Bug 218015

Summary: amdgpu powerplay: Spontaneous changes to power_dpm_force_performance_level
Product: Drivers Reporter: Roman Žilka (roman.zilka)
Component: Video(DRI - non Intel)Assignee: drivers_video-dri
Status: RESOLVED ANSWERED    
Severity: high CC: alexdeucher, christian.koenig
Priority: P3    
Hardware: All   
OS: Linux   
URL: https://gitlab.freedesktop.org/drm/amd/-/issues/2931
Kernel Version: 6.1.57 Subsystem:
Regression: No Bisected commit-id:
Attachments: config-6.1.57
lspci -vv

Description Roman Žilka 2023-10-16 07:55:01 UTC
On an RX 6500 XT card I set /sys/class/drm/card*/device/power_dpm_force_performance_level to "profile_standard" before I start any game. That is to avoid overheating (95+ °C) which occurs quickly when the perf level is left at the default of "auto" (it's some sort of OC card) and impacts the GPU as well as the nearby mobo components (I get a thermal warning in kernel log about the onboard audio chip). The perf level setting spontaneously reverts back to "auto" at random, however, and I'm forced to run a script which polls power_dpm_force_performance_level and corrects it when necessary. This happens 0-10 times per hour while playing.

MSI Radeon RX 6500 XT MECH 2X 4G OC. Kernel 6.1.57, x86_64, Gentoo, gcc 13.2.1 20230826, X11, xf86-video-amdgpu-23.0.0, linux-firmware-20230919.
Comment 1 Roman Žilka 2023-10-16 07:56:11 UTC
Created attachment 305225 [details]
config-6.1.57
Comment 2 Roman Žilka 2023-10-16 07:57:22 UTC
Created attachment 305226 [details]
lspci -vv
Comment 3 Artem S. Tashkinov 2023-10-16 08:34:44 UTC
Please report here instead:

https://gitlab.freedesktop.org/drm/amd/-/issues

6.1.x is really not the best kernel to use with this card.

You should try something fresher, i.e. 6.5.7.
Comment 4 Alex Deucher 2023-10-16 13:41:24 UTC
UMDs (e.g., mesa) can change the power profile as well.  E.g., if they want to capture an SQ trace, etc.  It's likely that is the cause.
Comment 5 Roman Žilka 2023-10-22 07:41:18 UTC
Thanks, I passed the info on.
https://gitlab.freedesktop.org/drm/amd/-/issues/2931