Disabling eviction of the mqd and fw objects in the amdgpu driver was necessary because when these buffers were evicted the hardware continued to reference invalid memory addresses, triggering page faults that could crash the kernel and freeze the system. The flaw results from a failure in the driver to enforce that these critical queue buffers remain resident, leading to a memory corruption scenario in the kernel that manifests as a system hang. This is a deterministic crash condition, not an information disclosure or escalation, but it prevents the affected system from functioning normally.

All Linux kernel builds that incorporate the AMDGPU graphics driver are potentially affected. No specific kernel versions are listed, so any kernel lacking the fix that pins the mqd and fw objects properly may experience the crash. The issue is confined to the amdgpu subsystem within the Linux kernel.

The CVSS score is 5.5, EPSS is less than 1%, and it is not listed in the CISA KEV catalog. Exploitation would require an actor to manipulate the GPU queue to cause eviction of the protected buffers; typical scenarios involve running privileged or malicious user‑space code that communicates with the GPU. Because the vulnerability is low probability of exploitation, the potential damage is high, but the low EPSS indicates that is unlikely. The attack vector is inferred to be local or involve privileged GPU access, not remote over the network.