[alsa-devel] [PATCH] dma: pl330: Add support for DMA_PAUSE command

[Lars-Peter Clausen]

On 05/16/2014 07:49 AM, Takashi Iwai wrote:
> At Thu, 15 May 2014 21:21:12 +0200,
> Lars-Peter Clausen wrote:
>>
>> On 05/15/2014 02:01 PM, Tushar Behera wrote:
>>> On 14 May 2014 17:54, Lars-Peter Clausen <lars at metafoo.de> wrote:
>>>> On 05/14/2014 02:07 PM, Tushar Behera wrote:
>>>>>
>>>>> On 14 May 2014 17:29, Jassi Brar <jassisinghbrar at gmail.com> wrote:
>>>>>>
>>>>>> On Wed, May 14, 2014 at 8:53 AM, Tushar Behera <tushar.behera at linaro.org>
>>>>>> wrote:
>>>>>>>
>>>>>>> While playing back audio, pmc_dmaengine requests the DMA channel to
>>>>>>> stop DMA transmission through DMA_PAUSE command.
>>>>>>>
>>>>>>> Currently PL330 driver doesn't support DMA pause command, leaving
>>>>>>> the DMA state inconsistent when the system resumes. Instead, it would
>>>>>>> be better to terminate the DMA transfer during suspend and restart
>>>>>>> again during resume.
>>>>>>>
>>>>>>> Tested with audio playback across a suspend-resume cycle.
>>>>>>>
>>>>>> What is pmc_dmaengine? How does DMA_PAUSE help, when there is no
>>>>>> DMA_RESUME?
>>>>>>
>>>>>
>>>>> Sorry, it is a typo.
>>>>>
>>>>> sound/core/pcm_dmaengine.c:snd_dmaengine_pcm_trigger() -->
>>>>> dmaengine_pause() is called during system suspend.
>>>>
>>>>
>>>> It is only called if the DMA driver has support for pausing and resuming DMA
>>>> transfers. Or at least that is the intention.
>>>>
>>>> - Lars
>>>
>>> During suspend, snd_dmaengine_pcm_trigger():SNDRV_PCM_TRIGGER_SUSPEND
>>> is called which unconditionally calls dmaengine_pause(). Should we
>>> update snd_dmaengine_pcm_trigger() to check for DMA pause/resume
>>> support and call dmaengine_pause() or dmaengine_terminate_all()
>>> accordingly?
>>
>> As far as I understand it we do not have to do anything for TRIGGER_SUSPEND
>> if we do not set the SNDRV_PCM_INFO_RESUME flag. It looks like
>> TRIGGER_SUSPEND is called unconditionally during suspend. But since the
>> error code is ignored it should be fine if we just call dmaengine_pause()
>> and that return -ENOSYS or similar.
>
> Well, TRIGGER_SUSPEND is issued by PCM core no matter whether
> SNDRV_PCM_INFO_RESUME flag is set or not.  The resume behavior depends
> on the flag (SNDRV_PCM_TRIGGER_RESUME is issued if the flag is set,
> otherwise the normal setup is done by alsa-lib at resume), but the
> suspend is always triggered in the same way.
>
> So, PCM core assumes that the driver stops the stream somehow by
> SNDRV_PCM_TRIGGER_SUSPEND.

Ok, so we should call terminate_all() when we get a 
SNDRV_PCM_TRIGGER_SUSPEND if the dmaengine driver does not support pauseing 
the transfer. Thanks for the clarification.

On a related note it seems like it is still possible to get 
PAUSE_PUSH/PAUSE_RELEASE events even if SNDRV_PCM_INFO_PAUSE is not set. Do 
you have an idea what should be the right thing to do in such a case? Return 
-ENOSYS?

- Lars