On Mon, 7 Sep 2020 10:41:37 +0200
Daniel Vetter <daniel@ffwll.ch> wrote:

> On Mon, Sep 07, 2020 at 08:37:31AM +0000, Simon Ser wrote:
> > On Monday, September 7, 2020 10:31 AM, Daniel Vetter <daniel@ffwll.ch> wrote:
> >   
> > > On Wed, Sep 02, 2020 at 02:59:49PM +0000, Simon Ser wrote:
> > >  
> > > > On Wednesday, September 2, 2020 4:29 PM, Daniel Vetter daniel.vetter@ffwll.ch wrote:
> > > >  
> > > > > On Wed, Sep 2, 2020 at 2:49 PM Simon Ser contact@emersion.fr wrote:
> > > > >  
> > > > > > On Wednesday, September 2, 2020 2:44 PM, Daniel Vetter daniel.vetter@ffwll.ch wrote:
> > > > > >  
> > > > > > > > I suppose something similar happens in user-space: gbm_bo_create
> > > > > > > > without modifiers needs to properly set the implicit modifier, ie.
> > > > > > > > gbm_bo_get_modifier needs to return the effective modifier. Is this
> > > > > > > > something already documented?  
> > > > > > >
> > > > > > > I don't think that happens, but it has come up in discussions. It's
> > > > > > > kinda different scenario though: getfb2 is for cross-compositor stuff,
> > > > > > > enabling smooth transitions at boot-up and when switching. So you have
> > > > > > > a legit reason for mixing modifier-aware userspace with
> > > > > > > non-modifier-aware userspace. And the modifier-aware userspace would
> > > > > > > like that everything works with modifiers consistently, including
> > > > > > > getfb2. But gbm is just within a single process, and that should
> > > > > > > either run all with modifiers, or not at all, since these worlds just
> > > > > > > dont mix well. Hence I'm not seeing much use for that, -modesetting
> > > > > > > being a confused mess nonwithstanding :-)  
> > > > > >
> > > > > > Well… There's also the case where some legacy Wayland client talks to a
> > > > > > modifier-aware compositor. gbm_bo_import would be called without a
> > > > > > modifier, but the compositor expects gbm_bo_get_modifier to work.
> > > > > > Also, wlroots will call gbm_bo_create without a modifier to only let
> > > > > > the driver pick "safe" modifiers in case passing the full list of
> > > > > > modifiers results in a black screen. Later on wlroots will call
> > > > > > gbm_bo_get_modifier to figure out what modifier the driver picked.  
> > > > >
> > > > > gbm_bo_import is a different thing from gbm_bo_create. Former I agree
> > > > > should figure out the right modifiers (and I think it does that, at
> > > > > least on intel mesa). For gbm_bo_create I'm not sure we should/need to
> > > > > require that.  
> > > >
> > > > I guess the compositor will just forward the value returned by
> > > > gbm_bo_get_modifier in any case, so returning INVALID would be fine
> > > > too (to mean "implicit modifier").
> > > > In both the create and import cases, other metadata like pitches and
> > > > offsets should be correctly set I think?  
> > >
> > > Well if you have a modifier format underneath, the non-modifiered offsets
> > > and pitches might be pure fiction. Also, they might not be sufficient, if
> > > the modifier adds more planes.  
> > 
> > In this case (gbm_bo_create without modifiers), we're discussing
> > whether we require gbm_bo_get_modifier to return a valid modifier, or
> > if INVALID is fine.  
> 
> Hm then I missed the use-case for a gbm_bo_create without modifiers, where
> afterwards userspace wants the modifiers. That sounds like a bug (and yes
> -modesetting is buggy that way).

I'm guessing that use case might be related to
https://gitlab.freedesktop.org/wayland/weston/-/issues/429

The weston issue is about
gbm_surface_create/gbm_surface_create_with_modifiers, but that's not
too different from gbm_bo_create/gbm_bo_create_with_modifiers?

Weston happens to have this code:
https://gitlab.freedesktop.org/wayland/weston/-/blob/9.0.0/libweston/backend-drm/drm-gbm.c#L209-230
and then it unconditionally calls gbm_bo_get_modifier(). However,
DRM_FORMAT_MOD_INVALID is handled specially:
https://gitlab.freedesktop.org/wayland/weston/-/blob/9.0.0/libweston/backend-drm/fb.c#L80-97


Thanks,
pq