GPU VulnDB

Database/NVIDIA / GPU stack

Imagination Graphics DDK: guest VM GPU commands write outside the virtualised GPU memory partition

CVE-2026-45199NVIDIA / GPU stackcurated

Impact

Kernel software inside a guest VM can post malformed commands to the GPU firmware that cause writes outside the guest's virtualised GPU memory, which the vendor states can be used to escalate privileges. This is the failure mode a GPU-virtualisation deployment is built to prevent: the isolation boundary between guests sharing one GPU is enforced by the firmware command path, and a write past it reaches memory belonging to another guest or to the host. Note the affected product is Imagination's PowerVR-class DDK, which is used in embedded and SoC platforms rather than in NVIDIA/AMD datacenter fleets - relevant only to operators running PowerVR-based accelerators with GPU virtualisation enabled. The record gives no CVSS score and no exploitation detail beyond the out-of-bounds write and privilege-escalation claim.

Who can reach it

Requires code running at kernel level inside a guest VM that has been assigned a virtualised share of the GPU - so a tenant who already controls their guest kernel, not an unauthenticated network attacker.

What to do

Imagination publishes fixed DDK versions on its GPU driver vulnerabilities page; the record names no specific version. Fixes reach operators through the SoC vendor's BSP, so the practical path is a platform-vendor driver/firmware update rather than a distro package. Applying it means replacing the kernel-mode GPU driver and firmware and rebooting the host, with guests stopped. Until an update is available for your platform, do not co-tenant untrusted guests on a shared virtualised GPU.

References

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.