CVE-2024-26916

What it actually does

Race condition in the amdgpu driver between GFXOFF control and SDMA ring callbacks during suspend entry, allowing concurrent SDMA traffic to trigger a kernel deadlock.

Why it matters to the business

Systems using AMD GPUs can freeze during suspend or lockscreen events, especially under GNOME, leading to downtime, lost work, and the need for hard reboots on affected Linux installations.

How an attack unfolds

An attacker with local access triggers a suspend operation that exercises the reverted GFXOFF path while SDMA traffic is active, deadlocking the kernel and denying service to the system.

How to fix it

Upgrade the Linux kernel to version 5.15.149 or later, or apply commit 65158edb0a3a8df23197d52cd24287e39eaf95d6 from the stable tree.

How to confirm the fix worked

Confirm the running kernel reports the fixed commit or version 5.15.149+, then perform repeated suspend/resume cycles under GNOME on AMD hardware while monitoring for deadlocks.

Typical attack path

1. Obtain local access or user session on an AMD GPU Linux system
2. Initiate suspend via desktop environment lockscreen or power management
3. Trigger concurrent SDMA operations during the GFXOFF transition window
4. Cause kernel deadlock and system unresponsiveness

How to detect it

Review kernel logs for amdgpu, sdma, and gfxoff messages around suspend events; correlate user-reported freezes on lockscreen with affected kernel versions.

What your security team should do

• Apply the kernel update to all AMD GPU systems immediately
• Test suspend behavior in a lab environment before broad rollout
• Add kernel version checks to vulnerability scanning and asset inventory