Re: [PATCH] drm/tegra: Fix Kconfig dependencies

[Guenter Roeck <linux-0h96xk9xTtrk1uMJSBkQmQ@public.gmane.org>]

On Wed, Jan 08, 2014 at 03:01:39PM +0100, Thierry Reding wrote:
> On Wed, Jan 08, 2014 at 05:48:44AM -0800, Guenter Roeck wrote:
> > On 01/08/2014 05:29 AM, Thierry Reding wrote:
> > >On Mon, Jan 06, 2014 at 08:20:55AM -0800, Guenter Roeck wrote:
> > >>On Mon, Jan 06, 2014 at 04:39:47PM +0100, Thierry Reding wrote:
> > >>>On Sun, Jan 05, 2014 at 10:22:51AM -0800, Guenter Roeck wrote:
> > >>>>arm:allmodconfig fails to build with several undefined drm and host1x symbols.
> > >>>>
> > >>>>drivers/gpu/drm/tegra/bus.c:52: undefined reference to `drm_dev_alloc'
> > >>>>drivers/gpu/drm/tegra/bus.c:56: undefined reference to `drm_dev_register'
> > >>>>drivers/gpu/drm/tegra/bus.c:67: undefined reference to `drm_dev_free'
> > >>>>drivers/gpu/drm/tegra/bus.c:75: undefined reference to `drm_put_dev'
> > >>>>drivers/gpu/drm/tegra/drm.c:445: undefined reference to `host1x_syncpt_get_base'
> > >>>>drivers/gpu/drm/tegra/drm.c:449: undefined reference to `host1x_syncpt_base_id'
> > >>>>
> > >>>>and so on.
> > >>>>
> > >>>>This is caused by DRM=m but DRM_TEGRA=y. DRM_TEGRA is bool while DRM is
> > >>>>tristate. DRM_TEGRA can not be set to tristate because it depends on unexported
> > >>>>host1x symbols (and possibly others). Fix by updating DRM_TEGRA dependency.
> > >>>>
> > >>>>Also, the DRM_TEGRA help text states that it can be built as module.
> > >>>>Remove that text since it is not (or no longer) correct.
> > >>>>
> > >>>>Signed-off-by: Guenter Roeck <linux-0h96xk9xTtrk1uMJSBkQmQ@public.gmane.org>
> > >>>>---
> > >>>>  drivers/gpu/drm/tegra/Kconfig |    5 +----
> > >>>>  1 file changed, 1 insertion(+), 4 deletions(-)
> > >>>
> > >>>Hi Guenter,
> > >>>
> > >>>I think this should be fixed in latest next, where it is now possible to
> > >>>build the Tegra DRM driver as a module. All needed host1x symbols are
> > >>>exported as well.
> > >>>
> > >>>One thing that strikes me as odd, though, is that I've always thought
> > >>>Kconfig would warn if a symbol selected as =y depended on a symbol
> > >>>selected as =m, precisely because of what you describe above. menuconfig
> > >>>complains about it pretty loudly as well.
> > >>>
> > >>>Oh, but I see that if I change DRM_TEGRA back to be bool, then Kconfig
> > >>>doesn't complain anymore, even if DRM_TEGRA=y and DRM=m. Should that
> > >>>perhaps be considered a bug?
> > >>>
> > >>No idea. All I know is that it causes a build failure ;-).
> > >>
> > >>Anyway, are there plans to push the patches changing DRM_TEGRA to tristate
> > >>into 3.13 ? This is one of the reasons for arm:allmodconfig to fail. If there
> > >>are no plans to push the patches into 3.13, I think it would make sense
> > >>to send my patch to Linus and apply the tristate changes on top of it.
> > >
> > >The patches to change DRM_TEGRA to tristate are part of what is queued
> > >for 3.14. While I agree that having allmodconfig working is a worthwhile
> > >goal in itself for build testing, I'm not sure it warrants a patch this
> > >late in the release cycle that will cause conflicts during the merge
> > >window.
> > >
> > >It also seems like there are other issues that prevent allmodconfig to
> > >build successfully on ARM not all of them being as easy to fix as this
> > >one.
> > >
> > 
> > All other problems are being addressed.
> 
> Okay, if this is indeed the only remaining issue then I think it makes
> sense indeed to go through the trouble of fixing it right away rather
> than waiting for 3.14.
> 
> > If arm:allmodconfig is not pursued as buildable, trying to build it is worthless,
> > and I'll drop it from my list of test builds for -stable. Thanks for letting me know.
> 
> Right, I hadn't considered -stable builds and this is an issue that's
> new in 3.13, isn't it? In that case, and since this is the only
> remaining issue, the above patch:
> 
> Acked-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> 
> Dave, since this is a single patch for 3.13 it might be easier for you
> to pick it up directly rather than have me send out a pull request.
> Either way is fine with me, though.
> 

For now I have dropped arm:allmodconfig from my list of build tests and replaced
it with a number of additional individual target builds. As I had this or
similar discussions ever since I started testing -stable builds, my conclusion
is that the arm community isn't much interested in keeping arm:allmodconfig
buildable. If that ever changes, I'll be happy to add it back in.

Guenter