diff mbox series

ASoC: core: only flush inited work during free

Message ID 20191128011358.39234-1-cujomalainey@chromium.org (mailing list archive)
State Accepted
Commit 9c9b65203492927cc4ae419e9601e837ecbd889e
Headers show
Series ASoC: core: only flush inited work during free | expand

Commit Message

Curtis Malainey Nov. 28, 2019, 1:13 a.m. UTC
There are many paths to soc_free_pcm_runtime which can both have and
have not yet inited the workqueue yet. When we flush the queue when we
have not yet inited the queue we cause warnings to be printed.

An example is soc_cleanup_card_resources which is called by
snd_soc_bind_card which has multiple failure points before and after
soc_link_init -> soc_new_pcm which is where the queue is inited.

Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>
---
 sound/soc/soc-core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Takashi Iwai Nov. 28, 2019, 6:39 a.m. UTC | #1
On Thu, 28 Nov 2019 02:13:58 +0100,
Curtis Malainey wrote:
> 
> There are many paths to soc_free_pcm_runtime which can both have and
> have not yet inited the workqueue yet. When we flush the queue when we
> have not yet inited the queue we cause warnings to be printed.
> 
> An example is soc_cleanup_card_resources which is called by
> snd_soc_bind_card which has multiple failure points before and after
> soc_link_init -> soc_new_pcm which is where the queue is inited.
> 
> Signed-off-by: Curtis Malainey <cujomalainey@chromium.org>

This patch itself isn't wrong, but there is a generic problem in the
current code in general.  Many fields in snd_soc_pcm_runtime object,
not only delayed_work, are initialized too late where
soc_free_pcm_runtime() may be called beforehand.  For example,
rtd->component_list is initialized after the error path of
rtd->codec_dais kmalloc, although soc_free_pcm_runtime() has a loop
over the link.

That said, at least the things like the linked list head and the work
struct must be initialized right after the allocation before any use
of the object.

For this delayed_work, the situation is a bit complex, though.
Usually the work is set up to point to a fixed function, but in the
case of ASoC, it seems serving for different purposes depending on the
component type.  I guess the cleaner way would be a redirect call
like:

static void rtd_delayed_work(struct work_struct *work)
{
	struct snd_soc_pcm_runtime *rtd =
		container_of(work, struct snd_soc_pcm_runtime, delayed_work.work);
	if (rtd->delayed_work_fn)
		rtd->delayed_work_fn(rtd);
}

static struct snd_soc_pcm_runtime *soc_new_pcm_runtime()
{
	....
	INIT_DELAYED_WORK(&rtd->delayed_work, rtd_delayed_work);
	....
}


thanks,

Takashi
Mark Brown Nov. 28, 2019, 1:49 p.m. UTC | #2
On Thu, Nov 28, 2019 at 07:39:30AM +0100, Takashi Iwai wrote:

> For this delayed_work, the situation is a bit complex, though.
> Usually the work is set up to point to a fixed function, but in the
> case of ASoC, it seems serving for different purposes depending on the
> component type.  I guess the cleaner way would be a redirect call
> like:

Yes, or just separate fields for each.
Curtis Malainey Nov. 28, 2019, 4:23 p.m. UTC | #3
On Thu, Nov 28, 2019 at 5:49 AM Mark Brown <broonie@kernel.org> wrote:
>
> On Thu, Nov 28, 2019 at 07:39:30AM +0100, Takashi Iwai wrote:
>
> > For this delayed_work, the situation is a bit complex, though.
> > Usually the work is set up to point to a fixed function, but in the
> > case of ASoC, it seems serving for different purposes depending on the
> > component type.  I guess the cleaner way would be a redirect call
> > like:
>
> Yes, or just separate fields for each.
Sounds good, I will refactor this change and send a new version next
week as US is on holiday rest of this week.
Mark Brown Nov. 28, 2019, 5:28 p.m. UTC | #4
On Thu, Nov 28, 2019 at 08:23:21AM -0800, Curtis Malainey wrote:
> On Thu, Nov 28, 2019 at 5:49 AM Mark Brown <broonie@kernel.org> wrote:

> > > For this delayed_work, the situation is a bit complex, though.
> > > Usually the work is set up to point to a fixed function, but in the
> > > case of ASoC, it seems serving for different purposes depending on the
> > > component type.  I guess the cleaner way would be a redirect call
> > > like:

> > Yes, or just separate fields for each.


> Sounds good, I will refactor this change and send a new version next
> week as US is on holiday rest of this week.

I applied the change as-is already since like Takashi says it is an
improvement in itself but obviously doing something more complete and
thorough on top of it would be great if you have the time!
Curtis Malainey Dec. 2, 2019, 10:50 p.m. UTC | #5
I sent a patch with the proxy function fix in it. I did not include a
revert for this patch as I didn't see it on any of your branches. Let
me know if I missed it and I can send you a revert as well.

Curtis

On Thu, Nov 28, 2019 at 9:28 AM Mark Brown <broonie@kernel.org> wrote:
>
> On Thu, Nov 28, 2019 at 08:23:21AM -0800, Curtis Malainey wrote:
> > On Thu, Nov 28, 2019 at 5:49 AM Mark Brown <broonie@kernel.org> wrote:
>
> > > > For this delayed_work, the situation is a bit complex, though.
> > > > Usually the work is set up to point to a fixed function, but in the
> > > > case of ASoC, it seems serving for different purposes depending on the
> > > > component type.  I guess the cleaner way would be a redirect call
> > > > like:
>
> > > Yes, or just separate fields for each.
>
>
> > Sounds good, I will refactor this change and send a new version next
> > week as US is on holiday rest of this week.
>
> I applied the change as-is already since like Takashi says it is an
> improvement in itself but obviously doing something more complete and
> thorough on top of it would be great if you have the time!
diff mbox series

Patch

diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c
index 062653ab03a37..3bf702b874a6d 100644
--- a/sound/soc/soc-core.c
+++ b/sound/soc/soc-core.c
@@ -419,7 +419,8 @@  static void soc_free_pcm_runtime(struct snd_soc_pcm_runtime *rtd)
 
 	list_del(&rtd->list);
 
-	flush_delayed_work(&rtd->delayed_work);
+	if (delayed_work_pending(&rtd->delayed_work))
+		flush_delayed_work(&rtd->delayed_work);
 	snd_soc_pcm_component_free(rtd);
 
 	/*