OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
power-suspend low very-common DISPUTE

Fix 'only powersave and performance available' and governors that reset at boot

archomarchyendeavouroscachyosmanjarolaptopdesktopamdintellimine
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

The diagnosis is right and the mechanics around it are wrong in four places. This workstation reproduces the headline symptom exactly and is real evidence, because it is a desktop with an Intel i9-9900K: /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver reads intel_pstate, scaling_available_governors reads performance powersave, scaling_governor reads powersave, energy_performance_preference reads performance, energy_performance_available_preferences reads default performance balance_performance balance_power power, and /sys/devices/system/cpu/intel_pstate/status reads active. I fetched https://www.kernel.org/doc/html/latest/admin-guide/pm/intel_pstate.html and https://www.kernel.org/doc/html/latest/admin-guide/pm/amd-pstate.html rather than recalling them, and the record's explanation of active versus passive mode matches the kernel's own wording closely, including that the two pseudo-governors share names with the generic ones but do not behave like them and that nothing is throttled. The first defect is the one that decides whether the fix survives a reboot at all, and the record does not mention it. power-profiles-daemon 0.30-1 is installed, enabled and active here, and Omarchy ships it deliberately: it is listed at /usr/share/omarchy/install/omarchy-base.packages:104, enabled at /usr/share/omarchy/install/config/enable-services.sh:13, exposed in the Super+Space menu through omarchy-powerprofiles-list and omarchy-powerprofiles-set, and restored at session start by hl.exec_cmd("omarchy-powerprofiles-init") in /usr/share/omarchy/default/hypr/autostart.lua:8. I read the daemon's own source at https://gitlab.freedesktop.org/upower/power-profiles-daemon/-/raw/main/src/ppd-driver-intel-pstate.c and https://gitlab.freedesktop.org/upower/power-profiles-daemon/-/raw/main/src/ppd-driver-amd-pstate.c: both write scaling_governor and then energy_performance_preference for every policy on every profile change, with the intel file carrying the comment "Force a scaling_governor where the preference can be written" and mapping the profiles to power, balance_power or balance_performance, and performance. That matches what this machine shows, profile performance with EPP performance. So the record's tmpfiles.d persistence is applied at boot and then overwritten, and the supported Omarchy route is to set the profile. The second defect is internal: the record tells the reader to test balance_performance and then writes balance_power into the tmpfiles file. The third is the Limine edit. The drop-in path and the KERNEL_CMDLINE[default]+= syntax are correct and match /etc/limine-entry-tool.d/omarchy-defaults.conf on this machine, and sudo limine-mkinitcpio is the right rebuild command, but the extra sudo tee -a /etc/default/limine < /etc/limine-entry-tool.d/pstate.conf is wrong twice over: /usr/lib/limine/limine-common-functions lines 99 to 129 load /usr/share/limine-entry-tool.d/*.conf, then /etc/limine-entry-tool.conf, then /etc/limine-entry-tool.d/*.conf, then /etc/default/limine last at highest priority, so with += in both files the parameter is appended to the cmdline twice, and /etc/default/limine is owned by no package (pacman -Qo confirms) so it is the wrong place to put a new setting. The fourth is that the record never says the EPP write is refused while the pseudo-governor is performance: the kernel doc states that in HWP plus performance the driver writes 0 to the EPP knob and "any attempts to change the EPP/EPB to a value different from 0 (performance) via sysfs in this configuration will be rejected". Things the record got right and I kept: /etc/default/cpupower-service.conf and cpupower.service are the correct names, confirmed against the Arch wiki CPU_frequency_scaling page and against the package file list from https://archlinux.org/packages/extra/x86_64/cpupower/json/ (cpupower 7.2.5-1 in extra, not installed here). sudo pacman -S cpupower is not blocked by the Omarchy guard: I read /usr/bin/omarchy-update-pacman-guard and it aborts only when both sync and sysupgrade are present, so a plain -S passes. I added that cpupower.service and power-profiles-daemon.service both write scaling_governor and must not both run, which the Arch wiki flags for the same reason for TLP and tuned. Severity low and frequency very-common are right and I left them, because nothing is actually broken. NOT exercised: I changed nothing, so I did not write an EPP value, did not add a Limine drop-in, did not reboot into passive mode, and did not install cpupower. The rejected-EPP-write behaviour and everything about amd_pstate are from the kernel documentation rather than from this machine, which has no /sys/devices/system/cpu/amd_pstate/.

The cause above was rewritten on 2026-09-12 to match this note; the fix was corrected by the audit itself.
RISK

Forcing the performance profile or EPP on a laptop raises temperatures and shortens battery life noticeably. Watch temperatures with sensors after changing it. The passive-mode step edits the kernel command line and limine-mkinitcpio rebuilds the UKI and the boot entry, so make sure you can reach the Limine menu and its snapshot entries before you reboot. Running cpupower.service alongside power-profiles-daemon.service leaves the governor at whichever of the two wrote last.

Symptom

cpupower frequency-set -g ondemand fails, and cpupower frequency-info lists only powersave and performance. Whatever I set reverts to powersave after a reboot, and I assume my CPU is being throttled.

Cause

intel_pstate and amd_pstate in *active* mode bypass the classic cpufreq governors and expose two pseudo-governors that happen to share their names. These are not the old governors. Both scale dynamically and map onto an Energy Performance Preference (EPP) hint, so nothing is being throttled and ondemand is not missing, it does not exist in this mode. Sysfs writes to scaling_governor and energy_performance_preference are also not persistent across boots on their own. On Omarchy 4 there is a second and larger reason a hand-set value does not stick: power-profiles-daemon is in the base package set, is enabled at install, and is driven from the Super+Space menu and from omarchy-powerprofiles-init at session start. Its intel_pstate and amd_pstate drivers write both scaling_governor and energy_performance_preference for every policy on every profile change, so they overwrite anything set by hand or from tmpfiles.d.

Fix

Check what you are actually running:


cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_preference
cat /sys/devices/system/cpu/intel_pstate/status 2>/dev/null   # Intel only

If scaling_driver reads intel_pstate or amd_pstate_epp you are in active mode, ondemand does not exist, and the knob to turn is the EPP hint.

**On Omarchy 4, power-profiles-daemon owns this setting.** It is installed and enabled by default and it rewrites both scaling_governor and energy_performance_preference on every profile change, so set the profile rather than the sysfs file:


powerprofilesctl get
powerprofilesctl list
omarchy-powerprofiles-set ac balanced           # remembered for mains
omarchy-powerprofiles-set battery power-saver   # remembered for battery

omarchy-powerprofiles-set saves the choice under ~/.local/state/omarchy/powerprofiles/ and omarchy-powerprofiles-init restores it at session start, so that is the persistence you want here. The profiles map onto EPP power for power-saver, balance_power or balance_performance for balanced, and performance for performance. The same choices are in the Super+Space menu.

**On plain Arch with no power-profiles-daemon**, write the EPP yourself:


cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_available_preferences
echo balance_performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference

That write is refused while the pseudo-governor is performance. With HWP the driver pins EPP to performance and rejects any other value, so switch the pseudo-governor first:


echo powersave | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

Make it persistent with the same value you just tested:


sudo tee /etc/tmpfiles.d/energy_performance_preference.conf <<'EOF'
w /sys/devices/system/cpu/cpufreq/policy*/energy_performance_preference - - - - balance_performance
EOF
sudo systemd-tmpfiles --create

If power-profiles-daemon is running, this file is applied at boot and then overwritten by the daemon. Use the profile route above or mask the daemon, but do not run both.

**If you genuinely want the classic governors**, switch the driver to passive mode with a kernel parameter: intel_pstate=passive or amd_pstate=passive, or amd_pstate=disable to fall back to acpi_cpufreq. On Omarchy 4 the kernel cmdline is assembled from Limine drop-ins. Put the parameter in a new file under /etc/limine-entry-tool.d/ and nowhere else. Do not also append it to /etc/default/limine: that file is owned by no package and is read last at the highest priority, so with += in both the parameter lands on the cmdline twice.


printf 'KERNEL_CMDLINE[default]+=" intel_pstate=passive"\n' | sudo tee /etc/limine-entry-tool.d/pstate.conf
sudo limine-mkinitcpio
sudo reboot

After the reboot, scaling_driver reads intel_cpufreq on Intel and scaling_available_governors lists the generic governors such as schedutil, ondemand and conservative. Only now is there a governor for cpupower to pin. cpupower is not installed on Omarchy 4, and on a fresh install the pacman sync databases are empty, so run omarchy update before installing anything:


sudo pacman -S cpupower
sudoedit /etc/default/cpupower-service.conf     # set governor='schedutil'
sudo systemctl enable --now cpupower.service

cpupower.service and power-profiles-daemon.service both write scaling_governor, so whichever runs last wins. Pick one. On Omarchy that means masking power-profiles-daemon.service, which also empties the power profile entry in the Super+Space menu.

Verify

After a reboot, cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver and cat /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference show what you set. On Omarchy 4 check powerprofilesctl get as well, because the daemon is what wrote the value and it is the thing that will change it again. Under load, watch -n1 'grep MHz /proc/cpuinfo' shows frequencies rising.

Sources

cpu-governor-not-persistent-pstate-active