CVE-2025-38011

What it actually does

The flaw permits an interruptible wait on the GPU VM lock during CSA unmap and VM teardown on process exit in the amdgpu driver, allowing signals to abort the operation and leave memory allocations unreleased, producing the amdgpu_driver_postclose_kms backtrace.

Why it matters to the business

Linux systems with AMD GPUs may experience gradual memory exhaustion or repeated kernel warnings after graphics process termination, increasing operational overhead and risk of instability on servers or workstations with frequent process churn.

How an attack unfolds

A local user starts a graphics process that allocates AMD GPU VM resources, then exits while delivering signals that interrupt the vm lock acquisition in the unmap path, triggering the memory leak condition before the uninterruptible fix was applied.

How to fix it

Upgrade the Linux kernel to a version containing commit 8d2979b9bb1be0f4a52dff600e56d780403e04ac or later stable releases; apply the Mageia imagemagick update to 6.8.9.9-4.2.mga5 where the package is present.

How to confirm the fix worked

Confirm the running kernel contains the referenced commit; reproduce process exit under signal load and verify absence of the amdgpu_driver_postclose_kms warning and associated memory growth.

Typical attack path

1. Obtain local shell access on an affected kernel
2. Launch an application that initializes amdgpu GPU contexts
3. Cause process termination while signals are pending
4. Trigger repeated exits to accumulate leaked GPU VM memory

How to detect it

Search kernel logs and dmesg for the exact backtrace containing amdgpu_driver_postclose_kms, drm_file_free, and get_signal calls that indicate an interrupted vm lock during exit.

What your security team should do

• Prioritize kernel package updates on all AMD GPU hosts
• Add kernel log monitoring for amdgpu postclose warnings
• Restrict local unprivileged access on systems that cannot be immediately patched