# Summary

power-profiles-daemon 0.30-2 fails with amd-pstate-epp on Ryzen 5 5600X, leaving all CPUs near 4.2 GHz at idle and causing severe load temperatures

# Category

Kali Package Bug

# Package

power-profiles-daemon 0.30-2

# Reproducibility

Always on the current boot with Linux 7.0.12-2kali1.

# Description

On Kali Linux with kernel 7.0.12+kali-amd64, power-profiles-daemon cannot activate its initial amd_pstate profile. It fails while writing the per-policy boost control for policy11:

```text
Core Failed to activate initial profile: Failed to activate CPU driver 'amd_pstate': Error writing '/sys/devices/system/cpu/cpufreq/policy11/boost': Invalid argument
```

The system then reports the `performance` profile. CPU policy is inconsistent, and all 12 logical CPUs remain near 4.2 GHz during a mostly idle sampling period. A local CPU-only Ollama workload increased CPU temperature from 70 C to 90 C in 23 seconds. A previous unmonitored workload was followed by an abrupt, unclean shutdown; this report does not claim that the daemon failure caused that shutdown.

The same machine completed a 60-second full-load test under Windows at 84 C. This is comparative context only, not proof of a Linux defect.

# System

```text
Kali kernel: Linux 7.0.12+kali-amd64 #1 SMP PREEMPT_DYNAMIC Kali 7.0.12-2kali1 (2026-06-18) x86_64
Kernel package: linux-image-7.0.12+kali-amd64 7.0.12-2kali1
libc6: 2.42-16
power-profiles-daemon: 0.30-2
CPU: AMD Ryzen 5 5600X 6-Core Processor, 6 cores / 12 threads
Motherboard: ASUSTeK TUF GAMING X570-PLUS (WI-FI), Rev X.0x
BIOS: 5044
RAM: 15 GiB
CPU driver: amd-pstate-epp
amd_pstate status: active
Kernel command line: BOOT_IMAGE=/boot/vmlinuz-7.0.12+kali-amd64 root=UUID=375a81af-13d1-42b2-bca1-024fb3946fe6 ro usbcore.autosuspend=-1 quiet splash
```

# Steps to reproduce

1. Boot Kali using linux-image-7.0.12+kali-amd64 on the hardware above.
2. Observe the power-profiles-daemon boot log.
3. Query the active profile and every cpufreq policy's governor, EPP, minimum, maximum, and current frequency.
4. Attempt to switch to power-saver.
5. Observe that the D-Bus request fails on `/sys/devices/system/cpu/cpufreq/policy11/boost` with `Invalid argument`.
6. Sample `scaling_cur_freq` while the system is mostly idle.
7. Run a bounded CPU workload with temperature monitoring and stop at 90 C.

# Actual result

```text
active profile: performance
driver: amd-pstate-epp
policy0..policy10: governor=powersave, EPP=balance_performance
policy11 before profile attempt: governor=performance, EPP=performance
all policies: scaling_min_freq=1972728, scaling_max_freq=5281819
```

The power-saver request returns:

```text
Failed to communicate with power-profiles-daemon: ... Failed to activate CPU driver 'amd_pstate': Error writing '/sys/devices/system/cpu/cpufreq/policy11/boost': Invalid argument (13)
```

After that failed request, policy11 partially changes to `governor=powersave, EPP=power`, while the other policies remain `balance_performance`; the global profile still reports `performance`.

Five idle samples, two seconds apart:

```text
policy0=4193965 policy1=4198188 policy10=4199007 policy11=4199379
policy0=4196956 policy1=4200097 policy10=4199855 policy11=4199327
policy0=4199228 policy1=4199850 policy10=4200078 policy11=4199849
policy0=4196712 policy1=4199599 policy10=4190769 policy11=4199061
policy0=4198384 policy1=4199955 policy10=4200156 policy11=4200317
```

Monitored CPU-only workload:

```text
00s 70 C
14s 86 C
16s 88 C
21s 89 C
23s 90 C - workload terminated automatically
```

# Expected result

power-profiles-daemon should apply one coherent profile to all CPU policies without a failed boost write. At idle, amd-pstate-epp should allow frequencies to fall rather than holding all logical CPUs near 4.2 GHz. Switching to power-saver should succeed.

# Additional information

The following older Kali kernels remain installed and can be used for an A/B test:

```text
linux-image-6.19.14+kali-amd64 6.19.14-1+kali1
linux-image-6.17.10+kali-amd64 6.17.10-1kali1
```

The 6.19.14 comparison is pending. No external workload or GPU acceleration was used in the monitored reproduction; Ollama reported CPU model weights and CPU KV cache.