A vulnerability in the Linux kernel's drm/xe component allowed non-fault-mode scratch virtual machines (VMs) to cause indefinite GPU retries by improperly handling page faults. The issue was that scratch-enabled VMs were incorrectly permitted through the ASID lookup, including dma-fence mode VMs, which could fault on valid VMAs without proper fault handling. This vulnerability has been resolved by restricting ASID lookup to only fault-mode VMs, maintaining support for fault-mode scratch VMs while rejecting faults from 3D VMs.
In the Linux kernel, the following vulnerability has been resolved:
nfsd: fix layout fence worker double-reference race
The workqueue core clears WORK_STRUCT_PENDING before the callback
is invoked, so delayed_work_pending() in lm_breaker_timedout() can
return false while the fence worker is already running. This lets
the breaker take a duplicate sc_count reference and schedule a new
worker that coalesces with the in-progress one. The extra reference
is never put, leaking the layout stateid.
Replace the racy delayed_work_pending() check with an
ls_fence_inflight boolean set atomically with
refcount_inc_not_zero() under ls_lock, and cleared under ls_lock
before the final nfs4_put_stid() on the dispose path; the retry
path intentionally retains it. Remove the self-rearm
mod_delayed_work() at the top of the worker.
In the Linux kernel, the following vulnerability has been resolved: dma-fence: Fix potential tracepoint null pointer dereferences Trace_dma_fence_signaled, trace_dma_fence_wait_end and trace_dma_fence_destroy can all currently dereference a null fence->ops pointer after it has been reset on fence signalling. Lets use the safe string getters for most tracepoints to avoid this class of a problem, while for the signal tracepoint we move it to before ops are cleared to avoid losing the driver and timeline name information. Apart from moving it we also need to add a new tracepoint class to bypass the safe name getters since the signaled bit is already set. For dma_fence_init we also need to use the new tracepoint class since the rcu read lock is not held there, and we can do the same for the enable signaling since there we are certain the fence cannot be signaled while we are holding the lock and have even validated the fence->ops.
CVE-2026-72095 is a high-severity vulnerability in the Linux kernel version 6.16 affecting the dma_fence_dedup_array() function. The function incorrectly returns 1 when called with zero fences, contradicting its documented behavior and causing potential dereferencing of uninitialized memory in certain callers like amdgpu_userq_wait_*. This flaw has been fixed by updating the function to return 0 early for zero input, preventing unsafe memory access and unnecessary operations.
A flaw was found in insights-core where the password redaction layer fails to recognize credentials not keyed under the literal string 'password'. This allows SSSD LDAP bind passwords (ldap_default_authtok) and Pacemaker fence device credentials to be included in cleartext in archives uploaded to console.redhat.com.
In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu/userq: fix indefinite fence wait during GPU reset pre_reset only force-completes fences of MAPPED queues. A queue in any other state (e.g. mid-eviction) keeps its last_fence pending; after a GPU reset that fence never signals, so the eviction/suspend worker and process teardown (amdgpu_evf_mgr_flush_suspend) wait on it forever and wedge the machine: INFO: task kworker/6:28 blocked for more than 120 seconds. Workqueue: events amdgpu_eviction_fence_suspend_worker [amdgpu] Call Trace: dma_fence_wait_timeout+0x7e/0x130 amdgpu_userq_evict+0x67/0x140 [amdgpu] amdgpu_eviction_fence_suspend_worker+0xd8/0x160 [amdgpu] process_scheduled_works+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)
CVE-2026-68384 is a race condition vulnerability in the Linux kernel's drm/xe/vf component involving the attach and detach operations of VF CCS batch buffers during buffer object (BO) moves. The flaw allows races with VF migration events that can observe or interfere with partially copied CCS metadata, potentially leading to incorrect save/restore states. The issue is fixed by moving the attach operation before job submission and detaching only after the copy fence signals, ensuring proper synchronization.
CVE-2026-68103 is a vulnerability in the Linux kernel's AMDGPU driver where the mapping of a reserved doorbell to a new user queue was improperly handled. This flaw allowed a new queue creation with the same doorbell buffer object handle and offset to overwrite existing queue and doorbell mappings. The issue could cause misrouting of fence interrupt requests and incorrect cleanup operations between queues. The vulnerability has been fixed by changing the mapping function to reject attempts to map a reserved doorbell to a new queue.
Software installed and run as a non-privileged user may conduct improper GPU system calls to manipulate the lifetimes of synchronisation objects in the kernel, leading to read/write UAFs.
During workload submission involving a fence exported by the GPU driver, the reference count of the underlying synchronisation primitive is not properly incremented. This can be exploited, by destroying the exported fence and prematurely release the underlying primitive, resulting in a potential use-after-free condition.
A memory leak vulnerability in the Linux kernel's Industrial I/O (iio) buffer subsystem was fixed. The function iio_buffer_enqueue_dmabuf() failed to release an initial reference to a DMA fence object, causing a permanent memory leak of 128-byte allocations. This leak accumulates rapidly under typical usage, leading to increased kernel memory consumption. The issue was resolved by properly releasing the DMA fence reference after it is added to the DMA reservation object.
CVE-2026-63856 is a high-severity vulnerability in the Linux kernel's AMDGPU driver affecting the VCN v2.0 encoder and decoder rings. The issue arises because these rings do not support 64-bit user fence writes, which could lead to improper handling of command submissions. The vulnerability has been addressed by setting the no_user_fence flag for these rings, causing the driver to reject command submissions that include user fences.
CVE-2026-63853 is a high-severity vulnerability in the Linux kernel's AMDGPU driver, specifically affecting the VCN v4.0 encoder ring. The issue arises because the VCN encoder and decoder rings do not support 64-bit user fence writes and reject command submissions that include user fences. This flaw was addressed by setting the no_user_fence flag for the VCN v4.0 encoder ring to prevent improper user fence usage.
CVE-2026-63852 is a high-severity vulnerability in the Linux kernel's AMDGPU driver, specifically affecting the VCN v4.0.3 encoder ring. The issue arises because the VCN encoder and decoder rings do not support 64-bit user fence writes, leading to rejection of command submissions that include user fences. This flaw has been addressed by setting the no_user_fence flag for the affected encoder ring in Linux kernel version 6.5.
CVE-2026-63850 is a high-severity vulnerability in the Linux kernel's AMDGPU driver affecting the VCN v5.0.0 encoder ring. The issue arises because the VCN encoder and decoder rings do not support 64-bit user fence writes and reject command submissions that include user fences. This flaw has been addressed by setting the no_user_fence flag for the VCN v5.0.0 encoder ring.
CVE-2026-63848 is a high-severity vulnerability in the Linux kernel's AMDGPU JPEG driver. It involves improper handling of user fence writes on the JPEG v2.0 ring, which does not support 64-bit user fence writes. The vulnerability is addressed by rejecting command submissions that include user fences for this ring. A fix has been applied in the Linux kernel source.
A vulnerability in the Linux kernel's AMDGPU JPEG driver (version 5.19) involves improper handling of user fence writes for the JPEG v4.0 ring. The JPEG rings do not support 64-bit user fence writes and reject command submissions that include user fences. This flaw has been addressed by setting the no_user_fence flag for the JPEG v4.0 ring.