On Tue, 17 Dec 2019, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
> On Tue, Dec 17, 2019 at 1:56 PM Steven Price <steven.price@xxxxxxx> wrote:
>> On 17/12/2019 06:37, Randy Dunlap wrote:
>> > On 12/16/19 9:42 PM, Sam Ravnborg wrote:
>> >> On Mon, Dec 16, 2019 at 08:25:11AM -0800, Randy Dunlap wrote:
>> >>> On 12/15/19 9:22 PM, Stephen Rothwell wrote:
>
>> >>> on x86_64:
>> >>>
>> >>> ld: drivers/gpu/drm/drm_panel.o: in function `drm_panel_of_backlight':
>> >>> (.text+0x2ee): undefined reference to `devm_of_find_backlight'
>> >>>
>> >>> ld: drivers/gpu/drm/i915/display/intel_panel.o: in function `intel_backlight_device_register':
>> >>> intel_panel.c:(.text+0x593e): undefined reference to `backlight_device_register'
>> >>> ld: drivers/gpu/drm/i915/display/intel_panel.o: in function `intel_backlight_device_unregister':
>> >>> intel_panel.c:(.text+0x5a04): undefined reference to `backlight_device_unregister'
>> >>>
>> >>> CONFIG_DRM_PANEL=y
>> >>> CONFIG_BACKLIGHT_CLASS_DEVICE=m
>> >>> CONFIG_DRM_I915=y
>> >>>
>> >>> Full randconfig file is attached.
>> >>
>> >> Can you please verify if you have:
>> >> 907aa265fde6589b8059dc51649c6d1f49ade2f3
>> >> ("drm/drm_panel: fix EXPORT of drm_panel_of_backlight")
>> >>
>> >> This commit is supposed to fix it.
>> >>
>> >> Sam
>> >>
>> >
>> > Hi Sam,
>> > I don't have the linux-next.git tree so I can't check that.
>> > I just built whatever is in linux-next of 20191216.
>> >
>>
>> 907aa265fde6589b8059dc51649c6d1f49ade2f3 ("drm/drm_panel: fix EXPORT of
>> drm_panel_of_backlight") is fixing drm_panel_of_backlight(), but the
>> error above is for backlight_device_register().
>>
>> From what I can tell, that commit is actually the cause of the error -
>> now intel_backlight_device_register() is being included in the kernel
>> even though it calls backlight_device_register() which is in a module.
>> Of course it also fixed the original error, so reverting it isn't any
>> use.
>>
>> The below Kconfig change fixes the build for me, but I've no idea
>> whether this is the correct fix.
>
> I think the proper one is to have s/IS_ENABLED/IS_REACHABLE/.
> It fixes issue for me.
As discussed off-line, this will allow silently building and linking a
configuration that's actually broken. (No backlight support despite
expectations.)
IMO deep down the problem is that we "select" BACKLIGHT_CLASS_DEVICE all
over the place, while we should "depends on" it. Everything else is just
duct tape that allows configurations where built-in code calls backlight
symbols in modules. It used to be more about an interaction with ACPI,
now we've added DRM_PANEL to the mix.
I've proposed a fix five years ago [1]. That's what it takes to fix
these recurring failures for good. I'm not really all that interested in
the whack-a-mole with the hacks.
BR,
Jani.
[1] http://lore.kernel.org/r/1413580403-16225-1-git-send-email-jani.nikula@xxxxxxxxx
--
Jani Nikula, Intel Open Source Graphics Center
