Index: head/sys/dev/bnxt/bnxt_txrx.c =================================================================== --- head/sys/dev/bnxt/bnxt_txrx.c (revision 365060) +++ head/sys/dev/bnxt/bnxt_txrx.c (revision 365061) @@ -1,677 +1,676 @@ /*- * Broadcom NetXtreme-C/E network driver. * * Copyright (c) 2016 Broadcom, All Rights Reserved. * The term Broadcom refers to Broadcom Limited and/or its subsidiaries * * 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 COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS' * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #include "bnxt.h" /* * Function prototypes */ static int bnxt_isc_txd_encap(void *sc, if_pkt_info_t pi); static void bnxt_isc_txd_flush(void *sc, uint16_t txqid, qidx_t pidx); static int bnxt_isc_txd_credits_update(void *sc, uint16_t txqid, bool clear); static void bnxt_isc_rxd_refill(void *sc, if_rxd_update_t iru); /* uint16_t rxqid, uint8_t flid, uint32_t pidx, uint64_t *paddrs, caddr_t *vaddrs, uint16_t count, uint16_t buf_size); */ static void bnxt_isc_rxd_flush(void *sc, uint16_t rxqid, uint8_t flid, qidx_t pidx); static int bnxt_isc_rxd_available(void *sc, uint16_t rxqid, qidx_t idx, qidx_t budget); static int bnxt_isc_rxd_pkt_get(void *sc, if_rxd_info_t ri); static int bnxt_intr(void *sc); struct if_txrx bnxt_txrx = { .ift_txd_encap = bnxt_isc_txd_encap, .ift_txd_flush = bnxt_isc_txd_flush, .ift_txd_credits_update = bnxt_isc_txd_credits_update, .ift_rxd_available = bnxt_isc_rxd_available, .ift_rxd_pkt_get = bnxt_isc_rxd_pkt_get, .ift_rxd_refill = bnxt_isc_rxd_refill, .ift_rxd_flush = bnxt_isc_rxd_flush, .ift_legacy_intr = bnxt_intr }; /* * Device Dependent Packet Transmit and Receive Functions */ static const uint16_t bnxt_tx_lhint[] = { TX_BD_SHORT_FLAGS_LHINT_LT512, TX_BD_SHORT_FLAGS_LHINT_LT1K, TX_BD_SHORT_FLAGS_LHINT_LT2K, TX_BD_SHORT_FLAGS_LHINT_LT2K, TX_BD_SHORT_FLAGS_LHINT_GTE2K, }; static int bnxt_isc_txd_encap(void *sc, if_pkt_info_t pi) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_ring *txr = &softc->tx_rings[pi->ipi_qsidx]; struct tx_bd_long *tbd; struct tx_bd_long_hi *tbdh; bool need_hi = false; uint16_t flags_type; uint16_t lflags; uint32_t cfa_meta; int seg = 0; /* If we have offloads enabled, we need to use two BDs. */ if ((pi->ipi_csum_flags & (CSUM_OFFLOAD | CSUM_TSO | CSUM_IP)) || pi->ipi_mflags & M_VLANTAG) need_hi = true; /* TODO: Devices before Cu+B1 need to not mix long and short BDs */ need_hi = true; pi->ipi_new_pidx = pi->ipi_pidx; tbd = &((struct tx_bd_long *)txr->vaddr)[pi->ipi_new_pidx]; pi->ipi_ndescs = 0; /* No need to byte-swap the opaque value */ tbd->opaque = ((pi->ipi_nsegs + need_hi) << 24) | pi->ipi_new_pidx; tbd->len = htole16(pi->ipi_segs[seg].ds_len); tbd->addr = htole64(pi->ipi_segs[seg++].ds_addr); flags_type = ((pi->ipi_nsegs + need_hi) << TX_BD_SHORT_FLAGS_BD_CNT_SFT) & TX_BD_SHORT_FLAGS_BD_CNT_MASK; if (pi->ipi_len >= 2048) flags_type |= TX_BD_SHORT_FLAGS_LHINT_GTE2K; else flags_type |= bnxt_tx_lhint[pi->ipi_len >> 9]; if (need_hi) { flags_type |= TX_BD_LONG_TYPE_TX_BD_LONG; pi->ipi_new_pidx = RING_NEXT(txr, pi->ipi_new_pidx); tbdh = &((struct tx_bd_long_hi *)txr->vaddr)[pi->ipi_new_pidx]; tbdh->mss = htole16(pi->ipi_tso_segsz); tbdh->hdr_size = htole16((pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen) >> 1); tbdh->cfa_action = 0; lflags = 0; cfa_meta = 0; if (pi->ipi_mflags & M_VLANTAG) { /* TODO: Do we need to byte-swap the vtag here? */ cfa_meta = TX_BD_LONG_CFA_META_KEY_VLAN_TAG | pi->ipi_vtag; cfa_meta |= TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100; } tbdh->cfa_meta = htole32(cfa_meta); if (pi->ipi_csum_flags & CSUM_TSO) { lflags |= TX_BD_LONG_LFLAGS_LSO | TX_BD_LONG_LFLAGS_T_IPID; } else if(pi->ipi_csum_flags & CSUM_OFFLOAD) { lflags |= TX_BD_LONG_LFLAGS_TCP_UDP_CHKSUM | TX_BD_LONG_LFLAGS_IP_CHKSUM; } else if(pi->ipi_csum_flags & CSUM_IP) { lflags |= TX_BD_LONG_LFLAGS_IP_CHKSUM; } tbdh->lflags = htole16(lflags); } else { flags_type |= TX_BD_SHORT_TYPE_TX_BD_SHORT; } for (; seg < pi->ipi_nsegs; seg++) { tbd->flags_type = htole16(flags_type); pi->ipi_new_pidx = RING_NEXT(txr, pi->ipi_new_pidx); tbd = &((struct tx_bd_long *)txr->vaddr)[pi->ipi_new_pidx]; tbd->len = htole16(pi->ipi_segs[seg].ds_len); tbd->addr = htole64(pi->ipi_segs[seg].ds_addr); flags_type = TX_BD_SHORT_TYPE_TX_BD_SHORT; } flags_type |= TX_BD_SHORT_FLAGS_PACKET_END; tbd->flags_type = htole16(flags_type); pi->ipi_new_pidx = RING_NEXT(txr, pi->ipi_new_pidx); return 0; } static void bnxt_isc_txd_flush(void *sc, uint16_t txqid, qidx_t pidx) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_ring *tx_ring = &softc->tx_rings[txqid]; /* pidx is what we last set ipi_new_pidx to */ BNXT_TX_DB(tx_ring, pidx); /* TODO: Cumulus+ doesn't need the double doorbell */ BNXT_TX_DB(tx_ring, pidx); return; } static int bnxt_isc_txd_credits_update(void *sc, uint16_t txqid, bool clear) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_cp_ring *cpr = &softc->tx_cp_rings[txqid]; struct tx_cmpl *cmpl = (struct tx_cmpl *)cpr->ring.vaddr; int avail = 0; uint32_t cons = cpr->cons; bool v_bit = cpr->v_bit; bool last_v_bit; uint32_t last_cons; uint16_t type; uint16_t err; for (;;) { last_cons = cons; last_v_bit = v_bit; NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmpl[cons], v_bit)) goto done; type = cmpl[cons].flags_type & TX_CMPL_TYPE_MASK; switch (type) { case TX_CMPL_TYPE_TX_L2: err = (le16toh(cmpl[cons].errors_v) & TX_CMPL_ERRORS_BUFFER_ERROR_MASK) >> TX_CMPL_ERRORS_BUFFER_ERROR_SFT; if (err) device_printf(softc->dev, "TX completion error %u\n", err); /* No need to byte-swap the opaque value */ avail += cmpl[cons].opaque >> 24; /* * If we're not clearing, iflib only cares if there's * at least one buffer. Don't scan the whole ring in * this case. */ if (!clear) goto done; break; default: if (type & 1) { NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); if (!CMP_VALID(&cmpl[cons], v_bit)) goto done; } device_printf(softc->dev, "Unhandled TX completion type %u\n", type); break; } } done: if (clear && avail) { cpr->cons = last_cons; cpr->v_bit = last_v_bit; BNXT_CP_IDX_DISABLE_DB(&cpr->ring, cpr->cons); } return avail; } static void bnxt_isc_rxd_refill(void *sc, if_rxd_update_t iru) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_ring *rx_ring; struct rx_prod_pkt_bd *rxbd; uint16_t type; uint16_t i; uint16_t rxqid; uint16_t count, len; uint32_t pidx; uint8_t flid; uint64_t *paddrs; qidx_t *frag_idxs; rxqid = iru->iru_qsidx; count = iru->iru_count; len = iru->iru_buf_size; pidx = iru->iru_pidx; flid = iru->iru_flidx; paddrs = iru->iru_paddrs; frag_idxs = iru->iru_idxs; if (flid == 0) { rx_ring = &softc->rx_rings[rxqid]; type = RX_PROD_PKT_BD_TYPE_RX_PROD_PKT; } else { rx_ring = &softc->ag_rings[rxqid]; type = RX_PROD_AGG_BD_TYPE_RX_PROD_AGG; } rxbd = (void *)rx_ring->vaddr; for (i=0; iring_size) pidx = 0; } return; } static void bnxt_isc_rxd_flush(void *sc, uint16_t rxqid, uint8_t flid, qidx_t pidx) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_ring *rx_ring; if (flid == 0) rx_ring = &softc->rx_rings[rxqid]; else rx_ring = &softc->ag_rings[rxqid]; /* * We *must* update the completion ring before updating the RX ring * or we will overrun the completion ring and the device will wedge for * RX. */ if (softc->rx_cp_rings[rxqid].cons != UINT32_MAX) BNXT_CP_IDX_DISABLE_DB(&softc->rx_cp_rings[rxqid].ring, softc->rx_cp_rings[rxqid].cons); - /* We're given the last filled RX buffer here, not the next empty one */ - BNXT_RX_DB(rx_ring, RING_NEXT(rx_ring, pidx)); + BNXT_RX_DB(rx_ring, pidx); /* TODO: Cumulus+ doesn't need the double doorbell */ - BNXT_RX_DB(rx_ring, RING_NEXT(rx_ring, pidx)); + BNXT_RX_DB(rx_ring, pidx); return; } static int bnxt_isc_rxd_available(void *sc, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_cp_ring *cpr = &softc->rx_cp_rings[rxqid]; struct rx_pkt_cmpl *rcp; struct rx_tpa_end_cmpl *rtpae; struct cmpl_base *cmp = (struct cmpl_base *)cpr->ring.vaddr; int avail = 0; uint32_t cons = cpr->cons; bool v_bit = cpr->v_bit; uint8_t ags; int i; uint16_t type; for (;;) { NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; type = le16toh(cmp[cons].type) & CMPL_BASE_TYPE_MASK; switch (type) { case CMPL_BASE_TYPE_RX_L2: rcp = (void *)&cmp[cons]; ags = (rcp->agg_bufs_v1 & RX_PKT_CMPL_AGG_BUFS_MASK) >> RX_PKT_CMPL_AGG_BUFS_SFT; NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; /* Now account for all the AG completions */ for (i=0; iring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; } avail++; break; case CMPL_BASE_TYPE_RX_TPA_END: rtpae = (void *)&cmp[cons]; ags = (rtpae->agg_bufs_v1 & RX_TPA_END_CMPL_AGG_BUFS_MASK) >> RX_TPA_END_CMPL_AGG_BUFS_SFT; NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; /* Now account for all the AG completions */ for (i=0; iring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; } avail++; break; case CMPL_BASE_TYPE_RX_TPA_START: NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; break; case CMPL_BASE_TYPE_RX_AGG: break; default: device_printf(softc->dev, "Unhandled completion type %d on RXQ %d\n", type, rxqid); /* Odd completion types use two completions */ if (type & 1) { NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); CMPL_PREFETCH_NEXT(cpr, cons); if (!CMP_VALID(&cmp[cons], v_bit)) goto cmpl_invalid; } break; } if (avail > budget) break; } cmpl_invalid: return avail; } static void bnxt_set_rsstype(if_rxd_info_t ri, uint8_t rss_hash_type) { uint8_t rss_profile_id; rss_profile_id = BNXT_GET_RSS_PROFILE_ID(rss_hash_type); switch (rss_profile_id) { case BNXT_RSS_HASH_TYPE_TCPV4: ri->iri_rsstype = M_HASHTYPE_RSS_TCP_IPV4; break; case BNXT_RSS_HASH_TYPE_UDPV4: ri->iri_rsstype = M_HASHTYPE_RSS_UDP_IPV4; break; case BNXT_RSS_HASH_TYPE_IPV4: ri->iri_rsstype = M_HASHTYPE_RSS_IPV4; break; case BNXT_RSS_HASH_TYPE_TCPV6: ri->iri_rsstype = M_HASHTYPE_RSS_TCP_IPV6; break; case BNXT_RSS_HASH_TYPE_UDPV6: ri->iri_rsstype = M_HASHTYPE_RSS_UDP_IPV6; break; case BNXT_RSS_HASH_TYPE_IPV6: ri->iri_rsstype = M_HASHTYPE_RSS_IPV6; break; default: ri->iri_rsstype = M_HASHTYPE_OPAQUE_HASH; break; } } static int bnxt_pkt_get_l2(struct bnxt_softc *softc, if_rxd_info_t ri, struct bnxt_cp_ring *cpr, uint16_t flags_type) { struct rx_pkt_cmpl *rcp; struct rx_pkt_cmpl_hi *rcph; struct rx_abuf_cmpl *acp; uint32_t flags2; uint32_t errors; uint8_t ags; int i; rcp = &((struct rx_pkt_cmpl *)cpr->ring.vaddr)[cpr->cons]; /* Extract from the first 16-byte BD */ if (flags_type & RX_PKT_CMPL_FLAGS_RSS_VALID) { ri->iri_flowid = le32toh(rcp->rss_hash); bnxt_set_rsstype(ri, rcp->rss_hash_type); } else { ri->iri_rsstype = M_HASHTYPE_NONE; } ags = (rcp->agg_bufs_v1 & RX_PKT_CMPL_AGG_BUFS_MASK) >> RX_PKT_CMPL_AGG_BUFS_SFT; ri->iri_nfrags = ags + 1; /* No need to byte-swap the opaque value */ ri->iri_frags[0].irf_flid = (rcp->opaque >> 16) & 0xff; ri->iri_frags[0].irf_idx = rcp->opaque & 0xffff; ri->iri_frags[0].irf_len = le16toh(rcp->len); ri->iri_len = le16toh(rcp->len); /* Now the second 16-byte BD */ NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); rcph = &((struct rx_pkt_cmpl_hi *)cpr->ring.vaddr)[cpr->cons]; flags2 = le32toh(rcph->flags2); errors = le16toh(rcph->errors_v2); if ((flags2 & RX_PKT_CMPL_FLAGS2_META_FORMAT_MASK) == RX_PKT_CMPL_FLAGS2_META_FORMAT_VLAN) { ri->iri_flags |= M_VLANTAG; /* TODO: Should this be the entire 16-bits? */ ri->iri_vtag = le32toh(rcph->metadata) & (RX_PKT_CMPL_METADATA_VID_MASK | RX_PKT_CMPL_METADATA_DE | RX_PKT_CMPL_METADATA_PRI_MASK); } if (flags2 & RX_PKT_CMPL_FLAGS2_IP_CS_CALC) { ri->iri_csum_flags |= CSUM_IP_CHECKED; if (!(errors & RX_PKT_CMPL_ERRORS_IP_CS_ERROR)) ri->iri_csum_flags |= CSUM_IP_VALID; } if (flags2 & (RX_PKT_CMPL_FLAGS2_L4_CS_CALC | RX_PKT_CMPL_FLAGS2_T_L4_CS_CALC)) { ri->iri_csum_flags |= CSUM_L4_CALC; if (!(errors & (RX_PKT_CMPL_ERRORS_L4_CS_ERROR | RX_PKT_CMPL_ERRORS_T_L4_CS_ERROR))) { ri->iri_csum_flags |= CSUM_L4_VALID; ri->iri_csum_data = 0xffff; } } /* And finally the ag ring stuff. */ for (i=1; i < ri->iri_nfrags; i++) { NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); acp = &((struct rx_abuf_cmpl *)cpr->ring.vaddr)[cpr->cons]; /* No need to byte-swap the opaque value */ ri->iri_frags[i].irf_flid = (acp->opaque >> 16 & 0xff); ri->iri_frags[i].irf_idx = acp->opaque & 0xffff; ri->iri_frags[i].irf_len = le16toh(acp->len); ri->iri_len += le16toh(acp->len); } return 0; } static int bnxt_pkt_get_tpa(struct bnxt_softc *softc, if_rxd_info_t ri, struct bnxt_cp_ring *cpr, uint16_t flags_type) { struct rx_tpa_end_cmpl *agend = &((struct rx_tpa_end_cmpl *)cpr->ring.vaddr)[cpr->cons]; struct rx_abuf_cmpl *acp; struct bnxt_full_tpa_start *tpas; uint32_t flags2; uint8_t ags; uint8_t agg_id; int i; /* Get the agg_id */ agg_id = (agend->agg_id & RX_TPA_END_CMPL_AGG_ID_MASK) >> RX_TPA_END_CMPL_AGG_ID_SFT; tpas = &(softc->rx_rings[ri->iri_qsidx].tpa_start[agg_id]); /* Extract from the first 16-byte BD */ if (le16toh(tpas->low.flags_type) & RX_TPA_START_CMPL_FLAGS_RSS_VALID) { ri->iri_flowid = le32toh(tpas->low.rss_hash); bnxt_set_rsstype(ri, tpas->low.rss_hash_type); } else { ri->iri_rsstype = M_HASHTYPE_NONE; } ags = (agend->agg_bufs_v1 & RX_TPA_END_CMPL_AGG_BUFS_MASK) >> RX_TPA_END_CMPL_AGG_BUFS_SFT; ri->iri_nfrags = ags + 1; /* No need to byte-swap the opaque value */ ri->iri_frags[0].irf_flid = ((tpas->low.opaque >> 16) & 0xff); ri->iri_frags[0].irf_idx = (tpas->low.opaque & 0xffff); ri->iri_frags[0].irf_len = le16toh(tpas->low.len); ri->iri_len = le16toh(tpas->low.len); /* Now the second 16-byte BD */ NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); flags2 = le32toh(tpas->high.flags2); if ((flags2 & RX_TPA_START_CMPL_FLAGS2_META_FORMAT_MASK) == RX_TPA_START_CMPL_FLAGS2_META_FORMAT_VLAN) { ri->iri_flags |= M_VLANTAG; /* TODO: Should this be the entire 16-bits? */ ri->iri_vtag = le32toh(tpas->high.metadata) & (RX_TPA_START_CMPL_METADATA_VID_MASK | RX_TPA_START_CMPL_METADATA_DE | RX_TPA_START_CMPL_METADATA_PRI_MASK); } if (flags2 & RX_TPA_START_CMPL_FLAGS2_IP_CS_CALC) { ri->iri_csum_flags |= CSUM_IP_CHECKED; ri->iri_csum_flags |= CSUM_IP_VALID; } if (flags2 & RX_TPA_START_CMPL_FLAGS2_L4_CS_CALC) { ri->iri_csum_flags |= CSUM_L4_CALC; ri->iri_csum_flags |= CSUM_L4_VALID; ri->iri_csum_data = 0xffff; } /* Now the ag ring stuff. */ for (i=1; i < ri->iri_nfrags; i++) { NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); acp = &((struct rx_abuf_cmpl *)cpr->ring.vaddr)[cpr->cons]; /* No need to byte-swap the opaque value */ ri->iri_frags[i].irf_flid = ((acp->opaque >> 16) & 0xff); ri->iri_frags[i].irf_idx = (acp->opaque & 0xffff); ri->iri_frags[i].irf_len = le16toh(acp->len); ri->iri_len += le16toh(acp->len); } /* And finally, the empty BD at the end... */ ri->iri_nfrags++; /* No need to byte-swap the opaque value */ ri->iri_frags[i].irf_flid = ((agend->opaque >> 16) & 0xff); ri->iri_frags[i].irf_idx = (agend->opaque & 0xffff); ri->iri_frags[i].irf_len = le16toh(agend->len); ri->iri_len += le16toh(agend->len); return 0; } /* If we return anything but zero, iflib will assert... */ static int bnxt_isc_rxd_pkt_get(void *sc, if_rxd_info_t ri) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; struct bnxt_cp_ring *cpr = &softc->rx_cp_rings[ri->iri_qsidx]; struct cmpl_base *cmp_q = (struct cmpl_base *)cpr->ring.vaddr; struct cmpl_base *cmp; struct rx_tpa_start_cmpl *rtpa; uint16_t flags_type; uint16_t type; uint8_t agg_id; for (;;) { NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); CMPL_PREFETCH_NEXT(cpr, cpr->cons); cmp = &((struct cmpl_base *)cpr->ring.vaddr)[cpr->cons]; flags_type = le16toh(cmp->type); type = flags_type & CMPL_BASE_TYPE_MASK; switch (type) { case CMPL_BASE_TYPE_RX_L2: return bnxt_pkt_get_l2(softc, ri, cpr, flags_type); case CMPL_BASE_TYPE_RX_TPA_END: return bnxt_pkt_get_tpa(softc, ri, cpr, flags_type); case CMPL_BASE_TYPE_RX_TPA_START: rtpa = (void *)&cmp_q[cpr->cons]; agg_id = (rtpa->agg_id & RX_TPA_START_CMPL_AGG_ID_MASK) >> RX_TPA_START_CMPL_AGG_ID_SFT; softc->rx_rings[ri->iri_qsidx].tpa_start[agg_id].low = *rtpa; NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); CMPL_PREFETCH_NEXT(cpr, cpr->cons); softc->rx_rings[ri->iri_qsidx].tpa_start[agg_id].high = ((struct rx_tpa_start_cmpl_hi *)cmp_q)[cpr->cons]; break; default: device_printf(softc->dev, "Unhandled completion type %d on RXQ %d get\n", type, ri->iri_qsidx); if (type & 1) { NEXT_CP_CONS_V(&cpr->ring, cpr->cons, cpr->v_bit); ri->iri_cidx = RING_NEXT(&cpr->ring, ri->iri_cidx); CMPL_PREFETCH_NEXT(cpr, cpr->cons); } break; } } return 0; } static int bnxt_intr(void *sc) { struct bnxt_softc *softc = (struct bnxt_softc *)sc; device_printf(softc->dev, "STUB: %s @ %s:%d\n", __func__, __FILE__, __LINE__); return ENOSYS; } Index: head/sys/dev/mgb/if_mgb.c =================================================================== --- head/sys/dev/mgb/if_mgb.c (revision 365060) +++ head/sys/dev/mgb/if_mgb.c (revision 365061) @@ -1,1634 +1,1639 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2019 The FreeBSD Foundation, Inc. * * This driver was written by Gerald ND Aryeetey * under sponsorship from the FreeBSD Foundation. * * 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("$FreeBSD$"); /* * Microchip LAN7430/LAN7431 PCIe to Gigabit Ethernet Controller driver. * * Product information: * LAN7430 https://www.microchip.com/wwwproducts/en/LAN7430 * - Integrated IEEE 802.3 compliant PHY * LAN7431 https://www.microchip.com/wwwproducts/en/LAN7431 * - RGMII Interface * * This driver uses the iflib interface and the default 'ukphy' PHY driver. * * UNIMPLEMENTED FEATURES * ---------------------- * A number of features supported by LAN743X device are not yet implemented in * this driver: * * - Multiple (up to 4) RX queues support * - Just needs to remove asserts and malloc multiple `rx_ring_data` * structs based on ncpus. * - RX/TX Checksum Offloading support * - VLAN support * - Receive Packet Filtering (Multicast Perfect/Hash Address) support * - Wake on LAN (WoL) support * - TX LSO support * - Receive Side Scaling (RSS) support * - Debugging Capabilities: * - Could include MAC statistics and * error status registers in sysctl. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ifdi_if.h" #include "miibus_if.h" static pci_vendor_info_t mgb_vendor_info_array[] = { PVID(MGB_MICROCHIP_VENDOR_ID, MGB_LAN7430_DEVICE_ID, "Microchip LAN7430 PCIe Gigabit Ethernet Controller"), PVID(MGB_MICROCHIP_VENDOR_ID, MGB_LAN7431_DEVICE_ID, "Microchip LAN7431 PCIe Gigabit Ethernet Controller"), PVID_END }; /* Device methods */ static device_register_t mgb_register; /* IFLIB methods */ static ifdi_attach_pre_t mgb_attach_pre; static ifdi_attach_post_t mgb_attach_post; static ifdi_detach_t mgb_detach; static ifdi_tx_queues_alloc_t mgb_tx_queues_alloc; static ifdi_rx_queues_alloc_t mgb_rx_queues_alloc; static ifdi_queues_free_t mgb_queues_free; static ifdi_init_t mgb_init; static ifdi_stop_t mgb_stop; static ifdi_msix_intr_assign_t mgb_msix_intr_assign; static ifdi_tx_queue_intr_enable_t mgb_tx_queue_intr_enable; static ifdi_rx_queue_intr_enable_t mgb_rx_queue_intr_enable; static ifdi_intr_enable_t mgb_intr_enable_all; static ifdi_intr_disable_t mgb_intr_disable_all; /* IFLIB_TXRX methods */ static int mgb_isc_txd_encap(void *, if_pkt_info_t); static void mgb_isc_txd_flush(void *, uint16_t, qidx_t); static int mgb_isc_txd_credits_update(void *, uint16_t, bool); static int mgb_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t); static int mgb_isc_rxd_pkt_get(void *, if_rxd_info_t); static void mgb_isc_rxd_refill(void *, if_rxd_update_t); static void mgb_isc_rxd_flush(void *, uint16_t, uint8_t, qidx_t); /* Interrupts */ static driver_filter_t mgb_legacy_intr; static driver_filter_t mgb_admin_intr; static driver_filter_t mgb_rxq_intr; static bool mgb_intr_test(struct mgb_softc *); /* MII methods */ static miibus_readreg_t mgb_miibus_readreg; static miibus_writereg_t mgb_miibus_writereg; static miibus_linkchg_t mgb_miibus_linkchg; static miibus_statchg_t mgb_miibus_statchg; static int mgb_media_change(if_t); static void mgb_media_status(if_t, struct ifmediareq *); /* Helper/Test functions */ static int mgb_test_bar(struct mgb_softc *); static int mgb_alloc_regs(struct mgb_softc *); static int mgb_release_regs(struct mgb_softc *); static void mgb_get_ethaddr(struct mgb_softc *, struct ether_addr *); static int mgb_wait_for_bits(struct mgb_softc *, int, int, int); /* H/W init, reset and teardown helpers */ static int mgb_hw_init(struct mgb_softc *); static int mgb_hw_teardown(struct mgb_softc *); static int mgb_hw_reset(struct mgb_softc *); static int mgb_mac_init(struct mgb_softc *); static int mgb_dmac_reset(struct mgb_softc *); static int mgb_phy_reset(struct mgb_softc *); static int mgb_dma_init(struct mgb_softc *); static int mgb_dma_tx_ring_init(struct mgb_softc *, int); static int mgb_dma_rx_ring_init(struct mgb_softc *, int); static int mgb_dmac_control(struct mgb_softc *, int, int, enum mgb_dmac_cmd); static int mgb_fct_control(struct mgb_softc *, int, int, enum mgb_fct_cmd); /********************************************************************* * FreeBSD Device Interface Entry Points *********************************************************************/ static device_method_t mgb_methods[] = { /* Device interface */ DEVMETHOD(device_register, mgb_register), DEVMETHOD(device_probe, iflib_device_probe), DEVMETHOD(device_attach, iflib_device_attach), DEVMETHOD(device_detach, iflib_device_detach), DEVMETHOD(device_shutdown, iflib_device_shutdown), DEVMETHOD(device_suspend, iflib_device_suspend), DEVMETHOD(device_resume, iflib_device_resume), /* MII Interface */ DEVMETHOD(miibus_readreg, mgb_miibus_readreg), DEVMETHOD(miibus_writereg, mgb_miibus_writereg), DEVMETHOD(miibus_linkchg, mgb_miibus_linkchg), DEVMETHOD(miibus_statchg, mgb_miibus_statchg), DEVMETHOD_END }; static driver_t mgb_driver = { "mgb", mgb_methods, sizeof(struct mgb_softc) }; devclass_t mgb_devclass; DRIVER_MODULE(mgb, pci, mgb_driver, mgb_devclass, NULL, NULL); IFLIB_PNP_INFO(pci, mgb, mgb_vendor_info_array); MODULE_VERSION(mgb, 1); #if 0 /* MIIBUS_DEBUG */ /* If MIIBUS debug stuff is in attach then order matters. Use below instead. */ DRIVER_MODULE_ORDERED(miibus, mgb, miibus_driver, miibus_devclass, NULL, NULL, SI_ORDER_ANY); #endif /* MIIBUS_DEBUG */ DRIVER_MODULE(miibus, mgb, miibus_driver, miibus_devclass, NULL, NULL); MODULE_DEPEND(mgb, pci, 1, 1, 1); MODULE_DEPEND(mgb, ether, 1, 1, 1); MODULE_DEPEND(mgb, miibus, 1, 1, 1); MODULE_DEPEND(mgb, iflib, 1, 1, 1); static device_method_t mgb_iflib_methods[] = { DEVMETHOD(ifdi_attach_pre, mgb_attach_pre), DEVMETHOD(ifdi_attach_post, mgb_attach_post), DEVMETHOD(ifdi_detach, mgb_detach), DEVMETHOD(ifdi_init, mgb_init), DEVMETHOD(ifdi_stop, mgb_stop), DEVMETHOD(ifdi_tx_queues_alloc, mgb_tx_queues_alloc), DEVMETHOD(ifdi_rx_queues_alloc, mgb_rx_queues_alloc), DEVMETHOD(ifdi_queues_free, mgb_queues_free), DEVMETHOD(ifdi_msix_intr_assign, mgb_msix_intr_assign), DEVMETHOD(ifdi_tx_queue_intr_enable, mgb_tx_queue_intr_enable), DEVMETHOD(ifdi_rx_queue_intr_enable, mgb_rx_queue_intr_enable), DEVMETHOD(ifdi_intr_enable, mgb_intr_enable_all), DEVMETHOD(ifdi_intr_disable, mgb_intr_disable_all), #if 0 /* Not yet implemented IFLIB methods */ /* * Set multicast addresses, mtu and promiscuous mode */ DEVMETHOD(ifdi_multi_set, mgb_multi_set), DEVMETHOD(ifdi_mtu_set, mgb_mtu_set), DEVMETHOD(ifdi_promisc_set, mgb_promisc_set), /* * Needed for VLAN support */ DEVMETHOD(ifdi_vlan_register, mgb_vlan_register), DEVMETHOD(ifdi_vlan_unregister, mgb_vlan_unregister), /* * Needed for WOL support * at the very least. */ DEVMETHOD(ifdi_shutdown, mgb_shutdown), DEVMETHOD(ifdi_suspend, mgb_suspend), DEVMETHOD(ifdi_resume, mgb_resume), #endif /* UNUSED_IFLIB_METHODS */ DEVMETHOD_END }; static driver_t mgb_iflib_driver = { "mgb", mgb_iflib_methods, sizeof(struct mgb_softc) }; struct if_txrx mgb_txrx = { .ift_txd_encap = mgb_isc_txd_encap, .ift_txd_flush = mgb_isc_txd_flush, .ift_txd_credits_update = mgb_isc_txd_credits_update, .ift_rxd_available = mgb_isc_rxd_available, .ift_rxd_pkt_get = mgb_isc_rxd_pkt_get, .ift_rxd_refill = mgb_isc_rxd_refill, .ift_rxd_flush = mgb_isc_rxd_flush, .ift_legacy_intr = mgb_legacy_intr }; struct if_shared_ctx mgb_sctx_init = { .isc_magic = IFLIB_MAGIC, .isc_q_align = PAGE_SIZE, .isc_admin_intrcnt = 1, .isc_flags = IFLIB_DRIVER_MEDIA /* | IFLIB_HAS_RXCQ | IFLIB_HAS_TXCQ*/, .isc_vendor_info = mgb_vendor_info_array, .isc_driver_version = "1", .isc_driver = &mgb_iflib_driver, /* 2 queues per set for TX and RX (ring queue, head writeback queue) */ .isc_ntxqs = 2, .isc_tx_maxsize = MGB_DMA_MAXSEGS * MCLBYTES, /* .isc_tx_nsegments = MGB_DMA_MAXSEGS, */ .isc_tx_maxsegsize = MCLBYTES, .isc_ntxd_min = {1, 1}, /* Will want to make this bigger */ .isc_ntxd_max = {MGB_DMA_RING_SIZE, 1}, .isc_ntxd_default = {MGB_DMA_RING_SIZE, 1}, .isc_nrxqs = 2, .isc_rx_maxsize = MCLBYTES, .isc_rx_nsegments = 1, .isc_rx_maxsegsize = MCLBYTES, .isc_nrxd_min = {1, 1}, /* Will want to make this bigger */ .isc_nrxd_max = {MGB_DMA_RING_SIZE, 1}, .isc_nrxd_default = {MGB_DMA_RING_SIZE, 1}, .isc_nfl = 1, /*one free list since there is only one queue */ #if 0 /* UNUSED_CTX */ .isc_tso_maxsize = MGB_TSO_MAXSIZE + sizeof(struct ether_vlan_header), .isc_tso_maxsegsize = MGB_TX_MAXSEGSIZE, #endif /* UNUSED_CTX */ }; /*********************************************************************/ static void * mgb_register(device_t dev) { return (&mgb_sctx_init); } static int mgb_attach_pre(if_ctx_t ctx) { struct mgb_softc *sc; if_softc_ctx_t scctx; int error, phyaddr, rid; struct ether_addr hwaddr; struct mii_data *miid; sc = iflib_get_softc(ctx); sc->ctx = ctx; sc->dev = iflib_get_dev(ctx); scctx = iflib_get_softc_ctx(ctx); /* IFLIB required setup */ scctx->isc_txrx = &mgb_txrx; scctx->isc_tx_nsegments = MGB_DMA_MAXSEGS; /* Ring desc queues */ scctx->isc_txqsizes[0] = sizeof(struct mgb_ring_desc) * scctx->isc_ntxd[0]; scctx->isc_rxqsizes[0] = sizeof(struct mgb_ring_desc) * scctx->isc_nrxd[0]; /* Head WB queues */ scctx->isc_txqsizes[1] = sizeof(uint32_t) * scctx->isc_ntxd[1]; scctx->isc_rxqsizes[1] = sizeof(uint32_t) * scctx->isc_nrxd[1]; /* XXX: Must have 1 txqset, but can have up to 4 rxqsets */ scctx->isc_nrxqsets = 1; scctx->isc_ntxqsets = 1; /* scctx->isc_tx_csum_flags = (CSUM_TCP | CSUM_UDP) | (CSUM_TCP_IPV6 | CSUM_UDP_IPV6) | CSUM_TSO */ scctx->isc_tx_csum_flags = 0; scctx->isc_capabilities = scctx->isc_capenable = 0; #if 0 /* * CSUM, TSO and VLAN support are TBD */ IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 | IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO | IFCAP_JUMBO_MTU; scctx->isc_capabilities |= IFCAP_LRO | IFCAP_VLAN_HWFILTER; #endif /* get the BAR */ error = mgb_alloc_regs(sc); if (error != 0) { device_printf(sc->dev, "Unable to allocate bus resource: registers.\n"); goto fail; } error = mgb_test_bar(sc); if (error != 0) goto fail; error = mgb_hw_init(sc); if (error != 0) { device_printf(sc->dev, "MGB device init failed. (err: %d)\n", error); goto fail; } switch (pci_get_device(sc->dev)) { case MGB_LAN7430_DEVICE_ID: phyaddr = 1; break; case MGB_LAN7431_DEVICE_ID: default: phyaddr = MII_PHY_ANY; break; } /* XXX: Would be nice(r) if locked methods were here */ error = mii_attach(sc->dev, &sc->miibus, iflib_get_ifp(ctx), mgb_media_change, mgb_media_status, BMSR_DEFCAPMASK, phyaddr, MII_OFFSET_ANY, MIIF_DOPAUSE); if (error != 0) { device_printf(sc->dev, "Failed to attach MII interface\n"); goto fail; } miid = device_get_softc(sc->miibus); scctx->isc_media = &miid->mii_media; scctx->isc_msix_bar = pci_msix_table_bar(sc->dev); /** Setup PBA BAR **/ rid = pci_msix_pba_bar(sc->dev); if (rid != scctx->isc_msix_bar) { sc->pba = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->pba == NULL) { error = ENXIO; device_printf(sc->dev, "Failed to setup PBA BAR\n"); goto fail; } } mgb_get_ethaddr(sc, &hwaddr); if (ETHER_IS_BROADCAST(hwaddr.octet) || ETHER_IS_MULTICAST(hwaddr.octet) || ETHER_IS_ZERO(hwaddr.octet)) ether_gen_addr(iflib_get_ifp(ctx), &hwaddr); /* * XXX: if the MAC address was generated the linux driver * writes it back to the device. */ iflib_set_mac(ctx, hwaddr.octet); /* Map all vectors to vector 0 (admin interrupts) by default. */ CSR_WRITE_REG(sc, MGB_INTR_VEC_RX_MAP, 0); CSR_WRITE_REG(sc, MGB_INTR_VEC_TX_MAP, 0); CSR_WRITE_REG(sc, MGB_INTR_VEC_OTHER_MAP, 0); return (0); fail: mgb_detach(ctx); return (error); } static int mgb_attach_post(if_ctx_t ctx) { struct mgb_softc *sc; sc = iflib_get_softc(ctx); device_printf(sc->dev, "Interrupt test: %s\n", (mgb_intr_test(sc) ? "PASS" : "FAIL")); return (0); } static int mgb_detach(if_ctx_t ctx) { struct mgb_softc *sc; int error; sc = iflib_get_softc(ctx); /* XXX: Should report errors but still detach everything. */ error = mgb_hw_teardown(sc); /* Release IRQs */ iflib_irq_free(ctx, &sc->rx_irq); iflib_irq_free(ctx, &sc->admin_irq); if (sc->miibus != NULL) device_delete_child(sc->dev, sc->miibus); if (sc->pba != NULL) error = bus_release_resource(sc->dev, SYS_RES_MEMORY, rman_get_rid(sc->pba), sc->pba); sc->pba = NULL; error = mgb_release_regs(sc); return (error); } static int mgb_media_change(if_t ifp) { struct mii_data *miid; struct mii_softc *miisc; struct mgb_softc *sc; if_ctx_t ctx; int needs_reset; ctx = if_getsoftc(ifp); sc = iflib_get_softc(ctx); miid = device_get_softc(sc->miibus); LIST_FOREACH(miisc, &miid->mii_phys, mii_list) PHY_RESET(miisc); needs_reset = mii_mediachg(miid); if (needs_reset != 0) ifp->if_init(ctx); return (needs_reset); } static void mgb_media_status(if_t ifp, struct ifmediareq *ifmr) { struct mgb_softc *sc; struct mii_data *miid; sc = iflib_get_softc(if_getsoftc(ifp)); miid = device_get_softc(sc->miibus); if ((if_getflags(ifp) & IFF_UP) == 0) return; mii_pollstat(miid); ifmr->ifm_active = miid->mii_media_active; ifmr->ifm_status = miid->mii_media_status; } static int mgb_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets) { struct mgb_softc *sc; struct mgb_ring_data *rdata; int q; sc = iflib_get_softc(ctx); KASSERT(ntxqsets == 1, ("ntxqsets = %d", ntxqsets)); rdata = &sc->tx_ring_data; for (q = 0; q < ntxqsets; q++) { KASSERT(ntxqs == 2, ("ntxqs = %d", ntxqs)); /* Ring */ rdata->ring = (struct mgb_ring_desc *) vaddrs[q * ntxqs + 0]; rdata->ring_bus_addr = paddrs[q * ntxqs + 0]; /* Head WB */ rdata->head_wb = (uint32_t *) vaddrs[q * ntxqs + 1]; rdata->head_wb_bus_addr = paddrs[q * ntxqs + 1]; } return 0; } static int mgb_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets) { struct mgb_softc *sc; struct mgb_ring_data *rdata; int q; sc = iflib_get_softc(ctx); KASSERT(nrxqsets == 1, ("nrxqsets = %d", nrxqsets)); rdata = &sc->rx_ring_data; for (q = 0; q < nrxqsets; q++) { KASSERT(nrxqs == 2, ("nrxqs = %d", nrxqs)); /* Ring */ rdata->ring = (struct mgb_ring_desc *) vaddrs[q * nrxqs + 0]; rdata->ring_bus_addr = paddrs[q * nrxqs + 0]; /* Head WB */ rdata->head_wb = (uint32_t *) vaddrs[q * nrxqs + 1]; rdata->head_wb_bus_addr = paddrs[q * nrxqs + 1]; } return 0; } static void mgb_queues_free(if_ctx_t ctx) { struct mgb_softc *sc; sc = iflib_get_softc(ctx); memset(&sc->rx_ring_data, 0, sizeof(struct mgb_ring_data)); memset(&sc->tx_ring_data, 0, sizeof(struct mgb_ring_data)); } static void mgb_init(if_ctx_t ctx) { struct mgb_softc *sc; struct mii_data *miid; int error; sc = iflib_get_softc(ctx); miid = device_get_softc(sc->miibus); device_printf(sc->dev, "running init ...\n"); mgb_dma_init(sc); /* XXX: Turn off perfect filtering, turn on (broad|multi|uni)cast rx */ CSR_CLEAR_REG(sc, MGB_RFE_CTL, MGB_RFE_ALLOW_PERFECT_FILTER); CSR_UPDATE_REG(sc, MGB_RFE_CTL, MGB_RFE_ALLOW_BROADCAST | MGB_RFE_ALLOW_UNICAST | MGB_RFE_ALLOW_UNICAST); error = mii_mediachg(miid); KASSERT(!error, ("mii_mediachg returned: %d", error)); } #ifdef DEBUG static void mgb_dump_some_stats(struct mgb_softc *sc) { int i; int first_stat = 0x1200; int last_stat = 0x12FC; for (i = first_stat; i <= last_stat; i += 4) if (CSR_READ_REG(sc, i) != 0) device_printf(sc->dev, "0x%04x: 0x%08x\n", i, CSR_READ_REG(sc, i)); char *stat_names[] = { "MAC_ERR_STS ", "FCT_INT_STS ", "DMAC_CFG ", "DMAC_CMD ", "DMAC_INT_STS ", "DMAC_INT_EN ", "DMAC_RX_ERR_STS0 ", "DMAC_RX_ERR_STS1 ", "DMAC_RX_ERR_STS2 ", "DMAC_RX_ERR_STS3 ", "INT_STS ", "INT_EN ", "INT_VEC_EN ", "INT_VEC_MAP0 ", "INT_VEC_MAP1 ", "INT_VEC_MAP2 ", "TX_HEAD0", "TX_TAIL0", "DMAC_TX_ERR_STS0 ", NULL }; int stats[] = { 0x114, 0xA0, 0xC00, 0xC0C, 0xC10, 0xC14, 0xC60, 0xCA0, 0xCE0, 0xD20, 0x780, 0x788, 0x794, 0x7A0, 0x7A4, 0x780, 0xD58, 0xD5C, 0xD60, 0x0 }; i = 0; printf("==============================\n"); while (stats[i++]) device_printf(sc->dev, "%s at offset 0x%04x = 0x%08x\n", stat_names[i - 1], stats[i - 1], CSR_READ_REG(sc, stats[i - 1])); printf("==== TX RING DESCS ====\n"); for (i = 0; i < MGB_DMA_RING_SIZE; i++) device_printf(sc->dev, "ring[%d].data0=0x%08x\n" "ring[%d].data1=0x%08x\n" "ring[%d].data2=0x%08x\n" "ring[%d].data3=0x%08x\n", i, sc->tx_ring_data.ring[i].ctl, i, sc->tx_ring_data.ring[i].addr.low, i, sc->tx_ring_data.ring[i].addr.high, i, sc->tx_ring_data.ring[i].sts); device_printf(sc->dev, "==== DUMP_TX_DMA_RAM ====\n"); int i; CSR_WRITE_REG(sc, 0x24, 0xF); // DP_SEL & TX_RAM_0 for (i = 0; i < 128; i++) { CSR_WRITE_REG(sc, 0x2C, i); // DP_ADDR CSR_WRITE_REG(sc, 0x28, 0); // DP_CMD while ((CSR_READ_REG(sc, 0x24) & 0x80000000) == 0) // DP_SEL & READY DELAY(1000); device_printf(sc->dev, "DMAC_TX_RAM_0[%u]=%08x\n", i, CSR_READ_REG(sc, 0x30)); // DP_DATA } } #endif static void mgb_stop(if_ctx_t ctx) { struct mgb_softc *sc ; if_softc_ctx_t scctx; int i; sc = iflib_get_softc(ctx); scctx = iflib_get_softc_ctx(ctx); /* XXX: Could potentially timeout */ for (i = 0; i < scctx->isc_nrxqsets; i++) { mgb_dmac_control(sc, MGB_DMAC_RX_START, 0, DMAC_STOP); mgb_fct_control(sc, MGB_FCT_RX_CTL, 0, FCT_DISABLE); } for (i = 0; i < scctx->isc_ntxqsets; i++) { mgb_dmac_control(sc, MGB_DMAC_TX_START, 0, DMAC_STOP); mgb_fct_control(sc, MGB_FCT_TX_CTL, 0, FCT_DISABLE); } } static int mgb_legacy_intr(void *xsc) { struct mgb_softc *sc; sc = xsc; iflib_admin_intr_deferred(sc->ctx); return (FILTER_HANDLED); } static int mgb_rxq_intr(void *xsc) { struct mgb_softc *sc; if_softc_ctx_t scctx; uint32_t intr_sts, intr_en; int qidx; sc = xsc; scctx = iflib_get_softc_ctx(sc->ctx); intr_sts = CSR_READ_REG(sc, MGB_INTR_STS); intr_en = CSR_READ_REG(sc, MGB_INTR_ENBL_SET); intr_sts &= intr_en; for (qidx = 0; qidx < scctx->isc_nrxqsets; qidx++) { if ((intr_sts & MGB_INTR_STS_RX(qidx))){ CSR_WRITE_REG(sc, MGB_INTR_ENBL_CLR, MGB_INTR_STS_RX(qidx)); CSR_WRITE_REG(sc, MGB_INTR_STS, MGB_INTR_STS_RX(qidx)); } } return (FILTER_SCHEDULE_THREAD); } static int mgb_admin_intr(void *xsc) { struct mgb_softc *sc; if_softc_ctx_t scctx; uint32_t intr_sts, intr_en; int qidx; sc = xsc; scctx = iflib_get_softc_ctx(sc->ctx); intr_sts = CSR_READ_REG(sc, MGB_INTR_STS); intr_en = CSR_READ_REG(sc, MGB_INTR_ENBL_SET); intr_sts &= intr_en; /* * NOTE: Debugging printfs here * will likely cause interrupt test failure. */ /* TODO: shouldn't continue if suspended */ if ((intr_sts & MGB_INTR_STS_ANY) == 0) { device_printf(sc->dev, "non-mgb interrupt triggered.\n"); return (FILTER_SCHEDULE_THREAD); } if ((intr_sts & MGB_INTR_STS_TEST) != 0) { sc->isr_test_flag = true; CSR_WRITE_REG(sc, MGB_INTR_STS, MGB_INTR_STS_TEST); return (FILTER_HANDLED); } if ((intr_sts & MGB_INTR_STS_RX_ANY) != 0) { for (qidx = 0; qidx < scctx->isc_nrxqsets; qidx++) { if ((intr_sts & MGB_INTR_STS_RX(qidx))){ iflib_rx_intr_deferred(sc->ctx, qidx); } } return (FILTER_HANDLED); } /* XXX: TX interrupts should not occur */ if ((intr_sts & MGB_INTR_STS_TX_ANY) != 0) { for (qidx = 0; qidx < scctx->isc_ntxqsets; qidx++) { if ((intr_sts & MGB_INTR_STS_RX(qidx))) { /* clear the interrupt sts and run handler */ CSR_WRITE_REG(sc, MGB_INTR_ENBL_CLR, MGB_INTR_STS_TX(qidx)); CSR_WRITE_REG(sc, MGB_INTR_STS, MGB_INTR_STS_TX(qidx)); iflib_tx_intr_deferred(sc->ctx, qidx); } } return (FILTER_HANDLED); } return (FILTER_SCHEDULE_THREAD); } static int mgb_msix_intr_assign(if_ctx_t ctx, int msix) { struct mgb_softc *sc; if_softc_ctx_t scctx; int error, i, vectorid; char irq_name[16]; sc = iflib_get_softc(ctx); scctx = iflib_get_softc_ctx(ctx); KASSERT(scctx->isc_nrxqsets == 1 && scctx->isc_ntxqsets == 1, ("num rxqsets/txqsets != 1 ")); /* * First vector should be admin interrupts, others vectors are TX/RX * * RIDs start at 1, and vector ids start at 0. */ vectorid = 0; error = iflib_irq_alloc_generic(ctx, &sc->admin_irq, vectorid + 1, IFLIB_INTR_ADMIN, mgb_admin_intr, sc, 0, "admin"); if (error) { device_printf(sc->dev, "Failed to register admin interrupt handler\n"); return (error); } for (i = 0; i < scctx->isc_nrxqsets; i++) { vectorid++; snprintf(irq_name, sizeof(irq_name), "rxq%d", i); error = iflib_irq_alloc_generic(ctx, &sc->rx_irq, vectorid + 1, IFLIB_INTR_RX, mgb_rxq_intr, sc, i, irq_name); if (error) { device_printf(sc->dev, "Failed to register rxq %d interrupt handler\n", i); return (error); } CSR_UPDATE_REG(sc, MGB_INTR_VEC_RX_MAP, MGB_INTR_VEC_MAP(vectorid, i)); } /* Not actually mapping hw TX interrupts ... */ for (i = 0; i < scctx->isc_ntxqsets; i++) { snprintf(irq_name, sizeof(irq_name), "txq%d", i); iflib_softirq_alloc_generic(ctx, NULL, IFLIB_INTR_TX, NULL, i, irq_name); } return (0); } static void mgb_intr_enable_all(if_ctx_t ctx) { struct mgb_softc *sc; if_softc_ctx_t scctx; int i, dmac_enable = 0, intr_sts = 0, vec_en = 0; sc = iflib_get_softc(ctx); scctx = iflib_get_softc_ctx(ctx); intr_sts |= MGB_INTR_STS_ANY; vec_en |= MGB_INTR_STS_ANY; for (i = 0; i < scctx->isc_nrxqsets; i++) { intr_sts |= MGB_INTR_STS_RX(i); dmac_enable |= MGB_DMAC_RX_INTR_ENBL(i); vec_en |= MGB_INTR_RX_VEC_STS(i); } /* TX interrupts aren't needed ... */ CSR_WRITE_REG(sc, MGB_INTR_ENBL_SET, intr_sts); CSR_WRITE_REG(sc, MGB_INTR_VEC_ENBL_SET, vec_en); CSR_WRITE_REG(sc, MGB_DMAC_INTR_STS, dmac_enable); CSR_WRITE_REG(sc, MGB_DMAC_INTR_ENBL_SET, dmac_enable); } static void mgb_intr_disable_all(if_ctx_t ctx) { struct mgb_softc *sc; sc = iflib_get_softc(ctx); CSR_WRITE_REG(sc, MGB_INTR_ENBL_CLR, UINT32_MAX); CSR_WRITE_REG(sc, MGB_INTR_VEC_ENBL_CLR, UINT32_MAX); CSR_WRITE_REG(sc, MGB_INTR_STS, UINT32_MAX); CSR_WRITE_REG(sc, MGB_DMAC_INTR_ENBL_CLR, UINT32_MAX); CSR_WRITE_REG(sc, MGB_DMAC_INTR_STS, UINT32_MAX); } static int mgb_rx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { /* called after successful rx isr */ struct mgb_softc *sc; sc = iflib_get_softc(ctx); CSR_WRITE_REG(sc, MGB_INTR_VEC_ENBL_SET, MGB_INTR_RX_VEC_STS(qid)); CSR_WRITE_REG(sc, MGB_INTR_ENBL_SET, MGB_INTR_STS_RX(qid)); CSR_WRITE_REG(sc, MGB_DMAC_INTR_STS, MGB_DMAC_RX_INTR_ENBL(qid)); CSR_WRITE_REG(sc, MGB_DMAC_INTR_ENBL_SET, MGB_DMAC_RX_INTR_ENBL(qid)); return (0); } static int mgb_tx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { /* XXX: not called (since tx interrupts not used) */ struct mgb_softc *sc; sc = iflib_get_softc(ctx); CSR_WRITE_REG(sc, MGB_INTR_ENBL_SET, MGB_INTR_STS_TX(qid)); CSR_WRITE_REG(sc, MGB_DMAC_INTR_STS, MGB_DMAC_TX_INTR_ENBL(qid)); CSR_WRITE_REG(sc, MGB_DMAC_INTR_ENBL_SET, MGB_DMAC_TX_INTR_ENBL(qid)); return (0); } static bool mgb_intr_test(struct mgb_softc *sc) { int i; sc->isr_test_flag = false; CSR_WRITE_REG(sc, MGB_INTR_STS, MGB_INTR_STS_TEST); CSR_WRITE_REG(sc, MGB_INTR_VEC_ENBL_SET, MGB_INTR_STS_ANY); CSR_WRITE_REG(sc, MGB_INTR_ENBL_SET, MGB_INTR_STS_ANY | MGB_INTR_STS_TEST); CSR_WRITE_REG(sc, MGB_INTR_SET, MGB_INTR_STS_TEST); if (sc->isr_test_flag) return true; for (i = 0; i < MGB_TIMEOUT; i++) { DELAY(10); if (sc->isr_test_flag) break; } CSR_WRITE_REG(sc, MGB_INTR_ENBL_CLR, MGB_INTR_STS_TEST); CSR_WRITE_REG(sc, MGB_INTR_STS, MGB_INTR_STS_TEST); return sc->isr_test_flag; } static int mgb_isc_txd_encap(void *xsc , if_pkt_info_t ipi) { struct mgb_softc *sc; if_softc_ctx_t scctx; struct mgb_ring_data *rdata; struct mgb_ring_desc *txd; bus_dma_segment_t *segs; qidx_t pidx, nsegs; int i; KASSERT(ipi->ipi_qsidx == 0, ("tried to refill TX Channel %d.\n", ipi->ipi_qsidx)); sc = xsc; scctx = iflib_get_softc_ctx(sc->ctx); rdata = &sc->tx_ring_data; pidx = ipi->ipi_pidx; segs = ipi->ipi_segs; nsegs = ipi->ipi_nsegs; /* For each seg, create a descriptor */ for (i = 0; i < nsegs; ++i) { KASSERT(nsegs == 1, ("Multisegment packet !!!!!\n")); txd = &rdata->ring[pidx]; txd->ctl = htole32( (segs[i].ds_len & MGB_DESC_CTL_BUFLEN_MASK ) | /* * XXX: This will be wrong in the multipacket case * I suspect FS should be for the first packet and * LS should be for the last packet */ MGB_TX_DESC_CTL_FS | MGB_TX_DESC_CTL_LS | MGB_DESC_CTL_FCS); txd->addr.low = htole32(CSR_TRANSLATE_ADDR_LOW32( segs[i].ds_addr)); txd->addr.high = htole32(CSR_TRANSLATE_ADDR_HIGH32( segs[i].ds_addr)); txd->sts = htole32( (segs[i].ds_len << 16) & MGB_DESC_FRAME_LEN_MASK); pidx = MGB_NEXT_RING_IDX(pidx); } ipi->ipi_new_pidx = pidx; return (0); } static void mgb_isc_txd_flush(void *xsc, uint16_t txqid, qidx_t pidx) { struct mgb_softc *sc; struct mgb_ring_data *rdata; KASSERT(txqid == 0, ("tried to flush TX Channel %d.\n", txqid)); sc = xsc; rdata = &sc->tx_ring_data; if (rdata->last_tail != pidx) { rdata->last_tail = pidx; CSR_WRITE_REG(sc, MGB_DMA_TX_TAIL(txqid), rdata->last_tail); } } static int mgb_isc_txd_credits_update(void *xsc, uint16_t txqid, bool clear) { struct mgb_softc *sc; struct mgb_ring_desc *txd; struct mgb_ring_data *rdata; int processed = 0; /* * > If clear is true, we need to report the number of TX command ring * > descriptors that have been processed by the device. If clear is * > false, we just need to report whether or not at least one TX * > command ring descriptor has been processed by the device. * - vmx driver */ KASSERT(txqid == 0, ("tried to credits_update TX Channel %d.\n", txqid)); sc = xsc; rdata = &sc->tx_ring_data; while (*(rdata->head_wb) != rdata->last_head) { if (!clear) return 1; txd = &rdata->ring[rdata->last_head]; memset(txd, 0, sizeof(struct mgb_ring_desc)); rdata->last_head = MGB_NEXT_RING_IDX(rdata->last_head); processed++; } return (processed); } static int mgb_isc_rxd_available(void *xsc, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct mgb_softc *sc; if_softc_ctx_t scctx; struct mgb_ring_data *rdata; int avail = 0; sc = xsc; KASSERT(rxqid == 0, ("tried to check availability in RX Channel %d.\n", rxqid)); rdata = &sc->rx_ring_data; scctx = iflib_get_softc_ctx(sc->ctx); for (; idx != *(rdata->head_wb); idx = MGB_NEXT_RING_IDX(idx)) { avail++; /* XXX: Could verify desc is device owned here */ if (avail == budget) break; } return (avail); } static int mgb_isc_rxd_pkt_get(void *xsc, if_rxd_info_t ri) { struct mgb_softc *sc; struct mgb_ring_data *rdata; struct mgb_ring_desc rxd; int total_len; KASSERT(ri->iri_qsidx == 0, ("tried to check availability in RX Channel %d\n", ri->iri_qsidx)); sc = xsc; total_len = 0; rdata = &sc->rx_ring_data; while (*(rdata->head_wb) != rdata->last_head) { /* copy ring desc and do swapping */ rxd = rdata->ring[rdata->last_head]; rxd.ctl = le32toh(rxd.ctl); rxd.addr.low = le32toh(rxd.ctl); rxd.addr.high = le32toh(rxd.ctl); rxd.sts = le32toh(rxd.ctl); if ((rxd.ctl & MGB_DESC_CTL_OWN) != 0) { device_printf(sc->dev, "Tried to read descriptor ... " "found that it's owned by the driver\n"); return EINVAL; } if ((rxd.ctl & MGB_RX_DESC_CTL_FS) == 0) { device_printf(sc->dev, "Tried to read descriptor ... " "found that FS is not set.\n"); device_printf(sc->dev, "Tried to read descriptor ... that it FS is not set.\n"); return EINVAL; } /* XXX: Multi-packet support */ if ((rxd.ctl & MGB_RX_DESC_CTL_LS) == 0) { device_printf(sc->dev, "Tried to read descriptor ... " "found that LS is not set. (Multi-buffer packets not yet supported)\n"); return EINVAL; } ri->iri_frags[0].irf_flid = 0; ri->iri_frags[0].irf_idx = rdata->last_head; ri->iri_frags[0].irf_len = MGB_DESC_GET_FRAME_LEN(&rxd); total_len += ri->iri_frags[0].irf_len; rdata->last_head = MGB_NEXT_RING_IDX(rdata->last_head); break; } ri->iri_nfrags = 1; ri->iri_len = total_len; return (0); } static void mgb_isc_rxd_refill(void *xsc, if_rxd_update_t iru) { if_softc_ctx_t scctx; struct mgb_softc *sc; struct mgb_ring_data *rdata; struct mgb_ring_desc *rxd; uint64_t *paddrs; qidx_t *idxs; qidx_t idx; int count, len; count = iru->iru_count; len = iru->iru_buf_size; idxs = iru->iru_idxs; paddrs = iru->iru_paddrs; KASSERT(iru->iru_qsidx == 0, ("tried to refill RX Channel %d.\n", iru->iru_qsidx)); sc = xsc; scctx = iflib_get_softc_ctx(sc->ctx); rdata = &sc->rx_ring_data; while (count > 0) { idx = idxs[--count]; rxd = &rdata->ring[idx]; rxd->sts = 0; rxd->addr.low = htole32(CSR_TRANSLATE_ADDR_LOW32(paddrs[count])); rxd->addr.high = htole32(CSR_TRANSLATE_ADDR_HIGH32(paddrs[count])); rxd->ctl = htole32(MGB_DESC_CTL_OWN | (len & MGB_DESC_CTL_BUFLEN_MASK)); } return; } static void mgb_isc_rxd_flush(void *xsc, uint16_t rxqid, uint8_t flid, qidx_t pidx) { struct mgb_softc *sc; sc = xsc; KASSERT(rxqid == 0, ("tried to flush RX Channel %d.\n", rxqid)); - sc->rx_ring_data.last_tail = pidx; + /* + * According to the programming guide, last_tail must be set to + * the last valid RX descriptor, rather than to the one past that. + * Note that this is not true for the TX ring! + */ + sc->rx_ring_data.last_tail = MGB_PREV_RING_IDX(pidx); CSR_WRITE_REG(sc, MGB_DMA_RX_TAIL(rxqid), sc->rx_ring_data.last_tail); return; } static int mgb_test_bar(struct mgb_softc *sc) { uint32_t id_rev, dev_id, rev; id_rev = CSR_READ_REG(sc, 0); dev_id = id_rev >> 16; rev = id_rev & 0xFFFF; if (dev_id == MGB_LAN7430_DEVICE_ID || dev_id == MGB_LAN7431_DEVICE_ID) { return 0; } else { device_printf(sc->dev, "ID check failed.\n"); return ENXIO; } } static int mgb_alloc_regs(struct mgb_softc *sc) { int rid; rid = PCIR_BAR(MGB_BAR); pci_enable_busmaster(sc->dev); sc->regs = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->regs == NULL) return ENXIO; return (0); } static int mgb_release_regs(struct mgb_softc *sc) { int error = 0; if (sc->regs != NULL) error = bus_release_resource(sc->dev, SYS_RES_MEMORY, rman_get_rid(sc->regs), sc->regs); sc->regs = NULL; pci_disable_busmaster(sc->dev); return error; } static int mgb_dma_init(struct mgb_softc *sc) { if_softc_ctx_t scctx; int ch, error = 0; scctx = iflib_get_softc_ctx(sc->ctx); for (ch = 0; ch < scctx->isc_nrxqsets; ch++) if ((error = mgb_dma_rx_ring_init(sc, ch))) goto fail; for (ch = 0; ch < scctx->isc_nrxqsets; ch++) if ((error = mgb_dma_tx_ring_init(sc, ch))) goto fail; fail: return error; } static int mgb_dma_rx_ring_init(struct mgb_softc *sc, int channel) { struct mgb_ring_data *rdata; int ring_config, error = 0; rdata = &sc->rx_ring_data; mgb_dmac_control(sc, MGB_DMAC_RX_START, 0, DMAC_RESET); KASSERT(MGB_DMAC_STATE_IS_INITIAL(sc, MGB_DMAC_RX_START, channel), ("Trying to init channels when not in init state\n")); /* write ring address */ if (rdata->ring_bus_addr == 0) { device_printf(sc->dev, "Invalid ring bus addr.\n"); goto fail; } CSR_WRITE_REG(sc, MGB_DMA_RX_BASE_H(channel), CSR_TRANSLATE_ADDR_HIGH32(rdata->ring_bus_addr)); CSR_WRITE_REG(sc, MGB_DMA_RX_BASE_L(channel), CSR_TRANSLATE_ADDR_LOW32(rdata->ring_bus_addr)); /* write head pointer writeback address */ if (rdata->head_wb_bus_addr == 0) { device_printf(sc->dev, "Invalid head wb bus addr.\n"); goto fail; } CSR_WRITE_REG(sc, MGB_DMA_RX_HEAD_WB_H(channel), CSR_TRANSLATE_ADDR_HIGH32(rdata->head_wb_bus_addr)); CSR_WRITE_REG(sc, MGB_DMA_RX_HEAD_WB_L(channel), CSR_TRANSLATE_ADDR_LOW32(rdata->head_wb_bus_addr)); /* Enable head pointer writeback */ CSR_WRITE_REG(sc, MGB_DMA_RX_CONFIG0(channel), MGB_DMA_HEAD_WB_ENBL); ring_config = CSR_READ_REG(sc, MGB_DMA_RX_CONFIG1(channel)); /* ring size */ ring_config &= ~MGB_DMA_RING_LEN_MASK; ring_config |= (MGB_DMA_RING_SIZE & MGB_DMA_RING_LEN_MASK); /* packet padding (PAD_2 is better for IP header alignment ...) */ ring_config &= ~MGB_DMA_RING_PAD_MASK; ring_config |= (MGB_DMA_RING_PAD_0 & MGB_DMA_RING_PAD_MASK); CSR_WRITE_REG(sc, MGB_DMA_RX_CONFIG1(channel), ring_config); rdata->last_head = CSR_READ_REG(sc, MGB_DMA_RX_HEAD(channel)); mgb_fct_control(sc, MGB_FCT_RX_CTL, channel, FCT_RESET); if (error != 0) { device_printf(sc->dev, "Failed to reset RX FCT.\n"); goto fail; } mgb_fct_control(sc, MGB_FCT_RX_CTL, channel, FCT_ENABLE); if (error != 0) { device_printf(sc->dev, "Failed to enable RX FCT.\n"); goto fail; } mgb_dmac_control(sc, MGB_DMAC_RX_START, channel, DMAC_START); if (error != 0) device_printf(sc->dev, "Failed to start RX DMAC.\n"); fail: return (error); } static int mgb_dma_tx_ring_init(struct mgb_softc *sc, int channel) { struct mgb_ring_data *rdata; int ring_config, error = 0; rdata = &sc->tx_ring_data; if ((error = mgb_fct_control(sc, MGB_FCT_TX_CTL, channel, FCT_RESET))) { device_printf(sc->dev, "Failed to reset TX FCT.\n"); goto fail; } if ((error = mgb_fct_control(sc, MGB_FCT_TX_CTL, channel, FCT_ENABLE))) { device_printf(sc->dev, "Failed to enable TX FCT.\n"); goto fail; } if ((error = mgb_dmac_control(sc, MGB_DMAC_TX_START, channel, DMAC_RESET))) { device_printf(sc->dev, "Failed to reset TX DMAC.\n"); goto fail; } KASSERT(MGB_DMAC_STATE_IS_INITIAL(sc, MGB_DMAC_TX_START, channel), ("Trying to init channels in not init state\n")); /* write ring address */ if (rdata->ring_bus_addr == 0) { device_printf(sc->dev, "Invalid ring bus addr.\n"); goto fail; } CSR_WRITE_REG(sc, MGB_DMA_TX_BASE_H(channel), CSR_TRANSLATE_ADDR_HIGH32(rdata->ring_bus_addr)); CSR_WRITE_REG(sc, MGB_DMA_TX_BASE_L(channel), CSR_TRANSLATE_ADDR_LOW32(rdata->ring_bus_addr)); /* write ring size */ ring_config = CSR_READ_REG(sc, MGB_DMA_TX_CONFIG1(channel)); ring_config &= ~MGB_DMA_RING_LEN_MASK; ring_config |= (MGB_DMA_RING_SIZE & MGB_DMA_RING_LEN_MASK); CSR_WRITE_REG(sc, MGB_DMA_TX_CONFIG1(channel), ring_config); /* Enable interrupt on completion and head pointer writeback */ ring_config = (MGB_DMA_HEAD_WB_LS_ENBL | MGB_DMA_HEAD_WB_ENBL); CSR_WRITE_REG(sc, MGB_DMA_TX_CONFIG0(channel), ring_config); /* write head pointer writeback address */ if (rdata->head_wb_bus_addr == 0) { device_printf(sc->dev, "Invalid head wb bus addr.\n"); goto fail; } CSR_WRITE_REG(sc, MGB_DMA_TX_HEAD_WB_H(channel), CSR_TRANSLATE_ADDR_HIGH32(rdata->head_wb_bus_addr)); CSR_WRITE_REG(sc, MGB_DMA_TX_HEAD_WB_L(channel), CSR_TRANSLATE_ADDR_LOW32(rdata->head_wb_bus_addr)); rdata->last_head = CSR_READ_REG(sc, MGB_DMA_TX_HEAD(channel)); KASSERT(rdata->last_head == 0, ("MGB_DMA_TX_HEAD was not reset.\n")); rdata->last_tail = 0; CSR_WRITE_REG(sc, MGB_DMA_TX_TAIL(channel), rdata->last_tail); if ((error = mgb_dmac_control(sc, MGB_DMAC_TX_START, channel, DMAC_START))) device_printf(sc->dev, "Failed to start TX DMAC.\n"); fail: return error; } static int mgb_dmac_control(struct mgb_softc *sc, int start, int channel, enum mgb_dmac_cmd cmd) { int error = 0; switch (cmd) { case DMAC_RESET: CSR_WRITE_REG(sc, MGB_DMAC_CMD, MGB_DMAC_CMD_RESET(start, channel)); error = mgb_wait_for_bits(sc, MGB_DMAC_CMD, 0, MGB_DMAC_CMD_RESET(start, channel)); break; case DMAC_START: /* * NOTE: this simplifies the logic, since it will never * try to start in STOP_PENDING, but it also increases work. */ error = mgb_dmac_control(sc, start, channel, DMAC_STOP); if (error != 0) return error; CSR_WRITE_REG(sc, MGB_DMAC_CMD, MGB_DMAC_CMD_START(start, channel)); break; case DMAC_STOP: CSR_WRITE_REG(sc, MGB_DMAC_CMD, MGB_DMAC_CMD_STOP(start, channel)); error = mgb_wait_for_bits(sc, MGB_DMAC_CMD, MGB_DMAC_CMD_STOP(start, channel), MGB_DMAC_CMD_START(start, channel)); break; } return error; } static int mgb_fct_control(struct mgb_softc *sc, int reg, int channel, enum mgb_fct_cmd cmd) { switch (cmd) { case FCT_RESET: CSR_WRITE_REG(sc, reg, MGB_FCT_RESET(channel)); return mgb_wait_for_bits(sc, reg, 0, MGB_FCT_RESET(channel)); case FCT_ENABLE: CSR_WRITE_REG(sc, reg, MGB_FCT_ENBL(channel)); return (0); case FCT_DISABLE: CSR_WRITE_REG(sc, reg, MGB_FCT_DSBL(channel)); return mgb_wait_for_bits(sc, reg, 0, MGB_FCT_ENBL(channel)); } } static int mgb_hw_teardown(struct mgb_softc *sc) { int err = 0; /* Stop MAC */ CSR_CLEAR_REG(sc, MGB_MAC_RX, MGB_MAC_ENBL); CSR_WRITE_REG(sc, MGB_MAC_TX, MGB_MAC_ENBL); if ((err = mgb_wait_for_bits(sc, MGB_MAC_RX, MGB_MAC_DSBL, 0))) return (err); if ((err = mgb_wait_for_bits(sc, MGB_MAC_TX, MGB_MAC_DSBL, 0))) return (err); return (err); } static int mgb_hw_init(struct mgb_softc *sc) { int error = 0; error = mgb_hw_reset(sc); if (error != 0) goto fail; mgb_mac_init(sc); error = mgb_phy_reset(sc); if (error != 0) goto fail; error = mgb_dmac_reset(sc); if (error != 0) goto fail; fail: return error; } static int mgb_hw_reset(struct mgb_softc *sc) { CSR_UPDATE_REG(sc, MGB_HW_CFG, MGB_LITE_RESET); return (mgb_wait_for_bits(sc, MGB_HW_CFG, 0, MGB_LITE_RESET)); } static int mgb_mac_init(struct mgb_softc *sc) { /** * enable automatic duplex detection and * automatic speed detection */ CSR_UPDATE_REG(sc, MGB_MAC_CR, MGB_MAC_ADD_ENBL | MGB_MAC_ASD_ENBL); CSR_UPDATE_REG(sc, MGB_MAC_TX, MGB_MAC_ENBL); CSR_UPDATE_REG(sc, MGB_MAC_RX, MGB_MAC_ENBL); return MGB_STS_OK; } static int mgb_phy_reset(struct mgb_softc *sc) { CSR_UPDATE_BYTE(sc, MGB_PMT_CTL, MGB_PHY_RESET); if (mgb_wait_for_bits(sc, MGB_PMT_CTL, 0, MGB_PHY_RESET) == MGB_STS_TIMEOUT) return MGB_STS_TIMEOUT; return (mgb_wait_for_bits(sc, MGB_PMT_CTL, MGB_PHY_READY, 0)); } static int mgb_dmac_reset(struct mgb_softc *sc) { CSR_WRITE_REG(sc, MGB_DMAC_CMD, MGB_DMAC_RESET); return (mgb_wait_for_bits(sc, MGB_DMAC_CMD, 0, MGB_DMAC_RESET)); } static int mgb_wait_for_bits(struct mgb_softc *sc, int reg, int set_bits, int clear_bits) { int i, val; i = 0; do { /* * XXX: Datasheets states delay should be > 5 microseconds * for device reset. */ DELAY(100); val = CSR_READ_REG(sc, reg); if ((val & set_bits) == set_bits && (val & clear_bits) == 0) return MGB_STS_OK; } while (i++ < MGB_TIMEOUT); return MGB_STS_TIMEOUT; } static void mgb_get_ethaddr(struct mgb_softc *sc, struct ether_addr *dest) { CSR_READ_REG_BYTES(sc, MGB_MAC_ADDR_BASE_L, &dest->octet[0], 4); CSR_READ_REG_BYTES(sc, MGB_MAC_ADDR_BASE_H, &dest->octet[4], 2); } static int mgb_miibus_readreg(device_t dev, int phy, int reg) { struct mgb_softc *sc; int mii_access; sc = iflib_get_softc(device_get_softc(dev)); if (mgb_wait_for_bits(sc, MGB_MII_ACCESS, 0, MGB_MII_BUSY) == MGB_STS_TIMEOUT) return EIO; mii_access = (phy & MGB_MII_PHY_ADDR_MASK) << MGB_MII_PHY_ADDR_SHIFT; mii_access |= (reg & MGB_MII_REG_ADDR_MASK) << MGB_MII_REG_ADDR_SHIFT; mii_access |= MGB_MII_BUSY | MGB_MII_READ; CSR_WRITE_REG(sc, MGB_MII_ACCESS, mii_access); if (mgb_wait_for_bits(sc, MGB_MII_ACCESS, 0, MGB_MII_BUSY) == MGB_STS_TIMEOUT) return EIO; return (CSR_READ_2_BYTES(sc, MGB_MII_DATA)); } static int mgb_miibus_writereg(device_t dev, int phy, int reg, int data) { struct mgb_softc *sc; int mii_access; sc = iflib_get_softc(device_get_softc(dev)); if (mgb_wait_for_bits(sc, MGB_MII_ACCESS, 0, MGB_MII_BUSY) == MGB_STS_TIMEOUT) return EIO; mii_access = (phy & MGB_MII_PHY_ADDR_MASK) << MGB_MII_PHY_ADDR_SHIFT; mii_access |= (reg & MGB_MII_REG_ADDR_MASK) << MGB_MII_REG_ADDR_SHIFT; mii_access |= MGB_MII_BUSY | MGB_MII_WRITE; CSR_WRITE_REG(sc, MGB_MII_DATA, data); CSR_WRITE_REG(sc, MGB_MII_ACCESS, mii_access); if (mgb_wait_for_bits(sc, MGB_MII_ACCESS, 0, MGB_MII_BUSY) == MGB_STS_TIMEOUT) return EIO; return 0; } /* XXX: May need to lock these up */ static void mgb_miibus_statchg(device_t dev) { struct mgb_softc *sc; struct mii_data *miid; sc = iflib_get_softc(device_get_softc(dev)); miid = device_get_softc(sc->miibus); /* Update baudrate in iflib */ sc->baudrate = ifmedia_baudrate(miid->mii_media_active); iflib_link_state_change(sc->ctx, sc->link_state, sc->baudrate); } static void mgb_miibus_linkchg(device_t dev) { struct mgb_softc *sc; struct mii_data *miid; int link_state; sc = iflib_get_softc(device_get_softc(dev)); miid = device_get_softc(sc->miibus); /* XXX: copied from miibus_linkchg **/ if (miid->mii_media_status & IFM_AVALID) { if (miid->mii_media_status & IFM_ACTIVE) link_state = LINK_STATE_UP; else link_state = LINK_STATE_DOWN; } else link_state = LINK_STATE_UNKNOWN; sc->link_state = link_state; iflib_link_state_change(sc->ctx, sc->link_state, sc->baudrate); } Index: head/sys/dev/mgb/if_mgb.h =================================================================== --- head/sys/dev/mgb/if_mgb.h (revision 365060) +++ head/sys/dev/mgb/if_mgb.h (revision 365061) @@ -1,332 +1,333 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2019 The FreeBSD Foundation, Inc. * * This driver was written by Gerald ND Aryeetey * under sponsorship from the FreeBSD Foundation. * * 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. * * $FreeBSD$ */ #ifndef _IF_MGB_H_ #define _IF_MGB_H_ #define MGB_MICROCHIP_VENDOR_ID 0x1055 #define MGB_LAN7430_DEVICE_ID 0x7430 #define MGB_LAN7431_DEVICE_ID 0x7431 #define MGB_TIMEOUT (500) /** Control/Status Registers **/ #define MGB_BAR 0 /* PCI Base Address */ /** Reset **/ #define MGB_HW_CFG 0x10 /** H/W Configuration Register **/ #define MGB_LITE_RESET 0x2 /** MAC **/ #define MGB_MAC_CR 0x0100 /** MAC Crontrol Register **/ #define MGB_MAC_ADD_ENBL 0x1000 /* Automatic Duplex Detection */ #define MGB_MAC_ASD_ENBL 0x0800 /* Automatic Speed Detection */ #define MGB_MAC_ADDR_BASE_L 0x11C /** MAC address lower 4 bytes (read) register **/ #define MGB_MAC_ADDR_BASE_H 0x118 /** MAC address upper 2 bytes (read) register **/ #define MGB_MAC_TX 0x0104 #define MGB_MAC_RX 0x0108 #define MGB_MAC_ENBL (1 << 0) #define MGB_MAC_DSBL (1 << 1) /** MAC Statistics **/ #define MGB_MAC_STAT_RX_FCS_ERR_CNT 0x1200 #define MGB_MAC_STAT_RX_ALIGN_ERR_CNT 0x1204 #define MGB_MAC_STAT_RX_FRAG_ERR_CNT 0x1208 #define MGB_MAC_STAT_RX_JABBER_ERR_CNT 0x120C #define MGB_MAC_STAT_RX_UNDER_ERR_CNT 0x1210 #define MGB_MAC_STAT_RX_OVER_ERR_CNT 0x1214 #define MGB_MAC_STAT_RX_DROPPED_CNT 0x1218 #define MGB_MAC_STAT_RX_BROADCAST_CNT1 0x1220 #define MGB_MAC_STAT_RX_BROADCAST_CNT 0x122C #define MGB_MAC_STAT_RX_FRAME_CNT 0x1254 #define MGB_MAC_STAT_RX_DROPPED_CNT 0x1218 #define MGB_MAC_STAT_RX_BROADCAST_CNT1 0x1220 #define MGB_MAC_STAT_RX_BROADCAST_CNT 0x122C #define MGB_MAC_STAT_RX_FRAME_CNT 0x1254 /* etc. */ /** Receive Filtering Engine **/ #define MGB_RFE_CTL 0x508 #define MGB_RFE_ALLOW_BROADCAST (1 << 10) #define MGB_RFE_ALLOW_MULTICAST (1 << 9) #define MGB_RFE_ALLOW_UNICAST (1 << 8) #define MGB_RFE_ALLOW_PERFECT_FILTER (1 << 1) /** PHY Reset (via power management control) **/ #define MGB_PMT_CTL 0x14 /** Power Management Control Register **/ #define MGB_PHY_RESET 0x10 #define MGB_PHY_READY 0x80 /** FIFO Controller **/ #define MGB_FCT_TX_CTL 0xC4 #define MGB_FCT_RX_CTL 0xAC #define MGB_FCT_ENBL(_channel) (1 << (28 + (_channel))) #define MGB_FCT_DSBL(_channel) (1 << (24 + (_channel))) #define MGB_FCT_RESET(_channel) (1 << (20 + (_channel))) /** DMA Controller **/ #define MGB_DMAC_CMD 0xC0C #define MGB_DMAC_RESET (1 << 31) #define MGB_DMAC_TX_START 16 #define MGB_DMAC_RX_START 0 #define MGB_DMAC_CMD_VAL(s, o, ch) (1 << ((s) + (o) + (ch))) #define MGB_DMAC_CMD_RESET(_s, _ch) MGB_DMAC_CMD_VAL(_s, 8, _ch) #define MGB_DMAC_CMD_START(_s, _ch) MGB_DMAC_CMD_VAL(_s, 4, _ch) #define MGB_DMAC_CMD_STOP( _s, _ch) MGB_DMAC_CMD_VAL(_s, 0, _ch) #define MGB_DMAC_STATE(_start, _stop) \ (((_start) ? 2 : 0) | ((_stop) ? 1 : 0)) #define MGB_DMAC_STATE_INITIAL MGB_DMAC_STATE(0, 0) #define MGB_DMAC_STATE_STARTED MGB_DMAC_STATE(1, 0) #define MGB_DMAC_STATE_STOP_PENDING MGB_DMAC_STATE(1, 1) #define MGB_DMAC_STATE_STOPPED MGB_DMAC_STATE(0, 1) #define MGB_DMAC_CMD_STATE(sc, _s, _ch) \ (MGB_DMAC_STATE( \ CSR_READ_REG(sc, MGB_DMAC_CMD) & MGB_DMAC_CMD_START(_s, _ch), \ CSR_READ_REG(sc, MGB_DMAC_CMD) & MGB_DMAC_CMD_STOP(_s, _ch))) #define MGB_DMAC_STATE_IS_INITIAL(sc, _s, _ch) \ (MGB_DMAC_CMD_STATE(sc, _s, _ch) == MGB_DMAC_STATE_INITIAL) #define MGB_DMAC_INTR_STS 0xC10 #define MGB_DMAC_INTR_ENBL_SET 0xC14 #define MGB_DMAC_INTR_ENBL_CLR 0xC18 #define MGB_DMAC_TX_INTR_ENBL(_ch) (1 << (_ch)) #define MGB_DMAC_RX_INTR_ENBL(_ch) (1 << (16 + (_ch))) /** DMA Rings **/ /** * Page size is 256 bytes * * Ring size, however, these could be tunable (for RX & TX) * to be a multiple of 4 (max is 65532) * **/ /* In linux driver these numbers are 50 and 65 for tx and rx .... */ #define MGB_DMA_RING_SIZE 16 /* in programming guide, this number is 100 */ #define MGB_DMA_MAXSEGS 32 #define MGB_DMA_REG(reg, _channel) ((reg) | ((_channel) << 6)) #define MGB_DMA_RING_LIST_SIZE \ (sizeof(struct mgb_ring_desc) * MGB_DMA_RING_SIZE) #define MGB_DMA_RING_INFO_SIZE \ (sizeof(uint32_t) + MGB_DMA_RING_LIST_SIZE) #define MGB_DMA_TX_CONFIG0(_channel) MGB_DMA_REG(0x0D40, _channel) #define MGB_DMA_TX_CONFIG1(_channel) MGB_DMA_REG(0x0D44, _channel) #define MGB_DMA_TX_BASE_H(_channel) MGB_DMA_REG(0x0D48, _channel) #define MGB_DMA_TX_BASE_L(_channel) MGB_DMA_REG(0x0D4C, _channel) #define MGB_DMA_TX_HEAD_WB_H(_channel) MGB_DMA_REG(0x0D50, _channel) /* head Writeback */ #define MGB_DMA_TX_HEAD_WB_L(_channel) MGB_DMA_REG(0x0D54, _channel) #define MGB_DMA_TX_HEAD(_channel) MGB_DMA_REG(0x0D58, _channel) #define MGB_DMA_TX_TAIL(_channel) MGB_DMA_REG(0x0D5C, _channel) #define MGB_DMA_RX_CONFIG0(_channel) MGB_DMA_REG(0x0C40, _channel) #define MGB_DMA_RX_CONFIG1(_channel) MGB_DMA_REG(0x0C44, _channel) #define MGB_DMA_RX_BASE_H(_channel) MGB_DMA_REG(0x0C48, _channel) #define MGB_DMA_RX_BASE_L(_channel) MGB_DMA_REG(0x0C4C, _channel) #define MGB_DMA_RX_HEAD_WB_H(_channel) MGB_DMA_REG(0x0C50, _channel) /* head Writeback */ #define MGB_DMA_RX_HEAD_WB_L(_channel) MGB_DMA_REG(0x0C54, _channel) #define MGB_DMA_RX_HEAD(_channel) MGB_DMA_REG(0x0C58, _channel) #define MGB_DMA_RX_TAIL(_channel) MGB_DMA_REG(0x0C5C, _channel) #define MGB_DMA_RING_LEN_MASK 0xFFFF #define MGB_DMA_IOC_ENBL 0x10000000 #define MGB_DMA_HEAD_WB_LS_ENBL 0x20000000 #define MGB_DMA_HEAD_WB_ENBL (1 << 5) #define MGB_DMA_RING_PAD_MASK 0x03000000 #define MGB_DMA_RING_PAD_0 0x00000000 #define MGB_DMA_RING_PAD_2 0x02000000 #define MGB_DESC_CTL_OWN (1 << 15) #define MGB_DESC_CTL_FCS (1 << 17) #define MGB_DESC_CTL_IOC (1 << 26) #define MGB_TX_DESC_CTL_LS (1 << 28) #define MGB_TX_DESC_CTL_FS (1 << 29) #define MGB_RX_DESC_CTL_LS (1 << 30) #define MGB_RX_DESC_CTL_FS (1 << 31) #define MGB_DESC_CTL_BUFLEN_MASK (0x0000FFFF) #define MGB_DESC_STS_BUFLEN_MASK (0x00003FFF) #define MGB_DESC_FRAME_LEN_MASK (0x3FFF0000) #define MGB_DESC_GET_FRAME_LEN(_desc) \ (((_desc)->ctl & MGB_DESC_FRAME_LEN_MASK) >> 16) -#define MGB_NEXT_RING_IDX(_idx) (((_idx) + 1) % MGB_DMA_RING_SIZE) +#define MGB_NEXT_RING_IDX(_idx) (((_idx) == MGB_DMA_RING_SIZE - 1) ? 0 : ((_idx_) + 1)) +#define MGB_PREV_RING_IDX(_idx) (((_idx) == 0) ? (MGB_DMA_RING_SIZE - 1) : ((_idx_) - 1)) #define MGB_RING_SPACE(_sc) \ ((((_sc)->tx_ring_data.last_head - (_sc)->tx_ring_data.last_tail - 1) \ + MGB_DMA_RING_SIZE ) % MGB_DMA_RING_SIZE ) /** PHY **/ #define MGB_MII_ACCESS 0x120 #define MGB_MII_DATA 0x124 #define MGB_MII_PHY_ADDR_MASK 0x1F #define MGB_MII_PHY_ADDR_SHIFT 11 #define MGB_MII_REG_ADDR_MASK 0x1F #define MGB_MII_REG_ADDR_SHIFT 6 #define MGB_MII_READ 0x0 #define MGB_MII_WRITE 0x2 #define MGB_MII_BUSY 0x1 /** Interrupt registers **/ #define MGB_INTR_STS 0x780 #define MGB_INTR_SET 0x784 /* This triggers a particular interrupt */ #define MGB_INTR_ENBL_SET 0x788 #define MGB_INTR_STS_ANY (0x1) #define MGB_INTR_STS_RX(_channel) (1 << (24 + (_channel))) #define MGB_INTR_STS_RX_ANY (0xF << 24) #define MGB_INTR_STS_TX(_channel) (1 << (16 + (_channel))) #define MGB_INTR_STS_TX_ANY (0xF << 16) #define MGB_INTR_STS_TEST (1 << 9) #define MGB_INTR_ENBL_CLR 0x78C #define MGB_INTR_VEC_ENBL_SET 0x794 #define MGB_INTR_VEC_ENBL_CLR 0x798 #define MGB_INTR_VEC_ENBL_AUTO_CLR 0x79C #define MGB_INTR_VEC_RX_MAP 0x7A0 #define MGB_INTR_VEC_TX_MAP 0x7A4 #define MGB_INTR_VEC_OTHER_MAP 0x7A8 #define MGB_INTR_VEC_MAP(_vsts, _ch) ((_vsts) << ((_ch) << 2)) #define MGB_INTR_VEC_STS(_v) (1 << (_v)) #define MGB_INTR_RX_VEC_STS(_qid) MGB_INTR_VEC_STS((_qid) + 1) #define MGB_STS_OK ( 0 ) #define MGB_STS_TIMEOUT (-1 ) #define CSR_READ_BYTE(sc, reg) \ bus_read_1((sc)->regs, reg) #define CSR_WRITE_BYTE(sc, reg, val) \ bus_write_1((sc)->regs, reg, val) #define CSR_UPDATE_BYTE(sc, reg, val) \ CSR_WRITE_BYTE(sc, reg, CSR_READ_BYTE(sc, reg) | (val)) #define CSR_READ_REG(sc, reg) \ bus_read_4((sc)->regs, reg) #define CSR_WRITE_REG(sc, reg, val) \ bus_write_4((sc)->regs, reg, val) #define CSR_CLEAR_REG(sc, reg, bits) \ CSR_WRITE_REG(sc, reg, CSR_READ_REG(sc, reg) & ~(bits)) #define CSR_UPDATE_REG(sc, reg, val) \ CSR_WRITE_REG(sc, reg, CSR_READ_REG(sc, reg) | (val)) #define CSR_READ_2_BYTES(sc, reg) \ bus_read_2((sc)->regs, reg) #define CSR_READ_REG_BYTES(sc, reg, dest, cnt) \ bus_read_region_1((sc)->regs, reg, dest, cnt) #define CSR_TRANSLATE_ADDR_LOW32(addr) ((uint64_t) (addr) & 0xFFFFFFFF) #define CSR_TRANSLATE_ADDR_HIGH32(addr) ((uint64_t) (addr) >> 32) struct mgb_irq { struct resource *res; void *handler; }; enum mgb_dmac_cmd { DMAC_RESET, DMAC_START, DMAC_STOP }; enum mgb_fct_cmd { FCT_RESET, FCT_ENABLE, FCT_DISABLE }; struct mgb_ring_desc_addr { uint32_t low; uint32_t high; } __packed; /* TODO: With descriptor bit information * this could be done without masks etc. * (using bitwise structs like vmx, * would have to separate rx/tx ring desc * definitions) */ struct mgb_ring_desc { uint32_t ctl; /* data0 */ struct mgb_ring_desc_addr addr; /* data(1|2) */ uint32_t sts; /* data3 */ } __packed; #if 0 struct mgb_ring_info { uint32_t head_wb; struct mgb_ring_desc *ring; } #endif #define MGB_HEAD_WB_PTR(_ring_info_ptr) \ ((uint32_t *)(_ring_info_ptr)) #define MGB_RING_PTR(_ring_info_ptr) \ ((struct mgb_ring_desc *)(MGB_HEAD_WB_PTR(_ring_info_ptr) + 1)) struct mgb_ring_data { uint32_t *head_wb; struct mgb_ring_desc *ring; bus_addr_t head_wb_bus_addr; bus_addr_t ring_bus_addr; uint32_t last_head; uint32_t last_tail; }; struct mgb_softc { if_ctx_t ctx; device_t dev; struct resource *regs; struct resource *pba; struct if_irq admin_irq; struct if_irq rx_irq; bool isr_test_flag; device_t miibus; int link_state; int baudrate; int if_flags; int ethaddr; int flags; struct mtx mtx; struct callout watchdog; int timer; bus_dma_tag_t dma_parent_tag; struct mgb_ring_data rx_ring_data; struct mgb_ring_data tx_ring_data; }; #endif /* _IF_MGB_H_ */ Index: head/sys/dev/vmware/vmxnet3/if_vmx.c =================================================================== --- head/sys/dev/vmware/vmxnet3/if_vmx.c (revision 365060) +++ head/sys/dev/vmware/vmxnet3/if_vmx.c (revision 365061) @@ -1,2547 +1,2540 @@ /*- * Copyright (c) 2013 Tsubai Masanari * Copyright (c) 2013 Bryan Venteicher * Copyright (c) 2018 Patrick Kelsey * * Permission to use, copy, modify, and distribute this software for any * purpose with or without fee is hereby granted, provided that the above * copyright notice and this permission notice appear in all copies. * * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. * * $OpenBSD: src/sys/dev/pci/if_vmx.c,v 1.11 2013/06/22 00:28:10 uebayasi Exp $ */ /* Driver for VMware vmxnet3 virtual ethernet devices. */ #include __FBSDID("$FreeBSD$"); #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef RSS #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ifdi_if.h" #include "if_vmxreg.h" #include "if_vmxvar.h" #include "opt_inet.h" #include "opt_inet6.h" #define VMXNET3_VMWARE_VENDOR_ID 0x15AD #define VMXNET3_VMWARE_DEVICE_ID 0x07B0 static pci_vendor_info_t vmxnet3_vendor_info_array[] = { PVID(VMXNET3_VMWARE_VENDOR_ID, VMXNET3_VMWARE_DEVICE_ID, "VMware VMXNET3 Ethernet Adapter"), /* required last entry */ PVID_END }; static void *vmxnet3_register(device_t); static int vmxnet3_attach_pre(if_ctx_t); static int vmxnet3_msix_intr_assign(if_ctx_t, int); static void vmxnet3_free_irqs(struct vmxnet3_softc *); static int vmxnet3_attach_post(if_ctx_t); static int vmxnet3_detach(if_ctx_t); static int vmxnet3_shutdown(if_ctx_t); static int vmxnet3_suspend(if_ctx_t); static int vmxnet3_resume(if_ctx_t); static int vmxnet3_alloc_resources(struct vmxnet3_softc *); static void vmxnet3_free_resources(struct vmxnet3_softc *); static int vmxnet3_check_version(struct vmxnet3_softc *); static void vmxnet3_set_interrupt_idx(struct vmxnet3_softc *); static int vmxnet3_queues_shared_alloc(struct vmxnet3_softc *); static void vmxnet3_init_txq(struct vmxnet3_softc *, int); static int vmxnet3_tx_queues_alloc(if_ctx_t, caddr_t *, uint64_t *, int, int); static void vmxnet3_init_rxq(struct vmxnet3_softc *, int, int); static int vmxnet3_rx_queues_alloc(if_ctx_t, caddr_t *, uint64_t *, int, int); static void vmxnet3_queues_free(if_ctx_t); static int vmxnet3_alloc_shared_data(struct vmxnet3_softc *); static void vmxnet3_free_shared_data(struct vmxnet3_softc *); static int vmxnet3_alloc_mcast_table(struct vmxnet3_softc *); static void vmxnet3_free_mcast_table(struct vmxnet3_softc *); static void vmxnet3_init_shared_data(struct vmxnet3_softc *); static void vmxnet3_reinit_rss_shared_data(struct vmxnet3_softc *); static void vmxnet3_reinit_shared_data(struct vmxnet3_softc *); static int vmxnet3_alloc_data(struct vmxnet3_softc *); static void vmxnet3_free_data(struct vmxnet3_softc *); static void vmxnet3_evintr(struct vmxnet3_softc *); static int vmxnet3_isc_txd_encap(void *, if_pkt_info_t); static void vmxnet3_isc_txd_flush(void *, uint16_t, qidx_t); static int vmxnet3_isc_txd_credits_update(void *, uint16_t, bool); static int vmxnet3_isc_rxd_available(void *, uint16_t, qidx_t, qidx_t); static int vmxnet3_isc_rxd_pkt_get(void *, if_rxd_info_t); static void vmxnet3_isc_rxd_refill(void *, if_rxd_update_t); static void vmxnet3_isc_rxd_flush(void *, uint16_t, uint8_t, qidx_t); static int vmxnet3_legacy_intr(void *); static int vmxnet3_rxq_intr(void *); static int vmxnet3_event_intr(void *); static void vmxnet3_stop(if_ctx_t); static void vmxnet3_txinit(struct vmxnet3_softc *, struct vmxnet3_txqueue *); static void vmxnet3_rxinit(struct vmxnet3_softc *, struct vmxnet3_rxqueue *); static void vmxnet3_reinit_queues(struct vmxnet3_softc *); static int vmxnet3_enable_device(struct vmxnet3_softc *); static void vmxnet3_reinit_rxfilters(struct vmxnet3_softc *); static void vmxnet3_init(if_ctx_t); static void vmxnet3_multi_set(if_ctx_t); static int vmxnet3_mtu_set(if_ctx_t, uint32_t); static void vmxnet3_media_status(if_ctx_t, struct ifmediareq *); static int vmxnet3_media_change(if_ctx_t); static int vmxnet3_promisc_set(if_ctx_t, int); static uint64_t vmxnet3_get_counter(if_ctx_t, ift_counter); static void vmxnet3_update_admin_status(if_ctx_t); static void vmxnet3_txq_timer(if_ctx_t, uint16_t); static void vmxnet3_update_vlan_filter(struct vmxnet3_softc *, int, uint16_t); static void vmxnet3_vlan_register(if_ctx_t, uint16_t); static void vmxnet3_vlan_unregister(if_ctx_t, uint16_t); static void vmxnet3_set_rxfilter(struct vmxnet3_softc *, int); static void vmxnet3_refresh_host_stats(struct vmxnet3_softc *); static int vmxnet3_link_is_up(struct vmxnet3_softc *); static void vmxnet3_link_status(struct vmxnet3_softc *); static void vmxnet3_set_lladdr(struct vmxnet3_softc *); static void vmxnet3_get_lladdr(struct vmxnet3_softc *); static void vmxnet3_setup_txq_sysctl(struct vmxnet3_txqueue *, struct sysctl_ctx_list *, struct sysctl_oid_list *); static void vmxnet3_setup_rxq_sysctl(struct vmxnet3_rxqueue *, struct sysctl_ctx_list *, struct sysctl_oid_list *); static void vmxnet3_setup_queue_sysctl(struct vmxnet3_softc *, struct sysctl_ctx_list *, struct sysctl_oid_list *); static void vmxnet3_setup_sysctl(struct vmxnet3_softc *); static void vmxnet3_write_bar0(struct vmxnet3_softc *, bus_size_t, uint32_t); static uint32_t vmxnet3_read_bar1(struct vmxnet3_softc *, bus_size_t); static void vmxnet3_write_bar1(struct vmxnet3_softc *, bus_size_t, uint32_t); static void vmxnet3_write_cmd(struct vmxnet3_softc *, uint32_t); static uint32_t vmxnet3_read_cmd(struct vmxnet3_softc *, uint32_t); static int vmxnet3_tx_queue_intr_enable(if_ctx_t, uint16_t); static int vmxnet3_rx_queue_intr_enable(if_ctx_t, uint16_t); static void vmxnet3_link_intr_enable(if_ctx_t); static void vmxnet3_enable_intr(struct vmxnet3_softc *, int); static void vmxnet3_disable_intr(struct vmxnet3_softc *, int); static void vmxnet3_intr_enable_all(if_ctx_t); static void vmxnet3_intr_disable_all(if_ctx_t); typedef enum { VMXNET3_BARRIER_RD, VMXNET3_BARRIER_WR, VMXNET3_BARRIER_RDWR, } vmxnet3_barrier_t; static void vmxnet3_barrier(struct vmxnet3_softc *, vmxnet3_barrier_t); static device_method_t vmxnet3_methods[] = { /* Device interface */ DEVMETHOD(device_register, vmxnet3_register), DEVMETHOD(device_probe, iflib_device_probe), DEVMETHOD(device_attach, iflib_device_attach), DEVMETHOD(device_detach, iflib_device_detach), DEVMETHOD(device_shutdown, iflib_device_shutdown), DEVMETHOD(device_suspend, iflib_device_suspend), DEVMETHOD(device_resume, iflib_device_resume), DEVMETHOD_END }; static driver_t vmxnet3_driver = { "vmx", vmxnet3_methods, sizeof(struct vmxnet3_softc) }; static devclass_t vmxnet3_devclass; DRIVER_MODULE(vmx, pci, vmxnet3_driver, vmxnet3_devclass, 0, 0); IFLIB_PNP_INFO(pci, vmx, vmxnet3_vendor_info_array); MODULE_VERSION(vmx, 2); MODULE_DEPEND(vmx, pci, 1, 1, 1); MODULE_DEPEND(vmx, ether, 1, 1, 1); MODULE_DEPEND(vmx, iflib, 1, 1, 1); static device_method_t vmxnet3_iflib_methods[] = { DEVMETHOD(ifdi_tx_queues_alloc, vmxnet3_tx_queues_alloc), DEVMETHOD(ifdi_rx_queues_alloc, vmxnet3_rx_queues_alloc), DEVMETHOD(ifdi_queues_free, vmxnet3_queues_free), DEVMETHOD(ifdi_attach_pre, vmxnet3_attach_pre), DEVMETHOD(ifdi_attach_post, vmxnet3_attach_post), DEVMETHOD(ifdi_detach, vmxnet3_detach), DEVMETHOD(ifdi_init, vmxnet3_init), DEVMETHOD(ifdi_stop, vmxnet3_stop), DEVMETHOD(ifdi_multi_set, vmxnet3_multi_set), DEVMETHOD(ifdi_mtu_set, vmxnet3_mtu_set), DEVMETHOD(ifdi_media_status, vmxnet3_media_status), DEVMETHOD(ifdi_media_change, vmxnet3_media_change), DEVMETHOD(ifdi_promisc_set, vmxnet3_promisc_set), DEVMETHOD(ifdi_get_counter, vmxnet3_get_counter), DEVMETHOD(ifdi_update_admin_status, vmxnet3_update_admin_status), DEVMETHOD(ifdi_timer, vmxnet3_txq_timer), DEVMETHOD(ifdi_tx_queue_intr_enable, vmxnet3_tx_queue_intr_enable), DEVMETHOD(ifdi_rx_queue_intr_enable, vmxnet3_rx_queue_intr_enable), DEVMETHOD(ifdi_link_intr_enable, vmxnet3_link_intr_enable), DEVMETHOD(ifdi_intr_enable, vmxnet3_intr_enable_all), DEVMETHOD(ifdi_intr_disable, vmxnet3_intr_disable_all), DEVMETHOD(ifdi_msix_intr_assign, vmxnet3_msix_intr_assign), DEVMETHOD(ifdi_vlan_register, vmxnet3_vlan_register), DEVMETHOD(ifdi_vlan_unregister, vmxnet3_vlan_unregister), DEVMETHOD(ifdi_shutdown, vmxnet3_shutdown), DEVMETHOD(ifdi_suspend, vmxnet3_suspend), DEVMETHOD(ifdi_resume, vmxnet3_resume), DEVMETHOD_END }; static driver_t vmxnet3_iflib_driver = { "vmx", vmxnet3_iflib_methods, sizeof(struct vmxnet3_softc) }; struct if_txrx vmxnet3_txrx = { .ift_txd_encap = vmxnet3_isc_txd_encap, .ift_txd_flush = vmxnet3_isc_txd_flush, .ift_txd_credits_update = vmxnet3_isc_txd_credits_update, .ift_rxd_available = vmxnet3_isc_rxd_available, .ift_rxd_pkt_get = vmxnet3_isc_rxd_pkt_get, .ift_rxd_refill = vmxnet3_isc_rxd_refill, .ift_rxd_flush = vmxnet3_isc_rxd_flush, .ift_legacy_intr = vmxnet3_legacy_intr }; static struct if_shared_ctx vmxnet3_sctx_init = { .isc_magic = IFLIB_MAGIC, .isc_q_align = 512, .isc_tx_maxsize = VMXNET3_TX_MAXSIZE, .isc_tx_maxsegsize = VMXNET3_TX_MAXSEGSIZE, .isc_tso_maxsize = VMXNET3_TSO_MAXSIZE + sizeof(struct ether_vlan_header), .isc_tso_maxsegsize = VMXNET3_TX_MAXSEGSIZE, /* * These values are used to configure the busdma tag used for * receive descriptors. Each receive descriptor only points to one * buffer. */ .isc_rx_maxsize = VMXNET3_RX_MAXSEGSIZE, /* One buf per descriptor */ .isc_rx_nsegments = 1, /* One mapping per descriptor */ .isc_rx_maxsegsize = VMXNET3_RX_MAXSEGSIZE, .isc_admin_intrcnt = 1, .isc_vendor_info = vmxnet3_vendor_info_array, .isc_driver_version = "2", .isc_driver = &vmxnet3_iflib_driver, .isc_flags = IFLIB_HAS_RXCQ | IFLIB_HAS_TXCQ | IFLIB_SINGLE_IRQ_RX_ONLY, /* * Number of receive queues per receive queue set, with associated * descriptor settings for each. */ .isc_nrxqs = 3, .isc_nfl = 2, /* one free list for each receive command queue */ .isc_nrxd_min = {VMXNET3_MIN_RX_NDESC, VMXNET3_MIN_RX_NDESC, VMXNET3_MIN_RX_NDESC}, .isc_nrxd_max = {VMXNET3_MAX_RX_NDESC, VMXNET3_MAX_RX_NDESC, VMXNET3_MAX_RX_NDESC}, .isc_nrxd_default = {VMXNET3_DEF_RX_NDESC, VMXNET3_DEF_RX_NDESC, VMXNET3_DEF_RX_NDESC}, /* * Number of transmit queues per transmit queue set, with associated * descriptor settings for each. */ .isc_ntxqs = 2, .isc_ntxd_min = {VMXNET3_MIN_TX_NDESC, VMXNET3_MIN_TX_NDESC}, .isc_ntxd_max = {VMXNET3_MAX_TX_NDESC, VMXNET3_MAX_TX_NDESC}, .isc_ntxd_default = {VMXNET3_DEF_TX_NDESC, VMXNET3_DEF_TX_NDESC}, }; static void * vmxnet3_register(device_t dev) { return (&vmxnet3_sctx_init); } static int trunc_powerof2(int val) { return (1U << (fls(val) - 1)); } static int vmxnet3_attach_pre(if_ctx_t ctx) { device_t dev; if_softc_ctx_t scctx; struct vmxnet3_softc *sc; uint32_t intr_config; int error; dev = iflib_get_dev(ctx); sc = iflib_get_softc(ctx); sc->vmx_dev = dev; sc->vmx_ctx = ctx; sc->vmx_sctx = iflib_get_sctx(ctx); sc->vmx_scctx = iflib_get_softc_ctx(ctx); sc->vmx_ifp = iflib_get_ifp(ctx); sc->vmx_media = iflib_get_media(ctx); scctx = sc->vmx_scctx; scctx->isc_tx_nsegments = VMXNET3_TX_MAXSEGS; scctx->isc_tx_tso_segments_max = VMXNET3_TX_MAXSEGS; /* isc_tx_tso_size_max doesn't include possible vlan header */ scctx->isc_tx_tso_size_max = VMXNET3_TSO_MAXSIZE; scctx->isc_tx_tso_segsize_max = VMXNET3_TX_MAXSEGSIZE; scctx->isc_txrx = &vmxnet3_txrx; /* If 0, the iflib tunable was not set, so set to the default */ if (scctx->isc_nrxqsets == 0) scctx->isc_nrxqsets = VMXNET3_DEF_RX_QUEUES; scctx->isc_nrxqsets = trunc_powerof2(scctx->isc_nrxqsets); scctx->isc_nrxqsets_max = min(VMXNET3_MAX_RX_QUEUES, mp_ncpus); scctx->isc_nrxqsets_max = trunc_powerof2(scctx->isc_nrxqsets_max); /* If 0, the iflib tunable was not set, so set to the default */ if (scctx->isc_ntxqsets == 0) scctx->isc_ntxqsets = VMXNET3_DEF_TX_QUEUES; scctx->isc_ntxqsets = trunc_powerof2(scctx->isc_ntxqsets); scctx->isc_ntxqsets_max = min(VMXNET3_MAX_TX_QUEUES, mp_ncpus); scctx->isc_ntxqsets_max = trunc_powerof2(scctx->isc_ntxqsets_max); /* * Enforce that the transmit completion queue descriptor count is * the same as the transmit command queue descriptor count. */ scctx->isc_ntxd[0] = scctx->isc_ntxd[1]; scctx->isc_txqsizes[0] = sizeof(struct vmxnet3_txcompdesc) * scctx->isc_ntxd[0]; scctx->isc_txqsizes[1] = sizeof(struct vmxnet3_txdesc) * scctx->isc_ntxd[1]; /* * Enforce that the receive completion queue descriptor count is the * sum of the receive command queue descriptor counts, and that the * second receive command queue descriptor count is the same as the * first one. */ scctx->isc_nrxd[2] = scctx->isc_nrxd[1]; scctx->isc_nrxd[0] = scctx->isc_nrxd[1] + scctx->isc_nrxd[2]; scctx->isc_rxqsizes[0] = sizeof(struct vmxnet3_rxcompdesc) * scctx->isc_nrxd[0]; scctx->isc_rxqsizes[1] = sizeof(struct vmxnet3_rxdesc) * scctx->isc_nrxd[1]; scctx->isc_rxqsizes[2] = sizeof(struct vmxnet3_rxdesc) * scctx->isc_nrxd[2]; /* * Initialize the max frame size and descriptor queue buffer * sizes. */ vmxnet3_mtu_set(ctx, if_getmtu(sc->vmx_ifp)); scctx->isc_rss_table_size = UPT1_RSS_MAX_IND_TABLE_SIZE; /* Map PCI BARs */ error = vmxnet3_alloc_resources(sc); if (error) goto fail; /* Check device versions */ error = vmxnet3_check_version(sc); if (error) goto fail; /* * The interrupt mode can be set in the hypervisor configuration via * the parameter ethernet.intrMode. */ intr_config = vmxnet3_read_cmd(sc, VMXNET3_CMD_GET_INTRCFG); sc->vmx_intr_mask_mode = (intr_config >> 2) & 0x03; /* * Configure the softc context to attempt to configure the interrupt * mode now indicated by intr_config. iflib will follow the usual * fallback path MSI-X -> MSI -> LEGACY, starting at the configured * starting mode. */ switch (intr_config & 0x03) { case VMXNET3_IT_AUTO: case VMXNET3_IT_MSIX: scctx->isc_msix_bar = pci_msix_table_bar(dev); break; case VMXNET3_IT_MSI: scctx->isc_msix_bar = -1; scctx->isc_disable_msix = 1; break; case VMXNET3_IT_LEGACY: scctx->isc_msix_bar = 0; break; } scctx->isc_tx_csum_flags = VMXNET3_CSUM_ALL_OFFLOAD; scctx->isc_capabilities = scctx->isc_capenable = IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 | IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO | IFCAP_JUMBO_MTU; /* These capabilities are not enabled by default. */ scctx->isc_capabilities |= IFCAP_LRO | IFCAP_VLAN_HWFILTER; vmxnet3_get_lladdr(sc); iflib_set_mac(ctx, sc->vmx_lladdr); return (0); fail: /* * We must completely clean up anything allocated above as iflib * will not invoke any other driver entry points as a result of this * failure. */ vmxnet3_free_resources(sc); return (error); } static int vmxnet3_msix_intr_assign(if_ctx_t ctx, int msix) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; struct vmxnet3_rxqueue *rxq; int error; int i; char irq_name[16]; sc = iflib_get_softc(ctx); scctx = sc->vmx_scctx; for (i = 0; i < scctx->isc_nrxqsets; i++) { snprintf(irq_name, sizeof(irq_name), "rxq%d", i); rxq = &sc->vmx_rxq[i]; error = iflib_irq_alloc_generic(ctx, &rxq->vxrxq_irq, i + 1, IFLIB_INTR_RX, vmxnet3_rxq_intr, rxq, i, irq_name); if (error) { device_printf(iflib_get_dev(ctx), "Failed to register rxq %d interrupt handler\n", i); return (error); } } for (i = 0; i < scctx->isc_ntxqsets; i++) { snprintf(irq_name, sizeof(irq_name), "txq%d", i); /* * Don't provide the corresponding rxq irq for reference - * we want the transmit task to be attached to a task queue * that is different from the one used by the corresponding * rxq irq. That is because the TX doorbell writes are very * expensive as virtualized MMIO operations, so we want to * be able to defer them to another core when possible so * that they don't steal receive processing cycles during * stack turnarounds like TCP ACK generation. The other * piece to this approach is enabling the iflib abdicate * option (currently via an interface-specific * tunable/sysctl). */ iflib_softirq_alloc_generic(ctx, NULL, IFLIB_INTR_TX, NULL, i, irq_name); } error = iflib_irq_alloc_generic(ctx, &sc->vmx_event_intr_irq, scctx->isc_nrxqsets + 1, IFLIB_INTR_ADMIN, vmxnet3_event_intr, sc, 0, "event"); if (error) { device_printf(iflib_get_dev(ctx), "Failed to register event interrupt handler\n"); return (error); } return (0); } static void vmxnet3_free_irqs(struct vmxnet3_softc *sc) { if_softc_ctx_t scctx; struct vmxnet3_rxqueue *rxq; int i; scctx = sc->vmx_scctx; for (i = 0; i < scctx->isc_nrxqsets; i++) { rxq = &sc->vmx_rxq[i]; iflib_irq_free(sc->vmx_ctx, &rxq->vxrxq_irq); } iflib_irq_free(sc->vmx_ctx, &sc->vmx_event_intr_irq); } static int vmxnet3_attach_post(if_ctx_t ctx) { device_t dev; if_softc_ctx_t scctx; struct vmxnet3_softc *sc; int error; dev = iflib_get_dev(ctx); scctx = iflib_get_softc_ctx(ctx); sc = iflib_get_softc(ctx); if (scctx->isc_nrxqsets > 1) sc->vmx_flags |= VMXNET3_FLAG_RSS; error = vmxnet3_alloc_data(sc); if (error) goto fail; vmxnet3_set_interrupt_idx(sc); vmxnet3_setup_sysctl(sc); ifmedia_add(sc->vmx_media, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(sc->vmx_media, IFM_ETHER | IFM_AUTO); fail: return (error); } static int vmxnet3_detach(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); vmxnet3_free_irqs(sc); vmxnet3_free_data(sc); vmxnet3_free_resources(sc); return (0); } static int vmxnet3_shutdown(if_ctx_t ctx) { return (0); } static int vmxnet3_suspend(if_ctx_t ctx) { return (0); } static int vmxnet3_resume(if_ctx_t ctx) { return (0); } static int vmxnet3_alloc_resources(struct vmxnet3_softc *sc) { device_t dev; int rid; dev = sc->vmx_dev; rid = PCIR_BAR(0); sc->vmx_res0 = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->vmx_res0 == NULL) { device_printf(dev, "could not map BAR0 memory\n"); return (ENXIO); } sc->vmx_iot0 = rman_get_bustag(sc->vmx_res0); sc->vmx_ioh0 = rman_get_bushandle(sc->vmx_res0); rid = PCIR_BAR(1); sc->vmx_res1 = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->vmx_res1 == NULL) { device_printf(dev, "could not map BAR1 memory\n"); return (ENXIO); } sc->vmx_iot1 = rman_get_bustag(sc->vmx_res1); sc->vmx_ioh1 = rman_get_bushandle(sc->vmx_res1); return (0); } static void vmxnet3_free_resources(struct vmxnet3_softc *sc) { device_t dev; dev = sc->vmx_dev; if (sc->vmx_res0 != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, rman_get_rid(sc->vmx_res0), sc->vmx_res0); sc->vmx_res0 = NULL; } if (sc->vmx_res1 != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, rman_get_rid(sc->vmx_res1), sc->vmx_res1); sc->vmx_res1 = NULL; } } static int vmxnet3_check_version(struct vmxnet3_softc *sc) { device_t dev; uint32_t version; dev = sc->vmx_dev; version = vmxnet3_read_bar1(sc, VMXNET3_BAR1_VRRS); if ((version & 0x01) == 0) { device_printf(dev, "unsupported hardware version %#x\n", version); return (ENOTSUP); } vmxnet3_write_bar1(sc, VMXNET3_BAR1_VRRS, 1); version = vmxnet3_read_bar1(sc, VMXNET3_BAR1_UVRS); if ((version & 0x01) == 0) { device_printf(dev, "unsupported UPT version %#x\n", version); return (ENOTSUP); } vmxnet3_write_bar1(sc, VMXNET3_BAR1_UVRS, 1); return (0); } static void vmxnet3_set_interrupt_idx(struct vmxnet3_softc *sc) { if_softc_ctx_t scctx; struct vmxnet3_txqueue *txq; struct vmxnet3_txq_shared *txs; struct vmxnet3_rxqueue *rxq; struct vmxnet3_rxq_shared *rxs; int intr_idx; int i; scctx = sc->vmx_scctx; /* * There is always one interrupt per receive queue, assigned * starting with the first interrupt. When there is only one * interrupt available, the event interrupt shares the receive queue * interrupt, otherwise it uses the interrupt following the last * receive queue interrupt. Transmit queues are not assigned * interrupts, so they are given indexes beyond the indexes that * correspond to the real interrupts. */ /* The event interrupt is always the last vector. */ sc->vmx_event_intr_idx = scctx->isc_vectors - 1; intr_idx = 0; for (i = 0; i < scctx->isc_nrxqsets; i++, intr_idx++) { rxq = &sc->vmx_rxq[i]; rxs = rxq->vxrxq_rs; rxq->vxrxq_intr_idx = intr_idx; rxs->intr_idx = rxq->vxrxq_intr_idx; } /* * Assign the tx queues interrupt indexes above what we are actually * using. These interrupts will never be enabled. */ intr_idx = scctx->isc_vectors; for (i = 0; i < scctx->isc_ntxqsets; i++, intr_idx++) { txq = &sc->vmx_txq[i]; txs = txq->vxtxq_ts; txq->vxtxq_intr_idx = intr_idx; txs->intr_idx = txq->vxtxq_intr_idx; } } static int vmxnet3_queues_shared_alloc(struct vmxnet3_softc *sc) { if_softc_ctx_t scctx; int size; int error; scctx = sc->vmx_scctx; /* * The txq and rxq shared data areas must be allocated contiguously * as vmxnet3_driver_shared contains only a single address member * for the shared queue data area. */ size = scctx->isc_ntxqsets * sizeof(struct vmxnet3_txq_shared) + scctx->isc_nrxqsets * sizeof(struct vmxnet3_rxq_shared); error = iflib_dma_alloc_align(sc->vmx_ctx, size, 128, &sc->vmx_qs_dma, 0); if (error) { device_printf(sc->vmx_dev, "cannot alloc queue shared memory\n"); return (error); } return (0); } static void vmxnet3_init_txq(struct vmxnet3_softc *sc, int q) { struct vmxnet3_txqueue *txq; struct vmxnet3_comp_ring *txc; struct vmxnet3_txring *txr; if_softc_ctx_t scctx; txq = &sc->vmx_txq[q]; txc = &txq->vxtxq_comp_ring; txr = &txq->vxtxq_cmd_ring; scctx = sc->vmx_scctx; snprintf(txq->vxtxq_name, sizeof(txq->vxtxq_name), "%s-tx%d", device_get_nameunit(sc->vmx_dev), q); txq->vxtxq_sc = sc; txq->vxtxq_id = q; txc->vxcr_ndesc = scctx->isc_ntxd[0]; txr->vxtxr_ndesc = scctx->isc_ntxd[1]; } static int vmxnet3_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets) { struct vmxnet3_softc *sc; int q; int error; caddr_t kva; sc = iflib_get_softc(ctx); /* Allocate the array of transmit queues */ sc->vmx_txq = malloc(sizeof(struct vmxnet3_txqueue) * ntxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vmx_txq == NULL) return (ENOMEM); /* Initialize driver state for each transmit queue */ for (q = 0; q < ntxqsets; q++) vmxnet3_init_txq(sc, q); /* * Allocate queue state that is shared with the device. This check * and call is performed in both vmxnet3_tx_queues_alloc() and * vmxnet3_rx_queues_alloc() so that we don't have to care which * order iflib invokes those routines in. */ if (sc->vmx_qs_dma.idi_size == 0) { error = vmxnet3_queues_shared_alloc(sc); if (error) return (error); } kva = sc->vmx_qs_dma.idi_vaddr; for (q = 0; q < ntxqsets; q++) { sc->vmx_txq[q].vxtxq_ts = (struct vmxnet3_txq_shared *) kva; kva += sizeof(struct vmxnet3_txq_shared); } /* Record descriptor ring vaddrs and paddrs */ for (q = 0; q < ntxqsets; q++) { struct vmxnet3_txqueue *txq; struct vmxnet3_txring *txr; struct vmxnet3_comp_ring *txc; txq = &sc->vmx_txq[q]; txc = &txq->vxtxq_comp_ring; txr = &txq->vxtxq_cmd_ring; /* Completion ring */ txc->vxcr_u.txcd = (struct vmxnet3_txcompdesc *) vaddrs[q * ntxqs + 0]; txc->vxcr_paddr = paddrs[q * ntxqs + 0]; /* Command ring */ txr->vxtxr_txd = (struct vmxnet3_txdesc *) vaddrs[q * ntxqs + 1]; txr->vxtxr_paddr = paddrs[q * ntxqs + 1]; } return (0); } static void vmxnet3_init_rxq(struct vmxnet3_softc *sc, int q, int nrxqs) { struct vmxnet3_rxqueue *rxq; struct vmxnet3_comp_ring *rxc; struct vmxnet3_rxring *rxr; if_softc_ctx_t scctx; int i; rxq = &sc->vmx_rxq[q]; rxc = &rxq->vxrxq_comp_ring; scctx = sc->vmx_scctx; snprintf(rxq->vxrxq_name, sizeof(rxq->vxrxq_name), "%s-rx%d", device_get_nameunit(sc->vmx_dev), q); rxq->vxrxq_sc = sc; rxq->vxrxq_id = q; /* * First rxq is the completion queue, so there are nrxqs - 1 command * rings starting at iflib queue id 1. */ rxc->vxcr_ndesc = scctx->isc_nrxd[0]; for (i = 0; i < nrxqs - 1; i++) { rxr = &rxq->vxrxq_cmd_ring[i]; rxr->vxrxr_ndesc = scctx->isc_nrxd[i + 1]; } } static int vmxnet3_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; int q; int i; int error; caddr_t kva; sc = iflib_get_softc(ctx); scctx = sc->vmx_scctx; /* Allocate the array of receive queues */ sc->vmx_rxq = malloc(sizeof(struct vmxnet3_rxqueue) * nrxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vmx_rxq == NULL) return (ENOMEM); /* Initialize driver state for each receive queue */ for (q = 0; q < nrxqsets; q++) vmxnet3_init_rxq(sc, q, nrxqs); /* * Allocate queue state that is shared with the device. This check * and call is performed in both vmxnet3_tx_queues_alloc() and * vmxnet3_rx_queues_alloc() so that we don't have to care which * order iflib invokes those routines in. */ if (sc->vmx_qs_dma.idi_size == 0) { error = vmxnet3_queues_shared_alloc(sc); if (error) return (error); } kva = sc->vmx_qs_dma.idi_vaddr + scctx->isc_ntxqsets * sizeof(struct vmxnet3_txq_shared); for (q = 0; q < nrxqsets; q++) { sc->vmx_rxq[q].vxrxq_rs = (struct vmxnet3_rxq_shared *) kva; kva += sizeof(struct vmxnet3_rxq_shared); } /* Record descriptor ring vaddrs and paddrs */ for (q = 0; q < nrxqsets; q++) { struct vmxnet3_rxqueue *rxq; struct vmxnet3_rxring *rxr; struct vmxnet3_comp_ring *rxc; rxq = &sc->vmx_rxq[q]; rxc = &rxq->vxrxq_comp_ring; /* Completion ring */ rxc->vxcr_u.rxcd = (struct vmxnet3_rxcompdesc *) vaddrs[q * nrxqs + 0]; rxc->vxcr_paddr = paddrs[q * nrxqs + 0]; /* Command ring(s) */ for (i = 0; i < nrxqs - 1; i++) { rxr = &rxq->vxrxq_cmd_ring[i]; rxr->vxrxr_rxd = (struct vmxnet3_rxdesc *) vaddrs[q * nrxqs + 1 + i]; rxr->vxrxr_paddr = paddrs[q * nrxqs + 1 + i]; } } return (0); } static void vmxnet3_queues_free(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); /* Free queue state area that is shared with the device */ if (sc->vmx_qs_dma.idi_size != 0) { iflib_dma_free(&sc->vmx_qs_dma); sc->vmx_qs_dma.idi_size = 0; } /* Free array of receive queues */ if (sc->vmx_rxq != NULL) { free(sc->vmx_rxq, M_DEVBUF); sc->vmx_rxq = NULL; } /* Free array of transmit queues */ if (sc->vmx_txq != NULL) { free(sc->vmx_txq, M_DEVBUF); sc->vmx_txq = NULL; } } static int vmxnet3_alloc_shared_data(struct vmxnet3_softc *sc) { device_t dev; size_t size; int error; dev = sc->vmx_dev; /* Top level state structure shared with the device */ size = sizeof(struct vmxnet3_driver_shared); error = iflib_dma_alloc_align(sc->vmx_ctx, size, 1, &sc->vmx_ds_dma, 0); if (error) { device_printf(dev, "cannot alloc shared memory\n"); return (error); } sc->vmx_ds = (struct vmxnet3_driver_shared *) sc->vmx_ds_dma.idi_vaddr; /* RSS table state shared with the device */ if (sc->vmx_flags & VMXNET3_FLAG_RSS) { size = sizeof(struct vmxnet3_rss_shared); error = iflib_dma_alloc_align(sc->vmx_ctx, size, 128, &sc->vmx_rss_dma, 0); if (error) { device_printf(dev, "cannot alloc rss shared memory\n"); return (error); } sc->vmx_rss = (struct vmxnet3_rss_shared *) sc->vmx_rss_dma.idi_vaddr; } return (0); } static void vmxnet3_free_shared_data(struct vmxnet3_softc *sc) { /* Free RSS table state shared with the device */ if (sc->vmx_rss != NULL) { iflib_dma_free(&sc->vmx_rss_dma); sc->vmx_rss = NULL; } /* Free top level state structure shared with the device */ if (sc->vmx_ds != NULL) { iflib_dma_free(&sc->vmx_ds_dma); sc->vmx_ds = NULL; } } static int vmxnet3_alloc_mcast_table(struct vmxnet3_softc *sc) { int error; /* Multicast table state shared with the device */ error = iflib_dma_alloc_align(sc->vmx_ctx, VMXNET3_MULTICAST_MAX * ETHER_ADDR_LEN, 32, &sc->vmx_mcast_dma, 0); if (error) device_printf(sc->vmx_dev, "unable to alloc multicast table\n"); else sc->vmx_mcast = sc->vmx_mcast_dma.idi_vaddr; return (error); } static void vmxnet3_free_mcast_table(struct vmxnet3_softc *sc) { /* Free multicast table state shared with the device */ if (sc->vmx_mcast != NULL) { iflib_dma_free(&sc->vmx_mcast_dma); sc->vmx_mcast = NULL; } } static void vmxnet3_init_shared_data(struct vmxnet3_softc *sc) { struct vmxnet3_driver_shared *ds; if_shared_ctx_t sctx; if_softc_ctx_t scctx; struct vmxnet3_txqueue *txq; struct vmxnet3_txq_shared *txs; struct vmxnet3_rxqueue *rxq; struct vmxnet3_rxq_shared *rxs; int i; ds = sc->vmx_ds; sctx = sc->vmx_sctx; scctx = sc->vmx_scctx; /* * Initialize fields of the shared data that remains the same across * reinits. Note the shared data is zero'd when allocated. */ ds->magic = VMXNET3_REV1_MAGIC; /* DriverInfo */ ds->version = VMXNET3_DRIVER_VERSION; ds->guest = VMXNET3_GOS_FREEBSD | #ifdef __LP64__ VMXNET3_GOS_64BIT; #else VMXNET3_GOS_32BIT; #endif ds->vmxnet3_revision = 1; ds->upt_version = 1; /* Misc. conf */ ds->driver_data = vtophys(sc); ds->driver_data_len = sizeof(struct vmxnet3_softc); ds->queue_shared = sc->vmx_qs_dma.idi_paddr; ds->queue_shared_len = sc->vmx_qs_dma.idi_size; ds->nrxsg_max = IFLIB_MAX_RX_SEGS; /* RSS conf */ if (sc->vmx_flags & VMXNET3_FLAG_RSS) { ds->rss.version = 1; ds->rss.paddr = sc->vmx_rss_dma.idi_paddr; ds->rss.len = sc->vmx_rss_dma.idi_size; } /* Interrupt control. */ ds->automask = sc->vmx_intr_mask_mode == VMXNET3_IMM_AUTO; /* * Total number of interrupt indexes we are using in the shared * config data, even though we don't actually allocate interrupt * resources for the tx queues. Some versions of the device will * fail to initialize successfully if interrupt indexes are used in * the shared config that exceed the number of interrupts configured * here. */ ds->nintr = (scctx->isc_vectors == 1) ? 2 : (scctx->isc_nrxqsets + scctx->isc_ntxqsets + 1); ds->evintr = sc->vmx_event_intr_idx; ds->ictrl = VMXNET3_ICTRL_DISABLE_ALL; for (i = 0; i < ds->nintr; i++) ds->modlevel[i] = UPT1_IMOD_ADAPTIVE; /* Receive filter. */ ds->mcast_table = sc->vmx_mcast_dma.idi_paddr; ds->mcast_tablelen = sc->vmx_mcast_dma.idi_size; /* Tx queues */ for (i = 0; i < scctx->isc_ntxqsets; i++) { txq = &sc->vmx_txq[i]; txs = txq->vxtxq_ts; txs->cmd_ring = txq->vxtxq_cmd_ring.vxtxr_paddr; txs->cmd_ring_len = txq->vxtxq_cmd_ring.vxtxr_ndesc; txs->comp_ring = txq->vxtxq_comp_ring.vxcr_paddr; txs->comp_ring_len = txq->vxtxq_comp_ring.vxcr_ndesc; txs->driver_data = vtophys(txq); txs->driver_data_len = sizeof(struct vmxnet3_txqueue); } /* Rx queues */ for (i = 0; i < scctx->isc_nrxqsets; i++) { rxq = &sc->vmx_rxq[i]; rxs = rxq->vxrxq_rs; rxs->cmd_ring[0] = rxq->vxrxq_cmd_ring[0].vxrxr_paddr; rxs->cmd_ring_len[0] = rxq->vxrxq_cmd_ring[0].vxrxr_ndesc; rxs->cmd_ring[1] = rxq->vxrxq_cmd_ring[1].vxrxr_paddr; rxs->cmd_ring_len[1] = rxq->vxrxq_cmd_ring[1].vxrxr_ndesc; rxs->comp_ring = rxq->vxrxq_comp_ring.vxcr_paddr; rxs->comp_ring_len = rxq->vxrxq_comp_ring.vxcr_ndesc; rxs->driver_data = vtophys(rxq); rxs->driver_data_len = sizeof(struct vmxnet3_rxqueue); } } static void vmxnet3_reinit_rss_shared_data(struct vmxnet3_softc *sc) { /* * Use the same key as the Linux driver until FreeBSD can do * RSS (presumably Toeplitz) in software. */ static const uint8_t rss_key[UPT1_RSS_MAX_KEY_SIZE] = { 0x3b, 0x56, 0xd1, 0x56, 0x13, 0x4a, 0xe7, 0xac, 0xe8, 0x79, 0x09, 0x75, 0xe8, 0x65, 0x79, 0x28, 0x35, 0x12, 0xb9, 0x56, 0x7c, 0x76, 0x4b, 0x70, 0xd8, 0x56, 0xa3, 0x18, 0x9b, 0x0a, 0xee, 0xf3, 0x96, 0xa6, 0x9f, 0x8f, 0x9e, 0x8c, 0x90, 0xc9, }; struct vmxnet3_driver_shared *ds; if_softc_ctx_t scctx; struct vmxnet3_rss_shared *rss; #ifdef RSS uint8_t rss_algo; #endif int i; ds = sc->vmx_ds; scctx = sc->vmx_scctx; rss = sc->vmx_rss; rss->hash_type = UPT1_RSS_HASH_TYPE_IPV4 | UPT1_RSS_HASH_TYPE_TCP_IPV4 | UPT1_RSS_HASH_TYPE_IPV6 | UPT1_RSS_HASH_TYPE_TCP_IPV6; rss->hash_func = UPT1_RSS_HASH_FUNC_TOEPLITZ; rss->hash_key_size = UPT1_RSS_MAX_KEY_SIZE; rss->ind_table_size = UPT1_RSS_MAX_IND_TABLE_SIZE; #ifdef RSS /* * If the software RSS is configured to anything else other than * Toeplitz, then just do Toeplitz in "hardware" for the sake of * the packet distribution, but report the hash as opaque to * disengage from the software RSS. */ rss_algo = rss_gethashalgo(); if (rss_algo == RSS_HASH_TOEPLITZ) { rss_getkey(rss->hash_key); for (i = 0; i < UPT1_RSS_MAX_IND_TABLE_SIZE; i++) { rss->ind_table[i] = rss_get_indirection_to_bucket(i) % scctx->isc_nrxqsets; } sc->vmx_flags |= VMXNET3_FLAG_SOFT_RSS; } else #endif { memcpy(rss->hash_key, rss_key, UPT1_RSS_MAX_KEY_SIZE); for (i = 0; i < UPT1_RSS_MAX_IND_TABLE_SIZE; i++) rss->ind_table[i] = i % scctx->isc_nrxqsets; sc->vmx_flags &= ~VMXNET3_FLAG_SOFT_RSS; } } static void vmxnet3_reinit_shared_data(struct vmxnet3_softc *sc) { struct ifnet *ifp; struct vmxnet3_driver_shared *ds; if_softc_ctx_t scctx; ifp = sc->vmx_ifp; ds = sc->vmx_ds; scctx = sc->vmx_scctx; ds->mtu = ifp->if_mtu; ds->ntxqueue = scctx->isc_ntxqsets; ds->nrxqueue = scctx->isc_nrxqsets; ds->upt_features = 0; if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) ds->upt_features |= UPT1_F_CSUM; if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) ds->upt_features |= UPT1_F_VLAN; if (ifp->if_capenable & IFCAP_LRO) ds->upt_features |= UPT1_F_LRO; if (sc->vmx_flags & VMXNET3_FLAG_RSS) { ds->upt_features |= UPT1_F_RSS; vmxnet3_reinit_rss_shared_data(sc); } vmxnet3_write_bar1(sc, VMXNET3_BAR1_DSL, sc->vmx_ds_dma.idi_paddr); vmxnet3_write_bar1(sc, VMXNET3_BAR1_DSH, (uint64_t) sc->vmx_ds_dma.idi_paddr >> 32); } static int vmxnet3_alloc_data(struct vmxnet3_softc *sc) { int error; error = vmxnet3_alloc_shared_data(sc); if (error) return (error); error = vmxnet3_alloc_mcast_table(sc); if (error) return (error); vmxnet3_init_shared_data(sc); return (0); } static void vmxnet3_free_data(struct vmxnet3_softc *sc) { vmxnet3_free_mcast_table(sc); vmxnet3_free_shared_data(sc); } static void vmxnet3_evintr(struct vmxnet3_softc *sc) { device_t dev; struct vmxnet3_txq_shared *ts; struct vmxnet3_rxq_shared *rs; uint32_t event; dev = sc->vmx_dev; /* Clear events. */ event = sc->vmx_ds->event; vmxnet3_write_bar1(sc, VMXNET3_BAR1_EVENT, event); if (event & VMXNET3_EVENT_LINK) vmxnet3_link_status(sc); if (event & (VMXNET3_EVENT_TQERROR | VMXNET3_EVENT_RQERROR)) { vmxnet3_read_cmd(sc, VMXNET3_CMD_GET_STATUS); ts = sc->vmx_txq[0].vxtxq_ts; if (ts->stopped != 0) device_printf(dev, "Tx queue error %#x\n", ts->error); rs = sc->vmx_rxq[0].vxrxq_rs; if (rs->stopped != 0) device_printf(dev, "Rx queue error %#x\n", rs->error); /* XXX - rely on liflib watchdog to reset us? */ device_printf(dev, "Rx/Tx queue error event ... " "waiting for iflib watchdog reset\n"); } if (event & VMXNET3_EVENT_DIC) device_printf(dev, "device implementation change event\n"); if (event & VMXNET3_EVENT_DEBUG) device_printf(dev, "debug event\n"); } static int vmxnet3_isc_txd_encap(void *vsc, if_pkt_info_t pi) { struct vmxnet3_softc *sc; struct vmxnet3_txqueue *txq; struct vmxnet3_txring *txr; struct vmxnet3_txdesc *txd, *sop; bus_dma_segment_t *segs; int nsegs; int pidx; int hdrlen; int i; int gen; sc = vsc; txq = &sc->vmx_txq[pi->ipi_qsidx]; txr = &txq->vxtxq_cmd_ring; segs = pi->ipi_segs; nsegs = pi->ipi_nsegs; pidx = pi->ipi_pidx; KASSERT(nsegs <= VMXNET3_TX_MAXSEGS, ("%s: packet with too many segments %d", __func__, nsegs)); sop = &txr->vxtxr_txd[pidx]; gen = txr->vxtxr_gen ^ 1; /* Owned by cpu (yet) */ for (i = 0; i < nsegs; i++) { txd = &txr->vxtxr_txd[pidx]; txd->addr = segs[i].ds_addr; txd->len = segs[i].ds_len; txd->gen = gen; txd->dtype = 0; txd->offload_mode = VMXNET3_OM_NONE; txd->offload_pos = 0; txd->hlen = 0; txd->eop = 0; txd->compreq = 0; txd->vtag_mode = 0; txd->vtag = 0; if (++pidx == txr->vxtxr_ndesc) { pidx = 0; txr->vxtxr_gen ^= 1; } gen = txr->vxtxr_gen; } txd->eop = 1; txd->compreq = !!(pi->ipi_flags & IPI_TX_INTR); pi->ipi_new_pidx = pidx; /* * VLAN */ if (pi->ipi_mflags & M_VLANTAG) { sop->vtag_mode = 1; sop->vtag = pi->ipi_vtag; } /* * TSO and checksum offloads */ hdrlen = pi->ipi_ehdrlen + pi->ipi_ip_hlen; if (pi->ipi_csum_flags & CSUM_TSO) { sop->offload_mode = VMXNET3_OM_TSO; sop->hlen = hdrlen + pi->ipi_tcp_hlen; sop->offload_pos = pi->ipi_tso_segsz; } else if (pi->ipi_csum_flags & (VMXNET3_CSUM_OFFLOAD | VMXNET3_CSUM_OFFLOAD_IPV6)) { sop->offload_mode = VMXNET3_OM_CSUM; sop->hlen = hdrlen; sop->offload_pos = hdrlen + ((pi->ipi_ipproto == IPPROTO_TCP) ? offsetof(struct tcphdr, th_sum) : offsetof(struct udphdr, uh_sum)); } /* Finally, change the ownership. */ vmxnet3_barrier(sc, VMXNET3_BARRIER_WR); sop->gen ^= 1; return (0); } static void vmxnet3_isc_txd_flush(void *vsc, uint16_t txqid, qidx_t pidx) { struct vmxnet3_softc *sc; struct vmxnet3_txqueue *txq; sc = vsc; txq = &sc->vmx_txq[txqid]; /* * pidx is what we last set ipi_new_pidx to in * vmxnet3_isc_txd_encap() */ /* * Avoid expensive register updates if the flush request is * redundant. */ if (txq->vxtxq_last_flush == pidx) return; txq->vxtxq_last_flush = pidx; vmxnet3_write_bar0(sc, VMXNET3_BAR0_TXH(txq->vxtxq_id), pidx); } static int vmxnet3_isc_txd_credits_update(void *vsc, uint16_t txqid, bool clear) { struct vmxnet3_softc *sc; struct vmxnet3_txqueue *txq; struct vmxnet3_comp_ring *txc; struct vmxnet3_txcompdesc *txcd; struct vmxnet3_txring *txr; int processed; sc = vsc; txq = &sc->vmx_txq[txqid]; txc = &txq->vxtxq_comp_ring; txr = &txq->vxtxq_cmd_ring; /* * If clear is true, we need to report the number of TX command ring * descriptors that have been processed by the device. If clear is * false, we just need to report whether or not at least one TX * command ring descriptor has been processed by the device. */ processed = 0; for (;;) { txcd = &txc->vxcr_u.txcd[txc->vxcr_next]; if (txcd->gen != txc->vxcr_gen) break; else if (!clear) return (1); vmxnet3_barrier(sc, VMXNET3_BARRIER_RD); if (++txc->vxcr_next == txc->vxcr_ndesc) { txc->vxcr_next = 0; txc->vxcr_gen ^= 1; } if (txcd->eop_idx < txr->vxtxr_next) processed += txr->vxtxr_ndesc - (txr->vxtxr_next - txcd->eop_idx) + 1; else processed += txcd->eop_idx - txr->vxtxr_next + 1; txr->vxtxr_next = (txcd->eop_idx + 1) % txr->vxtxr_ndesc; } return (processed); } static int vmxnet3_isc_rxd_available(void *vsc, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct vmxnet3_softc *sc; struct vmxnet3_rxqueue *rxq; struct vmxnet3_comp_ring *rxc; struct vmxnet3_rxcompdesc *rxcd; int avail; int completed_gen; #ifdef INVARIANTS int expect_sop = 1; #endif sc = vsc; rxq = &sc->vmx_rxq[rxqid]; rxc = &rxq->vxrxq_comp_ring; avail = 0; completed_gen = rxc->vxcr_gen; for (;;) { rxcd = &rxc->vxcr_u.rxcd[idx]; if (rxcd->gen != completed_gen) break; vmxnet3_barrier(sc, VMXNET3_BARRIER_RD); #ifdef INVARIANTS if (expect_sop) KASSERT(rxcd->sop, ("%s: expected sop", __func__)); else KASSERT(!rxcd->sop, ("%s: unexpected sop", __func__)); expect_sop = rxcd->eop; #endif if (rxcd->eop && (rxcd->len != 0)) avail++; if (avail > budget) break; if (++idx == rxc->vxcr_ndesc) { idx = 0; completed_gen ^= 1; } } return (avail); } static int vmxnet3_isc_rxd_pkt_get(void *vsc, if_rxd_info_t ri) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; struct vmxnet3_rxqueue *rxq; struct vmxnet3_comp_ring *rxc; struct vmxnet3_rxcompdesc *rxcd; struct vmxnet3_rxring *rxr; struct vmxnet3_rxdesc *rxd; if_rxd_frag_t frag; int cqidx; uint16_t total_len; uint8_t nfrags; uint8_t i; uint8_t flid; sc = vsc; scctx = sc->vmx_scctx; rxq = &sc->vmx_rxq[ri->iri_qsidx]; rxc = &rxq->vxrxq_comp_ring; /* * Get a single packet starting at the given index in the completion * queue. That we have been called indicates that * vmxnet3_isc_rxd_available() has already verified that either * there is a complete packet available starting at the given index, * or there are one or more zero length packets starting at the * given index followed by a complete packet, so no verification of * ownership of the descriptors (and no associated read barrier) is * required here. */ cqidx = ri->iri_cidx; rxcd = &rxc->vxcr_u.rxcd[cqidx]; while (rxcd->len == 0) { KASSERT(rxcd->sop && rxcd->eop, ("%s: zero-length packet without both sop and eop set", __func__)); rxc->vxcr_zero_length++; if (++cqidx == rxc->vxcr_ndesc) { cqidx = 0; rxc->vxcr_gen ^= 1; } rxcd = &rxc->vxcr_u.rxcd[cqidx]; } KASSERT(rxcd->sop, ("%s: expected sop", __func__)); /* * RSS and flow ID. * Types other than M_HASHTYPE_NONE and M_HASHTYPE_OPAQUE_HASH should * be used only if the software RSS is enabled and it uses the same * algorithm and the hash key as the "hardware". If the software RSS * is not enabled, then it's simply pointless to use those types. * If it's enabled but with different parameters, then hash values will * not match. */ ri->iri_flowid = rxcd->rss_hash; #ifdef RSS if ((sc->vmx_flags & VMXNET3_FLAG_SOFT_RSS) != 0) { switch (rxcd->rss_type) { case VMXNET3_RCD_RSS_TYPE_NONE: ri->iri_flowid = ri->iri_qsidx; ri->iri_rsstype = M_HASHTYPE_NONE; break; case VMXNET3_RCD_RSS_TYPE_IPV4: ri->iri_rsstype = M_HASHTYPE_RSS_IPV4; break; case VMXNET3_RCD_RSS_TYPE_TCPIPV4: ri->iri_rsstype = M_HASHTYPE_RSS_TCP_IPV4; break; case VMXNET3_RCD_RSS_TYPE_IPV6: ri->iri_rsstype = M_HASHTYPE_RSS_IPV6; break; case VMXNET3_RCD_RSS_TYPE_TCPIPV6: ri->iri_rsstype = M_HASHTYPE_RSS_TCP_IPV6; break; default: ri->iri_rsstype = M_HASHTYPE_OPAQUE_HASH; break; } } else #endif { switch (rxcd->rss_type) { case VMXNET3_RCD_RSS_TYPE_NONE: ri->iri_flowid = ri->iri_qsidx; ri->iri_rsstype = M_HASHTYPE_NONE; break; default: ri->iri_rsstype = M_HASHTYPE_OPAQUE_HASH; break; } } /* * The queue numbering scheme used for rxcd->qid is as follows: * - All of the command ring 0s are numbered [0, nrxqsets - 1] * - All of the command ring 1s are numbered [nrxqsets, 2*nrxqsets - 1] * * Thus, rxcd->qid less than nrxqsets indicates command ring (and * flid) 0, and rxcd->qid greater than or equal to nrxqsets * indicates command ring (and flid) 1. */ nfrags = 0; total_len = 0; do { rxcd = &rxc->vxcr_u.rxcd[cqidx]; KASSERT(rxcd->gen == rxc->vxcr_gen, ("%s: generation mismatch", __func__)); flid = (rxcd->qid >= scctx->isc_nrxqsets) ? 1 : 0; rxr = &rxq->vxrxq_cmd_ring[flid]; rxd = &rxr->vxrxr_rxd[rxcd->rxd_idx]; frag = &ri->iri_frags[nfrags]; frag->irf_flid = flid; frag->irf_idx = rxcd->rxd_idx; frag->irf_len = rxcd->len; total_len += rxcd->len; nfrags++; if (++cqidx == rxc->vxcr_ndesc) { cqidx = 0; rxc->vxcr_gen ^= 1; } } while (!rxcd->eop); ri->iri_cidx = cqidx; ri->iri_nfrags = nfrags; ri->iri_len = total_len; /* * If there's an error, the last descriptor in the packet will * have the error indicator set. In this case, set all * fragment lengths to zero. This will cause iflib to discard * the packet, but process all associated descriptors through * the refill mechanism. */ if (__predict_false(rxcd->error)) { rxc->vxcr_pkt_errors++; for (i = 0; i < nfrags; i++) { frag = &ri->iri_frags[i]; frag->irf_len = 0; } } else { /* Checksum offload information is in the last descriptor. */ if (!rxcd->no_csum) { uint32_t csum_flags = 0; if (rxcd->ipv4) { csum_flags |= CSUM_IP_CHECKED; if (rxcd->ipcsum_ok) csum_flags |= CSUM_IP_VALID; } if (!rxcd->fragment && (rxcd->tcp || rxcd->udp)) { csum_flags |= CSUM_L4_CALC; if (rxcd->csum_ok) { csum_flags |= CSUM_L4_VALID; ri->iri_csum_data = 0xffff; } } ri->iri_csum_flags = csum_flags; } /* VLAN information is in the last descriptor. */ if (rxcd->vlan) { ri->iri_flags |= M_VLANTAG; ri->iri_vtag = rxcd->vtag; } } return (0); } static void vmxnet3_isc_rxd_refill(void *vsc, if_rxd_update_t iru) { struct vmxnet3_softc *sc; struct vmxnet3_rxqueue *rxq; struct vmxnet3_rxring *rxr; struct vmxnet3_rxdesc *rxd; uint64_t *paddrs; int count; int len; int idx; int i; uint8_t flid; uint8_t btype; count = iru->iru_count; len = iru->iru_buf_size; flid = iru->iru_flidx; paddrs = iru->iru_paddrs; sc = vsc; rxq = &sc->vmx_rxq[iru->iru_qsidx]; rxr = &rxq->vxrxq_cmd_ring[flid]; rxd = rxr->vxrxr_rxd; /* * Command ring 0 is filled with BTYPE_HEAD descriptors, and * command ring 1 is filled with BTYPE_BODY descriptors. */ btype = (flid == 0) ? VMXNET3_BTYPE_HEAD : VMXNET3_BTYPE_BODY; /* * The refill entries from iflib will advance monotonically, * but the refilled descriptors may not be contiguous due to * earlier skipping of descriptors by the device. The refill * entries from iflib need an entire state update, while the * descriptors previously skipped by the device only need to * have their generation numbers updated. */ idx = rxr->vxrxr_refill_start; i = 0; do { if (idx == iru->iru_idxs[i]) { rxd[idx].addr = paddrs[i]; rxd[idx].len = len; rxd[idx].btype = btype; i++; } else rxr->vxrxr_desc_skips++; rxd[idx].gen = rxr->vxrxr_gen; if (++idx == rxr->vxrxr_ndesc) { idx = 0; rxr->vxrxr_gen ^= 1; } } while (i != count); rxr->vxrxr_refill_start = idx; } static void vmxnet3_isc_rxd_flush(void *vsc, uint16_t rxqid, uint8_t flid, qidx_t pidx) { struct vmxnet3_softc *sc; struct vmxnet3_rxqueue *rxq; struct vmxnet3_rxring *rxr; bus_size_t r; sc = vsc; rxq = &sc->vmx_rxq[rxqid]; rxr = &rxq->vxrxq_cmd_ring[flid]; if (flid == 0) r = VMXNET3_BAR0_RXH1(rxqid); else r = VMXNET3_BAR0_RXH2(rxqid); - /* - * pidx is the index of the last descriptor with a buffer the device - * can use, and the device needs to be told which index is one past - * that. - */ - if (++pidx == rxr->vxrxr_ndesc) - pidx = 0; vmxnet3_write_bar0(sc, r, pidx); } static int vmxnet3_legacy_intr(void *xsc) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; if_ctx_t ctx; sc = xsc; scctx = sc->vmx_scctx; ctx = sc->vmx_ctx; /* * When there is only a single interrupt configured, this routine * runs in fast interrupt context, following which the rxq 0 task * will be enqueued. */ if (scctx->isc_intr == IFLIB_INTR_LEGACY) { if (vmxnet3_read_bar1(sc, VMXNET3_BAR1_INTR) == 0) return (FILTER_HANDLED); } if (sc->vmx_intr_mask_mode == VMXNET3_IMM_ACTIVE) vmxnet3_intr_disable_all(ctx); if (sc->vmx_ds->event != 0) iflib_admin_intr_deferred(ctx); /* * XXX - When there is both rxq and event activity, do we care * whether the rxq 0 task or the admin task re-enables the interrupt * first? */ return (FILTER_SCHEDULE_THREAD); } static int vmxnet3_rxq_intr(void *vrxq) { struct vmxnet3_softc *sc; struct vmxnet3_rxqueue *rxq; rxq = vrxq; sc = rxq->vxrxq_sc; if (sc->vmx_intr_mask_mode == VMXNET3_IMM_ACTIVE) vmxnet3_disable_intr(sc, rxq->vxrxq_intr_idx); return (FILTER_SCHEDULE_THREAD); } static int vmxnet3_event_intr(void *vsc) { struct vmxnet3_softc *sc; sc = vsc; if (sc->vmx_intr_mask_mode == VMXNET3_IMM_ACTIVE) vmxnet3_disable_intr(sc, sc->vmx_event_intr_idx); /* * The work will be done via vmxnet3_update_admin_status(), and the * interrupt will be re-enabled in vmxnet3_link_intr_enable(). * * The interrupt will be re-enabled by vmxnet3_link_intr_enable(). */ return (FILTER_SCHEDULE_THREAD); } static void vmxnet3_stop(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); sc->vmx_link_active = 0; vmxnet3_write_cmd(sc, VMXNET3_CMD_DISABLE); vmxnet3_write_cmd(sc, VMXNET3_CMD_RESET); } static void vmxnet3_txinit(struct vmxnet3_softc *sc, struct vmxnet3_txqueue *txq) { struct vmxnet3_txring *txr; struct vmxnet3_comp_ring *txc; txq->vxtxq_last_flush = -1; txr = &txq->vxtxq_cmd_ring; txr->vxtxr_next = 0; txr->vxtxr_gen = VMXNET3_INIT_GEN; /* * iflib has zeroed out the descriptor array during the prior attach * or stop */ txc = &txq->vxtxq_comp_ring; txc->vxcr_next = 0; txc->vxcr_gen = VMXNET3_INIT_GEN; /* * iflib has zeroed out the descriptor array during the prior attach * or stop */ } static void vmxnet3_rxinit(struct vmxnet3_softc *sc, struct vmxnet3_rxqueue *rxq) { struct vmxnet3_rxring *rxr; struct vmxnet3_comp_ring *rxc; int i; /* * The descriptors will be populated with buffers during a * subsequent invocation of vmxnet3_isc_rxd_refill() */ for (i = 0; i < sc->vmx_sctx->isc_nrxqs - 1; i++) { rxr = &rxq->vxrxq_cmd_ring[i]; rxr->vxrxr_gen = VMXNET3_INIT_GEN; rxr->vxrxr_desc_skips = 0; rxr->vxrxr_refill_start = 0; /* * iflib has zeroed out the descriptor array during the * prior attach or stop */ } for (/**/; i < VMXNET3_RXRINGS_PERQ; i++) { rxr = &rxq->vxrxq_cmd_ring[i]; rxr->vxrxr_gen = 0; rxr->vxrxr_desc_skips = 0; rxr->vxrxr_refill_start = 0; bzero(rxr->vxrxr_rxd, rxr->vxrxr_ndesc * sizeof(struct vmxnet3_rxdesc)); } rxc = &rxq->vxrxq_comp_ring; rxc->vxcr_next = 0; rxc->vxcr_gen = VMXNET3_INIT_GEN; rxc->vxcr_zero_length = 0; rxc->vxcr_pkt_errors = 0; /* * iflib has zeroed out the descriptor array during the prior attach * or stop */ } static void vmxnet3_reinit_queues(struct vmxnet3_softc *sc) { if_softc_ctx_t scctx; int q; scctx = sc->vmx_scctx; for (q = 0; q < scctx->isc_ntxqsets; q++) vmxnet3_txinit(sc, &sc->vmx_txq[q]); for (q = 0; q < scctx->isc_nrxqsets; q++) vmxnet3_rxinit(sc, &sc->vmx_rxq[q]); } static int vmxnet3_enable_device(struct vmxnet3_softc *sc) { if_softc_ctx_t scctx; int q; scctx = sc->vmx_scctx; if (vmxnet3_read_cmd(sc, VMXNET3_CMD_ENABLE) != 0) { device_printf(sc->vmx_dev, "device enable command failed!\n"); return (1); } /* Reset the Rx queue heads. */ for (q = 0; q < scctx->isc_nrxqsets; q++) { vmxnet3_write_bar0(sc, VMXNET3_BAR0_RXH1(q), 0); vmxnet3_write_bar0(sc, VMXNET3_BAR0_RXH2(q), 0); } return (0); } static void vmxnet3_reinit_rxfilters(struct vmxnet3_softc *sc) { struct ifnet *ifp; ifp = sc->vmx_ifp; vmxnet3_set_rxfilter(sc, if_getflags(ifp)); if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) bcopy(sc->vmx_vlan_filter, sc->vmx_ds->vlan_filter, sizeof(sc->vmx_ds->vlan_filter)); else bzero(sc->vmx_ds->vlan_filter, sizeof(sc->vmx_ds->vlan_filter)); vmxnet3_write_cmd(sc, VMXNET3_CMD_VLAN_FILTER); } static void vmxnet3_init(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); /* Use the current MAC address. */ bcopy(IF_LLADDR(sc->vmx_ifp), sc->vmx_lladdr, ETHER_ADDR_LEN); vmxnet3_set_lladdr(sc); vmxnet3_reinit_shared_data(sc); vmxnet3_reinit_queues(sc); vmxnet3_enable_device(sc); vmxnet3_reinit_rxfilters(sc); vmxnet3_link_status(sc); } static void vmxnet3_multi_set(if_ctx_t ctx) { vmxnet3_set_rxfilter(iflib_get_softc(ctx), if_getflags(iflib_get_ifp(ctx))); } static int vmxnet3_mtu_set(if_ctx_t ctx, uint32_t mtu) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; sc = iflib_get_softc(ctx); scctx = sc->vmx_scctx; if (mtu > VMXNET3_TX_MAXSIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN + ETHER_CRC_LEN)) return (EINVAL); /* * Update the max frame size so that the rx mbuf size is * chosen based on the new mtu during the interface init that * will occur after this routine returns. */ scctx->isc_max_frame_size = mtu + ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN + ETHER_CRC_LEN; /* RX completion queue - n/a */ scctx->isc_rxd_buf_size[0] = 0; /* * For header-type descriptors (used for first segment of * packet), let iflib determine the buffer size based on the * max frame size. */ scctx->isc_rxd_buf_size[1] = 0; /* * For body-type descriptors (used for jumbo frames and LRO), * always use page-sized buffers. */ scctx->isc_rxd_buf_size[2] = MJUMPAGESIZE; return (0); } static void vmxnet3_media_status(if_ctx_t ctx, struct ifmediareq * ifmr) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (vmxnet3_link_is_up(sc) != 0) { ifmr->ifm_status |= IFM_ACTIVE; ifmr->ifm_active |= IFM_AUTO; } else ifmr->ifm_active |= IFM_NONE; } static int vmxnet3_media_change(if_ctx_t ctx) { /* Ignore. */ return (0); } static int vmxnet3_promisc_set(if_ctx_t ctx, int flags) { vmxnet3_set_rxfilter(iflib_get_softc(ctx), flags); return (0); } static uint64_t vmxnet3_get_counter(if_ctx_t ctx, ift_counter cnt) { if_t ifp = iflib_get_ifp(ctx); if (cnt < IFCOUNTERS) return if_get_counter_default(ifp, cnt); return (0); } static void vmxnet3_update_admin_status(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); if (sc->vmx_ds->event != 0) vmxnet3_evintr(sc); vmxnet3_refresh_host_stats(sc); } static void vmxnet3_txq_timer(if_ctx_t ctx, uint16_t qid) { /* Host stats refresh is global, so just trigger it on txq 0 */ if (qid == 0) vmxnet3_refresh_host_stats(iflib_get_softc(ctx)); } static void vmxnet3_update_vlan_filter(struct vmxnet3_softc *sc, int add, uint16_t tag) { int idx, bit; if (tag == 0 || tag > 4095) return; idx = (tag >> 5) & 0x7F; bit = tag & 0x1F; /* Update our private VLAN bitvector. */ if (add) sc->vmx_vlan_filter[idx] |= (1 << bit); else sc->vmx_vlan_filter[idx] &= ~(1 << bit); } static void vmxnet3_vlan_register(if_ctx_t ctx, uint16_t tag) { vmxnet3_update_vlan_filter(iflib_get_softc(ctx), 1, tag); } static void vmxnet3_vlan_unregister(if_ctx_t ctx, uint16_t tag) { vmxnet3_update_vlan_filter(iflib_get_softc(ctx), 0, tag); } static u_int vmxnet3_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int count) { struct vmxnet3_softc *sc = arg; if (count < VMXNET3_MULTICAST_MAX) bcopy(LLADDR(sdl), &sc->vmx_mcast[count * ETHER_ADDR_LEN], ETHER_ADDR_LEN); return (1); } static void vmxnet3_set_rxfilter(struct vmxnet3_softc *sc, int flags) { struct ifnet *ifp; struct vmxnet3_driver_shared *ds; u_int mode; ifp = sc->vmx_ifp; ds = sc->vmx_ds; mode = VMXNET3_RXMODE_UCAST | VMXNET3_RXMODE_BCAST; if (flags & IFF_PROMISC) mode |= VMXNET3_RXMODE_PROMISC; if (flags & IFF_ALLMULTI) mode |= VMXNET3_RXMODE_ALLMULTI; else { int cnt; cnt = if_foreach_llmaddr(ifp, vmxnet3_hash_maddr, sc); if (cnt >= VMXNET3_MULTICAST_MAX) { cnt = 0; mode |= VMXNET3_RXMODE_ALLMULTI; } else if (cnt > 0) mode |= VMXNET3_RXMODE_MCAST; ds->mcast_tablelen = cnt * ETHER_ADDR_LEN; } ds->rxmode = mode; vmxnet3_write_cmd(sc, VMXNET3_CMD_SET_FILTER); vmxnet3_write_cmd(sc, VMXNET3_CMD_SET_RXMODE); } static void vmxnet3_refresh_host_stats(struct vmxnet3_softc *sc) { vmxnet3_write_cmd(sc, VMXNET3_CMD_GET_STATS); } static int vmxnet3_link_is_up(struct vmxnet3_softc *sc) { uint32_t status; status = vmxnet3_read_cmd(sc, VMXNET3_CMD_GET_LINK); return !!(status & 0x1); } static void vmxnet3_link_status(struct vmxnet3_softc *sc) { if_ctx_t ctx; uint64_t speed; int link; ctx = sc->vmx_ctx; link = vmxnet3_link_is_up(sc); speed = IF_Gbps(10); if (link != 0 && sc->vmx_link_active == 0) { sc->vmx_link_active = 1; iflib_link_state_change(ctx, LINK_STATE_UP, speed); } else if (link == 0 && sc->vmx_link_active != 0) { sc->vmx_link_active = 0; iflib_link_state_change(ctx, LINK_STATE_DOWN, speed); } } static void vmxnet3_set_lladdr(struct vmxnet3_softc *sc) { uint32_t ml, mh; ml = sc->vmx_lladdr[0]; ml |= sc->vmx_lladdr[1] << 8; ml |= sc->vmx_lladdr[2] << 16; ml |= sc->vmx_lladdr[3] << 24; vmxnet3_write_bar1(sc, VMXNET3_BAR1_MACL, ml); mh = sc->vmx_lladdr[4]; mh |= sc->vmx_lladdr[5] << 8; vmxnet3_write_bar1(sc, VMXNET3_BAR1_MACH, mh); } static void vmxnet3_get_lladdr(struct vmxnet3_softc *sc) { uint32_t ml, mh; ml = vmxnet3_read_cmd(sc, VMXNET3_CMD_GET_MACL); mh = vmxnet3_read_cmd(sc, VMXNET3_CMD_GET_MACH); sc->vmx_lladdr[0] = ml; sc->vmx_lladdr[1] = ml >> 8; sc->vmx_lladdr[2] = ml >> 16; sc->vmx_lladdr[3] = ml >> 24; sc->vmx_lladdr[4] = mh; sc->vmx_lladdr[5] = mh >> 8; } static void vmxnet3_setup_txq_sysctl(struct vmxnet3_txqueue *txq, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child) { struct sysctl_oid *node, *txsnode; struct sysctl_oid_list *list, *txslist; struct UPT1_TxStats *txstats; char namebuf[16]; txstats = &txq->vxtxq_ts->stats; snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vxtxq_id); node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue"); txq->vxtxq_sysctl = list = SYSCTL_CHILDREN(node); /* * Add statistics reported by the host. These are updated by the * iflib txq timer on txq 0. */ txsnode = SYSCTL_ADD_NODE(ctx, list, OID_AUTO, "hstats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Host Statistics"); txslist = SYSCTL_CHILDREN(txsnode); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "tso_packets", CTLFLAG_RD, &txstats->TSO_packets, "TSO packets"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "tso_bytes", CTLFLAG_RD, &txstats->TSO_bytes, "TSO bytes"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "ucast_packets", CTLFLAG_RD, &txstats->ucast_packets, "Unicast packets"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "unicast_bytes", CTLFLAG_RD, &txstats->ucast_bytes, "Unicast bytes"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "mcast_packets", CTLFLAG_RD, &txstats->mcast_packets, "Multicast packets"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "mcast_bytes", CTLFLAG_RD, &txstats->mcast_bytes, "Multicast bytes"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "error", CTLFLAG_RD, &txstats->error, "Errors"); SYSCTL_ADD_UQUAD(ctx, txslist, OID_AUTO, "discard", CTLFLAG_RD, &txstats->discard, "Discards"); } static void vmxnet3_setup_rxq_sysctl(struct vmxnet3_rxqueue *rxq, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child) { struct sysctl_oid *node, *rxsnode; struct sysctl_oid_list *list, *rxslist; struct UPT1_RxStats *rxstats; char namebuf[16]; rxstats = &rxq->vxrxq_rs->stats; snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vxrxq_id); node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue"); rxq->vxrxq_sysctl = list = SYSCTL_CHILDREN(node); /* * Add statistics reported by the host. These are updated by the * iflib txq timer on txq 0. */ rxsnode = SYSCTL_ADD_NODE(ctx, list, OID_AUTO, "hstats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Host Statistics"); rxslist = SYSCTL_CHILDREN(rxsnode); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "lro_packets", CTLFLAG_RD, &rxstats->LRO_packets, "LRO packets"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "lro_bytes", CTLFLAG_RD, &rxstats->LRO_bytes, "LRO bytes"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "ucast_packets", CTLFLAG_RD, &rxstats->ucast_packets, "Unicast packets"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "unicast_bytes", CTLFLAG_RD, &rxstats->ucast_bytes, "Unicast bytes"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "mcast_packets", CTLFLAG_RD, &rxstats->mcast_packets, "Multicast packets"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "mcast_bytes", CTLFLAG_RD, &rxstats->mcast_bytes, "Multicast bytes"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "bcast_packets", CTLFLAG_RD, &rxstats->bcast_packets, "Broadcast packets"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "bcast_bytes", CTLFLAG_RD, &rxstats->bcast_bytes, "Broadcast bytes"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "nobuffer", CTLFLAG_RD, &rxstats->nobuffer, "No buffer"); SYSCTL_ADD_UQUAD(ctx, rxslist, OID_AUTO, "error", CTLFLAG_RD, &rxstats->error, "Errors"); } static void vmxnet3_setup_debug_sysctl(struct vmxnet3_softc *sc, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child) { if_softc_ctx_t scctx; struct sysctl_oid *node; struct sysctl_oid_list *list; int i; scctx = sc->vmx_scctx; for (i = 0; i < scctx->isc_ntxqsets; i++) { struct vmxnet3_txqueue *txq = &sc->vmx_txq[i]; node = SYSCTL_ADD_NODE(ctx, txq->vxtxq_sysctl, OID_AUTO, "debug", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); list = SYSCTL_CHILDREN(node); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "cmd_next", CTLFLAG_RD, &txq->vxtxq_cmd_ring.vxtxr_next, 0, ""); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "cmd_ndesc", CTLFLAG_RD, &txq->vxtxq_cmd_ring.vxtxr_ndesc, 0, ""); SYSCTL_ADD_INT(ctx, list, OID_AUTO, "cmd_gen", CTLFLAG_RD, &txq->vxtxq_cmd_ring.vxtxr_gen, 0, ""); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "comp_next", CTLFLAG_RD, &txq->vxtxq_comp_ring.vxcr_next, 0, ""); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "comp_ndesc", CTLFLAG_RD, &txq->vxtxq_comp_ring.vxcr_ndesc, 0,""); SYSCTL_ADD_INT(ctx, list, OID_AUTO, "comp_gen", CTLFLAG_RD, &txq->vxtxq_comp_ring.vxcr_gen, 0, ""); } for (i = 0; i < scctx->isc_nrxqsets; i++) { struct vmxnet3_rxqueue *rxq = &sc->vmx_rxq[i]; node = SYSCTL_ADD_NODE(ctx, rxq->vxrxq_sysctl, OID_AUTO, "debug", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); list = SYSCTL_CHILDREN(node); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "cmd0_ndesc", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[0].vxrxr_ndesc, 0, ""); SYSCTL_ADD_INT(ctx, list, OID_AUTO, "cmd0_gen", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[0].vxrxr_gen, 0, ""); SYSCTL_ADD_U64(ctx, list, OID_AUTO, "cmd0_desc_skips", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[0].vxrxr_desc_skips, 0, ""); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "cmd1_ndesc", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[1].vxrxr_ndesc, 0, ""); SYSCTL_ADD_INT(ctx, list, OID_AUTO, "cmd1_gen", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[1].vxrxr_gen, 0, ""); SYSCTL_ADD_U64(ctx, list, OID_AUTO, "cmd1_desc_skips", CTLFLAG_RD, &rxq->vxrxq_cmd_ring[1].vxrxr_desc_skips, 0, ""); SYSCTL_ADD_UINT(ctx, list, OID_AUTO, "comp_ndesc", CTLFLAG_RD, &rxq->vxrxq_comp_ring.vxcr_ndesc, 0,""); SYSCTL_ADD_INT(ctx, list, OID_AUTO, "comp_gen", CTLFLAG_RD, &rxq->vxrxq_comp_ring.vxcr_gen, 0, ""); SYSCTL_ADD_U64(ctx, list, OID_AUTO, "comp_zero_length", CTLFLAG_RD, &rxq->vxrxq_comp_ring.vxcr_zero_length, 0, ""); SYSCTL_ADD_U64(ctx, list, OID_AUTO, "comp_pkt_errors", CTLFLAG_RD, &rxq->vxrxq_comp_ring.vxcr_pkt_errors, 0, ""); } } static void vmxnet3_setup_queue_sysctl(struct vmxnet3_softc *sc, struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child) { if_softc_ctx_t scctx; int i; scctx = sc->vmx_scctx; for (i = 0; i < scctx->isc_ntxqsets; i++) vmxnet3_setup_txq_sysctl(&sc->vmx_txq[i], ctx, child); for (i = 0; i < scctx->isc_nrxqsets; i++) vmxnet3_setup_rxq_sysctl(&sc->vmx_rxq[i], ctx, child); vmxnet3_setup_debug_sysctl(sc, ctx, child); } static void vmxnet3_setup_sysctl(struct vmxnet3_softc *sc) { device_t dev; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; dev = sc->vmx_dev; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); vmxnet3_setup_queue_sysctl(sc, ctx, child); } static void vmxnet3_write_bar0(struct vmxnet3_softc *sc, bus_size_t r, uint32_t v) { bus_space_write_4(sc->vmx_iot0, sc->vmx_ioh0, r, v); } static uint32_t vmxnet3_read_bar1(struct vmxnet3_softc *sc, bus_size_t r) { return (bus_space_read_4(sc->vmx_iot1, sc->vmx_ioh1, r)); } static void vmxnet3_write_bar1(struct vmxnet3_softc *sc, bus_size_t r, uint32_t v) { bus_space_write_4(sc->vmx_iot1, sc->vmx_ioh1, r, v); } static void vmxnet3_write_cmd(struct vmxnet3_softc *sc, uint32_t cmd) { vmxnet3_write_bar1(sc, VMXNET3_BAR1_CMD, cmd); } static uint32_t vmxnet3_read_cmd(struct vmxnet3_softc *sc, uint32_t cmd) { vmxnet3_write_cmd(sc, cmd); bus_space_barrier(sc->vmx_iot1, sc->vmx_ioh1, 0, 0, BUS_SPACE_BARRIER_READ | BUS_SPACE_BARRIER_WRITE); return (vmxnet3_read_bar1(sc, VMXNET3_BAR1_CMD)); } static void vmxnet3_enable_intr(struct vmxnet3_softc *sc, int irq) { vmxnet3_write_bar0(sc, VMXNET3_BAR0_IMASK(irq), 0); } static void vmxnet3_disable_intr(struct vmxnet3_softc *sc, int irq) { vmxnet3_write_bar0(sc, VMXNET3_BAR0_IMASK(irq), 1); } static int vmxnet3_tx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { /* Not using interrupts for TX */ return (0); } static int vmxnet3_rx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); vmxnet3_enable_intr(sc, sc->vmx_rxq[qid].vxrxq_intr_idx); return (0); } static void vmxnet3_link_intr_enable(if_ctx_t ctx) { struct vmxnet3_softc *sc; sc = iflib_get_softc(ctx); vmxnet3_enable_intr(sc, sc->vmx_event_intr_idx); } static void vmxnet3_intr_enable_all(if_ctx_t ctx) { struct vmxnet3_softc *sc; if_softc_ctx_t scctx; int i; sc = iflib_get_softc(ctx); scctx = sc->vmx_scctx; sc->vmx_ds->ictrl &= ~VMXNET3_ICTRL_DISABLE_ALL; for (i = 0; i < scctx->isc_vectors; i++) vmxnet3_enable_intr(sc, i); } static void vmxnet3_intr_disable_all(if_ctx_t ctx) { struct vmxnet3_softc *sc; int i; sc = iflib_get_softc(ctx); /* * iflib may invoke this routine before vmxnet3_attach_post() has * run, which is before the top level shared data area is * initialized and the device made aware of it. */ if (sc->vmx_ds != NULL) sc->vmx_ds->ictrl |= VMXNET3_ICTRL_DISABLE_ALL; for (i = 0; i < VMXNET3_MAX_INTRS; i++) vmxnet3_disable_intr(sc, i); } /* * Since this is a purely paravirtualized device, we do not have * to worry about DMA coherency. But at times, we must make sure * both the compiler and CPU do not reorder memory operations. */ static inline void vmxnet3_barrier(struct vmxnet3_softc *sc, vmxnet3_barrier_t type) { switch (type) { case VMXNET3_BARRIER_RD: rmb(); break; case VMXNET3_BARRIER_WR: wmb(); break; case VMXNET3_BARRIER_RDWR: mb(); break; default: panic("%s: bad barrier type %d", __func__, type); } } Index: head/sys/net/iflib.c =================================================================== --- head/sys/net/iflib.c (revision 365060) +++ head/sys/net/iflib.c (revision 365061) @@ -1,6921 +1,6932 @@ /*- * Copyright (c) 2014-2018, Matthew 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. Neither the name of Matthew Macy nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_acpi.h" #include "opt_sched.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 #include #include #include "ifdi_if.h" #ifdef PCI_IOV #include #endif #include /* * enable accounting of every mbuf as it comes in to and goes out of * iflib's software descriptor references */ #define MEMORY_LOGGING 0 /* * Enable mbuf vectors for compressing long mbuf chains */ /* * NB: * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead * we prefetch needs to be determined by the time spent in m_free vis a vis * the cost of a prefetch. This will of course vary based on the workload: * - NFLX's m_free path is dominated by vm-based M_EXT manipulation which * is quite expensive, thus suggesting very little prefetch. * - small packet forwarding which is just returning a single mbuf to * UMA will typically be very fast vis a vis the cost of a memory * access. */ /* * File organization: * - private structures * - iflib private utility functions * - ifnet functions * - vlan registry and other exported functions * - iflib public core functions * * */ MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library"); #define IFLIB_RXEOF_MORE (1U << 0) #define IFLIB_RXEOF_EMPTY (2U << 0) struct iflib_txq; typedef struct iflib_txq *iflib_txq_t; struct iflib_rxq; typedef struct iflib_rxq *iflib_rxq_t; struct iflib_fl; typedef struct iflib_fl *iflib_fl_t; struct iflib_ctx; static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid); static void iflib_timer(void *arg); typedef struct iflib_filter_info { driver_filter_t *ifi_filter; void *ifi_filter_arg; struct grouptask *ifi_task; void *ifi_ctx; } *iflib_filter_info_t; struct iflib_ctx { KOBJ_FIELDS; /* * Pointer to hardware driver's softc */ void *ifc_softc; device_t ifc_dev; if_t ifc_ifp; cpuset_t ifc_cpus; if_shared_ctx_t ifc_sctx; struct if_softc_ctx ifc_softc_ctx; struct sx ifc_ctx_sx; struct mtx ifc_state_mtx; iflib_txq_t ifc_txqs; iflib_rxq_t ifc_rxqs; uint32_t ifc_if_flags; uint32_t ifc_flags; uint32_t ifc_max_fl_buf_size; uint32_t ifc_rx_mbuf_sz; int ifc_link_state; int ifc_watchdog_events; struct cdev *ifc_led_dev; struct resource *ifc_msix_mem; struct if_irq ifc_legacy_irq; struct grouptask ifc_admin_task; struct grouptask ifc_vflr_task; struct iflib_filter_info ifc_filter_info; struct ifmedia ifc_media; struct ifmedia *ifc_mediap; struct sysctl_oid *ifc_sysctl_node; uint16_t ifc_sysctl_ntxqs; uint16_t ifc_sysctl_nrxqs; uint16_t ifc_sysctl_qs_eq_override; uint16_t ifc_sysctl_rx_budget; uint16_t ifc_sysctl_tx_abdicate; uint16_t ifc_sysctl_core_offset; #define CORE_OFFSET_UNSPECIFIED 0xffff uint8_t ifc_sysctl_separate_txrx; qidx_t ifc_sysctl_ntxds[8]; qidx_t ifc_sysctl_nrxds[8]; struct if_txrx ifc_txrx; #define isc_txd_encap ifc_txrx.ift_txd_encap #define isc_txd_flush ifc_txrx.ift_txd_flush #define isc_txd_credits_update ifc_txrx.ift_txd_credits_update #define isc_rxd_available ifc_txrx.ift_rxd_available #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get #define isc_rxd_refill ifc_txrx.ift_rxd_refill #define isc_rxd_flush ifc_txrx.ift_rxd_flush #define isc_legacy_intr ifc_txrx.ift_legacy_intr eventhandler_tag ifc_vlan_attach_event; eventhandler_tag ifc_vlan_detach_event; struct ether_addr ifc_mac; }; void * iflib_get_softc(if_ctx_t ctx) { return (ctx->ifc_softc); } device_t iflib_get_dev(if_ctx_t ctx) { return (ctx->ifc_dev); } if_t iflib_get_ifp(if_ctx_t ctx) { return (ctx->ifc_ifp); } struct ifmedia * iflib_get_media(if_ctx_t ctx) { return (ctx->ifc_mediap); } uint32_t iflib_get_flags(if_ctx_t ctx) { return (ctx->ifc_flags); } void iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN]) { bcopy(mac, ctx->ifc_mac.octet, ETHER_ADDR_LEN); } if_softc_ctx_t iflib_get_softc_ctx(if_ctx_t ctx) { return (&ctx->ifc_softc_ctx); } if_shared_ctx_t iflib_get_sctx(if_ctx_t ctx) { return (ctx->ifc_sctx); } #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2) #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*)) #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1))) #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP) #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF) typedef struct iflib_sw_rx_desc_array { bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ struct mbuf **ifsd_m; /* pkthdr mbufs */ caddr_t *ifsd_cl; /* direct cluster pointer for rx */ bus_addr_t *ifsd_ba; /* bus addr of cluster for rx */ } iflib_rxsd_array_t; typedef struct iflib_sw_tx_desc_array { bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ bus_dmamap_t *ifsd_tso_map; /* bus_dma maps for TSO packet */ struct mbuf **ifsd_m; /* pkthdr mbufs */ } if_txsd_vec_t; /* magic number that should be high enough for any hardware */ #define IFLIB_MAX_TX_SEGS 128 #define IFLIB_RX_COPY_THRESH 128 #define IFLIB_MAX_RX_REFRESH 32 /* The minimum descriptors per second before we start coalescing */ #define IFLIB_MIN_DESC_SEC 16384 #define IFLIB_DEFAULT_TX_UPDATE_FREQ 16 #define IFLIB_QUEUE_IDLE 0 #define IFLIB_QUEUE_HUNG 1 #define IFLIB_QUEUE_WORKING 2 /* maximum number of txqs that can share an rx interrupt */ #define IFLIB_MAX_TX_SHARED_INTR 4 /* this should really scale with ring size - this is a fairly arbitrary value */ #define TX_BATCH_SIZE 32 #define IFLIB_RESTART_BUDGET 8 #define CSUM_OFFLOAD (CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \ CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \ CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP) struct iflib_txq { qidx_t ift_in_use; qidx_t ift_cidx; qidx_t ift_cidx_processed; qidx_t ift_pidx; uint8_t ift_gen; uint8_t ift_br_offset; uint16_t ift_npending; uint16_t ift_db_pending; uint16_t ift_rs_pending; /* implicit pad */ uint8_t ift_txd_size[8]; uint64_t ift_processed; uint64_t ift_cleaned; uint64_t ift_cleaned_prev; #if MEMORY_LOGGING uint64_t ift_enqueued; uint64_t ift_dequeued; #endif uint64_t ift_no_tx_dma_setup; uint64_t ift_no_desc_avail; uint64_t ift_mbuf_defrag_failed; uint64_t ift_mbuf_defrag; uint64_t ift_map_failed; uint64_t ift_txd_encap_efbig; uint64_t ift_pullups; uint64_t ift_last_timer_tick; struct mtx ift_mtx; struct mtx ift_db_mtx; /* constant values */ if_ctx_t ift_ctx; struct ifmp_ring *ift_br; struct grouptask ift_task; qidx_t ift_size; uint16_t ift_id; struct callout ift_timer; if_txsd_vec_t ift_sds; uint8_t ift_qstatus; uint8_t ift_closed; uint8_t ift_update_freq; struct iflib_filter_info ift_filter_info; bus_dma_tag_t ift_buf_tag; bus_dma_tag_t ift_tso_buf_tag; iflib_dma_info_t ift_ifdi; #define MTX_NAME_LEN 32 char ift_mtx_name[MTX_NAME_LEN]; bus_dma_segment_t ift_segs[IFLIB_MAX_TX_SEGS] __aligned(CACHE_LINE_SIZE); #ifdef IFLIB_DIAGNOSTICS uint64_t ift_cpu_exec_count[256]; #endif } __aligned(CACHE_LINE_SIZE); struct iflib_fl { qidx_t ifl_cidx; qidx_t ifl_pidx; qidx_t ifl_credits; uint8_t ifl_gen; uint8_t ifl_rxd_size; #if MEMORY_LOGGING uint64_t ifl_m_enqueued; uint64_t ifl_m_dequeued; uint64_t ifl_cl_enqueued; uint64_t ifl_cl_dequeued; #endif /* implicit pad */ bitstr_t *ifl_rx_bitmap; qidx_t ifl_fragidx; /* constant */ qidx_t ifl_size; uint16_t ifl_buf_size; uint16_t ifl_cltype; uma_zone_t ifl_zone; iflib_rxsd_array_t ifl_sds; iflib_rxq_t ifl_rxq; uint8_t ifl_id; bus_dma_tag_t ifl_buf_tag; iflib_dma_info_t ifl_ifdi; uint64_t ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE); qidx_t ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH]; } __aligned(CACHE_LINE_SIZE); static inline qidx_t get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen) { qidx_t used; if (pidx > cidx) used = pidx - cidx; else if (pidx < cidx) used = size - cidx + pidx; else if (gen == 0 && pidx == cidx) used = 0; else if (gen == 1 && pidx == cidx) used = size; else panic("bad state"); return (used); } #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen)) #define IDXDIFF(head, tail, wrap) \ ((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head)) struct iflib_rxq { if_ctx_t ifr_ctx; iflib_fl_t ifr_fl; uint64_t ifr_rx_irq; struct pfil_head *pfil; /* * If there is a separate completion queue (IFLIB_HAS_RXCQ), this is * the completion queue consumer index. Otherwise it's unused. */ qidx_t ifr_cq_cidx; uint16_t ifr_id; uint8_t ifr_nfl; uint8_t ifr_ntxqirq; uint8_t ifr_txqid[IFLIB_MAX_TX_SHARED_INTR]; uint8_t ifr_fl_offset; struct lro_ctrl ifr_lc; struct grouptask ifr_task; struct callout ifr_watchdog; struct iflib_filter_info ifr_filter_info; iflib_dma_info_t ifr_ifdi; /* dynamically allocate if any drivers need a value substantially larger than this */ struct if_rxd_frag ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE); #ifdef IFLIB_DIAGNOSTICS uint64_t ifr_cpu_exec_count[256]; #endif } __aligned(CACHE_LINE_SIZE); typedef struct if_rxsd { caddr_t *ifsd_cl; iflib_fl_t ifsd_fl; } *if_rxsd_t; /* multiple of word size */ #ifdef __LP64__ #define PKT_INFO_SIZE 6 #define RXD_INFO_SIZE 5 #define PKT_TYPE uint64_t #else #define PKT_INFO_SIZE 11 #define RXD_INFO_SIZE 8 #define PKT_TYPE uint32_t #endif #define PKT_LOOP_BOUND ((PKT_INFO_SIZE/3)*3) #define RXD_LOOP_BOUND ((RXD_INFO_SIZE/4)*4) typedef struct if_pkt_info_pad { PKT_TYPE pkt_val[PKT_INFO_SIZE]; } *if_pkt_info_pad_t; typedef struct if_rxd_info_pad { PKT_TYPE rxd_val[RXD_INFO_SIZE]; } *if_rxd_info_pad_t; CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info)); CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info)); static inline void pkt_info_zero(if_pkt_info_t pi) { if_pkt_info_pad_t pi_pad; pi_pad = (if_pkt_info_pad_t)pi; pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0; pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0; #ifndef __LP64__ pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0; pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0; #endif } static device_method_t iflib_pseudo_methods[] = { DEVMETHOD(device_attach, noop_attach), DEVMETHOD(device_detach, iflib_pseudo_detach), DEVMETHOD_END }; driver_t iflib_pseudodriver = { "iflib_pseudo", iflib_pseudo_methods, sizeof(struct iflib_ctx), }; static inline void rxd_info_zero(if_rxd_info_t ri) { if_rxd_info_pad_t ri_pad; int i; ri_pad = (if_rxd_info_pad_t)ri; for (i = 0; i < RXD_LOOP_BOUND; i += 4) { ri_pad->rxd_val[i] = 0; ri_pad->rxd_val[i+1] = 0; ri_pad->rxd_val[i+2] = 0; ri_pad->rxd_val[i+3] = 0; } #ifdef __LP64__ ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0; #endif } /* * Only allow a single packet to take up most 1/nth of the tx ring */ #define MAX_SINGLE_PACKET_FRACTION 12 #define IF_BAD_DMA (bus_addr_t)-1 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING)) #define CTX_LOCK_INIT(_sc) sx_init(&(_sc)->ifc_ctx_sx, "iflib ctx lock") #define CTX_LOCK(ctx) sx_xlock(&(ctx)->ifc_ctx_sx) #define CTX_UNLOCK(ctx) sx_xunlock(&(ctx)->ifc_ctx_sx) #define CTX_LOCK_DESTROY(ctx) sx_destroy(&(ctx)->ifc_ctx_sx) #define STATE_LOCK_INIT(_sc, _name) mtx_init(&(_sc)->ifc_state_mtx, _name, "iflib state lock", MTX_DEF) #define STATE_LOCK(ctx) mtx_lock(&(ctx)->ifc_state_mtx) #define STATE_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_state_mtx) #define STATE_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_state_mtx) #define CALLOUT_LOCK(txq) mtx_lock(&txq->ift_mtx) #define CALLOUT_UNLOCK(txq) mtx_unlock(&txq->ift_mtx) void iflib_set_detach(if_ctx_t ctx) { STATE_LOCK(ctx); ctx->ifc_flags |= IFC_IN_DETACH; STATE_UNLOCK(ctx); } /* Our boot-time initialization hook */ static int iflib_module_event_handler(module_t, int, void *); static moduledata_t iflib_moduledata = { "iflib", iflib_module_event_handler, NULL }; DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY); MODULE_VERSION(iflib, 1); MODULE_DEPEND(iflib, pci, 1, 1, 1); MODULE_DEPEND(iflib, ether, 1, 1, 1); TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1); TASKQGROUP_DEFINE(if_config_tqg, 1, 1); #ifndef IFLIB_DEBUG_COUNTERS #ifdef INVARIANTS #define IFLIB_DEBUG_COUNTERS 1 #else #define IFLIB_DEBUG_COUNTERS 0 #endif /* !INVARIANTS */ #endif static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "iflib driver parameters"); /* * XXX need to ensure that this can't accidentally cause the head to be moved backwards */ static int iflib_min_tx_latency = 0; SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW, &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput"); static int iflib_no_tx_batch = 0; SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW, &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput"); #if IFLIB_DEBUG_COUNTERS static int iflib_tx_seen; static int iflib_tx_sent; static int iflib_tx_encap; static int iflib_rx_allocs; static int iflib_fl_refills; static int iflib_fl_refills_large; static int iflib_tx_frees; SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD, &iflib_tx_seen, 0, "# TX mbufs seen"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD, &iflib_tx_sent, 0, "# TX mbufs sent"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD, &iflib_tx_encap, 0, "# TX mbufs encapped"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD, &iflib_tx_frees, 0, "# TX frees"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD, &iflib_rx_allocs, 0, "# RX allocations"); SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD, &iflib_fl_refills, 0, "# refills"); SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD, &iflib_fl_refills_large, 0, "# large refills"); static int iflib_txq_drain_flushing; static int iflib_txq_drain_oactive; static int iflib_txq_drain_notready; SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD, &iflib_txq_drain_flushing, 0, "# drain flushes"); SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD, &iflib_txq_drain_oactive, 0, "# drain oactives"); SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD, &iflib_txq_drain_notready, 0, "# drain notready"); static int iflib_encap_load_mbuf_fail; static int iflib_encap_pad_mbuf_fail; static int iflib_encap_txq_avail_fail; static int iflib_encap_txd_encap_fail; SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD, &iflib_encap_load_mbuf_fail, 0, "# busdma load failures"); SYSCTL_INT(_net_iflib, OID_AUTO, encap_pad_mbuf_fail, CTLFLAG_RD, &iflib_encap_pad_mbuf_fail, 0, "# runt frame pad failures"); SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD, &iflib_encap_txq_avail_fail, 0, "# txq avail failures"); SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD, &iflib_encap_txd_encap_fail, 0, "# driver encap failures"); static int iflib_task_fn_rxs; static int iflib_rx_intr_enables; static int iflib_fast_intrs; static int iflib_rx_unavail; static int iflib_rx_ctx_inactive; static int iflib_rx_if_input; static int iflib_rxd_flush; static int iflib_verbose_debug; SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD, &iflib_task_fn_rxs, 0, "# task_fn_rx calls"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD, &iflib_rx_intr_enables, 0, "# RX intr enables"); SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD, &iflib_fast_intrs, 0, "# fast_intr calls"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD, &iflib_rx_unavail, 0, "# times rxeof called with no available data"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD, &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD, &iflib_rx_if_input, 0, "# times rxeof called if_input"); SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD, &iflib_rxd_flush, 0, "# times rxd_flush called"); SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW, &iflib_verbose_debug, 0, "enable verbose debugging"); #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1) static void iflib_debug_reset(void) { iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs = iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees = iflib_txq_drain_flushing = iflib_txq_drain_oactive = iflib_txq_drain_notready = iflib_encap_load_mbuf_fail = iflib_encap_pad_mbuf_fail = iflib_encap_txq_avail_fail = iflib_encap_txd_encap_fail = iflib_task_fn_rxs = iflib_rx_intr_enables = iflib_fast_intrs = iflib_rx_unavail = iflib_rx_ctx_inactive = iflib_rx_if_input = iflib_rxd_flush = 0; } #else #define DBG_COUNTER_INC(name) static void iflib_debug_reset(void) {} #endif #define IFLIB_DEBUG 0 static void iflib_tx_structures_free(if_ctx_t ctx); static void iflib_rx_structures_free(if_ctx_t ctx); static int iflib_queues_alloc(if_ctx_t ctx); static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq); static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget); static int iflib_qset_structures_setup(if_ctx_t ctx); static int iflib_msix_init(if_ctx_t ctx); static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, const char *str); static void iflib_txq_check_drain(iflib_txq_t txq, int budget); static uint32_t iflib_txq_can_drain(struct ifmp_ring *); #ifdef ALTQ static void iflib_altq_if_start(if_t ifp); static int iflib_altq_if_transmit(if_t ifp, struct mbuf *m); #endif static int iflib_register(if_ctx_t); static void iflib_deregister(if_ctx_t); static void iflib_unregister_vlan_handlers(if_ctx_t ctx); static uint16_t iflib_get_mbuf_size_for(unsigned int size); static void iflib_init_locked(if_ctx_t ctx); static void iflib_add_device_sysctl_pre(if_ctx_t ctx); static void iflib_add_device_sysctl_post(if_ctx_t ctx); static void iflib_ifmp_purge(iflib_txq_t txq); static void _iflib_pre_assert(if_softc_ctx_t scctx); static void iflib_if_init_locked(if_ctx_t ctx); static void iflib_free_intr_mem(if_ctx_t ctx); #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * iflib_fixup_rx(struct mbuf *m); #endif static SLIST_HEAD(cpu_offset_list, cpu_offset) cpu_offsets = SLIST_HEAD_INITIALIZER(cpu_offsets); struct cpu_offset { SLIST_ENTRY(cpu_offset) entries; cpuset_t set; unsigned int refcount; uint16_t offset; }; static struct mtx cpu_offset_mtx; MTX_SYSINIT(iflib_cpu_offset, &cpu_offset_mtx, "iflib_cpu_offset lock", MTX_DEF); DEBUGNET_DEFINE(iflib); static int iflib_num_rx_descs(if_ctx_t ctx) { if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; uint16_t first_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0; return scctx->isc_nrxd[first_rxq]; } static int iflib_num_tx_descs(if_ctx_t ctx) { if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; uint16_t first_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0; return scctx->isc_ntxd[first_txq]; } #ifdef DEV_NETMAP #include #include #include MODULE_DEPEND(iflib, netmap, 1, 1, 1); static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, bool init); /* * device-specific sysctl variables: * * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it. * During regular operations the CRC is stripped, but on some * hardware reception of frames not multiple of 64 is slower, * so using crcstrip=0 helps in benchmarks. * * iflib_rx_miss, iflib_rx_miss_bufs: * count packets that might be missed due to lost interrupts. */ SYSCTL_DECL(_dev_netmap); /* * The xl driver by default strips CRCs and we do not override it. */ int iflib_crcstrip = 1; SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip, CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on RX frames"); int iflib_rx_miss, iflib_rx_miss_bufs; SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss, CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed RX intr"); SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs, CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed RX intr bufs"); /* * Register/unregister. We are already under netmap lock. * Only called on the first register or the last unregister. */ static int iflib_netmap_register(struct netmap_adapter *na, int onoff) { if_t ifp = na->ifp; if_ctx_t ctx = ifp->if_softc; int status; CTX_LOCK(ctx); IFDI_INTR_DISABLE(ctx); /* Tell the stack that the interface is no longer active */ ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); if (!CTX_IS_VF(ctx)) IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); iflib_stop(ctx); /* * Enable (or disable) netmap flags, and intercept (or restore) * ifp->if_transmit. This is done once the device has been stopped * to prevent race conditions. */ if (onoff) { nm_set_native_flags(na); } else { nm_clear_native_flags(na); } iflib_init_locked(ctx); IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ? status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1; if (status) nm_clear_native_flags(na); CTX_UNLOCK(ctx); return (status); } static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, bool init) { struct netmap_adapter *na = kring->na; u_int const lim = kring->nkr_num_slots - 1; u_int nm_i = kring->nr_hwcur; struct netmap_ring *ring = kring->ring; bus_dmamap_t *map; struct if_rxd_update iru; if_ctx_t ctx = rxq->ifr_ctx; iflib_fl_t fl = &rxq->ifr_fl[0]; u_int nic_i_first, nic_i; int i, n; #if IFLIB_DEBUG_COUNTERS int rf_count = 0; #endif /* * This function is used both at initialization and in rxsync. * At initialization we need to prepare (with isc_rxd_refill()) * all the (N) netmap buffers in the ring, in such a way to keep * fl->ifl_pidx and kring->nr_hwcur in sync (except for * kring->nkr_hwofs); at rxsync time, both indexes point to the * next buffer to be refilled. * In any case we publish (with isc_rxd_flush()) up to * (fl->ifl_pidx - 1) % N (included), to avoid the NIC tail/prod * pointer to overrun the head/cons pointer, although this is * not necessary for some NICs (e.g. vmx). */ if (__predict_false(init)) n = kring->nkr_num_slots; else { n = kring->rhead - nm_i; if (n == 0) return (0); /* Nothing to do. */ if (n < 0) n += kring->nkr_num_slots; } /* Start to refill from nr_hwcur, publishing n buffers. */ iru_init(&iru, rxq, 0 /* flid */); map = fl->ifl_sds.ifsd_map; nic_i = fl->ifl_pidx; MPASS(nic_i == netmap_idx_k2n(kring, nm_i)); DBG_COUNTER_INC(fl_refills); while (n > 0) { #if IFLIB_DEBUG_COUNTERS if (++rf_count == 9) DBG_COUNTER_INC(fl_refills_large); #endif nic_i_first = nic_i; for (i = 0; n > 0 && i < IFLIB_MAX_RX_REFRESH; n--, i++) { struct netmap_slot *slot = &ring->slot[nm_i]; void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[i]); MPASS(i < IFLIB_MAX_RX_REFRESH); if (addr == NETMAP_BUF_BASE(na)) /* bad buf */ return netmap_ring_reinit(kring); fl->ifl_rxd_idxs[i] = nic_i; if (__predict_false(init)) { netmap_load_map(na, fl->ifl_buf_tag, map[nic_i], addr); } else if (slot->flags & NS_BUF_CHANGED) { /* buffer has changed, reload map */ netmap_reload_map(na, fl->ifl_buf_tag, map[nic_i], addr); } bus_dmamap_sync(fl->ifl_buf_tag, map[nic_i], BUS_DMASYNC_PREREAD); slot->flags &= ~NS_BUF_CHANGED; nm_i = nm_next(nm_i, lim); nic_i = nm_next(nic_i, lim); } iru.iru_pidx = nic_i_first; iru.iru_count = i; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); } fl->ifl_pidx = nic_i; MPASS(!init || nm_i == 0); MPASS(nm_i == kring->rhead); kring->nr_hwcur = nm_i; bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nm_prev(nic_i, lim)); DBG_COUNTER_INC(rxd_flush); return (0); } /* * Reconcile kernel and user view of the transmit ring. * * All information is in the kring. * Userspace wants to send packets up to the one before kring->rhead, * kernel knows kring->nr_hwcur is the first unsent packet. * * Here we push packets out (as many as possible), and possibly * reclaim buffers from previously completed transmission. * * The caller (netmap) guarantees that there is only one instance * running at any time. Any interference with other driver * methods should be handled by the individual drivers. */ static int iflib_netmap_txsync(struct netmap_kring *kring, int flags) { struct netmap_adapter *na = kring->na; if_t ifp = na->ifp; struct netmap_ring *ring = kring->ring; u_int nm_i; /* index into the netmap kring */ u_int nic_i; /* index into the NIC ring */ u_int n; u_int const lim = kring->nkr_num_slots - 1; u_int const head = kring->rhead; struct if_pkt_info pi; /* * interrupts on every tx packet are expensive so request * them every half ring, or where NS_REPORT is set */ u_int report_frequency = kring->nkr_num_slots >> 1; /* device-specific */ if_ctx_t ctx = ifp->if_softc; iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id]; bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * First part: process new packets to send. * nm_i is the current index in the netmap kring, * nic_i is the corresponding index in the NIC ring. * * If we have packets to send (nm_i != head) * iterate over the netmap ring, fetch length and update * the corresponding slot in the NIC ring. Some drivers also * need to update the buffer's physical address in the NIC slot * even NS_BUF_CHANGED is not set (PNMB computes the addresses). * * The netmap_reload_map() calls is especially expensive, * even when (as in this case) the tag is 0, so do only * when the buffer has actually changed. * * If possible do not set the report/intr bit on all slots, * but only a few times per ring or when NS_REPORT is set. * * Finally, on 10G and faster drivers, it might be useful * to prefetch the next slot and txr entry. */ nm_i = kring->nr_hwcur; if (nm_i != head) { /* we have new packets to send */ pkt_info_zero(&pi); pi.ipi_segs = txq->ift_segs; pi.ipi_qsidx = kring->ring_id; nic_i = netmap_idx_k2n(kring, nm_i); __builtin_prefetch(&ring->slot[nm_i]); __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]); __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]); for (n = 0; nm_i != head; n++) { struct netmap_slot *slot = &ring->slot[nm_i]; u_int len = slot->len; uint64_t paddr; void *addr = PNMB(na, slot, &paddr); int flags = (slot->flags & NS_REPORT || nic_i == 0 || nic_i == report_frequency) ? IPI_TX_INTR : 0; /* device-specific */ pi.ipi_len = len; pi.ipi_segs[0].ds_addr = paddr; pi.ipi_segs[0].ds_len = len; pi.ipi_nsegs = 1; pi.ipi_ndescs = 0; pi.ipi_pidx = nic_i; pi.ipi_flags = flags; /* Fill the slot in the NIC ring. */ ctx->isc_txd_encap(ctx->ifc_softc, &pi); DBG_COUNTER_INC(tx_encap); /* prefetch for next round */ __builtin_prefetch(&ring->slot[nm_i + 1]); __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]); __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]); NM_CHECK_ADDR_LEN(na, addr, len); if (slot->flags & NS_BUF_CHANGED) { /* buffer has changed, reload map */ netmap_reload_map(na, txq->ift_buf_tag, txq->ift_sds.ifsd_map[nic_i], addr); } /* make sure changes to the buffer are synced */ bus_dmamap_sync(txq->ift_buf_tag, txq->ift_sds.ifsd_map[nic_i], BUS_DMASYNC_PREWRITE); slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED); nm_i = nm_next(nm_i, lim); nic_i = nm_next(nic_i, lim); } kring->nr_hwcur = nm_i; /* synchronize the NIC ring */ bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* (re)start the tx unit up to slot nic_i (excluded) */ ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i); } /* * Second part: reclaim buffers for completed transmissions. * * If there are unclaimed buffers, attempt to reclaim them. * If none are reclaimed, and TX IRQs are not in use, do an initial * minimal delay, then trigger the tx handler which will spin in the * group task queue. */ if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) { if (iflib_tx_credits_update(ctx, txq)) { /* some tx completed, increment avail */ nic_i = txq->ift_cidx_processed; kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim); } } if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) { callout_reset_on(&txq->ift_timer, hz < 2000 ? 1 : hz / 1000, iflib_timer, txq, txq->ift_timer.c_cpu); } return (0); } /* * Reconcile kernel and user view of the receive ring. * Same as for the txsync, this routine must be efficient. * The caller guarantees a single invocations, but races against * the rest of the driver should be handled here. * * On call, kring->rhead is the first packet that userspace wants * to keep, and kring->rcur is the wakeup point. * The kernel has previously reported packets up to kring->rtail. * * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective * of whether or not we received an interrupt. */ static int iflib_netmap_rxsync(struct netmap_kring *kring, int flags) { struct netmap_adapter *na = kring->na; struct netmap_ring *ring = kring->ring; if_t ifp = na->ifp; uint32_t nm_i; /* index into the netmap ring */ uint32_t nic_i; /* index into the NIC ring */ u_int n; u_int const lim = kring->nkr_num_slots - 1; int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR; if_ctx_t ctx = ifp->if_softc; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id]; iflib_fl_t fl = &rxq->ifr_fl[0]; struct if_rxd_info ri; qidx_t *cidxp; /* * netmap only uses free list 0, to avoid out of order consumption * of receive buffers */ bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * First part: import newly received packets. * * nm_i is the index of the next free slot in the netmap ring, * nic_i is the index of the next received packet in the NIC ring * (or in the free list 0 if IFLIB_HAS_RXCQ is set), and they may * differ in case if_init() has been called while * in netmap mode. For the receive ring we have * * nic_i = fl->ifl_cidx; * nm_i = kring->nr_hwtail (previous) * and * nm_i == (nic_i + kring->nkr_hwofs) % ring_size * * fl->ifl_cidx is set to 0 on a ring reinit */ if (netmap_no_pendintr || force_update) { uint32_t hwtail_lim = nm_prev(kring->nr_hwcur, lim); bool have_rxcq = sctx->isc_flags & IFLIB_HAS_RXCQ; int crclen = iflib_crcstrip ? 0 : 4; int error, avail; /* * For the free list consumer index, we use the same * logic as in iflib_rxeof(). */ if (have_rxcq) cidxp = &rxq->ifr_cq_cidx; else cidxp = &fl->ifl_cidx; avail = ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, *cidxp, USHRT_MAX); nic_i = fl->ifl_cidx; nm_i = netmap_idx_n2k(kring, nic_i); MPASS(nm_i == kring->nr_hwtail); for (n = 0; avail > 0 && nm_i != hwtail_lim; n++, avail--) { rxd_info_zero(&ri); ri.iri_frags = rxq->ifr_frags; ri.iri_qsidx = kring->ring_id; ri.iri_ifp = ctx->ifc_ifp; ri.iri_cidx = *cidxp; error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); ring->slot[nm_i].len = error ? 0 : ri.iri_len - crclen; ring->slot[nm_i].flags = 0; if (have_rxcq) { *cidxp = ri.iri_cidx; while (*cidxp >= scctx->isc_nrxd[0]) *cidxp -= scctx->isc_nrxd[0]; } bus_dmamap_sync(fl->ifl_buf_tag, fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD); nm_i = nm_next(nm_i, lim); fl->ifl_cidx = nic_i = nm_next(nic_i, lim); } if (n) { /* update the state variables */ if (netmap_no_pendintr && !force_update) { /* diagnostics */ iflib_rx_miss ++; iflib_rx_miss_bufs += n; } kring->nr_hwtail = nm_i; } kring->nr_kflags &= ~NKR_PENDINTR; } /* * Second part: skip past packets that userspace has released. * (kring->nr_hwcur to head excluded), * and make the buffers available for reception. * As usual nm_i is the index in the netmap ring, * nic_i is the index in the NIC ring, and * nm_i == (nic_i + kring->nkr_hwofs) % ring_size */ netmap_fl_refill(rxq, kring, false); return (0); } static void iflib_netmap_intr(struct netmap_adapter *na, int onoff) { if_ctx_t ctx = na->ifp->if_softc; CTX_LOCK(ctx); if (onoff) { IFDI_INTR_ENABLE(ctx); } else { IFDI_INTR_DISABLE(ctx); } CTX_UNLOCK(ctx); } static int iflib_netmap_attach(if_ctx_t ctx) { struct netmap_adapter na; bzero(&na, sizeof(na)); na.ifp = ctx->ifc_ifp; na.na_flags = NAF_BDG_MAYSLEEP; MPASS(ctx->ifc_softc_ctx.isc_ntxqsets); MPASS(ctx->ifc_softc_ctx.isc_nrxqsets); na.num_tx_desc = iflib_num_tx_descs(ctx); na.num_rx_desc = iflib_num_rx_descs(ctx); na.nm_txsync = iflib_netmap_txsync; na.nm_rxsync = iflib_netmap_rxsync; na.nm_register = iflib_netmap_register; na.nm_intr = iflib_netmap_intr; na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets; na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets; return (netmap_attach(&na)); } static int iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq) { struct netmap_adapter *na = NA(ctx->ifc_ifp); struct netmap_slot *slot; slot = netmap_reset(na, NR_TX, txq->ift_id, 0); if (slot == NULL) return (0); for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) { /* * In netmap mode, set the map for the packet buffer. * NOTE: Some drivers (not this one) also need to set * the physical buffer address in the NIC ring. * netmap_idx_n2k() maps a nic index, i, into the corresponding * netmap slot index, si */ int si = netmap_idx_n2k(na->tx_rings[txq->ift_id], i); netmap_load_map(na, txq->ift_buf_tag, txq->ift_sds.ifsd_map[i], NMB(na, slot + si)); } return (1); } static int iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq) { struct netmap_adapter *na = NA(ctx->ifc_ifp); struct netmap_kring *kring; struct netmap_slot *slot; slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0); if (slot == NULL) return (0); kring = na->rx_rings[rxq->ifr_id]; netmap_fl_refill(rxq, kring, true); return (1); } static void iflib_netmap_timer_adjust(if_ctx_t ctx, iflib_txq_t txq, uint32_t *reset_on) { struct netmap_kring *kring; uint16_t txqid; txqid = txq->ift_id; kring = netmap_kring_on(NA(ctx->ifc_ifp), txqid, NR_TX); if (kring == NULL) return; if (kring->nr_hwcur != nm_next(kring->nr_hwtail, kring->nkr_num_slots - 1)) { bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD); if (ctx->isc_txd_credits_update(ctx->ifc_softc, txqid, false)) netmap_tx_irq(ctx->ifc_ifp, txqid); if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ)) { if (hz < 2000) *reset_on = 1; else *reset_on = hz / 1000; } } } #define iflib_netmap_detach(ifp) netmap_detach(ifp) #else #define iflib_netmap_txq_init(ctx, txq) (0) #define iflib_netmap_rxq_init(ctx, rxq) (0) #define iflib_netmap_detach(ifp) #define iflib_netmap_attach(ctx) (0) #define netmap_rx_irq(ifp, qid, budget) (0) #define netmap_tx_irq(ifp, qid) do {} while (0) #define iflib_netmap_timer_adjust(ctx, txq, reset_on) #endif #if defined(__i386__) || defined(__amd64__) static __inline void prefetch(void *x) { __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); } static __inline void prefetch2cachelines(void *x) { __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); #if (CACHE_LINE_SIZE < 128) __asm volatile("prefetcht0 %0" :: "m" (*(((unsigned long *)x)+CACHE_LINE_SIZE/(sizeof(unsigned long))))); #endif } #else #define prefetch(x) #define prefetch2cachelines(x) #endif static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid) { iflib_fl_t fl; fl = &rxq->ifr_fl[flid]; iru->iru_paddrs = fl->ifl_bus_addrs; iru->iru_idxs = fl->ifl_rxd_idxs; iru->iru_qsidx = rxq->ifr_id; iru->iru_buf_size = fl->ifl_buf_size; iru->iru_flidx = fl->ifl_id; } static void _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err) { if (err) return; *(bus_addr_t *) arg = segs[0].ds_addr; } int iflib_dma_alloc_align(if_ctx_t ctx, int size, int align, iflib_dma_info_t dma, int mapflags) { int err; device_t dev = ctx->ifc_dev; err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ align, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ size, /* maxsize */ 1, /* nsegments */ size, /* maxsegsize */ BUS_DMA_ALLOCNOW, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &dma->idi_tag); if (err) { device_printf(dev, "%s: bus_dma_tag_create failed: %d\n", __func__, err); goto fail_0; } err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr, BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map); if (err) { device_printf(dev, "%s: bus_dmamem_alloc(%ju) failed: %d\n", __func__, (uintmax_t)size, err); goto fail_1; } dma->idi_paddr = IF_BAD_DMA; err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr, size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT); if (err || dma->idi_paddr == IF_BAD_DMA) { device_printf(dev, "%s: bus_dmamap_load failed: %d\n", __func__, err); goto fail_2; } dma->idi_size = size; return (0); fail_2: bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); fail_1: bus_dma_tag_destroy(dma->idi_tag); fail_0: dma->idi_tag = NULL; return (err); } int iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags) { if_shared_ctx_t sctx = ctx->ifc_sctx; KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized")); return (iflib_dma_alloc_align(ctx, size, sctx->isc_q_align, dma, mapflags)); } int iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count) { int i, err; iflib_dma_info_t *dmaiter; dmaiter = dmalist; for (i = 0; i < count; i++, dmaiter++) { if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0) break; } if (err) iflib_dma_free_multi(dmalist, i); return (err); } void iflib_dma_free(iflib_dma_info_t dma) { if (dma->idi_tag == NULL) return; if (dma->idi_paddr != IF_BAD_DMA) { bus_dmamap_sync(dma->idi_tag, dma->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(dma->idi_tag, dma->idi_map); dma->idi_paddr = IF_BAD_DMA; } if (dma->idi_vaddr != NULL) { bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); dma->idi_vaddr = NULL; } bus_dma_tag_destroy(dma->idi_tag); dma->idi_tag = NULL; } void iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count) { int i; iflib_dma_info_t *dmaiter = dmalist; for (i = 0; i < count; i++, dmaiter++) iflib_dma_free(*dmaiter); } static int iflib_fast_intr(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; int result; DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL) { result = info->ifi_filter(info->ifi_filter_arg); if ((result & FILTER_SCHEDULE_THREAD) == 0) return (result); } GROUPTASK_ENQUEUE(gtask); return (FILTER_HANDLED); } static int iflib_fast_intr_rxtx(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; if_ctx_t ctx; iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx; iflib_txq_t txq; void *sc; int i, cidx, result; qidx_t txqid; bool intr_enable, intr_legacy; DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL) { result = info->ifi_filter(info->ifi_filter_arg); if ((result & FILTER_SCHEDULE_THREAD) == 0) return (result); } ctx = rxq->ifr_ctx; sc = ctx->ifc_softc; intr_enable = false; intr_legacy = !!(ctx->ifc_flags & IFC_LEGACY); MPASS(rxq->ifr_ntxqirq); for (i = 0; i < rxq->ifr_ntxqirq; i++) { txqid = rxq->ifr_txqid[i]; txq = &ctx->ifc_txqs[txqid]; bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD); if (!ctx->isc_txd_credits_update(sc, txqid, false)) { if (intr_legacy) intr_enable = true; else IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid); continue; } GROUPTASK_ENQUEUE(&txq->ift_task); } if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ) cidx = rxq->ifr_cq_cidx; else cidx = rxq->ifr_fl[0].ifl_cidx; if (iflib_rxd_avail(ctx, rxq, cidx, 1)) GROUPTASK_ENQUEUE(gtask); else { if (intr_legacy) intr_enable = true; else IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); DBG_COUNTER_INC(rx_intr_enables); } if (intr_enable) IFDI_INTR_ENABLE(ctx); return (FILTER_HANDLED); } static int iflib_fast_intr_ctx(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; int result; DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL) { result = info->ifi_filter(info->ifi_filter_arg); if ((result & FILTER_SCHEDULE_THREAD) == 0) return (result); } GROUPTASK_ENQUEUE(gtask); return (FILTER_HANDLED); } static int _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, driver_filter_t filter, driver_intr_t handler, void *arg, const char *name) { struct resource *res; void *tag = NULL; device_t dev = ctx->ifc_dev; int flags, i, rc; flags = RF_ACTIVE; if (ctx->ifc_flags & IFC_LEGACY) flags |= RF_SHAREABLE; MPASS(rid < 512); i = rid; res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &i, flags); if (res == NULL) { device_printf(dev, "failed to allocate IRQ for rid %d, name %s.\n", rid, name); return (ENOMEM); } irq->ii_res = res; KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL")); rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET, filter, handler, arg, &tag); if (rc != 0) { device_printf(dev, "failed to setup interrupt for rid %d, name %s: %d\n", rid, name ? name : "unknown", rc); return (rc); } else if (name) bus_describe_intr(dev, res, tag, "%s", name); irq->ii_tag = tag; return (0); } /********************************************************************* * * Allocate DMA resources for TX buffers as well as memory for the TX * mbuf map. TX DMA maps (non-TSO/TSO) and TX mbuf map are kept in a * iflib_sw_tx_desc_array structure, storing all the information that * is needed to transmit a packet on the wire. This is called only * once at attach, setup is done every reset. * **********************************************************************/ static int iflib_txsd_alloc(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; bus_size_t tsomaxsize; int err, nsegments, ntsosegments; bool tso; nsegments = scctx->isc_tx_nsegments; ntsosegments = scctx->isc_tx_tso_segments_max; tsomaxsize = scctx->isc_tx_tso_size_max; if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_VLAN_MTU) tsomaxsize += sizeof(struct ether_vlan_header); MPASS(scctx->isc_ntxd[0] > 0); MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0); MPASS(nsegments > 0); if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) { MPASS(ntsosegments > 0); MPASS(sctx->isc_tso_maxsize >= tsomaxsize); } /* * Set up DMA tags for TX buffers. */ if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ sctx->isc_tx_maxsize, /* maxsize */ nsegments, /* nsegments */ sctx->isc_tx_maxsegsize, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txq->ift_buf_tag))) { device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err); device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n", (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize); goto fail; } tso = (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) != 0; if (tso && (err = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ tsomaxsize, /* maxsize */ ntsosegments, /* nsegments */ sctx->isc_tso_maxsegsize,/* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txq->ift_tso_buf_tag))) { device_printf(dev, "Unable to allocate TSO TX DMA tag: %d\n", err); goto fail; } /* Allocate memory for the TX mbuf map. */ if (!(txq->ift_sds.ifsd_m = (struct mbuf **) malloc(sizeof(struct mbuf *) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate TX mbuf map memory\n"); err = ENOMEM; goto fail; } /* * Create the DMA maps for TX buffers. */ if ((txq->ift_sds.ifsd_map = (bus_dmamap_t *)malloc( sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { device_printf(dev, "Unable to allocate TX buffer DMA map memory\n"); err = ENOMEM; goto fail; } if (tso && (txq->ift_sds.ifsd_tso_map = (bus_dmamap_t *)malloc( sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { device_printf(dev, "Unable to allocate TSO TX buffer map memory\n"); err = ENOMEM; goto fail; } for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) { err = bus_dmamap_create(txq->ift_buf_tag, 0, &txq->ift_sds.ifsd_map[i]); if (err != 0) { device_printf(dev, "Unable to create TX DMA map\n"); goto fail; } if (!tso) continue; err = bus_dmamap_create(txq->ift_tso_buf_tag, 0, &txq->ift_sds.ifsd_tso_map[i]); if (err != 0) { device_printf(dev, "Unable to create TSO TX DMA map\n"); goto fail; } } return (0); fail: /* We free all, it handles case where we are in the middle */ iflib_tx_structures_free(ctx); return (err); } static void iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i) { bus_dmamap_t map; if (txq->ift_sds.ifsd_map != NULL) { map = txq->ift_sds.ifsd_map[i]; bus_dmamap_sync(txq->ift_buf_tag, map, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_buf_tag, map); bus_dmamap_destroy(txq->ift_buf_tag, map); txq->ift_sds.ifsd_map[i] = NULL; } if (txq->ift_sds.ifsd_tso_map != NULL) { map = txq->ift_sds.ifsd_tso_map[i]; bus_dmamap_sync(txq->ift_tso_buf_tag, map, BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_tso_buf_tag, map); bus_dmamap_destroy(txq->ift_tso_buf_tag, map); txq->ift_sds.ifsd_tso_map[i] = NULL; } } static void iflib_txq_destroy(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; for (int i = 0; i < txq->ift_size; i++) iflib_txsd_destroy(ctx, txq, i); if (txq->ift_br != NULL) { ifmp_ring_free(txq->ift_br); txq->ift_br = NULL; } mtx_destroy(&txq->ift_mtx); if (txq->ift_sds.ifsd_map != NULL) { free(txq->ift_sds.ifsd_map, M_IFLIB); txq->ift_sds.ifsd_map = NULL; } if (txq->ift_sds.ifsd_tso_map != NULL) { free(txq->ift_sds.ifsd_tso_map, M_IFLIB); txq->ift_sds.ifsd_tso_map = NULL; } if (txq->ift_sds.ifsd_m != NULL) { free(txq->ift_sds.ifsd_m, M_IFLIB); txq->ift_sds.ifsd_m = NULL; } if (txq->ift_buf_tag != NULL) { bus_dma_tag_destroy(txq->ift_buf_tag); txq->ift_buf_tag = NULL; } if (txq->ift_tso_buf_tag != NULL) { bus_dma_tag_destroy(txq->ift_tso_buf_tag); txq->ift_tso_buf_tag = NULL; } if (txq->ift_ifdi != NULL) { free(txq->ift_ifdi, M_IFLIB); } } static void iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i) { struct mbuf **mp; mp = &txq->ift_sds.ifsd_m[i]; if (*mp == NULL) return; if (txq->ift_sds.ifsd_map != NULL) { bus_dmamap_sync(txq->ift_buf_tag, txq->ift_sds.ifsd_map[i], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[i]); } if (txq->ift_sds.ifsd_tso_map != NULL) { bus_dmamap_sync(txq->ift_tso_buf_tag, txq->ift_sds.ifsd_tso_map[i], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_tso_buf_tag, txq->ift_sds.ifsd_tso_map[i]); } m_free(*mp); DBG_COUNTER_INC(tx_frees); *mp = NULL; } static int iflib_txq_setup(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; iflib_dma_info_t di; int i; /* Set number of descriptors available */ txq->ift_qstatus = IFLIB_QUEUE_IDLE; /* XXX make configurable */ txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ; /* Reset indices */ txq->ift_cidx_processed = 0; txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0; txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset]; for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++) bzero((void *)di->idi_vaddr, di->idi_size); IFDI_TXQ_SETUP(ctx, txq->ift_id); for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++) bus_dmamap_sync(di->idi_tag, di->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } /********************************************************************* * * Allocate DMA resources for RX buffers as well as memory for the RX * mbuf map, direct RX cluster pointer map and RX cluster bus address * map. RX DMA map, RX mbuf map, direct RX cluster pointer map and * RX cluster map are kept in a iflib_sw_rx_desc_array structure. * Since we use use one entry in iflib_sw_rx_desc_array per received * packet, the maximum number of entries we'll need is equal to the * number of hardware receive descriptors that we've allocated. * **********************************************************************/ static int iflib_rxsd_alloc(iflib_rxq_t rxq) { if_ctx_t ctx = rxq->ifr_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; iflib_fl_t fl; int err; MPASS(scctx->isc_nrxd[0] > 0); MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0); fl = rxq->ifr_fl; for (int i = 0; i < rxq->ifr_nfl; i++, fl++) { fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */ /* Set up DMA tag for RX buffers. */ err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ sctx->isc_rx_maxsize, /* maxsize */ sctx->isc_rx_nsegments, /* nsegments */ sctx->isc_rx_maxsegsize, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &fl->ifl_buf_tag); if (err) { device_printf(dev, "Unable to allocate RX DMA tag: %d\n", err); goto fail; } /* Allocate memory for the RX mbuf map. */ if (!(fl->ifl_sds.ifsd_m = (struct mbuf **) malloc(sizeof(struct mbuf *) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX mbuf map memory\n"); err = ENOMEM; goto fail; } /* Allocate memory for the direct RX cluster pointer map. */ if (!(fl->ifl_sds.ifsd_cl = (caddr_t *) malloc(sizeof(caddr_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX cluster map memory\n"); err = ENOMEM; goto fail; } /* Allocate memory for the RX cluster bus address map. */ if (!(fl->ifl_sds.ifsd_ba = (bus_addr_t *) malloc(sizeof(bus_addr_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX bus address map memory\n"); err = ENOMEM; goto fail; } /* * Create the DMA maps for RX buffers. */ if (!(fl->ifl_sds.ifsd_map = (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX buffer DMA map memory\n"); err = ENOMEM; goto fail; } for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) { err = bus_dmamap_create(fl->ifl_buf_tag, 0, &fl->ifl_sds.ifsd_map[i]); if (err != 0) { device_printf(dev, "Unable to create RX buffer DMA map\n"); goto fail; } } } return (0); fail: iflib_rx_structures_free(ctx); return (err); } /* * Internal service routines */ struct rxq_refill_cb_arg { int error; bus_dma_segment_t seg; int nseg; }; static void _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) { struct rxq_refill_cb_arg *cb_arg = arg; cb_arg->error = error; cb_arg->seg = segs[0]; cb_arg->nseg = nseg; } /** * iflib_fl_refill - refill an rxq free-buffer list * @ctx: the iflib context * @fl: the free list to refill * @count: the number of new buffers to allocate * * (Re)populate an rxq free-buffer list with up to @count new packet buffers. - * The caller must assure that @count does not exceed the queue's capacity. + * The caller must assure that @count does not exceed the queue's capacity + * minus one (since we always leave a descriptor unavailable). */ static uint8_t iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count) { struct if_rxd_update iru; struct rxq_refill_cb_arg cb_arg; struct mbuf *m; caddr_t cl, *sd_cl; struct mbuf **sd_m; bus_dmamap_t *sd_map; bus_addr_t bus_addr, *sd_ba; int err, frag_idx, i, idx, n, pidx; qidx_t credits; + MPASS(count <= fl->ifl_size - fl->ifl_credits - 1); + sd_m = fl->ifl_sds.ifsd_m; sd_map = fl->ifl_sds.ifsd_map; sd_cl = fl->ifl_sds.ifsd_cl; sd_ba = fl->ifl_sds.ifsd_ba; pidx = fl->ifl_pidx; idx = pidx; frag_idx = fl->ifl_fragidx; credits = fl->ifl_credits; i = 0; n = count; MPASS(n > 0); MPASS(credits + n <= fl->ifl_size); if (pidx < fl->ifl_cidx) MPASS(pidx + n <= fl->ifl_cidx); if (pidx == fl->ifl_cidx && (credits < fl->ifl_size)) MPASS(fl->ifl_gen == 0); if (pidx > fl->ifl_cidx) MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx); DBG_COUNTER_INC(fl_refills); if (n > 8) DBG_COUNTER_INC(fl_refills_large); iru_init(&iru, fl->ifl_rxq, fl->ifl_id); while (n-- > 0) { /* * We allocate an uninitialized mbuf + cluster, mbuf is * initialized after rx. * * If the cluster is still set then we know a minimum sized * packet was received */ bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size, &frag_idx); if (frag_idx < 0) bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx); MPASS(frag_idx >= 0); if ((cl = sd_cl[frag_idx]) == NULL) { cl = uma_zalloc(fl->ifl_zone, M_NOWAIT); if (__predict_false(cl == NULL)) break; cb_arg.error = 0; MPASS(sd_map != NULL); err = bus_dmamap_load(fl->ifl_buf_tag, sd_map[frag_idx], cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, BUS_DMA_NOWAIT); if (__predict_false(err != 0 || cb_arg.error)) { uma_zfree(fl->ifl_zone, cl); break; } sd_ba[frag_idx] = bus_addr = cb_arg.seg.ds_addr; sd_cl[frag_idx] = cl; #if MEMORY_LOGGING fl->ifl_cl_enqueued++; #endif } else { bus_addr = sd_ba[frag_idx]; } bus_dmamap_sync(fl->ifl_buf_tag, sd_map[frag_idx], BUS_DMASYNC_PREREAD); if (sd_m[frag_idx] == NULL) { m = m_gethdr(M_NOWAIT, MT_NOINIT); if (__predict_false(m == NULL)) break; sd_m[frag_idx] = m; } bit_set(fl->ifl_rx_bitmap, frag_idx); #if MEMORY_LOGGING fl->ifl_m_enqueued++; #endif DBG_COUNTER_INC(rx_allocs); fl->ifl_rxd_idxs[i] = frag_idx; fl->ifl_bus_addrs[i] = bus_addr; credits++; i++; MPASS(credits <= fl->ifl_size); if (++idx == fl->ifl_size) { #ifdef INVARIANTS fl->ifl_gen = 1; #endif idx = 0; } if (n == 0 || i == IFLIB_MAX_RX_REFRESH) { iru.iru_pidx = pidx; iru.iru_count = i; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); fl->ifl_pidx = idx; fl->ifl_credits = credits; pidx = idx; i = 0; } } if (n < count - 1) { if (i != 0) { iru.iru_pidx = pidx; iru.iru_count = i; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); fl->ifl_pidx = idx; fl->ifl_credits = credits; } DBG_COUNTER_INC(rxd_flush); - if (fl->ifl_pidx == 0) - pidx = fl->ifl_size - 1; - else - pidx = fl->ifl_pidx - 1; - bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, - fl->ifl_id, pidx); + fl->ifl_id, fl->ifl_pidx); if (__predict_true(bit_test(fl->ifl_rx_bitmap, frag_idx))) { fl->ifl_fragidx = frag_idx + 1; if (fl->ifl_fragidx == fl->ifl_size) fl->ifl_fragidx = 0; } else { fl->ifl_fragidx = frag_idx; } } return (n == -1 ? 0 : IFLIB_RXEOF_EMPTY); } static inline uint8_t iflib_fl_refill_all(if_ctx_t ctx, iflib_fl_t fl) { - /* we avoid allowing pidx to catch up with cidx as it confuses ixl */ + /* + * We leave an unused descriptor to avoid pidx to catch up with cidx. + * This is important as it confuses most NICs. For instance, + * Intel NICs have (per receive ring) RDH and RDT registers, where + * RDH points to the next receive descriptor to be used by the NIC, + * and RDT for the next receive descriptor to be published by the + * driver to the NIC (RDT - 1 is thus the last valid one). + * The condition RDH == RDT means no descriptors are available to + * the NIC, and thus it would be ambiguous if it also meant that + * all the descriptors are available to the NIC. + */ int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1; #ifdef INVARIANTS int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1; #endif MPASS(fl->ifl_credits <= fl->ifl_size); MPASS(reclaimable == delta); if (reclaimable > 0) return (iflib_fl_refill(ctx, fl, reclaimable)); return (0); } uint8_t iflib_in_detach(if_ctx_t ctx) { bool in_detach; STATE_LOCK(ctx); in_detach = !!(ctx->ifc_flags & IFC_IN_DETACH); STATE_UNLOCK(ctx); return (in_detach); } static void iflib_fl_bufs_free(iflib_fl_t fl) { iflib_dma_info_t idi = fl->ifl_ifdi; bus_dmamap_t sd_map; uint32_t i; for (i = 0; i < fl->ifl_size; i++) { struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i]; caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i]; if (*sd_cl != NULL) { sd_map = fl->ifl_sds.ifsd_map[i]; bus_dmamap_sync(fl->ifl_buf_tag, sd_map, BUS_DMASYNC_POSTREAD); bus_dmamap_unload(fl->ifl_buf_tag, sd_map); uma_zfree(fl->ifl_zone, *sd_cl); *sd_cl = NULL; if (*sd_m != NULL) { m_init(*sd_m, M_NOWAIT, MT_DATA, 0); uma_zfree(zone_mbuf, *sd_m); *sd_m = NULL; } } else { MPASS(*sd_m == NULL); } #if MEMORY_LOGGING fl->ifl_m_dequeued++; fl->ifl_cl_dequeued++; #endif } #ifdef INVARIANTS for (i = 0; i < fl->ifl_size; i++) { MPASS(fl->ifl_sds.ifsd_cl[i] == NULL); MPASS(fl->ifl_sds.ifsd_m[i] == NULL); } #endif /* * Reset free list values */ fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0; bzero(idi->idi_vaddr, idi->idi_size); } /********************************************************************* * * Initialize a free list and its buffers. * **********************************************************************/ static int iflib_fl_setup(iflib_fl_t fl) { iflib_rxq_t rxq = fl->ifl_rxq; if_ctx_t ctx = rxq->ifr_ctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; int qidx; bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size - 1); /* ** Free current RX buffer structs and their mbufs */ iflib_fl_bufs_free(fl); /* Now replenish the mbufs */ MPASS(fl->ifl_credits == 0); qidx = rxq->ifr_fl_offset + fl->ifl_id; if (scctx->isc_rxd_buf_size[qidx] != 0) fl->ifl_buf_size = scctx->isc_rxd_buf_size[qidx]; else fl->ifl_buf_size = ctx->ifc_rx_mbuf_sz; /* * ifl_buf_size may be a driver-supplied value, so pull it up * to the selected mbuf size. */ fl->ifl_buf_size = iflib_get_mbuf_size_for(fl->ifl_buf_size); if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size) ctx->ifc_max_fl_buf_size = fl->ifl_buf_size; fl->ifl_cltype = m_gettype(fl->ifl_buf_size); fl->ifl_zone = m_getzone(fl->ifl_buf_size); - /* avoid pre-allocating zillions of clusters to an idle card - * potentially speeding up attach + /* + * Avoid pre-allocating zillions of clusters to an idle card + * potentially speeding up attach. In any case make sure + * to leave a descriptor unavailable. See the comment in + * iflib_fl_refill_all(). */ - (void)iflib_fl_refill(ctx, fl, min(128, fl->ifl_size)); - MPASS(min(128, fl->ifl_size) == fl->ifl_credits); - if (min(128, fl->ifl_size) != fl->ifl_credits) + MPASS(fl->ifl_size > 0); + (void)iflib_fl_refill(ctx, fl, min(128, fl->ifl_size - 1)); + if (min(128, fl->ifl_size - 1) != fl->ifl_credits) return (ENOBUFS); /* * handle failure */ MPASS(rxq != NULL); MPASS(fl->ifl_ifdi != NULL); bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } /********************************************************************* * * Free receive ring data structures * **********************************************************************/ static void iflib_rx_sds_free(iflib_rxq_t rxq) { iflib_fl_t fl; int i, j; if (rxq->ifr_fl != NULL) { for (i = 0; i < rxq->ifr_nfl; i++) { fl = &rxq->ifr_fl[i]; if (fl->ifl_buf_tag != NULL) { if (fl->ifl_sds.ifsd_map != NULL) { for (j = 0; j < fl->ifl_size; j++) { bus_dmamap_sync( fl->ifl_buf_tag, fl->ifl_sds.ifsd_map[j], BUS_DMASYNC_POSTREAD); bus_dmamap_unload( fl->ifl_buf_tag, fl->ifl_sds.ifsd_map[j]); bus_dmamap_destroy( fl->ifl_buf_tag, fl->ifl_sds.ifsd_map[j]); } } bus_dma_tag_destroy(fl->ifl_buf_tag); fl->ifl_buf_tag = NULL; } free(fl->ifl_sds.ifsd_m, M_IFLIB); free(fl->ifl_sds.ifsd_cl, M_IFLIB); free(fl->ifl_sds.ifsd_ba, M_IFLIB); free(fl->ifl_sds.ifsd_map, M_IFLIB); fl->ifl_sds.ifsd_m = NULL; fl->ifl_sds.ifsd_cl = NULL; fl->ifl_sds.ifsd_ba = NULL; fl->ifl_sds.ifsd_map = NULL; } free(rxq->ifr_fl, M_IFLIB); rxq->ifr_fl = NULL; free(rxq->ifr_ifdi, M_IFLIB); rxq->ifr_ifdi = NULL; rxq->ifr_cq_cidx = 0; } } /* * Timer routine */ static void iflib_timer(void *arg) { iflib_txq_t txq = arg; if_ctx_t ctx = txq->ift_ctx; if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; uint64_t this_tick = ticks; uint32_t reset_on = hz / 2; if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) return; /* ** Check on the state of the TX queue(s), this ** can be done without the lock because its RO ** and the HUNG state will be static if set. */ if (this_tick - txq->ift_last_timer_tick >= hz / 2) { txq->ift_last_timer_tick = this_tick; IFDI_TIMER(ctx, txq->ift_id); if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) && ((txq->ift_cleaned_prev == txq->ift_cleaned) || (sctx->isc_pause_frames == 0))) goto hung; if (txq->ift_qstatus != IFLIB_QUEUE_IDLE && ifmp_ring_is_stalled(txq->ift_br)) { KASSERT(ctx->ifc_link_state == LINK_STATE_UP, ("queue can't be marked as hung if interface is down")); txq->ift_qstatus = IFLIB_QUEUE_HUNG; } txq->ift_cleaned_prev = txq->ift_cleaned; } #ifdef DEV_NETMAP if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) iflib_netmap_timer_adjust(ctx, txq, &reset_on); #endif /* handle any laggards */ if (txq->ift_db_pending) GROUPTASK_ENQUEUE(&txq->ift_task); sctx->isc_pause_frames = 0; if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) callout_reset_on(&txq->ift_timer, reset_on, iflib_timer, txq, txq->ift_timer.c_cpu); return; hung: device_printf(ctx->ifc_dev, "Watchdog timeout (TX: %d desc avail: %d pidx: %d) -- resetting\n", txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx); STATE_LOCK(ctx); if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); ctx->ifc_flags |= (IFC_DO_WATCHDOG|IFC_DO_RESET); iflib_admin_intr_deferred(ctx); STATE_UNLOCK(ctx); } static uint16_t iflib_get_mbuf_size_for(unsigned int size) { if (size <= MCLBYTES) return (MCLBYTES); else return (MJUMPAGESIZE); } static void iflib_calc_rx_mbuf_sz(if_ctx_t ctx) { if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; /* * XXX don't set the max_frame_size to larger * than the hardware can handle */ ctx->ifc_rx_mbuf_sz = iflib_get_mbuf_size_for(sctx->isc_max_frame_size); } uint32_t iflib_get_rx_mbuf_sz(if_ctx_t ctx) { return (ctx->ifc_rx_mbuf_sz); } static void iflib_init_locked(if_ctx_t ctx) { if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_t ifp = ctx->ifc_ifp; iflib_fl_t fl; iflib_txq_t txq; iflib_rxq_t rxq; int i, j, tx_ip_csum_flags, tx_ip6_csum_flags; if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); IFDI_INTR_DISABLE(ctx); tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP); tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP); /* Set hardware offload abilities */ if_clearhwassist(ifp); if (if_getcapenable(ifp) & IFCAP_TXCSUM) if_sethwassistbits(ifp, tx_ip_csum_flags, 0); if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) if_sethwassistbits(ifp, tx_ip6_csum_flags, 0); if (if_getcapenable(ifp) & IFCAP_TSO4) if_sethwassistbits(ifp, CSUM_IP_TSO, 0); if (if_getcapenable(ifp) & IFCAP_TSO6) if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) { CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); iflib_netmap_txq_init(ctx, txq); } /* * Calculate a suitable Rx mbuf size prior to calling IFDI_INIT, so * that drivers can use the value when setting up the hardware receive * buffers. */ iflib_calc_rx_mbuf_sz(ctx); #ifdef INVARIANTS i = if_getdrvflags(ifp); #endif IFDI_INIT(ctx); MPASS(if_getdrvflags(ifp) == i); for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) { if (iflib_netmap_rxq_init(ctx, rxq) > 0) { /* This rxq is in netmap mode. Skip normal init. */ continue; } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { if (iflib_fl_setup(fl)) { device_printf(ctx->ifc_dev, "setting up free list %d failed - " "check cluster settings\n", j); goto done; } } } done: if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE); IFDI_INTR_ENABLE(ctx); txq = ctx->ifc_txqs; for (i = 0; i < sctx->isc_ntxqsets; i++, txq++) callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); } static int iflib_media_change(if_t ifp) { if_ctx_t ctx = if_getsoftc(ifp); int err; CTX_LOCK(ctx); if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0) iflib_init_locked(ctx); CTX_UNLOCK(ctx); return (err); } static void iflib_media_status(if_t ifp, struct ifmediareq *ifmr) { if_ctx_t ctx = if_getsoftc(ifp); CTX_LOCK(ctx); IFDI_UPDATE_ADMIN_STATUS(ctx); IFDI_MEDIA_STATUS(ctx, ifmr); CTX_UNLOCK(ctx); } void iflib_stop(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; iflib_rxq_t rxq = ctx->ifc_rxqs; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; iflib_dma_info_t di; iflib_fl_t fl; int i, j; /* Tell the stack that the interface is no longer active */ if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); IFDI_INTR_DISABLE(ctx); DELAY(1000); IFDI_STOP(ctx); DELAY(1000); iflib_debug_reset(); /* Wait for current tx queue users to exit to disarm watchdog timer. */ for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) { /* make sure all transmitters have completed before proceeding XXX */ CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); /* clean any enqueued buffers */ iflib_ifmp_purge(txq); /* Free any existing tx buffers. */ for (j = 0; j < txq->ift_size; j++) { iflib_txsd_free(ctx, txq, j); } txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0; txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0; txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0; txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0; txq->ift_pullups = 0; ifmp_ring_reset_stats(txq->ift_br); for (j = 0, di = txq->ift_ifdi; j < sctx->isc_ntxqs; j++, di++) bzero((void *)di->idi_vaddr, di->idi_size); } for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) { /* make sure all transmitters have completed before proceeding XXX */ rxq->ifr_cq_cidx = 0; for (j = 0, di = rxq->ifr_ifdi; j < sctx->isc_nrxqs; j++, di++) bzero((void *)di->idi_vaddr, di->idi_size); /* also resets the free lists pidx/cidx */ for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) iflib_fl_bufs_free(fl); } } static inline caddr_t calc_next_rxd(iflib_fl_t fl, int cidx) { qidx_t size; int nrxd; caddr_t start, end, cur, next; nrxd = fl->ifl_size; size = fl->ifl_rxd_size; start = fl->ifl_ifdi->idi_vaddr; if (__predict_false(size == 0)) return (start); cur = start + size*cidx; end = start + size*nrxd; next = CACHE_PTR_NEXT(cur); return (next < end ? next : start); } static inline void prefetch_pkts(iflib_fl_t fl, int cidx) { int nextptr; int nrxd = fl->ifl_size; caddr_t next_rxd; nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1); prefetch(&fl->ifl_sds.ifsd_m[nextptr]); prefetch(&fl->ifl_sds.ifsd_cl[nextptr]); next_rxd = calc_next_rxd(fl, cidx); prefetch(next_rxd); prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]); } static struct mbuf * rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, bool unload, if_rxsd_t sd, int *pf_rv, if_rxd_info_t ri) { bus_dmamap_t map; iflib_fl_t fl; caddr_t payload; struct mbuf *m; int flid, cidx, len, next; map = NULL; flid = irf->irf_flid; cidx = irf->irf_idx; fl = &rxq->ifr_fl[flid]; sd->ifsd_fl = fl; m = fl->ifl_sds.ifsd_m[cidx]; sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx]; fl->ifl_credits--; #if MEMORY_LOGGING fl->ifl_m_dequeued++; #endif if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH) prefetch_pkts(fl, cidx); next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1); prefetch(&fl->ifl_sds.ifsd_map[next]); map = fl->ifl_sds.ifsd_map[cidx]; bus_dmamap_sync(fl->ifl_buf_tag, map, BUS_DMASYNC_POSTREAD); if (rxq->pfil != NULL && PFIL_HOOKED_IN(rxq->pfil) && pf_rv != NULL && irf->irf_len != 0) { payload = *sd->ifsd_cl; payload += ri->iri_pad; len = ri->iri_len - ri->iri_pad; *pf_rv = pfil_run_hooks(rxq->pfil, payload, ri->iri_ifp, len | PFIL_MEMPTR | PFIL_IN, NULL); switch (*pf_rv) { case PFIL_DROPPED: case PFIL_CONSUMED: /* * The filter ate it. Everything is recycled. */ m = NULL; unload = 0; break; case PFIL_REALLOCED: /* * The filter copied it. Everything is recycled. */ m = pfil_mem2mbuf(payload); unload = 0; break; case PFIL_PASS: /* * Filter said it was OK, so receive like * normal */ fl->ifl_sds.ifsd_m[cidx] = NULL; break; default: MPASS(0); } } else { fl->ifl_sds.ifsd_m[cidx] = NULL; *pf_rv = PFIL_PASS; } if (unload && irf->irf_len != 0) bus_dmamap_unload(fl->ifl_buf_tag, map); fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1); if (__predict_false(fl->ifl_cidx == 0)) fl->ifl_gen = 0; bit_clear(fl->ifl_rx_bitmap, cidx); return (m); } static struct mbuf * assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd, int *pf_rv) { struct mbuf *m, *mh, *mt; caddr_t cl; int *pf_rv_ptr, flags, i, padlen; bool consumed; i = 0; mh = NULL; consumed = false; *pf_rv = PFIL_PASS; pf_rv_ptr = pf_rv; do { m = rxd_frag_to_sd(rxq, &ri->iri_frags[i], !consumed, sd, pf_rv_ptr, ri); MPASS(*sd->ifsd_cl != NULL); /* * Exclude zero-length frags & frags from * packets the filter has consumed or dropped */ if (ri->iri_frags[i].irf_len == 0 || consumed || *pf_rv == PFIL_CONSUMED || *pf_rv == PFIL_DROPPED) { if (mh == NULL) { /* everything saved here */ consumed = true; pf_rv_ptr = NULL; continue; } /* XXX we can save the cluster here, but not the mbuf */ m_init(m, M_NOWAIT, MT_DATA, 0); m_free(m); continue; } if (mh == NULL) { flags = M_PKTHDR|M_EXT; mh = mt = m; padlen = ri->iri_pad; } else { flags = M_EXT; mt->m_next = m; mt = m; /* assuming padding is only on the first fragment */ padlen = 0; } cl = *sd->ifsd_cl; *sd->ifsd_cl = NULL; /* Can these two be made one ? */ m_init(m, M_NOWAIT, MT_DATA, flags); m_cljset(m, cl, sd->ifsd_fl->ifl_cltype); /* * These must follow m_init and m_cljset */ m->m_data += padlen; ri->iri_len -= padlen; m->m_len = ri->iri_frags[i].irf_len; } while (++i < ri->iri_nfrags); return (mh); } /* * Process one software descriptor */ static struct mbuf * iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri) { struct if_rxsd sd; struct mbuf *m; int pf_rv; /* should I merge this back in now that the two paths are basically duplicated? */ if (ri->iri_nfrags == 1 && ri->iri_frags[0].irf_len != 0 && ri->iri_frags[0].irf_len <= MIN(IFLIB_RX_COPY_THRESH, MHLEN)) { m = rxd_frag_to_sd(rxq, &ri->iri_frags[0], false, &sd, &pf_rv, ri); if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED) return (m); if (pf_rv == PFIL_PASS) { m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR); #ifndef __NO_STRICT_ALIGNMENT if (!IP_ALIGNED(m)) m->m_data += 2; #endif memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len); m->m_len = ri->iri_frags[0].irf_len; } } else { m = assemble_segments(rxq, ri, &sd, &pf_rv); if (m == NULL) return (NULL); if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED) return (m); } m->m_pkthdr.len = ri->iri_len; m->m_pkthdr.rcvif = ri->iri_ifp; m->m_flags |= ri->iri_flags; m->m_pkthdr.ether_vtag = ri->iri_vtag; m->m_pkthdr.flowid = ri->iri_flowid; M_HASHTYPE_SET(m, ri->iri_rsstype); m->m_pkthdr.csum_flags = ri->iri_csum_flags; m->m_pkthdr.csum_data = ri->iri_csum_data; return (m); } #if defined(INET6) || defined(INET) static void iflib_get_ip_forwarding(struct lro_ctrl *lc, bool *v4, bool *v6) { CURVNET_SET(lc->ifp->if_vnet); #if defined(INET6) *v6 = V_ip6_forwarding; #endif #if defined(INET) *v4 = V_ipforwarding; #endif CURVNET_RESTORE(); } /* * Returns true if it's possible this packet could be LROed. * if it returns false, it is guaranteed that tcp_lro_rx() * would not return zero. */ static bool iflib_check_lro_possible(struct mbuf *m, bool v4_forwarding, bool v6_forwarding) { struct ether_header *eh; eh = mtod(m, struct ether_header *); switch (eh->ether_type) { #if defined(INET6) case htons(ETHERTYPE_IPV6): return (!v6_forwarding); #endif #if defined (INET) case htons(ETHERTYPE_IP): return (!v4_forwarding); #endif } return false; } #else static void iflib_get_ip_forwarding(struct lro_ctrl *lc __unused, bool *v4 __unused, bool *v6 __unused) { } #endif static void _task_fn_rx_watchdog(void *context) { iflib_rxq_t rxq = context; GROUPTASK_ENQUEUE(&rxq->ifr_task); } static uint8_t iflib_rxeof(iflib_rxq_t rxq, qidx_t budget) { if_t ifp; if_ctx_t ctx = rxq->ifr_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; int avail, i; qidx_t *cidxp; struct if_rxd_info ri; int err, budget_left, rx_bytes, rx_pkts; iflib_fl_t fl; int lro_enabled; bool v4_forwarding, v6_forwarding, lro_possible; uint8_t retval = 0; /* * XXX early demux data packets so that if_input processing only handles * acks in interrupt context */ struct mbuf *m, *mh, *mt, *mf; NET_EPOCH_ASSERT(); lro_possible = v4_forwarding = v6_forwarding = false; ifp = ctx->ifc_ifp; mh = mt = NULL; MPASS(budget > 0); rx_pkts = rx_bytes = 0; if (sctx->isc_flags & IFLIB_HAS_RXCQ) cidxp = &rxq->ifr_cq_cidx; else cidxp = &rxq->ifr_fl[0].ifl_cidx; if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) { for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) retval |= iflib_fl_refill_all(ctx, fl); DBG_COUNTER_INC(rx_unavail); return (retval); } /* pfil needs the vnet to be set */ CURVNET_SET_QUIET(ifp->if_vnet); for (budget_left = budget; budget_left > 0 && avail > 0;) { if (__predict_false(!CTX_ACTIVE(ctx))) { DBG_COUNTER_INC(rx_ctx_inactive); break; } /* * Reset client set fields to their default values */ rxd_info_zero(&ri); ri.iri_qsidx = rxq->ifr_id; ri.iri_cidx = *cidxp; ri.iri_ifp = ifp; ri.iri_frags = rxq->ifr_frags; err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); if (err) goto err; rx_pkts += 1; rx_bytes += ri.iri_len; if (sctx->isc_flags & IFLIB_HAS_RXCQ) { *cidxp = ri.iri_cidx; /* Update our consumer index */ /* XXX NB: shurd - check if this is still safe */ while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) rxq->ifr_cq_cidx -= scctx->isc_nrxd[0]; /* was this only a completion queue message? */ if (__predict_false(ri.iri_nfrags == 0)) continue; } MPASS(ri.iri_nfrags != 0); MPASS(ri.iri_len != 0); /* will advance the cidx on the corresponding free lists */ m = iflib_rxd_pkt_get(rxq, &ri); avail--; budget_left--; if (avail == 0 && budget_left) avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left); if (__predict_false(m == NULL)) continue; /* imm_pkt: -- cxgb */ if (mh == NULL) mh = mt = m; else { mt->m_nextpkt = m; mt = m; } } CURVNET_RESTORE(); /* make sure that we can refill faster than drain */ for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) retval |= iflib_fl_refill_all(ctx, fl); lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO); if (lro_enabled) iflib_get_ip_forwarding(&rxq->ifr_lc, &v4_forwarding, &v6_forwarding); mt = mf = NULL; while (mh != NULL) { m = mh; mh = mh->m_nextpkt; m->m_nextpkt = NULL; #ifndef __NO_STRICT_ALIGNMENT if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL) continue; #endif rx_bytes += m->m_pkthdr.len; rx_pkts++; #if defined(INET6) || defined(INET) if (lro_enabled) { if (!lro_possible) { lro_possible = iflib_check_lro_possible(m, v4_forwarding, v6_forwarding); if (lro_possible && mf != NULL) { ifp->if_input(ifp, mf); DBG_COUNTER_INC(rx_if_input); mt = mf = NULL; } } if ((m->m_pkthdr.csum_flags & (CSUM_L4_CALC|CSUM_L4_VALID)) == (CSUM_L4_CALC|CSUM_L4_VALID)) { if (lro_possible && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) continue; } } #endif if (lro_possible) { ifp->if_input(ifp, m); DBG_COUNTER_INC(rx_if_input); continue; } if (mf == NULL) mf = m; if (mt != NULL) mt->m_nextpkt = m; mt = m; } if (mf != NULL) { ifp->if_input(ifp, mf); DBG_COUNTER_INC(rx_if_input); } if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes); if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts); /* * Flush any outstanding LRO work */ #if defined(INET6) || defined(INET) tcp_lro_flush_all(&rxq->ifr_lc); #endif if (avail != 0 || iflib_rxd_avail(ctx, rxq, *cidxp, 1) != 0) retval |= IFLIB_RXEOF_MORE; return (retval); err: STATE_LOCK(ctx); ctx->ifc_flags |= IFC_DO_RESET; iflib_admin_intr_deferred(ctx); STATE_UNLOCK(ctx); return (0); } #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1) static inline qidx_t txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use) { qidx_t notify_count = TXD_NOTIFY_COUNT(txq); qidx_t minthresh = txq->ift_size / 8; if (in_use > 4*minthresh) return (notify_count); if (in_use > 2*minthresh) return (notify_count >> 1); if (in_use > minthresh) return (notify_count >> 3); return (0); } static inline qidx_t txq_max_rs_deferred(iflib_txq_t txq) { qidx_t notify_count = TXD_NOTIFY_COUNT(txq); qidx_t minthresh = txq->ift_size / 8; if (txq->ift_in_use > 4*minthresh) return (notify_count); if (txq->ift_in_use > 2*minthresh) return (notify_count >> 1); if (txq->ift_in_use > minthresh) return (notify_count >> 2); return (2); } #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags) #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG) #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use)) #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq) #define TXQ_MAX_DB_CONSUMED(size) (size >> 4) /* forward compatibility for cxgb */ #define FIRST_QSET(ctx) 0 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets) #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets) #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx)) #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments)) /* XXX we should be setting this to something other than zero */ #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh) #define MAX_TX_DESC(ctx) max((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max, \ (ctx)->ifc_softc_ctx.isc_tx_nsegments) static inline bool iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring, qidx_t in_use) { qidx_t dbval, max; bool rang; rang = false; max = TXQ_MAX_DB_DEFERRED(txq, in_use); if (ring || txq->ift_db_pending >= max) { dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx; bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval); txq->ift_db_pending = txq->ift_npending = 0; rang = true; } return (rang); } #ifdef PKT_DEBUG static void print_pkt(if_pkt_info_t pi) { printf("pi len: %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n", pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx); printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n", pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag); printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n", pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto); } #endif #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO) #define IS_TX_OFFLOAD4(pi) ((pi)->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP_TSO)) #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO) #define IS_TX_OFFLOAD6(pi) ((pi)->ipi_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_TSO)) static int iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp) { if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx; struct ether_vlan_header *eh; struct mbuf *m; m = *mp; if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) && M_WRITABLE(m) == 0) { if ((m = m_dup(m, M_NOWAIT)) == NULL) { return (ENOMEM); } else { m_freem(*mp); DBG_COUNTER_INC(tx_frees); *mp = m; } } /* * Determine where frame payload starts. * Jump over vlan headers if already present, * helpful for QinQ too. */ if (__predict_false(m->m_len < sizeof(*eh))) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL)) return (ENOMEM); } eh = mtod(m, struct ether_vlan_header *); if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { pi->ipi_etype = ntohs(eh->evl_proto); pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; } else { pi->ipi_etype = ntohs(eh->evl_encap_proto); pi->ipi_ehdrlen = ETHER_HDR_LEN; } switch (pi->ipi_etype) { #ifdef INET case ETHERTYPE_IP: { struct mbuf *n; struct ip *ip = NULL; struct tcphdr *th = NULL; int minthlen; minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th)); if (__predict_false(m->m_len < minthlen)) { /* * if this code bloat is causing too much of a hit * move it to a separate function and mark it noinline */ if (m->m_len == pi->ipi_ehdrlen) { n = m->m_next; MPASS(n); if (n->m_len >= sizeof(*ip)) { ip = (struct ip *)n->m_data; if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } else { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) return (ENOMEM); ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); } } else { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) return (ENOMEM); ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } } else { ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } pi->ipi_ip_hlen = ip->ip_hl << 2; pi->ipi_ipproto = ip->ip_p; pi->ipi_flags |= IPI_TX_IPV4; /* TCP checksum offload may require TCP header length */ if (IS_TX_OFFLOAD4(pi)) { if (__predict_true(pi->ipi_ipproto == IPPROTO_TCP)) { if (__predict_false(th == NULL)) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL)) return (ENOMEM); th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen); } pi->ipi_tcp_hflags = th->th_flags; pi->ipi_tcp_hlen = th->th_off << 2; pi->ipi_tcp_seq = th->th_seq; } if (IS_TSO4(pi)) { if (__predict_false(ip->ip_p != IPPROTO_TCP)) return (ENXIO); /* * TSO always requires hardware checksum offload. */ pi->ipi_csum_flags |= (CSUM_IP_TCP | CSUM_IP); th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; if (sctx->isc_flags & IFLIB_TSO_INIT_IP) { ip->ip_sum = 0; ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz); } } } if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP)) ip->ip_sum = 0; break; } #endif #ifdef INET6 case ETHERTYPE_IPV6: { struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen); struct tcphdr *th; pi->ipi_ip_hlen = sizeof(struct ip6_hdr); if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL)) return (ENOMEM); } th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen); /* XXX-BZ this will go badly in case of ext hdrs. */ pi->ipi_ipproto = ip6->ip6_nxt; pi->ipi_flags |= IPI_TX_IPV6; /* TCP checksum offload may require TCP header length */ if (IS_TX_OFFLOAD6(pi)) { if (pi->ipi_ipproto == IPPROTO_TCP) { if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL)) return (ENOMEM); } pi->ipi_tcp_hflags = th->th_flags; pi->ipi_tcp_hlen = th->th_off << 2; pi->ipi_tcp_seq = th->th_seq; } if (IS_TSO6(pi)) { if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP)) return (ENXIO); /* * TSO always requires hardware checksum offload. */ pi->ipi_csum_flags |= CSUM_IP6_TCP; th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0); pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; } } break; } #endif default: pi->ipi_csum_flags &= ~CSUM_OFFLOAD; pi->ipi_ip_hlen = 0; break; } *mp = m; return (0); } /* * If dodgy hardware rejects the scatter gather chain we've handed it * we'll need to remove the mbuf chain from ifsg_m[] before we can add the * m_defrag'd mbufs */ static __noinline struct mbuf * iflib_remove_mbuf(iflib_txq_t txq) { int ntxd, pidx; struct mbuf *m, **ifsd_m; ifsd_m = txq->ift_sds.ifsd_m; ntxd = txq->ift_size; pidx = txq->ift_pidx & (ntxd - 1); ifsd_m = txq->ift_sds.ifsd_m; m = ifsd_m[pidx]; ifsd_m[pidx] = NULL; bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[pidx]); if (txq->ift_sds.ifsd_tso_map != NULL) bus_dmamap_unload(txq->ift_tso_buf_tag, txq->ift_sds.ifsd_tso_map[pidx]); #if MEMORY_LOGGING txq->ift_dequeued++; #endif return (m); } static inline caddr_t calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid) { qidx_t size; int ntxd; caddr_t start, end, cur, next; ntxd = txq->ift_size; size = txq->ift_txd_size[qid]; start = txq->ift_ifdi[qid].idi_vaddr; if (__predict_false(size == 0)) return (start); cur = start + size*cidx; end = start + size*ntxd; next = CACHE_PTR_NEXT(cur); return (next < end ? next : start); } /* * Pad an mbuf to ensure a minimum ethernet frame size. * min_frame_size is the frame size (less CRC) to pad the mbuf to */ static __noinline int iflib_ether_pad(device_t dev, struct mbuf **m_head, uint16_t min_frame_size) { /* * 18 is enough bytes to pad an ARP packet to 46 bytes, and * and ARP message is the smallest common payload I can think of */ static char pad[18]; /* just zeros */ int n; struct mbuf *new_head; if (!M_WRITABLE(*m_head)) { new_head = m_dup(*m_head, M_NOWAIT); if (new_head == NULL) { m_freem(*m_head); device_printf(dev, "cannot pad short frame, m_dup() failed"); DBG_COUNTER_INC(encap_pad_mbuf_fail); DBG_COUNTER_INC(tx_frees); return ENOMEM; } m_freem(*m_head); *m_head = new_head; } for (n = min_frame_size - (*m_head)->m_pkthdr.len; n > 0; n -= sizeof(pad)) if (!m_append(*m_head, min(n, sizeof(pad)), pad)) break; if (n > 0) { m_freem(*m_head); device_printf(dev, "cannot pad short frame\n"); DBG_COUNTER_INC(encap_pad_mbuf_fail); DBG_COUNTER_INC(tx_frees); return (ENOBUFS); } return 0; } static int iflib_encap(iflib_txq_t txq, struct mbuf **m_headp) { if_ctx_t ctx; if_shared_ctx_t sctx; if_softc_ctx_t scctx; bus_dma_tag_t buf_tag; bus_dma_segment_t *segs; struct mbuf *m_head, **ifsd_m; void *next_txd; bus_dmamap_t map; struct if_pkt_info pi; int remap = 0; int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd; ctx = txq->ift_ctx; sctx = ctx->ifc_sctx; scctx = &ctx->ifc_softc_ctx; segs = txq->ift_segs; ntxd = txq->ift_size; m_head = *m_headp; map = NULL; /* * If we're doing TSO the next descriptor to clean may be quite far ahead */ cidx = txq->ift_cidx; pidx = txq->ift_pidx; if (ctx->ifc_flags & IFC_PREFETCH) { next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1); if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) { next_txd = calc_next_txd(txq, cidx, 0); prefetch(next_txd); } /* prefetch the next cache line of mbuf pointers and flags */ prefetch(&txq->ift_sds.ifsd_m[next]); prefetch(&txq->ift_sds.ifsd_map[next]); next = (cidx + CACHE_LINE_SIZE) & (ntxd-1); } map = txq->ift_sds.ifsd_map[pidx]; ifsd_m = txq->ift_sds.ifsd_m; if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { buf_tag = txq->ift_tso_buf_tag; max_segs = scctx->isc_tx_tso_segments_max; map = txq->ift_sds.ifsd_tso_map[pidx]; MPASS(buf_tag != NULL); MPASS(max_segs > 0); } else { buf_tag = txq->ift_buf_tag; max_segs = scctx->isc_tx_nsegments; map = txq->ift_sds.ifsd_map[pidx]; } if ((sctx->isc_flags & IFLIB_NEED_ETHER_PAD) && __predict_false(m_head->m_pkthdr.len < scctx->isc_min_frame_size)) { err = iflib_ether_pad(ctx->ifc_dev, m_headp, scctx->isc_min_frame_size); if (err) { DBG_COUNTER_INC(encap_txd_encap_fail); return err; } } m_head = *m_headp; pkt_info_zero(&pi); pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST)); pi.ipi_pidx = pidx; pi.ipi_qsidx = txq->ift_id; pi.ipi_len = m_head->m_pkthdr.len; pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags; pi.ipi_vtag = M_HAS_VLANTAG(m_head) ? m_head->m_pkthdr.ether_vtag : 0; /* deliberate bitwise OR to make one condition */ if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) { if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0)) { DBG_COUNTER_INC(encap_txd_encap_fail); return (err); } m_head = *m_headp; } retry: err = bus_dmamap_load_mbuf_sg(buf_tag, map, m_head, segs, &nsegs, BUS_DMA_NOWAIT); defrag: if (__predict_false(err)) { switch (err) { case EFBIG: /* try collapse once and defrag once */ if (remap == 0) { m_head = m_collapse(*m_headp, M_NOWAIT, max_segs); /* try defrag if collapsing fails */ if (m_head == NULL) remap++; } if (remap == 1) { txq->ift_mbuf_defrag++; m_head = m_defrag(*m_headp, M_NOWAIT); } /* * remap should never be >1 unless bus_dmamap_load_mbuf_sg * failed to map an mbuf that was run through m_defrag */ MPASS(remap <= 1); if (__predict_false(m_head == NULL || remap > 1)) goto defrag_failed; remap++; *m_headp = m_head; goto retry; break; case ENOMEM: txq->ift_no_tx_dma_setup++; break; default: txq->ift_no_tx_dma_setup++; m_freem(*m_headp); DBG_COUNTER_INC(tx_frees); *m_headp = NULL; break; } txq->ift_map_failed++; DBG_COUNTER_INC(encap_load_mbuf_fail); DBG_COUNTER_INC(encap_txd_encap_fail); return (err); } ifsd_m[pidx] = m_head; /* * XXX assumes a 1 to 1 relationship between segments and * descriptors - this does not hold true on all drivers, e.g. * cxgb */ if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) { txq->ift_no_desc_avail++; bus_dmamap_unload(buf_tag, map); DBG_COUNTER_INC(encap_txq_avail_fail); DBG_COUNTER_INC(encap_txd_encap_fail); if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0) GROUPTASK_ENQUEUE(&txq->ift_task); return (ENOBUFS); } /* * On Intel cards we can greatly reduce the number of TX interrupts * we see by only setting report status on every Nth descriptor. * However, this also means that the driver will need to keep track * of the descriptors that RS was set on to check them for the DD bit. */ txq->ift_rs_pending += nsegs + 1; if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) || iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs) <= MAX_TX_DESC(ctx) + 2) { pi.ipi_flags |= IPI_TX_INTR; txq->ift_rs_pending = 0; } pi.ipi_segs = segs; pi.ipi_nsegs = nsegs; MPASS(pidx >= 0 && pidx < txq->ift_size); #ifdef PKT_DEBUG print_pkt(&pi); #endif if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) { bus_dmamap_sync(buf_tag, map, BUS_DMASYNC_PREWRITE); DBG_COUNTER_INC(tx_encap); MPASS(pi.ipi_new_pidx < txq->ift_size); ndesc = pi.ipi_new_pidx - pi.ipi_pidx; if (pi.ipi_new_pidx < pi.ipi_pidx) { ndesc += txq->ift_size; txq->ift_gen = 1; } /* * drivers can need as many as * two sentinels */ MPASS(ndesc <= pi.ipi_nsegs + 2); MPASS(pi.ipi_new_pidx != pidx); MPASS(ndesc > 0); txq->ift_in_use += ndesc; /* * We update the last software descriptor again here because there may * be a sentinel and/or there may be more mbufs than segments */ txq->ift_pidx = pi.ipi_new_pidx; txq->ift_npending += pi.ipi_ndescs; } else { *m_headp = m_head = iflib_remove_mbuf(txq); if (err == EFBIG) { txq->ift_txd_encap_efbig++; if (remap < 2) { remap = 1; goto defrag; } } goto defrag_failed; } /* * err can't possibly be non-zero here, so we don't neet to test it * to see if we need to DBG_COUNTER_INC(encap_txd_encap_fail). */ return (err); defrag_failed: txq->ift_mbuf_defrag_failed++; txq->ift_map_failed++; m_freem(*m_headp); DBG_COUNTER_INC(tx_frees); *m_headp = NULL; DBG_COUNTER_INC(encap_txd_encap_fail); return (ENOMEM); } static void iflib_tx_desc_free(iflib_txq_t txq, int n) { uint32_t qsize, cidx, mask, gen; struct mbuf *m, **ifsd_m; bool do_prefetch; cidx = txq->ift_cidx; gen = txq->ift_gen; qsize = txq->ift_size; mask = qsize-1; ifsd_m = txq->ift_sds.ifsd_m; do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH); while (n-- > 0) { if (do_prefetch) { prefetch(ifsd_m[(cidx + 3) & mask]); prefetch(ifsd_m[(cidx + 4) & mask]); } if ((m = ifsd_m[cidx]) != NULL) { prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]); if (m->m_pkthdr.csum_flags & CSUM_TSO) { bus_dmamap_sync(txq->ift_tso_buf_tag, txq->ift_sds.ifsd_tso_map[cidx], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_tso_buf_tag, txq->ift_sds.ifsd_tso_map[cidx]); } else { bus_dmamap_sync(txq->ift_buf_tag, txq->ift_sds.ifsd_map[cidx], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[cidx]); } /* XXX we don't support any drivers that batch packets yet */ MPASS(m->m_nextpkt == NULL); m_freem(m); ifsd_m[cidx] = NULL; #if MEMORY_LOGGING txq->ift_dequeued++; #endif DBG_COUNTER_INC(tx_frees); } if (__predict_false(++cidx == qsize)) { cidx = 0; gen = 0; } } txq->ift_cidx = cidx; txq->ift_gen = gen; } static __inline int iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh) { int reclaim; if_ctx_t ctx = txq->ift_ctx; KASSERT(thresh >= 0, ("invalid threshold to reclaim")); MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size); /* * Need a rate-limiting check so that this isn't called every time */ iflib_tx_credits_update(ctx, txq); reclaim = DESC_RECLAIMABLE(txq); if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) { #ifdef INVARIANTS if (iflib_verbose_debug) { printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__, txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments, reclaim, thresh); } #endif return (0); } iflib_tx_desc_free(txq, reclaim); txq->ift_cleaned += reclaim; txq->ift_in_use -= reclaim; return (reclaim); } static struct mbuf ** _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining) { int next, size; struct mbuf **items; size = r->size; next = (cidx + CACHE_PTR_INCREMENT) & (size-1); items = __DEVOLATILE(struct mbuf **, &r->items[0]); prefetch(items[(cidx + offset) & (size-1)]); if (remaining > 1) { prefetch2cachelines(&items[next]); prefetch2cachelines(items[(cidx + offset + 1) & (size-1)]); prefetch2cachelines(items[(cidx + offset + 2) & (size-1)]); prefetch2cachelines(items[(cidx + offset + 3) & (size-1)]); } return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)])); } static void iflib_txq_check_drain(iflib_txq_t txq, int budget) { ifmp_ring_check_drainage(txq->ift_br, budget); } static uint32_t iflib_txq_can_drain(struct ifmp_ring *r) { iflib_txq_t txq = r->cookie; if_ctx_t ctx = txq->ift_ctx; if (TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2) return (1); bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD); return (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false)); } static uint32_t iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) { iflib_txq_t txq = r->cookie; if_ctx_t ctx = txq->ift_ctx; if_t ifp = ctx->ifc_ifp; struct mbuf *m, **mp; int avail, bytes_sent, consumed, count, err, i, in_use_prev; int mcast_sent, pkt_sent, reclaimed, txq_avail; bool do_prefetch, rang, ring; if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) || !LINK_ACTIVE(ctx))) { DBG_COUNTER_INC(txq_drain_notready); return (0); } reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); rang = iflib_txd_db_check(ctx, txq, reclaimed, txq->ift_in_use); avail = IDXDIFF(pidx, cidx, r->size); if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) { DBG_COUNTER_INC(txq_drain_flushing); for (i = 0; i < avail; i++) { if (__predict_true(r->items[(cidx + i) & (r->size-1)] != (void *)txq)) m_free(r->items[(cidx + i) & (r->size-1)]); r->items[(cidx + i) & (r->size-1)] = NULL; } return (avail); } if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) { txq->ift_qstatus = IFLIB_QUEUE_IDLE; CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); DBG_COUNTER_INC(txq_drain_oactive); return (0); } if (reclaimed) txq->ift_qstatus = IFLIB_QUEUE_IDLE; consumed = mcast_sent = bytes_sent = pkt_sent = 0; count = MIN(avail, TX_BATCH_SIZE); #ifdef INVARIANTS if (iflib_verbose_debug) printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__, avail, ctx->ifc_flags, TXQ_AVAIL(txq)); #endif do_prefetch = (ctx->ifc_flags & IFC_PREFETCH); txq_avail = TXQ_AVAIL(txq); err = 0; for (i = 0; i < count && txq_avail > MAX_TX_DESC(ctx) + 2; i++) { int rem = do_prefetch ? count - i : 0; mp = _ring_peek_one(r, cidx, i, rem); MPASS(mp != NULL && *mp != NULL); if (__predict_false(*mp == (struct mbuf *)txq)) { consumed++; continue; } in_use_prev = txq->ift_in_use; err = iflib_encap(txq, mp); if (__predict_false(err)) { /* no room - bail out */ if (err == ENOBUFS) break; consumed++; /* we can't send this packet - skip it */ continue; } consumed++; pkt_sent++; m = *mp; DBG_COUNTER_INC(tx_sent); bytes_sent += m->m_pkthdr.len; mcast_sent += !!(m->m_flags & M_MCAST); txq_avail = TXQ_AVAIL(txq); txq->ift_db_pending += (txq->ift_in_use - in_use_prev); ETHER_BPF_MTAP(ifp, m); if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING))) break; rang = iflib_txd_db_check(ctx, txq, false, in_use_prev); } /* deliberate use of bitwise or to avoid gratuitous short-circuit */ ring = rang ? false : (iflib_min_tx_latency | err) || (TXQ_AVAIL(txq) < MAX_TX_DESC(ctx)); iflib_txd_db_check(ctx, txq, ring, txq->ift_in_use); if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent); if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent); if (mcast_sent) if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent); #ifdef INVARIANTS if (iflib_verbose_debug) printf("consumed=%d\n", consumed); #endif return (consumed); } static uint32_t iflib_txq_drain_always(struct ifmp_ring *r) { return (1); } static uint32_t iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) { int i, avail; struct mbuf **mp; iflib_txq_t txq; txq = r->cookie; txq->ift_qstatus = IFLIB_QUEUE_IDLE; CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); avail = IDXDIFF(pidx, cidx, r->size); for (i = 0; i < avail; i++) { mp = _ring_peek_one(r, cidx, i, avail - i); if (__predict_false(*mp == (struct mbuf *)txq)) continue; m_freem(*mp); DBG_COUNTER_INC(tx_frees); } MPASS(ifmp_ring_is_stalled(r) == 0); return (avail); } static void iflib_ifmp_purge(iflib_txq_t txq) { struct ifmp_ring *r; r = txq->ift_br; r->drain = iflib_txq_drain_free; r->can_drain = iflib_txq_drain_always; ifmp_ring_check_drainage(r, r->size); r->drain = iflib_txq_drain; r->can_drain = iflib_txq_can_drain; } static void _task_fn_tx(void *context) { iflib_txq_t txq = context; if_ctx_t ctx = txq->ift_ctx; if_t ifp = ctx->ifc_ifp; int abdicate = ctx->ifc_sysctl_tx_abdicate; #ifdef IFLIB_DIAGNOSTICS txq->ift_cpu_exec_count[curcpu]++; #endif if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) return; #ifdef DEV_NETMAP if ((if_getcapenable(ifp) & IFCAP_NETMAP) && netmap_tx_irq(ifp, txq->ift_id)) goto skip_ifmp; #endif #ifdef ALTQ if (ALTQ_IS_ENABLED(&ifp->if_snd)) iflib_altq_if_start(ifp); #endif if (txq->ift_db_pending) ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE, abdicate); else if (!abdicate) ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); /* * When abdicating, we always need to check drainage, not just when we don't enqueue */ if (abdicate) ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); #ifdef DEV_NETMAP skip_ifmp: #endif if (ctx->ifc_flags & IFC_LEGACY) IFDI_INTR_ENABLE(ctx); else IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id); } static void _task_fn_rx(void *context) { iflib_rxq_t rxq = context; if_ctx_t ctx = rxq->ifr_ctx; uint8_t more; uint16_t budget; #ifdef DEV_NETMAP u_int work = 0; int nmirq; #endif #ifdef IFLIB_DIAGNOSTICS rxq->ifr_cpu_exec_count[curcpu]++; #endif DBG_COUNTER_INC(task_fn_rxs); if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) return; #ifdef DEV_NETMAP nmirq = netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work); if (nmirq != NM_IRQ_PASS) { more = (nmirq == NM_IRQ_RESCHED) ? IFLIB_RXEOF_MORE : 0; goto skip_rxeof; } #endif budget = ctx->ifc_sysctl_rx_budget; if (budget == 0) budget = 16; /* XXX */ more = iflib_rxeof(rxq, budget); #ifdef DEV_NETMAP skip_rxeof: #endif if ((more & IFLIB_RXEOF_MORE) == 0) { if (ctx->ifc_flags & IFC_LEGACY) IFDI_INTR_ENABLE(ctx); else IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); DBG_COUNTER_INC(rx_intr_enables); } if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) return; if (more & IFLIB_RXEOF_MORE) GROUPTASK_ENQUEUE(&rxq->ifr_task); else if (more & IFLIB_RXEOF_EMPTY) callout_reset_curcpu(&rxq->ifr_watchdog, 1, &_task_fn_rx_watchdog, rxq); } static void _task_fn_admin(void *context) { if_ctx_t ctx = context; if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; iflib_txq_t txq; int i; bool oactive, running, do_reset, do_watchdog, in_detach; uint32_t reset_on = hz / 2; STATE_LOCK(ctx); running = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING); oactive = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE); do_reset = (ctx->ifc_flags & IFC_DO_RESET); do_watchdog = (ctx->ifc_flags & IFC_DO_WATCHDOG); in_detach = (ctx->ifc_flags & IFC_IN_DETACH); ctx->ifc_flags &= ~(IFC_DO_RESET|IFC_DO_WATCHDOG); STATE_UNLOCK(ctx); if ((!running && !oactive) && !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN)) return; if (in_detach) return; CTX_LOCK(ctx); for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); } if (do_watchdog) { ctx->ifc_watchdog_events++; IFDI_WATCHDOG_RESET(ctx); } IFDI_UPDATE_ADMIN_STATUS(ctx); for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { #ifdef DEV_NETMAP reset_on = hz / 2; if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) iflib_netmap_timer_adjust(ctx, txq, &reset_on); #endif callout_reset_on(&txq->ift_timer, reset_on, iflib_timer, txq, txq->ift_timer.c_cpu); } IFDI_LINK_INTR_ENABLE(ctx); if (do_reset) iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); if (LINK_ACTIVE(ctx) == 0) return; for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); } static void _task_fn_iov(void *context) { if_ctx_t ctx = context; if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) && !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN)) return; CTX_LOCK(ctx); IFDI_VFLR_HANDLE(ctx); CTX_UNLOCK(ctx); } static int iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS) { int err; if_int_delay_info_t info; if_ctx_t ctx; info = (if_int_delay_info_t)arg1; ctx = info->iidi_ctx; info->iidi_req = req; info->iidi_oidp = oidp; CTX_LOCK(ctx); err = IFDI_SYSCTL_INT_DELAY(ctx, info); CTX_UNLOCK(ctx); return (err); } /********************************************************************* * * IFNET FUNCTIONS * **********************************************************************/ static void iflib_if_init_locked(if_ctx_t ctx) { iflib_stop(ctx); iflib_init_locked(ctx); } static void iflib_if_init(void *arg) { if_ctx_t ctx = arg; CTX_LOCK(ctx); iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); } static int iflib_if_transmit(if_t ifp, struct mbuf *m) { if_ctx_t ctx = if_getsoftc(ifp); iflib_txq_t txq; int err, qidx; int abdicate = ctx->ifc_sysctl_tx_abdicate; if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) { DBG_COUNTER_INC(tx_frees); m_freem(m); return (ENETDOWN); } MPASS(m->m_nextpkt == NULL); /* ALTQ-enabled interfaces always use queue 0. */ qidx = 0; if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m) && !ALTQ_IS_ENABLED(&ifp->if_snd)) qidx = QIDX(ctx, m); /* * XXX calculate buf_ring based on flowid (divvy up bits?) */ txq = &ctx->ifc_txqs[qidx]; #ifdef DRIVER_BACKPRESSURE if (txq->ift_closed) { while (m != NULL) { next = m->m_nextpkt; m->m_nextpkt = NULL; m_freem(m); DBG_COUNTER_INC(tx_frees); m = next; } return (ENOBUFS); } #endif #ifdef notyet qidx = count = 0; mp = marr; next = m; do { count++; next = next->m_nextpkt; } while (next != NULL); if (count > nitems(marr)) if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) { /* XXX check nextpkt */ m_freem(m); /* XXX simplify for now */ DBG_COUNTER_INC(tx_frees); return (ENOBUFS); } for (next = m, i = 0; next != NULL; i++) { mp[i] = next; next = next->m_nextpkt; mp[i]->m_nextpkt = NULL; } #endif DBG_COUNTER_INC(tx_seen); err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE, abdicate); if (abdicate) GROUPTASK_ENQUEUE(&txq->ift_task); if (err) { if (!abdicate) GROUPTASK_ENQUEUE(&txq->ift_task); /* support forthcoming later */ #ifdef DRIVER_BACKPRESSURE txq->ift_closed = TRUE; #endif ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); m_freem(m); DBG_COUNTER_INC(tx_frees); } return (err); } #ifdef ALTQ /* * The overall approach to integrating iflib with ALTQ is to continue to use * the iflib mp_ring machinery between the ALTQ queue(s) and the hardware * ring. Technically, when using ALTQ, queueing to an intermediate mp_ring * is redundant/unnecessary, but doing so minimizes the amount of * ALTQ-specific code required in iflib. It is assumed that the overhead of * redundantly queueing to an intermediate mp_ring is swamped by the * performance limitations inherent in using ALTQ. * * When ALTQ support is compiled in, all iflib drivers will use a transmit * routine, iflib_altq_if_transmit(), that checks if ALTQ is enabled for the * given interface. If ALTQ is enabled for an interface, then all * transmitted packets for that interface will be submitted to the ALTQ * subsystem via IFQ_ENQUEUE(). We don't use the legacy if_transmit() * implementation because it uses IFQ_HANDOFF(), which will duplicatively * update stats that the iflib machinery handles, and which is sensitve to * the disused IFF_DRV_OACTIVE flag. Additionally, iflib_altq_if_start() * will be installed as the start routine for use by ALTQ facilities that * need to trigger queue drains on a scheduled basis. * */ static void iflib_altq_if_start(if_t ifp) { struct ifaltq *ifq = &ifp->if_snd; struct mbuf *m; IFQ_LOCK(ifq); IFQ_DEQUEUE_NOLOCK(ifq, m); while (m != NULL) { iflib_if_transmit(ifp, m); IFQ_DEQUEUE_NOLOCK(ifq, m); } IFQ_UNLOCK(ifq); } static int iflib_altq_if_transmit(if_t ifp, struct mbuf *m) { int err; if (ALTQ_IS_ENABLED(&ifp->if_snd)) { IFQ_ENQUEUE(&ifp->if_snd, m, err); if (err == 0) iflib_altq_if_start(ifp); } else err = iflib_if_transmit(ifp, m); return (err); } #endif /* ALTQ */ static void iflib_if_qflush(if_t ifp) { if_ctx_t ctx = if_getsoftc(ifp); iflib_txq_t txq = ctx->ifc_txqs; int i; STATE_LOCK(ctx); ctx->ifc_flags |= IFC_QFLUSH; STATE_UNLOCK(ctx); for (i = 0; i < NTXQSETS(ctx); i++, txq++) while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br))) iflib_txq_check_drain(txq, 0); STATE_LOCK(ctx); ctx->ifc_flags &= ~IFC_QFLUSH; STATE_UNLOCK(ctx); /* * When ALTQ is enabled, this will also take care of purging the * ALTQ queue(s). */ if_qflush(ifp); } #define IFCAP_FLAGS (IFCAP_HWCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \ IFCAP_TSO | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \ IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | \ IFCAP_VLAN_HWTSO | IFCAP_VLAN_HWCSUM | IFCAP_NOMAP) static int iflib_if_ioctl(if_t ifp, u_long command, caddr_t data) { if_ctx_t ctx = if_getsoftc(ifp); struct ifreq *ifr = (struct ifreq *)data; #if defined(INET) || defined(INET6) struct ifaddr *ifa = (struct ifaddr *)data; #endif bool avoid_reset = false; int err = 0, reinit = 0, bits; switch (command) { case SIOCSIFADDR: #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) avoid_reset = true; #endif #ifdef INET6 if (ifa->ifa_addr->sa_family == AF_INET6) avoid_reset = true; #endif /* ** Calling init results in link renegotiation, ** so we avoid doing it when possible. */ if (avoid_reset) { if_setflagbits(ifp, IFF_UP,0); if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING)) reinit = 1; #ifdef INET if (!(if_getflags(ifp) & IFF_NOARP)) arp_ifinit(ifp, ifa); #endif } else err = ether_ioctl(ifp, command, data); break; case SIOCSIFMTU: CTX_LOCK(ctx); if (ifr->ifr_mtu == if_getmtu(ifp)) { CTX_UNLOCK(ctx); break; } bits = if_getdrvflags(ifp); /* stop the driver and free any clusters before proceeding */ iflib_stop(ctx); if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) { STATE_LOCK(ctx); if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size) ctx->ifc_flags |= IFC_MULTISEG; else ctx->ifc_flags &= ~IFC_MULTISEG; STATE_UNLOCK(ctx); err = if_setmtu(ifp, ifr->ifr_mtu); } iflib_init_locked(ctx); STATE_LOCK(ctx); if_setdrvflags(ifp, bits); STATE_UNLOCK(ctx); CTX_UNLOCK(ctx); break; case SIOCSIFFLAGS: CTX_LOCK(ctx); if (if_getflags(ifp) & IFF_UP) { if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { if ((if_getflags(ifp) ^ ctx->ifc_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) { CTX_UNLOCK(ctx); err = IFDI_PROMISC_SET(ctx, if_getflags(ifp)); CTX_LOCK(ctx); } } else reinit = 1; } else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { iflib_stop(ctx); } ctx->ifc_if_flags = if_getflags(ifp); CTX_UNLOCK(ctx); break; case SIOCADDMULTI: case SIOCDELMULTI: if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { CTX_LOCK(ctx); IFDI_INTR_DISABLE(ctx); IFDI_MULTI_SET(ctx); IFDI_INTR_ENABLE(ctx); CTX_UNLOCK(ctx); } break; case SIOCSIFMEDIA: CTX_LOCK(ctx); IFDI_MEDIA_SET(ctx); CTX_UNLOCK(ctx); /* FALLTHROUGH */ case SIOCGIFMEDIA: case SIOCGIFXMEDIA: err = ifmedia_ioctl(ifp, ifr, ctx->ifc_mediap, command); break; case SIOCGI2C: { struct ifi2creq i2c; err = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c)); if (err != 0) break; if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) { err = EINVAL; break; } if (i2c.len > sizeof(i2c.data)) { err = EINVAL; break; } if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0) err = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c)); break; } case SIOCSIFCAP: { int mask, setmask, oldmask; oldmask = if_getcapenable(ifp); mask = ifr->ifr_reqcap ^ oldmask; mask &= ctx->ifc_softc_ctx.isc_capabilities | IFCAP_NOMAP; setmask = 0; #ifdef TCP_OFFLOAD setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6); #endif setmask |= (mask & IFCAP_FLAGS); setmask |= (mask & IFCAP_WOL); /* * If any RX csum has changed, change all the ones that * are supported by the driver. */ if (setmask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { setmask |= ctx->ifc_softc_ctx.isc_capabilities & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6); } /* * want to ensure that traffic has stopped before we change any of the flags */ if (setmask) { CTX_LOCK(ctx); bits = if_getdrvflags(ifp); if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL) iflib_stop(ctx); STATE_LOCK(ctx); if_togglecapenable(ifp, setmask); STATE_UNLOCK(ctx); if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL) iflib_init_locked(ctx); STATE_LOCK(ctx); if_setdrvflags(ifp, bits); STATE_UNLOCK(ctx); CTX_UNLOCK(ctx); } if_vlancap(ifp); break; } case SIOCGPRIVATE_0: case SIOCSDRVSPEC: case SIOCGDRVSPEC: CTX_LOCK(ctx); err = IFDI_PRIV_IOCTL(ctx, command, data); CTX_UNLOCK(ctx); break; default: err = ether_ioctl(ifp, command, data); break; } if (reinit) iflib_if_init(ctx); return (err); } static uint64_t iflib_if_get_counter(if_t ifp, ift_counter cnt) { if_ctx_t ctx = if_getsoftc(ifp); return (IFDI_GET_COUNTER(ctx, cnt)); } /********************************************************************* * * OTHER FUNCTIONS EXPORTED TO THE STACK * **********************************************************************/ static void iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag) { if_ctx_t ctx = if_getsoftc(ifp); if ((void *)ctx != arg) return; if ((vtag == 0) || (vtag > 4095)) return; if (iflib_in_detach(ctx)) return; CTX_LOCK(ctx); /* Driver may need all untagged packets to be flushed */ if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG)) iflib_stop(ctx); IFDI_VLAN_REGISTER(ctx, vtag); /* Re-init to load the changes, if required */ if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG)) iflib_init_locked(ctx); CTX_UNLOCK(ctx); } static void iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag) { if_ctx_t ctx = if_getsoftc(ifp); if ((void *)ctx != arg) return; if ((vtag == 0) || (vtag > 4095)) return; CTX_LOCK(ctx); /* Driver may need all tagged packets to be flushed */ if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG)) iflib_stop(ctx); IFDI_VLAN_UNREGISTER(ctx, vtag); /* Re-init to load the changes, if required */ if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG)) iflib_init_locked(ctx); CTX_UNLOCK(ctx); } static void iflib_led_func(void *arg, int onoff) { if_ctx_t ctx = arg; CTX_LOCK(ctx); IFDI_LED_FUNC(ctx, onoff); CTX_UNLOCK(ctx); } /********************************************************************* * * BUS FUNCTION DEFINITIONS * **********************************************************************/ int iflib_device_probe(device_t dev) { const pci_vendor_info_t *ent; if_shared_ctx_t sctx; uint16_t pci_device_id, pci_rev_id, pci_subdevice_id, pci_subvendor_id; uint16_t pci_vendor_id; if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) return (ENOTSUP); pci_vendor_id = pci_get_vendor(dev); pci_device_id = pci_get_device(dev); pci_subvendor_id = pci_get_subvendor(dev); pci_subdevice_id = pci_get_subdevice(dev); pci_rev_id = pci_get_revid(dev); if (sctx->isc_parse_devinfo != NULL) sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id); ent = sctx->isc_vendor_info; while (ent->pvi_vendor_id != 0) { if (pci_vendor_id != ent->pvi_vendor_id) { ent++; continue; } if ((pci_device_id == ent->pvi_device_id) && ((pci_subvendor_id == ent->pvi_subvendor_id) || (ent->pvi_subvendor_id == 0)) && ((pci_subdevice_id == ent->pvi_subdevice_id) || (ent->pvi_subdevice_id == 0)) && ((pci_rev_id == ent->pvi_rev_id) || (ent->pvi_rev_id == 0))) { device_set_desc_copy(dev, ent->pvi_name); /* this needs to be changed to zero if the bus probing code * ever stops re-probing on best match because the sctx * may have its values over written by register calls * in subsequent probes */ return (BUS_PROBE_DEFAULT); } ent++; } return (ENXIO); } int iflib_device_probe_vendor(device_t dev) { int probe; probe = iflib_device_probe(dev); if (probe == BUS_PROBE_DEFAULT) return (BUS_PROBE_VENDOR); else return (probe); } static void iflib_reset_qvalues(if_ctx_t ctx) { if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; device_t dev = ctx->ifc_dev; int i; if (ctx->ifc_sysctl_ntxqs != 0) scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs; if (ctx->ifc_sysctl_nrxqs != 0) scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs; for (i = 0; i < sctx->isc_ntxqs; i++) { if (ctx->ifc_sysctl_ntxds[i] != 0) scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i]; else scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; } for (i = 0; i < sctx->isc_nrxqs; i++) { if (ctx->ifc_sysctl_nrxds[i] != 0) scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i]; else scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; } for (i = 0; i < sctx->isc_nrxqs; i++) { if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) { device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n", i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]); scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i]; } if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) { device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n", i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]); scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i]; } if (!powerof2(scctx->isc_nrxd[i])) { device_printf(dev, "nrxd%d: %d is not a power of 2 - using default value of %d\n", i, scctx->isc_nrxd[i], sctx->isc_nrxd_default[i]); scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; } } for (i = 0; i < sctx->isc_ntxqs; i++) { if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) { device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n", i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]); scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i]; } if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) { device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n", i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]); scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i]; } if (!powerof2(scctx->isc_ntxd[i])) { device_printf(dev, "ntxd%d: %d is not a power of 2 - using default value of %d\n", i, scctx->isc_ntxd[i], sctx->isc_ntxd_default[i]); scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; } } } static void iflib_add_pfil(if_ctx_t ctx) { struct pfil_head *pfil; struct pfil_head_args pa; iflib_rxq_t rxq; int i; pa.pa_version = PFIL_VERSION; pa.pa_flags = PFIL_IN; pa.pa_type = PFIL_TYPE_ETHERNET; pa.pa_headname = ctx->ifc_ifp->if_xname; pfil = pfil_head_register(&pa); for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { rxq->pfil = pfil; } } static void iflib_rem_pfil(if_ctx_t ctx) { struct pfil_head *pfil; iflib_rxq_t rxq; int i; rxq = ctx->ifc_rxqs; pfil = rxq->pfil; for (i = 0; i < NRXQSETS(ctx); i++, rxq++) { rxq->pfil = NULL; } pfil_head_unregister(pfil); } static uint16_t get_ctx_core_offset(if_ctx_t ctx) { if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; struct cpu_offset *op; uint16_t qc; uint16_t ret = ctx->ifc_sysctl_core_offset; if (ret != CORE_OFFSET_UNSPECIFIED) return (ret); if (ctx->ifc_sysctl_separate_txrx) qc = scctx->isc_ntxqsets + scctx->isc_nrxqsets; else qc = max(scctx->isc_ntxqsets, scctx->isc_nrxqsets); mtx_lock(&cpu_offset_mtx); SLIST_FOREACH(op, &cpu_offsets, entries) { if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) { ret = op->offset; op->offset += qc; MPASS(op->refcount < UINT_MAX); op->refcount++; break; } } if (ret == CORE_OFFSET_UNSPECIFIED) { ret = 0; op = malloc(sizeof(struct cpu_offset), M_IFLIB, M_NOWAIT | M_ZERO); if (op == NULL) { device_printf(ctx->ifc_dev, "allocation for cpu offset failed.\n"); } else { op->offset = qc; op->refcount = 1; CPU_COPY(&ctx->ifc_cpus, &op->set); SLIST_INSERT_HEAD(&cpu_offsets, op, entries); } } mtx_unlock(&cpu_offset_mtx); return (ret); } static void unref_ctx_core_offset(if_ctx_t ctx) { struct cpu_offset *op, *top; mtx_lock(&cpu_offset_mtx); SLIST_FOREACH_SAFE(op, &cpu_offsets, entries, top) { if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) { MPASS(op->refcount > 0); op->refcount--; if (op->refcount == 0) { SLIST_REMOVE(&cpu_offsets, op, cpu_offset, entries); free(op, M_IFLIB); } break; } } mtx_unlock(&cpu_offset_mtx); } int iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp) { if_ctx_t ctx; if_t ifp; if_softc_ctx_t scctx; kobjop_desc_t kobj_desc; kobj_method_t *kobj_method; int err, msix, rid; int num_txd, num_rxd; ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO); if (sc == NULL) { sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); device_set_softc(dev, ctx); ctx->ifc_flags |= IFC_SC_ALLOCATED; } ctx->ifc_sctx = sctx; ctx->ifc_dev = dev; ctx->ifc_softc = sc; if ((err = iflib_register(ctx)) != 0) { device_printf(dev, "iflib_register failed %d\n", err); goto fail_ctx_free; } iflib_add_device_sysctl_pre(ctx); scctx = &ctx->ifc_softc_ctx; ifp = ctx->ifc_ifp; iflib_reset_qvalues(ctx); CTX_LOCK(ctx); if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); goto fail_unlock; } _iflib_pre_assert(scctx); ctx->ifc_txrx = *scctx->isc_txrx; if (sctx->isc_flags & IFLIB_DRIVER_MEDIA) ctx->ifc_mediap = scctx->isc_media; #ifdef INVARIANTS if (scctx->isc_capabilities & IFCAP_TXCSUM) MPASS(scctx->isc_tx_csum_flags); #endif if_setcapabilities(ifp, scctx->isc_capabilities | IFCAP_HWSTATS | IFCAP_NOMAP); if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS | IFCAP_NOMAP); if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets)) scctx->isc_ntxqsets = scctx->isc_ntxqsets_max; if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets)) scctx->isc_nrxqsets = scctx->isc_nrxqsets_max; num_txd = iflib_num_tx_descs(ctx); num_rxd = iflib_num_rx_descs(ctx); /* XXX change for per-queue sizes */ device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n", num_txd, num_rxd); if (scctx->isc_tx_nsegments > num_txd / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_nsegments = max(1, num_txd / MAX_SINGLE_PACKET_FRACTION); if (scctx->isc_tx_tso_segments_max > num_txd / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_tso_segments_max = max(1, num_txd / MAX_SINGLE_PACKET_FRACTION); /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ if (if_getcapabilities(ifp) & IFCAP_TSO) { /* * The stack can't handle a TSO size larger than IP_MAXPACKET, * but some MACs do. */ if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max, IP_MAXPACKET)); /* * Take maximum number of m_pullup(9)'s in iflib_parse_header() * into account. In the worst case, each of these calls will * add another mbuf and, thus, the requirement for another DMA * segment. So for best performance, it doesn't make sense to * advertize a maximum of TSO segments that typically will * require defragmentation in iflib_encap(). */ if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3); if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max); } if (scctx->isc_rss_table_size == 0) scctx->isc_rss_table_size = 64; scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); /* XXX format name */ taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, NULL, NULL, "admin"); /* Set up cpu set. If it fails, use the set of all CPUs. */ if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) != 0) { device_printf(dev, "Unable to fetch CPU list\n"); CPU_COPY(&all_cpus, &ctx->ifc_cpus); } MPASS(CPU_COUNT(&ctx->ifc_cpus) > 0); /* ** Now set up MSI or MSI-X, should return us the number of supported ** vectors (will be 1 for a legacy interrupt and MSI). */ if (sctx->isc_flags & IFLIB_SKIP_MSIX) { msix = scctx->isc_vectors; } else if (scctx->isc_msix_bar != 0) /* * The simple fact that isc_msix_bar is not 0 does not mean we * we have a good value there that is known to work. */ msix = iflib_msix_init(ctx); else { scctx->isc_vectors = 1; scctx->isc_ntxqsets = 1; scctx->isc_nrxqsets = 1; scctx->isc_intr = IFLIB_INTR_LEGACY; msix = 0; } /* Get memory for the station queues */ if ((err = iflib_queues_alloc(ctx))) { device_printf(dev, "Unable to allocate queue memory\n"); goto fail_intr_free; } if ((err = iflib_qset_structures_setup(ctx))) goto fail_queues; /* * Now that we know how many queues there are, get the core offset. */ ctx->ifc_sysctl_core_offset = get_ctx_core_offset(ctx); if (msix > 1) { /* * When using MSI-X, ensure that ifdi_{r,t}x_queue_intr_enable * aren't the default NULL implementation. */ kobj_desc = &ifdi_rx_queue_intr_enable_desc; kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL, kobj_desc); if (kobj_method == &kobj_desc->deflt) { device_printf(dev, "MSI-X requires ifdi_rx_queue_intr_enable method"); err = EOPNOTSUPP; goto fail_queues; } kobj_desc = &ifdi_tx_queue_intr_enable_desc; kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL, kobj_desc); if (kobj_method == &kobj_desc->deflt) { device_printf(dev, "MSI-X requires ifdi_tx_queue_intr_enable method"); err = EOPNOTSUPP; goto fail_queues; } /* * Assign the MSI-X vectors. * Note that the default NULL ifdi_msix_intr_assign method will * fail here, too. */ err = IFDI_MSIX_INTR_ASSIGN(ctx, msix); if (err != 0) { device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", err); goto fail_queues; } } else if (scctx->isc_intr != IFLIB_INTR_MSIX) { rid = 0; if (scctx->isc_intr == IFLIB_INTR_MSI) { MPASS(msix == 1); rid = 1; } if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) { device_printf(dev, "iflib_legacy_setup failed %d\n", err); goto fail_queues; } } else { device_printf(dev, "Cannot use iflib with only 1 MSI-X interrupt!\n"); err = ENODEV; goto fail_intr_free; } ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); if ((err = IFDI_ATTACH_POST(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); goto fail_detach; } /* * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. * This must appear after the call to ether_ifattach() because * ether_ifattach() sets if_hdrlen to the default value. */ if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); if ((err = iflib_netmap_attach(ctx))) { device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err); goto fail_detach; } *ctxp = ctx; DEBUGNET_SET(ctx->ifc_ifp, iflib); if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); iflib_add_device_sysctl_post(ctx); iflib_add_pfil(ctx); ctx->ifc_flags |= IFC_INIT_DONE; CTX_UNLOCK(ctx); return (0); fail_detach: ether_ifdetach(ctx->ifc_ifp); fail_intr_free: iflib_free_intr_mem(ctx); fail_queues: iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); taskqgroup_detach(qgroup_if_config_tqg, &ctx->ifc_admin_task); IFDI_DETACH(ctx); fail_unlock: CTX_UNLOCK(ctx); iflib_deregister(ctx); fail_ctx_free: device_set_softc(ctx->ifc_dev, NULL); if (ctx->ifc_flags & IFC_SC_ALLOCATED) free(ctx->ifc_softc, M_IFLIB); free(ctx, M_IFLIB); return (err); } int iflib_pseudo_register(device_t dev, if_shared_ctx_t sctx, if_ctx_t *ctxp, struct iflib_cloneattach_ctx *clctx) { int num_txd, num_rxd; int err; if_ctx_t ctx; if_t ifp; if_softc_ctx_t scctx; int i; void *sc; ctx = malloc(sizeof(*ctx), M_IFLIB, M_WAITOK|M_ZERO); sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); ctx->ifc_flags |= IFC_SC_ALLOCATED; if (sctx->isc_flags & (IFLIB_PSEUDO|IFLIB_VIRTUAL)) ctx->ifc_flags |= IFC_PSEUDO; ctx->ifc_sctx = sctx; ctx->ifc_softc = sc; ctx->ifc_dev = dev; if ((err = iflib_register(ctx)) != 0) { device_printf(dev, "%s: iflib_register failed %d\n", __func__, err); goto fail_ctx_free; } iflib_add_device_sysctl_pre(ctx); scctx = &ctx->ifc_softc_ctx; ifp = ctx->ifc_ifp; iflib_reset_qvalues(ctx); CTX_LOCK(ctx); if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); goto fail_unlock; } if (sctx->isc_flags & IFLIB_GEN_MAC) ether_gen_addr(ifp, &ctx->ifc_mac); if ((err = IFDI_CLONEATTACH(ctx, clctx->cc_ifc, clctx->cc_name, clctx->cc_params)) != 0) { device_printf(dev, "IFDI_CLONEATTACH failed %d\n", err); goto fail_unlock; } #ifdef INVARIANTS if (scctx->isc_capabilities & IFCAP_TXCSUM) MPASS(scctx->isc_tx_csum_flags); #endif if_setcapabilities(ifp, scctx->isc_capabilities | IFCAP_HWSTATS | IFCAP_LINKSTATE); if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS | IFCAP_LINKSTATE); ifp->if_flags |= IFF_NOGROUP; if (sctx->isc_flags & IFLIB_PSEUDO) { ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO); if (sctx->isc_flags & IFLIB_PSEUDO_ETHER) { ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); } else { if_attach(ctx->ifc_ifp); bpfattach(ctx->ifc_ifp, DLT_NULL, sizeof(u_int32_t)); } if ((err = IFDI_ATTACH_POST(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); goto fail_detach; } *ctxp = ctx; /* * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. * This must appear after the call to ether_ifattach() because * ether_ifattach() sets if_hdrlen to the default value. */ if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); iflib_add_device_sysctl_post(ctx); ctx->ifc_flags |= IFC_INIT_DONE; CTX_UNLOCK(ctx); return (0); } ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL); ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO, 0, NULL); ifmedia_set(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO); _iflib_pre_assert(scctx); ctx->ifc_txrx = *scctx->isc_txrx; if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets)) scctx->isc_ntxqsets = scctx->isc_ntxqsets_max; if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets)) scctx->isc_nrxqsets = scctx->isc_nrxqsets_max; num_txd = iflib_num_tx_descs(ctx); num_rxd = iflib_num_rx_descs(ctx); /* XXX change for per-queue sizes */ device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n", num_txd, num_rxd); if (scctx->isc_tx_nsegments > num_txd / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_nsegments = max(1, num_txd / MAX_SINGLE_PACKET_FRACTION); if (scctx->isc_tx_tso_segments_max > num_txd / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_tso_segments_max = max(1, num_txd / MAX_SINGLE_PACKET_FRACTION); /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ if (if_getcapabilities(ifp) & IFCAP_TSO) { /* * The stack can't handle a TSO size larger than IP_MAXPACKET, * but some MACs do. */ if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max, IP_MAXPACKET)); /* * Take maximum number of m_pullup(9)'s in iflib_parse_header() * into account. In the worst case, each of these calls will * add another mbuf and, thus, the requirement for another DMA * segment. So for best performance, it doesn't make sense to * advertize a maximum of TSO segments that typically will * require defragmentation in iflib_encap(). */ if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3); if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max); } if (scctx->isc_rss_table_size == 0) scctx->isc_rss_table_size = 64; scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); /* XXX format name */ taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, NULL, NULL, "admin"); /* XXX --- can support > 1 -- but keep it simple for now */ scctx->isc_intr = IFLIB_INTR_LEGACY; /* Get memory for the station queues */ if ((err = iflib_queues_alloc(ctx))) { device_printf(dev, "Unable to allocate queue memory\n"); goto fail_iflib_detach; } if ((err = iflib_qset_structures_setup(ctx))) { device_printf(dev, "qset structure setup failed %d\n", err); goto fail_queues; } /* * XXX What if anything do we want to do about interrupts? */ ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet); if ((err = IFDI_ATTACH_POST(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); goto fail_detach; } /* * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported. * This must appear after the call to ether_ifattach() because * ether_ifattach() sets if_hdrlen to the default value. */ if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU) if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); /* XXX handle more than one queue */ for (i = 0; i < scctx->isc_nrxqsets; i++) IFDI_RX_CLSET(ctx, 0, i, ctx->ifc_rxqs[i].ifr_fl[0].ifl_sds.ifsd_cl); *ctxp = ctx; if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); iflib_add_device_sysctl_post(ctx); ctx->ifc_flags |= IFC_INIT_DONE; CTX_UNLOCK(ctx); return (0); fail_detach: ether_ifdetach(ctx->ifc_ifp); fail_queues: iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); fail_iflib_detach: IFDI_DETACH(ctx); fail_unlock: CTX_UNLOCK(ctx); iflib_deregister(ctx); fail_ctx_free: free(ctx->ifc_softc, M_IFLIB); free(ctx, M_IFLIB); return (err); } int iflib_pseudo_deregister(if_ctx_t ctx) { if_t ifp = ctx->ifc_ifp; if_shared_ctx_t sctx = ctx->ifc_sctx; iflib_txq_t txq; iflib_rxq_t rxq; int i, j; struct taskqgroup *tqg; iflib_fl_t fl; /* Unregister VLAN event handlers early */ iflib_unregister_vlan_handlers(ctx); if ((sctx->isc_flags & IFLIB_PSEUDO) && (sctx->isc_flags & IFLIB_PSEUDO_ETHER) == 0) { bpfdetach(ifp); if_detach(ifp); } else { ether_ifdetach(ifp); } /* XXX drain any dependent tasks */ tqg = qgroup_if_io_tqg; for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { callout_drain(&txq->ift_timer); if (txq->ift_task.gt_uniq != NULL) taskqgroup_detach(tqg, &txq->ift_task); } for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { callout_drain(&rxq->ifr_watchdog); if (rxq->ifr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &rxq->ifr_task); for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) free(fl->ifl_rx_bitmap, M_IFLIB); } tqg = qgroup_if_config_tqg; if (ctx->ifc_admin_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_admin_task); if (ctx->ifc_vflr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_vflr_task); iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); iflib_deregister(ctx); if (ctx->ifc_flags & IFC_SC_ALLOCATED) free(ctx->ifc_softc, M_IFLIB); free(ctx, M_IFLIB); return (0); } int iflib_device_attach(device_t dev) { if_ctx_t ctx; if_shared_ctx_t sctx; if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) return (ENOTSUP); pci_enable_busmaster(dev); return (iflib_device_register(dev, NULL, sctx, &ctx)); } int iflib_device_deregister(if_ctx_t ctx) { if_t ifp = ctx->ifc_ifp; iflib_txq_t txq; iflib_rxq_t rxq; device_t dev = ctx->ifc_dev; int i, j; struct taskqgroup *tqg; iflib_fl_t fl; /* Make sure VLANS are not using driver */ if (if_vlantrunkinuse(ifp)) { device_printf(dev, "Vlan in use, detach first\n"); return (EBUSY); } #ifdef PCI_IOV if (!CTX_IS_VF(ctx) && pci_iov_detach(dev) != 0) { device_printf(dev, "SR-IOV in use; detach first.\n"); return (EBUSY); } #endif STATE_LOCK(ctx); ctx->ifc_flags |= IFC_IN_DETACH; STATE_UNLOCK(ctx); /* Unregister VLAN handlers before calling iflib_stop() */ iflib_unregister_vlan_handlers(ctx); iflib_netmap_detach(ifp); ether_ifdetach(ifp); CTX_LOCK(ctx); iflib_stop(ctx); CTX_UNLOCK(ctx); iflib_rem_pfil(ctx); if (ctx->ifc_led_dev != NULL) led_destroy(ctx->ifc_led_dev); /* XXX drain any dependent tasks */ tqg = qgroup_if_io_tqg; for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { callout_drain(&txq->ift_timer); if (txq->ift_task.gt_uniq != NULL) taskqgroup_detach(tqg, &txq->ift_task); } for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { if (rxq->ifr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &rxq->ifr_task); for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) free(fl->ifl_rx_bitmap, M_IFLIB); } tqg = qgroup_if_config_tqg; if (ctx->ifc_admin_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_admin_task); if (ctx->ifc_vflr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_vflr_task); CTX_LOCK(ctx); IFDI_DETACH(ctx); CTX_UNLOCK(ctx); /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ iflib_free_intr_mem(ctx); bus_generic_detach(dev); iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); iflib_deregister(ctx); device_set_softc(ctx->ifc_dev, NULL); if (ctx->ifc_flags & IFC_SC_ALLOCATED) free(ctx->ifc_softc, M_IFLIB); unref_ctx_core_offset(ctx); free(ctx, M_IFLIB); return (0); } static void iflib_free_intr_mem(if_ctx_t ctx) { if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) { iflib_irq_free(ctx, &ctx->ifc_legacy_irq); } if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) { pci_release_msi(ctx->ifc_dev); } if (ctx->ifc_msix_mem != NULL) { bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY, rman_get_rid(ctx->ifc_msix_mem), ctx->ifc_msix_mem); ctx->ifc_msix_mem = NULL; } } int iflib_device_detach(device_t dev) { if_ctx_t ctx = device_get_softc(dev); return (iflib_device_deregister(ctx)); } int iflib_device_suspend(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_SUSPEND(ctx); CTX_UNLOCK(ctx); return bus_generic_suspend(dev); } int iflib_device_shutdown(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_SHUTDOWN(ctx); CTX_UNLOCK(ctx); return bus_generic_suspend(dev); } int iflib_device_resume(device_t dev) { if_ctx_t ctx = device_get_softc(dev); iflib_txq_t txq = ctx->ifc_txqs; CTX_LOCK(ctx); IFDI_RESUME(ctx); iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); for (int i = 0; i < NTXQSETS(ctx); i++, txq++) iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); return (bus_generic_resume(dev)); } int iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params) { int error; if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); error = IFDI_IOV_INIT(ctx, num_vfs, params); CTX_UNLOCK(ctx); return (error); } void iflib_device_iov_uninit(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_IOV_UNINIT(ctx); CTX_UNLOCK(ctx); } int iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params) { int error; if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); error = IFDI_IOV_VF_ADD(ctx, vfnum, params); CTX_UNLOCK(ctx); return (error); } /********************************************************************* * * MODULE FUNCTION DEFINITIONS * **********************************************************************/ /* * - Start a fast taskqueue thread for each core * - Start a taskqueue for control operations */ static int iflib_module_init(void) { return (0); } static int iflib_module_event_handler(module_t mod, int what, void *arg) { int err; switch (what) { case MOD_LOAD: if ((err = iflib_module_init()) != 0) return (err); break; case MOD_UNLOAD: return (EBUSY); default: return (EOPNOTSUPP); } return (0); } /********************************************************************* * * PUBLIC FUNCTION DEFINITIONS * ordered as in iflib.h * **********************************************************************/ static void _iflib_assert(if_shared_ctx_t sctx) { int i; MPASS(sctx->isc_tx_maxsize); MPASS(sctx->isc_tx_maxsegsize); MPASS(sctx->isc_rx_maxsize); MPASS(sctx->isc_rx_nsegments); MPASS(sctx->isc_rx_maxsegsize); MPASS(sctx->isc_nrxqs >= 1 && sctx->isc_nrxqs <= 8); for (i = 0; i < sctx->isc_nrxqs; i++) { MPASS(sctx->isc_nrxd_min[i]); MPASS(powerof2(sctx->isc_nrxd_min[i])); MPASS(sctx->isc_nrxd_max[i]); MPASS(powerof2(sctx->isc_nrxd_max[i])); MPASS(sctx->isc_nrxd_default[i]); MPASS(powerof2(sctx->isc_nrxd_default[i])); } MPASS(sctx->isc_ntxqs >= 1 && sctx->isc_ntxqs <= 8); for (i = 0; i < sctx->isc_ntxqs; i++) { MPASS(sctx->isc_ntxd_min[i]); MPASS(powerof2(sctx->isc_ntxd_min[i])); MPASS(sctx->isc_ntxd_max[i]); MPASS(powerof2(sctx->isc_ntxd_max[i])); MPASS(sctx->isc_ntxd_default[i]); MPASS(powerof2(sctx->isc_ntxd_default[i])); } } static void _iflib_pre_assert(if_softc_ctx_t scctx) { MPASS(scctx->isc_txrx->ift_txd_encap); MPASS(scctx->isc_txrx->ift_txd_flush); MPASS(scctx->isc_txrx->ift_txd_credits_update); MPASS(scctx->isc_txrx->ift_rxd_available); MPASS(scctx->isc_txrx->ift_rxd_pkt_get); MPASS(scctx->isc_txrx->ift_rxd_refill); MPASS(scctx->isc_txrx->ift_rxd_flush); } static int iflib_register(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; driver_t *driver = sctx->isc_driver; device_t dev = ctx->ifc_dev; if_t ifp; u_char type; int iflags; if ((sctx->isc_flags & IFLIB_PSEUDO) == 0) _iflib_assert(sctx); CTX_LOCK_INIT(ctx); STATE_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev)); if (sctx->isc_flags & IFLIB_PSEUDO) { if (sctx->isc_flags & IFLIB_PSEUDO_ETHER) type = IFT_ETHER; else type = IFT_PPP; } else type = IFT_ETHER; ifp = ctx->ifc_ifp = if_alloc(type); if (ifp == NULL) { device_printf(dev, "can not allocate ifnet structure\n"); return (ENOMEM); } /* * Initialize our context's device specific methods */ kobj_init((kobj_t) ctx, (kobj_class_t) driver); kobj_class_compile((kobj_class_t) driver); if_initname(ifp, device_get_name(dev), device_get_unit(dev)); if_setsoftc(ifp, ctx); if_setdev(ifp, dev); if_setinitfn(ifp, iflib_if_init); if_setioctlfn(ifp, iflib_if_ioctl); #ifdef ALTQ if_setstartfn(ifp, iflib_altq_if_start); if_settransmitfn(ifp, iflib_altq_if_transmit); if_setsendqready(ifp); #else if_settransmitfn(ifp, iflib_if_transmit); #endif if_setqflushfn(ifp, iflib_if_qflush); iflags = IFF_MULTICAST | IFF_KNOWSEPOCH; if ((sctx->isc_flags & IFLIB_PSEUDO) && (sctx->isc_flags & IFLIB_PSEUDO_ETHER) == 0) iflags |= IFF_POINTOPOINT; else iflags |= IFF_BROADCAST | IFF_SIMPLEX; if_setflags(ifp, iflags); ctx->ifc_vlan_attach_event = EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx, EVENTHANDLER_PRI_FIRST); ctx->ifc_vlan_detach_event = EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx, EVENTHANDLER_PRI_FIRST); if ((sctx->isc_flags & IFLIB_DRIVER_MEDIA) == 0) { ctx->ifc_mediap = &ctx->ifc_media; ifmedia_init(ctx->ifc_mediap, IFM_IMASK, iflib_media_change, iflib_media_status); } return (0); } static void iflib_unregister_vlan_handlers(if_ctx_t ctx) { /* Unregister VLAN events */ if (ctx->ifc_vlan_attach_event != NULL) { EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event); ctx->ifc_vlan_attach_event = NULL; } if (ctx->ifc_vlan_detach_event != NULL) { EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event); ctx->ifc_vlan_detach_event = NULL; } } static void iflib_deregister(if_ctx_t ctx) { if_t ifp = ctx->ifc_ifp; /* Remove all media */ ifmedia_removeall(&ctx->ifc_media); /* Ensure that VLAN event handlers are unregistered */ iflib_unregister_vlan_handlers(ctx); /* Release kobject reference */ kobj_delete((kobj_t) ctx, NULL); /* Free the ifnet structure */ if_free(ifp); STATE_LOCK_DESTROY(ctx); /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ CTX_LOCK_DESTROY(ctx); } static int iflib_queues_alloc(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; int nrxqsets = scctx->isc_nrxqsets; int ntxqsets = scctx->isc_ntxqsets; iflib_txq_t txq; iflib_rxq_t rxq; iflib_fl_t fl = NULL; int i, j, cpu, err, txconf, rxconf; iflib_dma_info_t ifdip; uint32_t *rxqsizes = scctx->isc_rxqsizes; uint32_t *txqsizes = scctx->isc_txqsizes; uint8_t nrxqs = sctx->isc_nrxqs; uint8_t ntxqs = sctx->isc_ntxqs; int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1; caddr_t *vaddrs; uint64_t *paddrs; KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1")); KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1")); /* Allocate the TX ring struct memory */ if (!(ctx->ifc_txqs = (iflib_txq_t) malloc(sizeof(struct iflib_txq) * ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate TX ring memory\n"); err = ENOMEM; goto fail; } /* Now allocate the RX */ if (!(ctx->ifc_rxqs = (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) * nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX ring memory\n"); err = ENOMEM; goto rx_fail; } txq = ctx->ifc_txqs; rxq = ctx->ifc_rxqs; /* * XXX handle allocation failure */ for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) { /* Set up some basics */ if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { device_printf(dev, "Unable to allocate TX DMA info memory\n"); err = ENOMEM; goto err_tx_desc; } txq->ift_ifdi = ifdip; for (j = 0; j < ntxqs; j++, ifdip++) { if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, 0)) { device_printf(dev, "Unable to allocate TX descriptors\n"); err = ENOMEM; goto err_tx_desc; } txq->ift_txd_size[j] = scctx->isc_txd_size[j]; bzero((void *)ifdip->idi_vaddr, txqsizes[j]); } txq->ift_ctx = ctx; txq->ift_id = i; if (sctx->isc_flags & IFLIB_HAS_TXCQ) { txq->ift_br_offset = 1; } else { txq->ift_br_offset = 0; } /* XXX fix this */ txq->ift_timer.c_cpu = cpu; if (iflib_txsd_alloc(txq)) { device_printf(dev, "Critical Failure setting up TX buffers\n"); err = ENOMEM; goto err_tx_desc; } /* Initialize the TX lock */ snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:TX(%d):callout", device_get_nameunit(dev), txq->ift_id); mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF); callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0); err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain, iflib_txq_can_drain, M_IFLIB, M_WAITOK); if (err) { /* XXX free any allocated rings */ device_printf(dev, "Unable to allocate buf_ring\n"); goto err_tx_desc; } } for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) { /* Set up some basics */ callout_init(&rxq->ifr_watchdog, 1); if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) { device_printf(dev, "Unable to allocate RX DMA info memory\n"); err = ENOMEM; goto err_tx_desc; } rxq->ifr_ifdi = ifdip; /* XXX this needs to be changed if #rx queues != #tx queues */ rxq->ifr_ntxqirq = 1; rxq->ifr_txqid[0] = i; for (j = 0; j < nrxqs; j++, ifdip++) { if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, 0)) { device_printf(dev, "Unable to allocate RX descriptors\n"); err = ENOMEM; goto err_tx_desc; } bzero((void *)ifdip->idi_vaddr, rxqsizes[j]); } rxq->ifr_ctx = ctx; rxq->ifr_id = i; if (sctx->isc_flags & IFLIB_HAS_RXCQ) { rxq->ifr_fl_offset = 1; } else { rxq->ifr_fl_offset = 0; } rxq->ifr_nfl = nfree_lists; if (!(fl = (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate free list memory\n"); err = ENOMEM; goto err_tx_desc; } rxq->ifr_fl = fl; for (j = 0; j < nfree_lists; j++) { fl[j].ifl_rxq = rxq; fl[j].ifl_id = j; fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset]; fl[j].ifl_rxd_size = scctx->isc_rxd_size[j]; } /* Allocate receive buffers for the ring */ if (iflib_rxsd_alloc(rxq)) { device_printf(dev, "Critical Failure setting up receive buffers\n"); err = ENOMEM; goto err_rx_desc; } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB, M_WAITOK); } /* TXQs */ vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); for (i = 0; i < ntxqsets; i++) { iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi; for (j = 0; j < ntxqs; j++, di++) { vaddrs[i*ntxqs + j] = di->idi_vaddr; paddrs[i*ntxqs + j] = di->idi_paddr; } } if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) { device_printf(ctx->ifc_dev, "Unable to allocate device TX queue\n"); iflib_tx_structures_free(ctx); free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); goto err_rx_desc; } free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); /* RXQs */ vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); for (i = 0; i < nrxqsets; i++) { iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi; for (j = 0; j < nrxqs; j++, di++) { vaddrs[i*nrxqs + j] = di->idi_vaddr; paddrs[i*nrxqs + j] = di->idi_paddr; } } if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) { device_printf(ctx->ifc_dev, "Unable to allocate device RX queue\n"); iflib_tx_structures_free(ctx); free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); goto err_rx_desc; } free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); return (0); /* XXX handle allocation failure changes */ err_rx_desc: err_tx_desc: rx_fail: if (ctx->ifc_rxqs != NULL) free(ctx->ifc_rxqs, M_IFLIB); ctx->ifc_rxqs = NULL; if (ctx->ifc_txqs != NULL) free(ctx->ifc_txqs, M_IFLIB); ctx->ifc_txqs = NULL; fail: return (err); } static int iflib_tx_structures_setup(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; int i; for (i = 0; i < NTXQSETS(ctx); i++, txq++) iflib_txq_setup(txq); return (0); } static void iflib_tx_structures_free(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; if_shared_ctx_t sctx = ctx->ifc_sctx; int i, j; for (i = 0; i < NTXQSETS(ctx); i++, txq++) { for (j = 0; j < sctx->isc_ntxqs; j++) iflib_dma_free(&txq->ift_ifdi[j]); iflib_txq_destroy(txq); } free(ctx->ifc_txqs, M_IFLIB); ctx->ifc_txqs = NULL; IFDI_QUEUES_FREE(ctx); } /********************************************************************* * * Initialize all receive rings. * **********************************************************************/ static int iflib_rx_structures_setup(if_ctx_t ctx) { iflib_rxq_t rxq = ctx->ifc_rxqs; int q; #if defined(INET6) || defined(INET) int err, i; #endif for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) { #if defined(INET6) || defined(INET) if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) { err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp, TCP_LRO_ENTRIES, min(1024, ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset])); if (err != 0) { device_printf(ctx->ifc_dev, "LRO Initialization failed!\n"); goto fail; } } #endif IFDI_RXQ_SETUP(ctx, rxq->ifr_id); } return (0); #if defined(INET6) || defined(INET) fail: /* * Free LRO resources allocated so far, we will only handle * the rings that completed, the failing case will have * cleaned up for itself. 'q' failed, so its the terminus. */ rxq = ctx->ifc_rxqs; for (i = 0; i < q; ++i, rxq++) { if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) tcp_lro_free(&rxq->ifr_lc); } return (err); #endif } /********************************************************************* * * Free all receive rings. * **********************************************************************/ static void iflib_rx_structures_free(if_ctx_t ctx) { iflib_rxq_t rxq = ctx->ifc_rxqs; if_shared_ctx_t sctx = ctx->ifc_sctx; int i, j; for (i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) { for (j = 0; j < sctx->isc_nrxqs; j++) iflib_dma_free(&rxq->ifr_ifdi[j]); iflib_rx_sds_free(rxq); #if defined(INET6) || defined(INET) if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) tcp_lro_free(&rxq->ifr_lc); #endif } free(ctx->ifc_rxqs, M_IFLIB); ctx->ifc_rxqs = NULL; } static int iflib_qset_structures_setup(if_ctx_t ctx) { int err; /* * It is expected that the caller takes care of freeing queues if this * fails. */ if ((err = iflib_tx_structures_setup(ctx)) != 0) { device_printf(ctx->ifc_dev, "iflib_tx_structures_setup failed: %d\n", err); return (err); } if ((err = iflib_rx_structures_setup(ctx)) != 0) device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err); return (err); } int iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, const char *name) { return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name)); } #ifdef SMP static int find_nth(if_ctx_t ctx, int qid) { cpuset_t cpus; int i, cpuid, eqid, count; CPU_COPY(&ctx->ifc_cpus, &cpus); count = CPU_COUNT(&cpus); eqid = qid % count; /* clear up to the qid'th bit */ for (i = 0; i < eqid; i++) { cpuid = CPU_FFS(&cpus); MPASS(cpuid != 0); CPU_CLR(cpuid-1, &cpus); } cpuid = CPU_FFS(&cpus); MPASS(cpuid != 0); return (cpuid-1); } #ifdef SCHED_ULE extern struct cpu_group *cpu_top; /* CPU topology */ static int find_child_with_core(int cpu, struct cpu_group *grp) { int i; if (grp->cg_children == 0) return -1; MPASS(grp->cg_child); for (i = 0; i < grp->cg_children; i++) { if (CPU_ISSET(cpu, &grp->cg_child[i].cg_mask)) return i; } return -1; } /* * Find the nth "close" core to the specified core * "close" is defined as the deepest level that shares * at least an L2 cache. With threads, this will be * threads on the same core. If the shared cache is L3 * or higher, simply returns the same core. */ static int find_close_core(int cpu, int core_offset) { struct cpu_group *grp; int i; int fcpu; cpuset_t cs; grp = cpu_top; if (grp == NULL) return cpu; i = 0; while ((i = find_child_with_core(cpu, grp)) != -1) { /* If the child only has one cpu, don't descend */ if (grp->cg_child[i].cg_count <= 1) break; grp = &grp->cg_child[i]; } /* If they don't share at least an L2 cache, use the same CPU */ if (grp->cg_level > CG_SHARE_L2 || grp->cg_level == CG_SHARE_NONE) return cpu; /* Now pick one */ CPU_COPY(&grp->cg_mask, &cs); /* Add the selected CPU offset to core offset. */ for (i = 0; (fcpu = CPU_FFS(&cs)) != 0; i++) { if (fcpu - 1 == cpu) break; CPU_CLR(fcpu - 1, &cs); } MPASS(fcpu); core_offset += i; CPU_COPY(&grp->cg_mask, &cs); for (i = core_offset % grp->cg_count; i > 0; i--) { MPASS(CPU_FFS(&cs)); CPU_CLR(CPU_FFS(&cs) - 1, &cs); } MPASS(CPU_FFS(&cs)); return CPU_FFS(&cs) - 1; } #else static int find_close_core(int cpu, int core_offset __unused) { return cpu; } #endif static int get_core_offset(if_ctx_t ctx, iflib_intr_type_t type, int qid) { switch (type) { case IFLIB_INTR_TX: /* TX queues get cores which share at least an L2 cache with the corresponding RX queue */ /* XXX handle multiple RX threads per core and more than two core per L2 group */ return qid / CPU_COUNT(&ctx->ifc_cpus) + 1; case IFLIB_INTR_RX: case IFLIB_INTR_RXTX: /* RX queues get the specified core */ return qid / CPU_COUNT(&ctx->ifc_cpus); default: return -1; } } #else #define get_core_offset(ctx, type, qid) CPU_FIRST() #define find_close_core(cpuid, tid) CPU_FIRST() #define find_nth(ctx, gid) CPU_FIRST() #endif /* Just to avoid copy/paste */ static inline int iflib_irq_set_affinity(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, int qid, struct grouptask *gtask, struct taskqgroup *tqg, void *uniq, const char *name) { device_t dev; int co, cpuid, err, tid; dev = ctx->ifc_dev; co = ctx->ifc_sysctl_core_offset; if (ctx->ifc_sysctl_separate_txrx && type == IFLIB_INTR_TX) co += ctx->ifc_softc_ctx.isc_nrxqsets; cpuid = find_nth(ctx, qid + co); tid = get_core_offset(ctx, type, qid); if (tid < 0) { device_printf(dev, "get_core_offset failed\n"); return (EOPNOTSUPP); } cpuid = find_close_core(cpuid, tid); err = taskqgroup_attach_cpu(tqg, gtask, uniq, cpuid, dev, irq->ii_res, name); if (err) { device_printf(dev, "taskqgroup_attach_cpu failed %d\n", err); return (err); } #ifdef notyet if (cpuid > ctx->ifc_cpuid_highest) ctx->ifc_cpuid_highest = cpuid; #endif return (0); } int iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid, iflib_intr_type_t type, driver_filter_t *filter, void *filter_arg, int qid, const char *name) { device_t dev; struct grouptask *gtask; struct taskqgroup *tqg; iflib_filter_info_t info; gtask_fn_t *fn; int tqrid, err; driver_filter_t *intr_fast; void *q; info = &ctx->ifc_filter_info; tqrid = rid; switch (type) { /* XXX merge tx/rx for netmap? */ case IFLIB_INTR_TX: q = &ctx->ifc_txqs[qid]; info = &ctx->ifc_txqs[qid].ift_filter_info; gtask = &ctx->ifc_txqs[qid].ift_task; tqg = qgroup_if_io_tqg; fn = _task_fn_tx; intr_fast = iflib_fast_intr; GROUPTASK_INIT(gtask, 0, fn, q); ctx->ifc_flags |= IFC_NETMAP_TX_IRQ; break; case IFLIB_INTR_RX: q = &ctx->ifc_rxqs[qid]; info = &ctx->ifc_rxqs[qid].ifr_filter_info; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; intr_fast = iflib_fast_intr; NET_GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_RXTX: q = &ctx->ifc_rxqs[qid]; info = &ctx->ifc_rxqs[qid].ifr_filter_info; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; intr_fast = iflib_fast_intr_rxtx; NET_GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_ADMIN: q = ctx; tqrid = -1; info = &ctx->ifc_filter_info; gtask = &ctx->ifc_admin_task; tqg = qgroup_if_config_tqg; fn = _task_fn_admin; intr_fast = iflib_fast_intr_ctx; break; default: device_printf(ctx->ifc_dev, "%s: unknown net intr type\n", __func__); return (EINVAL); } info->ifi_filter = filter; info->ifi_filter_arg = filter_arg; info->ifi_task = gtask; info->ifi_ctx = q; dev = ctx->ifc_dev; err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info, name); if (err != 0) { device_printf(dev, "_iflib_irq_alloc failed %d\n", err); return (err); } if (type == IFLIB_INTR_ADMIN) return (0); if (tqrid != -1) { err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg, q, name); if (err) return (err); } else { taskqgroup_attach(tqg, gtask, q, dev, irq->ii_res, name); } return (0); } void iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, void *arg, int qid, const char *name) { struct grouptask *gtask; struct taskqgroup *tqg; gtask_fn_t *fn; void *q; int err; switch (type) { case IFLIB_INTR_TX: q = &ctx->ifc_txqs[qid]; gtask = &ctx->ifc_txqs[qid].ift_task; tqg = qgroup_if_io_tqg; fn = _task_fn_tx; GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_RX: q = &ctx->ifc_rxqs[qid]; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; NET_GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_IOV: q = ctx; gtask = &ctx->ifc_vflr_task; tqg = qgroup_if_config_tqg; fn = _task_fn_iov; GROUPTASK_INIT(gtask, 0, fn, q); break; default: panic("unknown net intr type"); } if (irq != NULL) { err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg, q, name); if (err) taskqgroup_attach(tqg, gtask, q, ctx->ifc_dev, irq->ii_res, name); } else { taskqgroup_attach(tqg, gtask, q, NULL, NULL, name); } } void iflib_irq_free(if_ctx_t ctx, if_irq_t irq) { if (irq->ii_tag) bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag); if (irq->ii_res) bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, rman_get_rid(irq->ii_res), irq->ii_res); } static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, const char *name) { iflib_txq_t txq = ctx->ifc_txqs; iflib_rxq_t rxq = ctx->ifc_rxqs; if_irq_t irq = &ctx->ifc_legacy_irq; iflib_filter_info_t info; device_t dev; struct grouptask *gtask; struct resource *res; struct taskqgroup *tqg; void *q; int err, tqrid; bool rx_only; q = &ctx->ifc_rxqs[0]; info = &rxq[0].ifr_filter_info; gtask = &rxq[0].ifr_task; tqg = qgroup_if_io_tqg; tqrid = *rid; rx_only = (ctx->ifc_sctx->isc_flags & IFLIB_SINGLE_IRQ_RX_ONLY) != 0; ctx->ifc_flags |= IFC_LEGACY; info->ifi_filter = filter; info->ifi_filter_arg = filter_arg; info->ifi_task = gtask; info->ifi_ctx = rx_only ? ctx : q; dev = ctx->ifc_dev; /* We allocate a single interrupt resource */ err = _iflib_irq_alloc(ctx, irq, tqrid, rx_only ? iflib_fast_intr_ctx : iflib_fast_intr_rxtx, NULL, info, name); if (err != 0) return (err); NET_GROUPTASK_INIT(gtask, 0, _task_fn_rx, q); res = irq->ii_res; taskqgroup_attach(tqg, gtask, q, dev, res, name); GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq); taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, dev, res, "tx"); return (0); } void iflib_led_create(if_ctx_t ctx) { ctx->ifc_led_dev = led_create(iflib_led_func, ctx, device_get_nameunit(ctx->ifc_dev)); } void iflib_tx_intr_deferred(if_ctx_t ctx, int txqid) { GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task); } void iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid) { GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task); } void iflib_admin_intr_deferred(if_ctx_t ctx) { MPASS(ctx->ifc_admin_task.gt_taskqueue != NULL); GROUPTASK_ENQUEUE(&ctx->ifc_admin_task); } void iflib_iov_intr_deferred(if_ctx_t ctx) { GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task); } void iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, const char *name) { taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, NULL, NULL, name); } void iflib_config_gtask_init(void *ctx, struct grouptask *gtask, gtask_fn_t *fn, const char *name) { GROUPTASK_INIT(gtask, 0, fn, ctx); taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, NULL, NULL, name); } void iflib_config_gtask_deinit(struct grouptask *gtask) { taskqgroup_detach(qgroup_if_config_tqg, gtask); } void iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate) { if_t ifp = ctx->ifc_ifp; iflib_txq_t txq = ctx->ifc_txqs; if_setbaudrate(ifp, baudrate); if (baudrate >= IF_Gbps(10)) { STATE_LOCK(ctx); ctx->ifc_flags |= IFC_PREFETCH; STATE_UNLOCK(ctx); } /* If link down, disable watchdog */ if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) { for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++) txq->ift_qstatus = IFLIB_QUEUE_IDLE; } ctx->ifc_link_state = link_state; if_link_state_change(ifp, link_state); } static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq) { int credits; #ifdef INVARIANTS int credits_pre = txq->ift_cidx_processed; #endif bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD); if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0) return (0); txq->ift_processed += credits; txq->ift_cidx_processed += credits; MPASS(credits_pre + credits == txq->ift_cidx_processed); if (txq->ift_cidx_processed >= txq->ift_size) txq->ift_cidx_processed -= txq->ift_size; return (credits); } static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget) { iflib_fl_t fl; u_int i; for (i = 0, fl = &rxq->ifr_fl[0]; i < rxq->ifr_nfl; i++, fl++) bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx, budget)); } void iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name, const char *description, if_int_delay_info_t info, int offset, int value) { info->iidi_ctx = ctx; info->iidi_offset = offset; info->iidi_value = value; SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev), SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)), OID_AUTO, name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, info, 0, iflib_sysctl_int_delay, "I", description); } struct sx * iflib_ctx_lock_get(if_ctx_t ctx) { return (&ctx->ifc_ctx_sx); } static int iflib_msix_init(if_ctx_t ctx) { device_t dev = ctx->ifc_dev; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; int admincnt, bar, err, iflib_num_rx_queues, iflib_num_tx_queues; int msgs, queuemsgs, queues, rx_queues, tx_queues, vectors; iflib_num_tx_queues = ctx->ifc_sysctl_ntxqs; iflib_num_rx_queues = ctx->ifc_sysctl_nrxqs; if (bootverbose) device_printf(dev, "msix_init qsets capped at %d\n", imax(scctx->isc_ntxqsets, scctx->isc_nrxqsets)); /* Override by tuneable */ if (scctx->isc_disable_msix) goto msi; /* First try MSI-X */ if ((msgs = pci_msix_count(dev)) == 0) { if (bootverbose) device_printf(dev, "MSI-X not supported or disabled\n"); goto msi; } bar = ctx->ifc_softc_ctx.isc_msix_bar; /* * bar == -1 => "trust me I know what I'm doing" * Some drivers are for hardware that is so shoddily * documented that no one knows which bars are which * so the developer has to map all bars. This hack * allows shoddy garbage to use MSI-X in this framework. */ if (bar != -1) { ctx->ifc_msix_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &bar, RF_ACTIVE); if (ctx->ifc_msix_mem == NULL) { device_printf(dev, "Unable to map MSI-X table\n"); goto msi; } } admincnt = sctx->isc_admin_intrcnt; #if IFLIB_DEBUG /* use only 1 qset in debug mode */ queuemsgs = min(msgs - admincnt, 1); #else queuemsgs = msgs - admincnt; #endif #ifdef RSS queues = imin(queuemsgs, rss_getnumbuckets()); #else queues = queuemsgs; #endif queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues); if (bootverbose) device_printf(dev, "intr CPUs: %d queue msgs: %d admincnt: %d\n", CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt); #ifdef RSS /* If we're doing RSS, clamp at the number of RSS buckets */ if (queues > rss_getnumbuckets()) queues = rss_getnumbuckets(); #endif if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt) rx_queues = iflib_num_rx_queues; else rx_queues = queues; if (rx_queues > scctx->isc_nrxqsets) rx_queues = scctx->isc_nrxqsets; /* * We want this to be all logical CPUs by default */ if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues) tx_queues = iflib_num_tx_queues; else tx_queues = mp_ncpus; if (tx_queues > scctx->isc_ntxqsets) tx_queues = scctx->isc_ntxqsets; if (ctx->ifc_sysctl_qs_eq_override == 0) { #ifdef INVARIANTS if (tx_queues != rx_queues) device_printf(dev, "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n", min(rx_queues, tx_queues), min(rx_queues, tx_queues)); #endif tx_queues = min(rx_queues, tx_queues); rx_queues = min(rx_queues, tx_queues); } vectors = rx_queues + admincnt; if (msgs < vectors) { device_printf(dev, "insufficient number of MSI-X vectors " "(supported %d, need %d)\n", msgs, vectors); goto msi; } device_printf(dev, "Using %d RX queues %d TX queues\n", rx_queues, tx_queues); msgs = vectors; if ((err = pci_alloc_msix(dev, &vectors)) == 0) { if (vectors != msgs) { device_printf(dev, "Unable to allocate sufficient MSI-X vectors " "(got %d, need %d)\n", vectors, msgs); pci_release_msi(dev); if (bar != -1) { bus_release_resource(dev, SYS_RES_MEMORY, bar, ctx->ifc_msix_mem); ctx->ifc_msix_mem = NULL; } goto msi; } device_printf(dev, "Using MSI-X interrupts with %d vectors\n", vectors); scctx->isc_vectors = vectors; scctx->isc_nrxqsets = rx_queues; scctx->isc_ntxqsets = tx_queues; scctx->isc_intr = IFLIB_INTR_MSIX; return (vectors); } else { device_printf(dev, "failed to allocate %d MSI-X vectors, err: %d\n", vectors, err); if (bar != -1) { bus_release_resource(dev, SYS_RES_MEMORY, bar, ctx->ifc_msix_mem); ctx->ifc_msix_mem = NULL; } } msi: vectors = pci_msi_count(dev); scctx->isc_nrxqsets = 1; scctx->isc_ntxqsets = 1; scctx->isc_vectors = vectors; if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) { device_printf(dev,"Using an MSI interrupt\n"); scctx->isc_intr = IFLIB_INTR_MSI; } else { scctx->isc_vectors = 1; device_printf(dev,"Using a Legacy interrupt\n"); scctx->isc_intr = IFLIB_INTR_LEGACY; } return (vectors); } static const char *ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" }; static int mp_ring_state_handler(SYSCTL_HANDLER_ARGS) { int rc; uint16_t *state = ((uint16_t *)oidp->oid_arg1); struct sbuf *sb; const char *ring_state = "UNKNOWN"; /* XXX needed ? */ rc = sysctl_wire_old_buffer(req, 0); MPASS(rc == 0); if (rc != 0) return (rc); sb = sbuf_new_for_sysctl(NULL, NULL, 80, req); MPASS(sb != NULL); if (sb == NULL) return (ENOMEM); if (state[3] <= 3) ring_state = ring_states[state[3]]; sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s", state[0], state[1], state[2], ring_state); rc = sbuf_finish(sb); sbuf_delete(sb); return(rc); } enum iflib_ndesc_handler { IFLIB_NTXD_HANDLER, IFLIB_NRXD_HANDLER, }; static int mp_ndesc_handler(SYSCTL_HANDLER_ARGS) { if_ctx_t ctx = (void *)arg1; enum iflib_ndesc_handler type = arg2; char buf[256] = {0}; qidx_t *ndesc; char *p, *next; int nqs, rc, i; nqs = 8; switch(type) { case IFLIB_NTXD_HANDLER: ndesc = ctx->ifc_sysctl_ntxds; if (ctx->ifc_sctx) nqs = ctx->ifc_sctx->isc_ntxqs; break; case IFLIB_NRXD_HANDLER: ndesc = ctx->ifc_sysctl_nrxds; if (ctx->ifc_sctx) nqs = ctx->ifc_sctx->isc_nrxqs; break; default: printf("%s: unhandled type\n", __func__); return (EINVAL); } if (nqs == 0) nqs = 8; for (i=0; i<8; i++) { if (i >= nqs) break; if (i) strcat(buf, ","); sprintf(strchr(buf, 0), "%d", ndesc[i]); } rc = sysctl_handle_string(oidp, buf, sizeof(buf), req); if (rc || req->newptr == NULL) return rc; for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p; i++, p = strsep(&next, " ,")) { ndesc[i] = strtoul(p, NULL, 10); } return(rc); } #define NAME_BUFLEN 32 static void iflib_add_device_sysctl_pre(if_ctx_t ctx) { device_t dev = iflib_get_dev(ctx); struct sysctl_oid_list *child, *oid_list; struct sysctl_ctx_list *ctx_list; struct sysctl_oid *node; ctx_list = device_get_sysctl_ctx(dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "IFLIB fields"); oid_list = SYSCTL_CHILDREN(node); SYSCTL_ADD_CONST_STRING(ctx_list, oid_list, OID_AUTO, "driver_version", CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, "driver version"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs", CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0, "# of txqs to use, 0 => use default #"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs", CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0, "# of rxqs to use, 0 => use default #"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable", CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0, "permit #txq != #rxq"); SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix", CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0, "disable MSI-X (default 0)"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget", CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0, "set the RX budget"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "tx_abdicate", CTLFLAG_RWTUN, &ctx->ifc_sysctl_tx_abdicate, 0, "cause TX to abdicate instead of running to completion"); ctx->ifc_sysctl_core_offset = CORE_OFFSET_UNSPECIFIED; SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "core_offset", CTLFLAG_RDTUN, &ctx->ifc_sysctl_core_offset, 0, "offset to start using cores at"); SYSCTL_ADD_U8(ctx_list, oid_list, OID_AUTO, "separate_txrx", CTLFLAG_RDTUN, &ctx->ifc_sysctl_separate_txrx, 0, "use separate cores for TX and RX"); /* XXX change for per-queue sizes */ SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds", CTLTYPE_STRING | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, ctx, IFLIB_NTXD_HANDLER, mp_ndesc_handler, "A", "list of # of TX descriptors to use, 0 = use default #"); SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds", CTLTYPE_STRING | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, ctx, IFLIB_NRXD_HANDLER, mp_ndesc_handler, "A", "list of # of RX descriptors to use, 0 = use default #"); } static void iflib_add_device_sysctl_post(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = iflib_get_dev(ctx); struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx_list; iflib_fl_t fl; iflib_txq_t txq; iflib_rxq_t rxq; int i, j; char namebuf[NAME_BUFLEN]; char *qfmt; struct sysctl_oid *queue_node, *fl_node, *node; struct sysctl_oid_list *queue_list, *fl_list; ctx_list = device_get_sysctl_ctx(dev); node = ctx->ifc_sysctl_node; child = SYSCTL_CHILDREN(node); if (scctx->isc_ntxqsets > 100) qfmt = "txq%03d"; else if (scctx->isc_ntxqsets > 10) qfmt = "txq%02d"; else qfmt = "txq%d"; for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) { snprintf(namebuf, NAME_BUFLEN, qfmt, i); queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); #if MEMORY_LOGGING SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued", CTLFLAG_RD, &txq->ift_dequeued, "total mbufs freed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued", CTLFLAG_RD, &txq->ift_enqueued, "total mbufs enqueued"); #endif SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag", CTLFLAG_RD, &txq->ift_mbuf_defrag, "# of times m_defrag was called"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups", CTLFLAG_RD, &txq->ift_pullups, "# of times m_pullup was called"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed", CTLFLAG_RD, &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail", CTLFLAG_RD, &txq->ift_no_desc_avail, "# of times no descriptors were available"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed", CTLFLAG_RD, &txq->ift_map_failed, "# of times DMA map failed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig", CTLFLAG_RD, &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup", CTLFLAG_RD, &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx", CTLFLAG_RD, &txq->ift_pidx, 1, "Producer Index"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx", CTLFLAG_RD, &txq->ift_cidx, 1, "Consumer Index"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed", CTLFLAG_RD, &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use", CTLFLAG_RD, &txq->ift_in_use, 1, "descriptors in use"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed", CTLFLAG_RD, &txq->ift_processed, "descriptors procesed for clean"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned", CTLFLAG_RD, &txq->ift_cleaned, "total cleaned"); SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, __DEVOLATILE(uint64_t *, &txq->ift_br->state), 0, mp_ring_state_handler, "A", "soft ring state"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues", CTLFLAG_RD, &txq->ift_br->enqueues, "# of enqueues to the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops", CTLFLAG_RD, &txq->ift_br->drops, "# of drops in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts", CTLFLAG_RD, &txq->ift_br->starts, "# of normal consumer starts in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls", CTLFLAG_RD, &txq->ift_br->stalls, "# of consumer stalls in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts", CTLFLAG_RD, &txq->ift_br->restarts, "# of consumer restarts in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications", CTLFLAG_RD, &txq->ift_br->abdications, "# of consumer abdications in the mp_ring for this queue"); } if (scctx->isc_nrxqsets > 100) qfmt = "rxq%03d"; else if (scctx->isc_nrxqsets > 10) qfmt = "rxq%02d"; else qfmt = "rxq%d"; for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) { snprintf(namebuf, NAME_BUFLEN, qfmt, i); queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); if (sctx->isc_flags & IFLIB_HAS_RXCQ) { SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx", CTLFLAG_RD, &rxq->ifr_cq_cidx, 1, "Consumer Index"); } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j); fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "freelist Name"); fl_list = SYSCTL_CHILDREN(fl_node); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx", CTLFLAG_RD, &fl->ifl_pidx, 1, "Producer Index"); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx", CTLFLAG_RD, &fl->ifl_cidx, 1, "Consumer Index"); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits", CTLFLAG_RD, &fl->ifl_credits, 1, "credits available"); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "buf_size", CTLFLAG_RD, &fl->ifl_buf_size, 1, "buffer size"); #if MEMORY_LOGGING SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued", CTLFLAG_RD, &fl->ifl_m_enqueued, "mbufs allocated"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued", CTLFLAG_RD, &fl->ifl_m_dequeued, "mbufs freed"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued", CTLFLAG_RD, &fl->ifl_cl_enqueued, "clusters allocated"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued", CTLFLAG_RD, &fl->ifl_cl_dequeued, "clusters freed"); #endif } } } void iflib_request_reset(if_ctx_t ctx) { STATE_LOCK(ctx); ctx->ifc_flags |= IFC_DO_RESET; STATE_UNLOCK(ctx); } #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * iflib_fixup_rx(struct mbuf *m) { struct mbuf *n; if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) { bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len); m->m_data += ETHER_HDR_LEN; n = m; } else { MGETHDR(n, M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return (NULL); } bcopy(m->m_data, n->m_data, ETHER_HDR_LEN); m->m_data += ETHER_HDR_LEN; m->m_len -= ETHER_HDR_LEN; n->m_len = ETHER_HDR_LEN; M_MOVE_PKTHDR(n, m); n->m_next = m; } return (n); } #endif #ifdef DEBUGNET static void iflib_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize) { if_ctx_t ctx; ctx = if_getsoftc(ifp); CTX_LOCK(ctx); *nrxr = NRXQSETS(ctx); *ncl = ctx->ifc_rxqs[0].ifr_fl->ifl_size; *clsize = ctx->ifc_rxqs[0].ifr_fl->ifl_buf_size; CTX_UNLOCK(ctx); } static void iflib_debugnet_event(if_t ifp, enum debugnet_ev event) { if_ctx_t ctx; if_softc_ctx_t scctx; iflib_fl_t fl; iflib_rxq_t rxq; int i, j; ctx = if_getsoftc(ifp); scctx = &ctx->ifc_softc_ctx; switch (event) { case DEBUGNET_START: for (i = 0; i < scctx->isc_nrxqsets; i++) { rxq = &ctx->ifc_rxqs[i]; for (j = 0; j < rxq->ifr_nfl; j++) { fl = rxq->ifr_fl; fl->ifl_zone = m_getzone(fl->ifl_buf_size); } } iflib_no_tx_batch = 1; break; default: break; } } static int iflib_debugnet_transmit(if_t ifp, struct mbuf *m) { if_ctx_t ctx; iflib_txq_t txq; int error; ctx = if_getsoftc(ifp); if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return (EBUSY); txq = &ctx->ifc_txqs[0]; error = iflib_encap(txq, &m); if (error == 0) (void)iflib_txd_db_check(ctx, txq, true, txq->ift_in_use); return (error); } static int iflib_debugnet_poll(if_t ifp, int count) { struct epoch_tracker et; if_ctx_t ctx; if_softc_ctx_t scctx; iflib_txq_t txq; int i; ctx = if_getsoftc(ifp); scctx = &ctx->ifc_softc_ctx; if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return (EBUSY); txq = &ctx->ifc_txqs[0]; (void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); NET_EPOCH_ENTER(et); for (i = 0; i < scctx->isc_nrxqsets; i++) (void)iflib_rxeof(&ctx->ifc_rxqs[i], 16 /* XXX */); NET_EPOCH_EXIT(et); return (0); } #endif /* DEBUGNET */