Re: [PATCH] soc: mediatek: turn MTK_PMIC_WRAP into visible symbols

[Arnd Bergmann <arnd@arndb.de>]

On Tue, Sep 26, 2017 at 8:00 AM, Jean Delvare <jdelvare@suse.de> wrote:
> On Thu, 21 Sep 2017 17:01:05 +0800, sean.wang@mediatek.com wrote:
>> From: Sean Wang <sean.wang@mediatek.com>
>>
>> MTK_PMIC_WRAP is the basic and required configuration for those various
>> MediaTek PMICs, so turning MTK_PMIC_WRAP into visible symbols easily
>> allows users tending to have the enablement for those PMICs.
>
> I can't really make sense of the sentence above, sorry, and can't see
> how it matches the change below anyway. MTK_PMIC_WRAP is already a
> visible symbol before this change. The change is probably good in
> itself, but please try to come up with a better description.
>
>> Signed-off-by: Sean Wang <sean.wang@mediatek.com>
>> ---
>>  drivers/soc/mediatek/Kconfig | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/soc/mediatek/Kconfig b/drivers/soc/mediatek/Kconfig
>> index a2fcd7f..d513629 100644
>> --- a/drivers/soc/mediatek/Kconfig
>> +++ b/drivers/soc/mediatek/Kconfig
>> @@ -15,7 +15,7 @@ config MTK_INFRACFG
>>  config MTK_PMIC_WRAP
>>       tristate "MediaTek PMIC Wrapper Support"
>>       depends on ARCH_MEDIATEK
>> -     depends on RESET_CONTROLLER
>> +     select RESET_CONTROLLER
>>       select REGMAP

Your other patch now removes the ARCH_MEDIATEK dependency
and allows enabling the driver for COMPILE_TEST on architectures
that might not have ARCH_HAS_RESET_CONTROLLER set,
so you cannot 'select' the symbol in general.

I think a better solution is to leave the 'depends on
RESET_CONTROLLER' present here, but instead add
'select RESET_CONTROLLER' to the ARCH_MEDIATEK
definition.

Generally speaking, we don't want to mix 'select' and 'depends on'
statements for the same symbol, but if we do, it should be done
consistently in a very clear hierarchy. In case of RESET_CONTROLLER,
the rule seems to be to 'select' it from architectures and platforms,
while using 'depends on' from device drivers that absolutely require it.

Note that for compile-testing, it should be fine to rely on the fallback
definitions in include/linux/reset.h, so you might be able to just
remove the 'depends on' statement completely.

In any case, please try to be clearer about what the patch
actually tries to achieve. Did you run into a build error without
it? If so, please copy the exact build error message into the
patch description.

      Arnd