If any of the engines hang, you can end up with a deadlock if you schedule dependencies across rings since one of the engines is hung. If you try a newer kernel, GPU reset is enabled and you should be able to recover after a hang. At that point, you can either restart your applications or use a relevant robustness API to properly handle a reset.
Alex ________________________________ From: amd-gfx <[email protected]> on behalf of CCXIAOP <[email protected]> Sent: Wednesday, June 19, 2019 5:28 AM To: amd-gfx Subject: amdgpu vce crash We are using wx5100 for rendering and encoding operations, but sometimes we encounter vce timeout and crash. Is vce not a independent module in gpu?Why does it affect the rendering module? We hope that vce will not affect the rendering module when crashing. Can I prevent the use of the rendering module from being affected? linux kernel :4.19.34 mesa: 18.3.5 llvm: 7.0 firmware:18.50 As logs： 2019-06-15T15:33:32.133842+08:00|err|kernel[-]|[315248.172603] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring vce0 timeout, signaled seq=1173730, emitted seq=1173732 2019-06-15T15:33:32.133939+08:00|info|kernel[-]|[315248.172607] [drm] GPU recovery disabled.
_______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
