CVE-2026-98898

What it actually does

Attempt to schedule delayed work on an uninitialized workqueue when the amdgpu.dpm=0 parameter disables DPM initialization, resulting in a race condition that can trigger a kernel crash or undefined behavior in the thermal management path.

Why it matters to the business

Systems using AMD GPUs may experience unexpected kernel panics or instability when the DPM feature is intentionally disabled via module parameter, disrupting graphics workloads and requiring reboots.

How an attack unfolds

An attacker with local access loads the amdgpu driver with DPM disabled, then triggers thermal events that attempt to queue work on the uninitialized structure, causing a denial-of-service crash; no remote vector or privilege escalation is present.

How to fix it

Upgrade to a Linux kernel containing commit bd018d36171a695952c6d391471c279c9e05c8b2 or later; avoid using amdgpu.dpm=0 on vulnerable kernels until patched.

How to confirm the fix worked

Confirm kernel source contains the check that skips work scheduling when the thermal queue is uninitialized; boot with amdgpu.dpm=0 and verify no crashes occur under thermal load.

Typical attack path

1. Load amdgpu kernel module with amdgpu.dpm=0 parameter
2. Trigger GPU thermal or power management events
3. Kernel attempts to schedule work on uninitialized queue
4. Resulting crash or memory corruption causes denial of service

How to detect it

Review kernel logs for workqueue initialization failures or scheduling attempts on uninitialized queues when amdgpu is loaded with DPM disabled; correlate with module parameter settings at boot.

What your security team should do

• Prioritize kernel upgrade on all AMD GPU systems within standard patch windows
• Audit module parameters for amdgpu.dpm usage across fleets
• Monitor for kernel oops or workqueue-related panics in logs