On Sat, Sep 10, 2022 at 12:22:04PM +0200, Takashi Iwai wrote:
> On Fri, 09 Sep 2022 21:39:19 +0200,
> Ville Syrjälä wrote:
> >
> > On Fri, Sep 09, 2022 at 05:59:47PM +0200, Takashi Iwai wrote:
> > > On Fri, 09 Sep 2022 17:45:25 +0200,
> > > Ville Syrjälä wrote:
> > > >
> > > > Hi Takashi,
> > > >
> > > > commit 7206998f578d ("ALSA: hda: Fix potential deadlock at codec
> > > > unbinding") introduced a problem on at least one of my older machines.
> > > >
> > > > The problem happens when hda_codec_driver_remove() encounters a
> > > > codec without any pcms (and thus the refcount is 1) and tries to
> > > > call refcount_dec(). Turns out refcount_dec() doesn't like to be
> > > > used for dropping the refcount to 0, and instead if spews a warning
> > > > and does its saturate thing. The subsequent wait_event() is then
> > > > permanently stuck waiting on the saturated refcount.
> > > >
> > > > I've definitely seen the same kind of pattern used elsewhere
> > > > in the kernel as well, so the fact that refcount_t can't be used
> > > > to implement it is a bit of surprise to me. I guess most other
> > > > places still use atomic_t instead.
> > >
> > > Does the patch below work around it? It seem to be a subtle
> > > difference between refcount_dec() and refcount_dec_and_test().
> >
> > Aye, this works.
> >
> > Tested-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
>
> Good to hear.
>
> I think the below is slightly safer, assuring the other *_disconnect()
> calls.
>
> Could you give it a try again? Once after confirming it works, I'll
> re-submit and merge to my tree.
This works too. Thanks
Tested-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
>
>
> thanks,
>
> Takashi
>
> -- 8< --
> From: Takashi Iwai <tiwai@xxxxxxx>
> Subject: [PATCH] ALSA: hda: Fix hang at HD-audio codec unbinding due to refcount saturation
> MIME-Version: 1.0
> Content-Type: text/plain; charset=UTF-8
> Content-Transfer-Encoding: 8bit
>
> We fixed the potential deadlock at dynamic unbinding the HD-audio
> codec at the commit 7206998f578d ("ALSA: hda: Fix potential deadlock
> at codec unbinding"), but ironically, this caused another potential
> deadlock. The current code uses refcount_dec() and waits for the
> pending task with wait_event for dropping the refcount to 0. This
> works fine when PCMs are assigned and actually waiting for the
> refcount drop.
>
> Meanwhile, when there was no PCM assigned, the refcount_dec() call
> itself was supposed to drop to zero -- alas, it doesn't in reality;
> refcount_dec() complains, spews kernel warning and it saturates
> instead of dropping to 0, due to the nature of refcount_dec()
> implementation. This eventually blocks the wait_event() wakeup and
> the code get stuck there.
>
> For avoiding the problem, we call refcount_dec_and_test() and skips
> the sync-wait if it already reaches to zero.
>
> The patch does a slight code reshuffling to make sure to invoke other
> disconnect calls before the sync-wait, too.
>
> Fixes: 7206998f578d ("ALSA: hda: Fix potential deadlock at codec unbinding")
> Reported-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
> Cc: <stable@xxxxxxxxxxxxxxx>
> Link: https://lore.kernel.org/r/YxtflWQnslMHVlU7@xxxxxxxxx
> Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>
> ---
> sound/pci/hda/hda_bind.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/sound/pci/hda/hda_bind.c b/sound/pci/hda/hda_bind.c
> index cae9a975cbcc..1a868dd9dc4b 100644
> --- a/sound/pci/hda/hda_bind.c
> +++ b/sound/pci/hda/hda_bind.c
> @@ -157,10 +157,10 @@ static int hda_codec_driver_remove(struct device *dev)
> return codec->bus->core.ext_ops->hdev_detach(&codec->core);
> }
>
> - refcount_dec(&codec->pcm_ref);
> snd_hda_codec_disconnect_pcms(codec);
> snd_hda_jack_tbl_disconnect(codec);
> - wait_event(codec->remove_sleep, !refcount_read(&codec->pcm_ref));
> + if (!refcount_dec_and_test(&codec->pcm_ref))
> + wait_event(codec->remove_sleep, !refcount_read(&codec->pcm_ref));
> snd_power_sync_ref(codec->bus->card);
>
> if (codec->patch_ops.free)
> --
> 2.35.3
--
Ville Syrjälä
Intel
