CVE-2026-41158: GPU DDK - Backed sparse PMRs are not handled by deferred free mechanism after shrink

Synopsis

This is a use-after-free vulnerability in the Imagination Technologies Graphics DDK (GPU driver kernel module). A local attacker with a low-privilege user account can issue GPU system calls that read from or write to physical memory pages the kernel has already freed, because backed sparse PMRs (physical memory resources) are not routed through the deferred free mechanism after a shrink operation. Successful exploitation gives the attacker full read, write, and denial-of-service capability over the affected system. A patched-image rebuild at fix versions 1.18 RTM, 23.2 RTM, 24.2 RTM, and 26.2 RTM is available on HarborGuard for affected environments.

HarborGuard Coverage

Exploit Conditions

• Network reachabilityNot required

  The attacker needs an existing shell or process on the host; no network access is required to reach the vulnerable code path.

• AuthenticationRequired

  Any low-privilege local user account is sufficient; no elevated or administrative credentials are needed.

• Victim interactionNot required

  No action from another user or administrator is needed; the attacker triggers the bug entirely through their own GPU system calls.

• Attack complexityDetail

  Exploit reliability is high and no special environmental conditions, race wins, or memory-layout dependencies are required.

Blast Radius

• Reads arbitrary physical memory pages that the kernel has freed, exposing data belonging to other processes or the kernel itself, such as credentials, keys, or sensitive application state.
• Writes to those same freed physical pages, allowing corruption of kernel data structures or other processes' memory.
• Crashes the affected system or a specific process by corrupting memory in ways that cause invalid dereferences or inconsistent kernel state.
• Combines read and write primitives to escalate privileges or pivot to a full kernel compromise on the host.