CVE-2024-44961

What it actually does

The AMD GPU driver fails to forward soft recovery errors to userspace, allowing hanging command buffers to cascade into hard GPU resets.

Why it matters to the business

Systems using AMD GPUs on Linux may experience repeated hard resets, causing downtime, lost work, and operational disruption on graphics workloads.

How an attack unfolds

An attacker able to submit GPU command buffers triggers hangs; the missing error forwarding prevents graceful userspace handling and forces escalation to a hard reset.

How to fix it

Update the Linux kernel to a version containing commit 0da0b06165d83a8ecbb6582d9d5a135f9d38a52a.

How to confirm the fix worked

Verify the drm/amdgpu driver contains the soft-recovery forwarding logic and test that GPU errors are now reported to userspace instead of escalating.

Typical attack path

1. Obtain ability to submit workloads to the AMD GPU
2. Submit command buffers designed to hang
3. Trigger soft recovery path without userspace notification
4. Cascade into full hard GPU reset

How to detect it

Inspect kernel logs for repeated amdgpu soft recovery messages without corresponding userspace notifications and correlate with GPU reset events.

What your security team should do

• Apply the kernel fix to all AMD GPU hosts
• Monitor dmesg and journal logs for amdgpu reset events
• Prioritize patching on systems running graphics or compute workloads