CVE-2025-21645

What it actually does

The AMD PMC driver's suspend handler, registered via DEFINE_SIMPLE_DEV_PM_OPS for .suspend/.freeze/.poweroff, can decrement IRQ1 wake_depth below zero when i8042 only enabled it for .suspend, triggering an unbalanced wake disable warning in irq_set_irq_wake.

Why it matters to the business

Administrators see spurious kernel warnings during hibernation or other power transitions on affected AMD systems; no data loss or downtime occurs but log noise may slow incident response.

How an attack unfolds

No attacker exploitation path exists; the issue manifests only during legitimate S4 hibernation after a fresh boot without prior s2idle, due to mismatched PM handler registration between PMC and i8042 drivers.

How to fix it

Update the Linux kernel to version 6.1.140 or later, or apply commit ab47d72b736e78d3c2370b26e0bfc46eb0918391; for listed Red Hat packages apply the corresponding 1:10.0.26-2.el6 or el7 updates.

How to confirm the fix worked

After patching, confirm absence of the 'Unbalanced IRQ 1 wake disable' warning during S4 hibernation; verify kernel version or git commit matches the fixed release.

Typical attack path

1. N/A - not an exploitable vulnerability

How to detect it

Search kernel logs for the strings 'Unbalanced IRQ 1 wake disable' and 'Disabling IRQ1 wakeup source to avoid platform firmware bug' around hibernation events.

What your security team should do

• Apply the kernel or package updates listed in fixed_versions at the next maintenance window
• Monitor kernel logs for the unbalanced IRQ warning on AMD platforms during power testing