net/mlx5e: Sync txq2sq updates with mlx5e_xmit for HTB queues

This commit makes necessary changes to guarantee that txq2sq remains
stable while mlx5e_xmit is running. Proper synchronization is added for
HTB TX queues.

All updates to txq2sq are performed while the corresponding queue is
disabled (i.e. mlx5e_xmit doesn't run on that queue). smp_wmb after each
change guarantees that mlx5e_xmit can see the updated value after the
queue is enabled. Comments explaining this mechanism are added to
mlx5e_xmit.

When an HTB SQ can be deleted (after deleting an HTB node), synchronize
with RCU to wait for mlx5e_select_queue to finish and stop selecting
that queue, before we re-enable it to avoid TX timeout watchdog alarms.

Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

2 files changed, 33 insertions, 4 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
index c1e07496c89c..ff45840581e3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/qos.c
@@ -50,7 +50,6 @@ static int mlx5e_find_unused_qos_qid(struct mlx5e_priv *priv)
 
 struct mlx5e_qos_node {
 	struct hlist_node hnode;
-	struct rcu_head rcu;
 	struct mlx5e_qos_node *parent;
 	u64 rate;
 	u32 bw_share;
@@ -132,7 +131,11 @@ static void mlx5e_sw_node_delete(struct mlx5e_priv *priv, struct mlx5e_qos_node
 		__clear_bit(node->qid, priv->htb.qos_used_qids);
 		mlx5e_update_tx_netdev_queues(priv);
 	}
-	kfree_rcu(node, rcu);
+	/* Make sure this qid is no longer selected by mlx5e_select_queue, so
+	 * that mlx5e_reactivate_qos_sq can safely restart the netdev TX queue.
+	 */
+	synchronize_net();
+	kfree(node);
 }
 
 /* TX datapath API */
@@ -273,10 +276,18 @@ err_free_sq:
 static void mlx5e_activate_qos_sq(struct mlx5e_priv *priv, struct mlx5e_qos_node *node)
 {
 	struct mlx5e_txqsq *sq;
+	u16 qid;
 
 	sq = mlx5e_get_qos_sq(priv, node->qid);
 
-	WRITE_ONCE(priv->txq2sq[mlx5e_qid_from_qos(&priv->channels, node->qid)], sq);
+	qid = mlx5e_qid_from_qos(&priv->channels, node->qid);
+
+	/* If it's a new queue, it will be marked as started at this point.
+	 * Stop it before updating txq2sq.
+	 */
+	mlx5e_tx_disable_queue(netdev_get_tx_queue(priv->netdev, qid));
+
+	priv->txq2sq[qid] = sq;
 
 	/* Make the change to txq2sq visible before the queue is started.
 	 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
@@ -299,8 +310,13 @@ static void mlx5e_deactivate_qos_sq(struct mlx5e_priv *priv, u16 qid)
 	qos_dbg(priv->mdev, "Deactivate QoS SQ qid %u\n", qid);
 	mlx5e_deactivate_txqsq(sq);
 
-	/* The queue is disabled, no synchronization with datapath is needed. */
 	priv->txq2sq[mlx5e_qid_from_qos(&priv->channels, qid)] = NULL;
+
+	/* Make the change to txq2sq visible before the queue is started again.
+	 * As mlx5e_xmit runs under a spinlock, there is an implicit ACQUIRE,
+	 * which pairs with this barrier.
+	 */
+	smp_wmb();
 }
 
 static void mlx5e_close_qos_sq(struct mlx5e_priv *priv, u16 qid)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 9c91ef0e1ed2..726661774979 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -691,8 +691,21 @@ netdev_tx_t mlx5e_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct mlx5e_txqsq *sq;
 	u16 pi;
 
+	/* All changes to txq2sq are performed in sync with mlx5e_xmit, when the
+	 * queue being changed is disabled, and smp_wmb guarantees that the
+	 * changes are visible before mlx5e_xmit tries to read from txq2sq. It
+	 * guarantees that the value of txq2sq[qid] doesn't change while
+	 * mlx5e_xmit is running on queue number qid. smb_wmb is paired with
+	 * HARD_TX_LOCK around ndo_start_xmit, which serves as an ACQUIRE.
+	 */
 	sq = priv->txq2sq[skb_get_queue_mapping(skb)];
 	if (unlikely(!sq)) {
+		/* Two cases when sq can be NULL:
+		 * 1. The HTB node is registered, and mlx5e_select_queue
+		 * selected its queue ID, but the SQ itself is not yet created.
+		 * 2. HTB SQ creation failed. Similar to the previous case, but
+		 * the SQ won't be created.
+		 */
 		dev_kfree_skb_any(skb);
 		return NETDEV_TX_OK;
 	}