[PATCH] net/mlx5: fix age position in hairpin split

[PATCH] net/mlx5: fix age position in hairpin split

[Bing Zhao]

When splitting a hairpin rule implicitly, the count action will
be on either Tx or Rx subflow based on the encapsulation checking.

Once there is a flow rule with both count and age action, one counter
will be reused. If there is only age action and the ASO flow hit is
supported, the flow hit will be chosen instead of a counter.

In the previous flow splitting, the age would always be in the Rx
part, while the count would be on the Tx part when there is an encap.

Before this commit, 2 issues can be observed with a hairpin split:
  1. On the root table, one counter was used on both Rx and Tx parts
     for age and count actions. Then one ingress packet will be
     counted twice.
  2. On the non-root table, an extra ASO flow hit was used on the Rx
     part. This would cause some overhead.

The age and count actions should be in the same subflow instead of 2.

Fixes: daed4b6e3db2 ("net/mlx5: use aging by counter when counter exists")
Cc: michaelba@nvidia.com
Cc: stable@dpdk.org

Signed-off-by: Bing Zhao <bingz@nvidia.com>
Acked-by: Ori Kam <orika@nvidia.com>
---
 drivers/net/mlx5/mlx5_flow.c    | 1 +
 drivers/net/mlx5/mlx5_flow_dv.c | 3 +--
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 6484874c35..f31fdfbf3d 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -5399,6 +5399,7 @@ flow_hairpin_split(struct rte_eth_dev *dev,
 			}
 			break;
 		case RTE_FLOW_ACTION_TYPE_COUNT:
+		case RTE_FLOW_ACTION_TYPE_AGE:
 			if (encap) {
 				rte_memcpy(actions_tx, actions,
 					   sizeof(struct rte_flow_action));
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 18f09b22be..df23c8eb6c 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -19363,8 +19363,7 @@ flow_dv_get_aged_flows(struct rte_eth_dev *dev,
 	LIST_FOREACH(act, &age_info->aged_aso, next) {
 		nb_flows++;
 		if (nb_contexts) {
-			context[nb_flows - 1] =
-						act->age_params.context;
+			context[nb_flows - 1] = act->age_params.context;
 			if (!(--nb_contexts))
 				break;
 		}
-- 
2.34.1

RE: [PATCH] net/mlx5: fix age position in hairpin split

[Raslan Darawsheh]

Hi,

> -----Original Message-----
> From: Bing Zhao <bingz@nvidia.com>
> Sent: Thursday, March 7, 2024 10:09 AM
> To: Slava Ovsiienko <viacheslavo@nvidia.com>; dev@dpdk.org; Raslan
> Darawsheh <rasland@nvidia.com>
> Cc: Ori Kam <orika@nvidia.com>; Dariusz Sosnowski
> <dsosnowski@nvidia.com>; Suanming Mou <suanmingm@nvidia.com>;
> Matan Azrad <matan@nvidia.com>; Michael Baum <michaelba@nvidia.com>;
> stable@dpdk.org
> Subject: [PATCH] net/mlx5: fix age position in hairpin split
> 
> When splitting a hairpin rule implicitly, the count action will be on either Tx or
> Rx subflow based on the encapsulation checking.
> 
> Once there is a flow rule with both count and age action, one counter will be
> reused. If there is only age action and the ASO flow hit is supported, the flow
> hit will be chosen instead of a counter.
> 
> In the previous flow splitting, the age would always be in the Rx part, while the
> count would be on the Tx part when there is an encap.
> 
> Before this commit, 2 issues can be observed with a hairpin split:
>   1. On the root table, one counter was used on both Rx and Tx parts
>      for age and count actions. Then one ingress packet will be
>      counted twice.
>   2. On the non-root table, an extra ASO flow hit was used on the Rx
>      part. This would cause some overhead.
> 
> The age and count actions should be in the same subflow instead of 2.
> 
> Fixes: daed4b6e3db2 ("net/mlx5: use aging by counter when counter exists")
> Cc: michaelba@nvidia.com
> Cc: stable@dpdk.org
> 
> Signed-off-by: Bing Zhao <bingz@nvidia.com>
> Acked-by: Ori Kam <orika@nvidia.com>
Patch applied to next-net-mlx,

Kindest regards
Raslan Darawsheh