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

[Lars-Peter Clausen]

On 05/19/2014 10:37 AM, Takashi Iwai wrote:
> At Fri, 16 May 2014 12:51:51 +0200,
> Lars-Peter Clausen wrote:
>>
>> 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?
>
> That's weird.  We have already a check in snd_pcm_pre_pause() in
> pcm_native.c to filter only for substreams with SNDRV_PCM_INFO_PAUSE
> flag.
>
> Could you check in which code path it is triggered?

Uhm, yes, I was only looking at the code, but couldn't find the check that 
makes sure that the PAUSE_PUSH/PAUSE_RELEASE events are only triggered if 
SNDRV_PCM_INFO_PAUSE is set. Looks like I need a pair of glasses.

Thanks and sorry for the noise,
- Lars