[PATCH] sched/core: avoid spurious spinlock recursion splats

[Mark Rutland]

The runqueue locks are special in that the owner changes over a context switch.
To ensure that this is account for in CONFIG_DEBUG_SPINLOCK builds,
finish_lock_switch updates rq->lock.owner while the lock is held.

However, this happens *after* prev->on_cpu is cleared, which allows prev to be
scheduled on another CPU. If prev then attempts to acquire the same rq lock, it
will see itself as the owner.

This can be seen in virtual environments, where the vCPU can be preempted for
an arbitrarily long period between updating prev->on_cpu and rq->lock.owner.

We can avoid this issue by updating rq->lock.owner first. The release of
prev->on_cpu will ensure that the new owner is visible to prev if it is
scheduled on another CPU.

Signed-off-by: Mark Rutland <mark.rutland@xxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
---
 kernel/sched/sched.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index b19552a212de..4f0d2e3701c3 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1342,6 +1342,10 @@ static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
 
 static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 {
+#ifdef CONFIG_DEBUG_SPINLOCK
+	/* this is a valid case when another task releases the spinlock */
+	rq->lock.owner = current;
+#endif
 #ifdef CONFIG_SMP
 	/*
 	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
@@ -1355,10 +1359,6 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
 	 */
 	smp_store_release(&prev->on_cpu, 0);
 #endif
-#ifdef CONFIG_DEBUG_SPINLOCK
-	/* this is a valid case when another task releases the spinlock */
-	rq->lock.owner = current;
-#endif
 	/*
 	 * If we are tracking spinlock dependencies then we have to
 	 * fix up the runqueue lock - which gets 'carried over' from
-- 
2.11.0