MangoHud & CoreCtrl issues

This did not grab my attention until now, this is several days/weeks after having received and installed couple minor updates. This is since between 22 April and last week.
Something has regressed, but in what package I don’t know.

Issue: MangoHud started showing an incorrect GPU clock value, when using CoreCtrl to manage GPU clocks.
GPU current clocks are all still correctly shown/reported via sudo cat /sys/kernel/debug/dri/0/amdgpu_pm_info.

Mangohud shows the Min Mhz GPU clock, that is set by CoreCtrl in the GPU profile (auto or manual), when Performance Mode set to Advanced. And when using Performance Mode set to Fixed High, it will shows 1780Mhz and on Low 300Mhz.
Frequency Mhz does not fluctuate under a varying load as you would expect, like a game.

Only when disabling GPU management in the CoreCtrl profile, letting the GPU manage it self, does Manguhod show the correct gpu_core_clock behaviour.

My pacman config was set to keep only 2 version, so I can’t test downgrading Manguhud to 0.6.8.
Last update of CoreCtrl for me was on 5 April and the ManguHud was working as expected back then.

Relevant hardware/software information

  • Manjaro (Testing branch, update 23-04-27)
  • Kernel 6.2.13
  • MangoHud
  • CoreCtrl 1.3.5
  • GPU AMD RX 5600 XT

There are some open bug reports, I personally gave up after using git version and some non-working fixes, and switched to load% | temp | power.

"Using current_gfxclk is the correct way. Like i said in #954 seems to be a bug in the kernel. I haven’t got around to reporting it the amdgpu devs.

current_gfxclk seems to work fine with other GPUs expect RDNA3"

No big deal for me to wait, and hopefully amdgpu gets fixed.

Yep seen the two sort of similar bug reports as well and indeed the issue is not a big deal.

The two reports however do not mentioned the use of the combination of CoreCtrl managing GPU clocks (with needed amdgpu.ppfeaturemask boot parameter), resulted in incorrect clock values being displayed/read by MangoHud.
While correct clock values are being shown by current_gfxclk for my RDNA1 card.

Going by #954 ‘it’s seeming to be a kernel bug’. Is it?
The odd thing is, disabling GPU clock management in CoreCtrl’s on my RDNA1 card then results in correct clocks being displayed properly again by MangoHud in accordance to current_gfxclk.

Made new a separate bug report on their GitHub due to this difference. Interesting to see what they can find out on this.