[PATCH] drm: don't override possible_crtcs for primary/cursor planes

[Ville Syrjälä]

On Thu, Oct 20, 2016 at 03:06:06PM -0400, Rob Clark wrote:
> It is kind of a pointless restriction.  If userspace does silly things
> like using crtcA's cursor plane on crtcB, and then setcursor on crtcA,
> it will end up with the overlay disabled on crtcB.  But userspace is
> allowed to shoot itself like this.
> 
> Signed-off-by: Rob Clark <robdclark at gmail.com>
> ---
> Note that cursor and primary planes are slightly useless to non-atomic
> userspace, since userspace has no way to know *which* crtc a primary
> or cursor plane belongs to.  So lighting up a 2nd display may or may
> not make your overlay planes go *poof*.
> 
> So basically use of cursor/primary planes plus legacy ioctls is an
> undefined behavior if we go with this patch.
> 
> *Maybe* we want to restrict this to atomic userspace.  (Ie. advertise
> the more limited possible_crtcs for legacy userspace.)  Depends on
> whether we can (slightly retroactively) declare that mixing atomic
> and legacy APIs is undefined behavior.  Or if really needed, add
> DRM_CLIENT_CAP_REALLY_REALLY_ATOMIC_NO_LEGACY_PLS.  (Or just pass in
> value 2 for existing DRM_CLIENT_CAP_ATOMIC)
> 
>  drivers/gpu/drm/drm_crtc.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c
> index c215802..1e6d37f 100644
> --- a/drivers/gpu/drm/drm_crtc.c
> +++ b/drivers/gpu/drm/drm_crtc.c
> @@ -707,9 +707,9 @@ int drm_crtc_init_with_planes(struct drm_device *dev, struct drm_crtc *crtc,
>  	crtc->primary = primary;
>  	crtc->cursor = cursor;
>  	if (primary)
> -		primary->possible_crtcs = 1 << drm_crtc_index(crtc);
> +		WARN_ON(!(primary->possible_crtcs & (1 << drm_crtc_index(crtc))));
>  	if (cursor)
> -		cursor->possible_crtcs = 1 << drm_crtc_index(crtc);
> +		WARN_ON(!(cursor->possible_crtcs & (1 << drm_crtc_index(crtc))));

The slight issue here is that this would require the caller to know the
upcoming crtc index before its even assigned. So might be better just
having the caller populate these however it likes after the crtc_init()?

>  
>  	if (drm_core_check_feature(dev, DRIVER_ATOMIC)) {
>  		drm_object_attach_property(&crtc->base, config->prop_active, 0);
> -- 
> 2.7.4

-- 
Ville Syrjälä
Intel OTC