From 728a9972d1f290608e730b9ccec2061aa81f6609 Mon Sep 17 00:00:00 2001 From: Ajit Khaparde Date: Mon, 13 Apr 2009 15:41:22 -0700 Subject: be2net: changes for checksum and segmentation offload - Enables Rx checksum feature by default. - Disables support for ipv6 tso. - Changes in Rx path to handle Rx completions with various checksum options. Signed-off-by: Ajit Khaparde Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 28 ++++++++++++++++++++++------ 1 file changed, 22 insertions(+), 6 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 9b75aa630062..8994b03d80ac 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -637,6 +637,22 @@ static void be_rx_stats_update(struct be_adapter *adapter, stats->be_rx_bytes += pktsize; } +static inline bool do_pkt_csum(struct be_eth_rx_compl *rxcp, bool cso) +{ + u8 l4_cksm, ip_version, ipcksm, tcpf = 0, udpf = 0, ipv6_chk; + + l4_cksm = AMAP_GET_BITS(struct amap_eth_rx_compl, l4_cksm, rxcp); + ipcksm = AMAP_GET_BITS(struct amap_eth_rx_compl, ipcksm, rxcp); + ip_version = AMAP_GET_BITS(struct amap_eth_rx_compl, ip_version, rxcp); + if (ip_version) { + tcpf = AMAP_GET_BITS(struct amap_eth_rx_compl, tcpf, rxcp); + udpf = AMAP_GET_BITS(struct amap_eth_rx_compl, udpf, rxcp); + } + ipv6_chk = (ip_version && (tcpf || udpf)); + + return ((l4_cksm && ipv6_chk && ipcksm) && cso) ? false : true; +} + static struct be_rx_page_info * get_rx_page_info(struct be_adapter *adapter, u16 frag_idx) { @@ -752,9 +768,7 @@ static void be_rx_compl_process(struct be_adapter *adapter, { struct sk_buff *skb; u32 vtp, vid; - int l4_cksm; - l4_cksm = AMAP_GET_BITS(struct amap_eth_rx_compl, l4_cksm, rxcp); vtp = AMAP_GET_BITS(struct amap_eth_rx_compl, vtp, rxcp); skb = netdev_alloc_skb(adapter->netdev, BE_HDR_LEN + NET_IP_ALIGN); @@ -769,10 +783,10 @@ static void be_rx_compl_process(struct be_adapter *adapter, skb_fill_rx_data(adapter, skb, rxcp); - if (l4_cksm && adapter->rx_csum) - skb->ip_summed = CHECKSUM_UNNECESSARY; - else + if (do_pkt_csum(rxcp, adapter->rx_csum)) skb->ip_summed = CHECKSUM_NONE; + else + skb->ip_summed = CHECKSUM_UNNECESSARY; skb->truesize = skb->len + sizeof(struct sk_buff); skb->protocol = eth_type_trans(skb, adapter->netdev); @@ -1626,10 +1640,12 @@ static void be_netdev_init(struct net_device *netdev) netdev->features |= NETIF_F_SG | NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_FILTER | NETIF_F_IP_CSUM | - NETIF_F_IPV6_CSUM | NETIF_F_TSO6; + NETIF_F_IPV6_CSUM; netdev->flags |= IFF_MULTICAST; + adapter->rx_csum = true; + BE_SET_NETDEV_OPS(netdev, &be_netdev_ops); SET_ETHTOOL_OPS(netdev, &be_ethtool_ops); -- cgit v1.2.3 From 28679751a924c11f7135641f26e99249385de5b4 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Wed, 27 May 2009 19:26:37 +0000 Subject: net: dont update dev->trans_start in 10GB drivers Followup of commits 9d21493b4beb8f918ba248032fefa393074a5e2b and 08baf561083bc27a953aa087dd8a664bb2b88e8e (net: tx scalability works : trans_start) (net: txq_trans_update() helper) Now that core network takes care of trans_start updates, dont do it in drivers themselves, if possible. Multi queue drivers can avoid one cache miss (on dev->trans_start) in their start_xmit() handler. Exceptions are NETIF_F_LLTX drivers (vxge & tehuti) Signed-off-by: Eric Dumazet Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 2 -- 1 file changed, 2 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index ae2f6b58ba25..5f17d80300ae 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -478,8 +478,6 @@ static int be_xmit(struct sk_buff *skb, struct net_device *netdev) be_txq_notify(&adapter->ctrl, txq->id, wrb_cnt); - netdev->trans_start = jiffies; - be_tx_stats_update(adapter, wrb_cnt, copied, stopped); return NETDEV_TX_OK; } -- cgit v1.2.3 From ebc8d2ab61dde6cf775ae7bb1ed9e38dfe12ca65 Mon Sep 17 00:00:00 2001 From: "David S. Miller" Date: Tue, 9 Jun 2009 01:01:31 -0700 Subject: be2net: Remove unnecessary frag list traversing. This driver does not indicate support for frag lists. Furthermore, even if it did, the code is walking the frag lists incorrectly. The idiom is: for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) but it's doing: for (iter = skb_shinfo(skb)->frag_list; iter; iter = skb_shinfo(iter)->frag_list) which would never work. And this proves that this driver never saw an SKB with active frag lists. So just remove the code altogether and the driver TX path becomes much simpler. Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 56 ++++++++++++++++++++------------------------- 1 file changed, 25 insertions(+), 31 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 5f17d80300ae..347d002fa425 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -337,13 +337,10 @@ static void be_tx_stats_update(struct be_adapter *adapter, /* Determine number of WRB entries needed to xmit data in an skb */ static u32 wrb_cnt_for_skb(struct sk_buff *skb, bool *dummy) { - int cnt = 0; - while (skb) { - if (skb->len > skb->data_len) - cnt++; - cnt += skb_shinfo(skb)->nr_frags; - skb = skb_shinfo(skb)->frag_list; - } + int cnt = (skb->len > skb->data_len); + + cnt += skb_shinfo(skb)->nr_frags; + /* to account for hdr wrb */ cnt++; if (cnt & 1) { @@ -409,31 +406,28 @@ static int make_tx_wrbs(struct be_adapter *adapter, hdr = queue_head_node(txq); queue_head_inc(txq); - while (skb) { - if (skb->len > skb->data_len) { - int len = skb->len - skb->data_len; - busaddr = pci_map_single(pdev, skb->data, len, - PCI_DMA_TODEVICE); - wrb = queue_head_node(txq); - wrb_fill(wrb, busaddr, len); - be_dws_cpu_to_le(wrb, sizeof(*wrb)); - queue_head_inc(txq); - copied += len; - } + if (skb->len > skb->data_len) { + int len = skb->len - skb->data_len; + busaddr = pci_map_single(pdev, skb->data, len, + PCI_DMA_TODEVICE); + wrb = queue_head_node(txq); + wrb_fill(wrb, busaddr, len); + be_dws_cpu_to_le(wrb, sizeof(*wrb)); + queue_head_inc(txq); + copied += len; + } - for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { - struct skb_frag_struct *frag = - &skb_shinfo(skb)->frags[i]; - busaddr = pci_map_page(pdev, frag->page, - frag->page_offset, - frag->size, PCI_DMA_TODEVICE); - wrb = queue_head_node(txq); - wrb_fill(wrb, busaddr, frag->size); - be_dws_cpu_to_le(wrb, sizeof(*wrb)); - queue_head_inc(txq); - copied += frag->size; - } - skb = skb_shinfo(skb)->frag_list; + for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { + struct skb_frag_struct *frag = + &skb_shinfo(skb)->frags[i]; + busaddr = pci_map_page(pdev, frag->page, + frag->page_offset, + frag->size, PCI_DMA_TODEVICE); + wrb = queue_head_node(txq); + wrb_fill(wrb, busaddr, frag->size); + be_dws_cpu_to_le(wrb, sizeof(*wrb)); + queue_head_inc(txq); + copied += frag->size; } if (dummy_wrb) { -- cgit v1.2.3 From b305be78a044c5f6a9d146229a280d08db0c630a Mon Sep 17 00:00:00 2001 From: Sathya Perla Date: Wed, 10 Jun 2009 02:18:35 +0000 Subject: be2net: Use cancel_delayed_work_sync instead of cancel_delayed_work() Use cancel_delayed_work_sycn instead of cancel_delayed_work() to reliably kill be_worker() as it rearms itself. Signed-off-by: Sathya Perla Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 347d002fa425..69046a109ac3 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -1533,7 +1533,7 @@ static int be_close(struct net_device *netdev) struct be_eq_obj *tx_eq = &adapter->tx_eq; int vec; - cancel_delayed_work(&adapter->work); + cancel_delayed_work_sync(&adapter->work); netif_stop_queue(netdev); netif_carrier_off(netdev); -- cgit v1.2.3 From 6811086899f2740c08d0ade26f8b9d705708e0cc Mon Sep 17 00:00:00 2001 From: Sathya Perla Date: Wed, 10 Jun 2009 02:21:16 +0000 Subject: be2net: fix netdev stats rx_errors and rx_dropped Fix netdev stat rx_errors to cover length related errors and checksum errors and rx_dropped to the pkts dropped due to lack of buffers Signed-off-by: Sathya Perla Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 43 +++++++++++++++++++++++++++++-------------- 1 file changed, 29 insertions(+), 14 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 69046a109ac3..6487eb4bad10 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -168,6 +168,7 @@ static void netdev_stats_update(struct be_adapter *adapter) struct be_port_rxf_stats *port_stats = &rxf_stats->port[adapter->port_num]; struct net_device_stats *dev_stats = &adapter->stats.net_stats; + struct be_erx_stats *erx_stats = &hw_stats->erx; dev_stats->rx_packets = port_stats->rx_total_frames; dev_stats->tx_packets = port_stats->tx_unicastframes + @@ -181,29 +182,33 @@ static void netdev_stats_update(struct be_adapter *adapter) dev_stats->rx_errors = port_stats->rx_crc_errors + port_stats->rx_alignment_symbol_errors + port_stats->rx_in_range_errors + - port_stats->rx_out_range_errors + port_stats->rx_frame_too_long; - - /* packet transmit problems */ - dev_stats->tx_errors = 0; - - /* no space in linux buffers */ - dev_stats->rx_dropped = 0; - - /* no space available in linux */ - dev_stats->tx_dropped = 0; - - dev_stats->multicast = port_stats->tx_multicastframes; - dev_stats->collisions = 0; + port_stats->rx_out_range_errors + + port_stats->rx_frame_too_long + + port_stats->rx_dropped_too_small + + port_stats->rx_dropped_too_short + + port_stats->rx_dropped_header_too_small + + port_stats->rx_dropped_tcp_length + + port_stats->rx_dropped_runt + + port_stats->rx_tcp_checksum_errs + + port_stats->rx_ip_checksum_errs + + port_stats->rx_udp_checksum_errs; + + /* no space in linux buffers: best possible approximation */ + dev_stats->rx_dropped = erx_stats->rx_drops_no_fragments[0]; /* detailed rx errors */ dev_stats->rx_length_errors = port_stats->rx_in_range_errors + - port_stats->rx_out_range_errors + port_stats->rx_frame_too_long; + port_stats->rx_out_range_errors + + port_stats->rx_frame_too_long; + /* receive ring buffer overflow */ dev_stats->rx_over_errors = 0; + dev_stats->rx_crc_errors = port_stats->rx_crc_errors; /* frame alignment errors */ dev_stats->rx_frame_errors = port_stats->rx_alignment_symbol_errors; + /* receiver fifo overrun */ /* drops_no_pbuf is no per i/f, it's per BE card */ dev_stats->rx_fifo_errors = port_stats->rx_fifo_overflow + @@ -211,6 +216,16 @@ static void netdev_stats_update(struct be_adapter *adapter) rxf_stats->rx_drops_no_pbuf; /* receiver missed packetd */ dev_stats->rx_missed_errors = 0; + + /* packet transmit problems */ + dev_stats->tx_errors = 0; + + /* no space available in linux */ + dev_stats->tx_dropped = 0; + + dev_stats->multicast = port_stats->tx_multicastframes; + dev_stats->collisions = 0; + /* detailed tx_errors */ dev_stats->tx_aborted_errors = 0; dev_stats->tx_carrier_errors = 0; -- cgit v1.2.3 From 76fbb42919396b9bf68c9a03ceb037c971a02e2f Mon Sep 17 00:00:00 2001 From: Sathya Perla Date: Wed, 10 Jun 2009 02:21:56 +0000 Subject: be2net: Fix rx stats updation in non-lro path rx stats are not getting updated when an rx_compl with only one frag is rcvd in non-lro path. Signed-off-by: Sathya Perla Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 6487eb4bad10..5aadd71d4613 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -743,7 +743,7 @@ static void skb_fill_rx_data(struct be_adapter *adapter, if (pktsize <= rx_frag_size) { BUG_ON(num_rcvd != 1); - return; + goto done; } /* More frags present for this completion */ @@ -765,6 +765,7 @@ static void skb_fill_rx_data(struct be_adapter *adapter, memset(page_info, 0, sizeof(*page_info)); } +done: be_rx_stats_update(adapter, pktsize, num_rcvd); return; } -- cgit v1.2.3 From a7a0ef31def6b6badd94fc96c8f17c2e18d91513 Mon Sep 17 00:00:00 2001 From: Sathya Perla Date: Wed, 10 Jun 2009 02:23:28 +0000 Subject: be2net: Fix early reset of rx-completion be_rx_compl_get() must not reset(via the valid word) the rx_compl as the rx_compl is not processed yet; it must be reset after it is processed. Signed-off-by: Sathya Perla Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 5aadd71d4613..86f70af93d60 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -876,12 +876,19 @@ static struct be_eth_rx_compl *be_rx_compl_get(struct be_adapter *adapter) be_dws_le_to_cpu(rxcp, sizeof(*rxcp)); - rxcp->dw[offsetof(struct amap_eth_rx_compl, valid) / 32] = 0; - queue_tail_inc(&adapter->rx_obj.cq); return rxcp; } +/* To reset the valid bit, we need to reset the whole word as + * when walking the queue the valid entries are little-endian + * and invalid entries are host endian + */ +static inline void be_rx_compl_reset(struct be_eth_rx_compl *rxcp) +{ + rxcp->dw[offsetof(struct amap_eth_rx_compl, valid) / 32] = 0; +} + static inline struct page *be_alloc_pages(u32 size) { gfp_t alloc_flags = GFP_ATOMIC; @@ -1013,6 +1020,7 @@ static void be_rx_q_clean(struct be_adapter *adapter) /* First cleanup pending rx completions */ while ((rxcp = be_rx_compl_get(adapter)) != NULL) { be_rx_compl_discard(adapter, rxcp); + be_rx_compl_reset(rxcp); be_cq_notify(&adapter->ctrl, rx_cq->id, true, 1); } @@ -1294,6 +1302,8 @@ int be_poll_rx(struct napi_struct *napi, int budget) be_rx_compl_process_lro(adapter, rxcp); else be_rx_compl_process(adapter, rxcp); + + be_rx_compl_reset(rxcp); } lro_flush_all(&adapter->rx_obj.lro_mgr); -- cgit v1.2.3 From 934037bc2ed29a94bbde72aa6a2e66bdc5861b98 Mon Sep 17 00:00:00 2001 From: Sathya Perla Date: Wed, 10 Jun 2009 02:23:50 +0000 Subject: be2net: Fix be_tx_q_clean() being called on freed queues In the tx queue destroy path, be_tx_q_clean() is currently called after the tx queues are freed; it must be called before. Signed-off-by: Sathya Perla Signed-off-by: David S. Miller --- drivers/net/benet/be_main.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) (limited to 'drivers/net/benet/be_main.c') diff --git a/drivers/net/benet/be_main.c b/drivers/net/benet/be_main.c index 86f70af93d60..66bb56874d9b 100644 --- a/drivers/net/benet/be_main.c +++ b/drivers/net/benet/be_main.c @@ -1056,8 +1056,13 @@ static void be_tx_queues_destroy(struct be_adapter *adapter) struct be_queue_info *q; q = &adapter->tx_obj.q; - if (q->created) + if (q->created) { be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_TXQ); + + /* No more tx completions can be rcvd now; clean up if there + * are any pending completions or pending tx requests */ + be_tx_q_clean(adapter); + } be_queue_free(adapter, q); q = &adapter->tx_obj.cq; @@ -1065,10 +1070,6 @@ static void be_tx_queues_destroy(struct be_adapter *adapter) be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_CQ); be_queue_free(adapter, q); - /* No more tx completions can be rcvd now; clean up if there are - * any pending completions or pending tx requests */ - be_tx_q_clean(adapter); - q = &adapter->tx_eq.q; if (q->created) be_cmd_q_destroy(&adapter->ctrl, q, QTYPE_EQ); -- cgit v1.2.3