The Linux kernel’s AMDGPU driver miscomputes the minimum size required for a UVD decode image by using the width value instead of the pitch. When a pitch greater than 4096 is supplied, the calculation overflows, enabling a buffer overflow that can corrupt memory or cause a crash. This flaw is restricted to the UVD decoding path and does not affect normal CPU operations. Based on the description, it is inferred that an attacker would need local or privileged access to supply a decode request with an oversized pitch, as the vulnerable code runs in kernel space.

Any Linux system running a kernel predating the commit that corrects the pitch calculation is affected. The vulnerability manifests only when the AMDGPU driver is used for decoding UVD streams; systems that never employ the UVD hardware decoder or that run a kernel containing the patch are not vulnerable.

The CVSS score of 7.8 indicates a medium‑to‑high severity. EPSS score of <1% and the bug is not listed in the CISA KEV catalog, indicating no publicly known exploits. The likely attack vector is local exploitation via an application that can instruct the AMDGPU driver to produce a UVD decode request; successful exploitation could lead to a denial of service or, if the memory write occurs in privileged memory, potential privilege escalation. The vulnerability is limited to systems using the affected kernel and AMDGPU driver, and would not impact other kernel modules or unrelated hardware.