patches/net-move-xmit_recursion-to-per-task-variable-on-RT.patch

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Wed, 13 Jan 2016 15:55:02 +0100
Subject: net: move xmit_recursion to per-task variable on -RT

A softirq on -RT can be preempted. That means one task is in
__dev_queue_xmit(), gets preempted and another task may enter
__dev_queue_xmit() aw well. netperf together with a bridge device
will then trigger the `recursion alert` because each task increments
the xmit_recursion variable which is per-CPU.
A virtual device like br0 is required to trigger this warning.

This patch moves the lock owner and counter to be per task instead per-CPU so
it counts the recursion properly on -RT. The owner is also a task now and not a
CPU number.

Cc: stable-rt@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/netdevice.h |   95 ++++++++++++++++++++++++++++++++++++++++++----
 include/linux/sched.h     |    3 +
 net/core/dev.c            |   15 ++++---
 net/core/filter.c         |    6 +-
 4 files changed, 104 insertions(+), 15 deletions(-)

--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -573,7 +573,11 @@ struct netdev_queue {
  * write-mostly part
  */
 	spinlock_t		_xmit_lock ____cacheline_aligned_in_smp;
+#ifdef CONFIG_PREEMPT_RT_FULL
+	struct task_struct	*xmit_lock_owner;
+#else
 	int			xmit_lock_owner;
+#endif
 	/*
 	 * Time (in jiffies) of last Tx
 	 */
@@ -2561,14 +2565,53 @@ void netdev_freemem(struct net_device *d
 void synchronize_net(void);
 int init_dummy_netdev(struct net_device *dev);
 
-DECLARE_PER_CPU(int, xmit_recursion);
 #define XMIT_RECURSION_LIMIT	10
+#ifdef CONFIG_PREEMPT_RT_FULL
+static inline int dev_recursion_level(void)
+{
+	return current->xmit_recursion;
+}
+
+static inline int xmit_rec_read(void)
+{
+	return current->xmit_recursion;
+}
+
+static inline void xmit_rec_inc(void)
+{
+	current->xmit_recursion++;
+}
+
+static inline void xmit_rec_dec(void)
+{
+	current->xmit_recursion--;
+}
+
+#else
+
+DECLARE_PER_CPU(int, xmit_recursion);
 
 static inline int dev_recursion_level(void)
 {
 	return this_cpu_read(xmit_recursion);
 }
 
+static inline int xmit_rec_read(void)
+{
+	return __this_cpu_read(xmit_recursion);
+}
+
+static inline void xmit_rec_inc(void)
+{
+	__this_cpu_inc(xmit_recursion);
+}
+
+static inline void xmit_rec_dec(void)
+{
+	__this_cpu_dec(xmit_recursion);
+}
+#endif
+
 struct net_device *dev_get_by_index(struct net *net, int ifindex);
 struct net_device *__dev_get_by_index(struct net *net, int ifindex);
 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex);
@@ -3646,10 +3689,48 @@ static inline u32 netif_msg_init(int deb
 	return (1 << debug_value) - 1;
 }
 
+#ifdef CONFIG_PREEMPT_RT_FULL
+static inline void netdev_queue_set_owner(struct netdev_queue *txq, int cpu)
+{
+	txq->xmit_lock_owner = current;
+}
+
+static inline void netdev_queue_clear_owner(struct netdev_queue *txq)
+{
+	txq->xmit_lock_owner = NULL;
+}
+
+static inline bool netdev_queue_has_owner(struct netdev_queue *txq)
+{
+	if (txq->xmit_lock_owner != NULL)
+		return true;
+	return false;
+}
+
+#else
+
+static inline void netdev_queue_set_owner(struct netdev_queue *txq, int cpu)
+{
+	txq->xmit_lock_owner = cpu;
+}
+
+static inline void netdev_queue_clear_owner(struct netdev_queue *txq)
+{
+	txq->xmit_lock_owner = -1;
+}
+
+static inline bool netdev_queue_has_owner(struct netdev_queue *txq)
+{
+	if (txq->xmit_lock_owner != -1)
+		return true;
+	return false;
+}
+#endif
+
 static inline void __netif_tx_lock(struct netdev_queue *txq, int cpu)
 {
 	spin_lock(&txq->_xmit_lock);
-	txq->xmit_lock_owner = cpu;
+	netdev_queue_set_owner(txq, cpu);
 }
 
 static inline bool __netif_tx_acquire(struct netdev_queue *txq)
@@ -3666,32 +3747,32 @@ static inline void __netif_tx_release(st
 static inline void __netif_tx_lock_bh(struct netdev_queue *txq)
 {
 	spin_lock_bh(&txq->_xmit_lock);
-	txq->xmit_lock_owner = smp_processor_id();
+	netdev_queue_set_owner(txq, smp_processor_id());
 }
 
 static inline bool __netif_tx_trylock(struct netdev_queue *txq)
 {
 	bool ok = spin_trylock(&txq->_xmit_lock);
 	if (likely(ok))
-		txq->xmit_lock_owner = smp_processor_id();
+		netdev_queue_set_owner(txq, smp_processor_id());
 	return ok;
 }
 
 static inline void __netif_tx_unlock(struct netdev_queue *txq)
 {
-	txq->xmit_lock_owner = -1;
+	netdev_queue_clear_owner(txq);
 	spin_unlock(&txq->_xmit_lock);
 }
 
 static inline void __netif_tx_unlock_bh(struct netdev_queue *txq)
 {
-	txq->xmit_lock_owner = -1;
+	netdev_queue_clear_owner(txq);
 	spin_unlock_bh(&txq->_xmit_lock);
 }
 
 static inline void txq_trans_update(struct netdev_queue *txq)
 {
-	if (txq->xmit_lock_owner != -1)
+	if (netdev_queue_has_owner(txq))
 		txq->trans_start = jiffies;
 }
 
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1196,6 +1196,9 @@ struct task_struct {
 #ifdef CONFIG_DEBUG_ATOMIC_SLEEP
 	unsigned long			task_state_change;
 #endif
+#ifdef CONFIG_PREEMPT_RT_FULL
+	int				xmit_recursion;
+#endif
 	int				pagefault_disabled;
 #ifdef CONFIG_MMU
 	struct task_struct		*oom_reaper_list;
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3336,8 +3336,10 @@ static void skb_update_prio(struct sk_bu
 #define skb_update_prio(skb)
 #endif
 
+#ifndef CONFIG_PREEMPT_RT_FULL
 DEFINE_PER_CPU(int, xmit_recursion);
 EXPORT_SYMBOL(xmit_recursion);
+#endif
 
 /**
  *	dev_loopback_xmit - loop back @skb
@@ -3577,9 +3579,12 @@ static int __dev_queue_xmit(struct sk_bu
 	if (dev->flags & IFF_UP) {
 		int cpu = smp_processor_id(); /* ok because BHs are off */
 
+#ifdef CONFIG_PREEMPT_RT_FULL
+		if (txq->xmit_lock_owner != current) {
+#else
 		if (txq->xmit_lock_owner != cpu) {
-			if (unlikely(__this_cpu_read(xmit_recursion) >
-				     XMIT_RECURSION_LIMIT))
+#endif
+			if (unlikely(xmit_rec_read() > XMIT_RECURSION_LIMIT))
 				goto recursion_alert;
 
 			skb = validate_xmit_skb(skb, dev, &again);
@@ -3589,9 +3594,9 @@ static int __dev_queue_xmit(struct sk_bu
 			HARD_TX_LOCK(dev, txq, cpu);
 
 			if (!netif_xmit_stopped(txq)) {
-				__this_cpu_inc(xmit_recursion);
+				xmit_rec_inc();
 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
-				__this_cpu_dec(xmit_recursion);
+				xmit_rec_dec();
 				if (dev_xmit_complete(rc)) {
 					HARD_TX_UNLOCK(dev, txq);
 					goto out;
@@ -7882,7 +7887,7 @@ static void netdev_init_one_queue(struct
 	/* Initialize queue lock */
 	spin_lock_init(&queue->_xmit_lock);
 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
-	queue->xmit_lock_owner = -1;
+	netdev_queue_clear_owner(queue);
 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
 	queue->dev = dev;
 #ifdef CONFIG_BQL
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -1983,7 +1983,7 @@ static inline int __bpf_tx_skb(struct ne
 {
 	int ret;
 
-	if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
+	if (unlikely(xmit_rec_read() > XMIT_RECURSION_LIMIT)) {
 		net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
 		kfree_skb(skb);
 		return -ENETDOWN;
@@ -1991,9 +1991,9 @@ static inline int __bpf_tx_skb(struct ne
 
 	skb->dev = dev;
 
-	__this_cpu_inc(xmit_recursion);
+	xmit_rec_inc();
 	ret = dev_queue_xmit(skb);
-	__this_cpu_dec(xmit_recursion);
+	xmit_rec_dec();
 
 	return ret;
 }