Date: Thu, 31 Mar 2011 22:40:44 +0000 (UTC) From: David Christensen <davidch@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r220229 - head/sys/dev/bxe Message-ID: <201103312240.p2VMeiEb024871@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: davidch Date: Thu Mar 31 22:40:44 2011 New Revision: 220229 URL: http://svn.freebsd.org/changeset/base/220229 Log: - Freshened debug support code. - Renamed several RX variable for more consistent usage. - Fixed a potential problem when masking RX CQ producer value. MFC after: One week. Modified: head/sys/dev/bxe/if_bxe.c head/sys/dev/bxe/if_bxe.h Modified: head/sys/dev/bxe/if_bxe.c ============================================================================== --- head/sys/dev/bxe/if_bxe.c Thu Mar 31 22:04:00 2011 (r220228) +++ head/sys/dev/bxe/if_bxe.c Thu Mar 31 22:40:44 2011 (r220229) @@ -68,7 +68,8 @@ __FBSDID("$FreeBSD$"); /* BXE Debug Options */ #ifdef BXE_DEBUG - uint32_t bxe_debug = BXE_INFO; +uint32_t bxe_debug = BXE_INFO; + /* 0 = Never */ /* 1 = 1 in 2,147,483,648 */ @@ -345,8 +346,8 @@ int bxe_set_gpio_int(struct bxe_softc * static int bxe_sysctl_driver_state(SYSCTL_HANDLER_ARGS); static int bxe_sysctl_hw_state(SYSCTL_HANDLER_ARGS); static int bxe_sysctl_dump_fw(SYSCTL_HANDLER_ARGS); -static int bxe_sysctl_dump_cqe_chain(SYSCTL_HANDLER_ARGS); -static int bxe_sysctl_dump_rx_chain(SYSCTL_HANDLER_ARGS); +static int bxe_sysctl_dump_rx_cq_chain(SYSCTL_HANDLER_ARGS); +static int bxe_sysctl_dump_rx_bd_chain(SYSCTL_HANDLER_ARGS); static int bxe_sysctl_dump_tx_chain(SYSCTL_HANDLER_ARGS); static int bxe_sysctl_reg_read(SYSCTL_HANDLER_ARGS); static int bxe_sysctl_breakpoint(SYSCTL_HANDLER_ARGS); @@ -357,16 +358,16 @@ static void bxe_dump_enet(struct bxe_sof static void bxe_dump_mbuf (struct bxe_softc *, struct mbuf *); static void bxe_dump_tx_mbuf_chain(struct bxe_softc *, int, int); static void bxe_dump_rx_mbuf_chain(struct bxe_softc *, int, int); -static void bxe_dump_pbd_locked(struct bxe_softc *,int, +static void bxe_dump_tx_parsing_bd(struct bxe_fastpath *,int, struct eth_tx_parse_bd *); -static void bxe_dump_txbd_locked(struct bxe_fastpath *, int, - struct eth_tx_bd *); -static void bxe_dump_rxbd_locked(struct bxe_fastpath *, int, +static void bxe_dump_txbd(struct bxe_fastpath *, int, + union eth_tx_bd_types *); +static void bxe_dump_rxbd(struct bxe_fastpath *, int, struct eth_rx_bd *); -static void bxe_dump_cqe_locked(struct bxe_fastpath *, int, union eth_rx_cqe *); +static void bxe_dump_cqe(struct bxe_fastpath *, int, union eth_rx_cqe *); static void bxe_dump_tx_chain(struct bxe_fastpath *, int, int); -static void bxe_dump_cqe_chain(struct bxe_fastpath *, int, int); -static void bxe_dump_rx_chain(struct bxe_fastpath *, int, int); +static void bxe_dump_rx_cq_chain(struct bxe_fastpath *, int, int); +static void bxe_dump_rx_bd_chain(struct bxe_fastpath *, int, int); static void bxe_dump_status_block(struct bxe_softc *); static void bxe_dump_stats_block(struct bxe_softc *); static void bxe_dump_fp_state(struct bxe_fastpath *); @@ -404,6 +405,7 @@ static device_method_t bxe_methods[] = { KOBJMETHOD_END }; + static driver_t bxe_driver = { "bxe", bxe_methods, @@ -1725,11 +1727,11 @@ bxe_attach(device_t dev) * processor memory. */ rid = PCIR_BAR(0); - sc->bxe_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, - &rid, RF_ACTIVE); + sc->bxe_res = bus_alloc_resource_any(dev, + SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->bxe_res == NULL) { BXE_PRINTF("%s(%d):PCI BAR0 memory allocation failed\n", - __FILE__, __LINE__); + __FILE__, __LINE__); rc = ENXIO; goto bxe_attach_fail; } @@ -1744,11 +1746,11 @@ bxe_attach(device_t dev) * Doorbell (DB) memory. */ rid = PCIR_BAR(2); - sc->bxe_db_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, - &rid, RF_ACTIVE); + sc->bxe_db_res = bus_alloc_resource_any(dev, + SYS_RES_MEMORY, &rid, RF_ACTIVE); if (sc->bxe_db_res == NULL) { BXE_PRINTF("%s(%d): PCI BAR2 memory allocation failed\n", - __FILE__, __LINE__); + __FILE__, __LINE__); rc = ENXIO; goto bxe_attach_fail; } @@ -1834,7 +1836,6 @@ bxe_attach(device_t dev) #ifdef BXE_DEBUG - /* Allocate a memory buffer for grcdump output.*/ sc->grcdump_buffer = malloc(BXE_GRCDUMP_BUF_SIZE, M_TEMP, M_NOWAIT); if (sc->grcdump_buffer == NULL) { @@ -3738,8 +3739,8 @@ bxe_init_locked(struct bxe_softc *sc, in callout_reset(&sc->bxe_tick_callout, hz, bxe_tick, sc); /* Everything went OK, go ahead and exit. */ goto bxe_init_locked_exit; + /* Try and gracefully shutdown the device because of a failure. */ -/* Try and gracefully shutdown the device because of a failure. */ bxe_init_locked_failed4: for (i = 1; i < sc->num_queues; i++) @@ -4413,6 +4414,7 @@ bxe_write_dmae(struct bxe_softc *sc, bus DELAY(50); + /* Wait up to 200ms. */ timeout = 4000; while (*wb_comp != BXE_WB_COMP_VAL) { if (!timeout) { @@ -4433,18 +4435,18 @@ bxe_write_dmae_exit: /* -* Perform a DMAE read from to device memory. -* + * Perform a DMAE read from to device memory. + * * Some of the registers on the 577XX controller are 128bits wide. It is * required that when accessing those registers that they be read * atomically and that no intervening bus acceses to the device occur. * This could be handled by a lock held across all driver instances for * the device or it can be handled by performing a DMA operation when * reading from the device. This code implements the latter. -* -* Returns: -* None. -*/ + * + * Returns: + * None. + */ void bxe_read_dmae(struct bxe_softc *sc, uint32_t src_addr, uint32_t len32) @@ -7244,7 +7246,8 @@ bxe_stats_init(struct bxe_softc *sc) sizeof(struct ustorm_per_client_stats)); memset(&fp->old_xclient, 0, sizeof(struct xstorm_per_client_stats)); - memset(&fp->eth_q_stats, 0, sizeof(struct bxe_q_stats)); + memset(&fp->eth_q_stats, 0, + sizeof(struct bxe_q_stats)); } sc->stats_state = STATS_STATE_DISABLED; @@ -8539,6 +8542,7 @@ bxe_chktso_window(struct bxe_softc* sc, return (defrag); } + /* * Encapsultes an mbuf cluster into the tx_bd chain structure and * makes the memory visible to the controller. @@ -8557,7 +8561,7 @@ bxe_tx_encap(struct bxe_fastpath *fp, st bus_dma_segment_t segs[32]; bus_dmamap_t map; struct mbuf *m0; - struct eth_tx_parse_bd *pbd; + struct eth_tx_parse_bd *tx_parse_bd; struct eth_tx_bd *tx_data_bd; struct eth_tx_bd *tx_total_pkt_size_bd; struct eth_tx_start_bd *tx_start_bd; @@ -8580,7 +8584,7 @@ bxe_tx_encap(struct bxe_fastpath *fp, st tx_total_pkt_size_bd = NULL; tx_start_bd = NULL; tx_data_bd = NULL; - pbd = NULL; + tx_parse_bd = NULL; pkt_prod = fp->tx_pkt_prod; bd_prod = TX_BD(fp->tx_bd_prod); @@ -8693,6 +8697,7 @@ bxe_tx_encap(struct bxe_fastpath *fp, st } break; } + /* Last try */ if (m0->m_pkthdr.csum_flags & CSUM_TSO){ if (bxe_chktso_window(sc,nsegs,segs,m0)) @@ -8765,11 +8770,11 @@ bxe_tx_encap(struct bxe_fastpath *fp, st * however, it is only used for tso & chksum. */ bd_prod = TX_BD(NEXT_TX_BD(bd_prod)); - pbd = (struct eth_tx_parse_bd *) + tx_parse_bd = (struct eth_tx_parse_bd *) &fp->tx_bd_chain[TX_PAGE(bd_prod)][TX_IDX(bd_prod)].parse_bd; - memset(pbd, 0, sizeof(struct eth_tx_parse_bd)); + memset(tx_parse_bd, 0, sizeof(struct eth_tx_parse_bd)); - /* Gather all info about the packet and add to pbd */ + /* Gather all info about the packet and add to tx_parse_bd */ if (m0->m_pkthdr.csum_flags) { struct ether_vlan_header *eh; struct ip *ip = NULL; @@ -8790,8 +8795,9 @@ bxe_tx_encap(struct bxe_fastpath *fp, st } /* Set the Ethernet header length in 16 bit words. */ - pbd->global_data = (e_hlen + ovlan) >> 1; - pbd->global_data |= ((m0->m_flags & M_VLANTAG) << ETH_TX_PARSE_BD_LLC_SNAP_EN_SHIFT); + tx_parse_bd->global_data = (e_hlen + ovlan) >> 1; + tx_parse_bd->global_data |= ((m0->m_flags & M_VLANTAG) << + ETH_TX_PARSE_BD_LLC_SNAP_EN_SHIFT); switch (etype) { case ETHERTYPE_IP:{ @@ -8802,10 +8808,10 @@ bxe_tx_encap(struct bxe_fastpath *fp, st ip = (struct ip *)(m0->m_data + e_hlen); /* Calculate IP header length (16 bit words). */ - pbd->ip_hlen = (ip->ip_hl << 1); + tx_parse_bd->ip_hlen = (ip->ip_hl << 1); /* Calculate enet + IP header length (16 bit words). */ - pbd->total_hlen = pbd->ip_hlen + (e_hlen >> 1); + tx_parse_bd->total_hlen = tx_parse_bd->ip_hlen + (e_hlen >> 1); if (m0->m_pkthdr.csum_flags & CSUM_IP) { DBPRINT(sc, BXE_EXTREME_SEND, "%s(): IP checksum " @@ -8816,7 +8822,7 @@ bxe_tx_encap(struct bxe_fastpath *fp, st /* Handle any checksums requested by the stack. */ if ((m0->m_pkthdr.csum_flags & CSUM_TCP)|| - (m0->m_pkthdr.csum_flags & CSUM_TSO)){ + (m0->m_pkthdr.csum_flags & CSUM_TSO)){ /* Perform TCP checksum offload. */ DBPRINT(sc, BXE_EXTREME_SEND, "%s(): TCP checksum " @@ -8830,10 +8836,10 @@ bxe_tx_encap(struct bxe_fastpath *fp, st DBRUN(sc->debug_tcp_csum_offload_frames++); /* Update the enet + IP + TCP header length. */ - pbd->total_hlen += (uint16_t)(th->th_off << 1); + tx_parse_bd->total_hlen += (uint16_t)(th->th_off << 1); /* Get the pseudo header checksum. */ - pbd->tcp_pseudo_csum = ntohs(th->th_sum); + tx_parse_bd->tcp_pseudo_csum = ntohs(th->th_sum); } else if (m0->m_pkthdr.csum_flags & CSUM_UDP) { /* * The hardware doesn't actually support UDP checksum @@ -8861,7 +8867,7 @@ bxe_tx_encap(struct bxe_fastpath *fp, st /* Add the TCP checksum offload flag for UDP frames too. */ flags |= ETH_TX_BD_FLAGS_L4_CSUM; DBRUN(sc->debug_udp_csum_offload_frames++); - pbd->global_data |= ETH_TX_PARSE_BD_UDP_CS_FLG; + tx_parse_bd->global_data |= ETH_TX_PARSE_BD_UDP_CS_FLG; /* Get a pointer to the UDP header. */ uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); @@ -8878,8 +8884,8 @@ bxe_tx_encap(struct bxe_fastpath *fp, st ntohl((*(tmp_uh + 2)) & 0x0000FFFF)); /* Update the enet + IP + UDP header length. */ - pbd->total_hlen += (sizeof(struct udphdr) >> 1); - pbd->tcp_pseudo_csum = ~in_addword(uh->uh_sum, ~tmp_csum); + tx_parse_bd->total_hlen += (sizeof(struct udphdr) >> 1); + tx_parse_bd->tcp_pseudo_csum = ~in_addword(uh->uh_sum, ~tmp_csum); } /* Update the flags settings for VLAN/Offload. */ @@ -8900,9 +8906,10 @@ bxe_tx_encap(struct bxe_fastpath *fp, st /* Setup the Parsing BD with TSO specific info */ if (m0->m_pkthdr.csum_flags & CSUM_TSO) { - uint16_t hdr_len = pbd->total_hlen << 1; + uint16_t hdr_len = tx_parse_bd->total_hlen << 1; - DBPRINT(sc, BXE_EXTREME_SEND, "%s(): TSO is enabled.\n",__FUNCTION__); + DBPRINT(sc, BXE_EXTREME_SEND, "%s(): TSO is enabled.\n", + __FUNCTION__); tx_start_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO; @@ -8918,9 +8925,9 @@ bxe_tx_encap(struct bxe_fastpath *fp, st __FUNCTION__, tx_start_bd->nbytes, tx_start_bd->addr_hi, tx_start_bd->addr_lo, nbds); - bd_prod = TX_BD(NEXT_TX_BD(bd_prod)); + bd_prod = TX_BD(NEXT_TX_BD(bd_prod)); - /* Get a new transmit BD (after the pbd) and fill it. */ + /* Get a new transmit BD (after the tx_parse_bd) and fill it. */ tx_data_bd = &fp->tx_bd_chain[TX_PAGE(bd_prod)][TX_IDX(bd_prod)].reg_bd; tx_data_bd->addr_hi = htole32(U64_HI(segs[0].ds_addr + hdr_len)); tx_data_bd->addr_lo = htole32(U64_LO(segs[0].ds_addr + hdr_len)); @@ -8934,23 +8941,26 @@ bxe_tx_encap(struct bxe_fastpath *fp, st */ DBPRINT(sc, BXE_EXTREME_SEND, "%s(): TSO split data " - "size is %d (%x:%x)\n", __FUNCTION__, - tx_data_bd->nbytes, tx_data_bd->addr_hi, tx_data_bd->addr_lo); + "size is %d (%x:%x)\n", __FUNCTION__, + tx_data_bd->nbytes, tx_data_bd->addr_hi, + tx_data_bd->addr_lo); } /* * For TSO the controller needs the following info: * MSS, tcp_send_seq, ip_id, and tcp_pseudo_csum. */ - pbd->lso_mss = htole16(m0->m_pkthdr.tso_segsz); - pbd->tcp_send_seq = ntohl(th->th_seq); - pbd->tcp_flags = th->th_flags; - pbd->ip_id = ntohs(ip->ip_id); + tx_parse_bd->lso_mss = htole16(m0->m_pkthdr.tso_segsz); + tx_parse_bd->tcp_send_seq = ntohl(th->th_seq); + tx_parse_bd->tcp_flags = th->th_flags; + tx_parse_bd->ip_id = ntohs(ip->ip_id); + + tx_parse_bd->tcp_pseudo_csum = + ntohs(in_pseudo(ip->ip_src.s_addr, + ip->ip_dst.s_addr, htons(IPPROTO_TCP))); - pbd->tcp_pseudo_csum = ntohs(in_pseudo(ip->ip_src.s_addr, - ip->ip_dst.s_addr, htons(IPPROTO_TCP))); - - pbd->global_data |= ETH_TX_PARSE_BD_PSEUDO_CS_WITHOUT_LEN; + tx_parse_bd->global_data |= + ETH_TX_PARSE_BD_PSEUDO_CS_WITHOUT_LEN; } } @@ -8971,17 +8981,6 @@ bxe_tx_encap(struct bxe_fastpath *fp, st /* Update bd producer index value for next tx */ bd_prod = TX_BD(NEXT_TX_BD(bd_prod)); -/* -BXE_PRINTF("tx_start_bd: addrlo:0x%x, addrhi:0x%x, nbytes:0x%x, bitfield:0x%x, gendata:0x%x, nbd:0x%x\n", -tx_start_bd->addr_lo, -tx_start_bd->addr_hi, -tx_start_bd->nbytes, -tx_start_bd->bd_flags.as_bitfield, -tx_start_bd->general_data, -tx_start_bd->nbd ); - -bxe_dump_mbuf(sc, m0); -*/ DBRUNMSG(BXE_EXTREME_SEND, bxe_dump_tx_chain(fp, debug_prod, nbds)); /* @@ -9391,12 +9390,12 @@ bxe_ioctl(struct ifnet *ifp, u_long comm * The adjusted value of *fp->rx_cons_sb. */ static __inline uint16_t -bxe_rx_comp_cons(struct bxe_fastpath *fp) +bxe_rx_cq_cons(struct bxe_fastpath *fp) { volatile uint16_t rx_cons_sb = 0; rmb(); - rx_cons_sb = (volatile uint16_t)le16toh(*fp->rx_cons_sb); + rx_cons_sb = (volatile uint16_t) le16toh(*fp->rx_cons_sb); /* * It is valid for the hardware's copy of the completion @@ -9405,8 +9404,8 @@ bxe_rx_comp_cons(struct bxe_fastpath *fp * that it is pointing at the next available CQE so we * need to adjust the value accordingly. */ - if ((rx_cons_sb & TOTAL_RCQ_ENTRIES_PER_PAGE) == - TOTAL_RCQ_ENTRIES_PER_PAGE) + if ((rx_cons_sb & USABLE_RCQ_ENTRIES_PER_PAGE) == + USABLE_RCQ_ENTRIES_PER_PAGE) rx_cons_sb++; return (rx_cons_sb); @@ -9434,7 +9433,7 @@ bxe_has_rx_work(struct bxe_fastpath *fp) { rmb(); - return (bxe_rx_comp_cons(fp) != fp->rx_comp_cons); + return (bxe_rx_cq_cons(fp) != fp->rx_cq_cons); } /* @@ -10306,7 +10305,7 @@ bxe_init_rx_chains(struct bxe_softc *sc) struct eth_rx_bd *rx_bd; struct eth_rx_cqe_next_page *nextpg; uint16_t ring_prod, cqe_ring_prod; - int func, i, j, max_agg_queues; + int func, i, j, rcq_idx, rx_idx, rx_sge_idx, max_agg_queues; DBENTER(BXE_VERBOSE_LOAD | BXE_VERBOSE_RESET); @@ -10374,16 +10373,19 @@ bxe_init_rx_chains(struct bxe_softc *sc) /* Link the SGE Ring Pages to form SGE chain */ for (j = 0; j < NUM_RX_SGE_PAGES; j++) { - DBPRINT(sc, - (BXE_INSANE_LOAD | BXE_INSANE_RESET), - "%s(): Linking fp[%d] SGE ring[%d].\n", - __FUNCTION__, i, j); - + rx_sge_idx = ((j + 1) % NUM_RX_SGE_PAGES); sge = &fp->rx_sge_chain[j][MAX_RX_SGE_CNT]; + + DBPRINT(sc, (BXE_EXTREME_LOAD | BXE_EXTREME_RESET), + "%s(): fp[%02d].rx_sge_chain[%02d][0x%04X]=0x%jX\n", + __FUNCTION__, i, j, + (uint16_t) MAX_RX_SGE_CNT, + (uintmax_t) fp->rx_sge_chain_paddr[rx_sge_idx]); + sge->addr_hi = - htole32(U64_HI(fp->rx_sge_chain_paddr[(j + 1) % NUM_RX_SGE_PAGES])); + htole32(U64_HI(fp->rx_sge_chain_paddr[rx_sge_idx])); sge->addr_lo = - htole32(U64_LO(fp->rx_sge_chain_paddr[(j + 1) % NUM_RX_SGE_PAGES])); + htole32(U64_LO(fp->rx_sge_chain_paddr[rx_sge_idx])); } bxe_init_sge_ring_bit_mask(fp); @@ -10394,16 +10396,19 @@ bxe_init_rx_chains(struct bxe_softc *sc) /* Link the pages to form the RX BD Chain. */ for (j = 0; j < NUM_RX_PAGES; j++) { + rx_idx = ((j + 1) % NUM_RX_PAGES); rx_bd = &fp->rx_bd_chain[j][USABLE_RX_BD_PER_PAGE]; - DBPRINT(sc, (BXE_INSANE_LOAD | BXE_INSANE_RESET), - "%s(): Linking fp[%d] RX BD chain page[%d].\n", - __FUNCTION__, i, j); + DBPRINT(sc, (BXE_INFO_LOAD), + "%s(): fp[%02d].rx_bd_chain[%02d][0x%04X]=0x%jX\n", + __FUNCTION__, i, j, + (uint16_t) USABLE_RX_BD_PER_PAGE, + (uintmax_t) fp->rx_bd_chain_paddr[rx_idx]); rx_bd->addr_hi = - htole32(U64_HI(fp->rx_bd_chain_paddr[(j + 1) % NUM_RX_PAGES])); + htole32(U64_HI(fp->rx_bd_chain_paddr[rx_idx])); rx_bd->addr_lo = - htole32(U64_LO(fp->rx_bd_chain_paddr[(j + 1) % NUM_RX_PAGES])); + htole32(U64_LO(fp->rx_bd_chain_paddr[rx_idx])); } DBPRINT(sc, (BXE_INSANE_LOAD | BXE_INSANE_RESET), @@ -10412,17 +10417,20 @@ bxe_init_rx_chains(struct bxe_softc *sc) /* Link the pages to form the RX Completion Queue.*/ for (j = 0; j < NUM_RCQ_PAGES; j++) { + rcq_idx = ((j + 1) % NUM_RCQ_PAGES); nextpg = (struct eth_rx_cqe_next_page *) - &fp->rx_comp_chain[j][USABLE_RCQ_ENTRIES_PER_PAGE]; + &fp->rx_cq_chain[j][USABLE_RCQ_ENTRIES_PER_PAGE]; - DBPRINT(sc, (BXE_INSANE_LOAD | BXE_INSANE_RESET), - "%s(): Linking fp[%d] RX completion chain page[%d].\n", - __FUNCTION__, i, j); + DBPRINT(sc, (BXE_INFO_LOAD), + "%s(): fp[%02d].rx_cq_chain[%02d][0x%04X]=0x%jX\n", + __FUNCTION__, i, j, + (uint16_t) USABLE_RCQ_ENTRIES_PER_PAGE, + (uintmax_t) fp->rx_cq_chain_paddr[rcq_idx]); nextpg->addr_hi = - htole32(U64_HI(fp->rx_comp_chain_paddr[(j + 1) % NUM_RCQ_PAGES])); + htole32(U64_HI(fp->rx_cq_chain_paddr[rcq_idx])); nextpg->addr_lo = - htole32(U64_LO(fp->rx_comp_chain_paddr[(j + 1) % NUM_RCQ_PAGES])); + htole32(U64_LO(fp->rx_cq_chain_paddr[rcq_idx])); } if (TPA_ENABLED(sc)) { @@ -10432,7 +10440,8 @@ bxe_init_rx_chains(struct bxe_softc *sc) while (ring_prod < sc->rx_ring_size) { if (bxe_alloc_rx_sge(sc, fp, ring_prod) != 0) { BXE_PRINTF( - "%s(%d): Memory allocation failure! Disabling TPA for fp[%d].\n", + "%s(%d): Memory allocation failure! " + "Disabling TPA for fp[%d].\n", __FILE__, __LINE__, i); /* Cleanup already allocated elements */ @@ -10452,7 +10461,7 @@ bxe_init_rx_chains(struct bxe_softc *sc) * Allocate buffers for all the RX BDs in RX BD Chain. * Add completion queue entries at the same time. */ - fp->rx_comp_cons = ring_prod = cqe_ring_prod = 0; + fp->rx_cq_cons = ring_prod = cqe_ring_prod = 0; DBRUN(fp->free_rx_bd = USABLE_RX_BD); while (ring_prod < sc->rx_ring_size) { @@ -10469,9 +10478,9 @@ bxe_init_rx_chains(struct bxe_softc *sc) /* Update the driver's copy of the producer indices. */ fp->rx_bd_prod = ring_prod; - fp->rx_comp_prod = cqe_ring_prod; + fp->rx_cq_prod = cqe_ring_prod; /* - * fp->rx_comp_prod = + * fp->rx_cq_prod = * (uint16_t)min(NUM_RCQ_PAGES*TOTAL_RCQ_ENTRIES_PER_PAGE, * cqe_ring_prod); */ @@ -10485,8 +10494,8 @@ bxe_init_rx_chains(struct bxe_softc *sc) BUS_DMASYNC_PREWRITE); for (j = 0; j < NUM_RCQ_PAGES; j++) - bus_dmamap_sync(fp->rx_comp_chain_tag, - fp->rx_comp_chain_map[j], BUS_DMASYNC_PREREAD | + bus_dmamap_sync(fp->rx_cq_chain_tag, + fp->rx_cq_chain_map[j], BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* @@ -10494,7 +10503,7 @@ bxe_init_rx_chains(struct bxe_softc *sc) * Warning! this will generate an interrupt (to the TSTORM). * This must only be done when the controller is initialized. */ - bxe_update_rx_prod(sc, fp, ring_prod, fp->rx_comp_prod, + bxe_update_rx_prod(sc, fp, ring_prod, fp->rx_cq_prod, fp->rx_sge_prod); if (i != 0) @@ -10502,10 +10511,10 @@ bxe_init_rx_chains(struct bxe_softc *sc) REG_WR(sc, BAR_USTORM_INTMEM + USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func), - U64_LO(fp->rx_comp_chain_paddr[0])); + U64_LO(fp->rx_cq_chain_paddr[0])); REG_WR(sc, BAR_USTORM_INTMEM + USTORM_MEM_WORKAROUND_ADDRESS_OFFSET(func) + 4, - U64_HI(fp->rx_comp_chain_paddr[0])); + U64_HI(fp->rx_cq_chain_paddr[0])); } DBEXIT(BXE_VERBOSE_LOAD | BXE_VERBOSE_RESET); @@ -10521,7 +10530,7 @@ static void bxe_init_tx_chains(struct bxe_softc *sc) { struct bxe_fastpath *fp; - struct eth_tx_next_bd *tx_bd; + struct eth_tx_next_bd *tx_n_bd; int i, j; DBENTER(BXE_VERBOSE_LOAD | BXE_VERBOSE_RESET); @@ -10532,17 +10541,17 @@ bxe_init_tx_chains(struct bxe_softc *sc) "%s(): Linking fp[%d] TX chain pages.\n", __FUNCTION__, i); for (j = 0; j < NUM_TX_PAGES; j++) { - tx_bd = + tx_n_bd = &fp->tx_bd_chain[j][USABLE_TX_BD_PER_PAGE].next_bd; DBPRINT(sc, (BXE_INSANE_LOAD | BXE_INSANE_RESET), "%s(): Linking fp[%d] TX BD chain page[%d].\n", __FUNCTION__, i, j); - tx_bd->addr_hi = + tx_n_bd->addr_hi = htole32(U64_HI(fp->tx_bd_chain_paddr[(j + 1) % NUM_TX_PAGES])); - tx_bd->addr_lo = + tx_n_bd->addr_lo = htole32(U64_LO(fp->tx_bd_chain_paddr[(j + 1) % NUM_TX_PAGES])); } @@ -10609,8 +10618,8 @@ bxe_free_rx_chains(struct bxe_softc *sc) /* Clear each RX completion queue page. */ for (j = 0; j < NUM_RCQ_PAGES; j++) { - if (fp->rx_comp_chain[j] != NULL) - bzero((char *)fp->rx_comp_chain[j], + if (fp->rx_cq_chain[j] != NULL) + bzero((char *)fp->rx_cq_chain[j], BXE_RX_CHAIN_PAGE_SZ); } @@ -11190,14 +11199,14 @@ bxe_init_internal_func(struct bxe_softc for (i = 0; i < sc->num_queues; i++) { fp = &sc->fp[i]; nextpg = (struct eth_rx_cqe_next_page *) - &fp->rx_comp_chain[i][USABLE_RCQ_ENTRIES_PER_PAGE]; + &fp->rx_cq_chain[i][USABLE_RCQ_ENTRIES_PER_PAGE]; /* Program the completion queue address. */ REG_WR(sc, BAR_USTORM_INTMEM + USTORM_CQE_PAGE_BASE_OFFSET(port, fp->cl_id), - U64_LO(fp->rx_comp_chain_paddr[0])); + U64_LO(fp->rx_cq_chain_paddr[0])); REG_WR(sc, BAR_USTORM_INTMEM + USTORM_CQE_PAGE_BASE_OFFSET(port, fp->cl_id) + 4, - U64_HI(fp->rx_comp_chain_paddr[0])); + U64_HI(fp->rx_cq_chain_paddr[0])); /* Next page */ REG_WR(sc, BAR_USTORM_INTMEM + USTORM_CQE_PAGE_NEXT_OFFSET(port, @@ -12806,22 +12815,22 @@ bxe_dma_free(struct bxe_softc *sc) * Free, unmap and destroy all RX CQ * chain pages. */ - if (fp->rx_comp_chain_tag != NULL) { + if (fp->rx_cq_chain_tag != NULL) { for (j = 0; j < NUM_RCQ_PAGES; j++ ) { - if (fp->rx_comp_chain_map[j] != NULL) { - if (fp->rx_comp_chain[j] != NULL) - bus_dmamem_free(fp->rx_comp_chain_tag, - fp->rx_comp_chain[j], - fp->rx_comp_chain_map[j]); - - bus_dmamap_unload(fp->rx_comp_chain_tag, - fp->rx_comp_chain_map[j]); - bus_dmamap_destroy(fp->rx_comp_chain_tag, - fp->rx_comp_chain_map[j]); + if (fp->rx_cq_chain_map[j] != NULL) { + if (fp->rx_cq_chain[j] != NULL) + bus_dmamem_free(fp->rx_cq_chain_tag, + fp->rx_cq_chain[j], + fp->rx_cq_chain_map[j]); + + bus_dmamap_unload(fp->rx_cq_chain_tag, + fp->rx_cq_chain_map[j]); + bus_dmamap_destroy(fp->rx_cq_chain_tag, + fp->rx_cq_chain_map[j]); } } - bus_dma_tag_destroy(fp->rx_comp_chain_tag); + bus_dma_tag_destroy(fp->rx_cq_chain_tag); } /* Unload and destroy the TX mbuf maps. */ @@ -13093,7 +13102,7 @@ bxe_dma_alloc(device_t dev) NULL, /* filter f() */ NULL, /* filter f() arg */ MAXBSIZE, /* max map for this tag */ - BUS_SPACE_UNRESTRICTED, /* #of discontinuities */ + BUS_SPACE_UNRESTRICTED, /* # of discontinuities */ BUS_SPACE_MAXSIZE_32BIT, /* max seg size */ 0, /* flags */ NULL, /* lock f() */ @@ -13115,9 +13124,9 @@ bxe_dma_alloc(device_t dev) (long unsigned int)sizeof(struct bxe_fastpath)); /* - * Create a DMA tag for the status block, allocate and clear the - * memory, map the memory into DMA space, and fetch the physical - * address of the block. + * Create a DMA tag for the status block, allocate and + * clear the memory, map the memory into DMA space, and + * fetch the physical address of the block. */ if (bus_dma_tag_create(sc->parent_tag, @@ -13266,10 +13275,11 @@ bxe_dma_alloc(device_t dev) /* Create DMA maps for each the TX mbuf cluster(ext buf). */ for (j = 0; j < TOTAL_TX_BD; j++) { - if (bus_dmamap_create(fp->tx_mbuf_tag, BUS_DMA_NOWAIT, + if (bus_dmamap_create(fp->tx_mbuf_tag, + BUS_DMA_NOWAIT, &(fp->tx_mbuf_map[j]))) { - BXE_PRINTF( - "%s(%d): Unable to create fp[%d] TX mbuf DMA map!\n", + BXE_PRINTF("%s(%d): Unable to create fp[%d] " + "TX mbuf DMA map!\n", __FILE__, __LINE__, i); rc = ENOMEM; goto bxe_dma_alloc_exit; @@ -13363,8 +13373,8 @@ bxe_dma_alloc(device_t dev) /* Create DMA maps for the RX mbuf clusters. */ for (j = 0; j < TOTAL_RX_BD; j++) { - if (bus_dmamap_create(fp->rx_mbuf_tag, BUS_DMA_NOWAIT, - &(fp->rx_mbuf_map[j]))) { + if (bus_dmamap_create(fp->rx_mbuf_tag, + BUS_DMA_NOWAIT, &(fp->rx_mbuf_map[j]))) { BXE_PRINTF( "%s(%d): Unable to create fp[%d] RX mbuf DMA map!\n", __FILE__, __LINE__, i); @@ -13392,7 +13402,7 @@ bxe_dma_alloc(device_t dev) 0, /* flags */ NULL, /* lock f() */ NULL, /* lock f() arg */ - &fp->rx_comp_chain_tag)) { + &fp->rx_cq_chain_tag)) { BXE_PRINTF( "%s(%d): Could not allocate fp[%d] RX Completion Queue DMA tag!\n", __FILE__, __LINE__, i); @@ -13401,9 +13411,9 @@ bxe_dma_alloc(device_t dev) } for (j = 0; j < NUM_RCQ_PAGES; j++) { - if (bus_dmamem_alloc(fp->rx_comp_chain_tag, - (void **)&fp->rx_comp_chain[j], BUS_DMA_NOWAIT, - &fp->rx_comp_chain_map[j])) { + if (bus_dmamem_alloc(fp->rx_cq_chain_tag, + (void **)&fp->rx_cq_chain[j], BUS_DMA_NOWAIT, + &fp->rx_cq_chain_map[j])) { BXE_PRINTF( "%s(%d): Could not allocate fp[%d] RX Completion Queue DMA memory!\n", __FILE__, __LINE__, i); @@ -13411,11 +13421,11 @@ bxe_dma_alloc(device_t dev) goto bxe_dma_alloc_exit; } - bzero((char *)fp->rx_comp_chain[j], + bzero((char *)fp->rx_cq_chain[j], BXE_RX_CHAIN_PAGE_SZ); - error = bus_dmamap_load(fp->rx_comp_chain_tag, - fp->rx_comp_chain_map[j], fp->rx_comp_chain[j], + error = bus_dmamap_load(fp->rx_cq_chain_tag, + fp->rx_cq_chain_map[j], fp->rx_cq_chain[j], BXE_RX_CHAIN_PAGE_SZ, bxe_dma_map_addr, &busaddr, BUS_DMA_NOWAIT); @@ -13431,17 +13441,17 @@ bxe_dma_alloc(device_t dev) * Physical address of each page in the RX * Completion Chain. */ - fp->rx_comp_chain_paddr[j] = busaddr; + fp->rx_cq_chain_paddr[j] = busaddr; DBPRINT(sc, (BXE_EXTREME_LOAD | BXE_EXTREME_RESET), - "%s(): fp[%d]->rx_comp_chain_paddr[%d] = 0x%jX\n", + "%s(): fp[%d]->rx_cq_chain_paddr[%d] = 0x%jX\n", __FUNCTION__, i, j, (uintmax_t)busaddr); } if (TPA_ENABLED(sc)) { int tpa_pool_max = CHIP_IS_E1H(sc) ? - ETH_MAX_AGGREGATION_QUEUES_E1H : - ETH_MAX_AGGREGATION_QUEUES_E1; + ETH_MAX_AGGREGATION_QUEUES_E1H : + ETH_MAX_AGGREGATION_QUEUES_E1; /* * Create a DMA tag for the RX SGE Ring, @@ -14213,6 +14223,7 @@ bxe_ifmedia_upd(struct ifnet *ifp) ifm = &sc->bxe_ifmedia; rc = 0; + /* This is an Ethernet controller. */ if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) { rc = EINVAL; @@ -14672,27 +14683,27 @@ bxe_tpa_stop_exit: */ static __inline void bxe_update_rx_prod(struct bxe_softc *sc, struct bxe_fastpath *fp, - uint16_t bd_prod, uint16_t rx_comp_prod, uint16_t rx_sge_prod) + uint16_t bd_prod, uint16_t rx_cq_prod, uint16_t sge_prod) { volatile struct ustorm_eth_rx_producers rx_prods = {0}; int i; /* Update producers. */ - rx_prods.bd_prod = bd_prod;/* htole16(bd_prod);*/ - rx_prods.cqe_prod = rx_comp_prod;/*htole16(rx_comp_prod);*/ - rx_prods.sge_prod = rx_sge_prod;/*htole16(rx_sge_prod);*/ + rx_prods.bd_prod = bd_prod; + rx_prods.cqe_prod = rx_cq_prod; + rx_prods.sge_prod = sge_prod; wmb(); for (i = 0; i < sizeof(struct ustorm_eth_rx_producers) / 4; i++){ REG_WR(sc, BAR_USTORM_INTMEM + USTORM_RX_PRODS_OFFSET(BP_PORT(sc), fp->cl_id) + i * 4, - ((volatile uint32_t *)&rx_prods)[i]); + ((volatile uint32_t *) &rx_prods)[i]); } - DBPRINT(sc, BXE_EXTREME_RECV, "%s(): Wrote fp[%d] bd_prod = 0x%04X, " - "cqe_prod = 0x%04X, sge_prod = 0x%04X\n", __FUNCTION__, fp->index, - bd_prod, rx_comp_prod, rx_sge_prod); + DBPRINT(sc, BXE_EXTREME_RECV, "%s(%d): Wrote fp[%02d] bd_prod = 0x%04X, " + "rx_cq_prod = 0x%04X, sge_prod = 0x%04X\n", __FUNCTION__, curcpu, + fp->index, bd_prod, rx_cq_prod, sge_prod); } /* @@ -14706,32 +14717,38 @@ bxe_rxeof(struct bxe_fastpath *fp) { struct bxe_softc *sc; struct ifnet *ifp; - uint16_t bd_cons, bd_prod, bd_prod_fw; - uint16_t hw_comp_cons_idx, sw_comp_cons_idx, sw_comp_prod; - uint16_t comp_ring_cons; + uint16_t rx_bd_cons, rx_bd_cons_idx; + uint16_t rx_bd_prod, rx_bd_prod_idx; + uint16_t rx_cq_cons, rx_cq_cons_idx; + uint16_t rx_cq_prod, rx_cq_cons_sb; #ifdef BXE_DEBUG unsigned long rx_pkts = 0; #endif + sc = fp->sc; ifp = sc->bxe_ifp; DBENTER(BXE_EXTREME_RECV); - /* Get the sb's view of the RX completion consumer index. */ - hw_comp_cons_idx = le16toh(*fp->rx_cons_sb); - if ((hw_comp_cons_idx & USABLE_RCQ_ENTRIES_PER_PAGE) == - USABLE_RCQ_ENTRIES_PER_PAGE) - hw_comp_cons_idx++; - /* Get working copies of the driver's view of the RX indices. */ - bd_cons = fp->rx_bd_cons; - bd_prod = bd_prod_fw = fp->rx_bd_prod; - sw_comp_cons_idx = fp->rx_comp_cons; - sw_comp_prod = fp->rx_comp_prod; + + /* Get the status block's view of the RX completion consumer index. */ + rx_cq_cons_sb = bxe_rx_cq_cons(fp); + + /* + * Get working copies of the driver's view of the + * RX indices. These are 16 bit values that are + * expected to increment from from 0 to 65535 + * and then wrap-around to 0 again. + */ + rx_bd_cons = fp->rx_bd_cons; + rx_bd_prod = fp->rx_bd_prod; + rx_cq_cons = fp->rx_cq_cons; + rx_cq_prod = fp->rx_cq_prod; DBPRINT(sc, (BXE_EXTREME_RECV), - "%s(): fp[%d], bd_cons = 0x%04X, bd_prod = 0x%04X, " - "sw_comp_cons = 0x%04X, sw_comp_prod = 0x%04X\n", __FUNCTION__, - fp->index, bd_cons, bd_prod, sw_comp_cons_idx, sw_comp_prod); + "%s(%d): BEFORE: fp[%d], rx_bd_cons = 0x%04X, rx_bd_prod = 0x%04X, " + "rx_cq_cons_sw = 0x%04X, rx_cq_prod_sw = 0x%04X\n", __FUNCTION__, + curcpu, fp->index, rx_bd_cons, rx_bd_prod, rx_cq_cons, rx_cq_prod); /* * Memory barrier to prevent speculative reads of the RX buffer @@ -14743,20 +14760,25 @@ bxe_rxeof(struct bxe_fastpath *fp) * Scan through the receive chain as long * as there is work to do. */ - while (sw_comp_cons_idx != hw_comp_cons_idx) { + while (rx_cq_cons != rx_cq_cons_sb) { struct mbuf *m; union eth_rx_cqe *cqe; uint8_t cqe_fp_flags; uint16_t len, pad; - /* Convert the indices to an actual rx_bd index. */ - comp_ring_cons = RCQ_ENTRY(sw_comp_cons_idx); - bd_prod = RX_BD(bd_prod); - bd_cons = RX_BD(bd_cons); + /* + * Convert the 16 bit indices used by hardware + * into values that map to the arrays used by + * the driver (i.e. an index). + */ + rx_cq_cons_idx = RCQ_ENTRY(rx_cq_cons); + rx_bd_prod_idx = RX_BD(rx_bd_prod); + rx_bd_cons_idx = RX_BD(rx_bd_cons); wmb(); + /* Fetch the cookie. */ - cqe = (union eth_rx_cqe*) - &fp->rx_comp_chain[RCQ_PAGE(comp_ring_cons)][RCQ_IDX(comp_ring_cons)]; + cqe = (union eth_rx_cqe *) + &fp->rx_cq_chain[RCQ_PAGE(rx_cq_cons_idx)][RCQ_IDX(rx_cq_cons_idx)]; cqe_fp_flags = cqe->fast_path_cqe.type_error_flags; #ifdef BXE_DEBUG @@ -14769,9 +14791,10 @@ bxe_rxeof(struct bxe_fastpath *fp) } #endif - DBRUNIF((cqe_fp_flags == 0), BXE_PRINTF( - "%s(): CQE received with null type/error flags!\n", - __FUNCTION__)); + DBRUNIF((cqe_fp_flags == 0), + BXE_PRINTF("%s(): CQE received with null " + "type/error flags!\n", __FUNCTION__); + bxe_dump_cqe(fp, rx_cq_cons_idx, cqe)); /* Check the CQE type for slowpath or fastpath completion. */ if (__predict_false(CQE_TYPE(cqe_fp_flags) == @@ -14803,8 +14826,8 @@ bxe_rxeof(struct bxe_fastpath *fp) * Check if a TPA aggregation has been started. */ if (TPA_TYPE(cqe_fp_flags) == TPA_TYPE_START) { - bxe_tpa_start(fp, queue, bd_cons, - bd_prod); + bxe_tpa_start(fp, queue, + rx_bd_cons_idx, rx_bd_prod_idx); goto bxe_rxeof_next_rx; } @@ -14826,7 +14849,7 @@ bxe_rxeof(struct bxe_fastpath *fp) * the frame up. */ bxe_tpa_stop(sc, fp, queue, pad, len, - cqe, comp_ring_cons); + cqe, rx_cq_cons_idx); bxe_update_sge_prod(fp, &cqe->fast_path_cqe); goto bxe_rxeof_next_cqe; @@ -14834,8 +14857,8 @@ bxe_rxeof(struct bxe_fastpath *fp) } /* Remove the mbuf from the RX chain. */ - m = fp->rx_mbuf_ptr[bd_cons]; - fp->rx_mbuf_ptr[bd_cons] = NULL; + m = fp->rx_mbuf_ptr[rx_bd_cons_idx]; + fp->rx_mbuf_ptr[rx_bd_cons_idx] = NULL; DBRUN(fp->free_rx_bd++); DBRUNIF((fp->free_rx_bd > USABLE_RX_BD), @@ -14845,9 +14868,10 @@ bxe_rxeof(struct bxe_fastpath *fp) /* Unmap the mbuf from DMA space. */ bus_dmamap_sync(fp->rx_mbuf_tag, - fp->rx_mbuf_map[bd_cons], BUS_DMASYNC_POSTREAD); + fp->rx_mbuf_map[rx_bd_cons_idx], + BUS_DMASYNC_POSTREAD); bus_dmamap_unload(fp->rx_mbuf_tag, - fp->rx_mbuf_map[bd_cons]); + fp->rx_mbuf_map[rx_bd_cons_idx]); /* Check if the received frame has any errors. */ if (__predict_false(cqe_fp_flags & @@ -14860,11 +14884,13 @@ bxe_rxeof(struct bxe_fastpath *fp) fp->soft_rx_errors++; /* Reuse the mbuf for a new frame. */ - if (bxe_get_buf(fp, m, bd_prod)) { + if (bxe_get_buf(fp, m, rx_bd_prod_idx)) { DBPRINT(sc, BXE_FATAL, "%s(): Can't reuse RX mbuf!\n", __FUNCTION__); DBRUN(bxe_breakpoint(sc)); + + /* ToDo: Find alterntive to panic(). */ panic("bxe%d: Can't reuse RX mbuf!\n", sc->bxe_unit); } @@ -14885,7 +14911,8 @@ bxe_rxeof(struct bxe_fastpath *fp) */ /* Allocate a new mbuf for the receive chain. */ - if (__predict_false(bxe_get_buf(fp, NULL, bd_prod))) { + if (__predict_false(bxe_get_buf(fp, + NULL, rx_bd_prod_idx))) { /* * Drop the current frame if we can't get * a new mbuf. @@ -14897,12 +14924,11 @@ bxe_rxeof(struct bxe_fastpath *fp) * receive chain. */ if (__predict_false(bxe_get_buf(fp, m, - bd_prod))) { + rx_bd_prod_idx))) { /* This is really bad! */ DBPRINT(sc, BXE_FATAL, "%s(): Can't reuse RX mbuf!\n", __FUNCTION__); - DBRUN(bxe_breakpoint(sc)); /* ToDo: Find alterntive to panic(). */ @@ -14980,12 +15006,12 @@ bxe_rxeof(struct bxe_fastpath *fp) } /* Last chance to check for problems. */ - DBRUN(bxe_validate_rx_packet(fp, comp_ring_cons, cqe, - m)); + DBRUN(bxe_validate_rx_packet(fp, rx_cq_cons, cqe, m)); /* Pass the mbuf off to the upper layers. */ ifp->if_ipackets++; + /* ToDo: Any potential locking issues here? */ /* Pass the frame to the stack. */ (*ifp->if_input)(ifp, m); @@ -14993,14 +15019,13 @@ bxe_rxeof(struct bxe_fastpath *fp) } bxe_rxeof_next_rx: - bd_prod = NEXT_RX_BD(bd_prod); - bd_prod_fw = NEXT_RX_BD(bd_prod_fw); - bd_cons = NEXT_RX_BD(bd_cons); + rx_bd_prod = NEXT_RX_BD(rx_bd_prod); + rx_bd_cons = NEXT_RX_BD(rx_bd_cons); DBRUN(rx_pkts++); bxe_rxeof_next_cqe: - sw_comp_prod = NEXT_RCQ_IDX(sw_comp_prod); - sw_comp_cons_idx = NEXT_RCQ_IDX(sw_comp_cons_idx); + rx_cq_prod = NEXT_RCQ_IDX(rx_cq_prod); + rx_cq_cons = NEXT_RCQ_IDX(rx_cq_cons); /* * Memory barrier to prevent speculative reads of the RX buffer @@ -15010,18 +15035,19 @@ bxe_rxeof_next_cqe: } /* Update the driver copy of the fastpath indices. */ - fp->rx_bd_cons = bd_cons; - fp->rx_bd_prod = bd_prod_fw; - fp->rx_comp_cons = sw_comp_cons_idx; *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201103312240.p2VMeiEb024871>