[PATCH v2 1/3] ASoC: soc-pcm: add option to start DMA after DAI

Claudiu.Beznea at microchip.com Claudiu.Beznea at microchip.com
Fri Feb 17 05:14:13 PST 2023


On 17.02.2023 15:09, Amadeusz Sławiński wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the
> content is safe
> 
> On 2/17/2023 1:41 PM, Claudiu Beznea wrote:
>> Add option to start DMA component after DAI trigger. This is done
>> by filling the new struct snd_soc_component_driver::start_dma_last.
>>
>> Signed-off-by: Claudiu Beznea <claudiu.beznea at microchip.com>
>> ---
>>   include/sound/soc-component.h |  2 ++
>>   sound/soc/soc-pcm.c           | 27 ++++++++++++++++++++++-----
>>   2 files changed, 24 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/sound/soc-component.h b/include/sound/soc-component.h
>> index 3203d35bc8c1..0814ed143864 100644
>> --- a/include/sound/soc-component.h
>> +++ b/include/sound/soc-component.h
>> @@ -190,6 +190,8 @@ struct snd_soc_component_driver {
>>       bool use_dai_pcm_id;    /* use DAI link PCM ID as PCM device number */
>>       int be_pcm_base;        /* base device ID for all BE PCMs */
>>
>> +     unsigned int start_dma_last;
>> +
>>   #ifdef CONFIG_DEBUG_FS
>>       const char *debugfs_prefix;
>>   #endif
>> diff --git a/sound/soc/soc-pcm.c b/sound/soc/soc-pcm.c
>> index 005b179a770a..5eb056b942ce 100644
>> --- a/sound/soc/soc-pcm.c
>> +++ b/sound/soc/soc-pcm.c
>> @@ -1088,22 +1088,39 @@ static int soc_pcm_hw_params(struct
>> snd_pcm_substream *substream,
>>   static int soc_pcm_trigger(struct snd_pcm_substream *substream, int cmd)
>>   {
>>       struct snd_soc_pcm_runtime *rtd = asoc_substream_to_rtd(substream);
>> -     int ret = -EINVAL, _ret = 0;
>> +     struct snd_soc_component *component;
>> +     int ret = -EINVAL, _ret = 0, start_dma_last = 0, i;
>>       int rollback = 0;
>>
>>       switch (cmd) {
>>       case SNDRV_PCM_TRIGGER_START:
>>       case SNDRV_PCM_TRIGGER_RESUME:
>>       case SNDRV_PCM_TRIGGER_PAUSE_RELEASE:
>> +             /* Do we need to start dma last? */
>> +             for_each_rtd_components(rtd, i, component) {
>> +                     if (component->driver->start_dma_last) {
>> +                             start_dma_last = 1;
>> +                             break;
>> +                     }
>> +             }
>> +
>>               ret = snd_soc_link_trigger(substream, cmd, 0);
>>               if (ret < 0)
>>                       goto start_err;
>>
>> -             ret = snd_soc_pcm_component_trigger(substream, cmd, 0);
>> -             if (ret < 0)
>> -                     goto start_err;
>> +             if (start_dma_last) {
>> +                     ret = snd_soc_pcm_dai_trigger(substream, cmd, 0);
>> +                     if (ret < 0)
>> +                             goto start_err;
>> +
>> +                     ret = snd_soc_pcm_component_trigger(substream, cmd,
>> 0);
>> +             } else {
>> +                     ret = snd_soc_pcm_component_trigger(substream, cmd,
>> 0);
>> +                     if (ret < 0)
>> +                             goto start_err;
>>
>> -             ret = snd_soc_pcm_dai_trigger(substream, cmd, 0);
>> +                     ret = snd_soc_pcm_dai_trigger(substream, cmd, 0);
>> +             }
>>   start_err:
>>               if (ret < 0)
>>                       rollback = 1;
> 
> Can all of the above be implemented similarly to already present
> stop_dma_first? It looks similar and I don't see reason to have one flag
> in snd_soc_component_driver and other in snd_soc_dai_link.

That was the other solution identified; I mentioned it in v1; from v1 cover
letter/discussions:

The other solution that was identified for this was to extend the already
existing mechanism around struct snd_soc_dai_link::stop_dma_first. The downside
of this was that a potential struct snd_soc_dai_link::start_dma_last
would have to be populated on sound card driver thus, had to be taken
into account in all sound card drivers. At the moment, the mchp-pdmc is
used only with simple-audio-card. In case of simple-audio-card a new DT
binding would had to be introduced to specify this action on dai-link
descriptions (as of my investigation).

> 
> 



More information about the linux-arm-kernel mailing list