[NET]: Make NAPI polling independent of struct net_device objects.

Several devices have multiple independant RX queues per net
device, and some have a single interrupt doorbell for several
queues.

In either case, it's easier to support layouts like that if the
structure representing the poll is independant from the net
device itself.

The signature of the ->poll() call back goes from:

	int foo_poll(struct net_device *dev, int *budget)

to

	int foo_poll(struct napi_struct *napi, int budget)

The caller is returned the number of RX packets processed (or
the number of "NAPI credits" consumed if you want to get
abstract).  The callee no longer messes around bumping
dev->quota, *budget, etc. because that is all handled in the
caller upon return.

The napi_struct is to be embedded in the device driver private data
structures.

Furthermore, it is the driver's responsibility to disable all NAPI
instances in it's ->stop() device close handler.  Since the
napi_struct is privatized into the driver's private data structures,
only the driver knows how to get at all of the napi_struct instances
it may have per-device.

With lots of help and suggestions from Rusty Russell, Roland Dreier,
Michael Chan, Jeff Garzik, and Jamal Hadi Salim.

Bug fixes from Thomas Graf, Roland Dreier, Peter Zijlstra,
Joseph Fannin, Scott Wood, Hans J. Koch, and Michael Chan.

[ Ported to current tree and all drivers converted.  Integrated
  Stephen's follow-on kerneldoc additions, and restored poll_list
  handling to the old style to fix mutual exclusion issues.  -DaveM ]

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 131 insertions, 109 deletions

diff --git a/net/core/dev.c b/net/core/dev.c
index a76021c..29cf00c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -220,7 +220,8 @@ static RAW_NOTIFIER_HEAD(netdev_chain);
  *	Device drivers call our routines to queue packets here. We empty the
  *	queue in the local softnet handler.
  */
-DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
+
+DEFINE_PER_CPU(struct softnet_data, softnet_data);
 
 #ifdef CONFIG_SYSFS
 extern int netdev_sysfs_init(void);
@@ -1018,16 +1019,12 @@ int dev_close(struct net_device *dev)
 	clear_bit(__LINK_STATE_START, &dev->state);
 
 	/* Synchronize to scheduled poll. We cannot touch poll list,
-	 * it can be even on different cpu. So just clear netif_running(),
-	 * and wait when poll really will happen. Actually, the best place
-	 * for this is inside dev->stop() after device stopped its irq
-	 * engine, but this requires more changes in devices. */
-
+	 * it can be even on different cpu. So just clear netif_running().
+	 *
+	 * dev->stop() will invoke napi_disable() on all of it's
+	 * napi_struct instances on this device.
+	 */
 	smp_mb__after_clear_bit(); /* Commit netif_running(). */
-	while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
-		/* No hurry. */
-		msleep(1);
-	}
 
 	/*
 	 *	Call the device specific close. This cannot fail.
@@ -1233,21 +1230,21 @@ void __netif_schedule(struct net_device *dev)
 }
 EXPORT_SYMBOL(__netif_schedule);
 
-void __netif_rx_schedule(struct net_device *dev)
+void dev_kfree_skb_irq(struct sk_buff *skb)
 {
-	unsigned long flags;
+	if (atomic_dec_and_test(&skb->users)) {
+		struct softnet_data *sd;
+		unsigned long flags;
 
-	local_irq_save(flags);
-	dev_hold(dev);
-	list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
-	if (dev->quota < 0)
-		dev->quota += dev->weight;
-	else
-		dev->quota = dev->weight;
-	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
-	local_irq_restore(flags);
+		local_irq_save(flags);
+		sd = &__get_cpu_var(softnet_data);
+		skb->next = sd->completion_queue;
+		sd->completion_queue = skb;
+		raise_softirq_irqoff(NET_TX_SOFTIRQ);
+		local_irq_restore(flags);
+	}
 }
-EXPORT_SYMBOL(__netif_rx_schedule);
+EXPORT_SYMBOL(dev_kfree_skb_irq);
 
 void dev_kfree_skb_any(struct sk_buff *skb)
 {
@@ -1259,7 +1256,12 @@ void dev_kfree_skb_any(struct sk_buff *skb)
 EXPORT_SYMBOL(dev_kfree_skb_any);
 
 
-/* Hot-plugging. */
+/**
+ * netif_device_detach - mark device as removed
+ * @dev: network device
+ *
+ * Mark device as removed from system and therefore no longer available.
+ */
 void netif_device_detach(struct net_device *dev)
 {
 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
@@ -1269,6 +1271,12 @@ void netif_device_detach(struct net_device *dev)
 }
 EXPORT_SYMBOL(netif_device_detach);
 
+/**
+ * netif_device_attach - mark device as attached
+ * @dev: network device
+ *
+ * Mark device as attached from system and restart if needed.
+ */
 void netif_device_attach(struct net_device *dev)
 {
 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
@@ -1730,7 +1738,7 @@ enqueue:
 			return NET_RX_SUCCESS;
 		}
 
-		netif_rx_schedule(&queue->backlog_dev);
+		napi_schedule(&queue->backlog);
 		goto enqueue;
 	}
 
@@ -1771,6 +1779,7 @@ static inline struct net_device *skb_bond(struct sk_buff *skb)
 	return dev;
 }
 
+
 static void net_tx_action(struct softirq_action *h)
 {
 	struct softnet_data *sd = &__get_cpu_var(softnet_data);
@@ -1927,7 +1936,7 @@ int netif_receive_skb(struct sk_buff *skb)
 	__be16 type;
 
 	/* if we've gotten here through NAPI, check netpoll */
-	if (skb->dev->poll && netpoll_rx(skb))
+	if (netpoll_receive_skb(skb))
 		return NET_RX_DROP;
 
 	if (!skb->tstamp.tv64)
@@ -2017,22 +2026,25 @@ out:
 	return ret;
 }
 
-static int process_backlog(struct net_device *backlog_dev, int *budget)
+static int process_backlog(struct napi_struct *napi, int quota)
 {
 	int work = 0;
-	int quota = min(backlog_dev->quota, *budget);
 	struct softnet_data *queue = &__get_cpu_var(softnet_data);
 	unsigned long start_time = jiffies;
 
-	backlog_dev->weight = weight_p;
-	for (;;) {
+	napi->weight = weight_p;
+	do {
 		struct sk_buff *skb;
 		struct net_device *dev;
 
 		local_irq_disable();
 		skb = __skb_dequeue(&queue->input_pkt_queue);
-		if (!skb)
-			goto job_done;
+		if (!skb) {
+			__napi_complete(napi);
+			local_irq_enable();
+			break;
+		}
+
 		local_irq_enable();
 
 		dev = skb->dev;
@@ -2040,67 +2052,86 @@ static int process_backlog(struct net_device *backlog_dev, int *budget)
 		netif_receive_skb(skb);
 
 		dev_put(dev);
+	} while (++work < quota && jiffies == start_time);
 
-		work++;
-
-		if (work >= quota || jiffies - start_time > 1)
-			break;
-
-	}
-
-	backlog_dev->quota -= work;
-	*budget -= work;
-	return -1;
-
-job_done:
-	backlog_dev->quota -= work;
-	*budget -= work;
+	return work;
+}
 
-	list_del(&backlog_dev->poll_list);
-	smp_mb__before_clear_bit();
-	netif_poll_enable(backlog_dev);
+/**
+ * __napi_schedule - schedule for receive
+ * @napi: entry to schedule
+ *
+ * The entry's receive function will be scheduled to run
+ */
+void fastcall __napi_schedule(struct napi_struct *n)
+{
+	unsigned long flags;
 
-	local_irq_enable();
-	return 0;
+	local_irq_save(flags);
+	list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
+	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
+	local_irq_restore(flags);
 }
+EXPORT_SYMBOL(__napi_schedule);
+
 
 static void net_rx_action(struct softirq_action *h)
 {
-	struct softnet_data *queue = &__get_cpu_var(softnet_data);
+	struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
 	unsigned long start_time = jiffies;
 	int budget = netdev_budget;
 	void *have;
 
 	local_irq_disable();
 
-	while (!list_empty(&queue->poll_list)) {
-		struct net_device *dev;
+	while (!list_empty(list)) {
+		struct napi_struct *n;
+		int work, weight;
 
-		if (budget <= 0 || jiffies - start_time > 1)
+		/* If softirq window is exhuasted then punt.
+		 *
+		 * Note that this is a slight policy change from the
+		 * previous NAPI code, which would allow up to 2
+		 * jiffies to pass before breaking out.  The test
+		 * used to be "jiffies - start_time > 1".
+		 */
+		if (unlikely(budget <= 0 || jiffies != start_time))
 			goto softnet_break;
 
 		local_irq_enable();
 
-		dev = list_entry(queue->poll_list.next,
-				 struct net_device, poll_list);
-		have = netpoll_poll_lock(dev);
+		/* Even though interrupts have been re-enabled, this
+		 * access is safe because interrupts can only add new
+		 * entries to the tail of this list, and only ->poll()
+		 * calls can remove this head entry from the list.
+		 */
+		n = list_entry(list->next, struct napi_struct, poll_list);
 
-		if (dev->quota <= 0 || dev->poll(dev, &budget)) {
-			netpoll_poll_unlock(have);
-			local_irq_disable();
-			list_move_tail(&dev->poll_list, &queue->poll_list);
-			if (dev->quota < 0)
-				dev->quota += dev->weight;
-			else
-				dev->quota = dev->weight;
-		} else {
-			netpoll_poll_unlock(have);
-			dev_put(dev);
-			local_irq_disable();
-		}
+		have = netpoll_poll_lock(n);
+
+		weight = n->weight;
+
+		work = n->poll(n, weight);
+
+		WARN_ON_ONCE(work > weight);
+
+		budget -= work;
+
+		local_irq_disable();
+
+		/* Drivers must not modify the NAPI state if they
+		 * consume the entire weight.  In such cases this code
+		 * still "owns" the NAPI instance and therefore can
+		 * move the instance around on the list at-will.
+		 */
+		if (unlikely(work == weight))
+			list_move_tail(&n->poll_list, list);
+
+		netpoll_poll_unlock(have);
 	}
 out:
 	local_irq_enable();
+
 #ifdef CONFIG_NET_DMA
 	/*
 	 * There may not be any more sk_buffs coming right now, so push
@@ -2115,6 +2146,7 @@ out:
 		}
 	}
 #endif
+
 	return;
 
 softnet_break:
@@ -3704,6 +3736,7 @@ struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
 	dev->egress_subqueue_count = queue_count;
 
 	dev->get_stats = internal_stats;
+	netpoll_netdev_init(dev);
 	setup(dev);
 	strcpy(dev->name, name);
 	return dev;
@@ -4076,10 +4109,9 @@ static int __init net_dev_init(void)
 		skb_queue_head_init(&queue->input_pkt_queue);
 		queue->completion_queue = NULL;
 		INIT_LIST_HEAD(&queue->poll_list);
-		set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
-		queue->backlog_dev.weight = weight_p;
-		queue->backlog_dev.poll = process_backlog;
-		atomic_set(&queue->backlog_dev.refcnt, 1);
+
+		queue->backlog.poll = process_backlog;
+		queue->backlog.weight = weight_p;
 	}
 
 	netdev_dma_register();
diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index 5c19b06..79159db 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -216,20 +216,6 @@ static ssize_t store_tx_queue_len(struct device *dev,
 	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 }
 
-NETDEVICE_SHOW(weight, fmt_dec);
-
-static int change_weight(struct net_device *net, unsigned long new_weight)
-{
-	net->weight = new_weight;
-	return 0;
-}
-
-static ssize_t store_weight(struct device *dev, struct device_attribute *attr,
-			    const char *buf, size_t len)
-{
-	return netdev_store(dev, attr, buf, len, change_weight);
-}
-
 static struct device_attribute net_class_attributes[] = {
 	__ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 	__ATTR(iflink, S_IRUGO, show_iflink, NULL),
@@ -246,7 +232,6 @@ static struct device_attribute net_class_attributes[] = {
 	__ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 	__ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 	       store_tx_queue_len),
-	__ATTR(weight, S_IRUGO | S_IWUSR, show_weight, store_weight),
 	{}
 };
 
diff --git a/net/core/netpoll.c b/net/core/netpoll.c
index de1b26a..abe6e3a 100644
--- a/net/core/netpoll.c
+++ b/net/core/netpoll.c
@@ -119,19 +119,22 @@ static __sum16 checksum_udp(struct sk_buff *skb, struct udphdr *uh,
 static void poll_napi(struct netpoll *np)
 {
 	struct netpoll_info *npinfo = np->dev->npinfo;
+	struct napi_struct *napi;
 	int budget = 16;
 
-	if (test_bit(__LINK_STATE_RX_SCHED, &np->dev->state) &&
-	    npinfo->poll_owner != smp_processor_id() &&
-	    spin_trylock(&npinfo->poll_lock)) {
-		npinfo->rx_flags |= NETPOLL_RX_DROP;
-		atomic_inc(&trapped);
+	list_for_each_entry(napi, &np->dev->napi_list, dev_list) {
+		if (test_bit(NAPI_STATE_SCHED, &napi->state) &&
+		    napi->poll_owner != smp_processor_id() &&
+		    spin_trylock(&napi->poll_lock)) {
+			npinfo->rx_flags |= NETPOLL_RX_DROP;
+			atomic_inc(&trapped);
 
-		np->dev->poll(np->dev, &budget);
+			napi->poll(napi, budget);
 
-		atomic_dec(&trapped);
-		npinfo->rx_flags &= ~NETPOLL_RX_DROP;
-		spin_unlock(&npinfo->poll_lock);
+			atomic_dec(&trapped);
+			npinfo->rx_flags &= ~NETPOLL_RX_DROP;
+			spin_unlock(&napi->poll_lock);
+		}
 	}
 }
 
@@ -157,7 +160,7 @@ void netpoll_poll(struct netpoll *np)
 
 	/* Process pending work on NIC */
 	np->dev->poll_controller(np->dev);
-	if (np->dev->poll)
+	if (!list_empty(&np->dev->napi_list))
 		poll_napi(np);
 
 	service_arp_queue(np->dev->npinfo);
@@ -233,6 +236,17 @@ repeat:
 	return skb;
 }
 
+static int netpoll_owner_active(struct net_device *dev)
+{
+	struct napi_struct *napi;
+
+	list_for_each_entry(napi, &dev->napi_list, dev_list) {
+		if (napi->poll_owner == smp_processor_id())
+			return 1;
+	}
+	return 0;
+}
+
 static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 {
 	int status = NETDEV_TX_BUSY;
@@ -246,8 +260,7 @@ static void netpoll_send_skb(struct netpoll *np, struct sk_buff *skb)
 	}
 
 	/* don't get messages out of order, and no recursion */
-	if (skb_queue_len(&npinfo->txq) == 0 &&
-		    npinfo->poll_owner != smp_processor_id()) {
+	if (skb_queue_len(&npinfo->txq) == 0 && !netpoll_owner_active(dev)) {
 		unsigned long flags;
 
 		local_irq_save(flags);
@@ -652,8 +665,6 @@ int netpoll_setup(struct netpoll *np)
 
 		npinfo->rx_flags = 0;
 		npinfo->rx_np = NULL;
-		spin_lock_init(&npinfo->poll_lock);
-		npinfo->poll_owner = -1;
 
 		spin_lock_init(&npinfo->rx_lock);
 		skb_queue_head_init(&npinfo->arp_tx);
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index 4756d58..2b0b6fa 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -634,7 +634,6 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, struct net_device *dev,
 
 	NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name);
 	NLA_PUT_U32(skb, IFLA_TXQLEN, dev->tx_queue_len);
-	NLA_PUT_U32(skb, IFLA_WEIGHT, dev->weight);
 	NLA_PUT_U8(skb, IFLA_OPERSTATE,
 		   netif_running(dev) ? dev->operstate : IF_OPER_DOWN);
 	NLA_PUT_U8(skb, IFLA_LINKMODE, dev->link_mode);
@@ -834,9 +833,6 @@ static int do_setlink(struct net_device *dev, struct ifinfomsg *ifm,
 	if (tb[IFLA_TXQLEN])
 		dev->tx_queue_len = nla_get_u32(tb[IFLA_TXQLEN]);
 
-	if (tb[IFLA_WEIGHT])
-		dev->weight = nla_get_u32(tb[IFLA_WEIGHT]);
-
 	if (tb[IFLA_OPERSTATE])
 		set_operstate(dev, nla_get_u8(tb[IFLA_OPERSTATE]));
 
@@ -1074,8 +1070,6 @@ replay:
 			       nla_len(tb[IFLA_BROADCAST]));
 		if (tb[IFLA_TXQLEN])
 			dev->tx_queue_len = nla_get_u32(tb[IFLA_TXQLEN]);
-		if (tb[IFLA_WEIGHT])
-			dev->weight = nla_get_u32(tb[IFLA_WEIGHT]);
 		if (tb[IFLA_OPERSTATE])
 			set_operstate(dev, nla_get_u8(tb[IFLA_OPERSTATE]));
 		if (tb[IFLA_LINKMODE])