Re: [PATCH 4/4] drm/vc4: Validate framebuffer pixel format/modifier

[Ville Syrjälä]

On Fri, Mar 09, 2018 at 12:54:43PM -0800, Eric Anholt wrote:
> Ville Syrjala  writes:
> 
> > From: Ville Syrjälä 
> >
> > Only create framebuffers with supported format/modifier combinations by
> > checking that at least one plane supports the requested combination.
> >
> > Using drm_any_plane_has_format() is somewhat suboptimal for vc4 since
> > the planes have (mostly) uniform capabilities. But I was lazy and
> > didn't feel like exporting drm_plane_format_check() and hand rolling
> > anything better. Also I really just wanted to come up with another
> > user for drm_any_plane_has_format() ;)
> 
> I'm not excited about vc4 having error-return behavior that other
> drivers don't have.  Actually, I don't really see why we should be doing
> this check in fb create at all, given that you have to do so again at
> atomic_check time with the specific plane.
> 
> Could you just delete the i915 fb format check code?

I don't want unsupported formats to get anywhere near the driver
code. Makes life much less stressful when you know what can and
can't get in.

Also I see no good reason to allow userspace to create fbs it
can't actually use later on. Much better to reject early and tell
userspace to pick another format. I imagine pre-blobifier userspace
could even use this as a way to probe what's supported by the driver.

-- 
Ville Syrjälä
Intel OTC
___
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Re: [PATCH 4/4] drm/vc4: Validate framebuffer pixel format/modifier

[Eric Anholt]

Ville Syrjala  writes:

> From: Ville Syrjälä 
>
> Only create framebuffers with supported format/modifier combinations by
> checking that at least one plane supports the requested combination.
>
> Using drm_any_plane_has_format() is somewhat suboptimal for vc4 since
> the planes have (mostly) uniform capabilities. But I was lazy and
> didn't feel like exporting drm_plane_format_check() and hand rolling
> anything better. Also I really just wanted to come up with another
> user for drm_any_plane_has_format() ;)

I'm not excited about vc4 having error-return behavior that other
drivers don't have.  Actually, I don't really see why we should be doing
this check in fb create at all, given that you have to do so again at
atomic_check time with the specific plane.

Could you just delete the i915 fb format check code?


signature.asc
Description: PGP signature
___
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel