Re: [PATCH] sound: usb: pcm: fix incorrect power state when playing sound after PM_AUTO suspend

Re: [PATCH] sound: usb: pcm: fix incorrect power state when playing sound after PM_AUTO suspend

[Macpaul Lin]

On Wed, 2020-06-03 at 10:45 +0200, Takashi Iwai wrote:
> On Wed, 03 Jun 2020 08:54:51 +0200,
> Takashi Iwai wrote:
> > 
> > On Wed, 03 Jun 2020 08:28:09 +0200,
> > Takashi Iwai wrote:
> > > 
> > > And, the most suspicious case is the last one,
> > > chip->num_suspended-intf.  It means that the device has multiple
> > > USB interfaces and they went to suspend, while the resume isn't
> > > performed for the all suspended interfaces in return.
> > 
> > If this is the cause, a patch like below might help.
> > It gets/puts the all assigned interfaced instead of only the primary
> > one.
> 
> ... and considering of the problem again, rather the patch below might
> be the right answer.  Now the driver tries to remember at which state
> it entered into the system-suspend.  Upon resume, in return, when the
> state reaches back to that point, set the card state to D0.
> 
> The previous patch can be applied on the top, too, and it might be
> worth to apply both.
> 
> Let me know if any of those actually helps.
> 
> 
> Takashi

Thanks for your response so quickly.
I've just test this patch since it looks like enough for the issue.

This patch worked since the flag system_suspend will be set at the same
time when power state has been changed. I have 2 interface with the head
set. But actually the problem happened when primary one is suspended.
So I didn't test the earlier patch "suspend all interface instead of
only the primary one."

Will you resend this patch officially later? I think this solution is
required to send to stable, too. It's better to have it for other stable
kernel versions include android's.

> ---
> diff --git a/sound/usb/card.c b/sound/usb/card.c
> --- a/sound/usb/card.c
> +++ b/sound/usb/card.c
> @@ -843,9 +843,6 @@ static int usb_audio_suspend(struct usb_interface *intf, pm_message_t message)
>  	if (chip == (void *)-1L)
>  		return 0;
>  
> -	chip->autosuspended = !!PMSG_IS_AUTO(message);
> -	if (!chip->autosuspended)
> -		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D3hot);
>  	if (!chip->num_suspended_intf++) {
>  		list_for_each_entry(as, &chip->pcm_list, list) {
>  			snd_usb_pcm_suspend(as);
> @@ -858,6 +855,11 @@ static int usb_audio_suspend(struct usb_interface *intf, pm_message_t message)
>  			snd_usb_mixer_suspend(mixer);
>  	}
>  
> +	if (!PMSG_IS_AUTO(message) && !chip->system_suspend) {
> +		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D3hot);
> +		chip->system_suspend = chip->num_suspended_intf;
> +	}
> +
>  	return 0;
>  }
>  
> @@ -871,10 +873,10 @@ static int __usb_audio_resume(struct usb_interface *intf, bool reset_resume)
>  
>  	if (chip == (void *)-1L)
>  		return 0;
> -	if (--chip->num_suspended_intf)
> -		return 0;
>  
>  	atomic_inc(&chip->active); /* avoid autopm */
> +	if (chip->num_suspended_intf > 1)
> +		goto out;
>  
>  	list_for_each_entry(as, &chip->pcm_list, list) {
>  		err = snd_usb_pcm_resume(as);
> @@ -896,9 +898,12 @@ static int __usb_audio_resume(struct usb_interface *intf, bool reset_resume)
>  		snd_usbmidi_resume(p);
>  	}
>  
> -	if (!chip->autosuspended)
> + out:
> +	if (chip->num_suspended_intf == chip->system_suspend) {
>  		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D0);
> -	chip->autosuspended = 0;
> +		chip->system_suspend = 0;
> +	}
> +	chip->num_suspended_intf--;
>  
>  err_out:
>  	atomic_dec(&chip->active); /* allow autopm after this point */
> diff --git a/sound/usb/usbaudio.h b/sound/usb/usbaudio.h
> index 1c892c7f14d7..e0ebfb25fbd5 100644
> --- a/sound/usb/usbaudio.h
> +++ b/sound/usb/usbaudio.h
> @@ -26,7 +26,7 @@ struct snd_usb_audio {
>  	struct usb_interface *pm_intf;
>  	u32 usb_id;
>  	struct mutex mutex;
> -	unsigned int autosuspended:1;	
> +	unsigned int system_suspend;
>  	atomic_t active;
>  	atomic_t shutdown;
>  	atomic_t usage_count;
> 
> _______________________________________________

Thank you very much!

Best regards,
Macpaul Lin

Re: [PATCH] sound: usb: pcm: fix incorrect power state when playing sound after PM_AUTO suspend

[Takashi Iwai]

On Wed, 03 Jun 2020 14:39:24 +0200,
Macpaul Lin wrote:
> 
> On Wed, 2020-06-03 at 10:45 +0200, Takashi Iwai wrote:
> > On Wed, 03 Jun 2020 08:54:51 +0200,
> > Takashi Iwai wrote:
> > > 
> > > On Wed, 03 Jun 2020 08:28:09 +0200,
> > > Takashi Iwai wrote:
> > > > 
> > > > And, the most suspicious case is the last one,
> > > > chip->num_suspended-intf.  It means that the device has multiple
> > > > USB interfaces and they went to suspend, while the resume isn't
> > > > performed for the all suspended interfaces in return.
> > > 
> > > If this is the cause, a patch like below might help.
> > > It gets/puts the all assigned interfaced instead of only the primary
> > > one.
> > 
> > ... and considering of the problem again, rather the patch below might
> > be the right answer.  Now the driver tries to remember at which state
> > it entered into the system-suspend.  Upon resume, in return, when the
> > state reaches back to that point, set the card state to D0.
> > 
> > The previous patch can be applied on the top, too, and it might be
> > worth to apply both.
> > 
> > Let me know if any of those actually helps.
> > 
> > 
> > Takashi
> 
> Thanks for your response so quickly.
> I've just test this patch since it looks like enough for the issue.

Good to hear!

> This patch worked since the flag system_suspend will be set at the same
> time when power state has been changed. I have 2 interface with the head
> set. But actually the problem happened when primary one is suspended.

Currently the autosuspend is set only to the primary interface; IOW,
the other interfaces will never get autosuspend, and the another
suspend-all-intf patch should improve that situation.  But it won't
fix your actual bug, obviously :)

> So I didn't test the earlier patch "suspend all interface instead of
> only the primary one."

Could you try it one on top of the last patch?  At least I'd like to
see whether it causes any regression.

> Will you resend this patch officially later? I think this solution is
> required to send to stable, too. It's better to have it for other stable
> kernel versions include android's.

Yes, that's a general bug and worth to be merged quickly.
I'm going to submit a proper patch soon later.


thanks,

Takashi


> 
> > ---
> > diff --git a/sound/usb/card.c b/sound/usb/card.c
> > --- a/sound/usb/card.c
> > +++ b/sound/usb/card.c
> > @@ -843,9 +843,6 @@ static int usb_audio_suspend(struct usb_interface *intf, pm_message_t message)
> >  	if (chip == (void *)-1L)
> >  		return 0;
> >  
> > -	chip->autosuspended = !!PMSG_IS_AUTO(message);
> > -	if (!chip->autosuspended)
> > -		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D3hot);
> >  	if (!chip->num_suspended_intf++) {
> >  		list_for_each_entry(as, &chip->pcm_list, list) {
> >  			snd_usb_pcm_suspend(as);
> > @@ -858,6 +855,11 @@ static int usb_audio_suspend(struct usb_interface *intf, pm_message_t message)
> >  			snd_usb_mixer_suspend(mixer);
> >  	}
> >  
> > +	if (!PMSG_IS_AUTO(message) && !chip->system_suspend) {
> > +		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D3hot);
> > +		chip->system_suspend = chip->num_suspended_intf;
> > +	}
> > +
> >  	return 0;
> >  }
> >  
> > @@ -871,10 +873,10 @@ static int __usb_audio_resume(struct usb_interface *intf, bool reset_resume)
> >  
> >  	if (chip == (void *)-1L)
> >  		return 0;
> > -	if (--chip->num_suspended_intf)
> > -		return 0;
> >  
> >  	atomic_inc(&chip->active); /* avoid autopm */
> > +	if (chip->num_suspended_intf > 1)
> > +		goto out;
> >  
> >  	list_for_each_entry(as, &chip->pcm_list, list) {
> >  		err = snd_usb_pcm_resume(as);
> > @@ -896,9 +898,12 @@ static int __usb_audio_resume(struct usb_interface *intf, bool reset_resume)
> >  		snd_usbmidi_resume(p);
> >  	}
> >  
> > -	if (!chip->autosuspended)
> > + out:
> > +	if (chip->num_suspended_intf == chip->system_suspend) {
> >  		snd_power_change_state(chip->card, SNDRV_CTL_POWER_D0);
> > -	chip->autosuspended = 0;
> > +		chip->system_suspend = 0;
> > +	}
> > +	chip->num_suspended_intf--;
> >  
> >  err_out:
> >  	atomic_dec(&chip->active); /* allow autopm after this point */
> > diff --git a/sound/usb/usbaudio.h b/sound/usb/usbaudio.h
> > index 1c892c7f14d7..e0ebfb25fbd5 100644
> > --- a/sound/usb/usbaudio.h
> > +++ b/sound/usb/usbaudio.h
> > @@ -26,7 +26,7 @@ struct snd_usb_audio {
> >  	struct usb_interface *pm_intf;
> >  	u32 usb_id;
> >  	struct mutex mutex;
> > -	unsigned int autosuspended:1;	
> > +	unsigned int system_suspend;
> >  	atomic_t active;
> >  	atomic_t shutdown;
> >  	atomic_t usage_count;
> > 
> > _______________________________________________
> 
> Thank you very much!
> 
> Best regards,
> Macpaul Lin
> 
> 

Re: [PATCH] sound: usb: pcm: fix incorrect power state when playing sound after PM_AUTO suspend

[Macpaul Lin]

On Wed, 2020-06-03 at 14:47 +0200, Takashi Iwai wrote:
> On Wed, 03 Jun 2020 14:39:24 +0200,
> Macpaul Lin wrote:
> > 
> > On Wed, 2020-06-03 at 10:45 +0200, Takashi Iwai wrote:
> > > On Wed, 03 Jun 2020 08:54:51 +0200,
> > > Takashi Iwai wrote:
> > > > 
> > > > On Wed, 03 Jun 2020 08:28:09 +0200,
> > > > Takashi Iwai wrote:
> > > > > 
> > > > > And, the most suspicious case is the last one,
> > > > > chip->num_suspended-intf.  It means that the device has multiple
> > > > > USB interfaces and they went to suspend, while the resume isn't
> > > > > performed for the all suspended interfaces in return.
> > > > 
> > > > If this is the cause, a patch like below might help.
> > > > It gets/puts the all assigned interfaced instead of only the primary
> > > > one.
> > > 
> > > ... and considering of the problem again, rather the patch below might
> > > be the right answer.  Now the driver tries to remember at which state
> > > it entered into the system-suspend.  Upon resume, in return, when the
> > > state reaches back to that point, set the card state to D0.
> > > 
> > > The previous patch can be applied on the top, too, and it might be
> > > worth to apply both.
> > > 
> > > Let me know if any of those actually helps.
> > > 
> > > 
> > > Takashi
> > 
> > Thanks for your response so quickly.
> > I've just test this patch since it looks like enough for the issue.
> 
> Good to hear!
> 
> > This patch worked since the flag system_suspend will be set at the same
> > time when power state has been changed. I have 2 interface with the head
> > set. But actually the problem happened when primary one is suspended.
> 
> Currently the autosuspend is set only to the primary interface; IOW,
> the other interfaces will never get autosuspend, and the another
> suspend-all-intf patch should improve that situation.  But it won't
> fix your actual bug, obviously :)
> 
> > So I didn't test the earlier patch "suspend all interface instead of
> > only the primary one."
> 
> Could you try it one on top of the last patch?  At least I'd like to
> see whether it causes any regression.

I've tried both of these 2 patches together, and it looks okay.

> > Will you resend this patch officially later? I think this solution is
> > required to send to stable, too. It's better to have it for other stable
> > kernel versions include android's.
> 
> Yes, that's a general bug and worth to be merged quickly.
> I'm going to submit a proper patch soon later.
> 
> 
> thanks,
> 
> Takashi
> 

Thanks!
Macpaul Lin