Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
References: <20200723180533.220312-1-pierre-louis.bossart@linux.intel.com> <20200729154639.1983854-1-jbrunet@baylibre.com> <2ad13f95-434d-376a-bc38-b209623b461e@linux.intel.com> <1jft998jbe.fsf@starbuckisacylon.baylibre.com> <936d6e37-0ad0-b0d7-814a-1ace12087746@linux.intel.com>
User-agent: mu4e 1.3.3; emacs 26.3
From:   Jerome Brunet <jbrunet@baylibre.com>
To:     Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>,
        Mark Brown <broonie@kernel.org>,
        Liam Girdwood <lgirdwood@gmail.com>
Cc:     alsa-devel@alsa-project.org, Stephan Gerhold <stephan@gerhold.net>,
        Kevin Hilman <khilman@baylibre.com>,
        linux-kernel@vger.kernel.org, zhangn1985@outlook.com,
        linux-amlogic@lists.infradead.org
Subject: Re: [PATCH] ASoC: core: restore dpcm flags semantics
In-reply-to: <936d6e37-0ad0-b0d7-814a-1ace12087746@linux.intel.com>
Date:   Fri, 31 Jul 2020 10:06:03 +0200
Message-ID: <1ja6zg85xw.fsf@starbuckisacylon.baylibre.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On Thu 30 Jul 2020 at 18:06, Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> wrote:

> On 7/30/20 4:04 AM, Jerome Brunet wrote:
>>
>> On Wed 29 Jul 2020 at 17:56, Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> wrote:
>>
>>> On 7/29/20 10:46 AM, Jerome Brunet wrote:
>>>> commit b73287f0b0745 ('ASoC: soc-pcm: dpcm: fix playback/capture checks')
>>>> changed dpcm_playback and dpcm_capture semantic by throwing an error if
>>>> these flags are not aligned with DAIs capabilities on the link.
>>>>
>>>> The former semantic did not force the flags and DAI caps to be aligned.
>>>> The flag previously allowed card drivers to disable a stream direction on
>>>> a link (whether or not such feature is deemed useful).
>>>>
>>>> With change ('ASoC: core: use less strict tests for dailink capabilities')
>>>> an error is thrown if the flags and and the DAI caps are not aligned. Those
>>>> parameters were not meant to aligned initially. No technical reason was
>>>> given about why cards should now be considered "broken" in such condition
>>>> is not met, or why it should be considered to be an improvement to enforce
>>>> that.
>>>>
>>>> Forcing the flags to be aligned with DAI caps just make the information
>>>> the flag carry redundant with DAI caps, breaking a few cards along the way.
>>>>
>>>> This change drops the added error conditions and restore the initial flag
>>>> semantics.
>>>
>>> or rather lack thereof.
>>
>> Again, why ? All there is so far is your personal preference. no facts.
>
> What would be the meaning/purpose of a dailink with .dpcm_capture set, with
> only dais that support playback only?
>
> What would be the meaning/purpose of a dailink with .capture_only set, but
> with a dai that supports playback?

You get to throw an error in those case

>
> What happens if none of these flags are set?

I think I already suggested to throw an error in the initial review of
your patch

>
> What happens when all these flags are set?

I don't see the problem here

>
> No one seems to know, so my suggestion is to align first on consistent
> configurations, then see what can be removed.
>
>>   * What we had gave capabilities to the link, independent of the DAI
>>     components. ASoC just computes the intersection of all that to
>>     determine which direction needs to be enabled. Seems rather simple
>>     and straight forward.
>
> that's what my last patch did, and when there is no intersection it
> complains. Please clarify what you expect when there is no overlap between
> dai and dailink capabilities. Keep in mind that we have a mix of hard-codec
> configuration and DT-created ones, your case is not the general one.
>
>>   * It worked for every user of DPCM so a far.
>
> Not completely true, when Morimoto-san added snd_soc_dai_stream_valid() it
> exposed tons of cases where the information on direction was not provided
> in a reliable at the DAI level. I will assert that we are still finding out
> cases with broken DAI configurations, and as a result we will also find
> broken dailink configurations. Your picture of DPCM as a perfectly
> functional system that I broke is a distortion of reality.

If it was not working, it was certainly not clear in the changelog.
What's clear is the regression it caused

>
> The reality is that we have to work in steps, first make sure all DAIs are
> properly described, then work on the dailinks and optimize at a later
> point. we will need warnings to find out what the problem cases are, and
> move slowly.

Sure, have it your way