diff mbox series

[net,V2] net/mlx4_en: Force CHECKSUM_NONE for short ethernet frames

Message ID 1549901057-2614-1-git-send-email-tariqt@mellanox.com
State Accepted
Delegated to: David Miller
Headers show
Series [net,V2] net/mlx4_en: Force CHECKSUM_NONE for short ethernet frames | expand

Commit Message

Tariq Toukan Feb. 11, 2019, 4:04 p.m. UTC 
From: Saeed Mahameed <saeedm@mellanox.com>

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 usually zero's, which don't affect
checksum. However, it is not guaranteed. For example, switches might
choose to make other use of these octets.
This repeatedly causes kernel hardware checksum fault.

Prior to the cited commit below, skb checksum was forced to be
CHECKSUM_NONE when padding is detected. After it, we need to keep
skb->csum updated. However, fixing up CHECKSUM_COMPLETE requires to
verify and parse IP headers, it does not worth the effort as the packets
are so small that CHECKSUM_COMPLETE has no significant advantage.

Future work: when reporting checksum complete is not an option for
IP non-TCP/UDP packets, we can actually fallback to report checksum
unnecessary, by looking at cqe IPOK bit.

Fixes: 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends")
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx4/en_rx.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

Please queue for -stable >= v4.18.

v2: Changed the comment per the mailing list discussion.

Comments

Eric Dumazet Feb. 11, 2019, 5:24 p.m. UTC | #1 
On 02/11/2019 08:04 AM, Tariq Toukan wrote:
> From: Saeed Mahameed <saeedm@mellanox.com>
> 
> 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 usually zero's, which don't affect
> checksum. However, it is not guaranteed. For example, switches might
> choose to make other use of these octets.
> This repeatedly causes kernel hardware checksum fault.
> 
> Prior to the cited commit below, skb checksum was forced to be
> CHECKSUM_NONE when padding is detected. After it, we need to keep
> skb->csum updated. However, fixing up CHECKSUM_COMPLETE requires to
> verify and parse IP headers, it does not worth the effort as the packets
> are so small that CHECKSUM_COMPLETE has no significant advantage.
> 
> Future work: when reporting checksum complete is not an option for
> IP non-TCP/UDP packets, we can actually fallback to report checksum
> unnecessary, by looking at cqe IPOK bit.
> 
> Fixes: 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends")
> Cc: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
> Signed-off-by: Tariq Toukan <tariqt@mellanox.com>

Thanks Tariq and Saeed

Reviewed-by: Eric Dumazet <edumazet@google.com>
David Miller Feb. 12, 2019, 5:28 p.m. UTC | #2 
From: Tariq Toukan <tariqt@mellanox.com>
Date: Mon, 11 Feb 2019 18:04:17 +0200

> From: Saeed Mahameed <saeedm@mellanox.com>
> 
> 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 usually zero's, which don't affect
> checksum. However, it is not guaranteed. For example, switches might
> choose to make other use of these octets.
> This repeatedly causes kernel hardware checksum fault.
> 
> Prior to the cited commit below, skb checksum was forced to be
> CHECKSUM_NONE when padding is detected. After it, we need to keep
> skb->csum updated. However, fixing up CHECKSUM_COMPLETE requires to
> verify and parse IP headers, it does not worth the effort as the packets
> are so small that CHECKSUM_COMPLETE has no significant advantage.
> 
> Future work: when reporting checksum complete is not an option for
> IP non-TCP/UDP packets, we can actually fallback to report checksum
> unnecessary, by looking at cqe IPOK bit.
> 
> Fixes: 88078d98d1bb ("net: pskb_trim_rcsum() and CHECKSUM_COMPLETE are friends")
> Cc: Eric Dumazet <edumazet@google.com>
> Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
> Signed-off-by: Tariq Toukan <tariqt@mellanox.com>

Applied and queued up for -stable.
diff mbox series

Patch

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 9a0881cb7f51..6c01314e87b0 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -617,6 +617,8 @@  static int get_fixed_ipv6_csum(__wsum hw_checksum, struct sk_buff *skb,
 }
 #endif
 
+#define short_frame(size) ((size) <= ETH_ZLEN + ETH_FCS_LEN)
+
 /* We reach this function only after checking that any of
  * the (IPv4 | IPv6) bits are set in cqe->status.
  */
@@ -624,9 +626,20 @@  static int check_csum(struct mlx4_cqe *cqe, struct sk_buff *skb, void *va,
 		      netdev_features_t dev_features)
 {
 	__wsum hw_checksum = 0;
+	void *hdr;
+
+	/* 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.
+	 *
+	 * Detecting these padded frames requires to verify and parse
+	 * IP headers, so we simply force all those small frames to skip
+	 * checksum complete.
+	 */
+	if (short_frame(skb->len))
+		return -EINVAL;
 
-	void *hdr = (u8 *)va + sizeof(struct ethhdr);
-
+	hdr = (u8 *)va + sizeof(struct ethhdr);
 	hw_checksum = csum_unfold((__force __sum16)cqe->checksum);
 
 	if (cqe->vlan_my_qpn & cpu_to_be32(MLX4_CQE_CVLAN_PRESENT_MASK) &&
@@ -819,6 +832,11 @@  int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 		skb_record_rx_queue(skb, cq_ring);
 
 		if (likely(dev->features & NETIF_F_RXCSUM)) {
+			/* TODO: For IP non TCP/UDP packets when csum complete is
+			 * not an option (not supported or any other reason) we can
+			 * actually check cqe IPOK status bit and report
+			 * CHECKSUM_UNNECESSARY rather than CHECKSUM_NONE
+			 */
 			if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_TCP |
 						       MLX4_CQE_STATUS_UDP)) &&
 			    (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&