net/mlx5e: aRFS, Prevent repeated kernel rule migrations requests

aRFS rule movement requests from one Rx ring to other Rx ring arrive
from the kernel to ensure that packets are steered to the right Rx ring.
In the time interval until satisfying such a request, several more
requests might follow, for the same flow.

This patch detects and prevents repeated aRFS rules movement requests.

In mlx5e_rx_flow_steer() ndo, after finding the aRFS rule that have been
requested to move by the kernel, check if it's already requested to move
by calling work_busy(&arfs_rule->arfs_work) handler. IOW, if this
request is pending to be executed (in the work queue) or it's executing
now but hasn't finished yet, return current filter ID and don't issue a
new transition work.

Signed-off-by: Adham Faris <afaris@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
index 5aa51d7..67d8b19 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
@@ -740,7 +740,7 @@ int mlx5e_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
        spin_lock_bh(&arfs->arfs_lock);
        arfs_rule = arfs_find_rule(arfs_t, &fk);
        if (arfs_rule) {
-               if (arfs_rule->rxq == rxq_index) {
+               if (arfs_rule->rxq == rxq_index || work_busy(&arfs_rule->arfs_work)) {
                        spin_unlock_bh(&arfs->arfs_lock);
                        return arfs_rule->filter_id;
                }