https://bugs.freedesktop.org/show_bug.cgi?id=108118
Bug ID: 108118 Summary: AMDGPU sometimes hangs forever when running graphical applications Product: DRI Version: unspecified Hardware: x86-64 (AMD64) OS: Linux (All) Status: NEW Severity: major Priority: medium Component: DRM/AMDgpu Assignee: dri-devel@lists.freedesktop.org Reporter: duooratar@gmail.com
Sometimes when running a graphical application the display will freeze up but system sound will continue. The machine is still functioning and is accessible over ssh. Using ssh the following related messages are found in the output of dmesg:
[drm:admgpu_job_timeout [amdgpu]] *ERROR* ring gfx timeout, signalled seq=1256221, emitted seq=1256223 amdgpu 0000:10:00.0 GPU reset begin! [drm:amdgpu_dm_atomic_check [amdgpu]] *ERROR* [CRTC:43:crtc-0] hw_done or flip_done timed out
This is the output with amdgpu.gpu_recovery=1 set in the kernel launch parameters. Without that the output is the same except the last two messages are replaced with a message about GPU recovery being disabled. Any attempt to access /sys/kernel/debug/dri/0/amdgpu_gpu_recover in either state hangs forever. Magic SysRq keys still work and processes can be killed over SSH but killing the game/Xorg/etc. will not cause the display to start working again, a reset is required.
This has been observed with both Xorg and the KDE's Wayland compositor.
This has only been observed with Vulkan applications (native Dota 2's Vulkan mode and DXVK backed Wine games) but hasn't been confirmed to not occur with others.
This was observed with the libvulkan_radeon.so Mesa Vulkan driver. I couldn't confirm the behavior with AMDVLK because graphical applications failed to launch with it installed. I don't believe the GPU in question is faulty, I've used it for long periods of time on Windows and it's rock stable.
Observed on kernels 4.18.11 and 4.19.0-rc6. Searching around the Internet suggests this may have started with 4.18.0 but I haven't confirmed that yet.
Hardware: CPU: AMD R7 1800X GPU: AMD RX Vega 64