On Thursday, December 10th, 2020 at 4:56 PM, Daniel Vetter daniel@ffwll.ch wrote:
Huh so crtc are registered forward and planes backward? I guess adding amd people. And yeah sounds like defacto you can't figure out which primary plane goes to which crtc, and we just take whatever goes. Maybe that stricter approach with more guarantees just doesn't work, ship sailed already :-/
Yeah. Even if we fixed the amdgpu driver and added the check, user-space still couldn't have the guarantee that it can associate the n-th primary plane with the n-th CRTC, because it might run with an old kernel.
If we really wanted to allow user-space to discover the internal ->{primary,cursor} pointers, I think we should just expose a new property. That way, the uAPI would be a lot more explicit and a lot less guessing. The cost is that it wouldn't work on older kernels, but with amdgpu user-space can't rely on the implicit rule you've suggested anyways.