Re: [RFC 5/6] drm/rcar-du: fix selection of CMM driver

[Geert Uytterhoeven]

Hi Arnd,

On Wed, Apr 15, 2020 at 3:47 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Tue, Apr 14, 2020 at 10:52 PM Laurent Pinchart
> <laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
> > On Tue, Apr 14, 2020 at 10:38:27PM +0200, Arnd Bergmann wrote:
> > > On Tue, Apr 14, 2020 at 10:17 PM Laurent Pinchart wrote:
> > > > On Wed, Apr 08, 2020 at 10:27:10PM +0200, Arnd Bergmann wrote:
> > > > > The 'imply' statement does not seem to have an effect, as it's
> > > > > still possible to turn the CMM code into a loadable module
> > > > > in a randconfig build, leading to a link error:
> > > > >
> > > > > arm-linux-gnueabi-ld: drivers/gpu/drm/rcar-du/rcar_du_crtc.o: in function `rcar_du_crtc_atomic_enable':
> > > > > rcar_du_crtc.c:(.text+0xad4): undefined reference to `rcar_lvds_clk_enable'
> > > > > arm-linux-gnueabi-ld: drivers/gpu/drm/rcar-du/rcar_du_crtc.o: in function `rcar_du_crtc_atomic_disable':
> > > > > rcar_du_crtc.c:(.text+0xd7c): undefined reference to `rcar_lvds_clk_disable'
> > > > > arm-linux-gnueabi-ld: drivers/gpu/drm/rcar-du/rcar_du_drv.o: in function `rcar_du_init':
> > > > > rcar_du_drv.c:(.init.text+0x4): undefined reference to `rcar_du_of_init'
> > > > > arm-linux-gnueabi-ld: drivers/gpu/drm/rcar-du/rcar_du_encoder.o: in function `rcar_du_encoder_init':
> > > > >
> > > > > Remove the 'imply', and instead use a silent symbol that defaults
> > > > > to the correct setting.
> > > >
> > > > This will result in the CMM always being selected when DU is, increasing
> > > > the kernel size even for devices that don't need it. I believe we need a
> > > > better construct in Kconfig to fix this.
> > >
> > > I had expected this to have the same meaning that we had before the
> > > Kconfig change: whenever the dependencies are available, turn it on,
> > > otherwise leave it disabled.
> > >
> > > Can you describe what behavior you actually want instead?
> >
> > Doesn't "imply" mean it gets selected by default but can be manually
> > disabled ?
>
> That may be what it means now (I still don't understand how it's defined
> as of v5.7-rc1), but traditionally it was more like a 'select if all
> dependencies are met'.

That's still what it is supposed to mean right now ;-)
Except that now it should correctly handle the modular case, too.

> > > > > --- a/drivers/gpu/drm/rcar-du/Kconfig
> > > > > +++ b/drivers/gpu/drm/rcar-du/Kconfig
> > > > > @@ -4,7 +4,6 @@ config DRM_RCAR_DU
> > > > >       depends on DRM && OF
> > > > >       depends on ARM || ARM64
> > > > >       depends on ARCH_RENESAS || COMPILE_TEST
> > > > > -     imply DRM_RCAR_CMM
> > > > >       imply DRM_RCAR_LVDS
> > > > >       select DRM_KMS_HELPER
> > > > >       select DRM_KMS_CMA_HELPER
> > > > > @@ -15,9 +14,8 @@ config DRM_RCAR_DU
> > > > >         If M is selected the module will be called rcar-du-drm.
> > > > >
> > > > >  config DRM_RCAR_CMM
> > > > > -     tristate "R-Car DU Color Management Module (CMM) Support"
> > > > > +     def_tristate DRM_RCAR_DU
> > > > >       depends on DRM && OF
> > > > > -     depends on DRM_RCAR_DU
> > > > >       help
> > >
> > > It would be easy enough to make this a visible 'bool' symbol and
> > > build it into the rcu-du-drm.ko module itself. Would that help you?
> >
> > That could indeed simplify a few things. I wonder if it could introduce
> > a few small issues though (but likely nothing we can't fix). The two
> > that come to mind are the fact that the module would have two
> > MODULE_DESCRIPTION and MODULE_LICENSE entries (I have no idea if that
> > could cause breakages), and that it could make module unloading more
> > difficult as the CMM being used by the DU would increase the refcount on
> > the module. I think the latter could be worked around by manually
> > unbinding the DU device through sysfs before unloading the module (and I
> > can't say for sure that unloading the DU module is not broken today
> > *innocent and naive look* :-)).
>
> In that case, a Makefile trick could also work, doing
>
> ifdef CONFIG_DRM_RCAR_CMM
> obj-$(CONFIG_DRM_RCAR_DU) += rcar-cmm.o
> endif
>
> Thereby making the cmm module have the same state (y or m) as
> the du module whenever the option is enabled.

What about dropping the "imply DRM_RCAR_CMM", but defaulting to
enable CMM if DU is enabled?

    config DRM_RCAR_CMM
            tristate "R-Car DU Color Management Module (CMM) Support"
            depends on DRM_RCAR_DU && OF
            default DRM_RCAR_DU

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds