The bug lies in the AMDGPU driver for RDNA4 (GFX12) devices, where a zero‑size GDS, GWS, or OA on‑chip memory region is incorrectly initialized. The init code unconditionally creates a TTM range manager for every resource, even when the size is zero. When the size is zero, the manager initialization triggers an assertion that crashes the kernel during module loading. The result is a kernel panic or crash, effectively denying service on systems using affected GPUs. The weakness can be classified as an improper initialization of hardware resources.

All Linux kernel builds that ship an unpatched AMDGPU driver on RDNA4 hardware, such as Radeon RX 9070 XT and similar GPUs. The exact kernel version range is not specified, but any kernel that has the unrevised amdgpu_ttm_init_on_chip implementation before the commit that introduced the early return is vulnerable. Users of newer kernel releases that contain the patch are not affected.

Because the defect is triggered by loading or initializing the amdgpu driver module, the attack requires local privilege to load the module or to boot into a kernel that executes it. The vulnerability has a high impact (kernel crash) but a low exploitation complexity and currently has no publicly disclosed exploitation technique. The EPSS score is unavailable, and the vulnerability is not listed in the CISA KEV catalog, indicating limited exploitation risk at present. Nonetheless, failure to patch leaves a system vulnerable to intentional denial of service by any local or remote attacker who can load the AMDGPU module.