Merge branch 'add-checking-sq-is-full-inside-xdp-xmit'

Xuan Zhuo says:

====================
add checking sq is full inside xdp xmit

If the queue of xdp xmit is not an independent queue, then when the xdp
xmit used all the desc, the xmit from the __dev_queue_xmit() may encounter
the following error.

net ens4: Unexpected TXQ (0) queue failure: -28

This patch adds a check whether sq is full in XDP Xmit.
====================

Link: https://lore.kernel.org/r/20230308024935.91686-1-xuanzhuo@linux.alibaba.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 85 insertions, 70 deletions

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index fb5e68ed3ec2..1a309cfb4976 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -545,6 +545,87 @@ ok:
 	return skb;
 }
 
+static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
+{
+	unsigned int len;
+	unsigned int packets = 0;
+	unsigned int bytes = 0;
+	void *ptr;
+
+	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
+		if (likely(!is_xdp_frame(ptr))) {
+			struct sk_buff *skb = ptr;
+
+			pr_debug("Sent skb %p\n", skb);
+
+			bytes += skb->len;
+			napi_consume_skb(skb, in_napi);
+		} else {
+			struct xdp_frame *frame = ptr_to_xdp(ptr);
+
+			bytes += xdp_get_frame_len(frame);
+			xdp_return_frame(frame);
+		}
+		packets++;
+	}
+
+	/* Avoid overhead when no packets have been processed
+	 * happens when called speculatively from start_xmit.
+	 */
+	if (!packets)
+		return;
+
+	u64_stats_update_begin(&sq->stats.syncp);
+	sq->stats.bytes += bytes;
+	sq->stats.packets += packets;
+	u64_stats_update_end(&sq->stats.syncp);
+}
+
+static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
+{
+	if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
+		return false;
+	else if (q < vi->curr_queue_pairs)
+		return true;
+	else
+		return false;
+}
+
+static void check_sq_full_and_disable(struct virtnet_info *vi,
+				      struct net_device *dev,
+				      struct send_queue *sq)
+{
+	bool use_napi = sq->napi.weight;
+	int qnum;
+
+	qnum = sq - vi->sq;
+
+	/* If running out of space, stop queue to avoid getting packets that we
+	 * are then unable to transmit.
+	 * An alternative would be to force queuing layer to requeue the skb by
+	 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
+	 * returned in a normal path of operation: it means that driver is not
+	 * maintaining the TX queue stop/start state properly, and causes
+	 * the stack to do a non-trivial amount of useless work.
+	 * Since most packets only take 1 or 2 ring slots, stopping the queue
+	 * early means 16 slots are typically wasted.
+	 */
+	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
+		netif_stop_subqueue(dev, qnum);
+		if (use_napi) {
+			if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
+				virtqueue_napi_schedule(&sq->napi, sq->vq);
+		} else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
+			/* More just got used, free them then recheck. */
+			free_old_xmit_skbs(sq, false);
+			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
+				netif_start_subqueue(dev, qnum);
+				virtqueue_disable_cb(sq->vq);
+			}
+		}
+	}
+}
+
 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
 				   struct send_queue *sq,
 				   struct xdp_frame *xdpf)
@@ -686,6 +767,9 @@ static int virtnet_xdp_xmit(struct net_device *dev,
 	}
 	ret = nxmit;
 
+	if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq))
+		check_sq_full_and_disable(vi, dev, sq);
+
 	if (flags & XDP_XMIT_FLUSH) {
 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
 			kicks = 1;
@@ -1714,52 +1798,6 @@ static int virtnet_receive(struct receive_queue *rq, int budget,
 	return stats.packets;
 }
 
-static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
-{
-	unsigned int len;
-	unsigned int packets = 0;
-	unsigned int bytes = 0;
-	void *ptr;
-
-	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
-		if (likely(!is_xdp_frame(ptr))) {
-			struct sk_buff *skb = ptr;
-
-			pr_debug("Sent skb %p\n", skb);
-
-			bytes += skb->len;
-			napi_consume_skb(skb, in_napi);
-		} else {
-			struct xdp_frame *frame = ptr_to_xdp(ptr);
-
-			bytes += xdp_get_frame_len(frame);
-			xdp_return_frame(frame);
-		}
-		packets++;
-	}
-
-	/* Avoid overhead when no packets have been processed
-	 * happens when called speculatively from start_xmit.
-	 */
-	if (!packets)
-		return;
-
-	u64_stats_update_begin(&sq->stats.syncp);
-	sq->stats.bytes += bytes;
-	sq->stats.packets += packets;
-	u64_stats_update_end(&sq->stats.syncp);
-}
-
-static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
-{
-	if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
-		return false;
-	else if (q < vi->curr_queue_pairs)
-		return true;
-	else
-		return false;
-}
-
 static void virtnet_poll_cleantx(struct receive_queue *rq)
 {
 	struct virtnet_info *vi = rq->vq->vdev->priv;
@@ -1989,30 +2027,7 @@ static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 		nf_reset_ct(skb);
 	}
 
-	/* If running out of space, stop queue to avoid getting packets that we
-	 * are then unable to transmit.
-	 * An alternative would be to force queuing layer to requeue the skb by
-	 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
-	 * returned in a normal path of operation: it means that driver is not
-	 * maintaining the TX queue stop/start state properly, and causes
-	 * the stack to do a non-trivial amount of useless work.
-	 * Since most packets only take 1 or 2 ring slots, stopping the queue
-	 * early means 16 slots are typically wasted.
-	 */
-	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
-		netif_stop_subqueue(dev, qnum);
-		if (use_napi) {
-			if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
-				virtqueue_napi_schedule(&sq->napi, sq->vq);
-		} else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
-			/* More just got used, free them then recheck. */
-			free_old_xmit_skbs(sq, false);
-			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
-				netif_start_subqueue(dev, qnum);
-				virtqueue_disable_cb(sq->vq);
-			}
-		}
-	}
+	check_sq_full_and_disable(vi, dev, sq);
 
 	if (kick || netif_xmit_stopped(txq)) {
 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {