Re: [GIT PULL] Backlight for v6.1

[Randy Dunlap]

On 10/9/22 05:58, Hans de Goede wrote:
> Hi,
> 
> On 10/9/22 01:23, Linus Torvalds wrote:
>> On Sat, Oct 8, 2022 at 12:59 PM Hans de Goede <hdegoede@xxxxxxxxxx> wrote:
>>>>
>>>> Yeah, I don't expect he driver to work on real hardware,
>>>
>>> I'm not sure what you mean here. I guess you mean that you
>>> do not expect to be able to test the driver on real hw
>>> yourself?
>>
>> Well, that too, but I really along the lines of "make it build as a
>> module when I2C_DESIGNWARE_PLATFORM is a module"
> 
> Actually because the driver only works properly when builtin the
> driver itself is a bool option not a tristate. So I am not sure
> if even if we change the the:
> 
>         depends on I2C_DESIGNWARE_PLATFORM=y
> 
> to:
> 
>         depends on I2C_DESIGNWARE_PLATFORM=y || COMPILE_TEST

That change worked for me.

drivers/mfd/intel_soc_pmic_crc.o is built and
drivers/i2c/busses/i2c-designware-platform.ko is built.

> it still will not be build with an "allmodconfig" build? I must
> admit I don't know what that does with bool options.

I don't see a problem...

but I didn't test all of the other x86 PMIC drivers.
They are all built as loadable modules since I didn't make any changes
to their Kconfig entries.

> If it says yes to bool options, then yes adding the " || COMPILE_TEST"
> will give you build coverage with an "allmodconfig" build and
> I would be happy to submit a patch adding this for all the affected
> x86 PMIC drivers.
> 
>> Because if it depends on some symbols from I2C_DESIGNWARE_PLATFORM,
>> and that one can be a module, then the Intel PMIC driver also needs to
>> be built as a module to just get the build coverage, at least.

I don't see intel_soc_pmic_crc.c using any direct calls into
i2c-designware-platform code. If it calls into it, it must be thru some
indirect pointers (?).
It's not causing any build problems once the "|| COMPILE_TEST"
change is made.  (I am build testing on linux-next-20221019.)

> See above, currently these are bool-s I guess we may be able to
> come up with some Kconfig magic where they can be build as module
> only when COMPILE_TEST is set ? Anyone now the right Kconfig magic
> for this ?
> 
>> And I can imagine that that will not work very well on actual hardware
>> with some of these core drivers that may want to initialize early?
> 
> Right, we want to enforce things, including the i2c controller driver
> to be build in because that is necessary on actual hw and allowing
> these to be build as modules has lead to bug reports in the past.
> 
>> But I'd love to at least have the build coverage.
> 
> I agree, see above for some gotchas though.
> 
> FWIW as a side-project I actively work on keeping these platforms working
> well with the latest mainline kernel, so I do build these with the latest
> rc kernels all the time.


-- 
~Randy