Re: [PATCH] drm: fix up fbdev Kconfig defaults

["Arnd Bergmann" <arnd@arndb.de>]

On Tue, Sep 12, 2023, at 09:14, Geert Uytterhoeven wrote:
> On Mon, Sep 11, 2023 at 10:53 PM Arnd Bergmann <arnd@kernel.org> wrote:
>> --- a/drivers/gpu/drm/Kconfig
>> +++ b/drivers/gpu/drm/Kconfig
>> @@ -135,7 +135,7 @@ config DRM_FBDEV_EMULATION
>>         bool "Enable legacy fbdev support for your modesetting driver"
>>         depends on DRM
>>         select FRAMEBUFFER_CONSOLE_DETECT_PRIMARY if FRAMEBUFFER_CONSOLE
>> -       default y
>> +       default FB
>
> While this is true for existing configs, it is no longer true in general,
> as DRM_FBDEV_EMULATION is no longer related to FB.

I think it still makes some sense though, as configs that have
both DRM and FB enabled almost certainly want this enabled.

>> diff --git a/drivers/video/console/Kconfig b/drivers/video/console/Kconfig
>> index b575cf54174af..83c2d7329ca58 100644
>> --- a/drivers/video/console/Kconfig
>> +++ b/drivers/video/console/Kconfig
>> @@ -74,6 +74,7 @@ config DUMMY_CONSOLE_ROWS
>>  config FRAMEBUFFER_CONSOLE
>>         bool "Framebuffer Console support"
>>         depends on FB_CORE && !UML
>> +       default DRM_FBDEV_EMULATION
>
> Sounds good to me, although it looks a bit strange at first sight
> (FRAMEBUFFER_CONSOLE defaults to n on a system with real fbdev, but
> y on emulated fbdev?).
> So this is the fix for commit a5ae331edb02b ("drm: Drop select
> FRAMEBUFFER_CONSOLE for DRM_FBDEV_EMULATION").

Correct, this should restore the console on configs that
accidentally lost it. The real problem here is much older,
the assymetry between framebuffer-only configs (with console
default off) and DRM configs (with console selected
unconditionally) started back in 2009 with commit 6fcefd56f5060
("drm/kms: fix kms helper license + Kconfig").

I think that was a mistake, but there is little we can do
to fix that now without breaking users.

The only alternative I can think of would be to default-enable
or force-enable FRAMEBUFFER_CONSOLE for any config that includes
both VT_CONSOLE and FB_CORE. This would increase defconfig
builds for systems that currently only want CONFIG_FB for
either FB_DEVICE or LOGO but don't care about
FRAMEBUFFER_CONSOLE. I have no idea who uses such a config,
but I think Javier previously said this was an important
use case.

>> diff --git a/drivers/video/fbdev/core/Kconfig b/drivers/video/fbdev/core/Kconfig
>> index 114cb8aa6c8fd..804c2bec9b43c 100644
>> --- a/drivers/video/fbdev/core/Kconfig
>> +++ b/drivers/video/fbdev/core/Kconfig
>> @@ -28,7 +28,7 @@ config FIRMWARE_EDID
>>  config FB_DEVICE
>>         bool "Provide legacy /dev/fb* device"
>>         depends on FB_CORE
>> -       default y
>> +       default FB
>
> Changing this means possibly causing regressions on systems running
> an fbdev userspace.

How? FB_DEVICE is a new config that was just split out from
CONFIG_FB in 6.6-rc1, so nobody should have any defconfig
that disables CONFIG_FB but relies on the FB_DEVICE default yet.

    Arnd