CVE-2026-68237: drm/amdgpu/userq: fix indefinite fence wait during GPU reset
In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu/userq: fix indefinite fence wait during GPU reset
prereset only force-completes fences of MAPPED queues. A queue in any other state (e.g. mid-eviction) keeps its lastfence pending; after a GPU reset that fence never signals, so the eviction/suspend worker and process teardown (amdgpuevfmgrflushsuspend) wait on it forever and wedge the machine:
INFO: task kworker/6:28 blocked for more than 120 seconds. Workqueue: events amdgpuevictionfencesuspendworker [amdgpu] Call Trace: dmafencewaittimeout+0x7e/0x130 amdgpuuserqevict+0x67/0x140 [amdgpu] amdgpuevictionfencesuspendworker+0xd8/0x160 [amdgpu] processscheduledworks+0xa6/0x420
Force-complete every queue's fence regardless of state. The unmap and mark-hung step stays gated on MAPPED, since unmapping a queue that is not mapped is invalid.
(cherry picked from commit 9102b39fa924dcc3dc75a3137bfa9633c40b88c0)
Affected Software
Event History
Frequently Asked Questions
What is the severity of CVE-2026-68237?
CVE-2026-68237 has a risk score of 37, indicating a moderate level of severity.
How do I fix CVE-2026-68237?
To fix CVE-2026-68237, you should update to the latest version of the Linux kernel where the vulnerability has been resolved.
What types of systems are affected by CVE-2026-68237?
CVE-2026-68237 affects systems running the Linux kernel with the amdgpu driver.
What are the potential consequences of CVE-2026-68237?
The potential consequence of CVE-2026-68237 is an indefinite wait on GPU fences, which can lead to system instability during GPU resets.
When was CVE-2026-68237 published?
CVE-2026-68237 was published on August 10, 2026.