CVE-2024-35817

What it actually does

Failure to set the GTT bound flag in amdgpu_ttm_gart_bind leaves stale GART page table entries after a BO is released, allowing GPU reads of freed system pages instead of triggering a page fault.

Why it matters to the business

Kernel memory corruption in AMD GPU drivers can cause system instability, undefined behavior during graphics operations, and potential privilege escalation paths on affected Linux systems, increasing operational risk for graphics workloads.

How an attack unfolds

An attacker with local access crafts GPU operations that trigger GTT binding and subsequent release, then induces erroneous GPU access to the stale GART address to read undefined memory values, potentially leading to further corruption or information leaks.

How to fix it

Update the Linux kernel to a fixed version such as 5.10.197-1 or later for Debian 11 packages; apply vendor backports for other distributions and verify the amdgpu_ttm_gart_bind flag logic is present.

How to confirm the fix worked

Confirm the kernel version contains the commit that sets the GTT bound flag in amdgpu_ttm_gart_bind; inspect source or run targeted tests that allocate/release GTT BOs and verify no stale mappings remain.

Typical attack path

1. Obtain local access on a system with AMDGPU driver
2. Allocate and bind GTT buffer objects via DRM interfaces
3. Release the BO while GART mappings remain active due to missing flag
4. Trigger GPU access to the now-stale GART address
5. Observe or exploit resulting undefined memory reads for corruption

How to detect it

Review kernel logs for amdgpu GART-related warnings or faults around BO release times; examine DRM debug output and memory access patterns for evidence of stale mapping usage on affected kernels.

What your security team should do

• Prioritize kernel updates on all AMDGPU-equipped systems within standard patch windows
• Monitor upstream stable kernels for additional related fixes
• Restrict untrusted local access where GPU workloads are not required