During a GPU reset, the amdgpu_coredump routine counts unsignalled rings and then copies ring data into a buffer. A race condition allows the number of rings to change between the two loops, causing the copy index to exceed the allocated size and write past the end of the buffer. The resulting kernel memory corruption manifests as a KASAN out‑of‑bounds error and could destabilise the system.

Linux kernels that include the unpatched amdgpu driver code. The vendor metadata lists the os as Linux and no specific version range is provided, so any distribution running a kernel with the affected driver implementation is at risk until the patch is applied.

The CVSS score of 7.8 indicates a high severity for kernel integrity. The EPSS score of < 1% and absence from the CISA KEV catalog suggest that exploitation has not yet been observed and that the vulnerability is difficult to trigger because it requires precise timing of GPU fence signalling during a reset. Consequently, while the impact is significant, the practical exploitability is limited and likely requires an attacker with control over the GPU device or a system that incurs frequent resets.