12:53MrCooper: mareko: mesa!25043 caused GPU hangs with rusticl on Navi 14
12:57mceier: there're already 2 issues reported with exact commit causing that problem: 9915 (mine) and 9916
12:59MrCooper: heh, late to the party
15:00emersion: MrCooper: did you find an answer to your question?
15:00emersion: the plane one
15:00emersion: it seems your patch just works around it
15:01MrCooper: not sure what you mean by that, are you referring to https://patchwork.freedesktop.org/patch/560465/?series=124527&rev=1 ?
15:02emersion: yes
15:03MrCooper: it checks for any relevant changes, so it doesn't need to avoid drm_atomic_get_plane_state
15:06emersion: yeah, but this issue of pulling unrelated planes is something i've seen elsewhere too
15:06emersion: in i915 for instance
15:06emersion: it was about CRTCs actually
16:53lileo: emerson: MrCooper: a thought: what if, instead of unconditionally rejecting a cursor plane on top of a differently-scaled plane, we only reject if they're overlapping? https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c#L9900
16:55MrCooper: lileo: one issue with that is it may then be difficult to disable a plane, if it would result in the cursor plane overlapping another plane with different scaling
16:57lileo: right, the atomic check failure there wouldn't be obvious at all
16:57lileo: at the same time, we wouldn't be able to do video overlay planes at all, they're almost always scaled
16:57lileo: (with a cursor enabled, that is)
16:59lileo: well, what if we checked all underlying planes for intersection?
17:00lileo: and reject if any intersects & has different scale?