Badness in futex_wait at kernel/futex.c:509

Hi!

This happen every time I switch from X to console:

Badness in futex_wait at kernel/futex.c:509

Call Trace:
 [futex_wait+434/448] futex_wait+0x1b2/0x1c0
 [default_wake_function+0/32] default_wake_function+0x0/0x20
 [default_wake_function+0/32] default_wake_function+0x0/0x20
 [do_futex+112/128] do_futex+0x70/0x80
 [sys_futex+292/320] sys_futex+0x124/0x140
 [syscall_call+7/11] syscall_call+0x7/0xb

Others messages found in syslog:

Badness in futex_wait at kernel/futex.c:509

Call Trace:
 [futex_wait+434/448] futex_wait+0x1b2/0x1c0
 [common_interrupt+24/32] common_interrupt+0x18/0x20
 [default_wake_function+0/32] default_wake_function+0x0/0x20
 [default_wake_function+0/32] default_wake_function+0x0/0x20
 [do_futex+112/128] do_futex+0x70/0x80
 [sys_futex+292/320] sys_futex+0x124/0x140
 [syscall_call+7/11] syscall_call+0x7/0xb

I do not get these messages with 2.6.1.

If more information is needeed, please send me an email to [email protected]

Thanks!
Horacio de Oro

	/* A spurious wakeup should never happen. */
	WARN_ON(!signal_pending(current));

(looks at Rusty)

In message <[email protected]> you write:

You were the one who said spurious wakeups shouldn't happen 8)

This implies that the console code decided to do wake_up_process() on us. We returned -EINTR to userspace, but there was no signal, which is odd.

Anyone have any ideas *why*?
Rusty.
--
Anyone who quotes me in their sig is an idiot. -- Rusty Russell.

Well this would seem to imply that this task woke up and ran while it was still parked on q.waiters, yes?

Nope. This patch should identify the source of the wakeup. Horacio, could you please add it, see if you get a nice trace? The patch is a bit racy wrt task exit, but it should hold up for long enough to get a trace.

 include/linux/sched.h |    1 +
 kernel/futex.c        |    5 ++++-
 kernel/sched.c        |    3 +++
 3 files changed, 8 insertions(+), 1 deletion(-)

diff -puN kernel/futex.c~futex-wakeup-debug kernel/futex.c
@@ -491,8 +491,11 @@ static int futex_wait(unsigned long uadd
 	 * !list_empty() is safe here without any lock.
 	 * q.lock_ptr != 0 is not safe, because of ordering against wakeup.
 	 */
-	if (likely(!list_empty(&q.list)))
+	if (likely(!list_empty(&q.list))) {
+		current->flags |= PF_FUTEX_DEBUG;
 		time = schedule_timeout(time);
+		current->flags &= ~PF_FUTEX_DEBUG;
+	}
 	__set_current_state(TASK_RUNNING);
 
 	/*

diff -puN include/linux/sched.h~futex-wakeup-debug include/linux/sched.h
@@ -500,6 +500,7 @@ do { if (atomic_dec_and_test(&(tsk)->usa
 #define PF_SWAPOFF	0x00080000	/* I am in swapoff */
 #define PF_LESS_THROTTLE 0x00100000	/* Throttle me less: I clean memory */
 #define PF_SYNCWRITE	0x00200000	/* I am doing a sync write */
+#define PF_FUTEX_DEBUG	0x00400000
 
 #ifdef CONFIG_SMP
 #define SD_FLAG_NEWIDLE		1	/* Balance when about to become idle */

diff -puN kernel/sched.c~futex-wakeup-debug kernel/sched.c
@@ -770,6 +770,9 @@ static int try_to_wake_up(task_t * p, un
 	}
 	goto out_activate;
 
+	if (p->flags & PF_FUTEX_DEBUG)
+		WARN_ON(!signal_pending(p));
+
 repeat_lock_task:
 	task_rq_unlock(rq, &flags);
 	rq = task_rq_lock(p, &flags);

_