Linux Kernel AMD GPU Hibernation Recovery Page Fault Vulnerability

A vulnerability in the Linux kernel's AMD GPU driver can lead to a page fault after resuming from hibernation in a passthrough environment. During hibernation, a mode reset occurs, but the partition mode is not properly restored upon resume. This mismanagement causes the compute pipeline to access buffer objects with an incorrect stride size, leading to out-of-bounds access and a resulting page fault. The issue arises in the AMDGPU driver, specifically within the compute partition management after hibernation.

Impact

The vulnerability causes a GPU page fault, disrupting normal graphics processing and potentially leading to application crashes or graphical errors.

Reproduction

To reproduce this vulnerability, place the system in a passthrough environment and initiate a hibernation cycle. After the system resumes, the CoralGem application will trigger a GPU page fault due to the improper restoration of the compute partition mode, causing the compute pipeline to access buffer objects incorrectly.

Remediation

The vulnerability has been addressed by ensuring that the function to switch compute partitions is called when resuming from hibernation. Users can apply the latest patches available in the Linux kernel stable tree to mitigate this issue.