicaddr = BGE_STD_RX_RINGS; /* Write the standard receive producer ring control block. */ CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi); CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo); CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags); CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr); /* Reset the standard receive producer ring producer index. */ bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0); /* * Initialize the jumbo RX producer ring control * block. We set the 'ring disabled' bit in the * flags field until we're actually ready to start * using this ring (i.e. once we set the MTU * high enough to require it). */ if (BGE_IS_JUMBO_CAPABLE(sc)) { rcb = &sc->bge_ldata.bge_info.bge_jumbo_rx_rcb; /* Get the jumbo receive producer ring RCB parameters. */ rcb->bge_hostaddr.bge_addr_lo = BGE_ADDR_LO(sc->bge_ldata.bge_rx_jumbo_ring_paddr); rcb->bge_hostaddr.bge_addr_hi = BGE_ADDR_HI(sc->bge_ldata.bge_rx_jumbo_ring_paddr); bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREREAD); rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_USE_EXT_RX_BD | BGE_RCB_FLAG_RING_DISABLED); if (sc->bge_asicrev == BGE_ASICREV_BCM5717 || sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717; else rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS; CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi); CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo); /* Program the jumbo receive producer ring RCB parameters. */ CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags); CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr); /* Reset the jumbo receive producer ring producer index. */ bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0); } /* Disable the mini receive producer ring RCB. */ if (BGE_IS_5700_FAMILY(sc)) { rcb = &sc->bge_ldata.bge_info.bge_mini_rx_rcb; rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED); CSR_WRITE_4(sc, BGE_RX_MINI_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags); /* Reset the mini receive producer ring producer index. */ bge_writembx(sc, BGE_MBX_RX_MINI_PROD_LO, 0); } /* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5906) { if (sc->bge_chipid == BGE_CHIPID_BCM5906_A0 || sc->bge_chipid == BGE_CHIPID_BCM5906_A1 || sc->bge_chipid == BGE_CHIPID_BCM5906_A2) CSR_WRITE_4(sc, BGE_ISO_PKT_TX, (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2); } /* * The BD ring replenish thresholds control how often the * hardware fetches new BD's from the producer rings in host * memory. Setting the value too low on a busy system can * starve the hardware and recue the throughpout. * * Set the BD ring replentish thresholds. The recommended * values are 1/8th the number of descriptors allocated to * each ring. * XXX The 5754 requires a lower threshold, so it might be a * requirement of all 575x family chips. The Linux driver sets * the lower threshold for all 5705 family chips as well, but there * are reports that it might not need to be so strict. * * XXX Linux does some extra fiddling here for the 5906 parts as * well. */ if (BGE_IS_5705_PLUS(sc)) val = 8; else val = BGE_STD_RX_RING_CNT / 8; CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val); if (BGE_IS_JUMBO_CAPABLE(sc)) CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH, BGE_JUMBO_RX_RING_CNT/8); if (BGE_IS_5717_PLUS(sc)) { CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32); CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16); } /* * Disable all send rings by setting the 'ring disabled' bit * in the flags field of all the TX send ring control blocks, * located in NIC memory. */ if (!BGE_IS_5705_PLUS(sc)) /* 5700 to 5704 had 16 send rings. */ limit = BGE_TX_RINGS_EXTSSRAM_MAX; else limit = 1; vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB; for (i = 0; i < limit; i++) { RCB_WRITE_4(sc, vrcb, bge_maxlen_flags, BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED)); RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0); vrcb += sizeof(struct bge_rcb); } /* Configure send ring RCB 0 (we use only the first ring) */ vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB; BGE_HOSTADDR(taddr, sc->bge_ldata.bge_tx_ring_paddr); RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi); RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo); if (sc->bge_asicrev == BGE_ASICREV_BCM5717 || sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717); else RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT)); RCB_WRITE_4(sc, vrcb, bge_maxlen_flags, BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0)); /* * Disable all receive return rings by setting the * 'ring diabled' bit in the flags field of all the receive * return ring control blocks, located in NIC memory. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5717 || sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) { /* Should be 17, use 16 until we get an SRAM map. */ limit = 16; } else if (!BGE_IS_5705_PLUS(sc)) limit = BGE_RX_RINGS_MAX; else if (sc->bge_asicrev == BGE_ASICREV_BCM5755 || sc->bge_asicrev == BGE_ASICREV_BCM57765) limit = 4; else limit = 1; /* Disable all receive return rings. */ vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB; for (i = 0; i < limit; i++) { RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0); RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0); RCB_WRITE_4(sc, vrcb, bge_maxlen_flags, BGE_RCB_FLAG_RING_DISABLED); RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0); bge_writembx(sc, BGE_MBX_RX_CONS0_LO + (i * (sizeof(uint64_t))), 0); vrcb += sizeof(struct bge_rcb); } /* * Set up receive return ring 0. Note that the NIC address * for RX return rings is 0x0. The return rings live entirely * within the host, so the nicaddr field in the RCB isn't used. */ vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB; BGE_HOSTADDR(taddr, sc->bge_ldata.bge_rx_return_ring_paddr); RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi); RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo); RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0); RCB_WRITE_4(sc, vrcb, bge_maxlen_flags, BGE_RCB_MAXLEN_FLAGS(sc->bge_return_ring_cnt, 0)); /* Set random backoff seed for TX */ CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF, IF_LLADDR(sc->bge_ifp)[0] + IF_LLADDR(sc->bge_ifp)[1] + IF_LLADDR(sc->bge_ifp)[2] + IF_LLADDR(sc->bge_ifp)[3] + IF_LLADDR(sc->bge_ifp)[4] + IF_LLADDR(sc->bge_ifp)[5] + BGE_TX_BACKOFF_SEED_MASK); /* Set inter-packet gap */ val = 0x2620; if (sc->bge_asicrev == BGE_ASICREV_BCM5720) val |= CSR_READ_4(sc, BGE_TX_LENGTHS) & (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK); CSR_WRITE_4(sc, BGE_TX_LENGTHS, val); /* * Specify which ring to use for packets that don't match * any RX rules. */ CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08); /* * Configure number of RX lists. One interrupt distribution * list, sixteen active lists, one bad frames class. */ CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181); /* Inialize RX list placement stats mask. */ CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF); CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1); /* Disable host coalescing until we get it set up */ CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000); /* Poll to make sure it's shut down. */ for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE)) break; } if (i == BGE_TIMEOUT) { device_printf(sc->bge_dev, "host coalescing engine failed to idle\n"); return (ENXIO); } /* Set up host coalescing defaults */ CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bge_rx_coal_ticks); CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bge_tx_coal_ticks); CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bge_rx_max_coal_bds); CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bge_tx_max_coal_bds); if (!(BGE_IS_5705_PLUS(sc))) { CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS_INT, 0); CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS_INT, 0); } CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 1); CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 1); /* Set up address of statistics block */ if (!(BGE_IS_5705_PLUS(sc))) { CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_HI, BGE_ADDR_HI(sc->bge_ldata.bge_stats_paddr)); CSR_WRITE_4(sc, BGE_HCC_STATS_ADDR_LO, BGE_ADDR_LO(sc->bge_ldata.bge_stats_paddr)); CSR_WRITE_4(sc, BGE_HCC_STATS_BASEADDR, BGE_STATS_BLOCK); CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_BASEADDR, BGE_STATUS_BLOCK); CSR_WRITE_4(sc, BGE_HCC_STATS_TICKS, sc->bge_stat_ticks); } /* Set up address of status block */ CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI, BGE_ADDR_HI(sc->bge_ldata.bge_status_block_paddr)); CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO, BGE_ADDR_LO(sc->bge_ldata.bge_status_block_paddr)); /* Set up status block size. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_C0) { val = BGE_STATBLKSZ_FULL; bzero(sc->bge_ldata.bge_status_block, BGE_STATUS_BLK_SZ); } else { val = BGE_STATBLKSZ_32BYTE; bzero(sc->bge_ldata.bge_status_block, 32); } bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* Turn on host coalescing state machine */ CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE); /* Turn on RX BD completion state machine and enable attentions */ CSR_WRITE_4(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE | BGE_RBDCMODE_ATTN); /* Turn on RX list placement state machine */ CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE); /* Turn on RX list selector state machine. */ if (!(BGE_IS_5705_PLUS(sc))) CSR_WRITE_4(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE); val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB | BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR | BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB | BGE_MACMODE_FRMHDR_DMA_ENB; if (sc->bge_flags & BGE_FLAG_TBI) val |= BGE_PORTMODE_TBI; else if (sc->bge_flags & BGE_FLAG_MII_SERDES) val |= BGE_PORTMODE_GMII; else val |= BGE_PORTMODE_MII; /* Turn on DMA, clear stats */ CSR_WRITE_4(sc, BGE_MAC_MODE, val); /* Set misc. local control, enable interrupts on attentions */ CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN); #ifdef notdef /* Assert GPIO pins for PHY reset */ BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0 | BGE_MLC_MISCIO_OUT1 | BGE_MLC_MISCIO_OUT2); BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0 | BGE_MLC_MISCIO_OUTEN1 | BGE_MLC_MISCIO_OUTEN2); #endif /* Turn on DMA completion state machine */ if (!(BGE_IS_5705_PLUS(sc))) CSR_WRITE_4(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE); val = BGE_WDMAMODE_ENABLE | BGE_WDMAMODE_ALL_ATTNS; /* Enable host coalescing bug fix. */ if (BGE_IS_5755_PLUS(sc)) val |= BGE_WDMAMODE_STATUS_TAG_FIX; /* Request larger DMA burst size to get better performance. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5785) val |= BGE_WDMAMODE_BURST_ALL_DATA; /* Turn on write DMA state machine */ CSR_WRITE_4(sc, BGE_WDMA_MODE, val); DELAY(40); /* Turn on read DMA state machine */ val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS; if (sc->bge_asicrev == BGE_ASICREV_BCM5717) val |= BGE_RDMAMODE_MULT_DMA_RD_DIS; if (sc->bge_asicrev == BGE_ASICREV_BCM5784 || sc->bge_asicrev == BGE_ASICREV_BCM5785 || sc->bge_asicrev == BGE_ASICREV_BCM57780) val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN | BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN | BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN; if (sc->bge_flags & BGE_FLAG_PCIE) val |= BGE_RDMAMODE_FIFO_LONG_BURST; if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) { val |= BGE_RDMAMODE_TSO4_ENABLE; if (sc->bge_flags & BGE_FLAG_TSO3 || sc->bge_asicrev == BGE_ASICREV_BCM5785 || sc->bge_asicrev == BGE_ASICREV_BCM57780) val |= BGE_RDMAMODE_TSO6_ENABLE; } if (sc->bge_asicrev == BGE_ASICREV_BCM5720) { val |= CSR_READ_4(sc, BGE_RDMA_MODE) & BGE_RDMAMODE_H2BNC_VLAN_DET; /* * Allow multiple outstanding read requests from * non-LSO read DMA engine. */ val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS; } if (sc->bge_asicrev == BGE_ASICREV_BCM5761 || sc->bge_asicrev == BGE_ASICREV_BCM5784 || sc->bge_asicrev == BGE_ASICREV_BCM5785 || sc->bge_asicrev == BGE_ASICREV_BCM57780 || BGE_IS_5717_PLUS(sc)) { dmactl = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL); /* * Adjust tx margin to prevent TX data corruption and * fix internal FIFO overflow. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) { dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK | BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK | BGE_RDMA_RSRVCTRL_TXMRGN_MASK); dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K | BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K | BGE_RDMA_RSRVCTRL_TXMRGN_320B; } /* * Enable fix for read DMA FIFO overruns. * The fix is to limit the number of RX BDs * the hardware would fetch at a fime. */ CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL, dmactl | BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX); } if (sc->bge_asicrev == BGE_ASICREV_BCM5719) { CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) | BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K | BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K); } else if (sc->bge_asicrev == BGE_ASICREV_BCM5720) { /* * Allow 4KB burst length reads for non-LSO frames. * Enable 512B burst length reads for buffer descriptors. */ CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL, CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) | BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 | BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K); } CSR_WRITE_4(sc, BGE_RDMA_MODE, val); DELAY(40); /* Turn on RX data completion state machine */ CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE); /* Turn on RX BD initiator state machine */ CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE); /* Turn on RX data and RX BD initiator state machine */ CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE); /* Turn on Mbuf cluster free state machine */ if (!(BGE_IS_5705_PLUS(sc))) CSR_WRITE_4(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE); /* Turn on send BD completion state machine */ CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE); /* Turn on send data completion state machine */ val = BGE_SDCMODE_ENABLE; if (sc->bge_asicrev == BGE_ASICREV_BCM5761) val |= BGE_SDCMODE_CDELAY; CSR_WRITE_4(sc, BGE_SDC_MODE, val); /* Turn on send data initiator state machine */ if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE | BGE_SDIMODE_HW_LSO_PRE_DMA); else CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE); /* Turn on send BD initiator state machine */ CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE); /* Turn on send BD selector state machine */ CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE); CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF); CSR_WRITE_4(sc, BGE_SDI_STATS_CTL, BGE_SDISTATSCTL_ENABLE | BGE_SDISTATSCTL_FASTER); /* ack/clear link change events */ CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED | BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE | BGE_MACSTAT_LINK_CHANGED); CSR_WRITE_4(sc, BGE_MI_STS, 0); /* * Enable attention when the link has changed state for * devices that use auto polling. */ if (sc->bge_flags & BGE_FLAG_TBI) { CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK); } else { if (sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) { CSR_WRITE_4(sc, BGE_MI_MODE, sc->bge_mi_mode); DELAY(80); } if (sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_B2) CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT); } /* * Clear any pending link state attention. * Otherwise some link state change events may be lost until attention * is cleared by bge_intr() -> bge_link_upd() sequence. * It's not necessary on newer BCM chips - perhaps enabling link * state change attentions implies clearing pending attention. */ CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED | BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE | BGE_MACSTAT_LINK_CHANGED); /* Enable link state change attentions. */ BGE_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED); return (0); } const struct bge_revision * bge_lookup_rev(uint32_t chipid) { const struct bge_revision *br; for (br = bge_revisions; br->br_name != NULL; br++) { if (br->br_chipid == chipid) return (br); } for (br = bge_majorrevs; br->br_name != NULL; br++) { if (br->br_chipid == BGE_ASICREV(chipid)) return (br); } return (NULL); } const struct bge_vendor * bge_lookup_vendor(uint16_t vid) { const struct bge_vendor *v; for (v = bge_vendors; v->v_name != NULL; v++) if (v->v_id == vid) return (v); panic("%s: unknown vendor %d", __func__, vid); return (NULL); } /* * Probe for a Broadcom chip. Check the PCI vendor and device IDs * against our list and return its name if we find a match. * * Note that since the Broadcom controller contains VPD support, we * try to get the device name string from the controller itself instead * of the compiled-in string. It guarantees we'll always announce the * right product name. We fall back to the compiled-in string when * VPD is unavailable or corrupt. */ static int bge_probe(device_t dev) { char buf[96]; char model[64]; const struct bge_revision *br; const char *pname; struct bge_softc *sc = device_get_softc(dev); const struct bge_type *t = bge_devs; const struct bge_vendor *v; uint32_t id; uint16_t did, vid; sc->bge_dev = dev; vid = pci_get_vendor(dev); did = pci_get_device(dev); while(t->bge_vid != 0) { if ((vid == t->bge_vid) && (did == t->bge_did)) { id = pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >> BGE_PCIMISCCTL_ASICREV_SHIFT; if (BGE_ASICREV(id) == BGE_ASICREV_USE_PRODID_REG) { /* * Find the ASCI revision. Different chips * use different registers. */ switch (pci_get_device(dev)) { case BCOM_DEVICEID_BCM5717: case BCOM_DEVICEID_BCM5718: case BCOM_DEVICEID_BCM5719: case BCOM_DEVICEID_BCM5720: id = pci_read_config(dev, BGE_PCI_GEN2_PRODID_ASICREV, 4); break; case BCOM_DEVICEID_BCM57761: case BCOM_DEVICEID_BCM57765: case BCOM_DEVICEID_BCM57781: case BCOM_DEVICEID_BCM57785: case BCOM_DEVICEID_BCM57791: case BCOM_DEVICEID_BCM57795: id = pci_read_config(dev, BGE_PCI_GEN15_PRODID_ASICREV, 4); break; default: id = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4); } } br = bge_lookup_rev(id); v = bge_lookup_vendor(vid); if (bge_has_eaddr(sc) && pci_get_vpd_ident(dev, &pname) == 0) snprintf(model, 64, "%s", pname); else snprintf(model, 64, "%s %s", v->v_name, br != NULL ? br->br_name : "NetXtreme Ethernet Controller"); snprintf(buf, 96, "%s, %sASIC rev. %#08x", model, br != NULL ? "" : "unknown ", id); device_set_desc_copy(dev, buf); return (0); } t++; } return (ENXIO); } static void bge_dma_free(struct bge_softc *sc) { int i; /* Destroy DMA maps for RX buffers. */ for (i = 0; i < BGE_STD_RX_RING_CNT; i++) { if (sc->bge_cdata.bge_rx_std_dmamap[i]) bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_dmamap[i]); } if (sc->bge_cdata.bge_rx_std_sparemap) bus_dmamap_destroy(sc->bge_cdata.bge_rx_mtag, sc->bge_cdata.bge_rx_std_sparemap); /* Destroy DMA maps for jumbo RX buffers. */ for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) { if (sc->bge_cdata.bge_rx_jumbo_dmamap[i]) bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_dmamap[i]); } if (sc->bge_cdata.bge_rx_jumbo_sparemap) bus_dmamap_destroy(sc->bge_cdata.bge_mtag_jumbo, sc->bge_cdata.bge_rx_jumbo_sparemap); /* Destroy DMA maps for TX buffers. */ for (i = 0; i < BGE_TX_RING_CNT; i++) { if (sc->bge_cdata.bge_tx_dmamap[i]) bus_dmamap_destroy(sc->bge_cdata.bge_tx_mtag, sc->bge_cdata.bge_tx_dmamap[i]); } if (sc->bge_cdata.bge_rx_mtag) bus_dma_tag_destroy(sc->bge_cdata.bge_rx_mtag); if (sc->bge_cdata.bge_mtag_jumbo) bus_dma_tag_destroy(sc->bge_cdata.bge_mtag_jumbo); if (sc->bge_cdata.bge_tx_mtag) bus_dma_tag_destroy(sc->bge_cdata.bge_tx_mtag); /* Destroy standard RX ring. */ if (sc->bge_cdata.bge_rx_std_ring_map) bus_dmamap_unload(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_cdata.bge_rx_std_ring_map); if (sc->bge_cdata.bge_rx_std_ring_map && sc->bge_ldata.bge_rx_std_ring) bus_dmamem_free(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_ldata.bge_rx_std_ring, sc->bge_cdata.bge_rx_std_ring_map); if (sc->bge_cdata.bge_rx_std_ring_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_rx_std_ring_tag); /* Destroy jumbo RX ring. */ if (sc->bge_cdata.bge_rx_jumbo_ring_map) bus_dmamap_unload(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_cdata.bge_rx_jumbo_ring_map); if (sc->bge_cdata.bge_rx_jumbo_ring_map && sc->bge_ldata.bge_rx_jumbo_ring) bus_dmamem_free(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_ldata.bge_rx_jumbo_ring, sc->bge_cdata.bge_rx_jumbo_ring_map); if (sc->bge_cdata.bge_rx_jumbo_ring_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_rx_jumbo_ring_tag); /* Destroy RX return ring. */ if (sc->bge_cdata.bge_rx_return_ring_map) bus_dmamap_unload(sc->bge_cdata.bge_rx_return_ring_tag, sc->bge_cdata.bge_rx_return_ring_map); if (sc->bge_cdata.bge_rx_return_ring_map && sc->bge_ldata.bge_rx_return_ring) bus_dmamem_free(sc->bge_cdata.bge_rx_return_ring_tag, sc->bge_ldata.bge_rx_return_ring, sc->bge_cdata.bge_rx_return_ring_map); if (sc->bge_cdata.bge_rx_return_ring_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_rx_return_ring_tag); /* Destroy TX ring. */ if (sc->bge_cdata.bge_tx_ring_map) bus_dmamap_unload(sc->bge_cdata.bge_tx_ring_tag, sc->bge_cdata.bge_tx_ring_map); if (sc->bge_cdata.bge_tx_ring_map && sc->bge_ldata.bge_tx_ring) bus_dmamem_free(sc->bge_cdata.bge_tx_ring_tag, sc->bge_ldata.bge_tx_ring, sc->bge_cdata.bge_tx_ring_map); if (sc->bge_cdata.bge_tx_ring_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_tx_ring_tag); /* Destroy status block. */ if (sc->bge_cdata.bge_status_map) bus_dmamap_unload(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map); if (sc->bge_cdata.bge_status_map && sc->bge_ldata.bge_status_block) bus_dmamem_free(sc->bge_cdata.bge_status_tag, sc->bge_ldata.bge_status_block, sc->bge_cdata.bge_status_map); if (sc->bge_cdata.bge_status_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_status_tag); /* Destroy statistics block. */ if (sc->bge_cdata.bge_stats_map) bus_dmamap_unload(sc->bge_cdata.bge_stats_tag, sc->bge_cdata.bge_stats_map); if (sc->bge_cdata.bge_stats_map && sc->bge_ldata.bge_stats) bus_dmamem_free(sc->bge_cdata.bge_stats_tag, sc->bge_ldata.bge_stats, sc->bge_cdata.bge_stats_map); if (sc->bge_cdata.bge_stats_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_stats_tag); if (sc->bge_cdata.bge_buffer_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_buffer_tag); /* Destroy the parent tag. */ if (sc->bge_cdata.bge_parent_tag) bus_dma_tag_destroy(sc->bge_cdata.bge_parent_tag); } static int bge_dma_ring_alloc(struct bge_softc *sc, bus_size_t alignment, bus_size_t maxsize, bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map, bus_addr_t *paddr, const char *msg) { struct bge_dmamap_arg ctx; bus_addr_t lowaddr; bus_size_t ring_end; int error; lowaddr = BUS_SPACE_MAXADDR; again: error = bus_dma_tag_create(sc->bge_cdata.bge_parent_tag, alignment, 0, lowaddr, BUS_SPACE_MAXADDR, NULL, NULL, maxsize, 1, maxsize, 0, NULL, NULL, tag); if (error != 0) { device_printf(sc->bge_dev, "could not create %s dma tag\n", msg); return (ENOMEM); } /* Allocate DMA'able memory for ring. */ error = bus_dmamem_alloc(*tag, (void **)ring, BUS_DMA_NOWAIT | BUS_DMA_ZERO | BUS_DMA_COHERENT, map); if (error != 0) { device_printf(sc->bge_dev, "could not allocate DMA'able memory for %s\n", msg); return (ENOMEM); } /* Load the address of the ring. */ ctx.bge_busaddr = 0; error = bus_dmamap_load(*tag, *map, *ring, maxsize, bge_dma_map_addr, &ctx, BUS_DMA_NOWAIT); if (error != 0) { device_printf(sc->bge_dev, "could not load DMA'able memory for %s\n", msg); return (ENOMEM); } *paddr = ctx.bge_busaddr; ring_end = *paddr + maxsize; if ((sc->bge_flags & BGE_FLAG_4G_BNDRY_BUG) != 0 && BGE_ADDR_HI(*paddr) != BGE_ADDR_HI(ring_end)) { /* * 4GB boundary crossed. Limit maximum allowable DMA * address space to 32bit and try again. */ bus_dmamap_unload(*tag, *map); bus_dmamem_free(*tag, *ring, *map); bus_dma_tag_destroy(*tag); if (bootverbose) device_printf(sc->bge_dev, "4GB boundary crossed, " "limit DMA address space to 32bit for %s\n", msg); *ring = NULL; *tag = NULL; *map = NULL; lowaddr = BUS_SPACE_MAXADDR_32BIT; goto again; } return (0); } static int bge_dma_alloc(struct bge_softc *sc) { bus_addr_t lowaddr; bus_size_t boundary, sbsz, rxmaxsegsz, txsegsz, txmaxsegsz; int i, error; lowaddr = BUS_SPACE_MAXADDR; if ((sc->bge_flags & BGE_FLAG_40BIT_BUG) != 0) lowaddr = BGE_DMA_MAXADDR; /* * Allocate the parent bus DMA tag appropriate for PCI. */ error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev), 1, 0, lowaddr, BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL, &sc->bge_cdata.bge_parent_tag); if (error != 0) { device_printf(sc->bge_dev, "could not allocate parent dma tag\n"); return (ENOMEM); } /* Create tag for standard RX ring. */ error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STD_RX_RING_SZ, &sc->bge_cdata.bge_rx_std_ring_tag, (uint8_t **)&sc->bge_ldata.bge_rx_std_ring, &sc->bge_cdata.bge_rx_std_ring_map, &sc->bge_ldata.bge_rx_std_ring_paddr, "RX ring"); if (error) return (error); /* Create tag for RX return ring. */ error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_RX_RTN_RING_SZ(sc), &sc->bge_cdata.bge_rx_return_ring_tag, (uint8_t **)&sc->bge_ldata.bge_rx_return_ring, &sc->bge_cdata.bge_rx_return_ring_map, &sc->bge_ldata.bge_rx_return_ring_paddr, "RX return ring"); if (error) return (error); /* Create tag for TX ring. */ error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_TX_RING_SZ, &sc->bge_cdata.bge_tx_ring_tag, (uint8_t **)&sc->bge_ldata.bge_tx_ring, &sc->bge_cdata.bge_tx_ring_map, &sc->bge_ldata.bge_tx_ring_paddr, "TX ring"); if (error) return (error); /* * Create tag for status block. * Because we only use single Tx/Rx/Rx return ring, use * minimum status block size except BCM5700 AX/BX which * seems to want to see full status block size regardless * of configured number of ring. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_C0) sbsz = BGE_STATUS_BLK_SZ; else sbsz = 32; error = bge_dma_ring_alloc(sc, PAGE_SIZE, sbsz, &sc->bge_cdata.bge_status_tag, (uint8_t **)&sc->bge_ldata.bge_status_block, &sc->bge_cdata.bge_status_map, &sc->bge_ldata.bge_status_block_paddr, "status block"); if (error) return (error); /* Create tag for statistics block. */ error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_STATS_SZ, &sc->bge_cdata.bge_stats_tag, (uint8_t **)&sc->bge_ldata.bge_stats, &sc->bge_cdata.bge_stats_map, &sc->bge_ldata.bge_stats_paddr, "statistics block"); if (error) return (error); /* Create tag for jumbo RX ring. */ if (BGE_IS_JUMBO_CAPABLE(sc)) { error = bge_dma_ring_alloc(sc, PAGE_SIZE, BGE_JUMBO_RX_RING_SZ, &sc->bge_cdata.bge_rx_jumbo_ring_tag, (uint8_t **)&sc->bge_ldata.bge_rx_jumbo_ring, &sc->bge_cdata.bge_rx_jumbo_ring_map, &sc->bge_ldata.bge_rx_jumbo_ring_paddr, "jumbo RX ring"); if (error) return (error); } /* Create parent tag for buffers. */ boundary = 0; if ((sc->bge_flags & BGE_FLAG_4G_BNDRY_BUG) != 0) { boundary = BGE_DMA_BNDRY; /* * XXX * watchdog timeout issue was observed on BCM5704 which * lives behind PCI-X bridge(e.g AMD 8131 PCI-X bridge). * Both limiting DMA address space to 32bits and flushing * mailbox write seem to address the issue. */ if (sc->bge_pcixcap != 0) lowaddr = BUS_SPACE_MAXADDR_32BIT; } error = bus_dma_tag_create(bus_get_dma_tag(sc->bge_dev), 1, boundary, lowaddr, BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL, &sc->bge_cdata.bge_buffer_tag); if (error != 0) { device_printf(sc->bge_dev, "could not allocate buffer dma tag\n"); return (ENOMEM); } /* Create tag for Tx mbufs. */ if (sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) { txsegsz = BGE_TSOSEG_SZ; txmaxsegsz = 65535 + sizeof(struct ether_vlan_header); } else { txsegsz = MCLBYTES; txmaxsegsz = MCLBYTES * BGE_NSEG_NEW; } error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, txmaxsegsz, BGE_NSEG_NEW, txsegsz, 0, NULL, NULL, &sc->bge_cdata.bge_tx_mtag); if (error) { device_printf(sc->bge_dev, "could not allocate TX dma tag\n"); return (ENOMEM); } /* Create tag for Rx mbufs. */ if (sc->bge_flags & BGE_FLAG_JUMBO_STD) rxmaxsegsz = MJUM9BYTES; else rxmaxsegsz = MCLBYTES; error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, rxmaxsegsz, 1, rxmaxsegsz, 0, NULL, NULL, &sc->bge_cdata.bge_rx_mtag); if (error) { device_printf(sc->bge_dev, "could not allocate RX dma tag\n"); return (ENOMEM); } /* Create DMA maps for RX buffers. */ error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0, &sc->bge_cdata.bge_rx_std_sparemap); if (error) { device_printf(sc->bge_dev, "can't create spare DMA map for RX\n"); return (ENOMEM); } for (i = 0; i < BGE_STD_RX_RING_CNT; i++) { error = bus_dmamap_create(sc->bge_cdata.bge_rx_mtag, 0, &sc->bge_cdata.bge_rx_std_dmamap[i]); if (error) { device_printf(sc->bge_dev, "can't create DMA map for RX\n"); return (ENOMEM); } } /* Create DMA maps for TX buffers. */ for (i = 0; i < BGE_TX_RING_CNT; i++) { error = bus_dmamap_create(sc->bge_cdata.bge_tx_mtag, 0, &sc->bge_cdata.bge_tx_dmamap[i]); if (error) { device_printf(sc->bge_dev, "can't create DMA map for TX\n"); return (ENOMEM); } } /* Create tags for jumbo RX buffers. */ if (BGE_IS_JUMBO_CAPABLE(sc)) { error = bus_dma_tag_create(sc->bge_cdata.bge_buffer_tag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MJUM9BYTES, BGE_NSEG_JUMBO, PAGE_SIZE, 0, NULL, NULL, &sc->bge_cdata.bge_mtag_jumbo); if (error) { device_printf(sc->bge_dev, "could not allocate jumbo dma tag\n"); return (ENOMEM); } /* Create DMA maps for jumbo RX buffers. */ error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo, 0, &sc->bge_cdata.bge_rx_jumbo_sparemap); if (error) { device_printf(sc->bge_dev, "can't create spare DMA map for jumbo RX\n"); return (ENOMEM); } for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) { error = bus_dmamap_create(sc->bge_cdata.bge_mtag_jumbo, 0, &sc->bge_cdata.bge_rx_jumbo_dmamap[i]); if (error) { device_printf(sc->bge_dev, "can't create DMA map for jumbo RX\n"); return (ENOMEM); } } } return (0); } /* * Return true if this device has more than one port. */ static int bge_has_multiple_ports(struct bge_softc *sc) { device_t dev = sc->bge_dev; u_int b, d, f, fscan, s; d = pci_get_domain(dev); b = pci_get_bus(dev); s = pci_get_slot(dev); f = pci_get_function(dev); for (fscan = 0; fscan <= PCI_FUNCMAX; fscan++) if (fscan != f && pci_find_dbsf(d, b, s, fscan) != NULL) return (1); return (0); } /* * Return true if MSI can be used with this device. */ static int bge_can_use_msi(struct bge_softc *sc) { int can_use_msi = 0; if (sc->bge_msi == 0) return (0); /* Disable MSI for polling(4). */ #ifdef DEVICE_POLLING return (0); #endif switch (sc->bge_asicrev) { case BGE_ASICREV_BCM5714_A0: case BGE_ASICREV_BCM5714: /* * Apparently, MSI doesn't work when these chips are * configured in single-port mode. */ if (bge_has_multiple_ports(sc)) can_use_msi = 1; break; case BGE_ASICREV_BCM5750: if (sc->bge_chiprev != BGE_CHIPREV_5750_AX && sc->bge_chiprev != BGE_CHIPREV_5750_BX) can_use_msi = 1; break; default: if (BGE_IS_575X_PLUS(sc)) can_use_msi = 1; } return (can_use_msi); } static int bge_mbox_reorder(struct bge_softc *sc) { /* Lists of PCI bridges that are known to reorder mailbox writes. */ static const struct mbox_reorder { const uint16_t vendor; const uint16_t device; const char *desc; } const mbox_reorder_lists[] = { { 0x1022, 0x7450, "AMD-8131 PCI-X Bridge" }, }; devclass_t pci, pcib; device_t bus, dev; int count, i; count = sizeof(mbox_reorder_lists) / sizeof(mbox_reorder_lists[0]); pci = devclass_find("pci"); pcib = devclass_find("pcib"); dev = sc->bge_dev; bus = device_get_parent(dev); for (;;) { dev = device_get_parent(bus); bus = device_get_parent(dev); device_printf(sc->bge_dev, "dev : %s%d, bus : %s%d\n", device_get_name(dev), device_get_unit(dev), device_get_name(bus), device_get_unit(bus)); if (device_get_devclass(dev) != pcib) break; for (i = 0; i < count; i++) { device_printf(sc->bge_dev, "probing dev : %s%d, vendor : 0x%04x " "device : 0x%04x\n", device_get_name(dev), device_get_unit(dev), pci_get_vendor(dev), pci_get_device(dev)); if (pci_get_vendor(dev) == mbox_reorder_lists[i].vendor && pci_get_device(dev) == mbox_reorder_lists[i].device) { device_printf(sc->bge_dev, "enabling MBOX workaround for %s\n", mbox_reorder_lists[i].desc); return (1); } } if (device_get_devclass(bus) != pci) break; } return (0); } static void bge_devinfo(struct bge_softc *sc) { uint32_t cfg, clk; device_printf(sc->bge_dev, "CHIP ID 0x%08x; ASIC REV 0x%02x; CHIP REV 0x%02x; ", sc->bge_chipid, sc->bge_asicrev, sc->bge_chiprev); if (sc->bge_flags & BGE_FLAG_PCIE) printf("PCI-E\n"); else if (sc->bge_flags & BGE_FLAG_PCIX) { printf("PCI-X "); cfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK; if (cfg == BGE_MISCCFG_BOARD_ID_5704CIOBE) clk = 133; else { clk = CSR_READ_4(sc, BGE_PCI_CLKCTL) & 0x1F; switch (clk) { case 0: clk = 33; break; case 2: clk = 50; break; case 4: clk = 66; break; case 6: clk = 100; break; case 7: clk = 133; break; } } printf("%u MHz\n", clk); } else { if (sc->bge_pcixcap != 0) printf("PCI on PCI-X "); else printf("PCI "); cfg = pci_read_config(sc->bge_dev, BGE_PCI_PCISTATE, 4); if (cfg & BGE_PCISTATE_PCI_BUSSPEED) clk = 66; else clk = 33; if (cfg & BGE_PCISTATE_32BIT_BUS) printf("%u MHz; 32bit\n", clk); else printf("%u MHz; 64bit\n", clk); } } static int bge_attach(device_t dev) { struct ifnet *ifp; struct bge_softc *sc; uint32_t hwcfg = 0, misccfg; u_char eaddr[ETHER_ADDR_LEN]; int capmask, error, f, msicount, phy_addr, reg, rid, trys; sc = device_get_softc(dev); sc->bge_dev = dev; TASK_INIT(&sc->bge_intr_task, 0, bge_intr_task, sc); /* * Map control/status registers. */ pci_enable_busmaster(dev); rid = PCIR_BAR(0); sc->bge_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE | PCI_RF_DENSE); if (sc->bge_res == NULL) { device_printf (sc->bge_dev, "couldn't map memory\n"); error = ENXIO; goto fail; } /* Save various chip information. */ sc->bge_chipid = pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >> BGE_PCIMISCCTL_ASICREV_SHIFT; if (BGE_ASICREV(sc->bge_chipid) == BGE_ASICREV_USE_PRODID_REG) { /* * Find the ASCI revision. Different chips use different * registers. */ switch (pci_get_device(dev)) { case BCOM_DEVICEID_BCM5717: case BCOM_DEVICEID_BCM5718: case BCOM_DEVICEID_BCM5719: case BCOM_DEVICEID_BCM5720: sc->bge_chipid = pci_read_config(dev, BGE_PCI_GEN2_PRODID_ASICREV, 4); break; case BCOM_DEVICEID_BCM57761: case BCOM_DEVICEID_BCM57765: case BCOM_DEVICEID_BCM57781: case BCOM_DEVICEID_BCM57785: case BCOM_DEVICEID_BCM57791: case BCOM_DEVICEID_BCM57795: sc->bge_chipid = pci_read_config(dev, BGE_PCI_GEN15_PRODID_ASICREV, 4); break; default: sc->bge_chipid = pci_read_config(dev, BGE_PCI_PRODID_ASICREV, 4); } } sc->bge_asicrev = BGE_ASICREV(sc->bge_chipid); sc->bge_chiprev = BGE_CHIPREV(sc->bge_chipid); /* Set default PHY address. */ phy_addr = 1; /* * PHY address mapping for various devices. * * | F0 Cu | F0 Sr | F1 Cu | F1 Sr | * ---------+-------+-------+-------+-------+ * BCM57XX | 1 | X | X | X | * BCM5704 | 1 | X | 1 | X | * BCM5717 | 1 | 8 | 2 | 9 | * BCM5719 | 1 | 8 | 2 | 9 | * BCM5720 | 1 | 8 | 2 | 9 | * * Other addresses may respond but they are not * IEEE compliant PHYs and should be ignored. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5717 || sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) { f = pci_get_function(dev); if (sc->bge_chipid == BGE_CHIPID_BCM5717_A0) { if (CSR_READ_4(sc, BGE_SGDIG_STS) & BGE_SGDIGSTS_IS_SERDES) phy_addr = f + 8; else phy_addr = f + 1; } else { if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) & BGE_CPMU_PHY_STRAP_IS_SERDES) phy_addr = f + 8; else phy_addr = f + 1; } } /* * Don't enable Ethernet@WireSpeed for the 5700, 5906, or the * 5705 A0 and A1 chips. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5700 || (sc->bge_asicrev == BGE_ASICREV_BCM5705 && (sc->bge_chipid != BGE_CHIPID_BCM5705_A0 && sc->bge_chipid != BGE_CHIPID_BCM5705_A1)) || sc->bge_asicrev == BGE_ASICREV_BCM5906) sc->bge_phy_flags |= BGE_PHY_NO_WIRESPEED; if (bge_has_eaddr(sc)) sc->bge_flags |= BGE_FLAG_EADDR; /* Save chipset family. */ switch (sc->bge_asicrev) { case BGE_ASICREV_BCM5717: case BGE_ASICREV_BCM5719: case BGE_ASICREV_BCM5720: case BGE_ASICREV_BCM57765: sc->bge_flags |= BGE_FLAG_5717_PLUS | BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS | BGE_FLAG_5705_PLUS | BGE_FLAG_JUMBO | BGE_FLAG_JUMBO_FRAME; if (sc->bge_asicrev == BGE_ASICREV_BCM5719 && sc->bge_chipid == BGE_CHIPID_BCM5719_A0) { /* Jumbo frame on BCM5719 A0 does not work. */ sc->bge_flags &= ~BGE_FLAG_JUMBO; } break; case BGE_ASICREV_BCM5755: case BGE_ASICREV_BCM5761: case BGE_ASICREV_BCM5784: case BGE_ASICREV_BCM5785: case BGE_ASICREV_BCM5787: case BGE_ASICREV_BCM57780: sc->bge_flags |= BGE_FLAG_5755_PLUS | BGE_FLAG_575X_PLUS | BGE_FLAG_5705_PLUS; break; case BGE_ASICREV_BCM5700: case BGE_ASICREV_BCM5701: case BGE_ASICREV_BCM5703: case BGE_ASICREV_BCM5704: sc->bge_flags |= BGE_FLAG_5700_FAMILY | BGE_FLAG_JUMBO; break; case BGE_ASICREV_BCM5714_A0: case BGE_ASICREV_BCM5780: case BGE_ASICREV_BCM5714: sc->bge_flags |= BGE_FLAG_5714_FAMILY | BGE_FLAG_JUMBO_STD; /* FALLTHROUGH */ case BGE_ASICREV_BCM5750: case BGE_ASICREV_BCM5752: case BGE_ASICREV_BCM5906: sc->bge_flags |= BGE_FLAG_575X_PLUS; /* FALLTHROUGH */ case BGE_ASICREV_BCM5705: sc->bge_flags |= BGE_FLAG_5705_PLUS; break; } /* Add SYSCTLs, requires the chipset family to be set. */ bge_add_sysctls(sc); /* Set various PHY bug flags. */ if (sc->bge_chipid == BGE_CHIPID_BCM5701_A0 || sc->bge_chipid == BGE_CHIPID_BCM5701_B0) sc->bge_phy_flags |= BGE_PHY_CRC_BUG; if (sc->bge_chiprev == BGE_CHIPREV_5703_AX || sc->bge_chiprev == BGE_CHIPREV_5704_AX) sc->bge_phy_flags |= BGE_PHY_ADC_BUG; if (sc->bge_chipid == BGE_CHIPID_BCM5704_A0) sc->bge_phy_flags |= BGE_PHY_5704_A0_BUG; if (pci_get_subvendor(dev) == DELL_VENDORID) sc->bge_phy_flags |= BGE_PHY_NO_3LED; if ((BGE_IS_5705_PLUS(sc)) && sc->bge_asicrev != BGE_ASICREV_BCM5906 && sc->bge_asicrev != BGE_ASICREV_BCM5717 && sc->bge_asicrev != BGE_ASICREV_BCM5719 && sc->bge_asicrev != BGE_ASICREV_BCM5720 && sc->bge_asicrev != BGE_ASICREV_BCM5785 && sc->bge_asicrev != BGE_ASICREV_BCM57765 && sc->bge_asicrev != BGE_ASICREV_BCM57780) { if (sc->bge_asicrev == BGE_ASICREV_BCM5755 || sc->bge_asicrev == BGE_ASICREV_BCM5761 || sc->bge_asicrev == BGE_ASICREV_BCM5784 || sc->bge_asicrev == BGE_ASICREV_BCM5787) { if (pci_get_device(dev) != BCOM_DEVICEID_BCM5722 && pci_get_device(dev) != BCOM_DEVICEID_BCM5756) sc->bge_phy_flags |= BGE_PHY_JITTER_BUG; if (pci_get_device(dev) == BCOM_DEVICEID_BCM5755M) sc->bge_phy_flags |= BGE_PHY_ADJUST_TRIM; } else sc->bge_phy_flags |= BGE_PHY_BER_BUG; } /* Identify the chips that use an CPMU. */ if (BGE_IS_5717_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5784 || sc->bge_asicrev == BGE_ASICREV_BCM5761 || sc->bge_asicrev == BGE_ASICREV_BCM5785 || sc->bge_asicrev == BGE_ASICREV_BCM57780) sc->bge_flags |= BGE_FLAG_CPMU_PRESENT; if ((sc->bge_flags & BGE_FLAG_CPMU_PRESENT) != 0) sc->bge_mi_mode = BGE_MIMODE_500KHZ_CONST; else sc->bge_mi_mode = BGE_MIMODE_BASE; /* Enable auto polling for BCM570[0-5]. */ if (BGE_IS_5700_FAMILY(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5705) sc->bge_mi_mode |= BGE_MIMODE_AUTOPOLL; /* * All Broadcom controllers have 4GB boundary DMA bug. * Whenever an address crosses a multiple of the 4GB boundary * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA * state machine will lockup and cause the device to hang. */ sc->bge_flags |= BGE_FLAG_4G_BNDRY_BUG; /* BCM5755 or higher and BCM5906 have short DMA bug. */ if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906) sc->bge_flags |= BGE_FLAG_SHORT_DMA_BUG; /* * BCM5719 cannot handle DMA requests for DMA segments that * have larger than 4KB in size. However the maximum DMA * segment size created in DMA tag is 4KB for TSO, so we * wouldn't encounter the issue here. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5719) sc->bge_flags |= BGE_FLAG_4K_RDMA_BUG; misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK; if (sc->bge_asicrev == BGE_ASICREV_BCM5705) { if (misccfg == BGE_MISCCFG_BOARD_ID_5788 || misccfg == BGE_MISCCFG_BOARD_ID_5788M) sc->bge_flags |= BGE_FLAG_5788; } capmask = BMSR_DEFCAPMASK; if ((sc->bge_asicrev == BGE_ASICREV_BCM5703 && (misccfg == 0x4000 || misccfg == 0x8000)) || (sc->bge_asicrev == BGE_ASICREV_BCM5705 && pci_get_vendor(dev) == BCOM_VENDORID && (pci_get_device(dev) == BCOM_DEVICEID_BCM5901 || pci_get_device(dev) == BCOM_DEVICEID_BCM5901A2 || pci_get_device(dev) == BCOM_DEVICEID_BCM5705F)) || (pci_get_vendor(dev) == BCOM_VENDORID && (pci_get_device(dev) == BCOM_DEVICEID_BCM5751F || pci_get_device(dev) == BCOM_DEVICEID_BCM5753F || pci_get_device(dev) == BCOM_DEVICEID_BCM5787F)) || pci_get_device(dev) == BCOM_DEVICEID_BCM57790 || sc->bge_asicrev == BGE_ASICREV_BCM5906) { /* These chips are 10/100 only. */ capmask &= ~BMSR_EXTSTAT; } /* * Some controllers seem to require a special firmware to use * TSO. But the firmware is not available to FreeBSD and Linux * claims that the TSO performed by the firmware is slower than * hardware based TSO. Moreover the firmware based TSO has one * known bug which can't handle TSO if ethernet header + IP/TCP * header is greater than 80 bytes. The workaround for the TSO * bug exist but it seems it's too expensive than not using * TSO at all. Some hardwares also have the TSO bug so limit * the TSO to the controllers that are not affected TSO issues * (e.g. 5755 or higher). */ if (BGE_IS_5717_PLUS(sc)) { /* BCM5717 requires different TSO configuration. */ sc->bge_flags |= BGE_FLAG_TSO3; if (sc->bge_asicrev == BGE_ASICREV_BCM5719 && sc->bge_chipid == BGE_CHIPID_BCM5719_A0) { /* TSO on BCM5719 A0 does not work. */ sc->bge_flags &= ~BGE_FLAG_TSO3; } } else if (BGE_IS_5755_PLUS(sc)) { /* * BCM5754 and BCM5787 shares the same ASIC id so * explicit device id check is required. * Due to unknown reason TSO does not work on BCM5755M. */ if (pci_get_device(dev) != BCOM_DEVICEID_BCM5754 && pci_get_device(dev) != BCOM_DEVICEID_BCM5754M && pci_get_device(dev) != BCOM_DEVICEID_BCM5755M) sc->bge_flags |= BGE_FLAG_TSO; } /* * Check if this is a PCI-X or PCI Express device. */ if (pci_find_extcap(dev, PCIY_EXPRESS, ®) == 0) { /* * Found a PCI Express capabilities register, this * must be a PCI Express device. */ sc->bge_flags |= BGE_FLAG_PCIE; sc->bge_expcap = reg; if (sc->bge_asicrev == BGE_ASICREV_BCM5719 || sc->bge_asicrev == BGE_ASICREV_BCM5720) pci_set_max_read_req(dev, 2048); else if (pci_get_max_read_req(dev) != 4096) pci_set_max_read_req(dev, 4096); } else { /* * Check if the device is in PCI-X Mode. * (This bit is not valid on PCI Express controllers.) */ if (pci_find_extcap(dev, PCIY_PCIX, ®) == 0) sc->bge_pcixcap = reg; if ((pci_read_config(dev, BGE_PCI_PCISTATE, 4) & BGE_PCISTATE_PCI_BUSMODE) == 0) sc->bge_flags |= BGE_FLAG_PCIX; } /* * The 40bit DMA bug applies to the 5714/5715 controllers and is * not actually a MAC controller bug but an issue with the embedded * PCIe to PCI-X bridge in the device. Use 40bit DMA workaround. */ if (BGE_IS_5714_FAMILY(sc) && (sc->bge_flags & BGE_FLAG_PCIX)) sc->bge_flags |= BGE_FLAG_40BIT_BUG; /* * Some PCI-X bridges are known to trigger write reordering to * the mailbox registers. Typical phenomena is watchdog timeouts * caused by out-of-order TX completions. Enable workaround for * PCI-X devices that live behind these bridges. * Note, PCI-X controllers can run in PCI mode so we can't use * BGE_FLAG_PCIX flag to detect PCI-X controllers. */ if (sc->bge_pcixcap != 0 && bge_mbox_reorder(sc) != 0) sc->bge_flags |= BGE_FLAG_MBOX_REORDER; /* * Allocate the interrupt, using MSI if possible. These devices * support 8 MSI messages, but only the first one is used in * normal operation. */ rid = 0; if (pci_find_extcap(sc->bge_dev, PCIY_MSI, ®) == 0) { sc->bge_msicap = reg; if (bge_can_use_msi(sc)) { msicount = pci_msi_count(dev); if (msicount > 1) msicount = 1; } else msicount = 0; if (msicount == 1 && pci_alloc_msi(dev, &msicount) == 0) { rid = 1; sc->bge_flags |= BGE_FLAG_MSI; } } /* * All controllers except BCM5700 supports tagged status but * we use tagged status only for MSI case on BCM5717. Otherwise * MSI on BCM5717 does not work. */ #ifndef DEVICE_POLLING if (sc->bge_flags & BGE_FLAG_MSI && BGE_IS_5717_PLUS(sc)) sc->bge_flags |= BGE_FLAG_TAGGED_STATUS; #endif sc->bge_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE); if (sc->bge_irq == NULL) { device_printf(sc->bge_dev, "couldn't map interrupt\n"); error = ENXIO; goto fail; } bge_devinfo(sc); BGE_LOCK_INIT(sc, device_get_nameunit(dev)); /* Try to reset the chip. */ if (bge_reset(sc)) { device_printf(sc->bge_dev, "chip reset failed\n"); error = ENXIO; goto fail; } sc->bge_asf_mode = 0; if (bge_allow_asf && (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) == BGE_SRAM_DATA_SIG_MAGIC)) { if (bge_readmem_ind(sc, BGE_SRAM_DATA_CFG) & BGE_HWCFG_ASF) { sc->bge_asf_mode |= ASF_ENABLE; sc->bge_asf_mode |= ASF_STACKUP; if (BGE_IS_575X_PLUS(sc)) sc->bge_asf_mode |= ASF_NEW_HANDSHAKE; } } /* Try to reset the chip again the nice way. */ bge_stop_fw(sc); bge_sig_pre_reset(sc, BGE_RESET_STOP); if (bge_reset(sc)) { device_printf(sc->bge_dev, "chip reset failed\n"); error = ENXIO; goto fail; } bge_sig_legacy(sc, BGE_RESET_STOP); bge_sig_post_reset(sc, BGE_RESET_STOP); if (bge_chipinit(sc)) { device_printf(sc->bge_dev, "chip initialization failed\n"); error = ENXIO; goto fail; } error = bge_get_eaddr(sc, eaddr); if (error) { device_printf(sc->bge_dev, "failed to read station address\n"); error = ENXIO; goto fail; } /* 5705 limits RX return ring to 512 entries. */ if (BGE_IS_5717_PLUS(sc)) sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT; else if (BGE_IS_5705_PLUS(sc)) sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT_5705; else sc->bge_return_ring_cnt = BGE_RETURN_RING_CNT; if (bge_dma_alloc(sc)) { device_printf(sc->bge_dev, "failed to allocate DMA resources\n"); error = ENXIO; goto fail; } /* Set default tuneable values. */ sc->bge_stat_ticks = BGE_TICKS_PER_SEC; sc->bge_rx_coal_ticks = 150; sc->bge_tx_coal_ticks = 150; sc->bge_rx_max_coal_bds = 10; sc->bge_tx_max_coal_bds = 10; /* Initialize checksum features to use. */ sc->bge_csum_features = BGE_CSUM_FEATURES; if (sc->bge_forced_udpcsum != 0) sc->bge_csum_features |= CSUM_UDP; /* Set up ifnet structure */ ifp = sc->bge_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(sc->bge_dev, "failed to if_alloc()\n"); error = ENXIO; goto fail; } ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = bge_ioctl; ifp->if_start = bge_start; ifp->if_init = bge_init; ifp->if_snd.ifq_drv_maxlen = BGE_TX_RING_CNT - 1; IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen); IFQ_SET_READY(&ifp->if_snd); ifp->if_hwassist = sc->bge_csum_features; ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU; if ((sc->bge_flags & (BGE_FLAG_TSO | BGE_FLAG_TSO3)) != 0) { ifp->if_hwassist |= CSUM_TSO; ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_VLAN_HWTSO; } #ifdef IFCAP_VLAN_HWCSUM ifp->if_capabilities |= IFCAP_VLAN_HWCSUM; #endif ifp->if_capenable = ifp->if_capabilities; #ifdef DEVICE_POLLING ifp->if_capabilities |= IFCAP_POLLING; #endif /* * 5700 B0 chips do not support checksumming correctly due * to hardware bugs. */ if (sc->bge_chipid == BGE_CHIPID_BCM5700_B0) { ifp->if_capabilities &= ~IFCAP_HWCSUM; ifp->if_capenable &= ~IFCAP_HWCSUM; ifp->if_hwassist = 0; } /* * Figure out what sort of media we have by checking the * hardware config word in the first 32k of NIC internal memory, * or fall back to examining the EEPROM if necessary. * Note: on some BCM5700 cards, this value appears to be unset. * If that's the case, we have to rely on identifying the NIC * by its PCI subsystem ID, as we do below for the SysKonnect * SK-9D41. */ if (bge_readmem_ind(sc, BGE_SRAM_DATA_SIG) == BGE_SRAM_DATA_SIG_MAGIC) hwcfg = bge_readmem_ind(sc, BGE_SRAM_DATA_CFG); else if ((sc->bge_flags & BGE_FLAG_EADDR) && (sc->bge_asicrev != BGE_ASICREV_BCM5906)) { if (bge_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET, sizeof(hwcfg))) { device_printf(sc->bge_dev, "failed to read EEPROM\n"); error = ENXIO; goto fail; } hwcfg = ntohl(hwcfg); } /* The SysKonnect SK-9D41 is a 1000baseSX card. */ if ((pci_read_config(dev, BGE_PCI_SUBSYS, 4) >> 16) == SK_SUBSYSID_9D41 || (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER) { if (BGE_IS_5714_FAMILY(sc)) sc->bge_flags |= BGE_FLAG_MII_SERDES; else sc->bge_flags |= BGE_FLAG_TBI; } if (sc->bge_flags & BGE_FLAG_TBI) { ifmedia_init(&sc->bge_ifmedia, IFM_IMASK, bge_ifmedia_upd, bge_ifmedia_sts); ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX, 0, NULL); ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_1000_SX | IFM_FDX, 0, NULL); ifmedia_add(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&sc->bge_ifmedia, IFM_ETHER | IFM_AUTO); sc->bge_ifmedia.ifm_media = sc->bge_ifmedia.ifm_cur->ifm_media; } else { /* * Do transceiver setup and tell the firmware the * driver is down so we can try to get access the * probe if ASF is running. Retry a couple of times * if we get a conflict with the ASF firmware accessing * the PHY. */ trys = 0; BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); again: bge_asf_driver_up(sc); error = mii_attach(dev, &sc->bge_miibus, ifp, bge_ifmedia_upd, bge_ifmedia_sts, capmask, phy_addr, MII_OFFSET_ANY, MIIF_DOPAUSE | MIIF_FORCEPAUSE); if (error != 0) { if (trys++ < 4) { device_printf(sc->bge_dev, "Try again\n"); bge_miibus_writereg(sc->bge_dev, 1, MII_BMCR, BMCR_RESET); goto again; } device_printf(sc->bge_dev, "attaching PHYs failed\n"); goto fail; } /* * Now tell the firmware we are going up after probing the PHY */ if (sc->bge_asf_mode & ASF_STACKUP) BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); } /* * When using the BCM5701 in PCI-X mode, data corruption has * been observed in the first few bytes of some received packets. * Aligning the packet buffer in memory eliminates the corruption. * Unfortunately, this misaligns the packet payloads. On platforms * which do not support unaligned accesses, we will realign the * payloads by copying the received packets. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5701 && sc->bge_flags & BGE_FLAG_PCIX) sc->bge_flags |= BGE_FLAG_RX_ALIGNBUG; /* * Call MI attach routine. */ ether_ifattach(ifp, eaddr); callout_init_mtx(&sc->bge_stat_ch, &sc->bge_mtx, 0); /* Tell upper layer we support long frames. */ ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); /* * Hookup IRQ last. */ if (BGE_IS_5755_PLUS(sc) && sc->bge_flags & BGE_FLAG_MSI) { /* Take advantage of single-shot MSI. */ CSR_WRITE_4(sc, BGE_MSI_MODE, CSR_READ_4(sc, BGE_MSI_MODE) & ~BGE_MSIMODE_ONE_SHOT_DISABLE); sc->bge_tq = taskqueue_create_fast("bge_taskq", M_WAITOK, taskqueue_thread_enqueue, &sc->bge_tq); if (sc->bge_tq == NULL) { device_printf(dev, "could not create taskqueue.\n"); ether_ifdetach(ifp); error = ENXIO; goto fail; } taskqueue_start_threads(&sc->bge_tq, 1, PI_NET, "%s taskq", device_get_nameunit(sc->bge_dev)); error = bus_setup_intr(dev, sc->bge_irq, INTR_TYPE_NET | INTR_MPSAFE, bge_msi_intr, NULL, sc, &sc->bge_intrhand); if (error) ether_ifdetach(ifp); } else error = bus_setup_intr(dev, sc->bge_irq, INTR_TYPE_NET | INTR_MPSAFE, NULL, bge_intr, sc, &sc->bge_intrhand); if (error) { bge_detach(dev); device_printf(sc->bge_dev, "couldn't set up irq\n"); } return (0); fail: bge_release_resources(sc); return (error); } static int bge_detach(device_t dev) { struct bge_softc *sc; struct ifnet *ifp; sc = device_get_softc(dev); ifp = sc->bge_ifp; #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) ether_poll_deregister(ifp); #endif BGE_LOCK(sc); bge_stop(sc); bge_reset(sc); BGE_UNLOCK(sc); callout_drain(&sc->bge_stat_ch); if (sc->bge_tq) taskqueue_drain(sc->bge_tq, &sc->bge_intr_task); ether_ifdetach(ifp); if (sc->bge_flags & BGE_FLAG_TBI) { ifmedia_removeall(&sc->bge_ifmedia); } else { bus_generic_detach(dev); device_delete_child(dev, sc->bge_miibus); } bge_release_resources(sc); return (0); } static void bge_release_resources(struct bge_softc *sc) { device_t dev; dev = sc->bge_dev; if (sc->bge_tq != NULL) taskqueue_free(sc->bge_tq); if (sc->bge_intrhand != NULL) bus_teardown_intr(dev, sc->bge_irq, sc->bge_intrhand); if (sc->bge_irq != NULL) bus_release_resource(dev, SYS_RES_IRQ, sc->bge_flags & BGE_FLAG_MSI ? 1 : 0, sc->bge_irq); if (sc->bge_flags & BGE_FLAG_MSI) pci_release_msi(dev); if (sc->bge_res != NULL) bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(0), sc->bge_res); if (sc->bge_ifp != NULL) if_free(sc->bge_ifp); bge_dma_free(sc); if (mtx_initialized(&sc->bge_mtx)) /* XXX */ BGE_LOCK_DESTROY(sc); } static int bge_reset(struct bge_softc *sc) { device_t dev; uint32_t cachesize, command, pcistate, reset, val; void (*write_op)(struct bge_softc *, int, int); uint16_t devctl; int i; dev = sc->bge_dev; if (BGE_IS_575X_PLUS(sc) && !BGE_IS_5714_FAMILY(sc) && (sc->bge_asicrev != BGE_ASICREV_BCM5906)) { if (sc->bge_flags & BGE_FLAG_PCIE) write_op = bge_writemem_direct; else write_op = bge_writemem_ind; } else write_op = bge_writereg_ind; /* Save some important PCI state. */ cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4); command = pci_read_config(dev, BGE_PCI_CMD, 4); pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4); pci_write_config(dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR | BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4); /* Disable fastboot on controllers that support it. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5752 || BGE_IS_5755_PLUS(sc)) { if (bootverbose) device_printf(dev, "Disabling fastboot\n"); CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0); } /* * Write the magic number to SRAM at offset 0xB50. * When firmware finishes its initialization it will * write ~BGE_SRAM_FW_MB_MAGIC to the same location. */ bge_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC); reset = BGE_MISCCFG_RESET_CORE_CLOCKS | BGE_32BITTIME_66MHZ; /* XXX: Broadcom Linux driver. */ if (sc->bge_flags & BGE_FLAG_PCIE) { if (CSR_READ_4(sc, 0x7E2C) == 0x60) /* PCIE 1.0 */ CSR_WRITE_4(sc, 0x7E2C, 0x20); if (sc->bge_chipid != BGE_CHIPID_BCM5750_A0) { /* Prevent PCIE link training during global reset */ CSR_WRITE_4(sc, BGE_MISC_CFG, 1 << 29); reset |= 1 << 29; } } /* * Set GPHY Power Down Override to leave GPHY * powered up in D0 uninitialized. */ if (BGE_IS_5705_PLUS(sc) && (sc->bge_flags & BGE_FLAG_CPMU_PRESENT) == 0) reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE; /* Issue global reset */ write_op(sc, BGE_MISC_CFG, reset); if (sc->bge_asicrev == BGE_ASICREV_BCM5906) { val = CSR_READ_4(sc, BGE_VCPU_STATUS); CSR_WRITE_4(sc, BGE_VCPU_STATUS, val | BGE_VCPU_STATUS_DRV_RESET); val = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL); CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL, val & ~BGE_VCPU_EXT_CTRL_HALT_CPU); } DELAY(1000); /* XXX: Broadcom Linux driver. */ if (sc->bge_flags & BGE_FLAG_PCIE) { if (sc->bge_chipid == BGE_CHIPID_BCM5750_A0) { DELAY(500000); /* wait for link training to complete */ val = pci_read_config(dev, 0xC4, 4); pci_write_config(dev, 0xC4, val | (1 << 15), 4); } devctl = pci_read_config(dev, - sc->bge_expcap + PCIR_EXPRESS_DEVICE_CTL, 2); + sc->bge_expcap + PCIER_DEVICE_CTL, 2); /* Clear enable no snoop and disable relaxed ordering. */ - devctl &= ~(PCIM_EXP_CTL_RELAXED_ORD_ENABLE | - PCIM_EXP_CTL_NOSNOOP_ENABLE); + devctl &= ~(PCIEM_CTL_RELAXED_ORD_ENABLE | + PCIEM_CTL_NOSNOOP_ENABLE); /* Set PCIE max payload size to 128. */ - devctl &= ~PCIM_EXP_CTL_MAX_PAYLOAD; - pci_write_config(dev, sc->bge_expcap + PCIR_EXPRESS_DEVICE_CTL, + devctl &= ~PCIEM_CTL_MAX_PAYLOAD; + pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_CTL, devctl, 2); /* Clear error status. */ - pci_write_config(dev, sc->bge_expcap + PCIR_EXPRESS_DEVICE_STA, - PCIM_EXP_STA_CORRECTABLE_ERROR | - PCIM_EXP_STA_NON_FATAL_ERROR | PCIM_EXP_STA_FATAL_ERROR | - PCIM_EXP_STA_UNSUPPORTED_REQ, 2); + pci_write_config(dev, sc->bge_expcap + PCIER_DEVICE_STA, + PCIEM_STA_CORRECTABLE_ERROR | + PCIEM_STA_NON_FATAL_ERROR | PCIEM_STA_FATAL_ERROR | + PCIEM_STA_UNSUPPORTED_REQ, 2); } /* Reset some of the PCI state that got zapped by reset. */ pci_write_config(dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_INDIRECT_ACCESS | BGE_PCIMISCCTL_MASK_PCI_INTR | BGE_HIF_SWAP_OPTIONS | BGE_PCIMISCCTL_PCISTATE_RW, 4); pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4); pci_write_config(dev, BGE_PCI_CMD, command, 4); write_op(sc, BGE_MISC_CFG, BGE_32BITTIME_66MHZ); /* * Disable PCI-X relaxed ordering to ensure status block update * comes first then packet buffer DMA. Otherwise driver may * read stale status block. */ if (sc->bge_flags & BGE_FLAG_PCIX) { devctl = pci_read_config(dev, sc->bge_pcixcap + PCIXR_COMMAND, 2); devctl &= ~PCIXM_COMMAND_ERO; if (sc->bge_asicrev == BGE_ASICREV_BCM5703) { devctl &= ~PCIXM_COMMAND_MAX_READ; devctl |= PCIXM_COMMAND_MAX_READ_2048; } else if (sc->bge_asicrev == BGE_ASICREV_BCM5704) { devctl &= ~(PCIXM_COMMAND_MAX_SPLITS | PCIXM_COMMAND_MAX_READ); devctl |= PCIXM_COMMAND_MAX_READ_2048; } pci_write_config(dev, sc->bge_pcixcap + PCIXR_COMMAND, devctl, 2); } /* Re-enable MSI, if necessary, and enable the memory arbiter. */ if (BGE_IS_5714_FAMILY(sc)) { /* This chip disables MSI on reset. */ if (sc->bge_flags & BGE_FLAG_MSI) { val = pci_read_config(dev, sc->bge_msicap + PCIR_MSI_CTRL, 2); pci_write_config(dev, sc->bge_msicap + PCIR_MSI_CTRL, val | PCIM_MSICTRL_MSI_ENABLE, 2); val = CSR_READ_4(sc, BGE_MSI_MODE); CSR_WRITE_4(sc, BGE_MSI_MODE, val | BGE_MSIMODE_ENABLE); } val = CSR_READ_4(sc, BGE_MARB_MODE); CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE | val); } else CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE); if (sc->bge_asicrev == BGE_ASICREV_BCM5906) { for (i = 0; i < BGE_TIMEOUT; i++) { val = CSR_READ_4(sc, BGE_VCPU_STATUS); if (val & BGE_VCPU_STATUS_INIT_DONE) break; DELAY(100); } if (i == BGE_TIMEOUT) { device_printf(dev, "reset timed out\n"); return (1); } } else { /* * Poll until we see the 1's complement of the magic number. * This indicates that the firmware initialization is complete. * We expect this to fail if no chip containing the Ethernet * address is fitted though. */ for (i = 0; i < BGE_TIMEOUT; i++) { DELAY(10); val = bge_readmem_ind(sc, BGE_SRAM_FW_MB); if (val == ~BGE_SRAM_FW_MB_MAGIC) break; } if ((sc->bge_flags & BGE_FLAG_EADDR) && i == BGE_TIMEOUT) device_printf(dev, "firmware handshake timed out, found 0x%08x\n", val); /* BCM57765 A0 needs additional time before accessing. */ if (sc->bge_chipid == BGE_CHIPID_BCM57765_A0) DELAY(10 * 1000); /* XXX */ } /* * XXX Wait for the value of the PCISTATE register to * return to its original pre-reset state. This is a * fairly good indicator of reset completion. If we don't * wait for the reset to fully complete, trying to read * from the device's non-PCI registers may yield garbage * results. */ for (i = 0; i < BGE_TIMEOUT; i++) { if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate) break; DELAY(10); } /* Fix up byte swapping. */ CSR_WRITE_4(sc, BGE_MODE_CTL, bge_dma_swap_options(sc)); /* Tell the ASF firmware we are up */ if (sc->bge_asf_mode & ASF_STACKUP) BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); CSR_WRITE_4(sc, BGE_MAC_MODE, 0); /* * The 5704 in TBI mode apparently needs some special * adjustment to insure the SERDES drive level is set * to 1.2V. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5704 && sc->bge_flags & BGE_FLAG_TBI) { val = CSR_READ_4(sc, BGE_SERDES_CFG); val = (val & ~0xFFF) | 0x880; CSR_WRITE_4(sc, BGE_SERDES_CFG, val); } /* XXX: Broadcom Linux driver. */ if (sc->bge_flags & BGE_FLAG_PCIE && !BGE_IS_5717_PLUS(sc) && sc->bge_chipid != BGE_CHIPID_BCM5750_A0 && sc->bge_asicrev != BGE_ASICREV_BCM5785) { /* Enable Data FIFO protection. */ val = CSR_READ_4(sc, 0x7C00); CSR_WRITE_4(sc, 0x7C00, val | (1 << 25)); } DELAY(10000); if (sc->bge_asicrev == BGE_ASICREV_BCM5720) BGE_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE, CPMU_CLCK_ORIDE_MAC_ORIDE_EN); return (0); } static __inline void bge_rxreuse_std(struct bge_softc *sc, int i) { struct bge_rx_bd *r; r = &sc->bge_ldata.bge_rx_std_ring[sc->bge_std]; r->bge_flags = BGE_RXBDFLAG_END; r->bge_len = sc->bge_cdata.bge_rx_std_seglen[i]; r->bge_idx = i; BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT); } static __inline void bge_rxreuse_jumbo(struct bge_softc *sc, int i) { struct bge_extrx_bd *r; r = &sc->bge_ldata.bge_rx_jumbo_ring[sc->bge_jumbo]; r->bge_flags = BGE_RXBDFLAG_JUMBO_RING | BGE_RXBDFLAG_END; r->bge_len0 = sc->bge_cdata.bge_rx_jumbo_seglen[i][0]; r->bge_len1 = sc->bge_cdata.bge_rx_jumbo_seglen[i][1]; r->bge_len2 = sc->bge_cdata.bge_rx_jumbo_seglen[i][2]; r->bge_len3 = sc->bge_cdata.bge_rx_jumbo_seglen[i][3]; r->bge_idx = i; BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT); } /* * Frame reception handling. This is called if there's a frame * on the receive return list. * * Note: we have to be able to handle two possibilities here: * 1) the frame is from the jumbo receive ring * 2) the frame is from the standard receive ring */ static void bge_rxeof(struct bge_softc *sc, uint16_t rx_prod, int holdlck) { struct ifnet *ifp; int stdcnt = 0, jumbocnt = 0; uint16_t rx_cons; rx_cons = sc->bge_rx_saved_considx; /* Nothing to do. */ if (rx_cons == rx_prod) return; ifp = sc->bge_ifp; bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag, sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_POSTREAD); bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_POSTWRITE); if (BGE_IS_JUMBO_CAPABLE(sc) && ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN)) bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_POSTWRITE); while (rx_cons != rx_prod) { struct bge_rx_bd *cur_rx; uint32_t rxidx; struct mbuf *m = NULL; uint16_t vlan_tag = 0; int have_tag = 0; #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) { if (sc->rxcycles <= 0) break; sc->rxcycles--; } #endif cur_rx = &sc->bge_ldata.bge_rx_return_ring[rx_cons]; rxidx = cur_rx->bge_idx; BGE_INC(rx_cons, sc->bge_return_ring_cnt); if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING && cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) { have_tag = 1; vlan_tag = cur_rx->bge_vlan_tag; } if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) { jumbocnt++; m = sc->bge_cdata.bge_rx_jumbo_chain[rxidx]; if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) { bge_rxreuse_jumbo(sc, rxidx); continue; } if (bge_newbuf_jumbo(sc, rxidx) != 0) { bge_rxreuse_jumbo(sc, rxidx); ifp->if_iqdrops++; continue; } BGE_INC(sc->bge_jumbo, BGE_JUMBO_RX_RING_CNT); } else { stdcnt++; m = sc->bge_cdata.bge_rx_std_chain[rxidx]; if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) { bge_rxreuse_std(sc, rxidx); continue; } if (bge_newbuf_std(sc, rxidx) != 0) { bge_rxreuse_std(sc, rxidx); ifp->if_iqdrops++; continue; } BGE_INC(sc->bge_std, BGE_STD_RX_RING_CNT); } ifp->if_ipackets++; #ifndef __NO_STRICT_ALIGNMENT /* * For architectures with strict alignment we must make sure * the payload is aligned. */ if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) { bcopy(m->m_data, m->m_data + ETHER_ALIGN, cur_rx->bge_len); m->m_data += ETHER_ALIGN; } #endif m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN; m->m_pkthdr.rcvif = ifp; if (ifp->if_capenable & IFCAP_RXCSUM) bge_rxcsum(sc, cur_rx, m); /* * If we received a packet with a vlan tag, * attach that information to the packet. */ if (have_tag) { m->m_pkthdr.ether_vtag = vlan_tag; m->m_flags |= M_VLANTAG; } if (holdlck != 0) { BGE_UNLOCK(sc); (*ifp->if_input)(ifp, m); BGE_LOCK(sc); } else (*ifp->if_input)(ifp, m); if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) return; } bus_dmamap_sync(sc->bge_cdata.bge_rx_return_ring_tag, sc->bge_cdata.bge_rx_return_ring_map, BUS_DMASYNC_PREREAD); if (stdcnt > 0) bus_dmamap_sync(sc->bge_cdata.bge_rx_std_ring_tag, sc->bge_cdata.bge_rx_std_ring_map, BUS_DMASYNC_PREWRITE); if (jumbocnt > 0) bus_dmamap_sync(sc->bge_cdata.bge_rx_jumbo_ring_tag, sc->bge_cdata.bge_rx_jumbo_ring_map, BUS_DMASYNC_PREWRITE); sc->bge_rx_saved_considx = rx_cons; bge_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bge_rx_saved_considx); if (stdcnt) bge_writembx(sc, BGE_MBX_RX_STD_PROD_LO, (sc->bge_std + BGE_STD_RX_RING_CNT - 1) % BGE_STD_RX_RING_CNT); if (jumbocnt) bge_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, (sc->bge_jumbo + BGE_JUMBO_RX_RING_CNT - 1) % BGE_JUMBO_RX_RING_CNT); #ifdef notyet /* * This register wraps very quickly under heavy packet drops. * If you need correct statistics, you can enable this check. */ if (BGE_IS_5705_PLUS(sc)) ifp->if_ierrors += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS); #endif } static void bge_rxcsum(struct bge_softc *sc, struct bge_rx_bd *cur_rx, struct mbuf *m) { if (BGE_IS_5717_PLUS(sc)) { if ((cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) { if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) { m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED; if ((cur_rx->bge_error_flag & BGE_RXERRFLAG_IP_CSUM_NOK) == 0) m->m_pkthdr.csum_flags |= CSUM_IP_VALID; } if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) { m->m_pkthdr.csum_data = cur_rx->bge_tcp_udp_csum; m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; } } } else { if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) { m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED; if ((cur_rx->bge_ip_csum ^ 0xFFFF) == 0) m->m_pkthdr.csum_flags |= CSUM_IP_VALID; } if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM && m->m_pkthdr.len >= ETHER_MIN_NOPAD) { m->m_pkthdr.csum_data = cur_rx->bge_tcp_udp_csum; m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; } } } static void bge_txeof(struct bge_softc *sc, uint16_t tx_cons) { struct bge_tx_bd *cur_tx; struct ifnet *ifp; BGE_LOCK_ASSERT(sc); /* Nothing to do. */ if (sc->bge_tx_saved_considx == tx_cons) return; ifp = sc->bge_ifp; bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag, sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_POSTWRITE); /* * Go through our tx ring and free mbufs for those * frames that have been sent. */ while (sc->bge_tx_saved_considx != tx_cons) { uint32_t idx; idx = sc->bge_tx_saved_considx; cur_tx = &sc->bge_ldata.bge_tx_ring[idx]; if (cur_tx->bge_flags & BGE_TXBDFLAG_END) ifp->if_opackets++; if (sc->bge_cdata.bge_tx_chain[idx] != NULL) { bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, sc->bge_cdata.bge_tx_dmamap[idx], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, sc->bge_cdata.bge_tx_dmamap[idx]); m_freem(sc->bge_cdata.bge_tx_chain[idx]); sc->bge_cdata.bge_tx_chain[idx] = NULL; } sc->bge_txcnt--; BGE_INC(sc->bge_tx_saved_considx, BGE_TX_RING_CNT); } ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; if (sc->bge_txcnt == 0) sc->bge_timer = 0; } #ifdef DEVICE_POLLING static void bge_poll(struct ifnet *ifp, enum poll_cmd cmd, int count) { struct bge_softc *sc = ifp->if_softc; uint16_t rx_prod, tx_cons; uint32_t statusword; BGE_LOCK(sc); if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { BGE_UNLOCK(sc); return; } bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx; tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx; statusword = sc->bge_ldata.bge_status_block->bge_status; sc->bge_ldata.bge_status_block->bge_status = 0; bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* Note link event. It will be processed by POLL_AND_CHECK_STATUS. */ if (statusword & BGE_STATFLAG_LINKSTATE_CHANGED) sc->bge_link_evt++; if (cmd == POLL_AND_CHECK_STATUS) if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_B2) || sc->bge_link_evt || (sc->bge_flags & BGE_FLAG_TBI)) bge_link_upd(sc); sc->rxcycles = count; bge_rxeof(sc, rx_prod, 1); if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { BGE_UNLOCK(sc); return; } bge_txeof(sc, tx_cons); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) bge_start_locked(ifp); BGE_UNLOCK(sc); } #endif /* DEVICE_POLLING */ static int bge_msi_intr(void *arg) { struct bge_softc *sc; sc = (struct bge_softc *)arg; /* * This interrupt is not shared and controller already * disabled further interrupt. */ taskqueue_enqueue(sc->bge_tq, &sc->bge_intr_task); return (FILTER_HANDLED); } static void bge_intr_task(void *arg, int pending) { struct bge_softc *sc; struct ifnet *ifp; uint32_t status, status_tag; uint16_t rx_prod, tx_cons; sc = (struct bge_softc *)arg; ifp = sc->bge_ifp; BGE_LOCK(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { BGE_UNLOCK(sc); return; } /* Get updated status block. */ bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* Save producer/consumer indexess. */ rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx; tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx; status = sc->bge_ldata.bge_status_block->bge_status; status_tag = sc->bge_ldata.bge_status_block->bge_status_tag << 24; sc->bge_ldata.bge_status_block->bge_status = 0; bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); if ((sc->bge_flags & BGE_FLAG_TAGGED_STATUS) == 0) status_tag = 0; if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) != 0) bge_link_upd(sc); /* Let controller work. */ bge_writembx(sc, BGE_MBX_IRQ0_LO, status_tag); if (ifp->if_drv_flags & IFF_DRV_RUNNING && sc->bge_rx_saved_considx != rx_prod) { /* Check RX return ring producer/consumer. */ BGE_UNLOCK(sc); bge_rxeof(sc, rx_prod, 0); BGE_LOCK(sc); } if (ifp->if_drv_flags & IFF_DRV_RUNNING) { /* Check TX ring producer/consumer. */ bge_txeof(sc, tx_cons); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) bge_start_locked(ifp); } BGE_UNLOCK(sc); } static void bge_intr(void *xsc) { struct bge_softc *sc; struct ifnet *ifp; uint32_t statusword; uint16_t rx_prod, tx_cons; sc = xsc; BGE_LOCK(sc); ifp = sc->bge_ifp; #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) { BGE_UNLOCK(sc); return; } #endif /* * Ack the interrupt by writing something to BGE_MBX_IRQ0_LO. Don't * disable interrupts by writing nonzero like we used to, since with * our current organization this just gives complications and * pessimizations for re-enabling interrupts. We used to have races * instead of the necessary complications. Disabling interrupts * would just reduce the chance of a status update while we are * running (by switching to the interrupt-mode coalescence * parameters), but this chance is already very low so it is more * efficient to get another interrupt than prevent it. * * We do the ack first to ensure another interrupt if there is a * status update after the ack. We don't check for the status * changing later because it is more efficient to get another * interrupt than prevent it, not quite as above (not checking is * a smaller optimization than not toggling the interrupt enable, * since checking doesn't involve PCI accesses and toggling require * the status check). So toggling would probably be a pessimization * even with MSI. It would only be needed for using a task queue. */ bge_writembx(sc, BGE_MBX_IRQ0_LO, 0); /* * Do the mandatory PCI flush as well as get the link status. */ statusword = CSR_READ_4(sc, BGE_MAC_STS) & BGE_MACSTAT_LINK_CHANGED; /* Make sure the descriptor ring indexes are coherent. */ bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); rx_prod = sc->bge_ldata.bge_status_block->bge_idx[0].bge_rx_prod_idx; tx_cons = sc->bge_ldata.bge_status_block->bge_idx[0].bge_tx_cons_idx; sc->bge_ldata.bge_status_block->bge_status = 0; bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); if ((sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_B2) || statusword || sc->bge_link_evt) bge_link_upd(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { /* Check RX return ring producer/consumer. */ bge_rxeof(sc, rx_prod, 1); } if (ifp->if_drv_flags & IFF_DRV_RUNNING) { /* Check TX ring producer/consumer. */ bge_txeof(sc, tx_cons); } if (ifp->if_drv_flags & IFF_DRV_RUNNING && !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) bge_start_locked(ifp); BGE_UNLOCK(sc); } static void bge_asf_driver_up(struct bge_softc *sc) { if (sc->bge_asf_mode & ASF_STACKUP) { /* Send ASF heartbeat aprox. every 2s */ if (sc->bge_asf_count) sc->bge_asf_count --; else { sc->bge_asf_count = 2; bge_writemem_ind(sc, BGE_SRAM_FW_CMD_MB, BGE_FW_CMD_DRV_ALIVE); bge_writemem_ind(sc, BGE_SRAM_FW_CMD_LEN_MB, 4); bge_writemem_ind(sc, BGE_SRAM_FW_CMD_DATA_MB, BGE_FW_HB_TIMEOUT_SEC); CSR_WRITE_4(sc, BGE_RX_CPU_EVENT, CSR_READ_4(sc, BGE_RX_CPU_EVENT) | BGE_RX_CPU_DRV_EVENT); } } } static void bge_tick(void *xsc) { struct bge_softc *sc = xsc; struct mii_data *mii = NULL; BGE_LOCK_ASSERT(sc); /* Synchronize with possible callout reset/stop. */ if (callout_pending(&sc->bge_stat_ch) || !callout_active(&sc->bge_stat_ch)) return; if (BGE_IS_5705_PLUS(sc)) bge_stats_update_regs(sc); else bge_stats_update(sc); if ((sc->bge_flags & BGE_FLAG_TBI) == 0) { mii = device_get_softc(sc->bge_miibus); /* * Do not touch PHY if we have link up. This could break * IPMI/ASF mode or produce extra input errors * (extra errors was reported for bcm5701 & bcm5704). */ if (!sc->bge_link) mii_tick(mii); } else { /* * Since in TBI mode auto-polling can't be used we should poll * link status manually. Here we register pending link event * and trigger interrupt. */ #ifdef DEVICE_POLLING /* In polling mode we poll link state in bge_poll(). */ if (!(sc->bge_ifp->if_capenable & IFCAP_POLLING)) #endif { sc->bge_link_evt++; if (sc->bge_asicrev == BGE_ASICREV_BCM5700 || sc->bge_flags & BGE_FLAG_5788) BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET); else BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW); } } bge_asf_driver_up(sc); bge_watchdog(sc); callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc); } static void bge_stats_update_regs(struct bge_softc *sc) { struct ifnet *ifp; struct bge_mac_stats *stats; ifp = sc->bge_ifp; stats = &sc->bge_mac_stats; stats->ifHCOutOctets += CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS); stats->etherStatsCollisions += CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS); stats->outXonSent += CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT); stats->outXoffSent += CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT); stats->dot3StatsInternalMacTransmitErrors += CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS); stats->dot3StatsSingleCollisionFrames += CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL); stats->dot3StatsMultipleCollisionFrames += CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL); stats->dot3StatsDeferredTransmissions += CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED); stats->dot3StatsExcessiveCollisions += CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL); stats->dot3StatsLateCollisions += CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL); stats->ifHCOutUcastPkts += CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST); stats->ifHCOutMulticastPkts += CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST); stats->ifHCOutBroadcastPkts += CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST); stats->ifHCInOctets += CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS); stats->etherStatsFragments += CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS); stats->ifHCInUcastPkts += CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST); stats->ifHCInMulticastPkts += CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST); stats->ifHCInBroadcastPkts += CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST); stats->dot3StatsFCSErrors += CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS); stats->dot3StatsAlignmentErrors += CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS); stats->xonPauseFramesReceived += CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD); stats->xoffPauseFramesReceived += CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD); stats->macControlFramesReceived += CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD); stats->xoffStateEntered += CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED); stats->dot3StatsFramesTooLong += CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG); stats->etherStatsJabbers += CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS); stats->etherStatsUndersizePkts += CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE); stats->FramesDroppedDueToFilters += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP); stats->DmaWriteQueueFull += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL); stats->DmaWriteHighPriQueueFull += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL); stats->NoMoreRxBDs += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS); /* * XXX * Unlike other controllers, BGE_RXLP_LOCSTAT_IFIN_DROPS * counter of BCM5717, BCM5718, BCM5719 A0 and BCM5720 A0 * includes number of unwanted multicast frames. This comes * from silicon bug and known workaround to get rough(not * exact) counter is to enable interrupt on MBUF low water * attention. This can be accomplished by setting * BGE_HCCMODE_ATTN bit of BGE_HCC_MODE, * BGE_BMANMODE_LOMBUF_ATTN bit of BGE_BMAN_MODE and * BGE_MODECTL_FLOWCTL_ATTN_INTR bit of BGE_MODE_CTL. * However that change would generate more interrupts and * there are still possibilities of losing multiple frames * during BGE_MODECTL_FLOWCTL_ATTN_INTR interrupt handling. * Given that the workaround still would not get correct * counter I don't think it's worth to implement it. So * ignore reading the counter on controllers that have the * silicon bug. */ if (sc->bge_asicrev != BGE_ASICREV_BCM5717 && sc->bge_chipid != BGE_CHIPID_BCM5719_A0 && sc->bge_chipid != BGE_CHIPID_BCM5720_A0) stats->InputDiscards += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS); stats->InputErrors += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS); stats->RecvThresholdHit += CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT); ifp->if_collisions = (u_long)stats->etherStatsCollisions; ifp->if_ierrors = (u_long)(stats->NoMoreRxBDs + stats->InputDiscards + stats->InputErrors); } static void bge_stats_clear_regs(struct bge_softc *sc) { CSR_READ_4(sc, BGE_TX_MAC_STATS_OCTETS); CSR_READ_4(sc, BGE_TX_MAC_STATS_COLLS); CSR_READ_4(sc, BGE_TX_MAC_STATS_XON_SENT); CSR_READ_4(sc, BGE_TX_MAC_STATS_XOFF_SENT); CSR_READ_4(sc, BGE_TX_MAC_STATS_ERRORS); CSR_READ_4(sc, BGE_TX_MAC_STATS_SINGLE_COLL); CSR_READ_4(sc, BGE_TX_MAC_STATS_MULTI_COLL); CSR_READ_4(sc, BGE_TX_MAC_STATS_DEFERRED); CSR_READ_4(sc, BGE_TX_MAC_STATS_EXCESS_COLL); CSR_READ_4(sc, BGE_TX_MAC_STATS_LATE_COLL); CSR_READ_4(sc, BGE_TX_MAC_STATS_UCAST); CSR_READ_4(sc, BGE_TX_MAC_STATS_MCAST); CSR_READ_4(sc, BGE_TX_MAC_STATS_BCAST); CSR_READ_4(sc, BGE_RX_MAC_STATS_OCTESTS); CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAGMENTS); CSR_READ_4(sc, BGE_RX_MAC_STATS_UCAST); CSR_READ_4(sc, BGE_RX_MAC_STATS_MCAST); CSR_READ_4(sc, BGE_RX_MAC_STATS_BCAST); CSR_READ_4(sc, BGE_RX_MAC_STATS_FCS_ERRORS); CSR_READ_4(sc, BGE_RX_MAC_STATS_ALGIN_ERRORS); CSR_READ_4(sc, BGE_RX_MAC_STATS_XON_RCVD); CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_RCVD); CSR_READ_4(sc, BGE_RX_MAC_STATS_CTRL_RCVD); CSR_READ_4(sc, BGE_RX_MAC_STATS_XOFF_ENTERED); CSR_READ_4(sc, BGE_RX_MAC_STATS_FRAME_TOO_LONG); CSR_READ_4(sc, BGE_RX_MAC_STATS_JABBERS); CSR_READ_4(sc, BGE_RX_MAC_STATS_UNDERSIZE); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_FILTDROP); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_WRQ_FULL); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_DMA_HPWRQ_FULL); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_DROPS); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_IFIN_ERRORS); CSR_READ_4(sc, BGE_RXLP_LOCSTAT_RXTHRESH_HIT); } static void bge_stats_update(struct bge_softc *sc) { struct ifnet *ifp; bus_size_t stats; uint32_t cnt; /* current register value */ ifp = sc->bge_ifp; stats = BGE_MEMWIN_START + BGE_STATS_BLOCK; #define READ_STAT(sc, stats, stat) \ CSR_READ_4(sc, stats + offsetof(struct bge_stats, stat)) cnt = READ_STAT(sc, stats, txstats.etherStatsCollisions.bge_addr_lo); ifp->if_collisions += (uint32_t)(cnt - sc->bge_tx_collisions); sc->bge_tx_collisions = cnt; cnt = READ_STAT(sc, stats, nicNoMoreRxBDs.bge_addr_lo); ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_nobds); sc->bge_rx_nobds = cnt; cnt = READ_STAT(sc, stats, ifInErrors.bge_addr_lo); ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_inerrs); sc->bge_rx_inerrs = cnt; cnt = READ_STAT(sc, stats, ifInDiscards.bge_addr_lo); ifp->if_ierrors += (uint32_t)(cnt - sc->bge_rx_discards); sc->bge_rx_discards = cnt; cnt = READ_STAT(sc, stats, txstats.ifOutDiscards.bge_addr_lo); ifp->if_oerrors += (uint32_t)(cnt - sc->bge_tx_discards); sc->bge_tx_discards = cnt; #undef READ_STAT } /* * Pad outbound frame to ETHER_MIN_NOPAD for an unusual reason. * The bge hardware will pad out Tx runts to ETHER_MIN_NOPAD, * but when such padded frames employ the bge IP/TCP checksum offload, * the hardware checksum assist gives incorrect results (possibly * from incorporating its own padding into the UDP/TCP checksum; who knows). * If we pad such runts with zeros, the onboard checksum comes out correct. */ static __inline int bge_cksum_pad(struct mbuf *m) { int padlen = ETHER_MIN_NOPAD - m->m_pkthdr.len; struct mbuf *last; /* If there's only the packet-header and we can pad there, use it. */ if (m->m_pkthdr.len == m->m_len && M_WRITABLE(m) && M_TRAILINGSPACE(m) >= padlen) { last = m; } else { /* * Walk packet chain to find last mbuf. We will either * pad there, or append a new mbuf and pad it. */ for (last = m; last->m_next != NULL; last = last->m_next); if (!(M_WRITABLE(last) && M_TRAILINGSPACE(last) >= padlen)) { /* Allocate new empty mbuf, pad it. Compact later. */ struct mbuf *n; MGET(n, M_DONTWAIT, MT_DATA); if (n == NULL) return (ENOBUFS); n->m_len = 0; last->m_next = n; last = n; } } /* Now zero the pad area, to avoid the bge cksum-assist bug. */ memset(mtod(last, caddr_t) + last->m_len, 0, padlen); last->m_len += padlen; m->m_pkthdr.len += padlen; return (0); } static struct mbuf * bge_check_short_dma(struct mbuf *m) { struct mbuf *n; int found; /* * If device receive two back-to-back send BDs with less than * or equal to 8 total bytes then the device may hang. The two * back-to-back send BDs must in the same frame for this failure * to occur. Scan mbuf chains and see whether two back-to-back * send BDs are there. If this is the case, allocate new mbuf * and copy the frame to workaround the silicon bug. */ for (n = m, found = 0; n != NULL; n = n->m_next) { if (n->m_len < 8) { found++; if (found > 1) break; continue; } found = 0; } if (found > 1) { n = m_defrag(m, M_DONTWAIT); if (n == NULL) m_freem(m); } else n = m; return (n); } static struct mbuf * bge_setup_tso(struct bge_softc *sc, struct mbuf *m, uint16_t *mss, uint16_t *flags) { struct ip *ip; struct tcphdr *tcp; struct mbuf *n; uint16_t hlen; uint32_t poff; if (M_WRITABLE(m) == 0) { /* Get a writable copy. */ n = m_dup(m, M_DONTWAIT); m_freem(m); if (n == NULL) return (NULL); m = n; } m = m_pullup(m, sizeof(struct ether_header) + sizeof(struct ip)); if (m == NULL) return (NULL); ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header)); poff = sizeof(struct ether_header) + (ip->ip_hl << 2); m = m_pullup(m, poff + sizeof(struct tcphdr)); if (m == NULL) return (NULL); tcp = (struct tcphdr *)(mtod(m, char *) + poff); m = m_pullup(m, poff + (tcp->th_off << 2)); if (m == NULL) return (NULL); /* * It seems controller doesn't modify IP length and TCP pseudo * checksum. These checksum computed by upper stack should be 0. */ *mss = m->m_pkthdr.tso_segsz; ip = (struct ip *)(mtod(m, char *) + sizeof(struct ether_header)); ip->ip_sum = 0; ip->ip_len = htons(*mss + (ip->ip_hl << 2) + (tcp->th_off << 2)); /* Clear pseudo checksum computed by TCP stack. */ tcp = (struct tcphdr *)(mtod(m, char *) + poff); tcp->th_sum = 0; /* * Broadcom controllers uses different descriptor format for * TSO depending on ASIC revision. Due to TSO-capable firmware * license issue and lower performance of firmware based TSO * we only support hardware based TSO. */ /* Calculate header length, incl. TCP/IP options, in 32 bit units. */ hlen = ((ip->ip_hl << 2) + (tcp->th_off << 2)) >> 2; if (sc->bge_flags & BGE_FLAG_TSO3) { /* * For BCM5717 and newer controllers, hardware based TSO * uses the 14 lower bits of the bge_mss field to store the * MSS and the upper 2 bits to store the lowest 2 bits of * the IP/TCP header length. The upper 6 bits of the header * length are stored in the bge_flags[14:10,4] field. Jumbo * frames are supported. */ *mss |= ((hlen & 0x3) << 14); *flags |= ((hlen & 0xF8) << 7) | ((hlen & 0x4) << 2); } else { /* * For BCM5755 and newer controllers, hardware based TSO uses * the lower 11 bits to store the MSS and the upper 5 bits to * store the IP/TCP header length. Jumbo frames are not * supported. */ *mss |= (hlen << 11); } return (m); } /* * Encapsulate an mbuf chain in the tx ring by coupling the mbuf data * pointers to descriptors. */ static int bge_encap(struct bge_softc *sc, struct mbuf **m_head, uint32_t *txidx) { bus_dma_segment_t segs[BGE_NSEG_NEW]; bus_dmamap_t map; struct bge_tx_bd *d; struct mbuf *m = *m_head; uint32_t idx = *txidx; uint16_t csum_flags, mss, vlan_tag; int nsegs, i, error; csum_flags = 0; mss = 0; vlan_tag = 0; if ((sc->bge_flags & BGE_FLAG_SHORT_DMA_BUG) != 0 && m->m_next != NULL) { *m_head = bge_check_short_dma(m); if (*m_head == NULL) return (ENOBUFS); m = *m_head; } if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) { *m_head = m = bge_setup_tso(sc, m, &mss, &csum_flags); if (*m_head == NULL) return (ENOBUFS); csum_flags |= BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA; } else if ((m->m_pkthdr.csum_flags & sc->bge_csum_features) != 0) { if (m->m_pkthdr.csum_flags & CSUM_IP) csum_flags |= BGE_TXBDFLAG_IP_CSUM; if (m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP)) { csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM; if (m->m_pkthdr.len < ETHER_MIN_NOPAD && (error = bge_cksum_pad(m)) != 0) { m_freem(m); *m_head = NULL; return (error); } } if (m->m_flags & M_LASTFRAG) csum_flags |= BGE_TXBDFLAG_IP_FRAG_END; else if (m->m_flags & M_FRAG) csum_flags |= BGE_TXBDFLAG_IP_FRAG; } if ((m->m_pkthdr.csum_flags & CSUM_TSO) == 0) { if (sc->bge_flags & BGE_FLAG_JUMBO_FRAME && m->m_pkthdr.len > ETHER_MAX_LEN) csum_flags |= BGE_TXBDFLAG_JUMBO_FRAME; if (sc->bge_forced_collapse > 0 && (sc->bge_flags & BGE_FLAG_PCIE) != 0 && m->m_next != NULL) { /* * Forcedly collapse mbuf chains to overcome hardware * limitation which only support a single outstanding * DMA read operation. */ if (sc->bge_forced_collapse == 1) m = m_defrag(m, M_DONTWAIT); else m = m_collapse(m, M_DONTWAIT, sc->bge_forced_collapse); if (m == NULL) m = *m_head; *m_head = m; } } map = sc->bge_cdata.bge_tx_dmamap[idx]; error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map, m, segs, &nsegs, BUS_DMA_NOWAIT); if (error == EFBIG) { m = m_collapse(m, M_DONTWAIT, BGE_NSEG_NEW); if (m == NULL) { m_freem(*m_head); *m_head = NULL; return (ENOBUFS); } *m_head = m; error = bus_dmamap_load_mbuf_sg(sc->bge_cdata.bge_tx_mtag, map, m, segs, &nsegs, BUS_DMA_NOWAIT); if (error) { m_freem(m); *m_head = NULL; return (error); } } else if (error != 0) return (error); /* Check if we have enough free send BDs. */ if (sc->bge_txcnt + nsegs >= BGE_TX_RING_CNT) { bus_dmamap_unload(sc->bge_cdata.bge_tx_mtag, map); return (ENOBUFS); } bus_dmamap_sync(sc->bge_cdata.bge_tx_mtag, map, BUS_DMASYNC_PREWRITE); if (m->m_flags & M_VLANTAG) { csum_flags |= BGE_TXBDFLAG_VLAN_TAG; vlan_tag = m->m_pkthdr.ether_vtag; } for (i = 0; ; i++) { d = &sc->bge_ldata.bge_tx_ring[idx]; d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr); d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr); d->bge_len = segs[i].ds_len; d->bge_flags = csum_flags; d->bge_vlan_tag = vlan_tag; d->bge_mss = mss; if (i == nsegs - 1) break; BGE_INC(idx, BGE_TX_RING_CNT); } /* Mark the last segment as end of packet... */ d->bge_flags |= BGE_TXBDFLAG_END; /* * Insure that the map for this transmission * is placed at the array index of the last descriptor * in this chain. */ sc->bge_cdata.bge_tx_dmamap[*txidx] = sc->bge_cdata.bge_tx_dmamap[idx]; sc->bge_cdata.bge_tx_dmamap[idx] = map; sc->bge_cdata.bge_tx_chain[idx] = m; sc->bge_txcnt += nsegs; BGE_INC(idx, BGE_TX_RING_CNT); *txidx = idx; return (0); } /* * Main transmit routine. To avoid having to do mbuf copies, we put pointers * to the mbuf data regions directly in the transmit descriptors. */ static void bge_start_locked(struct ifnet *ifp) { struct bge_softc *sc; struct mbuf *m_head; uint32_t prodidx; int count; sc = ifp->if_softc; BGE_LOCK_ASSERT(sc); if (!sc->bge_link || (ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return; prodidx = sc->bge_tx_prodidx; for (count = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd);) { if (sc->bge_txcnt > BGE_TX_RING_CNT - 16) { ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; /* * XXX * The code inside the if() block is never reached since we * must mark CSUM_IP_FRAGS in our if_hwassist to start getting * requests to checksum TCP/UDP in a fragmented packet. * * XXX * safety overkill. If this is a fragmented packet chain * with delayed TCP/UDP checksums, then only encapsulate * it if we have enough descriptors to handle the entire * chain at once. * (paranoia -- may not actually be needed) */ if (m_head->m_flags & M_FIRSTFRAG && m_head->m_pkthdr.csum_flags & (CSUM_DELAY_DATA)) { if ((BGE_TX_RING_CNT - sc->bge_txcnt) < m_head->m_pkthdr.csum_data + 16) { IFQ_DRV_PREPEND(&ifp->if_snd, m_head); ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } } /* * Pack the data into the transmit ring. If we * don't have room, set the OACTIVE flag and wait * for the NIC to drain the ring. */ if (bge_encap(sc, &m_head, &prodidx)) { if (m_head == NULL) break; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } ++count; /* * If there's a BPF listener, bounce a copy of this frame * to him. */ #ifdef ETHER_BPF_MTAP ETHER_BPF_MTAP(ifp, m_head); #else BPF_MTAP(ifp, m_head); #endif } if (count > 0) { bus_dmamap_sync(sc->bge_cdata.bge_tx_ring_tag, sc->bge_cdata.bge_tx_ring_map, BUS_DMASYNC_PREWRITE); /* Transmit. */ bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx); /* 5700 b2 errata */ if (sc->bge_chiprev == BGE_CHIPREV_5700_BX) bge_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx); sc->bge_tx_prodidx = prodidx; /* * Set a timeout in case the chip goes out to lunch. */ sc->bge_timer = 5; } } /* * Main transmit routine. To avoid having to do mbuf copies, we put pointers * to the mbuf data regions directly in the transmit descriptors. */ static void bge_start(struct ifnet *ifp) { struct bge_softc *sc; sc = ifp->if_softc; BGE_LOCK(sc); bge_start_locked(ifp); BGE_UNLOCK(sc); } static void bge_init_locked(struct bge_softc *sc) { struct ifnet *ifp; uint16_t *m; uint32_t mode; BGE_LOCK_ASSERT(sc); ifp = sc->bge_ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; /* Cancel pending I/O and flush buffers. */ bge_stop(sc); bge_stop_fw(sc); bge_sig_pre_reset(sc, BGE_RESET_START); bge_reset(sc); bge_sig_legacy(sc, BGE_RESET_START); bge_sig_post_reset(sc, BGE_RESET_START); bge_chipinit(sc); /* * Init the various state machines, ring * control blocks and firmware. */ if (bge_blockinit(sc)) { device_printf(sc->bge_dev, "initialization failure\n"); return; } ifp = sc->bge_ifp; /* Specify MTU. */ CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + (ifp->if_capenable & IFCAP_VLAN_MTU ? ETHER_VLAN_ENCAP_LEN : 0)); /* Load our MAC address. */ m = (uint16_t *)IF_LLADDR(sc->bge_ifp); CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0])); CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2])); /* Program promiscuous mode. */ bge_setpromisc(sc); /* Program multicast filter. */ bge_setmulti(sc); /* Program VLAN tag stripping. */ bge_setvlan(sc); /* Override UDP checksum offloading. */ if (sc->bge_forced_udpcsum == 0) sc->bge_csum_features &= ~CSUM_UDP; else sc->bge_csum_features |= CSUM_UDP; if (ifp->if_capabilities & IFCAP_TXCSUM && ifp->if_capenable & IFCAP_TXCSUM) { ifp->if_hwassist &= ~(BGE_CSUM_FEATURES | CSUM_UDP); ifp->if_hwassist |= sc->bge_csum_features; } /* Init RX ring. */ if (bge_init_rx_ring_std(sc) != 0) { device_printf(sc->bge_dev, "no memory for std Rx buffers.\n"); bge_stop(sc); return; } /* * Workaround for a bug in 5705 ASIC rev A0. Poll the NIC's * memory to insure that the chip has in fact read the first * entry of the ring. */ if (sc->bge_chipid == BGE_CHIPID_BCM5705_A0) { uint32_t v, i; for (i = 0; i < 10; i++) { DELAY(20); v = bge_readmem_ind(sc, BGE_STD_RX_RINGS + 8); if (v == (MCLBYTES - ETHER_ALIGN)) break; } if (i == 10) device_printf (sc->bge_dev, "5705 A0 chip failed to load RX ring\n"); } /* Init jumbo RX ring. */ if (BGE_IS_JUMBO_CAPABLE(sc) && ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN + ETHER_VLAN_ENCAP_LEN > (MCLBYTES - ETHER_ALIGN)) { if (bge_init_rx_ring_jumbo(sc) != 0) { device_printf(sc->bge_dev, "no memory for jumbo Rx buffers.\n"); bge_stop(sc); return; } } /* Init our RX return ring index. */ sc->bge_rx_saved_considx = 0; /* Init our RX/TX stat counters. */ sc->bge_rx_discards = sc->bge_tx_discards = sc->bge_tx_collisions = 0; /* Init TX ring. */ bge_init_tx_ring(sc); /* Enable TX MAC state machine lockup fix. */ mode = CSR_READ_4(sc, BGE_TX_MODE); if (BGE_IS_5755_PLUS(sc) || sc->bge_asicrev == BGE_ASICREV_BCM5906) mode |= BGE_TXMODE_MBUF_LOCKUP_FIX; if (sc->bge_asicrev == BGE_ASICREV_BCM5720) { mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE); mode |= CSR_READ_4(sc, BGE_TX_MODE) & (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE); } /* Turn on transmitter. */ CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE); /* Turn on receiver. */ BGE_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE); /* * Set the number of good frames to receive after RX MBUF * Low Watermark has been reached. After the RX MAC receives * this number of frames, it will drop subsequent incoming * frames until the MBUF High Watermark is reached. */ if (sc->bge_asicrev == BGE_ASICREV_BCM57765) CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1); else CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2); /* Clear MAC statistics. */ if (BGE_IS_5705_PLUS(sc)) bge_stats_clear_regs(sc); /* Tell firmware we're alive. */ BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); #ifdef DEVICE_POLLING /* Disable interrupts if we are polling. */ if (ifp->if_capenable & IFCAP_POLLING) { BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR); bge_writembx(sc, BGE_MBX_IRQ0_LO, 1); } else #endif /* Enable host interrupts. */ { BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA); BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR); bge_writembx(sc, BGE_MBX_IRQ0_LO, 0); } bge_ifmedia_upd_locked(ifp); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; callout_reset(&sc->bge_stat_ch, hz, bge_tick, sc); } static void bge_init(void *xsc) { struct bge_softc *sc = xsc; BGE_LOCK(sc); bge_init_locked(sc); BGE_UNLOCK(sc); } /* * Set media options. */ static int bge_ifmedia_upd(struct ifnet *ifp) { struct bge_softc *sc = ifp->if_softc; int res; BGE_LOCK(sc); res = bge_ifmedia_upd_locked(ifp); BGE_UNLOCK(sc); return (res); } static int bge_ifmedia_upd_locked(struct ifnet *ifp) { struct bge_softc *sc = ifp->if_softc; struct mii_data *mii; struct mii_softc *miisc; struct ifmedia *ifm; BGE_LOCK_ASSERT(sc); ifm = &sc->bge_ifmedia; /* If this is a 1000baseX NIC, enable the TBI port. */ if (sc->bge_flags & BGE_FLAG_TBI) { if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); switch(IFM_SUBTYPE(ifm->ifm_media)) { case IFM_AUTO: /* * The BCM5704 ASIC appears to have a special * mechanism for programming the autoneg * advertisement registers in TBI mode. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5704) { uint32_t sgdig; sgdig = CSR_READ_4(sc, BGE_SGDIG_STS); if (sgdig & BGE_SGDIGSTS_DONE) { CSR_WRITE_4(sc, BGE_TX_TBI_AUTONEG, 0); sgdig = CSR_READ_4(sc, BGE_SGDIG_CFG); sgdig |= BGE_SGDIGCFG_AUTO | BGE_SGDIGCFG_PAUSE_CAP | BGE_SGDIGCFG_ASYM_PAUSE; CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig | BGE_SGDIGCFG_SEND); DELAY(5); CSR_WRITE_4(sc, BGE_SGDIG_CFG, sgdig); } } break; case IFM_1000_SX: if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) { BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX); } else { BGE_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX); } break; default: return (EINVAL); } return (0); } sc->bge_link_evt++; mii = device_get_softc(sc->bge_miibus); LIST_FOREACH(miisc, &mii->mii_phys, mii_list) mii_phy_reset(miisc); mii_mediachg(mii); /* * Force an interrupt so that we will call bge_link_upd * if needed and clear any pending link state attention. * Without this we are not getting any further interrupts * for link state changes and thus will not UP the link and * not be able to send in bge_start_locked. The only * way to get things working was to receive a packet and * get an RX intr. * bge_tick should help for fiber cards and we might not * need to do this here if BGE_FLAG_TBI is set but as * we poll for fiber anyway it should not harm. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5700 || sc->bge_flags & BGE_FLAG_5788) BGE_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET); else BGE_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW); return (0); } /* * Report current media status. */ static void bge_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { struct bge_softc *sc = ifp->if_softc; struct mii_data *mii; BGE_LOCK(sc); if (sc->bge_flags & BGE_FLAG_TBI) { ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (CSR_READ_4(sc, BGE_MAC_STS) & BGE_MACSTAT_TBI_PCS_SYNCHED) ifmr->ifm_status |= IFM_ACTIVE; else { ifmr->ifm_active |= IFM_NONE; BGE_UNLOCK(sc); return; } ifmr->ifm_active |= IFM_1000_SX; if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX) ifmr->ifm_active |= IFM_HDX; else ifmr->ifm_active |= IFM_FDX; BGE_UNLOCK(sc); return; } mii = device_get_softc(sc->bge_miibus); mii_pollstat(mii); ifmr->ifm_active = mii->mii_media_active; ifmr->ifm_status = mii->mii_media_status; BGE_UNLOCK(sc); } static int bge_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct bge_softc *sc = ifp->if_softc; struct ifreq *ifr = (struct ifreq *) data; struct mii_data *mii; int flags, mask, error = 0; switch (command) { case SIOCSIFMTU: if (BGE_IS_JUMBO_CAPABLE(sc) || (sc->bge_flags & BGE_FLAG_JUMBO_STD)) { if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > BGE_JUMBO_MTU) { error = EINVAL; break; } } else if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU) { error = EINVAL; break; } BGE_LOCK(sc); if (ifp->if_mtu != ifr->ifr_mtu) { ifp->if_mtu = ifr->ifr_mtu; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; bge_init_locked(sc); } } BGE_UNLOCK(sc); break; case SIOCSIFFLAGS: BGE_LOCK(sc); if (ifp->if_flags & IFF_UP) { /* * If only the state of the PROMISC flag changed, * then just use the 'set promisc mode' command * instead of reinitializing the entire NIC. Doing * a full re-init means reloading the firmware and * waiting for it to start up, which may take a * second or two. Similarly for ALLMULTI. */ if (ifp->if_drv_flags & IFF_DRV_RUNNING) { flags = ifp->if_flags ^ sc->bge_if_flags; if (flags & IFF_PROMISC) bge_setpromisc(sc); if (flags & IFF_ALLMULTI) bge_setmulti(sc); } else bge_init_locked(sc); } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { bge_stop(sc); } } sc->bge_if_flags = ifp->if_flags; BGE_UNLOCK(sc); error = 0; break; case SIOCADDMULTI: case SIOCDELMULTI: if (ifp->if_drv_flags & IFF_DRV_RUNNING) { BGE_LOCK(sc); bge_setmulti(sc); BGE_UNLOCK(sc); error = 0; } break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: if (sc->bge_flags & BGE_FLAG_TBI) { error = ifmedia_ioctl(ifp, ifr, &sc->bge_ifmedia, command); } else { mii = device_get_softc(sc->bge_miibus); error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command); } break; case SIOCSIFCAP: mask = ifr->ifr_reqcap ^ ifp->if_capenable; #ifdef DEVICE_POLLING if (mask & IFCAP_POLLING) { if (ifr->ifr_reqcap & IFCAP_POLLING) { error = ether_poll_register(bge_poll, ifp); if (error) return (error); BGE_LOCK(sc); BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR); bge_writembx(sc, BGE_MBX_IRQ0_LO, 1); ifp->if_capenable |= IFCAP_POLLING; BGE_UNLOCK(sc); } else { error = ether_poll_deregister(ifp); /* Enable interrupt even in error case */ BGE_LOCK(sc); BGE_CLRBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR); bge_writembx(sc, BGE_MBX_IRQ0_LO, 0); ifp->if_capenable &= ~IFCAP_POLLING; BGE_UNLOCK(sc); } } #endif if ((mask & IFCAP_TXCSUM) != 0 && (ifp->if_capabilities & IFCAP_TXCSUM) != 0) { ifp->if_capenable ^= IFCAP_TXCSUM; if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) ifp->if_hwassist |= sc->bge_csum_features; else ifp->if_hwassist &= ~sc->bge_csum_features; } if ((mask & IFCAP_RXCSUM) != 0 && (ifp->if_capabilities & IFCAP_RXCSUM) != 0) ifp->if_capenable ^= IFCAP_RXCSUM; if ((mask & IFCAP_TSO4) != 0 && (ifp->if_capabilities & IFCAP_TSO4) != 0) { ifp->if_capenable ^= IFCAP_TSO4; if ((ifp->if_capenable & IFCAP_TSO4) != 0) ifp->if_hwassist |= CSUM_TSO; else ifp->if_hwassist &= ~CSUM_TSO; } if (mask & IFCAP_VLAN_MTU) { ifp->if_capenable ^= IFCAP_VLAN_MTU; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; bge_init(sc); } if ((mask & IFCAP_VLAN_HWTSO) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if ((mask & IFCAP_VLAN_HWTAGGING) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) ifp->if_capenable &= ~IFCAP_VLAN_HWTSO; BGE_LOCK(sc); bge_setvlan(sc); BGE_UNLOCK(sc); } #ifdef VLAN_CAPABILITIES VLAN_CAPABILITIES(ifp); #endif break; default: error = ether_ioctl(ifp, command, data); break; } return (error); } static void bge_watchdog(struct bge_softc *sc) { struct ifnet *ifp; BGE_LOCK_ASSERT(sc); if (sc->bge_timer == 0 || --sc->bge_timer) return; ifp = sc->bge_ifp; if_printf(ifp, "watchdog timeout -- resetting\n"); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; bge_init_locked(sc); ifp->if_oerrors++; } static void bge_stop_block(struct bge_softc *sc, bus_size_t reg, uint32_t bit) { int i; BGE_CLRBIT(sc, reg, bit); for (i = 0; i < BGE_TIMEOUT; i++) { if ((CSR_READ_4(sc, reg) & bit) == 0) return; DELAY(100); } } /* * Stop the adapter and free any mbufs allocated to the * RX and TX lists. */ static void bge_stop(struct bge_softc *sc) { struct ifnet *ifp; BGE_LOCK_ASSERT(sc); ifp = sc->bge_ifp; callout_stop(&sc->bge_stat_ch); /* Disable host interrupts. */ BGE_SETBIT(sc, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_MASK_PCI_INTR); bge_writembx(sc, BGE_MBX_IRQ0_LO, 1); /* * Tell firmware we're shutting down. */ bge_stop_fw(sc); bge_sig_pre_reset(sc, BGE_RESET_STOP); /* * Disable all of the receiver blocks. */ bge_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE); bge_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE); bge_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE); if (BGE_IS_5700_FAMILY(sc)) bge_stop_block(sc, BGE_RXLS_MODE, BGE_RXLSMODE_ENABLE); bge_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE); bge_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE); bge_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE); /* * Disable all of the transmit blocks. */ bge_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE); bge_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE); bge_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE); bge_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE); bge_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE); if (BGE_IS_5700_FAMILY(sc)) bge_stop_block(sc, BGE_DMAC_MODE, BGE_DMACMODE_ENABLE); bge_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE); /* * Shut down all of the memory managers and related * state machines. */ bge_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE); bge_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE); if (BGE_IS_5700_FAMILY(sc)) bge_stop_block(sc, BGE_MBCF_MODE, BGE_MBCFMODE_ENABLE); CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF); CSR_WRITE_4(sc, BGE_FTQ_RESET, 0); if (!(BGE_IS_5705_PLUS(sc))) { BGE_CLRBIT(sc, BGE_BMAN_MODE, BGE_BMANMODE_ENABLE); BGE_CLRBIT(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE); } /* Update MAC statistics. */ if (BGE_IS_5705_PLUS(sc)) bge_stats_update_regs(sc); bge_reset(sc); bge_sig_legacy(sc, BGE_RESET_STOP); bge_sig_post_reset(sc, BGE_RESET_STOP); /* * Keep the ASF firmware running if up. */ if (sc->bge_asf_mode & ASF_STACKUP) BGE_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); else BGE_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP); /* Free the RX lists. */ bge_free_rx_ring_std(sc); /* Free jumbo RX list. */ if (BGE_IS_JUMBO_CAPABLE(sc)) bge_free_rx_ring_jumbo(sc); /* Free TX buffers. */ bge_free_tx_ring(sc); sc->bge_tx_saved_considx = BGE_TXCONS_UNSET; /* Clear MAC's link state (PHY may still have link UP). */ if (bootverbose && sc->bge_link) if_printf(sc->bge_ifp, "link DOWN\n"); sc->bge_link = 0; ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); } /* * Stop all chip I/O so that the kernel's probe routines don't * get confused by errant DMAs when rebooting. */ static int bge_shutdown(device_t dev) { struct bge_softc *sc; sc = device_get_softc(dev); BGE_LOCK(sc); bge_stop(sc); bge_reset(sc); BGE_UNLOCK(sc); return (0); } static int bge_suspend(device_t dev) { struct bge_softc *sc; sc = device_get_softc(dev); BGE_LOCK(sc); bge_stop(sc); BGE_UNLOCK(sc); return (0); } static int bge_resume(device_t dev) { struct bge_softc *sc; struct ifnet *ifp; sc = device_get_softc(dev); BGE_LOCK(sc); ifp = sc->bge_ifp; if (ifp->if_flags & IFF_UP) { bge_init_locked(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) bge_start_locked(ifp); } BGE_UNLOCK(sc); return (0); } static void bge_link_upd(struct bge_softc *sc) { struct mii_data *mii; uint32_t link, status; BGE_LOCK_ASSERT(sc); /* Clear 'pending link event' flag. */ sc->bge_link_evt = 0; /* * Process link state changes. * Grrr. The link status word in the status block does * not work correctly on the BCM5700 rev AX and BX chips, * according to all available information. Hence, we have * to enable MII interrupts in order to properly obtain * async link changes. Unfortunately, this also means that * we have to read the MAC status register to detect link * changes, thereby adding an additional register access to * the interrupt handler. * * XXX: perhaps link state detection procedure used for * BGE_CHIPID_BCM5700_B2 can be used for others BCM5700 revisions. */ if (sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_B2) { status = CSR_READ_4(sc, BGE_MAC_STS); if (status & BGE_MACSTAT_MI_INTERRUPT) { mii = device_get_softc(sc->bge_miibus); mii_pollstat(mii); if (!sc->bge_link && mii->mii_media_status & IFM_ACTIVE && IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) { sc->bge_link++; if (bootverbose) if_printf(sc->bge_ifp, "link UP\n"); } else if (sc->bge_link && (!(mii->mii_media_status & IFM_ACTIVE) || IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) { sc->bge_link = 0; if (bootverbose) if_printf(sc->bge_ifp, "link DOWN\n"); } /* Clear the interrupt. */ CSR_WRITE_4(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_MI_INTERRUPT); bge_miibus_readreg(sc->bge_dev, 1, BRGPHY_MII_ISR); bge_miibus_writereg(sc->bge_dev, 1, BRGPHY_MII_IMR, BRGPHY_INTRS); } return; } if (sc->bge_flags & BGE_FLAG_TBI) { status = CSR_READ_4(sc, BGE_MAC_STS); if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) { if (!sc->bge_link) { sc->bge_link++; if (sc->bge_asicrev == BGE_ASICREV_BCM5704) BGE_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_TBI_SEND_CFGS); CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF); if (bootverbose) if_printf(sc->bge_ifp, "link UP\n"); if_link_state_change(sc->bge_ifp, LINK_STATE_UP); } } else if (sc->bge_link) { sc->bge_link = 0; if (bootverbose) if_printf(sc->bge_ifp, "link DOWN\n"); if_link_state_change(sc->bge_ifp, LINK_STATE_DOWN); } } else if ((sc->bge_mi_mode & BGE_MIMODE_AUTOPOLL) != 0) { /* * Some broken BCM chips have BGE_STATFLAG_LINKSTATE_CHANGED bit * in status word always set. Workaround this bug by reading * PHY link status directly. */ link = (CSR_READ_4(sc, BGE_MI_STS) & BGE_MISTS_LINK) ? 1 : 0; if (link != sc->bge_link || sc->bge_asicrev == BGE_ASICREV_BCM5700) { mii = device_get_softc(sc->bge_miibus); mii_pollstat(mii); if (!sc->bge_link && mii->mii_media_status & IFM_ACTIVE && IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) { sc->bge_link++; if (bootverbose) if_printf(sc->bge_ifp, "link UP\n"); } else if (sc->bge_link && (!(mii->mii_media_status & IFM_ACTIVE) || IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) { sc->bge_link = 0; if (bootverbose) if_printf(sc->bge_ifp, "link DOWN\n"); } } } else { /* * For controllers that call mii_tick, we have to poll * link status. */ mii = device_get_softc(sc->bge_miibus); mii_pollstat(mii); bge_miibus_statchg(sc->bge_dev); } /* Clear the attention. */ CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED | BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE | BGE_MACSTAT_LINK_CHANGED); } static void bge_add_sysctls(struct bge_softc *sc) { struct sysctl_ctx_list *ctx; struct sysctl_oid_list *children; char tn[32]; int unit; ctx = device_get_sysctl_ctx(sc->bge_dev); children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->bge_dev)); #ifdef BGE_REGISTER_DEBUG SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "debug_info", CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_debug_info, "I", "Debug Information"); SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "reg_read", CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_reg_read, "I", "Register Read"); SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mem_read", CTLTYPE_INT | CTLFLAG_RW, sc, 0, bge_sysctl_mem_read, "I", "Memory Read"); #endif unit = device_get_unit(sc->bge_dev); /* * A common design characteristic for many Broadcom client controllers * is that they only support a single outstanding DMA read operation * on the PCIe bus. This means that it will take twice as long to fetch * a TX frame that is split into header and payload buffers as it does * to fetch a single, contiguous TX frame (2 reads vs. 1 read). For * these controllers, coalescing buffers to reduce the number of memory * reads is effective way to get maximum performance(about 940Mbps). * Without collapsing TX buffers the maximum TCP bulk transfer * performance is about 850Mbps. However forcing coalescing mbufs * consumes a lot of CPU cycles, so leave it off by default. */ sc->bge_forced_collapse = 0; snprintf(tn, sizeof(tn), "dev.bge.%d.forced_collapse", unit); TUNABLE_INT_FETCH(tn, &sc->bge_forced_collapse); SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_collapse", CTLFLAG_RW, &sc->bge_forced_collapse, 0, "Number of fragmented TX buffers of a frame allowed before " "forced collapsing"); sc->bge_msi = 1; snprintf(tn, sizeof(tn), "dev.bge.%d.msi", unit); TUNABLE_INT_FETCH(tn, &sc->bge_msi); SYSCTL_ADD_INT(ctx, children, OID_AUTO, "msi", CTLFLAG_RD, &sc->bge_msi, 0, "Enable MSI"); /* * It seems all Broadcom controllers have a bug that can generate UDP * datagrams with checksum value 0 when TX UDP checksum offloading is * enabled. Generating UDP checksum value 0 is RFC 768 violation. * Even though the probability of generating such UDP datagrams is * low, I don't want to see FreeBSD boxes to inject such datagrams * into network so disable UDP checksum offloading by default. Users * still override this behavior by setting a sysctl variable, * dev.bge.0.forced_udpcsum. */ sc->bge_forced_udpcsum = 0; snprintf(tn, sizeof(tn), "dev.bge.%d.bge_forced_udpcsum", unit); TUNABLE_INT_FETCH(tn, &sc->bge_forced_udpcsum); SYSCTL_ADD_INT(ctx, children, OID_AUTO, "forced_udpcsum", CTLFLAG_RW, &sc->bge_forced_udpcsum, 0, "Enable UDP checksum offloading even if controller can " "generate UDP checksum value 0"); if (BGE_IS_5705_PLUS(sc)) bge_add_sysctl_stats_regs(sc, ctx, children); else bge_add_sysctl_stats(sc, ctx, children); } #define BGE_SYSCTL_STAT(sc, ctx, desc, parent, node, oid) \ SYSCTL_ADD_PROC(ctx, parent, OID_AUTO, oid, CTLTYPE_UINT|CTLFLAG_RD, \ sc, offsetof(struct bge_stats, node), bge_sysctl_stats, "IU", \ desc) static void bge_add_sysctl_stats(struct bge_softc *sc, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *parent) { struct sysctl_oid *tree; struct sysctl_oid_list *children, *schildren; tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats", CTLFLAG_RD, NULL, "BGE Statistics"); schildren = children = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT(sc, ctx, "Frames Dropped Due To Filters", children, COSFramesDroppedDueToFilters, "FramesDroppedDueToFilters"); BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write Queue Full", children, nicDmaWriteQueueFull, "DmaWriteQueueFull"); BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Write High Priority Queue Full", children, nicDmaWriteHighPriQueueFull, "DmaWriteHighPriQueueFull"); BGE_SYSCTL_STAT(sc, ctx, "NIC No More RX Buffer Descriptors", children, nicNoMoreRxBDs, "NoMoreRxBDs"); BGE_SYSCTL_STAT(sc, ctx, "Discarded Input Frames", children, ifInDiscards, "InputDiscards"); BGE_SYSCTL_STAT(sc, ctx, "Input Errors", children, ifInErrors, "InputErrors"); BGE_SYSCTL_STAT(sc, ctx, "NIC Recv Threshold Hit", children, nicRecvThresholdHit, "RecvThresholdHit"); BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read Queue Full", children, nicDmaReadQueueFull, "DmaReadQueueFull"); BGE_SYSCTL_STAT(sc, ctx, "NIC DMA Read High Priority Queue Full", children, nicDmaReadHighPriQueueFull, "DmaReadHighPriQueueFull"); BGE_SYSCTL_STAT(sc, ctx, "NIC Send Data Complete Queue Full", children, nicSendDataCompQueueFull, "SendDataCompQueueFull"); BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Set Send Producer Index", children, nicRingSetSendProdIndex, "RingSetSendProdIndex"); BGE_SYSCTL_STAT(sc, ctx, "NIC Ring Status Update", children, nicRingStatusUpdate, "RingStatusUpdate"); BGE_SYSCTL_STAT(sc, ctx, "NIC Interrupts", children, nicInterrupts, "Interrupts"); BGE_SYSCTL_STAT(sc, ctx, "NIC Avoided Interrupts", children, nicAvoidedInterrupts, "AvoidedInterrupts"); BGE_SYSCTL_STAT(sc, ctx, "NIC Send Threshold Hit", children, nicSendThresholdHit, "SendThresholdHit"); tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "rx", CTLFLAG_RD, NULL, "BGE RX Statistics"); children = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT(sc, ctx, "Inbound Octets", children, rxstats.ifHCInOctets, "ifHCInOctets"); BGE_SYSCTL_STAT(sc, ctx, "Fragments", children, rxstats.etherStatsFragments, "Fragments"); BGE_SYSCTL_STAT(sc, ctx, "Inbound Unicast Packets", children, rxstats.ifHCInUcastPkts, "UnicastPkts"); BGE_SYSCTL_STAT(sc, ctx, "Inbound Multicast Packets", children, rxstats.ifHCInMulticastPkts, "MulticastPkts"); BGE_SYSCTL_STAT(sc, ctx, "FCS Errors", children, rxstats.dot3StatsFCSErrors, "FCSErrors"); BGE_SYSCTL_STAT(sc, ctx, "Alignment Errors", children, rxstats.dot3StatsAlignmentErrors, "AlignmentErrors"); BGE_SYSCTL_STAT(sc, ctx, "XON Pause Frames Received", children, rxstats.xonPauseFramesReceived, "xonPauseFramesReceived"); BGE_SYSCTL_STAT(sc, ctx, "XOFF Pause Frames Received", children, rxstats.xoffPauseFramesReceived, "xoffPauseFramesReceived"); BGE_SYSCTL_STAT(sc, ctx, "MAC Control Frames Received", children, rxstats.macControlFramesReceived, "ControlFramesReceived"); BGE_SYSCTL_STAT(sc, ctx, "XOFF State Entered", children, rxstats.xoffStateEntered, "xoffStateEntered"); BGE_SYSCTL_STAT(sc, ctx, "Frames Too Long", children, rxstats.dot3StatsFramesTooLong, "FramesTooLong"); BGE_SYSCTL_STAT(sc, ctx, "Jabbers", children, rxstats.etherStatsJabbers, "Jabbers"); BGE_SYSCTL_STAT(sc, ctx, "Undersized Packets", children, rxstats.etherStatsUndersizePkts, "UndersizePkts"); BGE_SYSCTL_STAT(sc, ctx, "Inbound Range Length Errors", children, rxstats.inRangeLengthError, "inRangeLengthError"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Range Length Errors", children, rxstats.outRangeLengthError, "outRangeLengthError"); tree = SYSCTL_ADD_NODE(ctx, schildren, OID_AUTO, "tx", CTLFLAG_RD, NULL, "BGE TX Statistics"); children = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT(sc, ctx, "Outbound Octets", children, txstats.ifHCOutOctets, "ifHCOutOctets"); BGE_SYSCTL_STAT(sc, ctx, "TX Collisions", children, txstats.etherStatsCollisions, "Collisions"); BGE_SYSCTL_STAT(sc, ctx, "XON Sent", children, txstats.outXonSent, "XonSent"); BGE_SYSCTL_STAT(sc, ctx, "XOFF Sent", children, txstats.outXoffSent, "XoffSent"); BGE_SYSCTL_STAT(sc, ctx, "Flow Control Done", children, txstats.flowControlDone, "flowControlDone"); BGE_SYSCTL_STAT(sc, ctx, "Internal MAC TX errors", children, txstats.dot3StatsInternalMacTransmitErrors, "InternalMacTransmitErrors"); BGE_SYSCTL_STAT(sc, ctx, "Single Collision Frames", children, txstats.dot3StatsSingleCollisionFrames, "SingleCollisionFrames"); BGE_SYSCTL_STAT(sc, ctx, "Multiple Collision Frames", children, txstats.dot3StatsMultipleCollisionFrames, "MultipleCollisionFrames"); BGE_SYSCTL_STAT(sc, ctx, "Deferred Transmissions", children, txstats.dot3StatsDeferredTransmissions, "DeferredTransmissions"); BGE_SYSCTL_STAT(sc, ctx, "Excessive Collisions", children, txstats.dot3StatsExcessiveCollisions, "ExcessiveCollisions"); BGE_SYSCTL_STAT(sc, ctx, "Late Collisions", children, txstats.dot3StatsLateCollisions, "LateCollisions"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Unicast Packets", children, txstats.ifHCOutUcastPkts, "UnicastPkts"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Multicast Packets", children, txstats.ifHCOutMulticastPkts, "MulticastPkts"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Broadcast Packets", children, txstats.ifHCOutBroadcastPkts, "BroadcastPkts"); BGE_SYSCTL_STAT(sc, ctx, "Carrier Sense Errors", children, txstats.dot3StatsCarrierSenseErrors, "CarrierSenseErrors"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Discards", children, txstats.ifOutDiscards, "Discards"); BGE_SYSCTL_STAT(sc, ctx, "Outbound Errors", children, txstats.ifOutErrors, "Errors"); } #undef BGE_SYSCTL_STAT #define BGE_SYSCTL_STAT_ADD64(c, h, n, p, d) \ SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d) static void bge_add_sysctl_stats_regs(struct bge_softc *sc, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *parent) { struct sysctl_oid *tree; struct sysctl_oid_list *child, *schild; struct bge_mac_stats *stats; stats = &sc->bge_mac_stats; tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "stats", CTLFLAG_RD, NULL, "BGE Statistics"); schild = child = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesDroppedDueToFilters", &stats->FramesDroppedDueToFilters, "Frames Dropped Due to Filters"); BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteQueueFull", &stats->DmaWriteQueueFull, "NIC DMA Write Queue Full"); BGE_SYSCTL_STAT_ADD64(ctx, child, "DmaWriteHighPriQueueFull", &stats->DmaWriteHighPriQueueFull, "NIC DMA Write High Priority Queue Full"); BGE_SYSCTL_STAT_ADD64(ctx, child, "NoMoreRxBDs", &stats->NoMoreRxBDs, "NIC No More RX Buffer Descriptors"); BGE_SYSCTL_STAT_ADD64(ctx, child, "InputDiscards", &stats->InputDiscards, "Discarded Input Frames"); BGE_SYSCTL_STAT_ADD64(ctx, child, "InputErrors", &stats->InputErrors, "Input Errors"); BGE_SYSCTL_STAT_ADD64(ctx, child, "RecvThresholdHit", &stats->RecvThresholdHit, "NIC Recv Threshold Hit"); tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "rx", CTLFLAG_RD, NULL, "BGE RX Statistics"); child = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCInOctets", &stats->ifHCInOctets, "Inbound Octets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "Fragments", &stats->etherStatsFragments, "Fragments"); BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts", &stats->ifHCInUcastPkts, "Inbound Unicast Packets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts", &stats->ifHCInMulticastPkts, "Inbound Multicast Packets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts", &stats->ifHCInBroadcastPkts, "Inbound Broadcast Packets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "FCSErrors", &stats->dot3StatsFCSErrors, "FCS Errors"); BGE_SYSCTL_STAT_ADD64(ctx, child, "AlignmentErrors", &stats->dot3StatsAlignmentErrors, "Alignment Errors"); BGE_SYSCTL_STAT_ADD64(ctx, child, "xonPauseFramesReceived", &stats->xonPauseFramesReceived, "XON Pause Frames Received"); BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffPauseFramesReceived", &stats->xoffPauseFramesReceived, "XOFF Pause Frames Received"); BGE_SYSCTL_STAT_ADD64(ctx, child, "ControlFramesReceived", &stats->macControlFramesReceived, "MAC Control Frames Received"); BGE_SYSCTL_STAT_ADD64(ctx, child, "xoffStateEntered", &stats->xoffStateEntered, "XOFF State Entered"); BGE_SYSCTL_STAT_ADD64(ctx, child, "FramesTooLong", &stats->dot3StatsFramesTooLong, "Frames Too Long"); BGE_SYSCTL_STAT_ADD64(ctx, child, "Jabbers", &stats->etherStatsJabbers, "Jabbers"); BGE_SYSCTL_STAT_ADD64(ctx, child, "UndersizePkts", &stats->etherStatsUndersizePkts, "Undersized Packets"); tree = SYSCTL_ADD_NODE(ctx, schild, OID_AUTO, "tx", CTLFLAG_RD, NULL, "BGE TX Statistics"); child = SYSCTL_CHILDREN(tree); BGE_SYSCTL_STAT_ADD64(ctx, child, "ifHCOutOctets", &stats->ifHCOutOctets, "Outbound Octets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "Collisions", &stats->etherStatsCollisions, "TX Collisions"); BGE_SYSCTL_STAT_ADD64(ctx, child, "XonSent", &stats->outXonSent, "XON Sent"); BGE_SYSCTL_STAT_ADD64(ctx, child, "XoffSent", &stats->outXoffSent, "XOFF Sent"); BGE_SYSCTL_STAT_ADD64(ctx, child, "InternalMacTransmitErrors", &stats->dot3StatsInternalMacTransmitErrors, "Internal MAC TX Errors"); BGE_SYSCTL_STAT_ADD64(ctx, child, "SingleCollisionFrames", &stats->dot3StatsSingleCollisionFrames, "Single Collision Frames"); BGE_SYSCTL_STAT_ADD64(ctx, child, "MultipleCollisionFrames", &stats->dot3StatsMultipleCollisionFrames, "Multiple Collision Frames"); BGE_SYSCTL_STAT_ADD64(ctx, child, "DeferredTransmissions", &stats->dot3StatsDeferredTransmissions, "Deferred Transmissions"); BGE_SYSCTL_STAT_ADD64(ctx, child, "ExcessiveCollisions", &stats->dot3StatsExcessiveCollisions, "Excessive Collisions"); BGE_SYSCTL_STAT_ADD64(ctx, child, "LateCollisions", &stats->dot3StatsLateCollisions, "Late Collisions"); BGE_SYSCTL_STAT_ADD64(ctx, child, "UnicastPkts", &stats->ifHCOutUcastPkts, "Outbound Unicast Packets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "MulticastPkts", &stats->ifHCOutMulticastPkts, "Outbound Multicast Packets"); BGE_SYSCTL_STAT_ADD64(ctx, child, "BroadcastPkts", &stats->ifHCOutBroadcastPkts, "Outbound Broadcast Packets"); } #undef BGE_SYSCTL_STAT_ADD64 static int bge_sysctl_stats(SYSCTL_HANDLER_ARGS) { struct bge_softc *sc; uint32_t result; int offset; sc = (struct bge_softc *)arg1; offset = arg2; result = CSR_READ_4(sc, BGE_MEMWIN_START + BGE_STATS_BLOCK + offset + offsetof(bge_hostaddr, bge_addr_lo)); return (sysctl_handle_int(oidp, &result, 0, req)); } #ifdef BGE_REGISTER_DEBUG static int bge_sysctl_debug_info(SYSCTL_HANDLER_ARGS) { struct bge_softc *sc; uint16_t *sbdata; int error, result, sbsz; int i, j; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || (req->newptr == NULL)) return (error); if (result == 1) { sc = (struct bge_softc *)arg1; if (sc->bge_asicrev == BGE_ASICREV_BCM5700 && sc->bge_chipid != BGE_CHIPID_BCM5700_C0) sbsz = BGE_STATUS_BLK_SZ; else sbsz = 32; sbdata = (uint16_t *)sc->bge_ldata.bge_status_block; printf("Status Block:\n"); BGE_LOCK(sc); bus_dmamap_sync(sc->bge_cdata.bge_status_tag, sc->bge_cdata.bge_status_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); for (i = 0x0; i < sbsz / sizeof(uint16_t); ) { printf("%06x:", i); for (j = 0; j < 8; j++) printf(" %04x", sbdata[i++]); printf("\n"); } printf("Registers:\n"); for (i = 0x800; i < 0xA00; ) { printf("%06x:", i); for (j = 0; j < 8; j++) { printf(" %08x", CSR_READ_4(sc, i)); i += 4; } printf("\n"); } BGE_UNLOCK(sc); printf("Hardware Flags:\n"); if (BGE_IS_5717_PLUS(sc)) printf(" - 5717 Plus\n"); if (BGE_IS_5755_PLUS(sc)) printf(" - 5755 Plus\n"); if (BGE_IS_575X_PLUS(sc)) printf(" - 575X Plus\n"); if (BGE_IS_5705_PLUS(sc)) printf(" - 5705 Plus\n"); if (BGE_IS_5714_FAMILY(sc)) printf(" - 5714 Family\n"); if (BGE_IS_5700_FAMILY(sc)) printf(" - 5700 Family\n"); if (sc->bge_flags & BGE_FLAG_JUMBO) printf(" - Supports Jumbo Frames\n"); if (sc->bge_flags & BGE_FLAG_PCIX) printf(" - PCI-X Bus\n"); if (sc->bge_flags & BGE_FLAG_PCIE) printf(" - PCI Express Bus\n"); if (sc->bge_phy_flags & BGE_PHY_NO_3LED) printf(" - No 3 LEDs\n"); if (sc->bge_flags & BGE_FLAG_RX_ALIGNBUG) printf(" - RX Alignment Bug\n"); } return (error); } static int bge_sysctl_reg_read(SYSCTL_HANDLER_ARGS) { struct bge_softc *sc; int error; uint16_t result; uint32_t val; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || (req->newptr == NULL)) return (error); if (result < 0x8000) { sc = (struct bge_softc *)arg1; val = CSR_READ_4(sc, result); printf("reg 0x%06X = 0x%08X\n", result, val); } return (error); } static int bge_sysctl_mem_read(SYSCTL_HANDLER_ARGS) { struct bge_softc *sc; int error; uint16_t result; uint32_t val; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || (req->newptr == NULL)) return (error); if (result < 0x8000) { sc = (struct bge_softc *)arg1; val = bge_readmem_ind(sc, result); printf("mem 0x%06X = 0x%08X\n", result, val); } return (error); } #endif static int bge_get_eaddr_fw(struct bge_softc *sc, uint8_t ether_addr[]) { if (sc->bge_flags & BGE_FLAG_EADDR) return (1); #ifdef __sparc64__ OF_getetheraddr(sc->bge_dev, ether_addr); return (0); #endif return (1); } static int bge_get_eaddr_mem(struct bge_softc *sc, uint8_t ether_addr[]) { uint32_t mac_addr; mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_HIGH_MB); if ((mac_addr >> 16) == 0x484b) { ether_addr[0] = (uint8_t)(mac_addr >> 8); ether_addr[1] = (uint8_t)mac_addr; mac_addr = bge_readmem_ind(sc, BGE_SRAM_MAC_ADDR_LOW_MB); ether_addr[2] = (uint8_t)(mac_addr >> 24); ether_addr[3] = (uint8_t)(mac_addr >> 16); ether_addr[4] = (uint8_t)(mac_addr >> 8); ether_addr[5] = (uint8_t)mac_addr; return (0); } return (1); } static int bge_get_eaddr_nvram(struct bge_softc *sc, uint8_t ether_addr[]) { int mac_offset = BGE_EE_MAC_OFFSET; if (sc->bge_asicrev == BGE_ASICREV_BCM5906) mac_offset = BGE_EE_MAC_OFFSET_5906; return (bge_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN)); } static int bge_get_eaddr_eeprom(struct bge_softc *sc, uint8_t ether_addr[]) { if (sc->bge_asicrev == BGE_ASICREV_BCM5906) return (1); return (bge_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2, ETHER_ADDR_LEN)); } static int bge_get_eaddr(struct bge_softc *sc, uint8_t eaddr[]) { static const bge_eaddr_fcn_t bge_eaddr_funcs[] = { /* NOTE: Order is critical */ bge_get_eaddr_fw, bge_get_eaddr_mem, bge_get_eaddr_nvram, bge_get_eaddr_eeprom, NULL }; const bge_eaddr_fcn_t *func; for (func = bge_eaddr_funcs; *func != NULL; ++func) { if ((*func)(sc, eaddr) == 0) break; } return (*func == NULL ? ENXIO : 0); } Index: stable/7/sys/dev/cxgb/cxgb_main.c =================================================================== --- stable/7/sys/dev/cxgb/cxgb_main.c (revision 242221) +++ stable/7/sys/dev/cxgb/cxgb_main.c (revision 242222) @@ -1,3436 +1,3436 @@ /************************************************************************** Copyright (c) 2007-2009, Chelsio Inc. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Neither the name of the Chelsio Corporation nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ***************************************************************************/ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef CONFIG_DEFINED #include #else #include #endif #ifdef PRIV_SUPPORTED #include #endif static int cxgb_setup_interrupts(adapter_t *); static void cxgb_teardown_interrupts(adapter_t *); static void cxgb_init(void *); static int cxgb_init_locked(struct port_info *); static int cxgb_uninit_locked(struct port_info *); static int cxgb_uninit_synchronized(struct port_info *); static int cxgb_ioctl(struct ifnet *, unsigned long, caddr_t); static int cxgb_media_change(struct ifnet *); static int cxgb_ifm_type(int); static void cxgb_build_medialist(struct port_info *); static void cxgb_media_status(struct ifnet *, struct ifmediareq *); static int setup_sge_qsets(adapter_t *); static void cxgb_async_intr(void *); static void cxgb_tick_handler(void *, int); static void cxgb_tick(void *); static void link_check_callout(void *); static void check_link_status(void *, int); static void setup_rss(adapter_t *sc); static int alloc_filters(struct adapter *); static int setup_hw_filters(struct adapter *); static int set_filter(struct adapter *, int, const struct filter_info *); static inline void mk_set_tcb_field(struct cpl_set_tcb_field *, unsigned int, unsigned int, u64, u64); static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *, unsigned int, unsigned int, u64, u64); /* Attachment glue for the PCI controller end of the device. Each port of * the device is attached separately, as defined later. */ static int cxgb_controller_probe(device_t); static int cxgb_controller_attach(device_t); static int cxgb_controller_detach(device_t); static void cxgb_free(struct adapter *); static __inline void reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start, unsigned int end); static void cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf); static int cxgb_get_regs_len(void); static int offload_open(struct port_info *pi); static void touch_bars(device_t dev); static int offload_close(struct t3cdev *tdev); static void cxgb_update_mac_settings(struct port_info *p); static device_method_t cxgb_controller_methods[] = { DEVMETHOD(device_probe, cxgb_controller_probe), DEVMETHOD(device_attach, cxgb_controller_attach), DEVMETHOD(device_detach, cxgb_controller_detach), /* bus interface */ DEVMETHOD(bus_print_child, bus_generic_print_child), DEVMETHOD(bus_driver_added, bus_generic_driver_added), { 0, 0 } }; static driver_t cxgb_controller_driver = { "cxgbc", cxgb_controller_methods, sizeof(struct adapter) }; static devclass_t cxgb_controller_devclass; DRIVER_MODULE(cxgbc, pci, cxgb_controller_driver, cxgb_controller_devclass, 0, 0); /* * Attachment glue for the ports. Attachment is done directly to the * controller device. */ static int cxgb_port_probe(device_t); static int cxgb_port_attach(device_t); static int cxgb_port_detach(device_t); static device_method_t cxgb_port_methods[] = { DEVMETHOD(device_probe, cxgb_port_probe), DEVMETHOD(device_attach, cxgb_port_attach), DEVMETHOD(device_detach, cxgb_port_detach), { 0, 0 } }; static driver_t cxgb_port_driver = { "cxgb", cxgb_port_methods, 0 }; static d_ioctl_t cxgb_extension_ioctl; static d_open_t cxgb_extension_open; static d_close_t cxgb_extension_close; static struct cdevsw cxgb_cdevsw = { .d_version = D_VERSION, .d_flags = 0, .d_open = cxgb_extension_open, .d_close = cxgb_extension_close, .d_ioctl = cxgb_extension_ioctl, .d_name = "cxgb", }; static devclass_t cxgb_port_devclass; DRIVER_MODULE(cxgb, cxgbc, cxgb_port_driver, cxgb_port_devclass, 0, 0); /* * The driver uses the best interrupt scheme available on a platform in the * order MSI-X, MSI, legacy pin interrupts. This parameter determines which * of these schemes the driver may consider as follows: * * msi = 2: choose from among all three options * msi = 1 : only consider MSI and pin interrupts * msi = 0: force pin interrupts */ static int msi_allowed = 2; TUNABLE_INT("hw.cxgb.msi_allowed", &msi_allowed); SYSCTL_NODE(_hw, OID_AUTO, cxgb, CTLFLAG_RD, 0, "CXGB driver parameters"); SYSCTL_UINT(_hw_cxgb, OID_AUTO, msi_allowed, CTLFLAG_RDTUN, &msi_allowed, 0, "MSI-X, MSI, INTx selector"); /* * The driver enables offload as a default. * To disable it, use ofld_disable = 1. */ static int ofld_disable = 0; TUNABLE_INT("hw.cxgb.ofld_disable", &ofld_disable); SYSCTL_UINT(_hw_cxgb, OID_AUTO, ofld_disable, CTLFLAG_RDTUN, &ofld_disable, 0, "disable ULP offload"); /* * The driver uses an auto-queue algorithm by default. * To disable it and force a single queue-set per port, use multiq = 0 */ static int multiq = 1; TUNABLE_INT("hw.cxgb.multiq", &multiq); SYSCTL_UINT(_hw_cxgb, OID_AUTO, multiq, CTLFLAG_RDTUN, &multiq, 0, "use min(ncpus/ports, 8) queue-sets per port"); /* * By default the driver will not update the firmware unless * it was compiled against a newer version * */ static int force_fw_update = 0; TUNABLE_INT("hw.cxgb.force_fw_update", &force_fw_update); SYSCTL_UINT(_hw_cxgb, OID_AUTO, force_fw_update, CTLFLAG_RDTUN, &force_fw_update, 0, "update firmware even if up to date"); int cxgb_use_16k_clusters = 1; TUNABLE_INT("hw.cxgb.use_16k_clusters", &cxgb_use_16k_clusters); SYSCTL_UINT(_hw_cxgb, OID_AUTO, use_16k_clusters, CTLFLAG_RDTUN, &cxgb_use_16k_clusters, 0, "use 16kB clusters for the jumbo queue "); /* * Tune the size of the output queue. */ int cxgb_snd_queue_len = IFQ_MAXLEN; TUNABLE_INT("hw.cxgb.snd_queue_len", &cxgb_snd_queue_len); SYSCTL_UINT(_hw_cxgb, OID_AUTO, snd_queue_len, CTLFLAG_RDTUN, &cxgb_snd_queue_len, 0, "send queue size "); static int nfilters = -1; TUNABLE_INT("hw.cxgb.nfilters", &nfilters); SYSCTL_INT(_hw_cxgb, OID_AUTO, nfilters, CTLFLAG_RDTUN, &nfilters, 0, "max number of entries in the filter table"); enum { MAX_TXQ_ENTRIES = 16384, MAX_CTRL_TXQ_ENTRIES = 1024, MAX_RSPQ_ENTRIES = 16384, MAX_RX_BUFFERS = 16384, MAX_RX_JUMBO_BUFFERS = 16384, MIN_TXQ_ENTRIES = 4, MIN_CTRL_TXQ_ENTRIES = 4, MIN_RSPQ_ENTRIES = 32, MIN_FL_ENTRIES = 32, MIN_FL_JUMBO_ENTRIES = 32 }; struct filter_info { u32 sip; u32 sip_mask; u32 dip; u16 sport; u16 dport; u32 vlan:12; u32 vlan_prio:3; u32 mac_hit:1; u32 mac_idx:4; u32 mac_vld:1; u32 pkt_type:2; u32 report_filter_id:1; u32 pass:1; u32 rss:1; u32 qset:3; u32 locked:1; u32 valid:1; }; enum { FILTER_NO_VLAN_PRI = 7 }; #define EEPROM_MAGIC 0x38E2F10C #define PORT_MASK ((1 << MAX_NPORTS) - 1) /* Table for probing the cards. The desc field isn't actually used */ struct cxgb_ident { uint16_t vendor; uint16_t device; int index; char *desc; } cxgb_identifiers[] = { {PCI_VENDOR_ID_CHELSIO, 0x0020, 0, "PE9000"}, {PCI_VENDOR_ID_CHELSIO, 0x0021, 1, "T302E"}, {PCI_VENDOR_ID_CHELSIO, 0x0022, 2, "T310E"}, {PCI_VENDOR_ID_CHELSIO, 0x0023, 3, "T320X"}, {PCI_VENDOR_ID_CHELSIO, 0x0024, 1, "T302X"}, {PCI_VENDOR_ID_CHELSIO, 0x0025, 3, "T320E"}, {PCI_VENDOR_ID_CHELSIO, 0x0026, 2, "T310X"}, {PCI_VENDOR_ID_CHELSIO, 0x0030, 2, "T3B10"}, {PCI_VENDOR_ID_CHELSIO, 0x0031, 3, "T3B20"}, {PCI_VENDOR_ID_CHELSIO, 0x0032, 1, "T3B02"}, {PCI_VENDOR_ID_CHELSIO, 0x0033, 4, "T3B04"}, {PCI_VENDOR_ID_CHELSIO, 0x0035, 6, "T3C10"}, {PCI_VENDOR_ID_CHELSIO, 0x0036, 3, "S320E-CR"}, {PCI_VENDOR_ID_CHELSIO, 0x0037, 7, "N320E-G2"}, {0, 0, 0, NULL} }; static int set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset); static __inline char t3rev2char(struct adapter *adapter) { char rev = 'z'; switch(adapter->params.rev) { case T3_REV_A: rev = 'a'; break; case T3_REV_B: case T3_REV_B2: rev = 'b'; break; case T3_REV_C: rev = 'c'; break; } return rev; } static struct cxgb_ident * cxgb_get_ident(device_t dev) { struct cxgb_ident *id; for (id = cxgb_identifiers; id->desc != NULL; id++) { if ((id->vendor == pci_get_vendor(dev)) && (id->device == pci_get_device(dev))) { return (id); } } return (NULL); } static const struct adapter_info * cxgb_get_adapter_info(device_t dev) { struct cxgb_ident *id; const struct adapter_info *ai; id = cxgb_get_ident(dev); if (id == NULL) return (NULL); ai = t3_get_adapter_info(id->index); return (ai); } static int cxgb_controller_probe(device_t dev) { const struct adapter_info *ai; char *ports, buf[80]; int nports; ai = cxgb_get_adapter_info(dev); if (ai == NULL) return (ENXIO); nports = ai->nports0 + ai->nports1; if (nports == 1) ports = "port"; else ports = "ports"; snprintf(buf, sizeof(buf), "%s, %d %s", ai->desc, nports, ports); device_set_desc_copy(dev, buf); return (BUS_PROBE_DEFAULT); } #define FW_FNAME "cxgb_t3fw" #define TPEEPROM_NAME "cxgb_t3%c_tp_eeprom" #define TPSRAM_NAME "cxgb_t3%c_protocol_sram" static int upgrade_fw(adapter_t *sc) { #ifdef FIRMWARE_LATEST const struct firmware *fw; #else struct firmware *fw; #endif int status; u32 vers; if ((fw = firmware_get(FW_FNAME)) == NULL) { device_printf(sc->dev, "Could not find firmware image %s\n", FW_FNAME); return (ENOENT); } else device_printf(sc->dev, "installing firmware on card\n"); status = t3_load_fw(sc, (const uint8_t *)fw->data, fw->datasize); if (status != 0) { device_printf(sc->dev, "failed to install firmware: %d\n", status); } else { t3_get_fw_version(sc, &vers); snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d", G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers)); } firmware_put(fw, FIRMWARE_UNLOAD); return (status); } /* * The cxgb_controller_attach function is responsible for the initial * bringup of the device. Its responsibilities include: * * 1. Determine if the device supports MSI or MSI-X. * 2. Allocate bus resources so that we can access the Base Address Register * 3. Create and initialize mutexes for the controller and its control * logic such as SGE and MDIO. * 4. Call hardware specific setup routine for the adapter as a whole. * 5. Allocate the BAR for doing MSI-X. * 6. Setup the line interrupt iff MSI-X is not supported. * 7. Create the driver's taskq. * 8. Start one task queue service thread. * 9. Check if the firmware and SRAM are up-to-date. They will be * auto-updated later (before FULL_INIT_DONE), if required. * 10. Create a child device for each MAC (port) * 11. Initialize T3 private state. * 12. Trigger the LED * 13. Setup offload iff supported. * 14. Reset/restart the tick callout. * 15. Attach sysctls * * NOTE: Any modification or deviation from this list MUST be reflected in * the above comment. Failure to do so will result in problems on various * error conditions including link flapping. */ static int cxgb_controller_attach(device_t dev) { device_t child; const struct adapter_info *ai; struct adapter *sc; int i, error = 0; uint32_t vers; int port_qsets = 1; #ifdef MSI_SUPPORTED int msi_needed, reg; #endif char buf[80]; sc = device_get_softc(dev); sc->dev = dev; sc->msi_count = 0; ai = cxgb_get_adapter_info(dev); /* * XXX not really related but a recent addition */ #ifdef MSI_SUPPORTED /* find the PCIe link width and set max read request to 4KB*/ if (pci_find_extcap(dev, PCIY_EXPRESS, ®) == 0) { uint16_t lnk; - lnk = pci_read_config(dev, reg + PCIR_EXPRESS_LINK_STA, 2); - sc->link_width = (lnk & PCIM_LINK_STA_WIDTH) >> 4; + lnk = pci_read_config(dev, reg + PCIER_LINK_STA, 2); + sc->link_width = (lnk & PCIEM_LINK_STA_WIDTH) >> 4; if (sc->link_width < 8 && (ai->caps & SUPPORTED_10000baseT_Full)) { device_printf(sc->dev, "PCIe x%d Link, expect reduced performance\n", sc->link_width); } pci_set_max_read_req(dev, 4096); } #endif touch_bars(dev); pci_enable_busmaster(dev); /* * Allocate the registers and make them available to the driver. * The registers that we care about for NIC mode are in BAR 0 */ sc->regs_rid = PCIR_BAR(0); if ((sc->regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->regs_rid, RF_ACTIVE)) == NULL) { device_printf(dev, "Cannot allocate BAR region 0\n"); return (ENXIO); } sc->udbs_rid = PCIR_BAR(2); sc->udbs_res = NULL; if (is_offload(sc) && ((sc->udbs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->udbs_rid, RF_ACTIVE)) == NULL)) { device_printf(dev, "Cannot allocate BAR region 1\n"); error = ENXIO; goto out; } snprintf(sc->lockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb controller lock %d", device_get_unit(dev)); ADAPTER_LOCK_INIT(sc, sc->lockbuf); snprintf(sc->reglockbuf, ADAPTER_LOCK_NAME_LEN, "SGE reg lock %d", device_get_unit(dev)); snprintf(sc->mdiolockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb mdio lock %d", device_get_unit(dev)); snprintf(sc->elmerlockbuf, ADAPTER_LOCK_NAME_LEN, "cxgb elmer lock %d", device_get_unit(dev)); MTX_INIT(&sc->sge.reg_lock, sc->reglockbuf, NULL, MTX_SPIN); MTX_INIT(&sc->mdio_lock, sc->mdiolockbuf, NULL, MTX_DEF); MTX_INIT(&sc->elmer_lock, sc->elmerlockbuf, NULL, MTX_DEF); sc->bt = rman_get_bustag(sc->regs_res); sc->bh = rman_get_bushandle(sc->regs_res); sc->mmio_len = rman_get_size(sc->regs_res); for (i = 0; i < MAX_NPORTS; i++) sc->port[i].adapter = sc; if (t3_prep_adapter(sc, ai, 1) < 0) { printf("prep adapter failed\n"); error = ENODEV; goto out; } /* Allocate the BAR for doing MSI-X. If it succeeds, try to allocate * enough messages for the queue sets. If that fails, try falling * back to MSI. If that fails, then try falling back to the legacy * interrupt pin model. */ #ifdef MSI_SUPPORTED sc->msix_regs_rid = 0x20; if ((msi_allowed >= 2) && (sc->msix_regs_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->msix_regs_rid, RF_ACTIVE)) != NULL) { if (multiq) port_qsets = min(SGE_QSETS/sc->params.nports, mp_ncpus); msi_needed = sc->msi_count = sc->params.nports * port_qsets + 1; if (pci_msix_count(dev) == 0 || (error = pci_alloc_msix(dev, &sc->msi_count)) != 0 || sc->msi_count != msi_needed) { device_printf(dev, "alloc msix failed - " "msi_count=%d, msi_needed=%d, err=%d; " "will try MSI\n", sc->msi_count, msi_needed, error); sc->msi_count = 0; port_qsets = 1; pci_release_msi(dev); bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_regs_rid, sc->msix_regs_res); sc->msix_regs_res = NULL; } else { sc->flags |= USING_MSIX; sc->cxgb_intr = cxgb_async_intr; device_printf(dev, "using MSI-X interrupts (%u vectors)\n", sc->msi_count); } } if ((msi_allowed >= 1) && (sc->msi_count == 0)) { sc->msi_count = 1; if ((error = pci_alloc_msi(dev, &sc->msi_count)) != 0) { device_printf(dev, "alloc msi failed - " "err=%d; will try INTx\n", error); sc->msi_count = 0; port_qsets = 1; pci_release_msi(dev); } else { sc->flags |= USING_MSI; sc->cxgb_intr = t3_intr_msi; device_printf(dev, "using MSI interrupts\n"); } } #endif if (sc->msi_count == 0) { device_printf(dev, "using line interrupts\n"); sc->cxgb_intr = t3b_intr; } /* Create a private taskqueue thread for handling driver events */ #ifdef TASKQUEUE_CURRENT sc->tq = taskqueue_create("cxgb_taskq", M_NOWAIT, taskqueue_thread_enqueue, &sc->tq); #else sc->tq = taskqueue_create_fast("cxgb_taskq", M_NOWAIT, taskqueue_thread_enqueue, &sc->tq); #endif if (sc->tq == NULL) { device_printf(dev, "failed to allocate controller task queue\n"); goto out; } taskqueue_start_threads(&sc->tq, 1, PI_NET, "%s taskq", device_get_nameunit(dev)); TASK_INIT(&sc->tick_task, 0, cxgb_tick_handler, sc); /* Create a periodic callout for checking adapter status */ callout_init(&sc->cxgb_tick_ch, TRUE); if (t3_check_fw_version(sc) < 0 || force_fw_update) { /* * Warn user that a firmware update will be attempted in init. */ device_printf(dev, "firmware needs to be updated to version %d.%d.%d\n", FW_VERSION_MAJOR, FW_VERSION_MINOR, FW_VERSION_MICRO); sc->flags &= ~FW_UPTODATE; } else { sc->flags |= FW_UPTODATE; } if (t3_check_tpsram_version(sc) < 0) { /* * Warn user that a firmware update will be attempted in init. */ device_printf(dev, "SRAM needs to be updated to version %c-%d.%d.%d\n", t3rev2char(sc), TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO); sc->flags &= ~TPS_UPTODATE; } else { sc->flags |= TPS_UPTODATE; } /* * Create a child device for each MAC. The ethernet attachment * will be done in these children. */ for (i = 0; i < (sc)->params.nports; i++) { struct port_info *pi; if ((child = device_add_child(dev, "cxgb", -1)) == NULL) { device_printf(dev, "failed to add child port\n"); error = EINVAL; goto out; } pi = &sc->port[i]; pi->adapter = sc; pi->nqsets = port_qsets; pi->first_qset = i*port_qsets; pi->port_id = i; pi->tx_chan = i >= ai->nports0; pi->txpkt_intf = pi->tx_chan ? 2 * (i - ai->nports0) + 1 : 2 * i; sc->rxpkt_map[pi->txpkt_intf] = i; sc->port[i].tx_chan = i >= ai->nports0; sc->portdev[i] = child; device_set_softc(child, pi); } if ((error = bus_generic_attach(dev)) != 0) goto out; /* initialize sge private state */ t3_sge_init_adapter(sc); t3_led_ready(sc); cxgb_offload_init(); if (is_offload(sc)) { setbit(&sc->registered_device_map, OFFLOAD_DEVMAP_BIT); cxgb_adapter_ofld(sc); } error = t3_get_fw_version(sc, &vers); if (error) goto out; snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d", G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers)); snprintf(buf, sizeof(buf), "%s %sNIC\t E/C: %s S/N: %s", ai->desc, is_offload(sc) ? "R" : "", sc->params.vpd.ec, sc->params.vpd.sn); device_set_desc_copy(dev, buf); snprintf(&sc->port_types[0], sizeof(sc->port_types), "%x%x%x%x", sc->params.vpd.port_type[0], sc->params.vpd.port_type[1], sc->params.vpd.port_type[2], sc->params.vpd.port_type[3]); device_printf(sc->dev, "Firmware Version %s\n", &sc->fw_version[0]); callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc); t3_add_attach_sysctls(sc); out: if (error) cxgb_free(sc); return (error); } /* * The cxgb_controller_detach routine is called with the device is * unloaded from the system. */ static int cxgb_controller_detach(device_t dev) { struct adapter *sc; sc = device_get_softc(dev); cxgb_free(sc); return (0); } /* * The cxgb_free() is called by the cxgb_controller_detach() routine * to tear down the structures that were built up in * cxgb_controller_attach(), and should be the final piece of work * done when fully unloading the driver. * * * 1. Shutting down the threads started by the cxgb_controller_attach() * routine. * 2. Stopping the lower level device and all callouts (cxgb_down_locked()). * 3. Detaching all of the port devices created during the * cxgb_controller_attach() routine. * 4. Removing the device children created via cxgb_controller_attach(). * 5. Releasing PCI resources associated with the device. * 6. Turning off the offload support, iff it was turned on. * 7. Destroying the mutexes created in cxgb_controller_attach(). * */ static void cxgb_free(struct adapter *sc) { int i, nqsets = 0; ADAPTER_LOCK(sc); sc->flags |= CXGB_SHUTDOWN; ADAPTER_UNLOCK(sc); cxgb_pcpu_shutdown_threads(sc); t3_sge_deinit_sw(sc); /* * Make sure all child devices are gone. */ bus_generic_detach(sc->dev); for (i = 0; i < (sc)->params.nports; i++) { if (sc->portdev[i] && device_delete_child(sc->dev, sc->portdev[i]) != 0) device_printf(sc->dev, "failed to delete child port\n"); nqsets += sc->port[i].nqsets; } /* * At this point, it is as if cxgb_port_detach has run on all ports, and * cxgb_down has run on the adapter. All interrupts have been silenced, * all open devices have been closed. */ KASSERT(sc->open_device_map == 0, ("%s: device(s) still open (%x)", __func__, sc->open_device_map)); for (i = 0; i < sc->params.nports; i++) { KASSERT(sc->port[i].ifp == NULL, ("%s: port %i undead!", __func__, i)); } /* * Finish off the adapter's callouts. */ callout_drain(&sc->cxgb_tick_ch); callout_drain(&sc->sge_timer_ch); /* * Release resources grabbed under FULL_INIT_DONE by cxgb_up. The * sysctls are cleaned up by the kernel linker. */ if (sc->flags & FULL_INIT_DONE) { t3_free_sge_resources(sc, nqsets); sc->flags &= ~FULL_INIT_DONE; } /* * Release all interrupt resources. */ cxgb_teardown_interrupts(sc); #ifdef MSI_SUPPORTED if (sc->flags & (USING_MSI | USING_MSIX)) { device_printf(sc->dev, "releasing msi message(s)\n"); pci_release_msi(sc->dev); } else { device_printf(sc->dev, "no msi message to release\n"); } if (sc->msix_regs_res != NULL) { bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->msix_regs_rid, sc->msix_regs_res); } #endif /* * Free the adapter's taskqueue. */ if (sc->tq != NULL) { taskqueue_free(sc->tq); sc->tq = NULL; } if (is_offload(sc)) { clrbit(&sc->registered_device_map, OFFLOAD_DEVMAP_BIT); cxgb_adapter_unofld(sc); } #ifdef notyet if (sc->flags & CXGB_OFLD_INIT) cxgb_offload_deactivate(sc); #endif free(sc->filters, M_DEVBUF); t3_sge_free(sc); cxgb_offload_exit(); if (sc->udbs_res != NULL) bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->udbs_rid, sc->udbs_res); if (sc->regs_res != NULL) bus_release_resource(sc->dev, SYS_RES_MEMORY, sc->regs_rid, sc->regs_res); MTX_DESTROY(&sc->mdio_lock); MTX_DESTROY(&sc->sge.reg_lock); MTX_DESTROY(&sc->elmer_lock); ADAPTER_LOCK_DEINIT(sc); } /** * setup_sge_qsets - configure SGE Tx/Rx/response queues * @sc: the controller softc * * Determines how many sets of SGE queues to use and initializes them. * We support multiple queue sets per port if we have MSI-X, otherwise * just one queue set per port. */ static int setup_sge_qsets(adapter_t *sc) { int i, j, err, irq_idx = 0, qset_idx = 0; u_int ntxq = SGE_TXQ_PER_SET; if ((err = t3_sge_alloc(sc)) != 0) { device_printf(sc->dev, "t3_sge_alloc returned %d\n", err); return (err); } if (sc->params.rev > 0 && !(sc->flags & USING_MSI)) irq_idx = -1; for (i = 0; i < (sc)->params.nports; i++) { struct port_info *pi = &sc->port[i]; for (j = 0; j < pi->nqsets; j++, qset_idx++) { err = t3_sge_alloc_qset(sc, qset_idx, (sc)->params.nports, (sc->flags & USING_MSIX) ? qset_idx + 1 : irq_idx, &sc->params.sge.qset[qset_idx], ntxq, pi); if (err) { t3_free_sge_resources(sc, qset_idx); device_printf(sc->dev, "t3_sge_alloc_qset failed with %d\n", err); return (err); } } } return (0); } static void cxgb_teardown_interrupts(adapter_t *sc) { int i; for (i = 0; i < SGE_QSETS; i++) { if (sc->msix_intr_tag[i] == NULL) { /* Should have been setup fully or not at all */ KASSERT(sc->msix_irq_res[i] == NULL && sc->msix_irq_rid[i] == 0, ("%s: half-done interrupt (%d).", __func__, i)); continue; } bus_teardown_intr(sc->dev, sc->msix_irq_res[i], sc->msix_intr_tag[i]); bus_release_resource(sc->dev, SYS_RES_IRQ, sc->msix_irq_rid[i], sc->msix_irq_res[i]); sc->msix_irq_res[i] = sc->msix_intr_tag[i] = NULL; sc->msix_irq_rid[i] = 0; } if (sc->intr_tag) { KASSERT(sc->irq_res != NULL, ("%s: half-done interrupt.", __func__)); bus_teardown_intr(sc->dev, sc->irq_res, sc->intr_tag); bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid, sc->irq_res); sc->irq_res = sc->intr_tag = NULL; sc->irq_rid = 0; } } static int cxgb_setup_interrupts(adapter_t *sc) { struct resource *res; void *tag; int i, rid, err, intr_flag = sc->flags & (USING_MSI | USING_MSIX); sc->irq_rid = intr_flag ? 1 : 0; sc->irq_res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &sc->irq_rid, RF_SHAREABLE | RF_ACTIVE); if (sc->irq_res == NULL) { device_printf(sc->dev, "Cannot allocate interrupt (%x, %u)\n", intr_flag, sc->irq_rid); err = EINVAL; sc->irq_rid = 0; } else { err = bus_setup_intr(sc->dev, sc->irq_res, INTR_MPSAFE | INTR_TYPE_NET, #ifdef INTR_FILTERS NULL, #endif sc->cxgb_intr, sc, &sc->intr_tag); if (err) { device_printf(sc->dev, "Cannot set up interrupt (%x, %u, %d)\n", intr_flag, sc->irq_rid, err); bus_release_resource(sc->dev, SYS_RES_IRQ, sc->irq_rid, sc->irq_res); sc->irq_res = sc->intr_tag = NULL; sc->irq_rid = 0; } } /* That's all for INTx or MSI */ if (!(intr_flag & USING_MSIX) || err) return (err); for (i = 0; i < sc->msi_count - 1; i++) { rid = i + 2; res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE); if (res == NULL) { device_printf(sc->dev, "Cannot allocate interrupt " "for message %d\n", rid); err = EINVAL; break; } err = bus_setup_intr(sc->dev, res, INTR_MPSAFE | INTR_TYPE_NET, #ifdef INTR_FILTERS NULL, #endif t3_intr_msix, &sc->sge.qs[i], &tag); if (err) { device_printf(sc->dev, "Cannot set up interrupt " "for message %d (%d)\n", rid, err); bus_release_resource(sc->dev, SYS_RES_IRQ, rid, res); break; } sc->msix_irq_rid[i] = rid; sc->msix_irq_res[i] = res; sc->msix_intr_tag[i] = tag; } if (err) cxgb_teardown_interrupts(sc); return (err); } static int cxgb_port_probe(device_t dev) { struct port_info *p; char buf[80]; const char *desc; p = device_get_softc(dev); desc = p->phy.desc; snprintf(buf, sizeof(buf), "Port %d %s", p->port_id, desc); device_set_desc_copy(dev, buf); return (0); } static int cxgb_makedev(struct port_info *pi) { pi->port_cdev = make_dev(&cxgb_cdevsw, pi->ifp->if_dunit, UID_ROOT, GID_WHEEL, 0600, "%s", if_name(pi->ifp)); if (pi->port_cdev == NULL) return (ENOMEM); pi->port_cdev->si_drv1 = (void *)pi; return (0); } #ifdef TSO_SUPPORTED #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | IFCAP_VLAN_HWTSO) /* Don't enable TSO6 yet */ #define CXGB_CAP_ENABLE (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM | IFCAP_TSO4 | IFCAP_JUMBO_MTU | IFCAP_LRO | IFCAP_VLAN_HWTSO) #else #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_JUMBO_MTU) /* Don't enable TSO6 yet */ #define CXGB_CAP_ENABLE (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_JUMBO_MTU) #define IFCAP_TSO4 0x0 #define IFCAP_TSO6 0x0 #define CSUM_TSO 0x0 #endif static int cxgb_port_attach(device_t dev) { struct port_info *p; struct ifnet *ifp; int err; struct adapter *sc; p = device_get_softc(dev); sc = p->adapter; snprintf(p->lockbuf, PORT_NAME_LEN, "cxgb port lock %d:%d", device_get_unit(device_get_parent(dev)), p->port_id); PORT_LOCK_INIT(p, p->lockbuf); callout_init(&p->link_check_ch, CALLOUT_MPSAFE); TASK_INIT(&p->link_check_task, 0, check_link_status, p); /* Allocate an ifnet object and set it up */ ifp = p->ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "Cannot allocate ifnet\n"); return (ENOMEM); } /* * Note that there is currently no watchdog timer. */ if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_init = cxgb_init; ifp->if_softc = p; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = cxgb_ioctl; ifp->if_start = cxgb_start; ifp->if_timer = 0; /* Disable ifnet watchdog */ ifp->if_watchdog = NULL; ifp->if_snd.ifq_drv_maxlen = max(cxgb_snd_queue_len, ifqmaxlen); IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen); IFQ_SET_READY(&ifp->if_snd); ifp->if_hwassist = ifp->if_capabilities = ifp->if_capenable = 0; ifp->if_capabilities |= CXGB_CAP; ifp->if_capenable |= CXGB_CAP_ENABLE; ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO); /* * disable TSO on 4-port - it isn't supported by the firmware yet */ if (sc->params.nports > 2) { ifp->if_capabilities &= ~(IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTSO); ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTSO); ifp->if_hwassist &= ~CSUM_TSO; } ether_ifattach(ifp, p->hw_addr); #ifdef IFNET_MULTIQUEUE ifp->if_transmit = cxgb_pcpu_transmit; #endif /* * Only default to jumbo frames on 10GigE */ if (p->adapter->params.nports <= 2) ifp->if_mtu = ETHERMTU_JUMBO; if ((err = cxgb_makedev(p)) != 0) { printf("makedev failed %d\n", err); return (err); } /* Create a list of media supported by this port */ ifmedia_init(&p->media, IFM_IMASK, cxgb_media_change, cxgb_media_status); cxgb_build_medialist(p); t3_sge_init_port(p); return (err); } /* * cxgb_port_detach() is called via the device_detach methods when * cxgb_free() calls the bus_generic_detach. It is responsible for * removing the device from the view of the kernel, i.e. from all * interfaces lists etc. This routine is only called when the driver is * being unloaded, not when the link goes down. */ static int cxgb_port_detach(device_t dev) { struct port_info *p; struct adapter *sc; p = device_get_softc(dev); sc = p->adapter; /* Tell cxgb_ioctl and if_init that the port is going away */ ADAPTER_LOCK(sc); SET_DOOMED(p); wakeup(&sc->flags); while (IS_BUSY(sc)) mtx_sleep(&sc->flags, &sc->lock, 0, "cxgbdtch", 0); SET_BUSY(sc); ADAPTER_UNLOCK(sc); if (p->port_cdev != NULL) destroy_dev(p->port_cdev); cxgb_uninit_synchronized(p); ether_ifdetach(p->ifp); PORT_LOCK_DEINIT(p); if_free(p->ifp); p->ifp = NULL; ADAPTER_LOCK(sc); CLR_BUSY(sc); wakeup_one(&sc->flags); ADAPTER_UNLOCK(sc); return (0); } void t3_fatal_err(struct adapter *sc) { u_int fw_status[4]; if (sc->flags & FULL_INIT_DONE) { t3_sge_stop(sc); t3_write_reg(sc, A_XGM_TX_CTRL, 0); t3_write_reg(sc, A_XGM_RX_CTRL, 0); t3_write_reg(sc, XGM_REG(A_XGM_TX_CTRL, 1), 0); t3_write_reg(sc, XGM_REG(A_XGM_RX_CTRL, 1), 0); t3_intr_disable(sc); } device_printf(sc->dev,"encountered fatal error, operation suspended\n"); if (!t3_cim_ctl_blk_read(sc, 0xa0, 4, fw_status)) device_printf(sc->dev, "FW_ status: 0x%x, 0x%x, 0x%x, 0x%x\n", fw_status[0], fw_status[1], fw_status[2], fw_status[3]); } int t3_os_find_pci_capability(adapter_t *sc, int cap) { device_t dev; struct pci_devinfo *dinfo; pcicfgregs *cfg; uint32_t status; uint8_t ptr; dev = sc->dev; dinfo = device_get_ivars(dev); cfg = &dinfo->cfg; status = pci_read_config(dev, PCIR_STATUS, 2); if (!(status & PCIM_STATUS_CAPPRESENT)) return (0); switch (cfg->hdrtype & PCIM_HDRTYPE) { case 0: case 1: ptr = PCIR_CAP_PTR; break; case 2: ptr = PCIR_CAP_PTR_2; break; default: return (0); break; } ptr = pci_read_config(dev, ptr, 1); while (ptr != 0) { if (pci_read_config(dev, ptr + PCICAP_ID, 1) == cap) return (ptr); ptr = pci_read_config(dev, ptr + PCICAP_NEXTPTR, 1); } return (0); } int t3_os_pci_save_state(struct adapter *sc) { device_t dev; struct pci_devinfo *dinfo; dev = sc->dev; dinfo = device_get_ivars(dev); pci_cfg_save(dev, dinfo, 0); return (0); } int t3_os_pci_restore_state(struct adapter *sc) { device_t dev; struct pci_devinfo *dinfo; dev = sc->dev; dinfo = device_get_ivars(dev); pci_cfg_restore(dev, dinfo); return (0); } /** * t3_os_link_changed - handle link status changes * @sc: the adapter associated with the link change * @port_id: the port index whose link status has changed * @link_status: the new status of the link * @speed: the new speed setting * @duplex: the new duplex setting * @fc: the new flow-control setting * * This is the OS-dependent handler for link status changes. The OS * neutral handler takes care of most of the processing for these events, * then calls this handler for any OS-specific processing. */ void t3_os_link_changed(adapter_t *adapter, int port_id, int link_status, int speed, int duplex, int fc, int mac_was_reset) { struct port_info *pi = &adapter->port[port_id]; struct ifnet *ifp = pi->ifp; /* no race with detach, so ifp should always be good */ KASSERT(ifp, ("%s: if detached.", __func__)); /* Reapply mac settings if they were lost due to a reset */ if (mac_was_reset) { PORT_LOCK(pi); cxgb_update_mac_settings(pi); PORT_UNLOCK(pi); } if (link_status) { ifp->if_baudrate = IF_Mbps(speed); if_link_state_change(ifp, LINK_STATE_UP); } else if_link_state_change(ifp, LINK_STATE_DOWN); } /** * t3_os_phymod_changed - handle PHY module changes * @phy: the PHY reporting the module change * @mod_type: new module type * * This is the OS-dependent handler for PHY module changes. It is * invoked when a PHY module is removed or inserted for any OS-specific * processing. */ void t3_os_phymod_changed(struct adapter *adap, int port_id) { static const char *mod_str[] = { NULL, "SR", "LR", "LRM", "TWINAX", "TWINAX-L", "unknown" }; struct port_info *pi = &adap->port[port_id]; int mod = pi->phy.modtype; if (mod != pi->media.ifm_cur->ifm_data) cxgb_build_medialist(pi); if (mod == phy_modtype_none) if_printf(pi->ifp, "PHY module unplugged\n"); else { KASSERT(mod < ARRAY_SIZE(mod_str), ("invalid PHY module type %d", mod)); if_printf(pi->ifp, "%s PHY module inserted\n", mod_str[mod]); } } void t3_os_set_hw_addr(adapter_t *adapter, int port_idx, u8 hw_addr[]) { /* * The ifnet might not be allocated before this gets called, * as this is called early on in attach by t3_prep_adapter * save the address off in the port structure */ if (cxgb_debug) printf("set_hw_addr on idx %d addr %6D\n", port_idx, hw_addr, ":"); bcopy(hw_addr, adapter->port[port_idx].hw_addr, ETHER_ADDR_LEN); } /* * Programs the XGMAC based on the settings in the ifnet. These settings * include MTU, MAC address, mcast addresses, etc. */ static void cxgb_update_mac_settings(struct port_info *p) { struct ifnet *ifp = p->ifp; struct t3_rx_mode rm; struct cmac *mac = &p->mac; int mtu, hwtagging; PORT_LOCK_ASSERT_OWNED(p); bcopy(IF_LLADDR(ifp), p->hw_addr, ETHER_ADDR_LEN); mtu = ifp->if_mtu; if (ifp->if_capenable & IFCAP_VLAN_MTU) mtu += ETHER_VLAN_ENCAP_LEN; hwtagging = (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0; t3_mac_set_mtu(mac, mtu); t3_set_vlan_accel(p->adapter, 1 << p->tx_chan, hwtagging); t3_mac_set_address(mac, 0, p->hw_addr); t3_init_rx_mode(&rm, p); t3_mac_set_rx_mode(mac, &rm); } static int await_mgmt_replies(struct adapter *adap, unsigned long init_cnt, unsigned long n) { int attempts = 5; while (adap->sge.qs[0].rspq.offload_pkts < init_cnt + n) { if (!--attempts) return (ETIMEDOUT); t3_os_sleep(10); } return 0; } static int init_tp_parity(struct adapter *adap) { int i; struct mbuf *m; struct cpl_set_tcb_field *greq; unsigned long cnt = adap->sge.qs[0].rspq.offload_pkts; t3_tp_set_offload_mode(adap, 1); for (i = 0; i < 16; i++) { struct cpl_smt_write_req *req; m = m_gethdr(M_WAITOK, MT_DATA); req = mtod(m, struct cpl_smt_write_req *); m->m_len = m->m_pkthdr.len = sizeof(*req); memset(req, 0, sizeof(*req)); req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, i)); req->iff = i; t3_mgmt_tx(adap, m); } for (i = 0; i < 2048; i++) { struct cpl_l2t_write_req *req; m = m_gethdr(M_WAITOK, MT_DATA); req = mtod(m, struct cpl_l2t_write_req *); m->m_len = m->m_pkthdr.len = sizeof(*req); memset(req, 0, sizeof(*req)); req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_L2T_WRITE_REQ, i)); req->params = htonl(V_L2T_W_IDX(i)); t3_mgmt_tx(adap, m); } for (i = 0; i < 2048; i++) { struct cpl_rte_write_req *req; m = m_gethdr(M_WAITOK, MT_DATA); req = mtod(m, struct cpl_rte_write_req *); m->m_len = m->m_pkthdr.len = sizeof(*req); memset(req, 0, sizeof(*req)); req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RTE_WRITE_REQ, i)); req->l2t_idx = htonl(V_L2T_W_IDX(i)); t3_mgmt_tx(adap, m); } m = m_gethdr(M_WAITOK, MT_DATA); greq = mtod(m, struct cpl_set_tcb_field *); m->m_len = m->m_pkthdr.len = sizeof(*greq); memset(greq, 0, sizeof(*greq)); greq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); OPCODE_TID(greq) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, 0)); greq->mask = htobe64(1); t3_mgmt_tx(adap, m); i = await_mgmt_replies(adap, cnt, 16 + 2048 + 2048 + 1); t3_tp_set_offload_mode(adap, 0); return (i); } /** * setup_rss - configure Receive Side Steering (per-queue connection demux) * @adap: the adapter * * Sets up RSS to distribute packets to multiple receive queues. We * configure the RSS CPU lookup table to distribute to the number of HW * receive queues, and the response queue lookup table to narrow that * down to the response queues actually configured for each port. * We always configure the RSS mapping for two ports since the mapping * table has plenty of entries. */ static void setup_rss(adapter_t *adap) { int i; u_int nq[2]; uint8_t cpus[SGE_QSETS + 1]; uint16_t rspq_map[RSS_TABLE_SIZE]; for (i = 0; i < SGE_QSETS; ++i) cpus[i] = i; cpus[SGE_QSETS] = 0xff; nq[0] = nq[1] = 0; for_each_port(adap, i) { const struct port_info *pi = adap2pinfo(adap, i); nq[pi->tx_chan] += pi->nqsets; } for (i = 0; i < RSS_TABLE_SIZE / 2; ++i) { rspq_map[i] = nq[0] ? i % nq[0] : 0; rspq_map[i + RSS_TABLE_SIZE / 2] = nq[1] ? i % nq[1] + nq[0] : 0; } /* Calculate the reverse RSS map table */ for (i = 0; i < SGE_QSETS; ++i) adap->rrss_map[i] = 0xff; for (i = 0; i < RSS_TABLE_SIZE; ++i) if (adap->rrss_map[rspq_map[i]] == 0xff) adap->rrss_map[rspq_map[i]] = i; t3_config_rss(adap, F_RQFEEDBACKENABLE | F_TNLLKPEN | F_TNLMAPEN | F_TNLPRTEN | F_TNL2TUPEN | F_TNL4TUPEN | F_OFDMAPEN | F_RRCPLMAPEN | V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map); } /* * Sends an mbuf to an offload queue driver * after dealing with any active network taps. */ static inline int offload_tx(struct t3cdev *tdev, struct mbuf *m) { int ret; ret = t3_offload_tx(tdev, m); return (ret); } static int write_smt_entry(struct adapter *adapter, int idx) { struct port_info *pi = &adapter->port[idx]; struct cpl_smt_write_req *req; struct mbuf *m; if ((m = m_gethdr(M_NOWAIT, MT_DATA)) == NULL) return (ENOMEM); req = mtod(m, struct cpl_smt_write_req *); m->m_pkthdr.len = m->m_len = sizeof(struct cpl_smt_write_req); req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SMT_WRITE_REQ, idx)); req->mtu_idx = NMTUS - 1; /* should be 0 but there's a T3 bug */ req->iff = idx; memset(req->src_mac1, 0, sizeof(req->src_mac1)); memcpy(req->src_mac0, pi->hw_addr, ETHER_ADDR_LEN); m_set_priority(m, 1); offload_tx(&adapter->tdev, m); return (0); } static int init_smt(struct adapter *adapter) { int i; for_each_port(adapter, i) write_smt_entry(adapter, i); return 0; } static void init_port_mtus(adapter_t *adapter) { unsigned int mtus = ETHERMTU | (ETHERMTU << 16); t3_write_reg(adapter, A_TP_MTU_PORT_TABLE, mtus); } static void send_pktsched_cmd(struct adapter *adap, int sched, int qidx, int lo, int hi, int port) { struct mbuf *m; struct mngt_pktsched_wr *req; m = m_gethdr(M_DONTWAIT, MT_DATA); if (m) { req = mtod(m, struct mngt_pktsched_wr *); req->wr_hi = htonl(V_WR_OP(FW_WROPCODE_MNGT)); req->mngt_opcode = FW_MNGTOPCODE_PKTSCHED_SET; req->sched = sched; req->idx = qidx; req->min = lo; req->max = hi; req->binding = port; m->m_len = m->m_pkthdr.len = sizeof(*req); t3_mgmt_tx(adap, m); } } static void bind_qsets(adapter_t *sc) { int i, j; cxgb_pcpu_startup_threads(sc); for (i = 0; i < (sc)->params.nports; ++i) { const struct port_info *pi = adap2pinfo(sc, i); for (j = 0; j < pi->nqsets; ++j) { send_pktsched_cmd(sc, 1, pi->first_qset + j, -1, -1, pi->tx_chan); } } } static void update_tpeeprom(struct adapter *adap) { #ifdef FIRMWARE_LATEST const struct firmware *tpeeprom; #else struct firmware *tpeeprom; #endif uint32_t version; unsigned int major, minor; int ret, len; char rev, name[32]; t3_seeprom_read(adap, TP_SRAM_OFFSET, &version); major = G_TP_VERSION_MAJOR(version); minor = G_TP_VERSION_MINOR(version); if (major == TP_VERSION_MAJOR && minor == TP_VERSION_MINOR) return; rev = t3rev2char(adap); snprintf(name, sizeof(name), TPEEPROM_NAME, rev); tpeeprom = firmware_get(name); if (tpeeprom == NULL) { device_printf(adap->dev, "could not load TP EEPROM: unable to load %s\n", name); return; } len = tpeeprom->datasize - 4; ret = t3_check_tpsram(adap, tpeeprom->data, tpeeprom->datasize); if (ret) goto release_tpeeprom; if (len != TP_SRAM_LEN) { device_printf(adap->dev, "%s length is wrong len=%d expected=%d\n", name, len, TP_SRAM_LEN); return; } ret = set_eeprom(&adap->port[0], tpeeprom->data, tpeeprom->datasize, TP_SRAM_OFFSET); if (!ret) { device_printf(adap->dev, "Protocol SRAM image updated in EEPROM to %d.%d.%d\n", TP_VERSION_MAJOR, TP_VERSION_MINOR, TP_VERSION_MICRO); } else device_printf(adap->dev, "Protocol SRAM image update in EEPROM failed\n"); release_tpeeprom: firmware_put(tpeeprom, FIRMWARE_UNLOAD); return; } static int update_tpsram(struct adapter *adap) { #ifdef FIRMWARE_LATEST const struct firmware *tpsram; #else struct firmware *tpsram; #endif int ret; char rev, name[32]; rev = t3rev2char(adap); snprintf(name, sizeof(name), TPSRAM_NAME, rev); update_tpeeprom(adap); tpsram = firmware_get(name); if (tpsram == NULL){ device_printf(adap->dev, "could not load TP SRAM\n"); return (EINVAL); } else device_printf(adap->dev, "updating TP SRAM\n"); ret = t3_check_tpsram(adap, tpsram->data, tpsram->datasize); if (ret) goto release_tpsram; ret = t3_set_proto_sram(adap, tpsram->data); if (ret) device_printf(adap->dev, "loading protocol SRAM failed\n"); release_tpsram: firmware_put(tpsram, FIRMWARE_UNLOAD); return ret; } /** * cxgb_up - enable the adapter * @adap: adapter being enabled * * Called when the first port is enabled, this function performs the * actions necessary to make an adapter operational, such as completing * the initialization of HW modules, and enabling interrupts. */ static int cxgb_up(struct adapter *sc) { int err = 0; unsigned int mxf = t3_mc5_size(&sc->mc5) - MC5_MIN_TIDS; KASSERT(sc->open_device_map == 0, ("%s: device(s) already open (%x)", __func__, sc->open_device_map)); if ((sc->flags & FULL_INIT_DONE) == 0) { ADAPTER_LOCK_ASSERT_NOTOWNED(sc); if ((sc->flags & FW_UPTODATE) == 0) if ((err = upgrade_fw(sc))) goto out; if ((sc->flags & TPS_UPTODATE) == 0) if ((err = update_tpsram(sc))) goto out; if (is_offload(sc) && nfilters != 0) { sc->params.mc5.nservers = 0; if (nfilters < 0) sc->params.mc5.nfilters = mxf; else sc->params.mc5.nfilters = min(nfilters, mxf); } err = t3_init_hw(sc, 0); if (err) goto out; t3_set_reg_field(sc, A_TP_PARA_REG5, 0, F_RXDDPOFFINIT); t3_write_reg(sc, A_ULPRX_TDDP_PSZ, V_HPZ0(PAGE_SHIFT - 12)); err = setup_sge_qsets(sc); if (err) goto out; alloc_filters(sc); setup_rss(sc); t3_intr_clear(sc); err = cxgb_setup_interrupts(sc); if (err) goto out; t3_add_configured_sysctls(sc); sc->flags |= FULL_INIT_DONE; } t3_intr_clear(sc); t3_sge_start(sc); t3_intr_enable(sc); if (sc->params.rev >= T3_REV_C && !(sc->flags & TP_PARITY_INIT) && is_offload(sc) && init_tp_parity(sc) == 0) sc->flags |= TP_PARITY_INIT; if (sc->flags & TP_PARITY_INIT) { t3_write_reg(sc, A_TP_INT_CAUSE, F_CMCACHEPERR | F_ARPLUTPERR); t3_write_reg(sc, A_TP_INT_ENABLE, 0x7fbfffff); } if (!(sc->flags & QUEUES_BOUND)) { bind_qsets(sc); setup_hw_filters(sc); sc->flags |= QUEUES_BOUND; } t3_sge_reset_adapter(sc); out: return (err); } /* * Called when the last open device is closed. Does NOT undo all of cxgb_up's * work. Specifically, the resources grabbed under FULL_INIT_DONE are released * during controller_detach, not here. */ static void cxgb_down(struct adapter *sc) { t3_sge_stop(sc); t3_intr_disable(sc); } static int offload_open(struct port_info *pi) { struct adapter *sc = pi->adapter; struct t3cdev *tdev = &sc->tdev; setbit(&sc->open_device_map, OFFLOAD_DEVMAP_BIT); t3_tp_set_offload_mode(sc, 1); tdev->lldev = pi->ifp; init_port_mtus(sc); t3_load_mtus(sc, sc->params.mtus, sc->params.a_wnd, sc->params.b_wnd, sc->params.rev == 0 ? sc->port[0].ifp->if_mtu : 0xffff); init_smt(sc); cxgb_add_clients(tdev); return (0); } static int offload_close(struct t3cdev *tdev) { struct adapter *adapter = tdev2adap(tdev); if (!isset(&adapter->open_device_map, OFFLOAD_DEVMAP_BIT)) return (0); /* Call back all registered clients */ cxgb_remove_clients(tdev); tdev->lldev = NULL; cxgb_set_dummy_ops(tdev); t3_tp_set_offload_mode(adapter, 0); clrbit(&adapter->open_device_map, OFFLOAD_DEVMAP_BIT); return (0); } /* * if_init for cxgb ports. */ static void cxgb_init(void *arg) { struct port_info *p = arg; struct adapter *sc = p->adapter; ADAPTER_LOCK(sc); cxgb_init_locked(p); /* releases adapter lock */ ADAPTER_LOCK_ASSERT_NOTOWNED(sc); } static int cxgb_init_locked(struct port_info *p) { struct adapter *sc = p->adapter; struct ifnet *ifp = p->ifp; struct cmac *mac = &p->mac; int rc = 0, may_sleep = 0, gave_up_lock = 0; ADAPTER_LOCK_ASSERT_OWNED(sc); while (!IS_DOOMED(p) && IS_BUSY(sc)) { gave_up_lock = 1; if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbinit", 0)) { rc = EINTR; goto done; } } if (IS_DOOMED(p)) { rc = ENXIO; goto done; } KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__)); /* * The code that runs during one-time adapter initialization can sleep * so it's important not to hold any locks across it. */ may_sleep = sc->flags & FULL_INIT_DONE ? 0 : 1; if (may_sleep) { SET_BUSY(sc); gave_up_lock = 1; ADAPTER_UNLOCK(sc); } if (sc->open_device_map == 0) { if ((rc = cxgb_up(sc)) != 0) goto done; if (is_offload(sc) && !ofld_disable && offload_open(p)) log(LOG_WARNING, "Could not initialize offload capabilities\n"); } PORT_LOCK(p); if (isset(&sc->open_device_map, p->port_id) && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { PORT_UNLOCK(p); goto done; } t3_port_intr_enable(sc, p->port_id); if (!mac->multiport) t3_mac_init(mac); cxgb_update_mac_settings(p); t3_link_start(&p->phy, mac, &p->link_config); t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; PORT_UNLOCK(p); /* all ok */ setbit(&sc->open_device_map, p->port_id); callout_reset(&p->link_check_ch, p->phy.caps & SUPPORTED_LINK_IRQ ? hz * 3 : hz / 4, link_check_callout, p); done: if (may_sleep) { ADAPTER_LOCK(sc); KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__)); CLR_BUSY(sc); } if (gave_up_lock) wakeup_one(&sc->flags); ADAPTER_UNLOCK(sc); return (rc); } static int cxgb_uninit_locked(struct port_info *p) { struct adapter *sc = p->adapter; int rc; ADAPTER_LOCK_ASSERT_OWNED(sc); while (!IS_DOOMED(p) && IS_BUSY(sc)) { if (mtx_sleep(&sc->flags, &sc->lock, PCATCH, "cxgbunin", 0)) { rc = EINTR; goto done; } } if (IS_DOOMED(p)) { rc = ENXIO; goto done; } KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__)); SET_BUSY(sc); ADAPTER_UNLOCK(sc); rc = cxgb_uninit_synchronized(p); ADAPTER_LOCK(sc); KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__)); CLR_BUSY(sc); wakeup_one(&sc->flags); done: ADAPTER_UNLOCK(sc); return (rc); } /* * Called on "ifconfig down", and from port_detach */ static int cxgb_uninit_synchronized(struct port_info *pi) { struct adapter *sc = pi->adapter; struct ifnet *ifp = pi->ifp; /* * taskqueue_drain may cause a deadlock if the adapter lock is held. */ ADAPTER_LOCK_ASSERT_NOTOWNED(sc); /* * Clear this port's bit from the open device map, and then drain all * the tasks that can access/manipulate this port's port_info or ifp. * We disable this port's interrupts here and so the slow/ext * interrupt tasks won't be enqueued. The tick task will continue to * be enqueued every second but the runs after this drain will not see * this port in the open device map. * * A well behaved task must take open_device_map into account and ignore * ports that are not open. */ clrbit(&sc->open_device_map, pi->port_id); t3_port_intr_disable(sc, pi->port_id); taskqueue_drain(sc->tq, &sc->slow_intr_task); taskqueue_drain(sc->tq, &sc->tick_task); callout_drain(&pi->link_check_ch); taskqueue_drain(sc->tq, &pi->link_check_task); PORT_LOCK(pi); ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); /* disable pause frames */ t3_set_reg_field(sc, A_XGM_TX_CFG + pi->mac.offset, F_TXPAUSEEN, 0); /* Reset RX FIFO HWM */ t3_set_reg_field(sc, A_XGM_RXFIFO_CFG + pi->mac.offset, V_RXFIFOPAUSEHWM(M_RXFIFOPAUSEHWM), 0); DELAY(100 * 1000); /* Wait for TXFIFO empty */ t3_wait_op_done(sc, A_XGM_TXFIFO_CFG + pi->mac.offset, F_TXFIFO_EMPTY, 1, 20, 5); DELAY(100 * 1000); t3_mac_disable(&pi->mac, MAC_DIRECTION_RX); pi->phy.ops->power_down(&pi->phy, 1); PORT_UNLOCK(pi); pi->link_config.link_ok = 0; t3_os_link_changed(sc, pi->port_id, 0, 0, 0, 0, 0); if ((sc->open_device_map & PORT_MASK) == 0) offload_close(&sc->tdev); if (sc->open_device_map == 0) cxgb_down(pi->adapter); return (0); } /* * Mark lro enabled or disabled in all qsets for this port */ static int cxgb_set_lro(struct port_info *p, int enabled) { int i; struct adapter *adp = p->adapter; struct sge_qset *q; for (i = 0; i < p->nqsets; i++) { q = &adp->sge.qs[p->first_qset + i]; q->lro.enabled = (enabled != 0); } return (0); } static int cxgb_ioctl(struct ifnet *ifp, unsigned long command, caddr_t data) { struct port_info *p = ifp->if_softc; struct adapter *sc = p->adapter; struct ifreq *ifr = (struct ifreq *)data; int flags, error = 0, mtu; uint32_t mask; switch (command) { case SIOCSIFMTU: ADAPTER_LOCK(sc); error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0); if (error) { fail: ADAPTER_UNLOCK(sc); return (error); } mtu = ifr->ifr_mtu; if ((mtu < ETHERMIN) || (mtu > ETHERMTU_JUMBO)) { error = EINVAL; } else { ifp->if_mtu = mtu; PORT_LOCK(p); cxgb_update_mac_settings(p); PORT_UNLOCK(p); } ADAPTER_UNLOCK(sc); break; case SIOCSIFFLAGS: ADAPTER_LOCK(sc); if (IS_DOOMED(p)) { error = ENXIO; goto fail; } if (ifp->if_flags & IFF_UP) { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { flags = p->if_flags; if (((ifp->if_flags ^ flags) & IFF_PROMISC) || ((ifp->if_flags ^ flags) & IFF_ALLMULTI)) { if (IS_BUSY(sc)) { error = EBUSY; goto fail; } PORT_LOCK(p); cxgb_update_mac_settings(p); PORT_UNLOCK(p); } ADAPTER_UNLOCK(sc); } else error = cxgb_init_locked(p); p->if_flags = ifp->if_flags; } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) error = cxgb_uninit_locked(p); else ADAPTER_UNLOCK(sc); ADAPTER_LOCK_ASSERT_NOTOWNED(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: ADAPTER_LOCK(sc); error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0); if (error) goto fail; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { PORT_LOCK(p); cxgb_update_mac_settings(p); PORT_UNLOCK(p); } ADAPTER_UNLOCK(sc); break; case SIOCSIFCAP: ADAPTER_LOCK(sc); error = IS_DOOMED(p) ? ENXIO : (IS_BUSY(sc) ? EBUSY : 0); if (error) goto fail; mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (mask & IFCAP_TXCSUM) { ifp->if_capenable ^= IFCAP_TXCSUM; ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP); if (IFCAP_TSO & ifp->if_capenable && !(IFCAP_TXCSUM & ifp->if_capenable)) { ifp->if_capenable &= ~IFCAP_TSO; ifp->if_hwassist &= ~CSUM_TSO; if_printf(ifp, "tso disabled due to -txcsum.\n"); } } if (mask & IFCAP_RXCSUM) ifp->if_capenable ^= IFCAP_RXCSUM; if (mask & IFCAP_TSO4) { ifp->if_capenable ^= IFCAP_TSO4; if (IFCAP_TSO & ifp->if_capenable) { if (IFCAP_TXCSUM & ifp->if_capenable) ifp->if_hwassist |= CSUM_TSO; else { ifp->if_capenable &= ~IFCAP_TSO; ifp->if_hwassist &= ~CSUM_TSO; if_printf(ifp, "enable txcsum first.\n"); error = EAGAIN; } } else ifp->if_hwassist &= ~CSUM_TSO; } if (mask & IFCAP_LRO) { ifp->if_capenable ^= IFCAP_LRO; /* Safe to do this even if cxgb_up not called yet */ cxgb_set_lro(p, ifp->if_capenable & IFCAP_LRO); } if (mask & IFCAP_VLAN_HWTAGGING) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { PORT_LOCK(p); cxgb_update_mac_settings(p); PORT_UNLOCK(p); } } if (mask & IFCAP_VLAN_MTU) { ifp->if_capenable ^= IFCAP_VLAN_MTU; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { PORT_LOCK(p); cxgb_update_mac_settings(p); PORT_UNLOCK(p); } } if (mask & IFCAP_VLAN_HWTSO) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if (mask & IFCAP_VLAN_HWCSUM) ifp->if_capenable ^= IFCAP_VLAN_HWCSUM; #ifdef VLAN_CAPABILITIES VLAN_CAPABILITIES(ifp); #endif ADAPTER_UNLOCK(sc); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &p->media, command); break; default: error = ether_ioctl(ifp, command, data); } return (error); } static int cxgb_media_change(struct ifnet *ifp) { return (EOPNOTSUPP); } /* * Translates phy->modtype to the correct Ethernet media subtype. */ static int cxgb_ifm_type(int mod) { switch (mod) { case phy_modtype_sr: return (IFM_10G_SR); case phy_modtype_lr: return (IFM_10G_LR); case phy_modtype_lrm: return (IFM_10G_LRM); case phy_modtype_twinax: return (IFM_10G_TWINAX); case phy_modtype_twinax_long: return (IFM_10G_TWINAX_LONG); case phy_modtype_none: return (IFM_NONE); case phy_modtype_unknown: return (IFM_UNKNOWN); } KASSERT(0, ("%s: modtype %d unknown", __func__, mod)); return (IFM_UNKNOWN); } /* * Rebuilds the ifmedia list for this port, and sets the current media. */ static void cxgb_build_medialist(struct port_info *p) { struct cphy *phy = &p->phy; struct ifmedia *media = &p->media; int mod = phy->modtype; int m = IFM_ETHER | IFM_FDX; PORT_LOCK(p); ifmedia_removeall(media); if (phy->caps & SUPPORTED_TP && phy->caps & SUPPORTED_Autoneg) { /* Copper (RJ45) */ if (phy->caps & SUPPORTED_10000baseT_Full) ifmedia_add(media, m | IFM_10G_T, mod, NULL); if (phy->caps & SUPPORTED_1000baseT_Full) ifmedia_add(media, m | IFM_1000_T, mod, NULL); if (phy->caps & SUPPORTED_100baseT_Full) ifmedia_add(media, m | IFM_100_TX, mod, NULL); if (phy->caps & SUPPORTED_10baseT_Full) ifmedia_add(media, m | IFM_10_T, mod, NULL); ifmedia_add(media, IFM_ETHER | IFM_AUTO, mod, NULL); ifmedia_set(media, IFM_ETHER | IFM_AUTO); } else if (phy->caps & SUPPORTED_TP) { /* Copper (CX4) */ KASSERT(phy->caps & SUPPORTED_10000baseT_Full, ("%s: unexpected cap 0x%x", __func__, phy->caps)); ifmedia_add(media, m | IFM_10G_CX4, mod, NULL); ifmedia_set(media, m | IFM_10G_CX4); } else if (phy->caps & SUPPORTED_FIBRE && phy->caps & SUPPORTED_10000baseT_Full) { /* 10G optical (but includes SFP+ twinax) */ m |= cxgb_ifm_type(mod); if (IFM_SUBTYPE(m) == IFM_NONE) m &= ~IFM_FDX; ifmedia_add(media, m, mod, NULL); ifmedia_set(media, m); } else if (phy->caps & SUPPORTED_FIBRE && phy->caps & SUPPORTED_1000baseT_Full) { /* 1G optical */ /* XXX: Lie and claim to be SX, could actually be any 1G-X */ ifmedia_add(media, m | IFM_1000_SX, mod, NULL); ifmedia_set(media, m | IFM_1000_SX); } else { KASSERT(0, ("%s: don't know how to handle 0x%x.", __func__, phy->caps)); } PORT_UNLOCK(p); } static void cxgb_media_status(struct ifnet *ifp, struct ifmediareq *ifmr) { struct port_info *p = ifp->if_softc; struct ifmedia_entry *cur = p->media.ifm_cur; int speed = p->link_config.speed; if (cur->ifm_data != p->phy.modtype) { cxgb_build_medialist(p); cur = p->media.ifm_cur; } ifmr->ifm_status = IFM_AVALID; if (!p->link_config.link_ok) return; ifmr->ifm_status |= IFM_ACTIVE; /* * active and current will differ iff current media is autoselect. That * can happen only for copper RJ45. */ if (IFM_SUBTYPE(cur->ifm_media) != IFM_AUTO) return; KASSERT(p->phy.caps & SUPPORTED_TP && p->phy.caps & SUPPORTED_Autoneg, ("%s: unexpected PHY caps 0x%x", __func__, p->phy.caps)); ifmr->ifm_active = IFM_ETHER | IFM_FDX; if (speed == SPEED_10000) ifmr->ifm_active |= IFM_10G_T; else if (speed == SPEED_1000) ifmr->ifm_active |= IFM_1000_T; else if (speed == SPEED_100) ifmr->ifm_active |= IFM_100_TX; else if (speed == SPEED_10) ifmr->ifm_active |= IFM_10_T; else KASSERT(0, ("%s: link up but speed unknown (%u)", __func__, speed)); } static void cxgb_async_intr(void *data) { adapter_t *sc = data; t3_write_reg(sc, A_PL_INT_ENABLE0, 0); (void) t3_read_reg(sc, A_PL_INT_ENABLE0); taskqueue_enqueue(sc->tq, &sc->slow_intr_task); } static void link_check_callout(void *arg) { struct port_info *pi = arg; struct adapter *sc = pi->adapter; if (!isset(&sc->open_device_map, pi->port_id)) return; taskqueue_enqueue(sc->tq, &pi->link_check_task); } static void check_link_status(void *arg, int pending) { struct port_info *pi = arg; struct adapter *sc = pi->adapter; if (!isset(&sc->open_device_map, pi->port_id)) return; t3_link_changed(sc, pi->port_id); if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ)) callout_reset(&pi->link_check_ch, hz, link_check_callout, pi); } void t3_os_link_intr(struct port_info *pi) { /* * Schedule a link check in the near future. If the link is flapping * rapidly we'll keep resetting the callout and delaying the check until * things stabilize a bit. */ callout_reset(&pi->link_check_ch, hz / 4, link_check_callout, pi); } static void check_t3b2_mac(struct adapter *sc) { int i; if (sc->flags & CXGB_SHUTDOWN) return; for_each_port(sc, i) { struct port_info *p = &sc->port[i]; int status; #ifdef INVARIANTS struct ifnet *ifp = p->ifp; #endif if (!isset(&sc->open_device_map, p->port_id) || p->link_fault || !p->link_config.link_ok) continue; KASSERT(ifp->if_drv_flags & IFF_DRV_RUNNING, ("%s: state mismatch (drv_flags %x, device_map %x)", __func__, ifp->if_drv_flags, sc->open_device_map)); PORT_LOCK(p); status = t3b2_mac_watchdog_task(&p->mac); if (status == 1) p->mac.stats.num_toggled++; else if (status == 2) { struct cmac *mac = &p->mac; cxgb_update_mac_settings(p); t3_link_start(&p->phy, mac, &p->link_config); t3_mac_enable(mac, MAC_DIRECTION_RX | MAC_DIRECTION_TX); t3_port_intr_enable(sc, p->port_id); p->mac.stats.num_resets++; } PORT_UNLOCK(p); } } static void cxgb_tick(void *arg) { adapter_t *sc = (adapter_t *)arg; if (sc->flags & CXGB_SHUTDOWN) return; taskqueue_enqueue(sc->tq, &sc->tick_task); callout_reset(&sc->cxgb_tick_ch, hz, cxgb_tick, sc); } static void cxgb_tick_handler(void *arg, int count) { adapter_t *sc = (adapter_t *)arg; const struct adapter_params *p = &sc->params; int i; uint32_t cause, reset; if (sc->flags & CXGB_SHUTDOWN || !(sc->flags & FULL_INIT_DONE)) return; if (p->rev == T3_REV_B2 && p->nports < 4 && sc->open_device_map) check_t3b2_mac(sc); cause = t3_read_reg(sc, A_SG_INT_CAUSE) & (F_RSPQSTARVE | F_FLEMPTY); if (cause) { struct sge_qset *qs = &sc->sge.qs[0]; uint32_t mask, v; v = t3_read_reg(sc, A_SG_RSPQ_FL_STATUS) & ~0xff00; mask = 1; for (i = 0; i < SGE_QSETS; i++) { if (v & mask) qs[i].rspq.starved++; mask <<= 1; } mask <<= SGE_QSETS; /* skip RSPQXDISABLED */ for (i = 0; i < SGE_QSETS * 2; i++) { if (v & mask) { qs[i / 2].fl[i % 2].empty++; } mask <<= 1; } /* clear */ t3_write_reg(sc, A_SG_RSPQ_FL_STATUS, v); t3_write_reg(sc, A_SG_INT_CAUSE, cause); } for (i = 0; i < sc->params.nports; i++) { struct port_info *pi = &sc->port[i]; struct ifnet *ifp = pi->ifp; struct cmac *mac = &pi->mac; struct mac_stats *mstats = &mac->stats; if (!isset(&sc->open_device_map, pi->port_id)) continue; PORT_LOCK(pi); t3_mac_update_stats(mac); PORT_UNLOCK(pi); ifp->if_opackets = mstats->tx_frames_64 + mstats->tx_frames_65_127 + mstats->tx_frames_128_255 + mstats->tx_frames_256_511 + mstats->tx_frames_512_1023 + mstats->tx_frames_1024_1518 + mstats->tx_frames_1519_max; ifp->if_ipackets = mstats->rx_frames_64 + mstats->rx_frames_65_127 + mstats->rx_frames_128_255 + mstats->rx_frames_256_511 + mstats->rx_frames_512_1023 + mstats->rx_frames_1024_1518 + mstats->rx_frames_1519_max; ifp->if_obytes = mstats->tx_octets; ifp->if_ibytes = mstats->rx_octets; ifp->if_omcasts = mstats->tx_mcast_frames; ifp->if_imcasts = mstats->rx_mcast_frames; ifp->if_collisions = mstats->tx_total_collisions; ifp->if_iqdrops = mstats->rx_cong_drops; ifp->if_oerrors = mstats->tx_excess_collisions + mstats->tx_underrun + mstats->tx_len_errs + mstats->tx_mac_internal_errs + mstats->tx_excess_deferral + mstats->tx_fcs_errs; ifp->if_ierrors = mstats->rx_jabber + mstats->rx_data_errs + mstats->rx_sequence_errs + mstats->rx_runt + mstats->rx_too_long + mstats->rx_mac_internal_errs + mstats->rx_short + mstats->rx_fcs_errs; if (mac->multiport) continue; /* Count rx fifo overflows, once per second */ cause = t3_read_reg(sc, A_XGM_INT_CAUSE + mac->offset); reset = 0; if (cause & F_RXFIFO_OVERFLOW) { mac->stats.rx_fifo_ovfl++; reset |= F_RXFIFO_OVERFLOW; } t3_write_reg(sc, A_XGM_INT_CAUSE + mac->offset, reset); } } static void touch_bars(device_t dev) { /* * Don't enable yet */ #if !defined(__LP64__) && 0 u32 v; pci_read_config_dword(pdev, PCI_BASE_ADDRESS_1, &v); pci_write_config_dword(pdev, PCI_BASE_ADDRESS_1, v); pci_read_config_dword(pdev, PCI_BASE_ADDRESS_3, &v); pci_write_config_dword(pdev, PCI_BASE_ADDRESS_3, v); pci_read_config_dword(pdev, PCI_BASE_ADDRESS_5, &v); pci_write_config_dword(pdev, PCI_BASE_ADDRESS_5, v); #endif } static int set_eeprom(struct port_info *pi, const uint8_t *data, int len, int offset) { uint8_t *buf; int err = 0; u32 aligned_offset, aligned_len, *p; struct adapter *adapter = pi->adapter; aligned_offset = offset & ~3; aligned_len = (len + (offset & 3) + 3) & ~3; if (aligned_offset != offset || aligned_len != len) { buf = malloc(aligned_len, M_DEVBUF, M_WAITOK|M_ZERO); if (!buf) return (ENOMEM); err = t3_seeprom_read(adapter, aligned_offset, (u32 *)buf); if (!err && aligned_len > 4) err = t3_seeprom_read(adapter, aligned_offset + aligned_len - 4, (u32 *)&buf[aligned_len - 4]); if (err) goto out; memcpy(buf + (offset & 3), data, len); } else buf = (uint8_t *)(uintptr_t)data; err = t3_seeprom_wp(adapter, 0); if (err) goto out; for (p = (u32 *)buf; !err && aligned_len; aligned_len -= 4, p++) { err = t3_seeprom_write(adapter, aligned_offset, *p); aligned_offset += 4; } if (!err) err = t3_seeprom_wp(adapter, 1); out: if (buf != data) free(buf, M_DEVBUF); return err; } static int in_range(int val, int lo, int hi) { return val < 0 || (val <= hi && val >= lo); } static int cxgb_extension_open(struct cdev *dev, int flags, int fmp, struct thread *td) { return (0); } static int cxgb_extension_close(struct cdev *dev, int flags, int fmt, struct thread *td) { return (0); } static int cxgb_extension_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag, struct thread *td) { int mmd, error = 0; struct port_info *pi = dev->si_drv1; adapter_t *sc = pi->adapter; #ifdef PRIV_SUPPORTED if (priv_check(td, PRIV_DRIVER)) { if (cxgb_debug) printf("user does not have access to privileged ioctls\n"); return (EPERM); } #else if (suser(td)) { if (cxgb_debug) printf("user does not have access to privileged ioctls\n"); return (EPERM); } #endif switch (cmd) { case CHELSIO_GET_MIIREG: { uint32_t val; struct cphy *phy = &pi->phy; struct ch_mii_data *mid = (struct ch_mii_data *)data; if (!phy->mdio_read) return (EOPNOTSUPP); if (is_10G(sc)) { mmd = mid->phy_id >> 8; if (!mmd) mmd = MDIO_DEV_PCS; else if (mmd > MDIO_DEV_VEND2) return (EINVAL); error = phy->mdio_read(sc, mid->phy_id & 0x1f, mmd, mid->reg_num, &val); } else error = phy->mdio_read(sc, mid->phy_id & 0x1f, 0, mid->reg_num & 0x1f, &val); if (error == 0) mid->val_out = val; break; } case CHELSIO_SET_MIIREG: { struct cphy *phy = &pi->phy; struct ch_mii_data *mid = (struct ch_mii_data *)data; if (!phy->mdio_write) return (EOPNOTSUPP); if (is_10G(sc)) { mmd = mid->phy_id >> 8; if (!mmd) mmd = MDIO_DEV_PCS; else if (mmd > MDIO_DEV_VEND2) return (EINVAL); error = phy->mdio_write(sc, mid->phy_id & 0x1f, mmd, mid->reg_num, mid->val_in); } else error = phy->mdio_write(sc, mid->phy_id & 0x1f, 0, mid->reg_num & 0x1f, mid->val_in); break; } case CHELSIO_SETREG: { struct ch_reg *edata = (struct ch_reg *)data; if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len) return (EFAULT); t3_write_reg(sc, edata->addr, edata->val); break; } case CHELSIO_GETREG: { struct ch_reg *edata = (struct ch_reg *)data; if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len) return (EFAULT); edata->val = t3_read_reg(sc, edata->addr); break; } case CHELSIO_GET_SGE_CONTEXT: { struct ch_cntxt *ecntxt = (struct ch_cntxt *)data; mtx_lock_spin(&sc->sge.reg_lock); switch (ecntxt->cntxt_type) { case CNTXT_TYPE_EGRESS: error = -t3_sge_read_ecntxt(sc, ecntxt->cntxt_id, ecntxt->data); break; case CNTXT_TYPE_FL: error = -t3_sge_read_fl(sc, ecntxt->cntxt_id, ecntxt->data); break; case CNTXT_TYPE_RSP: error = -t3_sge_read_rspq(sc, ecntxt->cntxt_id, ecntxt->data); break; case CNTXT_TYPE_CQ: error = -t3_sge_read_cq(sc, ecntxt->cntxt_id, ecntxt->data); break; default: error = EINVAL; break; } mtx_unlock_spin(&sc->sge.reg_lock); break; } case CHELSIO_GET_SGE_DESC: { struct ch_desc *edesc = (struct ch_desc *)data; int ret; if (edesc->queue_num >= SGE_QSETS * 6) return (EINVAL); ret = t3_get_desc(&sc->sge.qs[edesc->queue_num / 6], edesc->queue_num % 6, edesc->idx, edesc->data); if (ret < 0) return (EINVAL); edesc->size = ret; break; } case CHELSIO_GET_QSET_PARAMS: { struct qset_params *q; struct ch_qset_params *t = (struct ch_qset_params *)data; int q1 = pi->first_qset; int nqsets = pi->nqsets; int i; if (t->qset_idx >= nqsets) return EINVAL; i = q1 + t->qset_idx; q = &sc->params.sge.qset[i]; t->rspq_size = q->rspq_size; t->txq_size[0] = q->txq_size[0]; t->txq_size[1] = q->txq_size[1]; t->txq_size[2] = q->txq_size[2]; t->fl_size[0] = q->fl_size; t->fl_size[1] = q->jumbo_size; t->polling = q->polling; t->lro = q->lro; t->intr_lat = q->coalesce_usecs; t->cong_thres = q->cong_thres; t->qnum = i; if ((sc->flags & FULL_INIT_DONE) == 0) t->vector = 0; else if (sc->flags & USING_MSIX) t->vector = rman_get_start(sc->msix_irq_res[i]); else t->vector = rman_get_start(sc->irq_res); break; } case CHELSIO_GET_QSET_NUM: { struct ch_reg *edata = (struct ch_reg *)data; edata->val = pi->nqsets; break; } case CHELSIO_LOAD_FW: { uint8_t *fw_data; uint32_t vers; struct ch_mem_range *t = (struct ch_mem_range *)data; /* * You're allowed to load a firmware only before FULL_INIT_DONE * * FW_UPTODATE is also set so the rest of the initialization * will not overwrite what was loaded here. This gives you the * flexibility to load any firmware (and maybe shoot yourself in * the foot). */ ADAPTER_LOCK(sc); if (sc->open_device_map || sc->flags & FULL_INIT_DONE) { ADAPTER_UNLOCK(sc); return (EBUSY); } fw_data = malloc(t->len, M_DEVBUF, M_NOWAIT); if (!fw_data) error = ENOMEM; else error = copyin(t->buf, fw_data, t->len); if (!error) error = -t3_load_fw(sc, fw_data, t->len); if (t3_get_fw_version(sc, &vers) == 0) { snprintf(&sc->fw_version[0], sizeof(sc->fw_version), "%d.%d.%d", G_FW_VERSION_MAJOR(vers), G_FW_VERSION_MINOR(vers), G_FW_VERSION_MICRO(vers)); } if (!error) sc->flags |= FW_UPTODATE; free(fw_data, M_DEVBUF); ADAPTER_UNLOCK(sc); break; } case CHELSIO_LOAD_BOOT: { uint8_t *boot_data; struct ch_mem_range *t = (struct ch_mem_range *)data; boot_data = malloc(t->len, M_DEVBUF, M_NOWAIT); if (!boot_data) return ENOMEM; error = copyin(t->buf, boot_data, t->len); if (!error) error = -t3_load_boot(sc, boot_data, t->len); free(boot_data, M_DEVBUF); break; } case CHELSIO_GET_PM: { struct ch_pm *m = (struct ch_pm *)data; struct tp_params *p = &sc->params.tp; if (!is_offload(sc)) return (EOPNOTSUPP); m->tx_pg_sz = p->tx_pg_size; m->tx_num_pg = p->tx_num_pgs; m->rx_pg_sz = p->rx_pg_size; m->rx_num_pg = p->rx_num_pgs; m->pm_total = p->pmtx_size + p->chan_rx_size * p->nchan; break; } case CHELSIO_SET_PM: { struct ch_pm *m = (struct ch_pm *)data; struct tp_params *p = &sc->params.tp; if (!is_offload(sc)) return (EOPNOTSUPP); if (sc->flags & FULL_INIT_DONE) return (EBUSY); if (!m->rx_pg_sz || (m->rx_pg_sz & (m->rx_pg_sz - 1)) || !m->tx_pg_sz || (m->tx_pg_sz & (m->tx_pg_sz - 1))) return (EINVAL); /* not power of 2 */ if (!(m->rx_pg_sz & 0x14000)) return (EINVAL); /* not 16KB or 64KB */ if (!(m->tx_pg_sz & 0x1554000)) return (EINVAL); if (m->tx_num_pg == -1) m->tx_num_pg = p->tx_num_pgs; if (m->rx_num_pg == -1) m->rx_num_pg = p->rx_num_pgs; if (m->tx_num_pg % 24 || m->rx_num_pg % 24) return (EINVAL); if (m->rx_num_pg * m->rx_pg_sz > p->chan_rx_size || m->tx_num_pg * m->tx_pg_sz > p->chan_tx_size) return (EINVAL); p->rx_pg_size = m->rx_pg_sz; p->tx_pg_size = m->tx_pg_sz; p->rx_num_pgs = m->rx_num_pg; p->tx_num_pgs = m->tx_num_pg; break; } case CHELSIO_SETMTUTAB: { struct ch_mtus *m = (struct ch_mtus *)data; int i; if (!is_offload(sc)) return (EOPNOTSUPP); if (offload_running(sc)) return (EBUSY); if (m->nmtus != NMTUS) return (EINVAL); if (m->mtus[0] < 81) /* accommodate SACK */ return (EINVAL); /* * MTUs must be in ascending order */ for (i = 1; i < NMTUS; ++i) if (m->mtus[i] < m->mtus[i - 1]) return (EINVAL); memcpy(sc->params.mtus, m->mtus, sizeof(sc->params.mtus)); break; } case CHELSIO_GETMTUTAB: { struct ch_mtus *m = (struct ch_mtus *)data; if (!is_offload(sc)) return (EOPNOTSUPP); memcpy(m->mtus, sc->params.mtus, sizeof(m->mtus)); m->nmtus = NMTUS; break; } case CHELSIO_GET_MEM: { struct ch_mem_range *t = (struct ch_mem_range *)data; struct mc7 *mem; uint8_t *useraddr; u64 buf[32]; /* * Use these to avoid modifying len/addr in the return * struct */ uint32_t len = t->len, addr = t->addr; if (!is_offload(sc)) return (EOPNOTSUPP); if (!(sc->flags & FULL_INIT_DONE)) return (EIO); /* need the memory controllers */ if ((addr & 0x7) || (len & 0x7)) return (EINVAL); if (t->mem_id == MEM_CM) mem = &sc->cm; else if (t->mem_id == MEM_PMRX) mem = &sc->pmrx; else if (t->mem_id == MEM_PMTX) mem = &sc->pmtx; else return (EINVAL); /* * Version scheme: * bits 0..9: chip version * bits 10..15: chip revision */ t->version = 3 | (sc->params.rev << 10); /* * Read 256 bytes at a time as len can be large and we don't * want to use huge intermediate buffers. */ useraddr = (uint8_t *)t->buf; while (len) { unsigned int chunk = min(len, sizeof(buf)); error = t3_mc7_bd_read(mem, addr / 8, chunk / 8, buf); if (error) return (-error); if (copyout(buf, useraddr, chunk)) return (EFAULT); useraddr += chunk; addr += chunk; len -= chunk; } break; } case CHELSIO_READ_TCAM_WORD: { struct ch_tcam_word *t = (struct ch_tcam_word *)data; if (!is_offload(sc)) return (EOPNOTSUPP); if (!(sc->flags & FULL_INIT_DONE)) return (EIO); /* need MC5 */ return -t3_read_mc5_range(&sc->mc5, t->addr, 1, t->buf); break; } case CHELSIO_SET_TRACE_FILTER: { struct ch_trace *t = (struct ch_trace *)data; const struct trace_params *tp; tp = (const struct trace_params *)&t->sip; if (t->config_tx) t3_config_trace_filter(sc, tp, 0, t->invert_match, t->trace_tx); if (t->config_rx) t3_config_trace_filter(sc, tp, 1, t->invert_match, t->trace_rx); break; } case CHELSIO_SET_PKTSCHED: { struct ch_pktsched_params *p = (struct ch_pktsched_params *)data; if (sc->open_device_map == 0) return (EAGAIN); send_pktsched_cmd(sc, p->sched, p->idx, p->min, p->max, p->binding); break; } case CHELSIO_IFCONF_GETREGS: { struct ch_ifconf_regs *regs = (struct ch_ifconf_regs *)data; int reglen = cxgb_get_regs_len(); uint8_t *buf = malloc(reglen, M_DEVBUF, M_NOWAIT); if (buf == NULL) { return (ENOMEM); } if (regs->len > reglen) regs->len = reglen; else if (regs->len < reglen) error = ENOBUFS; if (!error) { cxgb_get_regs(sc, regs, buf); error = copyout(buf, regs->data, reglen); } free(buf, M_DEVBUF); break; } case CHELSIO_SET_HW_SCHED: { struct ch_hw_sched *t = (struct ch_hw_sched *)data; unsigned int ticks_per_usec = core_ticks_per_usec(sc); if ((sc->flags & FULL_INIT_DONE) == 0) return (EAGAIN); /* need TP to be initialized */ if (t->sched >= NTX_SCHED || !in_range(t->mode, 0, 1) || !in_range(t->channel, 0, 1) || !in_range(t->kbps, 0, 10000000) || !in_range(t->class_ipg, 0, 10000 * 65535 / ticks_per_usec) || !in_range(t->flow_ipg, 0, dack_ticks_to_usec(sc, 0x7ff))) return (EINVAL); if (t->kbps >= 0) { error = t3_config_sched(sc, t->kbps, t->sched); if (error < 0) return (-error); } if (t->class_ipg >= 0) t3_set_sched_ipg(sc, t->sched, t->class_ipg); if (t->flow_ipg >= 0) { t->flow_ipg *= 1000; /* us -> ns */ t3_set_pace_tbl(sc, &t->flow_ipg, t->sched, 1); } if (t->mode >= 0) { int bit = 1 << (S_TX_MOD_TIMER_MODE + t->sched); t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP, bit, t->mode ? bit : 0); } if (t->channel >= 0) t3_set_reg_field(sc, A_TP_TX_MOD_QUEUE_REQ_MAP, 1 << t->sched, t->channel << t->sched); break; } case CHELSIO_GET_EEPROM: { int i; struct ch_eeprom *e = (struct ch_eeprom *)data; uint8_t *buf = malloc(EEPROMSIZE, M_DEVBUF, M_NOWAIT); if (buf == NULL) { return (ENOMEM); } e->magic = EEPROM_MAGIC; for (i = e->offset & ~3; !error && i < e->offset + e->len; i += 4) error = -t3_seeprom_read(sc, i, (uint32_t *)&buf[i]); if (!error) error = copyout(buf + e->offset, e->data, e->len); free(buf, M_DEVBUF); break; } case CHELSIO_CLEAR_STATS: { if (!(sc->flags & FULL_INIT_DONE)) return EAGAIN; PORT_LOCK(pi); t3_mac_update_stats(&pi->mac); memset(&pi->mac.stats, 0, sizeof(pi->mac.stats)); PORT_UNLOCK(pi); break; } case CHELSIO_GET_UP_LA: { struct ch_up_la *la = (struct ch_up_la *)data; uint8_t *buf = malloc(LA_BUFSIZE, M_DEVBUF, M_NOWAIT); if (buf == NULL) { return (ENOMEM); } if (la->bufsize < LA_BUFSIZE) error = ENOBUFS; if (!error) error = -t3_get_up_la(sc, &la->stopped, &la->idx, &la->bufsize, buf); if (!error) error = copyout(buf, la->data, la->bufsize); free(buf, M_DEVBUF); break; } case CHELSIO_GET_UP_IOQS: { struct ch_up_ioqs *ioqs = (struct ch_up_ioqs *)data; uint8_t *buf = malloc(IOQS_BUFSIZE, M_DEVBUF, M_NOWAIT); uint32_t *v; if (buf == NULL) { return (ENOMEM); } if (ioqs->bufsize < IOQS_BUFSIZE) error = ENOBUFS; if (!error) error = -t3_get_up_ioqs(sc, &ioqs->bufsize, buf); if (!error) { v = (uint32_t *)buf; ioqs->ioq_rx_enable = *v++; ioqs->ioq_tx_enable = *v++; ioqs->ioq_rx_status = *v++; ioqs->ioq_tx_status = *v++; error = copyout(v, ioqs->data, ioqs->bufsize); } free(buf, M_DEVBUF); break; } case CHELSIO_SET_FILTER: { struct ch_filter *f = (struct ch_filter *)data;; struct filter_info *p; unsigned int nfilters = sc->params.mc5.nfilters; if (!is_offload(sc)) return (EOPNOTSUPP); /* No TCAM */ if (!(sc->flags & FULL_INIT_DONE)) return (EAGAIN); /* mc5 not setup yet */ if (nfilters == 0) return (EBUSY); /* TOE will use TCAM */ /* sanity checks */ if (f->filter_id >= nfilters || (f->val.dip && f->mask.dip != 0xffffffff) || (f->val.sport && f->mask.sport != 0xffff) || (f->val.dport && f->mask.dport != 0xffff) || (f->val.vlan && f->mask.vlan != 0xfff) || (f->val.vlan_prio && f->mask.vlan_prio != FILTER_NO_VLAN_PRI) || (f->mac_addr_idx != 0xffff && f->mac_addr_idx > 15) || f->qset >= SGE_QSETS || sc->rrss_map[f->qset] >= RSS_TABLE_SIZE) return (EINVAL); /* Was allocated with M_WAITOK */ KASSERT(sc->filters, ("filter table NULL\n")); p = &sc->filters[f->filter_id]; if (p->locked) return (EPERM); bzero(p, sizeof(*p)); p->sip = f->val.sip; p->sip_mask = f->mask.sip; p->dip = f->val.dip; p->sport = f->val.sport; p->dport = f->val.dport; p->vlan = f->mask.vlan ? f->val.vlan : 0xfff; p->vlan_prio = f->mask.vlan_prio ? (f->val.vlan_prio & 6) : FILTER_NO_VLAN_PRI; p->mac_hit = f->mac_hit; p->mac_vld = f->mac_addr_idx != 0xffff; p->mac_idx = f->mac_addr_idx; p->pkt_type = f->proto; p->report_filter_id = f->want_filter_id; p->pass = f->pass; p->rss = f->rss; p->qset = f->qset; error = set_filter(sc, f->filter_id, p); if (error == 0) p->valid = 1; break; } case CHELSIO_DEL_FILTER: { struct ch_filter *f = (struct ch_filter *)data; struct filter_info *p; unsigned int nfilters = sc->params.mc5.nfilters; if (!is_offload(sc)) return (EOPNOTSUPP); if (!(sc->flags & FULL_INIT_DONE)) return (EAGAIN); if (nfilters == 0 || sc->filters == NULL) return (EINVAL); if (f->filter_id >= nfilters) return (EINVAL); p = &sc->filters[f->filter_id]; if (p->locked) return (EPERM); if (!p->valid) return (EFAULT); /* Read "Bad address" as "Bad index" */ bzero(p, sizeof(*p)); p->sip = p->sip_mask = 0xffffffff; p->vlan = 0xfff; p->vlan_prio = FILTER_NO_VLAN_PRI; p->pkt_type = 1; error = set_filter(sc, f->filter_id, p); break; } case CHELSIO_GET_FILTER: { struct ch_filter *f = (struct ch_filter *)data; struct filter_info *p; unsigned int i, nfilters = sc->params.mc5.nfilters; if (!is_offload(sc)) return (EOPNOTSUPP); if (!(sc->flags & FULL_INIT_DONE)) return (EAGAIN); if (nfilters == 0 || sc->filters == NULL) return (EINVAL); i = f->filter_id == 0xffffffff ? 0 : f->filter_id + 1; for (; i < nfilters; i++) { p = &sc->filters[i]; if (!p->valid) continue; bzero(f, sizeof(*f)); f->filter_id = i; f->val.sip = p->sip; f->mask.sip = p->sip_mask; f->val.dip = p->dip; f->mask.dip = p->dip ? 0xffffffff : 0; f->val.sport = p->sport; f->mask.sport = p->sport ? 0xffff : 0; f->val.dport = p->dport; f->mask.dport = p->dport ? 0xffff : 0; f->val.vlan = p->vlan == 0xfff ? 0 : p->vlan; f->mask.vlan = p->vlan == 0xfff ? 0 : 0xfff; f->val.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ? 0 : p->vlan_prio; f->mask.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ? 0 : FILTER_NO_VLAN_PRI; f->mac_hit = p->mac_hit; f->mac_addr_idx = p->mac_vld ? p->mac_idx : 0xffff; f->proto = p->pkt_type; f->want_filter_id = p->report_filter_id; f->pass = p->pass; f->rss = p->rss; f->qset = p->qset; break; } if (i == nfilters) f->filter_id = 0xffffffff; break; } default: return (EOPNOTSUPP); break; } return (error); } static __inline void reg_block_dump(struct adapter *ap, uint8_t *buf, unsigned int start, unsigned int end) { uint32_t *p = (uint32_t *)(buf + start); for ( ; start <= end; start += sizeof(uint32_t)) *p++ = t3_read_reg(ap, start); } #define T3_REGMAP_SIZE (3 * 1024) static int cxgb_get_regs_len(void) { return T3_REGMAP_SIZE; } static void cxgb_get_regs(adapter_t *sc, struct ch_ifconf_regs *regs, uint8_t *buf) { /* * Version scheme: * bits 0..9: chip version * bits 10..15: chip revision * bit 31: set for PCIe cards */ regs->version = 3 | (sc->params.rev << 10) | (is_pcie(sc) << 31); /* * We skip the MAC statistics registers because they are clear-on-read. * Also reading multi-register stats would need to synchronize with the * periodic mac stats accumulation. Hard to justify the complexity. */ memset(buf, 0, cxgb_get_regs_len()); reg_block_dump(sc, buf, 0, A_SG_RSPQ_CREDIT_RETURN); reg_block_dump(sc, buf, A_SG_HI_DRB_HI_THRSH, A_ULPRX_PBL_ULIMIT); reg_block_dump(sc, buf, A_ULPTX_CONFIG, A_MPS_INT_CAUSE); reg_block_dump(sc, buf, A_CPL_SWITCH_CNTRL, A_CPL_MAP_TBL_DATA); reg_block_dump(sc, buf, A_SMB_GLOBAL_TIME_CFG, A_XGM_SERDES_STAT3); reg_block_dump(sc, buf, A_XGM_SERDES_STATUS0, XGM_REG(A_XGM_SERDES_STAT3, 1)); reg_block_dump(sc, buf, XGM_REG(A_XGM_SERDES_STATUS0, 1), XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1)); } static int alloc_filters(struct adapter *sc) { struct filter_info *p; unsigned int nfilters = sc->params.mc5.nfilters; if (nfilters == 0) return (0); p = malloc(sizeof(*p) * nfilters, M_DEVBUF, M_WAITOK | M_ZERO); sc->filters = p; p = &sc->filters[nfilters - 1]; p->vlan = 0xfff; p->vlan_prio = FILTER_NO_VLAN_PRI; p->pass = p->rss = p->valid = p->locked = 1; return (0); } static int setup_hw_filters(struct adapter *sc) { int i, rc; unsigned int nfilters = sc->params.mc5.nfilters; if (!sc->filters) return (0); t3_enable_filters(sc); for (i = rc = 0; i < nfilters && !rc; i++) { if (sc->filters[i].locked) rc = set_filter(sc, i, &sc->filters[i]); } return (rc); } static int set_filter(struct adapter *sc, int id, const struct filter_info *f) { int len; struct mbuf *m; struct ulp_txpkt *txpkt; struct work_request_hdr *wr; struct cpl_pass_open_req *oreq; struct cpl_set_tcb_field *sreq; len = sizeof(*wr) + sizeof(*oreq) + 2 * sizeof(*sreq); KASSERT(len <= MHLEN, ("filter request too big for an mbuf")); id += t3_mc5_size(&sc->mc5) - sc->params.mc5.nroutes - sc->params.mc5.nfilters; m = m_gethdr(M_WAITOK, MT_DATA); m->m_len = m->m_pkthdr.len = len; bzero(mtod(m, char *), len); wr = mtod(m, struct work_request_hdr *); wr->wr_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS) | F_WR_ATOMIC); oreq = (struct cpl_pass_open_req *)(wr + 1); txpkt = (struct ulp_txpkt *)oreq; txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT)); txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*oreq) / 8)); OPCODE_TID(oreq) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, id)); oreq->local_port = htons(f->dport); oreq->peer_port = htons(f->sport); oreq->local_ip = htonl(f->dip); oreq->peer_ip = htonl(f->sip); oreq->peer_netmask = htonl(f->sip_mask); oreq->opt0h = 0; oreq->opt0l = htonl(F_NO_OFFLOAD); oreq->opt1 = htonl(V_MAC_MATCH_VALID(f->mac_vld) | V_CONN_POLICY(CPL_CONN_POLICY_FILTER) | V_VLAN_PRI(f->vlan_prio >> 1) | V_VLAN_PRI_VALID(f->vlan_prio != FILTER_NO_VLAN_PRI) | V_PKT_TYPE(f->pkt_type) | V_OPT1_VLAN(f->vlan) | V_MAC_MATCH(f->mac_idx | (f->mac_hit << 4))); sreq = (struct cpl_set_tcb_field *)(oreq + 1); set_tcb_field_ulp(sreq, id, 1, 0x1800808000ULL, (f->report_filter_id << 15) | (1 << 23) | ((u64)f->pass << 35) | ((u64)!f->rss << 36)); set_tcb_field_ulp(sreq + 1, id, 0, 0xffffffff, (2 << 19) | 1); t3_mgmt_tx(sc, m); if (f->pass && !f->rss) { len = sizeof(*sreq); m = m_gethdr(M_WAITOK, MT_DATA); m->m_len = m->m_pkthdr.len = len; bzero(mtod(m, char *), len); sreq = mtod(m, struct cpl_set_tcb_field *); sreq->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); mk_set_tcb_field(sreq, id, 25, 0x3f80000, (u64)sc->rrss_map[f->qset] << 19); t3_mgmt_tx(sc, m); } return 0; } static inline void mk_set_tcb_field(struct cpl_set_tcb_field *req, unsigned int tid, unsigned int word, u64 mask, u64 val) { OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid)); req->reply = V_NO_REPLY(1); req->cpu_idx = 0; req->word = htons(word); req->mask = htobe64(mask); req->val = htobe64(val); } static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *req, unsigned int tid, unsigned int word, u64 mask, u64 val) { struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req; txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT)); txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*req) / 8)); mk_set_tcb_field(req, tid, word, mask, val); } Index: stable/7/sys/dev/cxgb/cxgb_osdep.h =================================================================== --- stable/7/sys/dev/cxgb/cxgb_osdep.h (revision 242221) +++ stable/7/sys/dev/cxgb/cxgb_osdep.h (revision 242222) @@ -1,501 +1,501 @@ /************************************************************************** Copyright (c) 2007-2008, Chelsio Inc. All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Neither the name of the Chelsio Corporation nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. $FreeBSD$ ***************************************************************************/ #include #include #include #include #include #include #include #include #ifdef CONFIG_DEFINED #include #include #else #include #include #endif #ifndef _CXGB_OSDEP_H_ #define _CXGB_OSDEP_H_ typedef struct adapter adapter_t; typedef struct port_info pinfo_t; struct sge_rspq; enum { TP_TMR_RES = 200, /* TP timer resolution in usec */ MAX_NPORTS = 4, /* max # of ports */ TP_SRAM_OFFSET = 4096, /* TP SRAM content offset in eeprom */ TP_SRAM_LEN = 2112, /* TP SRAM content offset in eeprom */ }; struct t3_mbuf_hdr { struct mbuf *mh_head; struct mbuf *mh_tail; }; #ifndef PANIC_IF #define PANIC_IF(exp) do { \ if (exp) \ panic("BUG: %s", #exp); \ } while (0) #endif #define m_get_priority(m) ((uintptr_t)(m)->m_pkthdr.rcvif) #define m_set_priority(m, pri) ((m)->m_pkthdr.rcvif = (struct ifnet *)((uintptr_t)pri)) #define m_set_sgl(m, sgl) ((m)->m_pkthdr.header = (sgl)) #define m_get_sgl(m) ((bus_dma_segment_t *)(m)->m_pkthdr.header) #define m_set_sgllen(m, len) ((m)->m_pkthdr.ether_vtag = len) #define m_get_sgllen(m) ((m)->m_pkthdr.ether_vtag) /* * XXX FIXME */ #define m_set_toep(m, a) ((m)->m_pkthdr.header = (a)) #define m_get_toep(m) ((m)->m_pkthdr.header) #define m_set_handler(m, handler) ((m)->m_pkthdr.header = (handler)) #define m_set_socket(m, a) ((m)->m_pkthdr.header = (a)) #define m_get_socket(m) ((m)->m_pkthdr.header) #define KTR_CXGB KTR_SPARE2 void cxgb_log_tcb(struct adapter *sc, unsigned int tid); #define MT_DONTFREE 128 #if __FreeBSD_version > 700030 #define INTR_FILTERS #define FIRMWARE_LATEST #endif #if ((__FreeBSD_version > 602103) && (__FreeBSD_version < 700000)) #define FIRMWARE_LATEST #endif #if __FreeBSD_version > 700000 #define MSI_SUPPORTED #define TSO_SUPPORTED #define VLAN_SUPPORTED #define TASKQUEUE_CURRENT #else #define if_name(ifp) (ifp)->if_xname #define M_SANITY(m, n) #endif #define __read_mostly __attribute__((__section__(".data.read_mostly"))) /* * Workaround for weird Chelsio issue */ #if __FreeBSD_version > 700029 #define PRIV_SUPPORTED #endif #define CXGB_TX_CLEANUP_THRESHOLD 32 #ifdef DEBUG_PRINT #define DPRINTF printf #else #define DPRINTF(...) #endif #define TX_MAX_SIZE (1 << 16) /* 64KB */ #define TX_MAX_SEGS 36 /* maximum supported by card */ #define TX_MAX_DESC 4 /* max descriptors per packet */ #define TX_START_MIN_DESC (TX_MAX_DESC << 2) #define TX_START_MAX_DESC (TX_MAX_DESC << 3) /* maximum number of descriptors * call to start used per */ #define TX_CLEAN_MAX_DESC (TX_MAX_DESC << 4) /* maximum tx descriptors * to clean per iteration */ #define TX_WR_SIZE_MAX 11*1024 /* the maximum total size of packets aggregated into a single * TX WR */ #define TX_WR_COUNT_MAX 7 /* the maximum total number of packets that can be * aggregated into a single TX WR */ #if defined(__i386__) || defined(__amd64__) #define mb() __asm volatile("mfence":::"memory") #define rmb() __asm volatile("lfence":::"memory") #define wmb() __asm volatile("sfence" ::: "memory") #define smp_mb() mb() #define L1_CACHE_BYTES 128 static __inline void prefetch(void *x) { __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); } extern void kdb_backtrace(void); #define WARN_ON(condition) do { \ if (__predict_false((condition)!=0)) { \ log(LOG_WARNING, "BUG: warning at %s:%d/%s()\n", __FILE__, __LINE__, __FUNCTION__); \ kdb_backtrace(); \ } \ } while (0) #else /* !i386 && !amd64 */ #define mb() #define rmb() #define wmb() #define smp_mb() #define prefetch(x) #define L1_CACHE_BYTES 32 #endif struct buf_ring { caddr_t *br_ring; volatile uint32_t br_cons; volatile uint32_t br_prod; int br_size; struct mtx br_lock; }; struct buf_ring *buf_ring_alloc(int count, int flags); void buf_ring_free(struct buf_ring *); static __inline int buf_ring_count(struct buf_ring *mr) { int size = mr->br_size; uint32_t mask = size - 1; return ((size + mr->br_prod - mr->br_cons) & mask); } static __inline int buf_ring_empty(struct buf_ring *mr) { return (mr->br_cons == mr->br_prod); } static __inline int buf_ring_full(struct buf_ring *mr) { uint32_t mask; mask = mr->br_size - 1; return (mr->br_cons == ((mr->br_prod + 1) & mask)); } /* * The producer and consumer are independently locked * this relies on the consumer providing his own serialization * */ static __inline void * buf_ring_dequeue(struct buf_ring *mr) { uint32_t prod, cons, mask; caddr_t *ring, m; ring = (caddr_t *)mr->br_ring; mask = mr->br_size - 1; cons = mr->br_cons; mb(); prod = mr->br_prod; m = NULL; if (cons != prod) { m = ring[cons]; ring[cons] = NULL; mr->br_cons = (cons + 1) & mask; mb(); } return (m); } #ifdef DEBUG_BUFRING static __inline void __buf_ring_scan(struct buf_ring *mr, void *m, char *file, int line) { int i; for (i = 0; i < mr->br_size; i++) if (m == mr->br_ring[i]) panic("%s:%d m=%p present prod=%d cons=%d idx=%d", file, line, m, mr->br_prod, mr->br_cons, i); } static __inline void buf_ring_scan(struct buf_ring *mr, void *m, char *file, int line) { mtx_lock(&mr->br_lock); __buf_ring_scan(mr, m, file, line); mtx_unlock(&mr->br_lock); } #else static __inline void __buf_ring_scan(struct buf_ring *mr, void *m, char *file, int line) { } static __inline void buf_ring_scan(struct buf_ring *mr, void *m, char *file, int line) { } #endif static __inline int __buf_ring_enqueue(struct buf_ring *mr, void *m, char *file, int line) { uint32_t prod, cons, mask; int err; mask = mr->br_size - 1; prod = mr->br_prod; mb(); cons = mr->br_cons; __buf_ring_scan(mr, m, file, line); if (((prod + 1) & mask) != cons) { KASSERT(mr->br_ring[prod] == NULL, ("overwriting entry")); mr->br_ring[prod] = m; mb(); mr->br_prod = (prod + 1) & mask; err = 0; } else err = ENOBUFS; return (err); } static __inline int buf_ring_enqueue_(struct buf_ring *mr, void *m, char *file, int line) { int err; mtx_lock(&mr->br_lock); err = __buf_ring_enqueue(mr, m, file, line); mtx_unlock(&mr->br_lock); return (err); } #define buf_ring_enqueue(mr, m) buf_ring_enqueue_((mr), (m), __FILE__, __LINE__) static __inline void * buf_ring_peek(struct buf_ring *mr) { int prod, cons, mask; caddr_t *ring, m; ring = (caddr_t *)mr->br_ring; mask = mr->br_size - 1; cons = mr->br_cons; prod = mr->br_prod; m = NULL; if (cons != prod) m = ring[cons]; return (m); } #define DBG_RX (1 << 0) static const int debug_flags = DBG_RX; #ifdef DEBUG_PRINT #define DBG(flag, msg) do { \ if ((flag & debug_flags)) \ printf msg; \ } while (0) #else #define DBG(...) #endif #include #define promisc_rx_mode(rm) ((rm)->port->ifp->if_flags & IFF_PROMISC) #define allmulti_rx_mode(rm) ((rm)->port->ifp->if_flags & IFF_ALLMULTI) #define CH_ERR(adap, fmt, ...) log(LOG_ERR, fmt, ##__VA_ARGS__) #define CH_WARN(adap, fmt, ...) log(LOG_WARNING, fmt, ##__VA_ARGS__) #define CH_ALERT(adap, fmt, ...) log(LOG_ALERT, fmt, ##__VA_ARGS__) #define t3_os_sleep(x) DELAY((x) * 1000) #define test_and_clear_bit(bit, p) atomic_cmpset_int((p), ((*(p)) | (1< #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "e1000_api.h" #include "e1000_82571.h" #include "if_em.h" /********************************************************************* * Set this to one to display debug statistics *********************************************************************/ int em_display_debug_stats = 0; /********************************************************************* * Driver version: *********************************************************************/ char em_driver_version[] = "7.1.9"; /********************************************************************* * PCI Device ID Table * * Used by probe to select devices to load on * Last field stores an index into e1000_strings * Last entry must be all 0s * * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index } *********************************************************************/ static em_vendor_info_t em_vendor_info_array[] = { /* Intel(R) PRO/1000 Network Connection */ { 0x8086, E1000_DEV_ID_82571EB_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_SERDES_DUAL, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_SERDES_QUAD, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER_LP, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571EB_QUAD_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82571PT_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82572EI_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82572EI_FIBER, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82572EI_SERDES, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82572EI, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82573E, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82573E_IAMT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82573L, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82583V, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_SPT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_SPT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_DPT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_DPT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IGP_M_AMT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IGP_AMT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IGP_C, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IFE, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IFE_GT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IFE_G, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_IGP_M, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH8_82567V_3, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IGP_M_AMT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IGP_AMT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IGP_C, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IGP_M, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IGP_M_V, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IFE, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IFE_GT, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_IFE_G, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH9_BM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82574L, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_82574LA, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_R_BM_LM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_R_BM_LF, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_R_BM_V, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_D_BM_LM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_D_BM_LF, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_ICH10_D_BM_V, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH_M_HV_LM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH_M_HV_LC, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH_D_HV_DM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH_D_HV_DC, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH2_LV_LM, PCI_ANY_ID, PCI_ANY_ID, 0}, { 0x8086, E1000_DEV_ID_PCH2_LV_V, PCI_ANY_ID, PCI_ANY_ID, 0}, /* required last entry */ { 0, 0, 0, 0, 0} }; /********************************************************************* * Table of branding strings for all supported NICs. *********************************************************************/ static char *em_strings[] = { "Intel(R) PRO/1000 Network Connection" }; /********************************************************************* * Function prototypes *********************************************************************/ static int em_probe(device_t); static int em_attach(device_t); static int em_detach(device_t); static int em_shutdown(device_t); static int em_suspend(device_t); static int em_resume(device_t); static void em_start(struct ifnet *); static void em_start_locked(struct ifnet *, struct tx_ring *); static int em_ioctl(struct ifnet *, u_long, caddr_t); static void em_init(void *); static void em_init_locked(struct adapter *); static void em_stop(void *); static void em_media_status(struct ifnet *, struct ifmediareq *); static int em_media_change(struct ifnet *); static void em_identify_hardware(struct adapter *); static int em_allocate_pci_resources(struct adapter *); static int em_allocate_legacy(struct adapter *); static int em_allocate_msix(struct adapter *); static int em_allocate_queues(struct adapter *); static int em_setup_msix(struct adapter *); static void em_free_pci_resources(struct adapter *); static void em_local_timer(void *); static void em_reset(struct adapter *); static int em_setup_interface(device_t, struct adapter *); static void em_setup_transmit_structures(struct adapter *); static void em_initialize_transmit_unit(struct adapter *); static int em_allocate_transmit_buffers(struct tx_ring *); static void em_free_transmit_structures(struct adapter *); static void em_free_transmit_buffers(struct tx_ring *); static int em_setup_receive_structures(struct adapter *); static int em_allocate_receive_buffers(struct rx_ring *); static void em_initialize_receive_unit(struct adapter *); static void em_free_receive_structures(struct adapter *); static void em_free_receive_buffers(struct rx_ring *); static void em_enable_intr(struct adapter *); static void em_disable_intr(struct adapter *); static void em_update_stats_counters(struct adapter *); static void em_add_hw_stats(struct adapter *adapter); static bool em_txeof(struct tx_ring *); static bool em_rxeof(struct rx_ring *, int, int *); #ifndef __NO_STRICT_ALIGNMENT static int em_fixup_rx(struct rx_ring *); #endif static void em_receive_checksum(struct e1000_rx_desc *, struct mbuf *); static void em_transmit_checksum_setup(struct tx_ring *, struct mbuf *, int, struct ip *, u32 *, u32 *); static void em_tso_setup(struct tx_ring *, struct mbuf *, int, struct ip *, struct tcphdr *, u32 *, u32 *); static void em_set_promisc(struct adapter *); static void em_disable_promisc(struct adapter *); static void em_set_multi(struct adapter *); static void em_update_link_status(struct adapter *); static void em_refresh_mbufs(struct rx_ring *, int); static void em_register_vlan(void *, struct ifnet *, u16); static void em_unregister_vlan(void *, struct ifnet *, u16); static void em_setup_vlan_hw_support(struct adapter *); static int em_xmit(struct tx_ring *, struct mbuf **); static int em_dma_malloc(struct adapter *, bus_size_t, struct em_dma_alloc *, int); static void em_dma_free(struct adapter *, struct em_dma_alloc *); static int em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS); static void em_print_nvm_info(struct adapter *); static int em_sysctl_debug_info(SYSCTL_HANDLER_ARGS); static void em_print_debug_info(struct adapter *); static int em_is_valid_ether_addr(u8 *); static int em_sysctl_int_delay(SYSCTL_HANDLER_ARGS); static void em_add_int_delay_sysctl(struct adapter *, const char *, const char *, struct em_int_delay_info *, int, int); /* Management and WOL Support */ static void em_init_manageability(struct adapter *); static void em_release_manageability(struct adapter *); static void em_get_hw_control(struct adapter *); static void em_release_hw_control(struct adapter *); static void em_get_wakeup(device_t); static void em_enable_wakeup(device_t); static int em_enable_phy_wakeup(struct adapter *); static void em_led_func(void *, int); static void em_disable_aspm(struct adapter *); static int em_irq_fast(void *); /* MSIX handlers */ static void em_msix_tx(void *); static void em_msix_rx(void *); static void em_msix_link(void *); static void em_handle_tx(void *context, int pending); static void em_handle_rx(void *context, int pending); static void em_handle_link(void *context, int pending); static void em_add_rx_process_limit(struct adapter *, const char *, const char *, int *, int); static void em_set_flow_cntrl(struct adapter *, const char *, const char *, int *, int); static __inline void em_rx_discard(struct rx_ring *, int); #ifdef DEVICE_POLLING static poll_handler_t em_poll; #endif /* POLLING */ /********************************************************************* * FreeBSD Device Interface Entry Points *********************************************************************/ static device_method_t em_methods[] = { /* Device interface */ DEVMETHOD(device_probe, em_probe), DEVMETHOD(device_attach, em_attach), DEVMETHOD(device_detach, em_detach), DEVMETHOD(device_shutdown, em_shutdown), DEVMETHOD(device_suspend, em_suspend), DEVMETHOD(device_resume, em_resume), {0, 0} }; static driver_t em_driver = { "em", em_methods, sizeof(struct adapter), }; devclass_t em_devclass; DRIVER_MODULE(em, pci, em_driver, em_devclass, 0, 0); MODULE_DEPEND(em, pci, 1, 1, 1); MODULE_DEPEND(em, ether, 1, 1, 1); /********************************************************************* * Tunable default values. *********************************************************************/ #define EM_TICKS_TO_USECS(ticks) ((1024 * (ticks) + 500) / 1000) #define EM_USECS_TO_TICKS(usecs) ((1000 * (usecs) + 512) / 1024) #define M_TSO_LEN 66 /* Allow common code without TSO */ #ifndef CSUM_TSO #define CSUM_TSO 0 #endif static int em_tx_int_delay_dflt = EM_TICKS_TO_USECS(EM_TIDV); static int em_rx_int_delay_dflt = EM_TICKS_TO_USECS(EM_RDTR); TUNABLE_INT("hw.em.tx_int_delay", &em_tx_int_delay_dflt); TUNABLE_INT("hw.em.rx_int_delay", &em_rx_int_delay_dflt); static int em_tx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_TADV); static int em_rx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_RADV); TUNABLE_INT("hw.em.tx_abs_int_delay", &em_tx_abs_int_delay_dflt); TUNABLE_INT("hw.em.rx_abs_int_delay", &em_rx_abs_int_delay_dflt); static int em_rxd = EM_DEFAULT_RXD; static int em_txd = EM_DEFAULT_TXD; TUNABLE_INT("hw.em.rxd", &em_rxd); TUNABLE_INT("hw.em.txd", &em_txd); static int em_smart_pwr_down = FALSE; TUNABLE_INT("hw.em.smart_pwr_down", &em_smart_pwr_down); /* Controls whether promiscuous also shows bad packets */ static int em_debug_sbp = FALSE; TUNABLE_INT("hw.em.sbp", &em_debug_sbp); static int em_enable_msix = TRUE; TUNABLE_INT("hw.em.enable_msix", &em_enable_msix); /* How many packets rxeof tries to clean at a time */ static int em_rx_process_limit = 100; TUNABLE_INT("hw.em.rx_process_limit", &em_rx_process_limit); /* Flow control setting - default to FULL */ static int em_fc_setting = e1000_fc_full; TUNABLE_INT("hw.em.fc_setting", &em_fc_setting); /* Global used in WOL setup with multiport cards */ static int global_quad_port_a = 0; /********************************************************************* * Device identification routine * * em_probe determines if the driver should be loaded on * adapter based on PCI vendor/device id of the adapter. * * return BUS_PROBE_DEFAULT on success, positive on failure *********************************************************************/ static int em_probe(device_t dev) { char adapter_name[60]; u16 pci_vendor_id = 0; u16 pci_device_id = 0; u16 pci_subvendor_id = 0; u16 pci_subdevice_id = 0; em_vendor_info_t *ent; INIT_DEBUGOUT("em_probe: begin"); pci_vendor_id = pci_get_vendor(dev); if (pci_vendor_id != EM_VENDOR_ID) return (ENXIO); pci_device_id = pci_get_device(dev); pci_subvendor_id = pci_get_subvendor(dev); pci_subdevice_id = pci_get_subdevice(dev); ent = em_vendor_info_array; while (ent->vendor_id != 0) { if ((pci_vendor_id == ent->vendor_id) && (pci_device_id == ent->device_id) && ((pci_subvendor_id == ent->subvendor_id) || (ent->subvendor_id == PCI_ANY_ID)) && ((pci_subdevice_id == ent->subdevice_id) || (ent->subdevice_id == PCI_ANY_ID))) { sprintf(adapter_name, "%s %s", em_strings[ent->index], em_driver_version); device_set_desc_copy(dev, adapter_name); return (BUS_PROBE_DEFAULT); } ent++; } return (ENXIO); } /********************************************************************* * Device initialization routine * * The attach entry point is called when the driver is being loaded. * This routine identifies the type of hardware, allocates all resources * and initializes the hardware. * * return 0 on success, positive on failure *********************************************************************/ static int em_attach(device_t dev) { struct adapter *adapter; int error = 0; INIT_DEBUGOUT("em_attach: begin"); adapter = device_get_softc(dev); adapter->dev = adapter->osdep.dev = dev; EM_CORE_LOCK_INIT(adapter, device_get_nameunit(dev)); /* SYSCTL stuff */ SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_sysctl_nvm_info, "I", "NVM Information"); SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_sysctl_debug_info, "I", "Debug Information"); callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0); /* Determine hardware and mac info */ em_identify_hardware(adapter); /* Setup PCI resources */ if (em_allocate_pci_resources(adapter)) { device_printf(dev, "Allocation of PCI resources failed\n"); error = ENXIO; goto err_pci; } /* ** For ICH8 and family we need to ** map the flash memory, and this ** must happen after the MAC is ** identified */ if ((adapter->hw.mac.type == e1000_ich8lan) || (adapter->hw.mac.type == e1000_ich9lan) || (adapter->hw.mac.type == e1000_ich10lan) || (adapter->hw.mac.type == e1000_pchlan) || (adapter->hw.mac.type == e1000_pch2lan)) { int rid = EM_BAR_TYPE_FLASH; adapter->flash = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (adapter->flash == NULL) { device_printf(dev, "Mapping of Flash failed\n"); error = ENXIO; goto err_pci; } /* This is used in the shared code */ adapter->hw.flash_address = (u8 *)adapter->flash; adapter->osdep.flash_bus_space_tag = rman_get_bustag(adapter->flash); adapter->osdep.flash_bus_space_handle = rman_get_bushandle(adapter->flash); } /* Do Shared Code initialization */ if (e1000_setup_init_funcs(&adapter->hw, TRUE)) { device_printf(dev, "Setup of Shared code failed\n"); error = ENXIO; goto err_pci; } e1000_get_bus_info(&adapter->hw); /* Set up some sysctls for the tunable interrupt delays */ em_add_int_delay_sysctl(adapter, "rx_int_delay", "receive interrupt delay in usecs", &adapter->rx_int_delay, E1000_REGISTER(&adapter->hw, E1000_RDTR), em_rx_int_delay_dflt); em_add_int_delay_sysctl(adapter, "tx_int_delay", "transmit interrupt delay in usecs", &adapter->tx_int_delay, E1000_REGISTER(&adapter->hw, E1000_TIDV), em_tx_int_delay_dflt); em_add_int_delay_sysctl(adapter, "rx_abs_int_delay", "receive interrupt delay limit in usecs", &adapter->rx_abs_int_delay, E1000_REGISTER(&adapter->hw, E1000_RADV), em_rx_abs_int_delay_dflt); em_add_int_delay_sysctl(adapter, "tx_abs_int_delay", "transmit interrupt delay limit in usecs", &adapter->tx_abs_int_delay, E1000_REGISTER(&adapter->hw, E1000_TADV), em_tx_abs_int_delay_dflt); /* Sysctl for limiting the amount of work done in the taskqueue */ em_add_rx_process_limit(adapter, "rx_processing_limit", "max number of rx packets to process", &adapter->rx_process_limit, em_rx_process_limit); /* Sysctl for setting the interface flow control */ em_set_flow_cntrl(adapter, "flow_control", "configure flow control", &adapter->fc_setting, em_fc_setting); /* * Validate number of transmit and receive descriptors. It * must not exceed hardware maximum, and must be multiple * of E1000_DBA_ALIGN. */ if (((em_txd * sizeof(struct e1000_tx_desc)) % EM_DBA_ALIGN) != 0 || (em_txd > EM_MAX_TXD) || (em_txd < EM_MIN_TXD)) { device_printf(dev, "Using %d TX descriptors instead of %d!\n", EM_DEFAULT_TXD, em_txd); adapter->num_tx_desc = EM_DEFAULT_TXD; } else adapter->num_tx_desc = em_txd; if (((em_rxd * sizeof(struct e1000_rx_desc)) % EM_DBA_ALIGN) != 0 || (em_rxd > EM_MAX_RXD) || (em_rxd < EM_MIN_RXD)) { device_printf(dev, "Using %d RX descriptors instead of %d!\n", EM_DEFAULT_RXD, em_rxd); adapter->num_rx_desc = EM_DEFAULT_RXD; } else adapter->num_rx_desc = em_rxd; adapter->hw.mac.autoneg = DO_AUTO_NEG; adapter->hw.phy.autoneg_wait_to_complete = FALSE; adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT; /* Copper options */ if (adapter->hw.phy.media_type == e1000_media_type_copper) { adapter->hw.phy.mdix = AUTO_ALL_MODES; adapter->hw.phy.disable_polarity_correction = FALSE; adapter->hw.phy.ms_type = EM_MASTER_SLAVE; } /* * Set the frame limits assuming * standard ethernet sized frames. */ adapter->max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHERNET_FCS_SIZE; adapter->min_frame_size = ETH_ZLEN + ETHERNET_FCS_SIZE; /* * This controls when hardware reports transmit completion * status. */ adapter->hw.mac.report_tx_early = 1; /* ** Get queue/ring memory */ if (em_allocate_queues(adapter)) { error = ENOMEM; goto err_pci; } /* Allocate multicast array memory. */ adapter->mta = malloc(sizeof(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES, M_DEVBUF, M_NOWAIT); if (adapter->mta == NULL) { device_printf(dev, "Can not allocate multicast setup array\n"); error = ENOMEM; goto err_late; } /* Check SOL/IDER usage */ if (e1000_check_reset_block(&adapter->hw)) device_printf(dev, "PHY reset is blocked" " due to SOL/IDER session.\n"); /* ** Start from a known state, this is ** important in reading the nvm and ** mac from that. */ e1000_reset_hw(&adapter->hw); /* Make sure we have a good EEPROM before we read from it */ if (e1000_validate_nvm_checksum(&adapter->hw) < 0) { /* ** Some PCI-E parts fail the first check due to ** the link being in sleep state, call it again, ** if it fails a second time its a real issue. */ if (e1000_validate_nvm_checksum(&adapter->hw) < 0) { device_printf(dev, "The EEPROM Checksum Is Not Valid\n"); error = EIO; goto err_late; } } /* Copy the permanent MAC address out of the EEPROM */ if (e1000_read_mac_addr(&adapter->hw) < 0) { device_printf(dev, "EEPROM read error while reading MAC" " address\n"); error = EIO; goto err_late; } if (!em_is_valid_ether_addr(adapter->hw.mac.addr)) { device_printf(dev, "Invalid MAC address\n"); error = EIO; goto err_late; } /* ** Do interrupt configuration */ if (adapter->msix > 1) /* Do MSIX */ error = em_allocate_msix(adapter); else /* MSI or Legacy */ error = em_allocate_legacy(adapter); if (error) goto err_late; /* * Get Wake-on-Lan and Management info for later use */ em_get_wakeup(dev); /* Setup OS specific network interface */ if (em_setup_interface(dev, adapter) != 0) goto err_late; em_reset(adapter); /* Initialize statistics */ em_update_stats_counters(adapter); adapter->hw.mac.get_link_status = 1; em_update_link_status(adapter); /* Register for VLAN events */ adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, em_register_vlan, adapter, EVENTHANDLER_PRI_FIRST); adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, em_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST); em_add_hw_stats(adapter); /* Non-AMT based hardware can now take control from firmware */ if (adapter->has_manage && !adapter->has_amt) em_get_hw_control(adapter); /* Tell the stack that the interface is not active */ adapter->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); adapter->led_dev = led_create(em_led_func, adapter, device_get_nameunit(dev)); INIT_DEBUGOUT("em_attach: end"); return (0); err_late: em_free_transmit_structures(adapter); em_free_receive_structures(adapter); em_release_hw_control(adapter); if (adapter->ifp != NULL) if_free(adapter->ifp); err_pci: em_free_pci_resources(adapter); free(adapter->mta, M_DEVBUF); EM_CORE_LOCK_DESTROY(adapter); return (error); } /********************************************************************* * Device removal routine * * The detach entry point is called when the driver is being removed. * This routine stops the adapter and deallocates all the resources * that were allocated for driver operation. * * return 0 on success, positive on failure *********************************************************************/ static int em_detach(device_t dev) { struct adapter *adapter = device_get_softc(dev); struct ifnet *ifp = adapter->ifp; INIT_DEBUGOUT("em_detach: begin"); /* Make sure VLANS are not using driver */ if (adapter->ifp->if_vlantrunk != NULL) { device_printf(dev,"Vlan in use, detach first\n"); return (EBUSY); } #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) ether_poll_deregister(ifp); #endif if (adapter->led_dev != NULL) led_destroy(adapter->led_dev); EM_CORE_LOCK(adapter); adapter->in_detach = 1; em_stop(adapter); EM_CORE_UNLOCK(adapter); EM_CORE_LOCK_DESTROY(adapter); e1000_phy_hw_reset(&adapter->hw); em_release_manageability(adapter); em_release_hw_control(adapter); /* Unregister VLAN events */ if (adapter->vlan_attach != NULL) EVENTHANDLER_DEREGISTER(vlan_config, adapter->vlan_attach); if (adapter->vlan_detach != NULL) EVENTHANDLER_DEREGISTER(vlan_unconfig, adapter->vlan_detach); ether_ifdetach(adapter->ifp); callout_drain(&adapter->timer); em_free_pci_resources(adapter); bus_generic_detach(dev); if_free(ifp); em_free_transmit_structures(adapter); em_free_receive_structures(adapter); em_release_hw_control(adapter); free(adapter->mta, M_DEVBUF); return (0); } /********************************************************************* * * Shutdown entry point * **********************************************************************/ static int em_shutdown(device_t dev) { return em_suspend(dev); } /* * Suspend/resume device methods. */ static int em_suspend(device_t dev) { struct adapter *adapter = device_get_softc(dev); EM_CORE_LOCK(adapter); em_release_manageability(adapter); em_release_hw_control(adapter); em_enable_wakeup(dev); EM_CORE_UNLOCK(adapter); return bus_generic_suspend(dev); } static int em_resume(device_t dev) { struct adapter *adapter = device_get_softc(dev); struct ifnet *ifp = adapter->ifp; EM_CORE_LOCK(adapter); em_init_locked(adapter); em_init_manageability(adapter); EM_CORE_UNLOCK(adapter); em_start(ifp); return bus_generic_resume(dev); } /********************************************************************* * Transmit entry point * * em_start is called by the stack to initiate a transmit. * The driver will remain in this routine as long as there are * packets to transmit and transmit resources are available. * In case resources are not available stack is notified and * the packet is requeued. **********************************************************************/ static void em_start_locked(struct ifnet *ifp, struct tx_ring *txr) { struct adapter *adapter = ifp->if_softc; struct mbuf *m_head; EM_TX_LOCK_ASSERT(txr); if ((ifp->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return; if (!adapter->link_active) return; /* Call cleanup if number of TX descriptors low */ if (txr->tx_avail <= EM_TX_CLEANUP_THRESHOLD) em_txeof(txr); while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { if (txr->tx_avail < EM_MAX_SCATTER) { ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; /* * Encapsulation can modify our pointer, and or make it * NULL on failure. In that event, we can't requeue. */ if (em_xmit(txr, &m_head)) { if (m_head == NULL) break; ifp->if_drv_flags |= IFF_DRV_OACTIVE; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); break; } /* Send a copy of the frame to the BPF listener */ ETHER_BPF_MTAP(ifp, m_head); /* Set timeout in case hardware has problems transmitting. */ txr->watchdog_time = ticks; txr->queue_status = EM_QUEUE_WORKING; } return; } static void em_start(struct ifnet *ifp) { struct adapter *adapter = ifp->if_softc; struct tx_ring *txr = adapter->tx_rings; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { EM_TX_LOCK(txr); em_start_locked(ifp, txr); EM_TX_UNLOCK(txr); } return; } /********************************************************************* * Ioctl entry point * * em_ioctl is called when the user wants to configure the * interface. * * return 0 on success, positive on failure **********************************************************************/ static int em_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct adapter *adapter = ifp->if_softc; struct ifreq *ifr = (struct ifreq *)data; #ifdef INET struct ifaddr *ifa = (struct ifaddr *)data; #endif int error = 0; if (adapter->in_detach) return (error); switch (command) { case SIOCSIFADDR: #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) { /* * XXX * Since resetting hardware takes a very long time * and results in link renegotiation we only * initialize the hardware only when it is absolutely * required. */ ifp->if_flags |= IFF_UP; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) { EM_CORE_LOCK(adapter); em_init_locked(adapter); EM_CORE_UNLOCK(adapter); } arp_ifinit(ifp, ifa); } else #endif error = ether_ioctl(ifp, command, data); break; case SIOCSIFMTU: { int max_frame_size; IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)"); EM_CORE_LOCK(adapter); switch (adapter->hw.mac.type) { case e1000_82571: case e1000_82572: case e1000_ich9lan: case e1000_ich10lan: case e1000_pch2lan: case e1000_82574: case e1000_80003es2lan: /* 9K Jumbo Frame size */ max_frame_size = 9234; break; case e1000_pchlan: max_frame_size = 4096; break; /* Adapters that do not support jumbo frames */ case e1000_82583: case e1000_ich8lan: max_frame_size = ETHER_MAX_LEN; break; default: max_frame_size = MAX_JUMBO_FRAME_SIZE; } if (ifr->ifr_mtu > max_frame_size - ETHER_HDR_LEN - ETHER_CRC_LEN) { EM_CORE_UNLOCK(adapter); error = EINVAL; break; } ifp->if_mtu = ifr->ifr_mtu; adapter->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN; em_init_locked(adapter); EM_CORE_UNLOCK(adapter); break; } case SIOCSIFFLAGS: IOCTL_DEBUGOUT("ioctl rcv'd:\ SIOCSIFFLAGS (Set Interface Flags)"); EM_CORE_LOCK(adapter); if (ifp->if_flags & IFF_UP) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) { if ((ifp->if_flags ^ adapter->if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) { em_disable_promisc(adapter); em_set_promisc(adapter); } } else em_init_locked(adapter); } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) em_stop(adapter); adapter->if_flags = ifp->if_flags; EM_CORE_UNLOCK(adapter); break; case SIOCADDMULTI: case SIOCDELMULTI: IOCTL_DEBUGOUT("ioctl rcv'd: SIOC(ADD|DEL)MULTI"); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { EM_CORE_LOCK(adapter); em_disable_intr(adapter); em_set_multi(adapter); #ifdef DEVICE_POLLING if (!(ifp->if_capenable & IFCAP_POLLING)) #endif em_enable_intr(adapter); EM_CORE_UNLOCK(adapter); } break; case SIOCSIFMEDIA: /* ** As the speed/duplex settings are being ** changed, we need to reset the PHY. */ adapter->hw.phy.reset_disable = FALSE; /* Check SOL/IDER usage */ EM_CORE_LOCK(adapter); if (e1000_check_reset_block(&adapter->hw)) { EM_CORE_UNLOCK(adapter); device_printf(adapter->dev, "Media change is" " blocked due to SOL/IDER session.\n"); break; } EM_CORE_UNLOCK(adapter); /* falls thru */ case SIOCGIFMEDIA: IOCTL_DEBUGOUT("ioctl rcv'd: \ SIOCxIFMEDIA (Get/Set Interface Media)"); error = ifmedia_ioctl(ifp, ifr, &adapter->media, command); break; case SIOCSIFCAP: { int mask, reinit; IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFCAP (Set Capabilities)"); reinit = 0; mask = ifr->ifr_reqcap ^ ifp->if_capenable; #ifdef DEVICE_POLLING if (mask & IFCAP_POLLING) { if (ifr->ifr_reqcap & IFCAP_POLLING) { error = ether_poll_register(em_poll, ifp); if (error) return (error); EM_CORE_LOCK(adapter); em_disable_intr(adapter); ifp->if_capenable |= IFCAP_POLLING; EM_CORE_UNLOCK(adapter); } else { error = ether_poll_deregister(ifp); /* Enable interrupt even in error case */ EM_CORE_LOCK(adapter); em_enable_intr(adapter); ifp->if_capenable &= ~IFCAP_POLLING; EM_CORE_UNLOCK(adapter); } } #endif if (mask & IFCAP_HWCSUM) { ifp->if_capenable ^= IFCAP_HWCSUM; reinit = 1; } if (mask & IFCAP_TSO4) { ifp->if_capenable ^= IFCAP_TSO4; reinit = 1; } if (mask & IFCAP_VLAN_HWTAGGING) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; reinit = 1; } if (mask & IFCAP_VLAN_HWFILTER) { ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; reinit = 1; } if ((mask & IFCAP_WOL) && (ifp->if_capabilities & IFCAP_WOL) != 0) { if (mask & IFCAP_WOL_MCAST) ifp->if_capenable ^= IFCAP_WOL_MCAST; if (mask & IFCAP_WOL_MAGIC) ifp->if_capenable ^= IFCAP_WOL_MAGIC; } if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) em_init(adapter); VLAN_CAPABILITIES(ifp); break; } default: error = ether_ioctl(ifp, command, data); break; } return (error); } /********************************************************************* * Init entry point * * This routine is used in two ways. It is used by the stack as * init entry point in network interface structure. It is also used * by the driver as a hw/sw initialization routine to get to a * consistent state. * * return 0 on success, positive on failure **********************************************************************/ static void em_init_locked(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; device_t dev = adapter->dev; u32 pba; INIT_DEBUGOUT("em_init: begin"); EM_CORE_LOCK_ASSERT(adapter); em_disable_intr(adapter); callout_stop(&adapter->timer); /* * Packet Buffer Allocation (PBA) * Writing PBA sets the receive portion of the buffer * the remainder is used for the transmit buffer. */ switch (adapter->hw.mac.type) { /* Total Packet Buffer on these is 48K */ case e1000_82571: case e1000_82572: case e1000_80003es2lan: pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */ break; case e1000_82573: /* 82573: Total Packet Buffer is 32K */ pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */ break; case e1000_82574: case e1000_82583: pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */ break; case e1000_ich8lan: pba = E1000_PBA_8K; break; case e1000_ich9lan: case e1000_ich10lan: pba = E1000_PBA_10K; break; case e1000_pchlan: case e1000_pch2lan: pba = E1000_PBA_26K; break; default: if (adapter->max_frame_size > 8192) pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */ else pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */ } INIT_DEBUGOUT1("em_init: pba=%dK",pba); E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba); /* Get the latest mac address, User can use a LAA */ bcopy(IF_LLADDR(adapter->ifp), adapter->hw.mac.addr, ETHER_ADDR_LEN); /* Put the address into the Receive Address Array */ e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0); /* * With the 82571 adapter, RAR[0] may be overwritten * when the other port is reset, we make a duplicate * in RAR[14] for that eventuality, this assures * the interface continues to function. */ if (adapter->hw.mac.type == e1000_82571) { e1000_set_laa_state_82571(&adapter->hw, TRUE); e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, E1000_RAR_ENTRIES - 1); } /* Initialize the hardware */ em_reset(adapter); em_update_link_status(adapter); /* Setup VLAN support, basic and offload if available */ E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN); /* Set hardware offload abilities */ ifp->if_hwassist = 0; if (ifp->if_capenable & IFCAP_TXCSUM) ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP); if (ifp->if_capenable & IFCAP_TSO4) ifp->if_hwassist |= CSUM_TSO; /* Configure for OS presence */ em_init_manageability(adapter); /* Prepare transmit descriptors and buffers */ em_setup_transmit_structures(adapter); em_initialize_transmit_unit(adapter); /* Setup Multicast table */ em_set_multi(adapter); /* ** Figure out the desired mbuf ** pool for doing jumbos */ if (adapter->max_frame_size <= 2048) adapter->rx_mbuf_sz = MCLBYTES; else if (adapter->max_frame_size <= 4096) adapter->rx_mbuf_sz = MJUMPAGESIZE; else adapter->rx_mbuf_sz = MJUM9BYTES; /* Prepare receive descriptors and buffers */ if (em_setup_receive_structures(adapter)) { device_printf(dev, "Could not setup receive structures\n"); em_stop(adapter); return; } em_initialize_receive_unit(adapter); /* Use real VLAN Filter support? */ if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) /* Use real VLAN Filter support */ em_setup_vlan_hw_support(adapter); else { u32 ctrl; ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); ctrl |= E1000_CTRL_VME; E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); } } /* Don't lose promiscuous settings */ em_set_promisc(adapter); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; callout_reset(&adapter->timer, hz, em_local_timer, adapter); e1000_clear_hw_cntrs_base_generic(&adapter->hw); /* MSI/X configuration for 82574 */ if (adapter->hw.mac.type == e1000_82574) { int tmp; tmp = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); tmp |= E1000_CTRL_EXT_PBA_CLR; E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, tmp); /* Set the IVAR - interrupt vector routing. */ E1000_WRITE_REG(&adapter->hw, E1000_IVAR, adapter->ivars); } #ifdef DEVICE_POLLING /* * Only enable interrupts if we are not polling, make sure * they are off otherwise. */ if (ifp->if_capenable & IFCAP_POLLING) em_disable_intr(adapter); else #endif /* DEVICE_POLLING */ em_enable_intr(adapter); /* AMT based hardware can now take control from firmware */ if (adapter->has_manage && adapter->has_amt) em_get_hw_control(adapter); /* Don't reset the phy next time init gets called */ adapter->hw.phy.reset_disable = TRUE; } static void em_init(void *arg) { struct adapter *adapter = arg; EM_CORE_LOCK(adapter); em_init_locked(adapter); EM_CORE_UNLOCK(adapter); } #ifdef DEVICE_POLLING /********************************************************************* * * Legacy polling routine: note this only works with single queue * *********************************************************************/ static void em_poll(struct ifnet *ifp, enum poll_cmd cmd, int count) { struct adapter *adapter = ifp->if_softc; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; u32 reg_icr; int rx_done; EM_CORE_LOCK(adapter); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { EM_CORE_UNLOCK(adapter); return; } if (cmd == POLL_AND_CHECK_STATUS) { reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR); if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { callout_stop(&adapter->timer); adapter->hw.mac.get_link_status = 1; em_update_link_status(adapter); callout_reset(&adapter->timer, hz, em_local_timer, adapter); } } EM_CORE_UNLOCK(adapter); em_rxeof(rxr, count, &rx_done); EM_TX_LOCK(txr); em_txeof(txr); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) em_start_locked(ifp, txr); EM_TX_UNLOCK(txr); return; } #endif /* DEVICE_POLLING */ /********************************************************************* * * Fast Legacy/MSI Combined Interrupt Service routine * *********************************************************************/ static int em_irq_fast(void *arg) { struct adapter *adapter = arg; struct ifnet *ifp; u32 reg_icr; ifp = adapter->ifp; reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR); /* Hot eject? */ if (reg_icr == 0xffffffff) return FILTER_STRAY; /* Definitely not our interrupt. */ if (reg_icr == 0x0) return FILTER_STRAY; /* * Starting with the 82571 chip, bit 31 should be used to * determine whether the interrupt belongs to us. */ if (adapter->hw.mac.type >= e1000_82571 && (reg_icr & E1000_ICR_INT_ASSERTED) == 0) return FILTER_STRAY; em_disable_intr(adapter); taskqueue_enqueue(adapter->tq, &adapter->que_task); /* Link status change */ if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { adapter->hw.mac.get_link_status = 1; taskqueue_enqueue(taskqueue_fast, &adapter->link_task); } if (reg_icr & E1000_ICR_RXO) adapter->rx_overruns++; return FILTER_HANDLED; } /* Combined RX/TX handler, used by Legacy and MSI */ static void em_handle_que(void *context, int pending) { struct adapter *adapter = context; struct ifnet *ifp = adapter->ifp; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; bool more; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { more = em_rxeof(rxr, adapter->rx_process_limit, NULL); EM_TX_LOCK(txr); em_txeof(txr); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) em_start_locked(ifp, txr); em_txeof(txr); EM_TX_UNLOCK(txr); if (more) { taskqueue_enqueue(adapter->tq, &adapter->que_task); return; } } em_enable_intr(adapter); return; } /********************************************************************* * * MSIX Interrupt Service Routines * **********************************************************************/ static void em_msix_tx(void *arg) { struct tx_ring *txr = arg; struct adapter *adapter = txr->adapter; bool more; ++txr->tx_irq; EM_TX_LOCK(txr); more = em_txeof(txr); EM_TX_UNLOCK(txr); if (more) taskqueue_enqueue(txr->tq, &txr->tx_task); else /* Reenable this interrupt */ E1000_WRITE_REG(&adapter->hw, E1000_IMS, txr->ims); return; } /********************************************************************* * * MSIX RX Interrupt Service routine * **********************************************************************/ static void em_msix_rx(void *arg) { struct rx_ring *rxr = arg; struct adapter *adapter = rxr->adapter; bool more; ++rxr->rx_irq; more = em_rxeof(rxr, adapter->rx_process_limit, NULL); if (more) taskqueue_enqueue(rxr->tq, &rxr->rx_task); else /* Reenable this interrupt */ E1000_WRITE_REG(&adapter->hw, E1000_IMS, rxr->ims); return; } /********************************************************************* * * MSIX Link Fast Interrupt Service routine * **********************************************************************/ static void em_msix_link(void *arg) { struct adapter *adapter = arg; u32 reg_icr; ++adapter->link_irq; reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR); if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { adapter->hw.mac.get_link_status = 1; em_handle_link(adapter, 0); } else E1000_WRITE_REG(&adapter->hw, E1000_IMS, EM_MSIX_LINK | E1000_IMS_LSC); return; } static void em_handle_rx(void *context, int pending) { struct rx_ring *rxr = context; struct adapter *adapter = rxr->adapter; bool more; more = em_rxeof(rxr, adapter->rx_process_limit, NULL); if (more) taskqueue_enqueue(rxr->tq, &rxr->rx_task); else /* Reenable this interrupt */ E1000_WRITE_REG(&adapter->hw, E1000_IMS, rxr->ims); } static void em_handle_tx(void *context, int pending) { struct tx_ring *txr = context; struct adapter *adapter = txr->adapter; struct ifnet *ifp = adapter->ifp; EM_TX_LOCK(txr); em_txeof(txr); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) em_start_locked(ifp, txr); em_txeof(txr); E1000_WRITE_REG(&adapter->hw, E1000_IMS, txr->ims); EM_TX_UNLOCK(txr); } static void em_handle_link(void *context, int pending) { struct adapter *adapter = context; struct ifnet *ifp = adapter->ifp; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) return; EM_CORE_LOCK(adapter); callout_stop(&adapter->timer); em_update_link_status(adapter); callout_reset(&adapter->timer, hz, em_local_timer, adapter); E1000_WRITE_REG(&adapter->hw, E1000_IMS, EM_MSIX_LINK | E1000_IMS_LSC); EM_CORE_UNLOCK(adapter); } /********************************************************************* * * Media Ioctl callback * * This routine is called whenever the user queries the status of * the interface using ifconfig. * **********************************************************************/ static void em_media_status(struct ifnet *ifp, struct ifmediareq *ifmr) { struct adapter *adapter = ifp->if_softc; u_char fiber_type = IFM_1000_SX; INIT_DEBUGOUT("em_media_status: begin"); EM_CORE_LOCK(adapter); em_update_link_status(adapter); ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (!adapter->link_active) { EM_CORE_UNLOCK(adapter); return; } ifmr->ifm_status |= IFM_ACTIVE; if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) { ifmr->ifm_active |= fiber_type | IFM_FDX; } else { switch (adapter->link_speed) { case 10: ifmr->ifm_active |= IFM_10_T; break; case 100: ifmr->ifm_active |= IFM_100_TX; break; case 1000: ifmr->ifm_active |= IFM_1000_T; break; } if (adapter->link_duplex == FULL_DUPLEX) ifmr->ifm_active |= IFM_FDX; else ifmr->ifm_active |= IFM_HDX; } EM_CORE_UNLOCK(adapter); } /********************************************************************* * * Media Ioctl callback * * This routine is called when the user changes speed/duplex using * media/mediopt option with ifconfig. * **********************************************************************/ static int em_media_change(struct ifnet *ifp) { struct adapter *adapter = ifp->if_softc; struct ifmedia *ifm = &adapter->media; INIT_DEBUGOUT("em_media_change: begin"); if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); EM_CORE_LOCK(adapter); switch (IFM_SUBTYPE(ifm->ifm_media)) { case IFM_AUTO: adapter->hw.mac.autoneg = DO_AUTO_NEG; adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT; break; case IFM_1000_LX: case IFM_1000_SX: case IFM_1000_T: adapter->hw.mac.autoneg = DO_AUTO_NEG; adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL; break; case IFM_100_TX: adapter->hw.mac.autoneg = FALSE; adapter->hw.phy.autoneg_advertised = 0; if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL; else adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF; break; case IFM_10_T: adapter->hw.mac.autoneg = FALSE; adapter->hw.phy.autoneg_advertised = 0; if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL; else adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF; break; default: device_printf(adapter->dev, "Unsupported media type\n"); } em_init_locked(adapter); EM_CORE_UNLOCK(adapter); return (0); } /********************************************************************* * * This routine maps the mbufs to tx descriptors. * * return 0 on success, positive on failure **********************************************************************/ static int em_xmit(struct tx_ring *txr, struct mbuf **m_headp) { struct adapter *adapter = txr->adapter; bus_dma_segment_t segs[EM_MAX_SCATTER]; bus_dmamap_t map; struct em_buffer *tx_buffer, *tx_buffer_mapped; struct e1000_tx_desc *ctxd = NULL; struct mbuf *m_head; struct ether_header *eh; struct ip *ip = NULL; struct tcphdr *tp = NULL; u32 txd_upper, txd_lower, txd_used, txd_saved; int ip_off, poff; int nsegs, i, j, first, last = 0; int error, do_tso, tso_desc = 0; m_head = *m_headp; txd_upper = txd_lower = txd_used = txd_saved = 0; do_tso = ((m_head->m_pkthdr.csum_flags & CSUM_TSO) != 0); ip_off = poff = 0; /* * Intel recommends entire IP/TCP header length reside in a single * buffer. If multiple descriptors are used to describe the IP and * TCP header, each descriptor should describe one or more * complete headers; descriptors referencing only parts of headers * are not supported. If all layer headers are not coalesced into * a single buffer, each buffer should not cross a 4KB boundary, * or be larger than the maximum read request size. * Controller also requires modifing IP/TCP header to make TSO work * so we firstly get a writable mbuf chain then coalesce ethernet/ * IP/TCP header into a single buffer to meet the requirement of * controller. This also simplifies IP/TCP/UDP checksum offloading * which also has similiar restrictions. */ if (do_tso || m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD) { if (do_tso || (m_head->m_next != NULL && m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD)) { if (M_WRITABLE(*m_headp) == 0) { m_head = m_dup(*m_headp, M_DONTWAIT); m_freem(*m_headp); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } *m_headp = m_head; } } /* * XXX * Assume IPv4, we don't have TSO/checksum offload support * for IPv6 yet. */ ip_off = sizeof(struct ether_header); m_head = m_pullup(m_head, ip_off); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } eh = mtod(m_head, struct ether_header *); if (eh->ether_type == htons(ETHERTYPE_VLAN)) { ip_off = sizeof(struct ether_vlan_header); m_head = m_pullup(m_head, ip_off); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } } m_head = m_pullup(m_head, ip_off + sizeof(struct ip)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m_head, char *) + ip_off); poff = ip_off + (ip->ip_hl << 2); if (do_tso) { m_head = m_pullup(m_head, poff + sizeof(struct tcphdr)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } tp = (struct tcphdr *)(mtod(m_head, char *) + poff); /* * TSO workaround: * pull 4 more bytes of data into it. */ m_head = m_pullup(m_head, poff + (tp->th_off << 2) + 4); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m_head, char *) + ip_off); ip->ip_len = 0; ip->ip_sum = 0; /* * The pseudo TCP checksum does not include TCP payload * length so driver should recompute the checksum here * what hardware expect to see. This is adherence of * Microsoft's Large Send specification. */ tp = (struct tcphdr *)(mtod(m_head, char *) + poff); tp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); } else if (m_head->m_pkthdr.csum_flags & CSUM_TCP) { m_head = m_pullup(m_head, poff + sizeof(struct tcphdr)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } tp = (struct tcphdr *)(mtod(m_head, char *) + poff); m_head = m_pullup(m_head, poff + (tp->th_off << 2)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m_head, char *) + ip_off); tp = (struct tcphdr *)(mtod(m_head, char *) + poff); } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) { m_head = m_pullup(m_head, poff + sizeof(struct udphdr)); if (m_head == NULL) { *m_headp = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m_head, char *) + ip_off); } *m_headp = m_head; } /* * Map the packet for DMA * * Capture the first descriptor index, * this descriptor will have the index * of the EOP which is the only one that * now gets a DONE bit writeback. */ first = txr->next_avail_desc; tx_buffer = &txr->tx_buffers[first]; tx_buffer_mapped = tx_buffer; map = tx_buffer->map; error = bus_dmamap_load_mbuf_sg(txr->txtag, map, *m_headp, segs, &nsegs, BUS_DMA_NOWAIT); /* * There are two types of errors we can (try) to handle: * - EFBIG means the mbuf chain was too long and bus_dma ran * out of segments. Defragment the mbuf chain and try again. * - ENOMEM means bus_dma could not obtain enough bounce buffers * at this point in time. Defer sending and try again later. * All other errors, in particular EINVAL, are fatal and prevent the * mbuf chain from ever going through. Drop it and report error. */ if (error == EFBIG) { struct mbuf *m; m = m_defrag(*m_headp, M_DONTWAIT); if (m == NULL) { adapter->mbuf_alloc_failed++; m_freem(*m_headp); *m_headp = NULL; return (ENOBUFS); } *m_headp = m; /* Try it again */ error = bus_dmamap_load_mbuf_sg(txr->txtag, map, *m_headp, segs, &nsegs, BUS_DMA_NOWAIT); if (error == ENOMEM) { adapter->no_tx_dma_setup++; return (error); } else if (error != 0) { adapter->no_tx_dma_setup++; m_freem(*m_headp); *m_headp = NULL; return (error); } } else if (error == ENOMEM) { adapter->no_tx_dma_setup++; return (error); } else if (error != 0) { adapter->no_tx_dma_setup++; m_freem(*m_headp); *m_headp = NULL; return (error); } /* * TSO Hardware workaround, if this packet is not * TSO, and is only a single descriptor long, and * it follows a TSO burst, then we need to add a * sentinel descriptor to prevent premature writeback. */ if ((do_tso == 0) && (txr->tx_tso == TRUE)) { if (nsegs == 1) tso_desc = TRUE; txr->tx_tso = FALSE; } if (nsegs > (txr->tx_avail - 2)) { txr->no_desc_avail++; bus_dmamap_unload(txr->txtag, map); return (ENOBUFS); } m_head = *m_headp; /* Do hardware assists */ if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { em_tso_setup(txr, m_head, ip_off, ip, tp, &txd_upper, &txd_lower); /* we need to make a final sentinel transmit desc */ tso_desc = TRUE; } else if (m_head->m_pkthdr.csum_flags & CSUM_OFFLOAD) em_transmit_checksum_setup(txr, m_head, ip_off, ip, &txd_upper, &txd_lower); i = txr->next_avail_desc; /* Set up our transmit descriptors */ for (j = 0; j < nsegs; j++) { bus_size_t seg_len; bus_addr_t seg_addr; tx_buffer = &txr->tx_buffers[i]; ctxd = &txr->tx_base[i]; seg_addr = segs[j].ds_addr; seg_len = segs[j].ds_len; /* ** TSO Workaround: ** If this is the last descriptor, we want to ** split it so we have a small final sentinel */ if (tso_desc && (j == (nsegs -1)) && (seg_len > 8)) { seg_len -= 4; ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32( adapter->txd_cmd | txd_lower | seg_len); ctxd->upper.data = htole32(txd_upper); if (++i == adapter->num_tx_desc) i = 0; /* Now make the sentinel */ ++txd_used; /* using an extra txd */ ctxd = &txr->tx_base[i]; tx_buffer = &txr->tx_buffers[i]; ctxd->buffer_addr = htole64(seg_addr + seg_len); ctxd->lower.data = htole32( adapter->txd_cmd | txd_lower | 4); ctxd->upper.data = htole32(txd_upper); last = i; if (++i == adapter->num_tx_desc) i = 0; } else { ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32( adapter->txd_cmd | txd_lower | seg_len); ctxd->upper.data = htole32(txd_upper); last = i; if (++i == adapter->num_tx_desc) i = 0; } tx_buffer->m_head = NULL; tx_buffer->next_eop = -1; } txr->next_avail_desc = i; txr->tx_avail -= nsegs; if (tso_desc) /* TSO used an extra for sentinel */ txr->tx_avail -= txd_used; if (m_head->m_flags & M_VLANTAG) { /* Set the vlan id. */ ctxd->upper.fields.special = htole16(m_head->m_pkthdr.ether_vtag); /* Tell hardware to add tag */ ctxd->lower.data |= htole32(E1000_TXD_CMD_VLE); } tx_buffer->m_head = m_head; tx_buffer_mapped->map = tx_buffer->map; tx_buffer->map = map; bus_dmamap_sync(txr->txtag, map, BUS_DMASYNC_PREWRITE); /* * Last Descriptor of Packet * needs End Of Packet (EOP) * and Report Status (RS) */ ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS); /* * Keep track in the first buffer which * descriptor will be written back */ tx_buffer = &txr->tx_buffers[first]; tx_buffer->next_eop = last; /* Update the watchdog time early and often */ txr->watchdog_time = ticks; /* * Advance the Transmit Descriptor Tail (TDT), this tells the E1000 * that this frame is available to transmit. */ bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), i); return (0); } static void em_set_promisc(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; u32 reg_rctl; reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); if (ifp->if_flags & IFF_PROMISC) { reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE); /* Turn this on if you want to see bad packets */ if (em_debug_sbp) reg_rctl |= E1000_RCTL_SBP; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } else if (ifp->if_flags & IFF_ALLMULTI) { reg_rctl |= E1000_RCTL_MPE; reg_rctl &= ~E1000_RCTL_UPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } } static void em_disable_promisc(struct adapter *adapter) { u32 reg_rctl; reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl &= (~E1000_RCTL_UPE); reg_rctl &= (~E1000_RCTL_MPE); reg_rctl &= (~E1000_RCTL_SBP); E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } /********************************************************************* * Multicast Update * * This routine is called whenever multicast address list is updated. * **********************************************************************/ static void em_set_multi(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; struct ifmultiaddr *ifma; u32 reg_rctl = 0; u8 *mta; /* Multicast array memory */ int mcnt = 0; IOCTL_DEBUGOUT("em_set_multi: begin"); mta = adapter->mta; bzero(mta, sizeof(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES); if (adapter->hw.mac.type == e1000_82542 && adapter->hw.revision_id == E1000_REVISION_2) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE) e1000_pci_clear_mwi(&adapter->hw); reg_rctl |= E1000_RCTL_RST; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); msec_delay(5); } IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; if (mcnt == MAX_NUM_MULTICAST_ADDRESSES) break; bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), &mta[mcnt * ETH_ADDR_LEN], ETH_ADDR_LEN); mcnt++; } IF_ADDR_UNLOCK(ifp); if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl |= E1000_RCTL_MPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } else e1000_update_mc_addr_list(&adapter->hw, mta, mcnt); if (adapter->hw.mac.type == e1000_82542 && adapter->hw.revision_id == E1000_REVISION_2) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl &= ~E1000_RCTL_RST; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); msec_delay(5); if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE) e1000_pci_set_mwi(&adapter->hw); } } /********************************************************************* * Timer routine * * This routine checks for link status and updates statistics. * **********************************************************************/ static void em_local_timer(void *arg) { struct adapter *adapter = arg; struct ifnet *ifp = adapter->ifp; struct tx_ring *txr = adapter->tx_rings; EM_CORE_LOCK_ASSERT(adapter); em_update_link_status(adapter); em_update_stats_counters(adapter); /* Reset LAA into RAR[0] on 82571 */ if ((adapter->hw.mac.type == e1000_82571) && e1000_get_laa_state_82571(&adapter->hw)) e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0); /* ** Don't do TX watchdog check if we've been paused */ if (adapter->pause_frames) { adapter->pause_frames = 0; goto out; } /* ** Check on the state of the TX queue(s), this ** can be done without the lock because its RO ** and the HUNG state will be static if set. */ for (int i = 0; i < adapter->num_queues; i++, txr++) if (txr->queue_status == EM_QUEUE_HUNG) goto hung; out: callout_reset(&adapter->timer, hz, em_local_timer, adapter); return; hung: /* Looks like we're hung */ device_printf(adapter->dev, "Watchdog timeout -- resetting\n"); device_printf(adapter->dev, "Queue(%d) tdh = %d, hw tdt = %d\n", txr->me, E1000_READ_REG(&adapter->hw, E1000_TDH(txr->me)), E1000_READ_REG(&adapter->hw, E1000_TDT(txr->me))); device_printf(adapter->dev,"TX(%d) desc avail = %d," "Next TX to Clean = %d\n", txr->me, txr->tx_avail, txr->next_to_clean); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; adapter->watchdog_events++; em_init_locked(adapter); } static void em_update_link_status(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; struct ifnet *ifp = adapter->ifp; device_t dev = adapter->dev; struct tx_ring *txr = adapter->tx_rings; u32 link_check = 0; /* Get the cached link value or read phy for real */ switch (hw->phy.media_type) { case e1000_media_type_copper: if (hw->mac.get_link_status) { /* Do the work to read phy */ e1000_check_for_link(hw); link_check = !hw->mac.get_link_status; if (link_check) /* ESB2 fix */ e1000_cfg_on_link_up(hw); } else link_check = TRUE; break; case e1000_media_type_fiber: e1000_check_for_link(hw); link_check = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU); break; case e1000_media_type_internal_serdes: e1000_check_for_link(hw); link_check = adapter->hw.mac.serdes_has_link; break; default: case e1000_media_type_unknown: break; } /* Now check for a transition */ if (link_check && (adapter->link_active == 0)) { e1000_get_speed_and_duplex(hw, &adapter->link_speed, &adapter->link_duplex); /* Check if we must disable SPEED_MODE bit on PCI-E */ if ((adapter->link_speed != SPEED_1000) && ((hw->mac.type == e1000_82571) || (hw->mac.type == e1000_82572))) { int tarc0; tarc0 = E1000_READ_REG(hw, E1000_TARC(0)); tarc0 &= ~SPEED_MODE_BIT; E1000_WRITE_REG(hw, E1000_TARC(0), tarc0); } if (bootverbose) device_printf(dev, "Link is up %d Mbps %s\n", adapter->link_speed, ((adapter->link_duplex == FULL_DUPLEX) ? "Full Duplex" : "Half Duplex")); adapter->link_active = 1; adapter->smartspeed = 0; ifp->if_baudrate = adapter->link_speed * 1000000; if_link_state_change(ifp, LINK_STATE_UP); } else if (!link_check && (adapter->link_active == 1)) { ifp->if_baudrate = adapter->link_speed = 0; adapter->link_duplex = 0; if (bootverbose) device_printf(dev, "Link is Down\n"); adapter->link_active = 0; /* Link down, disable watchdog */ for (int i = 0; i < adapter->num_queues; i++, txr++) txr->queue_status = EM_QUEUE_IDLE; if_link_state_change(ifp, LINK_STATE_DOWN); } } /********************************************************************* * * This routine disables all traffic on the adapter by issuing a * global reset on the MAC and deallocates TX/RX buffers. * * This routine should always be called with BOTH the CORE * and TX locks. **********************************************************************/ static void em_stop(void *arg) { struct adapter *adapter = arg; struct ifnet *ifp = adapter->ifp; struct tx_ring *txr = adapter->tx_rings; EM_CORE_LOCK_ASSERT(adapter); INIT_DEBUGOUT("em_stop: begin"); em_disable_intr(adapter); callout_stop(&adapter->timer); /* Tell the stack that the interface is no longer active */ ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); /* Unarm watchdog timer. */ for (int i = 0; i < adapter->num_queues; i++, txr++) { EM_TX_LOCK(txr); txr->queue_status = EM_QUEUE_IDLE; EM_TX_UNLOCK(txr); } e1000_reset_hw(&adapter->hw); E1000_WRITE_REG(&adapter->hw, E1000_WUC, 0); e1000_led_off(&adapter->hw); e1000_cleanup_led(&adapter->hw); } /********************************************************************* * * Determine hardware revision. * **********************************************************************/ static void em_identify_hardware(struct adapter *adapter) { device_t dev = adapter->dev; /* Make sure our PCI config space has the necessary stuff set */ adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2); if (!((adapter->hw.bus.pci_cmd_word & PCIM_CMD_BUSMASTEREN) && (adapter->hw.bus.pci_cmd_word & PCIM_CMD_MEMEN))) { device_printf(dev, "Memory Access and/or Bus Master bits " "were not set!\n"); adapter->hw.bus.pci_cmd_word |= (PCIM_CMD_BUSMASTEREN | PCIM_CMD_MEMEN); pci_write_config(dev, PCIR_COMMAND, adapter->hw.bus.pci_cmd_word, 2); } /* Save off the information about this board */ adapter->hw.vendor_id = pci_get_vendor(dev); adapter->hw.device_id = pci_get_device(dev); adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1); adapter->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2); adapter->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2); /* Do Shared Code Init and Setup */ if (e1000_set_mac_type(&adapter->hw)) { device_printf(dev, "Setup init failure\n"); return; } } static int em_allocate_pci_resources(struct adapter *adapter) { device_t dev = adapter->dev; int rid; rid = PCIR_BAR(0); adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (adapter->memory == NULL) { device_printf(dev, "Unable to allocate bus resource: memory\n"); return (ENXIO); } adapter->osdep.mem_bus_space_tag = rman_get_bustag(adapter->memory); adapter->osdep.mem_bus_space_handle = rman_get_bushandle(adapter->memory); adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle; /* Default to a single queue */ adapter->num_queues = 1; /* * Setup MSI/X or MSI if PCI Express */ adapter->msix = em_setup_msix(adapter); adapter->hw.back = &adapter->osdep; return (0); } /********************************************************************* * * Setup the Legacy or MSI Interrupt handler * **********************************************************************/ int em_allocate_legacy(struct adapter *adapter) { device_t dev = adapter->dev; int error, rid = 0; /* Manually turn off all interrupts */ E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff); if (adapter->msix == 1) /* using MSI */ rid = 1; /* We allocate a single interrupt resource */ adapter->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE); if (adapter->res == NULL) { device_printf(dev, "Unable to allocate bus resource: " "interrupt\n"); return (ENXIO); } /* * Allocate a fast interrupt and the associated * deferred processing contexts. */ TASK_INIT(&adapter->que_task, 0, em_handle_que, adapter); TASK_INIT(&adapter->link_task, 0, em_handle_link, adapter); adapter->tq = taskqueue_create_fast("em_taskq", M_NOWAIT, taskqueue_thread_enqueue, &adapter->tq); taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s taskq", device_get_nameunit(adapter->dev)); if ((error = bus_setup_intr(dev, adapter->res, INTR_TYPE_NET, em_irq_fast, NULL, adapter, &adapter->tag)) != 0) { device_printf(dev, "Failed to register fast interrupt " "handler: %d\n", error); taskqueue_free(adapter->tq); adapter->tq = NULL; return (error); } return (0); } /********************************************************************* * * Setup the MSIX Interrupt handlers * This is not really Multiqueue, rather * its just multiple interrupt vectors. * **********************************************************************/ int em_allocate_msix(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; int error, rid, vector = 0; /* Make sure all interrupts are disabled */ E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff); /* First set up ring resources */ for (int i = 0; i < adapter->num_queues; i++, txr++, rxr++) { /* RX ring */ rid = vector + 1; rxr->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_ACTIVE); if (rxr->res == NULL) { device_printf(dev, "Unable to allocate bus resource: " "RX MSIX Interrupt %d\n", i); return (ENXIO); } if ((error = bus_setup_intr(dev, rxr->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, em_msix_rx, rxr, &rxr->tag)) != 0) { device_printf(dev, "Failed to register RX handler"); return (error); } rxr->msix = vector++; /* NOTE increment vector for TX */ TASK_INIT(&rxr->rx_task, 0, em_handle_rx, rxr); rxr->tq = taskqueue_create_fast("em_rxq", M_NOWAIT, taskqueue_thread_enqueue, &rxr->tq); taskqueue_start_threads(&rxr->tq, 1, PI_NET, "%s rxq", device_get_nameunit(adapter->dev)); /* ** Set the bit to enable interrupt ** in E1000_IMS -- bits 20 and 21 ** are for RX0 and RX1, note this has ** NOTHING to do with the MSIX vector */ rxr->ims = 1 << (20 + i); adapter->ivars |= (8 | rxr->msix) << (i * 4); /* TX ring */ rid = vector + 1; txr->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_ACTIVE); if (txr->res == NULL) { device_printf(dev, "Unable to allocate bus resource: " "TX MSIX Interrupt %d\n", i); return (ENXIO); } if ((error = bus_setup_intr(dev, txr->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, em_msix_tx, txr, &txr->tag)) != 0) { device_printf(dev, "Failed to register TX handler"); return (error); } txr->msix = vector++; /* Increment vector for next pass */ TASK_INIT(&txr->tx_task, 0, em_handle_tx, txr); txr->tq = taskqueue_create_fast("em_txq", M_NOWAIT, taskqueue_thread_enqueue, &txr->tq); taskqueue_start_threads(&txr->tq, 1, PI_NET, "%s txq", device_get_nameunit(adapter->dev)); /* ** Set the bit to enable interrupt ** in E1000_IMS -- bits 22 and 23 ** are for TX0 and TX1, note this has ** NOTHING to do with the MSIX vector */ txr->ims = 1 << (22 + i); adapter->ivars |= (8 | txr->msix) << (8 + (i * 4)); } /* Link interrupt */ ++rid; adapter->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_ACTIVE); if (!adapter->res) { device_printf(dev,"Unable to allocate " "bus resource: Link interrupt [%d]\n", rid); return (ENXIO); } /* Set the link handler function */ error = bus_setup_intr(dev, adapter->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, em_msix_link, adapter, &adapter->tag); if (error) { adapter->res = NULL; device_printf(dev, "Failed to register LINK handler"); return (error); } adapter->linkvec = vector; adapter->ivars |= (8 | vector) << 16; adapter->ivars |= 0x80000000; return (0); } static void em_free_pci_resources(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr; struct rx_ring *rxr; int rid; /* ** Release all the queue interrupt resources: */ for (int i = 0; i < adapter->num_queues; i++) { txr = &adapter->tx_rings[i]; rxr = &adapter->rx_rings[i]; /* an early abort? */ if ((txr == NULL) || (rxr == NULL)) break; rid = txr->msix +1; if (txr->tag != NULL) { bus_teardown_intr(dev, txr->res, txr->tag); txr->tag = NULL; } if (txr->res != NULL) bus_release_resource(dev, SYS_RES_IRQ, rid, txr->res); rid = rxr->msix +1; if (rxr->tag != NULL) { bus_teardown_intr(dev, rxr->res, rxr->tag); rxr->tag = NULL; } if (rxr->res != NULL) bus_release_resource(dev, SYS_RES_IRQ, rid, rxr->res); } if (adapter->linkvec) /* we are doing MSIX */ rid = adapter->linkvec + 1; else (adapter->msix != 0) ? (rid = 1):(rid = 0); if (adapter->tag != NULL) { bus_teardown_intr(dev, adapter->res, adapter->tag); adapter->tag = NULL; } if (adapter->res != NULL) bus_release_resource(dev, SYS_RES_IRQ, rid, adapter->res); if (adapter->msix) pci_release_msi(dev); if (adapter->msix_mem != NULL) bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(EM_MSIX_BAR), adapter->msix_mem); if (adapter->memory != NULL) bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(0), adapter->memory); if (adapter->flash != NULL) bus_release_resource(dev, SYS_RES_MEMORY, EM_FLASH, adapter->flash); } /* * Setup MSI or MSI/X */ static int em_setup_msix(struct adapter *adapter) { device_t dev = adapter->dev; int val = 0; /* ** Setup MSI/X for Hartwell: tests have shown ** use of two queues to be unstable, and to ** provide no great gain anyway, so we simply ** seperate the interrupts and use a single queue. */ if ((adapter->hw.mac.type == e1000_82574) && (em_enable_msix == TRUE)) { /* Map the MSIX BAR */ int rid = PCIR_BAR(EM_MSIX_BAR); adapter->msix_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (!adapter->msix_mem) { /* May not be enabled */ device_printf(adapter->dev, "Unable to map MSIX table \n"); goto msi; } val = pci_msix_count(dev); if (val < 3) { bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(EM_MSIX_BAR), adapter->msix_mem); adapter->msix_mem = NULL; device_printf(adapter->dev, "MSIX: insufficient vectors, using MSI\n"); goto msi; } val = 3; adapter->num_queues = 1; if (pci_alloc_msix(dev, &val) == 0) { device_printf(adapter->dev, "Using MSIX interrupts " "with %d vectors\n", val); } return (val); } msi: val = pci_msi_count(dev); if (val == 1 && pci_alloc_msi(dev, &val) == 0) { adapter->msix = 1; device_printf(adapter->dev,"Using an MSI interrupt\n"); return (val); } /* Should only happen due to manual configuration */ device_printf(adapter->dev,"No MSI/MSIX using a Legacy IRQ\n"); return (0); } /********************************************************************* * * Initialize the hardware to a configuration * as specified by the adapter structure. * **********************************************************************/ static void em_reset(struct adapter *adapter) { device_t dev = adapter->dev; struct ifnet *ifp = adapter->ifp; struct e1000_hw *hw = &adapter->hw; u16 rx_buffer_size; INIT_DEBUGOUT("em_reset: begin"); /* Set up smart power down as default off on newer adapters. */ if (!em_smart_pwr_down && (hw->mac.type == e1000_82571 || hw->mac.type == e1000_82572)) { u16 phy_tmp = 0; /* Speed up time to link by disabling smart power down. */ e1000_read_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, &phy_tmp); phy_tmp &= ~IGP02E1000_PM_SPD; e1000_write_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, phy_tmp); } /* * These parameters control the automatic generation (Tx) and * response (Rx) to Ethernet PAUSE frames. * - High water mark should allow for at least two frames to be * received after sending an XOFF. * - Low water mark works best when it is very near the high water mark. * This allows the receiver to restart by sending XON when it has * drained a bit. Here we use an arbitary value of 1500 which will * restart after one full frame is pulled from the buffer. There * could be several smaller frames in the buffer and if so they will * not trigger the XON until their total number reduces the buffer * by 1500. * - The pause time is fairly large at 1000 x 512ns = 512 usec. */ rx_buffer_size = ((E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10 ); hw->fc.high_water = rx_buffer_size - roundup2(adapter->max_frame_size, 1024); hw->fc.low_water = hw->fc.high_water - 1500; if (hw->mac.type == e1000_80003es2lan) hw->fc.pause_time = 0xFFFF; else hw->fc.pause_time = EM_FC_PAUSE_TIME; hw->fc.send_xon = TRUE; /* Set Flow control, use the tunable location if sane */ hw->fc.requested_mode = adapter->fc_setting; /* Workaround: no TX flow ctrl for PCH */ if (hw->mac.type == e1000_pchlan) hw->fc.requested_mode = e1000_fc_rx_pause; /* Override - settings for PCH2LAN, ya its magic :) */ if (hw->mac.type == e1000_pch2lan) { hw->fc.high_water = 0x5C20; hw->fc.low_water = 0x5048; hw->fc.pause_time = 0x0650; hw->fc.refresh_time = 0x0400; /* Jumbos need adjusted PBA */ if (ifp->if_mtu > ETHERMTU) E1000_WRITE_REG(hw, E1000_PBA, 12); else E1000_WRITE_REG(hw, E1000_PBA, 26); } /* Issue a global reset */ e1000_reset_hw(hw); E1000_WRITE_REG(hw, E1000_WUC, 0); em_disable_aspm(adapter); if (e1000_init_hw(hw) < 0) { device_printf(dev, "Hardware Initialization Failed\n"); return; } E1000_WRITE_REG(hw, E1000_VET, ETHERTYPE_VLAN); e1000_get_phy_info(hw); e1000_check_for_link(hw); return; } /********************************************************************* * * Setup networking device structure and register an interface. * **********************************************************************/ static int em_setup_interface(device_t dev, struct adapter *adapter) { struct ifnet *ifp; INIT_DEBUGOUT("em_setup_interface: begin"); ifp = adapter->ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "can not allocate ifnet structure\n"); return (-1); } if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_mtu = ETHERMTU; ifp->if_init = em_init; ifp->if_softc = adapter; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = em_ioctl; ifp->if_start = em_start; IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 1); ifp->if_snd.ifq_drv_maxlen = adapter->num_tx_desc - 1; IFQ_SET_READY(&ifp->if_snd); ether_ifattach(ifp, adapter->hw.mac.addr); ifp->if_capabilities = ifp->if_capenable = 0; ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM; ifp->if_capenable |= IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM; /* Enable TSO by default, can disable with ifconfig */ ifp->if_capabilities |= IFCAP_TSO4; ifp->if_capenable |= IFCAP_TSO4; /* * Tell the upper layer(s) we * support full VLAN capability */ ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU; ifp->if_capenable |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU; /* ** Dont turn this on by default, if vlans are ** created on another pseudo device (eg. lagg) ** then vlan events are not passed thru, breaking ** operation, but with HW FILTER off it works. If ** using vlans directly on the em driver you can ** enable this and get full hardware tag filtering. */ ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; #ifdef DEVICE_POLLING ifp->if_capabilities |= IFCAP_POLLING; #endif /* Enable only WOL MAGIC by default */ if (adapter->wol) { ifp->if_capabilities |= IFCAP_WOL; ifp->if_capenable |= IFCAP_WOL_MAGIC; } /* * Specify the media types supported by this adapter and register * callbacks to update media and link information */ ifmedia_init(&adapter->media, IFM_IMASK, em_media_change, em_media_status); if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) { u_char fiber_type = IFM_1000_SX; /* default type */ ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL); } else { ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX, 0, NULL); if (adapter->hw.phy.type != e1000_phy_ife) { ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T, 0, NULL); } } ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO); return (0); } /* * Manage DMA'able memory. */ static void em_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) { if (error) return; *(bus_addr_t *) arg = segs[0].ds_addr; } static int em_dma_malloc(struct adapter *adapter, bus_size_t size, struct em_dma_alloc *dma, int mapflags) { int error; error = bus_dma_tag_create(bus_get_dma_tag(adapter->dev), /* parent */ EM_DBA_ALIGN, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ size, /* maxsize */ 1, /* nsegments */ size, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &dma->dma_tag); if (error) { device_printf(adapter->dev, "%s: bus_dma_tag_create failed: %d\n", __func__, error); goto fail_0; } error = bus_dmamem_alloc(dma->dma_tag, (void**) &dma->dma_vaddr, BUS_DMA_NOWAIT | BUS_DMA_COHERENT, &dma->dma_map); if (error) { device_printf(adapter->dev, "%s: bus_dmamem_alloc(%ju) failed: %d\n", __func__, (uintmax_t)size, error); goto fail_2; } dma->dma_paddr = 0; error = bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr, size, em_dmamap_cb, &dma->dma_paddr, mapflags | BUS_DMA_NOWAIT); if (error || dma->dma_paddr == 0) { device_printf(adapter->dev, "%s: bus_dmamap_load failed: %d\n", __func__, error); goto fail_3; } return (0); fail_3: bus_dmamap_unload(dma->dma_tag, dma->dma_map); fail_2: bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); bus_dma_tag_destroy(dma->dma_tag); fail_0: dma->dma_map = NULL; dma->dma_tag = NULL; return (error); } static void em_dma_free(struct adapter *adapter, struct em_dma_alloc *dma) { if (dma->dma_tag == NULL) return; if (dma->dma_map != NULL) { bus_dmamap_sync(dma->dma_tag, dma->dma_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(dma->dma_tag, dma->dma_map); bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map); dma->dma_map = NULL; } bus_dma_tag_destroy(dma->dma_tag); dma->dma_tag = NULL; } /********************************************************************* * * Allocate memory for the transmit and receive rings, and then * the descriptors associated with each, called only once at attach. * **********************************************************************/ static int em_allocate_queues(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr = NULL; struct rx_ring *rxr = NULL; int rsize, tsize, error = E1000_SUCCESS; int txconf = 0, rxconf = 0; /* Allocate the TX ring struct memory */ if (!(adapter->tx_rings = (struct tx_ring *) malloc(sizeof(struct tx_ring) * adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate TX ring memory\n"); error = ENOMEM; goto fail; } /* Now allocate the RX */ if (!(adapter->rx_rings = (struct rx_ring *) malloc(sizeof(struct rx_ring) * adapter->num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX ring memory\n"); error = ENOMEM; goto rx_fail; } tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc), EM_DBA_ALIGN); /* * Now set up the TX queues, txconf is needed to handle the * possibility that things fail midcourse and we need to * undo memory gracefully */ for (int i = 0; i < adapter->num_queues; i++, txconf++) { /* Set up some basics */ txr = &adapter->tx_rings[i]; txr->adapter = adapter; txr->me = i; /* Initialize the TX lock */ snprintf(txr->mtx_name, sizeof(txr->mtx_name), "%s:tx(%d)", device_get_nameunit(dev), txr->me); mtx_init(&txr->tx_mtx, txr->mtx_name, NULL, MTX_DEF); if (em_dma_malloc(adapter, tsize, &txr->txdma, BUS_DMA_NOWAIT)) { device_printf(dev, "Unable to allocate TX Descriptor memory\n"); error = ENOMEM; goto err_tx_desc; } txr->tx_base = (struct e1000_tx_desc *)txr->txdma.dma_vaddr; bzero((void *)txr->tx_base, tsize); if (em_allocate_transmit_buffers(txr)) { device_printf(dev, "Critical Failure setting up transmit buffers\n"); error = ENOMEM; goto err_tx_desc; } } /* * Next the RX queues... */ rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc), EM_DBA_ALIGN); for (int i = 0; i < adapter->num_queues; i++, rxconf++) { rxr = &adapter->rx_rings[i]; rxr->adapter = adapter; rxr->me = i; /* Initialize the RX lock */ snprintf(rxr->mtx_name, sizeof(rxr->mtx_name), "%s:rx(%d)", device_get_nameunit(dev), txr->me); mtx_init(&rxr->rx_mtx, rxr->mtx_name, NULL, MTX_DEF); if (em_dma_malloc(adapter, rsize, &rxr->rxdma, BUS_DMA_NOWAIT)) { device_printf(dev, "Unable to allocate RxDescriptor memory\n"); error = ENOMEM; goto err_rx_desc; } rxr->rx_base = (struct e1000_rx_desc *)rxr->rxdma.dma_vaddr; bzero((void *)rxr->rx_base, rsize); /* Allocate receive buffers for the ring*/ if (em_allocate_receive_buffers(rxr)) { device_printf(dev, "Critical Failure setting up receive buffers\n"); error = ENOMEM; goto err_rx_desc; } } return (0); err_rx_desc: for (rxr = adapter->rx_rings; rxconf > 0; rxr++, rxconf--) em_dma_free(adapter, &rxr->rxdma); err_tx_desc: for (txr = adapter->tx_rings; txconf > 0; txr++, txconf--) em_dma_free(adapter, &txr->txdma); free(adapter->rx_rings, M_DEVBUF); rx_fail: free(adapter->tx_rings, M_DEVBUF); fail: return (error); } /********************************************************************* * * Allocate memory for tx_buffer structures. The tx_buffer stores all * the information needed to transmit a packet on the wire. This is * called only once at attach, setup is done every reset. * **********************************************************************/ static int em_allocate_transmit_buffers(struct tx_ring *txr) { struct adapter *adapter = txr->adapter; device_t dev = adapter->dev; struct em_buffer *txbuf; int error, i; /* * Setup DMA descriptor areas. */ if ((error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ EM_TSO_SIZE, /* maxsize */ EM_MAX_SCATTER, /* nsegments */ PAGE_SIZE, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txr->txtag))) { device_printf(dev,"Unable to allocate TX DMA tag\n"); goto fail; } if (!(txr->tx_buffers = (struct em_buffer *) malloc(sizeof(struct em_buffer) * adapter->num_tx_desc, M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); error = ENOMEM; goto fail; } /* Create the descriptor buffer dma maps */ txbuf = txr->tx_buffers; for (i = 0; i < adapter->num_tx_desc; i++, txbuf++) { error = bus_dmamap_create(txr->txtag, 0, &txbuf->map); if (error != 0) { device_printf(dev, "Unable to create TX DMA map\n"); goto fail; } } return 0; fail: /* We free all, it handles case where we are in the middle */ em_free_transmit_structures(adapter); return (error); } /********************************************************************* * * Initialize a transmit ring. * **********************************************************************/ static void em_setup_transmit_ring(struct tx_ring *txr) { struct adapter *adapter = txr->adapter; struct em_buffer *txbuf; int i; /* Clear the old descriptor contents */ EM_TX_LOCK(txr); bzero((void *)txr->tx_base, (sizeof(struct e1000_tx_desc)) * adapter->num_tx_desc); /* Reset indices */ txr->next_avail_desc = 0; txr->next_to_clean = 0; /* Free any existing tx buffers. */ txbuf = txr->tx_buffers; for (i = 0; i < adapter->num_tx_desc; i++, txbuf++) { if (txbuf->m_head != NULL) { bus_dmamap_sync(txr->txtag, txbuf->map, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txr->txtag, txbuf->map); m_freem(txbuf->m_head); txbuf->m_head = NULL; } /* clear the watch index */ txbuf->next_eop = -1; } /* Set number of descriptors available */ txr->tx_avail = adapter->num_tx_desc; txr->queue_status = EM_QUEUE_IDLE; /* Clear checksum offload context. */ txr->last_hw_offload = 0; txr->last_hw_ipcss = 0; txr->last_hw_ipcso = 0; txr->last_hw_tucss = 0; txr->last_hw_tucso = 0; bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); EM_TX_UNLOCK(txr); } /********************************************************************* * * Initialize all transmit rings. * **********************************************************************/ static void em_setup_transmit_structures(struct adapter *adapter) { struct tx_ring *txr = adapter->tx_rings; for (int i = 0; i < adapter->num_queues; i++, txr++) em_setup_transmit_ring(txr); return; } /********************************************************************* * * Enable transmit unit. * **********************************************************************/ static void em_initialize_transmit_unit(struct adapter *adapter) { struct tx_ring *txr = adapter->tx_rings; struct e1000_hw *hw = &adapter->hw; u32 tctl, tarc, tipg = 0; INIT_DEBUGOUT("em_initialize_transmit_unit: begin"); for (int i = 0; i < adapter->num_queues; i++, txr++) { u64 bus_addr = txr->txdma.dma_paddr; /* Base and Len of TX Ring */ E1000_WRITE_REG(hw, E1000_TDLEN(i), adapter->num_tx_desc * sizeof(struct e1000_tx_desc)); E1000_WRITE_REG(hw, E1000_TDBAH(i), (u32)(bus_addr >> 32)); E1000_WRITE_REG(hw, E1000_TDBAL(i), (u32)bus_addr); /* Init the HEAD/TAIL indices */ E1000_WRITE_REG(hw, E1000_TDT(i), 0); E1000_WRITE_REG(hw, E1000_TDH(i), 0); HW_DEBUGOUT2("Base = %x, Length = %x\n", E1000_READ_REG(&adapter->hw, E1000_TDBAL(i)), E1000_READ_REG(&adapter->hw, E1000_TDLEN(i))); txr->queue_status = EM_QUEUE_IDLE; } /* Set the default values for the Tx Inter Packet Gap timer */ switch (adapter->hw.mac.type) { case e1000_82542: tipg = DEFAULT_82542_TIPG_IPGT; tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; break; case e1000_80003es2lan: tipg = DEFAULT_82543_TIPG_IPGR1; tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; break; default: if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) tipg = DEFAULT_82543_TIPG_IPGT_FIBER; else tipg = DEFAULT_82543_TIPG_IPGT_COPPER; tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; } E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg); E1000_WRITE_REG(&adapter->hw, E1000_TIDV, adapter->tx_int_delay.value); if(adapter->hw.mac.type >= e1000_82540) E1000_WRITE_REG(&adapter->hw, E1000_TADV, adapter->tx_abs_int_delay.value); if ((adapter->hw.mac.type == e1000_82571) || (adapter->hw.mac.type == e1000_82572)) { tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0)); tarc |= SPEED_MODE_BIT; E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); } else if (adapter->hw.mac.type == e1000_80003es2lan) { tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0)); tarc |= 1; E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(1)); tarc |= 1; E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc); } adapter->txd_cmd = E1000_TXD_CMD_IFCS; if (adapter->tx_int_delay.value > 0) adapter->txd_cmd |= E1000_TXD_CMD_IDE; /* Program the Transmit Control Register */ tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL); tctl &= ~E1000_TCTL_CT; tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN | (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT)); if (adapter->hw.mac.type >= e1000_82571) tctl |= E1000_TCTL_MULR; /* This write will effectively turn on the transmit unit. */ E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl); } /********************************************************************* * * Free all transmit rings. * **********************************************************************/ static void em_free_transmit_structures(struct adapter *adapter) { struct tx_ring *txr = adapter->tx_rings; for (int i = 0; i < adapter->num_queues; i++, txr++) { EM_TX_LOCK(txr); em_free_transmit_buffers(txr); em_dma_free(adapter, &txr->txdma); EM_TX_UNLOCK(txr); EM_TX_LOCK_DESTROY(txr); } free(adapter->tx_rings, M_DEVBUF); } /********************************************************************* * * Free transmit ring related data structures. * **********************************************************************/ static void em_free_transmit_buffers(struct tx_ring *txr) { struct adapter *adapter = txr->adapter; struct em_buffer *txbuf; INIT_DEBUGOUT("free_transmit_ring: begin"); if (txr->tx_buffers == NULL) return; for (int i = 0; i < adapter->num_tx_desc; i++) { txbuf = &txr->tx_buffers[i]; if (txbuf->m_head != NULL) { bus_dmamap_sync(txr->txtag, txbuf->map, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txr->txtag, txbuf->map); m_freem(txbuf->m_head); txbuf->m_head = NULL; if (txbuf->map != NULL) { bus_dmamap_destroy(txr->txtag, txbuf->map); txbuf->map = NULL; } } else if (txbuf->map != NULL) { bus_dmamap_unload(txr->txtag, txbuf->map); bus_dmamap_destroy(txr->txtag, txbuf->map); txbuf->map = NULL; } } if (txr->tx_buffers != NULL) { free(txr->tx_buffers, M_DEVBUF); txr->tx_buffers = NULL; } if (txr->txtag != NULL) { bus_dma_tag_destroy(txr->txtag); txr->txtag = NULL; } return; } /********************************************************************* * The offload context is protocol specific (TCP/UDP) and thus * only needs to be set when the protocol changes. The occasion * of a context change can be a performance detriment, and * might be better just disabled. The reason arises in the way * in which the controller supports pipelined requests from the * Tx data DMA. Up to four requests can be pipelined, and they may * belong to the same packet or to multiple packets. However all * requests for one packet are issued before a request is issued * for a subsequent packet and if a request for the next packet * requires a context change, that request will be stalled * until the previous request completes. This means setting up * a new context effectively disables pipelined Tx data DMA which * in turn greatly slow down performance to send small sized * frames. **********************************************************************/ static void em_transmit_checksum_setup(struct tx_ring *txr, struct mbuf *mp, int ip_off, struct ip *ip, u32 *txd_upper, u32 *txd_lower) { struct adapter *adapter = txr->adapter; struct e1000_context_desc *TXD = NULL; struct em_buffer *tx_buffer; int cur, hdr_len; u32 cmd = 0; u16 offload = 0; u8 ipcso, ipcss, tucso, tucss; ipcss = ipcso = tucss = tucso = 0; hdr_len = ip_off + (ip->ip_hl << 2); cur = txr->next_avail_desc; /* Setup of IP header checksum. */ if (mp->m_pkthdr.csum_flags & CSUM_IP) { *txd_upper |= E1000_TXD_POPTS_IXSM << 8; offload |= CSUM_IP; ipcss = ip_off; ipcso = ip_off + offsetof(struct ip, ip_sum); /* * Start offset for header checksum calculation. * End offset for header checksum calculation. * Offset of place to put the checksum. */ TXD = (struct e1000_context_desc *)&txr->tx_base[cur]; TXD->lower_setup.ip_fields.ipcss = ipcss; TXD->lower_setup.ip_fields.ipcse = htole16(hdr_len); TXD->lower_setup.ip_fields.ipcso = ipcso; cmd |= E1000_TXD_CMD_IP; } if (mp->m_pkthdr.csum_flags & CSUM_TCP) { *txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D; *txd_upper |= E1000_TXD_POPTS_TXSM << 8; offload |= CSUM_TCP; tucss = hdr_len; tucso = hdr_len + offsetof(struct tcphdr, th_sum); /* * Setting up new checksum offload context for every frames * takes a lot of processing time for hardware. This also * reduces performance a lot for small sized frames so avoid * it if driver can use previously configured checksum * offload context. */ if (txr->last_hw_offload == offload) { if (offload & CSUM_IP) { if (txr->last_hw_ipcss == ipcss && txr->last_hw_ipcso == ipcso && txr->last_hw_tucss == tucss && txr->last_hw_tucso == tucso) return; } else { if (txr->last_hw_tucss == tucss && txr->last_hw_tucso == tucso) return; } } txr->last_hw_offload = offload; txr->last_hw_tucss = tucss; txr->last_hw_tucso = tucso; /* * Start offset for payload checksum calculation. * End offset for payload checksum calculation. * Offset of place to put the checksum. */ TXD = (struct e1000_context_desc *)&txr->tx_base[cur]; TXD->upper_setup.tcp_fields.tucss = hdr_len; TXD->upper_setup.tcp_fields.tucse = htole16(0); TXD->upper_setup.tcp_fields.tucso = tucso; cmd |= E1000_TXD_CMD_TCP; } else if (mp->m_pkthdr.csum_flags & CSUM_UDP) { *txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D; *txd_upper |= E1000_TXD_POPTS_TXSM << 8; tucss = hdr_len; tucso = hdr_len + offsetof(struct udphdr, uh_sum); /* * Setting up new checksum offload context for every frames * takes a lot of processing time for hardware. This also * reduces performance a lot for small sized frames so avoid * it if driver can use previously configured checksum * offload context. */ if (txr->last_hw_offload == offload) { if (offload & CSUM_IP) { if (txr->last_hw_ipcss == ipcss && txr->last_hw_ipcso == ipcso && txr->last_hw_tucss == tucss && txr->last_hw_tucso == tucso) return; } else { if (txr->last_hw_tucss == tucss && txr->last_hw_tucso == tucso) return; } } txr->last_hw_offload = offload; txr->last_hw_tucss = tucss; txr->last_hw_tucso = tucso; /* * Start offset for header checksum calculation. * End offset for header checksum calculation. * Offset of place to put the checksum. */ TXD = (struct e1000_context_desc *)&txr->tx_base[cur]; TXD->upper_setup.tcp_fields.tucss = tucss; TXD->upper_setup.tcp_fields.tucse = htole16(0); TXD->upper_setup.tcp_fields.tucso = tucso; } if (offload & CSUM_IP) { txr->last_hw_ipcss = ipcss; txr->last_hw_ipcso = ipcso; } TXD->tcp_seg_setup.data = htole32(0); TXD->cmd_and_length = htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT | cmd); tx_buffer = &txr->tx_buffers[cur]; tx_buffer->m_head = NULL; tx_buffer->next_eop = -1; if (++cur == adapter->num_tx_desc) cur = 0; txr->tx_avail--; txr->next_avail_desc = cur; } /********************************************************************** * * Setup work for hardware segmentation offload (TSO) * **********************************************************************/ static void em_tso_setup(struct tx_ring *txr, struct mbuf *mp, int ip_off, struct ip *ip, struct tcphdr *tp, u32 *txd_upper, u32 *txd_lower) { struct adapter *adapter = txr->adapter; struct e1000_context_desc *TXD; struct em_buffer *tx_buffer; int cur, hdr_len; /* * In theory we can use the same TSO context if and only if * frame is the same type(IP/TCP) and the same MSS. However * checking whether a frame has the same IP/TCP structure is * hard thing so just ignore that and always restablish a * new TSO context. */ hdr_len = ip_off + (ip->ip_hl << 2) + (tp->th_off << 2); *txd_lower = (E1000_TXD_CMD_DEXT | /* Extended descr type */ E1000_TXD_DTYP_D | /* Data descr type */ E1000_TXD_CMD_TSE); /* Do TSE on this packet */ /* IP and/or TCP header checksum calculation and insertion. */ *txd_upper = (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8; cur = txr->next_avail_desc; tx_buffer = &txr->tx_buffers[cur]; TXD = (struct e1000_context_desc *) &txr->tx_base[cur]; /* * Start offset for header checksum calculation. * End offset for header checksum calculation. * Offset of place put the checksum. */ TXD->lower_setup.ip_fields.ipcss = ip_off; TXD->lower_setup.ip_fields.ipcse = htole16(ip_off + (ip->ip_hl << 2) - 1); TXD->lower_setup.ip_fields.ipcso = ip_off + offsetof(struct ip, ip_sum); /* * Start offset for payload checksum calculation. * End offset for payload checksum calculation. * Offset of place to put the checksum. */ TXD->upper_setup.tcp_fields.tucss = ip_off + (ip->ip_hl << 2); TXD->upper_setup.tcp_fields.tucse = 0; TXD->upper_setup.tcp_fields.tucso = ip_off + (ip->ip_hl << 2) + offsetof(struct tcphdr, th_sum); /* * Payload size per packet w/o any headers. * Length of all headers up to payload. */ TXD->tcp_seg_setup.fields.mss = htole16(mp->m_pkthdr.tso_segsz); TXD->tcp_seg_setup.fields.hdr_len = hdr_len; TXD->cmd_and_length = htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT | /* Extended descr */ E1000_TXD_CMD_TSE | /* TSE context */ E1000_TXD_CMD_IP | /* Do IP csum */ E1000_TXD_CMD_TCP | /* Do TCP checksum */ (mp->m_pkthdr.len - (hdr_len))); /* Total len */ tx_buffer->m_head = NULL; tx_buffer->next_eop = -1; if (++cur == adapter->num_tx_desc) cur = 0; txr->tx_avail--; txr->next_avail_desc = cur; txr->tx_tso = TRUE; } /********************************************************************** * * Examine each tx_buffer in the used queue. If the hardware is done * processing the packet then free associated resources. The * tx_buffer is put back on the free queue. * **********************************************************************/ static bool em_txeof(struct tx_ring *txr) { struct adapter *adapter = txr->adapter; int first, last, done, processed; struct em_buffer *tx_buffer; struct e1000_tx_desc *tx_desc, *eop_desc; struct ifnet *ifp = adapter->ifp; EM_TX_LOCK_ASSERT(txr); /* No work, make sure watchdog is off */ if (txr->tx_avail == adapter->num_tx_desc) { txr->queue_status = EM_QUEUE_IDLE; return (FALSE); } processed = 0; first = txr->next_to_clean; tx_desc = &txr->tx_base[first]; tx_buffer = &txr->tx_buffers[first]; last = tx_buffer->next_eop; eop_desc = &txr->tx_base[last]; /* * What this does is get the index of the * first descriptor AFTER the EOP of the * first packet, that way we can do the * simple comparison on the inner while loop. */ if (++last == adapter->num_tx_desc) last = 0; done = last; bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, BUS_DMASYNC_POSTREAD); while (eop_desc->upper.fields.status & E1000_TXD_STAT_DD) { /* We clean the range of the packet */ while (first != done) { tx_desc->upper.data = 0; tx_desc->lower.data = 0; tx_desc->buffer_addr = 0; ++txr->tx_avail; ++processed; if (tx_buffer->m_head) { bus_dmamap_sync(txr->txtag, tx_buffer->map, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txr->txtag, tx_buffer->map); m_freem(tx_buffer->m_head); tx_buffer->m_head = NULL; } tx_buffer->next_eop = -1; txr->watchdog_time = ticks; if (++first == adapter->num_tx_desc) first = 0; tx_buffer = &txr->tx_buffers[first]; tx_desc = &txr->tx_base[first]; } ++ifp->if_opackets; /* See if we can continue to the next packet */ last = tx_buffer->next_eop; if (last != -1) { eop_desc = &txr->tx_base[last]; /* Get new done point */ if (++last == adapter->num_tx_desc) last = 0; done = last; } else break; } bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); txr->next_to_clean = first; /* ** Watchdog calculation, we know there's ** work outstanding or the first return ** would have been taken, so none processed ** for too long indicates a hang. local timer ** will examine this and do a reset if needed. */ if ((!processed) && ((ticks - txr->watchdog_time) > EM_WATCHDOG)) txr->queue_status = EM_QUEUE_HUNG; /* * If we have enough room, clear IFF_DRV_OACTIVE * to tell the stack that it is OK to send packets. */ if (txr->tx_avail > EM_TX_CLEANUP_THRESHOLD) { ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; /* Disable watchdog if all clean */ if (txr->tx_avail == adapter->num_tx_desc) { txr->queue_status = EM_QUEUE_IDLE; return (FALSE); } } return (TRUE); } /********************************************************************* * * Refresh RX descriptor mbufs from system mbuf buffer pool. * **********************************************************************/ static void em_refresh_mbufs(struct rx_ring *rxr, int limit) { struct adapter *adapter = rxr->adapter; struct mbuf *m; bus_dma_segment_t segs[1]; struct em_buffer *rxbuf; int i, error, nsegs, cleaned; i = rxr->next_to_refresh; cleaned = -1; while (i != limit) { rxbuf = &rxr->rx_buffers[i]; if (rxbuf->m_head == NULL) { m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, adapter->rx_mbuf_sz); /* ** If we have a temporary resource shortage ** that causes a failure, just abort refresh ** for now, we will return to this point when ** reinvoked from em_rxeof. */ if (m == NULL) goto update; } else m = rxbuf->m_head; m->m_len = m->m_pkthdr.len = adapter->rx_mbuf_sz; m->m_flags |= M_PKTHDR; m->m_data = m->m_ext.ext_buf; /* Use bus_dma machinery to setup the memory mapping */ error = bus_dmamap_load_mbuf_sg(rxr->rxtag, rxbuf->map, m, segs, &nsegs, BUS_DMA_NOWAIT); if (error != 0) { printf("Refresh mbufs: hdr dmamap load" " failure - %d\n", error); m_free(m); rxbuf->m_head = NULL; goto update; } rxbuf->m_head = m; bus_dmamap_sync(rxr->rxtag, rxbuf->map, BUS_DMASYNC_PREREAD); rxr->rx_base[i].buffer_addr = htole64(segs[0].ds_addr); cleaned = i; /* Calculate next index */ if (++i == adapter->num_rx_desc) i = 0; rxr->next_to_refresh = i; } update: /* ** Update the tail pointer only if, ** and as far as we have refreshed. */ if (cleaned != -1) /* Update tail index */ E1000_WRITE_REG(&adapter->hw, E1000_RDT(rxr->me), cleaned); return; } /********************************************************************* * * Allocate memory for rx_buffer structures. Since we use one * rx_buffer per received packet, the maximum number of rx_buffer's * that we'll need is equal to the number of receive descriptors * that we've allocated. * **********************************************************************/ static int em_allocate_receive_buffers(struct rx_ring *rxr) { struct adapter *adapter = rxr->adapter; device_t dev = adapter->dev; struct em_buffer *rxbuf; int error; rxr->rx_buffers = malloc(sizeof(struct em_buffer) * adapter->num_rx_desc, M_DEVBUF, M_NOWAIT | M_ZERO); if (rxr->rx_buffers == NULL) { device_printf(dev, "Unable to allocate rx_buffer memory\n"); return (ENOMEM); } error = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ MJUM9BYTES, /* maxsize */ 1, /* nsegments */ MJUM9BYTES, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &rxr->rxtag); if (error) { device_printf(dev, "%s: bus_dma_tag_create failed %d\n", __func__, error); goto fail; } rxbuf = rxr->rx_buffers; for (int i = 0; i < adapter->num_rx_desc; i++, rxbuf++) { rxbuf = &rxr->rx_buffers[i]; error = bus_dmamap_create(rxr->rxtag, BUS_DMA_NOWAIT, &rxbuf->map); if (error) { device_printf(dev, "%s: bus_dmamap_create failed: %d\n", __func__, error); goto fail; } } return (0); fail: em_free_receive_structures(adapter); return (error); } /********************************************************************* * * Initialize a receive ring and its buffers. * **********************************************************************/ static int em_setup_receive_ring(struct rx_ring *rxr) { struct adapter *adapter = rxr->adapter; struct em_buffer *rxbuf; bus_dma_segment_t seg[1]; int rsize, nsegs, error; /* Clear the ring contents */ EM_RX_LOCK(rxr); rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc), EM_DBA_ALIGN); bzero((void *)rxr->rx_base, rsize); /* ** Free current RX buffer structs and their mbufs */ for (int i = 0; i < adapter->num_rx_desc; i++) { rxbuf = &rxr->rx_buffers[i]; if (rxbuf->m_head != NULL) { bus_dmamap_sync(rxr->rxtag, rxbuf->map, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(rxr->rxtag, rxbuf->map); m_freem(rxbuf->m_head); } } /* Now replenish the mbufs */ for (int j = 0; j != adapter->num_rx_desc; ++j) { rxbuf = &rxr->rx_buffers[j]; rxbuf->m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, adapter->rx_mbuf_sz); if (rxbuf->m_head == NULL) return (ENOBUFS); rxbuf->m_head->m_len = adapter->rx_mbuf_sz; rxbuf->m_head->m_flags &= ~M_HASFCS; /* we strip it */ rxbuf->m_head->m_pkthdr.len = adapter->rx_mbuf_sz; /* Get the memory mapping */ error = bus_dmamap_load_mbuf_sg(rxr->rxtag, rxbuf->map, rxbuf->m_head, seg, &nsegs, BUS_DMA_NOWAIT); if (error != 0) { m_freem(rxbuf->m_head); rxbuf->m_head = NULL; return (error); } bus_dmamap_sync(rxr->rxtag, rxbuf->map, BUS_DMASYNC_PREREAD); /* Update descriptor */ rxr->rx_base[j].buffer_addr = htole64(seg[0].ds_addr); } /* Setup our descriptor indices */ rxr->next_to_check = 0; rxr->next_to_refresh = 0; bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); EM_RX_UNLOCK(rxr); return (0); } /********************************************************************* * * Initialize all receive rings. * **********************************************************************/ static int em_setup_receive_structures(struct adapter *adapter) { struct rx_ring *rxr = adapter->rx_rings; int j; for (j = 0; j < adapter->num_queues; j++, rxr++) if (em_setup_receive_ring(rxr)) goto fail; return (0); fail: /* * Free RX buffers allocated so far, we will only handle * the rings that completed, the failing case will have * cleaned up for itself. 'j' failed, so its the terminus. */ for (int i = 0; i < j; ++i) { rxr = &adapter->rx_rings[i]; for (int n = 0; n < adapter->num_rx_desc; n++) { struct em_buffer *rxbuf; rxbuf = &rxr->rx_buffers[n]; if (rxbuf->m_head != NULL) { bus_dmamap_sync(rxr->rxtag, rxbuf->map, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(rxr->rxtag, rxbuf->map); m_freem(rxbuf->m_head); rxbuf->m_head = NULL; } } } return (ENOBUFS); } /********************************************************************* * * Free all receive rings. * **********************************************************************/ static void em_free_receive_structures(struct adapter *adapter) { struct rx_ring *rxr = adapter->rx_rings; for (int i = 0; i < adapter->num_queues; i++, rxr++) { em_free_receive_buffers(rxr); /* Free the ring memory as well */ em_dma_free(adapter, &rxr->rxdma); EM_RX_LOCK_DESTROY(rxr); } free(adapter->rx_rings, M_DEVBUF); } /********************************************************************* * * Free receive ring data structures * **********************************************************************/ static void em_free_receive_buffers(struct rx_ring *rxr) { struct adapter *adapter = rxr->adapter; struct em_buffer *rxbuf = NULL; INIT_DEBUGOUT("free_receive_buffers: begin"); if (rxr->rx_buffers != NULL) { for (int i = 0; i < adapter->num_rx_desc; i++) { rxbuf = &rxr->rx_buffers[i]; if (rxbuf->map != NULL) { bus_dmamap_sync(rxr->rxtag, rxbuf->map, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(rxr->rxtag, rxbuf->map); bus_dmamap_destroy(rxr->rxtag, rxbuf->map); } if (rxbuf->m_head != NULL) { m_freem(rxbuf->m_head); rxbuf->m_head = NULL; } } free(rxr->rx_buffers, M_DEVBUF); rxr->rx_buffers = NULL; } if (rxr->rxtag != NULL) { bus_dma_tag_destroy(rxr->rxtag); rxr->rxtag = NULL; } return; } /********************************************************************* * * Enable receive unit. * **********************************************************************/ #define MAX_INTS_PER_SEC 8000 #define DEFAULT_ITR 1000000000/(MAX_INTS_PER_SEC * 256) static void em_initialize_receive_unit(struct adapter *adapter) { struct rx_ring *rxr = adapter->rx_rings; struct ifnet *ifp = adapter->ifp; struct e1000_hw *hw = &adapter->hw; u64 bus_addr; u32 rctl, rxcsum; INIT_DEBUGOUT("em_initialize_receive_units: begin"); /* * Make sure receives are disabled while setting * up the descriptor ring */ rctl = E1000_READ_REG(hw, E1000_RCTL); E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN); E1000_WRITE_REG(&adapter->hw, E1000_RADV, adapter->rx_abs_int_delay.value); /* * Set the interrupt throttling rate. Value is calculated * as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns) */ E1000_WRITE_REG(hw, E1000_ITR, DEFAULT_ITR); /* ** When using MSIX interrupts we need to throttle ** using the EITR register (82574 only) */ if (hw->mac.type == e1000_82574) for (int i = 0; i < 4; i++) E1000_WRITE_REG(hw, E1000_EITR_82574(i), DEFAULT_ITR); /* Disable accelerated ackknowledge */ if (adapter->hw.mac.type == e1000_82574) E1000_WRITE_REG(hw, E1000_RFCTL, E1000_RFCTL_ACK_DIS); if (ifp->if_capenable & IFCAP_RXCSUM) { rxcsum = E1000_READ_REG(hw, E1000_RXCSUM); rxcsum |= (E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL); E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum); } /* ** XXX TEMPORARY WORKAROUND: on some systems with 82573 ** long latencies are observed, like Lenovo X60. This ** change eliminates the problem, but since having positive ** values in RDTR is a known source of problems on other ** platforms another solution is being sought. */ if (hw->mac.type == e1000_82573) E1000_WRITE_REG(hw, E1000_RDTR, 0x20); for (int i = 0; i < adapter->num_queues; i++, rxr++) { /* Setup the Base and Length of the Rx Descriptor Ring */ bus_addr = rxr->rxdma.dma_paddr; E1000_WRITE_REG(hw, E1000_RDLEN(i), adapter->num_rx_desc * sizeof(struct e1000_rx_desc)); E1000_WRITE_REG(hw, E1000_RDBAH(i), (u32)(bus_addr >> 32)); E1000_WRITE_REG(hw, E1000_RDBAL(i), (u32)bus_addr); /* Setup the Head and Tail Descriptor Pointers */ E1000_WRITE_REG(hw, E1000_RDH(i), 0); E1000_WRITE_REG(hw, E1000_RDT(i), adapter->num_rx_desc - 1); } /* Set early receive threshold on appropriate hw */ if (((adapter->hw.mac.type == e1000_ich9lan) || (adapter->hw.mac.type == e1000_pch2lan) || (adapter->hw.mac.type == e1000_ich10lan)) && (ifp->if_mtu > ETHERMTU)) { u32 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(0)); E1000_WRITE_REG(hw, E1000_RXDCTL(0), rxdctl | 3); E1000_WRITE_REG(hw, E1000_ERT, 0x100 | (1 << 13)); } if (adapter->hw.mac.type == e1000_pch2lan) { if (ifp->if_mtu > ETHERMTU) e1000_lv_jumbo_workaround_ich8lan(hw, TRUE); else e1000_lv_jumbo_workaround_ich8lan(hw, FALSE); } /* Setup the Receive Control Register */ rctl &= ~(3 << E1000_RCTL_MO_SHIFT); rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF | (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT); /* Strip the CRC */ rctl |= E1000_RCTL_SECRC; /* Make sure VLAN Filters are off */ rctl &= ~E1000_RCTL_VFE; rctl &= ~E1000_RCTL_SBP; if (adapter->rx_mbuf_sz == MCLBYTES) rctl |= E1000_RCTL_SZ_2048; else if (adapter->rx_mbuf_sz == MJUMPAGESIZE) rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX; else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX; if (ifp->if_mtu > ETHERMTU) rctl |= E1000_RCTL_LPE; else rctl &= ~E1000_RCTL_LPE; /* Write out the settings */ E1000_WRITE_REG(hw, E1000_RCTL, rctl); return; } /********************************************************************* * * This routine executes in interrupt context. It replenishes * the mbufs in the descriptor and sends data which has been * dma'ed into host memory to upper layer. * * We loop at most count times if count is > 0, or until done if * count < 0. * * For polling we also now return the number of cleaned packets *********************************************************************/ static bool em_rxeof(struct rx_ring *rxr, int count, int *done) { struct adapter *adapter = rxr->adapter; struct ifnet *ifp = adapter->ifp; struct mbuf *mp, *sendmp; u8 status = 0; u16 len; int i, processed, rxdone = 0; bool eop; struct e1000_rx_desc *cur; EM_RX_LOCK(rxr); for (i = rxr->next_to_check, processed = 0; count != 0;) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) break; bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); cur = &rxr->rx_base[i]; status = cur->status; mp = sendmp = NULL; if ((status & E1000_RXD_STAT_DD) == 0) break; len = le16toh(cur->length); eop = (status & E1000_RXD_STAT_EOP) != 0; if ((cur->errors & E1000_RXD_ERR_FRAME_ERR_MASK) || (rxr->discard == TRUE)) { adapter->dropped_pkts++; ++rxr->rx_discarded; if (!eop) /* Catch subsequent segs */ rxr->discard = TRUE; else rxr->discard = FALSE; em_rx_discard(rxr, i); goto next_desc; } /* Assign correct length to the current fragment */ mp = rxr->rx_buffers[i].m_head; mp->m_len = len; /* Trigger for refresh */ rxr->rx_buffers[i].m_head = NULL; /* First segment? */ if (rxr->fmp == NULL) { mp->m_pkthdr.len = len; rxr->fmp = rxr->lmp = mp; } else { /* Chain mbuf's together */ mp->m_flags &= ~M_PKTHDR; rxr->lmp->m_next = mp; rxr->lmp = mp; rxr->fmp->m_pkthdr.len += len; } if (eop) { --count; sendmp = rxr->fmp; sendmp->m_pkthdr.rcvif = ifp; ifp->if_ipackets++; em_receive_checksum(cur, sendmp); #ifndef __NO_STRICT_ALIGNMENT if (adapter->max_frame_size > (MCLBYTES - ETHER_ALIGN) && em_fixup_rx(rxr) != 0) goto skip; #endif if (status & E1000_RXD_STAT_VP) { sendmp->m_pkthdr.ether_vtag = (le16toh(cur->special) & E1000_RXD_SPC_VLAN_MASK); sendmp->m_flags |= M_VLANTAG; } #ifndef __NO_STRICT_ALIGNMENT skip: #endif rxr->fmp = rxr->lmp = NULL; } next_desc: /* Zero out the receive descriptors status. */ cur->status = 0; ++rxdone; /* cumulative for POLL */ ++processed; /* Advance our pointers to the next descriptor. */ if (++i == adapter->num_rx_desc) i = 0; /* Send to the stack */ if (sendmp != NULL) { rxr->next_to_check = i; EM_RX_UNLOCK(rxr); (*ifp->if_input)(ifp, sendmp); EM_RX_LOCK(rxr); i = rxr->next_to_check; } /* Only refresh mbufs every 8 descriptors */ if (processed == 8) { em_refresh_mbufs(rxr, i); processed = 0; } } /* Catch any remaining refresh work */ em_refresh_mbufs(rxr, i); rxr->next_to_check = i; if (done != NULL) *done = rxdone; EM_RX_UNLOCK(rxr); return ((status & E1000_RXD_STAT_DD) ? TRUE : FALSE); } static __inline void em_rx_discard(struct rx_ring *rxr, int i) { struct em_buffer *rbuf; rbuf = &rxr->rx_buffers[i]; /* Free any previous pieces */ if (rxr->fmp != NULL) { rxr->fmp->m_flags |= M_PKTHDR; m_freem(rxr->fmp); rxr->fmp = NULL; rxr->lmp = NULL; } /* ** Free buffer and allow em_refresh_mbufs() ** to clean up and recharge buffer. */ if (rbuf->m_head) { m_free(rbuf->m_head); rbuf->m_head = NULL; } return; } #ifndef __NO_STRICT_ALIGNMENT /* * When jumbo frames are enabled we should realign entire payload on * architecures with strict alignment. This is serious design mistake of 8254x * as it nullifies DMA operations. 8254x just allows RX buffer size to be * 2048/4096/8192/16384. What we really want is 2048 - ETHER_ALIGN to align its * payload. On architecures without strict alignment restrictions 8254x still * performs unaligned memory access which would reduce the performance too. * To avoid copying over an entire frame to align, we allocate a new mbuf and * copy ethernet header to the new mbuf. The new mbuf is prepended into the * existing mbuf chain. * * Be aware, best performance of the 8254x is achived only when jumbo frame is * not used at all on architectures with strict alignment. */ static int em_fixup_rx(struct rx_ring *rxr) { struct adapter *adapter = rxr->adapter; struct mbuf *m, *n; int error; error = 0; m = rxr->fmp; if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) { bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len); m->m_data += ETHER_HDR_LEN; } else { MGETHDR(n, M_DONTWAIT, MT_DATA); if (n != NULL) { bcopy(m->m_data, n->m_data, ETHER_HDR_LEN); m->m_data += ETHER_HDR_LEN; m->m_len -= ETHER_HDR_LEN; n->m_len = ETHER_HDR_LEN; M_MOVE_PKTHDR(n, m); n->m_next = m; rxr->fmp = n; } else { adapter->dropped_pkts++; m_freem(rxr->fmp); rxr->fmp = NULL; error = ENOMEM; } } return (error); } #endif /********************************************************************* * * Verify that the hardware indicated that the checksum is valid. * Inform the stack about the status of checksum so that stack * doesn't spend time verifying the checksum. * *********************************************************************/ static void em_receive_checksum(struct e1000_rx_desc *rx_desc, struct mbuf *mp) { /* Ignore Checksum bit is set */ if (rx_desc->status & E1000_RXD_STAT_IXSM) { mp->m_pkthdr.csum_flags = 0; return; } if (rx_desc->status & E1000_RXD_STAT_IPCS) { /* Did it pass? */ if (!(rx_desc->errors & E1000_RXD_ERR_IPE)) { /* IP Checksum Good */ mp->m_pkthdr.csum_flags = CSUM_IP_CHECKED; mp->m_pkthdr.csum_flags |= CSUM_IP_VALID; } else { mp->m_pkthdr.csum_flags = 0; } } if (rx_desc->status & E1000_RXD_STAT_TCPCS) { /* Did it pass? */ if (!(rx_desc->errors & E1000_RXD_ERR_TCPE)) { mp->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); mp->m_pkthdr.csum_data = htons(0xffff); } } } /* * This routine is run via an vlan * config EVENT */ static void em_register_vlan(void *arg, struct ifnet *ifp, u16 vtag) { struct adapter *adapter = ifp->if_softc; u32 index, bit; if (ifp->if_softc != arg) /* Not our event */ return; if ((vtag == 0) || (vtag > 4095)) /* Invalid ID */ return; EM_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] |= (1 << bit); ++adapter->num_vlans; /* Re-init to load the changes */ if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) em_init_locked(adapter); EM_CORE_UNLOCK(adapter); } /* * This routine is run via an vlan * unconfig EVENT */ static void em_unregister_vlan(void *arg, struct ifnet *ifp, u16 vtag) { struct adapter *adapter = ifp->if_softc; u32 index, bit; if (ifp->if_softc != arg) return; if ((vtag == 0) || (vtag > 4095)) /* Invalid */ return; EM_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] &= ~(1 << bit); --adapter->num_vlans; /* Re-init to load the changes */ if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) em_init_locked(adapter); EM_CORE_UNLOCK(adapter); } static void em_setup_vlan_hw_support(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; u32 reg; /* ** We get here thru init_locked, meaning ** a soft reset, this has already cleared ** the VFTA and other state, so if there ** have been no vlan's registered do nothing. */ if (adapter->num_vlans == 0) return; /* ** A soft reset zero's out the VFTA, so ** we need to repopulate it now. */ for (int i = 0; i < EM_VFTA_SIZE; i++) if (adapter->shadow_vfta[i] != 0) E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, adapter->shadow_vfta[i]); reg = E1000_READ_REG(hw, E1000_CTRL); reg |= E1000_CTRL_VME; E1000_WRITE_REG(hw, E1000_CTRL, reg); /* Enable the Filter Table */ reg = E1000_READ_REG(hw, E1000_RCTL); reg &= ~E1000_RCTL_CFIEN; reg |= E1000_RCTL_VFE; E1000_WRITE_REG(hw, E1000_RCTL, reg); } static void em_enable_intr(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; u32 ims_mask = IMS_ENABLE_MASK; if (hw->mac.type == e1000_82574) { E1000_WRITE_REG(hw, EM_EIAC, EM_MSIX_MASK); ims_mask |= EM_MSIX_MASK; } E1000_WRITE_REG(hw, E1000_IMS, ims_mask); } static void em_disable_intr(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; if (hw->mac.type == e1000_82574) E1000_WRITE_REG(hw, EM_EIAC, 0); E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff); } /* * Bit of a misnomer, what this really means is * to enable OS management of the system... aka * to disable special hardware management features */ static void em_init_manageability(struct adapter *adapter) { /* A shared code workaround */ #define E1000_82542_MANC2H E1000_MANC2H if (adapter->has_manage) { int manc2h = E1000_READ_REG(&adapter->hw, E1000_MANC2H); int manc = E1000_READ_REG(&adapter->hw, E1000_MANC); /* disable hardware interception of ARP */ manc &= ~(E1000_MANC_ARP_EN); /* enable receiving management packets to the host */ manc |= E1000_MANC_EN_MNG2HOST; #define E1000_MNG2HOST_PORT_623 (1 << 5) #define E1000_MNG2HOST_PORT_664 (1 << 6) manc2h |= E1000_MNG2HOST_PORT_623; manc2h |= E1000_MNG2HOST_PORT_664; E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h); E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc); } } /* * Give control back to hardware management * controller if there is one. */ static void em_release_manageability(struct adapter *adapter) { if (adapter->has_manage) { int manc = E1000_READ_REG(&adapter->hw, E1000_MANC); /* re-enable hardware interception of ARP */ manc |= E1000_MANC_ARP_EN; manc &= ~E1000_MANC_EN_MNG2HOST; E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc); } } /* * em_get_hw_control sets the {CTRL_EXT|FWSM}:DRV_LOAD bit. * For ASF and Pass Through versions of f/w this means * that the driver is loaded. For AMT version type f/w * this means that the network i/f is open. */ static void em_get_hw_control(struct adapter *adapter) { u32 ctrl_ext, swsm; if (adapter->hw.mac.type == e1000_82573) { swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM); E1000_WRITE_REG(&adapter->hw, E1000_SWSM, swsm | E1000_SWSM_DRV_LOAD); return; } /* else */ ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD); return; } /* * em_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit. * For ASF and Pass Through versions of f/w this means that * the driver is no longer loaded. For AMT versions of the * f/w this means that the network i/f is closed. */ static void em_release_hw_control(struct adapter *adapter) { u32 ctrl_ext, swsm; if (!adapter->has_manage) return; if (adapter->hw.mac.type == e1000_82573) { swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM); E1000_WRITE_REG(&adapter->hw, E1000_SWSM, swsm & ~E1000_SWSM_DRV_LOAD); return; } /* else */ ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD); return; } static int em_is_valid_ether_addr(u8 *addr) { char zero_addr[6] = { 0, 0, 0, 0, 0, 0 }; if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) { return (FALSE); } return (TRUE); } /* ** Parse the interface capabilities with regard ** to both system management and wake-on-lan for ** later use. */ static void em_get_wakeup(device_t dev) { struct adapter *adapter = device_get_softc(dev); u16 eeprom_data = 0, device_id, apme_mask; adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw); apme_mask = EM_EEPROM_APME; switch (adapter->hw.mac.type) { case e1000_82573: case e1000_82583: adapter->has_amt = TRUE; /* Falls thru */ case e1000_82571: case e1000_82572: case e1000_80003es2lan: if (adapter->hw.bus.func == 1) { e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data); break; } else e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data); break; case e1000_ich8lan: case e1000_ich9lan: case e1000_ich10lan: case e1000_pchlan: case e1000_pch2lan: apme_mask = E1000_WUC_APME; adapter->has_amt = TRUE; eeprom_data = E1000_READ_REG(&adapter->hw, E1000_WUC); break; default: e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data); break; } if (eeprom_data & apme_mask) adapter->wol = (E1000_WUFC_MAG | E1000_WUFC_MC); /* * We have the eeprom settings, now apply the special cases * where the eeprom may be wrong or the board won't support * wake on lan on a particular port */ device_id = pci_get_device(dev); switch (device_id) { case E1000_DEV_ID_82571EB_FIBER: /* Wake events only supported on port A for dual fiber * regardless of eeprom setting */ if (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_FUNC_1) adapter->wol = 0; break; case E1000_DEV_ID_82571EB_QUAD_COPPER: case E1000_DEV_ID_82571EB_QUAD_FIBER: case E1000_DEV_ID_82571EB_QUAD_COPPER_LP: /* if quad port adapter, disable WoL on all but port A */ if (global_quad_port_a != 0) adapter->wol = 0; /* Reset for multiple quad port adapters */ if (++global_quad_port_a == 4) global_quad_port_a = 0; break; } return; } /* * Enable PCI Wake On Lan capability */ static void em_enable_wakeup(device_t dev) { struct adapter *adapter = device_get_softc(dev); struct ifnet *ifp = adapter->ifp; u32 pmc, ctrl, ctrl_ext, rctl; u16 status; if ((pci_find_extcap(dev, PCIY_PMG, &pmc) != 0)) return; /* Advertise the wakeup capability */ ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); ctrl |= (E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN3); E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN); if ((adapter->hw.mac.type == e1000_ich8lan) || (adapter->hw.mac.type == e1000_pchlan) || (adapter->hw.mac.type == e1000_ich9lan) || (adapter->hw.mac.type == e1000_ich10lan)) { e1000_disable_gig_wol_ich8lan(&adapter->hw); e1000_hv_phy_powerdown_workaround_ich8lan(&adapter->hw); } /* Keep the laser running on Fiber adapters */ if (adapter->hw.phy.media_type == e1000_media_type_fiber || adapter->hw.phy.media_type == e1000_media_type_internal_serdes) { ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA; E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext); } /* ** Determine type of Wakeup: note that wol ** is set with all bits on by default. */ if ((ifp->if_capenable & IFCAP_WOL_MAGIC) == 0) adapter->wol &= ~E1000_WUFC_MAG; if ((ifp->if_capenable & IFCAP_WOL_MCAST) == 0) adapter->wol &= ~E1000_WUFC_MC; else { rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); rctl |= E1000_RCTL_MPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl); } if ((adapter->hw.mac.type == e1000_pchlan) || (adapter->hw.mac.type == e1000_pch2lan)) { if (em_enable_phy_wakeup(adapter)) return; } else { E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN); E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol); } if (adapter->hw.phy.type == e1000_phy_igp_3) e1000_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw); /* Request PME */ status = pci_read_config(dev, pmc + PCIR_POWER_STATUS, 2); status &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE); if (ifp->if_capenable & IFCAP_WOL) status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE; pci_write_config(dev, pmc + PCIR_POWER_STATUS, status, 2); return; } /* ** WOL in the newer chipset interfaces (pchlan) ** require thing to be copied into the phy */ static int em_enable_phy_wakeup(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; u32 mreg, ret = 0; u16 preg; /* copy MAC RARs to PHY RARs */ e1000_copy_rx_addrs_to_phy_ich8lan(hw); /* copy MAC MTA to PHY MTA */ for (int i = 0; i < adapter->hw.mac.mta_reg_count; i++) { mreg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i); e1000_write_phy_reg(hw, BM_MTA(i), (u16)(mreg & 0xFFFF)); e1000_write_phy_reg(hw, BM_MTA(i) + 1, (u16)((mreg >> 16) & 0xFFFF)); } /* configure PHY Rx Control register */ e1000_read_phy_reg(&adapter->hw, BM_RCTL, &preg); mreg = E1000_READ_REG(hw, E1000_RCTL); if (mreg & E1000_RCTL_UPE) preg |= BM_RCTL_UPE; if (mreg & E1000_RCTL_MPE) preg |= BM_RCTL_MPE; preg &= ~(BM_RCTL_MO_MASK); if (mreg & E1000_RCTL_MO_3) preg |= (((mreg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT) << BM_RCTL_MO_SHIFT); if (mreg & E1000_RCTL_BAM) preg |= BM_RCTL_BAM; if (mreg & E1000_RCTL_PMCF) preg |= BM_RCTL_PMCF; mreg = E1000_READ_REG(hw, E1000_CTRL); if (mreg & E1000_CTRL_RFCE) preg |= BM_RCTL_RFCE; e1000_write_phy_reg(&adapter->hw, BM_RCTL, preg); /* enable PHY wakeup in MAC register */ E1000_WRITE_REG(hw, E1000_WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN); E1000_WRITE_REG(hw, E1000_WUFC, adapter->wol); /* configure and enable PHY wakeup in PHY registers */ e1000_write_phy_reg(&adapter->hw, BM_WUFC, adapter->wol); e1000_write_phy_reg(&adapter->hw, BM_WUC, E1000_WUC_PME_EN); /* activate PHY wakeup */ ret = hw->phy.ops.acquire(hw); if (ret) { printf("Could not acquire PHY\n"); return ret; } e1000_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT, (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT)); ret = e1000_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &preg); if (ret) { printf("Could not read PHY page 769\n"); goto out; } preg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT; ret = e1000_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, preg); if (ret) printf("Could not set PHY Host Wakeup bit\n"); out: hw->phy.ops.release(hw); return ret; } static void em_led_func(void *arg, int onoff) { struct adapter *adapter = arg; EM_CORE_LOCK(adapter); if (onoff) { e1000_setup_led(&adapter->hw); e1000_led_on(&adapter->hw); } else { e1000_led_off(&adapter->hw); e1000_cleanup_led(&adapter->hw); } EM_CORE_UNLOCK(adapter); } /* ** Disable the L0S and L1 LINK states */ static void em_disable_aspm(struct adapter *adapter) { int base, reg; u16 link_cap,link_ctrl; device_t dev = adapter->dev; switch (adapter->hw.mac.type) { case e1000_82573: case e1000_82574: case e1000_82583: break; default: return; } if (pci_find_extcap(dev, PCIY_EXPRESS, &base) != 0) return; - reg = base + PCIR_EXPRESS_LINK_CAP; + reg = base + PCIER_LINK_CAP; link_cap = pci_read_config(dev, reg, 2); - if ((link_cap & PCIM_LINK_CAP_ASPM) == 0) + if ((link_cap & PCIEM_LINK_CAP_ASPM) == 0) return; - reg = base + PCIR_EXPRESS_LINK_CTL; + reg = base + PCIER_LINK_CTL; link_ctrl = pci_read_config(dev, reg, 2); link_ctrl &= 0xFFFC; /* turn off bit 1 and 2 */ pci_write_config(dev, reg, link_ctrl, 2); return; } /********************************************************************** * * Update the board statistics counters. * **********************************************************************/ static void em_update_stats_counters(struct adapter *adapter) { struct ifnet *ifp; if(adapter->hw.phy.media_type == e1000_media_type_copper || (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) { adapter->stats.symerrs += E1000_READ_REG(&adapter->hw, E1000_SYMERRS); adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC); } adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS); adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC); adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC); adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL); adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC); adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL); adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC); adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC); adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC); adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC); adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC); /* ** For watchdog management we need to know if we have been ** paused during the last interval, so capture that here. */ adapter->pause_frames = E1000_READ_REG(&adapter->hw, E1000_XOFFRXC); adapter->stats.xoffrxc += adapter->pause_frames; adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC); adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC); adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64); adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127); adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255); adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511); adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023); adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522); adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC); adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC); adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC); adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC); /* For the 64-bit byte counters the low dword must be read first. */ /* Both registers clear on the read of the high dword */ adapter->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCL) + ((u64)E1000_READ_REG(&adapter->hw, E1000_GORCH) << 32); adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCL) + ((u64)E1000_READ_REG(&adapter->hw, E1000_GOTCH) << 32); adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC); adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC); adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC); adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC); adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC); adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH); adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH); adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR); adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT); adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64); adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127); adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255); adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511); adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023); adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522); adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC); adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC); /* Interrupt Counts */ adapter->stats.iac += E1000_READ_REG(&adapter->hw, E1000_IAC); adapter->stats.icrxptc += E1000_READ_REG(&adapter->hw, E1000_ICRXPTC); adapter->stats.icrxatc += E1000_READ_REG(&adapter->hw, E1000_ICRXATC); adapter->stats.ictxptc += E1000_READ_REG(&adapter->hw, E1000_ICTXPTC); adapter->stats.ictxatc += E1000_READ_REG(&adapter->hw, E1000_ICTXATC); adapter->stats.ictxqec += E1000_READ_REG(&adapter->hw, E1000_ICTXQEC); adapter->stats.ictxqmtc += E1000_READ_REG(&adapter->hw, E1000_ICTXQMTC); adapter->stats.icrxdmtc += E1000_READ_REG(&adapter->hw, E1000_ICRXDMTC); adapter->stats.icrxoc += E1000_READ_REG(&adapter->hw, E1000_ICRXOC); if (adapter->hw.mac.type >= e1000_82543) { adapter->stats.algnerrc += E1000_READ_REG(&adapter->hw, E1000_ALGNERRC); adapter->stats.rxerrc += E1000_READ_REG(&adapter->hw, E1000_RXERRC); adapter->stats.tncrs += E1000_READ_REG(&adapter->hw, E1000_TNCRS); adapter->stats.cexterr += E1000_READ_REG(&adapter->hw, E1000_CEXTERR); adapter->stats.tsctc += E1000_READ_REG(&adapter->hw, E1000_TSCTC); adapter->stats.tsctfc += E1000_READ_REG(&adapter->hw, E1000_TSCTFC); } ifp = adapter->ifp; ifp->if_collisions = adapter->stats.colc; /* Rx Errors */ ifp->if_ierrors = adapter->dropped_pkts + adapter->stats.rxerrc + adapter->stats.crcerrs + adapter->stats.algnerrc + adapter->stats.ruc + adapter->stats.roc + adapter->stats.mpc + adapter->stats.cexterr; /* Tx Errors */ ifp->if_oerrors = adapter->stats.ecol + adapter->stats.latecol + adapter->watchdog_events; } /* Export a single 32-bit register via a read-only sysctl. */ static int em_sysctl_reg_handler(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; u_int val; adapter = oidp->oid_arg1; val = E1000_READ_REG(&adapter->hw, oidp->oid_arg2); return (sysctl_handle_int(oidp, &val, 0, req)); } /* * Add sysctl variables, one per statistic, to the system. */ static void em_add_hw_stats(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev); struct sysctl_oid *tree = device_get_sysctl_tree(dev); struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree); struct e1000_hw_stats *stats = &adapter->stats; struct sysctl_oid *stat_node, *queue_node, *int_node; struct sysctl_oid_list *stat_list, *queue_list, *int_list; #define QUEUE_NAME_LEN 32 char namebuf[QUEUE_NAME_LEN]; /* Driver Statistics */ SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "link_irq", CTLFLAG_RD, &adapter->link_irq, 0, "Link MSIX IRQ Handled"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_fail", CTLFLAG_RD, &adapter->mbuf_alloc_failed, "Std mbuf failed"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "cluster_alloc_fail", CTLFLAG_RD, &adapter->mbuf_cluster_failed, "Std mbuf cluster failed"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "dropped", CTLFLAG_RD, &adapter->dropped_pkts, "Driver dropped packets"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_dma_fail", CTLFLAG_RD, &adapter->no_tx_dma_setup, "Driver tx dma failure in xmit"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_overruns", CTLFLAG_RD, &adapter->rx_overruns, "RX overruns"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "watchdog_timeouts", CTLFLAG_RD, &adapter->watchdog_events, "Watchdog timeouts"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "device_control", CTLFLAG_RD, adapter, E1000_CTRL, em_sysctl_reg_handler, "IU", "Device Control Register"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_control", CTLFLAG_RD, adapter, E1000_RCTL, em_sysctl_reg_handler, "IU", "Receiver Control Register"); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_high_water", CTLFLAG_RD, &adapter->hw.fc.high_water, 0, "Flow Control High Watermark"); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_low_water", CTLFLAG_RD, &adapter->hw.fc.low_water, 0, "Flow Control Low Watermark"); for (int i = 0; i < adapter->num_queues; i++, rxr++, txr++) { snprintf(namebuf, QUEUE_NAME_LEN, "queue%d", i); queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head", CTLFLAG_RD, adapter, E1000_TDH(txr->me), em_sysctl_reg_handler, "IU", "Transmit Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_tail", CTLFLAG_RD, adapter, E1000_TDT(txr->me), em_sysctl_reg_handler, "IU", "Transmit Descriptor Tail"); SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "tx_irq", CTLFLAG_RD, &txr->tx_irq, "Queue MSI-X Transmit Interrupts"); SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "no_desc_avail", CTLFLAG_RD, &txr->no_desc_avail, "Queue No Descriptor Available"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_head", CTLFLAG_RD, adapter, E1000_RDH(rxr->me), em_sysctl_reg_handler, "IU", "Receive Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_tail", CTLFLAG_RD, adapter, E1000_RDT(rxr->me), em_sysctl_reg_handler, "IU", "Receive Descriptor Tail"); SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "rx_irq", CTLFLAG_RD, &rxr->rx_irq, "Queue MSI-X Receive Interrupts"); } /* MAC stats get their own sub node */ stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats", CTLFLAG_RD, NULL, "Statistics"); stat_list = SYSCTL_CHILDREN(stat_node); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "excess_coll", CTLFLAG_RD, &stats->ecol, "Excessive collisions"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "single_coll", CTLFLAG_RD, &stats->scc, "Single collisions"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "multiple_coll", CTLFLAG_RD, &stats->mcc, "Multiple collisions"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "late_coll", CTLFLAG_RD, &stats->latecol, "Late collisions"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "collision_count", CTLFLAG_RD, &stats->colc, "Collision Count"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "symbol_errors", CTLFLAG_RD, &adapter->stats.symerrs, "Symbol Errors"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "sequence_errors", CTLFLAG_RD, &adapter->stats.sec, "Sequence Errors"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "defer_count", CTLFLAG_RD, &adapter->stats.dc, "Defer Count"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "missed_packets", CTLFLAG_RD, &adapter->stats.mpc, "Missed Packets"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_no_buff", CTLFLAG_RD, &adapter->stats.rnbc, "Receive No Buffers"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_undersize", CTLFLAG_RD, &adapter->stats.ruc, "Receive Undersize"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_fragmented", CTLFLAG_RD, &adapter->stats.rfc, "Fragmented Packets Received "); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_oversize", CTLFLAG_RD, &adapter->stats.roc, "Oversized Packets Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_jabber", CTLFLAG_RD, &adapter->stats.rjc, "Recevied Jabber"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "recv_errs", CTLFLAG_RD, &adapter->stats.rxerrc, "Receive Errors"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "crc_errs", CTLFLAG_RD, &adapter->stats.crcerrs, "CRC errors"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "alignment_errs", CTLFLAG_RD, &adapter->stats.algnerrc, "Alignment Errors"); /* On 82575 these are collision counts */ SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "coll_ext_errs", CTLFLAG_RD, &adapter->stats.cexterr, "Collision/Carrier extension errors"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xon_recvd", CTLFLAG_RD, &adapter->stats.xonrxc, "XON Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xon_txd", CTLFLAG_RD, &adapter->stats.xontxc, "XON Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xoff_recvd", CTLFLAG_RD, &adapter->stats.xoffrxc, "XOFF Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "xoff_txd", CTLFLAG_RD, &adapter->stats.xofftxc, "XOFF Transmitted"); /* Packet Reception Stats */ SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_pkts_recvd", CTLFLAG_RD, &adapter->stats.tpr, "Total Packets Received "); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd", CTLFLAG_RD, &adapter->stats.gprc, "Good Packets Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_recvd", CTLFLAG_RD, &adapter->stats.bprc, "Broadcast Packets Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd", CTLFLAG_RD, &adapter->stats.mprc, "Multicast Packets Received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_64", CTLFLAG_RD, &adapter->stats.prc64, "64 byte frames received "); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127", CTLFLAG_RD, &adapter->stats.prc127, "65-127 byte frames received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255", CTLFLAG_RD, &adapter->stats.prc255, "128-255 byte frames received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511", CTLFLAG_RD, &adapter->stats.prc511, "256-511 byte frames received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023", CTLFLAG_RD, &adapter->stats.prc1023, "512-1023 byte frames received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522", CTLFLAG_RD, &adapter->stats.prc1522, "1023-1522 byte frames received"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd", CTLFLAG_RD, &adapter->stats.gorc, "Good Octets Received"); /* Packet Transmission Stats */ SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_octets_txd", CTLFLAG_RD, &adapter->stats.gotc, "Good Octets Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd", CTLFLAG_RD, &adapter->stats.tpt, "Total Packets Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd", CTLFLAG_RD, &adapter->stats.gptc, "Good Packets Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd", CTLFLAG_RD, &adapter->stats.bptc, "Broadcast Packets Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd", CTLFLAG_RD, &adapter->stats.mptc, "Multicast Packets Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_64", CTLFLAG_RD, &adapter->stats.ptc64, "64 byte frames transmitted "); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127", CTLFLAG_RD, &adapter->stats.ptc127, "65-127 byte frames transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255", CTLFLAG_RD, &adapter->stats.ptc255, "128-255 byte frames transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511", CTLFLAG_RD, &adapter->stats.ptc511, "256-511 byte frames transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023", CTLFLAG_RD, &adapter->stats.ptc1023, "512-1023 byte frames transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522", CTLFLAG_RD, &adapter->stats.ptc1522, "1024-1522 byte frames transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tso_txd", CTLFLAG_RD, &adapter->stats.tsctc, "TSO Contexts Transmitted"); SYSCTL_ADD_QUAD(ctx, stat_list, OID_AUTO, "tso_ctx_fail", CTLFLAG_RD, &adapter->stats.tsctfc, "TSO Contexts Failed"); /* Interrupt Stats */ int_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "interrupts", CTLFLAG_RD, NULL, "Interrupt Statistics"); int_list = SYSCTL_CHILDREN(int_node); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "asserts", CTLFLAG_RD, &adapter->stats.iac, "Interrupt Assertion Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_pkt_timer", CTLFLAG_RD, &adapter->stats.icrxptc, "Interrupt Cause Rx Pkt Timer Expire Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_abs_timer", CTLFLAG_RD, &adapter->stats.icrxatc, "Interrupt Cause Rx Abs Timer Expire Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_pkt_timer", CTLFLAG_RD, &adapter->stats.ictxptc, "Interrupt Cause Tx Pkt Timer Expire Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_abs_timer", CTLFLAG_RD, &adapter->stats.ictxatc, "Interrupt Cause Tx Abs Timer Expire Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_queue_empty", CTLFLAG_RD, &adapter->stats.ictxqec, "Interrupt Cause Tx Queue Empty Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "tx_queue_min_thresh", CTLFLAG_RD, &adapter->stats.ictxqmtc, "Interrupt Cause Tx Queue Min Thresh Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_desc_min_thresh", CTLFLAG_RD, &adapter->stats.icrxdmtc, "Interrupt Cause Rx Desc Min Thresh Count"); SYSCTL_ADD_QUAD(ctx, int_list, OID_AUTO, "rx_overrun", CTLFLAG_RD, &adapter->stats.icrxoc, "Interrupt Cause Receiver Overrun Count"); } /********************************************************************** * * This routine provides a way to dump out the adapter eeprom, * often a useful debug/service tool. This only dumps the first * 32 words, stuff that matters is in that extent. * **********************************************************************/ static int em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; int error; int result; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || !req->newptr) return (error); /* * This value will cause a hex dump of the * first 32 16-bit words of the EEPROM to * the screen. */ if (result == 1) { adapter = (struct adapter *)arg1; em_print_nvm_info(adapter); } return (error); } static void em_print_nvm_info(struct adapter *adapter) { u16 eeprom_data; int i, j, row = 0; /* Its a bit crude, but it gets the job done */ printf("\nInterface EEPROM Dump:\n"); printf("Offset\n0x0000 "); for (i = 0, j = 0; i < 32; i++, j++) { if (j == 8) { /* Make the offset block */ j = 0; ++row; printf("\n0x00%x0 ",row); } e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data); printf("%04x ", eeprom_data); } printf("\n"); } static int em_sysctl_int_delay(SYSCTL_HANDLER_ARGS) { struct em_int_delay_info *info; struct adapter *adapter; u32 regval; int error, usecs, ticks; info = (struct em_int_delay_info *)arg1; usecs = info->value; error = sysctl_handle_int(oidp, &usecs, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (usecs < 0 || usecs > EM_TICKS_TO_USECS(65535)) return (EINVAL); info->value = usecs; ticks = EM_USECS_TO_TICKS(usecs); adapter = info->adapter; EM_CORE_LOCK(adapter); regval = E1000_READ_OFFSET(&adapter->hw, info->offset); regval = (regval & ~0xffff) | (ticks & 0xffff); /* Handle a few special cases. */ switch (info->offset) { case E1000_RDTR: break; case E1000_TIDV: if (ticks == 0) { adapter->txd_cmd &= ~E1000_TXD_CMD_IDE; /* Don't write 0 into the TIDV register. */ regval++; } else adapter->txd_cmd |= E1000_TXD_CMD_IDE; break; } E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval); EM_CORE_UNLOCK(adapter); return (0); } static void em_add_int_delay_sysctl(struct adapter *adapter, const char *name, const char *description, struct em_int_delay_info *info, int offset, int value) { info->adapter = adapter; info->offset = offset; info->value = value; SYSCTL_ADD_PROC(device_get_sysctl_ctx(adapter->dev), SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)), OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, info, 0, em_sysctl_int_delay, "I", description); } static void em_add_rx_process_limit(struct adapter *adapter, const char *name, const char *description, int *limit, int value) { *limit = value; SYSCTL_ADD_INT(device_get_sysctl_ctx(adapter->dev), SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)), OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, limit, value, description); } static void em_set_flow_cntrl(struct adapter *adapter, const char *name, const char *description, int *limit, int value) { *limit = value; SYSCTL_ADD_INT(device_get_sysctl_ctx(adapter->dev), SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)), OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, limit, value, description); } static int em_sysctl_debug_info(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; int error; int result; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || !req->newptr) return (error); if (result == 1) { adapter = (struct adapter *)arg1; em_print_debug_info(adapter); } return (error); } /* ** This routine is meant to be fluid, add whatever is ** needed for debugging a problem. -jfv */ static void em_print_debug_info(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; if (adapter->ifp->if_drv_flags & IFF_DRV_RUNNING) printf("Interface is RUNNING "); else printf("Interface is NOT RUNNING\n"); if (adapter->ifp->if_drv_flags & IFF_DRV_OACTIVE) printf("and ACTIVE\n"); else printf("and INACTIVE\n"); device_printf(dev, "hw tdh = %d, hw tdt = %d\n", E1000_READ_REG(&adapter->hw, E1000_TDH(0)), E1000_READ_REG(&adapter->hw, E1000_TDT(0))); device_printf(dev, "hw rdh = %d, hw rdt = %d\n", E1000_READ_REG(&adapter->hw, E1000_RDH(0)), E1000_READ_REG(&adapter->hw, E1000_RDT(0))); device_printf(dev, "Tx Queue Status = %d\n", txr->queue_status); device_printf(dev, "TX descriptors avail = %d\n", txr->tx_avail); device_printf(dev, "Tx Descriptors avail failure = %ld\n", txr->no_desc_avail); device_printf(dev, "RX discarded packets = %ld\n", rxr->rx_discarded); device_printf(dev, "RX Next to Check = %d\n", rxr->next_to_check); device_printf(dev, "RX Next to Refresh = %d\n", rxr->next_to_refresh); } Index: stable/7/sys/dev/et/if_et.c =================================================================== --- stable/7/sys/dev/et/if_et.c (revision 242221) +++ stable/7/sys/dev/et/if_et.c (revision 242222) @@ -1,2731 +1,2731 @@ /*- * Copyright (c) 2007 Sepherosa Ziehau. All rights reserved. * * This code is derived from software contributed to The DragonFly Project * by Sepherosa Ziehau * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in * the documentation and/or other materials provided with the * distribution. * 3. Neither the name of The DragonFly Project nor the names of its * contributors may be used to endorse or promote products derived * from this software without specific, prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $DragonFly: src/sys/dev/netif/et/if_et.c,v 1.10 2008/05/18 07:47:14 sephe Exp $ */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "miibus_if.h" MODULE_DEPEND(et, pci, 1, 1, 1); MODULE_DEPEND(et, ether, 1, 1, 1); MODULE_DEPEND(et, miibus, 1, 1, 1); /* Tunables. */ static int msi_disable = 0; TUNABLE_INT("hw.et.msi_disable", &msi_disable); #define ET_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP) static int et_probe(device_t); static int et_attach(device_t); static int et_detach(device_t); static int et_shutdown(device_t); static int et_suspend(device_t); static int et_resume(device_t); static int et_miibus_readreg(device_t, int, int); static int et_miibus_writereg(device_t, int, int, int); static void et_miibus_statchg(device_t); static void et_init_locked(struct et_softc *); static void et_init(void *); static int et_ioctl(struct ifnet *, u_long, caddr_t); static void et_start_locked(struct ifnet *); static void et_start(struct ifnet *); static int et_watchdog(struct et_softc *); static int et_ifmedia_upd_locked(struct ifnet *); static int et_ifmedia_upd(struct ifnet *); static void et_ifmedia_sts(struct ifnet *, struct ifmediareq *); static void et_add_sysctls(struct et_softc *); static int et_sysctl_rx_intr_npkts(SYSCTL_HANDLER_ARGS); static int et_sysctl_rx_intr_delay(SYSCTL_HANDLER_ARGS); static void et_intr(void *); static void et_rxeof(struct et_softc *); static void et_txeof(struct et_softc *); static int et_dma_alloc(struct et_softc *); static void et_dma_free(struct et_softc *); static void et_dma_map_addr(void *, bus_dma_segment_t *, int, int); static int et_dma_ring_alloc(struct et_softc *, bus_size_t, bus_size_t, bus_dma_tag_t *, uint8_t **, bus_dmamap_t *, bus_addr_t *, const char *); static void et_dma_ring_free(struct et_softc *, bus_dma_tag_t *, uint8_t **, bus_dmamap_t *); static void et_init_tx_ring(struct et_softc *); static int et_init_rx_ring(struct et_softc *); static void et_free_tx_ring(struct et_softc *); static void et_free_rx_ring(struct et_softc *); static int et_encap(struct et_softc *, struct mbuf **); static int et_newbuf_cluster(struct et_rxbuf_data *, int); static int et_newbuf_hdr(struct et_rxbuf_data *, int); static void et_rxbuf_discard(struct et_rxbuf_data *, int); static void et_stop(struct et_softc *); static int et_chip_init(struct et_softc *); static void et_chip_attach(struct et_softc *); static void et_init_mac(struct et_softc *); static void et_init_rxmac(struct et_softc *); static void et_init_txmac(struct et_softc *); static int et_init_rxdma(struct et_softc *); static int et_init_txdma(struct et_softc *); static int et_start_rxdma(struct et_softc *); static int et_start_txdma(struct et_softc *); static int et_stop_rxdma(struct et_softc *); static int et_stop_txdma(struct et_softc *); static void et_reset(struct et_softc *); static int et_bus_config(struct et_softc *); static void et_get_eaddr(device_t, uint8_t[]); static void et_setmulti(struct et_softc *); static void et_tick(void *); static void et_stats_update(struct et_softc *); static const struct et_dev { uint16_t vid; uint16_t did; const char *desc; } et_devices[] = { { PCI_VENDOR_LUCENT, PCI_PRODUCT_LUCENT_ET1310, "Agere ET1310 Gigabit Ethernet" }, { PCI_VENDOR_LUCENT, PCI_PRODUCT_LUCENT_ET1310_FAST, "Agere ET1310 Fast Ethernet" }, { 0, 0, NULL } }; static device_method_t et_methods[] = { DEVMETHOD(device_probe, et_probe), DEVMETHOD(device_attach, et_attach), DEVMETHOD(device_detach, et_detach), DEVMETHOD(device_shutdown, et_shutdown), DEVMETHOD(device_suspend, et_suspend), DEVMETHOD(device_resume, et_resume), DEVMETHOD(bus_print_child, bus_generic_print_child), DEVMETHOD(bus_driver_added, bus_generic_driver_added), DEVMETHOD(miibus_readreg, et_miibus_readreg), DEVMETHOD(miibus_writereg, et_miibus_writereg), DEVMETHOD(miibus_statchg, et_miibus_statchg), { 0, 0 } }; static driver_t et_driver = { "et", et_methods, sizeof(struct et_softc) }; static devclass_t et_devclass; DRIVER_MODULE(et, pci, et_driver, et_devclass, 0, 0); DRIVER_MODULE(miibus, et, miibus_driver, miibus_devclass, 0, 0); static int et_rx_intr_npkts = 32; static int et_rx_intr_delay = 20; /* x10 usec */ static int et_tx_intr_nsegs = 126; static uint32_t et_timer = 1000 * 1000 * 1000; /* nanosec */ TUNABLE_INT("hw.et.timer", &et_timer); TUNABLE_INT("hw.et.rx_intr_npkts", &et_rx_intr_npkts); TUNABLE_INT("hw.et.rx_intr_delay", &et_rx_intr_delay); TUNABLE_INT("hw.et.tx_intr_nsegs", &et_tx_intr_nsegs); static int et_probe(device_t dev) { const struct et_dev *d; uint16_t did, vid; vid = pci_get_vendor(dev); did = pci_get_device(dev); for (d = et_devices; d->desc != NULL; ++d) { if (vid == d->vid && did == d->did) { device_set_desc(dev, d->desc); return (BUS_PROBE_DEFAULT); } } return (ENXIO); } static int et_attach(device_t dev) { struct et_softc *sc; struct ifnet *ifp; uint8_t eaddr[ETHER_ADDR_LEN]; uint32_t pmcfg; int cap, error, msic; sc = device_get_softc(dev); sc->dev = dev; mtx_init(&sc->sc_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK, MTX_DEF); callout_init_mtx(&sc->sc_tick, &sc->sc_mtx, 0); ifp = sc->ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "can not if_alloc()\n"); error = ENOSPC; goto fail; } /* * Initialize tunables */ sc->sc_rx_intr_npkts = et_rx_intr_npkts; sc->sc_rx_intr_delay = et_rx_intr_delay; sc->sc_tx_intr_nsegs = et_tx_intr_nsegs; sc->sc_timer = et_timer; /* Enable bus mastering */ pci_enable_busmaster(dev); /* * Allocate IO memory */ sc->sc_mem_rid = PCIR_BAR(0); sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &sc->sc_mem_rid, RF_ACTIVE); if (sc->sc_mem_res == NULL) { device_printf(dev, "can't allocate IO memory\n"); return (ENXIO); } msic = 0; if (pci_find_extcap(dev, PCIY_EXPRESS, &cap) == 0) { sc->sc_expcap = cap; sc->sc_flags |= ET_FLAG_PCIE; msic = pci_msi_count(dev); if (bootverbose) device_printf(dev, "MSI count: %d\n", msic); } if (msic > 0 && msi_disable == 0) { msic = 1; if (pci_alloc_msi(dev, &msic) == 0) { if (msic == 1) { device_printf(dev, "Using %d MSI message\n", msic); sc->sc_flags |= ET_FLAG_MSI; } else pci_release_msi(dev); } } /* * Allocate IRQ */ if ((sc->sc_flags & ET_FLAG_MSI) == 0) { sc->sc_irq_rid = 0; sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->sc_irq_rid, RF_SHAREABLE | RF_ACTIVE); } else { sc->sc_irq_rid = 1; sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->sc_irq_rid, RF_ACTIVE); } if (sc->sc_irq_res == NULL) { device_printf(dev, "can't allocate irq\n"); error = ENXIO; goto fail; } if (pci_get_device(dev) == PCI_PRODUCT_LUCENT_ET1310_FAST) sc->sc_flags |= ET_FLAG_FASTETHER; error = et_bus_config(sc); if (error) goto fail; et_get_eaddr(dev, eaddr); /* Take PHY out of COMA and enable clocks. */ pmcfg = ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE | ET_PM_RXCLK_GATE; if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0) pmcfg |= EM_PM_GIGEPHY_ENB; CSR_WRITE_4(sc, ET_PM, pmcfg); et_reset(sc); error = et_dma_alloc(sc); if (error) goto fail; ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_init = et_init; ifp->if_ioctl = et_ioctl; ifp->if_start = et_start; ifp->if_capabilities = IFCAP_TXCSUM | IFCAP_VLAN_MTU; ifp->if_capenable = ifp->if_capabilities; ifp->if_snd.ifq_drv_maxlen = ET_TX_NDESC - 1; IFQ_SET_MAXLEN(&ifp->if_snd, ET_TX_NDESC - 1); IFQ_SET_READY(&ifp->if_snd); et_chip_attach(sc); error = mii_attach(dev, &sc->sc_miibus, ifp, et_ifmedia_upd, et_ifmedia_sts, BMSR_DEFCAPMASK, MII_PHY_ANY, MII_OFFSET_ANY, MIIF_DOPAUSE); if (error) { device_printf(dev, "attaching PHYs failed\n"); goto fail; } ether_ifattach(ifp, eaddr); /* Tell the upper layer(s) we support long frames. */ ifp->if_hdrlen = sizeof(struct ether_vlan_header); error = bus_setup_intr(dev, sc->sc_irq_res, INTR_TYPE_NET | INTR_MPSAFE, NULL, et_intr, sc, &sc->sc_irq_handle); if (error) { ether_ifdetach(ifp); device_printf(dev, "can't setup intr\n"); goto fail; } et_add_sysctls(sc); return (0); fail: et_detach(dev); return (error); } static int et_detach(device_t dev) { struct et_softc *sc; sc = device_get_softc(dev); if (device_is_attached(dev)) { ether_ifdetach(sc->ifp); ET_LOCK(sc); et_stop(sc); ET_UNLOCK(sc); callout_drain(&sc->sc_tick); } if (sc->sc_miibus != NULL) device_delete_child(dev, sc->sc_miibus); bus_generic_detach(dev); if (sc->sc_irq_handle != NULL) bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_irq_handle); if (sc->sc_irq_res != NULL) bus_release_resource(dev, SYS_RES_IRQ, rman_get_rid(sc->sc_irq_res), sc->sc_irq_res); if ((sc->sc_flags & ET_FLAG_MSI) != 0) pci_release_msi(dev); if (sc->sc_mem_res != NULL) bus_release_resource(dev, SYS_RES_MEMORY, rman_get_rid(sc->sc_mem_res), sc->sc_mem_res); if (sc->ifp != NULL) if_free(sc->ifp); et_dma_free(sc); mtx_destroy(&sc->sc_mtx); return (0); } static int et_shutdown(device_t dev) { struct et_softc *sc; sc = device_get_softc(dev); ET_LOCK(sc); et_stop(sc); ET_UNLOCK(sc); return (0); } static int et_miibus_readreg(device_t dev, int phy, int reg) { struct et_softc *sc; uint32_t val; int i, ret; sc = device_get_softc(dev); /* Stop any pending operations */ CSR_WRITE_4(sc, ET_MII_CMD, 0); val = (phy << ET_MII_ADDR_PHY_SHIFT) & ET_MII_ADDR_PHY_MASK; val |= (reg << ET_MII_ADDR_REG_SHIFT) & ET_MII_ADDR_REG_MASK; CSR_WRITE_4(sc, ET_MII_ADDR, val); /* Start reading */ CSR_WRITE_4(sc, ET_MII_CMD, ET_MII_CMD_READ); #define NRETRY 50 for (i = 0; i < NRETRY; ++i) { val = CSR_READ_4(sc, ET_MII_IND); if ((val & (ET_MII_IND_BUSY | ET_MII_IND_INVALID)) == 0) break; DELAY(50); } if (i == NRETRY) { if_printf(sc->ifp, "read phy %d, reg %d timed out\n", phy, reg); ret = 0; goto back; } #undef NRETRY val = CSR_READ_4(sc, ET_MII_STAT); ret = val & ET_MII_STAT_VALUE_MASK; back: /* Make sure that the current operation is stopped */ CSR_WRITE_4(sc, ET_MII_CMD, 0); return (ret); } static int et_miibus_writereg(device_t dev, int phy, int reg, int val0) { struct et_softc *sc; uint32_t val; int i; sc = device_get_softc(dev); /* Stop any pending operations */ CSR_WRITE_4(sc, ET_MII_CMD, 0); val = (phy << ET_MII_ADDR_PHY_SHIFT) & ET_MII_ADDR_PHY_MASK; val |= (reg << ET_MII_ADDR_REG_SHIFT) & ET_MII_ADDR_REG_MASK; CSR_WRITE_4(sc, ET_MII_ADDR, val); /* Start writing */ CSR_WRITE_4(sc, ET_MII_CTRL, (val0 << ET_MII_CTRL_VALUE_SHIFT) & ET_MII_CTRL_VALUE_MASK); #define NRETRY 100 for (i = 0; i < NRETRY; ++i) { val = CSR_READ_4(sc, ET_MII_IND); if ((val & ET_MII_IND_BUSY) == 0) break; DELAY(50); } if (i == NRETRY) { if_printf(sc->ifp, "write phy %d, reg %d timed out\n", phy, reg); et_miibus_readreg(dev, phy, reg); } #undef NRETRY /* Make sure that the current operation is stopped */ CSR_WRITE_4(sc, ET_MII_CMD, 0); return (0); } static void et_miibus_statchg(device_t dev) { struct et_softc *sc; struct mii_data *mii; struct ifnet *ifp; uint32_t cfg1, cfg2, ctrl; int i; sc = device_get_softc(dev); mii = device_get_softc(sc->sc_miibus); ifp = sc->ifp; if (mii == NULL || ifp == NULL || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; sc->sc_flags &= ~ET_FLAG_LINK; if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: sc->sc_flags |= ET_FLAG_LINK; break; case IFM_1000_T: if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0) sc->sc_flags |= ET_FLAG_LINK; break; } } /* XXX Stop TX/RX MAC? */ if ((sc->sc_flags & ET_FLAG_LINK) == 0) return; /* Program MACs with resolved speed/duplex/flow-control. */ ctrl = CSR_READ_4(sc, ET_MAC_CTRL); ctrl &= ~(ET_MAC_CTRL_GHDX | ET_MAC_CTRL_MODE_MII); cfg1 = CSR_READ_4(sc, ET_MAC_CFG1); cfg1 &= ~(ET_MAC_CFG1_TXFLOW | ET_MAC_CFG1_RXFLOW | ET_MAC_CFG1_LOOPBACK); cfg2 = CSR_READ_4(sc, ET_MAC_CFG2); cfg2 &= ~(ET_MAC_CFG2_MODE_MII | ET_MAC_CFG2_MODE_GMII | ET_MAC_CFG2_FDX | ET_MAC_CFG2_BIGFRM); cfg2 |= ET_MAC_CFG2_LENCHK | ET_MAC_CFG2_CRC | ET_MAC_CFG2_PADCRC | ((7 << ET_MAC_CFG2_PREAMBLE_LEN_SHIFT) & ET_MAC_CFG2_PREAMBLE_LEN_MASK); if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T) cfg2 |= ET_MAC_CFG2_MODE_GMII; else { cfg2 |= ET_MAC_CFG2_MODE_MII; ctrl |= ET_MAC_CTRL_MODE_MII; } if (IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) { cfg2 |= ET_MAC_CFG2_FDX; /* * Controller lacks automatic TX pause frame * generation so it should be handled by driver. * Even though driver can send pause frame with * arbitrary pause time, controller does not * provide a way that tells how many free RX * buffers are available in controller. This * limitation makes it hard to generate XON frame * in time on driver side so don't enable TX flow * control. */ #ifdef notyet if (IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) cfg1 |= ET_MAC_CFG1_TXFLOW; #endif if (IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) cfg1 |= ET_MAC_CFG1_RXFLOW; } else ctrl |= ET_MAC_CTRL_GHDX; CSR_WRITE_4(sc, ET_MAC_CTRL, ctrl); CSR_WRITE_4(sc, ET_MAC_CFG2, cfg2); cfg1 |= ET_MAC_CFG1_TXEN | ET_MAC_CFG1_RXEN; CSR_WRITE_4(sc, ET_MAC_CFG1, cfg1); #define NRETRY 50 for (i = 0; i < NRETRY; ++i) { cfg1 = CSR_READ_4(sc, ET_MAC_CFG1); if ((cfg1 & (ET_MAC_CFG1_SYNC_TXEN | ET_MAC_CFG1_SYNC_RXEN)) == (ET_MAC_CFG1_SYNC_TXEN | ET_MAC_CFG1_SYNC_RXEN)) break; DELAY(100); } if (i == NRETRY) if_printf(ifp, "can't enable RX/TX\n"); sc->sc_flags |= ET_FLAG_TXRX_ENABLED; #undef NRETRY } static int et_ifmedia_upd_locked(struct ifnet *ifp) { struct et_softc *sc; struct mii_data *mii; struct mii_softc *miisc; sc = ifp->if_softc; mii = device_get_softc(sc->sc_miibus); LIST_FOREACH(miisc, &mii->mii_phys, mii_list) mii_phy_reset(miisc); return (mii_mediachg(mii)); } static int et_ifmedia_upd(struct ifnet *ifp) { struct et_softc *sc; int res; sc = ifp->if_softc; ET_LOCK(sc); res = et_ifmedia_upd_locked(ifp); ET_UNLOCK(sc); return (res); } static void et_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { struct et_softc *sc; struct mii_data *mii; sc = ifp->if_softc; ET_LOCK(sc); if ((ifp->if_flags & IFF_UP) == 0) { ET_UNLOCK(sc); return; } mii = device_get_softc(sc->sc_miibus); mii_pollstat(mii); ifmr->ifm_active = mii->mii_media_active; ifmr->ifm_status = mii->mii_media_status; ET_UNLOCK(sc); } static void et_stop(struct et_softc *sc) { struct ifnet *ifp; ET_LOCK_ASSERT(sc); ifp = sc->ifp; callout_stop(&sc->sc_tick); /* Disable interrupts. */ CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff); CSR_WRITE_4(sc, ET_MAC_CFG1, CSR_READ_4(sc, ET_MAC_CFG1) & ~( ET_MAC_CFG1_TXEN | ET_MAC_CFG1_RXEN)); DELAY(100); et_stop_rxdma(sc); et_stop_txdma(sc); et_stats_update(sc); et_free_tx_ring(sc); et_free_rx_ring(sc); sc->sc_tx = 0; sc->sc_tx_intr = 0; sc->sc_flags &= ~ET_FLAG_TXRX_ENABLED; sc->watchdog_timer = 0; ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); } static int et_bus_config(struct et_softc *sc) { uint32_t val, max_plsz; uint16_t ack_latency, replay_timer; /* * Test whether EEPROM is valid * NOTE: Read twice to get the correct value */ pci_read_config(sc->dev, ET_PCIR_EEPROM_STATUS, 1); val = pci_read_config(sc->dev, ET_PCIR_EEPROM_STATUS, 1); if (val & ET_PCIM_EEPROM_STATUS_ERROR) { device_printf(sc->dev, "EEPROM status error 0x%02x\n", val); return (ENXIO); } /* TODO: LED */ if ((sc->sc_flags & ET_FLAG_PCIE) == 0) return (0); /* * Configure ACK latency and replay timer according to * max playload size */ val = pci_read_config(sc->dev, - sc->sc_expcap + PCIR_EXPRESS_DEVICE_CAP, 4); - max_plsz = val & PCIM_EXP_CAP_MAX_PAYLOAD; + sc->sc_expcap + PCIER_DEVICE_CAP, 4); + max_plsz = val & PCIEM_CAP_MAX_PAYLOAD; switch (max_plsz) { case ET_PCIV_DEVICE_CAPS_PLSZ_128: ack_latency = ET_PCIV_ACK_LATENCY_128; replay_timer = ET_PCIV_REPLAY_TIMER_128; break; case ET_PCIV_DEVICE_CAPS_PLSZ_256: ack_latency = ET_PCIV_ACK_LATENCY_256; replay_timer = ET_PCIV_REPLAY_TIMER_256; break; default: ack_latency = pci_read_config(sc->dev, ET_PCIR_ACK_LATENCY, 2); replay_timer = pci_read_config(sc->dev, ET_PCIR_REPLAY_TIMER, 2); device_printf(sc->dev, "ack latency %u, replay timer %u\n", ack_latency, replay_timer); break; } if (ack_latency != 0) { pci_write_config(sc->dev, ET_PCIR_ACK_LATENCY, ack_latency, 2); pci_write_config(sc->dev, ET_PCIR_REPLAY_TIMER, replay_timer, 2); } /* * Set L0s and L1 latency timer to 2us */ val = pci_read_config(sc->dev, ET_PCIR_L0S_L1_LATENCY, 4); - val &= ~(PCIM_LINK_CAP_L0S_EXIT | PCIM_LINK_CAP_L1_EXIT); + val &= ~(PCIEM_LINK_CAP_L0S_EXIT | PCIEM_LINK_CAP_L1_EXIT); /* L0s exit latency : 2us */ val |= 0x00005000; /* L1 exit latency : 2us */ val |= 0x00028000; pci_write_config(sc->dev, ET_PCIR_L0S_L1_LATENCY, val, 4); /* * Set max read request size to 2048 bytes */ pci_set_max_read_req(sc->dev, 2048); return (0); } static void et_get_eaddr(device_t dev, uint8_t eaddr[]) { uint32_t val; int i; val = pci_read_config(dev, ET_PCIR_MAC_ADDR0, 4); for (i = 0; i < 4; ++i) eaddr[i] = (val >> (8 * i)) & 0xff; val = pci_read_config(dev, ET_PCIR_MAC_ADDR1, 2); for (; i < ETHER_ADDR_LEN; ++i) eaddr[i] = (val >> (8 * (i - 4))) & 0xff; } static void et_reset(struct et_softc *sc) { CSR_WRITE_4(sc, ET_MAC_CFG1, ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC | ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC | ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST); CSR_WRITE_4(sc, ET_SWRST, ET_SWRST_TXDMA | ET_SWRST_RXDMA | ET_SWRST_TXMAC | ET_SWRST_RXMAC | ET_SWRST_MAC | ET_SWRST_MAC_STAT | ET_SWRST_MMC); CSR_WRITE_4(sc, ET_MAC_CFG1, ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC | ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC); CSR_WRITE_4(sc, ET_MAC_CFG1, 0); /* Disable interrupts. */ CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff); } struct et_dmamap_arg { bus_addr_t et_busaddr; }; static void et_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error) { struct et_dmamap_arg *ctx; if (error) return; KASSERT(nseg == 1, ("%s: %d segments returned!", __func__, nseg)); ctx = arg; ctx->et_busaddr = segs->ds_addr; } static int et_dma_ring_alloc(struct et_softc *sc, bus_size_t alignment, bus_size_t maxsize, bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map, bus_addr_t *paddr, const char *msg) { struct et_dmamap_arg ctx; int error; error = bus_dma_tag_create(sc->sc_dtag, alignment, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, maxsize, 1, maxsize, 0, NULL, NULL, tag); if (error != 0) { device_printf(sc->dev, "could not create %s dma tag\n", msg); return (error); } /* Allocate DMA'able memory for ring. */ error = bus_dmamem_alloc(*tag, (void **)ring, BUS_DMA_NOWAIT | BUS_DMA_ZERO | BUS_DMA_COHERENT, map); if (error != 0) { device_printf(sc->dev, "could not allocate DMA'able memory for %s\n", msg); return (error); } /* Load the address of the ring. */ ctx.et_busaddr = 0; error = bus_dmamap_load(*tag, *map, *ring, maxsize, et_dma_map_addr, &ctx, BUS_DMA_NOWAIT); if (error != 0) { device_printf(sc->dev, "could not load DMA'able memory for %s\n", msg); return (error); } *paddr = ctx.et_busaddr; return (0); } static void et_dma_ring_free(struct et_softc *sc, bus_dma_tag_t *tag, uint8_t **ring, bus_dmamap_t *map) { if (*map != NULL) bus_dmamap_unload(*tag, *map); if (*map != NULL && *ring != NULL) { bus_dmamem_free(*tag, *ring, *map); *ring = NULL; *map = NULL; } if (*tag) { bus_dma_tag_destroy(*tag); *tag = NULL; } } static int et_dma_alloc(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_rxdesc_ring *rx_ring; struct et_rxstat_ring *rxst_ring; struct et_rxstatus_data *rxsd; struct et_rxbuf_data *rbd; struct et_txbuf_data *tbd; struct et_txstatus_data *txsd; int i, error; error = bus_dma_tag_create(bus_get_dma_tag(sc->dev), 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL, &sc->sc_dtag); if (error != 0) { device_printf(sc->dev, "could not allocate parent dma tag\n"); return (error); } /* TX ring. */ tx_ring = &sc->sc_tx_ring; error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_TX_RING_SIZE, &tx_ring->tr_dtag, (uint8_t **)&tx_ring->tr_desc, &tx_ring->tr_dmap, &tx_ring->tr_paddr, "TX ring"); if (error) return (error); /* TX status block. */ txsd = &sc->sc_tx_status; error = et_dma_ring_alloc(sc, ET_STATUS_ALIGN, sizeof(uint32_t), &txsd->txsd_dtag, (uint8_t **)&txsd->txsd_status, &txsd->txsd_dmap, &txsd->txsd_paddr, "TX status block"); if (error) return (error); /* RX ring 0, used as to recive small sized frames. */ rx_ring = &sc->sc_rx_ring[0]; error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RX_RING_SIZE, &rx_ring->rr_dtag, (uint8_t **)&rx_ring->rr_desc, &rx_ring->rr_dmap, &rx_ring->rr_paddr, "RX ring 0"); rx_ring->rr_posreg = ET_RX_RING0_POS; if (error) return (error); /* RX ring 1, used as to store normal sized frames. */ rx_ring = &sc->sc_rx_ring[1]; error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RX_RING_SIZE, &rx_ring->rr_dtag, (uint8_t **)&rx_ring->rr_desc, &rx_ring->rr_dmap, &rx_ring->rr_paddr, "RX ring 1"); rx_ring->rr_posreg = ET_RX_RING1_POS; if (error) return (error); /* RX stat ring. */ rxst_ring = &sc->sc_rxstat_ring; error = et_dma_ring_alloc(sc, ET_RING_ALIGN, ET_RXSTAT_RING_SIZE, &rxst_ring->rsr_dtag, (uint8_t **)&rxst_ring->rsr_stat, &rxst_ring->rsr_dmap, &rxst_ring->rsr_paddr, "RX stat ring"); if (error) return (error); /* RX status block. */ rxsd = &sc->sc_rx_status; error = et_dma_ring_alloc(sc, ET_STATUS_ALIGN, sizeof(struct et_rxstatus), &rxsd->rxsd_dtag, (uint8_t **)&rxsd->rxsd_status, &rxsd->rxsd_dmap, &rxsd->rxsd_paddr, "RX status block"); if (error) return (error); /* Create parent DMA tag for mbufs. */ error = bus_dma_tag_create(bus_get_dma_tag(sc->dev), 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL, &sc->sc_mbuf_dtag); if (error != 0) { device_printf(sc->dev, "could not allocate parent dma tag for mbuf\n"); return (error); } /* Create DMA tag for mini RX mbufs to use RX ring 0. */ error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MHLEN, 1, MHLEN, 0, NULL, NULL, &sc->sc_rx_mini_tag); if (error) { device_printf(sc->dev, "could not create mini RX dma tag\n"); return (error); } /* Create DMA tag for standard RX mbufs to use RX ring 1. */ error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->sc_rx_tag); if (error) { device_printf(sc->dev, "could not create RX dma tag\n"); return (error); } /* Create DMA tag for TX mbufs. */ error = bus_dma_tag_create(sc->sc_mbuf_dtag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES * ET_NSEG_MAX, ET_NSEG_MAX, MCLBYTES, 0, NULL, NULL, &sc->sc_tx_tag); if (error) { device_printf(sc->dev, "could not create TX dma tag\n"); return (error); } /* Initialize RX ring 0. */ rbd = &sc->sc_rx_data[0]; rbd->rbd_bufsize = ET_RXDMA_CTRL_RING0_128; rbd->rbd_newbuf = et_newbuf_hdr; rbd->rbd_discard = et_rxbuf_discard; rbd->rbd_softc = sc; rbd->rbd_ring = &sc->sc_rx_ring[0]; /* Create DMA maps for mini RX buffers, ring 0. */ for (i = 0; i < ET_RX_NDESC; i++) { error = bus_dmamap_create(sc->sc_rx_mini_tag, 0, &rbd->rbd_buf[i].rb_dmap); if (error) { device_printf(sc->dev, "could not create DMA map for mini RX mbufs\n"); return (error); } } /* Create a spare DMA map for mini RX buffers, ring 0. */ error = bus_dmamap_create(sc->sc_rx_mini_tag, 0, &sc->sc_rx_mini_sparemap); if (error) { device_printf(sc->dev, "could not create spare DMA map for mini RX mbuf\n"); return (error); } /* Initialize RX ring 1. */ rbd = &sc->sc_rx_data[1]; rbd->rbd_bufsize = ET_RXDMA_CTRL_RING1_2048; rbd->rbd_newbuf = et_newbuf_cluster; rbd->rbd_discard = et_rxbuf_discard; rbd->rbd_softc = sc; rbd->rbd_ring = &sc->sc_rx_ring[1]; /* Create DMA maps for standard RX buffers, ring 1. */ for (i = 0; i < ET_RX_NDESC; i++) { error = bus_dmamap_create(sc->sc_rx_tag, 0, &rbd->rbd_buf[i].rb_dmap); if (error) { device_printf(sc->dev, "could not create DMA map for mini RX mbufs\n"); return (error); } } /* Create a spare DMA map for standard RX buffers, ring 1. */ error = bus_dmamap_create(sc->sc_rx_tag, 0, &sc->sc_rx_sparemap); if (error) { device_printf(sc->dev, "could not create spare DMA map for RX mbuf\n"); return (error); } /* Create DMA maps for TX buffers. */ tbd = &sc->sc_tx_data; for (i = 0; i < ET_TX_NDESC; i++) { error = bus_dmamap_create(sc->sc_tx_tag, 0, &tbd->tbd_buf[i].tb_dmap); if (error) { device_printf(sc->dev, "could not create DMA map for TX mbufs\n"); return (error); } } return (0); } static void et_dma_free(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_rxdesc_ring *rx_ring; struct et_txstatus_data *txsd; struct et_rxstat_ring *rxst_ring; struct et_rxstatus_data *rxsd; struct et_rxbuf_data *rbd; struct et_txbuf_data *tbd; int i; /* Destroy DMA maps for mini RX buffers, ring 0. */ rbd = &sc->sc_rx_data[0]; for (i = 0; i < ET_RX_NDESC; i++) { if (rbd->rbd_buf[i].rb_dmap) { bus_dmamap_destroy(sc->sc_rx_mini_tag, rbd->rbd_buf[i].rb_dmap); rbd->rbd_buf[i].rb_dmap = NULL; } } if (sc->sc_rx_mini_sparemap) { bus_dmamap_destroy(sc->sc_rx_mini_tag, sc->sc_rx_mini_sparemap); sc->sc_rx_mini_sparemap = NULL; } if (sc->sc_rx_mini_tag) { bus_dma_tag_destroy(sc->sc_rx_mini_tag); sc->sc_rx_mini_tag = NULL; } /* Destroy DMA maps for standard RX buffers, ring 1. */ rbd = &sc->sc_rx_data[1]; for (i = 0; i < ET_RX_NDESC; i++) { if (rbd->rbd_buf[i].rb_dmap) { bus_dmamap_destroy(sc->sc_rx_tag, rbd->rbd_buf[i].rb_dmap); rbd->rbd_buf[i].rb_dmap = NULL; } } if (sc->sc_rx_sparemap) { bus_dmamap_destroy(sc->sc_rx_tag, sc->sc_rx_sparemap); sc->sc_rx_sparemap = NULL; } if (sc->sc_rx_tag) { bus_dma_tag_destroy(sc->sc_rx_tag); sc->sc_rx_tag = NULL; } /* Destroy DMA maps for TX buffers. */ tbd = &sc->sc_tx_data; for (i = 0; i < ET_TX_NDESC; i++) { if (tbd->tbd_buf[i].tb_dmap) { bus_dmamap_destroy(sc->sc_tx_tag, tbd->tbd_buf[i].tb_dmap); tbd->tbd_buf[i].tb_dmap = NULL; } } if (sc->sc_tx_tag) { bus_dma_tag_destroy(sc->sc_tx_tag); sc->sc_tx_tag = NULL; } /* Destroy mini RX ring, ring 0. */ rx_ring = &sc->sc_rx_ring[0]; et_dma_ring_free(sc, &rx_ring->rr_dtag, (void *)&rx_ring->rr_desc, &rx_ring->rr_dmap); /* Destroy standard RX ring, ring 1. */ rx_ring = &sc->sc_rx_ring[1]; et_dma_ring_free(sc, &rx_ring->rr_dtag, (void *)&rx_ring->rr_desc, &rx_ring->rr_dmap); /* Destroy RX stat ring. */ rxst_ring = &sc->sc_rxstat_ring; et_dma_ring_free(sc, &rxst_ring->rsr_dtag, (void *)&rxst_ring->rsr_stat, &rxst_ring->rsr_dmap); /* Destroy RX status block. */ rxsd = &sc->sc_rx_status; et_dma_ring_free(sc, &rxst_ring->rsr_dtag, (void *)&rxst_ring->rsr_stat, &rxst_ring->rsr_dmap); /* Destroy TX ring. */ tx_ring = &sc->sc_tx_ring; et_dma_ring_free(sc, &tx_ring->tr_dtag, (void *)&tx_ring->tr_desc, &tx_ring->tr_dmap); /* Destroy TX status block. */ txsd = &sc->sc_tx_status; et_dma_ring_free(sc, &txsd->txsd_dtag, (void *)&txsd->txsd_status, &txsd->txsd_dmap); /* Destroy the parent tag. */ if (sc->sc_dtag) { bus_dma_tag_destroy(sc->sc_dtag); sc->sc_dtag = NULL; } } static void et_chip_attach(struct et_softc *sc) { uint32_t val; /* * Perform minimal initialization */ /* Disable loopback */ CSR_WRITE_4(sc, ET_LOOPBACK, 0); /* Reset MAC */ CSR_WRITE_4(sc, ET_MAC_CFG1, ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC | ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC | ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST); /* * Setup half duplex mode */ val = (10 << ET_MAC_HDX_ALT_BEB_TRUNC_SHIFT) | (15 << ET_MAC_HDX_REXMIT_MAX_SHIFT) | (55 << ET_MAC_HDX_COLLWIN_SHIFT) | ET_MAC_HDX_EXC_DEFER; CSR_WRITE_4(sc, ET_MAC_HDX, val); /* Clear MAC control */ CSR_WRITE_4(sc, ET_MAC_CTRL, 0); /* Reset MII */ CSR_WRITE_4(sc, ET_MII_CFG, ET_MII_CFG_CLKRST); /* Bring MAC out of reset state */ CSR_WRITE_4(sc, ET_MAC_CFG1, 0); /* Enable memory controllers */ CSR_WRITE_4(sc, ET_MMC_CTRL, ET_MMC_CTRL_ENABLE); } static void et_intr(void *xsc) { struct et_softc *sc; struct ifnet *ifp; uint32_t status; sc = xsc; ET_LOCK(sc); ifp = sc->ifp; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) goto done; status = CSR_READ_4(sc, ET_INTR_STATUS); if ((status & ET_INTRS) == 0) goto done; /* Disable further interrupts. */ CSR_WRITE_4(sc, ET_INTR_MASK, 0xffffffff); if (status & (ET_INTR_RXDMA_ERROR | ET_INTR_TXDMA_ERROR)) { device_printf(sc->dev, "DMA error(0x%08x) -- resetting\n", status); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; et_init_locked(sc); ET_UNLOCK(sc); return; } if (status & ET_INTR_RXDMA) et_rxeof(sc); if (status & (ET_INTR_TXDMA | ET_INTR_TIMER)) et_txeof(sc); if (status & ET_INTR_TIMER) CSR_WRITE_4(sc, ET_TIMER, sc->sc_timer); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { CSR_WRITE_4(sc, ET_INTR_MASK, ~ET_INTRS); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) et_start_locked(ifp); } done: ET_UNLOCK(sc); } static void et_init_locked(struct et_softc *sc) { struct ifnet *ifp; int error; ET_LOCK_ASSERT(sc); ifp = sc->ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; et_stop(sc); et_reset(sc); et_init_tx_ring(sc); error = et_init_rx_ring(sc); if (error) return; error = et_chip_init(sc); if (error) goto fail; /* * Start TX/RX DMA engine */ error = et_start_rxdma(sc); if (error) return; error = et_start_txdma(sc); if (error) return; /* Enable interrupts. */ CSR_WRITE_4(sc, ET_INTR_MASK, ~ET_INTRS); CSR_WRITE_4(sc, ET_TIMER, sc->sc_timer); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; sc->sc_flags &= ~ET_FLAG_LINK; et_ifmedia_upd_locked(ifp); callout_reset(&sc->sc_tick, hz, et_tick, sc); fail: if (error) et_stop(sc); } static void et_init(void *xsc) { struct et_softc *sc = xsc; ET_LOCK(sc); et_init_locked(sc); ET_UNLOCK(sc); } static int et_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct et_softc *sc; struct mii_data *mii; struct ifreq *ifr; int error, mask, max_framelen; sc = ifp->if_softc; ifr = (struct ifreq *)data; error = 0; /* XXX LOCKSUSED */ switch (cmd) { case SIOCSIFFLAGS: ET_LOCK(sc); if (ifp->if_flags & IFF_UP) { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { if ((ifp->if_flags ^ sc->sc_if_flags) & (IFF_ALLMULTI | IFF_PROMISC | IFF_BROADCAST)) et_setmulti(sc); } else { et_init_locked(sc); } } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) et_stop(sc); } sc->sc_if_flags = ifp->if_flags; ET_UNLOCK(sc); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: mii = device_get_softc(sc->sc_miibus); error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd); break; case SIOCADDMULTI: case SIOCDELMULTI: if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ET_LOCK(sc); et_setmulti(sc); ET_UNLOCK(sc); } break; case SIOCSIFMTU: ET_LOCK(sc); #if 0 if (sc->sc_flags & ET_FLAG_JUMBO) max_framelen = ET_JUMBO_FRAMELEN; else #endif max_framelen = MCLBYTES - 1; if (ET_FRAMELEN(ifr->ifr_mtu) > max_framelen) { error = EOPNOTSUPP; ET_UNLOCK(sc); break; } if (ifp->if_mtu != ifr->ifr_mtu) { ifp->if_mtu = ifr->ifr_mtu; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; et_init_locked(sc); } } ET_UNLOCK(sc); break; case SIOCSIFCAP: ET_LOCK(sc); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if ((mask & IFCAP_TXCSUM) != 0 && (IFCAP_TXCSUM & ifp->if_capabilities) != 0) { ifp->if_capenable ^= IFCAP_TXCSUM; if ((IFCAP_TXCSUM & ifp->if_capenable) != 0) ifp->if_hwassist |= ET_CSUM_FEATURES; else ifp->if_hwassist &= ~ET_CSUM_FEATURES; } ET_UNLOCK(sc); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } static void et_start_locked(struct ifnet *ifp) { struct et_softc *sc; struct mbuf *m_head = NULL; struct et_txdesc_ring *tx_ring; struct et_txbuf_data *tbd; uint32_t tx_ready_pos; int enq; sc = ifp->if_softc; ET_LOCK_ASSERT(sc); if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING || (sc->sc_flags & (ET_FLAG_LINK | ET_FLAG_TXRX_ENABLED)) != (ET_FLAG_LINK | ET_FLAG_TXRX_ENABLED)) return; /* * Driver does not request TX completion interrupt for every * queued frames to prevent generating excessive interrupts. * This means driver may wait for TX completion interrupt even * though some frames were sucessfully transmitted. Reclaiming * transmitted frames will ensure driver see all available * descriptors. */ tbd = &sc->sc_tx_data; if (tbd->tbd_used > (ET_TX_NDESC * 2) / 3) et_txeof(sc); for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) { if (tbd->tbd_used + ET_NSEG_SPARE >= ET_TX_NDESC) { ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; if (et_encap(sc, &m_head)) { if (m_head == NULL) { ifp->if_oerrors++; break; } IFQ_DRV_PREPEND(&ifp->if_snd, m_head); if (tbd->tbd_used > 0) ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } enq++; ETHER_BPF_MTAP(ifp, m_head); } if (enq > 0) { tx_ring = &sc->sc_tx_ring; bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap, BUS_DMASYNC_PREWRITE); tx_ready_pos = tx_ring->tr_ready_index & ET_TX_READY_POS_INDEX_MASK; if (tx_ring->tr_ready_wrap) tx_ready_pos |= ET_TX_READY_POS_WRAP; CSR_WRITE_4(sc, ET_TX_READY_POS, tx_ready_pos); sc->watchdog_timer = 5; } } static void et_start(struct ifnet *ifp) { struct et_softc *sc; sc = ifp->if_softc; ET_LOCK(sc); et_start_locked(ifp); ET_UNLOCK(sc); } static int et_watchdog(struct et_softc *sc) { uint32_t status; ET_LOCK_ASSERT(sc); if (sc->watchdog_timer == 0 || --sc->watchdog_timer) return (0); bus_dmamap_sync(sc->sc_tx_status.txsd_dtag, sc->sc_tx_status.txsd_dmap, BUS_DMASYNC_POSTREAD); status = le32toh(*(sc->sc_tx_status.txsd_status)); if_printf(sc->ifp, "watchdog timed out (0x%08x) -- resetting\n", status); sc->ifp->if_oerrors++; sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING; et_init_locked(sc); return (EJUSTRETURN); } static int et_stop_rxdma(struct et_softc *sc) { CSR_WRITE_4(sc, ET_RXDMA_CTRL, ET_RXDMA_CTRL_HALT | ET_RXDMA_CTRL_RING1_ENABLE); DELAY(5); if ((CSR_READ_4(sc, ET_RXDMA_CTRL) & ET_RXDMA_CTRL_HALTED) == 0) { if_printf(sc->ifp, "can't stop RX DMA engine\n"); return (ETIMEDOUT); } return (0); } static int et_stop_txdma(struct et_softc *sc) { CSR_WRITE_4(sc, ET_TXDMA_CTRL, ET_TXDMA_CTRL_HALT | ET_TXDMA_CTRL_SINGLE_EPKT); return (0); } static void et_free_tx_ring(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_txbuf_data *tbd; struct et_txbuf *tb; int i; tbd = &sc->sc_tx_data; tx_ring = &sc->sc_tx_ring; for (i = 0; i < ET_TX_NDESC; ++i) { tb = &tbd->tbd_buf[i]; if (tb->tb_mbuf != NULL) { bus_dmamap_sync(sc->sc_tx_tag, tb->tb_dmap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->sc_mbuf_dtag, tb->tb_dmap); m_freem(tb->tb_mbuf); tb->tb_mbuf = NULL; } } } static void et_free_rx_ring(struct et_softc *sc) { struct et_rxbuf_data *rbd; struct et_rxdesc_ring *rx_ring; struct et_rxbuf *rb; int i; /* Ring 0 */ rx_ring = &sc->sc_rx_ring[0]; rbd = &sc->sc_rx_data[0]; for (i = 0; i < ET_RX_NDESC; ++i) { rb = &rbd->rbd_buf[i]; if (rb->rb_mbuf != NULL) { bus_dmamap_sync(sc->sc_rx_mini_tag, rx_ring->rr_dmap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->sc_rx_mini_tag, rb->rb_dmap); m_freem(rb->rb_mbuf); rb->rb_mbuf = NULL; } } /* Ring 1 */ rx_ring = &sc->sc_rx_ring[1]; rbd = &sc->sc_rx_data[1]; for (i = 0; i < ET_RX_NDESC; ++i) { rb = &rbd->rbd_buf[i]; if (rb->rb_mbuf != NULL) { bus_dmamap_sync(sc->sc_rx_tag, rx_ring->rr_dmap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->sc_rx_tag, rb->rb_dmap); m_freem(rb->rb_mbuf); rb->rb_mbuf = NULL; } } } static void et_setmulti(struct et_softc *sc) { struct ifnet *ifp; uint32_t hash[4] = { 0, 0, 0, 0 }; uint32_t rxmac_ctrl, pktfilt; struct ifmultiaddr *ifma; int i, count; ET_LOCK_ASSERT(sc); ifp = sc->ifp; pktfilt = CSR_READ_4(sc, ET_PKTFILT); rxmac_ctrl = CSR_READ_4(sc, ET_RXMAC_CTRL); pktfilt &= ~(ET_PKTFILT_BCAST | ET_PKTFILT_MCAST | ET_PKTFILT_UCAST); if (ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) { rxmac_ctrl |= ET_RXMAC_CTRL_NO_PKTFILT; goto back; } count = 0; IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { uint32_t *hp, h; if (ifma->ifma_addr->sa_family != AF_LINK) continue; h = ether_crc32_be(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), ETHER_ADDR_LEN); h = (h & 0x3f800000) >> 23; hp = &hash[0]; if (h >= 32 && h < 64) { h -= 32; hp = &hash[1]; } else if (h >= 64 && h < 96) { h -= 64; hp = &hash[2]; } else if (h >= 96) { h -= 96; hp = &hash[3]; } *hp |= (1 << h); ++count; } IF_ADDR_UNLOCK(ifp); for (i = 0; i < 4; ++i) CSR_WRITE_4(sc, ET_MULTI_HASH + (i * 4), hash[i]); if (count > 0) pktfilt |= ET_PKTFILT_MCAST; rxmac_ctrl &= ~ET_RXMAC_CTRL_NO_PKTFILT; back: CSR_WRITE_4(sc, ET_PKTFILT, pktfilt); CSR_WRITE_4(sc, ET_RXMAC_CTRL, rxmac_ctrl); } static int et_chip_init(struct et_softc *sc) { struct ifnet *ifp; uint32_t rxq_end; int error, frame_len, rxmem_size; ifp = sc->ifp; /* * Split 16Kbytes internal memory between TX and RX * according to frame length. */ frame_len = ET_FRAMELEN(ifp->if_mtu); if (frame_len < 2048) { rxmem_size = ET_MEM_RXSIZE_DEFAULT; } else if (frame_len <= ET_RXMAC_CUT_THRU_FRMLEN) { rxmem_size = ET_MEM_SIZE / 2; } else { rxmem_size = ET_MEM_SIZE - roundup(frame_len + ET_MEM_TXSIZE_EX, ET_MEM_UNIT); } rxq_end = ET_QUEUE_ADDR(rxmem_size); CSR_WRITE_4(sc, ET_RXQUEUE_START, ET_QUEUE_ADDR_START); CSR_WRITE_4(sc, ET_RXQUEUE_END, rxq_end); CSR_WRITE_4(sc, ET_TXQUEUE_START, rxq_end + 1); CSR_WRITE_4(sc, ET_TXQUEUE_END, ET_QUEUE_ADDR_END); /* No loopback */ CSR_WRITE_4(sc, ET_LOOPBACK, 0); /* Clear MSI configure */ if ((sc->sc_flags & ET_FLAG_MSI) == 0) CSR_WRITE_4(sc, ET_MSI_CFG, 0); /* Disable timer */ CSR_WRITE_4(sc, ET_TIMER, 0); /* Initialize MAC */ et_init_mac(sc); /* Enable memory controllers */ CSR_WRITE_4(sc, ET_MMC_CTRL, ET_MMC_CTRL_ENABLE); /* Initialize RX MAC */ et_init_rxmac(sc); /* Initialize TX MAC */ et_init_txmac(sc); /* Initialize RX DMA engine */ error = et_init_rxdma(sc); if (error) return (error); /* Initialize TX DMA engine */ error = et_init_txdma(sc); if (error) return (error); return (0); } static void et_init_tx_ring(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_txbuf_data *tbd; struct et_txstatus_data *txsd; tx_ring = &sc->sc_tx_ring; bzero(tx_ring->tr_desc, ET_TX_RING_SIZE); bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap, BUS_DMASYNC_PREWRITE); tbd = &sc->sc_tx_data; tbd->tbd_start_index = 0; tbd->tbd_start_wrap = 0; tbd->tbd_used = 0; txsd = &sc->sc_tx_status; bzero(txsd->txsd_status, sizeof(uint32_t)); bus_dmamap_sync(txsd->txsd_dtag, txsd->txsd_dmap, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static int et_init_rx_ring(struct et_softc *sc) { struct et_rxstatus_data *rxsd; struct et_rxstat_ring *rxst_ring; struct et_rxbuf_data *rbd; int i, error, n; for (n = 0; n < ET_RX_NRING; ++n) { rbd = &sc->sc_rx_data[n]; for (i = 0; i < ET_RX_NDESC; ++i) { error = rbd->rbd_newbuf(rbd, i); if (error) { if_printf(sc->ifp, "%d ring %d buf, " "newbuf failed: %d\n", n, i, error); return (error); } } } rxsd = &sc->sc_rx_status; bzero(rxsd->rxsd_status, sizeof(struct et_rxstatus)); bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); rxst_ring = &sc->sc_rxstat_ring; bzero(rxst_ring->rsr_stat, ET_RXSTAT_RING_SIZE); bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } static int et_init_rxdma(struct et_softc *sc) { struct et_rxstatus_data *rxsd; struct et_rxstat_ring *rxst_ring; struct et_rxdesc_ring *rx_ring; int error; error = et_stop_rxdma(sc); if (error) { if_printf(sc->ifp, "can't init RX DMA engine\n"); return (error); } /* * Install RX status */ rxsd = &sc->sc_rx_status; CSR_WRITE_4(sc, ET_RX_STATUS_HI, ET_ADDR_HI(rxsd->rxsd_paddr)); CSR_WRITE_4(sc, ET_RX_STATUS_LO, ET_ADDR_LO(rxsd->rxsd_paddr)); /* * Install RX stat ring */ rxst_ring = &sc->sc_rxstat_ring; CSR_WRITE_4(sc, ET_RXSTAT_HI, ET_ADDR_HI(rxst_ring->rsr_paddr)); CSR_WRITE_4(sc, ET_RXSTAT_LO, ET_ADDR_LO(rxst_ring->rsr_paddr)); CSR_WRITE_4(sc, ET_RXSTAT_CNT, ET_RX_NSTAT - 1); CSR_WRITE_4(sc, ET_RXSTAT_POS, 0); CSR_WRITE_4(sc, ET_RXSTAT_MINCNT, ((ET_RX_NSTAT * 15) / 100) - 1); /* Match ET_RXSTAT_POS */ rxst_ring->rsr_index = 0; rxst_ring->rsr_wrap = 0; /* * Install the 2nd RX descriptor ring */ rx_ring = &sc->sc_rx_ring[1]; CSR_WRITE_4(sc, ET_RX_RING1_HI, ET_ADDR_HI(rx_ring->rr_paddr)); CSR_WRITE_4(sc, ET_RX_RING1_LO, ET_ADDR_LO(rx_ring->rr_paddr)); CSR_WRITE_4(sc, ET_RX_RING1_CNT, ET_RX_NDESC - 1); CSR_WRITE_4(sc, ET_RX_RING1_POS, ET_RX_RING1_POS_WRAP); CSR_WRITE_4(sc, ET_RX_RING1_MINCNT, ((ET_RX_NDESC * 15) / 100) - 1); /* Match ET_RX_RING1_POS */ rx_ring->rr_index = 0; rx_ring->rr_wrap = 1; /* * Install the 1st RX descriptor ring */ rx_ring = &sc->sc_rx_ring[0]; CSR_WRITE_4(sc, ET_RX_RING0_HI, ET_ADDR_HI(rx_ring->rr_paddr)); CSR_WRITE_4(sc, ET_RX_RING0_LO, ET_ADDR_LO(rx_ring->rr_paddr)); CSR_WRITE_4(sc, ET_RX_RING0_CNT, ET_RX_NDESC - 1); CSR_WRITE_4(sc, ET_RX_RING0_POS, ET_RX_RING0_POS_WRAP); CSR_WRITE_4(sc, ET_RX_RING0_MINCNT, ((ET_RX_NDESC * 15) / 100) - 1); /* Match ET_RX_RING0_POS */ rx_ring->rr_index = 0; rx_ring->rr_wrap = 1; /* * RX intr moderation */ CSR_WRITE_4(sc, ET_RX_INTR_NPKTS, sc->sc_rx_intr_npkts); CSR_WRITE_4(sc, ET_RX_INTR_DELAY, sc->sc_rx_intr_delay); return (0); } static int et_init_txdma(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_txstatus_data *txsd; int error; error = et_stop_txdma(sc); if (error) { if_printf(sc->ifp, "can't init TX DMA engine\n"); return (error); } /* * Install TX descriptor ring */ tx_ring = &sc->sc_tx_ring; CSR_WRITE_4(sc, ET_TX_RING_HI, ET_ADDR_HI(tx_ring->tr_paddr)); CSR_WRITE_4(sc, ET_TX_RING_LO, ET_ADDR_LO(tx_ring->tr_paddr)); CSR_WRITE_4(sc, ET_TX_RING_CNT, ET_TX_NDESC - 1); /* * Install TX status */ txsd = &sc->sc_tx_status; CSR_WRITE_4(sc, ET_TX_STATUS_HI, ET_ADDR_HI(txsd->txsd_paddr)); CSR_WRITE_4(sc, ET_TX_STATUS_LO, ET_ADDR_LO(txsd->txsd_paddr)); CSR_WRITE_4(sc, ET_TX_READY_POS, 0); /* Match ET_TX_READY_POS */ tx_ring->tr_ready_index = 0; tx_ring->tr_ready_wrap = 0; return (0); } static void et_init_mac(struct et_softc *sc) { struct ifnet *ifp; const uint8_t *eaddr; uint32_t val; /* Reset MAC */ CSR_WRITE_4(sc, ET_MAC_CFG1, ET_MAC_CFG1_RST_TXFUNC | ET_MAC_CFG1_RST_RXFUNC | ET_MAC_CFG1_RST_TXMC | ET_MAC_CFG1_RST_RXMC | ET_MAC_CFG1_SIM_RST | ET_MAC_CFG1_SOFT_RST); /* * Setup inter packet gap */ val = (56 << ET_IPG_NONB2B_1_SHIFT) | (88 << ET_IPG_NONB2B_2_SHIFT) | (80 << ET_IPG_MINIFG_SHIFT) | (96 << ET_IPG_B2B_SHIFT); CSR_WRITE_4(sc, ET_IPG, val); /* * Setup half duplex mode */ val = (10 << ET_MAC_HDX_ALT_BEB_TRUNC_SHIFT) | (15 << ET_MAC_HDX_REXMIT_MAX_SHIFT) | (55 << ET_MAC_HDX_COLLWIN_SHIFT) | ET_MAC_HDX_EXC_DEFER; CSR_WRITE_4(sc, ET_MAC_HDX, val); /* Clear MAC control */ CSR_WRITE_4(sc, ET_MAC_CTRL, 0); /* Reset MII */ CSR_WRITE_4(sc, ET_MII_CFG, ET_MII_CFG_CLKRST); /* * Set MAC address */ ifp = sc->ifp; eaddr = IF_LLADDR(ifp); val = eaddr[2] | (eaddr[3] << 8) | (eaddr[4] << 16) | (eaddr[5] << 24); CSR_WRITE_4(sc, ET_MAC_ADDR1, val); val = (eaddr[0] << 16) | (eaddr[1] << 24); CSR_WRITE_4(sc, ET_MAC_ADDR2, val); /* Set max frame length */ CSR_WRITE_4(sc, ET_MAX_FRMLEN, ET_FRAMELEN(ifp->if_mtu)); /* Bring MAC out of reset state */ CSR_WRITE_4(sc, ET_MAC_CFG1, 0); } static void et_init_rxmac(struct et_softc *sc) { struct ifnet *ifp; const uint8_t *eaddr; uint32_t val; int i; /* Disable RX MAC and WOL */ CSR_WRITE_4(sc, ET_RXMAC_CTRL, ET_RXMAC_CTRL_WOL_DISABLE); /* * Clear all WOL related registers */ for (i = 0; i < 3; ++i) CSR_WRITE_4(sc, ET_WOL_CRC + (i * 4), 0); for (i = 0; i < 20; ++i) CSR_WRITE_4(sc, ET_WOL_MASK + (i * 4), 0); /* * Set WOL source address. XXX is this necessary? */ ifp = sc->ifp; eaddr = IF_LLADDR(ifp); val = (eaddr[2] << 24) | (eaddr[3] << 16) | (eaddr[4] << 8) | eaddr[5]; CSR_WRITE_4(sc, ET_WOL_SA_LO, val); val = (eaddr[0] << 8) | eaddr[1]; CSR_WRITE_4(sc, ET_WOL_SA_HI, val); /* Clear packet filters */ CSR_WRITE_4(sc, ET_PKTFILT, 0); /* No ucast filtering */ CSR_WRITE_4(sc, ET_UCAST_FILTADDR1, 0); CSR_WRITE_4(sc, ET_UCAST_FILTADDR2, 0); CSR_WRITE_4(sc, ET_UCAST_FILTADDR3, 0); if (ET_FRAMELEN(ifp->if_mtu) > ET_RXMAC_CUT_THRU_FRMLEN) { /* * In order to transmit jumbo packets greater than * ET_RXMAC_CUT_THRU_FRMLEN bytes, the FIFO between * RX MAC and RX DMA needs to be reduced in size to * (ET_MEM_SIZE - ET_MEM_TXSIZE_EX - framelen). In * order to implement this, we must use "cut through" * mode in the RX MAC, which chops packets down into * segments. In this case we selected 256 bytes, * since this is the size of the PCI-Express TLP's * that the ET1310 uses. */ val = (ET_RXMAC_SEGSZ(256) & ET_RXMAC_MC_SEGSZ_MAX_MASK) | ET_RXMAC_MC_SEGSZ_ENABLE; } else { val = 0; } CSR_WRITE_4(sc, ET_RXMAC_MC_SEGSZ, val); CSR_WRITE_4(sc, ET_RXMAC_MC_WATERMARK, 0); /* Initialize RX MAC management register */ CSR_WRITE_4(sc, ET_RXMAC_MGT, 0); CSR_WRITE_4(sc, ET_RXMAC_SPACE_AVL, 0); CSR_WRITE_4(sc, ET_RXMAC_MGT, ET_RXMAC_MGT_PASS_ECRC | ET_RXMAC_MGT_PASS_ELEN | ET_RXMAC_MGT_PASS_ETRUNC | ET_RXMAC_MGT_CHECK_PKT); /* * Configure runt filtering (may not work on certain chip generation) */ val = (ETHER_MIN_LEN << ET_PKTFILT_MINLEN_SHIFT) & ET_PKTFILT_MINLEN_MASK; val |= ET_PKTFILT_FRAG; CSR_WRITE_4(sc, ET_PKTFILT, val); /* Enable RX MAC but leave WOL disabled */ CSR_WRITE_4(sc, ET_RXMAC_CTRL, ET_RXMAC_CTRL_WOL_DISABLE | ET_RXMAC_CTRL_ENABLE); /* * Setup multicast hash and allmulti/promisc mode */ et_setmulti(sc); } static void et_init_txmac(struct et_softc *sc) { /* Disable TX MAC and FC(?) */ CSR_WRITE_4(sc, ET_TXMAC_CTRL, ET_TXMAC_CTRL_FC_DISABLE); /* * Initialize pause time. * This register should be set before XON/XOFF frame is * sent by driver. */ CSR_WRITE_4(sc, ET_TXMAC_FLOWCTRL, 0 << ET_TXMAC_FLOWCTRL_CFPT_SHIFT); /* Enable TX MAC but leave FC(?) diabled */ CSR_WRITE_4(sc, ET_TXMAC_CTRL, ET_TXMAC_CTRL_ENABLE | ET_TXMAC_CTRL_FC_DISABLE); } static int et_start_rxdma(struct et_softc *sc) { uint32_t val; val = (sc->sc_rx_data[0].rbd_bufsize & ET_RXDMA_CTRL_RING0_SIZE_MASK) | ET_RXDMA_CTRL_RING0_ENABLE; val |= (sc->sc_rx_data[1].rbd_bufsize & ET_RXDMA_CTRL_RING1_SIZE_MASK) | ET_RXDMA_CTRL_RING1_ENABLE; CSR_WRITE_4(sc, ET_RXDMA_CTRL, val); DELAY(5); if (CSR_READ_4(sc, ET_RXDMA_CTRL) & ET_RXDMA_CTRL_HALTED) { if_printf(sc->ifp, "can't start RX DMA engine\n"); return (ETIMEDOUT); } return (0); } static int et_start_txdma(struct et_softc *sc) { CSR_WRITE_4(sc, ET_TXDMA_CTRL, ET_TXDMA_CTRL_SINGLE_EPKT); return (0); } static void et_rxeof(struct et_softc *sc) { struct et_rxstatus_data *rxsd; struct et_rxstat_ring *rxst_ring; struct et_rxbuf_data *rbd; struct et_rxdesc_ring *rx_ring; struct et_rxstat *st; struct ifnet *ifp; struct mbuf *m; uint32_t rxstat_pos, rxring_pos; uint32_t rxst_info1, rxst_info2, rxs_stat_ring; int buflen, buf_idx, npost[2], ring_idx; int rxst_index, rxst_wrap; ET_LOCK_ASSERT(sc); ifp = sc->ifp; rxsd = &sc->sc_rx_status; rxst_ring = &sc->sc_rxstat_ring; if ((sc->sc_flags & ET_FLAG_TXRX_ENABLED) == 0) return; bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap, BUS_DMASYNC_POSTREAD); bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap, BUS_DMASYNC_POSTREAD); npost[0] = npost[1] = 0; rxs_stat_ring = le32toh(rxsd->rxsd_status->rxs_stat_ring); rxst_wrap = (rxs_stat_ring & ET_RXS_STATRING_WRAP) ? 1 : 0; rxst_index = (rxs_stat_ring & ET_RXS_STATRING_INDEX_MASK) >> ET_RXS_STATRING_INDEX_SHIFT; while (rxst_index != rxst_ring->rsr_index || rxst_wrap != rxst_ring->rsr_wrap) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) break; MPASS(rxst_ring->rsr_index < ET_RX_NSTAT); st = &rxst_ring->rsr_stat[rxst_ring->rsr_index]; rxst_info1 = le32toh(st->rxst_info1); rxst_info2 = le32toh(st->rxst_info2); buflen = (rxst_info2 & ET_RXST_INFO2_LEN_MASK) >> ET_RXST_INFO2_LEN_SHIFT; buf_idx = (rxst_info2 & ET_RXST_INFO2_BUFIDX_MASK) >> ET_RXST_INFO2_BUFIDX_SHIFT; ring_idx = (rxst_info2 & ET_RXST_INFO2_RINGIDX_MASK) >> ET_RXST_INFO2_RINGIDX_SHIFT; if (++rxst_ring->rsr_index == ET_RX_NSTAT) { rxst_ring->rsr_index = 0; rxst_ring->rsr_wrap ^= 1; } rxstat_pos = rxst_ring->rsr_index & ET_RXSTAT_POS_INDEX_MASK; if (rxst_ring->rsr_wrap) rxstat_pos |= ET_RXSTAT_POS_WRAP; CSR_WRITE_4(sc, ET_RXSTAT_POS, rxstat_pos); if (ring_idx >= ET_RX_NRING) { ifp->if_ierrors++; if_printf(ifp, "invalid ring index %d\n", ring_idx); continue; } if (buf_idx >= ET_RX_NDESC) { ifp->if_ierrors++; if_printf(ifp, "invalid buf index %d\n", buf_idx); continue; } rbd = &sc->sc_rx_data[ring_idx]; m = rbd->rbd_buf[buf_idx].rb_mbuf; if ((rxst_info1 & ET_RXST_INFO1_OK) == 0){ /* Discard errored frame. */ rbd->rbd_discard(rbd, buf_idx); } else if (rbd->rbd_newbuf(rbd, buf_idx) != 0) { /* No available mbufs, discard it. */ ifp->if_iqdrops++; rbd->rbd_discard(rbd, buf_idx); } else { buflen -= ETHER_CRC_LEN; if (buflen < ETHER_HDR_LEN) { m_freem(m); ifp->if_ierrors++; } else { m->m_pkthdr.len = m->m_len = buflen; m->m_pkthdr.rcvif = ifp; ET_UNLOCK(sc); ifp->if_input(ifp, m); ET_LOCK(sc); } } rx_ring = &sc->sc_rx_ring[ring_idx]; if (buf_idx != rx_ring->rr_index) { if_printf(ifp, "WARNING!! ring %d, buf_idx %d, rr_idx %d\n", ring_idx, buf_idx, rx_ring->rr_index); } MPASS(rx_ring->rr_index < ET_RX_NDESC); if (++rx_ring->rr_index == ET_RX_NDESC) { rx_ring->rr_index = 0; rx_ring->rr_wrap ^= 1; } rxring_pos = rx_ring->rr_index & ET_RX_RING_POS_INDEX_MASK; if (rx_ring->rr_wrap) rxring_pos |= ET_RX_RING_POS_WRAP; CSR_WRITE_4(sc, rx_ring->rr_posreg, rxring_pos); } bus_dmamap_sync(rxsd->rxsd_dtag, rxsd->rxsd_dmap, BUS_DMASYNC_PREREAD); bus_dmamap_sync(rxst_ring->rsr_dtag, rxst_ring->rsr_dmap, BUS_DMASYNC_PREREAD); } static int et_encap(struct et_softc *sc, struct mbuf **m0) { struct et_txdesc_ring *tx_ring; struct et_txbuf_data *tbd; struct et_txdesc *td; struct mbuf *m; bus_dma_segment_t segs[ET_NSEG_MAX]; bus_dmamap_t map; uint32_t csum_flags, last_td_ctrl2; int error, i, idx, first_idx, last_idx, nsegs; tx_ring = &sc->sc_tx_ring; MPASS(tx_ring->tr_ready_index < ET_TX_NDESC); tbd = &sc->sc_tx_data; first_idx = tx_ring->tr_ready_index; map = tbd->tbd_buf[first_idx].tb_dmap; error = bus_dmamap_load_mbuf_sg(sc->sc_tx_tag, map, *m0, segs, &nsegs, 0); if (error == EFBIG) { m = m_collapse(*m0, M_DONTWAIT, ET_NSEG_MAX); if (m == NULL) { m_freem(*m0); *m0 = NULL; return (ENOMEM); } *m0 = m; error = bus_dmamap_load_mbuf_sg(sc->sc_tx_tag, map, *m0, segs, &nsegs, 0); if (error != 0) { m_freem(*m0); *m0 = NULL; return (error); } } else if (error != 0) return (error); /* Check for descriptor overruns. */ if (tbd->tbd_used + nsegs > ET_TX_NDESC - 1) { bus_dmamap_unload(sc->sc_tx_tag, map); return (ENOBUFS); } bus_dmamap_sync(sc->sc_tx_tag, map, BUS_DMASYNC_PREWRITE); last_td_ctrl2 = ET_TDCTRL2_LAST_FRAG; sc->sc_tx += nsegs; if (sc->sc_tx / sc->sc_tx_intr_nsegs != sc->sc_tx_intr) { sc->sc_tx_intr = sc->sc_tx / sc->sc_tx_intr_nsegs; last_td_ctrl2 |= ET_TDCTRL2_INTR; } m = *m0; csum_flags = 0; if ((m->m_pkthdr.csum_flags & ET_CSUM_FEATURES) != 0) { if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0) csum_flags |= ET_TDCTRL2_CSUM_IP; if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0) csum_flags |= ET_TDCTRL2_CSUM_UDP; else if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0) csum_flags |= ET_TDCTRL2_CSUM_TCP; } last_idx = -1; for (i = 0; i < nsegs; ++i) { idx = (first_idx + i) % ET_TX_NDESC; td = &tx_ring->tr_desc[idx]; td->td_addr_hi = htole32(ET_ADDR_HI(segs[i].ds_addr)); td->td_addr_lo = htole32(ET_ADDR_LO(segs[i].ds_addr)); td->td_ctrl1 = htole32(segs[i].ds_len & ET_TDCTRL1_LEN_MASK); if (i == nsegs - 1) { /* Last frag */ td->td_ctrl2 = htole32(last_td_ctrl2 | csum_flags); last_idx = idx; } else td->td_ctrl2 = htole32(csum_flags); MPASS(tx_ring->tr_ready_index < ET_TX_NDESC); if (++tx_ring->tr_ready_index == ET_TX_NDESC) { tx_ring->tr_ready_index = 0; tx_ring->tr_ready_wrap ^= 1; } } td = &tx_ring->tr_desc[first_idx]; /* First frag */ td->td_ctrl2 |= htole32(ET_TDCTRL2_FIRST_FRAG); MPASS(last_idx >= 0); tbd->tbd_buf[first_idx].tb_dmap = tbd->tbd_buf[last_idx].tb_dmap; tbd->tbd_buf[last_idx].tb_dmap = map; tbd->tbd_buf[last_idx].tb_mbuf = m; tbd->tbd_used += nsegs; MPASS(tbd->tbd_used <= ET_TX_NDESC); return (0); } static void et_txeof(struct et_softc *sc) { struct et_txdesc_ring *tx_ring; struct et_txbuf_data *tbd; struct et_txbuf *tb; struct ifnet *ifp; uint32_t tx_done; int end, wrap; ET_LOCK_ASSERT(sc); ifp = sc->ifp; tx_ring = &sc->sc_tx_ring; tbd = &sc->sc_tx_data; if ((sc->sc_flags & ET_FLAG_TXRX_ENABLED) == 0) return; if (tbd->tbd_used == 0) return; bus_dmamap_sync(tx_ring->tr_dtag, tx_ring->tr_dmap, BUS_DMASYNC_POSTWRITE); tx_done = CSR_READ_4(sc, ET_TX_DONE_POS); end = tx_done & ET_TX_DONE_POS_INDEX_MASK; wrap = (tx_done & ET_TX_DONE_POS_WRAP) ? 1 : 0; while (tbd->tbd_start_index != end || tbd->tbd_start_wrap != wrap) { MPASS(tbd->tbd_start_index < ET_TX_NDESC); tb = &tbd->tbd_buf[tbd->tbd_start_index]; if (tb->tb_mbuf != NULL) { bus_dmamap_sync(sc->sc_tx_tag, tb->tb_dmap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->sc_tx_tag, tb->tb_dmap); m_freem(tb->tb_mbuf); tb->tb_mbuf = NULL; } if (++tbd->tbd_start_index == ET_TX_NDESC) { tbd->tbd_start_index = 0; tbd->tbd_start_wrap ^= 1; } MPASS(tbd->tbd_used > 0); tbd->tbd_used--; } if (tbd->tbd_used == 0) sc->watchdog_timer = 0; if (tbd->tbd_used + ET_NSEG_SPARE < ET_TX_NDESC) ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } static void et_tick(void *xsc) { struct et_softc *sc; struct ifnet *ifp; struct mii_data *mii; sc = xsc; ET_LOCK_ASSERT(sc); ifp = sc->ifp; mii = device_get_softc(sc->sc_miibus); mii_tick(mii); et_stats_update(sc); if (et_watchdog(sc) == EJUSTRETURN) return; callout_reset(&sc->sc_tick, hz, et_tick, sc); } static int et_newbuf_cluster(struct et_rxbuf_data *rbd, int buf_idx) { struct et_softc *sc; struct et_rxdesc *desc; struct et_rxbuf *rb; struct mbuf *m; bus_dma_segment_t segs[1]; bus_dmamap_t dmap; int nsegs; MPASS(buf_idx < ET_RX_NDESC); m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MCLBYTES; m_adj(m, ETHER_ALIGN); sc = rbd->rbd_softc; rb = &rbd->rbd_buf[buf_idx]; if (bus_dmamap_load_mbuf_sg(sc->sc_rx_tag, sc->sc_rx_sparemap, m, segs, &nsegs, 0) != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); if (rb->rb_mbuf != NULL) { bus_dmamap_sync(sc->sc_rx_tag, rb->rb_dmap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->sc_rx_tag, rb->rb_dmap); } dmap = rb->rb_dmap; rb->rb_dmap = sc->sc_rx_sparemap; sc->sc_rx_sparemap = dmap; bus_dmamap_sync(sc->sc_rx_tag, rb->rb_dmap, BUS_DMASYNC_PREREAD); rb->rb_mbuf = m; desc = &rbd->rbd_ring->rr_desc[buf_idx]; desc->rd_addr_hi = htole32(ET_ADDR_HI(segs[0].ds_addr)); desc->rd_addr_lo = htole32(ET_ADDR_LO(segs[0].ds_addr)); desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK); bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap, BUS_DMASYNC_PREWRITE); return (0); } static void et_rxbuf_discard(struct et_rxbuf_data *rbd, int buf_idx) { struct et_rxdesc *desc; desc = &rbd->rbd_ring->rr_desc[buf_idx]; desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK); bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap, BUS_DMASYNC_PREWRITE); } static int et_newbuf_hdr(struct et_rxbuf_data *rbd, int buf_idx) { struct et_softc *sc; struct et_rxdesc *desc; struct et_rxbuf *rb; struct mbuf *m; bus_dma_segment_t segs[1]; bus_dmamap_t dmap; int nsegs; MPASS(buf_idx < ET_RX_NDESC); MGETHDR(m, M_DONTWAIT, MT_DATA); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MHLEN; m_adj(m, ETHER_ALIGN); sc = rbd->rbd_softc; rb = &rbd->rbd_buf[buf_idx]; if (bus_dmamap_load_mbuf_sg(sc->sc_rx_mini_tag, sc->sc_rx_mini_sparemap, m, segs, &nsegs, 0) != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); if (rb->rb_mbuf != NULL) { bus_dmamap_sync(sc->sc_rx_mini_tag, rb->rb_dmap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->sc_rx_mini_tag, rb->rb_dmap); } dmap = rb->rb_dmap; rb->rb_dmap = sc->sc_rx_mini_sparemap; sc->sc_rx_mini_sparemap = dmap; bus_dmamap_sync(sc->sc_rx_mini_tag, rb->rb_dmap, BUS_DMASYNC_PREREAD); rb->rb_mbuf = m; desc = &rbd->rbd_ring->rr_desc[buf_idx]; desc->rd_addr_hi = htole32(ET_ADDR_HI(segs[0].ds_addr)); desc->rd_addr_lo = htole32(ET_ADDR_LO(segs[0].ds_addr)); desc->rd_ctrl = htole32(buf_idx & ET_RDCTRL_BUFIDX_MASK); bus_dmamap_sync(rbd->rbd_ring->rr_dtag, rbd->rbd_ring->rr_dmap, BUS_DMASYNC_PREWRITE); return (0); } #define ET_SYSCTL_STAT_ADD32(c, h, n, p, d) \ SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d) #define ET_SYSCTL_STAT_ADD64(c, h, n, p, d) \ SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d) /* * Create sysctl tree */ static void et_add_sysctls(struct et_softc * sc) { struct sysctl_ctx_list *ctx; struct sysctl_oid_list *children, *parent; struct sysctl_oid *tree; struct et_hw_stats *stats; ctx = device_get_sysctl_ctx(sc->dev); children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->dev)); SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rx_intr_npkts", CTLTYPE_INT | CTLFLAG_RW, sc, 0, et_sysctl_rx_intr_npkts, "I", "RX IM, # packets per RX interrupt"); SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rx_intr_delay", CTLTYPE_INT | CTLFLAG_RW, sc, 0, et_sysctl_rx_intr_delay, "I", "RX IM, RX interrupt delay (x10 usec)"); SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_intr_nsegs", CTLFLAG_RW, &sc->sc_tx_intr_nsegs, 0, "TX IM, # segments per TX interrupt"); SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "timer", CTLFLAG_RW, &sc->sc_timer, 0, "TX timer"); tree = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD, NULL, "ET statistics"); parent = SYSCTL_CHILDREN(tree); /* TX/RX statistics. */ stats = &sc->sc_stats; ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_64", &stats->pkts_64, "0 to 64 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_65_127", &stats->pkts_65, "65 to 127 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_128_255", &stats->pkts_128, "128 to 255 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_256_511", &stats->pkts_256, "256 to 511 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_512_1023", &stats->pkts_512, "512 to 1023 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_1024_1518", &stats->pkts_1024, "1024 to 1518 bytes frames"); ET_SYSCTL_STAT_ADD64(ctx, parent, "frames_1519_1522", &stats->pkts_1519, "1519 to 1522 bytes frames"); /* RX statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD, NULL, "RX MAC statistics"); children = SYSCTL_CHILDREN(tree); ET_SYSCTL_STAT_ADD64(ctx, children, "bytes", &stats->rx_bytes, "Good bytes"); ET_SYSCTL_STAT_ADD64(ctx, children, "frames", &stats->rx_frames, "Good frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "crc_errs", &stats->rx_crcerrs, "CRC errors"); ET_SYSCTL_STAT_ADD64(ctx, children, "mcast_frames", &stats->rx_mcast, "Multicast frames"); ET_SYSCTL_STAT_ADD64(ctx, children, "bcast_frames", &stats->rx_bcast, "Broadcast frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "control", &stats->rx_control, "Control frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "pause", &stats->rx_pause, "Pause frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "unknown_control", &stats->rx_unknown_control, "Unknown control frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "align_errs", &stats->rx_alignerrs, "Alignment errors"); ET_SYSCTL_STAT_ADD32(ctx, children, "len_errs", &stats->rx_lenerrs, "Frames with length mismatched"); ET_SYSCTL_STAT_ADD32(ctx, children, "code_errs", &stats->rx_codeerrs, "Frames with code error"); ET_SYSCTL_STAT_ADD32(ctx, children, "cs_errs", &stats->rx_cserrs, "Frames with carrier sense error"); ET_SYSCTL_STAT_ADD32(ctx, children, "runts", &stats->rx_runts, "Too short frames"); ET_SYSCTL_STAT_ADD64(ctx, children, "oversize", &stats->rx_oversize, "Oversized frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "fragments", &stats->rx_fragments, "Fragmented frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "jabbers", &stats->rx_jabbers, "Frames with jabber error"); ET_SYSCTL_STAT_ADD32(ctx, children, "drop", &stats->rx_drop, "Dropped frames"); /* TX statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD, NULL, "TX MAC statistics"); children = SYSCTL_CHILDREN(tree); ET_SYSCTL_STAT_ADD64(ctx, children, "bytes", &stats->tx_bytes, "Good bytes"); ET_SYSCTL_STAT_ADD64(ctx, children, "frames", &stats->tx_frames, "Good frames"); ET_SYSCTL_STAT_ADD64(ctx, children, "mcast_frames", &stats->tx_mcast, "Multicast frames"); ET_SYSCTL_STAT_ADD64(ctx, children, "bcast_frames", &stats->tx_bcast, "Broadcast frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "pause", &stats->tx_pause, "Pause frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "deferred", &stats->tx_deferred, "Deferred frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "excess_deferred", &stats->tx_excess_deferred, "Excessively deferred frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "single_colls", &stats->tx_single_colls, "Single collisions"); ET_SYSCTL_STAT_ADD32(ctx, children, "multi_colls", &stats->tx_multi_colls, "Multiple collisions"); ET_SYSCTL_STAT_ADD32(ctx, children, "late_colls", &stats->tx_late_colls, "Late collisions"); ET_SYSCTL_STAT_ADD32(ctx, children, "excess_colls", &stats->tx_excess_colls, "Excess collisions"); ET_SYSCTL_STAT_ADD32(ctx, children, "total_colls", &stats->tx_total_colls, "Total collisions"); ET_SYSCTL_STAT_ADD32(ctx, children, "pause_honored", &stats->tx_pause_honored, "Honored pause frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "drop", &stats->tx_drop, "Dropped frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "jabbers", &stats->tx_jabbers, "Frames with jabber errors"); ET_SYSCTL_STAT_ADD32(ctx, children, "crc_errs", &stats->tx_crcerrs, "Frames with CRC errors"); ET_SYSCTL_STAT_ADD32(ctx, children, "control", &stats->tx_control, "Control frames"); ET_SYSCTL_STAT_ADD64(ctx, children, "oversize", &stats->tx_oversize, "Oversized frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "undersize", &stats->tx_undersize, "Undersized frames"); ET_SYSCTL_STAT_ADD32(ctx, children, "fragments", &stats->tx_fragments, "Fragmented frames"); } #undef ET_SYSCTL_STAT_ADD32 #undef ET_SYSCTL_STAT_ADD64 static int et_sysctl_rx_intr_npkts(SYSCTL_HANDLER_ARGS) { struct et_softc *sc; struct ifnet *ifp; int error, v; sc = arg1; ifp = sc->ifp; v = sc->sc_rx_intr_npkts; error = sysctl_handle_int(oidp, &v, 0, req); if (error || req->newptr == NULL) goto back; if (v <= 0) { error = EINVAL; goto back; } if (sc->sc_rx_intr_npkts != v) { if (ifp->if_drv_flags & IFF_DRV_RUNNING) CSR_WRITE_4(sc, ET_RX_INTR_NPKTS, v); sc->sc_rx_intr_npkts = v; } back: return (error); } static int et_sysctl_rx_intr_delay(SYSCTL_HANDLER_ARGS) { struct et_softc *sc; struct ifnet *ifp; int error, v; sc = arg1; ifp = sc->ifp; v = sc->sc_rx_intr_delay; error = sysctl_handle_int(oidp, &v, 0, req); if (error || req->newptr == NULL) goto back; if (v <= 0) { error = EINVAL; goto back; } if (sc->sc_rx_intr_delay != v) { if (ifp->if_drv_flags & IFF_DRV_RUNNING) CSR_WRITE_4(sc, ET_RX_INTR_DELAY, v); sc->sc_rx_intr_delay = v; } back: return (error); } static void et_stats_update(struct et_softc *sc) { struct ifnet *ifp; struct et_hw_stats *stats; stats = &sc->sc_stats; stats->pkts_64 += CSR_READ_4(sc, ET_STAT_PKTS_64); stats->pkts_65 += CSR_READ_4(sc, ET_STAT_PKTS_65_127); stats->pkts_128 += CSR_READ_4(sc, ET_STAT_PKTS_128_255); stats->pkts_256 += CSR_READ_4(sc, ET_STAT_PKTS_256_511); stats->pkts_512 += CSR_READ_4(sc, ET_STAT_PKTS_512_1023); stats->pkts_1024 += CSR_READ_4(sc, ET_STAT_PKTS_1024_1518); stats->pkts_1519 += CSR_READ_4(sc, ET_STAT_PKTS_1519_1522); stats->rx_bytes += CSR_READ_4(sc, ET_STAT_RX_BYTES); stats->rx_frames += CSR_READ_4(sc, ET_STAT_RX_FRAMES); stats->rx_crcerrs += CSR_READ_4(sc, ET_STAT_RX_CRC_ERR); stats->rx_mcast += CSR_READ_4(sc, ET_STAT_RX_MCAST); stats->rx_bcast += CSR_READ_4(sc, ET_STAT_RX_BCAST); stats->rx_control += CSR_READ_4(sc, ET_STAT_RX_CTL); stats->rx_pause += CSR_READ_4(sc, ET_STAT_RX_PAUSE); stats->rx_unknown_control += CSR_READ_4(sc, ET_STAT_RX_UNKNOWN_CTL); stats->rx_alignerrs += CSR_READ_4(sc, ET_STAT_RX_ALIGN_ERR); stats->rx_lenerrs += CSR_READ_4(sc, ET_STAT_RX_LEN_ERR); stats->rx_codeerrs += CSR_READ_4(sc, ET_STAT_RX_CODE_ERR); stats->rx_cserrs += CSR_READ_4(sc, ET_STAT_RX_CS_ERR); stats->rx_runts += CSR_READ_4(sc, ET_STAT_RX_RUNT); stats->rx_oversize += CSR_READ_4(sc, ET_STAT_RX_OVERSIZE); stats->rx_fragments += CSR_READ_4(sc, ET_STAT_RX_FRAG); stats->rx_jabbers += CSR_READ_4(sc, ET_STAT_RX_JABBER); stats->rx_drop += CSR_READ_4(sc, ET_STAT_RX_DROP); stats->tx_bytes += CSR_READ_4(sc, ET_STAT_TX_BYTES); stats->tx_frames += CSR_READ_4(sc, ET_STAT_TX_FRAMES); stats->tx_mcast += CSR_READ_4(sc, ET_STAT_TX_MCAST); stats->tx_bcast += CSR_READ_4(sc, ET_STAT_TX_BCAST); stats->tx_pause += CSR_READ_4(sc, ET_STAT_TX_PAUSE); stats->tx_deferred += CSR_READ_4(sc, ET_STAT_TX_DEFER); stats->tx_excess_deferred += CSR_READ_4(sc, ET_STAT_TX_EXCESS_DEFER); stats->tx_single_colls += CSR_READ_4(sc, ET_STAT_TX_SINGLE_COL); stats->tx_multi_colls += CSR_READ_4(sc, ET_STAT_TX_MULTI_COL); stats->tx_late_colls += CSR_READ_4(sc, ET_STAT_TX_LATE_COL); stats->tx_excess_colls += CSR_READ_4(sc, ET_STAT_TX_EXCESS_COL); stats->tx_total_colls += CSR_READ_4(sc, ET_STAT_TX_TOTAL_COL); stats->tx_pause_honored += CSR_READ_4(sc, ET_STAT_TX_PAUSE_HONOR); stats->tx_drop += CSR_READ_4(sc, ET_STAT_TX_DROP); stats->tx_jabbers += CSR_READ_4(sc, ET_STAT_TX_JABBER); stats->tx_crcerrs += CSR_READ_4(sc, ET_STAT_TX_CRC_ERR); stats->tx_control += CSR_READ_4(sc, ET_STAT_TX_CTL); stats->tx_oversize += CSR_READ_4(sc, ET_STAT_TX_OVERSIZE); stats->tx_undersize += CSR_READ_4(sc, ET_STAT_TX_UNDERSIZE); stats->tx_fragments += CSR_READ_4(sc, ET_STAT_TX_FRAG); /* Update ifnet counters. */ ifp = sc->ifp; ifp->if_opackets = (u_long)stats->tx_frames; ifp->if_collisions = stats->tx_total_colls; ifp->if_oerrors = stats->tx_drop + stats->tx_jabbers + stats->tx_crcerrs + stats->tx_excess_deferred + stats->tx_late_colls; ifp->if_ipackets = (u_long)stats->rx_frames; ifp->if_ierrors = stats->rx_crcerrs + stats->rx_alignerrs + stats->rx_lenerrs + stats->rx_codeerrs + stats->rx_cserrs + stats->rx_runts + stats->rx_jabbers + stats->rx_drop; } static int et_suspend(device_t dev) { struct et_softc *sc; uint32_t pmcfg; sc = device_get_softc(dev); ET_LOCK(sc); if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) et_stop(sc); /* Diable all clocks and put PHY into COMA. */ pmcfg = CSR_READ_4(sc, ET_PM); pmcfg &= ~(EM_PM_GIGEPHY_ENB | ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE | ET_PM_RXCLK_GATE); pmcfg |= ET_PM_PHY_SW_COMA; CSR_WRITE_4(sc, ET_PM, pmcfg); ET_UNLOCK(sc); return (0); } static int et_resume(device_t dev) { struct et_softc *sc; uint32_t pmcfg; sc = device_get_softc(dev); ET_LOCK(sc); /* Take PHY out of COMA and enable clocks. */ pmcfg = ET_PM_SYSCLK_GATE | ET_PM_TXCLK_GATE | ET_PM_RXCLK_GATE; if ((sc->sc_flags & ET_FLAG_FASTETHER) == 0) pmcfg |= EM_PM_GIGEPHY_ENB; CSR_WRITE_4(sc, ET_PM, pmcfg); if ((sc->ifp->if_flags & IFF_UP) != 0) et_init_locked(sc); ET_UNLOCK(sc); return (0); } Index: stable/7/sys/dev/jme/if_jme.c =================================================================== --- stable/7/sys/dev/jme/if_jme.c (revision 242221) +++ stable/7/sys/dev/jme/if_jme.c (revision 242222) @@ -1,3466 +1,3466 @@ /*- * Copyright (c) 2008, Pyun YongHyeon * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* "device miibus" required. See GENERIC if you get errors here. */ #include "miibus_if.h" /* Define the following to disable printing Rx errors. */ #undef JME_SHOW_ERRORS #define JME_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP) MODULE_DEPEND(jme, pci, 1, 1, 1); MODULE_DEPEND(jme, ether, 1, 1, 1); MODULE_DEPEND(jme, miibus, 1, 1, 1); /* Tunables. */ static int msi_disable = 0; static int msix_disable = 0; TUNABLE_INT("hw.jme.msi_disable", &msi_disable); TUNABLE_INT("hw.jme.msix_disable", &msix_disable); /* * Devices supported by this driver. */ static struct jme_dev { uint16_t jme_vendorid; uint16_t jme_deviceid; const char *jme_name; } jme_devs[] = { { VENDORID_JMICRON, DEVICEID_JMC250, "JMicron Inc, JMC25x Gigabit Ethernet" }, { VENDORID_JMICRON, DEVICEID_JMC260, "JMicron Inc, JMC26x Fast Ethernet" }, }; static int jme_miibus_readreg(device_t, int, int); static int jme_miibus_writereg(device_t, int, int, int); static void jme_miibus_statchg(device_t); static void jme_mediastatus(struct ifnet *, struct ifmediareq *); static int jme_mediachange(struct ifnet *); static int jme_probe(device_t); static int jme_eeprom_read_byte(struct jme_softc *, uint8_t, uint8_t *); static int jme_eeprom_macaddr(struct jme_softc *); static int jme_efuse_macaddr(struct jme_softc *); static void jme_reg_macaddr(struct jme_softc *); static void jme_set_macaddr(struct jme_softc *, uint8_t *); static void jme_map_intr_vector(struct jme_softc *); static int jme_attach(device_t); static int jme_detach(device_t); static void jme_sysctl_node(struct jme_softc *); static void jme_dmamap_cb(void *, bus_dma_segment_t *, int, int); static int jme_dma_alloc(struct jme_softc *); static void jme_dma_free(struct jme_softc *); static int jme_shutdown(device_t); static void jme_setlinkspeed(struct jme_softc *); static void jme_setwol(struct jme_softc *); static int jme_suspend(device_t); static int jme_resume(device_t); static int jme_encap(struct jme_softc *, struct mbuf **); static void jme_start(struct ifnet *); static void jme_start_locked(struct ifnet *); static void jme_watchdog(struct jme_softc *); static int jme_ioctl(struct ifnet *, u_long, caddr_t); static void jme_mac_config(struct jme_softc *); static void jme_link_task(void *, int); static int jme_intr(void *); static void jme_int_task(void *, int); static void jme_txeof(struct jme_softc *); static __inline void jme_discard_rxbuf(struct jme_softc *, int); static void jme_rxeof(struct jme_softc *); static int jme_rxintr(struct jme_softc *, int); static void jme_tick(void *); static void jme_reset(struct jme_softc *); static void jme_init(void *); static void jme_init_locked(struct jme_softc *); static void jme_stop(struct jme_softc *); static void jme_stop_tx(struct jme_softc *); static void jme_stop_rx(struct jme_softc *); static int jme_init_rx_ring(struct jme_softc *); static void jme_init_tx_ring(struct jme_softc *); static void jme_init_ssb(struct jme_softc *); static int jme_newbuf(struct jme_softc *, struct jme_rxdesc *); static void jme_set_vlan(struct jme_softc *); static void jme_set_filter(struct jme_softc *); static void jme_stats_clear(struct jme_softc *); static void jme_stats_save(struct jme_softc *); static void jme_stats_update(struct jme_softc *); static void jme_phy_down(struct jme_softc *); static void jme_phy_up(struct jme_softc *); static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int); static int sysctl_hw_jme_tx_coal_to(SYSCTL_HANDLER_ARGS); static int sysctl_hw_jme_tx_coal_pkt(SYSCTL_HANDLER_ARGS); static int sysctl_hw_jme_rx_coal_to(SYSCTL_HANDLER_ARGS); static int sysctl_hw_jme_rx_coal_pkt(SYSCTL_HANDLER_ARGS); static int sysctl_hw_jme_proc_limit(SYSCTL_HANDLER_ARGS); static device_method_t jme_methods[] = { /* Device interface. */ DEVMETHOD(device_probe, jme_probe), DEVMETHOD(device_attach, jme_attach), DEVMETHOD(device_detach, jme_detach), DEVMETHOD(device_shutdown, jme_shutdown), DEVMETHOD(device_suspend, jme_suspend), DEVMETHOD(device_resume, jme_resume), /* MII interface. */ DEVMETHOD(miibus_readreg, jme_miibus_readreg), DEVMETHOD(miibus_writereg, jme_miibus_writereg), DEVMETHOD(miibus_statchg, jme_miibus_statchg), { NULL, NULL } }; static driver_t jme_driver = { "jme", jme_methods, sizeof(struct jme_softc) }; static devclass_t jme_devclass; DRIVER_MODULE(jme, pci, jme_driver, jme_devclass, 0, 0); DRIVER_MODULE(miibus, jme, miibus_driver, miibus_devclass, 0, 0); static struct resource_spec jme_res_spec_mem[] = { { SYS_RES_MEMORY, PCIR_BAR(0), RF_ACTIVE }, { -1, 0, 0 } }; static struct resource_spec jme_irq_spec_legacy[] = { { SYS_RES_IRQ, 0, RF_ACTIVE | RF_SHAREABLE }, { -1, 0, 0 } }; static struct resource_spec jme_irq_spec_msi[] = { { SYS_RES_IRQ, 1, RF_ACTIVE }, { -1, 0, 0 } }; /* * Read a PHY register on the MII of the JMC250. */ static int jme_miibus_readreg(device_t dev, int phy, int reg) { struct jme_softc *sc; uint32_t val; int i; sc = device_get_softc(dev); /* For FPGA version, PHY address 0 should be ignored. */ if ((sc->jme_flags & JME_FLAG_FPGA) != 0 && phy == 0) return (0); CSR_WRITE_4(sc, JME_SMI, SMI_OP_READ | SMI_OP_EXECUTE | SMI_PHY_ADDR(phy) | SMI_REG_ADDR(reg)); for (i = JME_PHY_TIMEOUT; i > 0; i--) { DELAY(1); if (((val = CSR_READ_4(sc, JME_SMI)) & SMI_OP_EXECUTE) == 0) break; } if (i == 0) { device_printf(sc->jme_dev, "phy read timeout : %d\n", reg); return (0); } return ((val & SMI_DATA_MASK) >> SMI_DATA_SHIFT); } /* * Write a PHY register on the MII of the JMC250. */ static int jme_miibus_writereg(device_t dev, int phy, int reg, int val) { struct jme_softc *sc; int i; sc = device_get_softc(dev); /* For FPGA version, PHY address 0 should be ignored. */ if ((sc->jme_flags & JME_FLAG_FPGA) != 0 && phy == 0) return (0); CSR_WRITE_4(sc, JME_SMI, SMI_OP_WRITE | SMI_OP_EXECUTE | ((val << SMI_DATA_SHIFT) & SMI_DATA_MASK) | SMI_PHY_ADDR(phy) | SMI_REG_ADDR(reg)); for (i = JME_PHY_TIMEOUT; i > 0; i--) { DELAY(1); if (((val = CSR_READ_4(sc, JME_SMI)) & SMI_OP_EXECUTE) == 0) break; } if (i == 0) device_printf(sc->jme_dev, "phy write timeout : %d\n", reg); return (0); } /* * Callback from MII layer when media changes. */ static void jme_miibus_statchg(device_t dev) { struct jme_softc *sc; sc = device_get_softc(dev); taskqueue_enqueue(taskqueue_swi, &sc->jme_link_task); } /* * Get the current interface media status. */ static void jme_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr) { struct jme_softc *sc; struct mii_data *mii; sc = ifp->if_softc; JME_LOCK(sc); if ((ifp->if_flags & IFF_UP) == 0) { JME_UNLOCK(sc); return; } mii = device_get_softc(sc->jme_miibus); mii_pollstat(mii); ifmr->ifm_status = mii->mii_media_status; ifmr->ifm_active = mii->mii_media_active; JME_UNLOCK(sc); } /* * Set hardware to newly-selected media. */ static int jme_mediachange(struct ifnet *ifp) { struct jme_softc *sc; struct mii_data *mii; struct mii_softc *miisc; int error; sc = ifp->if_softc; JME_LOCK(sc); mii = device_get_softc(sc->jme_miibus); LIST_FOREACH(miisc, &mii->mii_phys, mii_list) mii_phy_reset(miisc); error = mii_mediachg(mii); JME_UNLOCK(sc); return (error); } static int jme_probe(device_t dev) { struct jme_dev *sp; int i; uint16_t vendor, devid; vendor = pci_get_vendor(dev); devid = pci_get_device(dev); sp = jme_devs; for (i = 0; i < sizeof(jme_devs) / sizeof(jme_devs[0]); i++, sp++) { if (vendor == sp->jme_vendorid && devid == sp->jme_deviceid) { device_set_desc(dev, sp->jme_name); return (BUS_PROBE_DEFAULT); } } return (ENXIO); } static int jme_eeprom_read_byte(struct jme_softc *sc, uint8_t addr, uint8_t *val) { uint32_t reg; int i; *val = 0; for (i = JME_TIMEOUT; i > 0; i--) { reg = CSR_READ_4(sc, JME_SMBCSR); if ((reg & SMBCSR_HW_BUSY_MASK) == SMBCSR_HW_IDLE) break; DELAY(1); } if (i == 0) { device_printf(sc->jme_dev, "EEPROM idle timeout!\n"); return (ETIMEDOUT); } reg = ((uint32_t)addr << SMBINTF_ADDR_SHIFT) & SMBINTF_ADDR_MASK; CSR_WRITE_4(sc, JME_SMBINTF, reg | SMBINTF_RD | SMBINTF_CMD_TRIGGER); for (i = JME_TIMEOUT; i > 0; i--) { DELAY(1); reg = CSR_READ_4(sc, JME_SMBINTF); if ((reg & SMBINTF_CMD_TRIGGER) == 0) break; } if (i == 0) { device_printf(sc->jme_dev, "EEPROM read timeout!\n"); return (ETIMEDOUT); } reg = CSR_READ_4(sc, JME_SMBINTF); *val = (reg & SMBINTF_RD_DATA_MASK) >> SMBINTF_RD_DATA_SHIFT; return (0); } static int jme_eeprom_macaddr(struct jme_softc *sc) { uint8_t eaddr[ETHER_ADDR_LEN]; uint8_t fup, reg, val; uint32_t offset; int match; offset = 0; if (jme_eeprom_read_byte(sc, offset++, &fup) != 0 || fup != JME_EEPROM_SIG0) return (ENOENT); if (jme_eeprom_read_byte(sc, offset++, &fup) != 0 || fup != JME_EEPROM_SIG1) return (ENOENT); match = 0; do { if (jme_eeprom_read_byte(sc, offset, &fup) != 0) break; if (JME_EEPROM_MKDESC(JME_EEPROM_FUNC0, JME_EEPROM_PAGE_BAR1) == (fup & (JME_EEPROM_FUNC_MASK | JME_EEPROM_PAGE_MASK))) { if (jme_eeprom_read_byte(sc, offset + 1, ®) != 0) break; if (reg >= JME_PAR0 && reg < JME_PAR0 + ETHER_ADDR_LEN) { if (jme_eeprom_read_byte(sc, offset + 2, &val) != 0) break; eaddr[reg - JME_PAR0] = val; match++; } } /* Check for the end of EEPROM descriptor. */ if ((fup & JME_EEPROM_DESC_END) == JME_EEPROM_DESC_END) break; /* Try next eeprom descriptor. */ offset += JME_EEPROM_DESC_BYTES; } while (match != ETHER_ADDR_LEN && offset < JME_EEPROM_END); if (match == ETHER_ADDR_LEN) { bcopy(eaddr, sc->jme_eaddr, ETHER_ADDR_LEN); return (0); } return (ENOENT); } static int jme_efuse_macaddr(struct jme_softc *sc) { uint32_t reg; int i; reg = pci_read_config(sc->jme_dev, JME_EFUSE_CTL1, 4); if ((reg & (EFUSE_CTL1_AUTOLOAD_ERR | EFUSE_CTL1_AUTOLAOD_DONE)) != EFUSE_CTL1_AUTOLAOD_DONE) return (ENOENT); /* Reset eFuse controller. */ reg = pci_read_config(sc->jme_dev, JME_EFUSE_CTL2, 4); reg |= EFUSE_CTL2_RESET; pci_write_config(sc->jme_dev, JME_EFUSE_CTL2, reg, 4); reg = pci_read_config(sc->jme_dev, JME_EFUSE_CTL2, 4); reg &= ~EFUSE_CTL2_RESET; pci_write_config(sc->jme_dev, JME_EFUSE_CTL2, reg, 4); /* Have eFuse reload station address to MAC controller. */ reg = pci_read_config(sc->jme_dev, JME_EFUSE_CTL1, 4); reg &= ~EFUSE_CTL1_CMD_MASK; reg |= EFUSE_CTL1_CMD_AUTOLOAD | EFUSE_CTL1_EXECUTE; pci_write_config(sc->jme_dev, JME_EFUSE_CTL1, reg, 4); /* * Verify completion of eFuse autload command. It should be * completed within 108us. */ DELAY(110); for (i = 10; i > 0; i--) { reg = pci_read_config(sc->jme_dev, JME_EFUSE_CTL1, 4); if ((reg & (EFUSE_CTL1_AUTOLOAD_ERR | EFUSE_CTL1_AUTOLAOD_DONE)) != EFUSE_CTL1_AUTOLAOD_DONE) { DELAY(20); continue; } if ((reg & EFUSE_CTL1_EXECUTE) == 0) break; /* Station address loading is still in progress. */ DELAY(20); } if (i == 0) { device_printf(sc->jme_dev, "eFuse autoload timed out.\n"); return (ETIMEDOUT); } return (0); } static void jme_reg_macaddr(struct jme_softc *sc) { uint32_t par0, par1; /* Read station address. */ par0 = CSR_READ_4(sc, JME_PAR0); par1 = CSR_READ_4(sc, JME_PAR1); par1 &= 0xFFFF; if ((par0 == 0 && par1 == 0) || (par0 == 0xFFFFFFFF && par1 == 0xFFFF)) { device_printf(sc->jme_dev, "generating fake ethernet address.\n"); par0 = arc4random(); /* Set OUI to JMicron. */ sc->jme_eaddr[0] = 0x02; /* U/L bit set. */ sc->jme_eaddr[1] = 0x1B; sc->jme_eaddr[2] = 0x8C; sc->jme_eaddr[3] = (par0 >> 16) & 0xff; sc->jme_eaddr[4] = (par0 >> 8) & 0xff; sc->jme_eaddr[5] = par0 & 0xff; } else { /* * For controllers that use eFuse, the station address * could also be extracted from JME_PCI_PAR0 and * JME_PCI_PAR1 registers in PCI configuration space. * Each register holds exactly half of station address(24bits) * so use JME_PAR0, JME_PAR1 registers instead. */ sc->jme_eaddr[0] = (par0 >> 0) & 0xFF; sc->jme_eaddr[1] = (par0 >> 8) & 0xFF; sc->jme_eaddr[2] = (par0 >> 16) & 0xFF; sc->jme_eaddr[3] = (par0 >> 24) & 0xFF; sc->jme_eaddr[4] = (par1 >> 0) & 0xFF; sc->jme_eaddr[5] = (par1 >> 8) & 0xFF; } } static void jme_set_macaddr(struct jme_softc *sc, uint8_t *eaddr) { uint32_t val; int i; if ((sc->jme_flags & JME_FLAG_EFUSE) != 0) { /* * Avoid reprogramming station address if the address * is the same as previous one. Note, reprogrammed * station address is permanent as if it was written * to EEPROM. So if station address was changed by * admistrator it's possible to lose factory configured * address when driver fails to restore its address. * (e.g. reboot or system crash) */ if (bcmp(eaddr, sc->jme_eaddr, ETHER_ADDR_LEN) != 0) { for (i = 0; i < ETHER_ADDR_LEN; i++) { val = JME_EFUSE_EEPROM_FUNC0 << JME_EFUSE_EEPROM_FUNC_SHIFT; val |= JME_EFUSE_EEPROM_PAGE_BAR1 << JME_EFUSE_EEPROM_PAGE_SHIFT; val |= (JME_PAR0 + i) << JME_EFUSE_EEPROM_ADDR_SHIFT; val |= eaddr[i] << JME_EFUSE_EEPROM_DATA_SHIFT; pci_write_config(sc->jme_dev, JME_EFUSE_EEPROM, val | JME_EFUSE_EEPROM_WRITE, 4); } } } else { CSR_WRITE_4(sc, JME_PAR0, eaddr[3] << 24 | eaddr[2] << 16 | eaddr[1] << 8 | eaddr[0]); CSR_WRITE_4(sc, JME_PAR1, eaddr[5] << 8 | eaddr[4]); } } static void jme_map_intr_vector(struct jme_softc *sc) { uint32_t map[MSINUM_NUM_INTR_SOURCE / JME_MSI_MESSAGES]; bzero(map, sizeof(map)); /* Map Tx interrupts source to MSI/MSIX vector 2. */ map[MSINUM_REG_INDEX(N_INTR_TXQ0_COMP)] = MSINUM_INTR_SOURCE(2, N_INTR_TXQ0_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ1_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ1_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ2_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ2_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ3_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ3_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ4_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ4_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ4_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ5_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ6_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ6_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ7_COMP)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ7_COMP); map[MSINUM_REG_INDEX(N_INTR_TXQ_COAL)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ_COAL); map[MSINUM_REG_INDEX(N_INTR_TXQ_COAL_TO)] |= MSINUM_INTR_SOURCE(2, N_INTR_TXQ_COAL_TO); /* Map Rx interrupts source to MSI/MSIX vector 1. */ map[MSINUM_REG_INDEX(N_INTR_RXQ0_COMP)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ0_COMP); map[MSINUM_REG_INDEX(N_INTR_RXQ1_COMP)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ1_COMP); map[MSINUM_REG_INDEX(N_INTR_RXQ2_COMP)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ2_COMP); map[MSINUM_REG_INDEX(N_INTR_RXQ3_COMP)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ3_COMP); map[MSINUM_REG_INDEX(N_INTR_RXQ0_DESC_EMPTY)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ0_DESC_EMPTY); map[MSINUM_REG_INDEX(N_INTR_RXQ1_DESC_EMPTY)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ1_DESC_EMPTY); map[MSINUM_REG_INDEX(N_INTR_RXQ2_DESC_EMPTY)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ2_DESC_EMPTY); map[MSINUM_REG_INDEX(N_INTR_RXQ3_DESC_EMPTY)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ3_DESC_EMPTY); map[MSINUM_REG_INDEX(N_INTR_RXQ0_COAL)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ0_COAL); map[MSINUM_REG_INDEX(N_INTR_RXQ1_COAL)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ1_COAL); map[MSINUM_REG_INDEX(N_INTR_RXQ2_COAL)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ2_COAL); map[MSINUM_REG_INDEX(N_INTR_RXQ3_COAL)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ3_COAL); map[MSINUM_REG_INDEX(N_INTR_RXQ0_COAL_TO)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ0_COAL_TO); map[MSINUM_REG_INDEX(N_INTR_RXQ1_COAL_TO)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ1_COAL_TO); map[MSINUM_REG_INDEX(N_INTR_RXQ2_COAL_TO)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ2_COAL_TO); map[MSINUM_REG_INDEX(N_INTR_RXQ3_COAL_TO)] = MSINUM_INTR_SOURCE(1, N_INTR_RXQ3_COAL_TO); /* Map all other interrupts source to MSI/MSIX vector 0. */ CSR_WRITE_4(sc, JME_MSINUM_BASE + sizeof(uint32_t) * 0, map[0]); CSR_WRITE_4(sc, JME_MSINUM_BASE + sizeof(uint32_t) * 1, map[1]); CSR_WRITE_4(sc, JME_MSINUM_BASE + sizeof(uint32_t) * 2, map[2]); CSR_WRITE_4(sc, JME_MSINUM_BASE + sizeof(uint32_t) * 3, map[3]); } static int jme_attach(device_t dev) { struct jme_softc *sc; struct ifnet *ifp; struct mii_softc *miisc; struct mii_data *mii; uint32_t reg; uint16_t burst; int error, i, mii_flags, msic, msixc, pmc; error = 0; sc = device_get_softc(dev); sc->jme_dev = dev; mtx_init(&sc->jme_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK, MTX_DEF); callout_init_mtx(&sc->jme_tick_ch, &sc->jme_mtx, 0); TASK_INIT(&sc->jme_int_task, 0, jme_int_task, sc); TASK_INIT(&sc->jme_link_task, 0, jme_link_task, sc); /* * Map the device. JMC250 supports both memory mapped and I/O * register space access. Because I/O register access should * use different BARs to access registers it's waste of time * to use I/O register spce access. JMC250 uses 16K to map * entire memory space. */ pci_enable_busmaster(dev); sc->jme_res_spec = jme_res_spec_mem; sc->jme_irq_spec = jme_irq_spec_legacy; error = bus_alloc_resources(dev, sc->jme_res_spec, sc->jme_res); if (error != 0) { device_printf(dev, "cannot allocate memory resources.\n"); goto fail; } /* Allocate IRQ resources. */ msixc = pci_msix_count(dev); msic = pci_msi_count(dev); if (bootverbose) { device_printf(dev, "MSIX count : %d\n", msixc); device_printf(dev, "MSI count : %d\n", msic); } /* Use 1 MSI/MSI-X. */ if (msixc > 1) msixc = 1; if (msic > 1) msic = 1; /* Prefer MSIX over MSI. */ if (msix_disable == 0 || msi_disable == 0) { if (msix_disable == 0 && msixc > 0 && pci_alloc_msix(dev, &msixc) == 0) { if (msixc == 1) { device_printf(dev, "Using %d MSIX messages.\n", msixc); sc->jme_flags |= JME_FLAG_MSIX; sc->jme_irq_spec = jme_irq_spec_msi; } else pci_release_msi(dev); } if (msi_disable == 0 && (sc->jme_flags & JME_FLAG_MSIX) == 0 && msic > 0 && pci_alloc_msi(dev, &msic) == 0) { if (msic == 1) { device_printf(dev, "Using %d MSI messages.\n", msic); sc->jme_flags |= JME_FLAG_MSI; sc->jme_irq_spec = jme_irq_spec_msi; } else pci_release_msi(dev); } /* Map interrupt vector 0, 1 and 2. */ if ((sc->jme_flags & JME_FLAG_MSI) != 0 || (sc->jme_flags & JME_FLAG_MSIX) != 0) jme_map_intr_vector(sc); } error = bus_alloc_resources(dev, sc->jme_irq_spec, sc->jme_irq); if (error != 0) { device_printf(dev, "cannot allocate IRQ resources.\n"); goto fail; } sc->jme_rev = pci_get_device(dev); if ((sc->jme_rev & DEVICEID_JMC2XX_MASK) == DEVICEID_JMC260) { sc->jme_flags |= JME_FLAG_FASTETH; sc->jme_flags |= JME_FLAG_NOJUMBO; } reg = CSR_READ_4(sc, JME_CHIPMODE); sc->jme_chip_rev = (reg & CHIPMODE_REV_MASK) >> CHIPMODE_REV_SHIFT; if (((reg & CHIPMODE_FPGA_REV_MASK) >> CHIPMODE_FPGA_REV_SHIFT) != CHIPMODE_NOT_FPGA) sc->jme_flags |= JME_FLAG_FPGA; if (bootverbose) { device_printf(dev, "PCI device revision : 0x%04x\n", sc->jme_rev); device_printf(dev, "Chip revision : 0x%02x\n", sc->jme_chip_rev); if ((sc->jme_flags & JME_FLAG_FPGA) != 0) device_printf(dev, "FPGA revision : 0x%04x\n", (reg & CHIPMODE_FPGA_REV_MASK) >> CHIPMODE_FPGA_REV_SHIFT); } if (sc->jme_chip_rev == 0xFF) { device_printf(dev, "Unknown chip revision : 0x%02x\n", sc->jme_rev); error = ENXIO; goto fail; } /* Identify controller features and bugs. */ if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 2) { if ((sc->jme_rev & DEVICEID_JMC2XX_MASK) == DEVICEID_JMC260 && CHIPMODE_REVFM(sc->jme_chip_rev) == 2) sc->jme_flags |= JME_FLAG_DMA32BIT; if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 5) sc->jme_flags |= JME_FLAG_EFUSE | JME_FLAG_PCCPCD; sc->jme_flags |= JME_FLAG_TXCLK | JME_FLAG_RXCLK; sc->jme_flags |= JME_FLAG_HWMIB; } /* Reset the ethernet controller. */ jme_reset(sc); /* Get station address. */ if ((sc->jme_flags & JME_FLAG_EFUSE) != 0) { error = jme_efuse_macaddr(sc); if (error == 0) jme_reg_macaddr(sc); } else { error = ENOENT; reg = CSR_READ_4(sc, JME_SMBCSR); if ((reg & SMBCSR_EEPROM_PRESENT) != 0) error = jme_eeprom_macaddr(sc); if (error != 0 && bootverbose) device_printf(sc->jme_dev, "ethernet hardware address not found in EEPROM.\n"); if (error != 0) jme_reg_macaddr(sc); } /* * Save PHY address. * Integrated JR0211 has fixed PHY address whereas FPGA version * requires PHY probing to get correct PHY address. */ if ((sc->jme_flags & JME_FLAG_FPGA) == 0) { sc->jme_phyaddr = CSR_READ_4(sc, JME_GPREG0) & GPREG0_PHY_ADDR_MASK; if (bootverbose) device_printf(dev, "PHY is at address %d.\n", sc->jme_phyaddr); } else sc->jme_phyaddr = 0; /* Set max allowable DMA size. */ if (pci_find_extcap(dev, PCIY_EXPRESS, &i) == 0) { sc->jme_flags |= JME_FLAG_PCIE; - burst = pci_read_config(dev, i + PCIR_EXPRESS_DEVICE_CTL, 2); + burst = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2); if (bootverbose) { device_printf(dev, "Read request size : %d bytes.\n", 128 << ((burst >> 12) & 0x07)); device_printf(dev, "TLP payload size : %d bytes.\n", 128 << ((burst >> 5) & 0x07)); } switch ((burst >> 12) & 0x07) { case 0: sc->jme_tx_dma_size = TXCSR_DMA_SIZE_128; break; case 1: sc->jme_tx_dma_size = TXCSR_DMA_SIZE_256; break; default: sc->jme_tx_dma_size = TXCSR_DMA_SIZE_512; break; } sc->jme_rx_dma_size = RXCSR_DMA_SIZE_128; } else { sc->jme_tx_dma_size = TXCSR_DMA_SIZE_512; sc->jme_rx_dma_size = RXCSR_DMA_SIZE_128; } /* Create coalescing sysctl node. */ jme_sysctl_node(sc); if ((error = jme_dma_alloc(sc) != 0)) goto fail; ifp = sc->jme_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "cannot allocate ifnet structure.\n"); error = ENXIO; goto fail; } ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = jme_ioctl; ifp->if_start = jme_start; ifp->if_init = jme_init; ifp->if_snd.ifq_drv_maxlen = JME_TX_RING_CNT - 1; IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen); IFQ_SET_READY(&ifp->if_snd); /* JMC250 supports Tx/Rx checksum offload as well as TSO. */ ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4; ifp->if_hwassist = JME_CSUM_FEATURES | CSUM_TSO; if (pci_find_extcap(dev, PCIY_PMG, &pmc) == 0) { sc->jme_flags |= JME_FLAG_PMCAP; ifp->if_capabilities |= IFCAP_WOL_MAGIC; } ifp->if_capenable = ifp->if_capabilities; /* Wakeup PHY. */ jme_phy_up(sc); mii_flags = MIIF_DOPAUSE; /* Ask PHY calibration to PHY driver. */ if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 5) mii_flags |= MIIF_MACPRIV0; /* Set up MII bus. */ error = mii_attach(dev, &sc->jme_miibus, ifp, jme_mediachange, jme_mediastatus, BMSR_DEFCAPMASK, sc->jme_flags & JME_FLAG_FPGA ? MII_PHY_ANY : sc->jme_phyaddr, MII_OFFSET_ANY, mii_flags); if (error != 0) { device_printf(dev, "attaching PHYs failed\n"); goto fail; } /* * Force PHY to FPGA mode. */ if ((sc->jme_flags & JME_FLAG_FPGA) != 0) { mii = device_get_softc(sc->jme_miibus); if (mii->mii_instance != 0) { LIST_FOREACH(miisc, &mii->mii_phys, mii_list) { if (miisc->mii_phy != 0) { sc->jme_phyaddr = miisc->mii_phy; break; } } if (sc->jme_phyaddr != 0) { device_printf(sc->jme_dev, "FPGA PHY is at %d\n", sc->jme_phyaddr); /* vendor magic. */ jme_miibus_writereg(dev, sc->jme_phyaddr, 27, 0x0004); } } } ether_ifattach(ifp, sc->jme_eaddr); /* VLAN capability setup */ ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO; ifp->if_capenable = ifp->if_capabilities; /* Tell the upper layer(s) we support long frames. */ ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); /* Create local taskq. */ sc->jme_tq = taskqueue_create_fast("jme_taskq", M_WAITOK, taskqueue_thread_enqueue, &sc->jme_tq); if (sc->jme_tq == NULL) { device_printf(dev, "could not create taskqueue.\n"); ether_ifdetach(ifp); error = ENXIO; goto fail; } taskqueue_start_threads(&sc->jme_tq, 1, PI_NET, "%s taskq", device_get_nameunit(sc->jme_dev)); for (i = 0; i < 1; i++) { error = bus_setup_intr(dev, sc->jme_irq[i], INTR_TYPE_NET | INTR_MPSAFE, jme_intr, NULL, sc, &sc->jme_intrhand[i]); if (error != 0) break; } if (error != 0) { device_printf(dev, "could not set up interrupt handler.\n"); taskqueue_free(sc->jme_tq); sc->jme_tq = NULL; ether_ifdetach(ifp); goto fail; } fail: if (error != 0) jme_detach(dev); return (error); } static int jme_detach(device_t dev) { struct jme_softc *sc; struct ifnet *ifp; int i; sc = device_get_softc(dev); ifp = sc->jme_ifp; if (device_is_attached(dev)) { JME_LOCK(sc); sc->jme_flags |= JME_FLAG_DETACH; jme_stop(sc); JME_UNLOCK(sc); callout_drain(&sc->jme_tick_ch); taskqueue_drain(sc->jme_tq, &sc->jme_int_task); taskqueue_drain(taskqueue_swi, &sc->jme_link_task); /* Restore possibly modified station address. */ if ((sc->jme_flags & JME_FLAG_EFUSE) != 0) jme_set_macaddr(sc, sc->jme_eaddr); ether_ifdetach(ifp); } if (sc->jme_tq != NULL) { taskqueue_drain(sc->jme_tq, &sc->jme_int_task); taskqueue_free(sc->jme_tq); sc->jme_tq = NULL; } if (sc->jme_miibus != NULL) { device_delete_child(dev, sc->jme_miibus); sc->jme_miibus = NULL; } bus_generic_detach(dev); jme_dma_free(sc); if (ifp != NULL) { if_free(ifp); sc->jme_ifp = NULL; } for (i = 0; i < 1; i++) { if (sc->jme_intrhand[i] != NULL) { bus_teardown_intr(dev, sc->jme_irq[i], sc->jme_intrhand[i]); sc->jme_intrhand[i] = NULL; } } if (sc->jme_irq[0] != NULL) bus_release_resources(dev, sc->jme_irq_spec, sc->jme_irq); if ((sc->jme_flags & (JME_FLAG_MSIX | JME_FLAG_MSI)) != 0) pci_release_msi(dev); if (sc->jme_res[0] != NULL) bus_release_resources(dev, sc->jme_res_spec, sc->jme_res); mtx_destroy(&sc->jme_mtx); return (0); } #define JME_SYSCTL_STAT_ADD32(c, h, n, p, d) \ SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d) static void jme_sysctl_node(struct jme_softc *sc) { struct sysctl_ctx_list *ctx; struct sysctl_oid_list *child, *parent; struct sysctl_oid *tree; struct jme_hw_stats *stats; int error; stats = &sc->jme_stats; ctx = device_get_sysctl_ctx(sc->jme_dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->jme_dev)); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_coal_to", CTLTYPE_INT | CTLFLAG_RW, &sc->jme_tx_coal_to, 0, sysctl_hw_jme_tx_coal_to, "I", "jme tx coalescing timeout"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_coal_pkt", CTLTYPE_INT | CTLFLAG_RW, &sc->jme_tx_coal_pkt, 0, sysctl_hw_jme_tx_coal_pkt, "I", "jme tx coalescing packet"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_coal_to", CTLTYPE_INT | CTLFLAG_RW, &sc->jme_rx_coal_to, 0, sysctl_hw_jme_rx_coal_to, "I", "jme rx coalescing timeout"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_coal_pkt", CTLTYPE_INT | CTLFLAG_RW, &sc->jme_rx_coal_pkt, 0, sysctl_hw_jme_rx_coal_pkt, "I", "jme rx coalescing packet"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit", CTLTYPE_INT | CTLFLAG_RW, &sc->jme_process_limit, 0, sysctl_hw_jme_proc_limit, "I", "max number of Rx events to process"); /* Pull in device tunables. */ sc->jme_process_limit = JME_PROC_DEFAULT; error = resource_int_value(device_get_name(sc->jme_dev), device_get_unit(sc->jme_dev), "process_limit", &sc->jme_process_limit); if (error == 0) { if (sc->jme_process_limit < JME_PROC_MIN || sc->jme_process_limit > JME_PROC_MAX) { device_printf(sc->jme_dev, "process_limit value out of range; " "using default: %d\n", JME_PROC_DEFAULT); sc->jme_process_limit = JME_PROC_DEFAULT; } } sc->jme_tx_coal_to = PCCTX_COAL_TO_DEFAULT; error = resource_int_value(device_get_name(sc->jme_dev), device_get_unit(sc->jme_dev), "tx_coal_to", &sc->jme_tx_coal_to); if (error == 0) { if (sc->jme_tx_coal_to < PCCTX_COAL_TO_MIN || sc->jme_tx_coal_to > PCCTX_COAL_TO_MAX) { device_printf(sc->jme_dev, "tx_coal_to value out of range; " "using default: %d\n", PCCTX_COAL_TO_DEFAULT); sc->jme_tx_coal_to = PCCTX_COAL_TO_DEFAULT; } } sc->jme_tx_coal_pkt = PCCTX_COAL_PKT_DEFAULT; error = resource_int_value(device_get_name(sc->jme_dev), device_get_unit(sc->jme_dev), "tx_coal_pkt", &sc->jme_tx_coal_to); if (error == 0) { if (sc->jme_tx_coal_pkt < PCCTX_COAL_PKT_MIN || sc->jme_tx_coal_pkt > PCCTX_COAL_PKT_MAX) { device_printf(sc->jme_dev, "tx_coal_pkt value out of range; " "using default: %d\n", PCCTX_COAL_PKT_DEFAULT); sc->jme_tx_coal_pkt = PCCTX_COAL_PKT_DEFAULT; } } sc->jme_rx_coal_to = PCCRX_COAL_TO_DEFAULT; error = resource_int_value(device_get_name(sc->jme_dev), device_get_unit(sc->jme_dev), "rx_coal_to", &sc->jme_rx_coal_to); if (error == 0) { if (sc->jme_rx_coal_to < PCCRX_COAL_TO_MIN || sc->jme_rx_coal_to > PCCRX_COAL_TO_MAX) { device_printf(sc->jme_dev, "rx_coal_to value out of range; " "using default: %d\n", PCCRX_COAL_TO_DEFAULT); sc->jme_rx_coal_to = PCCRX_COAL_TO_DEFAULT; } } sc->jme_rx_coal_pkt = PCCRX_COAL_PKT_DEFAULT; error = resource_int_value(device_get_name(sc->jme_dev), device_get_unit(sc->jme_dev), "rx_coal_pkt", &sc->jme_rx_coal_to); if (error == 0) { if (sc->jme_rx_coal_pkt < PCCRX_COAL_PKT_MIN || sc->jme_rx_coal_pkt > PCCRX_COAL_PKT_MAX) { device_printf(sc->jme_dev, "tx_coal_pkt value out of range; " "using default: %d\n", PCCRX_COAL_PKT_DEFAULT); sc->jme_rx_coal_pkt = PCCRX_COAL_PKT_DEFAULT; } } if ((sc->jme_flags & JME_FLAG_HWMIB) == 0) return; tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD, NULL, "JME statistics"); parent = SYSCTL_CHILDREN(tree); /* Rx statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD, NULL, "Rx MAC statistics"); child = SYSCTL_CHILDREN(tree); JME_SYSCTL_STAT_ADD32(ctx, child, "good_frames", &stats->rx_good_frames, "Good frames"); JME_SYSCTL_STAT_ADD32(ctx, child, "crc_errs", &stats->rx_crc_errs, "CRC errors"); JME_SYSCTL_STAT_ADD32(ctx, child, "mii_errs", &stats->rx_mii_errs, "MII errors"); JME_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows", &stats->rx_fifo_oflows, "FIFO overflows"); JME_SYSCTL_STAT_ADD32(ctx, child, "desc_empty", &stats->rx_desc_empty, "Descriptor empty"); JME_SYSCTL_STAT_ADD32(ctx, child, "bad_frames", &stats->rx_bad_frames, "Bad frames"); /* Tx statistics. */ tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD, NULL, "Tx MAC statistics"); child = SYSCTL_CHILDREN(tree); JME_SYSCTL_STAT_ADD32(ctx, child, "good_frames", &stats->tx_good_frames, "Good frames"); JME_SYSCTL_STAT_ADD32(ctx, child, "bad_frames", &stats->tx_bad_frames, "Bad frames"); } #undef JME_SYSCTL_STAT_ADD32 struct jme_dmamap_arg { bus_addr_t jme_busaddr; }; static void jme_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct jme_dmamap_arg *ctx; if (error != 0) return; KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); ctx = (struct jme_dmamap_arg *)arg; ctx->jme_busaddr = segs[0].ds_addr; } static int jme_dma_alloc(struct jme_softc *sc) { struct jme_dmamap_arg ctx; struct jme_txdesc *txd; struct jme_rxdesc *rxd; bus_addr_t lowaddr, rx_ring_end, tx_ring_end; int error, i; lowaddr = BUS_SPACE_MAXADDR; if ((sc->jme_flags & JME_FLAG_DMA32BIT) != 0) lowaddr = BUS_SPACE_MAXADDR_32BIT; again: /* Create parent ring tag. */ error = bus_dma_tag_create(bus_get_dma_tag(sc->jme_dev),/* parent */ 1, 0, /* algnmnt, boundary */ lowaddr, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT, /* maxsize */ 0, /* nsegments */ BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_ring_tag); if (error != 0) { device_printf(sc->jme_dev, "could not create parent ring DMA tag.\n"); goto fail; } /* Create tag for Tx ring. */ error = bus_dma_tag_create(sc->jme_cdata.jme_ring_tag,/* parent */ JME_TX_RING_ALIGN, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ JME_TX_RING_SIZE, /* maxsize */ 1, /* nsegments */ JME_TX_RING_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_tx_ring_tag); if (error != 0) { device_printf(sc->jme_dev, "could not allocate Tx ring DMA tag.\n"); goto fail; } /* Create tag for Rx ring. */ error = bus_dma_tag_create(sc->jme_cdata.jme_ring_tag,/* parent */ JME_RX_RING_ALIGN, 0, /* algnmnt, boundary */ lowaddr, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ JME_RX_RING_SIZE, /* maxsize */ 1, /* nsegments */ JME_RX_RING_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_rx_ring_tag); if (error != 0) { device_printf(sc->jme_dev, "could not allocate Rx ring DMA tag.\n"); goto fail; } /* Allocate DMA'able memory and load the DMA map for Tx ring. */ error = bus_dmamem_alloc(sc->jme_cdata.jme_tx_ring_tag, (void **)&sc->jme_rdata.jme_tx_ring, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->jme_cdata.jme_tx_ring_map); if (error != 0) { device_printf(sc->jme_dev, "could not allocate DMA'able memory for Tx ring.\n"); goto fail; } ctx.jme_busaddr = 0; error = bus_dmamap_load(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map, sc->jme_rdata.jme_tx_ring, JME_TX_RING_SIZE, jme_dmamap_cb, &ctx, BUS_DMA_NOWAIT); if (error != 0 || ctx.jme_busaddr == 0) { device_printf(sc->jme_dev, "could not load DMA'able memory for Tx ring.\n"); goto fail; } sc->jme_rdata.jme_tx_ring_paddr = ctx.jme_busaddr; /* Allocate DMA'able memory and load the DMA map for Rx ring. */ error = bus_dmamem_alloc(sc->jme_cdata.jme_rx_ring_tag, (void **)&sc->jme_rdata.jme_rx_ring, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->jme_cdata.jme_rx_ring_map); if (error != 0) { device_printf(sc->jme_dev, "could not allocate DMA'able memory for Rx ring.\n"); goto fail; } ctx.jme_busaddr = 0; error = bus_dmamap_load(sc->jme_cdata.jme_rx_ring_tag, sc->jme_cdata.jme_rx_ring_map, sc->jme_rdata.jme_rx_ring, JME_RX_RING_SIZE, jme_dmamap_cb, &ctx, BUS_DMA_NOWAIT); if (error != 0 || ctx.jme_busaddr == 0) { device_printf(sc->jme_dev, "could not load DMA'able memory for Rx ring.\n"); goto fail; } sc->jme_rdata.jme_rx_ring_paddr = ctx.jme_busaddr; if (lowaddr != BUS_SPACE_MAXADDR_32BIT) { /* Tx/Rx descriptor queue should reside within 4GB boundary. */ tx_ring_end = sc->jme_rdata.jme_tx_ring_paddr + JME_TX_RING_SIZE; rx_ring_end = sc->jme_rdata.jme_rx_ring_paddr + JME_RX_RING_SIZE; if ((JME_ADDR_HI(tx_ring_end) != JME_ADDR_HI(sc->jme_rdata.jme_tx_ring_paddr)) || (JME_ADDR_HI(rx_ring_end) != JME_ADDR_HI(sc->jme_rdata.jme_rx_ring_paddr))) { device_printf(sc->jme_dev, "4GB boundary crossed, " "switching to 32bit DMA address mode.\n"); jme_dma_free(sc); /* Limit DMA address space to 32bit and try again. */ lowaddr = BUS_SPACE_MAXADDR_32BIT; goto again; } } lowaddr = BUS_SPACE_MAXADDR; if ((sc->jme_flags & JME_FLAG_DMA32BIT) != 0) lowaddr = BUS_SPACE_MAXADDR_32BIT; /* Create parent buffer tag. */ error = bus_dma_tag_create(bus_get_dma_tag(sc->jme_dev),/* parent */ 1, 0, /* algnmnt, boundary */ lowaddr, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ BUS_SPACE_MAXSIZE_32BIT, /* maxsize */ 0, /* nsegments */ BUS_SPACE_MAXSIZE_32BIT, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_buffer_tag); if (error != 0) { device_printf(sc->jme_dev, "could not create parent buffer DMA tag.\n"); goto fail; } /* Create shadow status block tag. */ error = bus_dma_tag_create(sc->jme_cdata.jme_buffer_tag,/* parent */ JME_SSB_ALIGN, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ JME_SSB_SIZE, /* maxsize */ 1, /* nsegments */ JME_SSB_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_ssb_tag); if (error != 0) { device_printf(sc->jme_dev, "could not create shared status block DMA tag.\n"); goto fail; } /* Create tag for Tx buffers. */ error = bus_dma_tag_create(sc->jme_cdata.jme_buffer_tag,/* parent */ 1, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ JME_TSO_MAXSIZE, /* maxsize */ JME_MAXTXSEGS, /* nsegments */ JME_TSO_MAXSEGSIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_tx_tag); if (error != 0) { device_printf(sc->jme_dev, "could not create Tx DMA tag.\n"); goto fail; } /* Create tag for Rx buffers. */ error = bus_dma_tag_create(sc->jme_cdata.jme_buffer_tag,/* parent */ JME_RX_BUF_ALIGN, 0, /* algnmnt, boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ MCLBYTES, /* maxsize */ 1, /* nsegments */ MCLBYTES, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockarg */ &sc->jme_cdata.jme_rx_tag); if (error != 0) { device_printf(sc->jme_dev, "could not create Rx DMA tag.\n"); goto fail; } /* * Allocate DMA'able memory and load the DMA map for shared * status block. */ error = bus_dmamem_alloc(sc->jme_cdata.jme_ssb_tag, (void **)&sc->jme_rdata.jme_ssb_block, BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT, &sc->jme_cdata.jme_ssb_map); if (error != 0) { device_printf(sc->jme_dev, "could not allocate DMA'able " "memory for shared status block.\n"); goto fail; } ctx.jme_busaddr = 0; error = bus_dmamap_load(sc->jme_cdata.jme_ssb_tag, sc->jme_cdata.jme_ssb_map, sc->jme_rdata.jme_ssb_block, JME_SSB_SIZE, jme_dmamap_cb, &ctx, BUS_DMA_NOWAIT); if (error != 0 || ctx.jme_busaddr == 0) { device_printf(sc->jme_dev, "could not load DMA'able memory " "for shared status block.\n"); goto fail; } sc->jme_rdata.jme_ssb_block_paddr = ctx.jme_busaddr; /* Create DMA maps for Tx buffers. */ for (i = 0; i < JME_TX_RING_CNT; i++) { txd = &sc->jme_cdata.jme_txdesc[i]; txd->tx_m = NULL; txd->tx_dmamap = NULL; error = bus_dmamap_create(sc->jme_cdata.jme_tx_tag, 0, &txd->tx_dmamap); if (error != 0) { device_printf(sc->jme_dev, "could not create Tx dmamap.\n"); goto fail; } } /* Create DMA maps for Rx buffers. */ if ((error = bus_dmamap_create(sc->jme_cdata.jme_rx_tag, 0, &sc->jme_cdata.jme_rx_sparemap)) != 0) { device_printf(sc->jme_dev, "could not create spare Rx dmamap.\n"); goto fail; } for (i = 0; i < JME_RX_RING_CNT; i++) { rxd = &sc->jme_cdata.jme_rxdesc[i]; rxd->rx_m = NULL; rxd->rx_dmamap = NULL; error = bus_dmamap_create(sc->jme_cdata.jme_rx_tag, 0, &rxd->rx_dmamap); if (error != 0) { device_printf(sc->jme_dev, "could not create Rx dmamap.\n"); goto fail; } } fail: return (error); } static void jme_dma_free(struct jme_softc *sc) { struct jme_txdesc *txd; struct jme_rxdesc *rxd; int i; /* Tx ring */ if (sc->jme_cdata.jme_tx_ring_tag != NULL) { if (sc->jme_cdata.jme_tx_ring_map) bus_dmamap_unload(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map); if (sc->jme_cdata.jme_tx_ring_map && sc->jme_rdata.jme_tx_ring) bus_dmamem_free(sc->jme_cdata.jme_tx_ring_tag, sc->jme_rdata.jme_tx_ring, sc->jme_cdata.jme_tx_ring_map); sc->jme_rdata.jme_tx_ring = NULL; sc->jme_cdata.jme_tx_ring_map = NULL; bus_dma_tag_destroy(sc->jme_cdata.jme_tx_ring_tag); sc->jme_cdata.jme_tx_ring_tag = NULL; } /* Rx ring */ if (sc->jme_cdata.jme_rx_ring_tag != NULL) { if (sc->jme_cdata.jme_rx_ring_map) bus_dmamap_unload(sc->jme_cdata.jme_rx_ring_tag, sc->jme_cdata.jme_rx_ring_map); if (sc->jme_cdata.jme_rx_ring_map && sc->jme_rdata.jme_rx_ring) bus_dmamem_free(sc->jme_cdata.jme_rx_ring_tag, sc->jme_rdata.jme_rx_ring, sc->jme_cdata.jme_rx_ring_map); sc->jme_rdata.jme_rx_ring = NULL; sc->jme_cdata.jme_rx_ring_map = NULL; bus_dma_tag_destroy(sc->jme_cdata.jme_rx_ring_tag); sc->jme_cdata.jme_rx_ring_tag = NULL; } /* Tx buffers */ if (sc->jme_cdata.jme_tx_tag != NULL) { for (i = 0; i < JME_TX_RING_CNT; i++) { txd = &sc->jme_cdata.jme_txdesc[i]; if (txd->tx_dmamap != NULL) { bus_dmamap_destroy(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap); txd->tx_dmamap = NULL; } } bus_dma_tag_destroy(sc->jme_cdata.jme_tx_tag); sc->jme_cdata.jme_tx_tag = NULL; } /* Rx buffers */ if (sc->jme_cdata.jme_rx_tag != NULL) { for (i = 0; i < JME_RX_RING_CNT; i++) { rxd = &sc->jme_cdata.jme_rxdesc[i]; if (rxd->rx_dmamap != NULL) { bus_dmamap_destroy(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap); rxd->rx_dmamap = NULL; } } if (sc->jme_cdata.jme_rx_sparemap != NULL) { bus_dmamap_destroy(sc->jme_cdata.jme_rx_tag, sc->jme_cdata.jme_rx_sparemap); sc->jme_cdata.jme_rx_sparemap = NULL; } bus_dma_tag_destroy(sc->jme_cdata.jme_rx_tag); sc->jme_cdata.jme_rx_tag = NULL; } /* Shared status block. */ if (sc->jme_cdata.jme_ssb_tag != NULL) { if (sc->jme_cdata.jme_ssb_map) bus_dmamap_unload(sc->jme_cdata.jme_ssb_tag, sc->jme_cdata.jme_ssb_map); if (sc->jme_cdata.jme_ssb_map && sc->jme_rdata.jme_ssb_block) bus_dmamem_free(sc->jme_cdata.jme_ssb_tag, sc->jme_rdata.jme_ssb_block, sc->jme_cdata.jme_ssb_map); sc->jme_rdata.jme_ssb_block = NULL; sc->jme_cdata.jme_ssb_map = NULL; bus_dma_tag_destroy(sc->jme_cdata.jme_ssb_tag); sc->jme_cdata.jme_ssb_tag = NULL; } if (sc->jme_cdata.jme_buffer_tag != NULL) { bus_dma_tag_destroy(sc->jme_cdata.jme_buffer_tag); sc->jme_cdata.jme_buffer_tag = NULL; } if (sc->jme_cdata.jme_ring_tag != NULL) { bus_dma_tag_destroy(sc->jme_cdata.jme_ring_tag); sc->jme_cdata.jme_ring_tag = NULL; } } /* * Make sure the interface is stopped at reboot time. */ static int jme_shutdown(device_t dev) { return (jme_suspend(dev)); } /* * Unlike other ethernet controllers, JMC250 requires * explicit resetting link speed to 10/100Mbps as gigabit * link will cunsume more power than 375mA. * Note, we reset the link speed to 10/100Mbps with * auto-negotiation but we don't know whether that operation * would succeed or not as we have no control after powering * off. If the renegotiation fail WOL may not work. Running * at 1Gbps draws more power than 375mA at 3.3V which is * specified in PCI specification and that would result in * complete shutdowning power to ethernet controller. * * TODO * Save current negotiated media speed/duplex/flow-control * to softc and restore the same link again after resuming. * PHY handling such as power down/resetting to 100Mbps * may be better handled in suspend method in phy driver. */ static void jme_setlinkspeed(struct jme_softc *sc) { struct mii_data *mii; int aneg, i; JME_LOCK_ASSERT(sc); mii = device_get_softc(sc->jme_miibus); mii_pollstat(mii); aneg = 0; if ((mii->mii_media_status & IFM_AVALID) != 0) { switch IFM_SUBTYPE(mii->mii_media_active) { case IFM_10_T: case IFM_100_TX: return; case IFM_1000_T: aneg++; default: break; } } jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, MII_100T2CR, 0); jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA); jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG); DELAY(1000); if (aneg != 0) { /* Poll link state until jme(4) get a 10/100 link. */ for (i = 0; i < MII_ANEGTICKS_GIGE; i++) { mii_pollstat(mii); if ((mii->mii_media_status & IFM_AVALID) != 0) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: jme_mac_config(sc); return; default: break; } } JME_UNLOCK(sc); pause("jmelnk", hz); JME_LOCK(sc); } if (i == MII_ANEGTICKS_GIGE) device_printf(sc->jme_dev, "establishing link failed, " "WOL may not work!"); } /* * No link, force MAC to have 100Mbps, full-duplex link. * This is the last resort and may/may not work. */ mii->mii_media_status = IFM_AVALID | IFM_ACTIVE; mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX; jme_mac_config(sc); } static void jme_setwol(struct jme_softc *sc) { struct ifnet *ifp; uint32_t gpr, pmcs; uint16_t pmstat; int pmc; JME_LOCK_ASSERT(sc); if (pci_find_extcap(sc->jme_dev, PCIY_PMG, &pmc) != 0) { /* Remove Tx MAC/offload clock to save more power. */ if ((sc->jme_flags & JME_FLAG_TXCLK) != 0) CSR_WRITE_4(sc, JME_GHC, CSR_READ_4(sc, JME_GHC) & ~(GHC_TX_OFFLD_CLK_100 | GHC_TX_MAC_CLK_100 | GHC_TX_OFFLD_CLK_1000 | GHC_TX_MAC_CLK_1000)); if ((sc->jme_flags & JME_FLAG_RXCLK) != 0) CSR_WRITE_4(sc, JME_GPREG1, CSR_READ_4(sc, JME_GPREG1) | GPREG1_RX_MAC_CLK_DIS); /* No PME capability, PHY power down. */ jme_phy_down(sc); return; } ifp = sc->jme_ifp; gpr = CSR_READ_4(sc, JME_GPREG0) & ~GPREG0_PME_ENB; pmcs = CSR_READ_4(sc, JME_PMCS); pmcs &= ~PMCS_WOL_ENB_MASK; if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0) { pmcs |= PMCS_MAGIC_FRAME | PMCS_MAGIC_FRAME_ENB; /* Enable PME message. */ gpr |= GPREG0_PME_ENB; /* For gigabit controllers, reset link speed to 10/100. */ if ((sc->jme_flags & JME_FLAG_FASTETH) == 0) jme_setlinkspeed(sc); } CSR_WRITE_4(sc, JME_PMCS, pmcs); CSR_WRITE_4(sc, JME_GPREG0, gpr); /* Remove Tx MAC/offload clock to save more power. */ if ((sc->jme_flags & JME_FLAG_TXCLK) != 0) CSR_WRITE_4(sc, JME_GHC, CSR_READ_4(sc, JME_GHC) & ~(GHC_TX_OFFLD_CLK_100 | GHC_TX_MAC_CLK_100 | GHC_TX_OFFLD_CLK_1000 | GHC_TX_MAC_CLK_1000)); /* Request PME. */ pmstat = pci_read_config(sc->jme_dev, pmc + PCIR_POWER_STATUS, 2); pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE); if ((ifp->if_capenable & IFCAP_WOL) != 0) pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE; pci_write_config(sc->jme_dev, pmc + PCIR_POWER_STATUS, pmstat, 2); if ((ifp->if_capenable & IFCAP_WOL) == 0) { /* No WOL, PHY power down. */ jme_phy_down(sc); } } static int jme_suspend(device_t dev) { struct jme_softc *sc; sc = device_get_softc(dev); JME_LOCK(sc); jme_stop(sc); jme_setwol(sc); JME_UNLOCK(sc); return (0); } static int jme_resume(device_t dev) { struct jme_softc *sc; struct ifnet *ifp; uint16_t pmstat; int pmc; sc = device_get_softc(dev); JME_LOCK(sc); if (pci_find_extcap(sc->jme_dev, PCIY_PMG, &pmc) != 0) { pmstat = pci_read_config(sc->jme_dev, pmc + PCIR_POWER_STATUS, 2); /* Disable PME clear PME status. */ pmstat &= ~PCIM_PSTAT_PMEENABLE; pci_write_config(sc->jme_dev, pmc + PCIR_POWER_STATUS, pmstat, 2); } /* Wakeup PHY. */ jme_phy_up(sc); ifp = sc->jme_ifp; if ((ifp->if_flags & IFF_UP) != 0) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; jme_init_locked(sc); } JME_UNLOCK(sc); return (0); } static int jme_encap(struct jme_softc *sc, struct mbuf **m_head) { struct jme_txdesc *txd; struct jme_desc *desc; struct mbuf *m; bus_dma_segment_t txsegs[JME_MAXTXSEGS]; int error, i, nsegs, prod; uint32_t cflags, tso_segsz; JME_LOCK_ASSERT(sc); M_ASSERTPKTHDR((*m_head)); if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) { /* * Due to the adherence to NDIS specification JMC250 * assumes upper stack computed TCP pseudo checksum * without including payload length. This breaks * checksum offload for TSO case so recompute TCP * pseudo checksum for JMC250. Hopefully this wouldn't * be much burden on modern CPUs. */ struct ether_header *eh; struct ip *ip; struct tcphdr *tcp; uint32_t ip_off, poff; if (M_WRITABLE(*m_head) == 0) { /* Get a writable copy. */ m = m_dup(*m_head, M_DONTWAIT); m_freem(*m_head); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } *m_head = m; } ip_off = sizeof(struct ether_header); m = m_pullup(*m_head, ip_off); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } eh = mtod(m, struct ether_header *); /* Check the existence of VLAN tag. */ if (eh->ether_type == htons(ETHERTYPE_VLAN)) { ip_off = sizeof(struct ether_vlan_header); m = m_pullup(m, ip_off); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } } m = m_pullup(m, ip_off + sizeof(struct ip)); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } ip = (struct ip *)(mtod(m, char *) + ip_off); poff = ip_off + (ip->ip_hl << 2); m = m_pullup(m, poff + sizeof(struct tcphdr)); if (m == NULL) { *m_head = NULL; return (ENOBUFS); } /* * Reset IP checksum and recompute TCP pseudo * checksum that NDIS specification requires. */ ip = (struct ip *)(mtod(m, char *) + ip_off); tcp = (struct tcphdr *)(mtod(m, char *) + poff); ip->ip_sum = 0; if (poff + (tcp->th_off << 2) == m->m_pkthdr.len) { tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((tcp->th_off << 2) + IPPROTO_TCP)); /* No need to TSO, force IP checksum offload. */ (*m_head)->m_pkthdr.csum_flags &= ~CSUM_TSO; (*m_head)->m_pkthdr.csum_flags |= CSUM_IP; } else tcp->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); *m_head = m; } prod = sc->jme_cdata.jme_tx_prod; txd = &sc->jme_cdata.jme_txdesc[prod]; error = bus_dmamap_load_mbuf_sg(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, *m_head, txsegs, &nsegs, 0); if (error == EFBIG) { m = m_collapse(*m_head, M_DONTWAIT, JME_MAXTXSEGS); if (m == NULL) { m_freem(*m_head); *m_head = NULL; return (ENOMEM); } *m_head = m; error = bus_dmamap_load_mbuf_sg(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, *m_head, txsegs, &nsegs, 0); if (error != 0) { m_freem(*m_head); *m_head = NULL; return (error); } } else if (error != 0) return (error); if (nsegs == 0) { m_freem(*m_head); *m_head = NULL; return (EIO); } /* * Check descriptor overrun. Leave one free descriptor. * Since we always use 64bit address mode for transmitting, * each Tx request requires one more dummy descriptor. */ if (sc->jme_cdata.jme_tx_cnt + nsegs + 1 > JME_TX_RING_CNT - 1) { bus_dmamap_unload(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap); return (ENOBUFS); } m = *m_head; cflags = 0; tso_segsz = 0; /* Configure checksum offload and TSO. */ if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) { tso_segsz = (uint32_t)m->m_pkthdr.tso_segsz << JME_TD_MSS_SHIFT; cflags |= JME_TD_TSO; } else { if ((m->m_pkthdr.csum_flags & CSUM_IP) != 0) cflags |= JME_TD_IPCSUM; if ((m->m_pkthdr.csum_flags & CSUM_TCP) != 0) cflags |= JME_TD_TCPCSUM; if ((m->m_pkthdr.csum_flags & CSUM_UDP) != 0) cflags |= JME_TD_UDPCSUM; } /* Configure VLAN. */ if ((m->m_flags & M_VLANTAG) != 0) { cflags |= (m->m_pkthdr.ether_vtag & JME_TD_VLAN_MASK); cflags |= JME_TD_VLAN_TAG; } desc = &sc->jme_rdata.jme_tx_ring[prod]; desc->flags = htole32(cflags); desc->buflen = htole32(tso_segsz); desc->addr_hi = htole32(m->m_pkthdr.len); desc->addr_lo = 0; sc->jme_cdata.jme_tx_cnt++; JME_DESC_INC(prod, JME_TX_RING_CNT); for (i = 0; i < nsegs; i++) { desc = &sc->jme_rdata.jme_tx_ring[prod]; desc->flags = htole32(JME_TD_OWN | JME_TD_64BIT); desc->buflen = htole32(txsegs[i].ds_len); desc->addr_hi = htole32(JME_ADDR_HI(txsegs[i].ds_addr)); desc->addr_lo = htole32(JME_ADDR_LO(txsegs[i].ds_addr)); sc->jme_cdata.jme_tx_cnt++; JME_DESC_INC(prod, JME_TX_RING_CNT); } /* Update producer index. */ sc->jme_cdata.jme_tx_prod = prod; /* * Finally request interrupt and give the first descriptor * owenership to hardware. */ desc = txd->tx_desc; desc->flags |= htole32(JME_TD_OWN | JME_TD_INTR); txd->tx_m = m; txd->tx_ndesc = nsegs + 1; /* Sync descriptors. */ bus_dmamap_sync(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, BUS_DMASYNC_PREWRITE); bus_dmamap_sync(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } static void jme_start(struct ifnet *ifp) { struct jme_softc *sc; sc = ifp->if_softc; JME_LOCK(sc); jme_start_locked(ifp); JME_UNLOCK(sc); } static void jme_start_locked(struct ifnet *ifp) { struct jme_softc *sc; struct mbuf *m_head; int enq; sc = ifp->if_softc; JME_LOCK_ASSERT(sc); if (sc->jme_cdata.jme_tx_cnt >= JME_TX_DESC_HIWAT) jme_txeof(sc); if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING || (sc->jme_flags & JME_FLAG_LINK) == 0) return; for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) { IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; /* * Pack the data into the transmit ring. If we * don't have room, set the OACTIVE flag and wait * for the NIC to drain the ring. */ if (jme_encap(sc, &m_head)) { if (m_head == NULL) break; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } enq++; /* * If there's a BPF listener, bounce a copy of this frame * to him. */ ETHER_BPF_MTAP(ifp, m_head); } if (enq > 0) { /* * Reading TXCSR takes very long time under heavy load * so cache TXCSR value and writes the ORed value with * the kick command to the TXCSR. This saves one register * access cycle. */ CSR_WRITE_4(sc, JME_TXCSR, sc->jme_txcsr | TXCSR_TX_ENB | TXCSR_TXQ_N_START(TXCSR_TXQ0)); /* Set a timeout in case the chip goes out to lunch. */ sc->jme_watchdog_timer = JME_TX_TIMEOUT; } } static void jme_watchdog(struct jme_softc *sc) { struct ifnet *ifp; JME_LOCK_ASSERT(sc); if (sc->jme_watchdog_timer == 0 || --sc->jme_watchdog_timer) return; ifp = sc->jme_ifp; if ((sc->jme_flags & JME_FLAG_LINK) == 0) { if_printf(sc->jme_ifp, "watchdog timeout (missed link)\n"); ifp->if_oerrors++; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; jme_init_locked(sc); return; } jme_txeof(sc); if (sc->jme_cdata.jme_tx_cnt == 0) { if_printf(sc->jme_ifp, "watchdog timeout (missed Tx interrupts) -- recovering\n"); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) jme_start_locked(ifp); return; } if_printf(sc->jme_ifp, "watchdog timeout\n"); ifp->if_oerrors++; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; jme_init_locked(sc); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) jme_start_locked(ifp); } static int jme_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct jme_softc *sc; struct ifreq *ifr; struct mii_data *mii; uint32_t reg; int error, mask; sc = ifp->if_softc; ifr = (struct ifreq *)data; error = 0; switch (cmd) { case SIOCSIFMTU: if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > JME_JUMBO_MTU || ((sc->jme_flags & JME_FLAG_NOJUMBO) != 0 && ifr->ifr_mtu > JME_MAX_MTU)) { error = EINVAL; break; } if (ifp->if_mtu != ifr->ifr_mtu) { /* * No special configuration is required when interface * MTU is changed but availability of TSO/Tx checksum * offload should be chcked against new MTU size as * FIFO size is just 2K. */ JME_LOCK(sc); if (ifr->ifr_mtu >= JME_TX_FIFO_SIZE) { ifp->if_capenable &= ~(IFCAP_TXCSUM | IFCAP_TSO4); ifp->if_hwassist &= ~(JME_CSUM_FEATURES | CSUM_TSO); VLAN_CAPABILITIES(ifp); } ifp->if_mtu = ifr->ifr_mtu; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; jme_init_locked(sc); } JME_UNLOCK(sc); } break; case SIOCSIFFLAGS: JME_LOCK(sc); if ((ifp->if_flags & IFF_UP) != 0) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if (((ifp->if_flags ^ sc->jme_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) jme_set_filter(sc); } else { if ((sc->jme_flags & JME_FLAG_DETACH) == 0) jme_init_locked(sc); } } else { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) jme_stop(sc); } sc->jme_if_flags = ifp->if_flags; JME_UNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: JME_LOCK(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) jme_set_filter(sc); JME_UNLOCK(sc); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: mii = device_get_softc(sc->jme_miibus); error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd); break; case SIOCSIFCAP: JME_LOCK(sc); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if ((mask & IFCAP_TXCSUM) != 0 && ifp->if_mtu < JME_TX_FIFO_SIZE) { if ((IFCAP_TXCSUM & ifp->if_capabilities) != 0) { ifp->if_capenable ^= IFCAP_TXCSUM; if ((IFCAP_TXCSUM & ifp->if_capenable) != 0) ifp->if_hwassist |= JME_CSUM_FEATURES; else ifp->if_hwassist &= ~JME_CSUM_FEATURES; } } if ((mask & IFCAP_RXCSUM) != 0 && (IFCAP_RXCSUM & ifp->if_capabilities) != 0) { ifp->if_capenable ^= IFCAP_RXCSUM; reg = CSR_READ_4(sc, JME_RXMAC); reg &= ~RXMAC_CSUM_ENB; if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) reg |= RXMAC_CSUM_ENB; CSR_WRITE_4(sc, JME_RXMAC, reg); } if ((mask & IFCAP_TSO4) != 0 && ifp->if_mtu < JME_TX_FIFO_SIZE) { if ((IFCAP_TSO4 & ifp->if_capabilities) != 0) { ifp->if_capenable ^= IFCAP_TSO4; if ((IFCAP_TSO4 & ifp->if_capenable) != 0) ifp->if_hwassist |= CSUM_TSO; else ifp->if_hwassist &= ~CSUM_TSO; } } if ((mask & IFCAP_WOL_MAGIC) != 0 && (IFCAP_WOL_MAGIC & ifp->if_capabilities) != 0) ifp->if_capenable ^= IFCAP_WOL_MAGIC; if ((mask & IFCAP_VLAN_HWCSUM) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWCSUM; if ((mask & IFCAP_VLAN_HWTSO) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if ((mask & IFCAP_VLAN_HWTAGGING) != 0 && (IFCAP_VLAN_HWTAGGING & ifp->if_capabilities) != 0) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; jme_set_vlan(sc); } JME_UNLOCK(sc); VLAN_CAPABILITIES(ifp); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } static void jme_mac_config(struct jme_softc *sc) { struct mii_data *mii; uint32_t ghc, gpreg, rxmac, txmac, txpause; uint32_t txclk; JME_LOCK_ASSERT(sc); mii = device_get_softc(sc->jme_miibus); CSR_WRITE_4(sc, JME_GHC, GHC_RESET); DELAY(10); CSR_WRITE_4(sc, JME_GHC, 0); ghc = 0; txclk = 0; rxmac = CSR_READ_4(sc, JME_RXMAC); rxmac &= ~RXMAC_FC_ENB; txmac = CSR_READ_4(sc, JME_TXMAC); txmac &= ~(TXMAC_CARRIER_EXT | TXMAC_FRAME_BURST); txpause = CSR_READ_4(sc, JME_TXPFC); txpause &= ~TXPFC_PAUSE_ENB; if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) { ghc |= GHC_FULL_DUPLEX; rxmac &= ~RXMAC_COLL_DET_ENB; txmac &= ~(TXMAC_COLL_ENB | TXMAC_CARRIER_SENSE | TXMAC_BACKOFF | TXMAC_CARRIER_EXT | TXMAC_FRAME_BURST); if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0) txpause |= TXPFC_PAUSE_ENB; if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0) rxmac |= RXMAC_FC_ENB; /* Disable retry transmit timer/retry limit. */ CSR_WRITE_4(sc, JME_TXTRHD, CSR_READ_4(sc, JME_TXTRHD) & ~(TXTRHD_RT_PERIOD_ENB | TXTRHD_RT_LIMIT_ENB)); } else { rxmac |= RXMAC_COLL_DET_ENB; txmac |= TXMAC_COLL_ENB | TXMAC_CARRIER_SENSE | TXMAC_BACKOFF; /* Enable retry transmit timer/retry limit. */ CSR_WRITE_4(sc, JME_TXTRHD, CSR_READ_4(sc, JME_TXTRHD) | TXTRHD_RT_PERIOD_ENB | TXTRHD_RT_LIMIT_ENB); } /* Reprogram Tx/Rx MACs with resolved speed/duplex. */ switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: ghc |= GHC_SPEED_10; txclk |= GHC_TX_OFFLD_CLK_100 | GHC_TX_MAC_CLK_100; break; case IFM_100_TX: ghc |= GHC_SPEED_100; txclk |= GHC_TX_OFFLD_CLK_100 | GHC_TX_MAC_CLK_100; break; case IFM_1000_T: if ((sc->jme_flags & JME_FLAG_FASTETH) != 0) break; ghc |= GHC_SPEED_1000; txclk |= GHC_TX_OFFLD_CLK_1000 | GHC_TX_MAC_CLK_1000; if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) == 0) txmac |= TXMAC_CARRIER_EXT | TXMAC_FRAME_BURST; break; default: break; } if (sc->jme_rev == DEVICEID_JMC250 && sc->jme_chip_rev == DEVICEREVID_JMC250_A2) { /* * Workaround occasional packet loss issue of JMC250 A2 * when it runs on half-duplex media. */ gpreg = CSR_READ_4(sc, JME_GPREG1); if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) gpreg &= ~GPREG1_HDPX_FIX; else gpreg |= GPREG1_HDPX_FIX; CSR_WRITE_4(sc, JME_GPREG1, gpreg); /* Workaround CRC errors at 100Mbps on JMC250 A2. */ if (IFM_SUBTYPE(mii->mii_media_active) == IFM_100_TX) { /* Extend interface FIFO depth. */ jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, 0x1B, 0x0000); } else { /* Select default interface FIFO depth. */ jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, 0x1B, 0x0004); } } if ((sc->jme_flags & JME_FLAG_TXCLK) != 0) ghc |= txclk; CSR_WRITE_4(sc, JME_GHC, ghc); CSR_WRITE_4(sc, JME_RXMAC, rxmac); CSR_WRITE_4(sc, JME_TXMAC, txmac); CSR_WRITE_4(sc, JME_TXPFC, txpause); } static void jme_link_task(void *arg, int pending) { struct jme_softc *sc; struct mii_data *mii; struct ifnet *ifp; struct jme_txdesc *txd; bus_addr_t paddr; int i; sc = (struct jme_softc *)arg; JME_LOCK(sc); mii = device_get_softc(sc->jme_miibus); ifp = sc->jme_ifp; if (mii == NULL || ifp == NULL || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { JME_UNLOCK(sc); return; } sc->jme_flags &= ~JME_FLAG_LINK; if ((mii->mii_media_status & IFM_AVALID) != 0) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: sc->jme_flags |= JME_FLAG_LINK; break; case IFM_1000_T: if ((sc->jme_flags & JME_FLAG_FASTETH) != 0) break; sc->jme_flags |= JME_FLAG_LINK; break; default: break; } } /* * Disabling Rx/Tx MACs have a side-effect of resetting * JME_TXNDA/JME_RXNDA register to the first address of * Tx/Rx descriptor address. So driver should reset its * internal procucer/consumer pointer and reclaim any * allocated resources. Note, just saving the value of * JME_TXNDA and JME_RXNDA registers before stopping MAC * and restoring JME_TXNDA/JME_RXNDA register is not * sufficient to make sure correct MAC state because * stopping MAC operation can take a while and hardware * might have updated JME_TXNDA/JME_RXNDA registers * during the stop operation. */ /* Block execution of task. */ taskqueue_block(sc->jme_tq); /* Disable interrupts and stop driver. */ CSR_WRITE_4(sc, JME_INTR_MASK_CLR, JME_INTRS); ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); callout_stop(&sc->jme_tick_ch); sc->jme_watchdog_timer = 0; /* Stop receiver/transmitter. */ jme_stop_rx(sc); jme_stop_tx(sc); /* XXX Drain all queued tasks. */ JME_UNLOCK(sc); taskqueue_drain(sc->jme_tq, &sc->jme_int_task); JME_LOCK(sc); if (sc->jme_cdata.jme_rxhead != NULL) m_freem(sc->jme_cdata.jme_rxhead); JME_RXCHAIN_RESET(sc); jme_txeof(sc); if (sc->jme_cdata.jme_tx_cnt != 0) { /* Remove queued packets for transmit. */ for (i = 0; i < JME_TX_RING_CNT; i++) { txd = &sc->jme_cdata.jme_txdesc[i]; if (txd->tx_m != NULL) { bus_dmamap_sync( sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload( sc->jme_cdata.jme_tx_tag, txd->tx_dmamap); m_freem(txd->tx_m); txd->tx_m = NULL; txd->tx_ndesc = 0; ifp->if_oerrors++; } } } /* * Reuse configured Rx descriptors and reset * producer/consumer index. */ sc->jme_cdata.jme_rx_cons = 0; sc->jme_morework = 0; jme_init_tx_ring(sc); /* Initialize shadow status block. */ jme_init_ssb(sc); /* Program MAC with resolved speed/duplex/flow-control. */ if ((sc->jme_flags & JME_FLAG_LINK) != 0) { jme_mac_config(sc); jme_stats_clear(sc); CSR_WRITE_4(sc, JME_RXCSR, sc->jme_rxcsr); CSR_WRITE_4(sc, JME_TXCSR, sc->jme_txcsr); /* Set Tx ring address to the hardware. */ paddr = JME_TX_RING_ADDR(sc, 0); CSR_WRITE_4(sc, JME_TXDBA_HI, JME_ADDR_HI(paddr)); CSR_WRITE_4(sc, JME_TXDBA_LO, JME_ADDR_LO(paddr)); /* Set Rx ring address to the hardware. */ paddr = JME_RX_RING_ADDR(sc, 0); CSR_WRITE_4(sc, JME_RXDBA_HI, JME_ADDR_HI(paddr)); CSR_WRITE_4(sc, JME_RXDBA_LO, JME_ADDR_LO(paddr)); /* Restart receiver/transmitter. */ CSR_WRITE_4(sc, JME_RXCSR, sc->jme_rxcsr | RXCSR_RX_ENB | RXCSR_RXQ_START); CSR_WRITE_4(sc, JME_TXCSR, sc->jme_txcsr | TXCSR_TX_ENB); /* Lastly enable TX/RX clock. */ if ((sc->jme_flags & JME_FLAG_TXCLK) != 0) CSR_WRITE_4(sc, JME_GHC, CSR_READ_4(sc, JME_GHC) & ~GHC_TX_MAC_CLK_DIS); if ((sc->jme_flags & JME_FLAG_RXCLK) != 0) CSR_WRITE_4(sc, JME_GPREG1, CSR_READ_4(sc, JME_GPREG1) & ~GPREG1_RX_MAC_CLK_DIS); } ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; callout_reset(&sc->jme_tick_ch, hz, jme_tick, sc); /* Unblock execution of task. */ taskqueue_unblock(sc->jme_tq); /* Reenable interrupts. */ CSR_WRITE_4(sc, JME_INTR_MASK_SET, JME_INTRS); JME_UNLOCK(sc); } static int jme_intr(void *arg) { struct jme_softc *sc; uint32_t status; sc = (struct jme_softc *)arg; status = CSR_READ_4(sc, JME_INTR_REQ_STATUS); if (status == 0 || status == 0xFFFFFFFF) return (FILTER_STRAY); /* Disable interrupts. */ CSR_WRITE_4(sc, JME_INTR_MASK_CLR, JME_INTRS); taskqueue_enqueue(sc->jme_tq, &sc->jme_int_task); return (FILTER_HANDLED); } static void jme_int_task(void *arg, int pending) { struct jme_softc *sc; struct ifnet *ifp; uint32_t status; int more; sc = (struct jme_softc *)arg; ifp = sc->jme_ifp; JME_LOCK(sc); status = CSR_READ_4(sc, JME_INTR_STATUS); if (sc->jme_morework != 0) { sc->jme_morework = 0; status |= INTR_RXQ_COAL | INTR_RXQ_COAL_TO; } if ((status & JME_INTRS) == 0 || status == 0xFFFFFFFF) goto done; /* Reset PCC counter/timer and Ack interrupts. */ status &= ~(INTR_TXQ_COMP | INTR_RXQ_COMP); if ((status & (INTR_TXQ_COAL | INTR_TXQ_COAL_TO)) != 0) status |= INTR_TXQ_COAL | INTR_TXQ_COAL_TO | INTR_TXQ_COMP; if ((status & (INTR_RXQ_COAL | INTR_RXQ_COAL_TO)) != 0) status |= INTR_RXQ_COAL | INTR_RXQ_COAL_TO | INTR_RXQ_COMP; CSR_WRITE_4(sc, JME_INTR_STATUS, status); more = 0; if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if ((status & (INTR_RXQ_COAL | INTR_RXQ_COAL_TO)) != 0) { more = jme_rxintr(sc, sc->jme_process_limit); if (more != 0) sc->jme_morework = 1; } if ((status & INTR_RXQ_DESC_EMPTY) != 0) { /* * Notify hardware availability of new Rx * buffers. * Reading RXCSR takes very long time under * heavy load so cache RXCSR value and writes * the ORed value with the kick command to * the RXCSR. This saves one register access * cycle. */ CSR_WRITE_4(sc, JME_RXCSR, sc->jme_rxcsr | RXCSR_RX_ENB | RXCSR_RXQ_START); } if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) jme_start_locked(ifp); } if (more != 0 || (CSR_READ_4(sc, JME_INTR_STATUS) & JME_INTRS) != 0) { taskqueue_enqueue(sc->jme_tq, &sc->jme_int_task); JME_UNLOCK(sc); return; } done: JME_UNLOCK(sc); /* Reenable interrupts. */ CSR_WRITE_4(sc, JME_INTR_MASK_SET, JME_INTRS); } static void jme_txeof(struct jme_softc *sc) { struct ifnet *ifp; struct jme_txdesc *txd; uint32_t status; int cons, nsegs; JME_LOCK_ASSERT(sc); ifp = sc->jme_ifp; cons = sc->jme_cdata.jme_tx_cons; if (cons == sc->jme_cdata.jme_tx_prod) return; bus_dmamap_sync(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * Go through our Tx list and free mbufs for those * frames which have been transmitted. */ for (; cons != sc->jme_cdata.jme_tx_prod;) { txd = &sc->jme_cdata.jme_txdesc[cons]; status = le32toh(txd->tx_desc->flags); if ((status & JME_TD_OWN) == JME_TD_OWN) break; if ((status & (JME_TD_TMOUT | JME_TD_RETRY_EXP)) != 0) ifp->if_oerrors++; else { ifp->if_opackets++; if ((status & JME_TD_COLLISION) != 0) ifp->if_collisions += le32toh(txd->tx_desc->buflen) & JME_TD_BUF_LEN_MASK; } /* * Only the first descriptor of multi-descriptor * transmission is updated so driver have to skip entire * chained buffers for the transmiited frame. In other * words, JME_TD_OWN bit is valid only at the first * descriptor of a multi-descriptor transmission. */ for (nsegs = 0; nsegs < txd->tx_ndesc; nsegs++) { sc->jme_rdata.jme_tx_ring[cons].flags = 0; JME_DESC_INC(cons, JME_TX_RING_CNT); } /* Reclaim transferred mbufs. */ bus_dmamap_sync(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap); KASSERT(txd->tx_m != NULL, ("%s: freeing NULL mbuf!\n", __func__)); m_freem(txd->tx_m); txd->tx_m = NULL; sc->jme_cdata.jme_tx_cnt -= txd->tx_ndesc; KASSERT(sc->jme_cdata.jme_tx_cnt >= 0, ("%s: Active Tx desc counter was garbled\n", __func__)); txd->tx_ndesc = 0; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } sc->jme_cdata.jme_tx_cons = cons; /* Unarm watchog timer when there is no pending descriptors in queue. */ if (sc->jme_cdata.jme_tx_cnt == 0) sc->jme_watchdog_timer = 0; bus_dmamap_sync(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static __inline void jme_discard_rxbuf(struct jme_softc *sc, int cons) { struct jme_desc *desc; desc = &sc->jme_rdata.jme_rx_ring[cons]; desc->flags = htole32(JME_RD_OWN | JME_RD_INTR | JME_RD_64BIT); desc->buflen = htole32(MCLBYTES); } /* Receive a frame. */ static void jme_rxeof(struct jme_softc *sc) { struct ifnet *ifp; struct jme_desc *desc; struct jme_rxdesc *rxd; struct mbuf *mp, *m; uint32_t flags, status; int cons, count, nsegs; JME_LOCK_ASSERT(sc); ifp = sc->jme_ifp; cons = sc->jme_cdata.jme_rx_cons; desc = &sc->jme_rdata.jme_rx_ring[cons]; flags = le32toh(desc->flags); status = le32toh(desc->buflen); nsegs = JME_RX_NSEGS(status); sc->jme_cdata.jme_rxlen = JME_RX_BYTES(status) - JME_RX_PAD_BYTES; if ((status & JME_RX_ERR_STAT) != 0) { ifp->if_ierrors++; jme_discard_rxbuf(sc, sc->jme_cdata.jme_rx_cons); #ifdef JME_SHOW_ERRORS device_printf(sc->jme_dev, "%s : receive error = 0x%b\n", __func__, JME_RX_ERR(status), JME_RX_ERR_BITS); #endif sc->jme_cdata.jme_rx_cons += nsegs; sc->jme_cdata.jme_rx_cons %= JME_RX_RING_CNT; return; } for (count = 0; count < nsegs; count++, JME_DESC_INC(cons, JME_RX_RING_CNT)) { rxd = &sc->jme_cdata.jme_rxdesc[cons]; mp = rxd->rx_m; /* Add a new receive buffer to the ring. */ if (jme_newbuf(sc, rxd) != 0) { ifp->if_iqdrops++; /* Reuse buffer. */ for (; count < nsegs; count++) { jme_discard_rxbuf(sc, cons); JME_DESC_INC(cons, JME_RX_RING_CNT); } if (sc->jme_cdata.jme_rxhead != NULL) { m_freem(sc->jme_cdata.jme_rxhead); JME_RXCHAIN_RESET(sc); } break; } /* * Assume we've received a full sized frame. * Actual size is fixed when we encounter the end of * multi-segmented frame. */ mp->m_len = MCLBYTES; /* Chain received mbufs. */ if (sc->jme_cdata.jme_rxhead == NULL) { sc->jme_cdata.jme_rxhead = mp; sc->jme_cdata.jme_rxtail = mp; } else { /* * Receive processor can receive a maximum frame * size of 65535 bytes. */ mp->m_flags &= ~M_PKTHDR; sc->jme_cdata.jme_rxtail->m_next = mp; sc->jme_cdata.jme_rxtail = mp; } if (count == nsegs - 1) { /* Last desc. for this frame. */ m = sc->jme_cdata.jme_rxhead; m->m_flags |= M_PKTHDR; m->m_pkthdr.len = sc->jme_cdata.jme_rxlen; if (nsegs > 1) { /* Set first mbuf size. */ m->m_len = MCLBYTES - JME_RX_PAD_BYTES; /* Set last mbuf size. */ mp->m_len = sc->jme_cdata.jme_rxlen - ((MCLBYTES - JME_RX_PAD_BYTES) + (MCLBYTES * (nsegs - 2))); } else m->m_len = sc->jme_cdata.jme_rxlen; m->m_pkthdr.rcvif = ifp; /* * Account for 10bytes auto padding which is used * to align IP header on 32bit boundary. Also note, * CRC bytes is automatically removed by the * hardware. */ m->m_data += JME_RX_PAD_BYTES; /* Set checksum information. */ if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 && (flags & JME_RD_IPV4) != 0) { m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED; if ((flags & JME_RD_IPCSUM) != 0) m->m_pkthdr.csum_flags |= CSUM_IP_VALID; if (((flags & JME_RD_MORE_FRAG) == 0) && ((flags & (JME_RD_TCP | JME_RD_TCPCSUM)) == (JME_RD_TCP | JME_RD_TCPCSUM) || (flags & (JME_RD_UDP | JME_RD_UDPCSUM)) == (JME_RD_UDP | JME_RD_UDPCSUM))) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } } /* Check for VLAN tagged packets. */ if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 && (flags & JME_RD_VLAN_TAG) != 0) { m->m_pkthdr.ether_vtag = flags & JME_RD_VLAN_MASK; m->m_flags |= M_VLANTAG; } ifp->if_ipackets++; /* Pass it on. */ JME_UNLOCK(sc); (*ifp->if_input)(ifp, m); JME_LOCK(sc); /* Reset mbuf chains. */ JME_RXCHAIN_RESET(sc); } } sc->jme_cdata.jme_rx_cons += nsegs; sc->jme_cdata.jme_rx_cons %= JME_RX_RING_CNT; } static int jme_rxintr(struct jme_softc *sc, int count) { struct jme_desc *desc; int nsegs, prog, pktlen; bus_dmamap_sync(sc->jme_cdata.jme_rx_ring_tag, sc->jme_cdata.jme_rx_ring_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); for (prog = 0; count > 0; prog++) { desc = &sc->jme_rdata.jme_rx_ring[sc->jme_cdata.jme_rx_cons]; if ((le32toh(desc->flags) & JME_RD_OWN) == JME_RD_OWN) break; if ((le32toh(desc->buflen) & JME_RD_VALID) == 0) break; nsegs = JME_RX_NSEGS(le32toh(desc->buflen)); /* * Check number of segments against received bytes. * Non-matching value would indicate that hardware * is still trying to update Rx descriptors. I'm not * sure whether this check is needed. */ pktlen = JME_RX_BYTES(le32toh(desc->buflen)); if (nsegs != ((pktlen + (MCLBYTES - 1)) / MCLBYTES)) break; prog++; /* Received a frame. */ jme_rxeof(sc); count -= nsegs; } if (prog > 0) bus_dmamap_sync(sc->jme_cdata.jme_rx_ring_tag, sc->jme_cdata.jme_rx_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (count > 0 ? 0 : EAGAIN); } static void jme_tick(void *arg) { struct jme_softc *sc; struct mii_data *mii; sc = (struct jme_softc *)arg; JME_LOCK_ASSERT(sc); mii = device_get_softc(sc->jme_miibus); mii_tick(mii); /* * Reclaim Tx buffers that have been completed. It's not * needed here but it would release allocated mbuf chains * faster and limit the maximum delay to a hz. */ jme_txeof(sc); jme_stats_update(sc); jme_watchdog(sc); callout_reset(&sc->jme_tick_ch, hz, jme_tick, sc); } static void jme_reset(struct jme_softc *sc) { uint32_t ghc, gpreg; /* Stop receiver, transmitter. */ jme_stop_rx(sc); jme_stop_tx(sc); /* Reset controller. */ CSR_WRITE_4(sc, JME_GHC, GHC_RESET); CSR_READ_4(sc, JME_GHC); DELAY(10); /* * Workaround Rx FIFO overruns seen under certain conditions. * Explicitly synchorize TX/RX clock. TX/RX clock should be * enabled only after enabling TX/RX MACs. */ if ((sc->jme_flags & (JME_FLAG_TXCLK | JME_FLAG_RXCLK)) != 0) { /* Disable TX clock. */ CSR_WRITE_4(sc, JME_GHC, GHC_RESET | GHC_TX_MAC_CLK_DIS); /* Disable RX clock. */ gpreg = CSR_READ_4(sc, JME_GPREG1); CSR_WRITE_4(sc, JME_GPREG1, gpreg | GPREG1_RX_MAC_CLK_DIS); gpreg = CSR_READ_4(sc, JME_GPREG1); /* De-assert RESET but still disable TX clock. */ CSR_WRITE_4(sc, JME_GHC, GHC_TX_MAC_CLK_DIS); ghc = CSR_READ_4(sc, JME_GHC); /* Enable TX clock. */ CSR_WRITE_4(sc, JME_GHC, ghc & ~GHC_TX_MAC_CLK_DIS); /* Enable RX clock. */ CSR_WRITE_4(sc, JME_GPREG1, gpreg & ~GPREG1_RX_MAC_CLK_DIS); CSR_READ_4(sc, JME_GPREG1); /* Disable TX/RX clock again. */ CSR_WRITE_4(sc, JME_GHC, GHC_TX_MAC_CLK_DIS); CSR_WRITE_4(sc, JME_GPREG1, gpreg | GPREG1_RX_MAC_CLK_DIS); } else CSR_WRITE_4(sc, JME_GHC, 0); CSR_READ_4(sc, JME_GHC); DELAY(10); } static void jme_init(void *xsc) { struct jme_softc *sc; sc = (struct jme_softc *)xsc; JME_LOCK(sc); jme_init_locked(sc); JME_UNLOCK(sc); } static void jme_init_locked(struct jme_softc *sc) { struct ifnet *ifp; struct mii_data *mii; bus_addr_t paddr; uint32_t reg; int error; JME_LOCK_ASSERT(sc); ifp = sc->jme_ifp; mii = device_get_softc(sc->jme_miibus); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) return; /* * Cancel any pending I/O. */ jme_stop(sc); /* * Reset the chip to a known state. */ jme_reset(sc); /* Init descriptors. */ error = jme_init_rx_ring(sc); if (error != 0) { device_printf(sc->jme_dev, "%s: initialization failed: no memory for Rx buffers.\n", __func__); jme_stop(sc); return; } jme_init_tx_ring(sc); /* Initialize shadow status block. */ jme_init_ssb(sc); /* Reprogram the station address. */ jme_set_macaddr(sc, IF_LLADDR(sc->jme_ifp)); /* * Configure Tx queue. * Tx priority queue weight value : 0 * Tx FIFO threshold for processing next packet : 16QW * Maximum Tx DMA length : 512 * Allow Tx DMA burst. */ sc->jme_txcsr = TXCSR_TXQ_N_SEL(TXCSR_TXQ0); sc->jme_txcsr |= TXCSR_TXQ_WEIGHT(TXCSR_TXQ_WEIGHT_MIN); sc->jme_txcsr |= TXCSR_FIFO_THRESH_16QW; sc->jme_txcsr |= sc->jme_tx_dma_size; sc->jme_txcsr |= TXCSR_DMA_BURST; CSR_WRITE_4(sc, JME_TXCSR, sc->jme_txcsr); /* Set Tx descriptor counter. */ CSR_WRITE_4(sc, JME_TXQDC, JME_TX_RING_CNT); /* Set Tx ring address to the hardware. */ paddr = JME_TX_RING_ADDR(sc, 0); CSR_WRITE_4(sc, JME_TXDBA_HI, JME_ADDR_HI(paddr)); CSR_WRITE_4(sc, JME_TXDBA_LO, JME_ADDR_LO(paddr)); /* Configure TxMAC parameters. */ reg = TXMAC_IFG1_DEFAULT | TXMAC_IFG2_DEFAULT | TXMAC_IFG_ENB; reg |= TXMAC_THRESH_1_PKT; reg |= TXMAC_CRC_ENB | TXMAC_PAD_ENB; CSR_WRITE_4(sc, JME_TXMAC, reg); /* * Configure Rx queue. * FIFO full threshold for transmitting Tx pause packet : 128T * FIFO threshold for processing next packet : 128QW * Rx queue 0 select * Max Rx DMA length : 128 * Rx descriptor retry : 32 * Rx descriptor retry time gap : 256ns * Don't receive runt/bad frame. */ sc->jme_rxcsr = RXCSR_FIFO_FTHRESH_128T; /* * Since Rx FIFO size is 4K bytes, receiving frames larger * than 4K bytes will suffer from Rx FIFO overruns. So * decrease FIFO threshold to reduce the FIFO overruns for * frames larger than 4000 bytes. * For best performance of standard MTU sized frames use * maximum allowable FIFO threshold, 128QW. Note these do * not hold on chip full mask verion >=2. For these * controllers 64QW and 128QW are not valid value. */ if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 2) sc->jme_rxcsr |= RXCSR_FIFO_THRESH_16QW; else { if ((ifp->if_mtu + ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN + ETHER_CRC_LEN) > JME_RX_FIFO_SIZE) sc->jme_rxcsr |= RXCSR_FIFO_THRESH_16QW; else sc->jme_rxcsr |= RXCSR_FIFO_THRESH_128QW; } sc->jme_rxcsr |= sc->jme_rx_dma_size | RXCSR_RXQ_N_SEL(RXCSR_RXQ0); sc->jme_rxcsr |= RXCSR_DESC_RT_CNT(RXCSR_DESC_RT_CNT_DEFAULT); sc->jme_rxcsr |= RXCSR_DESC_RT_GAP_256 & RXCSR_DESC_RT_GAP_MASK; CSR_WRITE_4(sc, JME_RXCSR, sc->jme_rxcsr); /* Set Rx descriptor counter. */ CSR_WRITE_4(sc, JME_RXQDC, JME_RX_RING_CNT); /* Set Rx ring address to the hardware. */ paddr = JME_RX_RING_ADDR(sc, 0); CSR_WRITE_4(sc, JME_RXDBA_HI, JME_ADDR_HI(paddr)); CSR_WRITE_4(sc, JME_RXDBA_LO, JME_ADDR_LO(paddr)); /* Clear receive filter. */ CSR_WRITE_4(sc, JME_RXMAC, 0); /* Set up the receive filter. */ jme_set_filter(sc); jme_set_vlan(sc); /* * Disable all WOL bits as WOL can interfere normal Rx * operation. Also clear WOL detection status bits. */ reg = CSR_READ_4(sc, JME_PMCS); reg &= ~PMCS_WOL_ENB_MASK; CSR_WRITE_4(sc, JME_PMCS, reg); reg = CSR_READ_4(sc, JME_RXMAC); /* * Pad 10bytes right before received frame. This will greatly * help Rx performance on strict-alignment architectures as * it does not need to copy the frame to align the payload. */ reg |= RXMAC_PAD_10BYTES; if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) reg |= RXMAC_CSUM_ENB; CSR_WRITE_4(sc, JME_RXMAC, reg); /* Configure general purpose reg0 */ reg = CSR_READ_4(sc, JME_GPREG0); reg &= ~GPREG0_PCC_UNIT_MASK; /* Set PCC timer resolution to micro-seconds unit. */ reg |= GPREG0_PCC_UNIT_US; /* * Disable all shadow register posting as we have to read * JME_INTR_STATUS register in jme_int_task. Also it seems * that it's hard to synchronize interrupt status between * hardware and software with shadow posting due to * requirements of bus_dmamap_sync(9). */ reg |= GPREG0_SH_POST_DW7_DIS | GPREG0_SH_POST_DW6_DIS | GPREG0_SH_POST_DW5_DIS | GPREG0_SH_POST_DW4_DIS | GPREG0_SH_POST_DW3_DIS | GPREG0_SH_POST_DW2_DIS | GPREG0_SH_POST_DW1_DIS | GPREG0_SH_POST_DW0_DIS; /* Disable posting of DW0. */ reg &= ~GPREG0_POST_DW0_ENB; /* Clear PME message. */ reg &= ~GPREG0_PME_ENB; /* Set PHY address. */ reg &= ~GPREG0_PHY_ADDR_MASK; reg |= sc->jme_phyaddr; CSR_WRITE_4(sc, JME_GPREG0, reg); /* Configure Tx queue 0 packet completion coalescing. */ reg = (sc->jme_tx_coal_to << PCCTX_COAL_TO_SHIFT) & PCCTX_COAL_TO_MASK; reg |= (sc->jme_tx_coal_pkt << PCCTX_COAL_PKT_SHIFT) & PCCTX_COAL_PKT_MASK; reg |= PCCTX_COAL_TXQ0; CSR_WRITE_4(sc, JME_PCCTX, reg); /* Configure Rx queue 0 packet completion coalescing. */ reg = (sc->jme_rx_coal_to << PCCRX_COAL_TO_SHIFT) & PCCRX_COAL_TO_MASK; reg |= (sc->jme_rx_coal_pkt << PCCRX_COAL_PKT_SHIFT) & PCCRX_COAL_PKT_MASK; CSR_WRITE_4(sc, JME_PCCRX0, reg); /* * Configure PCD(Packet Completion Deferring). It seems PCD * generates an interrupt when the time interval between two * back-to-back incoming/outgoing packet is long enough for * it to reach its timer value 0. The arrival of new packets * after timer has started causes the PCD timer to restart. * Unfortunately, it's not clear how PCD is useful at this * moment, so just use the same of PCC parameters. */ if ((sc->jme_flags & JME_FLAG_PCCPCD) != 0) { sc->jme_rx_pcd_to = sc->jme_rx_coal_to; if (sc->jme_rx_coal_to > PCDRX_TO_MAX) sc->jme_rx_pcd_to = PCDRX_TO_MAX; sc->jme_tx_pcd_to = sc->jme_tx_coal_to; if (sc->jme_tx_coal_to > PCDTX_TO_MAX) sc->jme_tx_pcd_to = PCDTX_TO_MAX; reg = sc->jme_rx_pcd_to << PCDRX0_TO_THROTTLE_SHIFT; reg |= sc->jme_rx_pcd_to << PCDRX0_TO_SHIFT; CSR_WRITE_4(sc, PCDRX_REG(0), reg); reg = sc->jme_tx_pcd_to << PCDTX_TO_THROTTLE_SHIFT; reg |= sc->jme_tx_pcd_to << PCDTX_TO_SHIFT; CSR_WRITE_4(sc, JME_PCDTX, reg); } /* Configure shadow status block but don't enable posting. */ paddr = sc->jme_rdata.jme_ssb_block_paddr; CSR_WRITE_4(sc, JME_SHBASE_ADDR_HI, JME_ADDR_HI(paddr)); CSR_WRITE_4(sc, JME_SHBASE_ADDR_LO, JME_ADDR_LO(paddr)); /* Disable Timer 1 and Timer 2. */ CSR_WRITE_4(sc, JME_TIMER1, 0); CSR_WRITE_4(sc, JME_TIMER2, 0); /* Configure retry transmit period, retry limit value. */ CSR_WRITE_4(sc, JME_TXTRHD, ((TXTRHD_RT_PERIOD_DEFAULT << TXTRHD_RT_PERIOD_SHIFT) & TXTRHD_RT_PERIOD_MASK) | ((TXTRHD_RT_LIMIT_DEFAULT << TXTRHD_RT_LIMIT_SHIFT) & TXTRHD_RT_LIMIT_SHIFT)); /* Disable RSS. */ CSR_WRITE_4(sc, JME_RSSC, RSSC_DIS_RSS); /* Initialize the interrupt mask. */ CSR_WRITE_4(sc, JME_INTR_MASK_SET, JME_INTRS); CSR_WRITE_4(sc, JME_INTR_STATUS, 0xFFFFFFFF); /* * Enabling Tx/Rx DMA engines and Rx queue processing is * done after detection of valid link in jme_link_task. */ sc->jme_flags &= ~JME_FLAG_LINK; /* Set the current media. */ mii_mediachg(mii); callout_reset(&sc->jme_tick_ch, hz, jme_tick, sc); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } static void jme_stop(struct jme_softc *sc) { struct ifnet *ifp; struct jme_txdesc *txd; struct jme_rxdesc *rxd; int i; JME_LOCK_ASSERT(sc); /* * Mark the interface down and cancel the watchdog timer. */ ifp = sc->jme_ifp; ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); sc->jme_flags &= ~JME_FLAG_LINK; callout_stop(&sc->jme_tick_ch); sc->jme_watchdog_timer = 0; /* * Disable interrupts. */ CSR_WRITE_4(sc, JME_INTR_MASK_CLR, JME_INTRS); CSR_WRITE_4(sc, JME_INTR_STATUS, 0xFFFFFFFF); /* Disable updating shadow status block. */ CSR_WRITE_4(sc, JME_SHBASE_ADDR_LO, CSR_READ_4(sc, JME_SHBASE_ADDR_LO) & ~SHBASE_POST_ENB); /* Stop receiver, transmitter. */ jme_stop_rx(sc); jme_stop_tx(sc); /* Reclaim Rx/Tx buffers that have been completed. */ jme_rxintr(sc, JME_RX_RING_CNT); if (sc->jme_cdata.jme_rxhead != NULL) m_freem(sc->jme_cdata.jme_rxhead); JME_RXCHAIN_RESET(sc); jme_txeof(sc); /* * Free RX and TX mbufs still in the queues. */ for (i = 0; i < JME_RX_RING_CNT; i++) { rxd = &sc->jme_cdata.jme_rxdesc[i]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap); m_freem(rxd->rx_m); rxd->rx_m = NULL; } } for (i = 0; i < JME_TX_RING_CNT; i++) { txd = &sc->jme_cdata.jme_txdesc[i]; if (txd->tx_m != NULL) { bus_dmamap_sync(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->jme_cdata.jme_tx_tag, txd->tx_dmamap); m_freem(txd->tx_m); txd->tx_m = NULL; txd->tx_ndesc = 0; } } jme_stats_update(sc); jme_stats_save(sc); } static void jme_stop_tx(struct jme_softc *sc) { uint32_t reg; int i; reg = CSR_READ_4(sc, JME_TXCSR); if ((reg & TXCSR_TX_ENB) == 0) return; reg &= ~TXCSR_TX_ENB; CSR_WRITE_4(sc, JME_TXCSR, reg); for (i = JME_TIMEOUT; i > 0; i--) { DELAY(1); if ((CSR_READ_4(sc, JME_TXCSR) & TXCSR_TX_ENB) == 0) break; } if (i == 0) device_printf(sc->jme_dev, "stopping transmitter timeout!\n"); } static void jme_stop_rx(struct jme_softc *sc) { uint32_t reg; int i; reg = CSR_READ_4(sc, JME_RXCSR); if ((reg & RXCSR_RX_ENB) == 0) return; reg &= ~RXCSR_RX_ENB; CSR_WRITE_4(sc, JME_RXCSR, reg); for (i = JME_TIMEOUT; i > 0; i--) { DELAY(1); if ((CSR_READ_4(sc, JME_RXCSR) & RXCSR_RX_ENB) == 0) break; } if (i == 0) device_printf(sc->jme_dev, "stopping recevier timeout!\n"); } static void jme_init_tx_ring(struct jme_softc *sc) { struct jme_ring_data *rd; struct jme_txdesc *txd; int i; sc->jme_cdata.jme_tx_prod = 0; sc->jme_cdata.jme_tx_cons = 0; sc->jme_cdata.jme_tx_cnt = 0; rd = &sc->jme_rdata; bzero(rd->jme_tx_ring, JME_TX_RING_SIZE); for (i = 0; i < JME_TX_RING_CNT; i++) { txd = &sc->jme_cdata.jme_txdesc[i]; txd->tx_m = NULL; txd->tx_desc = &rd->jme_tx_ring[i]; txd->tx_ndesc = 0; } bus_dmamap_sync(sc->jme_cdata.jme_tx_ring_tag, sc->jme_cdata.jme_tx_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static void jme_init_ssb(struct jme_softc *sc) { struct jme_ring_data *rd; rd = &sc->jme_rdata; bzero(rd->jme_ssb_block, JME_SSB_SIZE); bus_dmamap_sync(sc->jme_cdata.jme_ssb_tag, sc->jme_cdata.jme_ssb_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } static int jme_init_rx_ring(struct jme_softc *sc) { struct jme_ring_data *rd; struct jme_rxdesc *rxd; int i; sc->jme_cdata.jme_rx_cons = 0; JME_RXCHAIN_RESET(sc); sc->jme_morework = 0; rd = &sc->jme_rdata; bzero(rd->jme_rx_ring, JME_RX_RING_SIZE); for (i = 0; i < JME_RX_RING_CNT; i++) { rxd = &sc->jme_cdata.jme_rxdesc[i]; rxd->rx_m = NULL; rxd->rx_desc = &rd->jme_rx_ring[i]; if (jme_newbuf(sc, rxd) != 0) return (ENOBUFS); } bus_dmamap_sync(sc->jme_cdata.jme_rx_ring_tag, sc->jme_cdata.jme_rx_ring_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } static int jme_newbuf(struct jme_softc *sc, struct jme_rxdesc *rxd) { struct jme_desc *desc; struct mbuf *m; bus_dma_segment_t segs[1]; bus_dmamap_t map; int nsegs; m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); if (m == NULL) return (ENOBUFS); /* * JMC250 has 64bit boundary alignment limitation so jme(4) * takes advantage of 10 bytes padding feature of hardware * in order not to copy entire frame to align IP header on * 32bit boundary. */ m->m_len = m->m_pkthdr.len = MCLBYTES; if (bus_dmamap_load_mbuf_sg(sc->jme_cdata.jme_rx_tag, sc->jme_cdata.jme_rx_sparemap, m, segs, &nsegs, 0) != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs)); if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap); } map = rxd->rx_dmamap; rxd->rx_dmamap = sc->jme_cdata.jme_rx_sparemap; sc->jme_cdata.jme_rx_sparemap = map; bus_dmamap_sync(sc->jme_cdata.jme_rx_tag, rxd->rx_dmamap, BUS_DMASYNC_PREREAD); rxd->rx_m = m; desc = rxd->rx_desc; desc->buflen = htole32(segs[0].ds_len); desc->addr_lo = htole32(JME_ADDR_LO(segs[0].ds_addr)); desc->addr_hi = htole32(JME_ADDR_HI(segs[0].ds_addr)); desc->flags = htole32(JME_RD_OWN | JME_RD_INTR | JME_RD_64BIT); return (0); } static void jme_set_vlan(struct jme_softc *sc) { struct ifnet *ifp; uint32_t reg; JME_LOCK_ASSERT(sc); ifp = sc->jme_ifp; reg = CSR_READ_4(sc, JME_RXMAC); reg &= ~RXMAC_VLAN_ENB; if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) reg |= RXMAC_VLAN_ENB; CSR_WRITE_4(sc, JME_RXMAC, reg); } static void jme_set_filter(struct jme_softc *sc) { struct ifnet *ifp; struct ifmultiaddr *ifma; uint32_t crc; uint32_t mchash[2]; uint32_t rxcfg; JME_LOCK_ASSERT(sc); ifp = sc->jme_ifp; rxcfg = CSR_READ_4(sc, JME_RXMAC); rxcfg &= ~ (RXMAC_BROADCAST | RXMAC_PROMISC | RXMAC_MULTICAST | RXMAC_ALLMULTI); /* Always accept frames destined to our station address. */ rxcfg |= RXMAC_UNICAST; if ((ifp->if_flags & IFF_BROADCAST) != 0) rxcfg |= RXMAC_BROADCAST; if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) { if ((ifp->if_flags & IFF_PROMISC) != 0) rxcfg |= RXMAC_PROMISC; if ((ifp->if_flags & IFF_ALLMULTI) != 0) rxcfg |= RXMAC_ALLMULTI; CSR_WRITE_4(sc, JME_MAR0, 0xFFFFFFFF); CSR_WRITE_4(sc, JME_MAR1, 0xFFFFFFFF); CSR_WRITE_4(sc, JME_RXMAC, rxcfg); return; } /* * Set up the multicast address filter by passing all multicast * addresses through a CRC generator, and then using the low-order * 6 bits as an index into the 64 bit multicast hash table. The * high order bits select the register, while the rest of the bits * select the bit within the register. */ rxcfg |= RXMAC_MULTICAST; bzero(mchash, sizeof(mchash)); IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &sc->jme_ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; crc = ether_crc32_be(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), ETHER_ADDR_LEN); /* Just want the 6 least significant bits. */ crc &= 0x3f; /* Set the corresponding bit in the hash table. */ mchash[crc >> 5] |= 1 << (crc & 0x1f); } IF_ADDR_UNLOCK(ifp); CSR_WRITE_4(sc, JME_MAR0, mchash[0]); CSR_WRITE_4(sc, JME_MAR1, mchash[1]); CSR_WRITE_4(sc, JME_RXMAC, rxcfg); } static void jme_stats_clear(struct jme_softc *sc) { JME_LOCK_ASSERT(sc); if ((sc->jme_flags & JME_FLAG_HWMIB) == 0) return; /* Disable and clear counters. */ CSR_WRITE_4(sc, JME_STATCSR, 0xFFFFFFFF); /* Activate hw counters. */ CSR_WRITE_4(sc, JME_STATCSR, 0); CSR_READ_4(sc, JME_STATCSR); bzero(&sc->jme_stats, sizeof(struct jme_hw_stats)); } static void jme_stats_save(struct jme_softc *sc) { JME_LOCK_ASSERT(sc); if ((sc->jme_flags & JME_FLAG_HWMIB) == 0) return; /* Save current counters. */ bcopy(&sc->jme_stats, &sc->jme_ostats, sizeof(struct jme_hw_stats)); /* Disable and clear counters. */ CSR_WRITE_4(sc, JME_STATCSR, 0xFFFFFFFF); } static void jme_stats_update(struct jme_softc *sc) { struct jme_hw_stats *stat, *ostat; uint32_t reg; JME_LOCK_ASSERT(sc); if ((sc->jme_flags & JME_FLAG_HWMIB) == 0) return; stat = &sc->jme_stats; ostat = &sc->jme_ostats; stat->tx_good_frames = CSR_READ_4(sc, JME_STAT_TXGOOD); stat->rx_good_frames = CSR_READ_4(sc, JME_STAT_RXGOOD); reg = CSR_READ_4(sc, JME_STAT_CRCMII); stat->rx_crc_errs = (reg & STAT_RX_CRC_ERR_MASK) >> STAT_RX_CRC_ERR_SHIFT; stat->rx_mii_errs = (reg & STAT_RX_MII_ERR_MASK) >> STAT_RX_MII_ERR_SHIFT; reg = CSR_READ_4(sc, JME_STAT_RXERR); stat->rx_fifo_oflows = (reg & STAT_RXERR_OFLOW_MASK) >> STAT_RXERR_OFLOW_SHIFT; stat->rx_desc_empty = (reg & STAT_RXERR_MPTY_MASK) >> STAT_RXERR_MPTY_SHIFT; reg = CSR_READ_4(sc, JME_STAT_FAIL); stat->rx_bad_frames = (reg & STAT_FAIL_RX_MASK) >> STAT_FAIL_RX_SHIFT; stat->tx_bad_frames = (reg & STAT_FAIL_TX_MASK) >> STAT_FAIL_TX_SHIFT; /* Account for previous counters. */ stat->rx_good_frames += ostat->rx_good_frames; stat->rx_crc_errs += ostat->rx_crc_errs; stat->rx_mii_errs += ostat->rx_mii_errs; stat->rx_fifo_oflows += ostat->rx_fifo_oflows; stat->rx_desc_empty += ostat->rx_desc_empty; stat->rx_bad_frames += ostat->rx_bad_frames; stat->tx_good_frames += ostat->tx_good_frames; stat->tx_bad_frames += ostat->tx_bad_frames; } static void jme_phy_down(struct jme_softc *sc) { uint32_t reg; jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, MII_BMCR, BMCR_PDOWN); if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 5) { reg = CSR_READ_4(sc, JME_PHYPOWDN); reg |= 0x0000000F; CSR_WRITE_4(sc, JME_PHYPOWDN, reg); reg = pci_read_config(sc->jme_dev, JME_PCI_PE1, 4); reg &= ~PE1_GIGA_PDOWN_MASK; reg |= PE1_GIGA_PDOWN_D3; pci_write_config(sc->jme_dev, JME_PCI_PE1, reg, 4); } } static void jme_phy_up(struct jme_softc *sc) { uint32_t reg; uint16_t bmcr; bmcr = jme_miibus_readreg(sc->jme_dev, sc->jme_phyaddr, MII_BMCR); bmcr &= ~BMCR_PDOWN; jme_miibus_writereg(sc->jme_dev, sc->jme_phyaddr, MII_BMCR, bmcr); if (CHIPMODE_REVFM(sc->jme_chip_rev) >= 5) { reg = CSR_READ_4(sc, JME_PHYPOWDN); reg &= ~0x0000000F; CSR_WRITE_4(sc, JME_PHYPOWDN, reg); reg = pci_read_config(sc->jme_dev, JME_PCI_PE1, 4); reg &= ~PE1_GIGA_PDOWN_MASK; reg |= PE1_GIGA_PDOWN_DIS; pci_write_config(sc->jme_dev, JME_PCI_PE1, reg, 4); } } static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high) { int error, value; if (arg1 == NULL) return (EINVAL); value = *(int *)arg1; error = sysctl_handle_int(oidp, &value, 0, req); if (error || req->newptr == NULL) return (error); if (value < low || value > high) return (EINVAL); *(int *)arg1 = value; return (0); } static int sysctl_hw_jme_tx_coal_to(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, PCCTX_COAL_TO_MIN, PCCTX_COAL_TO_MAX)); } static int sysctl_hw_jme_tx_coal_pkt(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, PCCTX_COAL_PKT_MIN, PCCTX_COAL_PKT_MAX)); } static int sysctl_hw_jme_rx_coal_to(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, PCCRX_COAL_TO_MIN, PCCRX_COAL_TO_MAX)); } static int sysctl_hw_jme_rx_coal_pkt(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, PCCRX_COAL_PKT_MIN, PCCRX_COAL_PKT_MAX)); } static int sysctl_hw_jme_proc_limit(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, JME_PROC_MIN, JME_PROC_MAX)); } Index: stable/7/sys/dev/pci/pci.c =================================================================== --- stable/7/sys/dev/pci/pci.c (revision 242221) +++ stable/7/sys/dev/pci/pci.c (revision 242222) @@ -1,3956 +1,3956 @@ /*- * Copyright (c) 1997, Stefan Esser * Copyright (c) 2000, Michael Smith * Copyright (c) 2000, BSDi * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_bus.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(__i386__) || defined(__amd64__) #include #endif #include #include #include #include #include "pcib_if.h" #include "pci_if.h" #ifdef __HAVE_ACPI #include #include "acpi_if.h" #else #define ACPI_PWR_FOR_SLEEP(x, y, z) #endif static pci_addr_t pci_mapbase(uint64_t mapreg); static const char *pci_maptype(uint64_t mapreg); static int pci_mapsize(uint64_t testval); static int pci_maprange(uint64_t mapreg); static void pci_fixancient(pcicfgregs *cfg); static int pci_porten(device_t dev); static int pci_memen(device_t dev); static void pci_assign_interrupt(device_t bus, device_t dev, int force_route); static int pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl, int force, int prefetch); static int pci_probe(device_t dev); static int pci_attach(device_t dev); static void pci_load_vendor_data(void); static int pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc); static char *pci_describe_device(device_t dev); static int pci_modevent(module_t mod, int what, void *arg); static void pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg); static void pci_read_cap(device_t pcib, pcicfgregs *cfg); static int pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data); #if 0 static int pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data); #endif static void pci_read_vpd(device_t pcib, pcicfgregs *cfg); static void pci_disable_msi(device_t dev); static void pci_enable_msi(device_t dev, uint64_t address, uint16_t data); static void pci_enable_msix(device_t dev, u_int index, uint64_t address, uint32_t data); static void pci_mask_msix(device_t dev, u_int index); static void pci_unmask_msix(device_t dev, u_int index); static int pci_msi_blacklisted(void); static void pci_resume_msi(device_t dev); static void pci_resume_msix(device_t dev); static device_method_t pci_methods[] = { /* Device interface */ DEVMETHOD(device_probe, pci_probe), DEVMETHOD(device_attach, pci_attach), DEVMETHOD(device_detach, bus_generic_detach), DEVMETHOD(device_shutdown, bus_generic_shutdown), DEVMETHOD(device_suspend, pci_suspend), DEVMETHOD(device_resume, pci_resume), /* Bus interface */ DEVMETHOD(bus_print_child, pci_print_child), DEVMETHOD(bus_probe_nomatch, pci_probe_nomatch), DEVMETHOD(bus_read_ivar, pci_read_ivar), DEVMETHOD(bus_write_ivar, pci_write_ivar), DEVMETHOD(bus_driver_added, pci_driver_added), DEVMETHOD(bus_setup_intr, pci_setup_intr), DEVMETHOD(bus_teardown_intr, pci_teardown_intr), DEVMETHOD(bus_get_resource_list,pci_get_resource_list), DEVMETHOD(bus_set_resource, bus_generic_rl_set_resource), DEVMETHOD(bus_get_resource, bus_generic_rl_get_resource), DEVMETHOD(bus_delete_resource, pci_delete_resource), DEVMETHOD(bus_alloc_resource, pci_alloc_resource), DEVMETHOD(bus_release_resource, bus_generic_rl_release_resource), DEVMETHOD(bus_activate_resource, pci_activate_resource), DEVMETHOD(bus_deactivate_resource, bus_generic_deactivate_resource), DEVMETHOD(bus_child_pnpinfo_str, pci_child_pnpinfo_str_method), DEVMETHOD(bus_child_location_str, pci_child_location_str_method), /* PCI interface */ DEVMETHOD(pci_read_config, pci_read_config_method), DEVMETHOD(pci_write_config, pci_write_config_method), DEVMETHOD(pci_enable_busmaster, pci_enable_busmaster_method), DEVMETHOD(pci_disable_busmaster, pci_disable_busmaster_method), DEVMETHOD(pci_enable_io, pci_enable_io_method), DEVMETHOD(pci_disable_io, pci_disable_io_method), DEVMETHOD(pci_get_vpd_ident, pci_get_vpd_ident_method), DEVMETHOD(pci_get_vpd_readonly, pci_get_vpd_readonly_method), DEVMETHOD(pci_get_powerstate, pci_get_powerstate_method), DEVMETHOD(pci_set_powerstate, pci_set_powerstate_method), DEVMETHOD(pci_assign_interrupt, pci_assign_interrupt_method), DEVMETHOD(pci_find_extcap, pci_find_extcap_method), DEVMETHOD(pci_alloc_msi, pci_alloc_msi_method), DEVMETHOD(pci_alloc_msix, pci_alloc_msix_method), DEVMETHOD(pci_remap_msix, pci_remap_msix_method), DEVMETHOD(pci_release_msi, pci_release_msi_method), DEVMETHOD(pci_msi_count, pci_msi_count_method), DEVMETHOD(pci_msix_count, pci_msix_count_method), DEVMETHOD_END }; DEFINE_CLASS_0(pci, pci_driver, pci_methods, 0); static devclass_t pci_devclass; DRIVER_MODULE(pci, pcib, pci_driver, pci_devclass, pci_modevent, 0); MODULE_VERSION(pci, 1); static char *pci_vendordata; static size_t pci_vendordata_size; struct pci_quirk { uint32_t devid; /* Vendor/device of the card */ int type; #define PCI_QUIRK_MAP_REG 1 /* PCI map register in weird place */ #define PCI_QUIRK_DISABLE_MSI 2 /* MSI/MSI-X doesn't work */ #define PCI_QUIRK_ENABLE_MSI_VM 3 /* Older chipset in VM where MSI works */ int arg1; int arg2; }; static const struct pci_quirk const pci_quirks[] = { /* The Intel 82371AB and 82443MX has a map register at offset 0x90. */ { 0x71138086, PCI_QUIRK_MAP_REG, 0x90, 0 }, { 0x719b8086, PCI_QUIRK_MAP_REG, 0x90, 0 }, /* As does the Serverworks OSB4 (the SMBus mapping register) */ { 0x02001166, PCI_QUIRK_MAP_REG, 0x90, 0 }, /* * MSI doesn't work with the ServerWorks CNB20-HE Host Bridge * or the CMIC-SL (AKA ServerWorks GC_LE). */ { 0x00141166, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x00171166, PCI_QUIRK_DISABLE_MSI, 0, 0 }, /* * MSI doesn't work on earlier Intel chipsets including * E7500, E7501, E7505, 845, 865, 875/E7210, and 855. */ { 0x25408086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x254c8086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x25508086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x25608086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x25708086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x25788086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, { 0x35808086, PCI_QUIRK_DISABLE_MSI, 0, 0 }, /* * MSI doesn't work with devices behind the AMD 8131 HT-PCIX * bridge. */ { 0x74501022, PCI_QUIRK_DISABLE_MSI, 0, 0 }, /* * MSI-X doesn't work with at least LSI SAS1068E passed through by * VMware. */ { 0x079015ad, PCI_QUIRK_DISABLE_MSI, 0, 0 }, /* * Some virtualization environments emulate an older chipset * but support MSI just fine. QEMU uses the Intel 82440. */ { 0x12378086, PCI_QUIRK_ENABLE_MSI_VM, 0, 0 }, { 0 } }; /* map register information */ #define PCI_MAPMEM 0x01 /* memory map */ #define PCI_MAPMEMP 0x02 /* prefetchable memory map */ #define PCI_MAPPORT 0x04 /* port map */ struct devlist pci_devq; uint32_t pci_generation; uint32_t pci_numdevs = 0; static int pcie_chipset, pcix_chipset; /* sysctl vars */ SYSCTL_NODE(_hw, OID_AUTO, pci, CTLFLAG_RD, 0, "PCI bus tuning parameters"); static int pci_enable_io_modes = 1; TUNABLE_INT("hw.pci.enable_io_modes", &pci_enable_io_modes); SYSCTL_INT(_hw_pci, OID_AUTO, enable_io_modes, CTLFLAG_RW, &pci_enable_io_modes, 1, "Enable I/O and memory bits in the config register. Some BIOSes do not\n\ enable these bits correctly. We'd like to do this all the time, but there\n\ are some peripherals that this causes problems with."); static int pci_do_power_nodriver = 0; TUNABLE_INT("hw.pci.do_power_nodriver", &pci_do_power_nodriver); SYSCTL_INT(_hw_pci, OID_AUTO, do_power_nodriver, CTLFLAG_RW, &pci_do_power_nodriver, 0, "Place a function into D3 state when no driver attaches to it. 0 means\n\ disable. 1 means conservatively place devices into D3 state. 2 means\n\ agressively place devices into D3 state. 3 means put absolutely everything\n\ in D3 state."); static int pci_do_power_resume = 1; TUNABLE_INT("hw.pci.do_power_resume", &pci_do_power_resume); SYSCTL_INT(_hw_pci, OID_AUTO, do_power_resume, CTLFLAG_RW, &pci_do_power_resume, 1, "Transition from D3 -> D0 on resume."); static int pci_do_msi = 1; TUNABLE_INT("hw.pci.enable_msi", &pci_do_msi); SYSCTL_INT(_hw_pci, OID_AUTO, enable_msi, CTLFLAG_RW, &pci_do_msi, 1, "Enable support for MSI interrupts"); static int pci_do_msix = 1; TUNABLE_INT("hw.pci.enable_msix", &pci_do_msix); SYSCTL_INT(_hw_pci, OID_AUTO, enable_msix, CTLFLAG_RW, &pci_do_msix, 1, "Enable support for MSI-X interrupts"); static int pci_honor_msi_blacklist = 1; TUNABLE_INT("hw.pci.honor_msi_blacklist", &pci_honor_msi_blacklist); SYSCTL_INT(_hw_pci, OID_AUTO, honor_msi_blacklist, CTLFLAG_RD, &pci_honor_msi_blacklist, 1, "Honor chipset blacklist for MSI"); /* Find a device_t by bus/slot/function in domain 0 */ device_t pci_find_bsf(uint8_t bus, uint8_t slot, uint8_t func) { return (pci_find_dbsf(0, bus, slot, func)); } /* Find a device_t by domain/bus/slot/function */ device_t pci_find_dbsf(uint32_t domain, uint8_t bus, uint8_t slot, uint8_t func) { struct pci_devinfo *dinfo; STAILQ_FOREACH(dinfo, &pci_devq, pci_links) { if ((dinfo->cfg.domain == domain) && (dinfo->cfg.bus == bus) && (dinfo->cfg.slot == slot) && (dinfo->cfg.func == func)) { return (dinfo->cfg.dev); } } return (NULL); } /* Find a device_t by vendor/device ID */ device_t pci_find_device(uint16_t vendor, uint16_t device) { struct pci_devinfo *dinfo; STAILQ_FOREACH(dinfo, &pci_devq, pci_links) { if ((dinfo->cfg.vendor == vendor) && (dinfo->cfg.device == device)) { return (dinfo->cfg.dev); } } return (NULL); } /* return base address of memory or port map */ static pci_addr_t pci_mapbase(uint64_t mapreg) { if (PCI_BAR_MEM(mapreg)) return (mapreg & PCIM_BAR_MEM_BASE); else return (mapreg & PCIM_BAR_IO_BASE); } /* return map type of memory or port map */ static const char * pci_maptype(uint64_t mapreg) { if (PCI_BAR_IO(mapreg)) return ("I/O Port"); if (mapreg & PCIM_BAR_MEM_PREFETCH) return ("Prefetchable Memory"); return ("Memory"); } /* return log2 of map size decoded for memory or port map */ static int pci_mapsize(uint64_t testval) { int ln2size; testval = pci_mapbase(testval); ln2size = 0; if (testval != 0) { while ((testval & 1) == 0) { ln2size++; testval >>= 1; } } return (ln2size); } /* return log2 of address range supported by map register */ static int pci_maprange(uint64_t mapreg) { int ln2range = 0; if (PCI_BAR_IO(mapreg)) ln2range = 32; else switch (mapreg & PCIM_BAR_MEM_TYPE) { case PCIM_BAR_MEM_32: ln2range = 32; break; case PCIM_BAR_MEM_1MB: ln2range = 20; break; case PCIM_BAR_MEM_64: ln2range = 64; break; } return (ln2range); } /* adjust some values from PCI 1.0 devices to match 2.0 standards ... */ static void pci_fixancient(pcicfgregs *cfg) { if (cfg->hdrtype != 0) return; /* PCI to PCI bridges use header type 1 */ if (cfg->baseclass == PCIC_BRIDGE && cfg->subclass == PCIS_BRIDGE_PCI) cfg->hdrtype = 1; } /* extract header type specific config data */ static void pci_hdrtypedata(device_t pcib, int b, int s, int f, pcicfgregs *cfg) { #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w) switch (cfg->hdrtype) { case 0: cfg->subvendor = REG(PCIR_SUBVEND_0, 2); cfg->subdevice = REG(PCIR_SUBDEV_0, 2); cfg->nummaps = PCI_MAXMAPS_0; break; case 1: cfg->nummaps = PCI_MAXMAPS_1; break; case 2: cfg->subvendor = REG(PCIR_SUBVEND_2, 2); cfg->subdevice = REG(PCIR_SUBDEV_2, 2); cfg->nummaps = PCI_MAXMAPS_2; break; } #undef REG } /* read configuration header into pcicfgregs structure */ struct pci_devinfo * pci_read_device(device_t pcib, int d, int b, int s, int f, size_t size) { #define REG(n, w) PCIB_READ_CONFIG(pcib, b, s, f, n, w) pcicfgregs *cfg = NULL; struct pci_devinfo *devlist_entry; struct devlist *devlist_head; devlist_head = &pci_devq; devlist_entry = NULL; if (REG(PCIR_DEVVENDOR, 4) != -1) { devlist_entry = malloc(size, M_DEVBUF, M_WAITOK | M_ZERO); if (devlist_entry == NULL) return (NULL); cfg = &devlist_entry->cfg; cfg->domain = d; cfg->bus = b; cfg->slot = s; cfg->func = f; cfg->vendor = REG(PCIR_VENDOR, 2); cfg->device = REG(PCIR_DEVICE, 2); cfg->cmdreg = REG(PCIR_COMMAND, 2); cfg->statreg = REG(PCIR_STATUS, 2); cfg->baseclass = REG(PCIR_CLASS, 1); cfg->subclass = REG(PCIR_SUBCLASS, 1); cfg->progif = REG(PCIR_PROGIF, 1); cfg->revid = REG(PCIR_REVID, 1); cfg->hdrtype = REG(PCIR_HDRTYPE, 1); cfg->cachelnsz = REG(PCIR_CACHELNSZ, 1); cfg->lattimer = REG(PCIR_LATTIMER, 1); cfg->intpin = REG(PCIR_INTPIN, 1); cfg->intline = REG(PCIR_INTLINE, 1); cfg->mingnt = REG(PCIR_MINGNT, 1); cfg->maxlat = REG(PCIR_MAXLAT, 1); cfg->mfdev = (cfg->hdrtype & PCIM_MFDEV) != 0; cfg->hdrtype &= ~PCIM_MFDEV; pci_fixancient(cfg); pci_hdrtypedata(pcib, b, s, f, cfg); if (REG(PCIR_STATUS, 2) & PCIM_STATUS_CAPPRESENT) pci_read_cap(pcib, cfg); STAILQ_INSERT_TAIL(devlist_head, devlist_entry, pci_links); devlist_entry->conf.pc_sel.pc_domain = cfg->domain; devlist_entry->conf.pc_sel.pc_bus = cfg->bus; devlist_entry->conf.pc_sel.pc_dev = cfg->slot; devlist_entry->conf.pc_sel.pc_func = cfg->func; devlist_entry->conf.pc_hdr = cfg->hdrtype; devlist_entry->conf.pc_subvendor = cfg->subvendor; devlist_entry->conf.pc_subdevice = cfg->subdevice; devlist_entry->conf.pc_vendor = cfg->vendor; devlist_entry->conf.pc_device = cfg->device; devlist_entry->conf.pc_class = cfg->baseclass; devlist_entry->conf.pc_subclass = cfg->subclass; devlist_entry->conf.pc_progif = cfg->progif; devlist_entry->conf.pc_revid = cfg->revid; pci_numdevs++; pci_generation++; } return (devlist_entry); #undef REG } static void pci_read_cap(device_t pcib, pcicfgregs *cfg) { #define REG(n, w) PCIB_READ_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, w) #define WREG(n, v, w) PCIB_WRITE_CONFIG(pcib, cfg->bus, cfg->slot, cfg->func, n, v, w) #if defined(__i386__) || defined(__amd64__) uint64_t addr; #endif uint32_t val; int ptr, nextptr, ptrptr; switch (cfg->hdrtype & PCIM_HDRTYPE) { case 0: case 1: ptrptr = PCIR_CAP_PTR; break; case 2: ptrptr = PCIR_CAP_PTR_2; /* cardbus capabilities ptr */ break; default: return; /* no extended capabilities support */ } nextptr = REG(ptrptr, 1); /* sanity check? */ /* * Read capability entries. */ while (nextptr != 0) { /* Sanity check */ if (nextptr > 255) { printf("illegal PCI extended capability offset %d\n", nextptr); return; } /* Find the next entry */ ptr = nextptr; nextptr = REG(ptr + PCICAP_NEXTPTR, 1); /* Process this entry */ switch (REG(ptr + PCICAP_ID, 1)) { case PCIY_PMG: /* PCI power management */ if (cfg->pp.pp_cap == 0) { cfg->pp.pp_cap = REG(ptr + PCIR_POWER_CAP, 2); cfg->pp.pp_status = ptr + PCIR_POWER_STATUS; cfg->pp.pp_pmcsr = ptr + PCIR_POWER_PMCSR; if ((nextptr - ptr) > PCIR_POWER_DATA) cfg->pp.pp_data = ptr + PCIR_POWER_DATA; } break; #if defined(__i386__) || defined(__amd64__) case PCIY_HT: /* HyperTransport */ /* Determine HT-specific capability type. */ val = REG(ptr + PCIR_HT_COMMAND, 2); switch (val & PCIM_HTCMD_CAP_MASK) { case PCIM_HTCAP_MSI_MAPPING: if (!(val & PCIM_HTCMD_MSI_FIXED)) { /* Sanity check the mapping window. */ addr = REG(ptr + PCIR_HTMSI_ADDRESS_HI, 4); addr <<= 32; addr |= REG(ptr + PCIR_HTMSI_ADDRESS_LO, 4); if (addr != MSI_INTEL_ADDR_BASE) device_printf(pcib, "HT Bridge at pci%d:%d:%d:%d has non-default MSI window 0x%llx\n", cfg->domain, cfg->bus, cfg->slot, cfg->func, (long long)addr); } else addr = MSI_INTEL_ADDR_BASE; cfg->ht.ht_msimap = ptr; cfg->ht.ht_msictrl = val; cfg->ht.ht_msiaddr = addr; break; } break; #endif case PCIY_MSI: /* PCI MSI */ cfg->msi.msi_location = ptr; cfg->msi.msi_ctrl = REG(ptr + PCIR_MSI_CTRL, 2); cfg->msi.msi_msgnum = 1 << ((cfg->msi.msi_ctrl & PCIM_MSICTRL_MMC_MASK)>>1); break; case PCIY_MSIX: /* PCI MSI-X */ cfg->msix.msix_location = ptr; cfg->msix.msix_ctrl = REG(ptr + PCIR_MSIX_CTRL, 2); cfg->msix.msix_msgnum = (cfg->msix.msix_ctrl & PCIM_MSIXCTRL_TABLE_SIZE) + 1; val = REG(ptr + PCIR_MSIX_TABLE, 4); cfg->msix.msix_table_bar = PCIR_BAR(val & PCIM_MSIX_BIR_MASK); cfg->msix.msix_table_offset = val & ~PCIM_MSIX_BIR_MASK; val = REG(ptr + PCIR_MSIX_PBA, 4); cfg->msix.msix_pba_bar = PCIR_BAR(val & PCIM_MSIX_BIR_MASK); cfg->msix.msix_pba_offset = val & ~PCIM_MSIX_BIR_MASK; break; case PCIY_VPD: /* PCI Vital Product Data */ cfg->vpd.vpd_reg = ptr; break; case PCIY_SUBVENDOR: /* Should always be true. */ if ((cfg->hdrtype & PCIM_HDRTYPE) == 1) { val = REG(ptr + PCIR_SUBVENDCAP_ID, 4); cfg->subvendor = val & 0xffff; cfg->subdevice = val >> 16; } break; case PCIY_PCIX: /* PCI-X */ /* * Assume we have a PCI-X chipset if we have * at least one PCI-PCI bridge with a PCI-X * capability. Note that some systems with * PCI-express or HT chipsets might match on * this check as well. */ if ((cfg->hdrtype & PCIM_HDRTYPE) == 1) pcix_chipset = 1; break; case PCIY_EXPRESS: /* PCI-express */ /* * Assume we have a PCI-express chipset if we have * at least one PCI-express device. */ pcie_chipset = 1; break; default: break; } } /* REG and WREG use carry through to next functions */ } /* * PCI Vital Product Data */ #define PCI_VPD_TIMEOUT 1000000 static int pci_read_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t *data) { int count = PCI_VPD_TIMEOUT; KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned")); WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg, 2); while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) != 0x8000) { if (--count < 0) return (ENXIO); DELAY(1); /* limit looping */ } *data = (REG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, 4)); return (0); } #if 0 static int pci_write_vpd_reg(device_t pcib, pcicfgregs *cfg, int reg, uint32_t data) { int count = PCI_VPD_TIMEOUT; KASSERT((reg & 3) == 0, ("VPD register must by 4 byte aligned")); WREG(cfg->vpd.vpd_reg + PCIR_VPD_DATA, data, 4); WREG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, reg | 0x8000, 2); while ((REG(cfg->vpd.vpd_reg + PCIR_VPD_ADDR, 2) & 0x8000) == 0x8000) { if (--count < 0) return (ENXIO); DELAY(1); /* limit looping */ } return (0); } #endif #undef PCI_VPD_TIMEOUT struct vpd_readstate { device_t pcib; pcicfgregs *cfg; uint32_t val; int bytesinval; int off; uint8_t cksum; }; static int vpd_nextbyte(struct vpd_readstate *vrs, uint8_t *data) { uint32_t reg; uint8_t byte; if (vrs->bytesinval == 0) { if (pci_read_vpd_reg(vrs->pcib, vrs->cfg, vrs->off, ®)) return (ENXIO); vrs->val = le32toh(reg); vrs->off += 4; byte = vrs->val & 0xff; vrs->bytesinval = 3; } else { vrs->val = vrs->val >> 8; byte = vrs->val & 0xff; vrs->bytesinval--; } vrs->cksum += byte; *data = byte; return (0); } static void pci_read_vpd(device_t pcib, pcicfgregs *cfg) { struct vpd_readstate vrs; int state; int name; int remain; int i; int alloc, off; /* alloc/off for RO/W arrays */ int cksumvalid; int dflen; uint8_t byte; uint8_t byte2; /* init vpd reader */ vrs.bytesinval = 0; vrs.off = 0; vrs.pcib = pcib; vrs.cfg = cfg; vrs.cksum = 0; state = 0; name = remain = i = 0; /* shut up stupid gcc */ alloc = off = 0; /* shut up stupid gcc */ dflen = 0; /* shut up stupid gcc */ cksumvalid = -1; while (state >= 0) { if (vpd_nextbyte(&vrs, &byte)) { state = -2; break; } #if 0 printf("vpd: val: %#x, off: %d, bytesinval: %d, byte: %#hhx, " \ "state: %d, remain: %d, name: %#x, i: %d\n", vrs.val, vrs.off, vrs.bytesinval, byte, state, remain, name, i); #endif switch (state) { case 0: /* item name */ if (byte & 0x80) { if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } remain = byte2; if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } remain |= byte2 << 8; if (remain > (0x7f*4 - vrs.off)) { state = -1; printf( "pci%d:%d:%d:%d: invalid VPD data, remain %#x\n", cfg->domain, cfg->bus, cfg->slot, cfg->func, remain); } name = byte & 0x7f; } else { remain = byte & 0x7; name = (byte >> 3) & 0xf; } switch (name) { case 0x2: /* String */ cfg->vpd.vpd_ident = malloc(remain + 1, M_DEVBUF, M_WAITOK); i = 0; state = 1; break; case 0xf: /* End */ state = -1; break; case 0x10: /* VPD-R */ alloc = 8; off = 0; cfg->vpd.vpd_ros = malloc(alloc * sizeof(*cfg->vpd.vpd_ros), M_DEVBUF, M_WAITOK | M_ZERO); state = 2; break; case 0x11: /* VPD-W */ alloc = 8; off = 0; cfg->vpd.vpd_w = malloc(alloc * sizeof(*cfg->vpd.vpd_w), M_DEVBUF, M_WAITOK | M_ZERO); state = 5; break; default: /* Invalid data, abort */ state = -1; break; } break; case 1: /* Identifier String */ cfg->vpd.vpd_ident[i++] = byte; remain--; if (remain == 0) { cfg->vpd.vpd_ident[i] = '\0'; state = 0; } break; case 2: /* VPD-R Keyword Header */ if (off == alloc) { cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros, (alloc *= 2) * sizeof(*cfg->vpd.vpd_ros), M_DEVBUF, M_WAITOK | M_ZERO); } cfg->vpd.vpd_ros[off].keyword[0] = byte; if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } cfg->vpd.vpd_ros[off].keyword[1] = byte2; if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } dflen = byte2; if (dflen == 0 && strncmp(cfg->vpd.vpd_ros[off].keyword, "RV", 2) == 0) { /* * if this happens, we can't trust the rest * of the VPD. */ printf( "pci%d:%d:%d:%d: bad keyword length: %d\n", cfg->domain, cfg->bus, cfg->slot, cfg->func, dflen); cksumvalid = 0; state = -1; break; } else if (dflen == 0) { cfg->vpd.vpd_ros[off].value = malloc(1 * sizeof(*cfg->vpd.vpd_ros[off].value), M_DEVBUF, M_WAITOK); cfg->vpd.vpd_ros[off].value[0] = '\x00'; } else cfg->vpd.vpd_ros[off].value = malloc( (dflen + 1) * sizeof(*cfg->vpd.vpd_ros[off].value), M_DEVBUF, M_WAITOK); remain -= 3; i = 0; /* keep in sync w/ state 3's transistions */ if (dflen == 0 && remain == 0) state = 0; else if (dflen == 0) state = 2; else state = 3; break; case 3: /* VPD-R Keyword Value */ cfg->vpd.vpd_ros[off].value[i++] = byte; if (strncmp(cfg->vpd.vpd_ros[off].keyword, "RV", 2) == 0 && cksumvalid == -1) { if (vrs.cksum == 0) cksumvalid = 1; else { if (bootverbose) printf( "pci%d:%d:%d:%d: bad VPD cksum, remain %hhu\n", cfg->domain, cfg->bus, cfg->slot, cfg->func, vrs.cksum); cksumvalid = 0; state = -1; break; } } dflen--; remain--; /* keep in sync w/ state 2's transistions */ if (dflen == 0) cfg->vpd.vpd_ros[off++].value[i++] = '\0'; if (dflen == 0 && remain == 0) { cfg->vpd.vpd_rocnt = off; cfg->vpd.vpd_ros = reallocf(cfg->vpd.vpd_ros, off * sizeof(*cfg->vpd.vpd_ros), M_DEVBUF, M_WAITOK | M_ZERO); state = 0; } else if (dflen == 0) state = 2; break; case 4: remain--; if (remain == 0) state = 0; break; case 5: /* VPD-W Keyword Header */ if (off == alloc) { cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w, (alloc *= 2) * sizeof(*cfg->vpd.vpd_w), M_DEVBUF, M_WAITOK | M_ZERO); } cfg->vpd.vpd_w[off].keyword[0] = byte; if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } cfg->vpd.vpd_w[off].keyword[1] = byte2; if (vpd_nextbyte(&vrs, &byte2)) { state = -2; break; } cfg->vpd.vpd_w[off].len = dflen = byte2; cfg->vpd.vpd_w[off].start = vrs.off - vrs.bytesinval; cfg->vpd.vpd_w[off].value = malloc((dflen + 1) * sizeof(*cfg->vpd.vpd_w[off].value), M_DEVBUF, M_WAITOK); remain -= 3; i = 0; /* keep in sync w/ state 6's transistions */ if (dflen == 0 && remain == 0) state = 0; else if (dflen == 0) state = 5; else state = 6; break; case 6: /* VPD-W Keyword Value */ cfg->vpd.vpd_w[off].value[i++] = byte; dflen--; remain--; /* keep in sync w/ state 5's transistions */ if (dflen == 0) cfg->vpd.vpd_w[off++].value[i++] = '\0'; if (dflen == 0 && remain == 0) { cfg->vpd.vpd_wcnt = off; cfg->vpd.vpd_w = reallocf(cfg->vpd.vpd_w, off * sizeof(*cfg->vpd.vpd_w), M_DEVBUF, M_WAITOK | M_ZERO); state = 0; } else if (dflen == 0) state = 5; break; default: printf("pci%d:%d:%d:%d: invalid state: %d\n", cfg->domain, cfg->bus, cfg->slot, cfg->func, state); state = -1; break; } } if (cksumvalid == 0 || state < -1) { /* read-only data bad, clean up */ if (cfg->vpd.vpd_ros != NULL) { for (off = 0; cfg->vpd.vpd_ros[off].value; off++) free(cfg->vpd.vpd_ros[off].value, M_DEVBUF); free(cfg->vpd.vpd_ros, M_DEVBUF); cfg->vpd.vpd_ros = NULL; } } if (state < -1) { /* I/O error, clean up */ printf("pci%d:%d:%d:%d: failed to read VPD data.\n", cfg->domain, cfg->bus, cfg->slot, cfg->func); if (cfg->vpd.vpd_ident != NULL) { free(cfg->vpd.vpd_ident, M_DEVBUF); cfg->vpd.vpd_ident = NULL; } if (cfg->vpd.vpd_w != NULL) { for (off = 0; cfg->vpd.vpd_w[off].value; off++) free(cfg->vpd.vpd_w[off].value, M_DEVBUF); free(cfg->vpd.vpd_w, M_DEVBUF); cfg->vpd.vpd_w = NULL; } } cfg->vpd.vpd_cached = 1; #undef REG #undef WREG } int pci_get_vpd_ident_method(device_t dev, device_t child, const char **identptr) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0) pci_read_vpd(device_get_parent(dev), cfg); *identptr = cfg->vpd.vpd_ident; if (*identptr == NULL) return (ENXIO); return (0); } int pci_get_vpd_readonly_method(device_t dev, device_t child, const char *kw, const char **vptr) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; int i; if (!cfg->vpd.vpd_cached && cfg->vpd.vpd_reg != 0) pci_read_vpd(device_get_parent(dev), cfg); for (i = 0; i < cfg->vpd.vpd_rocnt; i++) if (memcmp(kw, cfg->vpd.vpd_ros[i].keyword, sizeof(cfg->vpd.vpd_ros[i].keyword)) == 0) { *vptr = cfg->vpd.vpd_ros[i].value; } if (i != cfg->vpd.vpd_rocnt) return (0); *vptr = NULL; return (ENXIO); } /* * Return the offset in configuration space of the requested extended * capability entry or 0 if the specified capability was not found. */ int pci_find_extcap_method(device_t dev, device_t child, int capability, int *capreg) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; u_int32_t status; u_int8_t ptr; /* * Check the CAP_LIST bit of the PCI status register first. */ status = pci_read_config(child, PCIR_STATUS, 2); if (!(status & PCIM_STATUS_CAPPRESENT)) return (ENXIO); /* * Determine the start pointer of the capabilities list. */ switch (cfg->hdrtype & PCIM_HDRTYPE) { case 0: case 1: ptr = PCIR_CAP_PTR; break; case 2: ptr = PCIR_CAP_PTR_2; break; default: /* XXX: panic? */ return (ENXIO); /* no extended capabilities support */ } ptr = pci_read_config(child, ptr, 1); /* * Traverse the capabilities list. */ while (ptr != 0) { if (pci_read_config(child, ptr + PCICAP_ID, 1) == capability) { if (capreg != NULL) *capreg = ptr; return (0); } ptr = pci_read_config(child, ptr + PCICAP_NEXTPTR, 1); } return (ENOENT); } /* * Support for MSI-X message interrupts. */ void pci_enable_msix(device_t dev, u_int index, uint64_t address, uint32_t data) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msix *msix = &dinfo->cfg.msix; uint32_t offset; KASSERT(msix->msix_table_len > index, ("bogus index")); offset = msix->msix_table_offset + index * 16; bus_write_4(msix->msix_table_res, offset, address & 0xffffffff); bus_write_4(msix->msix_table_res, offset + 4, address >> 32); bus_write_4(msix->msix_table_res, offset + 8, data); /* Enable MSI -> HT mapping. */ pci_ht_map_msi(dev, address); } void pci_mask_msix(device_t dev, u_int index) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msix *msix = &dinfo->cfg.msix; uint32_t offset, val; KASSERT(msix->msix_msgnum > index, ("bogus index")); offset = msix->msix_table_offset + index * 16 + 12; val = bus_read_4(msix->msix_table_res, offset); if (!(val & PCIM_MSIX_VCTRL_MASK)) { val |= PCIM_MSIX_VCTRL_MASK; bus_write_4(msix->msix_table_res, offset, val); } } void pci_unmask_msix(device_t dev, u_int index) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msix *msix = &dinfo->cfg.msix; uint32_t offset, val; KASSERT(msix->msix_table_len > index, ("bogus index")); offset = msix->msix_table_offset + index * 16 + 12; val = bus_read_4(msix->msix_table_res, offset); if (val & PCIM_MSIX_VCTRL_MASK) { val &= ~PCIM_MSIX_VCTRL_MASK; bus_write_4(msix->msix_table_res, offset, val); } } int pci_pending_msix(device_t dev, u_int index) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msix *msix = &dinfo->cfg.msix; uint32_t offset, bit; KASSERT(msix->msix_table_len > index, ("bogus index")); offset = msix->msix_pba_offset + (index / 32) * 4; bit = 1 << index % 32; return (bus_read_4(msix->msix_pba_res, offset) & bit); } /* * Restore MSI-X registers and table during resume. If MSI-X is * enabled then walk the virtual table to restore the actual MSI-X * table. */ static void pci_resume_msix(device_t dev) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msix *msix = &dinfo->cfg.msix; struct msix_table_entry *mte; struct msix_vector *mv; int i; if (msix->msix_alloc > 0) { /* First, mask all vectors. */ for (i = 0; i < msix->msix_msgnum; i++) pci_mask_msix(dev, i); /* Second, program any messages with at least one handler. */ for (i = 0; i < msix->msix_table_len; i++) { mte = &msix->msix_table[i]; if (mte->mte_vector == 0 || mte->mte_handlers == 0) continue; mv = &msix->msix_vectors[mte->mte_vector - 1]; pci_enable_msix(dev, i, mv->mv_address, mv->mv_data); pci_unmask_msix(dev, i); } } pci_write_config(dev, msix->msix_location + PCIR_MSIX_CTRL, msix->msix_ctrl, 2); } /* * Attempt to allocate *count MSI-X messages. The actual number allocated is * returned in *count. After this function returns, each message will be * available to the driver as SYS_RES_IRQ resources starting at rid 1. */ int pci_alloc_msix_method(device_t dev, device_t child, int *count) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; struct resource_list_entry *rle; int actual, error, i, irq, max; /* Don't let count == 0 get us into trouble. */ if (*count == 0) return (EINVAL); /* If rid 0 is allocated, then fail. */ rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0); if (rle != NULL && rle->res != NULL) return (ENXIO); /* Already have allocated messages? */ if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0) return (ENXIO); /* If MSI is blacklisted for this system, fail. */ if (pci_msi_blacklisted()) return (ENXIO); /* MSI-X capability present? */ if (cfg->msix.msix_location == 0 || !pci_do_msix) return (ENODEV); /* Make sure the appropriate BARs are mapped. */ rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY, cfg->msix.msix_table_bar); if (rle == NULL || rle->res == NULL || !(rman_get_flags(rle->res) & RF_ACTIVE)) return (ENXIO); cfg->msix.msix_table_res = rle->res; if (cfg->msix.msix_pba_bar != cfg->msix.msix_table_bar) { rle = resource_list_find(&dinfo->resources, SYS_RES_MEMORY, cfg->msix.msix_pba_bar); if (rle == NULL || rle->res == NULL || !(rman_get_flags(rle->res) & RF_ACTIVE)) return (ENXIO); } cfg->msix.msix_pba_res = rle->res; if (bootverbose) device_printf(child, "attempting to allocate %d MSI-X vectors (%d supported)\n", *count, cfg->msix.msix_msgnum); max = min(*count, cfg->msix.msix_msgnum); for (i = 0; i < max; i++) { /* Allocate a message. */ error = PCIB_ALLOC_MSIX(device_get_parent(dev), child, &irq); if (error) { if (i == 0) return (error); break; } resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq, irq, 1); } actual = i; if (bootverbose) { rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 1); if (actual == 1) device_printf(child, "using IRQ %lu for MSI-X\n", rle->start); else { int run; /* * Be fancy and try to print contiguous runs of * IRQ values as ranges. 'irq' is the previous IRQ. * 'run' is true if we are in a range. */ device_printf(child, "using IRQs %lu", rle->start); irq = rle->start; run = 0; for (i = 1; i < actual; i++) { rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); /* Still in a run? */ if (rle->start == irq + 1) { run = 1; irq++; continue; } /* Finish previous range. */ if (run) { printf("-%d", irq); run = 0; } /* Start new range. */ printf(",%lu", rle->start); irq = rle->start; } /* Unfinished range? */ if (run) printf("-%d", irq); printf(" for MSI-X\n"); } } /* Mask all vectors. */ for (i = 0; i < cfg->msix.msix_msgnum; i++) pci_mask_msix(child, i); /* Allocate and initialize vector data and virtual table. */ cfg->msix.msix_vectors = malloc(sizeof(struct msix_vector) * actual, M_DEVBUF, M_WAITOK | M_ZERO); cfg->msix.msix_table = malloc(sizeof(struct msix_table_entry) * actual, M_DEVBUF, M_WAITOK | M_ZERO); for (i = 0; i < actual; i++) { rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); cfg->msix.msix_vectors[i].mv_irq = rle->start; cfg->msix.msix_table[i].mte_vector = i + 1; } /* Update control register to enable MSI-X. */ cfg->msix.msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE; pci_write_config(child, cfg->msix.msix_location + PCIR_MSIX_CTRL, cfg->msix.msix_ctrl, 2); /* Update counts of alloc'd messages. */ cfg->msix.msix_alloc = actual; cfg->msix.msix_table_len = actual; *count = actual; return (0); } /* * By default, pci_alloc_msix() will assign the allocated IRQ * resources consecutively to the first N messages in the MSI-X table. * However, device drivers may want to use different layouts if they * either receive fewer messages than they asked for, or they wish to * populate the MSI-X table sparsely. This method allows the driver * to specify what layout it wants. It must be called after a * successful pci_alloc_msix() but before any of the associated * SYS_RES_IRQ resources are allocated via bus_alloc_resource(). * * The 'vectors' array contains 'count' message vectors. The array * maps directly to the MSI-X table in that index 0 in the array * specifies the vector for the first message in the MSI-X table, etc. * The vector value in each array index can either be 0 to indicate * that no vector should be assigned to a message slot, or it can be a * number from 1 to N (where N is the count returned from a * succcessful call to pci_alloc_msix()) to indicate which message * vector (IRQ) to be used for the corresponding message. * * On successful return, each message with a non-zero vector will have * an associated SYS_RES_IRQ whose rid is equal to the array index + * 1. Additionally, if any of the IRQs allocated via the previous * call to pci_alloc_msix() are not used in the mapping, those IRQs * will be freed back to the system automatically. * * For example, suppose a driver has a MSI-X table with 6 messages and * asks for 6 messages, but pci_alloc_msix() only returns a count of * 3. Call the three vectors allocated by pci_alloc_msix() A, B, and * C. After the call to pci_alloc_msix(), the device will be setup to * have an MSI-X table of ABC--- (where - means no vector assigned). * If the driver ten passes a vector array of { 1, 0, 1, 2, 0, 2 }, * then the MSI-X table will look like A-AB-B, and the 'C' vector will * be freed back to the system. This device will also have valid * SYS_RES_IRQ rids of 1, 3, 4, and 6. * * In any case, the SYS_RES_IRQ rid X will always map to the message * at MSI-X table index X - 1 and will only be valid if a vector is * assigned to that table entry. */ int pci_remap_msix_method(device_t dev, device_t child, int count, const u_int *vectors) { struct pci_devinfo *dinfo = device_get_ivars(child); struct pcicfg_msix *msix = &dinfo->cfg.msix; struct resource_list_entry *rle; int i, irq, j, *used; /* * Have to have at least one message in the table but the * table can't be bigger than the actual MSI-X table in the * device. */ if (count == 0 || count > msix->msix_msgnum) return (EINVAL); /* Sanity check the vectors. */ for (i = 0; i < count; i++) if (vectors[i] > msix->msix_alloc) return (EINVAL); /* * Make sure there aren't any holes in the vectors to be used. * It's a big pain to support it, and it doesn't really make * sense anyway. Also, at least one vector must be used. */ used = malloc(sizeof(int) * msix->msix_alloc, M_DEVBUF, M_WAITOK | M_ZERO); for (i = 0; i < count; i++) if (vectors[i] != 0) used[vectors[i] - 1] = 1; for (i = 0; i < msix->msix_alloc - 1; i++) if (used[i] == 0 && used[i + 1] == 1) { free(used, M_DEVBUF); return (EINVAL); } if (used[0] != 1) { free(used, M_DEVBUF); return (EINVAL); } /* Make sure none of the resources are allocated. */ for (i = 0; i < msix->msix_table_len; i++) { if (msix->msix_table[i].mte_vector == 0) continue; if (msix->msix_table[i].mte_handlers > 0) return (EBUSY); rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); KASSERT(rle != NULL, ("missing resource")); if (rle->res != NULL) return (EBUSY); } /* Free the existing resource list entries. */ for (i = 0; i < msix->msix_table_len; i++) { if (msix->msix_table[i].mte_vector == 0) continue; resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1); } /* * Build the new virtual table keeping track of which vectors are * used. */ free(msix->msix_table, M_DEVBUF); msix->msix_table = malloc(sizeof(struct msix_table_entry) * count, M_DEVBUF, M_WAITOK | M_ZERO); for (i = 0; i < count; i++) msix->msix_table[i].mte_vector = vectors[i]; msix->msix_table_len = count; /* Free any unused IRQs and resize the vectors array if necessary. */ j = msix->msix_alloc - 1; if (used[j] == 0) { struct msix_vector *vec; while (used[j] == 0) { PCIB_RELEASE_MSIX(device_get_parent(dev), child, msix->msix_vectors[j].mv_irq); j--; } vec = malloc(sizeof(struct msix_vector) * (j + 1), M_DEVBUF, M_WAITOK); bcopy(msix->msix_vectors, vec, sizeof(struct msix_vector) * (j + 1)); free(msix->msix_vectors, M_DEVBUF); msix->msix_vectors = vec; msix->msix_alloc = j + 1; } free(used, M_DEVBUF); /* Map the IRQs onto the rids. */ for (i = 0; i < count; i++) { if (vectors[i] == 0) continue; irq = msix->msix_vectors[vectors[i]].mv_irq; resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irq, irq, 1); } if (bootverbose) { device_printf(child, "Remapped MSI-X IRQs as: "); for (i = 0; i < count; i++) { if (i != 0) printf(", "); if (vectors[i] == 0) printf("---"); else printf("%d", msix->msix_vectors[vectors[i]].mv_irq); } printf("\n"); } return (0); } static int pci_release_msix(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); struct pcicfg_msix *msix = &dinfo->cfg.msix; struct resource_list_entry *rle; int i; /* Do we have any messages to release? */ if (msix->msix_alloc == 0) return (ENODEV); /* Make sure none of the resources are allocated. */ for (i = 0; i < msix->msix_table_len; i++) { if (msix->msix_table[i].mte_vector == 0) continue; if (msix->msix_table[i].mte_handlers > 0) return (EBUSY); rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); KASSERT(rle != NULL, ("missing resource")); if (rle->res != NULL) return (EBUSY); } /* Update control register to disable MSI-X. */ msix->msix_ctrl &= ~PCIM_MSIXCTRL_MSIX_ENABLE; pci_write_config(child, msix->msix_location + PCIR_MSIX_CTRL, msix->msix_ctrl, 2); /* Free the resource list entries. */ for (i = 0; i < msix->msix_table_len; i++) { if (msix->msix_table[i].mte_vector == 0) continue; resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1); } free(msix->msix_table, M_DEVBUF); msix->msix_table_len = 0; /* Release the IRQs. */ for (i = 0; i < msix->msix_alloc; i++) PCIB_RELEASE_MSIX(device_get_parent(dev), child, msix->msix_vectors[i].mv_irq); free(msix->msix_vectors, M_DEVBUF); msix->msix_alloc = 0; return (0); } /* * Return the max supported MSI-X messages this device supports. * Basically, assuming the MD code can alloc messages, this function * should return the maximum value that pci_alloc_msix() can return. * Thus, it is subject to the tunables, etc. */ int pci_msix_count_method(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); struct pcicfg_msix *msix = &dinfo->cfg.msix; if (pci_do_msix && msix->msix_location != 0) return (msix->msix_msgnum); return (0); } /* * HyperTransport MSI mapping control */ void pci_ht_map_msi(device_t dev, uint64_t addr) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_ht *ht = &dinfo->cfg.ht; if (!ht->ht_msimap) return; if (addr && !(ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) && ht->ht_msiaddr >> 20 == addr >> 20) { /* Enable MSI -> HT mapping. */ ht->ht_msictrl |= PCIM_HTCMD_MSI_ENABLE; pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND, ht->ht_msictrl, 2); } if (!addr && ht->ht_msictrl & PCIM_HTCMD_MSI_ENABLE) { /* Disable MSI -> HT mapping. */ ht->ht_msictrl &= ~PCIM_HTCMD_MSI_ENABLE; pci_write_config(dev, ht->ht_msimap + PCIR_HT_COMMAND, ht->ht_msictrl, 2); } } int pci_get_max_read_req(device_t dev) { int cap; uint16_t val; if (pci_find_cap(dev, PCIY_EXPRESS, &cap) != 0) return (0); - val = pci_read_config(dev, cap + PCIR_EXPRESS_DEVICE_CTL, 2); - val &= PCIM_EXP_CTL_MAX_READ_REQUEST; + val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2); + val &= PCIEM_CTL_MAX_READ_REQUEST; val >>= 12; return (1 << (val + 7)); } int pci_set_max_read_req(device_t dev, int size) { int cap; uint16_t val; if (pci_find_cap(dev, PCIY_EXPRESS, &cap) != 0) return (0); if (size < 128) size = 128; if (size > 4096) size = 4096; size = (1 << (fls(size) - 1)); - val = pci_read_config(dev, cap + PCIR_EXPRESS_DEVICE_CTL, 2); - val &= ~PCIM_EXP_CTL_MAX_READ_REQUEST; + val = pci_read_config(dev, cap + PCIER_DEVICE_CTL, 2); + val &= ~PCIEM_CTL_MAX_READ_REQUEST; val |= (fls(size) - 8) << 12; - pci_write_config(dev, cap + PCIR_EXPRESS_DEVICE_CTL, val, 2); + pci_write_config(dev, cap + PCIER_DEVICE_CTL, val, 2); return (size); } /* * Support for MSI message signalled interrupts. */ void pci_enable_msi(device_t dev, uint64_t address, uint16_t data) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msi *msi = &dinfo->cfg.msi; /* Write data and address values. */ pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR, address & 0xffffffff, 4); if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) { pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR_HIGH, address >> 32, 4); pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA_64BIT, data, 2); } else pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA, data, 2); /* Enable MSI in the control register. */ msi->msi_ctrl |= PCIM_MSICTRL_MSI_ENABLE; pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl, 2); /* Enable MSI -> HT mapping. */ pci_ht_map_msi(dev, address); } void pci_disable_msi(device_t dev) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msi *msi = &dinfo->cfg.msi; /* Disable MSI -> HT mapping. */ pci_ht_map_msi(dev, 0); /* Disable MSI in the control register. */ msi->msi_ctrl &= ~PCIM_MSICTRL_MSI_ENABLE; pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl, 2); } /* * Restore MSI registers during resume. If MSI is enabled then * restore the data and address registers in addition to the control * register. */ static void pci_resume_msi(device_t dev) { struct pci_devinfo *dinfo = device_get_ivars(dev); struct pcicfg_msi *msi = &dinfo->cfg.msi; uint64_t address; uint16_t data; if (msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE) { address = msi->msi_addr; data = msi->msi_data; pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR, address & 0xffffffff, 4); if (msi->msi_ctrl & PCIM_MSICTRL_64BIT) { pci_write_config(dev, msi->msi_location + PCIR_MSI_ADDR_HIGH, address >> 32, 4); pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA_64BIT, data, 2); } else pci_write_config(dev, msi->msi_location + PCIR_MSI_DATA, data, 2); } pci_write_config(dev, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl, 2); } int pci_remap_msi_irq(device_t dev, u_int irq) { struct pci_devinfo *dinfo = device_get_ivars(dev); pcicfgregs *cfg = &dinfo->cfg; struct resource_list_entry *rle; struct msix_table_entry *mte; struct msix_vector *mv; device_t bus; uint64_t addr; uint32_t data; int error, i, j; bus = device_get_parent(dev); /* * Handle MSI first. We try to find this IRQ among our list * of MSI IRQs. If we find it, we request updated address and * data registers and apply the results. */ if (cfg->msi.msi_alloc > 0) { /* If we don't have any active handlers, nothing to do. */ if (cfg->msi.msi_handlers == 0) return (0); for (i = 0; i < cfg->msi.msi_alloc; i++) { rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); if (rle->start == irq) { error = PCIB_MAP_MSI(device_get_parent(bus), dev, irq, &addr, &data); if (error) return (error); pci_disable_msi(dev); dinfo->cfg.msi.msi_addr = addr; dinfo->cfg.msi.msi_data = data; pci_enable_msi(dev, addr, data); return (0); } } return (ENOENT); } /* * For MSI-X, we check to see if we have this IRQ. If we do, * we request the updated mapping info. If that works, we go * through all the slots that use this IRQ and update them. */ if (cfg->msix.msix_alloc > 0) { for (i = 0; i < cfg->msix.msix_alloc; i++) { mv = &cfg->msix.msix_vectors[i]; if (mv->mv_irq == irq) { error = PCIB_MAP_MSI(device_get_parent(bus), dev, irq, &addr, &data); if (error) return (error); mv->mv_address = addr; mv->mv_data = data; for (j = 0; j < cfg->msix.msix_table_len; j++) { mte = &cfg->msix.msix_table[j]; if (mte->mte_vector != i + 1) continue; if (mte->mte_handlers == 0) continue; pci_mask_msix(dev, j); pci_enable_msix(dev, j, addr, data); pci_unmask_msix(dev, j); } } } return (ENOENT); } return (ENOENT); } /* * Returns true if the specified device is blacklisted because MSI * doesn't work. */ int pci_msi_device_blacklisted(device_t dev) { const struct pci_quirk *q; if (!pci_honor_msi_blacklist) return (0); for (q = &pci_quirks[0]; q->devid; q++) { if (q->devid == pci_get_devid(dev) && q->type == PCI_QUIRK_DISABLE_MSI) return (1); } return (0); } /* * Returns true if a specified chipset supports MSI when it is * emulated hardware in a virtual machine. */ static int pci_msi_vm_chipset(device_t dev) { const struct pci_quirk *q; for (q = &pci_quirks[0]; q->devid; q++) { if (q->devid == pci_get_devid(dev) && q->type == PCI_QUIRK_ENABLE_MSI_VM) return (1); } return (0); } /* * Determine if MSI is blacklisted globally on this sytem. Currently, * we just check for blacklisted chipsets as represented by the * host-PCI bridge at device 0:0:0. In the future, it may become * necessary to check other system attributes, such as the kenv values * that give the motherboard manufacturer and model number. */ static int pci_msi_blacklisted(void) { device_t dev; if (!pci_honor_msi_blacklist) return (0); /* Blacklist all non-PCI-express and non-PCI-X chipsets. */ if (!(pcie_chipset || pcix_chipset)) { if (vm_guest != VM_GUEST_NO) { dev = pci_find_bsf(0, 0, 0); if (dev != NULL) return (pci_msi_vm_chipset(dev) == 0); } return (1); } dev = pci_find_bsf(0, 0, 0); if (dev != NULL) return (pci_msi_device_blacklisted(dev)); return (0); } /* * Attempt to allocate *count MSI messages. The actual number allocated is * returned in *count. After this function returns, each message will be * available to the driver as SYS_RES_IRQ resources starting at a rid 1. */ int pci_alloc_msi_method(device_t dev, device_t child, int *count) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; struct resource_list_entry *rle; int actual, error, i, irqs[32]; uint16_t ctrl; /* Don't let count == 0 get us into trouble. */ if (*count == 0) return (EINVAL); /* If rid 0 is allocated, then fail. */ rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, 0); if (rle != NULL && rle->res != NULL) return (ENXIO); /* Already have allocated messages? */ if (cfg->msi.msi_alloc != 0 || cfg->msix.msix_alloc != 0) return (ENXIO); /* If MSI is blacklisted for this system, fail. */ if (pci_msi_blacklisted()) return (ENXIO); /* MSI capability present? */ if (cfg->msi.msi_location == 0 || !pci_do_msi) return (ENODEV); if (bootverbose) device_printf(child, "attempting to allocate %d MSI vectors (%d supported)\n", *count, cfg->msi.msi_msgnum); /* Don't ask for more than the device supports. */ actual = min(*count, cfg->msi.msi_msgnum); /* Don't ask for more than 32 messages. */ actual = min(actual, 32); /* MSI requires power of 2 number of messages. */ if (!powerof2(actual)) return (EINVAL); for (;;) { /* Try to allocate N messages. */ error = PCIB_ALLOC_MSI(device_get_parent(dev), child, actual, actual, irqs); if (error == 0) break; if (actual == 1) return (error); /* Try N / 2. */ actual >>= 1; } /* * We now have N actual messages mapped onto SYS_RES_IRQ * resources in the irqs[] array, so add new resources * starting at rid 1. */ for (i = 0; i < actual; i++) resource_list_add(&dinfo->resources, SYS_RES_IRQ, i + 1, irqs[i], irqs[i], 1); if (bootverbose) { if (actual == 1) device_printf(child, "using IRQ %d for MSI\n", irqs[0]); else { int run; /* * Be fancy and try to print contiguous runs * of IRQ values as ranges. 'run' is true if * we are in a range. */ device_printf(child, "using IRQs %d", irqs[0]); run = 0; for (i = 1; i < actual; i++) { /* Still in a run? */ if (irqs[i] == irqs[i - 1] + 1) { run = 1; continue; } /* Finish previous range. */ if (run) { printf("-%d", irqs[i - 1]); run = 0; } /* Start new range. */ printf(",%d", irqs[i]); } /* Unfinished range? */ if (run) printf("-%d", irqs[actual - 1]); printf(" for MSI\n"); } } /* Update control register with actual count. */ ctrl = cfg->msi.msi_ctrl; ctrl &= ~PCIM_MSICTRL_MME_MASK; ctrl |= (ffs(actual) - 1) << 4; cfg->msi.msi_ctrl = ctrl; pci_write_config(child, cfg->msi.msi_location + PCIR_MSI_CTRL, ctrl, 2); /* Update counts of alloc'd messages. */ cfg->msi.msi_alloc = actual; cfg->msi.msi_handlers = 0; *count = actual; return (0); } /* Release the MSI messages associated with this device. */ int pci_release_msi_method(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); struct pcicfg_msi *msi = &dinfo->cfg.msi; struct resource_list_entry *rle; int error, i, irqs[32]; /* Try MSI-X first. */ error = pci_release_msix(dev, child); if (error != ENODEV) return (error); /* Do we have any messages to release? */ if (msi->msi_alloc == 0) return (ENODEV); KASSERT(msi->msi_alloc <= 32, ("more than 32 alloc'd messages")); /* Make sure none of the resources are allocated. */ if (msi->msi_handlers > 0) return (EBUSY); for (i = 0; i < msi->msi_alloc; i++) { rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, i + 1); KASSERT(rle != NULL, ("missing MSI resource")); if (rle->res != NULL) return (EBUSY); irqs[i] = rle->start; } /* Update control register with 0 count. */ KASSERT(!(msi->msi_ctrl & PCIM_MSICTRL_MSI_ENABLE), ("%s: MSI still enabled", __func__)); msi->msi_ctrl &= ~PCIM_MSICTRL_MME_MASK; pci_write_config(child, msi->msi_location + PCIR_MSI_CTRL, msi->msi_ctrl, 2); /* Release the messages. */ PCIB_RELEASE_MSI(device_get_parent(dev), child, msi->msi_alloc, irqs); for (i = 0; i < msi->msi_alloc; i++) resource_list_delete(&dinfo->resources, SYS_RES_IRQ, i + 1); /* Update alloc count. */ msi->msi_alloc = 0; msi->msi_addr = 0; msi->msi_data = 0; return (0); } /* * Return the max supported MSI messages this device supports. * Basically, assuming the MD code can alloc messages, this function * should return the maximum value that pci_alloc_msi() can return. * Thus, it is subject to the tunables, etc. */ int pci_msi_count_method(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); struct pcicfg_msi *msi = &dinfo->cfg.msi; if (pci_do_msi && msi->msi_location != 0) return (msi->msi_msgnum); return (0); } /* free pcicfgregs structure and all depending data structures */ int pci_freecfg(struct pci_devinfo *dinfo) { struct devlist *devlist_head; int i; devlist_head = &pci_devq; if (dinfo->cfg.vpd.vpd_reg) { free(dinfo->cfg.vpd.vpd_ident, M_DEVBUF); for (i = 0; i < dinfo->cfg.vpd.vpd_rocnt; i++) free(dinfo->cfg.vpd.vpd_ros[i].value, M_DEVBUF); free(dinfo->cfg.vpd.vpd_ros, M_DEVBUF); for (i = 0; i < dinfo->cfg.vpd.vpd_wcnt; i++) free(dinfo->cfg.vpd.vpd_w[i].value, M_DEVBUF); free(dinfo->cfg.vpd.vpd_w, M_DEVBUF); } STAILQ_REMOVE(devlist_head, dinfo, pci_devinfo, pci_links); free(dinfo, M_DEVBUF); /* increment the generation count */ pci_generation++; /* we're losing one device */ pci_numdevs--; return (0); } /* * PCI power manangement */ int pci_set_powerstate_method(device_t dev, device_t child, int state) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; uint16_t status; int result, oldstate, highest, delay; if (cfg->pp.pp_cap == 0) return (EOPNOTSUPP); /* * Optimize a no state change request away. While it would be OK to * write to the hardware in theory, some devices have shown odd * behavior when going from D3 -> D3. */ oldstate = pci_get_powerstate(child); if (oldstate == state) return (0); /* * The PCI power management specification states that after a state * transition between PCI power states, system software must * guarantee a minimal delay before the function accesses the device. * Compute the worst case delay that we need to guarantee before we * access the device. Many devices will be responsive much more * quickly than this delay, but there are some that don't respond * instantly to state changes. Transitions to/from D3 state require * 10ms, while D2 requires 200us, and D0/1 require none. The delay * is done below with DELAY rather than a sleeper function because * this function can be called from contexts where we cannot sleep. */ highest = (oldstate > state) ? oldstate : state; if (highest == PCI_POWERSTATE_D3) delay = 10000; else if (highest == PCI_POWERSTATE_D2) delay = 200; else delay = 0; status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2) & ~PCIM_PSTAT_DMASK; result = 0; switch (state) { case PCI_POWERSTATE_D0: status |= PCIM_PSTAT_D0; break; case PCI_POWERSTATE_D1: if ((cfg->pp.pp_cap & PCIM_PCAP_D1SUPP) == 0) return (EOPNOTSUPP); status |= PCIM_PSTAT_D1; break; case PCI_POWERSTATE_D2: if ((cfg->pp.pp_cap & PCIM_PCAP_D2SUPP) == 0) return (EOPNOTSUPP); status |= PCIM_PSTAT_D2; break; case PCI_POWERSTATE_D3: status |= PCIM_PSTAT_D3; break; default: return (EINVAL); } if (bootverbose) printf( "pci%d:%d:%d:%d: Transition from D%d to D%d\n", dinfo->cfg.domain, dinfo->cfg.bus, dinfo->cfg.slot, dinfo->cfg.func, oldstate, state); PCI_WRITE_CONFIG(dev, child, cfg->pp.pp_status, status, 2); if (delay) DELAY(delay); return (0); } int pci_get_powerstate_method(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; uint16_t status; int result; if (cfg->pp.pp_cap != 0) { status = PCI_READ_CONFIG(dev, child, cfg->pp.pp_status, 2); switch (status & PCIM_PSTAT_DMASK) { case PCIM_PSTAT_D0: result = PCI_POWERSTATE_D0; break; case PCIM_PSTAT_D1: result = PCI_POWERSTATE_D1; break; case PCIM_PSTAT_D2: result = PCI_POWERSTATE_D2; break; case PCIM_PSTAT_D3: result = PCI_POWERSTATE_D3; break; default: result = PCI_POWERSTATE_UNKNOWN; break; } } else { /* No support, device is always at D0 */ result = PCI_POWERSTATE_D0; } return (result); } /* * Some convenience functions for PCI device drivers. */ static __inline void pci_set_command_bit(device_t dev, device_t child, uint16_t bit) { uint16_t command; command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2); command |= bit; PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2); } static __inline void pci_clear_command_bit(device_t dev, device_t child, uint16_t bit) { uint16_t command; command = PCI_READ_CONFIG(dev, child, PCIR_COMMAND, 2); command &= ~bit; PCI_WRITE_CONFIG(dev, child, PCIR_COMMAND, command, 2); } int pci_enable_busmaster_method(device_t dev, device_t child) { pci_set_command_bit(dev, child, PCIM_CMD_BUSMASTEREN); return (0); } int pci_disable_busmaster_method(device_t dev, device_t child) { pci_clear_command_bit(dev, child, PCIM_CMD_BUSMASTEREN); return (0); } int pci_enable_io_method(device_t dev, device_t child, int space) { uint16_t bit; switch(space) { case SYS_RES_IOPORT: bit = PCIM_CMD_PORTEN; break; case SYS_RES_MEMORY: bit = PCIM_CMD_MEMEN; break; default: return (EINVAL); } pci_set_command_bit(dev, child, bit); return (0); } int pci_disable_io_method(device_t dev, device_t child, int space) { uint16_t bit; switch(space) { case SYS_RES_IOPORT: bit = PCIM_CMD_PORTEN; break; case SYS_RES_MEMORY: bit = PCIM_CMD_MEMEN; break; default: return (EINVAL); } pci_clear_command_bit(dev, child, bit); return (0); } /* * New style pci driver. Parent device is either a pci-host-bridge or a * pci-pci-bridge. Both kinds are represented by instances of pcib. */ void pci_print_verbose(struct pci_devinfo *dinfo) { if (bootverbose) { pcicfgregs *cfg = &dinfo->cfg; printf("found->\tvendor=0x%04x, dev=0x%04x, revid=0x%02x\n", cfg->vendor, cfg->device, cfg->revid); printf("\tdomain=%d, bus=%d, slot=%d, func=%d\n", cfg->domain, cfg->bus, cfg->slot, cfg->func); printf("\tclass=%02x-%02x-%02x, hdrtype=0x%02x, mfdev=%d\n", cfg->baseclass, cfg->subclass, cfg->progif, cfg->hdrtype, cfg->mfdev); printf("\tcmdreg=0x%04x, statreg=0x%04x, cachelnsz=%d (dwords)\n", cfg->cmdreg, cfg->statreg, cfg->cachelnsz); printf("\tlattimer=0x%02x (%d ns), mingnt=0x%02x (%d ns), maxlat=0x%02x (%d ns)\n", cfg->lattimer, cfg->lattimer * 30, cfg->mingnt, cfg->mingnt * 250, cfg->maxlat, cfg->maxlat * 250); if (cfg->intpin > 0) printf("\tintpin=%c, irq=%d\n", cfg->intpin +'a' -1, cfg->intline); if (cfg->pp.pp_cap) { uint16_t status; status = pci_read_config(cfg->dev, cfg->pp.pp_status, 2); printf("\tpowerspec %d supports D0%s%s D3 current D%d\n", cfg->pp.pp_cap & PCIM_PCAP_SPEC, cfg->pp.pp_cap & PCIM_PCAP_D1SUPP ? " D1" : "", cfg->pp.pp_cap & PCIM_PCAP_D2SUPP ? " D2" : "", status & PCIM_PSTAT_DMASK); } if (cfg->msi.msi_location) { int ctrl; ctrl = cfg->msi.msi_ctrl; printf("\tMSI supports %d message%s%s%s\n", cfg->msi.msi_msgnum, (cfg->msi.msi_msgnum == 1) ? "" : "s", (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "", (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks":""); } if (cfg->msix.msix_location) { printf("\tMSI-X supports %d message%s ", cfg->msix.msix_msgnum, (cfg->msix.msix_msgnum == 1) ? "" : "s"); if (cfg->msix.msix_table_bar == cfg->msix.msix_pba_bar) printf("in map 0x%x\n", cfg->msix.msix_table_bar); else printf("in maps 0x%x and 0x%x\n", cfg->msix.msix_table_bar, cfg->msix.msix_pba_bar); } } } static int pci_porten(device_t dev) { return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_PORTEN) != 0; } static int pci_memen(device_t dev) { return (pci_read_config(dev, PCIR_COMMAND, 2) & PCIM_CMD_MEMEN) != 0; } static void pci_read_bar(device_t dev, int reg, pci_addr_t *mapp, pci_addr_t *testvalp) { pci_addr_t map, testval; int ln2range; uint16_t cmd; map = pci_read_config(dev, reg, 4); ln2range = pci_maprange(map); if (ln2range == 64) map |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32; /* * Disable decoding via the command register before * determining the BAR's length since we will be placing it in * a weird state. */ cmd = pci_read_config(dev, PCIR_COMMAND, 2); pci_write_config(dev, PCIR_COMMAND, cmd & ~(PCI_BAR_MEM(map) ? PCIM_CMD_MEMEN : PCIM_CMD_PORTEN), 2); /* * Determine the BAR's length by writing all 1's. The bottom * log_2(size) bits of the BAR will stick as 0 when we read * the value back. */ pci_write_config(dev, reg, 0xffffffff, 4); testval = pci_read_config(dev, reg, 4); if (ln2range == 64) { pci_write_config(dev, reg + 4, 0xffffffff, 4); testval |= (pci_addr_t)pci_read_config(dev, reg + 4, 4) << 32; } /* * Restore the original value of the BAR. We may have reprogrammed * the BAR of the low-level console device and when booting verbose, * we need the console device addressable. */ pci_write_config(dev, reg, map, 4); if (ln2range == 64) pci_write_config(dev, reg + 4, map >> 32, 4); pci_write_config(dev, PCIR_COMMAND, cmd, 2); *mapp = map; *testvalp = testval; } static void pci_write_bar(device_t dev, int reg, pci_addr_t base) { pci_addr_t map; int ln2range; map = pci_read_config(dev, reg, 4); ln2range = pci_maprange(map); pci_write_config(dev, reg, base, 4); if (ln2range == 64) pci_write_config(dev, reg + 4, base >> 32, 4); } /* * Add a resource based on a pci map register. Return 1 if the map * register is a 32bit map register or 2 if it is a 64bit register. */ static int pci_add_map(device_t bus, device_t dev, int reg, struct resource_list *rl, int force, int prefetch) { pci_addr_t base, map, testval; pci_addr_t start, end, count; int barlen, basezero, maprange, mapsize, type; uint16_t cmd; struct resource *res; pci_read_bar(dev, reg, &map, &testval); if (PCI_BAR_MEM(map)) { type = SYS_RES_MEMORY; if (map & PCIM_BAR_MEM_PREFETCH) prefetch = 1; } else type = SYS_RES_IOPORT; mapsize = pci_mapsize(testval); base = pci_mapbase(map); #ifdef __PCI_BAR_ZERO_VALID basezero = 0; #else basezero = base == 0; #endif maprange = pci_maprange(map); barlen = maprange == 64 ? 2 : 1; /* * For I/O registers, if bottom bit is set, and the next bit up * isn't clear, we know we have a BAR that doesn't conform to the * spec, so ignore it. Also, sanity check the size of the data * areas to the type of memory involved. Memory must be at least * 16 bytes in size, while I/O ranges must be at least 4. */ if (PCI_BAR_IO(testval) && (testval & PCIM_BAR_IO_RESERVED) != 0) return (barlen); if ((type == SYS_RES_MEMORY && mapsize < 4) || (type == SYS_RES_IOPORT && mapsize < 2)) return (barlen); if (bootverbose) { printf("\tmap[%02x]: type %s, range %2d, base %#jx, size %2d", reg, pci_maptype(map), maprange, (uintmax_t)base, mapsize); if (type == SYS_RES_IOPORT && !pci_porten(dev)) printf(", port disabled\n"); else if (type == SYS_RES_MEMORY && !pci_memen(dev)) printf(", memory disabled\n"); else printf(", enabled\n"); } /* * If base is 0, then we have problems if this architecture does * not allow that. It is best to ignore such entries for the * moment. These will be allocated later if the driver specifically * requests them. However, some removable busses look better when * all resources are allocated, so allow '0' to be overriden. * * Similarly treat maps whose values is the same as the test value * read back. These maps have had all f's written to them by the * BIOS in an attempt to disable the resources. */ if (!force && (basezero || map == testval)) return (barlen); if ((u_long)base != base) { device_printf(bus, "pci%d:%d:%d:%d bar %#x too many address bits", pci_get_domain(dev), pci_get_bus(dev), pci_get_slot(dev), pci_get_function(dev), reg); return (barlen); } /* * This code theoretically does the right thing, but has * undesirable side effects in some cases where peripherals * respond oddly to having these bits enabled. Let the user * be able to turn them off (since pci_enable_io_modes is 1 by * default). */ if (pci_enable_io_modes) { /* Turn on resources that have been left off by a lazy BIOS */ if (type == SYS_RES_IOPORT && !pci_porten(dev)) { cmd = pci_read_config(dev, PCIR_COMMAND, 2); cmd |= PCIM_CMD_PORTEN; pci_write_config(dev, PCIR_COMMAND, cmd, 2); } if (type == SYS_RES_MEMORY && !pci_memen(dev)) { cmd = pci_read_config(dev, PCIR_COMMAND, 2); cmd |= PCIM_CMD_MEMEN; pci_write_config(dev, PCIR_COMMAND, cmd, 2); } } else { if (type == SYS_RES_IOPORT && !pci_porten(dev)) return (barlen); if (type == SYS_RES_MEMORY && !pci_memen(dev)) return (barlen); } count = (pci_addr_t)1 << mapsize; if (basezero || base == pci_mapbase(testval)) { start = 0; /* Let the parent decide. */ end = ~0ULL; } else { start = base; end = base + count - 1; } resource_list_add(rl, type, reg, start, end, count); /* * Try to allocate the resource for this BAR from our parent * so that this resource range is already reserved. The * driver for this device will later inherit this resource in * pci_alloc_resource(). */ res = resource_list_alloc(rl, bus, dev, type, ®, start, end, count, prefetch ? RF_PREFETCHABLE : 0); if (res == NULL) { /* * If the allocation fails, clear the BAR and delete * the resource list entry to force * pci_alloc_resource() to allocate resources from the * parent. */ resource_list_delete(rl, type, reg); start = 0; } else { start = rman_get_start(res); rman_set_device(res, bus); } pci_write_bar(dev, reg, start); return (barlen); } /* * For ATA devices we need to decide early what addressing mode to use. * Legacy demands that the primary and secondary ATA ports sits on the * same addresses that old ISA hardware did. This dictates that we use * those addresses and ignore the BAR's if we cannot set PCI native * addressing mode. */ static void pci_ata_maps(device_t bus, device_t dev, struct resource_list *rl, int force, uint32_t prefetchmask) { struct resource *r; int rid, type, progif; #if 0 /* if this device supports PCI native addressing use it */ progif = pci_read_config(dev, PCIR_PROGIF, 1); if ((progif & 0x8a) == 0x8a) { if (pci_mapbase(pci_read_config(dev, PCIR_BAR(0), 4)) && pci_mapbase(pci_read_config(dev, PCIR_BAR(2), 4))) { printf("Trying ATA native PCI addressing mode\n"); pci_write_config(dev, PCIR_PROGIF, progif | 0x05, 1); } } #endif progif = pci_read_config(dev, PCIR_PROGIF, 1); type = SYS_RES_IOPORT; if (progif & PCIP_STORAGE_IDE_MODEPRIM) { pci_add_map(bus, dev, PCIR_BAR(0), rl, force, prefetchmask & (1 << 0)); pci_add_map(bus, dev, PCIR_BAR(1), rl, force, prefetchmask & (1 << 1)); } else { rid = PCIR_BAR(0); resource_list_add(rl, type, rid, 0x1f0, 0x1f7, 8); r = resource_list_alloc(rl, bus, dev, type, &rid, 0x1f0, 0x1f7, 8, 0); rman_set_device(r, bus); rid = PCIR_BAR(1); resource_list_add(rl, type, rid, 0x3f6, 0x3f6, 1); r = resource_list_alloc(rl, bus, dev, type, &rid, 0x3f6, 0x3f6, 1, 0); rman_set_device(r, bus); } if (progif & PCIP_STORAGE_IDE_MODESEC) { pci_add_map(bus, dev, PCIR_BAR(2), rl, force, prefetchmask & (1 << 2)); pci_add_map(bus, dev, PCIR_BAR(3), rl, force, prefetchmask & (1 << 3)); } else { rid = PCIR_BAR(2); resource_list_add(rl, type, rid, 0x170, 0x177, 8); r = resource_list_alloc(rl, bus, dev, type, &rid, 0x170, 0x177, 8, 0); rman_set_device(r, bus); rid = PCIR_BAR(3); resource_list_add(rl, type, rid, 0x376, 0x376, 1); r = resource_list_alloc(rl, bus, dev, type, &rid, 0x376, 0x376, 1, 0); rman_set_device(r, bus); } pci_add_map(bus, dev, PCIR_BAR(4), rl, force, prefetchmask & (1 << 4)); pci_add_map(bus, dev, PCIR_BAR(5), rl, force, prefetchmask & (1 << 5)); } static void pci_assign_interrupt(device_t bus, device_t dev, int force_route) { struct pci_devinfo *dinfo = device_get_ivars(dev); pcicfgregs *cfg = &dinfo->cfg; char tunable_name[64]; int irq; /* Has to have an intpin to have an interrupt. */ if (cfg->intpin == 0) return; /* Let the user override the IRQ with a tunable. */ irq = PCI_INVALID_IRQ; snprintf(tunable_name, sizeof(tunable_name), "hw.pci%d.%d.%d.INT%c.irq", cfg->domain, cfg->bus, cfg->slot, cfg->intpin + 'A' - 1); if (TUNABLE_INT_FETCH(tunable_name, &irq) && (irq >= 255 || irq <= 0)) irq = PCI_INVALID_IRQ; /* * If we didn't get an IRQ via the tunable, then we either use the * IRQ value in the intline register or we ask the bus to route an * interrupt for us. If force_route is true, then we only use the * value in the intline register if the bus was unable to assign an * IRQ. */ if (!PCI_INTERRUPT_VALID(irq)) { if (!PCI_INTERRUPT_VALID(cfg->intline) || force_route) irq = PCI_ASSIGN_INTERRUPT(bus, dev); if (!PCI_INTERRUPT_VALID(irq)) irq = cfg->intline; } /* If after all that we don't have an IRQ, just bail. */ if (!PCI_INTERRUPT_VALID(irq)) return; /* Update the config register if it changed. */ if (irq != cfg->intline) { cfg->intline = irq; pci_write_config(dev, PCIR_INTLINE, irq, 1); } /* Add this IRQ as rid 0 interrupt resource. */ resource_list_add(&dinfo->resources, SYS_RES_IRQ, 0, irq, irq, 1); } void pci_add_resources(device_t bus, device_t dev, int force, uint32_t prefetchmask) { struct pci_devinfo *dinfo = device_get_ivars(dev); pcicfgregs *cfg = &dinfo->cfg; struct resource_list *rl = &dinfo->resources; const struct pci_quirk *q; int i; /* ATA devices needs special map treatment */ if ((pci_get_class(dev) == PCIC_STORAGE) && (pci_get_subclass(dev) == PCIS_STORAGE_IDE) && ((pci_get_progif(dev) & PCIP_STORAGE_IDE_MASTERDEV) || (!pci_read_config(dev, PCIR_BAR(0), 4) && !pci_read_config(dev, PCIR_BAR(2), 4))) ) pci_ata_maps(bus, dev, rl, force, prefetchmask); else for (i = 0; i < cfg->nummaps;) i += pci_add_map(bus, dev, PCIR_BAR(i), rl, force, prefetchmask & (1 << i)); /* * Add additional, quirked resources. */ for (q = &pci_quirks[0]; q->devid; q++) { if (q->devid == ((cfg->device << 16) | cfg->vendor) && q->type == PCI_QUIRK_MAP_REG) pci_add_map(bus, dev, q->arg1, rl, force, 0); } if (cfg->intpin > 0 && PCI_INTERRUPT_VALID(cfg->intline)) { #ifdef __PCI_REROUTE_INTERRUPT /* * Try to re-route interrupts. Sometimes the BIOS or * firmware may leave bogus values in these registers. * If the re-route fails, then just stick with what we * have. */ pci_assign_interrupt(bus, dev, 1); #else pci_assign_interrupt(bus, dev, 0); #endif } } void pci_add_children(device_t dev, int domain, int busno, size_t dinfo_size) { #define REG(n, w) PCIB_READ_CONFIG(pcib, busno, s, f, n, w) device_t pcib = device_get_parent(dev); struct pci_devinfo *dinfo; int maxslots; int s, f, pcifunchigh; uint8_t hdrtype; KASSERT(dinfo_size >= sizeof(struct pci_devinfo), ("dinfo_size too small")); maxslots = PCIB_MAXSLOTS(pcib); for (s = 0; s <= maxslots; s++) { pcifunchigh = 0; f = 0; DELAY(1); hdrtype = REG(PCIR_HDRTYPE, 1); if ((hdrtype & PCIM_HDRTYPE) > PCI_MAXHDRTYPE) continue; if (hdrtype & PCIM_MFDEV) pcifunchigh = PCI_FUNCMAX; for (f = 0; f <= pcifunchigh; f++) { dinfo = pci_read_device(pcib, domain, busno, s, f, dinfo_size); if (dinfo != NULL) { pci_add_child(dev, dinfo); } } } #undef REG } void pci_add_child(device_t bus, struct pci_devinfo *dinfo) { dinfo->cfg.dev = device_add_child(bus, NULL, -1); device_set_ivars(dinfo->cfg.dev, dinfo); resource_list_init(&dinfo->resources); pci_cfg_save(dinfo->cfg.dev, dinfo, 0); pci_cfg_restore(dinfo->cfg.dev, dinfo); pci_print_verbose(dinfo); pci_add_resources(bus, dinfo->cfg.dev, 0, 0); } static int pci_probe(device_t dev) { device_set_desc(dev, "PCI bus"); /* Allow other subclasses to override this driver. */ return (-1000); } static int pci_attach(device_t dev) { int busno, domain; /* * Since there can be multiple independantly numbered PCI * busses on systems with multiple PCI domains, we can't use * the unit number to decide which bus we are probing. We ask * the parent pcib what our domain and bus numbers are. */ domain = pcib_get_domain(dev); busno = pcib_get_bus(dev); if (bootverbose) device_printf(dev, "domain=%d, physical bus=%d\n", domain, busno); pci_add_children(dev, domain, busno, sizeof(struct pci_devinfo)); return (bus_generic_attach(dev)); } int pci_suspend(device_t dev) { int dstate, error, i, numdevs; device_t acpi_dev, child, *devlist; struct pci_devinfo *dinfo; /* * Save the PCI configuration space for each child and set the * device in the appropriate power state for this sleep state. */ acpi_dev = NULL; if (pci_do_power_resume) acpi_dev = devclass_get_device(devclass_find("acpi"), 0); device_get_children(dev, &devlist, &numdevs); for (i = 0; i < numdevs; i++) { child = devlist[i]; dinfo = (struct pci_devinfo *) device_get_ivars(child); pci_cfg_save(child, dinfo, 0); } /* Suspend devices before potentially powering them down. */ error = bus_generic_suspend(dev); if (error) { free(devlist, M_TEMP); return (error); } /* * Always set the device to D3. If ACPI suggests a different * power state, use it instead. If ACPI is not present, the * firmware is responsible for managing device power. Skip * children who aren't attached since they are powered down * separately. Only manage type 0 devices for now. */ for (i = 0; acpi_dev && i < numdevs; i++) { child = devlist[i]; dinfo = (struct pci_devinfo *) device_get_ivars(child); if (device_is_attached(child) && dinfo->cfg.hdrtype == 0) { dstate = PCI_POWERSTATE_D3; ACPI_PWR_FOR_SLEEP(acpi_dev, child, &dstate); pci_set_powerstate(child, dstate); } } free(devlist, M_TEMP); return (0); } int pci_resume(device_t dev) { int i, numdevs; device_t acpi_dev, child, *devlist; struct pci_devinfo *dinfo; /* * Set each child to D0 and restore its PCI configuration space. */ acpi_dev = NULL; if (pci_do_power_resume) acpi_dev = devclass_get_device(devclass_find("acpi"), 0); device_get_children(dev, &devlist, &numdevs); for (i = 0; i < numdevs; i++) { /* * Notify ACPI we're going to D0 but ignore the result. If * ACPI is not present, the firmware is responsible for * managing device power. Only manage type 0 devices for now. */ child = devlist[i]; dinfo = (struct pci_devinfo *) device_get_ivars(child); if (acpi_dev && device_is_attached(child) && dinfo->cfg.hdrtype == 0) { ACPI_PWR_FOR_SLEEP(acpi_dev, child, NULL); pci_set_powerstate(child, PCI_POWERSTATE_D0); } /* Now the device is powered up, restore its config space. */ pci_cfg_restore(child, dinfo); } free(devlist, M_TEMP); return (bus_generic_resume(dev)); } static void pci_load_vendor_data(void) { caddr_t vendordata, info; if ((vendordata = preload_search_by_type("pci_vendor_data")) != NULL) { info = preload_search_info(vendordata, MODINFO_ADDR); pci_vendordata = *(char **)info; info = preload_search_info(vendordata, MODINFO_SIZE); pci_vendordata_size = *(size_t *)info; /* terminate the database */ pci_vendordata[pci_vendordata_size] = '\n'; } } void pci_driver_added(device_t dev, driver_t *driver) { int numdevs; device_t *devlist; device_t child; struct pci_devinfo *dinfo; int i; if (bootverbose) device_printf(dev, "driver added\n"); DEVICE_IDENTIFY(driver, dev); device_get_children(dev, &devlist, &numdevs); for (i = 0; i < numdevs; i++) { child = devlist[i]; if (device_get_state(child) != DS_NOTPRESENT) continue; dinfo = device_get_ivars(child); pci_print_verbose(dinfo); if (bootverbose) printf("pci%d:%d:%d:%d: reprobing on driver added\n", dinfo->cfg.domain, dinfo->cfg.bus, dinfo->cfg.slot, dinfo->cfg.func); pci_cfg_restore(child, dinfo); if (device_probe_and_attach(child) != 0) pci_cfg_save(child, dinfo, 1); } free(devlist, M_TEMP); } int pci_setup_intr(device_t dev, device_t child, struct resource *irq, int flags, driver_filter_t *filter, driver_intr_t *intr, void *arg, void **cookiep) { struct pci_devinfo *dinfo; struct msix_table_entry *mte; struct msix_vector *mv; uint64_t addr; uint32_t data; void *cookie; int error, rid; error = bus_generic_setup_intr(dev, child, irq, flags, filter, intr, arg, &cookie); if (error) return (error); /* If this is not a direct child, just bail out. */ if (device_get_parent(child) != dev) { *cookiep = cookie; return(0); } rid = rman_get_rid(irq); if (rid == 0) { /* Make sure that INTx is enabled */ pci_clear_command_bit(dev, child, PCIM_CMD_INTxDIS); } else { /* * Check to see if the interrupt is MSI or MSI-X. * Ask our parent to map the MSI and give * us the address and data register values. * If we fail for some reason, teardown the * interrupt handler. */ dinfo = device_get_ivars(child); if (dinfo->cfg.msi.msi_alloc > 0) { if (dinfo->cfg.msi.msi_addr == 0) { KASSERT(dinfo->cfg.msi.msi_handlers == 0, ("MSI has handlers, but vectors not mapped")); error = PCIB_MAP_MSI(device_get_parent(dev), child, rman_get_start(irq), &addr, &data); if (error) goto bad; dinfo->cfg.msi.msi_addr = addr; dinfo->cfg.msi.msi_data = data; } if (dinfo->cfg.msi.msi_handlers == 0) pci_enable_msi(child, dinfo->cfg.msi.msi_addr, dinfo->cfg.msi.msi_data); dinfo->cfg.msi.msi_handlers++; } else { KASSERT(dinfo->cfg.msix.msix_alloc > 0, ("No MSI or MSI-X interrupts allocated")); KASSERT(rid <= dinfo->cfg.msix.msix_table_len, ("MSI-X index too high")); mte = &dinfo->cfg.msix.msix_table[rid - 1]; KASSERT(mte->mte_vector != 0, ("no message vector")); mv = &dinfo->cfg.msix.msix_vectors[mte->mte_vector - 1]; KASSERT(mv->mv_irq == rman_get_start(irq), ("IRQ mismatch")); if (mv->mv_address == 0) { KASSERT(mte->mte_handlers == 0, ("MSI-X table entry has handlers, but vector not mapped")); error = PCIB_MAP_MSI(device_get_parent(dev), child, rman_get_start(irq), &addr, &data); if (error) goto bad; mv->mv_address = addr; mv->mv_data = data; } if (mte->mte_handlers == 0) { pci_enable_msix(child, rid - 1, mv->mv_address, mv->mv_data); pci_unmask_msix(child, rid - 1); } mte->mte_handlers++; } /* Make sure that INTx is disabled if we are using MSI/MSIX */ pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS); bad: if (error) { (void)bus_generic_teardown_intr(dev, child, irq, cookie); return (error); } } *cookiep = cookie; return (0); } int pci_teardown_intr(device_t dev, device_t child, struct resource *irq, void *cookie) { struct msix_table_entry *mte; struct resource_list_entry *rle; struct pci_devinfo *dinfo; int error, rid; if (irq == NULL || !(rman_get_flags(irq) & RF_ACTIVE)) return (EINVAL); /* If this isn't a direct child, just bail out */ if (device_get_parent(child) != dev) return(bus_generic_teardown_intr(dev, child, irq, cookie)); rid = rman_get_rid(irq); if (rid == 0) { /* Mask INTx */ pci_set_command_bit(dev, child, PCIM_CMD_INTxDIS); } else { /* * Check to see if the interrupt is MSI or MSI-X. If so, * decrement the appropriate handlers count and mask the * MSI-X message, or disable MSI messages if the count * drops to 0. */ dinfo = device_get_ivars(child); rle = resource_list_find(&dinfo->resources, SYS_RES_IRQ, rid); if (rle->res != irq) return (EINVAL); if (dinfo->cfg.msi.msi_alloc > 0) { KASSERT(rid <= dinfo->cfg.msi.msi_alloc, ("MSI-X index too high")); if (dinfo->cfg.msi.msi_handlers == 0) return (EINVAL); dinfo->cfg.msi.msi_handlers--; if (dinfo->cfg.msi.msi_handlers == 0) pci_disable_msi(child); } else { KASSERT(dinfo->cfg.msix.msix_alloc > 0, ("No MSI or MSI-X interrupts allocated")); KASSERT(rid <= dinfo->cfg.msix.msix_table_len, ("MSI-X index too high")); mte = &dinfo->cfg.msix.msix_table[rid - 1]; if (mte->mte_handlers == 0) return (EINVAL); mte->mte_handlers--; if (mte->mte_handlers == 0) pci_mask_msix(child, rid - 1); } } error = bus_generic_teardown_intr(dev, child, irq, cookie); if (rid > 0) KASSERT(error == 0, ("%s: generic teardown failed for MSI/MSI-X", __func__)); return (error); } int pci_print_child(device_t dev, device_t child) { struct pci_devinfo *dinfo; struct resource_list *rl; int retval = 0; dinfo = device_get_ivars(child); rl = &dinfo->resources; retval += bus_print_child_header(dev, child); retval += resource_list_print_type(rl, "port", SYS_RES_IOPORT, "%#lx"); retval += resource_list_print_type(rl, "mem", SYS_RES_MEMORY, "%#lx"); retval += resource_list_print_type(rl, "irq", SYS_RES_IRQ, "%ld"); if (device_get_flags(dev)) retval += printf(" flags %#x", device_get_flags(dev)); retval += printf(" at device %d.%d", pci_get_slot(child), pci_get_function(child)); retval += bus_print_child_footer(dev, child); return (retval); } static struct { int class; int subclass; char *desc; } pci_nomatch_tab[] = { {PCIC_OLD, -1, "old"}, {PCIC_OLD, PCIS_OLD_NONVGA, "non-VGA display device"}, {PCIC_OLD, PCIS_OLD_VGA, "VGA-compatible display device"}, {PCIC_STORAGE, -1, "mass storage"}, {PCIC_STORAGE, PCIS_STORAGE_SCSI, "SCSI"}, {PCIC_STORAGE, PCIS_STORAGE_IDE, "ATA"}, {PCIC_STORAGE, PCIS_STORAGE_FLOPPY, "floppy disk"}, {PCIC_STORAGE, PCIS_STORAGE_IPI, "IPI"}, {PCIC_STORAGE, PCIS_STORAGE_RAID, "RAID"}, {PCIC_STORAGE, PCIS_STORAGE_ATA_ADMA, "ATA (ADMA)"}, {PCIC_STORAGE, PCIS_STORAGE_SATA, "SATA"}, {PCIC_STORAGE, PCIS_STORAGE_SAS, "SAS"}, {PCIC_STORAGE, PCIS_STORAGE_NVM, "NVM"}, {PCIC_NETWORK, -1, "network"}, {PCIC_NETWORK, PCIS_NETWORK_ETHERNET, "ethernet"}, {PCIC_NETWORK, PCIS_NETWORK_TOKENRING, "token ring"}, {PCIC_NETWORK, PCIS_NETWORK_FDDI, "fddi"}, {PCIC_NETWORK, PCIS_NETWORK_ATM, "ATM"}, {PCIC_NETWORK, PCIS_NETWORK_ISDN, "ISDN"}, {PCIC_DISPLAY, -1, "display"}, {PCIC_DISPLAY, PCIS_DISPLAY_VGA, "VGA"}, {PCIC_DISPLAY, PCIS_DISPLAY_XGA, "XGA"}, {PCIC_DISPLAY, PCIS_DISPLAY_3D, "3D"}, {PCIC_MULTIMEDIA, -1, "multimedia"}, {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_VIDEO, "video"}, {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_AUDIO, "audio"}, {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_TELE, "telephony"}, {PCIC_MULTIMEDIA, PCIS_MULTIMEDIA_HDA, "HDA"}, {PCIC_MEMORY, -1, "memory"}, {PCIC_MEMORY, PCIS_MEMORY_RAM, "RAM"}, {PCIC_MEMORY, PCIS_MEMORY_FLASH, "flash"}, {PCIC_BRIDGE, -1, "bridge"}, {PCIC_BRIDGE, PCIS_BRIDGE_HOST, "HOST-PCI"}, {PCIC_BRIDGE, PCIS_BRIDGE_ISA, "PCI-ISA"}, {PCIC_BRIDGE, PCIS_BRIDGE_EISA, "PCI-EISA"}, {PCIC_BRIDGE, PCIS_BRIDGE_MCA, "PCI-MCA"}, {PCIC_BRIDGE, PCIS_BRIDGE_PCI, "PCI-PCI"}, {PCIC_BRIDGE, PCIS_BRIDGE_PCMCIA, "PCI-PCMCIA"}, {PCIC_BRIDGE, PCIS_BRIDGE_NUBUS, "PCI-NuBus"}, {PCIC_BRIDGE, PCIS_BRIDGE_CARDBUS, "PCI-CardBus"}, {PCIC_BRIDGE, PCIS_BRIDGE_RACEWAY, "PCI-RACEway"}, {PCIC_SIMPLECOMM, -1, "simple comms"}, {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_UART, "UART"}, /* could detect 16550 */ {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_PAR, "parallel port"}, {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MULSER, "multiport serial"}, {PCIC_SIMPLECOMM, PCIS_SIMPLECOMM_MODEM, "generic modem"}, {PCIC_BASEPERIPH, -1, "base peripheral"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PIC, "interrupt controller"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_DMA, "DMA controller"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_TIMER, "timer"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_RTC, "realtime clock"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_PCIHOT, "PCI hot-plug controller"}, {PCIC_BASEPERIPH, PCIS_BASEPERIPH_SDHC, "SD host controller"}, {PCIC_INPUTDEV, -1, "input device"}, {PCIC_INPUTDEV, PCIS_INPUTDEV_KEYBOARD, "keyboard"}, {PCIC_INPUTDEV, PCIS_INPUTDEV_DIGITIZER,"digitizer"}, {PCIC_INPUTDEV, PCIS_INPUTDEV_MOUSE, "mouse"}, {PCIC_INPUTDEV, PCIS_INPUTDEV_SCANNER, "scanner"}, {PCIC_INPUTDEV, PCIS_INPUTDEV_GAMEPORT, "gameport"}, {PCIC_DOCKING, -1, "docking station"}, {PCIC_PROCESSOR, -1, "processor"}, {PCIC_SERIALBUS, -1, "serial bus"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_FW, "FireWire"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_ACCESS, "AccessBus"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_SSA, "SSA"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_USB, "USB"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_FC, "Fibre Channel"}, {PCIC_SERIALBUS, PCIS_SERIALBUS_SMBUS, "SMBus"}, {PCIC_WIRELESS, -1, "wireless controller"}, {PCIC_WIRELESS, PCIS_WIRELESS_IRDA, "iRDA"}, {PCIC_WIRELESS, PCIS_WIRELESS_IR, "IR"}, {PCIC_WIRELESS, PCIS_WIRELESS_RF, "RF"}, {PCIC_INTELLIIO, -1, "intelligent I/O controller"}, {PCIC_INTELLIIO, PCIS_INTELLIIO_I2O, "I2O"}, {PCIC_SATCOM, -1, "satellite communication"}, {PCIC_SATCOM, PCIS_SATCOM_TV, "sat TV"}, {PCIC_SATCOM, PCIS_SATCOM_AUDIO, "sat audio"}, {PCIC_SATCOM, PCIS_SATCOM_VOICE, "sat voice"}, {PCIC_SATCOM, PCIS_SATCOM_DATA, "sat data"}, {PCIC_CRYPTO, -1, "encrypt/decrypt"}, {PCIC_CRYPTO, PCIS_CRYPTO_NETCOMP, "network/computer crypto"}, {PCIC_CRYPTO, PCIS_CRYPTO_ENTERTAIN, "entertainment crypto"}, {PCIC_DASP, -1, "dasp"}, {PCIC_DASP, PCIS_DASP_DPIO, "DPIO module"}, {0, 0, NULL} }; void pci_probe_nomatch(device_t dev, device_t child) { int i; char *cp, *scp, *device; /* * Look for a listing for this device in a loaded device database. */ if ((device = pci_describe_device(child)) != NULL) { device_printf(dev, "<%s>", device); free(device, M_DEVBUF); } else { /* * Scan the class/subclass descriptions for a general * description. */ cp = "unknown"; scp = NULL; for (i = 0; pci_nomatch_tab[i].desc != NULL; i++) { if (pci_nomatch_tab[i].class == pci_get_class(child)) { if (pci_nomatch_tab[i].subclass == -1) { cp = pci_nomatch_tab[i].desc; } else if (pci_nomatch_tab[i].subclass == pci_get_subclass(child)) { scp = pci_nomatch_tab[i].desc; } } } device_printf(dev, "<%s%s%s>", cp ? cp : "", ((cp != NULL) && (scp != NULL)) ? ", " : "", scp ? scp : ""); } printf(" at device %d.%d (no driver attached)\n", pci_get_slot(child), pci_get_function(child)); pci_cfg_save(child, (struct pci_devinfo *)device_get_ivars(child), 1); return; } /* * Parse the PCI device database, if loaded, and return a pointer to a * description of the device. * * The database is flat text formatted as follows: * * Any line not in a valid format is ignored. * Lines are terminated with newline '\n' characters. * * A VENDOR line consists of the 4 digit (hex) vendor code, a TAB, then * the vendor name. * * A DEVICE line is entered immediately below the corresponding VENDOR ID. * - devices cannot be listed without a corresponding VENDOR line. * A DEVICE line consists of a TAB, the 4 digit (hex) device code, * another TAB, then the device name. */ /* * Assuming (ptr) points to the beginning of a line in the database, * return the vendor or device and description of the next entry. * The value of (vendor) or (device) inappropriate for the entry type * is set to -1. Returns nonzero at the end of the database. * * Note that this is slightly unrobust in the face of corrupt data; * we attempt to safeguard against this by spamming the end of the * database with a newline when we initialise. */ static int pci_describe_parse_line(char **ptr, int *vendor, int *device, char **desc) { char *cp = *ptr; int left; *device = -1; *vendor = -1; **desc = '\0'; for (;;) { left = pci_vendordata_size - (cp - pci_vendordata); if (left <= 0) { *ptr = cp; return(1); } /* vendor entry? */ if (*cp != '\t' && sscanf(cp, "%x\t%80[^\n]", vendor, *desc) == 2) break; /* device entry? */ if (*cp == '\t' && sscanf(cp, "%x\t%80[^\n]", device, *desc) == 2) break; /* skip to next line */ while (*cp != '\n' && left > 0) { cp++; left--; } if (*cp == '\n') { cp++; left--; } } /* skip to next line */ while (*cp != '\n' && left > 0) { cp++; left--; } if (*cp == '\n' && left > 0) cp++; *ptr = cp; return(0); } static char * pci_describe_device(device_t dev) { int vendor, device; char *desc, *vp, *dp, *line; desc = vp = dp = NULL; /* * If we have no vendor data, we can't do anything. */ if (pci_vendordata == NULL) goto out; /* * Scan the vendor data looking for this device */ line = pci_vendordata; if ((vp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL) goto out; for (;;) { if (pci_describe_parse_line(&line, &vendor, &device, &vp)) goto out; if (vendor == pci_get_vendor(dev)) break; } if ((dp = malloc(80, M_DEVBUF, M_NOWAIT)) == NULL) goto out; for (;;) { if (pci_describe_parse_line(&line, &vendor, &device, &dp)) { *dp = 0; break; } if (vendor != -1) { *dp = 0; break; } if (device == pci_get_device(dev)) break; } if (dp[0] == '\0') snprintf(dp, 80, "0x%x", pci_get_device(dev)); if ((desc = malloc(strlen(vp) + strlen(dp) + 3, M_DEVBUF, M_NOWAIT)) != NULL) sprintf(desc, "%s, %s", vp, dp); out: if (vp != NULL) free(vp, M_DEVBUF); if (dp != NULL) free(dp, M_DEVBUF); return(desc); } int pci_read_ivar(device_t dev, device_t child, int which, uintptr_t *result) { struct pci_devinfo *dinfo; pcicfgregs *cfg; dinfo = device_get_ivars(child); cfg = &dinfo->cfg; switch (which) { case PCI_IVAR_ETHADDR: /* * The generic accessor doesn't deal with failure, so * we set the return value, then return an error. */ *((uint8_t **) result) = NULL; return (EINVAL); case PCI_IVAR_SUBVENDOR: *result = cfg->subvendor; break; case PCI_IVAR_SUBDEVICE: *result = cfg->subdevice; break; case PCI_IVAR_VENDOR: *result = cfg->vendor; break; case PCI_IVAR_DEVICE: *result = cfg->device; break; case PCI_IVAR_DEVID: *result = (cfg->device << 16) | cfg->vendor; break; case PCI_IVAR_CLASS: *result = cfg->baseclass; break; case PCI_IVAR_SUBCLASS: *result = cfg->subclass; break; case PCI_IVAR_PROGIF: *result = cfg->progif; break; case PCI_IVAR_REVID: *result = cfg->revid; break; case PCI_IVAR_INTPIN: *result = cfg->intpin; break; case PCI_IVAR_IRQ: *result = cfg->intline; break; case PCI_IVAR_DOMAIN: *result = cfg->domain; break; case PCI_IVAR_BUS: *result = cfg->bus; break; case PCI_IVAR_SLOT: *result = cfg->slot; break; case PCI_IVAR_FUNCTION: *result = cfg->func; break; case PCI_IVAR_CMDREG: *result = cfg->cmdreg; break; case PCI_IVAR_CACHELNSZ: *result = cfg->cachelnsz; break; case PCI_IVAR_MINGNT: *result = cfg->mingnt; break; case PCI_IVAR_MAXLAT: *result = cfg->maxlat; break; case PCI_IVAR_LATTIMER: *result = cfg->lattimer; break; default: return (ENOENT); } return (0); } int pci_write_ivar(device_t dev, device_t child, int which, uintptr_t value) { struct pci_devinfo *dinfo; dinfo = device_get_ivars(child); switch (which) { case PCI_IVAR_INTPIN: dinfo->cfg.intpin = value; return (0); case PCI_IVAR_ETHADDR: case PCI_IVAR_SUBVENDOR: case PCI_IVAR_SUBDEVICE: case PCI_IVAR_VENDOR: case PCI_IVAR_DEVICE: case PCI_IVAR_DEVID: case PCI_IVAR_CLASS: case PCI_IVAR_SUBCLASS: case PCI_IVAR_PROGIF: case PCI_IVAR_REVID: case PCI_IVAR_IRQ: case PCI_IVAR_DOMAIN: case PCI_IVAR_BUS: case PCI_IVAR_SLOT: case PCI_IVAR_FUNCTION: return (EINVAL); /* disallow for now */ default: return (ENOENT); } } #include "opt_ddb.h" #ifdef DDB #include #include /* * List resources based on pci map registers, used for within ddb */ DB_SHOW_COMMAND(pciregs, db_pci_dump) { struct pci_devinfo *dinfo; struct devlist *devlist_head; struct pci_conf *p; const char *name; int i, error, none_count; none_count = 0; /* get the head of the device queue */ devlist_head = &pci_devq; /* * Go through the list of devices and print out devices */ for (error = 0, i = 0, dinfo = STAILQ_FIRST(devlist_head); (dinfo != NULL) && (error == 0) && (i < pci_numdevs) && !db_pager_quit; dinfo = STAILQ_NEXT(dinfo, pci_links), i++) { /* Populate pd_name and pd_unit */ name = NULL; if (dinfo->cfg.dev) name = device_get_name(dinfo->cfg.dev); p = &dinfo->conf; db_printf("%s%d@pci%d:%d:%d:%d:\tclass=0x%06x card=0x%08x " "chip=0x%08x rev=0x%02x hdr=0x%02x\n", (name && *name) ? name : "none", (name && *name) ? (int)device_get_unit(dinfo->cfg.dev) : none_count++, p->pc_sel.pc_domain, p->pc_sel.pc_bus, p->pc_sel.pc_dev, p->pc_sel.pc_func, (p->pc_class << 16) | (p->pc_subclass << 8) | p->pc_progif, (p->pc_subdevice << 16) | p->pc_subvendor, (p->pc_device << 16) | p->pc_vendor, p->pc_revid, p->pc_hdr); } } #endif /* DDB */ static struct resource * pci_alloc_map(device_t dev, device_t child, int type, int *rid, u_long start, u_long end, u_long count, u_int flags) { struct pci_devinfo *dinfo = device_get_ivars(child); struct resource_list *rl = &dinfo->resources; struct resource_list_entry *rle; struct resource *res; pci_addr_t map, testval; int mapsize; /* * Weed out the bogons, and figure out how large the BAR/map * is. Bars that read back 0 here are bogus and unimplemented. * Note: atapci in legacy mode are special and handled elsewhere * in the code. If you have a atapci device in legacy mode and * it fails here, that other code is broken. */ res = NULL; pci_read_bar(child, *rid, &map, &testval); /* Ignore a BAR with a base of 0. */ if (pci_mapbase(testval) == 0) goto out; if (PCI_BAR_MEM(testval)) { if (type != SYS_RES_MEMORY) { if (bootverbose) device_printf(dev, "child %s requested type %d for rid %#x," " but the BAR says it is an memio\n", device_get_nameunit(child), type, *rid); goto out; } } else { if (type != SYS_RES_IOPORT) { if (bootverbose) device_printf(dev, "child %s requested type %d for rid %#x," " but the BAR says it is an ioport\n", device_get_nameunit(child), type, *rid); goto out; } } /* * For real BARs, we need to override the size that * the driver requests, because that's what the BAR * actually uses and we would otherwise have a * situation where we might allocate the excess to * another driver, which won't work. */ mapsize = pci_mapsize(testval); count = (pci_addr_t)1 << mapsize; if (RF_ALIGNMENT(flags) < mapsize) flags = (flags & ~RF_ALIGNMENT_MASK) | RF_ALIGNMENT_LOG2(mapsize); if (PCI_BAR_MEM(testval) && (testval & PCIM_BAR_MEM_PREFETCH)) flags |= RF_PREFETCHABLE; /* * Allocate enough resource, and then write back the * appropriate bar for that resource. */ res = BUS_ALLOC_RESOURCE(device_get_parent(dev), child, type, rid, start, end, count, flags & ~RF_ACTIVE); if (res == NULL) { device_printf(child, "%#lx bytes of rid %#x res %d failed (%#lx, %#lx).\n", count, *rid, type, start, end); goto out; } rman_set_device(res, dev); resource_list_add(rl, type, *rid, start, end, count); rle = resource_list_find(rl, type, *rid); if (rle == NULL) panic("pci_alloc_map: unexpectedly can't find resource."); rle->res = res; rle->start = rman_get_start(res); rle->end = rman_get_end(res); rle->count = count; if (bootverbose) device_printf(child, "Lazy allocation of %#lx bytes rid %#x type %d at %#lx\n", count, *rid, type, rman_get_start(res)); map = rman_get_start(res); pci_write_bar(child, *rid, map); out:; return (res); } struct resource * pci_alloc_resource(device_t dev, device_t child, int type, int *rid, u_long start, u_long end, u_long count, u_int flags) { struct pci_devinfo *dinfo = device_get_ivars(child); struct resource_list *rl = &dinfo->resources; struct resource_list_entry *rle; struct resource *res; pcicfgregs *cfg = &dinfo->cfg; if (device_get_parent(child) != dev) return (BUS_ALLOC_RESOURCE(device_get_parent(dev), child, type, rid, start, end, count, flags)); /* * Perform lazy resource allocation */ switch (type) { case SYS_RES_IRQ: /* * Can't alloc legacy interrupt once MSI messages have * been allocated. */ if (*rid == 0 && (cfg->msi.msi_alloc > 0 || cfg->msix.msix_alloc > 0)) return (NULL); /* * If the child device doesn't have an interrupt * routed and is deserving of an interrupt, try to * assign it one. */ if (*rid == 0 && !PCI_INTERRUPT_VALID(cfg->intline) && (cfg->intpin != 0)) pci_assign_interrupt(dev, child, 0); break; case SYS_RES_IOPORT: case SYS_RES_MEMORY: /* Allocate resources for this BAR if needed. */ rle = resource_list_find(rl, type, *rid); if (rle == NULL) { res = pci_alloc_map(dev, child, type, rid, start, end, count, flags); if (res == NULL) return (NULL); rle = resource_list_find(rl, type, *rid); } /* * If the resource belongs to the bus, then give it to * the child. We need to activate it if requested * since the bus always allocates inactive resources. */ if (rle != NULL && rle->res != NULL && rman_get_device(rle->res) == dev) { if (bootverbose) device_printf(child, "Reserved %#lx bytes for rid %#x type %d at %#lx\n", rman_get_size(rle->res), *rid, type, rman_get_start(rle->res)); rman_set_device(rle->res, child); if ((flags & RF_ACTIVE) && bus_activate_resource(child, type, *rid, rle->res) != 0) return (NULL); return (rle->res); } } return (resource_list_alloc(rl, dev, child, type, rid, start, end, count, flags)); } int pci_release_resource(device_t dev, device_t child, int type, int rid, struct resource *r) { int error; if (device_get_parent(child) != dev) return (BUS_RELEASE_RESOURCE(device_get_parent(dev), child, type, rid, r)); /* * For BARs we don't actually want to release the resource. * Instead, we deactivate the resource if needed and then give * ownership of the BAR back to the bus. */ switch (type) { case SYS_RES_IOPORT: case SYS_RES_MEMORY: if (rman_get_device(r) != child) return (EINVAL); if (rman_get_flags(r) & RF_ACTIVE) { error = bus_deactivate_resource(child, type, rid, r); if (error) return (error); } rman_set_device(r, dev); return (0); } return (bus_generic_rl_release_resource(dev, child, type, rid, r)); } int pci_activate_resource(device_t dev, device_t child, int type, int rid, struct resource *r) { int error; error = bus_generic_activate_resource(dev, child, type, rid, r); if (error) return (error); /* Enable decoding in the command register when activating BARs. */ if (device_get_parent(child) == dev) { switch (type) { case SYS_RES_IOPORT: case SYS_RES_MEMORY: error = PCI_ENABLE_IO(dev, child, type); break; } } return (error); } void pci_delete_resource(device_t dev, device_t child, int type, int rid) { struct pci_devinfo *dinfo; struct resource_list *rl; struct resource_list_entry *rle; if (device_get_parent(child) != dev) return; dinfo = device_get_ivars(child); rl = &dinfo->resources; rle = resource_list_find(rl, type, rid); if (rle == NULL) return; if (rle->res) { if (rman_get_device(rle->res) != dev || rman_get_flags(rle->res) & RF_ACTIVE) { device_printf(dev, "delete_resource: " "Resource still owned by child, oops. " "(type=%d, rid=%d, addr=%lx)\n", rle->type, rle->rid, rman_get_start(rle->res)); return; } #ifndef __PCI_BAR_ZERO_VALID /* * If this is a BAR, clear the BAR so it stops * decoding before releasing the resource. */ switch (type) { case SYS_RES_IOPORT: case SYS_RES_MEMORY: pci_write_bar(child, rid, 0); break; } #endif bus_release_resource(dev, type, rid, rle->res); } resource_list_delete(rl, type, rid); } struct resource_list * pci_get_resource_list (device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); return (&dinfo->resources); } uint32_t pci_read_config_method(device_t dev, device_t child, int reg, int width) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; return (PCIB_READ_CONFIG(device_get_parent(dev), cfg->bus, cfg->slot, cfg->func, reg, width)); } void pci_write_config_method(device_t dev, device_t child, int reg, uint32_t val, int width) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; PCIB_WRITE_CONFIG(device_get_parent(dev), cfg->bus, cfg->slot, cfg->func, reg, val, width); } int pci_child_location_str_method(device_t dev, device_t child, char *buf, size_t buflen) { snprintf(buf, buflen, "slot=%d function=%d", pci_get_slot(child), pci_get_function(child)); return (0); } int pci_child_pnpinfo_str_method(device_t dev, device_t child, char *buf, size_t buflen) { struct pci_devinfo *dinfo; pcicfgregs *cfg; dinfo = device_get_ivars(child); cfg = &dinfo->cfg; snprintf(buf, buflen, "vendor=0x%04x device=0x%04x subvendor=0x%04x " "subdevice=0x%04x class=0x%02x%02x%02x", cfg->vendor, cfg->device, cfg->subvendor, cfg->subdevice, cfg->baseclass, cfg->subclass, cfg->progif); return (0); } int pci_assign_interrupt_method(device_t dev, device_t child) { struct pci_devinfo *dinfo = device_get_ivars(child); pcicfgregs *cfg = &dinfo->cfg; return (PCIB_ROUTE_INTERRUPT(device_get_parent(dev), child, cfg->intpin)); } static int pci_modevent(module_t mod, int what, void *arg) { static struct cdev *pci_cdev; switch (what) { case MOD_LOAD: STAILQ_INIT(&pci_devq); pci_generation = 0; pci_cdev = make_dev(&pcicdev, 0, UID_ROOT, GID_WHEEL, 0644, "pci"); pci_load_vendor_data(); break; case MOD_UNLOAD: destroy_dev(pci_cdev); break; } return (0); } void pci_cfg_restore(device_t dev, struct pci_devinfo *dinfo) { int i; /* * Only do header type 0 devices. Type 1 devices are bridges, * which we know need special treatment. Type 2 devices are * cardbus bridges which also require special treatment. * Other types are unknown, and we err on the side of safety * by ignoring them. */ if (dinfo->cfg.hdrtype != 0) return; /* * Restore the device to full power mode. We must do this * before we restore the registers because moving from D3 to * D0 will cause the chip's BARs and some other registers to * be reset to some unknown power on reset values. Cut down * the noise on boot by doing nothing if we are already in * state D0. */ if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) { pci_set_powerstate(dev, PCI_POWERSTATE_D0); } for (i = 0; i < dinfo->cfg.nummaps; i++) pci_write_config(dev, PCIR_BAR(i), dinfo->cfg.bar[i], 4); pci_write_config(dev, PCIR_BIOS, dinfo->cfg.bios, 4); pci_write_config(dev, PCIR_COMMAND, dinfo->cfg.cmdreg, 2); pci_write_config(dev, PCIR_INTLINE, dinfo->cfg.intline, 1); pci_write_config(dev, PCIR_INTPIN, dinfo->cfg.intpin, 1); pci_write_config(dev, PCIR_MINGNT, dinfo->cfg.mingnt, 1); pci_write_config(dev, PCIR_MAXLAT, dinfo->cfg.maxlat, 1); pci_write_config(dev, PCIR_CACHELNSZ, dinfo->cfg.cachelnsz, 1); pci_write_config(dev, PCIR_LATTIMER, dinfo->cfg.lattimer, 1); pci_write_config(dev, PCIR_PROGIF, dinfo->cfg.progif, 1); pci_write_config(dev, PCIR_REVID, dinfo->cfg.revid, 1); /* Restore MSI and MSI-X configurations if they are present. */ if (dinfo->cfg.msi.msi_location != 0) pci_resume_msi(dev); if (dinfo->cfg.msix.msix_location != 0) pci_resume_msix(dev); } void pci_cfg_save(device_t dev, struct pci_devinfo *dinfo, int setstate) { int i; uint32_t cls; int ps; /* * Only do header type 0 devices. Type 1 devices are bridges, which * we know need special treatment. Type 2 devices are cardbus bridges * which also require special treatment. Other types are unknown, and * we err on the side of safety by ignoring them. Powering down * bridges should not be undertaken lightly. */ if (dinfo->cfg.hdrtype != 0) return; for (i = 0; i < dinfo->cfg.nummaps; i++) dinfo->cfg.bar[i] = pci_read_config(dev, PCIR_BAR(i), 4); dinfo->cfg.bios = pci_read_config(dev, PCIR_BIOS, 4); /* * Some drivers apparently write to these registers w/o updating our * cached copy. No harm happens if we update the copy, so do so here * so we can restore them. The COMMAND register is modified by the * bus w/o updating the cache. This should represent the normally * writable portion of the 'defined' part of type 0 headers. In * theory we also need to save/restore the PCI capability structures * we know about, but apart from power we don't know any that are * writable. */ dinfo->cfg.subvendor = pci_read_config(dev, PCIR_SUBVEND_0, 2); dinfo->cfg.subdevice = pci_read_config(dev, PCIR_SUBDEV_0, 2); dinfo->cfg.vendor = pci_read_config(dev, PCIR_VENDOR, 2); dinfo->cfg.device = pci_read_config(dev, PCIR_DEVICE, 2); dinfo->cfg.cmdreg = pci_read_config(dev, PCIR_COMMAND, 2); dinfo->cfg.intline = pci_read_config(dev, PCIR_INTLINE, 1); dinfo->cfg.intpin = pci_read_config(dev, PCIR_INTPIN, 1); dinfo->cfg.mingnt = pci_read_config(dev, PCIR_MINGNT, 1); dinfo->cfg.maxlat = pci_read_config(dev, PCIR_MAXLAT, 1); dinfo->cfg.cachelnsz = pci_read_config(dev, PCIR_CACHELNSZ, 1); dinfo->cfg.lattimer = pci_read_config(dev, PCIR_LATTIMER, 1); dinfo->cfg.baseclass = pci_read_config(dev, PCIR_CLASS, 1); dinfo->cfg.subclass = pci_read_config(dev, PCIR_SUBCLASS, 1); dinfo->cfg.progif = pci_read_config(dev, PCIR_PROGIF, 1); dinfo->cfg.revid = pci_read_config(dev, PCIR_REVID, 1); /* * don't set the state for display devices, base peripherals and * memory devices since bad things happen when they are powered down. * We should (a) have drivers that can easily detach and (b) use * generic drivers for these devices so that some device actually * attaches. We need to make sure that when we implement (a) we don't * power the device down on a reattach. */ cls = pci_get_class(dev); if (!setstate) return; switch (pci_do_power_nodriver) { case 0: /* NO powerdown at all */ return; case 1: /* Conservative about what to power down */ if (cls == PCIC_STORAGE) return; /*FALLTHROUGH*/ case 2: /* Agressive about what to power down */ if (cls == PCIC_DISPLAY || cls == PCIC_MEMORY || cls == PCIC_BASEPERIPH) return; /*FALLTHROUGH*/ case 3: /* Power down everything */ break; } /* * PCI spec says we can only go into D3 state from D0 state. * Transition from D[12] into D0 before going to D3 state. */ ps = pci_get_powerstate(dev); if (ps != PCI_POWERSTATE_D0 && ps != PCI_POWERSTATE_D3) pci_set_powerstate(dev, PCI_POWERSTATE_D0); if (pci_get_powerstate(dev) != PCI_POWERSTATE_D3) pci_set_powerstate(dev, PCI_POWERSTATE_D3); } Index: stable/7/sys/dev/pci/pcireg.h =================================================================== --- stable/7/sys/dev/pci/pcireg.h (revision 242221) +++ stable/7/sys/dev/pci/pcireg.h (revision 242222) @@ -1,753 +1,955 @@ /*- * Copyright (c) 1997, Stefan Esser * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ * */ /* * PCIM_xxx: mask to locate subfield in register * PCIR_xxx: config register offset * PCIC_xxx: device class * PCIS_xxx: device subclass * PCIP_xxx: device programming interface * PCIV_xxx: PCI vendor ID (only required to fixup ancient devices) * PCID_xxx: device ID * PCIY_xxx: capability identification number * PCIZ_xxx: extended capability identification number */ /* some PCI bus constants */ #define PCI_DOMAINMAX 65535 /* highest supported domain number */ #define PCI_BUSMAX 255 /* highest supported bus number */ #define PCI_SLOTMAX 31 /* highest supported slot number */ #define PCI_FUNCMAX 7 /* highest supported function number */ #define PCI_REGMAX 255 /* highest supported config register addr. */ #define PCIE_REGMAX 4095 /* highest supported config register addr. */ #define PCI_MAXHDRTYPE 2 /* PCI config header registers for all devices */ #define PCIR_DEVVENDOR 0x00 #define PCIR_VENDOR 0x00 #define PCIR_DEVICE 0x02 #define PCIR_COMMAND 0x04 #define PCIM_CMD_PORTEN 0x0001 #define PCIM_CMD_MEMEN 0x0002 #define PCIM_CMD_BUSMASTEREN 0x0004 #define PCIM_CMD_SPECIALEN 0x0008 #define PCIM_CMD_MWRICEN 0x0010 #define PCIM_CMD_PERRESPEN 0x0040 #define PCIM_CMD_SERRESPEN 0x0100 #define PCIM_CMD_BACKTOBACK 0x0200 #define PCIM_CMD_INTxDIS 0x0400 #define PCIR_STATUS 0x06 #define PCIM_STATUS_CAPPRESENT 0x0010 #define PCIM_STATUS_66CAPABLE 0x0020 #define PCIM_STATUS_BACKTOBACK 0x0080 #define PCIM_STATUS_MDPERR 0x0100 #define PCIM_STATUS_PERRREPORT PCIM_STATUS_MDPERR #define PCIM_STATUS_SEL_FAST 0x0000 #define PCIM_STATUS_SEL_MEDIMUM 0x0200 #define PCIM_STATUS_SEL_SLOW 0x0400 #define PCIM_STATUS_SEL_MASK 0x0600 #define PCIM_STATUS_STABORT 0x0800 #define PCIM_STATUS_RTABORT 0x1000 #define PCIM_STATUS_RMABORT 0x2000 #define PCIM_STATUS_SERR 0x4000 #define PCIM_STATUS_PERR 0x8000 #define PCIR_REVID 0x08 #define PCIR_PROGIF 0x09 #define PCIR_SUBCLASS 0x0a #define PCIR_CLASS 0x0b #define PCIR_CACHELNSZ 0x0c #define PCIR_LATTIMER 0x0d #define PCIR_HDRTYPE 0x0e #define PCIM_HDRTYPE 0x7f #define PCIM_HDRTYPE_NORMAL 0x00 #define PCIM_HDRTYPE_BRIDGE 0x01 #define PCIM_HDRTYPE_CARDBUS 0x02 #define PCIM_MFDEV 0x80 #define PCIR_BIST 0x0f /* Capability Register Offsets */ #define PCICAP_ID 0x0 #define PCICAP_NEXTPTR 0x1 /* Capability Identification Numbers */ #define PCIY_PMG 0x01 /* PCI Power Management */ #define PCIY_AGP 0x02 /* AGP */ #define PCIY_VPD 0x03 /* Vital Product Data */ #define PCIY_SLOTID 0x04 /* Slot Identification */ #define PCIY_MSI 0x05 /* Message Signaled Interrupts */ #define PCIY_CHSWP 0x06 /* CompactPCI Hot Swap */ #define PCIY_PCIX 0x07 /* PCI-X */ #define PCIY_HT 0x08 /* HyperTransport */ #define PCIY_VENDOR 0x09 /* Vendor Unique */ #define PCIY_DEBUG 0x0a /* Debug port */ #define PCIY_CRES 0x0b /* CompactPCI central resource control */ #define PCIY_HOTPLUG 0x0c /* PCI Hot-Plug */ #define PCIY_SUBVENDOR 0x0d /* PCI-PCI bridge subvendor ID */ #define PCIY_AGP8X 0x0e /* AGP 8x */ #define PCIY_SECDEV 0x0f /* Secure Device */ #define PCIY_EXPRESS 0x10 /* PCI Express */ #define PCIY_MSIX 0x11 /* MSI-X */ #define PCIY_SATA 0x12 /* SATA */ #define PCIY_PCIAF 0x13 /* PCI Advanced Features */ /* Extended Capability Register Fields */ #define PCIR_EXTCAP 0x100 #define PCIM_EXTCAP_ID 0x0000ffff #define PCIM_EXTCAP_VER 0x000f0000 #define PCIM_EXTCAP_NEXTPTR 0xfff00000 #define PCI_EXTCAP_ID(ecap) ((ecap) & PCIM_EXTCAP_ID) #define PCI_EXTCAP_VER(ecap) (((ecap) & PCIM_EXTCAP_VER) >> 16) #define PCI_EXTCAP_NEXTPTR(ecap) (((ecap) & PCIM_EXTCAP_NEXTPTR) >> 20) /* Extended Capability Identification Numbers */ #define PCIZ_AER 0x0001 /* Advanced Error Reporting */ #define PCIZ_VC 0x0002 /* Virtual Channel */ #define PCIZ_SERNUM 0x0003 /* Device Serial Number */ #define PCIZ_PWRBDGT 0x0004 /* Power Budgeting */ #define PCIZ_VENDOR 0x000b /* Vendor Unique */ #define PCIZ_ACS 0x000d /* Access Control Services */ #define PCIZ_ARI 0x000e /* Alternative Routing-ID Interpretation */ #define PCIZ_ATS 0x000f /* Address Translation Services */ #define PCIZ_SRIOV 0x0010 /* Single Root IO Virtualization */ /* config registers for header type 0 devices */ #define PCIR_BARS 0x10 #define PCIR_BAR(x) (PCIR_BARS + (x) * 4) #define PCIR_MAX_BAR_0 5 #define PCI_RID2BAR(rid) (((rid) - PCIR_BARS) / 4) #define PCI_BAR_IO(x) (((x) & PCIM_BAR_SPACE) == PCIM_BAR_IO_SPACE) #define PCI_BAR_MEM(x) (((x) & PCIM_BAR_SPACE) == PCIM_BAR_MEM_SPACE) #define PCIM_BAR_SPACE 0x00000001 #define PCIM_BAR_MEM_SPACE 0 #define PCIM_BAR_IO_SPACE 1 #define PCIM_BAR_MEM_TYPE 0x00000006 #define PCIM_BAR_MEM_32 0 #define PCIM_BAR_MEM_1MB 2 /* Locate below 1MB in PCI <= 2.1 */ #define PCIM_BAR_MEM_64 4 #define PCIM_BAR_MEM_PREFETCH 0x00000008 #define PCIM_BAR_MEM_BASE 0xfffffffffffffff0ULL #define PCIM_BAR_IO_RESERVED 0x00000002 #define PCIM_BAR_IO_BASE 0xfffffffc #define PCIR_CIS 0x28 #define PCIM_CIS_ASI_MASK 0x7 #define PCIM_CIS_ASI_CONFIG 0 #define PCIM_CIS_ASI_BAR0 1 #define PCIM_CIS_ASI_BAR1 2 #define PCIM_CIS_ASI_BAR2 3 #define PCIM_CIS_ASI_BAR3 4 #define PCIM_CIS_ASI_BAR4 5 #define PCIM_CIS_ASI_BAR5 6 #define PCIM_CIS_ASI_ROM 7 #define PCIM_CIS_ADDR_MASK 0x0ffffff8 #define PCIM_CIS_ROM_MASK 0xf0000000 #define PCIM_CIS_CONFIG_MASK 0xff #define PCIR_SUBVEND_0 0x2c #define PCIR_SUBDEV_0 0x2e #define PCIR_BIOS 0x30 #define PCIM_BIOS_ENABLE 0x01 #define PCIM_BIOS_ADDR_MASK 0xfffff800 #define PCIR_CAP_PTR 0x34 #define PCIR_INTLINE 0x3c #define PCIR_INTPIN 0x3d #define PCIR_MINGNT 0x3e #define PCIR_MAXLAT 0x3f /* config registers for header type 1 (PCI-to-PCI bridge) devices */ #define PCIR_MAX_BAR_1 1 #define PCIR_SECSTAT_1 0x1e #define PCIR_PRIBUS_1 0x18 #define PCIR_SECBUS_1 0x19 #define PCIR_SUBBUS_1 0x1a #define PCIR_SECLAT_1 0x1b #define PCIR_IOBASEL_1 0x1c #define PCIR_IOLIMITL_1 0x1d #define PCIR_IOBASEH_1 0x30 #define PCIR_IOLIMITH_1 0x32 #define PCIM_BRIO_16 0x0 #define PCIM_BRIO_32 0x1 #define PCIM_BRIO_MASK 0xf #define PCIR_MEMBASE_1 0x20 #define PCIR_MEMLIMIT_1 0x22 #define PCIR_PMBASEL_1 0x24 #define PCIR_PMLIMITL_1 0x26 #define PCIR_PMBASEH_1 0x28 #define PCIR_PMLIMITH_1 0x2c #define PCIM_BRPM_32 0x0 #define PCIM_BRPM_64 0x1 #define PCIM_BRPM_MASK 0xf #define PCIR_BRIDGECTL_1 0x3e /* config registers for header type 2 (CardBus) devices */ #define PCIR_MAX_BAR_2 0 #define PCIR_CAP_PTR_2 0x14 #define PCIR_SECSTAT_2 0x16 #define PCIR_PRIBUS_2 0x18 #define PCIR_SECBUS_2 0x19 #define PCIR_SUBBUS_2 0x1a #define PCIR_SECLAT_2 0x1b #define PCIR_MEMBASE0_2 0x1c #define PCIR_MEMLIMIT0_2 0x20 #define PCIR_MEMBASE1_2 0x24 #define PCIR_MEMLIMIT1_2 0x28 #define PCIR_IOBASE0_2 0x2c #define PCIR_IOLIMIT0_2 0x30 #define PCIR_IOBASE1_2 0x34 #define PCIR_IOLIMIT1_2 0x38 #define PCIR_BRIDGECTL_2 0x3e #define PCIR_SUBVEND_2 0x40 #define PCIR_SUBDEV_2 0x42 #define PCIR_PCCARDIF_2 0x44 /* PCI device class, subclass and programming interface definitions */ #define PCIC_OLD 0x00 #define PCIS_OLD_NONVGA 0x00 #define PCIS_OLD_VGA 0x01 #define PCIC_STORAGE 0x01 #define PCIS_STORAGE_SCSI 0x00 #define PCIS_STORAGE_IDE 0x01 #define PCIP_STORAGE_IDE_MODEPRIM 0x01 #define PCIP_STORAGE_IDE_PROGINDPRIM 0x02 #define PCIP_STORAGE_IDE_MODESEC 0x04 #define PCIP_STORAGE_IDE_PROGINDSEC 0x08 #define PCIP_STORAGE_IDE_MASTERDEV 0x80 #define PCIS_STORAGE_FLOPPY 0x02 #define PCIS_STORAGE_IPI 0x03 #define PCIS_STORAGE_RAID 0x04 #define PCIS_STORAGE_ATA_ADMA 0x05 #define PCIS_STORAGE_SATA 0x06 #define PCIP_STORAGE_SATA_AHCI_1_0 0x01 #define PCIS_STORAGE_SAS 0x07 #define PCIS_STORAGE_NVM 0x08 #define PCIP_STORAGE_NVM_NVMHCI_1_0 0x01 #define PCIP_STORAGE_NVM_ENTERPRISE_NVMHCI_1_0 0x02 #define PCIS_STORAGE_OTHER 0x80 #define PCIC_NETWORK 0x02 #define PCIS_NETWORK_ETHERNET 0x00 #define PCIS_NETWORK_TOKENRING 0x01 #define PCIS_NETWORK_FDDI 0x02 #define PCIS_NETWORK_ATM 0x03 #define PCIS_NETWORK_ISDN 0x04 #define PCIS_NETWORK_WORLDFIP 0x05 #define PCIS_NETWORK_PICMG 0x06 #define PCIS_NETWORK_OTHER 0x80 #define PCIC_DISPLAY 0x03 #define PCIS_DISPLAY_VGA 0x00 #define PCIS_DISPLAY_XGA 0x01 #define PCIS_DISPLAY_3D 0x02 #define PCIS_DISPLAY_OTHER 0x80 #define PCIC_MULTIMEDIA 0x04 #define PCIS_MULTIMEDIA_VIDEO 0x00 #define PCIS_MULTIMEDIA_AUDIO 0x01 #define PCIS_MULTIMEDIA_TELE 0x02 #define PCIS_MULTIMEDIA_HDA 0x03 #define PCIS_MULTIMEDIA_OTHER 0x80 #define PCIC_MEMORY 0x05 #define PCIS_MEMORY_RAM 0x00 #define PCIS_MEMORY_FLASH 0x01 #define PCIS_MEMORY_OTHER 0x80 #define PCIC_BRIDGE 0x06 #define PCIS_BRIDGE_HOST 0x00 #define PCIS_BRIDGE_ISA 0x01 #define PCIS_BRIDGE_EISA 0x02 #define PCIS_BRIDGE_MCA 0x03 #define PCIS_BRIDGE_PCI 0x04 #define PCIP_BRIDGE_PCI_SUBTRACTIVE 0x01 #define PCIS_BRIDGE_PCMCIA 0x05 #define PCIS_BRIDGE_NUBUS 0x06 #define PCIS_BRIDGE_CARDBUS 0x07 #define PCIS_BRIDGE_RACEWAY 0x08 #define PCIS_BRIDGE_PCI_TRANSPARENT 0x09 #define PCIS_BRIDGE_INFINIBAND 0x0a #define PCIS_BRIDGE_OTHER 0x80 #define PCIC_SIMPLECOMM 0x07 #define PCIS_SIMPLECOMM_UART 0x00 #define PCIP_SIMPLECOMM_UART_8250 0x00 #define PCIP_SIMPLECOMM_UART_16450A 0x01 #define PCIP_SIMPLECOMM_UART_16550A 0x02 #define PCIP_SIMPLECOMM_UART_16650A 0x03 #define PCIP_SIMPLECOMM_UART_16750A 0x04 #define PCIP_SIMPLECOMM_UART_16850A 0x05 #define PCIP_SIMPLECOMM_UART_16950A 0x06 #define PCIS_SIMPLECOMM_PAR 0x01 #define PCIS_SIMPLECOMM_MULSER 0x02 #define PCIS_SIMPLECOMM_MODEM 0x03 #define PCIS_SIMPLECOMM_GPIB 0x04 #define PCIS_SIMPLECOMM_SMART_CARD 0x05 #define PCIS_SIMPLECOMM_OTHER 0x80 #define PCIC_BASEPERIPH 0x08 #define PCIS_BASEPERIPH_PIC 0x00 #define PCIP_BASEPERIPH_PIC_8259A 0x00 #define PCIP_BASEPERIPH_PIC_ISA 0x01 #define PCIP_BASEPERIPH_PIC_EISA 0x02 #define PCIP_BASEPERIPH_PIC_IO_APIC 0x10 #define PCIP_BASEPERIPH_PIC_IOX_APIC 0x20 #define PCIS_BASEPERIPH_DMA 0x01 #define PCIS_BASEPERIPH_TIMER 0x02 #define PCIS_BASEPERIPH_RTC 0x03 #define PCIS_BASEPERIPH_PCIHOT 0x04 #define PCIS_BASEPERIPH_SDHC 0x05 #define PCIS_BASEPERIPH_OTHER 0x80 #define PCIC_INPUTDEV 0x09 #define PCIS_INPUTDEV_KEYBOARD 0x00 #define PCIS_INPUTDEV_DIGITIZER 0x01 #define PCIS_INPUTDEV_MOUSE 0x02 #define PCIS_INPUTDEV_SCANNER 0x03 #define PCIS_INPUTDEV_GAMEPORT 0x04 #define PCIS_INPUTDEV_OTHER 0x80 #define PCIC_DOCKING 0x0a #define PCIS_DOCKING_GENERIC 0x00 #define PCIS_DOCKING_OTHER 0x80 #define PCIC_PROCESSOR 0x0b #define PCIS_PROCESSOR_386 0x00 #define PCIS_PROCESSOR_486 0x01 #define PCIS_PROCESSOR_PENTIUM 0x02 #define PCIS_PROCESSOR_ALPHA 0x10 #define PCIS_PROCESSOR_POWERPC 0x20 #define PCIS_PROCESSOR_MIPS 0x30 #define PCIS_PROCESSOR_COPROC 0x40 #define PCIC_SERIALBUS 0x0c #define PCIS_SERIALBUS_FW 0x00 #define PCIS_SERIALBUS_ACCESS 0x01 #define PCIS_SERIALBUS_SSA 0x02 #define PCIS_SERIALBUS_USB 0x03 #define PCIP_SERIALBUS_USB_UHCI 0x00 #define PCIP_SERIALBUS_USB_OHCI 0x10 #define PCIP_SERIALBUS_USB_EHCI 0x20 #define PCIP_SERIALBUS_USB_DEVICE 0xfe #define PCIS_SERIALBUS_FC 0x04 #define PCIS_SERIALBUS_SMBUS 0x05 #define PCIS_SERIALBUS_INFINIBAND 0x06 #define PCIS_SERIALBUS_IPMI 0x07 #define PCIP_SERIALBUS_IPMI_SMIC 0x00 #define PCIP_SERIALBUS_IPMI_KCS 0x01 #define PCIP_SERIALBUS_IPMI_BT 0x02 #define PCIS_SERIALBUS_SERCOS 0x08 #define PCIS_SERIALBUS_CANBUS 0x09 #define PCIC_WIRELESS 0x0d #define PCIS_WIRELESS_IRDA 0x00 #define PCIS_WIRELESS_IR 0x01 #define PCIS_WIRELESS_RF 0x10 #define PCIS_WIRELESS_BLUETOOTH 0x11 #define PCIS_WIRELESS_BROADBAND 0x12 #define PCIS_WIRELESS_80211A 0x20 #define PCIS_WIRELESS_80211B 0x21 #define PCIS_WIRELESS_OTHER 0x80 #define PCIC_INTELLIIO 0x0e #define PCIS_INTELLIIO_I2O 0x00 #define PCIC_SATCOM 0x0f #define PCIS_SATCOM_TV 0x01 #define PCIS_SATCOM_AUDIO 0x02 #define PCIS_SATCOM_VOICE 0x03 #define PCIS_SATCOM_DATA 0x04 #define PCIC_CRYPTO 0x10 #define PCIS_CRYPTO_NETCOMP 0x00 #define PCIS_CRYPTO_ENTERTAIN 0x10 #define PCIS_CRYPTO_OTHER 0x80 #define PCIC_DASP 0x11 #define PCIS_DASP_DPIO 0x00 #define PCIS_DASP_PERFCNTRS 0x01 #define PCIS_DASP_COMM_SYNC 0x10 #define PCIS_DASP_MGMT_CARD 0x20 #define PCIS_DASP_OTHER 0x80 #define PCIC_OTHER 0xff /* Bridge Control Values. */ #define PCIB_BCR_PERR_ENABLE 0x0001 #define PCIB_BCR_SERR_ENABLE 0x0002 #define PCIB_BCR_ISA_ENABLE 0x0004 #define PCIB_BCR_VGA_ENABLE 0x0008 #define PCIB_BCR_MASTER_ABORT_MODE 0x0020 #define PCIB_BCR_SECBUS_RESET 0x0040 #define PCIB_BCR_SECBUS_BACKTOBACK 0x0080 #define PCIB_BCR_PRI_DISCARD_TIMEOUT 0x0100 #define PCIB_BCR_SEC_DISCARD_TIMEOUT 0x0200 #define PCIB_BCR_DISCARD_TIMER_STATUS 0x0400 #define PCIB_BCR_DISCARD_TIMER_SERREN 0x0800 /* PCI power manangement */ #define PCIR_POWER_CAP 0x2 #define PCIM_PCAP_SPEC 0x0007 #define PCIM_PCAP_PMEREQCLK 0x0008 #define PCIM_PCAP_PMEREQPWR 0x0010 #define PCIM_PCAP_DEVSPECINIT 0x0020 #define PCIM_PCAP_DYNCLOCK 0x0040 #define PCIM_PCAP_SECCLOCK 0x00c0 #define PCIM_PCAP_CLOCKMASK 0x00c0 #define PCIM_PCAP_REQFULLCLOCK 0x0100 #define PCIM_PCAP_D1SUPP 0x0200 #define PCIM_PCAP_D2SUPP 0x0400 #define PCIM_PCAP_D0PME 0x0800 #define PCIM_PCAP_D1PME 0x1000 #define PCIM_PCAP_D2PME 0x2000 #define PCIM_PCAP_D3PME_HOT 0x4000 #define PCIM_PCAP_D3PME_COLD 0x8000 #define PCIR_POWER_STATUS 0x4 #define PCIM_PSTAT_D0 0x0000 #define PCIM_PSTAT_D1 0x0001 #define PCIM_PSTAT_D2 0x0002 #define PCIM_PSTAT_D3 0x0003 #define PCIM_PSTAT_DMASK 0x0003 #define PCIM_PSTAT_REPENABLE 0x0010 #define PCIM_PSTAT_PMEENABLE 0x0100 #define PCIM_PSTAT_D0POWER 0x0000 #define PCIM_PSTAT_D1POWER 0x0200 #define PCIM_PSTAT_D2POWER 0x0400 #define PCIM_PSTAT_D3POWER 0x0600 #define PCIM_PSTAT_D0HEAT 0x0800 #define PCIM_PSTAT_D1HEAT 0x1000 #define PCIM_PSTAT_D2HEAT 0x1200 #define PCIM_PSTAT_D3HEAT 0x1400 #define PCIM_PSTAT_DATAUNKN 0x0000 #define PCIM_PSTAT_DATADIV10 0x2000 #define PCIM_PSTAT_DATADIV100 0x4000 #define PCIM_PSTAT_DATADIV1000 0x6000 #define PCIM_PSTAT_DATADIVMASK 0x6000 #define PCIM_PSTAT_PME 0x8000 #define PCIR_POWER_PMCSR 0x6 #define PCIM_PMCSR_DCLOCK 0x10 #define PCIM_PMCSR_B2SUPP 0x20 #define PCIM_BMCSR_B3SUPP 0x40 #define PCIM_BMCSR_BPCE 0x80 #define PCIR_POWER_DATA 0x7 /* VPD capability registers */ #define PCIR_VPD_ADDR 0x2 #define PCIR_VPD_DATA 0x4 /* PCI Message Signalled Interrupts (MSI) */ #define PCIR_MSI_CTRL 0x2 #define PCIM_MSICTRL_VECTOR 0x0100 #define PCIM_MSICTRL_64BIT 0x0080 #define PCIM_MSICTRL_MME_MASK 0x0070 #define PCIM_MSICTRL_MME_1 0x0000 #define PCIM_MSICTRL_MME_2 0x0010 #define PCIM_MSICTRL_MME_4 0x0020 #define PCIM_MSICTRL_MME_8 0x0030 #define PCIM_MSICTRL_MME_16 0x0040 #define PCIM_MSICTRL_MME_32 0x0050 #define PCIM_MSICTRL_MMC_MASK 0x000E #define PCIM_MSICTRL_MMC_1 0x0000 #define PCIM_MSICTRL_MMC_2 0x0002 #define PCIM_MSICTRL_MMC_4 0x0004 #define PCIM_MSICTRL_MMC_8 0x0006 #define PCIM_MSICTRL_MMC_16 0x0008 #define PCIM_MSICTRL_MMC_32 0x000A #define PCIM_MSICTRL_MSI_ENABLE 0x0001 #define PCIR_MSI_ADDR 0x4 #define PCIR_MSI_ADDR_HIGH 0x8 #define PCIR_MSI_DATA 0x8 #define PCIR_MSI_DATA_64BIT 0xc #define PCIR_MSI_MASK 0x10 #define PCIR_MSI_PENDING 0x14 /* PCI-X definitions */ /* For header type 0 devices */ #define PCIXR_COMMAND 0x2 #define PCIXM_COMMAND_DPERR_E 0x0001 /* Data Parity Error Recovery */ #define PCIXM_COMMAND_ERO 0x0002 /* Enable Relaxed Ordering */ #define PCIXM_COMMAND_MAX_READ 0x000c /* Maximum Burst Read Count */ #define PCIXM_COMMAND_MAX_READ_512 0x0000 #define PCIXM_COMMAND_MAX_READ_1024 0x0004 #define PCIXM_COMMAND_MAX_READ_2048 0x0008 #define PCIXM_COMMAND_MAX_READ_4096 0x000c #define PCIXM_COMMAND_MAX_SPLITS 0x0070 /* Maximum Split Transactions */ #define PCIXM_COMMAND_MAX_SPLITS_1 0x0000 #define PCIXM_COMMAND_MAX_SPLITS_2 0x0010 #define PCIXM_COMMAND_MAX_SPLITS_3 0x0020 #define PCIXM_COMMAND_MAX_SPLITS_4 0x0030 #define PCIXM_COMMAND_MAX_SPLITS_8 0x0040 #define PCIXM_COMMAND_MAX_SPLITS_12 0x0050 #define PCIXM_COMMAND_MAX_SPLITS_16 0x0060 #define PCIXM_COMMAND_MAX_SPLITS_32 0x0070 #define PCIXM_COMMAND_VERSION 0x3000 #define PCIXR_STATUS 0x4 #define PCIXM_STATUS_DEVFN 0x000000FF #define PCIXM_STATUS_BUS 0x0000FF00 #define PCIXM_STATUS_64BIT 0x00010000 #define PCIXM_STATUS_133CAP 0x00020000 #define PCIXM_STATUS_SC_DISCARDED 0x00040000 #define PCIXM_STATUS_UNEXP_SC 0x00080000 #define PCIXM_STATUS_COMPLEX_DEV 0x00100000 #define PCIXM_STATUS_MAX_READ 0x00600000 #define PCIXM_STATUS_MAX_READ_512 0x00000000 #define PCIXM_STATUS_MAX_READ_1024 0x00200000 #define PCIXM_STATUS_MAX_READ_2048 0x00400000 #define PCIXM_STATUS_MAX_READ_4096 0x00600000 #define PCIXM_STATUS_MAX_SPLITS 0x03800000 #define PCIXM_STATUS_MAX_SPLITS_1 0x00000000 #define PCIXM_STATUS_MAX_SPLITS_2 0x00800000 #define PCIXM_STATUS_MAX_SPLITS_3 0x01000000 #define PCIXM_STATUS_MAX_SPLITS_4 0x01800000 #define PCIXM_STATUS_MAX_SPLITS_8 0x02000000 #define PCIXM_STATUS_MAX_SPLITS_12 0x02800000 #define PCIXM_STATUS_MAX_SPLITS_16 0x03000000 #define PCIXM_STATUS_MAX_SPLITS_32 0x03800000 #define PCIXM_STATUS_MAX_CUM_READ 0x1C000000 #define PCIXM_STATUS_RCVD_SC_ERR 0x20000000 #define PCIXM_STATUS_266CAP 0x40000000 #define PCIXM_STATUS_533CAP 0x80000000 /* For header type 1 devices (PCI-X bridges) */ #define PCIXR_SEC_STATUS 0x2 #define PCIXM_SEC_STATUS_64BIT 0x0001 #define PCIXM_SEC_STATUS_133CAP 0x0002 #define PCIXM_SEC_STATUS_SC_DISC 0x0004 #define PCIXM_SEC_STATUS_UNEXP_SC 0x0008 #define PCIXM_SEC_STATUS_SC_OVERRUN 0x0010 #define PCIXM_SEC_STATUS_SR_DELAYED 0x0020 #define PCIXM_SEC_STATUS_BUS_MODE 0x03c0 #define PCIXM_SEC_STATUS_VERSION 0x3000 #define PCIXM_SEC_STATUS_266CAP 0x4000 #define PCIXM_SEC_STATUS_533CAP 0x8000 #define PCIXR_BRIDGE_STATUS 0x4 #define PCIXM_BRIDGE_STATUS_DEVFN 0x000000FF #define PCIXM_BRIDGE_STATUS_BUS 0x0000FF00 #define PCIXM_BRIDGE_STATUS_64BIT 0x00010000 #define PCIXM_BRIDGE_STATUS_133CAP 0x00020000 #define PCIXM_BRIDGE_STATUS_SC_DISCARDED 0x00040000 #define PCIXM_BRIDGE_STATUS_UNEXP_SC 0x00080000 #define PCIXM_BRIDGE_STATUS_SC_OVERRUN 0x00100000 #define PCIXM_BRIDGE_STATUS_SR_DELAYED 0x00200000 #define PCIXM_BRIDGE_STATUS_DEVID_MSGCAP 0x20000000 #define PCIXM_BRIDGE_STATUS_266CAP 0x40000000 #define PCIXM_BRIDGE_STATUS_533CAP 0x80000000 /* HT (HyperTransport) Capability definitions */ #define PCIR_HT_COMMAND 0x2 #define PCIM_HTCMD_CAP_MASK 0xf800 /* Capability type. */ #define PCIM_HTCAP_SLAVE 0x0000 /* 000xx */ #define PCIM_HTCAP_HOST 0x2000 /* 001xx */ #define PCIM_HTCAP_SWITCH 0x4000 /* 01000 */ #define PCIM_HTCAP_INTERRUPT 0x8000 /* 10000 */ #define PCIM_HTCAP_REVISION_ID 0x8800 /* 10001 */ #define PCIM_HTCAP_UNITID_CLUMPING 0x9000 /* 10010 */ #define PCIM_HTCAP_EXT_CONFIG_SPACE 0x9800 /* 10011 */ #define PCIM_HTCAP_ADDRESS_MAPPING 0xa000 /* 10100 */ #define PCIM_HTCAP_MSI_MAPPING 0xa800 /* 10101 */ #define PCIM_HTCAP_DIRECT_ROUTE 0xb000 /* 10110 */ #define PCIM_HTCAP_VCSET 0xb800 /* 10111 */ #define PCIM_HTCAP_RETRY_MODE 0xc000 /* 11000 */ #define PCIM_HTCAP_X86_ENCODING 0xc800 /* 11001 */ /* HT MSI Mapping Capability definitions. */ #define PCIM_HTCMD_MSI_ENABLE 0x0001 #define PCIM_HTCMD_MSI_FIXED 0x0002 #define PCIR_HTMSI_ADDRESS_LO 0x4 #define PCIR_HTMSI_ADDRESS_HI 0x8 /* PCI Vendor capability definitions */ #define PCIR_VENDOR_LENGTH 0x2 #define PCIR_VENDOR_DATA 0x3 /* PCI EHCI Debug Port definitions */ #define PCIR_DEBUG_PORT 0x2 #define PCIM_DEBUG_PORT_OFFSET 0x1FFF #define PCIM_DEBUG_PORT_BAR 0xe000 /* PCI-PCI Bridge Subvendor definitions */ #define PCIR_SUBVENDCAP_ID 0x4 /* PCI Express definitions */ -#define PCIR_EXPRESS_FLAGS 0x2 -#define PCIM_EXP_FLAGS_VERSION 0x000F -#define PCIM_EXP_FLAGS_TYPE 0x00F0 -#define PCIM_EXP_TYPE_ENDPOINT 0x0000 -#define PCIM_EXP_TYPE_LEGACY_ENDPOINT 0x0010 -#define PCIM_EXP_TYPE_ROOT_PORT 0x0040 -#define PCIM_EXP_TYPE_UPSTREAM_PORT 0x0050 -#define PCIM_EXP_TYPE_DOWNSTREAM_PORT 0x0060 -#define PCIM_EXP_TYPE_PCI_BRIDGE 0x0070 -#define PCIM_EXP_TYPE_PCIE_BRIDGE 0x0080 -#define PCIM_EXP_TYPE_ROOT_INT_EP 0x0090 -#define PCIM_EXP_TYPE_ROOT_EC 0x00a0 -#define PCIM_EXP_FLAGS_SLOT 0x0100 -#define PCIM_EXP_FLAGS_IRQ 0x3e00 -#define PCIR_EXPRESS_DEVICE_CAP 0x4 -#define PCIM_EXP_CAP_MAX_PAYLOAD 0x0007 -#define PCIR_EXPRESS_DEVICE_CTL 0x8 -#define PCIM_EXP_CTL_NFER_ENABLE 0x0002 -#define PCIM_EXP_CTL_FER_ENABLE 0x0004 -#define PCIM_EXP_CTL_URR_ENABLE 0x0008 -#define PCIM_EXP_CTL_RELAXED_ORD_ENABLE 0x0010 -#define PCIM_EXP_CTL_MAX_PAYLOAD 0x00e0 -#define PCIM_EXP_CTL_NOSNOOP_ENABLE 0x0800 -#define PCIM_EXP_CTL_MAX_READ_REQUEST 0x7000 -#define PCIR_EXPRESS_DEVICE_STA 0xa -#define PCIM_EXP_STA_CORRECTABLE_ERROR 0x0001 -#define PCIM_EXP_STA_NON_FATAL_ERROR 0x0002 -#define PCIM_EXP_STA_FATAL_ERROR 0x0004 -#define PCIM_EXP_STA_UNSUPPORTED_REQ 0x0008 -#define PCIM_EXP_STA_AUX_POWER 0x0010 -#define PCIM_EXP_STA_TRANSACTION_PND 0x0020 -#define PCIR_EXPRESS_LINK_CAP 0xc -#define PCIM_LINK_CAP_MAX_SPEED 0x0000000f -#define PCIM_LINK_CAP_MAX_WIDTH 0x000003f0 -#define PCIM_LINK_CAP_ASPM 0x00000c00 -#define PCIM_LINK_CAP_L0S_EXIT 0x00007000 -#define PCIM_LINK_CAP_L1_EXIT 0x00038000 -#define PCIM_LINK_CAP_PORT 0xff000000 -#define PCIR_EXPRESS_LINK_CTL 0x10 -#define PCIR_EXPRESS_LINK_STA 0x12 -#define PCIM_LINK_STA_SPEED 0x000f -#define PCIM_LINK_STA_WIDTH 0x03f0 -#define PCIM_LINK_STA_TRAINING_ERROR 0x0400 -#define PCIM_LINK_STA_TRAINING 0x0800 -#define PCIM_LINK_STA_SLOT_CLOCK 0x1000 -#define PCIR_EXPRESS_SLOT_CAP 0x14 -#define PCIR_EXPRESS_SLOT_CTL 0x18 -#define PCIR_EXPRESS_SLOT_STA 0x1a -#define PCIR_EXPRESS_ROOT_CTL 0x1c -#define PCIR_EXPRESS_ROOT_STA 0x20 +#define PCIER_FLAGS 0x2 +#define PCIEM_FLAGS_VERSION 0x000F +#define PCIEM_FLAGS_TYPE 0x00F0 +#define PCIEM_TYPE_ENDPOINT 0x0000 +#define PCIEM_TYPE_LEGACY_ENDPOINT 0x0010 +#define PCIEM_TYPE_ROOT_PORT 0x0040 +#define PCIEM_TYPE_UPSTREAM_PORT 0x0050 +#define PCIEM_TYPE_DOWNSTREAM_PORT 0x0060 +#define PCIEM_TYPE_PCI_BRIDGE 0x0070 +#define PCIEM_TYPE_PCIE_BRIDGE 0x0080 +#define PCIEM_TYPE_ROOT_INT_EP 0x0090 +#define PCIEM_TYPE_ROOT_EC 0x00a0 +#define PCIEM_FLAGS_SLOT 0x0100 +#define PCIEM_FLAGS_IRQ 0x3e00 +#define PCIER_DEVICE_CAP 0x4 +#define PCIEM_CAP_MAX_PAYLOAD 0x00000007 +#define PCIEM_CAP_PHANTHOM_FUNCS 0x00000018 +#define PCIEM_CAP_EXT_TAG_FIELD 0x00000020 +#define PCIEM_CAP_L0S_LATENCY 0x000001c0 +#define PCIEM_CAP_L1_LATENCY 0x00000e00 +#define PCIEM_CAP_ROLE_ERR_RPT 0x00008000 +#define PCIEM_CAP_SLOT_PWR_LIM_VAL 0x03fc0000 +#define PCIEM_CAP_SLOT_PWR_LIM_SCALE 0x0c000000 +#define PCIEM_CAP_FLR 0x10000000 +#define PCIER_DEVICE_CTL 0x8 +#define PCIEM_CTL_COR_ENABLE 0x0001 +#define PCIEM_CTL_NFER_ENABLE 0x0002 +#define PCIEM_CTL_FER_ENABLE 0x0004 +#define PCIEM_CTL_URR_ENABLE 0x0008 +#define PCIEM_CTL_RELAXED_ORD_ENABLE 0x0010 +#define PCIEM_CTL_MAX_PAYLOAD 0x00e0 +#define PCIEM_CTL_EXT_TAG_FIELD 0x0100 +#define PCIEM_CTL_PHANTHOM_FUNCS 0x0200 +#define PCIEM_CTL_AUX_POWER_PM 0x0400 +#define PCIEM_CTL_NOSNOOP_ENABLE 0x0800 +#define PCIEM_CTL_MAX_READ_REQUEST 0x7000 +#define PCIEM_CTL_BRDG_CFG_RETRY 0x8000 /* PCI-E - PCI/PCI-X bridges */ +#define PCIEM_CTL_INITIATE_FLR 0x8000 /* FLR capable endpoints */ +#define PCIER_DEVICE_STA 0xa +#define PCIEM_STA_CORRECTABLE_ERROR 0x0001 +#define PCIEM_STA_NON_FATAL_ERROR 0x0002 +#define PCIEM_STA_FATAL_ERROR 0x0004 +#define PCIEM_STA_UNSUPPORTED_REQ 0x0008 +#define PCIEM_STA_AUX_POWER 0x0010 +#define PCIEM_STA_TRANSACTION_PND 0x0020 +#define PCIER_LINK_CAP 0xc +#define PCIEM_LINK_CAP_MAX_SPEED 0x0000000f +#define PCIEM_LINK_CAP_MAX_WIDTH 0x000003f0 +#define PCIEM_LINK_CAP_ASPM 0x00000c00 +#define PCIEM_LINK_CAP_L0S_EXIT 0x00007000 +#define PCIEM_LINK_CAP_L1_EXIT 0x00038000 +#define PCIEM_LINK_CAP_CLOCK_PM 0x00040000 +#define PCIEM_LINK_CAP_SURPRISE_DOWN 0x00080000 +#define PCIEM_LINK_CAP_DL_ACTIVE 0x00100000 +#define PCIEM_LINK_CAP_LINK_BW_NOTIFY 0x00200000 +#define PCIEM_LINK_CAP_ASPM_COMPLIANCE 0x00400000 +#define PCIEM_LINK_CAP_PORT 0xff000000 +#define PCIER_LINK_CTL 0x10 +#define PCIEM_LINK_CTL_ASPMC_DIS 0x0000 +#define PCIEM_LINK_CTL_ASPMC_L0S 0x0001 +#define PCIEM_LINK_CTL_ASPMC_L1 0x0002 +#define PCIEM_LINK_CTL_ASPMC 0x0003 +#define PCIEM_LINK_CTL_RCB 0x0008 +#define PCIEM_LINK_CTL_LINK_DIS 0x0010 +#define PCIEM_LINK_CTL_RETRAIN_LINK 0x0020 +#define PCIEM_LINK_CTL_COMMON_CLOCK 0x0040 +#define PCIEM_LINK_CTL_EXTENDED_SYNC 0x0080 +#define PCIEM_LINK_CTL_ECPM 0x0100 +#define PCIEM_LINK_CTL_HAWD 0x0200 +#define PCIEM_LINK_CTL_LBMIE 0x0400 +#define PCIEM_LINK_CTL_LABIE 0x0800 +#define PCIER_LINK_STA 0x12 +#define PCIEM_LINK_STA_SPEED 0x000f +#define PCIEM_LINK_STA_WIDTH 0x03f0 +#define PCIEM_LINK_STA_TRAINING_ERROR 0x0400 +#define PCIEM_LINK_STA_TRAINING 0x0800 +#define PCIEM_LINK_STA_SLOT_CLOCK 0x1000 +#define PCIEM_LINK_STA_DL_ACTIVE 0x2000 +#define PCIEM_LINK_STA_LINK_BW_MGMT 0x4000 +#define PCIEM_LINK_STA_LINK_AUTO_BW 0x8000 +#define PCIER_SLOT_CAP 0x14 +#define PCIEM_SLOT_CAP_APB 0x00000001 +#define PCIEM_SLOT_CAP_PCP 0x00000002 +#define PCIEM_SLOT_CAP_MRLSP 0x00000004 +#define PCIEM_SLOT_CAP_AIP 0x00000008 +#define PCIEM_SLOT_CAP_PIP 0x00000010 +#define PCIEM_SLOT_CAP_HPS 0x00000020 +#define PCIEM_SLOT_CAP_HPC 0x00000040 +#define PCIEM_SLOT_CAP_SPLV 0x00007f80 +#define PCIEM_SLOT_CAP_SPLS 0x00018000 +#define PCIEM_SLOT_CAP_EIP 0x00020000 +#define PCIEM_SLOT_CAP_NCCS 0x00040000 +#define PCIEM_SLOT_CAP_PSN 0xfff80000 +#define PCIER_SLOT_CTL 0x18 +#define PCIEM_SLOT_CTL_ABPE 0x0001 +#define PCIEM_SLOT_CTL_PFDE 0x0002 +#define PCIEM_SLOT_CTL_MRLSCE 0x0004 +#define PCIEM_SLOT_CTL_PDCE 0x0008 +#define PCIEM_SLOT_CTL_CCIE 0x0010 +#define PCIEM_SLOT_CTL_HPIE 0x0020 +#define PCIEM_SLOT_CTL_AIC 0x00c0 +#define PCIEM_SLOT_CTL_PIC 0x0300 +#define PCIEM_SLOT_CTL_PCC 0x0400 +#define PCIEM_SLOT_CTL_EIC 0x0800 +#define PCIEM_SLOT_CTL_DLLSCE 0x1000 +#define PCIER_SLOT_STA 0x1a +#define PCIEM_SLOT_STA_ABP 0x0001 +#define PCIEM_SLOT_STA_PFD 0x0002 +#define PCIEM_SLOT_STA_MRLSC 0x0004 +#define PCIEM_SLOT_STA_PDC 0x0008 +#define PCIEM_SLOT_STA_CC 0x0010 +#define PCIEM_SLOT_STA_MRLSS 0x0020 +#define PCIEM_SLOT_STA_PDS 0x0040 +#define PCIEM_SLOT_STA_EIS 0x0080 +#define PCIEM_SLOT_STA_DLLSC 0x0100 +#define PCIER_ROOT_CTL 0x1c +#define PCIER_ROOT_CAP 0x1e +#define PCIER_ROOT_STA 0x20 +#define PCIER_DEVICE_CAP2 0x24 +#define PCIER_DEVICE_CTL2 0x28 +#define PCIEM_CTL2_COMP_TIMEOUT_VAL 0x000f +#define PCIEM_CTL2_COMP_TIMEOUT_DIS 0x0010 +#define PCIEM_CTL2_ARI 0x0020 +#define PCIEM_CTL2_ATOMIC_REQ_ENABLE 0x0040 +#define PCIEM_CTL2_ATOMIC_EGR_BLOCK 0x0080 +#define PCIEM_CTL2_ID_ORDERED_REQ_EN 0x0100 +#define PCIEM_CTL2_ID_ORDERED_CMP_EN 0x0200 +#define PCIEM_CTL2_LTR_ENABLE 0x0400 +#define PCIEM_CTL2_OBFF 0x6000 +#define PCIEM_OBFF_DISABLE 0x0000 +#define PCIEM_OBFF_MSGA_ENABLE 0x2000 +#define PCIEM_OBFF_MSGB_ENABLE 0x4000 +#define PCIEM_OBFF_WAKE_ENABLE 0x6000 +#define PCIEM_CTL2_END2END_TLP 0x8000 +#define PCIER_DEVICE_STA2 0x2a +#define PCIER_LINK_CAP2 0x2c +#define PCIER_LINK_CTL2 0x30 +#define PCIER_LINK_STA2 0x32 +#define PCIER_SLOT_CAP2 0x34 +#define PCIER_SLOT_CTL2 0x38 +#define PCIER_SLOT_STA2 0x3a + +/* Old compatibility definitions for PCI Express registers */ +#define PCIR_EXPRESS_FLAGS PCIER_FLAGS +#define PCIM_EXP_FLAGS_VERSION PCIEM_FLAGS_VERSION +#define PCIM_EXP_FLAGS_TYPE PCIEM_FLAGS_TYPE +#define PCIM_EXP_TYPE_ENDPOINT PCIEM_TYPE_ENDPOINT +#define PCIM_EXP_TYPE_LEGACY_ENDPOINT PCIEM_TYPE_LEGACY_ENDPOINT +#define PCIM_EXP_TYPE_ROOT_PORT PCIEM_TYPE_ROOT_PORT +#define PCIM_EXP_TYPE_UPSTREAM_PORT PCIEM_TYPE_UPSTREAM_PORT +#define PCIM_EXP_TYPE_DOWNSTREAM_PORT PCIEM_TYPE_DOWNSTREAM_PORT +#define PCIM_EXP_TYPE_PCI_BRIDGE PCIEM_TYPE_PCI_BRIDGE +#define PCIM_EXP_TYPE_PCIE_BRIDGE PCIEM_TYPE_PCIE_BRIDGE +#define PCIM_EXP_TYPE_ROOT_INT_EP PCIEM_TYPE_ROOT_INT_EP +#define PCIM_EXP_TYPE_ROOT_EC PCIEM_TYPE_ROOT_EC +#define PCIM_EXP_FLAGS_SLOT PCIEM_FLAGS_SLOT +#define PCIM_EXP_FLAGS_IRQ PCIEM_FLAGS_IRQ +#define PCIR_EXPRESS_DEVICE_CAP PCIER_DEVICE_CAP +#define PCIM_EXP_CAP_MAX_PAYLOAD PCIEM_CAP_MAX_PAYLOAD +#define PCIM_EXP_CAP_PHANTHOM_FUNCS PCIEM_CAP_PHANTHOM_FUNCS +#define PCIM_EXP_CAP_EXT_TAG_FIELD PCIEM_CAP_EXT_TAG_FIELD +#define PCIM_EXP_CAP_L0S_LATENCY PCIEM_CAP_L0S_LATENCY +#define PCIM_EXP_CAP_L1_LATENCY PCIEM_CAP_L1_LATENCY +#define PCIM_EXP_CAP_ROLE_ERR_RPT PCIEM_CAP_ROLE_ERR_RPT +#define PCIM_EXP_CAP_SLOT_PWR_LIM_VAL PCIEM_CAP_SLOT_PWR_LIM_VAL +#define PCIM_EXP_CAP_SLOT_PWR_LIM_SCALE PCIEM_CAP_SLOT_PWR_LIM_SCALE +#define PCIM_EXP_CAP_FLR PCIEM_CAP_FLR +#define PCIR_EXPRESS_DEVICE_CTL PCIER_DEVICE_CTL +#define PCIM_EXP_CTL_COR_ENABLE PCIEM_CTL_COR_ENABLE +#define PCIM_EXP_CTL_NFER_ENABLE PCIEM_CTL_NFER_ENABLE +#define PCIM_EXP_CTL_FER_ENABLE PCIEM_CTL_FER_ENABLE +#define PCIM_EXP_CTL_URR_ENABLE PCIEM_CTL_URR_ENABLE +#define PCIM_EXP_CTL_RELAXED_ORD_ENABLE PCIEM_CTL_RELAXED_ORD_ENABLE +#define PCIM_EXP_CTL_MAX_PAYLOAD PCIEM_CTL_MAX_PAYLOAD +#define PCIM_EXP_CTL_EXT_TAG_FIELD PCIEM_CTL_EXT_TAG_FIELD +#define PCIM_EXP_CTL_PHANTHOM_FUNCS PCIEM_CTL_PHANTHOM_FUNCS +#define PCIM_EXP_CTL_AUX_POWER_PM PCIEM_CTL_AUX_POWER_PM +#define PCIM_EXP_CTL_NOSNOOP_ENABLE PCIEM_CTL_NOSNOOP_ENABLE +#define PCIM_EXP_CTL_MAX_READ_REQUEST PCIEM_CTL_MAX_READ_REQUEST +#define PCIM_EXP_CTL_BRDG_CFG_RETRY PCIEM_CTL_BRDG_CFG_RETRY +#define PCIM_EXP_CTL_INITIATE_FLR PCIEM_CTL_INITIATE_FLR +#define PCIR_EXPRESS_DEVICE_STA PCIER_DEVICE_STA +#define PCIM_EXP_STA_CORRECTABLE_ERROR PCIEM_STA_CORRECTABLE_ERROR +#define PCIM_EXP_STA_NON_FATAL_ERROR PCIEM_STA_NON_FATAL_ERROR +#define PCIM_EXP_STA_FATAL_ERROR PCIEM_STA_FATAL_ERROR +#define PCIM_EXP_STA_UNSUPPORTED_REQ PCIEM_STA_UNSUPPORTED_REQ +#define PCIM_EXP_STA_AUX_POWER PCIEM_STA_AUX_POWER +#define PCIM_EXP_STA_TRANSACTION_PND PCIEM_STA_TRANSACTION_PND +#define PCIR_EXPRESS_LINK_CAP PCIER_LINK_CAP +#define PCIM_LINK_CAP_MAX_SPEED PCIEM_LINK_CAP_MAX_SPEED +#define PCIM_LINK_CAP_MAX_WIDTH PCIEM_LINK_CAP_MAX_WIDTH +#define PCIM_LINK_CAP_ASPM PCIEM_LINK_CAP_ASPM +#define PCIM_LINK_CAP_L0S_EXIT PCIEM_LINK_CAP_L0S_EXIT +#define PCIM_LINK_CAP_L1_EXIT PCIEM_LINK_CAP_L1_EXIT +#define PCIM_LINK_CAP_CLOCK_PM PCIEM_LINK_CAP_CLOCK_PM +#define PCIM_LINK_CAP_SURPRISE_DOWN PCIEM_LINK_CAP_SURPRISE_DOWN +#define PCIM_LINK_CAP_DL_ACTIVE PCIEM_LINK_CAP_DL_ACTIVE +#define PCIM_LINK_CAP_LINK_BW_NOTIFY PCIEM_LINK_CAP_LINK_BW_NOTIFY +#define PCIM_LINK_CAP_ASPM_COMPLIANCE PCIEM_LINK_CAP_ASPM_COMPLIANCE +#define PCIM_LINK_CAP_PORT PCIEM_LINK_CAP_PORT +#define PCIR_EXPRESS_LINK_CTL PCIER_LINK_CTL +#define PCIM_EXP_LINK_CTL_ASPMC_DIS PCIEM_LINK_CTL_ASPMC_DIS +#define PCIM_EXP_LINK_CTL_ASPMC_L0S PCIEM_LINK_CTL_ASPMC_L0S +#define PCIM_EXP_LINK_CTL_ASPMC_L1 PCIEM_LINK_CTL_ASPMC_L1 +#define PCIM_EXP_LINK_CTL_ASPMC PCIEM_LINK_CTL_ASPMC +#define PCIM_EXP_LINK_CTL_RCB PCIEM_LINK_CTL_RCB +#define PCIM_EXP_LINK_CTL_LINK_DIS PCIEM_LINK_CTL_LINK_DIS +#define PCIM_EXP_LINK_CTL_RETRAIN_LINK PCIEM_LINK_CTL_RETRAIN_LINK +#define PCIM_EXP_LINK_CTL_COMMON_CLOCK PCIEM_LINK_CTL_COMMON_CLOCK +#define PCIM_EXP_LINK_CTL_EXTENDED_SYNC PCIEM_LINK_CTL_EXTENDED_SYNC +#define PCIM_EXP_LINK_CTL_ECPM PCIEM_LINK_CTL_ECPM +#define PCIM_EXP_LINK_CTL_HAWD PCIEM_LINK_CTL_HAWD +#define PCIM_EXP_LINK_CTL_LBMIE PCIEM_LINK_CTL_LBMIE +#define PCIM_EXP_LINK_CTL_LABIE PCIEM_LINK_CTL_LABIE +#define PCIR_EXPRESS_LINK_STA PCIER_LINK_STA +#define PCIM_LINK_STA_SPEED PCIEM_LINK_STA_SPEED +#define PCIM_LINK_STA_WIDTH PCIEM_LINK_STA_WIDTH +#define PCIM_LINK_STA_TRAINING_ERROR PCIEM_LINK_STA_TRAINING_ERROR +#define PCIM_LINK_STA_TRAINING PCIEM_LINK_STA_TRAINING +#define PCIM_LINK_STA_SLOT_CLOCK PCIEM_LINK_STA_SLOT_CLOCK +#define PCIM_LINK_STA_DL_ACTIVE PCIEM_LINK_STA_DL_ACTIVE +#define PCIM_LINK_STA_LINK_BW_MGMT PCIEM_LINK_STA_LINK_BW_MGMT +#define PCIM_LINK_STA_LINK_AUTO_BW PCIEM_LINK_STA_LINK_AUTO_BW +#define PCIR_EXPRESS_SLOT_CAP PCIER_SLOT_CAP +#define PCIR_EXPRESS_SLOT_CTL PCIER_SLOT_CTL +#define PCIR_EXPRESS_SLOT_STA PCIER_SLOT_STA +#define PCIR_EXPRESS_ROOT_CTL PCIER_ROOT_CTL +#define PCIR_EXPRESS_ROOT_CAP PCIER_ROOT_CAP +#define PCIR_EXPRESS_ROOT_STA PCIER_ROOT_STA +#define PCIR_EXPRESS_DEVICE_CAP2 PCIER_DEVICE_CAP2 +#define PCIR_EXPRESS_DEVICE_CTL2 PCIER_DEVICE_CTL2 +#define PCIM_EXP_CTL2_COMP_TIMEOUT_VAL PCIEM_CTL2_COMP_TIMEOUT_VAL +#define PCIM_EXP_CTL2_COMP_TIMEOUT_DIS PCIEM_CTL2_COMP_TIMEOUT_DIS +#define PCIM_EXP_CTL2_ARI PCIEM_CTL2_ARI +#define PCIM_EXP_CTL2_ATOMIC_REQ_ENABLE PCIEM_CTL2_ATOMIC_REQ_ENABLE +#define PCIM_EXP_CTL2_ATOMIC_EGR_BLOCK PCIEM_CTL2_ATOMIC_EGR_BLOCK +#define PCIM_EXP_CTL2_ID_ORDERED_REQ_EN PCIEM_CTL2_ID_ORDERED_REQ_EN +#define PCIM_EXP_CTL2_ID_ORDERED_CMP_EN PCIEM_CTL2_ID_ORDERED_CMP_EN +#define PCIM_EXP_CTL2_LTR_ENABLE PCIEM_CTL2_LTR_ENABLE +#define PCIM_EXP_CTL2_OBFF PCIEM_CTL2_OBFF +#define PCIM_EXP_OBFF_DISABLE PCIEM_OBFF_DISABLE +#define PCIM_EXP_OBFF_MSGA_ENABLE PCIEM_OBFF_MSGA_ENABLE +#define PCIM_EXP_OBFF_MSGB_ENABLE PCIEM_OBFF_MSGB_ENABLE +#define PCIM_EXP_OBFF_WAKE_ENABLE PCIEM_OBFF_WAKE_ENABLE +#define PCIM_EXP_CTL2_END2END_TLP PCIEM_CTL2_END2END_TLP +#define PCIR_EXPRESS_DEVICE_STA2 PCIER_DEVICE_STA2 +#define PCIR_EXPRESS_LINK_CAP2 PCIER_LINK_CAP2 +#define PCIR_EXPRESS_LINK_CTL2 PCIER_LINK_CTL2 +#define PCIR_EXPRESS_LINK_STA2 PCIER_LINK_STA2 +#define PCIR_EXPRESS_SLOT_CAP2 PCIER_SLOT_CAP2 +#define PCIR_EXPRESS_SLOT_CTL2 PCIER_SLOT_CTL2 +#define PCIR_EXPRESS_SLOT_STA2 PCIER_SLOT_STA2 /* MSI-X definitions */ #define PCIR_MSIX_CTRL 0x2 #define PCIM_MSIXCTRL_MSIX_ENABLE 0x8000 #define PCIM_MSIXCTRL_FUNCTION_MASK 0x4000 #define PCIM_MSIXCTRL_TABLE_SIZE 0x07FF #define PCIR_MSIX_TABLE 0x4 #define PCIR_MSIX_PBA 0x8 #define PCIM_MSIX_BIR_MASK 0x7 #define PCIM_MSIX_BIR_BAR_10 0 #define PCIM_MSIX_BIR_BAR_14 1 #define PCIM_MSIX_BIR_BAR_18 2 #define PCIM_MSIX_BIR_BAR_1C 3 #define PCIM_MSIX_BIR_BAR_20 4 #define PCIM_MSIX_BIR_BAR_24 5 #define PCIM_MSIX_VCTRL_MASK 0x1 /* PCI Advanced Features definitions */ #define PCIR_PCIAF_CAP 0x3 #define PCIM_PCIAFCAP_TP 0x01 #define PCIM_PCIAFCAP_FLR 0x02 #define PCIR_PCIAF_CTRL 0x4 #define PCIR_PCIAFCTRL_FLR 0x01 #define PCIR_PCIAF_STATUS 0x5 #define PCIR_PCIAFSTATUS_TP 0x01 /* Advanced Error Reporting */ #define PCIR_AER_UC_STATUS 0x04 #define PCIM_AER_UC_TRAINING_ERROR 0x00000001 #define PCIM_AER_UC_DL_PROTOCOL_ERROR 0x00000010 #define PCIM_AER_UC_POISONED_TLP 0x00001000 #define PCIM_AER_UC_FC_PROTOCOL_ERROR 0x00002000 #define PCIM_AER_UC_COMPLETION_TIMEOUT 0x00004000 #define PCIM_AER_UC_COMPLETER_ABORT 0x00008000 #define PCIM_AER_UC_UNEXPECTED_COMPLETION 0x00010000 #define PCIM_AER_UC_RECEIVER_OVERFLOW 0x00020000 #define PCIM_AER_UC_MALFORMED_TLP 0x00040000 #define PCIM_AER_UC_ECRC_ERROR 0x00080000 #define PCIM_AER_UC_UNSUPPORTED_REQUEST 0x00100000 #define PCIM_AER_UC_ACS_VIOLATION 0x00200000 #define PCIR_AER_UC_MASK 0x08 /* Shares bits with UC_STATUS */ #define PCIR_AER_UC_SEVERITY 0x0c /* Shares bits with UC_STATUS */ #define PCIR_AER_COR_STATUS 0x10 #define PCIM_AER_COR_RECEIVER_ERROR 0x00000001 #define PCIM_AER_COR_BAD_TLP 0x00000040 #define PCIM_AER_COR_BAD_DLLP 0x00000080 #define PCIM_AER_COR_REPLAY_ROLLOVER 0x00000100 #define PCIM_AER_COR_REPLAY_TIMEOUT 0x00001000 #define PCIR_AER_COR_MASK 0x14 /* Shares bits with COR_STATUS */ #define PCIR_AER_CAP_CONTROL 0x18 #define PCIM_AER_FIRST_ERROR_PTR 0x0000001f #define PCIM_AER_ECRC_GEN_CAPABLE 0x00000020 #define PCIM_AER_ECRC_GEN_ENABLE 0x00000040 #define PCIM_AER_ECRC_CHECK_CAPABLE 0x00000080 #define PCIM_AER_ECRC_CHECK_ENABLE 0x00000100 #define PCIR_AER_HEADER_LOG 0x1c #define PCIR_AER_ROOTERR_CMD 0x2c /* Only for root complex ports */ #define PCIM_AER_ROOTERR_COR_ENABLE 0x00000001 #define PCIM_AER_ROOTERR_NF_ENABLE 0x00000002 #define PCIM_AER_ROOTERR_F_ENABLE 0x00000004 #define PCIR_AER_ROOTERR_STATUS 0x30 /* Only for root complex ports */ #define PCIM_AER_ROOTERR_COR_ERR 0x00000001 #define PCIM_AER_ROOTERR_MULTI_COR_ERR 0x00000002 #define PCIM_AER_ROOTERR_UC_ERR 0x00000004 #define PCIM_AER_ROOTERR_MULTI_UC_ERR 0x00000008 #define PCIM_AER_ROOTERR_FIRST_UC_FATAL 0x00000010 #define PCIM_AER_ROOTERR_NF_ERR 0x00000020 #define PCIM_AER_ROOTERR_F_ERR 0x00000040 #define PCIM_AER_ROOTERR_INT_MESSAGE 0xf8000000 #define PCIR_AER_COR_SOURCE_ID 0x34 /* Only for root complex ports */ #define PCIR_AER_ERR_SOURCE_ID 0x36 /* Only for root complex ports */ /* Virtual Channel definitions */ #define PCIR_VC_CAP1 0x04 #define PCIM_VC_CAP1_EXT_COUNT 0x00000007 #define PCIM_VC_CAP1_LOWPRI_EXT_COUNT 0x00000070 #define PCIR_VC_CAP2 0x08 #define PCIR_VC_CONTROL 0x0C #define PCIR_VC_STATUS 0x0E #define PCIR_VC_RESOURCE_CAP(n) (0x10 + (n) * 0x0C) #define PCIR_VC_RESOURCE_CTL(n) (0x14 + (n) * 0x0C) #define PCIR_VC_RESOURCE_STA(n) (0x18 + (n) * 0x0C) /* Serial Number definitions */ #define PCIR_SERIAL_LOW 0x04 #define PCIR_SERIAL_HIGH 0x08 Index: stable/7/sys/dev/re/if_re.c =================================================================== --- stable/7/sys/dev/re/if_re.c (revision 242221) +++ stable/7/sys/dev/re/if_re.c (revision 242222) @@ -1,3967 +1,3967 @@ /*- * Copyright (c) 1997, 1998-2003 * Bill Paul . All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Bill Paul. * 4. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); /* * RealTek 8139C+/8169/8169S/8110S/8168/8111/8101E PCI NIC driver * * Written by Bill Paul * Senior Networking Software Engineer * Wind River Systems */ /* * This driver is designed to support RealTek's next generation of * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently * seven devices in this family: the RTL8139C+, the RTL8169, the RTL8169S, * RTL8110S, the RTL8168, the RTL8111 and the RTL8101E. * * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible * with the older 8139 family, however it also supports a special * C+ mode of operation that provides several new performance enhancing * features. These include: * * o Descriptor based DMA mechanism. Each descriptor represents * a single packet fragment. Data buffers may be aligned on * any byte boundary. * * o 64-bit DMA * * o TCP/IP checksum offload for both RX and TX * * o High and normal priority transmit DMA rings * * o VLAN tag insertion and extraction * * o TCP large send (segmentation offload) * * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+ * programming API is fairly straightforward. The RX filtering, EEPROM * access and PHY access is the same as it is on the older 8139 series * chips. * * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the * same programming API and feature set as the 8139C+ with the following * differences and additions: * * o 1000Mbps mode * * o Jumbo frames * * o GMII and TBI ports/registers for interfacing with copper * or fiber PHYs * * o RX and TX DMA rings can have up to 1024 descriptors * (the 8139C+ allows a maximum of 64) * * o Slight differences in register layout from the 8139C+ * * The TX start and timer interrupt registers are at different locations * on the 8169 than they are on the 8139C+. Also, the status word in the * RX descriptor has a slightly different bit layout. The 8169 does not * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska' * copper gigE PHY. * * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs * (the 'S' stands for 'single-chip'). These devices have the same * programming API as the older 8169, but also have some vendor-specific * registers for the on-board PHY. The 8110S is a LAN-on-motherboard * part designed to be pin-compatible with the RealTek 8100 10/100 chip. * * This driver takes advantage of the RX and TX checksum offload and * VLAN tag insertion/extraction features. It also implements TX * interrupt moderation using the timer interrupt registers, which * significantly reduces TX interrupt load. There is also support * for jumbo frames, however the 8169/8169S/8110S can not transmit * jumbo frames larger than 7440, so the max MTU possible with this * driver is 7422 bytes. */ #ifdef HAVE_KERNEL_OPTION_HEADERS #include "opt_device_polling.h" #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MODULE_DEPEND(re, pci, 1, 1, 1); MODULE_DEPEND(re, ether, 1, 1, 1); MODULE_DEPEND(re, miibus, 1, 1, 1); /* "device miibus" required. See GENERIC if you get errors here. */ #include "miibus_if.h" /* Tunables. */ static int intr_filter = 0; TUNABLE_INT("hw.re.intr_filter", &intr_filter); static int msi_disable = 0; TUNABLE_INT("hw.re.msi_disable", &msi_disable); static int msix_disable = 0; TUNABLE_INT("hw.re.msix_disable", &msix_disable); static int prefer_iomap = 0; TUNABLE_INT("hw.re.prefer_iomap", &prefer_iomap); #define RE_CSUM_FEATURES (CSUM_IP | CSUM_TCP | CSUM_UDP) /* * Various supported device vendors/types and their names. */ static struct rl_type re_devs[] = { { DLINK_VENDORID, DLINK_DEVICEID_528T, 0, "D-Link DGE-528(T) Gigabit Ethernet Adapter" }, { DLINK_VENDORID, DLINK_DEVICEID_530T_REVC, 0, "D-Link DGE-530(T) Gigabit Ethernet Adapter" }, { RT_VENDORID, RT_DEVICEID_8139, 0, "RealTek 8139C+ 10/100BaseTX" }, { RT_VENDORID, RT_DEVICEID_8101E, 0, "RealTek 810xE PCIe 10/100baseTX" }, { RT_VENDORID, RT_DEVICEID_8168, 0, "RealTek 8168/8111 B/C/CP/D/DP/E/F PCIe Gigabit Ethernet" }, { RT_VENDORID, RT_DEVICEID_8169, 0, "RealTek 8169/8169S/8169SB(L)/8110S/8110SB(L) Gigabit Ethernet" }, { RT_VENDORID, RT_DEVICEID_8169SC, 0, "RealTek 8169SC/8110SC Single-chip Gigabit Ethernet" }, { COREGA_VENDORID, COREGA_DEVICEID_CGLAPCIGT, 0, "Corega CG-LAPCIGT (RTL8169S) Gigabit Ethernet" }, { LINKSYS_VENDORID, LINKSYS_DEVICEID_EG1032, 0, "Linksys EG1032 (RTL8169S) Gigabit Ethernet" }, { USR_VENDORID, USR_DEVICEID_997902, 0, "US Robotics 997902 (RTL8169S) Gigabit Ethernet" } }; static struct rl_hwrev re_hwrevs[] = { { RL_HWREV_8139, RL_8139, "", RL_MTU }, { RL_HWREV_8139A, RL_8139, "A", RL_MTU }, { RL_HWREV_8139AG, RL_8139, "A-G", RL_MTU }, { RL_HWREV_8139B, RL_8139, "B", RL_MTU }, { RL_HWREV_8130, RL_8139, "8130", RL_MTU }, { RL_HWREV_8139C, RL_8139, "C", RL_MTU }, { RL_HWREV_8139D, RL_8139, "8139D/8100B/8100C", RL_MTU }, { RL_HWREV_8139CPLUS, RL_8139CPLUS, "C+", RL_MTU }, { RL_HWREV_8168B_SPIN1, RL_8169, "8168", RL_JUMBO_MTU }, { RL_HWREV_8169, RL_8169, "8169", RL_JUMBO_MTU }, { RL_HWREV_8169S, RL_8169, "8169S", RL_JUMBO_MTU }, { RL_HWREV_8110S, RL_8169, "8110S", RL_JUMBO_MTU }, { RL_HWREV_8169_8110SB, RL_8169, "8169SB/8110SB", RL_JUMBO_MTU }, { RL_HWREV_8169_8110SC, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU }, { RL_HWREV_8169_8110SBL, RL_8169, "8169SBL/8110SBL", RL_JUMBO_MTU }, { RL_HWREV_8169_8110SCE, RL_8169, "8169SC/8110SC", RL_JUMBO_MTU }, { RL_HWREV_8100, RL_8139, "8100", RL_MTU }, { RL_HWREV_8101, RL_8139, "8101", RL_MTU }, { RL_HWREV_8100E, RL_8169, "8100E", RL_MTU }, { RL_HWREV_8101E, RL_8169, "8101E", RL_MTU }, { RL_HWREV_8102E, RL_8169, "8102E", RL_MTU }, { RL_HWREV_8102EL, RL_8169, "8102EL", RL_MTU }, { RL_HWREV_8102EL_SPIN1, RL_8169, "8102EL", RL_MTU }, { RL_HWREV_8103E, RL_8169, "8103E", RL_MTU }, { RL_HWREV_8401E, RL_8169, "8401E", RL_MTU }, { RL_HWREV_8402, RL_8169, "8402", RL_MTU }, { RL_HWREV_8105E, RL_8169, "8105E", RL_MTU }, { RL_HWREV_8105E_SPIN1, RL_8169, "8105E", RL_MTU }, { RL_HWREV_8168B_SPIN2, RL_8169, "8168", RL_JUMBO_MTU }, { RL_HWREV_8168B_SPIN3, RL_8169, "8168", RL_JUMBO_MTU }, { RL_HWREV_8168C, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K }, { RL_HWREV_8168C_SPIN2, RL_8169, "8168C/8111C", RL_JUMBO_MTU_6K }, { RL_HWREV_8168CP, RL_8169, "8168CP/8111CP", RL_JUMBO_MTU_6K }, { RL_HWREV_8168D, RL_8169, "8168D/8111D", RL_JUMBO_MTU_9K }, { RL_HWREV_8168DP, RL_8169, "8168DP/8111DP", RL_JUMBO_MTU_9K }, { RL_HWREV_8168E, RL_8169, "8168E/8111E", RL_JUMBO_MTU_9K}, { RL_HWREV_8168E_VL, RL_8169, "8168E/8111E-VL", RL_JUMBO_MTU_6K}, { RL_HWREV_8168F, RL_8169, "8168F/8111F", RL_JUMBO_MTU_9K}, { RL_HWREV_8411, RL_8169, "8411", RL_JUMBO_MTU_9K}, { 0, 0, NULL, 0 } }; static int re_probe (device_t); static int re_attach (device_t); static int re_detach (device_t); static int re_encap (struct rl_softc *, struct mbuf **); static void re_dma_map_addr (void *, bus_dma_segment_t *, int, int); static int re_allocmem (device_t, struct rl_softc *); static __inline void re_discard_rxbuf (struct rl_softc *, int); static int re_newbuf (struct rl_softc *, int); static int re_jumbo_newbuf (struct rl_softc *, int); static int re_rx_list_init (struct rl_softc *); static int re_jrx_list_init (struct rl_softc *); static int re_tx_list_init (struct rl_softc *); #ifdef RE_FIXUP_RX static __inline void re_fixup_rx (struct mbuf *); #endif static int re_rxeof (struct rl_softc *); static void re_txeof (struct rl_softc *); #ifdef DEVICE_POLLING static void re_poll (struct ifnet *, enum poll_cmd, int); static void re_poll_locked (struct ifnet *, enum poll_cmd, int); #endif static int re_intr (void *); static void re_intr_msi (void *); static void re_tick (void *); static void re_int_task (void *, int); static void re_start (struct ifnet *); static void re_start_locked (struct ifnet *); static int re_ioctl (struct ifnet *, u_long, caddr_t); static void re_init (void *); static void re_init_locked (struct rl_softc *); static void re_stop (struct rl_softc *); static void re_watchdog (struct rl_softc *); static int re_suspend (device_t); static int re_resume (device_t); static int re_shutdown (device_t); static int re_ifmedia_upd (struct ifnet *); static void re_ifmedia_sts (struct ifnet *, struct ifmediareq *); static void re_eeprom_putbyte (struct rl_softc *, int); static void re_eeprom_getword (struct rl_softc *, int, u_int16_t *); static void re_read_eeprom (struct rl_softc *, caddr_t, int, int); static int re_gmii_readreg (device_t, int, int); static int re_gmii_writereg (device_t, int, int, int); static int re_miibus_readreg (device_t, int, int); static int re_miibus_writereg (device_t, int, int, int); static void re_miibus_statchg (device_t); static void re_set_jumbo (struct rl_softc *, int); static void re_set_rxmode (struct rl_softc *); static void re_reset (struct rl_softc *); static void re_setwol (struct rl_softc *); static void re_clrwol (struct rl_softc *); static void re_set_linkspeed (struct rl_softc *); #ifdef RE_DIAG static int re_diag (struct rl_softc *); #endif static void re_add_sysctls (struct rl_softc *); static int re_sysctl_stats (SYSCTL_HANDLER_ARGS); static int sysctl_int_range (SYSCTL_HANDLER_ARGS, int, int); static int sysctl_hw_re_int_mod (SYSCTL_HANDLER_ARGS); static device_method_t re_methods[] = { /* Device interface */ DEVMETHOD(device_probe, re_probe), DEVMETHOD(device_attach, re_attach), DEVMETHOD(device_detach, re_detach), DEVMETHOD(device_suspend, re_suspend), DEVMETHOD(device_resume, re_resume), DEVMETHOD(device_shutdown, re_shutdown), /* bus interface */ DEVMETHOD(bus_print_child, bus_generic_print_child), DEVMETHOD(bus_driver_added, bus_generic_driver_added), /* MII interface */ DEVMETHOD(miibus_readreg, re_miibus_readreg), DEVMETHOD(miibus_writereg, re_miibus_writereg), DEVMETHOD(miibus_statchg, re_miibus_statchg), { 0, 0 } }; static driver_t re_driver = { "re", re_methods, sizeof(struct rl_softc) }; static devclass_t re_devclass; DRIVER_MODULE(re, pci, re_driver, re_devclass, 0, 0); DRIVER_MODULE(re, cardbus, re_driver, re_devclass, 0, 0); DRIVER_MODULE(miibus, re, miibus_driver, miibus_devclass, 0, 0); #define EE_SET(x) \ CSR_WRITE_1(sc, RL_EECMD, \ CSR_READ_1(sc, RL_EECMD) | x) #define EE_CLR(x) \ CSR_WRITE_1(sc, RL_EECMD, \ CSR_READ_1(sc, RL_EECMD) & ~x) /* * Send a read command and address to the EEPROM, check for ACK. */ static void re_eeprom_putbyte(struct rl_softc *sc, int addr) { int d, i; d = addr | (RL_9346_READ << sc->rl_eewidth); /* * Feed in each bit and strobe the clock. */ for (i = 1 << (sc->rl_eewidth + 3); i; i >>= 1) { if (d & i) { EE_SET(RL_EE_DATAIN); } else { EE_CLR(RL_EE_DATAIN); } DELAY(100); EE_SET(RL_EE_CLK); DELAY(150); EE_CLR(RL_EE_CLK); DELAY(100); } } /* * Read a word of data stored in the EEPROM at address 'addr.' */ static void re_eeprom_getword(struct rl_softc *sc, int addr, u_int16_t *dest) { int i; u_int16_t word = 0; /* * Send address of word we want to read. */ re_eeprom_putbyte(sc, addr); /* * Start reading bits from EEPROM. */ for (i = 0x8000; i; i >>= 1) { EE_SET(RL_EE_CLK); DELAY(100); if (CSR_READ_1(sc, RL_EECMD) & RL_EE_DATAOUT) word |= i; EE_CLR(RL_EE_CLK); DELAY(100); } *dest = word; } /* * Read a sequence of words from the EEPROM. */ static void re_read_eeprom(struct rl_softc *sc, caddr_t dest, int off, int cnt) { int i; u_int16_t word = 0, *ptr; CSR_SETBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM); DELAY(100); for (i = 0; i < cnt; i++) { CSR_SETBIT_1(sc, RL_EECMD, RL_EE_SEL); re_eeprom_getword(sc, off + i, &word); CSR_CLRBIT_1(sc, RL_EECMD, RL_EE_SEL); ptr = (u_int16_t *)(dest + (i * 2)); *ptr = word; } CSR_CLRBIT_1(sc, RL_EECMD, RL_EEMODE_PROGRAM); } static int re_gmii_readreg(device_t dev, int phy, int reg) { struct rl_softc *sc; u_int32_t rval; int i; sc = device_get_softc(dev); /* Let the rgephy driver read the GMEDIASTAT register */ if (reg == RL_GMEDIASTAT) { rval = CSR_READ_1(sc, RL_GMEDIASTAT); return (rval); } CSR_WRITE_4(sc, RL_PHYAR, reg << 16); for (i = 0; i < RL_PHY_TIMEOUT; i++) { rval = CSR_READ_4(sc, RL_PHYAR); if (rval & RL_PHYAR_BUSY) break; DELAY(25); } if (i == RL_PHY_TIMEOUT) { device_printf(sc->rl_dev, "PHY read failed\n"); return (0); } /* * Controller requires a 20us delay to process next MDIO request. */ DELAY(20); return (rval & RL_PHYAR_PHYDATA); } static int re_gmii_writereg(device_t dev, int phy, int reg, int data) { struct rl_softc *sc; u_int32_t rval; int i; sc = device_get_softc(dev); CSR_WRITE_4(sc, RL_PHYAR, (reg << 16) | (data & RL_PHYAR_PHYDATA) | RL_PHYAR_BUSY); for (i = 0; i < RL_PHY_TIMEOUT; i++) { rval = CSR_READ_4(sc, RL_PHYAR); if (!(rval & RL_PHYAR_BUSY)) break; DELAY(25); } if (i == RL_PHY_TIMEOUT) { device_printf(sc->rl_dev, "PHY write failed\n"); return (0); } /* * Controller requires a 20us delay to process next MDIO request. */ DELAY(20); return (0); } static int re_miibus_readreg(device_t dev, int phy, int reg) { struct rl_softc *sc; u_int16_t rval = 0; u_int16_t re8139_reg = 0; sc = device_get_softc(dev); if (sc->rl_type == RL_8169) { rval = re_gmii_readreg(dev, phy, reg); return (rval); } switch (reg) { case MII_BMCR: re8139_reg = RL_BMCR; break; case MII_BMSR: re8139_reg = RL_BMSR; break; case MII_ANAR: re8139_reg = RL_ANAR; break; case MII_ANER: re8139_reg = RL_ANER; break; case MII_ANLPAR: re8139_reg = RL_LPAR; break; case MII_PHYIDR1: case MII_PHYIDR2: return (0); /* * Allow the rlphy driver to read the media status * register. If we have a link partner which does not * support NWAY, this is the register which will tell * us the results of parallel detection. */ case RL_MEDIASTAT: rval = CSR_READ_1(sc, RL_MEDIASTAT); return (rval); default: device_printf(sc->rl_dev, "bad phy register\n"); return (0); } rval = CSR_READ_2(sc, re8139_reg); if (sc->rl_type == RL_8139CPLUS && re8139_reg == RL_BMCR) { /* 8139C+ has different bit layout. */ rval &= ~(BMCR_LOOP | BMCR_ISO); } return (rval); } static int re_miibus_writereg(device_t dev, int phy, int reg, int data) { struct rl_softc *sc; u_int16_t re8139_reg = 0; int rval = 0; sc = device_get_softc(dev); if (sc->rl_type == RL_8169) { rval = re_gmii_writereg(dev, phy, reg, data); return (rval); } switch (reg) { case MII_BMCR: re8139_reg = RL_BMCR; if (sc->rl_type == RL_8139CPLUS) { /* 8139C+ has different bit layout. */ data &= ~(BMCR_LOOP | BMCR_ISO); } break; case MII_BMSR: re8139_reg = RL_BMSR; break; case MII_ANAR: re8139_reg = RL_ANAR; break; case MII_ANER: re8139_reg = RL_ANER; break; case MII_ANLPAR: re8139_reg = RL_LPAR; break; case MII_PHYIDR1: case MII_PHYIDR2: return (0); break; default: device_printf(sc->rl_dev, "bad phy register\n"); return (0); } CSR_WRITE_2(sc, re8139_reg, data); return (0); } static void re_miibus_statchg(device_t dev) { struct rl_softc *sc; struct ifnet *ifp; struct mii_data *mii; sc = device_get_softc(dev); mii = device_get_softc(sc->rl_miibus); ifp = sc->rl_ifp; if (mii == NULL || ifp == NULL || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; sc->rl_flags &= ~RL_FLAG_LINK; if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: sc->rl_flags |= RL_FLAG_LINK; break; case IFM_1000_T: if ((sc->rl_flags & RL_FLAG_FASTETHER) != 0) break; sc->rl_flags |= RL_FLAG_LINK; break; default: break; } } /* * RealTek controllers does not provide any interface to * Tx/Rx MACs for resolved speed, duplex and flow-control * parameters. */ } /* * Set the RX configuration and 64-bit multicast hash filter. */ static void re_set_rxmode(struct rl_softc *sc) { struct ifnet *ifp; struct ifmultiaddr *ifma; uint32_t hashes[2] = { 0, 0 }; uint32_t h, rxfilt; RL_LOCK_ASSERT(sc); ifp = sc->rl_ifp; rxfilt = RL_RXCFG_CONFIG | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_BROAD; if (ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) { if (ifp->if_flags & IFF_PROMISC) rxfilt |= RL_RXCFG_RX_ALLPHYS; /* * Unlike other hardwares, we have to explicitly set * RL_RXCFG_RX_MULTI to receive multicast frames in * promiscuous mode. */ rxfilt |= RL_RXCFG_RX_MULTI; hashes[0] = hashes[1] = 0xffffffff; goto done; } IF_ADDR_LOCK(ifp); TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; h = ether_crc32_be(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), ETHER_ADDR_LEN) >> 26; if (h < 32) hashes[0] |= (1 << h); else hashes[1] |= (1 << (h - 32)); } IF_ADDR_UNLOCK(ifp); if (hashes[0] != 0 || hashes[1] != 0) { /* * For some unfathomable reason, RealTek decided to * reverse the order of the multicast hash registers * in the PCI Express parts. This means we have to * write the hash pattern in reverse order for those * devices. */ if ((sc->rl_flags & RL_FLAG_PCIE) != 0) { h = bswap32(hashes[0]); hashes[0] = bswap32(hashes[1]); hashes[1] = h; } rxfilt |= RL_RXCFG_RX_MULTI; } done: CSR_WRITE_4(sc, RL_MAR0, hashes[0]); CSR_WRITE_4(sc, RL_MAR4, hashes[1]); CSR_WRITE_4(sc, RL_RXCFG, rxfilt); } static void re_reset(struct rl_softc *sc) { int i; RL_LOCK_ASSERT(sc); CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RESET); for (i = 0; i < RL_TIMEOUT; i++) { DELAY(10); if (!(CSR_READ_1(sc, RL_COMMAND) & RL_CMD_RESET)) break; } if (i == RL_TIMEOUT) device_printf(sc->rl_dev, "reset never completed!\n"); if ((sc->rl_flags & RL_FLAG_MACRESET) != 0) CSR_WRITE_1(sc, 0x82, 1); if (sc->rl_hwrev->rl_rev == RL_HWREV_8169S) re_gmii_writereg(sc->rl_dev, 1, 0x0b, 0); } #ifdef RE_DIAG /* * The following routine is designed to test for a defect on some * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64# * lines connected to the bus, however for a 32-bit only card, they * should be pulled high. The result of this defect is that the * NIC will not work right if you plug it into a 64-bit slot: DMA * operations will be done with 64-bit transfers, which will fail * because the 64-bit data lines aren't connected. * * There's no way to work around this (short of talking a soldering * iron to the board), however we can detect it. The method we use * here is to put the NIC into digital loopback mode, set the receiver * to promiscuous mode, and then try to send a frame. We then compare * the frame data we sent to what was received. If the data matches, * then the NIC is working correctly, otherwise we know the user has * a defective NIC which has been mistakenly plugged into a 64-bit PCI * slot. In the latter case, there's no way the NIC can work correctly, * so we print out a message on the console and abort the device attach. */ static int re_diag(struct rl_softc *sc) { struct ifnet *ifp = sc->rl_ifp; struct mbuf *m0; struct ether_header *eh; struct rl_desc *cur_rx; u_int16_t status; u_int32_t rxstat; int total_len, i, error = 0, phyaddr; u_int8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' }; u_int8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' }; /* Allocate a single mbuf */ MGETHDR(m0, M_DONTWAIT, MT_DATA); if (m0 == NULL) return (ENOBUFS); RL_LOCK(sc); /* * Initialize the NIC in test mode. This sets the chip up * so that it can send and receive frames, but performs the * following special functions: * - Puts receiver in promiscuous mode * - Enables digital loopback mode * - Leaves interrupts turned off */ ifp->if_flags |= IFF_PROMISC; sc->rl_testmode = 1; ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); sc->rl_flags |= RL_FLAG_LINK; if (sc->rl_type == RL_8169) phyaddr = 1; else phyaddr = 0; re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_RESET); for (i = 0; i < RL_TIMEOUT; i++) { status = re_miibus_readreg(sc->rl_dev, phyaddr, MII_BMCR); if (!(status & BMCR_RESET)) break; } re_miibus_writereg(sc->rl_dev, phyaddr, MII_BMCR, BMCR_LOOP); CSR_WRITE_2(sc, RL_ISR, RL_INTRS); DELAY(100000); /* Put some data in the mbuf */ eh = mtod(m0, struct ether_header *); bcopy ((char *)&dst, eh->ether_dhost, ETHER_ADDR_LEN); bcopy ((char *)&src, eh->ether_shost, ETHER_ADDR_LEN); eh->ether_type = htons(ETHERTYPE_IP); m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN; /* * Queue the packet, start transmission. * Note: IF_HANDOFF() ultimately calls re_start() for us. */ CSR_WRITE_2(sc, RL_ISR, 0xFFFF); RL_UNLOCK(sc); /* XXX: re_diag must not be called when in ALTQ mode */ IF_HANDOFF(&ifp->if_snd, m0, ifp); RL_LOCK(sc); m0 = NULL; /* Wait for it to propagate through the chip */ DELAY(100000); for (i = 0; i < RL_TIMEOUT; i++) { status = CSR_READ_2(sc, RL_ISR); CSR_WRITE_2(sc, RL_ISR, status); if ((status & (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) == (RL_ISR_TIMEOUT_EXPIRED|RL_ISR_RX_OK)) break; DELAY(10); } if (i == RL_TIMEOUT) { device_printf(sc->rl_dev, "diagnostic failed, failed to receive packet in" " loopback mode\n"); error = EIO; goto done; } /* * The packet should have been dumped into the first * entry in the RX DMA ring. Grab it from there. */ bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, BUS_DMASYNC_POSTREAD); bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, sc->rl_ldata.rl_rx_desc[0].rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, sc->rl_ldata.rl_rx_desc[0].rx_dmamap); m0 = sc->rl_ldata.rl_rx_desc[0].rx_m; sc->rl_ldata.rl_rx_desc[0].rx_m = NULL; eh = mtod(m0, struct ether_header *); cur_rx = &sc->rl_ldata.rl_rx_list[0]; total_len = RL_RXBYTES(cur_rx); rxstat = le32toh(cur_rx->rl_cmdstat); if (total_len != ETHER_MIN_LEN) { device_printf(sc->rl_dev, "diagnostic failed, received short packet\n"); error = EIO; goto done; } /* Test that the received packet data matches what we sent. */ if (bcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) || bcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) || ntohs(eh->ether_type) != ETHERTYPE_IP) { device_printf(sc->rl_dev, "WARNING, DMA FAILURE!\n"); device_printf(sc->rl_dev, "expected TX data: %6D/%6D/0x%x\n", dst, ":", src, ":", ETHERTYPE_IP); device_printf(sc->rl_dev, "received RX data: %6D/%6D/0x%x\n", eh->ether_dhost, ":", eh->ether_shost, ":", ntohs(eh->ether_type)); device_printf(sc->rl_dev, "You may have a defective 32-bit " "NIC plugged into a 64-bit PCI slot.\n"); device_printf(sc->rl_dev, "Please re-install the NIC in a " "32-bit slot for proper operation.\n"); device_printf(sc->rl_dev, "Read the re(4) man page for more " "details.\n"); error = EIO; } done: /* Turn interface off, release resources */ sc->rl_testmode = 0; sc->rl_flags &= ~RL_FLAG_LINK; ifp->if_flags &= ~IFF_PROMISC; re_stop(sc); if (m0 != NULL) m_freem(m0); RL_UNLOCK(sc); return (error); } #endif /* * Probe for a RealTek 8139C+/8169/8110 chip. Check the PCI vendor and device * IDs against our list and return a device name if we find a match. */ static int re_probe(device_t dev) { struct rl_type *t; uint16_t devid, vendor; uint16_t revid, sdevid; int i; vendor = pci_get_vendor(dev); devid = pci_get_device(dev); revid = pci_get_revid(dev); sdevid = pci_get_subdevice(dev); if (vendor == LINKSYS_VENDORID && devid == LINKSYS_DEVICEID_EG1032) { if (sdevid != LINKSYS_SUBDEVICE_EG1032_REV3) { /* * Only attach to rev. 3 of the Linksys EG1032 adapter. * Rev. 2 is supported by sk(4). */ return (ENXIO); } } if (vendor == RT_VENDORID && devid == RT_DEVICEID_8139) { if (revid != 0x20) { /* 8139, let rl(4) take care of this device. */ return (ENXIO); } } t = re_devs; for (i = 0; i < sizeof(re_devs) / sizeof(re_devs[0]); i++, t++) { if (vendor == t->rl_vid && devid == t->rl_did) { device_set_desc(dev, t->rl_name); return (BUS_PROBE_DEFAULT); } } return (ENXIO); } /* * Map a single buffer address. */ static void re_dma_map_addr(void *arg, bus_dma_segment_t *segs, int nseg, int error) { bus_addr_t *addr; if (error) return; KASSERT(nseg == 1, ("too many DMA segments, %d should be 1", nseg)); addr = arg; *addr = segs->ds_addr; } static int re_allocmem(device_t dev, struct rl_softc *sc) { bus_addr_t lowaddr; bus_size_t rx_list_size, tx_list_size; int error; int i; rx_list_size = sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc); tx_list_size = sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc); /* * Allocate the parent bus DMA tag appropriate for PCI. * In order to use DAC, RL_CPLUSCMD_PCI_DAC bit of RL_CPLUS_CMD * register should be set. However some RealTek chips are known * to be buggy on DAC handling, therefore disable DAC by limiting * DMA address space to 32bit. PCIe variants of RealTek chips * may not have the limitation. */ lowaddr = BUS_SPACE_MAXADDR; if ((sc->rl_flags & RL_FLAG_PCIE) == 0) lowaddr = BUS_SPACE_MAXADDR_32BIT; error = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, lowaddr, BUS_SPACE_MAXADDR, NULL, NULL, BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT, 0, NULL, NULL, &sc->rl_parent_tag); if (error) { device_printf(dev, "could not allocate parent DMA tag\n"); return (error); } /* * Allocate map for TX mbufs. */ error = bus_dma_tag_create(sc->rl_parent_tag, 1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES * RL_NTXSEGS, RL_NTXSEGS, 4096, 0, NULL, NULL, &sc->rl_ldata.rl_tx_mtag); if (error) { device_printf(dev, "could not allocate TX DMA tag\n"); return (error); } /* * Allocate map for RX mbufs. */ if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) { error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MJUM9BYTES, 1, MJUM9BYTES, 0, NULL, NULL, &sc->rl_ldata.rl_jrx_mtag); if (error) { device_printf(dev, "could not allocate jumbo RX DMA tag\n"); return (error); } } error = bus_dma_tag_create(sc->rl_parent_tag, sizeof(uint64_t), 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, MCLBYTES, 1, MCLBYTES, 0, NULL, NULL, &sc->rl_ldata.rl_rx_mtag); if (error) { device_printf(dev, "could not allocate RX DMA tag\n"); return (error); } /* * Allocate map for TX descriptor list. */ error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN, 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, tx_list_size, 1, tx_list_size, 0, NULL, NULL, &sc->rl_ldata.rl_tx_list_tag); if (error) { device_printf(dev, "could not allocate TX DMA ring tag\n"); return (error); } /* Allocate DMA'able memory for the TX ring */ error = bus_dmamem_alloc(sc->rl_ldata.rl_tx_list_tag, (void **)&sc->rl_ldata.rl_tx_list, BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO, &sc->rl_ldata.rl_tx_list_map); if (error) { device_printf(dev, "could not allocate TX DMA ring\n"); return (error); } /* Load the map for the TX ring. */ sc->rl_ldata.rl_tx_list_addr = 0; error = bus_dmamap_load(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list_map, sc->rl_ldata.rl_tx_list, tx_list_size, re_dma_map_addr, &sc->rl_ldata.rl_tx_list_addr, BUS_DMA_NOWAIT); if (error != 0 || sc->rl_ldata.rl_tx_list_addr == 0) { device_printf(dev, "could not load TX DMA ring\n"); return (ENOMEM); } /* Create DMA maps for TX buffers */ for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) { error = bus_dmamap_create(sc->rl_ldata.rl_tx_mtag, 0, &sc->rl_ldata.rl_tx_desc[i].tx_dmamap); if (error) { device_printf(dev, "could not create DMA map for TX\n"); return (error); } } /* * Allocate map for RX descriptor list. */ error = bus_dma_tag_create(sc->rl_parent_tag, RL_RING_ALIGN, 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, rx_list_size, 1, rx_list_size, 0, NULL, NULL, &sc->rl_ldata.rl_rx_list_tag); if (error) { device_printf(dev, "could not create RX DMA ring tag\n"); return (error); } /* Allocate DMA'able memory for the RX ring */ error = bus_dmamem_alloc(sc->rl_ldata.rl_rx_list_tag, (void **)&sc->rl_ldata.rl_rx_list, BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO, &sc->rl_ldata.rl_rx_list_map); if (error) { device_printf(dev, "could not allocate RX DMA ring\n"); return (error); } /* Load the map for the RX ring. */ sc->rl_ldata.rl_rx_list_addr = 0; error = bus_dmamap_load(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, sc->rl_ldata.rl_rx_list, rx_list_size, re_dma_map_addr, &sc->rl_ldata.rl_rx_list_addr, BUS_DMA_NOWAIT); if (error != 0 || sc->rl_ldata.rl_rx_list_addr == 0) { device_printf(dev, "could not load RX DMA ring\n"); return (ENOMEM); } /* Create DMA maps for RX buffers */ if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) { error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0, &sc->rl_ldata.rl_jrx_sparemap); if (error) { device_printf(dev, "could not create spare DMA map for jumbo RX\n"); return (error); } for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { error = bus_dmamap_create(sc->rl_ldata.rl_jrx_mtag, 0, &sc->rl_ldata.rl_jrx_desc[i].rx_dmamap); if (error) { device_printf(dev, "could not create DMA map for jumbo RX\n"); return (error); } } } error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0, &sc->rl_ldata.rl_rx_sparemap); if (error) { device_printf(dev, "could not create spare DMA map for RX\n"); return (error); } for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { error = bus_dmamap_create(sc->rl_ldata.rl_rx_mtag, 0, &sc->rl_ldata.rl_rx_desc[i].rx_dmamap); if (error) { device_printf(dev, "could not create DMA map for RX\n"); return (error); } } /* Create DMA map for statistics. */ error = bus_dma_tag_create(sc->rl_parent_tag, RL_DUMP_ALIGN, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, sizeof(struct rl_stats), 1, sizeof(struct rl_stats), 0, NULL, NULL, &sc->rl_ldata.rl_stag); if (error) { device_printf(dev, "could not create statistics DMA tag\n"); return (error); } /* Allocate DMA'able memory for statistics. */ error = bus_dmamem_alloc(sc->rl_ldata.rl_stag, (void **)&sc->rl_ldata.rl_stats, BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO, &sc->rl_ldata.rl_smap); if (error) { device_printf(dev, "could not allocate statistics DMA memory\n"); return (error); } /* Load the map for statistics. */ sc->rl_ldata.rl_stats_addr = 0; error = bus_dmamap_load(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap, sc->rl_ldata.rl_stats, sizeof(struct rl_stats), re_dma_map_addr, &sc->rl_ldata.rl_stats_addr, BUS_DMA_NOWAIT); if (error != 0 || sc->rl_ldata.rl_stats_addr == 0) { device_printf(dev, "could not load statistics DMA memory\n"); return (ENOMEM); } return (0); } /* * Attach the interface. Allocate softc structures, do ifmedia * setup and ethernet/BPF attach. */ static int re_attach(device_t dev) { u_char eaddr[ETHER_ADDR_LEN]; u_int16_t as[ETHER_ADDR_LEN / 2]; struct rl_softc *sc; struct ifnet *ifp; struct rl_hwrev *hw_rev; u_int32_t cap, ctl; int hwrev; u_int16_t devid, re_did = 0; int error = 0, i, phy, rid; int msic, msixc, reg; uint8_t cfg; sc = device_get_softc(dev); sc->rl_dev = dev; mtx_init(&sc->rl_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK, MTX_DEF); callout_init_mtx(&sc->rl_stat_callout, &sc->rl_mtx, 0); /* * Map control/status registers. */ pci_enable_busmaster(dev); devid = pci_get_device(dev); /* * Prefer memory space register mapping over IO space. * Because RTL8169SC does not seem to work when memory mapping * is used always activate io mapping. */ if (devid == RT_DEVICEID_8169SC) prefer_iomap = 1; if (prefer_iomap == 0) { sc->rl_res_id = PCIR_BAR(1); sc->rl_res_type = SYS_RES_MEMORY; /* RTL8168/8101E seems to use different BARs. */ if (devid == RT_DEVICEID_8168 || devid == RT_DEVICEID_8101E) sc->rl_res_id = PCIR_BAR(2); } else { sc->rl_res_id = PCIR_BAR(0); sc->rl_res_type = SYS_RES_IOPORT; } sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type, &sc->rl_res_id, RF_ACTIVE); if (sc->rl_res == NULL && prefer_iomap == 0) { sc->rl_res_id = PCIR_BAR(0); sc->rl_res_type = SYS_RES_IOPORT; sc->rl_res = bus_alloc_resource_any(dev, sc->rl_res_type, &sc->rl_res_id, RF_ACTIVE); } if (sc->rl_res == NULL) { device_printf(dev, "couldn't map ports/memory\n"); error = ENXIO; goto fail; } sc->rl_btag = rman_get_bustag(sc->rl_res); sc->rl_bhandle = rman_get_bushandle(sc->rl_res); msic = pci_msi_count(dev); msixc = pci_msix_count(dev); if (pci_find_extcap(dev, PCIY_EXPRESS, ®) == 0) { sc->rl_flags |= RL_FLAG_PCIE; sc->rl_expcap = reg; } if (bootverbose) { device_printf(dev, "MSI count : %d\n", msic); device_printf(dev, "MSI-X count : %d\n", msixc); } if (msix_disable > 0) msixc = 0; if (msi_disable > 0) msic = 0; /* Prefer MSI-X to MSI. */ if (msixc > 0) { msixc = 1; rid = PCIR_BAR(4); sc->rl_res_pba = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->rl_res_pba == NULL) { device_printf(sc->rl_dev, "could not allocate MSI-X PBA resource\n"); } if (sc->rl_res_pba != NULL && pci_alloc_msix(dev, &msixc) == 0) { if (msixc == 1) { device_printf(dev, "Using %d MSI-X message\n", msixc); sc->rl_flags |= RL_FLAG_MSIX; } else pci_release_msi(dev); } if ((sc->rl_flags & RL_FLAG_MSIX) == 0) { if (sc->rl_res_pba != NULL) bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba); sc->rl_res_pba = NULL; msixc = 0; } } /* Prefer MSI to INTx. */ if (msixc == 0 && msic > 0) { msic = 1; if (pci_alloc_msi(dev, &msic) == 0) { if (msic == RL_MSI_MESSAGES) { device_printf(dev, "Using %d MSI message\n", msic); sc->rl_flags |= RL_FLAG_MSI; /* Explicitly set MSI enable bit. */ CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE); cfg = CSR_READ_1(sc, RL_CFG2); cfg |= RL_CFG2_MSI; CSR_WRITE_1(sc, RL_CFG2, cfg); CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); } else pci_release_msi(dev); } if ((sc->rl_flags & RL_FLAG_MSI) == 0) msic = 0; } /* Allocate interrupt */ if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) { rid = 0; sc->rl_irq[0] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE); if (sc->rl_irq[0] == NULL) { device_printf(dev, "couldn't allocate IRQ resources\n"); error = ENXIO; goto fail; } } else { for (i = 0, rid = 1; i < RL_MSI_MESSAGES; i++, rid++) { sc->rl_irq[i] = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_ACTIVE); if (sc->rl_irq[i] == NULL) { device_printf(dev, "couldn't llocate IRQ resources for " "message %d\n", rid); error = ENXIO; goto fail; } } } if ((sc->rl_flags & RL_FLAG_MSI) == 0) { CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE); cfg = CSR_READ_1(sc, RL_CFG2); if ((cfg & RL_CFG2_MSI) != 0) { device_printf(dev, "turning off MSI enable bit.\n"); cfg &= ~RL_CFG2_MSI; CSR_WRITE_1(sc, RL_CFG2, cfg); } CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); } /* Disable ASPM L0S/L1. */ if (sc->rl_expcap != 0) { cap = pci_read_config(dev, sc->rl_expcap + - PCIR_EXPRESS_LINK_CAP, 2); - if ((cap & PCIM_LINK_CAP_ASPM) != 0) { + PCIER_LINK_CAP, 2); + if ((cap & PCIEM_LINK_CAP_ASPM) != 0) { ctl = pci_read_config(dev, sc->rl_expcap + - PCIR_EXPRESS_LINK_CTL, 2); + PCIER_LINK_CTL, 2); if ((ctl & 0x0003) != 0) { ctl &= ~0x0003; pci_write_config(dev, sc->rl_expcap + - PCIR_EXPRESS_LINK_CTL, ctl, 2); + PCIER_LINK_CTL, ctl, 2); device_printf(dev, "ASPM disabled\n"); } } else device_printf(dev, "no ASPM capability\n"); } hw_rev = re_hwrevs; hwrev = CSR_READ_4(sc, RL_TXCFG); switch (hwrev & 0x70000000) { case 0x00000000: case 0x10000000: device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0xfc800000); hwrev &= (RL_TXCFG_HWREV | 0x80000000); break; default: device_printf(dev, "Chip rev. 0x%08x\n", hwrev & 0x7c800000); hwrev &= RL_TXCFG_HWREV; break; } device_printf(dev, "MAC rev. 0x%08x\n", hwrev & 0x00700000); while (hw_rev->rl_desc != NULL) { if (hw_rev->rl_rev == hwrev) { sc->rl_type = hw_rev->rl_type; sc->rl_hwrev = hw_rev; break; } hw_rev++; } if (hw_rev->rl_desc == NULL) { device_printf(dev, "Unknown H/W revision: 0x%08x\n", hwrev); error = ENXIO; goto fail; } switch (hw_rev->rl_rev) { case RL_HWREV_8139CPLUS: sc->rl_flags |= RL_FLAG_FASTETHER | RL_FLAG_AUTOPAD; break; case RL_HWREV_8100E: case RL_HWREV_8101E: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_FASTETHER; break; case RL_HWREV_8102E: case RL_HWREV_8102EL: case RL_HWREV_8102EL_SPIN1: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD; break; case RL_HWREV_8103E: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_MACSLEEP; break; case RL_HWREV_8401E: case RL_HWREV_8105E: case RL_HWREV_8105E_SPIN1: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD; break; case RL_HWREV_8402: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_FASTETHER | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_CMDSTOP_WAIT_TXQ; break; case RL_HWREV_8168B_SPIN1: case RL_HWREV_8168B_SPIN2: sc->rl_flags |= RL_FLAG_WOLRXENB; /* FALLTHROUGH */ case RL_HWREV_8168B_SPIN3: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_MACSTAT; break; case RL_HWREV_8168C_SPIN2: sc->rl_flags |= RL_FLAG_MACSLEEP; /* FALLTHROUGH */ case RL_HWREV_8168C: if ((hwrev & 0x00700000) == 0x00200000) sc->rl_flags |= RL_FLAG_MACSLEEP; /* FALLTHROUGH */ case RL_HWREV_8168CP: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK; break; case RL_HWREV_8168D: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK; break; case RL_HWREV_8168DP: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WAIT_TXPOLL | RL_FLAG_WOL_MANLINK; break; case RL_HWREV_8168E: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PHYWAKE_PM | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_WOL_MANLINK; break; case RL_HWREV_8168E_VL: case RL_HWREV_8168F: case RL_HWREV_8411: sc->rl_flags |= RL_FLAG_PHYWAKE | RL_FLAG_PAR | RL_FLAG_DESCV2 | RL_FLAG_MACSTAT | RL_FLAG_CMDSTOP | RL_FLAG_AUTOPAD | RL_FLAG_JUMBOV2 | RL_FLAG_CMDSTOP_WAIT_TXQ | RL_FLAG_WOL_MANLINK; break; case RL_HWREV_8169_8110SB: case RL_HWREV_8169_8110SBL: case RL_HWREV_8169_8110SC: case RL_HWREV_8169_8110SCE: sc->rl_flags |= RL_FLAG_PHYWAKE; /* FALLTHROUGH */ case RL_HWREV_8169: case RL_HWREV_8169S: case RL_HWREV_8110S: sc->rl_flags |= RL_FLAG_MACRESET; break; default: break; } if (sc->rl_hwrev->rl_rev == RL_HWREV_8139CPLUS) { sc->rl_cfg0 = RL_8139_CFG0; sc->rl_cfg1 = RL_8139_CFG1; sc->rl_cfg2 = 0; sc->rl_cfg3 = RL_8139_CFG3; sc->rl_cfg4 = RL_8139_CFG4; sc->rl_cfg5 = RL_8139_CFG5; } else { sc->rl_cfg0 = RL_CFG0; sc->rl_cfg1 = RL_CFG1; sc->rl_cfg2 = RL_CFG2; sc->rl_cfg3 = RL_CFG3; sc->rl_cfg4 = RL_CFG4; sc->rl_cfg5 = RL_CFG5; } /* Reset the adapter. */ RL_LOCK(sc); re_reset(sc); RL_UNLOCK(sc); /* Enable PME. */ CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE); cfg = CSR_READ_1(sc, sc->rl_cfg1); cfg |= RL_CFG1_PME; CSR_WRITE_1(sc, sc->rl_cfg1, cfg); cfg = CSR_READ_1(sc, sc->rl_cfg5); cfg &= RL_CFG5_PME_STS; CSR_WRITE_1(sc, sc->rl_cfg5, cfg); CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); if ((sc->rl_flags & RL_FLAG_PAR) != 0) { /* * XXX Should have a better way to extract station * address from EEPROM. */ for (i = 0; i < ETHER_ADDR_LEN; i++) eaddr[i] = CSR_READ_1(sc, RL_IDR0 + i); } else { sc->rl_eewidth = RL_9356_ADDR_LEN; re_read_eeprom(sc, (caddr_t)&re_did, 0, 1); if (re_did != 0x8129) sc->rl_eewidth = RL_9346_ADDR_LEN; /* * Get station address from the EEPROM. */ re_read_eeprom(sc, (caddr_t)as, RL_EE_EADDR, 3); for (i = 0; i < ETHER_ADDR_LEN / 2; i++) as[i] = le16toh(as[i]); bcopy(as, eaddr, sizeof(eaddr)); } if (sc->rl_type == RL_8169) { /* Set RX length mask and number of descriptors. */ sc->rl_rxlenmask = RL_RDESC_STAT_GFRAGLEN; sc->rl_txstart = RL_GTXSTART; sc->rl_ldata.rl_tx_desc_cnt = RL_8169_TX_DESC_CNT; sc->rl_ldata.rl_rx_desc_cnt = RL_8169_RX_DESC_CNT; } else { /* Set RX length mask and number of descriptors. */ sc->rl_rxlenmask = RL_RDESC_STAT_FRAGLEN; sc->rl_txstart = RL_TXSTART; sc->rl_ldata.rl_tx_desc_cnt = RL_8139_TX_DESC_CNT; sc->rl_ldata.rl_rx_desc_cnt = RL_8139_RX_DESC_CNT; } error = re_allocmem(dev, sc); if (error) goto fail; re_add_sysctls(sc); ifp = sc->rl_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "can not if_alloc()\n"); error = ENOSPC; goto fail; } /* Take controller out of deep sleep mode. */ if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) { if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80) CSR_WRITE_1(sc, RL_GPIO, CSR_READ_1(sc, RL_GPIO) | 0x01); else CSR_WRITE_1(sc, RL_GPIO, CSR_READ_1(sc, RL_GPIO) & ~0x01); } /* Take PHY out of power down mode. */ if ((sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) { CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) | 0x80); if (hw_rev->rl_rev == RL_HWREV_8401E) CSR_WRITE_1(sc, 0xD1, CSR_READ_1(sc, 0xD1) & ~0x08); } if ((sc->rl_flags & RL_FLAG_PHYWAKE) != 0) { re_gmii_writereg(dev, 1, 0x1f, 0); re_gmii_writereg(dev, 1, 0x0e, 0); } ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = re_ioctl; ifp->if_start = re_start; /* * RTL8168/8111C generates wrong IP checksummed frame if the * packet has IP options so disable TX IP checksum offloading. */ if (sc->rl_hwrev->rl_rev == RL_HWREV_8168C || sc->rl_hwrev->rl_rev == RL_HWREV_8168C_SPIN2) ifp->if_hwassist = CSUM_TCP | CSUM_UDP; else ifp->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP; ifp->if_hwassist |= CSUM_TSO; ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_TSO4; ifp->if_capenable = ifp->if_capabilities; ifp->if_init = re_init; IFQ_SET_MAXLEN(&ifp->if_snd, RL_IFQ_MAXLEN); ifp->if_snd.ifq_drv_maxlen = RL_IFQ_MAXLEN; IFQ_SET_READY(&ifp->if_snd); TASK_INIT(&sc->rl_inttask, 0, re_int_task, sc); #define RE_PHYAD_INTERNAL 0 /* Do MII setup. */ phy = RE_PHYAD_INTERNAL; if (sc->rl_type == RL_8169) phy = 1; error = mii_attach(dev, &sc->rl_miibus, ifp, re_ifmedia_upd, re_ifmedia_sts, BMSR_DEFCAPMASK, phy, MII_OFFSET_ANY, MIIF_DOPAUSE); if (error != 0) { device_printf(dev, "attaching PHYs failed\n"); goto fail; } /* * Call MI attach routine. */ ether_ifattach(ifp, eaddr); /* VLAN capability setup */ ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING; if (ifp->if_capabilities & IFCAP_HWCSUM) ifp->if_capabilities |= IFCAP_VLAN_HWCSUM; /* Enable WOL if PM is supported. */ if (pci_find_extcap(sc->rl_dev, PCIY_PMG, ®) == 0) ifp->if_capabilities |= IFCAP_WOL; ifp->if_capenable = ifp->if_capabilities; ifp->if_capenable &= ~(IFCAP_WOL_UCAST | IFCAP_WOL_MCAST); /* * Don't enable TSO by default. It is known to generate * corrupted TCP segments(bad TCP options) under certain * circumtances. */ ifp->if_hwassist &= ~CSUM_TSO; ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO); #ifdef DEVICE_POLLING ifp->if_capabilities |= IFCAP_POLLING; #endif /* * Tell the upper layer(s) we support long frames. * Must appear after the call to ether_ifattach() because * ether_ifattach() sets ifi_hdrlen to the default value. */ ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); #ifdef RE_DIAG /* * Perform hardware diagnostic on the original RTL8169. * Some 32-bit cards were incorrectly wired and would * malfunction if plugged into a 64-bit slot. */ if (hwrev == RL_HWREV_8169) { error = re_diag(sc); if (error) { device_printf(dev, "attach aborted due to hardware diag failure\n"); ether_ifdetach(ifp); goto fail; } } #endif #ifdef RE_TX_MODERATION intr_filter = 1; #endif /* Hook interrupt last to avoid having to lock softc */ if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 && intr_filter == 0) { error = bus_setup_intr(dev, sc->rl_irq[0], INTR_TYPE_NET | INTR_MPSAFE, NULL, re_intr_msi, sc, &sc->rl_intrhand[0]); } else { error = bus_setup_intr(dev, sc->rl_irq[0], INTR_TYPE_NET | INTR_MPSAFE, re_intr, NULL, sc, &sc->rl_intrhand[0]); } if (error) { device_printf(dev, "couldn't set up irq\n"); ether_ifdetach(ifp); } fail: if (error) re_detach(dev); return (error); } /* * Shutdown hardware and free up resources. This can be called any * time after the mutex has been initialized. It is called in both * the error case in attach and the normal detach case so it needs * to be careful about only freeing resources that have actually been * allocated. */ static int re_detach(device_t dev) { struct rl_softc *sc; struct ifnet *ifp; int i, rid; sc = device_get_softc(dev); ifp = sc->rl_ifp; KASSERT(mtx_initialized(&sc->rl_mtx), ("re mutex not initialized")); /* These should only be active if attach succeeded */ if (device_is_attached(dev)) { #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) ether_poll_deregister(ifp); #endif RL_LOCK(sc); #if 0 sc->suspended = 1; #endif re_stop(sc); RL_UNLOCK(sc); callout_drain(&sc->rl_stat_callout); taskqueue_drain(taskqueue_fast, &sc->rl_inttask); /* * Force off the IFF_UP flag here, in case someone * still had a BPF descriptor attached to this * interface. If they do, ether_ifdetach() will cause * the BPF code to try and clear the promisc mode * flag, which will bubble down to re_ioctl(), * which will try to call re_init() again. This will * turn the NIC back on and restart the MII ticker, * which will panic the system when the kernel tries * to invoke the re_tick() function that isn't there * anymore. */ ifp->if_flags &= ~IFF_UP; ether_ifdetach(ifp); } if (sc->rl_miibus) device_delete_child(dev, sc->rl_miibus); bus_generic_detach(dev); /* * The rest is resource deallocation, so we should already be * stopped here. */ if (sc->rl_intrhand[0] != NULL) { bus_teardown_intr(dev, sc->rl_irq[0], sc->rl_intrhand[0]); sc->rl_intrhand[0] = NULL; } if (ifp != NULL) if_free(ifp); if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) rid = 0; else rid = 1; if (sc->rl_irq[0] != NULL) { bus_release_resource(dev, SYS_RES_IRQ, rid, sc->rl_irq[0]); sc->rl_irq[0] = NULL; } if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0) pci_release_msi(dev); if (sc->rl_res_pba) { rid = PCIR_BAR(4); bus_release_resource(dev, SYS_RES_MEMORY, rid, sc->rl_res_pba); } if (sc->rl_res) bus_release_resource(dev, sc->rl_res_type, sc->rl_res_id, sc->rl_res); /* Unload and free the RX DMA ring memory and map */ if (sc->rl_ldata.rl_rx_list_tag) { if (sc->rl_ldata.rl_rx_list_map) bus_dmamap_unload(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map); if (sc->rl_ldata.rl_rx_list_map && sc->rl_ldata.rl_rx_list) bus_dmamem_free(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list, sc->rl_ldata.rl_rx_list_map); bus_dma_tag_destroy(sc->rl_ldata.rl_rx_list_tag); } /* Unload and free the TX DMA ring memory and map */ if (sc->rl_ldata.rl_tx_list_tag) { if (sc->rl_ldata.rl_tx_list_map) bus_dmamap_unload(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list_map); if (sc->rl_ldata.rl_tx_list_map && sc->rl_ldata.rl_tx_list) bus_dmamem_free(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list, sc->rl_ldata.rl_tx_list_map); bus_dma_tag_destroy(sc->rl_ldata.rl_tx_list_tag); } /* Destroy all the RX and TX buffer maps */ if (sc->rl_ldata.rl_tx_mtag) { for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) { if (sc->rl_ldata.rl_tx_desc[i].tx_dmamap) bus_dmamap_destroy(sc->rl_ldata.rl_tx_mtag, sc->rl_ldata.rl_tx_desc[i].tx_dmamap); } bus_dma_tag_destroy(sc->rl_ldata.rl_tx_mtag); } if (sc->rl_ldata.rl_rx_mtag) { for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { if (sc->rl_ldata.rl_rx_desc[i].rx_dmamap) bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag, sc->rl_ldata.rl_rx_desc[i].rx_dmamap); } if (sc->rl_ldata.rl_rx_sparemap) bus_dmamap_destroy(sc->rl_ldata.rl_rx_mtag, sc->rl_ldata.rl_rx_sparemap); bus_dma_tag_destroy(sc->rl_ldata.rl_rx_mtag); } if (sc->rl_ldata.rl_jrx_mtag) { for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { if (sc->rl_ldata.rl_jrx_desc[i].rx_dmamap) bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag, sc->rl_ldata.rl_jrx_desc[i].rx_dmamap); } if (sc->rl_ldata.rl_jrx_sparemap) bus_dmamap_destroy(sc->rl_ldata.rl_jrx_mtag, sc->rl_ldata.rl_jrx_sparemap); bus_dma_tag_destroy(sc->rl_ldata.rl_jrx_mtag); } /* Unload and free the stats buffer and map */ if (sc->rl_ldata.rl_stag) { if (sc->rl_ldata.rl_smap) bus_dmamap_unload(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap); if (sc->rl_ldata.rl_smap && sc->rl_ldata.rl_stats) bus_dmamem_free(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_stats, sc->rl_ldata.rl_smap); bus_dma_tag_destroy(sc->rl_ldata.rl_stag); } if (sc->rl_parent_tag) bus_dma_tag_destroy(sc->rl_parent_tag); mtx_destroy(&sc->rl_mtx); return (0); } static __inline void re_discard_rxbuf(struct rl_softc *sc, int idx) { struct rl_desc *desc; struct rl_rxdesc *rxd; uint32_t cmdstat; if (sc->rl_ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0) rxd = &sc->rl_ldata.rl_jrx_desc[idx]; else rxd = &sc->rl_ldata.rl_rx_desc[idx]; desc = &sc->rl_ldata.rl_rx_list[idx]; desc->rl_vlanctl = 0; cmdstat = rxd->rx_size; if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1) cmdstat |= RL_RDESC_CMD_EOR; desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN); } static int re_newbuf(struct rl_softc *sc, int idx) { struct mbuf *m; struct rl_rxdesc *rxd; bus_dma_segment_t segs[1]; bus_dmamap_t map; struct rl_desc *desc; uint32_t cmdstat; int error, nsegs; m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MCLBYTES; #ifdef RE_FIXUP_RX /* * This is part of an evil trick to deal with non-x86 platforms. * The RealTek chip requires RX buffers to be aligned on 64-bit * boundaries, but that will hose non-x86 machines. To get around * this, we leave some empty space at the start of each buffer * and for non-x86 hosts, we copy the buffer back six bytes * to achieve word alignment. This is slightly more efficient * than allocating a new buffer, copying the contents, and * discarding the old buffer. */ m_adj(m, RE_ETHER_ALIGN); #endif error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_rx_mtag, sc->rl_ldata.rl_rx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT); if (error != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs)); rxd = &sc->rl_ldata.rl_rx_desc[idx]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap); } rxd->rx_m = m; map = rxd->rx_dmamap; rxd->rx_dmamap = sc->rl_ldata.rl_rx_sparemap; rxd->rx_size = segs[0].ds_len; sc->rl_ldata.rl_rx_sparemap = map; bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap, BUS_DMASYNC_PREREAD); desc = &sc->rl_ldata.rl_rx_list[idx]; desc->rl_vlanctl = 0; desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr)); desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr)); cmdstat = segs[0].ds_len; if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1) cmdstat |= RL_RDESC_CMD_EOR; desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN); return (0); } static int re_jumbo_newbuf(struct rl_softc *sc, int idx) { struct mbuf *m; struct rl_rxdesc *rxd; bus_dma_segment_t segs[1]; bus_dmamap_t map; struct rl_desc *desc; uint32_t cmdstat; int error, nsegs; m = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, MJUM9BYTES); if (m == NULL) return (ENOBUFS); m->m_len = m->m_pkthdr.len = MJUM9BYTES; #ifdef RE_FIXUP_RX m_adj(m, RE_ETHER_ALIGN); #endif error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_jrx_mtag, sc->rl_ldata.rl_jrx_sparemap, m, segs, &nsegs, BUS_DMA_NOWAIT); if (error != 0) { m_freem(m); return (ENOBUFS); } KASSERT(nsegs == 1, ("%s: %d segment returned!", __func__, nsegs)); rxd = &sc->rl_ldata.rl_jrx_desc[idx]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap); } rxd->rx_m = m; map = rxd->rx_dmamap; rxd->rx_dmamap = sc->rl_ldata.rl_jrx_sparemap; rxd->rx_size = segs[0].ds_len; sc->rl_ldata.rl_jrx_sparemap = map; bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap, BUS_DMASYNC_PREREAD); desc = &sc->rl_ldata.rl_rx_list[idx]; desc->rl_vlanctl = 0; desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[0].ds_addr)); desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[0].ds_addr)); cmdstat = segs[0].ds_len; if (idx == sc->rl_ldata.rl_rx_desc_cnt - 1) cmdstat |= RL_RDESC_CMD_EOR; desc->rl_cmdstat = htole32(cmdstat | RL_RDESC_CMD_OWN); return (0); } #ifdef RE_FIXUP_RX static __inline void re_fixup_rx(struct mbuf *m) { int i; uint16_t *src, *dst; src = mtod(m, uint16_t *); dst = src - (RE_ETHER_ALIGN - ETHER_ALIGN) / sizeof *src; for (i = 0; i < (m->m_len / sizeof(uint16_t) + 1); i++) *dst++ = *src++; m->m_data -= RE_ETHER_ALIGN - ETHER_ALIGN; } #endif static int re_tx_list_init(struct rl_softc *sc) { struct rl_desc *desc; int i; RL_LOCK_ASSERT(sc); bzero(sc->rl_ldata.rl_tx_list, sc->rl_ldata.rl_tx_desc_cnt * sizeof(struct rl_desc)); for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) sc->rl_ldata.rl_tx_desc[i].tx_m = NULL; /* Set EOR. */ desc = &sc->rl_ldata.rl_tx_list[sc->rl_ldata.rl_tx_desc_cnt - 1]; desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOR); bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); sc->rl_ldata.rl_tx_prodidx = 0; sc->rl_ldata.rl_tx_considx = 0; sc->rl_ldata.rl_tx_free = sc->rl_ldata.rl_tx_desc_cnt; return (0); } static int re_rx_list_init(struct rl_softc *sc) { int error, i; bzero(sc->rl_ldata.rl_rx_list, sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc)); for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { sc->rl_ldata.rl_rx_desc[i].rx_m = NULL; if ((error = re_newbuf(sc, i)) != 0) return (error); } /* Flush the RX descriptors */ bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD); sc->rl_ldata.rl_rx_prodidx = 0; sc->rl_head = sc->rl_tail = NULL; sc->rl_int_rx_act = 0; return (0); } static int re_jrx_list_init(struct rl_softc *sc) { int error, i; bzero(sc->rl_ldata.rl_rx_list, sc->rl_ldata.rl_rx_desc_cnt * sizeof(struct rl_desc)); for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { sc->rl_ldata.rl_jrx_desc[i].rx_m = NULL; if ((error = re_jumbo_newbuf(sc, i)) != 0) return (error); } bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, BUS_DMASYNC_PREWRITE | BUS_DMASYNC_PREREAD); sc->rl_ldata.rl_rx_prodidx = 0; sc->rl_head = sc->rl_tail = NULL; sc->rl_int_rx_act = 0; return (0); } /* * RX handler for C+ and 8169. For the gigE chips, we support * the reception of jumbo frames that have been fragmented * across multiple 2K mbuf cluster buffers. */ static int re_rxeof(struct rl_softc *sc) { struct mbuf *m; struct ifnet *ifp; int i, rxerr, total_len; struct rl_desc *cur_rx; u_int32_t rxstat, rxvlan; int jumbo, maxpkt = 16; RL_LOCK_ASSERT(sc); ifp = sc->rl_ifp; if (ifp->if_mtu > RL_MTU && (sc->rl_flags & RL_FLAG_JUMBOV2) != 0) jumbo = 1; else jumbo = 0; /* Invalidate the descriptor memory */ bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); for (i = sc->rl_ldata.rl_rx_prodidx; maxpkt > 0; i = RL_RX_DESC_NXT(sc, i)) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) break; cur_rx = &sc->rl_ldata.rl_rx_list[i]; rxstat = le32toh(cur_rx->rl_cmdstat); if ((rxstat & RL_RDESC_STAT_OWN) != 0) break; total_len = rxstat & sc->rl_rxlenmask; rxvlan = le32toh(cur_rx->rl_vlanctl); if (jumbo != 0) m = sc->rl_ldata.rl_jrx_desc[i].rx_m; else m = sc->rl_ldata.rl_rx_desc[i].rx_m; if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 && (rxstat & (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) != (RL_RDESC_STAT_SOF | RL_RDESC_STAT_EOF)) { /* * RTL8168C or later controllers do not * support multi-fragment packet. */ re_discard_rxbuf(sc, i); continue; } else if ((rxstat & RL_RDESC_STAT_EOF) == 0) { if (re_newbuf(sc, i) != 0) { /* * If this is part of a multi-fragment packet, * discard all the pieces. */ if (sc->rl_head != NULL) { m_freem(sc->rl_head); sc->rl_head = sc->rl_tail = NULL; } re_discard_rxbuf(sc, i); continue; } m->m_len = RE_RX_DESC_BUFLEN; if (sc->rl_head == NULL) sc->rl_head = sc->rl_tail = m; else { m->m_flags &= ~M_PKTHDR; sc->rl_tail->m_next = m; sc->rl_tail = m; } continue; } /* * NOTE: for the 8139C+, the frame length field * is always 12 bits in size, but for the gigE chips, * it is 13 bits (since the max RX frame length is 16K). * Unfortunately, all 32 bits in the status word * were already used, so to make room for the extra * length bit, RealTek took out the 'frame alignment * error' bit and shifted the other status bits * over one slot. The OWN, EOR, FS and LS bits are * still in the same places. We have already extracted * the frame length and checked the OWN bit, so rather * than using an alternate bit mapping, we shift the * status bits one space to the right so we can evaluate * them using the 8169 status as though it was in the * same format as that of the 8139C+. */ if (sc->rl_type == RL_8169) rxstat >>= 1; /* * if total_len > 2^13-1, both _RXERRSUM and _GIANT will be * set, but if CRC is clear, it will still be a valid frame. */ if ((rxstat & RL_RDESC_STAT_RXERRSUM) != 0) { rxerr = 1; if ((sc->rl_flags & RL_FLAG_JUMBOV2) == 0 && total_len > 8191 && (rxstat & RL_RDESC_STAT_ERRS) == RL_RDESC_STAT_GIANT) rxerr = 0; if (rxerr != 0) { ifp->if_ierrors++; /* * If this is part of a multi-fragment packet, * discard all the pieces. */ if (sc->rl_head != NULL) { m_freem(sc->rl_head); sc->rl_head = sc->rl_tail = NULL; } re_discard_rxbuf(sc, i); continue; } } /* * If allocating a replacement mbuf fails, * reload the current one. */ if (jumbo != 0) rxerr = re_jumbo_newbuf(sc, i); else rxerr = re_newbuf(sc, i); if (rxerr != 0) { ifp->if_iqdrops++; if (sc->rl_head != NULL) { m_freem(sc->rl_head); sc->rl_head = sc->rl_tail = NULL; } re_discard_rxbuf(sc, i); continue; } if (sc->rl_head != NULL) { if (jumbo != 0) m->m_len = total_len; else { m->m_len = total_len % RE_RX_DESC_BUFLEN; if (m->m_len == 0) m->m_len = RE_RX_DESC_BUFLEN; } /* * Special case: if there's 4 bytes or less * in this buffer, the mbuf can be discarded: * the last 4 bytes is the CRC, which we don't * care about anyway. */ if (m->m_len <= ETHER_CRC_LEN) { sc->rl_tail->m_len -= (ETHER_CRC_LEN - m->m_len); m_freem(m); } else { m->m_len -= ETHER_CRC_LEN; m->m_flags &= ~M_PKTHDR; sc->rl_tail->m_next = m; } m = sc->rl_head; sc->rl_head = sc->rl_tail = NULL; m->m_pkthdr.len = total_len - ETHER_CRC_LEN; } else m->m_pkthdr.len = m->m_len = (total_len - ETHER_CRC_LEN); #ifdef RE_FIXUP_RX re_fixup_rx(m); #endif ifp->if_ipackets++; m->m_pkthdr.rcvif = ifp; /* Do RX checksumming if enabled */ if (ifp->if_capenable & IFCAP_RXCSUM) { if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) { /* Check IP header checksum */ if (rxstat & RL_RDESC_STAT_PROTOID) m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED; if (!(rxstat & RL_RDESC_STAT_IPSUMBAD)) m->m_pkthdr.csum_flags |= CSUM_IP_VALID; /* Check TCP/UDP checksum */ if ((RL_TCPPKT(rxstat) && !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) || (RL_UDPPKT(rxstat) && !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID|CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } } else { /* * RTL8168C/RTL816CP/RTL8111C/RTL8111CP */ if ((rxstat & RL_RDESC_STAT_PROTOID) && (rxvlan & RL_RDESC_IPV4)) m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED; if (!(rxstat & RL_RDESC_STAT_IPSUMBAD) && (rxvlan & RL_RDESC_IPV4)) m->m_pkthdr.csum_flags |= CSUM_IP_VALID; if (((rxstat & RL_RDESC_STAT_TCP) && !(rxstat & RL_RDESC_STAT_TCPSUMBAD)) || ((rxstat & RL_RDESC_STAT_UDP) && !(rxstat & RL_RDESC_STAT_UDPSUMBAD))) { m->m_pkthdr.csum_flags |= CSUM_DATA_VALID|CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xffff; } } } maxpkt--; if (rxvlan & RL_RDESC_VLANCTL_TAG) { m->m_pkthdr.ether_vtag = bswap16((rxvlan & RL_RDESC_VLANCTL_DATA)); m->m_flags |= M_VLANTAG; } RL_UNLOCK(sc); (*ifp->if_input)(ifp, m); RL_LOCK(sc); } /* Flush the RX DMA ring */ bus_dmamap_sync(sc->rl_ldata.rl_rx_list_tag, sc->rl_ldata.rl_rx_list_map, BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD); sc->rl_ldata.rl_rx_prodidx = i; if (maxpkt) return (EAGAIN); return (0); } static void re_txeof(struct rl_softc *sc) { struct ifnet *ifp; struct rl_txdesc *txd; u_int32_t txstat; int cons; cons = sc->rl_ldata.rl_tx_considx; if (cons == sc->rl_ldata.rl_tx_prodidx) return; ifp = sc->rl_ifp; /* Invalidate the TX descriptor list */ bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); for (; cons != sc->rl_ldata.rl_tx_prodidx; cons = RL_TX_DESC_NXT(sc, cons)) { txstat = le32toh(sc->rl_ldata.rl_tx_list[cons].rl_cmdstat); if (txstat & RL_TDESC_STAT_OWN) break; /* * We only stash mbufs in the last descriptor * in a fragment chain, which also happens to * be the only place where the TX status bits * are valid. */ if (txstat & RL_TDESC_CMD_EOF) { txd = &sc->rl_ldata.rl_tx_desc[cons]; bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap); KASSERT(txd->tx_m != NULL, ("%s: freeing NULL mbufs!", __func__)); m_freem(txd->tx_m); txd->tx_m = NULL; if (txstat & (RL_TDESC_STAT_EXCESSCOL| RL_TDESC_STAT_COLCNT)) ifp->if_collisions++; if (txstat & RL_TDESC_STAT_TXERRSUM) ifp->if_oerrors++; else ifp->if_opackets++; } sc->rl_ldata.rl_tx_free++; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; } sc->rl_ldata.rl_tx_considx = cons; /* No changes made to the TX ring, so no flush needed */ if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) { #ifdef RE_TX_MODERATION /* * If not all descriptors have been reaped yet, reload * the timer so that we will eventually get another * interrupt that will cause us to re-enter this routine. * This is done in case the transmitter has gone idle. */ CSR_WRITE_4(sc, RL_TIMERCNT, 1); #endif } else sc->rl_watchdog_timer = 0; } static void re_tick(void *xsc) { struct rl_softc *sc; struct mii_data *mii; sc = xsc; RL_LOCK_ASSERT(sc); mii = device_get_softc(sc->rl_miibus); mii_tick(mii); if ((sc->rl_flags & RL_FLAG_LINK) == 0) re_miibus_statchg(sc->rl_dev); /* * Reclaim transmitted frames here. Technically it is not * necessary to do here but it ensures periodic reclamation * regardless of Tx completion interrupt which seems to be * lost on PCIe based controllers under certain situations. */ re_txeof(sc); re_watchdog(sc); callout_reset(&sc->rl_stat_callout, hz, re_tick, sc); } #ifdef DEVICE_POLLING static void re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count) { struct rl_softc *sc = ifp->if_softc; RL_LOCK(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) re_poll_locked(ifp, cmd, count); RL_UNLOCK(sc); } static void re_poll_locked(struct ifnet *ifp, enum poll_cmd cmd, int count) { struct rl_softc *sc = ifp->if_softc; RL_LOCK_ASSERT(sc); sc->rxcycles = count; re_rxeof(sc); re_txeof(sc); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) re_start_locked(ifp); if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */ u_int16_t status; status = CSR_READ_2(sc, RL_ISR); if (status == 0xffff) return; if (status) CSR_WRITE_2(sc, RL_ISR, status); if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) && (sc->rl_flags & RL_FLAG_PCIE)) CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START); /* * XXX check behaviour on receiver stalls. */ if (status & RL_ISR_SYSTEM_ERR) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); } } } #endif /* DEVICE_POLLING */ static int re_intr(void *arg) { struct rl_softc *sc; uint16_t status; sc = arg; status = CSR_READ_2(sc, RL_ISR); if (status == 0xFFFF || (status & RL_INTRS_CPLUS) == 0) return (FILTER_STRAY); CSR_WRITE_2(sc, RL_IMR, 0); taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask); return (FILTER_HANDLED); } static void re_int_task(void *arg, int npending) { struct rl_softc *sc; struct ifnet *ifp; u_int16_t status; int rval = 0; sc = arg; ifp = sc->rl_ifp; RL_LOCK(sc); status = CSR_READ_2(sc, RL_ISR); CSR_WRITE_2(sc, RL_ISR, status); if (sc->suspended || (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { RL_UNLOCK(sc); return; } #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) { RL_UNLOCK(sc); return; } #endif if (status & (RL_ISR_RX_OK|RL_ISR_RX_ERR|RL_ISR_FIFO_OFLOW)) rval = re_rxeof(sc); /* * Some chips will ignore a second TX request issued * while an existing transmission is in progress. If * the transmitter goes idle but there are still * packets waiting to be sent, we need to restart the * channel here to flush them out. This only seems to * be required with the PCIe devices. */ if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) && (sc->rl_flags & RL_FLAG_PCIE)) CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START); if (status & ( #ifdef RE_TX_MODERATION RL_ISR_TIMEOUT_EXPIRED| #else RL_ISR_TX_OK| #endif RL_ISR_TX_ERR|RL_ISR_TX_DESC_UNAVAIL)) re_txeof(sc); if (status & RL_ISR_SYSTEM_ERR) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); } if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) re_start_locked(ifp); RL_UNLOCK(sc); if ((CSR_READ_2(sc, RL_ISR) & RL_INTRS_CPLUS) || rval) { taskqueue_enqueue_fast(taskqueue_fast, &sc->rl_inttask); return; } CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS); } static void re_intr_msi(void *xsc) { struct rl_softc *sc; struct ifnet *ifp; uint16_t intrs, status; sc = xsc; RL_LOCK(sc); ifp = sc->rl_ifp; #ifdef DEVICE_POLLING if (ifp->if_capenable & IFCAP_POLLING) { RL_UNLOCK(sc); return; } #endif /* Disable interrupts. */ CSR_WRITE_2(sc, RL_IMR, 0); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { RL_UNLOCK(sc); return; } intrs = RL_INTRS_CPLUS; status = CSR_READ_2(sc, RL_ISR); CSR_WRITE_2(sc, RL_ISR, status); if (sc->rl_int_rx_act > 0) { intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN); status &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN); } if (status & (RL_ISR_TIMEOUT_EXPIRED | RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) { re_rxeof(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if (sc->rl_int_rx_mod != 0 && (status & (RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN)) != 0) { /* Rearm one-shot timer. */ CSR_WRITE_4(sc, RL_TIMERCNT, 1); intrs &= ~(RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN); sc->rl_int_rx_act = 1; } else { intrs |= RL_ISR_RX_OK | RL_ISR_RX_ERR | RL_ISR_FIFO_OFLOW | RL_ISR_RX_OVERRUN; sc->rl_int_rx_act = 0; } } } /* * Some chips will ignore a second TX request issued * while an existing transmission is in progress. If * the transmitter goes idle but there are still * packets waiting to be sent, we need to restart the * channel here to flush them out. This only seems to * be required with the PCIe devices. */ if ((status & (RL_ISR_TX_OK | RL_ISR_TX_DESC_UNAVAIL)) && (sc->rl_flags & RL_FLAG_PCIE)) CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START); if (status & (RL_ISR_TX_OK | RL_ISR_TX_ERR | RL_ISR_TX_DESC_UNAVAIL)) re_txeof(sc); if (status & RL_ISR_SYSTEM_ERR) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); } if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) re_start_locked(ifp); CSR_WRITE_2(sc, RL_IMR, intrs); } RL_UNLOCK(sc); } static int re_encap(struct rl_softc *sc, struct mbuf **m_head) { struct rl_txdesc *txd, *txd_last; bus_dma_segment_t segs[RL_NTXSEGS]; bus_dmamap_t map; struct mbuf *m_new; struct rl_desc *desc; int nsegs, prod; int i, error, ei, si; int padlen; uint32_t cmdstat, csum_flags, vlanctl; RL_LOCK_ASSERT(sc); M_ASSERTPKTHDR((*m_head)); /* * With some of the RealTek chips, using the checksum offload * support in conjunction with the autopadding feature results * in the transmission of corrupt frames. For example, if we * need to send a really small IP fragment that's less than 60 * bytes in size, and IP header checksumming is enabled, the * resulting ethernet frame that appears on the wire will * have garbled payload. To work around this, if TX IP checksum * offload is enabled, we always manually pad short frames out * to the minimum ethernet frame size. */ if ((sc->rl_flags & RL_FLAG_AUTOPAD) == 0 && (*m_head)->m_pkthdr.len < RL_IP4CSUMTX_PADLEN && ((*m_head)->m_pkthdr.csum_flags & CSUM_IP) != 0) { padlen = RL_MIN_FRAMELEN - (*m_head)->m_pkthdr.len; if (M_WRITABLE(*m_head) == 0) { /* Get a writable copy. */ m_new = m_dup(*m_head, M_DONTWAIT); m_freem(*m_head); if (m_new == NULL) { *m_head = NULL; return (ENOBUFS); } *m_head = m_new; } if ((*m_head)->m_next != NULL || M_TRAILINGSPACE(*m_head) < padlen) { m_new = m_defrag(*m_head, M_DONTWAIT); if (m_new == NULL) { m_freem(*m_head); *m_head = NULL; return (ENOBUFS); } } else m_new = *m_head; /* * Manually pad short frames, and zero the pad space * to avoid leaking data. */ bzero(mtod(m_new, char *) + m_new->m_pkthdr.len, padlen); m_new->m_pkthdr.len += padlen; m_new->m_len = m_new->m_pkthdr.len; *m_head = m_new; } prod = sc->rl_ldata.rl_tx_prodidx; txd = &sc->rl_ldata.rl_tx_desc[prod]; error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT); if (error == EFBIG) { m_new = m_collapse(*m_head, M_DONTWAIT, RL_NTXSEGS); if (m_new == NULL) { m_freem(*m_head); *m_head = NULL; return (ENOBUFS); } *m_head = m_new; error = bus_dmamap_load_mbuf_sg(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap, *m_head, segs, &nsegs, BUS_DMA_NOWAIT); if (error != 0) { m_freem(*m_head); *m_head = NULL; return (error); } } else if (error != 0) return (error); if (nsegs == 0) { m_freem(*m_head); *m_head = NULL; return (EIO); } /* Check for number of available descriptors. */ if (sc->rl_ldata.rl_tx_free - nsegs <= 1) { bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap); return (ENOBUFS); } bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap, BUS_DMASYNC_PREWRITE); /* * Set up checksum offload. Note: checksum offload bits must * appear in all descriptors of a multi-descriptor transmit * attempt. This is according to testing done with an 8169 * chip. This is a requirement. */ vlanctl = 0; csum_flags = 0; if (((*m_head)->m_pkthdr.csum_flags & CSUM_TSO) != 0) { if ((sc->rl_flags & RL_FLAG_DESCV2) != 0) { csum_flags |= RL_TDESC_CMD_LGSEND; vlanctl |= ((uint32_t)(*m_head)->m_pkthdr.tso_segsz << RL_TDESC_CMD_MSSVALV2_SHIFT); } else { csum_flags |= RL_TDESC_CMD_LGSEND | ((uint32_t)(*m_head)->m_pkthdr.tso_segsz << RL_TDESC_CMD_MSSVAL_SHIFT); } } else { /* * Unconditionally enable IP checksum if TCP or UDP * checksum is required. Otherwise, TCP/UDP checksum * does't make effects. */ if (((*m_head)->m_pkthdr.csum_flags & RE_CSUM_FEATURES) != 0) { if ((sc->rl_flags & RL_FLAG_DESCV2) == 0) { csum_flags |= RL_TDESC_CMD_IPCSUM; if (((*m_head)->m_pkthdr.csum_flags & CSUM_TCP) != 0) csum_flags |= RL_TDESC_CMD_TCPCSUM; if (((*m_head)->m_pkthdr.csum_flags & CSUM_UDP) != 0) csum_flags |= RL_TDESC_CMD_UDPCSUM; } else { vlanctl |= RL_TDESC_CMD_IPCSUMV2; if (((*m_head)->m_pkthdr.csum_flags & CSUM_TCP) != 0) vlanctl |= RL_TDESC_CMD_TCPCSUMV2; if (((*m_head)->m_pkthdr.csum_flags & CSUM_UDP) != 0) vlanctl |= RL_TDESC_CMD_UDPCSUMV2; } } } /* * Set up hardware VLAN tagging. Note: vlan tag info must * appear in all descriptors of a multi-descriptor * transmission attempt. */ if ((*m_head)->m_flags & M_VLANTAG) vlanctl |= bswap16((*m_head)->m_pkthdr.ether_vtag) | RL_TDESC_VLANCTL_TAG; si = prod; for (i = 0; i < nsegs; i++, prod = RL_TX_DESC_NXT(sc, prod)) { desc = &sc->rl_ldata.rl_tx_list[prod]; desc->rl_vlanctl = htole32(vlanctl); desc->rl_bufaddr_lo = htole32(RL_ADDR_LO(segs[i].ds_addr)); desc->rl_bufaddr_hi = htole32(RL_ADDR_HI(segs[i].ds_addr)); cmdstat = segs[i].ds_len; if (i != 0) cmdstat |= RL_TDESC_CMD_OWN; if (prod == sc->rl_ldata.rl_tx_desc_cnt - 1) cmdstat |= RL_TDESC_CMD_EOR; desc->rl_cmdstat = htole32(cmdstat | csum_flags); sc->rl_ldata.rl_tx_free--; } /* Update producer index. */ sc->rl_ldata.rl_tx_prodidx = prod; /* Set EOF on the last descriptor. */ ei = RL_TX_DESC_PRV(sc, prod); desc = &sc->rl_ldata.rl_tx_list[ei]; desc->rl_cmdstat |= htole32(RL_TDESC_CMD_EOF); desc = &sc->rl_ldata.rl_tx_list[si]; /* Set SOF and transfer ownership of packet to the chip. */ desc->rl_cmdstat |= htole32(RL_TDESC_CMD_OWN | RL_TDESC_CMD_SOF); /* * Insure that the map for this transmission * is placed at the array index of the last descriptor * in this chain. (Swap last and first dmamaps.) */ txd_last = &sc->rl_ldata.rl_tx_desc[ei]; map = txd->tx_dmamap; txd->tx_dmamap = txd_last->tx_dmamap; txd_last->tx_dmamap = map; txd_last->tx_m = *m_head; return (0); } static void re_start(struct ifnet *ifp) { struct rl_softc *sc; sc = ifp->if_softc; RL_LOCK(sc); re_start_locked(ifp); RL_UNLOCK(sc); } /* * Main transmit routine for C+ and gigE NICs. */ static void re_start_locked(struct ifnet *ifp) { struct rl_softc *sc; struct mbuf *m_head; int queued; sc = ifp->if_softc; if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING || (sc->rl_flags & RL_FLAG_LINK) == 0) return; for (queued = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd) && sc->rl_ldata.rl_tx_free > 1;) { IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; if (re_encap(sc, &m_head) != 0) { if (m_head == NULL) break; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); ifp->if_drv_flags |= IFF_DRV_OACTIVE; break; } /* * If there's a BPF listener, bounce a copy of this frame * to him. */ ETHER_BPF_MTAP(ifp, m_head); queued++; } if (queued == 0) { #ifdef RE_TX_MODERATION if (sc->rl_ldata.rl_tx_free != sc->rl_ldata.rl_tx_desc_cnt) CSR_WRITE_4(sc, RL_TIMERCNT, 1); #endif return; } /* Flush the TX descriptors */ bus_dmamap_sync(sc->rl_ldata.rl_tx_list_tag, sc->rl_ldata.rl_tx_list_map, BUS_DMASYNC_PREWRITE|BUS_DMASYNC_PREREAD); CSR_WRITE_1(sc, sc->rl_txstart, RL_TXSTART_START); #ifdef RE_TX_MODERATION /* * Use the countdown timer for interrupt moderation. * 'TX done' interrupts are disabled. Instead, we reset the * countdown timer, which will begin counting until it hits * the value in the TIMERINT register, and then trigger an * interrupt. Each time we write to the TIMERCNT register, * the timer count is reset to 0. */ CSR_WRITE_4(sc, RL_TIMERCNT, 1); #endif /* * Set a timeout in case the chip goes out to lunch. */ sc->rl_watchdog_timer = 5; } static void re_set_jumbo(struct rl_softc *sc, int jumbo) { if (sc->rl_hwrev->rl_rev == RL_HWREV_8168E_VL) { pci_set_max_read_req(sc->rl_dev, 4096); return; } CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG); if (jumbo != 0) { CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) | RL_CFG3_JUMBO_EN0); switch (sc->rl_hwrev->rl_rev) { case RL_HWREV_8168DP: break; case RL_HWREV_8168E: CSR_WRITE_1(sc, sc->rl_cfg4, CSR_READ_1(sc, sc->rl_cfg4) | 0x01); break; default: CSR_WRITE_1(sc, sc->rl_cfg4, CSR_READ_1(sc, sc->rl_cfg4) | RL_CFG4_JUMBO_EN1); } } else { CSR_WRITE_1(sc, sc->rl_cfg3, CSR_READ_1(sc, sc->rl_cfg3) & ~RL_CFG3_JUMBO_EN0); switch (sc->rl_hwrev->rl_rev) { case RL_HWREV_8168DP: break; case RL_HWREV_8168E: CSR_WRITE_1(sc, sc->rl_cfg4, CSR_READ_1(sc, sc->rl_cfg4) & ~0x01); break; default: CSR_WRITE_1(sc, sc->rl_cfg4, CSR_READ_1(sc, sc->rl_cfg4) & ~RL_CFG4_JUMBO_EN1); } } CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); switch (sc->rl_hwrev->rl_rev) { case RL_HWREV_8168DP: pci_set_max_read_req(sc->rl_dev, 4096); break; default: if (jumbo != 0) pci_set_max_read_req(sc->rl_dev, 512); else pci_set_max_read_req(sc->rl_dev, 4096); } } static void re_init(void *xsc) { struct rl_softc *sc = xsc; RL_LOCK(sc); re_init_locked(sc); RL_UNLOCK(sc); } static void re_init_locked(struct rl_softc *sc) { struct ifnet *ifp = sc->rl_ifp; struct mii_data *mii; uint32_t reg; uint16_t cfg; union { uint32_t align_dummy; u_char eaddr[ETHER_ADDR_LEN]; } eaddr; RL_LOCK_ASSERT(sc); mii = device_get_softc(sc->rl_miibus); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) return; /* * Cancel pending I/O and free all RX/TX buffers. */ re_stop(sc); /* Put controller into known state. */ re_reset(sc); /* * For C+ mode, initialize the RX descriptors and mbufs. */ if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) { if (ifp->if_mtu > RL_MTU) { if (re_jrx_list_init(sc) != 0) { device_printf(sc->rl_dev, "no memory for jumbo RX buffers\n"); re_stop(sc); return; } /* Disable checksum offloading for jumbo frames. */ ifp->if_capenable &= ~(IFCAP_HWCSUM | IFCAP_TSO4); ifp->if_hwassist &= ~(RE_CSUM_FEATURES | CSUM_TSO); } else { if (re_rx_list_init(sc) != 0) { device_printf(sc->rl_dev, "no memory for RX buffers\n"); re_stop(sc); return; } } re_set_jumbo(sc, ifp->if_mtu > RL_MTU); } else { if (re_rx_list_init(sc) != 0) { device_printf(sc->rl_dev, "no memory for RX buffers\n"); re_stop(sc); return; } if ((sc->rl_flags & RL_FLAG_PCIE) != 0 && pci_get_device(sc->rl_dev) != RT_DEVICEID_8101E) { if (ifp->if_mtu > RL_MTU) pci_set_max_read_req(sc->rl_dev, 512); else pci_set_max_read_req(sc->rl_dev, 4096); } } re_tx_list_init(sc); /* * Enable C+ RX and TX mode, as well as VLAN stripping and * RX checksum offload. We must configure the C+ register * before all others. */ cfg = RL_CPLUSCMD_PCI_MRW; if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) cfg |= RL_CPLUSCMD_RXCSUM_ENB; if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) cfg |= RL_CPLUSCMD_VLANSTRIP; if ((sc->rl_flags & RL_FLAG_MACSTAT) != 0) { cfg |= RL_CPLUSCMD_MACSTAT_DIS; /* XXX magic. */ cfg |= 0x0001; } else cfg |= RL_CPLUSCMD_RXENB | RL_CPLUSCMD_TXENB; CSR_WRITE_2(sc, RL_CPLUS_CMD, cfg); if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SC || sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) { reg = 0x000fff00; if ((CSR_READ_1(sc, sc->rl_cfg2) & RL_CFG2_PCI66MHZ) != 0) reg |= 0x000000ff; if (sc->rl_hwrev->rl_rev == RL_HWREV_8169_8110SCE) reg |= 0x00f00000; CSR_WRITE_4(sc, 0x7c, reg); /* Disable interrupt mitigation. */ CSR_WRITE_2(sc, 0xe2, 0); } /* * Disable TSO if interface MTU size is greater than MSS * allowed in controller. */ if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) { ifp->if_capenable &= ~IFCAP_TSO4; ifp->if_hwassist &= ~CSUM_TSO; } /* * Init our MAC address. Even though the chipset * documentation doesn't mention it, we need to enter "Config * register write enable" mode to modify the ID registers. */ /* Copy MAC address on stack to align. */ bcopy(IF_LLADDR(ifp), eaddr.eaddr, ETHER_ADDR_LEN); CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_WRITECFG); CSR_WRITE_4(sc, RL_IDR0, htole32(*(u_int32_t *)(&eaddr.eaddr[0]))); CSR_WRITE_4(sc, RL_IDR4, htole32(*(u_int32_t *)(&eaddr.eaddr[4]))); CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); /* * Load the addresses of the RX and TX lists into the chip. */ CSR_WRITE_4(sc, RL_RXLIST_ADDR_HI, RL_ADDR_HI(sc->rl_ldata.rl_rx_list_addr)); CSR_WRITE_4(sc, RL_RXLIST_ADDR_LO, RL_ADDR_LO(sc->rl_ldata.rl_rx_list_addr)); CSR_WRITE_4(sc, RL_TXLIST_ADDR_HI, RL_ADDR_HI(sc->rl_ldata.rl_tx_list_addr)); CSR_WRITE_4(sc, RL_TXLIST_ADDR_LO, RL_ADDR_LO(sc->rl_ldata.rl_tx_list_addr)); /* * Enable transmit and receive. */ CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB); /* * Set the initial TX configuration. */ if (sc->rl_testmode) { if (sc->rl_type == RL_8169) CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG|RL_LOOPTEST_ON); else CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG|RL_LOOPTEST_ON_CPLUS); } else CSR_WRITE_4(sc, RL_TXCFG, RL_TXCFG_CONFIG); CSR_WRITE_1(sc, RL_EARLY_TX_THRESH, 16); /* * Set the initial RX configuration. */ re_set_rxmode(sc); /* Configure interrupt moderation. */ if (sc->rl_type == RL_8169) { /* Magic from vendor. */ CSR_WRITE_2(sc, RL_INTRMOD, 0x5100); } #ifdef DEVICE_POLLING /* * Disable interrupts if we are polling. */ if (ifp->if_capenable & IFCAP_POLLING) CSR_WRITE_2(sc, RL_IMR, 0); else /* otherwise ... */ #endif /* * Enable interrupts. */ if (sc->rl_testmode) CSR_WRITE_2(sc, RL_IMR, 0); else CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS); CSR_WRITE_2(sc, RL_ISR, RL_INTRS_CPLUS); /* Set initial TX threshold */ sc->rl_txthresh = RL_TX_THRESH_INIT; /* Start RX/TX process. */ CSR_WRITE_4(sc, RL_MISSEDPKT, 0); #ifdef notdef /* Enable receiver and transmitter. */ CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_TX_ENB|RL_CMD_RX_ENB); #endif /* * Initialize the timer interrupt register so that * a timer interrupt will be generated once the timer * reaches a certain number of ticks. The timer is * reloaded on each transmit. */ #ifdef RE_TX_MODERATION /* * Use timer interrupt register to moderate TX interrupt * moderation, which dramatically improves TX frame rate. */ if (sc->rl_type == RL_8169) CSR_WRITE_4(sc, RL_TIMERINT_8169, 0x800); else CSR_WRITE_4(sc, RL_TIMERINT, 0x400); #else /* * Use timer interrupt register to moderate RX interrupt * moderation. */ if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) != 0 && intr_filter == 0) { if (sc->rl_type == RL_8169) CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(sc->rl_int_rx_mod)); } else { if (sc->rl_type == RL_8169) CSR_WRITE_4(sc, RL_TIMERINT_8169, RL_USECS(0)); } #endif /* * For 8169 gigE NICs, set the max allowed RX packet * size so we can receive jumbo frames. */ if (sc->rl_type == RL_8169) { if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) { /* * For controllers that use new jumbo frame scheme, * set maximum size of jumbo frame depedning on * controller revisions. */ if (ifp->if_mtu > RL_MTU) CSR_WRITE_2(sc, RL_MAXRXPKTLEN, sc->rl_hwrev->rl_max_mtu + ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN + ETHER_CRC_LEN); else CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN); } else if ((sc->rl_flags & RL_FLAG_PCIE) != 0 && sc->rl_hwrev->rl_max_mtu == RL_MTU) { /* RTL810x has no jumbo frame support. */ CSR_WRITE_2(sc, RL_MAXRXPKTLEN, RE_RX_DESC_BUFLEN); } else CSR_WRITE_2(sc, RL_MAXRXPKTLEN, 16383); } if (sc->rl_testmode) return; CSR_WRITE_1(sc, sc->rl_cfg1, CSR_READ_1(sc, sc->rl_cfg1) | RL_CFG1_DRVLOAD); ifp->if_drv_flags |= IFF_DRV_RUNNING; ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; sc->rl_flags &= ~RL_FLAG_LINK; mii_mediachg(mii); sc->rl_watchdog_timer = 0; callout_reset(&sc->rl_stat_callout, hz, re_tick, sc); } /* * Set media options. */ static int re_ifmedia_upd(struct ifnet *ifp) { struct rl_softc *sc; struct mii_data *mii; int error; sc = ifp->if_softc; mii = device_get_softc(sc->rl_miibus); RL_LOCK(sc); error = mii_mediachg(mii); RL_UNLOCK(sc); return (error); } /* * Report current media status. */ static void re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { struct rl_softc *sc; struct mii_data *mii; sc = ifp->if_softc; mii = device_get_softc(sc->rl_miibus); RL_LOCK(sc); mii_pollstat(mii); ifmr->ifm_active = mii->mii_media_active; ifmr->ifm_status = mii->mii_media_status; RL_UNLOCK(sc); } static int re_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct rl_softc *sc = ifp->if_softc; struct ifreq *ifr = (struct ifreq *) data; struct mii_data *mii; uint32_t rev; int error = 0; switch (command) { case SIOCSIFMTU: if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > sc->rl_hwrev->rl_max_mtu) { error = EINVAL; break; } RL_LOCK(sc); if (ifp->if_mtu != ifr->ifr_mtu) { ifp->if_mtu = ifr->ifr_mtu; if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 && (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); } if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) { ifp->if_capenable &= ~(IFCAP_TSO4 | IFCAP_VLAN_HWTSO); ifp->if_hwassist &= ~CSUM_TSO; } VLAN_CAPABILITIES(ifp); } RL_UNLOCK(sc); break; case SIOCSIFFLAGS: RL_LOCK(sc); if ((ifp->if_flags & IFF_UP) != 0) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) { if (((ifp->if_flags ^ sc->rl_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) re_set_rxmode(sc); } else re_init_locked(sc); } else { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) re_stop(sc); } sc->rl_if_flags = ifp->if_flags; RL_UNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: RL_LOCK(sc); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) re_set_rxmode(sc); RL_UNLOCK(sc); break; case SIOCGIFMEDIA: case SIOCSIFMEDIA: mii = device_get_softc(sc->rl_miibus); error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, command); break; case SIOCSIFCAP: { int mask, reinit; mask = ifr->ifr_reqcap ^ ifp->if_capenable; reinit = 0; #ifdef DEVICE_POLLING if (mask & IFCAP_POLLING) { if (ifr->ifr_reqcap & IFCAP_POLLING) { error = ether_poll_register(re_poll, ifp); if (error) return (error); RL_LOCK(sc); /* Disable interrupts */ CSR_WRITE_2(sc, RL_IMR, 0x0000); ifp->if_capenable |= IFCAP_POLLING; RL_UNLOCK(sc); } else { error = ether_poll_deregister(ifp); /* Enable interrupts. */ RL_LOCK(sc); CSR_WRITE_2(sc, RL_IMR, RL_INTRS_CPLUS); ifp->if_capenable &= ~IFCAP_POLLING; RL_UNLOCK(sc); } } #endif /* DEVICE_POLLING */ RL_LOCK(sc); if ((mask & IFCAP_TXCSUM) != 0 && (ifp->if_capabilities & IFCAP_TXCSUM) != 0) { ifp->if_capenable ^= IFCAP_TXCSUM; if ((ifp->if_capenable & IFCAP_TXCSUM) != 0) { rev = sc->rl_hwrev->rl_rev; if (rev == RL_HWREV_8168C || rev == RL_HWREV_8168C_SPIN2) ifp->if_hwassist |= CSUM_TCP | CSUM_UDP; else ifp->if_hwassist |= RE_CSUM_FEATURES; } else ifp->if_hwassist &= ~RE_CSUM_FEATURES; reinit = 1; } if ((mask & IFCAP_RXCSUM) != 0 && (ifp->if_capabilities & IFCAP_RXCSUM) != 0) { ifp->if_capenable ^= IFCAP_RXCSUM; reinit = 1; } if ((mask & IFCAP_TSO4) != 0 && (ifp->if_capabilities & IFCAP_TSO4) != 0) { ifp->if_capenable ^= IFCAP_TSO4; if ((IFCAP_TSO4 & ifp->if_capenable) != 0) ifp->if_hwassist |= CSUM_TSO; else ifp->if_hwassist &= ~CSUM_TSO; if (ifp->if_mtu > RL_TSO_MTU && (ifp->if_capenable & IFCAP_TSO4) != 0) { ifp->if_capenable &= ~IFCAP_TSO4; ifp->if_hwassist &= ~CSUM_TSO; } } if ((mask & IFCAP_VLAN_HWTSO) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if ((mask & IFCAP_VLAN_HWTAGGING) != 0 && (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) { ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; /* TSO over VLAN requires VLAN hardware tagging. */ if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0) ifp->if_capenable &= ~IFCAP_VLAN_HWTSO; reinit = 1; } if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0 && (mask & (IFCAP_HWCSUM | IFCAP_TSO4 | IFCAP_VLAN_HWTSO)) != 0) reinit = 1; if ((mask & IFCAP_WOL) != 0 && (ifp->if_capabilities & IFCAP_WOL) != 0) { if ((mask & IFCAP_WOL_UCAST) != 0) ifp->if_capenable ^= IFCAP_WOL_UCAST; if ((mask & IFCAP_WOL_MCAST) != 0) ifp->if_capenable ^= IFCAP_WOL_MCAST; if ((mask & IFCAP_WOL_MAGIC) != 0) ifp->if_capenable ^= IFCAP_WOL_MAGIC; } if (reinit && ifp->if_drv_flags & IFF_DRV_RUNNING) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); } RL_UNLOCK(sc); VLAN_CAPABILITIES(ifp); } break; default: error = ether_ioctl(ifp, command, data); break; } return (error); } static void re_watchdog(struct rl_softc *sc) { struct ifnet *ifp; RL_LOCK_ASSERT(sc); if (sc->rl_watchdog_timer == 0 || --sc->rl_watchdog_timer != 0) return; ifp = sc->rl_ifp; re_txeof(sc); if (sc->rl_ldata.rl_tx_free == sc->rl_ldata.rl_tx_desc_cnt) { if_printf(ifp, "watchdog timeout (missed Tx interrupts) " "-- recovering\n"); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) re_start_locked(ifp); return; } if_printf(ifp, "watchdog timeout\n"); ifp->if_oerrors++; re_rxeof(sc); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; re_init_locked(sc); if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) re_start_locked(ifp); } /* * Stop the adapter and free any mbufs allocated to the * RX and TX lists. */ static void re_stop(struct rl_softc *sc) { int i; struct ifnet *ifp; struct rl_txdesc *txd; struct rl_rxdesc *rxd; RL_LOCK_ASSERT(sc); ifp = sc->rl_ifp; sc->rl_watchdog_timer = 0; callout_stop(&sc->rl_stat_callout); ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); /* * Disable accepting frames to put RX MAC into idle state. * Otherwise it's possible to get frames while stop command * execution is in progress and controller can DMA the frame * to already freed RX buffer during that period. */ CSR_WRITE_4(sc, RL_RXCFG, CSR_READ_4(sc, RL_RXCFG) & ~(RL_RXCFG_RX_ALLPHYS | RL_RXCFG_RX_INDIV | RL_RXCFG_RX_MULTI | RL_RXCFG_RX_BROAD)); if ((sc->rl_flags & RL_FLAG_WAIT_TXPOLL) != 0) { for (i = RL_TIMEOUT; i > 0; i--) { if ((CSR_READ_1(sc, sc->rl_txstart) & RL_TXSTART_START) == 0) break; DELAY(20); } if (i == 0) device_printf(sc->rl_dev, "stopping TX poll timed out!\n"); CSR_WRITE_1(sc, RL_COMMAND, 0x00); } else if ((sc->rl_flags & RL_FLAG_CMDSTOP) != 0) { CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_STOPREQ | RL_CMD_TX_ENB | RL_CMD_RX_ENB); if ((sc->rl_flags & RL_FLAG_CMDSTOP_WAIT_TXQ) != 0) { for (i = RL_TIMEOUT; i > 0; i--) { if ((CSR_READ_4(sc, RL_TXCFG) & RL_TXCFG_QUEUE_EMPTY) != 0) break; DELAY(100); } if (i == 0) device_printf(sc->rl_dev, "stopping TXQ timed out!\n"); } } else CSR_WRITE_1(sc, RL_COMMAND, 0x00); DELAY(1000); CSR_WRITE_2(sc, RL_IMR, 0x0000); CSR_WRITE_2(sc, RL_ISR, 0xFFFF); if (sc->rl_head != NULL) { m_freem(sc->rl_head); sc->rl_head = sc->rl_tail = NULL; } /* Free the TX list buffers. */ for (i = 0; i < sc->rl_ldata.rl_tx_desc_cnt; i++) { txd = &sc->rl_ldata.rl_tx_desc[i]; if (txd->tx_m != NULL) { bus_dmamap_sync(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(sc->rl_ldata.rl_tx_mtag, txd->tx_dmamap); m_freem(txd->tx_m); txd->tx_m = NULL; } } /* Free the RX list buffers. */ for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { rxd = &sc->rl_ldata.rl_rx_desc[i]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->rl_ldata.rl_rx_mtag, rxd->rx_dmamap); m_freem(rxd->rx_m); rxd->rx_m = NULL; } } if ((sc->rl_flags & RL_FLAG_JUMBOV2) != 0) { for (i = 0; i < sc->rl_ldata.rl_rx_desc_cnt; i++) { rxd = &sc->rl_ldata.rl_jrx_desc[i]; if (rxd->rx_m != NULL) { bus_dmamap_sync(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(sc->rl_ldata.rl_jrx_mtag, rxd->rx_dmamap); m_freem(rxd->rx_m); rxd->rx_m = NULL; } } } } /* * Device suspend routine. Stop the interface and save some PCI * settings in case the BIOS doesn't restore them properly on * resume. */ static int re_suspend(device_t dev) { struct rl_softc *sc; sc = device_get_softc(dev); RL_LOCK(sc); re_stop(sc); re_setwol(sc); sc->suspended = 1; RL_UNLOCK(sc); return (0); } /* * Device resume routine. Restore some PCI settings in case the BIOS * doesn't, re-enable busmastering, and restart the interface if * appropriate. */ static int re_resume(device_t dev) { struct rl_softc *sc; struct ifnet *ifp; sc = device_get_softc(dev); RL_LOCK(sc); ifp = sc->rl_ifp; /* Take controller out of sleep mode. */ if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) { if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80) CSR_WRITE_1(sc, RL_GPIO, CSR_READ_1(sc, RL_GPIO) | 0x01); } /* * Clear WOL matching such that normal Rx filtering * wouldn't interfere with WOL patterns. */ re_clrwol(sc); /* reinitialize interface if necessary */ if (ifp->if_flags & IFF_UP) re_init_locked(sc); sc->suspended = 0; RL_UNLOCK(sc); return (0); } /* * Stop all chip I/O so that the kernel's probe routines don't * get confused by errant DMAs when rebooting. */ static int re_shutdown(device_t dev) { struct rl_softc *sc; sc = device_get_softc(dev); RL_LOCK(sc); re_stop(sc); /* * Mark interface as down since otherwise we will panic if * interrupt comes in later on, which can happen in some * cases. */ sc->rl_ifp->if_flags &= ~IFF_UP; re_setwol(sc); RL_UNLOCK(sc); return (0); } static void re_set_linkspeed(struct rl_softc *sc) { struct mii_softc *miisc; struct mii_data *mii; int aneg, i, phyno; RL_LOCK_ASSERT(sc); mii = device_get_softc(sc->rl_miibus); mii_pollstat(mii); aneg = 0; if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch IFM_SUBTYPE(mii->mii_media_active) { case IFM_10_T: case IFM_100_TX: return; case IFM_1000_T: aneg++; break; default: break; } } miisc = LIST_FIRST(&mii->mii_phys); phyno = miisc->mii_phy; LIST_FOREACH(miisc, &mii->mii_phys, mii_list) mii_phy_reset(miisc); re_miibus_writereg(sc->rl_dev, phyno, MII_100T2CR, 0); re_miibus_writereg(sc->rl_dev, phyno, MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA); re_miibus_writereg(sc->rl_dev, phyno, MII_BMCR, BMCR_AUTOEN | BMCR_STARTNEG); DELAY(1000); if (aneg != 0) { /* * Poll link state until re(4) get a 10/100Mbps link. */ for (i = 0; i < MII_ANEGTICKS_GIGE; i++) { mii_pollstat(mii); if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) == (IFM_ACTIVE | IFM_AVALID)) { switch (IFM_SUBTYPE(mii->mii_media_active)) { case IFM_10_T: case IFM_100_TX: return; default: break; } } RL_UNLOCK(sc); pause("relnk", hz); RL_LOCK(sc); } if (i == MII_ANEGTICKS_GIGE) device_printf(sc->rl_dev, "establishing a link failed, WOL may not work!"); } /* * No link, force MAC to have 100Mbps, full-duplex link. * MAC does not require reprogramming on resolved speed/duplex, * so this is just for completeness. */ mii->mii_media_status = IFM_AVALID | IFM_ACTIVE; mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX; } static void re_setwol(struct rl_softc *sc) { struct ifnet *ifp; int pmc; uint16_t pmstat; uint8_t v; RL_LOCK_ASSERT(sc); if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0) return; ifp = sc->rl_ifp; /* Put controller into sleep mode. */ if ((sc->rl_flags & RL_FLAG_MACSLEEP) != 0) { if ((CSR_READ_1(sc, RL_MACDBG) & 0x80) == 0x80) CSR_WRITE_1(sc, RL_GPIO, CSR_READ_1(sc, RL_GPIO) & ~0x01); } if ((ifp->if_capenable & IFCAP_WOL) != 0) { re_set_rxmode(sc); if ((sc->rl_flags & RL_FLAG_WOL_MANLINK) != 0) re_set_linkspeed(sc); if ((sc->rl_flags & RL_FLAG_WOLRXENB) != 0) CSR_WRITE_1(sc, RL_COMMAND, RL_CMD_RX_ENB); } /* Enable config register write. */ CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE); /* Enable PME. */ v = CSR_READ_1(sc, sc->rl_cfg1); v &= ~RL_CFG1_PME; if ((ifp->if_capenable & IFCAP_WOL) != 0) v |= RL_CFG1_PME; CSR_WRITE_1(sc, sc->rl_cfg1, v); v = CSR_READ_1(sc, sc->rl_cfg3); v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC); if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0) v |= RL_CFG3_WOL_MAGIC; CSR_WRITE_1(sc, sc->rl_cfg3, v); v = CSR_READ_1(sc, sc->rl_cfg5); v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST | RL_CFG5_WOL_LANWAKE); if ((ifp->if_capenable & IFCAP_WOL_UCAST) != 0) v |= RL_CFG5_WOL_UCAST; if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0) v |= RL_CFG5_WOL_MCAST | RL_CFG5_WOL_BCAST; if ((ifp->if_capenable & IFCAP_WOL) != 0) v |= RL_CFG5_WOL_LANWAKE; CSR_WRITE_1(sc, sc->rl_cfg5, v); /* Config register write done. */ CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); if ((ifp->if_capenable & IFCAP_WOL) == 0 && (sc->rl_flags & RL_FLAG_PHYWAKE_PM) != 0) CSR_WRITE_1(sc, RL_PMCH, CSR_READ_1(sc, RL_PMCH) & ~0x80); /* * It seems that hardware resets its link speed to 100Mbps in * power down mode so switching to 100Mbps in driver is not * needed. */ /* Request PME if WOL is requested. */ pmstat = pci_read_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, 2); pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE); if ((ifp->if_capenable & IFCAP_WOL) != 0) pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE; pci_write_config(sc->rl_dev, pmc + PCIR_POWER_STATUS, pmstat, 2); } static void re_clrwol(struct rl_softc *sc) { int pmc; uint8_t v; RL_LOCK_ASSERT(sc); if (pci_find_extcap(sc->rl_dev, PCIY_PMG, &pmc) != 0) return; /* Enable config register write. */ CSR_WRITE_1(sc, RL_EECMD, RL_EE_MODE); v = CSR_READ_1(sc, sc->rl_cfg3); v &= ~(RL_CFG3_WOL_LINK | RL_CFG3_WOL_MAGIC); CSR_WRITE_1(sc, sc->rl_cfg3, v); /* Config register write done. */ CSR_WRITE_1(sc, RL_EECMD, RL_EEMODE_OFF); v = CSR_READ_1(sc, sc->rl_cfg5); v &= ~(RL_CFG5_WOL_BCAST | RL_CFG5_WOL_MCAST | RL_CFG5_WOL_UCAST); v &= ~RL_CFG5_WOL_LANWAKE; CSR_WRITE_1(sc, sc->rl_cfg5, v); } static void re_add_sysctls(struct rl_softc *sc) { struct sysctl_ctx_list *ctx; struct sysctl_oid_list *children; int error; ctx = device_get_sysctl_ctx(sc->rl_dev); children = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->rl_dev)); SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "stats", CTLTYPE_INT | CTLFLAG_RW, sc, 0, re_sysctl_stats, "I", "Statistics Information"); if ((sc->rl_flags & (RL_FLAG_MSI | RL_FLAG_MSIX)) == 0) return; SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "int_rx_mod", CTLTYPE_INT | CTLFLAG_RW, &sc->rl_int_rx_mod, 0, sysctl_hw_re_int_mod, "I", "re RX interrupt moderation"); /* Pull in device tunables. */ sc->rl_int_rx_mod = RL_TIMER_DEFAULT; error = resource_int_value(device_get_name(sc->rl_dev), device_get_unit(sc->rl_dev), "int_rx_mod", &sc->rl_int_rx_mod); if (error == 0) { if (sc->rl_int_rx_mod < RL_TIMER_MIN || sc->rl_int_rx_mod > RL_TIMER_MAX) { device_printf(sc->rl_dev, "int_rx_mod value out of " "range; using default: %d\n", RL_TIMER_DEFAULT); sc->rl_int_rx_mod = RL_TIMER_DEFAULT; } } } static int re_sysctl_stats(SYSCTL_HANDLER_ARGS) { struct rl_softc *sc; struct rl_stats *stats; int error, i, result; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || req->newptr == NULL) return (error); if (result == 1) { sc = (struct rl_softc *)arg1; RL_LOCK(sc); if ((sc->rl_ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { RL_UNLOCK(sc); goto done; } bus_dmamap_sync(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap, BUS_DMASYNC_PREREAD); CSR_WRITE_4(sc, RL_DUMPSTATS_HI, RL_ADDR_HI(sc->rl_ldata.rl_stats_addr)); CSR_WRITE_4(sc, RL_DUMPSTATS_LO, RL_ADDR_LO(sc->rl_ldata.rl_stats_addr)); CSR_WRITE_4(sc, RL_DUMPSTATS_LO, RL_ADDR_LO(sc->rl_ldata.rl_stats_addr | RL_DUMPSTATS_START)); for (i = RL_TIMEOUT; i > 0; i--) { if ((CSR_READ_4(sc, RL_DUMPSTATS_LO) & RL_DUMPSTATS_START) == 0) break; DELAY(1000); } bus_dmamap_sync(sc->rl_ldata.rl_stag, sc->rl_ldata.rl_smap, BUS_DMASYNC_POSTREAD); RL_UNLOCK(sc); if (i == 0) { device_printf(sc->rl_dev, "DUMP statistics request timedout\n"); return (ETIMEDOUT); } done: stats = sc->rl_ldata.rl_stats; printf("%s statistics:\n", device_get_nameunit(sc->rl_dev)); printf("Tx frames : %ju\n", (uintmax_t)le64toh(stats->rl_tx_pkts)); printf("Rx frames : %ju\n", (uintmax_t)le64toh(stats->rl_rx_pkts)); printf("Tx errors : %ju\n", (uintmax_t)le64toh(stats->rl_tx_errs)); printf("Rx errors : %u\n", le32toh(stats->rl_rx_errs)); printf("Rx missed frames : %u\n", (uint32_t)le16toh(stats->rl_missed_pkts)); printf("Rx frame alignment errs : %u\n", (uint32_t)le16toh(stats->rl_rx_framealign_errs)); printf("Tx single collisions : %u\n", le32toh(stats->rl_tx_onecoll)); printf("Tx multiple collisions : %u\n", le32toh(stats->rl_tx_multicolls)); printf("Rx unicast frames : %ju\n", (uintmax_t)le64toh(stats->rl_rx_ucasts)); printf("Rx broadcast frames : %ju\n", (uintmax_t)le64toh(stats->rl_rx_bcasts)); printf("Rx multicast frames : %u\n", le32toh(stats->rl_rx_mcasts)); printf("Tx aborts : %u\n", (uint32_t)le16toh(stats->rl_tx_aborts)); printf("Tx underruns : %u\n", (uint32_t)le16toh(stats->rl_rx_underruns)); } return (error); } static int sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high) { int error, value; if (arg1 == NULL) return (EINVAL); value = *(int *)arg1; error = sysctl_handle_int(oidp, &value, 0, req); if (error || req->newptr == NULL) return (error); if (value < low || value > high) return (EINVAL); *(int *)arg1 = value; return (0); } static int sysctl_hw_re_int_mod(SYSCTL_HANDLER_ARGS) { return (sysctl_int_range(oidp, arg1, arg2, req, RL_TIMER_MIN, RL_TIMER_MAX)); } Index: stable/7/sys =================================================================== --- stable/7/sys (revision 242221) +++ stable/7/sys (revision 242222) Property changes on: stable/7/sys ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/sys:r240680 Index: stable/7/usr.sbin/pciconf/cap.c =================================================================== --- stable/7/usr.sbin/pciconf/cap.c (revision 242221) +++ stable/7/usr.sbin/pciconf/cap.c (revision 242222) @@ -1,632 +1,632 @@ /*- * Copyright (c) 2007 Yahoo!, Inc. * All rights reserved. * Written by: John Baldwin * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef lint static const char rcsid[] = "$FreeBSD$"; #endif /* not lint */ #include #include #include #include #include #include #include #include "pciconf.h" static void list_ecaps(int fd, struct pci_conf *p); static void cap_power(int fd, struct pci_conf *p, uint8_t ptr) { uint16_t cap, status; cap = read_config(fd, &p->pc_sel, ptr + PCIR_POWER_CAP, 2); status = read_config(fd, &p->pc_sel, ptr + PCIR_POWER_STATUS, 2); printf("powerspec %d supports D0%s%s D3 current D%d", cap & PCIM_PCAP_SPEC, cap & PCIM_PCAP_D1SUPP ? " D1" : "", cap & PCIM_PCAP_D2SUPP ? " D2" : "", status & PCIM_PSTAT_DMASK); } static void cap_agp(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t status, command; status = read_config(fd, &p->pc_sel, ptr + AGP_STATUS, 4); command = read_config(fd, &p->pc_sel, ptr + AGP_CAPID, 4); printf("AGP "); if (AGP_MODE_GET_MODE_3(status)) { printf("v3 "); if (AGP_MODE_GET_RATE(status) & AGP_MODE_V3_RATE_8x) printf("8x "); if (AGP_MODE_GET_RATE(status) & AGP_MODE_V3_RATE_4x) printf("4x "); } else { if (AGP_MODE_GET_RATE(status) & AGP_MODE_V2_RATE_4x) printf("4x "); if (AGP_MODE_GET_RATE(status) & AGP_MODE_V2_RATE_2x) printf("2x "); if (AGP_MODE_GET_RATE(status) & AGP_MODE_V2_RATE_1x) printf("1x "); } if (AGP_MODE_GET_SBA(status)) printf("SBA "); if (AGP_MODE_GET_AGP(command)) { printf("enabled at "); if (AGP_MODE_GET_MODE_3(command)) { printf("v3 "); switch (AGP_MODE_GET_RATE(command)) { case AGP_MODE_V3_RATE_8x: printf("8x "); break; case AGP_MODE_V3_RATE_4x: printf("4x "); break; } } else switch (AGP_MODE_GET_RATE(command)) { case AGP_MODE_V2_RATE_4x: printf("4x "); break; case AGP_MODE_V2_RATE_2x: printf("2x "); break; case AGP_MODE_V2_RATE_1x: printf("1x "); break; } if (AGP_MODE_GET_SBA(command)) printf("SBA "); } else printf("disabled"); } static void cap_vpd(int fd, struct pci_conf *p, uint8_t ptr) { printf("VPD"); } static void cap_msi(int fd, struct pci_conf *p, uint8_t ptr) { uint16_t ctrl; int msgnum; ctrl = read_config(fd, &p->pc_sel, ptr + PCIR_MSI_CTRL, 2); msgnum = 1 << ((ctrl & PCIM_MSICTRL_MMC_MASK) >> 1); printf("MSI supports %d message%s%s%s ", msgnum, (msgnum == 1) ? "" : "s", (ctrl & PCIM_MSICTRL_64BIT) ? ", 64 bit" : "", (ctrl & PCIM_MSICTRL_VECTOR) ? ", vector masks" : ""); if (ctrl & PCIM_MSICTRL_MSI_ENABLE) { msgnum = 1 << ((ctrl & PCIM_MSICTRL_MME_MASK) >> 4); printf("enabled with %d message%s", msgnum, (msgnum == 1) ? "" : "s"); } } static void cap_pcix(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t status; int comma, max_splits, max_burst_read; status = read_config(fd, &p->pc_sel, ptr + PCIXR_STATUS, 4); printf("PCI-X "); if (status & PCIXM_STATUS_64BIT) printf("64-bit "); if ((p->pc_hdr & PCIM_HDRTYPE) == 1) printf("bridge "); if ((p->pc_hdr & PCIM_HDRTYPE) != 1 || (status & (PCIXM_STATUS_133CAP | PCIXM_STATUS_266CAP | PCIXM_STATUS_533CAP)) != 0) printf("supports"); comma = 0; if (status & PCIXM_STATUS_133CAP) { printf("%s 133MHz", comma ? "," : ""); comma = 1; } if (status & PCIXM_STATUS_266CAP) { printf("%s 266MHz", comma ? "," : ""); comma = 1; } if (status & PCIXM_STATUS_533CAP) { printf("%s 533MHz", comma ? "," : ""); comma = 1; } if ((p->pc_hdr & PCIM_HDRTYPE) == 1) return; switch (status & PCIXM_STATUS_MAX_READ) { case PCIXM_STATUS_MAX_READ_512: max_burst_read = 512; break; case PCIXM_STATUS_MAX_READ_1024: max_burst_read = 1024; break; case PCIXM_STATUS_MAX_READ_2048: max_burst_read = 2048; break; case PCIXM_STATUS_MAX_READ_4096: max_burst_read = 4096; break; } switch (status & PCIXM_STATUS_MAX_SPLITS) { case PCIXM_STATUS_MAX_SPLITS_1: max_splits = 1; break; case PCIXM_STATUS_MAX_SPLITS_2: max_splits = 2; break; case PCIXM_STATUS_MAX_SPLITS_3: max_splits = 3; break; case PCIXM_STATUS_MAX_SPLITS_4: max_splits = 4; break; case PCIXM_STATUS_MAX_SPLITS_8: max_splits = 8; break; case PCIXM_STATUS_MAX_SPLITS_12: max_splits = 12; break; case PCIXM_STATUS_MAX_SPLITS_16: max_splits = 16; break; case PCIXM_STATUS_MAX_SPLITS_32: max_splits = 32; break; } printf("%s %d burst read, %d split transaction%s", comma ? "," : "", max_burst_read, max_splits, max_splits == 1 ? "" : "s"); } static void cap_ht(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t reg; uint16_t command; command = read_config(fd, &p->pc_sel, ptr + PCIR_HT_COMMAND, 2); printf("HT "); if ((command & 0xe000) == PCIM_HTCAP_SLAVE) printf("slave"); else if ((command & 0xe000) == PCIM_HTCAP_HOST) printf("host"); else switch (command & PCIM_HTCMD_CAP_MASK) { case PCIM_HTCAP_SWITCH: printf("switch"); break; case PCIM_HTCAP_INTERRUPT: printf("interrupt"); break; case PCIM_HTCAP_REVISION_ID: printf("revision ID"); break; case PCIM_HTCAP_UNITID_CLUMPING: printf("unit ID clumping"); break; case PCIM_HTCAP_EXT_CONFIG_SPACE: printf("extended config space"); break; case PCIM_HTCAP_ADDRESS_MAPPING: printf("address mapping"); break; case PCIM_HTCAP_MSI_MAPPING: printf("MSI %saddress window %s at 0x", command & PCIM_HTCMD_MSI_FIXED ? "fixed " : "", command & PCIM_HTCMD_MSI_ENABLE ? "enabled" : "disabled"); if (command & PCIM_HTCMD_MSI_FIXED) printf("fee00000"); else { reg = read_config(fd, &p->pc_sel, ptr + PCIR_HTMSI_ADDRESS_HI, 4); if (reg != 0) printf("%08x", reg); reg = read_config(fd, &p->pc_sel, ptr + PCIR_HTMSI_ADDRESS_LO, 4); printf("%08x", reg); } break; case PCIM_HTCAP_DIRECT_ROUTE: printf("direct route"); break; case PCIM_HTCAP_VCSET: printf("VC set"); break; case PCIM_HTCAP_RETRY_MODE: printf("retry mode"); break; case PCIM_HTCAP_X86_ENCODING: printf("X86 encoding"); break; default: printf("unknown %02x", command); break; } } static void cap_vendor(int fd, struct pci_conf *p, uint8_t ptr) { uint8_t length; length = read_config(fd, &p->pc_sel, ptr + PCIR_VENDOR_LENGTH, 1); printf("vendor (length %d)", length); if (p->pc_vendor == 0x8086) { /* Intel */ uint8_t version; version = read_config(fd, &p->pc_sel, ptr + PCIR_VENDOR_DATA, 1); printf(" Intel cap %d version %d", version >> 4, version & 0xf); if (version >> 4 == 1 && length == 12) { /* Feature Detection */ uint32_t fvec; int comma; comma = 0; fvec = read_config(fd, &p->pc_sel, ptr + PCIR_VENDOR_DATA + 5, 4); printf("\n\t\t features:"); if (fvec & (1 << 0)) { printf(" AMT"); comma = 1; } fvec = read_config(fd, &p->pc_sel, ptr + PCIR_VENDOR_DATA + 1, 4); if (fvec & (1 << 21)) { printf("%s Quick Resume", comma ? "," : ""); comma = 1; } if (fvec & (1 << 18)) { printf("%s SATA RAID-5", comma ? "," : ""); comma = 1; } if (fvec & (1 << 9)) { printf("%s Mobile", comma ? "," : ""); comma = 1; } if (fvec & (1 << 7)) { printf("%s 6 PCI-e x1 slots", comma ? "," : ""); comma = 1; } else { printf("%s 4 PCI-e x1 slots", comma ? "," : ""); comma = 1; } if (fvec & (1 << 5)) { printf("%s SATA RAID-0/1/10", comma ? "," : ""); comma = 1; } if (fvec & (1 << 3)) { printf("%s SATA AHCI", comma ? "," : ""); comma = 1; } } } } static void cap_debug(int fd, struct pci_conf *p, uint8_t ptr) { uint16_t debug_port; debug_port = read_config(fd, &p->pc_sel, ptr + PCIR_DEBUG_PORT, 2); printf("EHCI Debug Port at offset 0x%x in map 0x%x", debug_port & PCIM_DEBUG_PORT_OFFSET, PCIR_BAR(debug_port >> 13)); } static void cap_subvendor(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t id; id = read_config(fd, &p->pc_sel, ptr + PCIR_SUBVENDCAP_ID, 4); printf("PCI Bridge card=0x%08x", id); } #define MAX_PAYLOAD(field) (128 << (field)) static void cap_express(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t val; uint16_t flags; - flags = read_config(fd, &p->pc_sel, ptr + PCIR_EXPRESS_FLAGS, 2); - printf("PCI-Express %d ", flags & PCIM_EXP_FLAGS_VERSION); - switch (flags & PCIM_EXP_FLAGS_TYPE) { - case PCIM_EXP_TYPE_ENDPOINT: + flags = read_config(fd, &p->pc_sel, ptr + PCIER_FLAGS, 2); + printf("PCI-Express %d ", flags & PCIEM_FLAGS_VERSION); + switch (flags & PCIEM_FLAGS_TYPE) { + case PCIEM_TYPE_ENDPOINT: printf("endpoint"); break; - case PCIM_EXP_TYPE_LEGACY_ENDPOINT: + case PCIEM_TYPE_LEGACY_ENDPOINT: printf("legacy endpoint"); break; - case PCIM_EXP_TYPE_ROOT_PORT: + case PCIEM_TYPE_ROOT_PORT: printf("root port"); break; - case PCIM_EXP_TYPE_UPSTREAM_PORT: + case PCIEM_TYPE_UPSTREAM_PORT: printf("upstream port"); break; - case PCIM_EXP_TYPE_DOWNSTREAM_PORT: + case PCIEM_TYPE_DOWNSTREAM_PORT: printf("downstream port"); break; - case PCIM_EXP_TYPE_PCI_BRIDGE: + case PCIEM_TYPE_PCI_BRIDGE: printf("PCI bridge"); break; - case PCIM_EXP_TYPE_PCIE_BRIDGE: + case PCIEM_TYPE_PCIE_BRIDGE: printf("PCI to PCIe bridge"); break; - case PCIM_EXP_TYPE_ROOT_INT_EP: + case PCIEM_TYPE_ROOT_INT_EP: printf("root endpoint"); break; - case PCIM_EXP_TYPE_ROOT_EC: + case PCIEM_TYPE_ROOT_EC: printf("event collector"); break; default: - printf("type %d", (flags & PCIM_EXP_FLAGS_TYPE) >> 4); + printf("type %d", (flags & PCIEM_FLAGS_TYPE) >> 4); break; } - if (flags & PCIM_EXP_FLAGS_IRQ) - printf(" IRQ %d", (flags & PCIM_EXP_FLAGS_IRQ) >> 8); - val = read_config(fd, &p->pc_sel, ptr + PCIR_EXPRESS_DEVICE_CAP, 4); - flags = read_config(fd, &p->pc_sel, ptr + PCIR_EXPRESS_DEVICE_CTL, 2); + if (flags & PCIEM_FLAGS_IRQ) + printf(" IRQ %d", (flags & PCIEM_FLAGS_IRQ) >> 8); + val = read_config(fd, &p->pc_sel, ptr + PCIER_DEVICE_CAP, 4); + flags = read_config(fd, &p->pc_sel, ptr + PCIER_DEVICE_CTL, 2); printf(" max data %d(%d)", - MAX_PAYLOAD((flags & PCIM_EXP_CTL_MAX_PAYLOAD) >> 5), - MAX_PAYLOAD(val & PCIM_EXP_CAP_MAX_PAYLOAD)); - val = read_config(fd, &p->pc_sel, ptr + PCIR_EXPRESS_LINK_CAP, 4); - flags = read_config(fd, &p->pc_sel, ptr+ PCIR_EXPRESS_LINK_STA, 2); - printf(" link x%d(x%d)", (flags & PCIM_LINK_STA_WIDTH) >> 4, - (val & PCIM_LINK_CAP_MAX_WIDTH) >> 4); + MAX_PAYLOAD((flags & PCIEM_CTL_MAX_PAYLOAD) >> 5), + MAX_PAYLOAD(val & PCIEM_CAP_MAX_PAYLOAD)); + val = read_config(fd, &p->pc_sel, ptr + PCIER_LINK_CAP, 4); + flags = read_config(fd, &p->pc_sel, ptr+ PCIER_LINK_STA, 2); + printf(" link x%d(x%d)", (flags & PCIEM_LINK_STA_WIDTH) >> 4, + (val & PCIEM_LINK_CAP_MAX_WIDTH) >> 4); } static void cap_msix(int fd, struct pci_conf *p, uint8_t ptr) { uint32_t val; uint16_t ctrl; int msgnum, table_bar, pba_bar; ctrl = read_config(fd, &p->pc_sel, ptr + PCIR_MSIX_CTRL, 2); msgnum = (ctrl & PCIM_MSIXCTRL_TABLE_SIZE) + 1; val = read_config(fd, &p->pc_sel, ptr + PCIR_MSIX_TABLE, 4); table_bar = PCIR_BAR(val & PCIM_MSIX_BIR_MASK); val = read_config(fd, &p->pc_sel, ptr + PCIR_MSIX_PBA, 4); pba_bar = PCIR_BAR(val & PCIM_MSIX_BIR_MASK); printf("MSI-X supports %d message%s ", msgnum, (msgnum == 1) ? "" : "s"); if (table_bar == pba_bar) printf("in map 0x%x", table_bar); else printf("in maps 0x%x and 0x%x", table_bar, pba_bar); if (ctrl & PCIM_MSIXCTRL_MSIX_ENABLE) printf(" enabled"); } static void cap_sata(int fd, struct pci_conf *p, uint8_t ptr) { printf("SATA Index-Data Pair"); } static void cap_pciaf(int fd, struct pci_conf *p, uint8_t ptr) { uint8_t cap; cap = read_config(fd, &p->pc_sel, ptr + PCIR_PCIAF_CAP, 1); printf("PCI Advanced Features:%s%s", cap & PCIM_PCIAFCAP_FLR ? " FLR" : "", cap & PCIM_PCIAFCAP_TP ? " TP" : ""); } void list_caps(int fd, struct pci_conf *p) { int express; uint16_t sta; uint8_t ptr, cap; /* Are capabilities present for this device? */ sta = read_config(fd, &p->pc_sel, PCIR_STATUS, 2); if (!(sta & PCIM_STATUS_CAPPRESENT)) return; switch (p->pc_hdr & PCIM_HDRTYPE) { case PCIM_HDRTYPE_NORMAL: case PCIM_HDRTYPE_BRIDGE: ptr = PCIR_CAP_PTR; break; case PCIM_HDRTYPE_CARDBUS: ptr = PCIR_CAP_PTR_2; break; default: errx(1, "list_caps: bad header type"); } /* Walk the capability list. */ express = 0; ptr = read_config(fd, &p->pc_sel, ptr, 1); while (ptr != 0 && ptr != 0xff) { cap = read_config(fd, &p->pc_sel, ptr + PCICAP_ID, 1); printf(" cap %02x[%02x] = ", cap, ptr); switch (cap) { case PCIY_PMG: cap_power(fd, p, ptr); break; case PCIY_AGP: cap_agp(fd, p, ptr); break; case PCIY_VPD: cap_vpd(fd, p, ptr); break; case PCIY_MSI: cap_msi(fd, p, ptr); break; case PCIY_PCIX: cap_pcix(fd, p, ptr); break; case PCIY_HT: cap_ht(fd, p, ptr); break; case PCIY_VENDOR: cap_vendor(fd, p, ptr); break; case PCIY_DEBUG: cap_debug(fd, p, ptr); break; case PCIY_SUBVENDOR: cap_subvendor(fd, p, ptr); break; case PCIY_EXPRESS: express = 1; cap_express(fd, p, ptr); break; case PCIY_MSIX: cap_msix(fd, p, ptr); break; case PCIY_SATA: cap_sata(fd, p, ptr); break; case PCIY_PCIAF: cap_pciaf(fd, p, ptr); break; default: printf("unknown"); break; } printf("\n"); ptr = read_config(fd, &p->pc_sel, ptr + PCICAP_NEXTPTR, 1); } if (express) list_ecaps(fd, p); } /* From . */ static __inline uint32_t bitcount32(uint32_t x) { x = (x & 0x55555555) + ((x & 0xaaaaaaaa) >> 1); x = (x & 0x33333333) + ((x & 0xcccccccc) >> 2); x = (x + (x >> 4)) & 0x0f0f0f0f; x = (x + (x >> 8)); x = (x + (x >> 16)) & 0x000000ff; return (x); } static void ecap_aer(int fd, struct pci_conf *p, uint16_t ptr, uint8_t ver) { uint32_t sta, mask; printf("AER %d", ver); if (ver != 1) return; sta = read_config(fd, &p->pc_sel, ptr + PCIR_AER_UC_STATUS, 4); mask = read_config(fd, &p->pc_sel, ptr + PCIR_AER_UC_SEVERITY, 4); printf(" %d fatal", bitcount32(sta & mask)); printf(" %d non-fatal", bitcount32(sta & ~mask)); sta = read_config(fd, &p->pc_sel, ptr + PCIR_AER_COR_STATUS, 4); printf(" %d corrected", bitcount32(sta)); } static void ecap_vc(int fd, struct pci_conf *p, uint16_t ptr, uint8_t ver) { uint32_t cap1; printf("VC %d", ver); if (ver != 1) return; cap1 = read_config(fd, &p->pc_sel, ptr + PCIR_VC_CAP1, 4); printf(" max VC%d", cap1 & PCIM_VC_CAP1_EXT_COUNT); if ((cap1 & PCIM_VC_CAP1_LOWPRI_EXT_COUNT) != 0) printf(" lowpri VC0-VC%d", (cap1 & PCIM_VC_CAP1_LOWPRI_EXT_COUNT) >> 4); } static void ecap_sernum(int fd, struct pci_conf *p, uint16_t ptr, uint8_t ver) { uint32_t high, low; printf("Serial %d", ver); if (ver != 1) return; low = read_config(fd, &p->pc_sel, ptr + PCIR_SERIAL_LOW, 4); high = read_config(fd, &p->pc_sel, ptr + PCIR_SERIAL_HIGH, 4); printf(" %08x%08x", high, low); } static void list_ecaps(int fd, struct pci_conf *p) { uint32_t ecap; uint16_t ptr; ptr = PCIR_EXTCAP; ecap = read_config(fd, &p->pc_sel, ptr, 4); if (ecap == 0xffffffff || ecap == 0) return; for (;;) { printf("ecap %04x[%03x] = ", PCI_EXTCAP_ID(ecap), ptr); switch (PCI_EXTCAP_ID(ecap)) { case PCIZ_AER: ecap_aer(fd, p, ptr, PCI_EXTCAP_VER(ecap)); break; case PCIZ_VC: ecap_vc(fd, p, ptr, PCI_EXTCAP_VER(ecap)); break; case PCIZ_SERNUM: ecap_sernum(fd, p, ptr, PCI_EXTCAP_VER(ecap)); break; default: printf("unknown %d", PCI_EXTCAP_VER(ecap)); break; } printf("\n"); ptr = PCI_EXTCAP_NEXTPTR(ecap); if (ptr == 0) break; ecap = read_config(fd, &p->pc_sel, ptr, 4); } } Index: stable/7/usr.sbin/pciconf =================================================================== --- stable/7/usr.sbin/pciconf (revision 242221) +++ stable/7/usr.sbin/pciconf (revision 242222) Property changes on: stable/7/usr.sbin/pciconf ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head/usr.sbin/pciconf:r240680