[PATCH 0/7] ASoC: Merge CPU/Codec DAIs

[Pierre-Louis Bossart]

On 3/10/20 9:27 PM, Kuninori Morimoto wrote:
> 
> Hi Pierre-Louis
> 
>>> 	rtd->cpu_dais   = [][][][][][][][][]
>>> 	rtd->codec_dais = [][][][][][][][][]
> (snip)
>>> 	rtd->dais = [][][][][][][][][][][][][][][][][][]
>>> 	            ^cpu_dais         ^codec_dais
>>> 	            |--- num_cpus ---|--- num_codecs --|
> (snip)
>> Is the end-goal to remove the cpu_dais and codec_dais, and fold them
>> as non-descript 'dais'? This is what I understand by "it will be no
>> longer good much for modern device"
> 
> Yes. We want to have non-descript DAIs in the future.
> I think this was indicated by Lars-Peter before at ELCE.
> But, I think we *can't* do it right now.
> Because many drivers are considering CPU and Codec separately.
> 
>> Or is this 'merge' a simple data handling change to avoid using two
>> "for" loops instead of one, and we are going to keep the distinction
>> between dais?
> 
> Yes.
> There are some functions which is doing something only for CPU or Codec.
> This patch-set do nothing to such functions.
> Maybe it can be updated in the future, maybe not (can't).
> 
> I hope this patch-set can be 1st step for non-descript DAIs.
> But the main purpose of this patch so far is that
> keeping current CPU / Codec DAIs method,
> but, simply merge such code *if possible*.
> 
>> more specifically I am concerned about the tons of code we have,
>> e.g. a random machine driver:
>>
>> 	struct snd_soc_dai *codec_dai = rtd->codec_dai;
>> 	struct snd_soc_jack *jack;
>> 	int ret;
>>
>> 	/* Configure sysclk for codec */
>> 	ret = snd_soc_dai_set_sysclk(codec_dai, DA7219_CLKSRC_MCLK,
>>
>> If the rtd structure only has an array of dais, how would the codecs
>> be configured then?
> 
> The image is like this.
> It still can use rtd->cpu_dais, rtd->codec_dais.
> Of course for_each_rtd_cpu/codec_dais() macro too.
> 
> 	rtd->dais = [][][][][][][][][][][][][][][][][][]
> 	            ^cpu_dais         ^codec_dais
> 	            |--- num_cpus ---|--- num_codecs --|
> 
> 	rtd->cpu_dais   = &rtd->dais[0];
> 	rtd->codec_dais = &rtd->dais[dai_link->num_cpus];
> 
> So we can use/keep existing code/method same as before.
> Is this good answer for you ?

Thanks for the precisions.

I have no objections to the addition of the rtd->dais, it can indeed 
simplify the code by just processing all dais in the same loop.

I would still like to make sure we have an broadbrush idea of what the 
2nd step might be. It seems to me it's not possible to avoid having a 
notion of source/sink inside of a dailink (the wording is probably not 
right e.g. for full duplex, maybe domain1/domain2 component1/component2 
are more accurate). The dais are exposed by different components and are 
really the hook by which the components can be configured with 
compatible settings.

Thanks
-Pierre