CVE-2026-68110: Linux Kernel AMDGPU SDMA Vulnerability

CVE-2026-68110 Overview

CVE-2026-68110 affects the Linux kernel's AMD GPU (amdgpu) driver, specifically the SDMA (System DMA) 4.4.2 component. The vulnerability stems from the use of BUG_ON() macros in error handling paths within the drm/amdgpu/sdma4.4.2 code. When triggered, these macros cause an unrecoverable kernel panic rather than gracefully handling the error condition. The upstream fix replaces BUG_ON() calls with WARN_ON(), which logs the condition without crashing the kernel. This vulnerability was resolved as a cherry-picked commit from the mainline Linux kernel tree.

Critical Impact
Local conditions on systems with AMD GPUs using the SDMA 4.4.2 IP block can trigger a kernel panic, resulting in a denial-of-service condition affecting system availability.

Affected Products

Linux kernel versions containing the drm/amdgpu/sdma4.4.2 driver prior to the fix
Systems using AMD GPU hardware with SDMA 4.4.2 IP blocks
Multiple stable kernel branches receiving the cherry-picked backport

Discovery Timeline

2026-08-10 - CVE-2026-68110 published to NVD
2026-08-13 - Last updated in NVD database

Technical Details for CVE-2026-68110

Vulnerability Analysis

The vulnerability resides in the AMD GPU driver's SDMA 4.4.2 implementation within the Direct Rendering Manager (DRM) subsystem. The affected code used BUG_ON() assertions to handle unexpected conditions during SDMA operations. When a BUG_ON() condition evaluates true, the kernel invokes BUG(), which triggers a kernel panic on most configurations. This behavior is disproportionate for conditions that the driver can recover from or safely ignore. The kernel maintainers determined that these code paths did not warrant crashing the entire system. The fix replaces BUG_ON() with WARN_ON(), which emits a warning and stack trace to the kernel log without halting execution.

Root Cause

The root cause is improper error handling in the amdgpu SDMA 4.4.2 driver code. Developers used BUG_ON() for defensive assertions in situations where continued execution was safe. This falls under [CWE-476]-adjacent defensive programming issues, where a recoverable condition triggers a fatal fault. The kernel community's coding guidelines discourage BUG_ON() outside of situations where continued execution would cause memory corruption or security issues.

Attack Vector

The attack vector requires local conditions that exercise the specific code paths guarded by the BUG_ON() assertions in the SDMA 4.4.2 driver. Any workload interacting with AMD GPU hardware through the DRM interface could potentially trigger these paths under abnormal conditions. Successful exploitation results in a kernel panic, forcing an unplanned system reboot. Detailed exploitation code is not published in the referenced advisories.

See the upstream commits referenced in the Kernel Git Commit dc3f5da1 advisory for the specific code changes.

Detection Methods for CVE-2026-68110

Indicators of Compromise

Unexpected kernel panics referencing amdgpu or sdma_v4_4_2 functions in crash logs
dmesg output showing BUG: traces originating from the SDMA 4.4.2 driver
Sudden system reboots on hosts equipped with AMD GPU hardware using SDMA 4.4.2 IP blocks

Detection Strategies

Monitor kernel ring buffer output via journalctl -k for panic messages tied to the amdgpu module
Collect and inspect crash dumps using kdump and crash utilities for stack frames in SDMA 4.4.2 code
Correlate host reboot events with GPU workload activity across managed Linux fleets

Monitoring Recommendations

Track kernel version inventory across Linux endpoints to identify hosts running unpatched builds
Alert on repeated crashes originating from the same kernel module to identify systemic issues
Use Linux endpoint telemetry to correlate GPU driver events with system availability metrics

How to Mitigate CVE-2026-68110

Immediate Actions Required

Apply the latest stable kernel updates from your Linux distribution that include the cherry-picked SDMA 4.4.2 fix
Inventory systems running AMD GPU workloads to prioritize patching where SDMA 4.4.2 hardware is present
Enable kdump collection so any recurrence is captured for root-cause analysis

Patch Information

The fix was cherry-picked from mainline commit fa4f86a148271e325e95287630a3a15a9cd35fdc. Stable branch backports are available in the following commits: Kernel Git Commit 256d6f48, Kernel Git Commit 40cdbe9f, Kernel Git Commit ca50e541, Kernel Git Commit cbe3b293, and Kernel Git Commit dc3f5da1. Update to a distribution kernel that incorporates one of these backports.

Workarounds

No official workaround is documented; apply the vendor kernel update as soon as available
Where patching is delayed, limit untrusted local workloads that exercise GPU-intensive operations on affected systems
Consider disabling non-essential AMD GPU workloads on production hosts until the patched kernel is deployed

bash

# Verify current kernel version and check for the amdgpu module
uname -r
lsmod | grep amdgpu

# Update to a patched kernel (example for Debian/Ubuntu)
sudo apt update && sudo apt install --only-upgrade linux-image-generic
sudo reboot