CVE-2026-53169: Linux Ethos-U NPU Driver DoS via Unimplemented Command
CVE-2026-53169 is a denial-of-service vulnerability in the Linux kernel's Ethos-U NPU driver. When userspace sends an unsupported NPU_OP_RESIZE command through the DRM interface, the driver's placeholder code unconditionally triggers a kernel warning. On systems configured to panic on warnings, any user with access to the DRM device can crash the kernel. The fix replaces the warning trigger with a proper error rejection, preventing both the spam and potential panic.
Source data · NVD / CISA · public domain
- CVSS
- 3.1 · 5.5 MEDIUM · CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H
- Weaknesses (CWE)
- CWE-617
- Affected products
- 7 configuration(s)
- Published / Modified
- 2026-06-25 / 2026-07-06
NVD description (verbatim)
In the Linux kernel, the following vulnerability has been resolved: accel/ethosu: reject NPU_OP_RESIZE commands from userspace NPU_OP_RESIZE is a U85-only command that the driver does not yet implement. The existing WARN_ON(1) placeholder fires unconditionally whenever userspace submits this command via DRM_IOCTL_ETHOSU_GEM_CREATE, causing unbounded kernel log spam. If panic_on_warn is set the kernel panics, giving any unprivileged user with access to the DRM device a trivial denial-of-service primitive. Replace the WARN_ON(1) with an explicit -EINVAL return so the ioctl rejects the command before it reaches hardware.
2 reference(s) · View on NVD →
SEC.co analysis · AI-assisted, reviewed against source
Technical summary
The accel/ethosu driver contains a WARN_ON(1) placeholder in the NPU_OP_RESIZE command path. This command is U85-specific and not yet implemented. When userspace submits the command via DRM_IOCTL_ETHOSU_GEM_CREATE ioctl, the warning fires unconditionally, filling kernel logs and—if panic_on_warn is enabled—panicking the kernel. The vulnerability stems from using a debug assertion as a temporary safeguard rather than returning a proper error code. The resolution replaces WARN_ON(1) with explicit -EINVAL validation, rejecting the unsupported command before it reaches hardware.
Business impact
Systems running affected Linux kernels with panic_on_warn enabled face elevated availability risk. Any unprivileged user with DRM device access can trigger kernel panics, disrupting services and user sessions. Organizations running machine-learning or AI workloads on Ethos-U hardware (common in embedded and edge deployments) face immediate DoS exposure. Non-panicking systems experience degraded observability through unbounded kernel log spam, potentially masking genuine issues.
Affected systems
The vulnerability affects Linux kernel versions containing the accel/ethosu driver code. The impact is most acute in systems with panic_on_warn configured and those granting unprivileged users access to DRM devices—typical in containers, multi-user workstations, and embedded systems running Ethos-U accelerators. The actual attack surface depends on kernel build configuration and DRM device permissions in your deployment.
Exploitability
Exploitation is straightforward: an unprivileged user with DRM device access (often the default) invokes a single ioctl with an NPU_OP_RESIZE command. On panic_on_warn systems, this immediately crashes the kernel with no additional privileges required. Even without panic_on_warn, repeated invocation floods kernel logs. No special tools or race conditions are necessary—this is a direct, reliable trigger.
Remediation
Update to a Linux kernel version containing the fix that replaces the WARN_ON(1) placeholder with -EINVAL validation. Verify the specific version against your distribution's advisory. As an interim control, restrict DRM device access to trusted users only; limit access via device permissions, SELinux, or AppArmor rules. Disable panic_on_warn if your security posture permits, reducing the severity to log spam rather than availability loss.
Patch guidance
Consult your Linux distribution (Red Hat, Debian, Ubuntu, SUSE, etc.) for kernel updates addressing CVE-2026-53169. Patches will vary by distribution version and release. Apply updates following your organization's kernel patching cycle. For custom or embedded kernels, cherry-pick the fix from upstream Linux or backport from the distribution patch. Test in non-production environments first, as kernel updates may require reboot.
Detection guidance
Monitor for repeated DRM_IOCTL_ETHOSU_GEM_CREATE ioctl invocations with NPU_OP_RESIZE commands from non-privileged processes. In kernel logs, look for WARN messages from the ethosu driver (search for 'ethosu' and 'WARN'). On panic_on_warn systems, kernel panics traceable to the ethosu driver indicate exploitation. Audit DRM device permissions using 'ls -l /dev/dri/*' and restrict as needed. Enable comprehensive DRM/GPU auditing if your organization deploys Ethos-U accelerators.
Why prioritize this
Although CVSS 5.5 is medium severity, the ease of exploitation—a simple ioctl from an unprivileged user—and the availability impact on panic_on_warn systems elevate practical risk. This is especially critical for organizations deploying Ethos-U accelerators in multi-tenant or untrusted-user environments. Prioritize patching systems with panic_on_warn enabled or those granting broad DRM access.
Risk score, explained
CVSS 3.1 assigns 5.5 (MEDIUM) due to local attack vector, low privilege requirement, no complexity, and high availability impact but no confidentiality or integrity harm. The score appropriately reflects that unprivileged users can cause denial-of-service. The lack of KEV designation indicates no known active exploitation, though the simplicity of the attack vector suggests risk is elevated in real-world deployments.
Frequently asked questions
Who can exploit this vulnerability?
Any user with read-write access to DRM devices (typically /dev/dri/renderD128 or similar). In multi-user systems, unprivileged local users often have default access. Containers and embedded systems are common targets.
What happens if panic_on_warn is not enabled?
The kernel will not panic, but the ioctl succeeds and fills kernel logs with warnings. Repeated exploitation degrades observability and system performance through log spam, but does not crash the kernel directly.
Do I need to reboot after applying the patch?
Yes. Like most kernel security updates, you will need to reboot to run the patched kernel. Coordinate reboots with your change management process to minimize service disruption.
Is this relevant to non-Ethos-U systems?
Only if you are running a kernel version containing the accel/ethosu driver and have enabled that driver. Most general-purpose distributions include it as a module, but it is inactive unless you use Ethos-U hardware or explicitly load the module.
This analysis is based on the published CVE description and available metadata. Specific patch versions, distribution advisories, and deployment impact depend on your environment. Always verify patch applicability against your Linux distribution's official security guidance before deploying. The vulnerability requires local access and is not exploitable remotely. This document does not constitute legal or compliance advice. Source: NVD (public-domain), retrieved 2026-08-03. Analysis generated by SEC.co (claude-haiku-4-5).
Related vulnerabilities
- CVE-2026-46220MEDIUMLinux AMDGPU Kernel Panic DoS Vulnerability
- CVE-2026-46287MEDIUMLinux txgbe Driver RTNL Locking Defect
- CVE-2026-52961MEDIUMLinux Kernel Ceph Filesystem Race Condition Causing Kernel Panic
- CVE-2026-53039MEDIUMLinux OCFS2 Local Denial of Service via Unvalidated Group Add
- CVE-2026-46117HIGHLinux RDMA/mana Kernel Privilege Escalation via Invalid Queue Pair Configuration
- CVE-2026-52952HIGHLinux Kernel IOMMU Use-After-Free in Domain Attachment
- CVE-2026-52954HIGHLinux Kernel Ceph Client Denial of Service via Malformed OSD Map
- CVE-2026-46542MEDIUMNimiq Ed25519 Denial-of-Service Vulnerability (Version 1.4.0 Patch)