KernelScan.io

HIGH Introduced in 2.6.12

amdkfd Checkpoint OOB

CVE-2026-68447

CVSS 7.0 / 10.0 KernelScan AI

CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:H

01

In the Linux kernel, the following vulnerability has been resolved: drm/amdkfd: clamp v9 CRIU control stack checkpoint copy to BO size CRIU checkpoint copies the MQD control stack using cp_hqd_cntl_stack_size from hardware without bounding it to the allocated BO region. If the HW field is larger than the queue's control stack allocation, memcpy reads past the BO into adjacent GTT memory and can leak kernel data to userspace. Store the page-aligned control stack BO size in mqd_manager and clamp checkpoint copies and reported checkpoint sizes to min(cp_hqd_cntl_stack_size, mm->ctl_stack_size). Apply the same bound for multi-XCC v9.4.3 checkpoint layout. (cherry picked from commit 6c2abd0ec09e86c6323010673766f76050e28aa3)

02

Engine v0.6.0

Risk summary

A local user with access to an AMD GPU compute device can trigger a CRIU checkpoint operation that copies more data than was allocated for the control stack buffer, reading into adjacent kernel memory. The excess data is returned to userspace in the checkpoint, leaking kernel memory contents. This requires AMD GPU hardware with CWSR enabled and access to the GPU render device.

Affecteddrivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c (amdkfd)

Vulnerability analysis

During CRIU checkpoint of a GPU compute queue, the driver copies the MQD control stack using a size field read directly from hardware without checking it against the actual allocated buffer object size. When the hardware-reported size exceeds the allocation, the copy reads past the buffer into adjacent GTT memory, and the checkpoint data is returned to userspace, exposing kernel memory. The fix stores the page-aligned control stack allocation size at initialization time and clamps every checkpoint copy and reported checkpoint size to the minimum of the hardware field and the actual allocation, preventing the over-read. This path is reachable by a local user who has access to the AMD GPU compute device (typically via the render device node), and requires AMD GPU hardware with CWSR enabled.

03

BranchIntroducedFixed inPatch commit
7.12.6.127.1.6a0d87beb2660
mainline2.6.127.2-rc2426ffae6ecc7