CVE-2026-98914

What it actually does

Unbounded recursion in the AMD GPU driver's ATOM command table parser allows a crafted table to exceed stack depth, triggering a kernel crash or denial of service.

Why it matters to the business

Attackers with local access can crash graphics workloads or the entire system on affected Linux kernels, causing downtime and loss of GPU-accelerated services; the issue is two days old with negligible observed exploitation.

How an attack unfolds

An attacker supplies a malicious ATOM command table containing more than 32 levels of nested execute commands, causing the drm/amdgpu parser to recurse until the stack is exhausted and the kernel panics.

How to fix it

Update to a Linux kernel containing the fix that caps ATOM nesting depth at 32 and returns -ELOOP; apply the patch or newer distribution kernel as soon as available.

How to confirm the fix worked

Confirm the running kernel includes the drm/amdgpu execute_depth check and -ELOOP return; test with a synthetic deeply-nested table to verify the cap is enforced without panic.

Typical attack path

1. Obtain local access or ability to load a custom GPU firmware/BIOS image
2. Construct an ATOM command table with excessive nesting depth
3. Trigger the drm/amdgpu driver to parse and execute the table
4. Cause kernel stack overflow and system crash

How to detect it

Review kernel logs for drm/amdgpu stack traces mentioning ATOM table execution or -ELOOP; look for unexpected GPU driver crashes on systems running unpatched kernels.

What your security team should do

• Prioritize kernel updates on systems with AMD GPUs within the next patch cycle
• Monitor upstream kernel changelogs for backports to supported LTS branches
• Restrict unprivileged users from loading custom firmware until patched