[navi21] Sleep and resume breaks switching to tty and causes a GPU hang by plasmashell in KDE

I encountered a session breaking amdgpu hang when switching to tty and back after resume:

1. Put computer to sleep (running KDE Wayland session).
2. Resume (session still works OK so far).
3. Switch to tty2 - it's not working properly (same for tty3, etc.).
4. Switch back to tty1 which is supposed to be Wayland session - it's not working either (none of the virtual consoles are).

Remote ssh log-in shows this in dmesg:

[  110.419058] [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
[  110.684801] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx_0.0.0 timeout, signaled seq=6145, emitted seq=6147
[  110.684987] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process plasmashell pid 1599 thread plasmashel:cs0 pid 1723
[  110.685145] amdgpu 0000:0f:00.0: amdgpu: GPU reset begin!
[  111.120500] amdgpu 0000:0f:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring kiq_2.1.0 test failed (-110)
[  111.120575] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* KGQ disable failed
[  111.400258] amdgpu 0000:0f:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring kiq_2.1.0 test failed (-110)
[  111.400320] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* KCQ disable failed
[  111.680112] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* failed to halt cp gfx
[  111.710758] [drm] free PSP TMR buffer
[  111.758498] CPU: 10 PID: 2783 Comm: kworker/u64:80 Not tainted 5.18.15 #1
[  111.758499] Hardware name: To Be Filled By O.E.M. X570 Taichi/X570 Taichi, BIOS P4.80 02/16/2022
[  111.758500] Workqueue: amdgpu-reset-dev drm_sched_job_timedout [gpu_sched]
[  111.758504] Call Trace:
[  111.758506]  <TASK>
[  111.758508]  dump_stack_lvl+0x45/0x5e
[  111.758511]  amdgpu_do_asic_reset+0x28/0x44e [amdgpu]
[  111.758605]  amdgpu_device_gpu_recover_imp.cold+0x613/0x8e7 [amdgpu]
[  111.758687]  amdgpu_job_timedout+0x153/0x190 [amdgpu]
[  111.758766]  drm_sched_job_timedout+0x76/0x110 [gpu_sched]
[  111.758769]  process_one_work+0x1e5/0x3b0
[  111.758771]  worker_thread+0x50/0x3a0
[  111.758772]  ? rescuer_thread+0x390/0x390
[  111.758773]  kthread+0xe8/0x110
[  111.758774]  ? kthread_complete_and_exit+0x20/0x20
[  111.758776]  ret_from_fork+0x22/0x30
[  111.758778]  </TASK>
[  111.758779] amdgpu 0000:0f:00.0: amdgpu: MODE1 reset
[  111.758781] amdgpu 0000:0f:00.0: amdgpu: GPU mode1 reset
[  111.758856] amdgpu 0000:0f:00.0: amdgpu: GPU smu mode1 reset
[  112.268204] amdgpu 0000:0f:00.0: amdgpu: GPU reset succeeded, trying to resume
[  112.268539] [drm] PCIE GART of 512M enabled (table at 0x0000008000E10000).
[  112.268567] [drm] VRAM is lost due to GPU reset!
...

Configuration:

• GPU: Sapphire Pulse RX 6800 XT
• Motherboard: Asrock X570 Taichi (UEFI 4.80)
• Linux/amdgpu: 5.18.15
• Mesa: 22.2.0-rc1
• amdgpu firmware: latest (release 22.20).
• KDE Plasma / kwin: 5.25.4