CVE-2026-99183

accel/rocket: search every core slot when looking up a scheduler

Published Oct 28, 2026 CVSS 3.1 · 4.4 Medium EPSS < 1% KEV · Not listed
Description
In the Linux kernel, the following vulnerability has been resolved:

accel/rocket: search every core slot when looking up a scheduler

sched_to_core() walks rdev->cores[] up to rdev->num_cores, and
rocket_remove() decrements num_cores for every core it removes. Unbind a
core that is not the last one and the cores behind it fall outside the
search, so sched_to_core() returns NULL for a core that is still bound and
still running jobs. Neither caller checks the result:

rocket_job_run(): rocket_fence_create(core), core->dev
rocket_job_timedout(): dev_err(core->dev, "NPU job timed out")

Unbinding the middle core of the three on an RK3588 while three clients are
submitting to all of them faults twice, once from the surviving core's
job queue and once from its reset work:

KASAN: null-ptr-deref in range [0x0000000000000220-0x0000000000000227]
Workqueue: fdad0000.npu drm_sched_run_job_work [gpu_sched]
pc : rocket_job_run+0x234/0x838 [rocket]
Call trace:
rocket_job_run+0x234/0x838 [rocket]
drm_sched_run_job_work+0x2cc/0xad8 [gpu_sched]
process_one_work+0x640/0x14f0

KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007]
Workqueue: rocket-reset-2 drm_sched_job_timedout [gpu_sched]
pc : rocket_job_timedout+0xf0/0x1e0 [rocket]
Call trace:
rocket_job_timedout+0xf0/0x1e0 [rocket]
drm_sched_job_timedout+0x188/0x6a0 [gpu_sched]

Both are the third core: the workqueue names are its device and its
core->index, and it was left at slot 2 while num_cores had dropped to 2.

Search all the slots that were allocated, the way find_core_for_dev() now
does. A core that is still bound is then found, and the two callers get
the pointer they already assume they have.

This does not make unbinding one core out of several safe. An open client
keeps an entity pointing at the scheduler of the core that went away:
drm_sched reports it as not ready for every job that lands on it, and the
client waits in dma_fence_default_wait for a fence that will never signal.
Stopping the NULL dereference is what belongs in a fix; the rest wants
more thought.

Analysis

AI-generated

Technical impact

This kernel flaw traps the rocket accelerator subsystem in a state where an inactive core remains bound but its address is omitted from the active search list after an intermediate core is removed. The lookup function then returns a null pointer, and subsequent code paths—such as job execution and timeout handling—dereference this pointer without verification. The faults manifest as kernel memory errors and, if not mitigated, can lead to full system crashes or loss of GPU service. The core problem is a classic null pointer dereference caused by incorrect array bounds handling within the scheduling code.

Affected Systems

The issue resides in the Linux kernel’s rocket accelerator (accel/rocket) module. It applies to any distribution or build that uses the rocket scheduler, including newer platforms such as the RK3588 where the accelerator is active. No specific kernel series or patch level is provided, so all kernels that incorporate the rocket subsystem before the commit that adds the bounds check are potentially vulnerable.

Risk and Exploitability

The kernel panic manifests as a denial‑of‑service. The flaw also causes clients to wait on fences that never complete. No public exploit data exists. The CVSS score of 4.4 indicates moderate severity. The EPSS score of < 1% indicates a low likelihood of exploitation. The vulnerability remains a priority for all affected kernel distributions.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Apply the latest Linux kernel update that includes the fix for the rocket scheduler lookup bug.
  • If an immediate update is not possible, ensure that no GPU clients remain active before disabling or unbinding any rocket accelerator cores; the kernel will otherwise crash.
  • As a temporary protection, disable the rocket scheduling module or the entire rocket accelerator driver so that no job requests are scheduled against the problematic core. Do this by blacklisting the module or using any vendor-provided disable flag.

Generated by OpenCVE AI on November 2, 2026 at 14:09 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
References
Metrics threat_severity

None

cvssV3_1

{'score': 4.4, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:H/UI:N/S:U/C:N/I:N/A:H'}

threat_severity

Moderate

Type Values Removed Values Added
First Time appeared Linux linux
Vendors & Products Linux linux
Type Values Removed Values Added
Weaknesses CWE-362
CWE-476
Type Values Removed Values Added
Description In the Linux kernel, the following vulnerability has been resolved: accel/rocket: search every core slot when looking up a scheduler sched_to_core() walks rdev->cores[] up to rdev->num_cores, and rocket_remove() decrements num_cores for every core it removes. Unbind a core that is not the last one and the cores behind it fall outside the search, so sched_to_core() returns NULL for a core that is still bound and still running jobs. Neither caller checks the result: rocket_job_run(): rocket_fence_create(core), core->dev rocket_job_timedout(): dev_err(core->dev, "NPU job timed out") Unbinding the middle core of the three on an RK3588 while three clients are submitting to all of them faults twice, once from the surviving core's job queue and once from its reset work: KASAN: null-ptr-deref in range [0x0000000000000220-0x0000000000000227] Workqueue: fdad0000.npu drm_sched_run_job_work [gpu_sched] pc : rocket_job_run+0x234/0x838 [rocket] Call trace: rocket_job_run+0x234/0x838 [rocket] drm_sched_run_job_work+0x2cc/0xad8 [gpu_sched] process_one_work+0x640/0x14f0 KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] Workqueue: rocket-reset-2 drm_sched_job_timedout [gpu_sched] pc : rocket_job_timedout+0xf0/0x1e0 [rocket] Call trace: rocket_job_timedout+0xf0/0x1e0 [rocket] drm_sched_job_timedout+0x188/0x6a0 [gpu_sched] Both are the third core: the workqueue names are its device and its core->index, and it was left at slot 2 while num_cores had dropped to 2. Search all the slots that were allocated, the way find_core_for_dev() now does. A core that is still bound is then found, and the two callers get the pointer they already assume they have. This does not make unbinding one core out of several safe. An open client keeps an entity pointing at the scheduler of the core that went away: drm_sched reports it as not ready for every job that lands on it, and the client waits in dma_fence_default_wait for a fence that will never signal. Stopping the NULL dereference is what belongs in a fix; the rest wants more thought.
Title accel/rocket: search every core slot when looking up a scheduler
First Time appeared Linux
Linux linux Kernel
CPEs cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*
Vendors & Products Linux
Linux linux Kernel
References
Weaknesses
  • CWE-362

    Concurrent Execution using Shared Resource with Improper Synchronization ('Race Condition')

  • CWE-476

    NULL Pointer Dereference

Sources

Status
PUBLISHED
Assigner
Linux
Published
Updated
Nov 6, 2026, 5:15 PM
Reserved
Sep 25, 2026, 10:25 AM
Status
Received
Published
Oct 28, 2026, 11:17 AM
Modified
Nov 6, 2026, 6:18 PM
Severity
Moderate
Public date
Oct 28, 2026, 10:35 AM