https://bugs.freedesktop.org/show_bug.cgi?id=107045

            Bug ID: 107045
           Summary: [4.18rc2] RX470 dGPU on hybrid laptop freezes screen
                    after use
           Product: DRI
           Version: DRI git
          Hardware: x86-64 (AMD64)
                OS: Linux (All)
            Status: NEW
          Severity: critical
          Priority: medium
         Component: DRM/AMDgpu
          Assignee: dri-devel@lists.freedesktop.org
          Reporter: taijian@posteo.de

Created attachment 140343
  --> https://bugs.freedesktop.org/attachment.cgi?id=140343&action=edit
full dmesg output via 'journalctl -kb -1'

With kernel 4.18rc dpm on my hybrid laptop is finally working (thanks again to
everyone who helped out getting it this far - see bug #104064, bug #106597 and
bug #105760). However, it is still not actually usable, because any activation
of the dGPU leads to a system freeze shortly thereafter. It does not really
matter HOW the dGPU is triggered - be it running a graphical application via
DRI_PRIME=1 or just querying #lshw. The dGPU activates and then tries to power
down again, but that seems to be problematic as the (graphical) system will
invariable freeze and become unresponsive.

The best I have been able to reconstruct from the journal is that whenever the
dGPU is called, it will first do this:

  [drm] PCIE GART of 256M enabled (table at 0x000000F400000000).

followed by a bunch of this: 

  amdgpu: [powerplay] 
           failed to send message 62 ret is 0 
  amdgpu: [powerplay] 
           last message was failed ret is 0
  amdgpu: [powerplay] 
           failed to send message 18f ret is 0 

and then finally: 

  [drm] UVD and UVD ENC initialized successfully.
  [drm] VCE initialized successfully.
  [drm] Cannot find any crtc or sizes
  amdgpu 0000:01:00.0: GPU pci config reset

Sometimes, this will work more than once, sometimes the first invocation after
boot will crash. This seems to be somewhat random. However, the crash will
always come in the "[powerplay] ...failed to send message 18f ret is 0" phase.

Any ideas how to tackle that or extract some more usable error message from my
system?

-- 
You are receiving this mail because:
You are the assignee for the bug.