ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

[Lars-Peter Clausen]

Hi,

The tegra pcm driver tries to emulate SNDRV_PCM_TRIGGER_PAUSE and
SNDRV_PCM_TRIGGER_RESUME. By passing SNDRV_PCM_TRIGGER_START and
SNDRV_PCM_TRIGGER_STOP to snd_dmaengine_pcm_trigger. But does this really
work? The old non-dmaengine based driver implemented this properly. I'd
reset the dma position pointer for SNDRV_PCM_TRIGGER_START, but keep it
where it was for SNDRV_PCM_TRIGGER_RESUME. But as far as I can see with the
dmaengine based driver the pointer always gets reset to 0 for both
SNDRV_PCM_TRIGGER_START and SNDRV_PCM_TRIGGER_RESUME.

- Lars

Re: ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

[Stephen Warren]

On 03/22/2013 10:26 AM, Lars-Peter Clausen wrote:
> Hi,
> 
> The tegra pcm driver tries to emulate SNDRV_PCM_TRIGGER_PAUSE and
> SNDRV_PCM_TRIGGER_RESUME. By passing SNDRV_PCM_TRIGGER_START and
> SNDRV_PCM_TRIGGER_STOP to snd_dmaengine_pcm_trigger. But does this really
> work? The old non-dmaengine based driver implemented this properly. I'd
> reset the dma position pointer for SNDRV_PCM_TRIGGER_START, but keep it
> where it was for SNDRV_PCM_TRIGGER_RESUME. But as far as I can see with the
> dmaengine based driver the pointer always gets reset to 0 for both
> SNDRV_PCM_TRIGGER_START and SNDRV_PCM_TRIGGER_RESUME.

It's quite possible this hasn't been tested, or tested recently. I don't
recall how to test PAUSE/RESUME. Can you enlighten me?

Laxman, can you comment here?

Re: ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

[Lars-Peter Clausen]

On 03/23/2013 12:35 AM, Stephen Warren wrote:
> On 03/22/2013 10:26 AM, Lars-Peter Clausen wrote:
>> Hi,
>>
>> The tegra pcm driver tries to emulate SNDRV_PCM_TRIGGER_PAUSE and
>> SNDRV_PCM_TRIGGER_RESUME. By passing SNDRV_PCM_TRIGGER_START and
>> SNDRV_PCM_TRIGGER_STOP to snd_dmaengine_pcm_trigger. But does this really
>> work? The old non-dmaengine based driver implemented this properly. I'd
>> reset the dma position pointer for SNDRV_PCM_TRIGGER_START, but keep it
>> where it was for SNDRV_PCM_TRIGGER_RESUME. But as far as I can see with the
>> dmaengine based driver the pointer always gets reset to 0 for both
>> SNDRV_PCM_TRIGGER_START and SNDRV_PCM_TRIGGER_RESUME.
> 
> It's quite possible this hasn't been tested, or tested recently. I don't
> recall how to test PAUSE/RESUME. Can you enlighten me?

Hm, I don't know of an application that is specificity designed to test
pause/resume, but any application using snd_pcm_pause, should be fine.

- Lars

Re: ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

[Lars-Peter Clausen]

On 03/23/2013 02:16 PM, Laxman Dewangan wrote:
> On Saturday 23 March 2013 05:05 AM, Stephen Warren wrote:
>> On 03/22/2013 10:26 AM, Lars-Peter Clausen wrote:
>>> Hi,
>>>
>>> The tegra pcm driver tries to emulate SNDRV_PCM_TRIGGER_PAUSE and
>>> SNDRV_PCM_TRIGGER_RESUME. By passing SNDRV_PCM_TRIGGER_START and
>>> SNDRV_PCM_TRIGGER_STOP to snd_dmaengine_pcm_trigger. But does this really
>>> work? The old non-dmaengine based driver implemented this properly. I'd
>>> reset the dma position pointer for SNDRV_PCM_TRIGGER_START, but keep it
>>> where it was for SNDRV_PCM_TRIGGER_RESUME. But as far as I can see with the
>>> dmaengine based driver the pointer always gets reset to 0 for both
>>> SNDRV_PCM_TRIGGER_START and SNDRV_PCM_TRIGGER_RESUME.
>> It's quite possible this hasn't been tested, or tested recently. I don't
>> recall how to test PAUSE/RESUME. Can you enlighten me?
>>
>> Laxman, can you comment here?
> The tegra20/Tegra30 does not support the pause of the channel. It was
> through global pause. hence the pause and resume is implemented like stop
> transfer and again restart transfer. hence you are seeing this as the  reset
> pointer as 0.
> 

Yea, but in that case you shouldn't implement pause/resume at all. You've
changed the semantics on pause/resume on which applications rely. If
pause/resume are not implemented applications usually workaround it.

- Lars

> So pause is stop and resume is the start again.
> 
> If it is require to implement the dmaengine_pause/resume() then it can not
> be done truely, it can be done atelast in resolution of one buffer transfer.
> We can not stop dma in mid of buffer transfer and start from there. We can
> do cancel of current buffer transfer and then start from begining of current
> buffer, not from the point it was paused.
> 
> However, tegra114 support this pause and resume truely.

Re: ASoC: tegra: Does the PCM_TRIGGER_{PAUSE, RESUME} emulation really work?

[Lars-Peter Clausen]

On 03/23/2013 02:28 PM, Laxman Dewangan wrote:
> On Saturday 23 March 2013 06:59 PM, Lars-Peter Clausen wrote:
>> On 03/23/2013 02:16 PM, Laxman Dewangan wrote:
>>> On Saturday 23 March 2013 05:05 AM, Stephen Warren wrote:
>>>> On 03/22/2013 10:26 AM, Lars-Peter Clausen wrote:
>>>>> Hi,
>>>>>
>>>>> The tegra pcm driver tries to emulate SNDRV_PCM_TRIGGER_PAUSE and
>>>>> SNDRV_PCM_TRIGGER_RESUME. By passing SNDRV_PCM_TRIGGER_START and
>>>>> SNDRV_PCM_TRIGGER_STOP to snd_dmaengine_pcm_trigger. But does this really
>>>>> work? The old non-dmaengine based driver implemented this properly. I'd
>>>>> reset the dma position pointer for SNDRV_PCM_TRIGGER_START, but keep it
>>>>> where it was for SNDRV_PCM_TRIGGER_RESUME. But as far as I can see with
>>>>> the
>>>>> dmaengine based driver the pointer always gets reset to 0 for both
>>>>> SNDRV_PCM_TRIGGER_START and SNDRV_PCM_TRIGGER_RESUME.
>>>> It's quite possible this hasn't been tested, or tested recently. I don't
>>>> recall how to test PAUSE/RESUME. Can you enlighten me?
>>>>
>>>> Laxman, can you comment here?
>>> The tegra20/Tegra30 does not support the pause of the channel. It was
>>> through global pause. hence the pause and resume is implemented like stop
>>> transfer and again restart transfer. hence you are seeing this as the  reset
>>> pointer as 0.
>>>
>> Yea, but in that case you shouldn't implement pause/resume at all. You've
>> changed the semantics on pause/resume on which applications rely. If
>> pause/resume are not implemented applications usually workaround it.
>>
>> - Lars
>>
> 
> So should we return -EINVAL in case of following?
> 
>         case SNDRV_PCM_TRIGGER_RESUME:
>         case SNDRV_PCM_TRIGGER_PAUSE_PUSH:
> 

Don't set the SNDRV_PCM_INFO_PAUSE and SNDRV_PCM_INFO_RESUME bits in the
pcm_hardware struct and use snd_dmaengine_pcm_trigger as your trigger
callback. The ALSA core will take care of everything else.

- Lars