Re: [BUG][tip/master] kernel panic while locking selftest at qspinlock_paravirt.h:137!

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Jul 10, 2015 at 08:32:46PM +0900, Masami Hiramatsu wrote:

>                     double unlock:
> ------------[ cut here ]------------
> kernel BUG at /home/mhiramat/ksrc/linux-3/kernel/locking/qspinlock_paravirt.h:137!

> Call Trace:
>  [<ffffffff81114a59>] __raw_callee_save___pv_queued_spin_unlock+0x11/0x1e
>  [<ffffffff81117133>] ? do_raw_spin_unlock+0xfa/0x10c
>  [<ffffffff817cd3f7>] _raw_spin_unlock+0x44/0x64
>  [<ffffffff814603ee>] double_unlock_spin+0x3d/0x46

Cute, but somewhat expected. A double unlock really is a BUG and the PV
spinlock code cannot deal with it.

Do we want to make double unlock non-fatal unconditionally?

---
 kernel/locking/qspinlock_paravirt.h | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
index 04ab18151cc8..172deeaf1311 100644
--- a/kernel/locking/qspinlock_paravirt.h
+++ b/kernel/locking/qspinlock_paravirt.h
@@ -286,15 +286,22 @@ __visible void __pv_queued_spin_unlock(struct qspinlock *lock)
 {
 	struct __qspinlock *l = (void *)lock;
 	struct pv_node *node;
+	u8 locked;
 
 	/*
 	 * We must not unlock if SLOW, because in that case we must first
 	 * unhash. Otherwise it would be possible to have multiple @lock
 	 * entries, which would be BAD.
 	 */
-	if (likely(cmpxchg(&l->locked, _Q_LOCKED_VAL, 0) == _Q_LOCKED_VAL))
+	locked = cmpxchg(&l->locked, _Q_LOCKED_VAL, 0);
+	if (likely(locked == _Q_LOCKED_VAL))
 		return;
 
+#ifdef CONFIG_DEBUG_LOCKING_API_SELFTESTS
+	if (unlikely(!locked))
+		return;
+#endif
+
 	/*
 	 * Since the above failed to release, this must be the SLOW path.
 	 * Therefore start by looking up the blocked node and unhashing it.