qdisc: validate skb without holding lock

Validation of skb can be pretty expensive :

GSO segmentation and/or checksum computations.

We can do this without holding qdisc lock, so that other cpus
can queue additional packets.

Trick is that requeued packets were already validated, so we carry
a boolean so that sch_direct_xmit() can validate a fresh skb list,
or directly use an old one.

Tested on 40Gb NIC (8 TX queues) and 200 concurrent flows, 48 threads
host.

Turning TSO on or off had no effect on throughput, only few more cpu
cycles. Lock contention on qdisc lock disappeared.

Same if disabling TX checksum offload.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 28 insertions, 33 deletions

diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 797ebef..2b349a4 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -56,40 +56,34 @@ static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
 	return 0;
 }
 
-static struct sk_buff *try_bulk_dequeue_skb(struct Qdisc *q,
-					    struct sk_buff *head_skb,
-					    int bytelimit)
+static void try_bulk_dequeue_skb(struct Qdisc *q,
+				 struct sk_buff *skb,
+				 const struct netdev_queue *txq)
 {
-	struct sk_buff *skb, *tail_skb = head_skb;
+	int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
 
 	while (bytelimit > 0) {
-		skb = q->dequeue(q);
-		if (!skb)
-			break;
+		struct sk_buff *nskb = q->dequeue(q);
 
-		bytelimit -= skb->len; /* covers GSO len */
-		skb = validate_xmit_skb(skb, qdisc_dev(q));
-		if (!skb)
+		if (!nskb)
 			break;
 
-		while (tail_skb->next) /* GSO list goto tail */
-			tail_skb = tail_skb->next;
-
-		tail_skb->next = skb;
-		tail_skb = skb;
+		bytelimit -= nskb->len; /* covers GSO len */
+		skb->next = nskb;
+		skb = nskb;
 	}
-
-	return head_skb;
+	skb->next = NULL;
 }
 
 /* Note that dequeue_skb can possibly return a SKB list (via skb->next).
  * A requeued skb (via q->gso_skb) can also be a SKB list.
  */
-static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
+static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
 {
 	struct sk_buff *skb = q->gso_skb;
 	const struct netdev_queue *txq = q->dev_queue;
 
+	*validate = true;
 	if (unlikely(skb)) {
 		/* check the reason of requeuing without tx lock first */
 		txq = skb_get_tx_queue(txq->dev, skb);
@@ -98,21 +92,16 @@ static inline struct sk_buff *dequeue_skb(struct Qdisc *q)
 			q->q.qlen--;
 		} else
 			skb = NULL;
+		/* skb in gso_skb were already validated */
+		*validate = false;
 	} else {
 		if (!(q->flags & TCQ_F_ONETXQUEUE) ||
 		    !netif_xmit_frozen_or_stopped(txq)) {
-			int bytelimit = qdisc_avail_bulklimit(txq);
-
 			skb = q->dequeue(q);
-			if (skb) {
-				bytelimit -= skb->len;
-				skb = validate_xmit_skb(skb, qdisc_dev(q));
-			}
 			if (skb && qdisc_may_bulk(q))
-				skb = try_bulk_dequeue_skb(q, skb, bytelimit);
+				try_bulk_dequeue_skb(q, skb, txq);
 		}
 	}
-
 	return skb;
 }
 
@@ -156,19 +145,24 @@ static inline int handle_dev_cpu_collision(struct sk_buff *skb,
  */
 int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
 		    struct net_device *dev, struct netdev_queue *txq,
-		    spinlock_t *root_lock)
+		    spinlock_t *root_lock, bool validate)
 {
 	int ret = NETDEV_TX_BUSY;
 
 	/* And release qdisc */
 	spin_unlock(root_lock);
 
-	HARD_TX_LOCK(dev, txq, smp_processor_id());
-	if (!netif_xmit_frozen_or_stopped(txq))
-		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
+	/* Note that we validate skb (GSO, checksum, ...) outside of locks */
+	if (validate)
+		skb = validate_xmit_skb_list(skb, dev);
 
-	HARD_TX_UNLOCK(dev, txq);
+	if (skb) {
+		HARD_TX_LOCK(dev, txq, smp_processor_id());
+		if (!netif_xmit_frozen_or_stopped(txq))
+			skb = dev_hard_start_xmit(skb, dev, txq, &ret);
 
+		HARD_TX_UNLOCK(dev, txq);
+	}
 	spin_lock(root_lock);
 
 	if (dev_xmit_complete(ret)) {
@@ -217,9 +211,10 @@ static inline int qdisc_restart(struct Qdisc *q)
 	struct net_device *dev;
 	spinlock_t *root_lock;
 	struct sk_buff *skb;
+	bool validate;
 
 	/* Dequeue packet */
-	skb = dequeue_skb(q);
+	skb = dequeue_skb(q, &validate);
 	if (unlikely(!skb))
 		return 0;
 
@@ -229,7 +224,7 @@ static inline int qdisc_restart(struct Qdisc *q)
 	dev = qdisc_dev(q);
 	txq = skb_get_tx_queue(dev, skb);
 
-	return sch_direct_xmit(skb, q, dev, txq, root_lock);
+	return sch_direct_xmit(skb, q, dev, txq, root_lock, validate);
 }
 
 void __qdisc_run(struct Qdisc *q)