CVE-2026-97928

A flaw of Low impact was found in the AMD GPU (amdgpu) driver of the Linux kernel, affecting Red Hat Enterprise Linux 9 (from the 9.6 minor release onward), Red Hat Enterprise Linux 10, Red Hat In-Vehicle OS 2, and OpenShift Container Platform 4.19 and later versions that ship Red Hat Enterprise Linux CoreOS 9. During VRAM clear-on-release operations, the amdgpu_ttm_map_buffer() code path reaches video memory through its direct MC address without programming a GART window, yet the wipe still forces an unnecessary VMID 0 flush. On GFX11 hardware (such as Navi33) this spurious SDMA flush can wedge the GPU engine. A local user with access to the amdgpu graphics device and its render nodes can trigger this condition through ordinary graphics memory operations, causing the GPU engine to hang and resulting in a denial of service. The issue affects availability only; there is no impact on confidentiality or integrity.

Red Hat Enterprise Linux 7 and 8, earlier Red Hat Enterprise Linux 9 minor releases, and OpenShift Container Platform versions shipping kernels without the vulnerable code are not affected, because the affected code (introduced upstream in the 6.10 kernel DRM changes) is not present in those streams. While the CVSS:3.1 base score for this flaw is 5.5 (Moderate), Red Hat rates its impact as Low: exploitation requires local access to GPU render nodes on affected hardware, only the availability of the graphics subsystem is impacted, and no memory disclosure or corruption primitive is involved.

For the affected streams the fix is deferred: as a Low-severity, availability-only issue, it does not warrant an immediate out-of-cycle kernel update, and the upstream correction will be picked up through future regularly scheduled kernel updates. Until such an update ships, the residual risk is that a local user on systems with affected AMD GFX11 GPUs can wedge the GPU engine and hang graphics workloads until the system is recovered.