ALSA system timer backend stops the timer via del_timer() without sync and leaves del_timer_sync() at the close instead. This is because of the restriction by the design of ALSA timer: namely, the stop callback may be called from the timer handler, and calling the sync shall lead to a hangup. However, this also triggers a kernel BUG() when the timer is rearmed immediately after stopping without sync: kernel BUG at kernel/time/timer.c:966! Call Trace: <IRQ> [<ffffffff8239c94e>] snd_timer_s_start+0x13e/0x1a0 [<ffffffff8239e1f4>] snd_timer_interrupt+0x504/0xec0 [<ffffffff8122fca0>] ? debug_check_no_locks_freed+0x290/0x290 [<ffffffff8239ec64>] snd_timer_s_function+0xb4/0x120 [<ffffffff81296b72>] call_timer_fn+0x162/0x520 [<ffffffff81296add>] ? call_timer_fn+0xcd/0x520 [<ffffffff8239ebb0>] ? snd_timer_interrupt+0xec0/0xec0 ....
It's the place where add_timer() checks the pending timer. It's clear that this may happen after the immediate restart without sync in our cases.
So, the workaround here is just to use mod_timer() instead of add_timer(). This looks like a band-aid fix, but it's a right move, as snd_timer_interrupt() takes care of the continuous rearm of timer.
Reported-by: Jiri Slaby jslaby@suse.cz Cc: stable@vger.kernel.org Signed-off-by: Takashi Iwai tiwai@suse.de --- sound/core/timer.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/sound/core/timer.c b/sound/core/timer.c index ea4d999113ef..6469bedda2f3 100644 --- a/sound/core/timer.c +++ b/sound/core/timer.c @@ -1019,8 +1019,8 @@ static int snd_timer_s_start(struct snd_timer * timer) njiff += timer->sticks - priv->correction; priv->correction = 0; } - priv->last_expires = priv->tlist.expires = njiff; - add_timer(&priv->tlist); + priv->last_expires = njiff; + mod_timer(&priv->tlist, njiff); return 0; }