AMDGPU.gpu_recovery does recover desktop to an unusable state

Submitted by Dávid Bíró @dbiro97

Assigned to Default DRI bug account

Link to original bug (#112297)

Description

I have got Radeon RX 5700 XT which has unstable driver. So that amdgpu.gpu_recovery=1 flag would be helpful if even that could work well.
Unfortunately it doesn't.

My kernel's latest commit ID: 53dbc27ad5a93932ff1892a8e4ef266827d74a0f ( fix AVFS handling with custom powerplay table) On branch drm-next-5.5
Display server: X11
WM: BSPWM
DM: Slim

Journalctl output:
https://pastebin.com/if4PYda7

The error is the "average" ring sdma0 timeout. After reset, the actual application that caused the crash, stays on the screen, The screen is flickering, vibrating, I'm frequently able to see some green rectangle at the top of the screen with full width. But I can able to move mouse. Also I cannot do anything else, because then BSPWM can switch workspaces (I can see it by the type of the mouse cursor), but the crashed application stays on the screen anyway, so I have to restart.

At least, I can save my work if I remember which workspace contained my work.