Re: [PATCH 0/4] ASoC: Intel: Mark BE DAIs as nonatomic for hsw and

[Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>]

On 6/27/22 10:41, Cezary Rojewski wrote:
> On 2022-06-27 4:45 PM, Pierre-Louis Bossart wrote:
>> On 6/25/22 03:29, Cezary Rojewski wrote:
> 
>>> Hmm.. that's a good feedback. Isn't ASoC's FE<->BE treated as a single
>>> PCM substream in sound/core/pcm_native.c though? If so, does it even
>>> make sense for card's BE DAI to be atomic, if it's FE counterpart is
>>> nonatomic already? Especially if it is specifying platform and cpu_dai
>>> that matches Intel's components which we know communicate using IPCs.
>>
>> I guess it depends on the cpu_dai implementation. Not all
>> implementations implement a delay in the .trigger callback and/or rely
>> on IPCs.
>>
>>> Warning is one thing, but will you be also getting rid of the
>>> if-statement in soc-pcm.c that actually forces nonatomic=1 on BE when FE
>>> is already declared as such? If the if-statement stays, I believe the
>>> declaring BE DAIs 'correctly' in the way to go.
>>
>> I meant just removing the dev_warn() only.
>>
>> See https://github.com/thesofproject/linux/pull/3723
> 
> So the framework is still fixing the flag for the driver. Ideally we
> would like to have all the drivers assign correct values to ->nonatomic
> flag themselves.
> 
> Now when I think about it, the message seems useful - at least as
> dev_dbg(). It _guides_ driver developer to the desired approach: setting
> the ->nonatomic flag for BE to '1' if the corresponding FE is already
> configured as such.

that would result in unnecessary changes to all machine drivers to get
rid of the message...