Skip to contentSkip to conversation

ifb: add multiqueue support

PATCH In review Changli Gao 2 participants 4 messages Dec 14, 2010 – Dec 15, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 4/5 ifb: add multiqueue support

Each ifb NIC has nr_cpu_ids rx queues and nr_cpu_ids queues. Packets transmitted to ifb are enqueued to the corresponding per cpu tx queues, and processed in the corresponding per cpu tasklet latter.

The stats are converted to the u64 ones.

tq is a stack variable now. It makes ifb_q_private smaller and tx queue locked only once in ri_tasklet.

The tx_queue_len is multiplied by the number of online CPUs.

---
v2: addressed the comments from Eirc. Thank him.

 drivers/net/ifb.c |  211 +++++++++++++++++++++++++++++++++++-------------------
 1 file changed, 138 insertions(+), 73 deletions(-)
drivers/net/ifb.c +138 -73
index 57c5cfb..918a38e 100644
@@ -37,56 +37,60 @@
 #include <net/net_namespace.h>
 
 #define TX_Q_LIMIT    32
-struct ifb_private {
-	struct tasklet_struct   ifb_tasklet;
-	int     tasklet_pending;
-	struct sk_buff_head     rq;
-	struct sk_buff_head     tq;
+struct ifb_q_private {
+	struct net_device	*dev;
+	struct netdev_queue	*txq;
+	struct tasklet_struct	ifb_tasklet;
+	struct sk_buff_head	rq;
+	struct u64_stats_sync	syncp;
+	u64			rx_packets;
+	u64			rx_bytes;
+	u64			rx_dropped;
 };
 
+#define ifb_priv(dev) ((struct ifb_q_private __percpu *)(dev)->ml_priv)
+
 static int numifbs = 2;
 
-static void ri_tasklet(unsigned long dev)
+static void ri_tasklet(unsigned long arg)
 {
-
-	struct net_device *_dev = (struct net_device *)dev;
-	struct ifb_private *dp = netdev_priv(_dev);
-	struct net_device_stats *stats = &_dev->stats;
-	struct netdev_queue *txq;
+	struct ifb_q_private *qp = (struct ifb_q_private *)arg;
+	struct net_device *dev = qp->dev;
+	struct netdev_queue *txq = qp->txq;
 	struct sk_buff *skb;
+	struct sk_buff_head tq;
 
-	txq = netdev_get_tx_queue(_dev, 0);
-	skb = skb_peek(&dp->tq);
-	if (skb == NULL) {
-		if (__netif_tx_trylock(txq)) {
-			skb_queue_splice_tail_init(&dp->rq, &dp->tq);
-			__netif_tx_unlock(txq);
-		} else {
-			/* reschedule */
-			goto resched;
-		}
+	__skb_queue_head_init(&tq);
+	if (!__netif_tx_trylock(txq)) {
+		tasklet_schedule(&qp->ifb_tasklet);
+		return;
 	}
+	skb_queue_splice_tail_init(&qp->rq, &tq);
+	if (netif_tx_queue_stopped(txq))
+		netif_tx_wake_queue(txq);
+	__netif_tx_unlock(txq);
 
-	while ((skb = skb_dequeue(&dp->tq)) != NULL) {
+	while ((skb = __skb_dequeue(&tq)) != NULL) {
 		u32 from = G_TC_FROM(skb->tc_verd);
 
 		skb->tc_verd = 0;
 		skb->tc_verd = SET_TC_NCLS(skb->tc_verd);
-		stats->tx_packets++;
-		stats->tx_bytes += skb->len;
+		u64_stats_update_begin(&qp->syncp);
+		txq->tx_packets++;
+		txq->tx_bytes += skb->len;
 
 		rcu_read_lock();
 		skb->dev = dev_get_by_index_rcu(&init_net, skb->skb_iif);
 		if (!skb->dev) {
 			rcu_read_unlock();
+			txq->tx_dropped++;
+			u64_stats_update_end(&qp->syncp);
 			dev_kfree_skb(skb);
-			stats->tx_dropped++;
-			if (skb_queue_len(&dp->tq) != 0)
-				goto resched;
-			break;
+			continue;
 		}
 		rcu_read_unlock();
-		skb->skb_iif = _dev->ifindex;
+		u64_stats_update_end(&qp->syncp);
+		skb->skb_iif = dev->ifindex;
 
 		if (from & AT_EGRESS) {
 			dev_queue_xmit(skb);
@@ -96,82 +100,135 @@ static void ri_tasklet(unsigned long dev)
 		} else
 			BUG();
 	}
-
-	if (__netif_tx_trylock(txq)) {
-		skb = skb_peek(&dp->rq);
-		if (skb == NULL) {
-			dp->tasklet_pending = 0;
-			if (netif_queue_stopped(_dev))
-				netif_wake_queue(_dev);
-		} else {
-			__netif_tx_unlock(txq);
-			goto resched;
-		}
-		__netif_tx_unlock(txq);
-	} else {
-resched:
-		dp->tasklet_pending = 1;
-		tasklet_schedule(&dp->ifb_tasklet);
-	}
-
 }
 
 static netdev_tx_t ifb_xmit(struct sk_buff *skb, struct net_device *dev)
 {
-	struct ifb_private *dp = netdev_priv(dev);
-	struct net_device_stats *stats = &dev->stats;
+	int qid = skb_get_queue_mapping(skb);
+	struct ifb_q_private *qp = per_cpu_ptr(ifb_priv(dev), qid);
 	u32 from = G_TC_FROM(skb->tc_verd);
 
-	stats->rx_packets++;
-	stats->rx_bytes += skb->len;
+	WARN_ON(qid != smp_processor_id());
+
+	u64_stats_update_begin(&qp->syncp);
+	qp->rx_packets++;
+	qp->rx_bytes += skb->len;
 
 	if (!(from & (AT_INGRESS|AT_EGRESS)) || !skb->skb_iif) {
+		qp->rx_dropped++;
+		u64_stats_update_end(&qp->syncp);
 		dev_kfree_skb(skb);
-		stats->rx_dropped++;
 		return NETDEV_TX_OK;
 	}
+	u64_stats_update_end(&qp->syncp);
 
-	__skb_queue_tail(&dp->rq, skb);
-	if (!dp->tasklet_pending) {
-		dp->tasklet_pending = 1;
-		tasklet_schedule(&dp->ifb_tasklet);
-	}
+	__skb_queue_tail(&qp->rq, skb);
+	if (skb_queue_len(&qp->rq) == 1)
+		tasklet_schedule(&qp->ifb_tasklet);
 
-	if (skb_queue_len(&dp->rq) >= dev->tx_queue_len)
-		netif_stop_queue(dev);
+	if (skb_queue_len(&qp->rq) >= dev->tx_queue_len)
+		netif_tx_stop_queue(qp->txq);
 
 	return NETDEV_TX_OK;
 }
 
 static int ifb_close(struct net_device *dev)
 {
-	struct ifb_private *dp = netdev_priv(dev);
-
-	tasklet_kill(&dp->ifb_tasklet);
-	netif_stop_queue(dev);
-	__skb_queue_purge(&dp->rq);
-	__skb_queue_purge(&dp->tq);
+	struct ifb_q_private *qp;
+	int cpu;
+
+	for_each_possible_cpu(cpu) {
+		qp = per_cpu_ptr(ifb_priv(dev), cpu);
+		tasklet_kill(&qp->ifb_tasklet);
+		netif_tx_stop_queue(qp->txq);
+		__skb_queue_purge(&qp->rq);
+	}
 
 	return 0;
 }
 
 static int ifb_open(struct net_device *dev)
 {
-	struct ifb_private *dp = netdev_priv(dev);
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		netif_tx_start_queue(netdev_get_tx_queue(dev, cpu));
 
-	tasklet_init(&dp->ifb_tasklet, ri_tasklet, (unsigned long)dev);
-	__skb_queue_head_init(&dp->rq);
-	__skb_queue_head_init(&dp->tq);
-	netif_start_queue(dev);
+	return 0;
+}
+
+static int ifb_init(struct net_device *dev)
+{
+	struct ifb_q_private *q;
+	int cpu;
+
+	dev->ml_priv = alloc_percpu(struct ifb_q_private);
+	if (!ifb_priv(dev))
+		return -ENOMEM;
+	for_each_possible_cpu(cpu) {
+		q = per_cpu_ptr(ifb_priv(dev), cpu);
+		__skb_queue_head_init(&q->rq);
+		q->txq = netdev_get_tx_queue(dev, cpu);
+		q->dev = dev;
+		tasklet_init(&q->ifb_tasklet, ri_tasklet, (unsigned long)q);
+		netdev_queue_numa_node_write(q->txq, cpu_to_node(cpu));
+	}
 
 	return 0;
 }
 
+static void ifb_uninit(struct net_device *dev)
+{
+	free_percpu(ifb_priv(dev));
+}
+
+static u16 ifb_select_queue(struct net_device *dev, struct sk_buff *skb)
+{
+	return smp_processor_id();
+}
+
+static struct rtnl_link_stats64 *ifb_get_stats64(struct net_device *dev,
+		struct rtnl_link_stats64 *stats)
+{
+	struct ifb_q_private *q;
+	struct netdev_queue *txq;
+	int cpu;
+	u64 rx_packets, rx_bytes, rx_dropped;
+	u64 tx_packets, tx_bytes, tx_dropped;
+	unsigned int start;
+
+	for_each_possible_cpu(cpu) {
+		q = per_cpu_ptr(ifb_priv(dev), cpu);
+		txq = q->txq;
+		do {
+			start = u64_stats_fetch_begin_bh(&q->syncp);
+			rx_packets = q->rx_packets;
+			rx_bytes = q->rx_bytes;
+			rx_dropped = q->rx_dropped;
+			tx_packets = txq->tx_packets;
+			tx_bytes = txq->tx_bytes;
+			tx_dropped = txq->tx_dropped;
+		} while (u64_stats_fetch_retry_bh(&q->syncp, start));
+		stats->rx_packets += rx_packets;
+		stats->rx_bytes += rx_bytes;
+		stats->rx_dropped += rx_dropped;
+		stats->tx_packets += tx_packets;
+		stats->tx_bytes += tx_bytes;
+		stats->tx_dropped += tx_dropped;
+	}
+
+	return stats;
+}
+
 static const struct net_device_ops ifb_netdev_ops = {
+	.ndo_init		= ifb_init,
+	.ndo_uninit		= ifb_uninit,
 	.ndo_open		= ifb_open,
 	.ndo_stop		= ifb_close,
 	.ndo_start_xmit		= ifb_xmit,
 	.ndo_validate_addr	= eth_validate_addr,
+	.ndo_select_queue	= ifb_select_queue,
+	.ndo_get_stats64	= ifb_get_stats64,
 };
 
 static void ifb_setup(struct net_device *dev)
@@ -202,11 +259,20 @@ static int ifb_validate(struct nlattr *tb[], struct nlattr *data[])
 	return 0;
 }
 
+static int ifb_get_tx_queues(struct net *net, struct nlattr *tb[],
+			     unsigned int *tx_queues,
+			     unsigned int *real_tx_queues)
+{
+	*real_tx_queues = *tx_queues = nr_cpu_ids;
+
+	return 0;
+}
+
 static struct rtnl_link_ops ifb_link_ops __read_mostly = {
 	.kind		= "ifb",
-	.priv_size	= sizeof(struct ifb_private),
 	.setup		= ifb_setup,
 	.validate	= ifb_validate,
+	.get_tx_queues	= ifb_get_tx_queues,
 };
 
 /* Number of ifb devices to be set up by this module. */
@@ -218,8 +284,7 @@ static int __init ifb_init_one(int index)
 	struct net_device *dev_ifb;
 	int err;
 
-	dev_ifb = alloc_netdev(sizeof(struct ifb_private), "ifb%d", ifb_setup);
-
+	dev_ifb = alloc_netdev_mq(0, "ifb%d", ifb_setup, nr_cpu_ids);
 	if (!dev_ifb)
 		return -ENOMEM;
 
#1 lore raw
Quoted 23 lines · Changli Gao
drivers/net/ifb.c +23 -0
+static struct rtnl_link_stats64 *ifb_get_stats64(struct net_device *dev,
+		struct rtnl_link_stats64 *stats)
+{
+	struct ifb_q_private *q;
+	struct netdev_queue *txq;
+	int cpu;
+	u64 rx_packets, rx_bytes, rx_dropped;
+	u64 tx_packets, tx_bytes, tx_dropped;
+	unsigned int start;
+
+	for_each_possible_cpu(cpu) {
+		q = per_cpu_ptr(ifb_priv(dev), cpu);
+		txq = q->txq;
+		do {
+			start = u64_stats_fetch_begin_bh(&q->syncp);
+			rx_packets = q->rx_packets;
+			rx_bytes = q->rx_bytes;
+			rx_dropped = q->rx_dropped;
+			tx_packets = txq->tx_packets;
+			tx_bytes = txq->tx_bytes;
+			tx_dropped = txq->tx_dropped;
+		} while (u64_stats_fetch_retry_bh(&q->syncp, start));
+		stats->rx_packets += rx_packets;

drivers/net/ifb.c static void ri_tasklet(unsigned long dev) · line 222

Quoted 10 lines
drivers/net/ifb.c +10 -0
+		stats->rx_bytes += rx_bytes;
+		stats->rx_dropped += rx_dropped;
+		stats->tx_packets += tx_packets;
+		stats->tx_bytes += tx_bytes;
+		stats->tx_dropped += tx_dropped;
+	}
+
+	return stats;
+}
+

There is a problem here.

You should sum in the loop rx_counters only, (the counters syncp protected), and use dev_txq_stats_fold() to get the tx_counters from core network.

dev_txq_stats_fold(dev, stats);

for_each_possible_cpu(cpu) {
	q = per_cpu_ptr(ifb_priv(dev), cpu);
	txq = q->txq;
	do {
		start = u64_stats_fetch_begin_bh(&q->syncp);
		rx_packets = q->rx_packets;
		rx_bytes   = q->rx_bytes;
		rx_dropped = q->rx_dropped;
	} while (u64_stats_fetch_retry_bh(&q->syncp, start));
	stats->rx_packets += rx_packets;
	stats->rx_bytes   += rx_bytes;
	stats->rx_dropped += rx_dropped;
}

return stats;

#2 lore raw
Quoted 7 lines · Eric Dumazet

There is a problem here.

You should sum in the loop rx_counters only, (the counters syncp protected), and use dev_txq_stats_fold() to get the tx_counters from core network.

Hmm, in ifb, txq->tx_counters aren't used in ifb_xmit() but ri_tasklet(). IOW, rx and tx are reversed. Here I use txq->tx_counters to reduce duplicate counters in ifb_q_private. Unless these variables might be accessed outside of ifb, I can use them freely? Thanks.

Signature · 2 lines
Regards,
Changli Gao([email protected])
#3 lore raw
Quoted 5 lines · Changli Gao
Quoted 4 lines · Eric Dumazet

There is a problem here.

Quoted 9 lines
Quoted 3 lines

You should sum in the loop rx_counters only, (the counters syncp protected), and use dev_txq_stats_fold() to get the tx_counters from core network.

Hmm, in ifb, txq->tx_counters aren't used in ifb_xmit() but ri_tasklet(). IOW, rx and tx are reversed. Here I use txq->tx_counters to reduce duplicate counters in ifb_q_private. Unless these variables might be accessed outside of ifb, I can use them freely? Thanks.

Yes, you are right, and the txq lock is not necessary to read their values as well (I thought it was hold at write time)