[PATCH net] mlx4: TCP/UDP packets have L4 hash

From: Eric Dumazet <redacted>

Mellanox driver has the knowledge if rxhash is a L4 hash,
if it receives a non fragmented TCP or UDP frame and
NETIF_F_RXCSUM is enabled on netdev.

ip_summed value is CHECKSUM_UNNECESSARY in this case.

Signed-off-by: Eric Dumazet <redacted>
Cc: Amir Vadai <redacted>
Cc: Ido Shamay <redacted>
---
 drivers/net/ethernet/mellanox/mlx4/en_rx.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 7a4f20bb7fcb..12c65e1ad6a9 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c

@@ -917,7 +917,9 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 			if (dev->features & NETIF_F_RXHASH)
 				skb_set_hash(gro_skb,
 					     be32_to_cpu(cqe->immed_rss_invalid),
-					     PKT_HASH_TYPE_L3);
+					     (ip_summed == CHECKSUM_UNNECESSARY) ?
+						PKT_HASH_TYPE_L4 :
+						PKT_HASH_TYPE_L3);
 
 			skb_record_rx_queue(gro_skb, cq->ring);
 			skb_mark_napi_id(gro_skb, &cq->napi);

@@ -963,7 +965,9 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 		if (dev->features & NETIF_F_RXHASH)
 			skb_set_hash(skb,
 				     be32_to_cpu(cqe->immed_rss_invalid),
-				     PKT_HASH_TYPE_L3);
+				     (ip_summed == CHECKSUM_UNNECESSARY) ?
+					PKT_HASH_TYPE_L4 :
+					PKT_HASH_TYPE_L3);
 
 		if ((be32_to_cpu(cqe->vlan_my_qpn) &
 		    MLX4_CQE_VLAN_PRESENT_MASK) &&

On 7/2/2015 2:24 PM, Eric Dumazet wrote:

Thanks Eric,

We can have a relaxation of the condition  by looking only at TCP/UDP CQE indication (without check-sum indications)
This can cover us also when device rx-checksuming feature is off.
Do we want it or a correlation between check-sum and l4_hash is needed?

Ido

On Mon, 2015-07-06 at 00:16 +0300, Ido Shamay wrote:

We can have a relaxation of the condition  by looking only at TCP/UDP 
CQE indication (without check-sum indications)
This can cover us also when device rx-checksuming feature is off.
Do we want it or a correlation between check-sum and l4_hash is needed?

I thought about that, but this was adding a more complex test in fast
path.

Not sure we should care here, as nobody would disable hardware checksum
if they care about performance.

On 7/6/2015 12:33 AM, Eric Dumazet wrote:

On Mon, 2015-07-06 at 00:16 +0300, Ido Shamay wrote:

quoted

We can have a relaxation of the condition  by looking only at TCP/UDP
CQE indication (without check-sum indications)
This can cover us also when device rx-checksuming feature is off.
Do we want it or a correlation between check-sum and l4_hash is needed?

I thought about that, but this was adding a more complex test in fast
path.

Not sure we should care here, as nobody would disable hardware checksum
if they care about performance.

I agree, thank you Eric
Acked-by: Ido Shamay <redacted>

From: Eric Dumazet <redacted>
Date: Thu, 02 Jul 2015 13:24:44 +0200

From: Eric Dumazet <redacted>

Mellanox driver has the knowledge if rxhash is a L4 hash,
if it receives a non fragmented TCP or UDP frame and
NETIF_F_RXCSUM is enabled on netdev.

ip_summed value is CHECKSUM_UNNECESSARY in this case.

Signed-off-by: Eric Dumazet <redacted>

Applied to net-next, thanks.