Thanks for the report, this is a kernel bug.

This was actually a regression caused from:

"drm/amd/display: PIP overlay corruption"

which was fixed by:

"drm/amd/display: Fix negative cursor pos programming"

...and has another fix for overlay specific cases 

"drm/amd/display: Prevent cursor hotspot overflow for RV overlay planes"

It actually looks like these 3 patches are in Linux next-git... but in the
wrong order or not applied correctly, since there are still remnants of the
first commit leftover that were removed by the second.

My guess is this was a bad patch autoselect, but I'll follow up with the
maintainers to try and get this fixed.

FWIW, you shouldn't see this in amd-staging-drm-next.