snd_pcm_nonblock() is called as snd_pcm_abort(). Since
snd_pcm_abort() is called often from a signal handler to clean things
up (e.g. aplay does it), we may face a deadlock if the signal is
raised during the locked operation.
There can be some way to check the deadlock state, but they would cost
much. Since the race condition of snd_pcm_nonblock() is quite small,
let's just drop the locking inside snd_pcm_nonblock() as a
workaround.
Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>
---
src/pcm/pcm.c | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/src/pcm/pcm.c b/src/pcm/pcm.c
index 0cf740f854a9..c136d5579320 100644
--- a/src/pcm/pcm.c
+++ b/src/pcm/pcm.c
@@ -759,7 +759,10 @@ int snd_pcm_nonblock(snd_pcm_t *pcm, int nonblock)
int err = 0;
assert(pcm);
- __snd_pcm_lock(pcm); /* forced lock due to pcm field change */
+ /* FIXME: __snd_pcm_lock() call below is commented out because of the
+ * the possible deadlock in signal handler calling snd_pcm_abort()
+ */
+ /* __snd_pcm_lock(pcm); */ /* forced lock due to pcm field change */
if ((err = pcm->ops->nonblock(pcm->op_arg, nonblock)) < 0)
goto unlock;
if (nonblock == 2) {
@@ -775,7 +778,7 @@ int snd_pcm_nonblock(snd_pcm_t *pcm, int nonblock)
pcm->mode &= ~SND_PCM_NONBLOCK;
}
unlock:
- __snd_pcm_unlock(pcm);
+ /* __snd_pcm_unlock(pcm); */ /* FIXME: see above */
return err;
}
--
2.11.0
_______________________________________________
Alsa-devel mailing list
Alsa-devel@xxxxxxxxxxxxxxxx
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
