On Wed, 19 Apr 2017, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Wed, Apr 19, 2017 at 10:21 PM, Laurent Pinchart
> <laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
>>>
>>> This adds a dependency like we have for the other panel drivers.
>>
>> I believe the dependency should be made optional. DPI panels that don't need
>> backlight control should be supported by a kernel that has backlight support
>> compiled out.
>
> That would be nice in principle, but I fear this would cause additional
> problems.
>
>> --- a/include/linux/backlight.h
>> +++ b/include/linux/backlight.h
>> @@ -162,7 +162,7 @@ struct generic_bl_info {
>> void (*kick_battery)(void);
>> };
>>
>> -#ifdef CONFIG_OF
>> +#if IS_ENABLED(CONFIG_OF) && IS_ENABLED(CONFIG_BACKLIGHT_CLASS_DEVICE)
>> struct backlight_device *of_find_backlight_by_node(struct device_node *node);
>> #else
>> static inline struct backlight_device *
>>
>>
>> We might need to create stubs for backlight_force_update() and
>> backlight_device_set_brightness() too.
>>
>
> With BACKLIGHT_CLASS_DEVICE=m, you still get a link error when the user is
> in a built-in driver. Using 'depends on' usually solves this (except for drivers
> that cannot be modules).
>
> There are three possible workarounds for this that I can think of:
>
> - Use 'depends on BACKLIGHT_CLASS_DEVICE || BACKLIGHT_CLASS_DEVICE=n'
> in each driver that implements optional backlight support. We do
> this elsewhere, but
> it's confusing and easy to get wrong.
FWIW I think this is the fix, and not a workaround.
"depends on FOO || FOO=n" is an expression used throughout the kernel,
and it accurately describes the dependency here. Of course, all drivers
implementing this must still wrap backlight class usage around
IS_ENABLED().
BR,
Jani.
>
> - use IS_REACHABLE() instead of IS_ENABLED() when testing for
> backlight support. This will always result in a kernel that builds cleanly,
> but can be surprising for users when backlight support is a module that
> gets loaded at boot, but it is still not used.
>
> - Make BACKLIGHT_CLASS_DEVICE a 'bool' symbol instead, and force the
> core API code to always be built-in or completely disabled. This makes
> it really easy to use, at the expense of a larger kernel image for those that
> currently use a loadable module.
>
> Arnd
> _______________________________________________
> dri-devel mailing list
> dri-devel@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
--
Jani Nikula, Intel Open Source Technology Center
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
