mlxbf_gige: fix receive packet race condition

Under heavy traffic, the BlueField Gigabit interface can
become unresponsive. This is due to a possible race condition
in the mlxbf_gige_rx_packet function, where the function exits
with producer and consumer indices equal but there are remaining
packet(s) to be processed. In order to prevent this situation,
read receive consumer index *before* the HW replenish so that
the mlxbf_gige_rx_packet function returns an accurate return
value even if a packet is received into just-replenished buffer
prior to exiting this routine. If the just-replenished buffer
is received and occupies the last RX ring entry, the interface
would not recover and instead would encounter RX packet drops
related to internal buffer shortages since the driver RX logic
is not being triggered to drain the RX ring. This patch will
address and prevent this "ring full" condition.

Fixes: f92e1869d74e ("Add Mellanox BlueField Gigabit Ethernet driver")
Reviewed-by: Asmaa Mnebhi <asmaa@nvidia.com>
Signed-off-by: David Thompson <davthompson@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 7 insertions, 2 deletions

diff --git a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
index 0d5a41a2ae0109..227d01cace3f0d 100644
--- a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
+++ b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
@@ -267,6 +267,13 @@ static bool mlxbf_gige_rx_packet(struct mlxbf_gige *priv, int *rx_pkts)
 		priv->stats.rx_truncate_errors++;
 	}
 
+	/* Read receive consumer index before replenish so that this routine
+	 * returns accurate return value even if packet is received into
+	 * just-replenished buffer prior to exiting this routine.
+	 */
+	rx_ci = readq(priv->base + MLXBF_GIGE_RX_CQE_PACKET_CI);
+	rx_ci_rem = rx_ci % priv->rx_q_entries;
+
 	/* Let hardware know we've replenished one buffer */
 	rx_pi++;
 
@@ -279,8 +286,6 @@ static bool mlxbf_gige_rx_packet(struct mlxbf_gige *priv, int *rx_pkts)
 	rx_pi_rem = rx_pi % priv->rx_q_entries;
 	if (rx_pi_rem == 0)
 		priv->valid_polarity ^= 1;
-	rx_ci = readq(priv->base + MLXBF_GIGE_RX_CQE_PACKET_CI);
-	rx_ci_rem = rx_ci % priv->rx_q_entries;
 
 	if (skb)
 		netif_receive_skb(skb);