pkt_sched: Schedule qdiscs instead of netdev_queue.

When we have shared qdiscs, packets come out of the qdiscs
for multiple transmit queues.

Therefore it doesn't make any sense to schedule the transmit
queue when logically we cannot know ahead of time the TX
queue of the SKB that the qdisc->dequeue() will give us.

Just for sanity I added a BUG check to make sure we never
get into a state where the noop_qdisc is scheduled.

Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 44 insertions, 59 deletions

diff --git a/net/core/dev.c b/net/core/dev.c
index 467bfb325123..0b909b74f698 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -1323,18 +1323,18 @@ static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
 }
 
 
-void __netif_schedule(struct netdev_queue *txq)
+void __netif_schedule(struct Qdisc *q)
 {
-	struct net_device *dev = txq->dev;
+	BUG_ON(q == &noop_qdisc);
 
-	if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
+	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) {
 		struct softnet_data *sd;
 		unsigned long flags;
 
 		local_irq_save(flags);
 		sd = &__get_cpu_var(softnet_data);
-		txq->next_sched = sd->output_queue;
-		sd->output_queue = txq;
+		q->next_sched = sd->output_queue;
+		sd->output_queue = q;
 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
 		local_irq_restore(flags);
 	}
@@ -1771,37 +1771,23 @@ gso:
 	rcu_read_lock_bh();
 
 	txq = dev_pick_tx(dev, skb);
-	spin_lock_prefetch(&txq->lock);
-
-	/* Updates of qdisc are serialized by queue->lock.
-	 * The struct Qdisc which is pointed to by qdisc is now a
-	 * rcu structure - it may be accessed without acquiring
-	 * a lock (but the structure may be stale.) The freeing of the
-	 * qdisc will be deferred until it's known that there are no
-	 * more references to it.
-	 *
-	 * If the qdisc has an enqueue function, we still need to
-	 * hold the queue->lock before calling it, since queue->lock
-	 * also serializes access to the device queue.
-	 */
-
 	q = rcu_dereference(txq->qdisc);
+
 #ifdef CONFIG_NET_CLS_ACT
 	skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
 #endif
 	if (q->enqueue) {
-		/* Grab device queue */
-		spin_lock(&txq->lock);
-		q = txq->qdisc;
-		if (q->enqueue) {
-			rc = q->enqueue(skb, q);
-			qdisc_run(txq);
-			spin_unlock(&txq->lock);
-
-			rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
-			goto out;
-		}
-		spin_unlock(&txq->lock);
+		spinlock_t *root_lock = qdisc_root_lock(q);
+
+		spin_lock(root_lock);
+
+		rc = q->enqueue(skb, q);
+		qdisc_run(q);
+
+		spin_unlock(root_lock);
+
+		rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
+		goto out;
 	}
 
 	/* The device has no queue. Common case for software devices:
@@ -1974,7 +1960,7 @@ static void net_tx_action(struct softirq_action *h)
 	}
 
 	if (sd->output_queue) {
-		struct netdev_queue *head;
+		struct Qdisc *head;
 
 		local_irq_disable();
 		head = sd->output_queue;
@@ -1982,18 +1968,20 @@ static void net_tx_action(struct softirq_action *h)
 		local_irq_enable();
 
 		while (head) {
-			struct netdev_queue *txq = head;
-			struct net_device *dev = txq->dev;
+			struct Qdisc *q = head;
+			spinlock_t *root_lock;
+
 			head = head->next_sched;
 
 			smp_mb__before_clear_bit();
-			clear_bit(__LINK_STATE_SCHED, &dev->state);
+			clear_bit(__QDISC_STATE_SCHED, &q->state);
 
-			if (spin_trylock(&txq->lock)) {
-				qdisc_run(txq);
-				spin_unlock(&txq->lock);
+			root_lock = qdisc_root_lock(q);
+			if (spin_trylock(root_lock)) {
+				qdisc_run(q);
+				spin_unlock(root_lock);
 			} else {
-				netif_schedule_queue(txq);
+				__netif_schedule(q);
 			}
 		}
 	}
@@ -4459,7 +4447,7 @@ static int dev_cpu_callback(struct notifier_block *nfb,
 			    void *ocpu)
 {
 	struct sk_buff **list_skb;
-	struct netdev_queue **list_net;
+	struct Qdisc **list_net;
 	struct sk_buff *skb;
 	unsigned int cpu, oldcpu = (unsigned long)ocpu;
 	struct softnet_data *sd, *oldsd;
diff --git a/net/sched/sch_api.c b/net/sched/sch_api.c
index 19c244a00839..8e8c5becc348 100644
--- a/net/sched/sch_api.c
+++ b/net/sched/sch_api.c
@@ -294,11 +294,10 @@ static enum hrtimer_restart qdisc_watchdog(struct hrtimer *timer)
 {
 	struct qdisc_watchdog *wd = container_of(timer, struct qdisc_watchdog,
 						 timer);
-	struct netdev_queue *txq = wd->qdisc->dev_queue;
 
 	wd->qdisc->flags &= ~TCQ_F_THROTTLED;
 	smp_wmb();
-	netif_schedule_queue(txq);
+	__netif_schedule(wd->qdisc);
 
 	return HRTIMER_NORESTART;
 }
diff --git a/net/sched/sch_cbq.c b/net/sched/sch_cbq.c
index 37ae653db683..a3953bbe2d79 100644
--- a/net/sched/sch_cbq.c
+++ b/net/sched/sch_cbq.c
@@ -650,7 +650,7 @@ static enum hrtimer_restart cbq_undelay(struct hrtimer *timer)
 	}
 
 	sch->flags &= ~TCQ_F_THROTTLED;
-	netif_schedule_queue(sch->dev_queue);
+	__netif_schedule(sch);
 	return HRTIMER_NORESTART;
 }
 
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 739a8711ab30..dd5c4e70abe4 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -72,16 +72,14 @@ static inline int qdisc_qlen(struct Qdisc *q)
 	return q->q.qlen;
 }
 
-static inline int dev_requeue_skb(struct sk_buff *skb,
-				  struct netdev_queue *dev_queue,
-				  struct Qdisc *q)
+static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
 {
 	if (unlikely(skb->next))
 		q->gso_skb = skb;
 	else
 		q->ops->requeue(skb, q);
 
-	netif_schedule_queue(dev_queue);
+	__netif_schedule(q);
 	return 0;
 }
 
@@ -121,7 +119,7 @@ static inline int handle_dev_cpu_collision(struct sk_buff *skb,
 		 * some time.
 		 */
 		__get_cpu_var(netdev_rx_stat).cpu_collision++;
-		ret = dev_requeue_skb(skb, dev_queue, q);
+		ret = dev_requeue_skb(skb, q);
 	}
 
 	return ret;
@@ -146,9 +144,9 @@ static inline int handle_dev_cpu_collision(struct sk_buff *skb,
  *				>0 - queue is not empty.
  *
  */
-static inline int qdisc_restart(struct netdev_queue *txq,
-				struct Qdisc *q)
+static inline int qdisc_restart(struct Qdisc *q)
 {
+	struct netdev_queue *txq;
 	int ret = NETDEV_TX_BUSY;
 	struct net_device *dev;
 	spinlock_t *root_lock;
@@ -163,7 +161,8 @@ static inline int qdisc_restart(struct netdev_queue *txq,
 	/* And release qdisc */
 	spin_unlock(root_lock);
 
-	dev = txq->dev;
+	dev = qdisc_dev(q);
+	txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
 
 	HARD_TX_LOCK(dev, txq, smp_processor_id());
 	if (!netif_subqueue_stopped(dev, skb))
@@ -189,29 +188,28 @@ static inline int qdisc_restart(struct netdev_queue *txq,
 			printk(KERN_WARNING "BUG %s code %d qlen %d\n",
 			       dev->name, ret, q->q.qlen);
 
-		ret = dev_requeue_skb(skb, txq, q);
+		ret = dev_requeue_skb(skb, q);
 		break;
 	}
 
+	if (ret && netif_tx_queue_stopped(txq))
+		ret = 0;
+
 	return ret;
 }
 
-void __qdisc_run(struct netdev_queue *txq)
+void __qdisc_run(struct Qdisc *q)
 {
 	unsigned long start_time = jiffies;
-	struct Qdisc *q = txq->qdisc;
-
-	while (qdisc_restart(txq, q)) {
-		if (netif_tx_queue_stopped(txq))
-			break;
 
+	while (qdisc_restart(q)) {
 		/*
 		 * Postpone processing if
 		 * 1. another process needs the CPU;
 		 * 2. we've been doing it for too long.
 		 */
 		if (need_resched() || jiffies != start_time) {
-			netif_schedule_queue(txq);
+			__netif_schedule(q);
 			break;
 		}
 	}