Index: head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c =================================================================== --- head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c (revision 297912) +++ head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c (revision 297913) @@ -1,2904 +1,2897 @@ /*- * Copyright (c) 2010-2012 Citrix Inc. * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2012 NetApp 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 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. */ /*- * Copyright (c) 2004-2006 Kip Macy * 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. * * 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 "opt_inet6.h" #include "opt_inet.h" #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 #include #include #include #include "hv_net_vsc.h" #include "hv_rndis.h" #include "hv_rndis_filter.h" #define hv_chan_rxr hv_chan_priv1 #define hv_chan_txr hv_chan_priv2 /* Short for Hyper-V network interface */ #define NETVSC_DEVNAME "hn" /* * It looks like offset 0 of buf is reserved to hold the softc pointer. * The sc pointer evidently not needed, and is not presently populated. * The packet offset is where the netvsc_packet starts in the buffer. */ #define HV_NV_SC_PTR_OFFSET_IN_BUF 0 #define HV_NV_PACKET_OFFSET_IN_BUF 16 /* YYY should get it from the underlying channel */ #define HN_TX_DESC_CNT 512 #define HN_LROENT_CNT_DEF 128 #define HN_RING_CNT_DEF_MAX 8 #define HN_RNDIS_MSG_LEN \ (sizeof(rndis_msg) + \ RNDIS_HASH_PPI_SIZE + \ RNDIS_VLAN_PPI_SIZE + \ RNDIS_TSO_PPI_SIZE + \ RNDIS_CSUM_PPI_SIZE) #define HN_RNDIS_MSG_BOUNDARY PAGE_SIZE #define HN_RNDIS_MSG_ALIGN CACHE_LINE_SIZE #define HN_TX_DATA_BOUNDARY PAGE_SIZE #define HN_TX_DATA_MAXSIZE IP_MAXPACKET #define HN_TX_DATA_SEGSIZE PAGE_SIZE #define HN_TX_DATA_SEGCNT_MAX \ (NETVSC_PACKET_MAXPAGE - HV_RF_NUM_TX_RESERVED_PAGE_BUFS) #define HN_DIRECT_TX_SIZE_DEF 128 struct hn_txdesc { #ifndef HN_USE_TXDESC_BUFRING SLIST_ENTRY(hn_txdesc) link; #endif struct mbuf *m; struct hn_tx_ring *txr; int refs; uint32_t flags; /* HN_TXD_FLAG_ */ netvsc_packet netvsc_pkt; /* XXX to be removed */ bus_dmamap_t data_dmap; bus_addr_t rndis_msg_paddr; rndis_msg *rndis_msg; bus_dmamap_t rndis_msg_dmap; }; #define HN_TXD_FLAG_ONLIST 0x1 #define HN_TXD_FLAG_DMAMAP 0x2 /* * Only enable UDP checksum offloading when it is on 2012R2 or * later. UDP checksum offloading doesn't work on earlier * Windows releases. */ #define HN_CSUM_ASSIST_WIN8 (CSUM_IP | CSUM_TCP) #define HN_CSUM_ASSIST (CSUM_IP | CSUM_UDP | CSUM_TCP) #define HN_LRO_LENLIM_MULTIRX_DEF (12 * ETHERMTU) #define HN_LRO_LENLIM_DEF (25 * ETHERMTU) /* YYY 2*MTU is a bit rough, but should be good enough. */ #define HN_LRO_LENLIM_MIN(ifp) (2 * (ifp)->if_mtu) #define HN_LRO_ACKCNT_DEF 1 /* * Be aware that this sleepable mutex will exhibit WITNESS errors when * certain TCP and ARP code paths are taken. This appears to be a * well-known condition, as all other drivers checked use a sleeping * mutex to protect their transmit paths. * Also Be aware that mutexes do not play well with semaphores, and there * is a conflicting semaphore in a certain channel code path. */ #define NV_LOCK_INIT(_sc, _name) \ mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF) #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock) #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED) #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock) #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock) /* * Globals */ int hv_promisc_mode = 0; /* normal mode by default */ SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD, NULL, "Hyper-V network interface"); /* Trust tcp segements verification on host side. */ static int hn_trust_hosttcp = 1; SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN, &hn_trust_hosttcp, 0, "Trust tcp segement verification on host side, " "when csum info is missing (global setting)"); /* Trust udp datagrams verification on host side. */ static int hn_trust_hostudp = 1; SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN, &hn_trust_hostudp, 0, "Trust udp datagram verification on host side, " "when csum info is missing (global setting)"); /* Trust ip packets verification on host side. */ static int hn_trust_hostip = 1; SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN, &hn_trust_hostip, 0, "Trust ip packet verification on host side, " "when csum info is missing (global setting)"); #if __FreeBSD_version >= 1100045 /* Limit TSO burst size */ static int hn_tso_maxlen = 0; SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN, &hn_tso_maxlen, 0, "TSO burst limit"); #endif /* Limit chimney send size */ static int hn_tx_chimney_size = 0; SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN, &hn_tx_chimney_size, 0, "Chimney send packet size limit"); /* Limit the size of packet for direct transmission */ static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF; SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN, &hn_direct_tx_size, 0, "Size of the packet for direct transmission"); #if defined(INET) || defined(INET6) #if __FreeBSD_version >= 1100095 static int hn_lro_entry_count = HN_LROENT_CNT_DEF; SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN, &hn_lro_entry_count, 0, "LRO entry count"); #endif #endif static int hn_share_tx_taskq = 0; SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN, &hn_share_tx_taskq, 0, "Enable shared TX taskqueue"); static struct taskqueue *hn_tx_taskq; #ifndef HN_USE_TXDESC_BUFRING static int hn_use_txdesc_bufring = 0; #else static int hn_use_txdesc_bufring = 1; #endif SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD, &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors"); static int hn_bind_tx_taskq = -1; SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN, &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu"); static int hn_use_if_start = 0; SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN, &hn_use_if_start, 0, "Use if_start TX method"); static int hn_chan_cnt = 0; SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN, &hn_chan_cnt, 0, "# of channels to use; each channel has one RX ring and one TX ring"); static int hn_tx_ring_cnt = 0; SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN, &hn_tx_ring_cnt, 0, "# of TX rings to use"); static u_int hn_cpu_index; /* * Forward declarations */ static void hn_stop(hn_softc_t *sc); static void hn_ifinit_locked(hn_softc_t *sc); static void hn_ifinit(void *xsc); static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data); static int hn_start_locked(struct hn_tx_ring *txr, int len); static void hn_start(struct ifnet *ifp); static void hn_start_txeof(struct hn_tx_ring *); static int hn_ifmedia_upd(struct ifnet *ifp); static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); #if __FreeBSD_version >= 1100099 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS); static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS); #endif static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS); static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS); static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS); static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS); static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS); static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS); static int hn_check_iplen(const struct mbuf *, int); static int hn_create_tx_ring(struct hn_softc *, int); static void hn_destroy_tx_ring(struct hn_tx_ring *); static int hn_create_tx_data(struct hn_softc *, int); static void hn_destroy_tx_data(struct hn_softc *); static void hn_start_taskfunc(void *, int); static void hn_start_txeof_taskfunc(void *, int); static void hn_stop_tx_tasks(struct hn_softc *); static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **); static void hn_create_rx_data(struct hn_softc *sc, int); static void hn_destroy_rx_data(struct hn_softc *sc); static void hn_set_tx_chimney_size(struct hn_softc *, int); static void hn_channel_attach(struct hn_softc *, struct hv_vmbus_channel *); static int hn_transmit(struct ifnet *, struct mbuf *); static void hn_xmit_qflush(struct ifnet *); static int hn_xmit(struct hn_tx_ring *, int); static void hn_xmit_txeof(struct hn_tx_ring *); static void hn_xmit_taskfunc(void *, int); static void hn_xmit_txeof_taskfunc(void *, int); #if __FreeBSD_version >= 1100099 static void hn_set_lro_lenlim(struct hn_softc *sc, int lenlim) { int i; for (i = 0; i < sc->hn_rx_ring_inuse; ++i) sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim; } #endif static int hn_ifmedia_upd(struct ifnet *ifp __unused) { return EOPNOTSUPP; } static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { struct hn_softc *sc = ifp->if_softc; ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (!sc->hn_carrier) { ifmr->ifm_active |= IFM_NONE; return; } ifmr->ifm_status |= IFM_ACTIVE; ifmr->ifm_active |= IFM_10G_T | IFM_FDX; } /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */ static const hv_guid g_net_vsc_device_type = { .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46, 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E} }; /* * Standard probe entry point. * */ static int netvsc_probe(device_t dev) { const char *p; p = vmbus_get_type(dev); if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) { device_set_desc(dev, "Synthetic Network Interface"); if (bootverbose) printf("Netvsc probe... DONE \n"); return (BUS_PROBE_DEFAULT); } return (ENXIO); } /* * Standard attach entry point. * * Called when the driver is loaded. It allocates needed resources, * and initializes the "hardware" and software. */ static int netvsc_attach(device_t dev) { struct hv_device *device_ctx = vmbus_get_devctx(dev); struct hv_vmbus_channel *chan; netvsc_device_info device_info; hn_softc_t *sc; int unit = device_get_unit(dev); struct ifnet *ifp = NULL; int error, ring_cnt, tx_ring_cnt; #if __FreeBSD_version >= 1100045 int tso_maxlen; #endif sc = device_get_softc(dev); - if (sc == NULL) { - return (ENOMEM); - } bzero(sc, sizeof(hn_softc_t)); sc->hn_unit = unit; sc->hn_dev = dev; if (hn_tx_taskq == NULL) { sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK, taskqueue_thread_enqueue, &sc->hn_tx_taskq); if (hn_bind_tx_taskq >= 0) { int cpu = hn_bind_tx_taskq; cpuset_t cpu_set; if (cpu > mp_ncpus - 1) cpu = mp_ncpus - 1; CPU_SETOF(cpu, &cpu_set); taskqueue_start_threads_cpuset(&sc->hn_tx_taskq, 1, PI_NET, &cpu_set, "%s tx", device_get_nameunit(dev)); } else { taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx", device_get_nameunit(dev)); } } else { sc->hn_tx_taskq = hn_tx_taskq; } NV_LOCK_INIT(sc, "NetVSCLock"); sc->hn_dev_obj = device_ctx; ifp = sc->hn_ifp = if_alloc(IFT_ETHER); ifp->if_softc = sc; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); /* * Figure out the # of RX rings (ring_cnt) and the # of TX rings * to use (tx_ring_cnt). * * NOTE: * The # of RX rings to use is same as the # of channels to use. */ ring_cnt = hn_chan_cnt; if (ring_cnt <= 0) { /* Default */ ring_cnt = mp_ncpus; if (ring_cnt > HN_RING_CNT_DEF_MAX) ring_cnt = HN_RING_CNT_DEF_MAX; } else if (ring_cnt > mp_ncpus) { ring_cnt = mp_ncpus; } tx_ring_cnt = hn_tx_ring_cnt; if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt) tx_ring_cnt = ring_cnt; if (hn_use_if_start) { /* ifnet.if_start only needs one TX ring. */ tx_ring_cnt = 1; } /* * Set the leader CPU for channels. */ sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus; error = hn_create_tx_data(sc, tx_ring_cnt); if (error) goto failed; hn_create_rx_data(sc, ring_cnt); /* * Associate the first TX/RX ring w/ the primary channel. */ chan = device_ctx->channel; KASSERT(HV_VMBUS_CHAN_ISPRIMARY(chan), ("not primary channel")); KASSERT(chan->offer_msg.offer.sub_channel_index == 0, ("primary channel subidx %u", chan->offer_msg.offer.sub_channel_index)); hn_channel_attach(sc, chan); ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = hn_ioctl; ifp->if_init = hn_ifinit; /* needed by hv_rf_on_device_add() code */ ifp->if_mtu = ETHERMTU; if (hn_use_if_start) { ifp->if_start = hn_start; IFQ_SET_MAXLEN(&ifp->if_snd, 512); ifp->if_snd.ifq_drv_maxlen = 511; IFQ_SET_READY(&ifp->if_snd); } else { ifp->if_transmit = hn_transmit; ifp->if_qflush = hn_xmit_qflush; } ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts); ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO); /* XXX ifmedia_set really should do this for us */ sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media; /* * Tell upper layers that we support full VLAN capability. */ ifp->if_hdrlen = sizeof(struct ether_vlan_header); ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO | IFCAP_LRO; ifp->if_capenable |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO | IFCAP_LRO; ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO; error = hv_rf_on_device_add(device_ctx, &device_info, ring_cnt); if (error) goto failed; KASSERT(sc->net_dev->num_channel > 0 && sc->net_dev->num_channel <= sc->hn_rx_ring_inuse, ("invalid channel count %u, should be less than %d", sc->net_dev->num_channel, sc->hn_rx_ring_inuse)); /* * Set the # of TX/RX rings that could be used according to * the # of channels that host offered. */ if (sc->hn_tx_ring_inuse > sc->net_dev->num_channel) sc->hn_tx_ring_inuse = sc->net_dev->num_channel; sc->hn_rx_ring_inuse = sc->net_dev->num_channel; device_printf(dev, "%d TX ring, %d RX ring\n", sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse); #if __FreeBSD_version >= 1100099 if (sc->hn_rx_ring_inuse > 1) { /* * Reduce TCP segment aggregation limit for multiple * RX rings to increase ACK timeliness. */ hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF); } #endif if (device_info.link_state == 0) { sc->hn_carrier = 1; } #if __FreeBSD_version >= 1100045 tso_maxlen = hn_tso_maxlen; if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET) tso_maxlen = IP_MAXPACKET; ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX; ifp->if_hw_tsomaxsegsize = PAGE_SIZE; ifp->if_hw_tsomax = tso_maxlen - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); #endif ether_ifattach(ifp, device_info.mac_addr); #if __FreeBSD_version >= 1100045 if_printf(ifp, "TSO: %u/%u/%u\n", ifp->if_hw_tsomax, ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize); #endif sc->hn_tx_chimney_max = sc->net_dev->send_section_size; hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max); if (hn_tx_chimney_size > 0 && hn_tx_chimney_size < sc->hn_tx_chimney_max) hn_set_tx_chimney_size(sc, hn_tx_chimney_size); return (0); failed: hn_destroy_tx_data(sc); if (ifp != NULL) if_free(ifp); return (error); } /* * Standard detach entry point */ static int netvsc_detach(device_t dev) { struct hn_softc *sc = device_get_softc(dev); struct hv_device *hv_device = vmbus_get_devctx(dev); if (bootverbose) printf("netvsc_detach\n"); /* * XXXKYS: Need to clean up all our * driver state; this is the driver * unloading. */ /* * XXXKYS: Need to stop outgoing traffic and unregister * the netdevice. */ hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL); hn_stop_tx_tasks(sc); ifmedia_removeall(&sc->hn_media); hn_destroy_rx_data(sc); hn_destroy_tx_data(sc); if (sc->hn_tx_taskq != hn_tx_taskq) taskqueue_free(sc->hn_tx_taskq); return (0); } /* * Standard shutdown entry point */ static int netvsc_shutdown(device_t dev) { return (0); } static __inline int hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs) { struct mbuf *m = *m_head; int error; error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT); if (error == EFBIG) { struct mbuf *m_new; m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX); if (m_new == NULL) return ENOBUFS; else *m_head = m = m_new; txr->hn_tx_collapsed++; error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT); } if (!error) { bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap, BUS_DMASYNC_PREWRITE); txd->flags |= HN_TXD_FLAG_DMAMAP; } return error; } static __inline void hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd) { if (txd->flags & HN_TXD_FLAG_DMAMAP) { bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txr->hn_tx_data_dtag, txd->data_dmap); txd->flags &= ~HN_TXD_FLAG_DMAMAP; } } static __inline int hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd) { KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0, ("put an onlist txd %#x", txd->flags)); KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs)); if (atomic_fetchadd_int(&txd->refs, -1) != 1) return 0; hn_txdesc_dmamap_unload(txr, txd); if (txd->m != NULL) { m_freem(txd->m); txd->m = NULL; } txd->flags |= HN_TXD_FLAG_ONLIST; #ifndef HN_USE_TXDESC_BUFRING mtx_lock_spin(&txr->hn_txlist_spin); KASSERT(txr->hn_txdesc_avail >= 0 && txr->hn_txdesc_avail < txr->hn_txdesc_cnt, ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail)); txr->hn_txdesc_avail++; SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link); mtx_unlock_spin(&txr->hn_txlist_spin); #else atomic_add_int(&txr->hn_txdesc_avail, 1); buf_ring_enqueue(txr->hn_txdesc_br, txd); #endif return 1; } static __inline struct hn_txdesc * hn_txdesc_get(struct hn_tx_ring *txr) { struct hn_txdesc *txd; #ifndef HN_USE_TXDESC_BUFRING mtx_lock_spin(&txr->hn_txlist_spin); txd = SLIST_FIRST(&txr->hn_txlist); if (txd != NULL) { KASSERT(txr->hn_txdesc_avail > 0, ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail)); txr->hn_txdesc_avail--; SLIST_REMOVE_HEAD(&txr->hn_txlist, link); } mtx_unlock_spin(&txr->hn_txlist_spin); #else txd = buf_ring_dequeue_sc(txr->hn_txdesc_br); #endif if (txd != NULL) { #ifdef HN_USE_TXDESC_BUFRING atomic_subtract_int(&txr->hn_txdesc_avail, 1); #endif KASSERT(txd->m == NULL && txd->refs == 0 && (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd")); txd->flags &= ~HN_TXD_FLAG_ONLIST; txd->refs = 1; } return txd; } static __inline void hn_txdesc_hold(struct hn_txdesc *txd) { /* 0->1 transition will never work */ KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs)); atomic_add_int(&txd->refs, 1); } static void hn_tx_done(struct hv_vmbus_channel *chan, void *xpkt) { netvsc_packet *packet = xpkt; struct hn_txdesc *txd; struct hn_tx_ring *txr; txd = (struct hn_txdesc *)(uintptr_t) packet->compl.send.send_completion_tid; txr = txd->txr; KASSERT(txr->hn_chan == chan, ("channel mismatch, on channel%u, should be channel%u", chan->offer_msg.offer.sub_channel_index, txr->hn_chan->offer_msg.offer.sub_channel_index)); txr->hn_has_txeof = 1; hn_txdesc_put(txr, txd); } void netvsc_channel_rollup(struct hv_vmbus_channel *chan) { struct hn_tx_ring *txr = chan->hv_chan_txr; #if defined(INET) || defined(INET6) struct hn_rx_ring *rxr = chan->hv_chan_rxr; tcp_lro_flush_all(&rxr->hn_lro); #endif /* * NOTE: * 'txr' could be NULL, if multiple channels and * ifnet.if_start method are enabled. */ if (txr == NULL || !txr->hn_has_txeof) return; txr->hn_has_txeof = 0; txr->hn_txeof(txr); } /* * NOTE: * If this function fails, then both txd and m_head0 will be freed. */ static int hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0) { bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX]; int error, nsegs, i; struct mbuf *m_head = *m_head0; netvsc_packet *packet; rndis_msg *rndis_mesg; rndis_packet *rndis_pkt; rndis_per_packet_info *rppi; struct ndis_hash_info *hash_info; uint32_t rndis_msg_size; packet = &txd->netvsc_pkt; packet->is_data_pkt = TRUE; packet->tot_data_buf_len = m_head->m_pkthdr.len; /* * extension points to the area reserved for the * rndis_filter_packet, which is placed just after * the netvsc_packet (and rppi struct, if present; * length is updated later). */ rndis_mesg = txd->rndis_msg; /* XXX not necessary */ memset(rndis_mesg, 0, HN_RNDIS_MSG_LEN); rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG; rndis_pkt = &rndis_mesg->msg.packet; rndis_pkt->data_offset = sizeof(rndis_packet); rndis_pkt->data_length = packet->tot_data_buf_len; rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet); rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet); /* * Set the hash info for this packet, so that the host could * dispatch the TX done event for this packet back to this TX * ring's channel. */ rndis_msg_size += RNDIS_HASH_PPI_SIZE; rppi = hv_set_rppi_data(rndis_mesg, RNDIS_HASH_PPI_SIZE, nbl_hash_value); hash_info = (struct ndis_hash_info *)((uint8_t *)rppi + rppi->per_packet_info_offset); hash_info->hash = txr->hn_tx_idx; if (m_head->m_flags & M_VLANTAG) { ndis_8021q_info *rppi_vlan_info; rndis_msg_size += RNDIS_VLAN_PPI_SIZE; rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE, ieee_8021q_info); rppi_vlan_info = (ndis_8021q_info *)((uint8_t *)rppi + rppi->per_packet_info_offset); rppi_vlan_info->u1.s1.vlan_id = m_head->m_pkthdr.ether_vtag & 0xfff; } if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { rndis_tcp_tso_info *tso_info; struct ether_vlan_header *eh; int ether_len; /* * XXX need m_pullup and use mtodo */ eh = mtod(m_head, struct ether_vlan_header*); if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; else ether_len = ETHER_HDR_LEN; rndis_msg_size += RNDIS_TSO_PPI_SIZE; rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE, tcp_large_send_info); tso_info = (rndis_tcp_tso_info *)((uint8_t *)rppi + rppi->per_packet_info_offset); tso_info->lso_v2_xmit.type = RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE; #ifdef INET if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) { struct ip *ip = (struct ip *)(m_head->m_data + ether_len); unsigned long iph_len = ip->ip_hl << 2; struct tcphdr *th = (struct tcphdr *)((caddr_t)ip + iph_len); tso_info->lso_v2_xmit.ip_version = RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4; ip->ip_len = 0; ip->ip_sum = 0; th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET6 { struct ip6_hdr *ip6 = (struct ip6_hdr *) (m_head->m_data + ether_len); struct tcphdr *th = (struct tcphdr *)(ip6 + 1); tso_info->lso_v2_xmit.ip_version = RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6; ip6->ip6_plen = 0; th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0); } #endif tso_info->lso_v2_xmit.tcp_header_offset = 0; tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz; } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) { rndis_tcp_ip_csum_info *csum_info; rndis_msg_size += RNDIS_CSUM_PPI_SIZE; rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE, tcpip_chksum_info); csum_info = (rndis_tcp_ip_csum_info *)((uint8_t *)rppi + rppi->per_packet_info_offset); csum_info->xmit.is_ipv4 = 1; if (m_head->m_pkthdr.csum_flags & CSUM_IP) csum_info->xmit.ip_header_csum = 1; if (m_head->m_pkthdr.csum_flags & CSUM_TCP) { csum_info->xmit.tcp_csum = 1; csum_info->xmit.tcp_header_offset = 0; } else if (m_head->m_pkthdr.csum_flags & CSUM_UDP) { csum_info->xmit.udp_csum = 1; } } rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size; packet->tot_data_buf_len = rndis_mesg->msg_len; /* * Chimney send, if the packet could fit into one chimney buffer. * * TODO: vRSS, chimney buffer should be per-channel. */ if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) { netvsc_dev *net_dev = txr->hn_sc->net_dev; uint32_t send_buf_section_idx; send_buf_section_idx = hv_nv_get_next_send_section(net_dev); if (send_buf_section_idx != NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) { uint8_t *dest = ((uint8_t *)net_dev->send_buf + (send_buf_section_idx * net_dev->send_section_size)); memcpy(dest, rndis_mesg, rndis_msg_size); dest += rndis_msg_size; m_copydata(m_head, 0, m_head->m_pkthdr.len, dest); packet->send_buf_section_idx = send_buf_section_idx; packet->send_buf_section_size = packet->tot_data_buf_len; packet->page_buf_count = 0; txr->hn_tx_chimney++; goto done; } } error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs); if (error) { int freed; /* * This mbuf is not linked w/ the txd yet, so free it now. */ m_freem(m_head); *m_head0 = NULL; freed = hn_txdesc_put(txr, txd); KASSERT(freed != 0, ("fail to free txd upon txdma error")); txr->hn_txdma_failed++; if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1); return error; } *m_head0 = m_head; packet->page_buf_count = nsegs + HV_RF_NUM_TX_RESERVED_PAGE_BUFS; /* send packet with page buffer */ packet->page_buffers[0].pfn = atop(txd->rndis_msg_paddr); packet->page_buffers[0].offset = txd->rndis_msg_paddr & PAGE_MASK; packet->page_buffers[0].length = rndis_msg_size; /* * Fill the page buffers with mbuf info starting at index * HV_RF_NUM_TX_RESERVED_PAGE_BUFS. */ for (i = 0; i < nsegs; ++i) { hv_vmbus_page_buffer *pb = &packet->page_buffers[ i + HV_RF_NUM_TX_RESERVED_PAGE_BUFS]; pb->pfn = atop(segs[i].ds_addr); pb->offset = segs[i].ds_addr & PAGE_MASK; pb->length = segs[i].ds_len; } packet->send_buf_section_idx = NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX; packet->send_buf_section_size = 0; done: txd->m = m_head; /* Set the completion routine */ packet->compl.send.on_send_completion = hn_tx_done; packet->compl.send.send_completion_context = packet; packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)txd; return 0; } /* * NOTE: * If this function fails, then txd will be freed, but the mbuf * associated w/ the txd will _not_ be freed. */ static int hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd) { int error, send_failed = 0; again: /* * Make sure that txd is not freed before ETHER_BPF_MTAP. */ hn_txdesc_hold(txd); error = hv_nv_on_send(txr->hn_chan, &txd->netvsc_pkt); if (!error) { ETHER_BPF_MTAP(ifp, txd->m); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); if (!hn_use_if_start) { if_inc_counter(ifp, IFCOUNTER_OBYTES, txd->m->m_pkthdr.len); if (txd->m->m_flags & M_MCAST) if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); } txr->hn_pkts++; } hn_txdesc_put(txr, txd); if (__predict_false(error)) { int freed; /* * This should "really rarely" happen. * * XXX Too many RX to be acked or too many sideband * commands to run? Ask netvsc_channel_rollup() * to kick start later. */ txr->hn_has_txeof = 1; if (!send_failed) { txr->hn_send_failed++; send_failed = 1; /* * Try sending again after set hn_has_txeof; * in case that we missed the last * netvsc_channel_rollup(). */ goto again; } if_printf(ifp, "send failed\n"); /* * Caller will perform further processing on the * associated mbuf, so don't free it in hn_txdesc_put(); * only unload it from the DMA map in hn_txdesc_put(), * if it was loaded. */ txd->m = NULL; freed = hn_txdesc_put(txr, txd); KASSERT(freed != 0, ("fail to free txd upon send error")); txr->hn_send_failed++; } return error; } /* * Start a transmit of one or more packets */ static int hn_start_locked(struct hn_tx_ring *txr, int len) { struct hn_softc *sc = txr->hn_sc; struct ifnet *ifp = sc->hn_ifp; KASSERT(hn_use_if_start, ("hn_start_locked is called, when if_start is disabled")); KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring")); mtx_assert(&txr->hn_tx_lock, MA_OWNED); if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return 0; while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { struct hn_txdesc *txd; struct mbuf *m_head; int error; IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head); if (m_head == NULL) break; if (len > 0 && m_head->m_pkthdr.len > len) { /* * This sending could be time consuming; let callers * dispatch this packet sending (and sending of any * following up packets) to tx taskqueue. */ IFQ_DRV_PREPEND(&ifp->if_snd, m_head); return 1; } txd = hn_txdesc_get(txr); if (txd == NULL) { txr->hn_no_txdescs++; IFQ_DRV_PREPEND(&ifp->if_snd, m_head); atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); break; } error = hn_encap(txr, txd, &m_head); if (error) { /* Both txd and m_head are freed */ continue; } error = hn_send_pkt(ifp, txr, txd); if (__predict_false(error)) { /* txd is freed, but m_head is not */ IFQ_DRV_PREPEND(&ifp->if_snd, m_head); atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); break; } } return 0; } /* * Link up/down notification */ void netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status) { hn_softc_t *sc = device_get_softc(device_obj->device); - - if (sc == NULL) { - return; - } if (status == 1) { sc->hn_carrier = 1; } else { sc->hn_carrier = 0; } } /* * Append the specified data to the indicated mbuf chain, * Extend the mbuf chain if the new data does not fit in * existing space. * * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c. * There should be an equivalent in the kernel mbuf code, * but there does not appear to be one yet. * * Differs from m_append() in that additional mbufs are * allocated with cluster size MJUMPAGESIZE, and filled * accordingly. * * Return 1 if able to complete the job; otherwise 0. */ static int hv_m_append(struct mbuf *m0, int len, c_caddr_t cp) { struct mbuf *m, *n; int remainder, space; for (m = m0; m->m_next != NULL; m = m->m_next) ; remainder = len; space = M_TRAILINGSPACE(m); if (space > 0) { /* * Copy into available space. */ if (space > remainder) space = remainder; bcopy(cp, mtod(m, caddr_t) + m->m_len, space); m->m_len += space; cp += space; remainder -= space; } while (remainder > 0) { /* * Allocate a new mbuf; could check space * and allocate a cluster instead. */ n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE); if (n == NULL) break; n->m_len = min(MJUMPAGESIZE, remainder); bcopy(cp, mtod(n, caddr_t), n->m_len); cp += n->m_len; remainder -= n->m_len; m->m_next = n; m = n; } if (m0->m_flags & M_PKTHDR) m0->m_pkthdr.len += len - remainder; return (remainder == 0); } /* * Called when we receive a data packet from the "wire" on the * specified device * * Note: This is no longer used as a callback */ int netvsc_recv(struct hv_vmbus_channel *chan, netvsc_packet *packet, rndis_tcp_ip_csum_info *csum_info) { struct hn_rx_ring *rxr = chan->hv_chan_rxr; struct ifnet *ifp = rxr->hn_ifp; struct mbuf *m_new; int size, do_lro = 0, do_csum = 1; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) return (0); /* * Bail out if packet contains more data than configured MTU. */ if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) { return (0); } else if (packet->tot_data_buf_len <= MHLEN) { m_new = m_gethdr(M_NOWAIT, MT_DATA); if (m_new == NULL) { if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1); return (0); } memcpy(mtod(m_new, void *), packet->data, packet->tot_data_buf_len); m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len; rxr->hn_small_pkts++; } else { /* * Get an mbuf with a cluster. For packets 2K or less, * get a standard 2K cluster. For anything larger, get a * 4K cluster. Any buffers larger than 4K can cause problems * if looped around to the Hyper-V TX channel, so avoid them. */ size = MCLBYTES; if (packet->tot_data_buf_len > MCLBYTES) { /* 4096 */ size = MJUMPAGESIZE; } m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size); if (m_new == NULL) { if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1); return (0); } hv_m_append(m_new, packet->tot_data_buf_len, packet->data); } m_new->m_pkthdr.rcvif = ifp; if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0)) do_csum = 0; /* receive side checksum offload */ if (csum_info != NULL) { /* IP csum offload */ if (csum_info->receive.ip_csum_succeeded && do_csum) { m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); rxr->hn_csum_ip++; } /* TCP/UDP csum offload */ if ((csum_info->receive.tcp_csum_succeeded || csum_info->receive.udp_csum_succeeded) && do_csum) { m_new->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m_new->m_pkthdr.csum_data = 0xffff; if (csum_info->receive.tcp_csum_succeeded) rxr->hn_csum_tcp++; else rxr->hn_csum_udp++; } if (csum_info->receive.ip_csum_succeeded && csum_info->receive.tcp_csum_succeeded) do_lro = 1; } else { const struct ether_header *eh; uint16_t etype; int hoff; hoff = sizeof(*eh); if (m_new->m_len < hoff) goto skip; eh = mtod(m_new, struct ether_header *); etype = ntohs(eh->ether_type); if (etype == ETHERTYPE_VLAN) { const struct ether_vlan_header *evl; hoff = sizeof(*evl); if (m_new->m_len < hoff) goto skip; evl = mtod(m_new, struct ether_vlan_header *); etype = ntohs(evl->evl_proto); } if (etype == ETHERTYPE_IP) { int pr; pr = hn_check_iplen(m_new, hoff); if (pr == IPPROTO_TCP) { if (do_csum && (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_TCP)) { rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m_new->m_pkthdr.csum_data = 0xffff; } /* Rely on SW csum verification though... */ do_lro = 1; } else if (pr == IPPROTO_UDP) { if (do_csum && (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_UDP)) { rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m_new->m_pkthdr.csum_data = 0xffff; } } else if (pr != IPPROTO_DONE && do_csum && (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) { rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); } } } skip: if ((packet->vlan_tci != 0) && (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) { m_new->m_pkthdr.ether_vtag = packet->vlan_tci; m_new->m_flags |= M_VLANTAG; } m_new->m_pkthdr.flowid = rxr->hn_rx_idx; M_HASHTYPE_SET(m_new, M_HASHTYPE_OPAQUE); /* * Note: Moved RX completion back to hv_nv_on_receive() so all * messages (not just data messages) will trigger a response. */ if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); rxr->hn_pkts++; if ((ifp->if_capenable & IFCAP_LRO) && do_lro) { #if defined(INET) || defined(INET6) struct lro_ctrl *lro = &rxr->hn_lro; if (lro->lro_cnt) { rxr->hn_lro_tried++; if (tcp_lro_rx(lro, m_new, 0) == 0) { /* DONE! */ return 0; } } #endif } /* We're not holding the lock here, so don't release it */ (*ifp->if_input)(ifp, m_new); return (0); } /* * Rules for using sc->temp_unusable: * 1. sc->temp_unusable can only be read or written while holding NV_LOCK() * 2. code reading sc->temp_unusable under NV_LOCK(), and finding * sc->temp_unusable set, must release NV_LOCK() and exit * 3. to retain exclusive control of the interface, * sc->temp_unusable must be set by code before releasing NV_LOCK() * 4. only code setting sc->temp_unusable can clear sc->temp_unusable * 5. code setting sc->temp_unusable must eventually clear sc->temp_unusable */ /* * Standard ioctl entry point. Called when the user wants to configure * the interface. */ static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { hn_softc_t *sc = ifp->if_softc; struct ifreq *ifr = (struct ifreq *)data; #ifdef INET struct ifaddr *ifa = (struct ifaddr *)data; #endif netvsc_device_info device_info; struct hv_device *hn_dev; int mask, error = 0; int retry_cnt = 500; switch(cmd) { case SIOCSIFADDR: #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) { ifp->if_flags |= IFF_UP; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) hn_ifinit(sc); arp_ifinit(ifp, ifa); } else #endif error = ether_ioctl(ifp, cmd, data); break; case SIOCSIFMTU: hn_dev = vmbus_get_devctx(sc->hn_dev); /* Check MTU value change */ if (ifp->if_mtu == ifr->ifr_mtu) break; if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) { error = EINVAL; break; } /* Obtain and record requested MTU */ ifp->if_mtu = ifr->ifr_mtu; #if __FreeBSD_version >= 1100099 /* * Make sure that LRO aggregation length limit is still * valid, after the MTU change. */ NV_LOCK(sc); if (sc->hn_rx_ring[0].hn_lro.lro_length_lim < HN_LRO_LENLIM_MIN(ifp)) hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp)); NV_UNLOCK(sc); #endif do { NV_LOCK(sc); if (!sc->temp_unusable) { sc->temp_unusable = TRUE; retry_cnt = -1; } NV_UNLOCK(sc); if (retry_cnt > 0) { retry_cnt--; DELAY(5 * 1000); } } while (retry_cnt > 0); if (retry_cnt == 0) { error = EINVAL; break; } /* We must remove and add back the device to cause the new * MTU to take effect. This includes tearing down, but not * deleting the channel, then bringing it back up. */ error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL); if (error) { NV_LOCK(sc); sc->temp_unusable = FALSE; NV_UNLOCK(sc); break; } error = hv_rf_on_device_add(hn_dev, &device_info, sc->hn_rx_ring_inuse); if (error) { NV_LOCK(sc); sc->temp_unusable = FALSE; NV_UNLOCK(sc); break; } sc->hn_tx_chimney_max = sc->net_dev->send_section_size; if (sc->hn_tx_ring[0].hn_tx_chimney_size > sc->hn_tx_chimney_max) hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max); hn_ifinit_locked(sc); NV_LOCK(sc); sc->temp_unusable = FALSE; NV_UNLOCK(sc); break; case SIOCSIFFLAGS: do { NV_LOCK(sc); if (!sc->temp_unusable) { sc->temp_unusable = TRUE; retry_cnt = -1; } NV_UNLOCK(sc); if (retry_cnt > 0) { retry_cnt--; DELAY(5 * 1000); } } while (retry_cnt > 0); if (retry_cnt == 0) { error = EINVAL; break; } 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. */ #ifdef notyet /* Fixme: Promiscuous mode? */ if (ifp->if_drv_flags & IFF_DRV_RUNNING && ifp->if_flags & IFF_PROMISC && !(sc->hn_if_flags & IFF_PROMISC)) { /* do something here for Hyper-V */ } else if (ifp->if_drv_flags & IFF_DRV_RUNNING && !(ifp->if_flags & IFF_PROMISC) && sc->hn_if_flags & IFF_PROMISC) { /* do something here for Hyper-V */ } else #endif hn_ifinit_locked(sc); } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { hn_stop(sc); } } NV_LOCK(sc); sc->temp_unusable = FALSE; NV_UNLOCK(sc); sc->hn_if_flags = ifp->if_flags; error = 0; break; case SIOCSIFCAP: NV_LOCK(sc); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (mask & IFCAP_TXCSUM) { ifp->if_capenable ^= IFCAP_TXCSUM; if (ifp->if_capenable & IFCAP_TXCSUM) { ifp->if_hwassist |= sc->hn_tx_ring[0].hn_csum_assist; } else { ifp->if_hwassist &= ~sc->hn_tx_ring[0].hn_csum_assist; } } if (mask & IFCAP_RXCSUM) ifp->if_capenable ^= IFCAP_RXCSUM; if (mask & IFCAP_LRO) ifp->if_capenable ^= IFCAP_LRO; if (mask & IFCAP_TSO4) { ifp->if_capenable ^= IFCAP_TSO4; if (ifp->if_capenable & IFCAP_TSO4) ifp->if_hwassist |= CSUM_IP_TSO; else ifp->if_hwassist &= ~CSUM_IP_TSO; } if (mask & IFCAP_TSO6) { ifp->if_capenable ^= IFCAP_TSO6; if (ifp->if_capenable & IFCAP_TSO6) ifp->if_hwassist |= CSUM_IP6_TSO; else ifp->if_hwassist &= ~CSUM_IP6_TSO; } NV_UNLOCK(sc); error = 0; break; case SIOCADDMULTI: case SIOCDELMULTI: #ifdef notyet /* Fixme: Multicast mode? */ if (ifp->if_drv_flags & IFF_DRV_RUNNING) { NV_LOCK(sc); netvsc_setmulti(sc); NV_UNLOCK(sc); error = 0; } #endif error = EINVAL; break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd); break; default: error = ether_ioctl(ifp, cmd, data); break; } return (error); } /* * */ static void hn_stop(hn_softc_t *sc) { struct ifnet *ifp; int ret, i; struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev); ifp = sc->hn_ifp; if (bootverbose) printf(" Closing Device ...\n"); atomic_clear_int(&ifp->if_drv_flags, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)); for (i = 0; i < sc->hn_tx_ring_inuse; ++i) sc->hn_tx_ring[i].hn_oactive = 0; if_link_state_change(ifp, LINK_STATE_DOWN); sc->hn_initdone = 0; ret = hv_rf_on_close(device_ctx); } /* * FreeBSD transmit entry point */ static void hn_start(struct ifnet *ifp) { struct hn_softc *sc = ifp->if_softc; struct hn_tx_ring *txr = &sc->hn_tx_ring[0]; if (txr->hn_sched_tx) goto do_sched; if (mtx_trylock(&txr->hn_tx_lock)) { int sched; sched = hn_start_locked(txr, txr->hn_direct_tx_size); mtx_unlock(&txr->hn_tx_lock); if (!sched) return; } do_sched: taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task); } static void hn_start_txeof(struct hn_tx_ring *txr) { struct hn_softc *sc = txr->hn_sc; struct ifnet *ifp = sc->hn_ifp; KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring")); if (txr->hn_sched_tx) goto do_sched; if (mtx_trylock(&txr->hn_tx_lock)) { int sched; atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); sched = hn_start_locked(txr, txr->hn_direct_tx_size); mtx_unlock(&txr->hn_tx_lock); if (sched) { taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task); } } else { do_sched: /* * Release the OACTIVE earlier, with the hope, that * others could catch up. The task will clear the * flag again with the hn_tx_lock to avoid possible * races. */ atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task); } } /* * */ static void hn_ifinit_locked(hn_softc_t *sc) { struct ifnet *ifp; struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev); int ret, i; ifp = sc->hn_ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { return; } hv_promisc_mode = 1; ret = hv_rf_on_open(device_ctx); if (ret != 0) { return; } else { sc->hn_initdone = 1; } atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); for (i = 0; i < sc->hn_tx_ring_inuse; ++i) sc->hn_tx_ring[i].hn_oactive = 0; atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING); if_link_state_change(ifp, LINK_STATE_UP); } /* * */ static void hn_ifinit(void *xsc) { hn_softc_t *sc = xsc; NV_LOCK(sc); if (sc->temp_unusable) { NV_UNLOCK(sc); return; } sc->temp_unusable = TRUE; NV_UNLOCK(sc); hn_ifinit_locked(sc); NV_LOCK(sc); sc->temp_unusable = FALSE; NV_UNLOCK(sc); } #ifdef LATER /* * */ static void hn_watchdog(struct ifnet *ifp) { hn_softc_t *sc; sc = ifp->if_softc; printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit); hn_ifinit(sc); /*???*/ if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); } #endif #if __FreeBSD_version >= 1100099 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; unsigned int lenlim; int error; lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim; error = sysctl_handle_int(oidp, &lenlim, 0, req); if (error || req->newptr == NULL) return error; if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) || lenlim > TCP_LRO_LENGTH_MAX) return EINVAL; NV_LOCK(sc); hn_set_lro_lenlim(sc, lenlim); NV_UNLOCK(sc); return 0; } static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int ackcnt, error, i; /* * lro_ackcnt_lim is append count limit, * +1 to turn it into aggregation limit. */ ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1; error = sysctl_handle_int(oidp, &ackcnt, 0, req); if (error || req->newptr == NULL) return error; if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1)) return EINVAL; /* * Convert aggregation limit back to append * count limit. */ --ackcnt; NV_LOCK(sc); for (i = 0; i < sc->hn_rx_ring_inuse; ++i) sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt; NV_UNLOCK(sc); return 0; } #endif static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int hcsum = arg2; int on, error, i; on = 0; if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum) on = 1; error = sysctl_handle_int(oidp, &on, 0, req); if (error || req->newptr == NULL) return error; NV_LOCK(sc); for (i = 0; i < sc->hn_rx_ring_inuse; ++i) { struct hn_rx_ring *rxr = &sc->hn_rx_ring[i]; if (on) rxr->hn_trust_hcsum |= hcsum; else rxr->hn_trust_hcsum &= ~hcsum; } NV_UNLOCK(sc); return 0; } static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int chimney_size, error; chimney_size = sc->hn_tx_ring[0].hn_tx_chimney_size; error = sysctl_handle_int(oidp, &chimney_size, 0, req); if (error || req->newptr == NULL) return error; if (chimney_size > sc->hn_tx_chimney_max || chimney_size <= 0) return EINVAL; hn_set_tx_chimney_size(sc, chimney_size); return 0; } static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int ofs = arg2, i, error; struct hn_rx_ring *rxr; u_long stat; stat = 0; for (i = 0; i < sc->hn_rx_ring_inuse; ++i) { rxr = &sc->hn_rx_ring[i]; stat += *((u_long *)((uint8_t *)rxr + ofs)); } error = sysctl_handle_long(oidp, &stat, 0, req); if (error || req->newptr == NULL) return error; /* Zero out this stat. */ for (i = 0; i < sc->hn_rx_ring_inuse; ++i) { rxr = &sc->hn_rx_ring[i]; *((u_long *)((uint8_t *)rxr + ofs)) = 0; } return 0; } static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int ofs = arg2, i, error; struct hn_rx_ring *rxr; uint64_t stat; stat = 0; for (i = 0; i < sc->hn_rx_ring_inuse; ++i) { rxr = &sc->hn_rx_ring[i]; stat += *((uint64_t *)((uint8_t *)rxr + ofs)); } error = sysctl_handle_64(oidp, &stat, 0, req); if (error || req->newptr == NULL) return error; /* Zero out this stat. */ for (i = 0; i < sc->hn_rx_ring_inuse; ++i) { rxr = &sc->hn_rx_ring[i]; *((uint64_t *)((uint8_t *)rxr + ofs)) = 0; } return 0; } static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int ofs = arg2, i, error; struct hn_tx_ring *txr; u_long stat; stat = 0; for (i = 0; i < sc->hn_tx_ring_inuse; ++i) { txr = &sc->hn_tx_ring[i]; stat += *((u_long *)((uint8_t *)txr + ofs)); } error = sysctl_handle_long(oidp, &stat, 0, req); if (error || req->newptr == NULL) return error; /* Zero out this stat. */ for (i = 0; i < sc->hn_tx_ring_inuse; ++i) { txr = &sc->hn_tx_ring[i]; *((u_long *)((uint8_t *)txr + ofs)) = 0; } return 0; } static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS) { struct hn_softc *sc = arg1; int ofs = arg2, i, error, conf; struct hn_tx_ring *txr; txr = &sc->hn_tx_ring[0]; conf = *((int *)((uint8_t *)txr + ofs)); error = sysctl_handle_int(oidp, &conf, 0, req); if (error || req->newptr == NULL) return error; NV_LOCK(sc); for (i = 0; i < sc->hn_tx_ring_inuse; ++i) { txr = &sc->hn_tx_ring[i]; *((int *)((uint8_t *)txr + ofs)) = conf; } NV_UNLOCK(sc); return 0; } static int hn_check_iplen(const struct mbuf *m, int hoff) { const struct ip *ip; int len, iphlen, iplen; const struct tcphdr *th; int thoff; /* TCP data offset */ len = hoff + sizeof(struct ip); /* The packet must be at least the size of an IP header. */ if (m->m_pkthdr.len < len) return IPPROTO_DONE; /* The fixed IP header must reside completely in the first mbuf. */ if (m->m_len < len) return IPPROTO_DONE; ip = mtodo(m, hoff); /* Bound check the packet's stated IP header length. */ iphlen = ip->ip_hl << 2; if (iphlen < sizeof(struct ip)) /* minimum header length */ return IPPROTO_DONE; /* The full IP header must reside completely in the one mbuf. */ if (m->m_len < hoff + iphlen) return IPPROTO_DONE; iplen = ntohs(ip->ip_len); /* * Check that the amount of data in the buffers is as * at least much as the IP header would have us expect. */ if (m->m_pkthdr.len < hoff + iplen) return IPPROTO_DONE; /* * Ignore IP fragments. */ if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF)) return IPPROTO_DONE; /* * The TCP/IP or UDP/IP header must be entirely contained within * the first fragment of a packet. */ switch (ip->ip_p) { case IPPROTO_TCP: if (iplen < iphlen + sizeof(struct tcphdr)) return IPPROTO_DONE; if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) return IPPROTO_DONE; th = (const struct tcphdr *)((const uint8_t *)ip + iphlen); thoff = th->th_off << 2; if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen) return IPPROTO_DONE; if (m->m_len < hoff + iphlen + thoff) return IPPROTO_DONE; break; case IPPROTO_UDP: if (iplen < iphlen + sizeof(struct udphdr)) return IPPROTO_DONE; if (m->m_len < hoff + iphlen + sizeof(struct udphdr)) return IPPROTO_DONE; break; default: if (iplen < iphlen) return IPPROTO_DONE; break; } return ip->ip_p; } static void hn_dma_map_paddr(void *arg, bus_dma_segment_t *segs, int nseg, int error) { bus_addr_t *paddr = arg; if (error) return; KASSERT(nseg == 1, ("too many segments %d!", nseg)); *paddr = segs->ds_addr; } static void hn_create_rx_data(struct hn_softc *sc, int ring_cnt) { struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx; device_t dev = sc->hn_dev; #if defined(INET) || defined(INET6) #if __FreeBSD_version >= 1100095 int lroent_cnt; #endif #endif int i; sc->hn_rx_ring_cnt = ring_cnt; sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt; sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt, M_NETVSC, M_WAITOK | M_ZERO); #if defined(INET) || defined(INET6) #if __FreeBSD_version >= 1100095 lroent_cnt = hn_lro_entry_count; if (lroent_cnt < TCP_LRO_ENTRIES) lroent_cnt = TCP_LRO_ENTRIES; device_printf(dev, "LRO: entry count %d\n", lroent_cnt); #endif #endif /* INET || INET6 */ ctx = device_get_sysctl_ctx(dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); /* Create dev.hn.UNIT.rx sysctl tree */ sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx", CTLFLAG_RD, 0, ""); for (i = 0; i < sc->hn_rx_ring_cnt; ++i) { struct hn_rx_ring *rxr = &sc->hn_rx_ring[i]; if (hn_trust_hosttcp) rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP; if (hn_trust_hostudp) rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP; if (hn_trust_hostip) rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP; rxr->hn_ifp = sc->hn_ifp; rxr->hn_rx_idx = i; /* * Initialize LRO. */ #if defined(INET) || defined(INET6) #if __FreeBSD_version >= 1100095 tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt, 0); #else tcp_lro_init(&rxr->hn_lro); rxr->hn_lro.ifp = sc->hn_ifp; #endif #if __FreeBSD_version >= 1100099 rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF; rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF; #endif #endif /* INET || INET6 */ if (sc->hn_rx_sysctl_tree != NULL) { char name[16]; /* * Create per RX ring sysctl tree: * dev.hn.UNIT.rx.RINGID */ snprintf(name, sizeof(name), "%d", i); rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree), OID_AUTO, name, CTLFLAG_RD, 0, ""); if (rxr->hn_rx_sysctl_tree != NULL) { SYSCTL_ADD_ULONG(ctx, SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree), OID_AUTO, "packets", CTLFLAG_RW, &rxr->hn_pkts, "# of packets received"); } } } SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued", CTLTYPE_U64 | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_lro.lro_queued), hn_rx_stat_u64_sysctl, "LU", "LRO queued"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed", CTLTYPE_U64 | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_lro.lro_flushed), hn_rx_stat_u64_sysctl, "LU", "LRO flushed"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_lro_tried), hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries"); #if __FreeBSD_version >= 1100099 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim", CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU", "Max # of data bytes to be aggregated by LRO"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim", CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I", "Max # of ACKs to be aggregated by LRO"); #endif SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp", CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP, hn_trust_hcsum_sysctl, "I", "Trust tcp segement verification on host side, " "when csum info is missing"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp", CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP, hn_trust_hcsum_sysctl, "I", "Trust udp datagram verification on host side, " "when csum info is missing"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip", CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP, hn_trust_hcsum_sysctl, "I", "Trust ip packet verification on host side, " "when csum info is missing"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_csum_ip), hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_csum_tcp), hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_csum_udp), hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_csum_trusted), hn_rx_stat_ulong_sysctl, "LU", "# of packets that we trust host's csum verification"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_rx_ring, hn_small_pkts), hn_rx_stat_ulong_sysctl, "LU", "# of small packets received"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt", CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse", CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings"); } static void hn_destroy_rx_data(struct hn_softc *sc) { #if defined(INET) || defined(INET6) int i; #endif if (sc->hn_rx_ring_cnt == 0) return; #if defined(INET) || defined(INET6) for (i = 0; i < sc->hn_rx_ring_cnt; ++i) tcp_lro_free(&sc->hn_rx_ring[i].hn_lro); #endif free(sc->hn_rx_ring, M_NETVSC); sc->hn_rx_ring = NULL; sc->hn_rx_ring_cnt = 0; sc->hn_rx_ring_inuse = 0; } static int hn_create_tx_ring(struct hn_softc *sc, int id) { struct hn_tx_ring *txr = &sc->hn_tx_ring[id]; bus_dma_tag_t parent_dtag; int error, i; txr->hn_sc = sc; txr->hn_tx_idx = id; #ifndef HN_USE_TXDESC_BUFRING mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN); #endif mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF); txr->hn_txdesc_cnt = HN_TX_DESC_CNT; txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt, M_NETVSC, M_WAITOK | M_ZERO); #ifndef HN_USE_TXDESC_BUFRING SLIST_INIT(&txr->hn_txlist); #else txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC, M_WAITOK, &txr->hn_tx_lock); #endif txr->hn_tx_taskq = sc->hn_tx_taskq; if (hn_use_if_start) { txr->hn_txeof = hn_start_txeof; TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr); TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr); } else { txr->hn_txeof = hn_xmit_txeof; TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr); TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr); txr->hn_mbuf_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC, M_WAITOK, &txr->hn_tx_lock); } txr->hn_direct_tx_size = hn_direct_tx_size; if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1) txr->hn_csum_assist = HN_CSUM_ASSIST; else txr->hn_csum_assist = HN_CSUM_ASSIST_WIN8; /* * Always schedule transmission instead of trying to do direct * transmission. This one gives the best performance so far. */ txr->hn_sched_tx = 1; parent_dtag = bus_get_dma_tag(sc->hn_dev); /* DMA tag for RNDIS messages. */ error = bus_dma_tag_create(parent_dtag, /* parent */ HN_RNDIS_MSG_ALIGN, /* alignment */ HN_RNDIS_MSG_BOUNDARY, /* boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ HN_RNDIS_MSG_LEN, /* maxsize */ 1, /* nsegments */ HN_RNDIS_MSG_LEN, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txr->hn_tx_rndis_dtag); if (error) { device_printf(sc->hn_dev, "failed to create rndis dmatag\n"); return error; } /* DMA tag for data. */ error = bus_dma_tag_create(parent_dtag, /* parent */ 1, /* alignment */ HN_TX_DATA_BOUNDARY, /* boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ HN_TX_DATA_MAXSIZE, /* maxsize */ HN_TX_DATA_SEGCNT_MAX, /* nsegments */ HN_TX_DATA_SEGSIZE, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txr->hn_tx_data_dtag); if (error) { device_printf(sc->hn_dev, "failed to create data dmatag\n"); return error; } for (i = 0; i < txr->hn_txdesc_cnt; ++i) { struct hn_txdesc *txd = &txr->hn_txdesc[i]; txd->txr = txr; /* * Allocate and load RNDIS messages. */ error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag, (void **)&txd->rndis_msg, BUS_DMA_WAITOK | BUS_DMA_COHERENT, &txd->rndis_msg_dmap); if (error) { device_printf(sc->hn_dev, "failed to allocate rndis_msg, %d\n", i); return error; } error = bus_dmamap_load(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap, txd->rndis_msg, HN_RNDIS_MSG_LEN, hn_dma_map_paddr, &txd->rndis_msg_paddr, BUS_DMA_NOWAIT); if (error) { device_printf(sc->hn_dev, "failed to load rndis_msg, %d\n", i); bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg, txd->rndis_msg_dmap); return error; } /* DMA map for TX data. */ error = bus_dmamap_create(txr->hn_tx_data_dtag, 0, &txd->data_dmap); if (error) { device_printf(sc->hn_dev, "failed to allocate tx data dmamap\n"); bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap); bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg, txd->rndis_msg_dmap); return error; } /* All set, put it to list */ txd->flags |= HN_TXD_FLAG_ONLIST; #ifndef HN_USE_TXDESC_BUFRING SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link); #else buf_ring_enqueue(txr->hn_txdesc_br, txd); #endif } txr->hn_txdesc_avail = txr->hn_txdesc_cnt; if (sc->hn_tx_sysctl_tree != NULL) { struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx; char name[16]; /* * Create per TX ring sysctl tree: * dev.hn.UNIT.tx.RINGID */ ctx = device_get_sysctl_ctx(sc->hn_dev); child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree); snprintf(name, sizeof(name), "%d", id); txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, name, CTLFLAG_RD, 0, ""); if (txr->hn_tx_sysctl_tree != NULL) { child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail", CTLFLAG_RD, &txr->hn_txdesc_avail, 0, "# of available TX descs"); if (!hn_use_if_start) { SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive", CTLFLAG_RD, &txr->hn_oactive, 0, "over active"); } SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets", CTLFLAG_RW, &txr->hn_pkts, "# of packets transmitted"); } } return 0; } static void hn_txdesc_dmamap_destroy(struct hn_txdesc *txd) { struct hn_tx_ring *txr = txd->txr; KASSERT(txd->m == NULL, ("still has mbuf installed")); KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped")); bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap); bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg, txd->rndis_msg_dmap); bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap); } static void hn_destroy_tx_ring(struct hn_tx_ring *txr) { struct hn_txdesc *txd; if (txr->hn_txdesc == NULL) return; #ifndef HN_USE_TXDESC_BUFRING while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) { SLIST_REMOVE_HEAD(&txr->hn_txlist, link); hn_txdesc_dmamap_destroy(txd); } #else mtx_lock(&txr->hn_tx_lock); while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL) hn_txdesc_dmamap_destroy(txd); mtx_unlock(&txr->hn_tx_lock); #endif if (txr->hn_tx_data_dtag != NULL) bus_dma_tag_destroy(txr->hn_tx_data_dtag); if (txr->hn_tx_rndis_dtag != NULL) bus_dma_tag_destroy(txr->hn_tx_rndis_dtag); #ifdef HN_USE_TXDESC_BUFRING buf_ring_free(txr->hn_txdesc_br, M_NETVSC); #endif free(txr->hn_txdesc, M_NETVSC); txr->hn_txdesc = NULL; if (txr->hn_mbuf_br != NULL) buf_ring_free(txr->hn_mbuf_br, M_NETVSC); #ifndef HN_USE_TXDESC_BUFRING mtx_destroy(&txr->hn_txlist_spin); #endif mtx_destroy(&txr->hn_tx_lock); } static int hn_create_tx_data(struct hn_softc *sc, int ring_cnt) { struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx; int i; sc->hn_tx_ring_cnt = ring_cnt; sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt; sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt, M_NETVSC, M_WAITOK | M_ZERO); ctx = device_get_sysctl_ctx(sc->hn_dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev)); /* Create dev.hn.UNIT.tx sysctl tree */ sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx", CTLFLAG_RD, 0, ""); for (i = 0; i < sc->hn_tx_ring_cnt; ++i) { int error; error = hn_create_tx_ring(sc, i); if (error) return error; } SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_no_txdescs), hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_send_failed), hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_txdma_failed), hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_tx_collapsed), hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney", CTLTYPE_ULONG | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_tx_chimney), hn_tx_stat_ulong_sysctl, "LU", "# of chimney send"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt", CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0, "# of total TX descs"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max", CTLFLAG_RD, &sc->hn_tx_chimney_max, 0, "Chimney send packet size upper boundary"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size", CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl, "I", "Chimney send packet size limit"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size", CTLTYPE_INT | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_direct_tx_size), hn_tx_conf_int_sysctl, "I", "Size of the packet for direct transmission"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx", CTLTYPE_INT | CTLFLAG_RW, sc, __offsetof(struct hn_tx_ring, hn_sched_tx), hn_tx_conf_int_sysctl, "I", "Always schedule transmission " "instead of doing direct transmission"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt", CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse", CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings"); return 0; } static void hn_set_tx_chimney_size(struct hn_softc *sc, int chimney_size) { int i; NV_LOCK(sc); for (i = 0; i < sc->hn_tx_ring_inuse; ++i) sc->hn_tx_ring[i].hn_tx_chimney_size = chimney_size; NV_UNLOCK(sc); } static void hn_destroy_tx_data(struct hn_softc *sc) { int i; if (sc->hn_tx_ring_cnt == 0) return; for (i = 0; i < sc->hn_tx_ring_cnt; ++i) hn_destroy_tx_ring(&sc->hn_tx_ring[i]); free(sc->hn_tx_ring, M_NETVSC); sc->hn_tx_ring = NULL; sc->hn_tx_ring_cnt = 0; sc->hn_tx_ring_inuse = 0; } static void hn_start_taskfunc(void *xtxr, int pending __unused) { struct hn_tx_ring *txr = xtxr; mtx_lock(&txr->hn_tx_lock); hn_start_locked(txr, 0); mtx_unlock(&txr->hn_tx_lock); } static void hn_start_txeof_taskfunc(void *xtxr, int pending __unused) { struct hn_tx_ring *txr = xtxr; mtx_lock(&txr->hn_tx_lock); atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE); hn_start_locked(txr, 0); mtx_unlock(&txr->hn_tx_lock); } static void hn_stop_tx_tasks(struct hn_softc *sc) { int i; for (i = 0; i < sc->hn_tx_ring_inuse; ++i) { struct hn_tx_ring *txr = &sc->hn_tx_ring[i]; taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task); taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task); } } static int hn_xmit(struct hn_tx_ring *txr, int len) { struct hn_softc *sc = txr->hn_sc; struct ifnet *ifp = sc->hn_ifp; struct mbuf *m_head; mtx_assert(&txr->hn_tx_lock, MA_OWNED); KASSERT(hn_use_if_start == 0, ("hn_xmit is called, when if_start is enabled")); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive) return 0; while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) { struct hn_txdesc *txd; int error; if (len > 0 && m_head->m_pkthdr.len > len) { /* * This sending could be time consuming; let callers * dispatch this packet sending (and sending of any * following up packets) to tx taskqueue. */ drbr_putback(ifp, txr->hn_mbuf_br, m_head); return 1; } txd = hn_txdesc_get(txr); if (txd == NULL) { txr->hn_no_txdescs++; drbr_putback(ifp, txr->hn_mbuf_br, m_head); txr->hn_oactive = 1; break; } error = hn_encap(txr, txd, &m_head); if (error) { /* Both txd and m_head are freed; discard */ drbr_advance(ifp, txr->hn_mbuf_br); continue; } error = hn_send_pkt(ifp, txr, txd); if (__predict_false(error)) { /* txd is freed, but m_head is not */ drbr_putback(ifp, txr->hn_mbuf_br, m_head); txr->hn_oactive = 1; break; } /* Sent */ drbr_advance(ifp, txr->hn_mbuf_br); } return 0; } static int hn_transmit(struct ifnet *ifp, struct mbuf *m) { struct hn_softc *sc = ifp->if_softc; struct hn_tx_ring *txr; int error, idx = 0; /* * Select the TX ring based on flowid */ if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse; txr = &sc->hn_tx_ring[idx]; error = drbr_enqueue(ifp, txr->hn_mbuf_br, m); if (error) return error; if (txr->hn_oactive) return 0; if (txr->hn_sched_tx) goto do_sched; if (mtx_trylock(&txr->hn_tx_lock)) { int sched; sched = hn_xmit(txr, txr->hn_direct_tx_size); mtx_unlock(&txr->hn_tx_lock); if (!sched) return 0; } do_sched: taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task); return 0; } static void hn_xmit_qflush(struct ifnet *ifp) { struct hn_softc *sc = ifp->if_softc; int i; for (i = 0; i < sc->hn_tx_ring_inuse; ++i) { struct hn_tx_ring *txr = &sc->hn_tx_ring[i]; struct mbuf *m; mtx_lock(&txr->hn_tx_lock); while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL) m_freem(m); mtx_unlock(&txr->hn_tx_lock); } if_qflush(ifp); } static void hn_xmit_txeof(struct hn_tx_ring *txr) { if (txr->hn_sched_tx) goto do_sched; if (mtx_trylock(&txr->hn_tx_lock)) { int sched; txr->hn_oactive = 0; sched = hn_xmit(txr, txr->hn_direct_tx_size); mtx_unlock(&txr->hn_tx_lock); if (sched) { taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task); } } else { do_sched: /* * Release the oactive earlier, with the hope, that * others could catch up. The task will clear the * oactive again with the hn_tx_lock to avoid possible * races. */ txr->hn_oactive = 0; taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task); } } static void hn_xmit_taskfunc(void *xtxr, int pending __unused) { struct hn_tx_ring *txr = xtxr; mtx_lock(&txr->hn_tx_lock); hn_xmit(txr, 0); mtx_unlock(&txr->hn_tx_lock); } static void hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused) { struct hn_tx_ring *txr = xtxr; mtx_lock(&txr->hn_tx_lock); txr->hn_oactive = 0; hn_xmit(txr, 0); mtx_unlock(&txr->hn_tx_lock); } static void hn_channel_attach(struct hn_softc *sc, struct hv_vmbus_channel *chan) { struct hn_rx_ring *rxr; int idx; idx = chan->offer_msg.offer.sub_channel_index; KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse, ("invalid channel index %d, should > 0 && < %d", idx, sc->hn_rx_ring_inuse)); rxr = &sc->hn_rx_ring[idx]; KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0, ("RX ring %d already attached", idx)); rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED; chan->hv_chan_rxr = rxr; if_printf(sc->hn_ifp, "link RX ring %d to channel%u\n", idx, chan->offer_msg.child_rel_id); if (idx < sc->hn_tx_ring_inuse) { struct hn_tx_ring *txr = &sc->hn_tx_ring[idx]; KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0, ("TX ring %d already attached", idx)); txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED; chan->hv_chan_txr = txr; txr->hn_chan = chan; if_printf(sc->hn_ifp, "link TX ring %d to channel%u\n", idx, chan->offer_msg.child_rel_id); } /* Bind channel to a proper CPU */ vmbus_channel_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus); } void netvsc_subchan_callback(struct hn_softc *sc, struct hv_vmbus_channel *chan) { KASSERT(!HV_VMBUS_CHAN_ISPRIMARY(chan), ("subchannel callback on primary channel")); KASSERT(chan->offer_msg.offer.sub_channel_index > 0, ("invalid channel subidx %u", chan->offer_msg.offer.sub_channel_index)); hn_channel_attach(sc, chan); } static void hn_tx_taskq_create(void *arg __unused) { if (!hn_share_tx_taskq) return; hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK, taskqueue_thread_enqueue, &hn_tx_taskq); if (hn_bind_tx_taskq >= 0) { int cpu = hn_bind_tx_taskq; cpuset_t cpu_set; if (cpu > mp_ncpus - 1) cpu = mp_ncpus - 1; CPU_SETOF(cpu, &cpu_set); taskqueue_start_threads_cpuset(&hn_tx_taskq, 1, PI_NET, &cpu_set, "hn tx"); } else { taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx"); } } SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST, hn_tx_taskq_create, NULL); static void hn_tx_taskq_destroy(void *arg __unused) { if (hn_tx_taskq != NULL) taskqueue_free(hn_tx_taskq); } SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST, hn_tx_taskq_destroy, NULL); static device_method_t netvsc_methods[] = { /* Device interface */ DEVMETHOD(device_probe, netvsc_probe), DEVMETHOD(device_attach, netvsc_attach), DEVMETHOD(device_detach, netvsc_detach), DEVMETHOD(device_shutdown, netvsc_shutdown), { 0, 0 } }; static driver_t netvsc_driver = { NETVSC_DEVNAME, netvsc_methods, sizeof(hn_softc_t) }; static devclass_t netvsc_devclass; DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0); MODULE_VERSION(hn, 1); MODULE_DEPEND(hn, vmbus, 1, 1, 1); Index: head/sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c =================================================================== --- head/sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c (revision 297912) +++ head/sys/dev/hyperv/storvsc/hv_storvsc_drv_freebsd.c (revision 297913) @@ -1,2137 +1,2130 @@ /*- * Copyright (c) 2009-2012 Microsoft Corp. * Copyright (c) 2012 NetApp Inc. * Copyright (c) 2012 Citrix 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 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. */ /** * StorVSC driver for Hyper-V. This driver presents a SCSI HBA interface * to the Comman Access Method (CAM) layer. CAM control blocks (CCBs) are * converted into VSCSI protocol messages which are delivered to the parent * partition StorVSP driver over the Hyper-V VMBUS. */ #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 "hv_vstorage.h" #define STORVSC_RINGBUFFER_SIZE (20*PAGE_SIZE) #define STORVSC_MAX_LUNS_PER_TARGET (64) #define STORVSC_MAX_IO_REQUESTS (STORVSC_MAX_LUNS_PER_TARGET * 2) #define BLKVSC_MAX_IDE_DISKS_PER_TARGET (1) #define BLKVSC_MAX_IO_REQUESTS STORVSC_MAX_IO_REQUESTS #define STORVSC_MAX_TARGETS (2) #define STORVSC_WIN7_MAJOR 4 #define STORVSC_WIN7_MINOR 2 #define STORVSC_WIN8_MAJOR 5 #define STORVSC_WIN8_MINOR 1 #define VSTOR_PKT_SIZE (sizeof(struct vstor_packet) - vmscsi_size_delta) #define HV_ALIGN(x, a) roundup2(x, a) struct storvsc_softc; struct hv_sgl_node { LIST_ENTRY(hv_sgl_node) link; struct sglist *sgl_data; }; struct hv_sgl_page_pool{ LIST_HEAD(, hv_sgl_node) in_use_sgl_list; LIST_HEAD(, hv_sgl_node) free_sgl_list; boolean_t is_init; } g_hv_sgl_page_pool; #define STORVSC_MAX_SG_PAGE_CNT STORVSC_MAX_IO_REQUESTS * HV_MAX_MULTIPAGE_BUFFER_COUNT enum storvsc_request_type { WRITE_TYPE, READ_TYPE, UNKNOWN_TYPE }; struct hv_storvsc_request { LIST_ENTRY(hv_storvsc_request) link; struct vstor_packet vstor_packet; hv_vmbus_multipage_buffer data_buf; void *sense_data; uint8_t sense_info_len; uint8_t retries; union ccb *ccb; struct storvsc_softc *softc; struct callout callout; struct sema synch_sema; /*Synchronize the request/response if needed */ struct sglist *bounce_sgl; unsigned int bounce_sgl_count; uint64_t not_aligned_seg_bits; }; struct storvsc_softc { struct hv_device *hs_dev; LIST_HEAD(, hv_storvsc_request) hs_free_list; struct mtx hs_lock; struct storvsc_driver_props *hs_drv_props; int hs_unit; uint32_t hs_frozen; struct cam_sim *hs_sim; struct cam_path *hs_path; uint32_t hs_num_out_reqs; boolean_t hs_destroy; boolean_t hs_drain_notify; boolean_t hs_open_multi_channel; struct sema hs_drain_sema; struct hv_storvsc_request hs_init_req; struct hv_storvsc_request hs_reset_req; }; /** * HyperV storvsc timeout testing cases: * a. IO returned after first timeout; * b. IO returned after second timeout and queue freeze; * c. IO returned while timer handler is running * The first can be tested by "sg_senddiag -vv /dev/daX", * and the second and third can be done by * "sg_wr_mode -v -p 08 -c 0,1a -m 0,ff /dev/daX". */ #define HVS_TIMEOUT_TEST 0 /* * Bus/adapter reset functionality on the Hyper-V host is * buggy and it will be disabled until * it can be further tested. */ #define HVS_HOST_RESET 0 struct storvsc_driver_props { char *drv_name; char *drv_desc; uint8_t drv_max_luns_per_target; uint8_t drv_max_ios_per_target; uint32_t drv_ringbuffer_size; }; enum hv_storage_type { DRIVER_BLKVSC, DRIVER_STORVSC, DRIVER_UNKNOWN }; #define HS_MAX_ADAPTERS 10 #define HV_STORAGE_SUPPORTS_MULTI_CHANNEL 0x1 /* {ba6163d9-04a1-4d29-b605-72e2ffb1dc7f} */ static const hv_guid gStorVscDeviceType={ .data = {0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d, 0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f} }; /* {32412632-86cb-44a2-9b5c-50d1417354f5} */ static const hv_guid gBlkVscDeviceType={ .data = {0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44, 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5} }; static struct storvsc_driver_props g_drv_props_table[] = { {"blkvsc", "Hyper-V IDE Storage Interface", BLKVSC_MAX_IDE_DISKS_PER_TARGET, BLKVSC_MAX_IO_REQUESTS, STORVSC_RINGBUFFER_SIZE}, {"storvsc", "Hyper-V SCSI Storage Interface", STORVSC_MAX_LUNS_PER_TARGET, STORVSC_MAX_IO_REQUESTS, STORVSC_RINGBUFFER_SIZE} }; /* * Sense buffer size changed in win8; have a run-time * variable to track the size we should use. */ static int sense_buffer_size; /* * The size of the vmscsi_request has changed in win8. The * additional size is for the newly added elements in the * structure. These elements are valid only when we are talking * to a win8 host. * Track the correct size we need to apply. */ static int vmscsi_size_delta; static int storvsc_current_major; static int storvsc_current_minor; /* static functions */ static int storvsc_probe(device_t dev); static int storvsc_attach(device_t dev); static int storvsc_detach(device_t dev); static void storvsc_poll(struct cam_sim * sim); static void storvsc_action(struct cam_sim * sim, union ccb * ccb); static int create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp); static void storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp); static enum hv_storage_type storvsc_get_storage_type(device_t dev); static void hv_storvsc_rescan_target(struct storvsc_softc *sc); static void hv_storvsc_on_channel_callback(void *context); static void hv_storvsc_on_iocompletion( struct storvsc_softc *sc, struct vstor_packet *vstor_packet, struct hv_storvsc_request *request); static int hv_storvsc_connect_vsp(struct hv_device *device); static void storvsc_io_done(struct hv_storvsc_request *reqp); static void storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl, bus_dma_segment_t *orig_sgl, unsigned int orig_sgl_count, uint64_t seg_bits); void storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl, unsigned int dest_sgl_count, struct sglist* src_sgl, uint64_t seg_bits); static device_method_t storvsc_methods[] = { /* Device interface */ DEVMETHOD(device_probe, storvsc_probe), DEVMETHOD(device_attach, storvsc_attach), DEVMETHOD(device_detach, storvsc_detach), DEVMETHOD(device_shutdown, bus_generic_shutdown), DEVMETHOD_END }; static driver_t storvsc_driver = { "storvsc", storvsc_methods, sizeof(struct storvsc_softc), }; static devclass_t storvsc_devclass; DRIVER_MODULE(storvsc, vmbus, storvsc_driver, storvsc_devclass, 0, 0); MODULE_VERSION(storvsc, 1); MODULE_DEPEND(storvsc, vmbus, 1, 1, 1); /** * The host is capable of sending messages to us that are * completely unsolicited. So, we need to address the race * condition where we may be in the process of unloading the * driver when the host may send us an unsolicited message. * We address this issue by implementing a sequentially * consistent protocol: * * 1. Channel callback is invoked while holding the the channel lock * and an unloading driver will reset the channel callback under * the protection of this channel lock. * * 2. To ensure bounded wait time for unloading a driver, we don't * permit outgoing traffic once the device is marked as being * destroyed. * * 3. Once the device is marked as being destroyed, we only * permit incoming traffic to properly account for * packets already sent out. */ static inline struct storvsc_softc * get_stor_device(struct hv_device *device, boolean_t outbound) { struct storvsc_softc *sc; sc = device_get_softc(device->device); - if (sc == NULL) { - return NULL; - } if (outbound) { /* * Here we permit outgoing I/O only * if the device is not being destroyed. */ if (sc->hs_destroy) { sc = NULL; } } else { /* * inbound case; if being destroyed * only permit to account for * messages already sent out. */ if (sc->hs_destroy && (sc->hs_num_out_reqs == 0)) { sc = NULL; } } return sc; } /** * @brief Callback handler, will be invoked when receive mutil-channel offer * * @param context new multi-channel */ static void storvsc_handle_sc_creation(void *context) { hv_vmbus_channel *new_channel; struct hv_device *device; struct storvsc_softc *sc; struct vmstor_chan_props props; int ret = 0; new_channel = (hv_vmbus_channel *)context; device = new_channel->device; sc = get_stor_device(device, TRUE); if (sc == NULL) return; if (FALSE == sc->hs_open_multi_channel) return; memset(&props, 0, sizeof(props)); ret = hv_vmbus_channel_open(new_channel, sc->hs_drv_props->drv_ringbuffer_size, sc->hs_drv_props->drv_ringbuffer_size, (void *)&props, sizeof(struct vmstor_chan_props), hv_storvsc_on_channel_callback, new_channel); return; } /** * @brief Send multi-channel creation request to host * * @param device a Hyper-V device pointer * @param max_chans the max channels supported by vmbus */ static void storvsc_send_multichannel_request(struct hv_device *dev, int max_chans) { struct storvsc_softc *sc; struct hv_storvsc_request *request; struct vstor_packet *vstor_packet; int request_channels_cnt = 0; int ret; /* get multichannels count that need to create */ request_channels_cnt = MIN(max_chans, mp_ncpus); sc = get_stor_device(dev, TRUE); if (sc == NULL) { printf("Storvsc_error: get sc failed while send mutilchannel " "request\n"); return; } request = &sc->hs_init_req; /* Establish a handler for multi-channel */ dev->channel->sc_creation_callback = storvsc_handle_sc_creation; /* request the host to create multi-channel */ memset(request, 0, sizeof(struct hv_storvsc_request)); sema_init(&request->synch_sema, 0, ("stor_synch_sema")); vstor_packet = &request->vstor_packet; vstor_packet->operation = VSTOR_OPERATION_CREATE_MULTI_CHANNELS; vstor_packet->flags = REQUEST_COMPLETION_FLAG; vstor_packet->u.multi_channels_cnt = request_channels_cnt; ret = hv_vmbus_channel_send_packet( dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); /* wait for 5 seconds */ ret = sema_timedwait(&request->synch_sema, 5 * hz); if (ret != 0) { printf("Storvsc_error: create multi-channel timeout, %d\n", ret); return; } if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO || vstor_packet->status != 0) { printf("Storvsc_error: create multi-channel invalid operation " "(%d) or statue (%u)\n", vstor_packet->operation, vstor_packet->status); return; } sc->hs_open_multi_channel = TRUE; if (bootverbose) printf("Storvsc create multi-channel success!\n"); } /** * @brief initialize channel connection to parent partition * * @param dev a Hyper-V device pointer * @returns 0 on success, non-zero error on failure */ static int hv_storvsc_channel_init(struct hv_device *dev) { int ret = 0; struct hv_storvsc_request *request; struct vstor_packet *vstor_packet; struct storvsc_softc *sc; uint16_t max_chans = 0; boolean_t support_multichannel = FALSE; max_chans = 0; support_multichannel = FALSE; sc = get_stor_device(dev, TRUE); if (sc == NULL) return (ENODEV); request = &sc->hs_init_req; memset(request, 0, sizeof(struct hv_storvsc_request)); vstor_packet = &request->vstor_packet; request->softc = sc; /** * Initiate the vsc/vsp initialization protocol on the open channel */ sema_init(&request->synch_sema, 0, ("stor_synch_sema")); vstor_packet->operation = VSTOR_OPERATION_BEGININITIALIZATION; vstor_packet->flags = REQUEST_COMPLETION_FLAG; ret = hv_vmbus_channel_send_packet( dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) goto cleanup; /* wait 5 seconds */ ret = sema_timedwait(&request->synch_sema, 5 * hz); if (ret != 0) goto cleanup; if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO || vstor_packet->status != 0) { goto cleanup; } /* reuse the packet for version range supported */ memset(vstor_packet, 0, sizeof(struct vstor_packet)); vstor_packet->operation = VSTOR_OPERATION_QUERYPROTOCOLVERSION; vstor_packet->flags = REQUEST_COMPLETION_FLAG; vstor_packet->u.version.major_minor = VMSTOR_PROTOCOL_VERSION(storvsc_current_major, storvsc_current_minor); /* revision is only significant for Windows guests */ vstor_packet->u.version.revision = 0; ret = hv_vmbus_channel_send_packet( dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) goto cleanup; /* wait 5 seconds */ ret = sema_timedwait(&request->synch_sema, 5 * hz); if (ret) goto cleanup; /* TODO: Check returned version */ if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO || vstor_packet->status != 0) goto cleanup; /** * Query channel properties */ memset(vstor_packet, 0, sizeof(struct vstor_packet)); vstor_packet->operation = VSTOR_OPERATION_QUERYPROPERTIES; vstor_packet->flags = REQUEST_COMPLETION_FLAG; ret = hv_vmbus_channel_send_packet( dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if ( ret != 0) goto cleanup; /* wait 5 seconds */ ret = sema_timedwait(&request->synch_sema, 5 * hz); if (ret != 0) goto cleanup; /* TODO: Check returned version */ if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO || vstor_packet->status != 0) { goto cleanup; } /* multi-channels feature is supported by WIN8 and above version */ max_chans = vstor_packet->u.chan_props.max_channel_cnt; if ((hv_vmbus_protocal_version != HV_VMBUS_VERSION_WIN7) && (hv_vmbus_protocal_version != HV_VMBUS_VERSION_WS2008) && (vstor_packet->u.chan_props.flags & HV_STORAGE_SUPPORTS_MULTI_CHANNEL)) { support_multichannel = TRUE; } memset(vstor_packet, 0, sizeof(struct vstor_packet)); vstor_packet->operation = VSTOR_OPERATION_ENDINITIALIZATION; vstor_packet->flags = REQUEST_COMPLETION_FLAG; ret = hv_vmbus_channel_send_packet( dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) { goto cleanup; } /* wait 5 seconds */ ret = sema_timedwait(&request->synch_sema, 5 * hz); if (ret != 0) goto cleanup; if (vstor_packet->operation != VSTOR_OPERATION_COMPLETEIO || vstor_packet->status != 0) goto cleanup; /* * If multi-channel is supported, send multichannel create * request to host. */ if (support_multichannel) storvsc_send_multichannel_request(dev, max_chans); cleanup: sema_destroy(&request->synch_sema); return (ret); } /** * @brief Open channel connection to paraent partition StorVSP driver * * Open and initialize channel connection to parent partition StorVSP driver. * * @param pointer to a Hyper-V device * @returns 0 on success, non-zero error on failure */ static int hv_storvsc_connect_vsp(struct hv_device *dev) { int ret = 0; struct vmstor_chan_props props; struct storvsc_softc *sc; sc = device_get_softc(dev->device); memset(&props, 0, sizeof(struct vmstor_chan_props)); /* * Open the channel */ ret = hv_vmbus_channel_open( dev->channel, sc->hs_drv_props->drv_ringbuffer_size, sc->hs_drv_props->drv_ringbuffer_size, (void *)&props, sizeof(struct vmstor_chan_props), hv_storvsc_on_channel_callback, dev->channel); if (ret != 0) { return ret; } ret = hv_storvsc_channel_init(dev); return (ret); } #if HVS_HOST_RESET static int hv_storvsc_host_reset(struct hv_device *dev) { int ret = 0; struct storvsc_softc *sc; struct hv_storvsc_request *request; struct vstor_packet *vstor_packet; sc = get_stor_device(dev, TRUE); if (sc == NULL) { return ENODEV; } request = &sc->hs_reset_req; request->softc = sc; vstor_packet = &request->vstor_packet; sema_init(&request->synch_sema, 0, "stor synch sema"); vstor_packet->operation = VSTOR_OPERATION_RESETBUS; vstor_packet->flags = REQUEST_COMPLETION_FLAG; ret = hv_vmbus_channel_send_packet(dev->channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)&sc->hs_reset_req, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); if (ret != 0) { goto cleanup; } ret = sema_timedwait(&request->synch_sema, 5 * hz); /* KYS 5 seconds */ if (ret) { goto cleanup; } /* * At this point, all outstanding requests in the adapter * should have been flushed out and return to us */ cleanup: sema_destroy(&request->synch_sema); return (ret); } #endif /* HVS_HOST_RESET */ /** * @brief Function to initiate an I/O request * * @param device Hyper-V device pointer * @param request pointer to a request structure * @returns 0 on success, non-zero error on failure */ static int hv_storvsc_io_request(struct hv_device *device, struct hv_storvsc_request *request) { struct storvsc_softc *sc; struct vstor_packet *vstor_packet = &request->vstor_packet; struct hv_vmbus_channel* outgoing_channel = NULL; int ret = 0; sc = get_stor_device(device, TRUE); if (sc == NULL) { return ENODEV; } vstor_packet->flags |= REQUEST_COMPLETION_FLAG; vstor_packet->u.vm_srb.length = VSTOR_PKT_SIZE; vstor_packet->u.vm_srb.sense_info_len = sense_buffer_size; vstor_packet->u.vm_srb.transfer_len = request->data_buf.length; vstor_packet->operation = VSTOR_OPERATION_EXECUTESRB; outgoing_channel = vmbus_select_outgoing_channel(device->channel); mtx_unlock(&request->softc->hs_lock); if (request->data_buf.length) { ret = hv_vmbus_channel_send_packet_multipagebuffer( outgoing_channel, &request->data_buf, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request); } else { ret = hv_vmbus_channel_send_packet( outgoing_channel, vstor_packet, VSTOR_PKT_SIZE, (uint64_t)(uintptr_t)request, HV_VMBUS_PACKET_TYPE_DATA_IN_BAND, HV_VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); } mtx_lock(&request->softc->hs_lock); if (ret != 0) { printf("Unable to send packet %p ret %d", vstor_packet, ret); } else { atomic_add_int(&sc->hs_num_out_reqs, 1); } return (ret); } /** * Process IO_COMPLETION_OPERATION and ready * the result to be completed for upper layer * processing by the CAM layer. */ static void hv_storvsc_on_iocompletion(struct storvsc_softc *sc, struct vstor_packet *vstor_packet, struct hv_storvsc_request *request) { struct vmscsi_req *vm_srb; vm_srb = &vstor_packet->u.vm_srb; if (((vm_srb->scsi_status & 0xFF) == SCSI_STATUS_CHECK_COND) && (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID)) { /* Autosense data available */ KASSERT(vm_srb->sense_info_len <= request->sense_info_len, ("vm_srb->sense_info_len <= " "request->sense_info_len")); memcpy(request->sense_data, vm_srb->u.sense_data, vm_srb->sense_info_len); request->sense_info_len = vm_srb->sense_info_len; } /* Complete request by passing to the CAM layer */ storvsc_io_done(request); atomic_subtract_int(&sc->hs_num_out_reqs, 1); if (sc->hs_drain_notify && (sc->hs_num_out_reqs == 0)) { sema_post(&sc->hs_drain_sema); } } static void hv_storvsc_rescan_target(struct storvsc_softc *sc) { path_id_t pathid; target_id_t targetid; union ccb *ccb; pathid = cam_sim_path(sc->hs_sim); targetid = CAM_TARGET_WILDCARD; /* * Allocate a CCB and schedule a rescan. */ ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { printf("unable to alloc CCB for rescan\n"); return; } if (xpt_create_path(&ccb->ccb_h.path, NULL, pathid, targetid, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { printf("unable to create path for rescan, pathid: %u," "targetid: %u\n", pathid, targetid); xpt_free_ccb(ccb); return; } if (targetid == CAM_TARGET_WILDCARD) ccb->ccb_h.func_code = XPT_SCAN_BUS; else ccb->ccb_h.func_code = XPT_SCAN_TGT; xpt_rescan(ccb); } static void hv_storvsc_on_channel_callback(void *context) { int ret = 0; hv_vmbus_channel *channel = (hv_vmbus_channel *)context; struct hv_device *device = NULL; struct storvsc_softc *sc; uint32_t bytes_recvd; uint64_t request_id; uint8_t packet[roundup2(sizeof(struct vstor_packet), 8)]; struct hv_storvsc_request *request; struct vstor_packet *vstor_packet; device = channel->device; KASSERT(device, ("device is NULL")); sc = get_stor_device(device, FALSE); if (sc == NULL) { printf("Storvsc_error: get stor device failed.\n"); return; } ret = hv_vmbus_channel_recv_packet( channel, packet, roundup2(VSTOR_PKT_SIZE, 8), &bytes_recvd, &request_id); while ((ret == 0) && (bytes_recvd > 0)) { request = (struct hv_storvsc_request *)(uintptr_t)request_id; if ((request == &sc->hs_init_req) || (request == &sc->hs_reset_req)) { memcpy(&request->vstor_packet, packet, sizeof(struct vstor_packet)); sema_post(&request->synch_sema); } else { vstor_packet = (struct vstor_packet *)packet; switch(vstor_packet->operation) { case VSTOR_OPERATION_COMPLETEIO: if (request == NULL) panic("VMBUS: storvsc received a " "packet with NULL request id in " "COMPLETEIO operation."); hv_storvsc_on_iocompletion(sc, vstor_packet, request); break; case VSTOR_OPERATION_REMOVEDEVICE: printf("VMBUS: storvsc operation %d not " "implemented.\n", vstor_packet->operation); /* TODO: implement */ break; case VSTOR_OPERATION_ENUMERATE_BUS: hv_storvsc_rescan_target(sc); break; default: break; } } ret = hv_vmbus_channel_recv_packet( channel, packet, roundup2(VSTOR_PKT_SIZE, 8), &bytes_recvd, &request_id); } } /** * @brief StorVSC probe function * * Device probe function. Returns 0 if the input device is a StorVSC * device. Otherwise, a ENXIO is returned. If the input device is * for BlkVSC (paravirtual IDE) device and this support is disabled in * favor of the emulated ATA/IDE device, return ENXIO. * * @param a device * @returns 0 on success, ENXIO if not a matcing StorVSC device */ static int storvsc_probe(device_t dev) { int ata_disk_enable = 0; int ret = ENXIO; if (hv_vmbus_protocal_version == HV_VMBUS_VERSION_WS2008 || hv_vmbus_protocal_version == HV_VMBUS_VERSION_WIN7) { sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE; vmscsi_size_delta = sizeof(struct vmscsi_win8_extension); storvsc_current_major = STORVSC_WIN7_MAJOR; storvsc_current_minor = STORVSC_WIN7_MINOR; } else { sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE; vmscsi_size_delta = 0; storvsc_current_major = STORVSC_WIN8_MAJOR; storvsc_current_minor = STORVSC_WIN8_MINOR; } switch (storvsc_get_storage_type(dev)) { case DRIVER_BLKVSC: if(bootverbose) device_printf(dev, "DRIVER_BLKVSC-Emulated ATA/IDE probe\n"); if (!getenv_int("hw.ata.disk_enable", &ata_disk_enable)) { if(bootverbose) device_printf(dev, "Enlightened ATA/IDE detected\n"); ret = BUS_PROBE_DEFAULT; } else if(bootverbose) device_printf(dev, "Emulated ATA/IDE set (hw.ata.disk_enable set)\n"); break; case DRIVER_STORVSC: if(bootverbose) device_printf(dev, "Enlightened SCSI device detected\n"); ret = BUS_PROBE_DEFAULT; break; default: ret = ENXIO; } return (ret); } /** * @brief StorVSC attach function * * Function responsible for allocating per-device structures, * setting up CAM interfaces and scanning for available LUNs to * be used for SCSI device peripherals. * * @param a device * @returns 0 on success or an error on failure */ static int storvsc_attach(device_t dev) { struct hv_device *hv_dev = vmbus_get_devctx(dev); enum hv_storage_type stor_type; struct storvsc_softc *sc; struct cam_devq *devq; int ret, i, j; struct hv_storvsc_request *reqp; struct root_hold_token *root_mount_token = NULL; struct hv_sgl_node *sgl_node = NULL; void *tmp_buff = NULL; /* * We need to serialize storvsc attach calls. */ root_mount_token = root_mount_hold("storvsc"); sc = device_get_softc(dev); - if (sc == NULL) { - ret = ENOMEM; - goto cleanup; - } stor_type = storvsc_get_storage_type(dev); if (stor_type == DRIVER_UNKNOWN) { ret = ENODEV; goto cleanup; } bzero(sc, sizeof(struct storvsc_softc)); /* fill in driver specific properties */ sc->hs_drv_props = &g_drv_props_table[stor_type]; /* fill in device specific properties */ sc->hs_unit = device_get_unit(dev); sc->hs_dev = hv_dev; device_set_desc(dev, g_drv_props_table[stor_type].drv_desc); LIST_INIT(&sc->hs_free_list); mtx_init(&sc->hs_lock, "hvslck", NULL, MTX_DEF); for (i = 0; i < sc->hs_drv_props->drv_max_ios_per_target; ++i) { reqp = malloc(sizeof(struct hv_storvsc_request), M_DEVBUF, M_WAITOK|M_ZERO); reqp->softc = sc; LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link); } /* create sg-list page pool */ if (FALSE == g_hv_sgl_page_pool.is_init) { g_hv_sgl_page_pool.is_init = TRUE; LIST_INIT(&g_hv_sgl_page_pool.in_use_sgl_list); LIST_INIT(&g_hv_sgl_page_pool.free_sgl_list); /* * Pre-create SG list, each SG list with * HV_MAX_MULTIPAGE_BUFFER_COUNT segments, each * segment has one page buffer */ for (i = 0; i < STORVSC_MAX_IO_REQUESTS; i++) { sgl_node = malloc(sizeof(struct hv_sgl_node), M_DEVBUF, M_WAITOK|M_ZERO); sgl_node->sgl_data = sglist_alloc(HV_MAX_MULTIPAGE_BUFFER_COUNT, M_WAITOK|M_ZERO); for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) { tmp_buff = malloc(PAGE_SIZE, M_DEVBUF, M_WAITOK|M_ZERO); sgl_node->sgl_data->sg_segs[j].ss_paddr = (vm_paddr_t)tmp_buff; } LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list, sgl_node, link); } } sc->hs_destroy = FALSE; sc->hs_drain_notify = FALSE; sc->hs_open_multi_channel = FALSE; sema_init(&sc->hs_drain_sema, 0, "Store Drain Sema"); ret = hv_storvsc_connect_vsp(hv_dev); if (ret != 0) { goto cleanup; } /* * Create the device queue. * Hyper-V maps each target to one SCSI HBA */ devq = cam_simq_alloc(sc->hs_drv_props->drv_max_ios_per_target); if (devq == NULL) { device_printf(dev, "Failed to alloc device queue\n"); ret = ENOMEM; goto cleanup; } sc->hs_sim = cam_sim_alloc(storvsc_action, storvsc_poll, sc->hs_drv_props->drv_name, sc, sc->hs_unit, &sc->hs_lock, 1, sc->hs_drv_props->drv_max_ios_per_target, devq); if (sc->hs_sim == NULL) { device_printf(dev, "Failed to alloc sim\n"); cam_simq_free(devq); ret = ENOMEM; goto cleanup; } mtx_lock(&sc->hs_lock); /* bus_id is set to 0, need to get it from VMBUS channel query? */ if (xpt_bus_register(sc->hs_sim, dev, 0) != CAM_SUCCESS) { cam_sim_free(sc->hs_sim, /*free_devq*/TRUE); mtx_unlock(&sc->hs_lock); device_printf(dev, "Unable to register SCSI bus\n"); ret = ENXIO; goto cleanup; } if (xpt_create_path(&sc->hs_path, /*periph*/NULL, cam_sim_path(sc->hs_sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { xpt_bus_deregister(cam_sim_path(sc->hs_sim)); cam_sim_free(sc->hs_sim, /*free_devq*/TRUE); mtx_unlock(&sc->hs_lock); device_printf(dev, "Unable to create path\n"); ret = ENXIO; goto cleanup; } mtx_unlock(&sc->hs_lock); root_mount_rel(root_mount_token); return (0); cleanup: root_mount_rel(root_mount_token); while (!LIST_EMPTY(&sc->hs_free_list)) { reqp = LIST_FIRST(&sc->hs_free_list); LIST_REMOVE(reqp, link); free(reqp, M_DEVBUF); } while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) { sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list); LIST_REMOVE(sgl_node, link); for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++) { if (NULL != (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) { free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF); } } sglist_free(sgl_node->sgl_data); free(sgl_node, M_DEVBUF); } return (ret); } /** * @brief StorVSC device detach function * * This function is responsible for safely detaching a * StorVSC device. This includes waiting for inbound responses * to complete and freeing associated per-device structures. * * @param dev a device * returns 0 on success */ static int storvsc_detach(device_t dev) { struct storvsc_softc *sc = device_get_softc(dev); struct hv_storvsc_request *reqp = NULL; struct hv_device *hv_device = vmbus_get_devctx(dev); struct hv_sgl_node *sgl_node = NULL; int j = 0; sc->hs_destroy = TRUE; /* * At this point, all outbound traffic should be disabled. We * only allow inbound traffic (responses) to proceed so that * outstanding requests can be completed. */ sc->hs_drain_notify = TRUE; sema_wait(&sc->hs_drain_sema); sc->hs_drain_notify = FALSE; /* * Since we have already drained, we don't need to busy wait. * The call to close the channel will reset the callback * under the protection of the incoming channel lock. */ hv_vmbus_channel_close(hv_device->channel); mtx_lock(&sc->hs_lock); while (!LIST_EMPTY(&sc->hs_free_list)) { reqp = LIST_FIRST(&sc->hs_free_list); LIST_REMOVE(reqp, link); free(reqp, M_DEVBUF); } mtx_unlock(&sc->hs_lock); while (!LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) { sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list); LIST_REMOVE(sgl_node, link); for (j = 0; j < HV_MAX_MULTIPAGE_BUFFER_COUNT; j++){ if (NULL != (void*)sgl_node->sgl_data->sg_segs[j].ss_paddr) { free((void*)sgl_node->sgl_data->sg_segs[j].ss_paddr, M_DEVBUF); } } sglist_free(sgl_node->sgl_data); free(sgl_node, M_DEVBUF); } return (0); } #if HVS_TIMEOUT_TEST /** * @brief unit test for timed out operations * * This function provides unit testing capability to simulate * timed out operations. Recompilation with HV_TIMEOUT_TEST=1 * is required. * * @param reqp pointer to a request structure * @param opcode SCSI operation being performed * @param wait if 1, wait for I/O to complete */ static void storvsc_timeout_test(struct hv_storvsc_request *reqp, uint8_t opcode, int wait) { int ret; union ccb *ccb = reqp->ccb; struct storvsc_softc *sc = reqp->softc; if (reqp->vstor_packet.vm_srb.cdb[0] != opcode) { return; } if (wait) { mtx_lock(&reqp->event.mtx); } ret = hv_storvsc_io_request(sc->hs_dev, reqp); if (ret != 0) { if (wait) { mtx_unlock(&reqp->event.mtx); } printf("%s: io_request failed with %d.\n", __func__, ret); ccb->ccb_h.status = CAM_PROVIDE_FAIL; mtx_lock(&sc->hs_lock); storvsc_free_request(sc, reqp); xpt_done(ccb); mtx_unlock(&sc->hs_lock); return; } if (wait) { xpt_print(ccb->ccb_h.path, "%u: %s: waiting for IO return.\n", ticks, __func__); ret = cv_timedwait(&reqp->event.cv, &reqp->event.mtx, 60*hz); mtx_unlock(&reqp->event.mtx); xpt_print(ccb->ccb_h.path, "%u: %s: %s.\n", ticks, __func__, (ret == 0)? "IO return detected" : "IO return not detected"); /* * Now both the timer handler and io done are running * simultaneously. We want to confirm the io done always * finishes after the timer handler exits. So reqp used by * timer handler is not freed or stale. Do busy loop for * another 1/10 second to make sure io done does * wait for the timer handler to complete. */ DELAY(100*1000); mtx_lock(&sc->hs_lock); xpt_print(ccb->ccb_h.path, "%u: %s: finishing, queue frozen %d, " "ccb status 0x%x scsi_status 0x%x.\n", ticks, __func__, sc->hs_frozen, ccb->ccb_h.status, ccb->csio.scsi_status); mtx_unlock(&sc->hs_lock); } } #endif /* HVS_TIMEOUT_TEST */ /** * @brief timeout handler for requests * * This function is called as a result of a callout expiring. * * @param arg pointer to a request */ static void storvsc_timeout(void *arg) { struct hv_storvsc_request *reqp = arg; struct storvsc_softc *sc = reqp->softc; union ccb *ccb = reqp->ccb; if (reqp->retries == 0) { mtx_lock(&sc->hs_lock); xpt_print(ccb->ccb_h.path, "%u: IO timed out (req=0x%p), wait for another %u secs.\n", ticks, reqp, ccb->ccb_h.timeout / 1000); cam_error_print(ccb, CAM_ESF_ALL, CAM_EPF_ALL); mtx_unlock(&sc->hs_lock); reqp->retries++; callout_reset_sbt(&reqp->callout, SBT_1MS * ccb->ccb_h.timeout, 0, storvsc_timeout, reqp, 0); #if HVS_TIMEOUT_TEST storvsc_timeout_test(reqp, SEND_DIAGNOSTIC, 0); #endif return; } mtx_lock(&sc->hs_lock); xpt_print(ccb->ccb_h.path, "%u: IO (reqp = 0x%p) did not return for %u seconds, %s.\n", ticks, reqp, ccb->ccb_h.timeout * (reqp->retries+1) / 1000, (sc->hs_frozen == 0)? "freezing the queue" : "the queue is already frozen"); if (sc->hs_frozen == 0) { sc->hs_frozen = 1; xpt_freeze_simq(xpt_path_sim(ccb->ccb_h.path), 1); } mtx_unlock(&sc->hs_lock); #if HVS_TIMEOUT_TEST storvsc_timeout_test(reqp, MODE_SELECT_10, 1); #endif } /** * @brief StorVSC device poll function * * This function is responsible for servicing requests when * interrupts are disabled (i.e when we are dumping core.) * * @param sim a pointer to a CAM SCSI interface module */ static void storvsc_poll(struct cam_sim *sim) { struct storvsc_softc *sc = cam_sim_softc(sim); mtx_assert(&sc->hs_lock, MA_OWNED); mtx_unlock(&sc->hs_lock); hv_storvsc_on_channel_callback(sc->hs_dev->channel); mtx_lock(&sc->hs_lock); } /** * @brief StorVSC device action function * * This function is responsible for handling SCSI operations which * are passed from the CAM layer. The requests are in the form of * CAM control blocks which indicate the action being performed. * Not all actions require converting the request to a VSCSI protocol * message - these actions can be responded to by this driver. * Requests which are destined for a backend storage device are converted * to a VSCSI protocol message and sent on the channel connection associated * with this device. * * @param sim pointer to a CAM SCSI interface module * @param ccb pointer to a CAM control block */ static void storvsc_action(struct cam_sim *sim, union ccb *ccb) { struct storvsc_softc *sc = cam_sim_softc(sim); int res; mtx_assert(&sc->hs_lock, MA_OWNED); switch (ccb->ccb_h.func_code) { case XPT_PATH_INQ: { struct ccb_pathinq *cpi = &ccb->cpi; cpi->version_num = 1; cpi->hba_inquiry = PI_TAG_ABLE|PI_SDTR_ABLE; cpi->target_sprt = 0; cpi->hba_misc = PIM_NOBUSRESET; cpi->hba_eng_cnt = 0; cpi->max_target = STORVSC_MAX_TARGETS; cpi->max_lun = sc->hs_drv_props->drv_max_luns_per_target; cpi->initiator_id = cpi->max_target; cpi->bus_id = cam_sim_bus(sim); cpi->base_transfer_speed = 300000; cpi->transport = XPORT_SAS; cpi->transport_version = 0; cpi->protocol = PROTO_SCSI; cpi->protocol_version = SCSI_REV_SPC2; strncpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN); strncpy(cpi->hba_vid, sc->hs_drv_props->drv_name, HBA_IDLEN); strncpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN); cpi->unit_number = cam_sim_unit(sim); ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); return; } case XPT_GET_TRAN_SETTINGS: { struct ccb_trans_settings *cts = &ccb->cts; cts->transport = XPORT_SAS; cts->transport_version = 0; cts->protocol = PROTO_SCSI; cts->protocol_version = SCSI_REV_SPC2; /* enable tag queuing and disconnected mode */ cts->proto_specific.valid = CTS_SCSI_VALID_TQ; cts->proto_specific.scsi.valid = CTS_SCSI_VALID_TQ; cts->proto_specific.scsi.flags = CTS_SCSI_FLAGS_TAG_ENB; cts->xport_specific.valid = CTS_SPI_VALID_DISC; cts->xport_specific.spi.flags = CTS_SPI_FLAGS_DISC_ENB; ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); return; } case XPT_SET_TRAN_SETTINGS: { ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); return; } case XPT_CALC_GEOMETRY:{ cam_calc_geometry(&ccb->ccg, 1); xpt_done(ccb); return; } case XPT_RESET_BUS: case XPT_RESET_DEV:{ #if HVS_HOST_RESET if ((res = hv_storvsc_host_reset(sc->hs_dev)) != 0) { xpt_print(ccb->ccb_h.path, "hv_storvsc_host_reset failed with %d\n", res); ccb->ccb_h.status = CAM_PROVIDE_FAIL; xpt_done(ccb); return; } ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); return; #else xpt_print(ccb->ccb_h.path, "%s reset not supported.\n", (ccb->ccb_h.func_code == XPT_RESET_BUS)? "bus" : "dev"); ccb->ccb_h.status = CAM_REQ_INVALID; xpt_done(ccb); return; #endif /* HVS_HOST_RESET */ } case XPT_SCSI_IO: case XPT_IMMED_NOTIFY: { struct hv_storvsc_request *reqp = NULL; if (ccb->csio.cdb_len == 0) { panic("cdl_len is 0\n"); } if (LIST_EMPTY(&sc->hs_free_list)) { ccb->ccb_h.status = CAM_REQUEUE_REQ; if (sc->hs_frozen == 0) { sc->hs_frozen = 1; xpt_freeze_simq(sim, /* count*/1); } xpt_done(ccb); return; } reqp = LIST_FIRST(&sc->hs_free_list); LIST_REMOVE(reqp, link); bzero(reqp, sizeof(struct hv_storvsc_request)); reqp->softc = sc; ccb->ccb_h.status |= CAM_SIM_QUEUED; if ((res = create_storvsc_request(ccb, reqp)) != 0) { ccb->ccb_h.status = CAM_REQ_INVALID; xpt_done(ccb); return; } if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) { callout_init(&reqp->callout, 1); callout_reset_sbt(&reqp->callout, SBT_1MS * ccb->ccb_h.timeout, 0, storvsc_timeout, reqp, 0); #if HVS_TIMEOUT_TEST cv_init(&reqp->event.cv, "storvsc timeout cv"); mtx_init(&reqp->event.mtx, "storvsc timeout mutex", NULL, MTX_DEF); switch (reqp->vstor_packet.vm_srb.cdb[0]) { case MODE_SELECT_10: case SEND_DIAGNOSTIC: /* To have timer send the request. */ return; default: break; } #endif /* HVS_TIMEOUT_TEST */ } if ((res = hv_storvsc_io_request(sc->hs_dev, reqp)) != 0) { xpt_print(ccb->ccb_h.path, "hv_storvsc_io_request failed with %d\n", res); ccb->ccb_h.status = CAM_PROVIDE_FAIL; storvsc_free_request(sc, reqp); xpt_done(ccb); return; } return; } default: ccb->ccb_h.status = CAM_REQ_INVALID; xpt_done(ccb); return; } } /** * @brief destroy bounce buffer * * This function is responsible for destroy a Scatter/Gather list * that create by storvsc_create_bounce_buffer() * * @param sgl- the Scatter/Gather need be destroy * @param sg_count- page count of the SG list. * */ static void storvsc_destroy_bounce_buffer(struct sglist *sgl) { struct hv_sgl_node *sgl_node = NULL; if (LIST_EMPTY(&g_hv_sgl_page_pool.in_use_sgl_list)) { printf("storvsc error: not enough in use sgl\n"); return; } sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.in_use_sgl_list); LIST_REMOVE(sgl_node, link); sgl_node->sgl_data = sgl; LIST_INSERT_HEAD(&g_hv_sgl_page_pool.free_sgl_list, sgl_node, link); } /** * @brief create bounce buffer * * This function is responsible for create a Scatter/Gather list, * which hold several pages that can be aligned with page size. * * @param seg_count- SG-list segments count * @param write - if WRITE_TYPE, set SG list page used size to 0, * otherwise set used size to page size. * * return NULL if create failed */ static struct sglist * storvsc_create_bounce_buffer(uint16_t seg_count, int write) { int i = 0; struct sglist *bounce_sgl = NULL; unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE); struct hv_sgl_node *sgl_node = NULL; /* get struct sglist from free_sgl_list */ if (LIST_EMPTY(&g_hv_sgl_page_pool.free_sgl_list)) { printf("storvsc error: not enough free sgl\n"); return NULL; } sgl_node = LIST_FIRST(&g_hv_sgl_page_pool.free_sgl_list); LIST_REMOVE(sgl_node, link); bounce_sgl = sgl_node->sgl_data; LIST_INSERT_HEAD(&g_hv_sgl_page_pool.in_use_sgl_list, sgl_node, link); bounce_sgl->sg_maxseg = seg_count; if (write == WRITE_TYPE) bounce_sgl->sg_nseg = 0; else bounce_sgl->sg_nseg = seg_count; for (i = 0; i < seg_count; i++) bounce_sgl->sg_segs[i].ss_len = buf_len; return bounce_sgl; } /** * @brief copy data from SG list to bounce buffer * * This function is responsible for copy data from one SG list's segments * to another SG list which used as bounce buffer. * * @param bounce_sgl - the destination SG list * @param orig_sgl - the segment of the source SG list. * @param orig_sgl_count - the count of segments. * @param orig_sgl_count - indicate which segment need bounce buffer, * set 1 means need. * */ static void storvsc_copy_sgl_to_bounce_buf(struct sglist *bounce_sgl, bus_dma_segment_t *orig_sgl, unsigned int orig_sgl_count, uint64_t seg_bits) { int src_sgl_idx = 0; for (src_sgl_idx = 0; src_sgl_idx < orig_sgl_count; src_sgl_idx++) { if (seg_bits & (1 << src_sgl_idx)) { memcpy((void*)bounce_sgl->sg_segs[src_sgl_idx].ss_paddr, (void*)orig_sgl[src_sgl_idx].ds_addr, orig_sgl[src_sgl_idx].ds_len); bounce_sgl->sg_segs[src_sgl_idx].ss_len = orig_sgl[src_sgl_idx].ds_len; } } } /** * @brief copy data from SG list which used as bounce to another SG list * * This function is responsible for copy data from one SG list with bounce * buffer to another SG list's segments. * * @param dest_sgl - the destination SG list's segments * @param dest_sgl_count - the count of destination SG list's segment. * @param src_sgl - the source SG list. * @param seg_bits - indicate which segment used bounce buffer of src SG-list. * */ void storvsc_copy_from_bounce_buf_to_sgl(bus_dma_segment_t *dest_sgl, unsigned int dest_sgl_count, struct sglist* src_sgl, uint64_t seg_bits) { int sgl_idx = 0; for (sgl_idx = 0; sgl_idx < dest_sgl_count; sgl_idx++) { if (seg_bits & (1 << sgl_idx)) { memcpy((void*)(dest_sgl[sgl_idx].ds_addr), (void*)(src_sgl->sg_segs[sgl_idx].ss_paddr), src_sgl->sg_segs[sgl_idx].ss_len); } } } /** * @brief check SG list with bounce buffer or not * * This function is responsible for check if need bounce buffer for SG list. * * @param sgl - the SG list's segments * @param sg_count - the count of SG list's segment. * @param bits - segmengs number that need bounce buffer * * return -1 if SG list needless bounce buffer */ static int storvsc_check_bounce_buffer_sgl(bus_dma_segment_t *sgl, unsigned int sg_count, uint64_t *bits) { int i = 0; int offset = 0; uint64_t phys_addr = 0; uint64_t tmp_bits = 0; boolean_t found_hole = FALSE; boolean_t pre_aligned = TRUE; if (sg_count < 2){ return -1; } *bits = 0; phys_addr = vtophys(sgl[0].ds_addr); offset = phys_addr - trunc_page(phys_addr); if (offset != 0) { pre_aligned = FALSE; tmp_bits |= 1; } for (i = 1; i < sg_count; i++) { phys_addr = vtophys(sgl[i].ds_addr); offset = phys_addr - trunc_page(phys_addr); if (offset == 0) { if (FALSE == pre_aligned){ /* * This segment is aligned, if the previous * one is not aligned, find a hole */ found_hole = TRUE; } pre_aligned = TRUE; } else { tmp_bits |= 1 << i; if (!pre_aligned) { if (phys_addr != vtophys(sgl[i-1].ds_addr + sgl[i-1].ds_len)) { /* * Check whether connect to previous * segment,if not, find the hole */ found_hole = TRUE; } } else { found_hole = TRUE; } pre_aligned = FALSE; } } if (!found_hole) { return (-1); } else { *bits = tmp_bits; return 0; } } /** * @brief Fill in a request structure based on a CAM control block * * Fills in a request structure based on the contents of a CAM control * block. The request structure holds the payload information for * VSCSI protocol request. * * @param ccb pointer to a CAM contorl block * @param reqp pointer to a request structure */ static int create_storvsc_request(union ccb *ccb, struct hv_storvsc_request *reqp) { struct ccb_scsiio *csio = &ccb->csio; uint64_t phys_addr; uint32_t bytes_to_copy = 0; uint32_t pfn_num = 0; uint32_t pfn; uint64_t not_aligned_seg_bits = 0; /* refer to struct vmscsi_req for meanings of these two fields */ reqp->vstor_packet.u.vm_srb.port = cam_sim_unit(xpt_path_sim(ccb->ccb_h.path)); reqp->vstor_packet.u.vm_srb.path_id = cam_sim_bus(xpt_path_sim(ccb->ccb_h.path)); reqp->vstor_packet.u.vm_srb.target_id = ccb->ccb_h.target_id; reqp->vstor_packet.u.vm_srb.lun = ccb->ccb_h.target_lun; reqp->vstor_packet.u.vm_srb.cdb_len = csio->cdb_len; if(ccb->ccb_h.flags & CAM_CDB_POINTER) { memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_ptr, csio->cdb_len); } else { memcpy(&reqp->vstor_packet.u.vm_srb.u.cdb, csio->cdb_io.cdb_bytes, csio->cdb_len); } switch (ccb->ccb_h.flags & CAM_DIR_MASK) { case CAM_DIR_OUT: reqp->vstor_packet.u.vm_srb.data_in = WRITE_TYPE; break; case CAM_DIR_IN: reqp->vstor_packet.u.vm_srb.data_in = READ_TYPE; break; case CAM_DIR_NONE: reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE; break; default: reqp->vstor_packet.u.vm_srb.data_in = UNKNOWN_TYPE; break; } reqp->sense_data = &csio->sense_data; reqp->sense_info_len = csio->sense_len; reqp->ccb = ccb; if (0 == csio->dxfer_len) { return (0); } reqp->data_buf.length = csio->dxfer_len; switch (ccb->ccb_h.flags & CAM_DATA_MASK) { case CAM_DATA_VADDR: { bytes_to_copy = csio->dxfer_len; phys_addr = vtophys(csio->data_ptr); reqp->data_buf.offset = phys_addr & PAGE_MASK; while (bytes_to_copy != 0) { int bytes, page_offset; phys_addr = vtophys(&csio->data_ptr[reqp->data_buf.length - bytes_to_copy]); pfn = phys_addr >> PAGE_SHIFT; reqp->data_buf.pfn_array[pfn_num] = pfn; page_offset = phys_addr & PAGE_MASK; bytes = min(PAGE_SIZE - page_offset, bytes_to_copy); bytes_to_copy -= bytes; pfn_num++; } break; } case CAM_DATA_SG: { int i = 0; int offset = 0; int ret; bus_dma_segment_t *storvsc_sglist = (bus_dma_segment_t *)ccb->csio.data_ptr; u_int16_t storvsc_sg_count = ccb->csio.sglist_cnt; printf("Storvsc: get SG I/O operation, %d\n", reqp->vstor_packet.u.vm_srb.data_in); if (storvsc_sg_count > HV_MAX_MULTIPAGE_BUFFER_COUNT){ printf("Storvsc: %d segments is too much, " "only support %d segments\n", storvsc_sg_count, HV_MAX_MULTIPAGE_BUFFER_COUNT); return (EINVAL); } /* * We create our own bounce buffer function currently. Idealy * we should use BUS_DMA(9) framework. But with current BUS_DMA * code there is no callback API to check the page alignment of * middle segments before busdma can decide if a bounce buffer * is needed for particular segment. There is callback, * "bus_dma_filter_t *filter", but the parrameters are not * sufficient for storvsc driver. * TODO: * Add page alignment check in BUS_DMA(9) callback. Once * this is complete, switch the following code to use * BUS_DMA(9) for storvsc bounce buffer support. */ /* check if we need to create bounce buffer */ ret = storvsc_check_bounce_buffer_sgl(storvsc_sglist, storvsc_sg_count, ¬_aligned_seg_bits); if (ret != -1) { reqp->bounce_sgl = storvsc_create_bounce_buffer(storvsc_sg_count, reqp->vstor_packet.u.vm_srb.data_in); if (NULL == reqp->bounce_sgl) { printf("Storvsc_error: " "create bounce buffer failed.\n"); return (ENOMEM); } reqp->bounce_sgl_count = storvsc_sg_count; reqp->not_aligned_seg_bits = not_aligned_seg_bits; /* * if it is write, we need copy the original data *to bounce buffer */ if (WRITE_TYPE == reqp->vstor_packet.u.vm_srb.data_in) { storvsc_copy_sgl_to_bounce_buf( reqp->bounce_sgl, storvsc_sglist, storvsc_sg_count, reqp->not_aligned_seg_bits); } /* transfer virtual address to physical frame number */ if (reqp->not_aligned_seg_bits & 0x1){ phys_addr = vtophys(reqp->bounce_sgl->sg_segs[0].ss_paddr); }else{ phys_addr = vtophys(storvsc_sglist[0].ds_addr); } reqp->data_buf.offset = phys_addr & PAGE_MASK; pfn = phys_addr >> PAGE_SHIFT; reqp->data_buf.pfn_array[0] = pfn; for (i = 1; i < storvsc_sg_count; i++) { if (reqp->not_aligned_seg_bits & (1 << i)) { phys_addr = vtophys(reqp->bounce_sgl->sg_segs[i].ss_paddr); } else { phys_addr = vtophys(storvsc_sglist[i].ds_addr); } pfn = phys_addr >> PAGE_SHIFT; reqp->data_buf.pfn_array[i] = pfn; } } else { phys_addr = vtophys(storvsc_sglist[0].ds_addr); reqp->data_buf.offset = phys_addr & PAGE_MASK; for (i = 0; i < storvsc_sg_count; i++) { phys_addr = vtophys(storvsc_sglist[i].ds_addr); pfn = phys_addr >> PAGE_SHIFT; reqp->data_buf.pfn_array[i] = pfn; } /* check the last segment cross boundary or not */ offset = phys_addr & PAGE_MASK; if (offset) { phys_addr = vtophys(storvsc_sglist[i-1].ds_addr + PAGE_SIZE - offset); pfn = phys_addr >> PAGE_SHIFT; reqp->data_buf.pfn_array[i] = pfn; } reqp->bounce_sgl_count = 0; } break; } default: printf("Unknow flags: %d\n", ccb->ccb_h.flags); return(EINVAL); } return(0); } /* * Modified based on scsi_print_inquiry which is responsible to * print the detail information for scsi_inquiry_data. * * Return 1 if it is valid, 0 otherwise. */ static inline int is_inquiry_valid(const struct scsi_inquiry_data *inq_data) { uint8_t type; char vendor[16], product[48], revision[16]; /* * Check device type and qualifier */ if (!(SID_QUAL_IS_VENDOR_UNIQUE(inq_data) || SID_QUAL(inq_data) == SID_QUAL_LU_CONNECTED)) return (0); type = SID_TYPE(inq_data); switch (type) { case T_DIRECT: case T_SEQUENTIAL: case T_PRINTER: case T_PROCESSOR: case T_WORM: case T_CDROM: case T_SCANNER: case T_OPTICAL: case T_CHANGER: case T_COMM: case T_STORARRAY: case T_ENCLOSURE: case T_RBC: case T_OCRW: case T_OSD: case T_ADC: break; case T_NODEVICE: default: return (0); } /* * Check vendor, product, and revision */ cam_strvis(vendor, inq_data->vendor, sizeof(inq_data->vendor), sizeof(vendor)); cam_strvis(product, inq_data->product, sizeof(inq_data->product), sizeof(product)); cam_strvis(revision, inq_data->revision, sizeof(inq_data->revision), sizeof(revision)); if (strlen(vendor) == 0 || strlen(product) == 0 || strlen(revision) == 0) return (0); return (1); } /** * @brief completion function before returning to CAM * * I/O process has been completed and the result needs * to be passed to the CAM layer. * Free resources related to this request. * * @param reqp pointer to a request structure */ static void storvsc_io_done(struct hv_storvsc_request *reqp) { union ccb *ccb = reqp->ccb; struct ccb_scsiio *csio = &ccb->csio; struct storvsc_softc *sc = reqp->softc; struct vmscsi_req *vm_srb = &reqp->vstor_packet.u.vm_srb; bus_dma_segment_t *ori_sglist = NULL; int ori_sg_count = 0; /* destroy bounce buffer if it is used */ if (reqp->bounce_sgl_count) { ori_sglist = (bus_dma_segment_t *)ccb->csio.data_ptr; ori_sg_count = ccb->csio.sglist_cnt; /* * If it is READ operation, we should copy back the data * to original SG list. */ if (READ_TYPE == reqp->vstor_packet.u.vm_srb.data_in) { storvsc_copy_from_bounce_buf_to_sgl(ori_sglist, ori_sg_count, reqp->bounce_sgl, reqp->not_aligned_seg_bits); } storvsc_destroy_bounce_buffer(reqp->bounce_sgl); reqp->bounce_sgl_count = 0; } if (reqp->retries > 0) { mtx_lock(&sc->hs_lock); #if HVS_TIMEOUT_TEST xpt_print(ccb->ccb_h.path, "%u: IO returned after timeout, " "waking up timer handler if any.\n", ticks); mtx_lock(&reqp->event.mtx); cv_signal(&reqp->event.cv); mtx_unlock(&reqp->event.mtx); #endif reqp->retries = 0; xpt_print(ccb->ccb_h.path, "%u: IO returned after timeout, " "stopping timer if any.\n", ticks); mtx_unlock(&sc->hs_lock); } /* * callout_drain() will wait for the timer handler to finish * if it is running. So we don't need any lock to synchronize * between this routine and the timer handler. * Note that we need to make sure reqp is not freed when timer * handler is using or will use it. */ if (ccb->ccb_h.timeout != CAM_TIME_INFINITY) { callout_drain(&reqp->callout); } ccb->ccb_h.status &= ~CAM_SIM_QUEUED; ccb->ccb_h.status &= ~CAM_STATUS_MASK; if (vm_srb->scsi_status == SCSI_STATUS_OK) { const struct scsi_generic *cmd; /* * Check whether the data for INQUIRY cmd is valid or * not. Windows 10 and Windows 2016 send all zero * inquiry data to VM even for unpopulated slots. */ cmd = (const struct scsi_generic *) ((ccb->ccb_h.flags & CAM_CDB_POINTER) ? csio->cdb_io.cdb_ptr : csio->cdb_io.cdb_bytes); if (cmd->opcode == INQUIRY && is_inquiry_valid( (const struct scsi_inquiry_data *)csio->data_ptr) == 0) { ccb->ccb_h.status |= CAM_DEV_NOT_THERE; if (bootverbose) { mtx_lock(&sc->hs_lock); xpt_print(ccb->ccb_h.path, "storvsc uninstalled device\n"); mtx_unlock(&sc->hs_lock); } } else { ccb->ccb_h.status |= CAM_REQ_CMP; } } else { mtx_lock(&sc->hs_lock); xpt_print(ccb->ccb_h.path, "storvsc scsi_status = %d\n", vm_srb->scsi_status); mtx_unlock(&sc->hs_lock); ccb->ccb_h.status |= CAM_SCSI_STATUS_ERROR; } ccb->csio.scsi_status = (vm_srb->scsi_status & 0xFF); ccb->csio.resid = ccb->csio.dxfer_len - vm_srb->transfer_len; if (reqp->sense_info_len != 0) { csio->sense_resid = csio->sense_len - reqp->sense_info_len; ccb->ccb_h.status |= CAM_AUTOSNS_VALID; } mtx_lock(&sc->hs_lock); if (reqp->softc->hs_frozen == 1) { xpt_print(ccb->ccb_h.path, "%u: storvsc unfreezing softc 0x%p.\n", ticks, reqp->softc); ccb->ccb_h.status |= CAM_RELEASE_SIMQ; reqp->softc->hs_frozen = 0; } storvsc_free_request(sc, reqp); xpt_done(ccb); mtx_unlock(&sc->hs_lock); } /** * @brief Free a request structure * * Free a request structure by returning it to the free list * * @param sc pointer to a softc * @param reqp pointer to a request structure */ static void storvsc_free_request(struct storvsc_softc *sc, struct hv_storvsc_request *reqp) { LIST_INSERT_HEAD(&sc->hs_free_list, reqp, link); } /** * @brief Determine type of storage device from GUID * * Using the type GUID, determine if this is a StorVSC (paravirtual * SCSI or BlkVSC (paravirtual IDE) device. * * @param dev a device * returns an enum */ static enum hv_storage_type storvsc_get_storage_type(device_t dev) { const char *p = vmbus_get_type(dev); if (!memcmp(p, &gBlkVscDeviceType, sizeof(hv_guid))) { return DRIVER_BLKVSC; } else if (!memcmp(p, &gStorVscDeviceType, sizeof(hv_guid))) { return DRIVER_STORVSC; } return (DRIVER_UNKNOWN); }