ASoC: SOF: handle paused streams during system suspend by ranj063 · Pull Request #3119 · thesofproject/linux

ranj063 · 2021-08-26T00:09:17Z

During system suspend, paused streams do not get suspended.
Therefore, we need to explicitly free these PCMs in the DSP
and free the associated DAPM widgets so that they can be set
up again during resume.

Fixes #3116

plbossart

questions below

sound/soc/sof/sof-audio.c

sound/soc/sof/pm.c

sound/soc/sof/sof-audio.c

There is no restore_stream flag anymmore. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>

plbossart · 2021-08-26T18:19:08Z

sound/soc/sof/sof-audio.c

+					swidget->widget->name);
+				return ret;
+			}
+		}


I am still unclear on how the tests above apply to the PAUSED case only.

the 'free the paused PCMS and associated DAPM widgets' is at best done with an indirect set of criteria?

@plbossart Im not sure Im following your question. But tear_dwn_pipelines() will only be called during system suspend and more specifically when the DSP should be in D3 during system suspend. In this case, all running streams will be suspended. The only spcm's that will have a widget list that is non-empty will be the paused ones. I could use the state check for paused streams but I think its unnecessary. But if you think it makes following the logic easier, I will make the change

I would feel more comfortable if we test the state and flag an error if the status is not paused. That said, if this is racy then it'd be a bad idea and we should only add a comment along the lines of what you wrote "The only spcm's that will have a widget list that is non-empty will be the paused ones"

i dont think it is a racy. let me modify it

but the reason I did not do that is because we are manually forcing the actions of the FE suspend and BE suspend on any streams that were left behind. I cant check the state for the for the first loop to walk through the spcms but the second one to free the DAI widgets doesnt have that option.

If it makes it better, I could change the name of the function to not call it out tear_down_paused_pipelines() to tear_down_all_pipelines() or something similar and add in the comment that it will take care of paused pipelines.

see e.g. soc_pcm_hw_params(), the comment says "This function can also be called multiple times" and internally it calls hw_params for all CPU dais.

I was wrong. I was getting confused with the hw_free where we check if the configred flag is set before freeing the widget.

But when the second BE hw_params comes along, it checks if the state of the all connected FE's are not active in

linux/sound/soc/soc-pcm.c

Line 2890 in c4bbda2

int snd_soc_dpcm_can_be_params(struct snd_soc_pcm_runtime *fe,

I dont see a second BE hw_params in SOF and I'm thinking it is because of this.

There is still a corner case where the application could do multiple hw_params on the two devices, before starting one.
I think it'd need dedicated hacks to test.

But even if that happens, I think the only issue is at the CPU dai level. if we do an internal hw_free to keep the refcounts consistent, regardless of what happens in the ALSA/ASoC core, we should be in good shape. It's done for SoundWire, I am not sure about the other DAIs.

There is still a corner case where the application could do multiple hw_params on the two devices, before starting one.
I think it'd need dedicated hacks to test.

agree. Let me add a fix for this.

just added a patch for this @plbossart

During system suspend, paused streams do not get suspended. Therefore, we need to explicitly free these PCMs in the DSP and free the associated DAPM widgets so that they can be set up again during resume. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>

It is not unusual for ALSA/ASoC hw_params callbacks to be invoked multiple times. Reset and free the DAI widget before reconfiguring it to keep the DAI widget use_count balanced. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>

kv2019i

Thanks. I do wonder whether this should be fixed in ASoC core as well. Whole point of paused state is that some resources/state is kept (otherwise you would just stop), so ignoring this at suspend, just seems like a gap. Otoh, given current interfaces, this fix seems both needed and correct, so approving.

kv2019i · 2021-08-27T09:10:14Z

sound/soc/sof/intel/hda.c

@@ -58,6 +58,13 @@ int hda_ctrl_dai_widget_setup(struct snd_soc_dapm_widget *w)
 		return -EINVAL;


The wording "unusual" is a bit unusual. :) I mean it is allowed and it does happen, so this is fixing a bug. But yeah, I get the point.

bardliao

Thanks @ranj063 It looks reasonable.

ranj063 requested review from bardliao, dbaluta, juimonen, keyonjie, kv2019i, lgirdwood, lyakh and plbossart as code owners August 26, 2021 00:09

ranj063 mentioned this pull request Aug 26, 2021

[SDW] Input/output error is seen if we do suspend in pause #3116

Closed

ranj063 force-pushed the fix/pause_resume branch 4 times, most recently from dcb8591 to 542ebe6 Compare August 26, 2021 02:58

paulstelian97 previously approved these changes Aug 26, 2021

View reviewed changes

plbossart reviewed Aug 26, 2021

View reviewed changes

sound/soc/sof/sof-audio.c Outdated Show resolved Hide resolved

sound/soc/sof/sof-audio.c Show resolved Hide resolved

sound/soc/sof/pm.c Show resolved Hide resolved

sound/soc/sof/sof-audio.c Show resolved Hide resolved

ASoC: SOF: pm: fix a stale comment

f7f9c3a

There is no restore_stream flag anymmore. Signed-off-by: Ranjani Sridharan <ranjani.sridharan@linux.intel.com>

ranj063 dismissed paulstelian97’s stale review via fe024e2 August 26, 2021 14:18

ranj063 force-pushed the fix/pause_resume branch from 542ebe6 to fe024e2 Compare August 26, 2021 14:18

ranj063 requested review from paulstelian97 and plbossart August 26, 2021 15:02

plbossart reviewed Aug 26, 2021

View reviewed changes

ranj063 force-pushed the fix/pause_resume branch from fe024e2 to 2ac6502 Compare August 26, 2021 20:11

ranj063 requested a review from libinyang as a code owner August 26, 2021 23:50

paulstelian97 approved these changes Aug 27, 2021

View reviewed changes

kv2019i approved these changes Aug 27, 2021

View reviewed changes

bardliao approved these changes Aug 27, 2021

View reviewed changes

plbossart approved these changes Aug 27, 2021

View reviewed changes

plbossart merged commit 973aa02 into thesofproject:topic/sof-dev Aug 27, 2021

ranj063 deleted the fix/pause_resume branch August 30, 2021 20:11

		@@ -58,6 +58,13 @@ int hda_ctrl_dai_widget_setup(struct snd_soc_dapm_widget *w)
		return -EINVAL;

Conversation

ranj063 commented Aug 26, 2021

Uh oh!

plbossart left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ranj063 Aug 26, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ranj063 Aug 26, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kv2019i left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bardliao left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ranj063 Aug 26, 2021 •

edited

Loading

ranj063 Aug 26, 2021 •

edited

Loading