diff mbox series

ASoC: soc-compress: Fix deadlock in soc_compr_open_fe

Message ID 20230613062350.271107-1-yixuanjiang@google.com
State Accepted
Commit 2222214749a9969e09454b9ba7febfdfb09c1c8d
Headers show
Series ASoC: soc-compress: Fix deadlock in soc_compr_open_fe | expand

Commit Message

yixuanjiang June 13, 2023, 6:23 a.m. UTC
Modify the error handling flow by release lock.
The require pcm_mutex will keep holding if open fail.

Fixes: aa9ff6a4955f ("ASoC: soc-compress: Reposition and add pcm_mutex")
Signed-off-by: yixuanjiang <yixuanjiang@google.com>
Cc: stable@vger.kernel.org # v5.15+
---
 sound/soc/soc-compress.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Mark Brown June 15, 2023, 12:56 a.m. UTC | #1
On Tue, Jun 13, 2023 at 02:23:50PM +0800, yixuanjiang wrote:
> Modify the error handling flow by release lock.
> The require pcm_mutex will keep holding if open fail.

> +++ b/sound/soc/soc-compress.c
> @@ -166,6 +166,7 @@ static int soc_compr_open_fe(struct snd_compr_stream *cstream)
>  	snd_soc_dai_compr_shutdown(cpu_dai, cstream, 1);
>  out:
>  	dpcm_path_put(&list);
> +	mutex_unlock(&fe->card->pcm_mutex);
>  be_err:

This is really hard to follow due to the lack of any mutex_lock()s in
the function, I think because this is intended to undo
snd_soc_dpcm_mutex_lock(fe) but if that's the case why is it not using
snd_soc_dpcm_mutex_unlock(fe) like the success path does?  Given the use
of classes not doing that looks like it'll create lockdep issues.

I'd expect the unlock to match the lock.
Carlos Llamas June 15, 2023, 3:26 p.m. UTC | #2
On Thu, Jun 15, 2023 at 01:56:35AM +0100, Mark Brown wrote:
> On Tue, Jun 13, 2023 at 02:23:50PM +0800, yixuanjiang wrote:
> > Modify the error handling flow by release lock.
> > The require pcm_mutex will keep holding if open fail.
> 
> > +++ b/sound/soc/soc-compress.c
> > @@ -166,6 +166,7 @@ static int soc_compr_open_fe(struct snd_compr_stream *cstream)
> >  	snd_soc_dai_compr_shutdown(cpu_dai, cstream, 1);
> >  out:
> >  	dpcm_path_put(&list);
> > +	mutex_unlock(&fe->card->pcm_mutex);
> >  be_err:
> 
> This is really hard to follow due to the lack of any mutex_lock()s in
> the function, I think because this is intended to undo
> snd_soc_dpcm_mutex_lock(fe) but if that's the case why is it not using
> snd_soc_dpcm_mutex_unlock(fe) like the success path does?  Given the use
> of classes not doing that looks like it'll create lockdep issues.
> 
> I'd expect the unlock to match the lock.

Yes, and judging from the context of the patch I believe this was based
off of stable 5.15.y tree. The locking has been refactored since. So
Yixuan, please rebase/adjust your patch on top of Linus's mainline tree
and resend. Thanks!

--
Carlos Llamas
diff mbox series

Patch

diff --git a/sound/soc/soc-compress.c b/sound/soc/soc-compress.c
index 256e45001f85..b6727b91dd85 100644
--- a/sound/soc/soc-compress.c
+++ b/sound/soc/soc-compress.c
@@ -166,6 +166,7 @@  static int soc_compr_open_fe(struct snd_compr_stream *cstream)
 	snd_soc_dai_compr_shutdown(cpu_dai, cstream, 1);
 out:
 	dpcm_path_put(&list);
+	mutex_unlock(&fe->card->pcm_mutex);
 be_err:
 	fe->dpcm[stream].runtime_update = SND_SOC_DPCM_UPDATE_NO;
 	mutex_unlock(&fe->card->mutex);