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

Diagnose a CPU stuck at low clocks while the machine is cool

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

Diagnostics and mechanisms are right, and the throttlestop script is copied faithfully from wiki.archlinux.org/title/Lenovo_ThinkPad_T480#CPU_stuck_at_minimum_frequency (including the reg%2 test and the msr-tools note). processor.ignore_ppc=1 / /sys/module/processor/parameters/ignore_ppc and the bios_limit path are verbatim from CPU frequency scaling. throttled (extra 0.12), turbostat, msr-tools, lm_sensors all exist. Three defects. (1) Wrong config path: the current Arch throttled package ships etc/throttled.conf and usr/lib/systemd/system/throttled.service. There is no /etc/lenovo_fix.conf. That name is the pre-rename lenovo_throttling_fix era. A user editing it will see no effect at all. (2) The cause is wrong on a safety-relevant point. The Arch wiki says BD PROCHOT 'is meant to protect the system and can be triggered by many reasons—the CPU temperature rising above 60 °C, using a third party battery, etc.', so 'completely independent of core temperature' is false, and clearing the bit can remove a live thermal protection. (3) The record strips the wiki's explicit warning on ignore_ppc and presents it as merely 'reversible, no reboot'. The wiki carries a Warning box: 'CPU frequency limitation is a safety feature of your BIOS and should not need to be bypassed in most cases' and 'This can seriously damage your hardware: use at your own risk.' Given these commands go into a root shell, that omission has to be fixed.

The cause above was not rewritten and may still contain the error described. The fix below is the corrected version.
RISK

This is the most hazardous fix in the set. BD PROCHOT and firmware RAPL limits are electrical and thermal *protections*: the Arch wiki's own warning on the BIOS frequency-limitation section is that bypassing them "can seriously damage your hardware." On a machine with a swollen or failing battery, an undersized or counterfeit charger, or a heatsink full of dust, clearing PROCHOT or raising PL1 can overheat the VRM/CPU or brown out the system under load. Do not apply any of this to a machine that is actually running hot. Verify with turbostat first. Do not run throttled.service alongside a hand-rolled MSR script or another undervolting tool. They overwrite each other and the resulting state is unpredictable. wrmsr writes to a CPU model-specific register. A typo in the register number can hang or destabilise the machine.

Symptom

The laptop is barely warm (sensors says 55 °C) but everything crawls. Under full load watch -n1 "grep 'MHz' /proc/cpuinfo" sits at 400 to 800 MHz even with the performance governor and boost enabled. On a ThinkPad it drops to exactly 400 MHz the moment I plug into a dock or a non-Lenovo charger. On other machines the package sticks at ~15 W when the chip is rated for 45 W. It is clearly not heat, so what is limiting it?

Cause

You are being power-limited or signal-limited rather than limited by the CPU's own reported core temperature. Three mechanisms produce identical symptoms. (a) **BD PROCHOT**, a hardware line the embedded controller can assert to force the CPU to its minimum P-state. It is asserted by the EC, not by the CPU's thermal control, so sensors can look cool while it is active, but it is a protection mechanism and the EC asserts it for real reasons: a third-party or undersized battery/charger, a hot VRM or chassis sensor, a dock, and on many ThinkPads a CPU temperature crossing an EC threshold as low as ~60 °C (well below Tjmax). Clearing the bit disables that protection. (b) A **BIOS _PPC limit** exposed as /sys/devices/system/cpu/cpu0/cpufreq/bios_limit, typically set when the firmware sees a failing battery or an undersized adapter. The Arch wiki treats overriding it as a hardware risk. (c) **RAPL power limits** (PL1/PL2) programmed low by firmware or Intel DPTF, which the kernel honours faithfully. The governor cannot override any of these.

Fix

Step 3: add the wiki's warning before the command:

> ⚠️ **Risk.** Per wiki.archlinux.org/title/CPU_frequency_scaling: 'CPU frequency limitation is a safety feature of your BIOS and should not need to be bypassed in most cases' and 'This can seriously damage your hardware: use at your own risk.' Before overriding it, check the BIOS for a manually-set frequency cap or thermal preference, and check whether the battery is failing or the adapter is undersized. Those are the usual legitimate triggers.

Step 4: fix the config path and add the same caveat:


omarchy pkg add throttled          # or: sudo pacman -Syu --needed throttled
sudo systemctl enable --now throttled.service
sudo systemctl status throttled.service

Its config is **/etc/throttled.conf** (the current package ships exactly etc/throttled.conf and usr/lib/systemd/system/throttled.service). /etc/lenovo_fix.conf was the old lenovo_throttling_fix name and editing it does nothing.

> ⚠️ **Risk.** BD PROCHOT is a protection signal. Before clearing it, watch PkgTmp and per-component temperatures in turbostat/sensors under load. If the EC is asserting it because of a hot VRM, a third-party charger or a dock, clearing the bit lets the machine run hot instead of slow. Do not run the hand-rolled throttlestop service *and* throttled at the same time. Pick one, since both write MSR 0x1FC and RAPL limits.

Everything else (turbostat/RAPL inspection, powerprofilesctl, tlp-stat -p, the Limine persistence step, and the thermald pointer) is accurate as written.

Verify

Under a sustained load (stress-ng --cpu $(nproc) --timeout 120s), sudo turbostat --quiet --interval 5 --show Bzy_MHz,PkgTmp,PkgWatt shows Bzy_MHz at or near the rated all-core turbo and PkgWatt at the expected TDP, with PkgTmp still below the throttle point. sudo rdmsr -d 0x1FC returns an even value once BD PROCHOT is cleared, and cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit (if present) no longer caps you.

Sources

cpu-power-limit-throttling-low-clocks