CVE-2026-53171: Linux Kernel Ethos-U NPU Driver Integer Arithmetic Vulnerability
A Linux kernel vulnerability in the Arm Ethos-U NPU driver's DMA length calculation function allows arithmetic wraparound to corrupt internal memory region size tracking. This corruption bypasses security checks that prevent invalid memory access, potentially enabling privilege escalation or system compromise on systems running vulnerable kernel versions with the Ethos-U accelerator driver loaded.
Source data · NVD / CISA · public domain
- CVSS
- 3.1 · 8.8 HIGH · CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H
- Weaknesses (CWE)
- —
- 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: fix arithmetic issues in dma_length() dma_length() derives DMA region usage from command stream values and updates region_size[]: len = ((len + stride[0]) * size0 + stride[1]) * size1 region_size[region] = max(..., len + dma->offset) Several arithmetic issues can corrupt the derived region size: - signed stride values may underflow when added to len - intermediate multiplications may overflow - len + dma->offset may overflow during region_size updates - dma_length() error returns were not validated by the caller region_size[] is later used by ethosu_job.c to validate command stream accesses against GEM buffer sizes. Arithmetic wraparound can therefore under-report region usage and bypass the bounds validation. Fix by validating signed additions, using overflow helpers for multiplications and offset updates, and propagating dma_length() failures to the caller.
2 reference(s) · View on NVD →
SEC.co analysis · AI-assisted, reviewed against source
Technical summary
CVE-2026-53171 affects the dma_length() function in accel/ethosu, which derives DMA region usage from command stream parameters. The function compounds four arithmetic issues: (1) signed stride values can underflow when added to the length variable, (2) intermediate multiplications lack overflow detection, (3) the final region_size update (len + dma->offset) can overflow, and (4) error returns from dma_length() were previously unvalidated by callers. These issues cause region_size[] array entries to reflect incorrect bounds, which are subsequently used by ethosu_job.c to validate command stream access against GEM buffer allocations. Arithmetic wraparound allows actual memory usage to exceed the recorded region_size, defeating bounds validation and enabling out-of-bounds access.
Business impact
Organizations deploying Linux-based systems with Arm Ethos-U NPU accelerators (common in AI/ML inference platforms, automotive edge devices, and mobile infrastructure) face elevated risk of kernel privilege escalation. A local attacker with unprivileged access can craft malicious command streams to trigger arithmetic wraparound, corrupt region tracking, and access memory outside allocated buffers. Successful exploitation may lead to information disclosure (reading kernel memory), data modification, or denial of service. The impact scales with deployment breadth—particularly critical for service providers offering shared compute resources.
Affected systems
The Linux kernel's Arm Ethos-U NPU accelerator driver (accel/ethosu) is affected. Vulnerable systems include Linux distributions deployed on platforms with integrated or discrete Ethos-U accelerators, such as Arm-based SoCs used in edge AI appliances, automotive compute modules, and some mobile devices. The vulnerability requires local code execution context; remote exploitation is not possible. Systems without the Ethos-U driver compiled or loaded are unaffected. Verify your kernel configuration and loaded modules to determine exposure.
Exploitability
Exploitation requires local unprivileged code execution on the target system. An attacker must craft a malicious Ethos-U command stream (typically via ioctl or similar driver interface) with specific stride and size parameters that trigger signed integer underflow, multiplication overflow, or offset overflow in dma_length(). The arithmetic conditions needed are deterministic but require precise calculation of parameters based on GEM buffer layout. No known public exploits exist, and the vulnerability is not tracked on the CISA KEV catalog. The technical complexity is moderate—a knowledgeable attacker with local access and knowledge of the driver's internals can likely succeed.
Remediation
Apply Linux kernel patches that add signed addition validation, use kernel overflow-checking helpers (e.g., check_add_overflow, check_mul_overflow) for intermediate calculations, and enforce validation of dma_length() return codes before consuming results. The fix ensures that any arithmetic anomaly is detected and propagated as an error rather than silently corrupting region_size[] tracking. Verify your kernel version against vendor advisories for backport availability to stable release branches.
Patch guidance
Update your Linux kernel to a version that includes the arithmetic validation fixes for accel/ethosu/dma_length(). Check with your distribution (Red Hat, Canonical, SUSE, etc.) for backported patches to in-service kernel versions; the fix is likely to appear in 6.10-rc or later mainline before final release, then backported to stable branches (6.9.z, 6.8.z, etc.). If you maintain a custom kernel build, apply the upstream patch directly from the Linux kernel mailing list or git repository. If the Ethos-U driver is not compiled into your kernel, you can safely disable it during build if it is not a required component for your platform.
Detection guidance
Monitor kernel logs for any error messages or warnings from accel/ethosu driver during command stream processing; a patched kernel may now reject previously accepted (but malformed) streams. Review kernel dmesg output for failed region_size validation. On vulnerable systems, enable audit rules to log all ioctl calls to Ethos-U device nodes (/dev/ethosu*) to identify suspicious command stream injection attempts. Inspect any system with Ethos-U accelerators in your inventory for kernel version and patch status. Network-based detection is not applicable since exploitation requires local code execution.
Why prioritize this
This vulnerability merits prompt patching because (1) it affects a security-critical function (memory bounds validation), (2) exploitation is locally reachable by unprivileged users, (3) successful exploitation can escalate privileges or read kernel memory, and (4) Ethos-U deployments are expanding in edge and automotive sectors where systems may be less frequently patched. The HIGH CVSS score (8.8) reflects the combination of local access requirement, high impact across confidentiality, integrity, and availability, and the lack of required user interaction. Organizations should prioritize Ethos-U-equipped systems in patch scheduling, especially in multi-tenant or high-availability environments.
Risk score, explained
The CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H score of 8.8 (HIGH) reflects: local attack vector (AV:L), low attack complexity requiring no special conditions (AC:L), low privileges required (PR:L) allowing unprivileged local users, no user interaction needed (UI:N), scope changed (S:C) allowing impact beyond the vulnerable component to other kernel subsystems or processes, and high impact on confidentiality, integrity, and availability (C:H, I:H, A:H) via memory corruption. The primary constraint is the requirement for local code execution; remote exploitation is not possible.
Frequently asked questions
Do I need to patch if I don't use Ethos-U accelerators?
No. If your kernel does not include accel/ethosu drivers or if the driver is not loaded, this vulnerability does not affect you. Check your kernel config and lsmod output to confirm. Systems without Ethos-U hardware or with the driver disabled are unaffected.
Can this vulnerability be exploited remotely?
No. The vulnerability requires local unprivileged code execution to craft malicious command streams and trigger the arithmetic issues. Remote exploitation is not possible. However, any system that allows unprivileged local access (shared hosting, multi-user systems, compromised applications) is at risk.
What is the difference between this and a buffer overflow?
This is an integer arithmetic vulnerability that corrupts security-critical metadata (region_size tracking), rather than a direct buffer overflow. The wraparound causes the kernel's bounds checker to underestimate memory usage, allowing out-of-bounds access. It is more subtle and harder to detect than a classical buffer overflow, but the impact is similar: unauthorized memory access.
Will patching break my existing applications?
Unlikely. The patch adds validation and error checking; legitimate command streams that previously succeeded will continue to work. Only malformed or malicious streams that exploited the arithmetic issues will now be rejected with an error. Well-behaved applications should see no change in behavior.
This analysis is provided for informational purposes and represents a best-effort assessment based on publicly available vulnerability data. Actual patch availability, affected kernel versions, and remediation timelines vary by Linux distribution and are subject to change. Organizations should consult official vendor security advisories (Red Hat Security Advisories, Ubuntu Security Notices, etc.) for definitive patch status and testing guidance. This document does not constitute professional security advice. Test all patches in non-production environments before deployment. SEC.co makes no warranty regarding the accuracy or completeness of this intelligence. Source: NVD (public-domain), retrieved 2026-08-03. Analysis generated by SEC.co (claude-haiku-4-5).
Affected vendors
Related vulnerabilities
- CVE-2026-0270HIGHCortex XSOAR Path Traversal on Linux — Exploit Requirements & Patching Guide
- CVE-2026-0271HIGHPalo Alto Networks Prisma Access Agent Linux Privilege Escalation
- CVE-2026-10001HIGHChrome Sandbox Escape via PerformanceManager Use-After-Free
- CVE-2026-10002HIGHGoogle Chrome PDFium Use-After-Free Vulnerability (CVSS 8.8)
- CVE-2026-10003HIGHChrome Use-After-Free Code Execution Vulnerability Analysis
- CVE-2026-10006HIGHChrome WebAudio Race Condition Remote Code Execution
- CVE-2026-10007HIGHChrome Use-After-Free in SVG Arbitrary Code Execution (CVSS 8.8)
- CVE-2026-10009HIGHChrome Skia Integer Overflow Sandbox Escape – Patch Guidance