[alsa-devel] Question about FE & BE DPCM

[Daniel Baluta]

On Sat, Oct 12, 2019 at 12:57 AM Sridharan, Ranjani
<ranjani.sridharan at intel.com> wrote:
>
>
>
> On Fri, Oct 11, 2019 at 2:05 PM Daniel Baluta <daniel.baluta at gmail.com> wrote:
>>
>> Hi Morimoto-san,
>>
>> I'm really confused on what exactly is Front End and Back End.
>>
>> Is FE a DAI or a DAI link or part of a DAI link?
>>
>> In simple-card.c there is simple_dai_link_of_dpcm function:
>>
>> struct snd_soc_dai_link_component *codecs = dai_link->codecs;
>>
>> »       /* BE is dummy */
>> »       »       codecs->of_node»»       = NULL;
>> »       »       codecs->dai_name»       = "snd-soc-dummy-dai";
>> »       »       codecs->name»   »       = "snd-soc-dummy";
>>
>> »       »       /* FE settings */
>> »       »       dai_link->dynamic»      »       = 1;
>> »       »       dai_link->dpcm_merged_format»   = 1;
>>
>> So, from here it looks like BE/FE are components being part of the same link.
>>
>> Another thing is:
>>
>> asoc_simple_canonicalize_platform function where it does say:
>>
>> »        * DPCM BE can be no platform.
>> »        * Alloced memory will be waste, but not leak.
>> »        */
>> »       if (!dai_link->platforms->of_node)
>> »       »       dai_link->num_platforms = 0;
>>
>> But I do want to create a BE with a platform. So, should I try to
>> revert the following
>> patch:
>>
>> commit 794fcee8da3c0c8a01b08ecad1c181cb0a622868
>> Author: Kuninori Morimoto <kuninori.morimoto.gx at renesas.com>
>> Date:   Wed Jul 10 17:01:12 2019 +0900
>>
>>     ASoC: simple-card-utils: care no Platform for DPCM
>>
>>     commit 34614739988ad ("ASoC: soc-core: support dai_link with
>>     platforms_num != 1") supports multi Platform, and
>>     commit 9f3eb91753451 ("ASoC: simple-card-utils: consider CPU-Platform
>>     possibility") removed no Platform from simple-card.
>>
>>     Multi Platform is now checking both Platform name/of_node are NULL case.
>>     But in normal case, DPCM be doesn't have Platform.
>>
>>
>>
>> Basically, what I want to do is: after the DAI link is created to call:
>> snd_soc_fixup_dai_links_platform_name function to set a proper name
>> for BE platform.
>>
>> The problem is that above mentioned patch sets dai_link->num_platforms = 0;
>> so even the name is set the platform component is not probed.
>
>
> Hi Daniel,
> Have you looked into how SOF does it at the time the dai_link elem in topology is parsed?
> https://github.com/thesofproject/linux/blob/5edcfdf3907354de7a260a451ba611f22c5e6a07/sound/soc/sof/topology.c#L2926
>
> Is this what you are asking about?

Hi Ranjani,

Not really. Due to this patch:

Author: Kuninori Morimoto <kuninori.morimoto.gx at renesas.com>
Date:   Wed Jul 10 17:01:12 2019 +0900

    ASoC: simple-card-utils: care no Platform for DPCM

I cannot probe platform component created by SOF.

Even if I set the correct name for BE's platform name because the
above patch sets
 dai_link->num_platforms = 0 the platform component won't be probed.
So, the topology won't be loaded.

I don't understand the full story here. There are these 3 patches:

(3) 794fcee8da3c ASoC: simple-card-utils: care no Platform for DPCM
(2) 9f3eb9175345 ASoC: simple-card-utils: consider CPU-Platform possibility
(1) 6f0437445735 ASoC: simple-card-utils: don't select unnecessary Platform

which I think they are doing as follows:

(1) Allows a DAI link to have no platform component attached.
(2) There is an usecase where Platform can't be NULL. So, if
platform->of_node is NULL
then is set to platforms->of_node = cpus->of_node.
(3) Now, the problem with patch (2) is when also cpus->of_node is
NULL. In this case
platforms->of_node becomes NULL. And the platforms is
left unspecified (because platforms->name is also NULL). So, I think
(3) is just an
workaround by setting link->num_platforms the function soc_bind_dai_link will
actually skip iterating over platforms.

My suggestion is that asoc_simple_canonicalize_platform should also
check if platform has a name
and in that case to skip zeroing dai_link->num_platforms.

Will send an RFC.