CachyOS / linux-cachyos

Archlinux Kernel based on different schedulers and some other performance improvements.
https://cachyos.org
GNU General Public License v3.0
790 stars 33 forks source link

Processor is running in the highest frequency with schedutil - TT Scheduler #1

Closed alitajelsir closed 2 years ago

alitajelsir commented 2 years ago
ptr1337 commented 2 years ago
  • I tried linux-tt 5.15.3-4 and linux-cacule 5.15.5-2 from AUR in Arch Linux, and in both the processor is running in the highest frequency with schedutil in intel_pstate=passive

    • The same happens with powersave and performance in intel_pstate=active

    • I'm using tlp to set the governor

    • This doesn't happen with ondemand

    • Strangely enough, if I change from schedutil to ondemand and back to schedutil, the scaling will work as expected, but it revert to the highest frequency after reboot.

    • If it helps, this doesn't happen with linux zen 5.15.4.zen1-1

    • Any information I can provide to help?

    • Thanks in advance!

How about if you using as cmdline intel_pstate=hwp ? Could be also a problem form the tt-scheduler.

Check if the cmdline and call back please.

alitajelsir commented 2 years ago
ptr1337 commented 2 years ago

Its a issue from the scheduler when no_hz_full is used i think.

See here:

https://github.com/hamadmarri/TT-CPU-Scheduler/issues/5

@hamadmarri

ptr1337 commented 2 years ago
  • In linux-tt and linux-cacule:

    • With intel_pstate=hwp or intel_pstate=active and powersave or performance the processor is running at the highest frequency without turbo (1.8 GHz in i5-8250U)
    • With intel_pstate=passive highest frequency with turbo (3.4 GHz)
    • In linux-zen:

    • With intel_pstate=passive (400 MHz)

You can also try 'intel_pstate=enable'. If its not getting solved, we need to switch to tickless idle only mode instead of completly tickless.

alitajelsir commented 2 years ago
ptr1337 commented 2 years ago

That’s a known issue at the tt kernel. Maybe it’s gonna fixed soon.

I thought first it’s the no_hz_full and changed this back to the old default

Same with the preempt behavior. But if there are no changes for you, it can be changed back.

Maybe do a issue at the tt scheduler repo.