Linux Kernel AMD GPU Deadlock Vulnerability in SVM Range Restoration

A deadlock vulnerability has been identified in the Linux kernel's handling of AMD GPUs, specifically within the SVM (Shared Virtual Memory) range restoration process. This issue arises when a process exits and the associated work to restore SVM ranges is flushed. If the flushing process releases the last reference to the process's memory, it triggers a deadlock by conflicting with the notification release process that handles memory management. The deadlock can be avoided by moving the SVM range flushing to a different stage in the process exit workflow, ensuring that the necessary memory references are maintained while the ranges are validated and mapped to the GPU.

Impact

Exiting processes can cause a deadlock, halting system operations that rely on the affected processes, which could lead to a broader system freeze or unresponsiveness.

Remediation

Users can upgrade to the latest version of the Linux kernel where this vulnerability has been addressed. Consult the Linux kernel changelog or your distribution's update notes for specific version information.