In the Linux kernel AMDGPU driver, a resource cleanup ordering issue was reintroduced when the kernel reverted a previous fix. When the driver is torn down, the device resource associated with the aperture is released before the DRM device release callback runs. During that callback, firmware state is updated through a pointer derived from the aperture base address, which is already unmapped, leading to a kernel page fault and a subsequent panic. This results in an unplanned reboot and loss of service.

The flaw impacts all Linux kernel builds that contain the reverted commit sequence, which includes recent kernel releases that still rely on the older aperture cleanup logic. Systems running the AMDGPU driver in kernel mode are at risk, especially those that use the hardware with firmware that triggers the crash on removal or failure states. Vulnerable versions are those before the re‑application of the correct cleanup code, but the exact list may vary by distribution.

The CVSS score is not listed, and the EPSS value is unavailable. The vulnerability requires local kernel execution, so it is not directly exploitable over a network, but any privileged process that can provoke driver reinitialization or unload the AMDGPU module could trigger the fault. The risk is high for environments where kernel stability is critical, and a current KEV listing confirms it is not flagged as widely exploited. Prompt patching mitigates the issue; without a fix, the only way to reduce risk is to avoid using the driver until the bug is corrected.