Linux Kernel EUVD-2026-28568

Requires Linux system with AMD GPU configured in SR-IOV Virtual Function (VF) mode using VCNv2.5 hardware acceleration AND triggering amdgpu driver module unload via rmmod/modprobe -r or system shutdown sequence that calls vcn_v2_5_hw_fini(). Exploitation limited to virtualized GPU deployments (cloud GPU instances, GPU passthrough VMs, containerized GPU workloads) where VF initialization does not enable VCN poison interrupts but cleanup code attempts release. Attack surface restricted by local access requirement (AV:L) and low privileges (PR:L) needed to unload kernel modules, typically requiring membership in sudoers or specific capability grants. Default desktop/server configurations without SR-IOV GPU virtualization are NOT affected-the vulnerable code path is inactive when GPU operates in physical function mode or when VCN features are not utilized.

Real-world risk is LOW despite the CVSS 5.5 (Medium) rating. CVSS vector (AV:L/AC:L/PR:L/UI:N) indicates local access with low privileges required and low attack complexity, but the technical impact is limited to availability (A:H) with no confidentiality or integrity impact. The EPSS score of 0.02% (4th percentile) confirms extremely low observed exploitation probability in the wild. No entries in CISA KEV and no public proof-of-concept indicate this is not being actively targeted. The vulnerability requires specific conditions: running amdgpu driver in a virtual function configuration with VCNv2.5 hardware AND triggering module unload/system shutdown. The impact is a kernel warning and potential denial of service through driver failure during cleanup-not arbitrary code execution or privilege escalation. For production systems, this presents minimal risk unless frequently hot-plugging GPU VFs or repeatedly loading/unloading the amdgpu module in virtualized environments. The availability of patches across multiple kernel stable branches (6.18.16, 6.19.6, 7.0) and the defensive nature of the fix (adding a conditional check) further reduces risk. Priority should be MEDIUM for virtualized GPU infrastructure, LOW for typical desktop/server Linux systems.

An attacker with local access and low-level privileges on a Linux system using AMD GPU virtualization triggers a kernel warning and potential denial of service by repeatedly loading and unloading the amdgpu kernel module (via modprobe -r amdgpu) in a virtual function environment. Each unload attempt calls vcn_v2_5_hw_fini(), which incorrectly tries to release an uninitialized VCN poison interrupt, triggering the WARNING at amdgpu_irq.c:641 and potentially destabilizing the driver or causing a system crash. No public proof-of-concept is currently available, but exploitation requires only standard module management commands (rmmod) available to users with sufficient privileges.