A race condition in the Linux kernel amdkfd driver may allow local authenticated users to trigger memory corruption or denial of service.

The Micro Engine Scheduler (MES) fails to perform heavy-weight TLB invalidation after unmapping queues. This flaw, which requires local user privileges (PR:L), allows in-flight DMA descriptors to access memory that has already been unmapped or migrated.

The vulnerability poses a significant risk to system stability and data integrity, particularly for environments utilizing GPU compute workloads. Because the flaw can lead to page faults and GPU queue hangs, it may result in unplanned system downtime for critical services relying on GPU acceleration. Furthermore, the potential for unauthorized memory access could be leveraged to compromise the confidentiality and integrity of sensitive data processed within the kernel context.

Immediate Action: Update the Linux kernel to version 6.18.51, 7.2.5, or a later patched release provided by your distribution vendor.

Proactive Monitoring: Monitor system logs for repeated GPU-related page faults or kernel panic events that might indicate attempts to trigger this race condition.

Compensating Controls: Restrict access to GPU compute resources to authorized users only, as the vulnerability requires local access to the system to exploit.

Public Exploit Available: Unknown

Given the potential for system instability and the severity of kernel-level memory corruption, organizations should prioritize patching affected systems during the next scheduled maintenance window. System administrators operating high-performance computing clusters or environments with heavy GPU utilization should treat this update with elevated urgency to maintain operational continuity and security.