CVE-2026-43400: Linux Kernel AMDGPU Driver DoS Flaw

CVE-2026-43400 Overview

CVE-2026-43400 is a Linux kernel vulnerability in the AMD GPU driver (drm/amdgpu). The amdgpu_userq_signal_ioctl function fails to validate upper bounds on user-supplied input values. Large input values trigger excessive kernel memory allocation, leading to an out-of-memory (OOM) condition. Local users with access to the AMD GPU device node can exploit this to exhaust kernel memory and cause denial of service. The upstream fix introduces validation against the AMDGPU_USERQ_MAX_HANDLES constant to constrain allocation sizes.

Critical Impact

Local attackers can trigger kernel OOM conditions through unbounded allocation in the amdgpu user queue signal IOCTL, resulting in denial of service on systems with AMD GPUs.

Affected Products

• Linux kernel versions including the vulnerable amdgpu_userq_signal_ioctl implementation
• Systems with AMD GPU hardware using the amdgpu Direct Rendering Manager (DRM) driver
• Distributions shipping kernels prior to the backported fix commits

Discovery Timeline

• 2026-05-08 - CVE-2026-43400 published to the National Vulnerability Database (NVD)
• 2026-05-12 - Last updated in NVD database

Technical Details for CVE-2026-43400

Vulnerability Analysis

The vulnerability resides in the AMD GPU kernel driver's user queue subsystem. The amdgpu_userq_signal_ioctl handler processes user-supplied parameters that drive kernel memory allocation. Without an upper bound check, an attacker can pass arbitrarily large input values that cause the kernel to attempt allocations sized well beyond legitimate use cases.

The resulting allocation either fails outright or consumes enough memory to trigger the kernel OOM killer. Either outcome produces denial of service, with potential collateral termination of unrelated processes when the OOM killer activates. The patch adds a sanity check comparing the input against AMDGPU_USERQ_MAX_HANDLES, which the maintainers describe as sufficient for legitimate workloads.

Root Cause

The root cause is missing input validation [CWE-1284] in an IOCTL handler. The vulnerable code path trusts user-controlled count or size values and feeds them directly into allocation routines. This is a classic resource exhaustion pattern in kernel ioctl interfaces where untrusted userspace input drives kernel-side allocation without sanity bounds.

Attack Vector

Exploitation requires local access to the system and permission to open the /dev/dri/renderD* device node exposed by the amdgpu driver. On most desktop and workstation Linux distributions, members of the video or render group hold this access. An attacker invokes the user queue signal IOCTL with crafted oversized input values, triggering the unbounded allocation path. No special privileges beyond device access are required.

No verified public proof-of-concept exploit is currently available. Technical details are documented in the upstream kernel commits referenced in the Kernel Git Commit Fix.

Detection Methods for CVE-2026-43400

Indicators of Compromise

• Kernel log entries showing OOM killer activation on systems with AMD GPUs and no corresponding legitimate memory pressure workload
• Repeated process terminations by oom-killer correlated with non-privileged user activity
• Unexpected amdgpu driver allocation failures or warnings in dmesg

Detection Strategies

• Audit running kernel versions against the patched commits 46630d966b99, 6fff5204d8aa, and ea78f8c68f4f to identify unpatched hosts
• Monitor /proc/meminfo and slab allocator metrics for abnormal kernel memory consumption tied to GPU driver activity
• Correlate IOCTL syscall telemetry against /dev/dri/renderD* device descriptors to identify unusual access patterns from non-graphical processes

Monitoring Recommendations

• Enable kernel auditing of IOCTL operations on DRM render nodes for high-value or multi-tenant hosts
• Track OOM killer events through journalctl -k and forward to centralized logging for anomaly detection
• Inventory hosts with AMD GPU hardware to scope exposure and prioritize patching across desktop, workstation, and GPU compute fleets

How to Mitigate CVE-2026-43400

Immediate Actions Required

• Update affected Linux kernels to a version containing the upstream fix or the stable backports referenced in the kernel.org commits
• Restrict access to /dev/dri/renderD* device nodes by removing untrusted users from the render and video groups
• On multi-tenant or shared compute hosts, prioritize patching as local users may trigger denial of service against co-tenants

Patch Information

The fix adds an upper bound check against AMDGPU_USERQ_MAX_HANDLES in amdgpu_userq_signal_ioctl. Apply the upstream patches from the Kernel Git Commit Fix, Kernel Git Commit Update, and Kernel Git Commit Change. Distribution vendors will ship the fix through their standard stable kernel update channels.

Workarounds

• Where the amdgpu driver is not required, blacklist the module to remove the attack surface entirely
• Apply Linux control groups (cgroups v2) memory limits to user sessions to contain the blast radius of OOM conditions
• Restrict device node permissions using udev rules so only trusted system users can open AMD GPU render nodes

# Example udev rule restricting amdgpu render node access
# Place in /etc/udev/rules.d/99-amdgpu-restrict.rules
SUBSYSTEM=="drm", KERNEL=="renderD*", GROUP="render", MODE="0660"

# Verify installed kernel includes the fix commits
zgrep -i "amdgpu_userq_signal_ioctl" /proc/config.gz
uname -r