Skip to contentSkip to conversation

Power management state flow in ALSA

DISCUSSION Nobin Mathew 3 participants 11 messages May 30, 2007 – Jun 6, 2007 1 message not in the archive

lore
Share & tools
lore
#0 lore raw

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

I am using linux kernel 2.6.17
ALSA driver version 1.0.11rc4
ASoC patch version 0.12.4

#1 lore raw

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#2 lore raw

In suspend () the application is dead (freezed state) before ALSA driver suspend() is called, so in this there is no way application will get to know the SUSPENDED state of driver.

In resume () ALSA driver resume () (changes the state of driver) is called first and then applications are activated.

So how the application will get to know the SUSPENDED state of driver through syscall.No syscall () from ALSA apps(freezed) is happening during the SUSPENDED duration of ALSA driver.

Am i going wrong ?

Quoted 32 lines · Takashi Iwai

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#3 lore raw

At Wed, 30 May 2007 15:36:33 +0530,

Quoted 6 lines · Nobin Mathew

In suspend () the application is dead (freezed state) before ALSA driver suspend() is called, so in this there is no way application will get to know the SUSPENDED state of driver.

Quoted 6 lines

In resume () ALSA driver resume () (changes the state of driver) is called first and then applications are activated.

So how the application will get to know the SUSPENDED state of driver through syscall.No syscall () from ALSA apps(freezed) is happening during the SUSPENDED duration of ALSA driver.

Your app shall issue syscalls sooner or later, otherwise you'll have no I/O :)

The concept of the (PCM) resume in ALSA is a passive way. The driver does _NOT_ resume streams by itself. It waits until the app requests to resume. This is designed so because usually the hardware cannot be recovered in 100% identical state as before, and often the app needs to reset something for the proper restart.

So, when resume callback is executed and the whole kernel PM thing is finished, the user-process restarts again. Then it issues syscalls, and gets to know to know that the stream is in the suspended state. Now it calls alsa-lib snd_pcm_resume() function which issues RESUME ioctl to restart.

Takashi

Quoted 37 lines · Takashi Iwai
Quoted 32 lines

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#4 lore raw

Sorry for my blunders.

driver resume () is not calling snd_pcm_resume().

Suspend only calls snd_pcm_suspend ().

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

Quoted 71 lines · Takashi Iwai

At Wed, 30 May 2007 15:36:33 +0530,

Quoted 6 lines · Nobin Mathew

In suspend () the application is dead (freezed state) before ALSA driver suspend() is called, so in this there is no way application will get to know the SUSPENDED state of driver.

Quoted 6 lines

In resume () ALSA driver resume () (changes the state of driver) is called first and then applications are activated.

So how the application will get to know the SUSPENDED state of driver through syscall.No syscall () from ALSA apps(freezed) is happening during the SUSPENDED duration of ALSA driver.

Your app shall issue syscalls sooner or later, otherwise you'll have no I/O :)

The concept of the (PCM) resume in ALSA is a passive way. The driver does _NOT_ resume streams by itself. It waits until the app requests to resume. This is designed so because usually the hardware cannot be recovered in 100% identical state as before, and often the app needs to reset something for the proper restart.

So, when resume callback is executed and the whole kernel PM thing is finished, the user-process restarts again. Then it issues syscalls, and gets to know to know that the stream is in the suspended state. Now it calls alsa-lib snd_pcm_resume() function which issues RESUME ioctl to restart.

Takashi

Quoted 37 lines · Takashi Iwai
Quoted 32 lines

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#5 lore raw

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 4 lines · Nobin Mathew

Sorry for my blunders.

driver resume () is not calling snd_pcm_resume().

As I wrote: *alsa-lib* snd_pcm_resume() function.
It's no function in the kernel.

Takashi

Quoted 88 lines

Suspend only calls snd_pcm_suspend ().

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

Quoted 71 lines · Takashi Iwai

At Wed, 30 May 2007 15:36:33 +0530,

Quoted 6 lines · Nobin Mathew

In suspend () the application is dead (freezed state) before ALSA driver suspend() is called, so in this there is no way application will get to know the SUSPENDED state of driver.

Quoted 6 lines

In resume () ALSA driver resume () (changes the state of driver) is called first and then applications are activated.

So how the application will get to know the SUSPENDED state of driver through syscall.No syscall () from ALSA apps(freezed) is happening during the SUSPENDED duration of ALSA driver.

Your app shall issue syscalls sooner or later, otherwise you'll have no I/O :)

The concept of the (PCM) resume in ALSA is a passive way. The driver does _NOT_ resume streams by itself. It waits until the app requests to resume. This is designed so because usually the hardware cannot be recovered in 100% identical state as before, and often the app needs to reset something for the proper restart.

So, when resume callback is executed and the whole kernel PM thing is finished, the user-process restarts again. Then it issues syscalls, and gets to know to know that the stream is in the suspended state. Now it calls alsa-lib snd_pcm_resume() function which issues RESUME ioctl to restart.

Takashi

Quoted 37 lines · Takashi Iwai
Quoted 32 lines

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#6 lore raw

It is there in kernel see the file sound/core/pcm_native.c. but it is static, not exported.

So no driver can call directly, may be only Ioctl. I got it.

#ifdef CONFIG_PM
/* resume */

static int snd_pcm_pre_resume(struct snd_pcm_substream *substream, int state)

{
	struct snd_pcm_runtime *runtime = substream->runtime;
	if (!(runtime->info & SNDRV_PCM_INFO_RESUME))
		return -ENOSYS;
	runtime->trigger_master = substream;
	return 0;
}

static int snd_pcm_do_resume(struct snd_pcm_substream *substream, int state)

{
	struct snd_pcm_runtime *runtime = substream->runtime;
	if (runtime->trigger_master != substream)
		return 0;
	/* DMA not running previously? */
	if (runtime->status->suspended_state != SNDRV_PCM_STATE_RUNNING &&
	    (runtime->status->suspended_state != SNDRV_PCM_STATE_DRAINING ||
	     substream->stream != SNDRV_PCM_STREAM_PLAYBACK))
		return 0;
	return substream->ops->trigger(substream, SNDRV_PCM_TRIGGER_RESUME);
}

static void snd_pcm_undo_resume(struct snd_pcm_substream *substream, int state)

{
	if (substream->runtime->trigger_master == substream &&
	    snd_pcm_running(substream))
		substream->ops->trigger(substream, SNDRV_PCM_TRIGGER_SUSPEND);
}

static void snd_pcm_post_resume(struct snd_pcm_substream *substream, int state)

{
	struct snd_pcm_runtime *runtime = substream->runtime;
	snd_pcm_trigger_tstamp(substream);
	if (substream->timer)
		snd_timer_notify(substream->timer, SNDRV_TIMER_EVENT_MRESUME,
				 &runtime->trigger_tstamp);
	runtime->status->state = runtime->status->suspended_state;
	if (runtime->sleep_min)
		snd_pcm_tick_prepare(substream);
}

static struct action_ops snd_pcm_action_resume = {
.pre_action = snd_pcm_pre_resume, .do_action = snd_pcm_do_resume, .undo_action = snd_pcm_undo_resume, .post_action = snd_pcm_post_resume
};

static int snd_pcm_resume(struct snd_pcm_substream *substream)

{
	struct snd_card *card = substream->pcm->card;
	int res;

	snd_power_lock(card);
	if ((res = snd_power_wait(card, SNDRV_CTL_POWER_D0)) >= 0)
		res = snd_pcm_action_lock_irq(&snd_pcm_action_resume, substream, 0);
	snd_power_unlock(card);
	return res;
}

#else

static int snd_pcm_resume(struct snd_pcm_substream *substream)

{
	return -ENOSYS;
}

#endif /* CONFIG_PM */

Quoted 102 lines · Takashi Iwai

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 4 lines · Nobin Mathew

Sorry for my blunders.

driver resume () is not calling snd_pcm_resume().

As I wrote: *alsa-lib* snd_pcm_resume() function.
It's no function in the kernel.

Takashi

Quoted 88 lines

Suspend only calls snd_pcm_suspend ().

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

Quoted 71 lines · Takashi Iwai

At Wed, 30 May 2007 15:36:33 +0530,

Quoted 6 lines · Nobin Mathew

In suspend () the application is dead (freezed state) before ALSA driver suspend() is called, so in this there is no way application will get to know the SUSPENDED state of driver.

Quoted 6 lines

In resume () ALSA driver resume () (changes the state of driver) is called first and then applications are activated.

So how the application will get to know the SUSPENDED state of driver through syscall.No syscall () from ALSA apps(freezed) is happening during the SUSPENDED duration of ALSA driver.

Your app shall issue syscalls sooner or later, otherwise you'll have no I/O :)

The concept of the (PCM) resume in ALSA is a passive way. The driver does _NOT_ resume streams by itself. It waits until the app requests to resume. This is designed so because usually the hardware cannot be recovered in 100% identical state as before, and often the app needs to reset something for the proper restart.

So, when resume callback is executed and the whole kernel PM thing is finished, the user-process restarts again. Then it issues syscalls, and gets to know to know that the stream is in the suspended state. Now it calls alsa-lib snd_pcm_resume() function which issues RESUME ioctl to restart.

Takashi

Quoted 37 lines · Takashi Iwai
Quoted 32 lines

At Wed, 30 May 2007 11:52:31 +0530,

Quoted 16 lines · Nobin Mathew

I am having a doubt regarding ALSA power management.

My understanding of APM suspend() is like this.

Freeze the ALSA apps

Call ALSA driver suspend ()

in the ALSA suspend() function it saves the current state of substream and changes the state of substream to SUSPENDED.

My understanding of APM resume() is like this

Call ALSA driver resume ()

Quoted 7 lines

Activate the ALSA apps

In ALSA resume function it restores the saved state of substream.

So my question is when ALSA app will get to know the SUSPENDED state of substream.???

When issuing any syscalls. Then you'll get ESTRPIPE error, which indicitaes the stream is in the SUSPEND state.

Takashi

#7 lore raw
In reply to Nobin Mathew

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 10 lines · Nobin Mathew

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

I suppose so.
In soc_suspend() and soc_resume(), you can get the card instance from codec->card, at least.

But currently SoC doesn't keep the PCM instances, so it cannot be implemented as it is... Liam, any plan or known issue regarding this?

Takashi

#8 lore raw
Quoted 11 lines · Takashi Iwai

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 9 lines · Nobin Mathew

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

Quoted 9 lines
Quoted 1 line

In soc-core.c ?

I suppose so.
In soc_suspend() and soc_resume(), you can get the card instance from codec->card, at least.

But currently SoC doesn't keep the PCM instances, so it cannot be implemented as it is... Liam, any plan or known issue regarding this?

I've just logged a bug :-

https://bugtrack.alsa-project.org/alsa-bug/view.php?id=3131

Should get some time to look at this later in the week.

Liam

#9 lore raw
In reply to Takashi Iwai

Nobin,

Can you try the following patch :-

http://opensource.wolfsonmicro.com/cgi-bin/gitweb/gitweb.cgi?p=linux-2.6-asoc;a=commit;h=6c868238a5e083dca4d74439a7fd467b5c7726b0

This will notify pcm's of their suspend state.

Liam

Quoted 22 lines · Takashi Iwai

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 10 lines · Nobin Mathew

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

I suppose so.
In soc_suspend() and soc_resume(), you can get the card instance from codec->card, at least.

But currently SoC doesn't keep the PCM instances, so it cannot be implemented as it is... Liam, any plan or known issue regarding this?

Takashi

A message that is not in the archive (sent off-list or not yet received): [email protected]

#11 lore raw

Liam,

In soc_resume()
we should do this

          snd_power_change_state(codec->card, SNDRV_CTL_POWER_D0);
instead of
        snd_power_change_state(codec->card, SNDRV_CTL_POWER_D3hot);

Because snd_pcm_resume in pcm_native.c expects the card to be in SNDRV_CTL_POWER_D0 state. Other ioctl will wait forever.

static int snd_pcm_resume(struct snd_pcm_substream *substream)

{
	struct snd_card *card = substream->pcm->card;
	int res;

	snd_power_lock(card);
	if ((res = snd_power_wait(card, SNDRV_CTL_POWER_D0)) >= 0)
		res = snd_pcm_action_lock_irq(&snd_pcm_action_resume, substream, 0);
	snd_power_unlock(card);
	return res;
}

One more thing why we are using

	snd_power_change_state(codec->card, SNDRV_CTL_POWER_D3cold);

every other driver is doing snd_power_change_state(card, SNDRV_CTL_POWER_D3hot);

SNDRV_CTL_POWER_D3cold is true for deep sleep where power to the controller is switched off .

In sleep, only clock to the controller is switched off.

Nobin Mathew

Quoted 41 lines · Nobin Mathew

Liam

I will get back, i need to set up the board

Quoted 35 lines · Liam Girdwood

Nobin,

Can you try the following patch :-

http://opensource.wolfsonmicro.com/cgi-bin/gitweb/gitweb.cgi?p=linux-2.6-asoc;a=commit;h=6c868238a5e083dca4d74439a7fd467b5c7726b0

This will notify pcm's of their suspend state.

Liam

Quoted 22 lines · Takashi Iwai

At Wed, 30 May 2007 16:06:54 +0530,

Quoted 10 lines · Nobin Mathew

I am writing an ASoC driver, where i can place these calls

snd_power_change_state(card, SNDRV_CTL_POWER_D3hot); snd_pcm_suspend_all(chip->pcm[i]);

and snd_power_change_state(card, SNDRV_CTL_POWER_D0);

In soc-core.c ?

I suppose so.
In soc_suspend() and soc_resume(), you can get the card instance from codec->card, at least.

But currently SoC doesn't keep the PCM instances, so it cannot be implemented as it is... Liam, any plan or known issue regarding this?

Takashi