GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux KVM nested SVM: INIT during L2 with paging off leaves L1 running on a stale MMU context

CVSS 8.8CVE-2026-99312Kernel, userspace & hypervisorcurated

Impact

When KVM emulates an INIT while a nested (L2) guest runs with paging disabled, it skips the MMU reset, so the L1 guest resumes on an MMU context built for L2. A guest can therefore end up with the wrong address translation. The record scores this as a scope change with high confidentiality, integrity and availability impact, but the description does not say how far the corruption reaches. The bug only affects AMD hosts (nested SVM). On VMX, INIT is always intercepted, so Intel hosts are not exposed. It matters on multi-tenant AMD GPU hosts that let tenant VMs run their own nested hypervisors.

Who can reach it

Local, from inside a guest VM: a low-privileged guest running a nested L2 guest on an AMD host where nested virtualization is enabled. No host credentials are needed. Hosts with nested SVM disabled, and Intel hosts, cannot be reached.

What to do

Install a kernel that carries the fixes from the stable commits listed, then drain each AMD KVM host and reboot it. Until then, you can block the path by turning off nested virtualization (kvm_amd nested=0). That means reloading the module or rebooting, and it breaks any tenant workload that runs nested guests.

References

Related entries

All Kernel, userspace & hypervisor entries

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.