lkml.org 
[lkml]   [2021]   [Oct]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v1 3/4] ASoC: Intel: bytcr_rt5651: use devm_clk_get_optional() for mclk
On Wed, Oct 6, 2021 at 8:12 PM Hans de Goede <hdegoede@redhat.com> wrote:
> On 10/6/21 6:37 PM, Pierre-Louis Bossart wrote:
> > On 10/6/21 11:23 AM, Andy Shevchenko wrote:
> >> On Wed, Oct 06, 2021 at 10:51:52AM -0500, Pierre-Louis Bossart wrote:
> >>> On 10/6/21 10:04 AM, Andy Shevchenko wrote:

...

> >>> I don't get why you removed the test on the BYT_RT5651_MCLK_EN quirk,
> >>> see below it was designed as a fall-back mode. We don't want to return
> >>> an error when we know the clock is not present/desired.
> >>
> >> Why should we do a unneeded test? When we switch to the optional, there
> >> will be no error from these CCF APIs. Besides that it drops indentation
> >> level and makes code neat.
> >
> > By looking at this code only one cannot really visualize that it's a
> > no-op. I personally prefer to see explicit intent rather than have to
> > dig hundreds of lines below what this clock is optional.
> >
> > I am also not even sure that in real products this clock is actually
> > optional, the default is to make use of it:
> >
> > #define BYT_RT5651_DEFAULT_QUIRKS (BYT_RT5651_MCLK_EN | \
> >
> > and the only platform without this clock is "Minnowboard Max B3" -
> > probably not used by anyone. I fried mine a long time ago.
> >
> > We'd need to Hans to comment on this since he's really the only one
> > maintaining this code.
>
> So as Mark wrote in his later reply:
>
> "AIUI with the clock API the idiomatic thing is that any optionality is
> handled at the point where the clock is acquired - if the clock is
> optional you end up with NULL which in the clock API is a dummy clock
> and ignored. The rest of the code then doesn't need to worry about any
> of this stuff and the handling can only be in one place."
>
> Combined with there pretty much always actually being an mclk I believe
> that this patch from Andy results in a nice cleanup so I'm in favor with
> this. And the other cleanups also look sensible to me

Thanks!

> I would like to run a small smoke-test with both the series to make
> sure nothing regresses (should be fine but better safe then sorry).

Thanks ahead!

> Andy I believe that there is going to be a v2 to address a couple
> of nitpicks, right ?

Right.

> Note for testing I would prefer a full v2 series, even if some
> patches don't change. And I assume the same applies to Mark for
> applying this.
>
> Sending partial series with only changed patches on the v2
> send turns things into a puzzle, which is not ideal IMHO.

I'll do it tomorrow.


--
With Best Regards,
Andy Shevchenko

\
 
 \ /
  Last update: 2021-10-06 19:21    [W:0.051 / U:0.276 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site