net/mlx5e: Overcome slow response for first IPsec ASO WQE

First ASO WQE causes to cache miss in hardware, which can't return
result immediately. It causes to the situation where such WQE is polled
earlier than it is needed. Add logic to retry ASO CQ polling operation.

Link: https://lore.kernel.org/r/eb92a758c533ff3f058e0dcb4f8d2324355304ad.1680162300.git.leonro@nvidia.com
Reviewed-by: Raed Salem <raeds@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec_offload.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec_offload.c
index 684de97..6971e5e 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec_offload.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec_offload.c
@@ -568,6 +568,7 @@ int mlx5e_ipsec_aso_query(struct mlx5e_ipsec_sa_entry *sa_entry,
        struct mlx5_wqe_aso_ctrl_seg *ctrl;
        struct mlx5e_hw_objs *res;
        struct mlx5_aso_wqe *wqe;
+       unsigned long expires;
        u8 ds_cnt;
        int ret;
 
@@ -589,7 +590,12 @@ int mlx5e_ipsec_aso_query(struct mlx5e_ipsec_sa_entry *sa_entry,
        mlx5e_ipsec_aso_copy(ctrl, data);
 
        mlx5_aso_post_wqe(aso->aso, false, &wqe->ctrl);
-       ret = mlx5_aso_poll_cq(aso->aso, false);
+       expires = jiffies + msecs_to_jiffies(10);
+       do {
+               ret = mlx5_aso_poll_cq(aso->aso, false);
+               if (ret)
+                       usleep_range(2, 10);
+       } while (ret && time_is_after_jiffies(expires));
        spin_unlock_bh(&aso->lock);
        return ret;
 }