The kernel’s AMDGPU driver contains a logic flaw that unconditionally triggers a VMID 0 flush when clearing VRAM, even when a GART window is not used. That spurious SDMA flush can block the graphics engine on GFX11 GPUs, causing the device to hang and rendering GPU‑accelerated applications unusable. The weakness is an improper initialization that leads to resource mismanagement (CWE‑665).

The vulnerability affects Linux systems running the Linux kernel with the AMDGPU driver on GFX11 class GPUs, such as the Navi33 series. Any kernel version that has not incorporated the upstream commit a306e406e570b74318ff7d80e5b07b540ca1d3a9 is potentially affected. No specific kernel version range is supplied, so upgrading to a kernel that contains the fix is required.

The flaw is localized in the AMDGPU kernel driver and can be triggered during normal use when a process maps VRAM via amdgpu_ttm_map_buffer(). The path to the vulnerability runs through kernel code, so a local kernel context is required; this inference comes from the fact that the driver functions operate in kernel mode. The description does not explicitly state a privilege escalation requirement, so it is presumed that any process able to load modules and use the driver can trigger the path. While no exploitation evidence has been reported and the EPSS score is unavailable, the impact is a high‑impact denial of service: the unnecessary VMID 0 flush can wedge the graphics engine on GFX11 GPUs, rendering the GPU unusable. The vulnerability is not listed in the CISA KEV catalog, but the potential for widespread impact on GFX11 GPU users warrants attention.