From mboxrd@z Thu Jan 1 00:00:00 1970 From: Cong Wang Subject: [Patch net v2] mlx5: fixup checksum for short ethernet frame padding Date: Tue, 27 Nov 2018 22:10:13 -0800 Message-ID: <20181128061013.3885-1-xiyou.wangcong@gmail.com> Mime-Version: 1.0 Content-Transfer-Encoding: 8bit Cc: Cong Wang , Saeed Mahameed , Eric Dumazet To: netdev@vger.kernel.org Return-path: Received: from mail-pl1-f196.google.com ([209.85.214.196]:45966 "EHLO mail-pl1-f196.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1727107AbeK1RLG (ORCPT ); Wed, 28 Nov 2018 12:11:06 -0500 Received: by mail-pl1-f196.google.com with SMTP id a14so16836689plm.12 for ; Tue, 27 Nov 2018 22:10:38 -0800 (PST) Sender: netdev-owner@vger.kernel.org List-ID: When an ethernet frame is padded to meet the minimum ethernet frame size, the padding octets are not covered by the hardware checksum. Fortunately the padding octets are ususally zero's, which don't affect checksum. However, we have a switch which pads non-zero octets, this causes kernel hardware checksum fault repeatedly. Prior to commit 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends"), skb checksum is forced to be CHECKSUM_NONE when padding is detected. After it, we need to keep skb->csum updated, like what we do for FCS. The logic is a bit complicated when dealing with both FCS and padding, so I wrap it up in a helper function mlx5e_csum_padding(). I tested this patch with RXFCS on and off, it works fine without any warning in both cases. Fixes: 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends"), Cc: Saeed Mahameed Cc: Eric Dumazet Signed-off-by: Cong Wang --- .../net/ethernet/mellanox/mlx5/core/en_rx.c | 39 +++++++++++++++++++ 1 file changed, 39 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c index 16985ca3248d..9b6bd2b51556 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c @@ -732,6 +732,37 @@ static u8 get_ip_proto(struct sk_buff *skb, __be16 proto) ((struct ipv6hdr *)ip_p)->nexthdr; } +static void mlx5e_csum_padding(struct sk_buff *skb, int network_depth, + __be16 proto, bool has_fcs) +{ + u32 frame_len = has_fcs ? skb->len - ETH_FCS_LEN : skb->len; + u32 pad_offset, pad_len; + void *pad, *ip_p; + + /* We only handle short frames here, and we know they are linear. */ + if (likely(frame_len > ETH_ZLEN)) + return; + + ip_p = skb->data + network_depth; + if (proto == htons(ETH_P_IP)) { + struct iphdr *ipv4 = ip_p; + + pad_offset = network_depth + be16_to_cpu(ipv4->tot_len); + } else { /* Either IPv4 or IPv6 here. */ + struct ipv6hdr *ipv6 = ip_p; + + pad_offset = network_depth + sizeof(struct ipv6hdr) + + be16_to_cpu(ipv6->payload_len); + } + pad_len = frame_len - pad_offset; + if (pad_len == 0) + return; + + pad = skb->data + pad_offset; + skb->csum = csum_block_add(skb->csum, csum_partial(pad, pad_len, 0), + pad_offset); +} + static inline void mlx5e_handle_csum(struct net_device *netdev, struct mlx5_cqe64 *cqe, struct mlx5e_rq *rq, @@ -772,6 +803,14 @@ static inline void mlx5e_handle_csum(struct net_device *netdev, skb->csum = csum_block_add(skb->csum, (__force __wsum)mlx5e_get_fcs(skb), skb->len - ETH_FCS_LEN); + + /* CQE csum doesn't cover padding octets in short ethernet + * frames. And the pad field is appended prior to calculating + * and appending the FCS field. + */ + mlx5e_csum_padding(skb, network_depth, proto, + !!(netdev->features & NETIF_F_RXFCS)); + stats->csum_complete++; return; } -- 2.19.1