net/mlx5e: When changing XDP program without reset, take refs for XSK RQs

Each RQ (including XSK RQs) takes a reference to the XDP program. When
an XDP program is attached or detached, the channels and queues are
recreated, however, there is a special flow for changing an active XDP
program to another one. In that flow, channels and queues stay alive,
but the refcounts of the old and new XDP programs are adjusted. This
flow didn't increment refcount by the number of active XSK RQs, and this
commit fixes it.

Fixes: db05815b36cb ("net/mlx5e: Add XSK zero-copy support")
Signed-off-by: Maxim Mikityanskiy <maximmi@mellanox.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_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 66d23cd275c1fea73c210df4eeac81cca7a341ce..6b761fb8d2697ff58467d6c7948b3171253bcd33 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4695,8 +4695,10 @@ static int mlx5e_xdp_set(struct net_device *netdev, struct bpf_prog *prog)
                struct mlx5e_channel *c = priv->channels.c[i];
 
                mlx5e_rq_replace_xdp_prog(&c->rq, prog);
-               if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state))
+               if (test_bit(MLX5E_CHANNEL_STATE_XSK, c->state)) {
+                       bpf_prog_inc(prog);
                        mlx5e_rq_replace_xdp_prog(&c->xskrq, prog);
+               }
        }
 
 unlock: