KVM/arm64 nested virt flaw allows guest escape to host

A guest-to-host escape in the Linux kernel's KVM hypervisor for arm64 has been disclosed as CVE-2026-89775. On hosts with nested virtualization enabled, a malicious guest can obtain unrestricted 64-bit reads and writes to a freed host kernel page at a fixed address, with no trap or VM exit.

Hyunwoo Kim reported the issue. Incorrect handling of stage-1 walk size information caused a required mapping invalidation to be skipped, so the page stayed writable to the guest after it should have been cleared. The practical consequence is a reliable escape path on multi-tenant arm64 public clouds that offer nested virtualization: an attacker who can create such an instance can break out to the host.

On distributions such as RHEL, where /dev/kvm is world-writable, the same condition also serves as local privilege escalation. Any unprivileged user can open the device; if nested virtualization is enabled on the host, the bug yields root.

The flaw was present in mainline from May 2025 until it was patched in August 2026.