CVE-2025-6349: Arm 5th Gen GPU Driver Use After Free Flaw

CVE-2025-6349 Overview

CVE-2025-6349 is a Use After Free (UAF) vulnerability affecting Arm Ltd's Valhall GPU Kernel Driver and Arm 5th Gen GPU Architecture Kernel Driver. This memory corruption flaw allows a local non-privileged user process to perform improper GPU memory processing operations, enabling unauthorized access to already freed memory regions. The vulnerability poses a risk to confidentiality and integrity of affected systems, though exploitation requires local access.

Critical Impact

Local attackers can exploit improper GPU memory processing to access freed memory, potentially leading to information disclosure or memory corruption on devices using Arm Mali GPUs.

Affected Products

• Arm Valhall GPU Kernel Driver versions r53p0 through r54p1
• Arm 5th Gen GPU Architecture Kernel Driver versions r53p0 through r54p1
• Devices utilizing Mali GPUs with vulnerable kernel driver versions

Discovery Timeline

• December 1, 2025 - CVE-2025-6349 published to NVD
• December 2, 2025 - Last updated in NVD database

Technical Details for CVE-2025-6349

Vulnerability Analysis

This vulnerability is classified as CWE-416 (Use After Free) with a CVSS 3.1 score of 5.1 (Medium severity). The CVSS vector CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:N indicates:

The EPSS (Exploit Prediction Scoring System) score is 0.016% with a percentile of 2.563, indicating a relatively low probability of active exploitation in the wild.

Root Cause

The vulnerability stems from improper memory management within the GPU kernel drivers. Specifically, the drivers fail to properly track and validate memory allocations during GPU memory processing operations. When memory is freed, references to that memory region may persist, allowing subsequent operations to access the deallocated memory space. This classic Use After Free condition occurs when the driver continues to reference memory objects after they have been released back to the system's memory pool.

Attack Vector

The attack vector is local, requiring an attacker to have execution capabilities on the target system. A malicious local process running without elevated privileges can trigger the vulnerability by performing specific sequences of GPU memory operations that exploit the race condition or improper reference handling in the driver. The attacker can craft memory allocation and deallocation patterns that leave dangling pointers, which can then be dereferenced to access freed memory containing sensitive data or to corrupt memory structures.

The exploitation scenario typically involves:

1. Allocating GPU memory buffers through the kernel driver interface
2. Triggering the free operation while maintaining implicit references
3. Causing the driver to access the freed memory through these stale references
4. Reading potentially sensitive data from the freed memory region or corrupting subsequent allocations

Detection Methods for CVE-2025-6349

Indicators of Compromise

• Unexpected GPU driver crashes or kernel panics related to Mali GPU operations
• Anomalous memory access patterns in GPU-related system calls
• Unusual process behavior involving repeated GPU memory allocation/deallocation cycles
• System log entries indicating GPU memory corruption or invalid memory accesses

Detection Strategies

Organizations should implement kernel-level monitoring to detect exploitation attempts:

1. Kernel Audit Logging: Enable comprehensive auditing of GPU driver interactions, particularly memory allocation and deallocation operations through the Mali driver interfaces.

2. Memory Corruption Detection: Deploy kernel hardening features such as KASAN (Kernel Address Sanitizer) in development environments to detect use-after-free conditions.

3. Behavioral Analysis: Monitor for processes that exhibit unusual patterns of GPU memory operations, particularly rapid allocation/free cycles that could indicate exploitation attempts.

4. SentinelOne Protection: SentinelOne's Singularity Platform provides real-time kernel-level visibility and can detect anomalous driver behavior patterns indicative of memory corruption exploitation attempts.

Monitoring Recommendations

Security teams should prioritize monitoring devices with Mali GPUs, particularly mobile devices and embedded systems running Android or Linux. Key monitoring points include:

• GPU driver initialization and memory management logs
• System calls related to the Mali GPU kernel interface (/dev/mali*)
• Kernel messages indicating memory access violations in GPU driver context
• Process behavior analysis for applications interacting heavily with GPU resources

How to Mitigate CVE-2025-6349

Immediate Actions Required

• Update Arm GPU kernel drivers to versions newer than r54p1 when available
• Review Arm's security advisory at the vendor documentation portal
• Implement application sandboxing to limit untrusted code's GPU access
• Enable additional kernel hardening features where supported

Patch Information

Arm has published a security advisory addressing this vulnerability. Organizations should consult the official vendor advisory for detailed patch information:

• Vendor Advisory: https://developer.arm.com/documentation/110697/latest/

The vulnerable driver versions span from r53p0 through r54p1 for both the Valhall GPU Kernel Driver and the Arm 5th Gen GPU Architecture Kernel Driver. Device manufacturers and OEMs should work with Arm to obtain updated driver versions and push updates to affected devices.

Workarounds

If immediate patching is not possible, consider the following temporary mitigations:

• Restrict GPU Access: Limit GPU driver access to trusted applications only using SELinux or AppArmor policies
• Application Isolation: Utilize containerization or sandboxing for untrusted applications to prevent direct GPU driver interaction
• Access Control: Implement strict file permission controls on GPU device nodes to limit which processes can interact with the driver

Device administrators should restrict access to the Mali GPU device interfaces by adjusting permissions:

# Restrict GPU device access to specific groups
chmod 660 /dev/mali*
chown root:gpu /dev/mali*

# Add only trusted users to the gpu group
usermod -aG gpu trusted_user

For Android devices, work with device manufacturers to obtain OTA updates containing patched GPU drivers. Enterprise mobile device management (MDM) solutions can be used to enforce update policies and restrict vulnerable devices from accessing sensitive resources until patched.