CVE-2026-23264: Linux Kernel AMD GPU ASPM DOS Vulnerability

CVE-2026-23264 Overview

A regression vulnerability has been identified in the Linux kernel's AMD GPU driver (drm/amd) related to Active State Power Management (ASPM) handling. The vulnerability was introduced when a previously reverted commit was erroneously re-applied after refactoring work on the amdgpu_aspm functionality. This results in hard-to-debug system crashes on systems configured with two AMD GPUs where only one supports ASPM.

Critical Impact
Systems with dual AMD GPU configurations where ASPM support differs between cards may experience unpredictable kernel crashes, leading to system instability and potential denial of service.

Affected Products

Linux kernel with AMD GPU driver (drm/amd)
Systems with dual AMD GPU configurations
Kernel versions containing the erroneously re-applied commit 7294863a6f01248d72b61d38478978d638641bee

Discovery Timeline

2026-03-18 - CVE CVE-2026-23264 published to NVD
2026-03-19 - Last updated in NVD database

Technical Details for CVE-2026-23264

Vulnerability Analysis

This vulnerability stems from a regression introduced through improper commit management in the Linux kernel's AMD GPU driver subsystem. The problematic commit 7294863a6f01248d72b61d38478978d638641bee was originally applied to check if ASPM is enabled from the PCIe subsystem. However, this commit was later removed by commit 0ab5d711ec74 which refactored amdgpu_aspm to be evaluated on a per-device basis.

The issue arose when the original commit was erroneously reintroduced after the refactoring, creating a conflict in how ASPM states are evaluated across multiple GPUs. When a system contains two AMD GPUs with differing ASPM support capabilities, the conflicting logic causes the kernel to improperly handle power state transitions, resulting in system crashes that are difficult to diagnose due to the race-condition-like nature of the failure.

Root Cause

The root cause is a configuration regression introduced by re-applying a commit that had been intentionally removed during refactoring. The original commit assumed a global ASPM state check, which conflicts with the per-device evaluation model introduced by the amdgpu_aspm refactoring. When one GPU supports ASPM and another does not, the conflicting evaluation logic leads to undefined behavior in the PCIe power management subsystem.

Attack Vector

This is a local denial of service vulnerability affecting system stability. The vulnerability is triggered through normal system operation when:

A system has two or more AMD GPUs installed
The GPUs have differing ASPM support capabilities
The kernel attempts to evaluate ASPM states during power management operations

The attack vector is local, requiring physical access to hardware configuration or the ability to influence driver loading and GPU enumeration. Exploitation does not require user interaction beyond normal system operation.

Detection Methods for CVE-2026-23264

Indicators of Compromise

Unexpected kernel crashes or system hangs on multi-GPU AMD systems
Kernel panic messages referencing drm/amd or amdgpu driver components
System instability that correlates with PCIe power state transitions
Log entries indicating ASPM-related failures in dmesg output

Detection Strategies

Monitor kernel logs for crashes related to drm/amd, amdgpu, or PCIe subsystem errors
Check kernel version and verify if the problematic commit 7294863a6f01248d72b61d38478978d638641bee is present
Review system configuration for dual AMD GPU setups with heterogeneous ASPM support
Implement automated kernel crash analysis to identify patterns consistent with this regression

Monitoring Recommendations

Enable kernel crash dump collection for forensic analysis of system crashes
Monitor dmesg output for ASPM-related warnings or errors
Track system uptime metrics for unexpected restarts on affected hardware configurations
Set up alerts for kernel panic events on systems with dual AMD GPUs

How to Mitigate CVE-2026-23264

Immediate Actions Required

Identify systems with dual AMD GPU configurations where ASPM support differs between cards
Review current kernel version against available patches
Consider temporarily disabling ASPM via kernel boot parameters as a workaround
Apply the revert patches from the stable kernel commits when available

Patch Information

Multiple patch commits have been released to address this vulnerability by reverting the erroneously re-applied commit. The fixes are available in the stable kernel branches:

The patch cherry-picks commit 97a9689300eb2b393ba5efc17c8e5db835917080 which properly reverts the problematic changes.

Workarounds

Disable ASPM globally via kernel boot parameter pcie_aspm=off
Use kernel module parameter to disable ASPM for the amdgpu driver specifically
Temporarily remove one GPU from multi-GPU systems until patches can be applied
Configure system to use only one AMD GPU until the kernel is updated

bash

# Configuration example
# Disable ASPM via GRUB boot parameters
# Edit /etc/default/grub and add to GRUB_CMDLINE_LINUX:
GRUB_CMDLINE_LINUX="pcie_aspm=off"

# Then update GRUB configuration:
sudo update-grub

# Alternatively, disable at runtime (temporary):
echo 0 | sudo tee /sys/module/pcie_aspm/parameters/policy