Hi,
AddFB is single-planar, with no offset and only depth/bpp rather than
an explicit format, so we now have AddFB2 which can carry multiple
planes, an explicit format and also a modifier. At the time we never
actually did GetFB2 to go with GetFB.
This submission rights that historical wrong, which allows Xorg
-background none to continue to work in the face of exotic buffers.
I've written patches to Xorg to use this as UABI verification, and
I'll post a link to that very shortly.
This is also available in git form, in the wip/2018-03/getfb2 branches of:
https://gitlab.collabora.com/daniels/linux
https://gitlab.collabora.com/daniels/libdrm
https://gitlab.collabora.com/daniels/igt-gpu-tools
https://gitlab.collabora.com/daniels/xserver
Note that the kernel tree is based on top of drm-tip, and has a bunch
of patches to convert drivers to placing objects (rather than creating
handles). At the moment this is just missing amdgpu and nouveau to
complete the set of drivers, and also any testing whatsoever; I'm
sending a patch to convert Intel as that's what I've actually tested
on.
The conversion is necessary since GetFB2 ensures that it does _not_
return duplicate handles if multiple planes of a single buffer share
the same BO. This is impossible if we just use create_handle from the
driver, so instead we just rely on drivers placing the GEM object
pointers in the drm_framebuffer directly. This has the added advantage
of cleaning up a ton of unnecessary code in drivers.
Cheers,
Daniel
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel
