Impact clarification: this is not a cosmetic or minor usability issue. On this hardware, normal sustained CPU workloads are currently unsafe on Kali: a monitored CPU-only workload rose from 70 C to 90 C in 23 seconds and was terminated automatically; a prior session reached 107 C; and the machine suffered an abrupt, unclean shutdown during the same period. I can hardly use the system for routine local workloads without risking another thermal shutdown.

The power-profiles-daemon amd_pstate activation error is reproducible on every boot/profile-change attempt observed with kernel 7.0.12-2kali1. Switching to power-saver also fails at policy11/boost with Invalid argument and leaves inconsistent policy state. Windows on the same hardware completed a 60-second full-load test at a reported maximum of 84 C.

Please raise severity/priority accordingly. An A/B test against the installed Kali 6.19.14 kernel is still pending.

A/B kernel update: the same failure reproduces on both installed older Kali kernels.

6.17.10+kali-amd64:

• power-profiles-daemon: Failed to activate CPU driver amd_pstate
• write to /sys/devices/system/cpu/cpufreq/policy11/boost fails with Invalid argument
• policy11 remains performance/performance while policies 0-10 are powersave/balance_performance
• most sampled policy frequencies remained near 4.2 GHz

6.19.14+kali-amd64:

• identical policy11/boost Invalid argument at daemon startup
• identical inconsistent policy state
• all sampled policy frequencies were near 4.2 GHz
• idle Tctl was approximately 68 C

Therefore this is not exclusive to kernel 7.0.12-2kali1. The common failure appears to involve power-profiles-daemon, amd-pstate-epp, firmware state, or their interaction across these Kali kernels. The earlier thermal and abrupt-shutdown impact remains unchanged.

Update: root cause isolated, narrow patch developed, and tested on the affected host

Summary: power-profiles-daemon 0.30-2 performs per-policy boost writes on this AMD amd-pstate-epp system even when the global cpufreq boost control is disabled (/sys/devices/system/cpu/cpufreq/boost = 0). Linux rejects the policy11/boost write with EINVAL, aborting the profile transaction and leaving policy state inconsistent.

The proposed patch checks the global boost control. When it exists and is disabled, the daemon skips only the invalid per-policy boost writes while continuing to apply governor, EPP, and minimum-frequency settings. The patch does not enable boost or alter the global boost setting.

Validation completed:

• Isolated test suite: 123 passed, 0 failed, 6 skipped.
• Official Kali 2026.2 QEMU image: patched package installation, profile switching, daemon operation, and offline rollback all passed.
• Real host: balanced, power-saver switching, clean-session balanced soak, and performance profile all completed without a daemon failure or policy inconsistency while patched.
• Clean-session balanced soak: 40 samples over 20 minutes; 52 C minimum, 53.6 C average, 63 C maximum; 0 policy inconsistencies.
• Performance profile with VS Code and Codex active: 36 samples over 3 minutes; 56 C minimum, 57.1 C average, 62 C maximum; average load 3.11; 0 policy inconsistencies.
• Performance mode correctly applied scaling_governor=performance and energy_performance_preference=performance to all 12 policies.
• Before rollback, the patched daemon successfully restored all policies to balanced (powersave / balance_performance).
• Every host test used an armed systemd rollback service with a conservative 75 C abort threshold.
• Rollback repeatedly restored the exact original Kali 0.30-2 package and daemon binary hash. The original policy11/boost EINVAL returned only after the unmodified package was restored.

Safety limitation / unpatched baseline:

I will not perform another sustained-load or sustained-idle comparison using the unmodified package on this physical machine. A prior unpatched run reached a user-observed 107 C and was associated with machine instability/shutdown. Repeating that condition would create an unacceptable risk to the hardware and filesystem. The monitored evidence already attached to this report independently shows the unpatched transaction failure and a rapid rise to 90 C before the workload was terminated. Therefore, thermal results from the patched trials are supporting evidence, not a controlled patched-versus-unpatched power or temperature comparison, and I am not claiming a precise thermal or power-saving percentage.

Maintainer-facing conclusion:

The patch fixes the failed AMD power-profile transaction on this host and remains stable across balanced, power-saver, clean-session, and performance-profile testing. Long-term thermal and package-power effects remain under investigation.

Attaching PROPOSED-FIX.patch referenced in note 0021881 at bottom since i cant upload patch or zips its a gist.

This patch changes the amd_pstate driver so that when the global cpufreq boost control exists and is disabled, power-profiles-daemon skips only the invalid per-policy boost writes. Governor, EPP, and minimum-frequency settings continue to be applied normally. The patch does not enable boost or modify the global boost setting.

Validation:

• 123 tests passed, 0 failed, 6 skipped
• Kali 2026.2 QEMU deployment and rollback passed
• Real-host balanced, power-saver, clean-session, and performance-profile tests passed
• No policy inconsistencies or patched-daemon failures
• Verified rollback restored the exact original Kali 0.30-2 package and binary

love yall
https://gist.github.com/c0ldn3s5/845f65e3e645ed7737b5c7f0734bae0b