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

[Daniel Vetter]

On Wed, Apr 15, 2020 at 11:22 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Wed, Apr 15, 2020 at 11:12 PM Laurent Pinchart
> <laurent.pinchart@xxxxxxxxxxxxxxxx> wrote:
> > On Wed, Apr 15, 2020 at 09:07:14PM +0200, Arnd Bergmann wrote:
> > > On Wed, Apr 15, 2020 at 5:18 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > > > On Wed, Apr 15, 2020 at 4:13 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > > > 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:
> > > > > > > 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.
> > > >
> > > > Then there is a bug. If I run 'make menuconfig' now on a mainline kernel
> > > > and enable CONFIG_DRM_RCAR_DU, I can set
> > > > DRM_RCAR_CMM and DRM_RCAR_LVDS to 'y', 'n' or 'm' regardless
> > > > of whether CONFIG_DRM_RCAR_DU is 'm' or 'y'. The 'implies'
> > > > statement seems to be ignored entirely, except as reverse 'default'
> > > > setting.
> > >
> > > Here is another version that should do what we want and is only
> > > half-ugly. I can send that as a proper patch if it passes my testing
> > > and nobody hates it too much.
> >
> > This may be a stupid question, but doesn't this really call for fixing
> > Kconfig ? This seems to be such a common pattern that requiring
> > constructs similar to the ones below will be a never-ending chase of
> > offenders.
>
> Maybe, I suppose the hardest part here would be to come up with
> an appropriate name for the keyword ;-)
>
> Any suggestions?
>
> This specific issue is fairly rare though, in most cases the dependencies
> are in the right order so a Kconfig symbol 'depends on' a second one
> when the corresponding loadable module uses symbols from that second
> module. The problem here is that the two are mixed up.
>
> The much more common problem is the one where one needs to
> wrong
>
> config FOO
>        depends on BAR || !BAR
>
> To ensure the dependency is either met or BAR is disabled, but
> not FOO=y with BAR=m. If you have any suggestions for a keyword
> for that thing, we can clean up hundreds of such instances.

Some ideas:

config FOO
    can use  BAR
    maybe BAR
    optional BAR

We should probably double-check that this is only ever used for when
both FOO and BAR are tri-state, since without that it doesn't make
much sense.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch