Re: [PATCH v5 2/5] i2c: piix4: Depends on X86

[Hans de Goede <hansg@kernel.org>]

Hi Geert,

On 10-Jun-25 18:59, Geert Uytterhoeven wrote:
> Hi Hans,
> 
> On Tue, 10 Jun 2025 at 16:55, Hans de Goede <hansg@kernel.org> wrote:
>> On 10-Jun-25 16:53, Hans de Goede wrote:
>>> On 10-Jun-25 16:12, Mario Limonciello wrote:
>>>> On 6/10/2025 2:24 AM, Huacai Chen wrote:
>>>>> On Tue, Jun 10, 2025 at 5:16 PM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>>>>>> On Wed, 23 Apr 2025 at 01:49, Mario Limonciello <superm1@kernel.org> wrote:
>>>>>>> From: Mario Limonciello <mario.limonciello@amd.com>
>>>>>>>
>>>>>>> PIIX4 and compatible controllers are only for X86. As some headers are
>>>>>>> being moved into x86 specific headers PIIX4 won't compile on non-x86.
>>>>>>>
>>>>>>> Suggested-by: Ingo Molnar <mingo@kernel.org>
>>>>>>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>>>>>>
>>>>>> Thanks for your patch, which is now commit 7e173eb82ae97175
>>>>>> ("i2c: piix4: Make CONFIG_I2C_PIIX4 dependent on CONFIG_X86")
>>>>>> in v6.16-rc1.
>>>>>>
>>>>>>> --- a/drivers/i2c/busses/Kconfig
>>>>>>> +++ b/drivers/i2c/busses/Kconfig
>>>>>>> @@ -200,7 +200,7 @@ config I2C_ISMT
>>>>>>>
>>>>>>>   config I2C_PIIX4
>>>>>>>          tristate "Intel PIIX4 and compatible (ATI/AMD/Serverworks/Broadcom/SMSC)"
>>>>>>> -       depends on PCI && HAS_IOPORT
>>>>>>> +       depends on PCI && HAS_IOPORT && X86
>>>>>>
>>>>>> Are you sure this south-bridge is not used on non-x86 platforms?
>>>>>> It is enabled in several non-x86 defconfigs:
>>>>>>
>>>>>>      arch/loongarch/configs/loongson3_defconfig:CONFIG_I2C_PIIX4=y
>>>>>>      arch/mips/configs/ip27_defconfig:CONFIG_I2C_PIIX4=m
>>>>>>      arch/mips/configs/loongson2k_defconfig:CONFIG_I2C_PIIX4=y
>>>>>>      arch/mips/configs/loongson3_defconfig:CONFIG_I2C_PIIX4=y
>>>>>>
>>>>>> The loongarch and loongson entries are probably bogus, but I wouldn't
>>>>>> be surprised if the SGI Onyx and Origin do use Intel south-bridges.
>>>>> Loongson can use AMD SB700/SB800 south bridges, which have I2C_PIIX4.
>>>>
>>>> Well we could revert this patch, but it's going to be a compile failure because of 624b0d5696a89b138408d385899dd35372db324b and other patches that go on top of that.
>>>>
>>>> My current leaning is we make a dummy fch.h header for these archs with #defines for 0.
>>>
>>> Move "fch.h" to include/linux/platform/x86/ so that it is available on all arches
>>> and if necessary ifdef out anything x86 specific in the C-code referencing it ?
>>
>> Correction that should be include/linux/platform_data/x86/
> 
> Is that a good location?

It is a location typically used for x86 headers which we want to be
available when not building on x86 too.

> It is not x86-specific, and I see no platform_data (struct) definitions in
> arch/x86/include/asm/amd/fch.h?

If it is not x86 specific then maybe just include/linux/amd_fch.h ?

Anyways I don't really give much about the exact name, the essence of
my suggestion is that we can fix this by moving fch.h to some place
(and maybe a new name) under include/linux so that the header is also
available when not building for x86.

Regards,

Hans





> 
> Gr{oetje,eeting}s,
> 
>                         Geert
>