From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rasesh Mody Subject: [PATCH 06/14] bna: TX Path and RX Path Changes Date: Tue, 16 Aug 2011 14:19:43 -0700 Message-ID: <1313529591-3718-7-git-send-email-rmody@brocade.com> References: <1313529591-3718-1-git-send-email-rmody@brocade.com> Mime-Version: 1.0 Content-Type: text/plain Cc: , Rasesh Mody , Gurunatha Karaje To: , Return-path: Received: from mx0b-000f0801.pphosted.com ([67.231.152.113]:47906 "EHLO mx0b-000f0801.pphosted.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752823Ab1HPVUb (ORCPT ); Tue, 16 Aug 2011 17:20:31 -0400 In-Reply-To: <1313529591-3718-1-git-send-email-rmody@brocade.com> Sender: netdev-owner@vger.kernel.org List-ID: Change details: - Disable and enable interrupts from the same polling context to prevent reordering in Rx path. - Add Rx NAPI debug counters. - Make NAPI budget check more generic - Add a macro bnad_dim_timer_stop for DIM(Dynamic Interrupt Moderation) timer stop - Handle reduced MSI-X vectors case in bnad_enable_msix - Replace existing checks with macros and add more checks for illegal skbs in transmit path. Add more tx_skb counters for dropped skbs. - Check for single frame TSO skbs and send them out as non-TSO. - Put memory barrier after bna_txq_prod_indx_doorbell() Signed-off-by: Gurunatha Karaje Signed-off-by: Rasesh Mody --- drivers/net/ethernet/brocade/bna/bnad.c | 207 +++++++++++++++++-------------- drivers/net/ethernet/brocade/bna/bnad.h | 33 +++++- 2 files changed, 148 insertions(+), 92 deletions(-) diff --git a/drivers/net/ethernet/brocade/bna/bnad.c b/drivers/net/ethernet/brocade/bna/bnad.c index 64e2106..0faa8a1 100644 --- a/drivers/net/ethernet/brocade/bna/bnad.c +++ b/drivers/net/ethernet/brocade/bna/bnad.c @@ -532,7 +532,7 @@ bnad_poll_cq(struct bnad *bnad, struct bna_ccb *ccb, int budget) (flags & BNA_CQ_EF_L4_CKSUM_OK))) skb->ip_summed = CHECKSUM_UNNECESSARY; else - skb_checksum_none_assert(skb); + skb->ip_summed = CHECKSUM_NONE; rcb->rxq->rx_packets++; rcb->rxq->rx_bytes += skb->len; @@ -555,7 +555,8 @@ next: BNA_QE_INDX_ADD(ccb->producer_index, wis, ccb->q_depth); if (likely(test_bit(BNAD_RXQ_STARTED, &ccb->rcb[0]->flags))) - bna_ib_ack(ccb->i_dbell, packets); + bna_ib_ack_disable_irq(ccb->i_dbell, packets); + bnad_refill_rxq(bnad, ccb->rcb[0]); if (ccb->rcb[1]) bnad_refill_rxq(bnad, ccb->rcb[1]); @@ -593,10 +594,9 @@ bnad_netif_rx_schedule_poll(struct bnad *bnad, struct bna_ccb *ccb) struct napi_struct *napi = &rx_ctrl->napi; if (likely(napi_schedule_prep(napi))) { - bnad_disable_rx_irq(bnad, ccb); __napi_schedule(napi); + rx_ctrl->rx_schedule++; } - BNAD_UPDATE_CTR(bnad, netif_rx_schedule); } /* MSIX Rx Path Handler */ @@ -605,8 +605,10 @@ bnad_msix_rx(int irq, void *data) { struct bna_ccb *ccb = (struct bna_ccb *)data; - if (ccb) + if (ccb) { + ((struct bnad_rx_ctrl *)(ccb->ctrl))->rx_intr_ctr++; bnad_netif_rx_schedule_poll(ccb->bnad, ccb); + } return IRQ_HANDLED; } @@ -1675,22 +1677,23 @@ bnad_napi_poll_rx(struct napi_struct *napi, int budget) struct bnad *bnad = rx_ctrl->bnad; int rcvd = 0; + rx_ctrl->rx_poll_ctr++; if (!netif_carrier_ok(bnad->netdev)) goto poll_exit; rcvd = bnad_poll_cq(bnad, rx_ctrl->ccb, budget); - if (rcvd == budget) + if (rcvd >= budget) return rcvd; poll_exit: napi_complete((napi)); - BNAD_UPDATE_CTR(bnad, netif_rx_complete); - + rx_ctrl->rx_complete++; if (rx_ctrl->ccb) - bnad_enable_rx_irq(bnad, rx_ctrl->ccb); + bnad_enable_rx_irq_unsafe(rx_ctrl->ccb); + return rcvd; } @@ -1894,20 +1897,14 @@ bnad_cleanup_rx(struct bnad *bnad, u32 rx_id) struct bna_rx_config *rx_config = &bnad->rx_config[rx_id]; struct bna_res_info *res_info = &bnad->rx_res_info[rx_id].res_info[0]; unsigned long flags; - int dim_timer_del = 0; if (!rx_info->rx) return; - if (0 == rx_id) { - spin_lock_irqsave(&bnad->bna_lock, flags); - dim_timer_del = bnad_dim_timer_running(bnad); - if (dim_timer_del) - clear_bit(BNAD_RF_DIM_TIMER_RUNNING, &bnad->run_flags); - spin_unlock_irqrestore(&bnad->bna_lock, flags); - if (dim_timer_del) - del_timer_sync(&bnad->dim_timer); - } + spin_lock_irqsave(&bnad->bna_lock, flags); + if (0 == rx_id) + bnad_dim_timer_stop(bnad, flags); + spin_unlock_irqrestore(&bnad->bna_lock, flags); init_completion(&bnad->bnad_completions.rx_comp); spin_lock_irqsave(&bnad->bna_lock, flags); @@ -2408,12 +2405,11 @@ bnad_enable_msix(struct bnad *bnad) spin_lock_irqsave(&bnad->bna_lock, flags); /* ret = #of vectors that we got */ - bnad_q_num_adjust(bnad, ret, 0); + bnad_q_num_adjust(bnad, (ret - BNAD_MAILBOX_MSIX_VECTORS) / 2, + (ret - BNAD_MAILBOX_MSIX_VECTORS) / 2); spin_unlock_irqrestore(&bnad->bna_lock, flags); - bnad->msix_num = (bnad->num_tx * bnad->num_txq_per_tx) - + (bnad->num_rx - * bnad->num_rxp_per_rx) + + bnad->msix_num = BNAD_NUM_TXQ + BNAD_NUM_RXP + BNAD_MAILBOX_MSIX_VECTORS; if (bnad->msix_num > ret) @@ -2571,26 +2567,21 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) u32 unmap_prod, wis, wis_used, wi_range; u32 vectors, vect_id, i, acked; int err; + unsigned int len; dma_addr_t dma_addr; struct bna_txq_entry *txqent; u16 flags; - if (unlikely - (skb->len <= ETH_HLEN || skb->len > BFI_TX_MAX_DATA_PER_PKT)) { - dev_kfree_skb(skb); - return NETDEV_TX_OK; - } + BNAD_DROP_AND_RETURN_IF(skb->len <= ETH_HLEN, tx_skb_too_short); + BNAD_DROP_AND_RETURN_IF(skb_headlen(skb) > BFI_TX_MAX_DATA_PER_VECTOR, + tx_skb_headlen_too_long); + BNAD_DROP_AND_RETURN_IF(skb_headlen(skb) == 0, tx_skb_headlen_zero); txq_id = skb_get_queue_mapping(skb); tcb = bnad->tx_info[0].tcb[txq_id]; - if (unlikely(!tcb)) { - dev_kfree_skb(skb); - return NETDEV_TX_OK; - } - unmap_q = tcb->unmap_q; /* * Takes care of the Tx that is scheduled between clearing the flag @@ -2638,8 +2629,6 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) } unmap_prod = unmap_q->producer_index; - wis_used = 1; - vect_id = 0; flags = 0; txq_prod = tcb->producer_index; @@ -2647,9 +2636,6 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) BUG_ON(!(wi_range <= tcb->q_depth)); txqent->hdr.wi.reserved = 0; txqent->hdr.wi.num_vectors = vectors; - txqent->hdr.wi.opcode = - htons((skb_is_gso(skb) ? BNA_TXQ_WI_SEND_LSO : - BNA_TXQ_WI_SEND)); if (vlan_tx_tag_present(skb)) { vlan_tag = (u16) vlan_tx_tag_get(skb); @@ -2665,62 +2651,74 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) txqent->hdr.wi.vlan_tag = htons(vlan_tag); if (skb_is_gso(skb)) { - err = bnad_tso_prepare(bnad, skb); - if (err) { - dev_kfree_skb(skb); - return NETDEV_TX_OK; + BNAD_DROP_AND_RETURN_IF(skb_is_gso(skb) > netdev->mtu, + tx_skb_mss_too_long); + if (unlikely((skb_is_gso(skb) + skb_transport_offset(skb) + + tcp_hdrlen(skb)) >= skb->len)) { + txqent->hdr.wi.opcode = + __constant_htons(BNA_TXQ_WI_SEND); + txqent->hdr.wi.lso_mss = 0; + BNAD_UPDATE_CTR(bnad, tx_skb_tso_too_short); + } else { + txqent->hdr.wi.opcode = + __constant_htons(BNA_TXQ_WI_SEND_LSO); + txqent->hdr.wi.lso_mss = htons(skb_is_gso(skb)); } - txqent->hdr.wi.lso_mss = htons(skb_is_gso(skb)); + + err = bnad_tso_prepare(bnad, skb); + BNAD_DROP_AND_RETURN_IF(err, tx_skb_tso_prepare); flags |= (BNA_TXQ_WI_CF_IP_CKSUM | BNA_TXQ_WI_CF_TCP_CKSUM); txqent->hdr.wi.l4_hdr_size_n_offset = htons(BNA_TXQ_WI_L4_HDR_N_OFFSET (tcp_hdrlen(skb) >> 2, skb_transport_offset(skb))); - } else if (skb->ip_summed == CHECKSUM_PARTIAL) { - u8 proto = 0; - + } else { + txqent->hdr.wi.opcode = __constant_htons(BNA_TXQ_WI_SEND); txqent->hdr.wi.lso_mss = 0; - if (skb->protocol == htons(ETH_P_IP)) - proto = ip_hdr(skb)->protocol; - else if (skb->protocol == htons(ETH_P_IPV6)) { - /* nexthdr may not be TCP immediately. */ - proto = ipv6_hdr(skb)->nexthdr; - } - if (proto == IPPROTO_TCP) { - flags |= BNA_TXQ_WI_CF_TCP_CKSUM; - txqent->hdr.wi.l4_hdr_size_n_offset = - htons(BNA_TXQ_WI_L4_HDR_N_OFFSET - (0, skb_transport_offset(skb))); - - BNAD_UPDATE_CTR(bnad, tcpcsum_offload); - - BUG_ON(!(skb_headlen(skb) >= - skb_transport_offset(skb) + tcp_hdrlen(skb))); + BNAD_DROP_AND_RETURN_IF(skb->len > (netdev->mtu + ETH_HLEN), + tx_skb_non_tso_too_long); - } else if (proto == IPPROTO_UDP) { - flags |= BNA_TXQ_WI_CF_UDP_CKSUM; - txqent->hdr.wi.l4_hdr_size_n_offset = - htons(BNA_TXQ_WI_L4_HDR_N_OFFSET - (0, skb_transport_offset(skb))); + if (skb->ip_summed == CHECKSUM_PARTIAL) { + u8 proto = 0; - BNAD_UPDATE_CTR(bnad, udpcsum_offload); - - BUG_ON(!(skb_headlen(skb) >= - skb_transport_offset(skb) + - sizeof(struct udphdr))); - } else { - err = skb_checksum_help(skb); - BNAD_UPDATE_CTR(bnad, csum_help); - if (err) { - dev_kfree_skb(skb); - BNAD_UPDATE_CTR(bnad, csum_help_err); - return NETDEV_TX_OK; + if (skb->protocol == __constant_htons(ETH_P_IP)) + proto = ip_hdr(skb)->protocol; + else if (skb->protocol == + __constant_htons(ETH_P_IPV6)) { + /* nexthdr may not be TCP immediately. */ + proto = ipv6_hdr(skb)->nexthdr; + } + if (proto == IPPROTO_TCP) { + flags |= BNA_TXQ_WI_CF_TCP_CKSUM; + txqent->hdr.wi.l4_hdr_size_n_offset = + htons(BNA_TXQ_WI_L4_HDR_N_OFFSET + (0, skb_transport_offset(skb))); + + BNAD_UPDATE_CTR(bnad, tcpcsum_offload); + + BNAD_DROP_AND_RETURN_IF(skb_headlen(skb) < + skb_transport_offset(skb) + + tcp_hdrlen(skb), tx_skb_tcp_hdr); + + } else if (proto == IPPROTO_UDP) { + flags |= BNA_TXQ_WI_CF_UDP_CKSUM; + txqent->hdr.wi.l4_hdr_size_n_offset = + htons(BNA_TXQ_WI_L4_HDR_N_OFFSET + (0, skb_transport_offset(skb))); + + BNAD_UPDATE_CTR(bnad, udpcsum_offload); + BNAD_DROP_AND_RETURN_IF(skb_headlen(skb) < + skb_transport_offset(skb) + + sizeof(struct udphdr), + tx_skb_udp_hdr); + + } else { + BNAD_DROP_AND_RETURN(tx_skb_csum_err); } + } else { + txqent->hdr.wi.l4_hdr_size_n_offset = 0; } - } else { - txqent->hdr.wi.lso_mss = 0; - txqent->hdr.wi.l4_hdr_size_n_offset = 0; } txqent->hdr.wi.flags = htons(flags); @@ -2728,20 +2726,36 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) txqent->hdr.wi.frame_length = htonl(skb->len); unmap_q->unmap_array[unmap_prod].skb = skb; - BUG_ON(!(skb_headlen(skb) <= BFI_TX_MAX_DATA_PER_VECTOR)); - txqent->vector[vect_id].length = htons(skb_headlen(skb)); + len = skb_headlen(skb); + txqent->vector[0].length = htons(len); dma_addr = dma_map_single(&bnad->pcidev->dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE); dma_unmap_addr_set(&unmap_q->unmap_array[unmap_prod], dma_addr, dma_addr); - BNA_SET_DMA_ADDR(dma_addr, &txqent->vector[vect_id].host_addr); + BNA_SET_DMA_ADDR(dma_addr, &txqent->vector[0].host_addr); BNA_QE_INDX_ADD(unmap_prod, 1, unmap_q->q_depth); + vect_id = 0; + wis_used = 1; + for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i]; u16 size = frag->size; + if (unlikely(size == 0)) { + unmap_prod = unmap_q->producer_index; + prefetch(&unmap_q->unmap_array[unmap_prod + 1]); + + BNAD_PCI_UNMAP_SKB(&bnad->pcidev->dev, + unmap_q->unmap_array, + unmap_prod, unmap_q->q_depth, skb, + i); + BNAD_DROP_AND_RETURN(tx_skb_frag_zero); + } + + len += size; + if (++vect_id == BFI_TX_MAX_VECTORS_PER_WI) { vect_id = 0; if (--wi_range) @@ -2752,10 +2766,10 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) wis_used = 0; BNA_TXQ_QPGE_PTR_GET(txq_prod, tcb->sw_qpt, txqent, wi_range); - BUG_ON(!(wi_range <= tcb->q_depth)); } wis_used++; - txqent->hdr.wi_ext.opcode = htons(BNA_TXQ_WI_EXTENSION); + txqent->hdr.wi_ext.opcode = + __constant_htons(BNA_TXQ_WI_EXTENSION); } BUG_ON(!(size <= BFI_TX_MAX_DATA_PER_VECTOR)); @@ -2768,6 +2782,16 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) BNA_QE_INDX_ADD(unmap_prod, 1, unmap_q->q_depth); } + if (unlikely(len != skb->len)) { + unmap_prod = unmap_q->producer_index; + prefetch(&unmap_q->unmap_array[unmap_prod + 1]); + + BNAD_PCI_UNMAP_SKB(&bnad->pcidev->dev, unmap_q->unmap_array, + unmap_prod, unmap_q->q_depth, skb, + skb_shinfo(skb)->nr_frags); + BNAD_DROP_AND_RETURN(tx_skb_len_mismatch); + } + unmap_q->producer_index = unmap_prod; BNA_QE_INDX_ADD(txq_prod, wis_used, tcb->q_depth); tcb->producer_index = txq_prod; @@ -2778,6 +2802,7 @@ bnad_start_xmit(struct sk_buff *skb, struct net_device *netdev) return NETDEV_TX_OK; bna_txq_prod_indx_doorbell(tcb); + smp_mb(); if ((u16) (*tcb->hw_consumer_index) != tcb->consumer_index) tasklet_schedule(&bnad->tx_free_tasklet); @@ -2888,6 +2913,9 @@ bnad_set_rx_mode(struct net_device *netdev) } } + if (bnad->rx_info[0].rx == NULL) + goto unlock; + bna_rx_mode_set(bnad->rx_info[0].rx, new_mask, valid_mask, NULL); if (!netdev_mc_empty(netdev)) { @@ -3040,12 +3068,9 @@ bnad_netpoll(struct net_device *netdev) continue; for (j = 0; j < bnad->num_rxp_per_rx; j++) { rx_ctrl = &rx_info->rx_ctrl[j]; - if (rx_ctrl->ccb) { - bnad_disable_rx_irq(bnad, - rx_ctrl->ccb); + if (rx_ctrl->ccb) bnad_netif_rx_schedule_poll(bnad, rx_ctrl->ccb); - } } } } diff --git a/drivers/net/ethernet/brocade/bna/bnad.h b/drivers/net/ethernet/brocade/bna/bnad.h index 6074c6d..f402149 100644 --- a/drivers/net/ethernet/brocade/bna/bnad.h +++ b/drivers/net/ethernet/brocade/bna/bnad.h @@ -55,6 +55,11 @@ struct bnad_rx_ctrl { struct bnad *bnad; unsigned long flags; struct napi_struct napi; + u64 rx_intr_ctr; + u64 rx_poll_ctr; + u64 rx_schedule; + u64 rx_keep_poll; + u64 rx_complete; }; #define BNAD_RXMODE_PROMISC_DEFAULT BNA_RXMODE_PROMISC @@ -150,8 +155,20 @@ struct bnad_drv_stats { u64 udpcsum_offload; u64 csum_help; u64 csum_help_err; + u64 tx_skb_too_short; u64 tx_skb_stopping; u64 tx_skb_max_vectors; + u64 tx_skb_mss_too_long; + u64 tx_skb_tso_too_short; + u64 tx_skb_tso_prepare; + u64 tx_skb_non_tso_too_long; + u64 tx_skb_tcp_hdr; + u64 tx_skb_udp_hdr; + u64 tx_skb_csum_err; + u64 tx_skb_headlen_too_long; + u64 tx_skb_headlen_zero; + u64 tx_skb_frag_zero; + u64 tx_skb_len_mismatch; u64 hw_stats_updates; u64 netif_rx_schedule; @@ -359,7 +376,7 @@ extern void bnad_netdev_hwstats_fill(struct bnad *bnad, #define bnad_enable_rx_irq_unsafe(_ccb) \ { \ - if (likely(test_bit(BNAD_RXQ_STARTED, &ccb->rcb[0]->flags))) {\ + if (likely(test_bit(BNAD_RXQ_STARTED, &(_ccb)->rcb[0]->flags))) {\ bna_ib_coalescing_timer_set((_ccb)->i_dbell, \ (_ccb)->rx_coalescing_timeo); \ bna_ib_ack((_ccb)->i_dbell, 0); \ @@ -369,5 +386,19 @@ extern void bnad_netdev_hwstats_fill(struct bnad *bnad, #define bnad_dim_timer_running(_bnad) \ (((_bnad)->cfg_flags & BNAD_CF_DIM_ENABLED) && \ (test_bit(BNAD_RF_DIM_TIMER_RUNNING, &((_bnad)->run_flags)))) +#define bnad_dim_timer_stop(_bnad, _flags) \ +do { \ + int to_del = 0; \ + \ + if ((_bnad)->cfg_flags & BNAD_CF_DIM_ENABLED && \ + test_bit(BNAD_RF_DIM_TIMER_RUNNING, &(_bnad)->run_flags)) {\ + clear_bit(BNAD_RF_DIM_TIMER_RUNNING, &(_bnad)->run_flags);\ + to_del = 1; \ + } \ + spin_unlock_irqrestore(&(_bnad)->bna_lock, (_flags)); \ + if (to_del) \ + del_timer_sync(&(_bnad)->dim_timer); \ + spin_lock_irqsave(&(_bnad)->bna_lock, (_flags));\ +} while (0) #endif /* __BNAD_H__ */ -- 1.7.1