On Wed, 2 Sep 2020 13:15:49 +0200
Daniel Vetter <daniel.vetter@ffwll.ch> wrote:

> On Wed, Sep 2, 2020 at 1:02 PM Pekka Paalanen <ppaalanen@gmail.com> wrote:
> >
> > On Wed,  2 Sep 2020 12:24:40 +0200
> > Daniel Vetter <daniel.vetter@ffwll.ch> wrote:
> >  
> > > Even for legacy userspace, since otherwise GETFB2 is broken and if you
> > > switch between modifier-less and modifier-aware compositors, smooth
> > > transitions break.
> > >
> > > Also it's just best practice to make sure modifiers are invariant for
> > > a given drm_fb, and that a modifier-aware kms drivers only has one
> > > place to store them, ignoring any old implicit bo flags or whatever
> > > else might float around.
> > >
> > > Motivated by some irc discussion with Bas about amdgpu modifier
> > > support.
> > >
> > > Fixes: 455e00f1412f ("drm: Add getfb2 ioctl")
> > > Cc: Daniel Stone <daniels@collabora.com>
> > > Cc: Juston Li <juston.li@intel.com>
> > > Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> > > Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
> > > Cc: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
> > > Cc: Marek Olšák <maraeo@gmail.com>
> > > Cc: "Wentland, Harry" <harry.wentland@amd.com>
> > > Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> > > ---
> > >  include/drm/drm_mode_config.h | 13 +++++++++++++
> > >  1 file changed, 13 insertions(+)
> > >
> > > diff --git a/include/drm/drm_mode_config.h b/include/drm/drm_mode_config.h
> > > index a18f73eb3cf6..5ffbb4ed5b35 100644
> > > --- a/include/drm/drm_mode_config.h
> > > +++ b/include/drm/drm_mode_config.h
> > > @@ -58,6 +58,12 @@ struct drm_mode_config_funcs {
> > >        * actual modifier used if the request doesn't have it specified,
> > >        * ie. when (@mode_cmd->flags & DRM_MODE_FB_MODIFIERS) == 0.
> > >        *
> > > +      * IMPORTANT: These implied modifiers for legacy userspace must be
> > > +      * stored in struct &drm_framebuffer, including all relevant metadata
> > > +      * like &drm_framebuffer.pitches and &drm_framebuffer.offsets if the
> > > +      * modifier enables additional planes beyond the fourcc pixel format
> > > +      * code. This is required by the GETFB2 ioctl.
> > > +      *
> > >        * If the parameters are deemed valid and the backing storage objects in
> > >        * the underlying memory manager all exist, then the driver allocates
> > >        * a new &drm_framebuffer structure, subclassed to contain
> > > @@ -915,6 +921,13 @@ struct drm_mode_config {
> > >        * @allow_fb_modifiers:
> > >        *
> > >        * Whether the driver supports fb modifiers in the ADDFB2.1 ioctl call.
> > > +      *
> > > +      * IMPORTANT:
> > > +      *
> > > +      * If this is set the driver must fill out the full implicit modifier
> > > +      * information in their &drm_mode_config_funcs.fb_create hook for legacy
> > > +      * userspace which does not set modifiers. Otherwise the GETFB2 ioctl is
> > > +      * broken for modifier aware userspace.
> > >        */
> > >       bool allow_fb_modifiers;
> > >  
> >
> > Hi,
> >
> > are there any drivers that would infer this information at
> > modeset/pageflip/atomic ioctl time instead of AddFB/AddFB2 time?  
> 
> Currently no, the only driver that infers anything for legacy is i915.
> Proposed amdgpu modifier patches don't work like that, but I think Bas
> is working on adding the modifier inference at addfb time for legacy
> userspace.
> 
> > Userspace may be creating the FB once per buffer and keep re-using
> > that over several render/display cycles. If a driver was changing the
> > "effective modifiers" dynamically, userspace could break.  
> 
> Yeah this is why I want to lock this all down, since effective
> modifiers that change for a drm_fb object which is supposed to have
> all invariant metadata just isn't great uapi.

Sounds good to me!

Acked-by: Pekka Paalanen <pekka.paalanen@collabora.com>


Thanks,
pq