Re: [PATCH] ASoC: Intel: bytcr_wm5102: Fix MCLK leak on platform_clock_control error

["Cássio Gabriel Monteiro Pires" <cassiogabrielcontato@gmail.com>]

[-- Attachment #1.1: Type: text/plain, Size: 1422 bytes --]

On 4/28/26 07:28, Cássio Gabriel Monteiro Pires wrote:
> On 4/28/26 03:55, Andy Shevchenko wrote:
>>
>> There are 6 drivers that do the same, why is only this one special?
>> Have you checked the flow on the error path of the caller of this
>> `platform_clock_control()`? Maybe there it calls with the opposite
>> event to shut the clock down?
>>
>> TL;DR: If it's a real issue, it has to be fixed for all affected drivers.
> 
> Yes, I'm going to check the other drivers.
> 

I checked the other platform_clock_control() users under
sound/soc/intel/boards/.

The same bug pattern exists when the ON path does:

        clk_prepare_enable()
        <fallible codec PLL/sysclk setup>
        return error without clk_disable_unprepare()

The affected drivers are bytcr_rt5640, bytcr_rt5651, cht_bsw_rt5672 and
bytcr_wm5102. The first three were already fixed by a02496a29463,
b022e5c142ef and dced5a373a96. This patch fixes the remaining one.

The other two MCLK platform_clock_control() users, cht_bsw_rt5645 and
cht_bsw_max98090_ti, only enable MCLK in the ON path and do not perform a
later fallible codec-clock setup in that same path, so I do not see the
same leak there.

I also checked the DAPM event path; failed widget callbacks are logged,
but there is no opposite-event rollback guarantee that would balance the
clock enable from the failed callback.

-- 
Thanks,
Cássio


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 236 bytes --]