Linux Kernel GPU Hard Reset Sequence Vulnerability Causes Kernel Crash

A vulnerability in the Linux kernel's GPU hard reset sequence can lead to a kernel crash. This issue arises in the DRM (Direct Rendering Manager) subsystem for Imagination GPUs. The hard reset process improperly uses the 'pm_runtime_force_suspend()' and 'pm_runtime_force_resume()' functions, which should only be applied during system-wide power management transitions. Depending on the internal runtime power management state, 'pm_runtime_force_resume()' may not reactivate the GPU unless necessary. If this occurs, the expected resume callback is not triggered, leaving the GPU clocks disabled. The kernel then crashes when it tries to access GPU registers during the power-on sequence. The vulnerability has been addressed by replacing the problematic power management calls with direct references to the driver's specific power management callbacks, ensuring the GPU clocks are properly re-enabled and preventing the kernel crash.

Impact

The vulnerability can cause a kernel crash by disrupting the GPU power management process, leading to a failure when the system attempts to access GPU registers.

Reproduction

The vulnerability can be reproduced by initiating a hard reset sequence on an affected Imagination GPU within the Linux kernel. The process will involve the GPU's runtime power management functions, which, due to the vulnerability, will not correctly resume the GPU's operation. This mismanagement can be observed when the kernel subsequently tries to access the GPU registers as part of the normal power-on sequence, resulting in a crash.

Remediation

Users can upgrade to the latest version of the Linux kernel where this vulnerability has been fixed. Instructions for downloading the patched version are available on the official Linux kernel website.