https://bugs.freedesktop.org/show_bug.cgi?id=94158

Bug ID: 94158 Summary: [amdgpu][tonga] GPU randomly gets stuck at 100% load and high clocks Product: DRI Version: DRI git Hardware: x86-64 (AMD64) OS: Linux (All) Status: NEW Severity: normal Priority: medium Component: DRM/AMDgpu Assignee: dri-devel@lists.freedesktop.org Reporter: csaba.halasz@gmail.com

I have noticed that my GPU fans start running at high speed after some time. Investigating, I see:

# cat /sys/kernel/debug/dri/0/amdgpu_pm_info

[ mclk ]: 1425 MHz [ sclk ]: 990 MHz [GPU load]: 100%

Of course I am not running any graphics intensive programs and radeontop confirms <1% load. Even shutting down Xorg altogether doesn't help. It may be related to dpms going to "off" state, the last time it happened while the monitors were off, or possibly just as they were switching off. I will try to collect additional data points. Nothing suspicious shows in logs. Powerplay seems to work fine until the issue occurs. Afterwards, I can manually lower clocks, but load stays 100%. Side note: dpms transitions seem to reset the manual power_dpm_force_performance_level back to auto and hence high clock.

System: r9 380, amdgpu git agd5f/drm-fixes-4.5 cc1de6e8 merged with agd5f/drm-next-4.6-wip fa9e6991, llvm git ae316aa0, mesa git fe14110f, Xorg 1.18