[PATCH 00/14] ASoC: Intel/SOF: extend run-time driver selection to ACPI devices

[Rojewski, Cezary]

On 2020-11-24 3:15 PM, Takashi Iwai wrote:
> On Tue, 24 Nov 2020 15:01:19 +0100, Mark Brown wrote:
>>
>> On Tue, Nov 24, 2020 at 11:56:36AM +0000, Rojewski, Cezary wrote:
>>
>>> What the patchset presents catpt vs SOF. /sof/ runs through SOF firmware
>>> so it cannot be account as old-implementation. It's a mix of not
>>> recommended fw + incorrect sw flow. As old /haswell/ is no more, there
>>> is no worrying about catpt deployment - it's your only option. As there
>>> is no userspace involved (lack of topology files), base firmware binary
>>> remains the same and amixer kcontrols behave 1:1 when compared to its
>>> predecessor, compatibility is left intact.
>>
>>> That's exactly why we should be explicit in driver selection. Pretty
>>> sure hsw/bdw case is still mistakenly addressed to as if it was
>>> atom-based platform.
>>
>> It's not just the userspace interface that worries people, it's also any
>> board specific quirks that might turn up.  A good chunk of the work with
>> x86 sound support is quirking around platform specifics - look at all
>> the patches Hans sends for example.  In an ideal world this would just
>> be people worrying too much but the general history with getting generic
>> code working well on a wide range of x86 hardware it's hard to blame
>> anyone for being conservative about substantial changes in the software
>> stack.

Mark, there is not a single word I don't agree with in your statement.

In regard to quirks - I was surprised how much detail Hans found out
regarding atom platforms. That's a lot of good input. And that's
probably one of the key reasons why atom is properly supported in linux.

My point has more "basic" nature.
 
> I guess Cezary's point is that CATPT is the only driver for Haswell,
> hence the intel-dsp-config is useless for it.

This! and..

> But I thought CATPT also covers Broadwell, and Broadwell can be
> supported by both CATPT and SOF?  If so, the dynamic switching makes
> sense.

..more. Dynamic selection made sense if you're in transition period as
it is the case for atoms. There is no transition period for hsw/bdw. BDW
as "supported" by SOF would be a strong claim. There is no commitment
and Intel does not recommend using it for hsw/bdw for any scenario. And
as such, selection-subject does not apply here.

Believe removal of /sof/intel/bdw.c is in order?

Czarek