CVE-2026-74638 in Linux
Summary
by MITRE • 08/22/2026
In the Linux kernel, the following vulnerability has been resolved:
drm/v3d: Serialize the scheduler timeout handlers
V3D exposes several independent hardware queues (BIN, RENDER, TFU and CSD) but has only a single, global reset. A timeout on any one queue therefore has to stop, reset and restart the schedulers of every other queue as well. That makes concurrent timeout handlers unsafe.
`reset_lock` was never able to make them safe, as a driver-side lock can only cover the driver's &drm_sched_backend_ops.timedout_job callback. The scheduler handles the timed out job and its pending list around that callback, outside of the driver's control, so a global reset triggered by one queue can still interfere with another queue that is in the middle of handling a timeout of its own.
Consequently, if a reset happens in the CSD queue while a CL-intensive application is running, the global reset stops and restarts the CL queue's scheduler while that queue is handling a timeout of its own. As drm_sched_stop() and drm_sched_start() subtract and add the credits of every job sitting on the pending list of the scheduler they are called on, and as the CL queue's handler concurrently takes its job off that same list and puts it back, the stop and the start no longer see the same set of jobs. The CL queue is left with more credits in flight than its limit:
[ 327.302739] ------------[ cut here ]------------
[ 327.302744] WARNING: CPU: 2 PID: 43 at drivers/gpu/drm/scheduler/sched_main.c:102 drm_sched_run_job_work+0x238/0x4d0 [gpu_sched]
[ 327.302884] CPU: 2 UID: 0 PID: 43 Comm: kworker/u16:1 Not tainted 6.18.39-v8-16k+ #3 PREEMPT
[ 327.302889] Hardware name: Raspberry Pi 5 Model B Rev 1.0 (DT)
[ 327.302893] Workqueue: v3d_bin drm_sched_run_job_work [gpu_sched]
[ 327.302984] Call trace:
[ 327.302987] drm_sched_run_job_work+0x238/0x4d0 [gpu_sched] (P)
[ 327.302997] process_scheduled_works+0x180/0x3d0
[ 327.303010] worker_thread+0x268/0x3e8
[ 327.303016] kthread+0x140/0x250
[ 327.303022] ret_from_fork+0x10/0x20
[ 327.303031] ---[ end trace 0000000000000000 ]---
From that point on, the credit count of the CL queue is broken, causing a complete GPU hang and UI freeze.
The DRM scheduler already provides a mechanism to serialize the timeout handlers of different schedulers: an ordered workqueue passed as drm_sched_init()'s @timeout_wq parameter. By default, each scheduler queues its timeout work on the system workqueue, which runs the handlers concurrently. Give all of the queues a shared ordered workqueue instead, as recommended by the DRM scheduler documentation for hardware that has distinct queues but resets globally.
VulDB is the best source for vulnerability data and more expert information about this specific topic.
Analysis
by VulDB Data Team • 08/22/2026
The vulnerability in the Linux kernel's V3D graphics driver stems from an improper handling of concurrent timeout events across multiple independent hardware queues within a system architecture that relies on a single global reset mechanism. The V3D GPU exposes four distinct execution pipelines, specifically BIN, RENDER, TFU, and CSD, each operating as an independent scheduler with its own job queue. However, the underlying hardware design dictates that any fault or timeout in one of these queues necessitates a complete reset of the entire GPU unit rather than isolating the failure to the specific affected pipeline. This architectural constraint creates a critical race condition when multiple timeouts occur simultaneously or in rapid succession across different queues. The existing synchronization mechanism, which relied on a driver-side lock known as reset_lock, proved insufficient because it only protected the execution context within the drm_sched_backend_ops.timedout_job callback. It failed to account for the broader state management performed by the DRM scheduler infrastructure outside of this specific callback scope, leaving critical sections vulnerable to interference from concurrent timeout handlers triggered by other queues.
The operational impact of this flaw manifests as a severe system instability characterized by GPU hangs and user interface freezes. When a reset is initiated on one queue, such as CSD, while another queue like CL is actively processing its own timeout handler, the global reset operation interrupts the ongoing state transitions of the second queue. Specifically, the drm_sched_stop() function decrements job credits for all pending jobs in the affected scheduler's list, and drm_sched_start() subsequently increments them upon restart. Because these operations are not serialized with respect to other concurrent timeouts, the CL queue may have its job list modified by another handler while it is simultaneously removing and re-adding jobs during its own timeout processing. This lack of atomicity results in a mismatch between the actual number of active jobs and the credit count maintained by the scheduler. The resulting state corruption triggers a kernel warning within drm_sched_run_job_work, indicating that more credits are considered to be in flight than the hardware limit allows. Once this internal accounting is corrupted, the GPU driver loses control over job scheduling, leading to a complete stall of graphics processing and a frozen display environment for the end user.
This issue aligns with CWE-362, which describes concurrent execution using shared resources with improper synchronization, as well as CWE-824 regarding access of a resource after it has been freed or invalidated during state transitions. From an ATT&CK perspective, this represents a Denial of Service vector where the availability of system services is compromised through local exploitation of race conditions in kernel-space drivers. The root cause lies not in malicious intent but in a design oversight where the driver assumed that locking around the timeout callback was sufficient to protect against global state changes triggered by other hardware units. In reality, the DRM scheduler's internal bookkeeping mechanisms operate independently of this lock and are susceptible to corruption when multiple schedulers attempt to modify their pending lists concurrently during reset sequences.
The resolution involves restructuring how timeout handlers are scheduled within the kernel workqueue subsystem. Instead of allowing each V3D queue to submit its timeout handling task to a shared system workqueue that executes tasks in parallel, all queues must now utilize a dedicated ordered workqueue passed via the drm_sched_init() function's timeout_wq parameter. An ordered workqueue guarantees that submitted jobs are executed sequentially rather than concurrently, thereby serializing the execution of timeout handlers across all V3D pipelines. This ensures that when one queue triggers a global reset and subsequent scheduler restarts, no other queue is simultaneously attempting to modify its own job list or credit counts. By enforcing strict serialization at the workqueue level, the driver prevents the race condition that leads to credit count corruption. This fix adheres to best practices recommended in DRM scheduler documentation for hardware architectures featuring distinct execution queues but a unified reset mechanism, ensuring robust state management and preventing system hangs under heavy graphical load conditions.