GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux KVM nSVM: KVM_SET_NESTED_STATE accepts LMA=1/LME=0, shadow MMU walks past its paging arrays

UnscoredCVE-2026-99345Kernel, userspace & hypervisorcurated

Impact

KVM_SET_NESTED_STATE on AMD did not reject L1 state with EFER.LMA=1 and LME=0. If CR4.PAE is also clear, KVM computes a 4- or 5-level paging role but wires the shadow MMU to the 2-level paging32 template, which leads to accesses past the guest walker's arrays in host kernel memory. The existing long-mode-without-PAE sanity check does not catch it because L2 is active when the fault happens. The result is host kernel memory corruption or a crash on the hypervisor node.

Who can reach it

A local process with access to /dev/kvm that sets nested state on a vCPU (a VMM, or a pod handed /dev/kvm). It is not described as reachable from inside a guest.

What to do

Install a kernel with the stable fix (commits linked) on AMD KVM hosts, then migrate or drain guests and reboot. Limit /dev/kvm access to trusted VMM processes. The record does not give fixed versions.

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.