CVE-2026-43195 Overview
CVE-2026-43195 affects the Linux kernel's AMD GPU driver (drm/amdgpu). The vulnerability stems from missing validation of user queue size constraints when allocating GPU ring buffers. The driver did not enforce that user-supplied queue sizes were a power of two or at least AMDGPU_GPU_PAGE_SIZE. Invalid configurations can produce undersized allocations and incorrect ring buffer wrapping. The upstream fix adds explicit size validation before the kernel accepts a user queue request.
Critical Impact
Invalid user queue sizes accepted by the amdgpu driver can trigger GPU faults or unexpected driver behavior on affected Linux systems.
Affected Products
- Linux kernel drm/amdgpu driver
- Systems using AMD GPUs with user queue submission paths
- Distributions shipping the unpatched kernel versions referenced in the upstream commits
Discovery Timeline
- 2026-05-06 - CVE-2026-43195 published to NVD
- 2026-05-06 - Last updated in NVD database
Technical Details for CVE-2026-43195
Vulnerability Analysis
The AMD GPU kernel driver exposes user queue creation interfaces that accept a caller-specified queue size. Hardware ring buffer logic depends on power-of-two sizes for efficient wrap-around using bitmask arithmetic. The pre-patch driver did not reject non-power-of-two values or sizes smaller than AMDGPU_GPU_PAGE_SIZE. Accepting such values leads to allocations that do not satisfy the hardware's ring buffer assumptions. The result is GPU faults, driver instability, or undefined queue behavior on the affected device.
Root Cause
The root cause is missing input validation in the user queue creation path of drm/amdgpu. The driver trusted the caller-supplied size without checking two required invariants: that the size is a power of two and that it meets the minimum GPU page size. The fix introduces explicit checks that reject sizes failing either constraint before any allocation proceeds.
Attack Vector
The attack surface is local to a host with an AMD GPU and a user account permitted to issue amdgpu user queue ioctls. A local user supplies a malformed queue size to the driver, which previously accepted the value and allocated a non-conforming ring buffer. The resulting GPU fault path or undersized allocation produces driver-level misbehavior. The vulnerability does not describe a remote vector. Exact impact and privilege requirements are not enumerated in the published advisory.
No verified public exploit code is available. See the upstream commits referenced below for the exact validation logic added by the maintainers.
Detection Methods for CVE-2026-43195
Indicators of Compromise
- Kernel log entries from the amdgpu driver reporting GPU faults, ring buffer errors, or queue allocation failures correlated with userspace processes issuing queue creation ioctls.
- Unexpected GPU resets or hangs on systems running unpatched AMD GPU kernel modules.
Detection Strategies
- Inventory running kernel versions and compare against the patched commits 8079b87c02e5, 9f6cc309cd15, and cf2a37be899d on git.kernel.org.
- Monitor dmesg and journald for amdgpu subsystem warnings tied to user queue creation paths.
- Audit local processes that interact with /dev/dri/renderD* for anomalous ioctl patterns.
Monitoring Recommendations
- Forward kernel logs to a central log platform and alert on repeated amdgpu ring or queue errors.
- Track GPU reset counters exposed via sysfs to identify hosts with recurring driver instability.
- Correlate GPU fault events with the userspace process tree to identify the originating workload.
How to Mitigate CVE-2026-43195
Immediate Actions Required
- Apply the upstream Linux kernel updates that include the drm/amdgpu user queue size validation fix.
- Identify hosts running AMD GPUs and confirm whether their kernel includes one of the fixed commits.
- Restrict access to GPU device nodes to trusted users where feasible.
Patch Information
The fix is published in the upstream Linux kernel as three stable commits that add power-of-two and minimum-size validation to user queue creation in drm/amdgpu. See the Kernel Git Commit 8079b87c, Kernel Git Commit 9f6cc309, and Kernel Git Commit cf2a37be. Rebuild and reboot into a kernel that contains these changes.
Workarounds
- Limit access to /dev/dri/renderD* and /dev/kfd to trusted local users until patched kernels are deployed.
- Avoid running untrusted GPU compute workloads on hosts with unpatched amdgpu drivers.
- Where possible, disable user queue submission features on affected systems pending the kernel update.
# Verify the running kernel and amdgpu module version
uname -r
modinfo amdgpu | grep -E '^(version|srcversion|filename):'
# Restrict DRI render node access to a trusted group
sudo groupadd -f gpu-trusted
sudo chgrp gpu-trusted /dev/dri/renderD*
sudo chmod 0660 /dev/dri/renderD*
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
