Happens on linux > 4.16 (also on the amd-staging-4.17-wip) but not on 4.15

Here are the relevant lines from dmesg:

[   33.835186] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835188] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835189] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835193] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835195] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835196] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835200] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835202] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835203] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835207] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835208] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835210] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835214] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835215] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835217] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835220] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835222] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835223] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835227] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835229] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835230] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   33.835234] amdgpu 0000:26:00.0: [gfxhub] VMC page fault (src_id:0 ring:158
vmid:1 pas_id:0)
[   33.835235] amdgpu 0000:26:00.0:   at page 0x0000000100000000 from 27
[   33.835237] amdgpu 0000:26:00.0: VM_L2_PROTECTION_FAULT_STATUS:0x00000000
[   43.998837] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout,
last signaled seq=6, last emitted seq=7
[   43.998848] [drm] No hardware hang detected. Did some blocks stall?