On 30/03/2018 3:34 AM, Saeed Mahameed wrote:
On Tue, Mar 27, 2018 at 2:21 PM, Eric Dumazet <eduma...@google.com> wrote:
Refine the RX check summing handling to propagate the
hardware provided checksum so that we do not have to
compute it later in software.

Signed-off-by: Eric Dumazet <eduma...@google.com>
Cc: Willem de Bruijn <will...@google.com>
Cc: Tariq Toukan <tar...@mellanox.com>
---
  drivers/net/ethernet/mellanox/mlx4/en_rx.c | 10 ++++------
  1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c 
b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 
05787efef492b1c0c6ce540ef73647fad91ce282..5c613c6663da51a4ae792eeb4d8956b54655786b
 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -821,14 +821,12 @@ 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)) {
-                       if (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_TCP |
-                                                     MLX4_CQE_STATUS_UDP)) {
+                       if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_TCP |
+                                                      MLX4_CQE_STATUS_UDP)) &&
+                           (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&
+                           cqe->checksum == cpu_to_be16(0xffff)) {
                                 bool l2_tunnel;


LGTM, this code even aligns better with the mlx4 HW documentation:

"When L4_CSUM field is not supported, L4 checksum for TCP/UDP packets
can be validated by: (IP_OK && (TCP || UDP)) && (checksum ==
0xFFFF))."


Hi,
Thanks Eric for your patch.
Thanks Saeed for your review while I'm off-work.

In IP fragments, MLX4_CQE_STATUS_IPOK is not set in cqe->status.
This patch suggests falling back to CHECKSUM_COMPLETE instead of CHECKSUM_NONE.
Looks good to me.

in the code we don't even consider L4_CSUM at the moment, As a future
patch, it could be a nice acceleration for the above 3 steps
condition.


Right. Good idea. I'll add it to plans.

Small comment, if we expect that  cqe->checksum is NOT likely to be
0xffff for UDP/TCP packets, maybe it is better performance wise to
move (cqe->checksum == cpu_to_be16(0xffff)) to be evaluated first in
the condition.


It _is_ likely  for common TCP/UDP (no IP fragments).
Let's keep it this way.

-                               if (!((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) 
&&
-                                     cqe->checksum == cpu_to_be16(0xffff)))
-                                       goto csum_none;
-
                                 l2_tunnel = (dev->hw_enc_features & NETIF_F_RXCSUM) 
&&
                                         (cqe->vlan_my_qpn & 
cpu_to_be32(MLX4_CQE_L2_TUNNEL));
                                 ip_summed = CHECKSUM_UNNECESSARY;
--
2.17.0.rc1.321.gba9d0f2565-goog


Acked-by: Tariq Toukan <tar...@mellanox.com>

Reply via email to