Critical Linux KVM Flaw Enables Guest-to-Host Escape on ARM64 Systems

A critical vulnerability in the Linux Kernel-based Virtual Machine (KVM) for ARM64 systems could let attackers escape a virtual machine and gain read and write access to host kernel memory.

This flaw, tracked as CVE-2026-89775, specifically affects ARM64 hosts with nested virtualization enabled and has been addressed in the mainline Linux kernel.

Security researcher Hyunwoo Kim disclosed the issue after the coordinated-disclosure embargo expired. The bug arises from KVM/arm64’s handling of virtual CPU nested virtualization control structures, which can leave a freed host memory page mapped as writable for a malicious guest.

Critical Linux KVM Flaw

The vulnerability stems from a type truncation that affects the stage-1 page-table walk level. During address-translation handling, this truncation causes a size calculation to return zero, which is intended to represent an unknown size.

However, the Virtual Nested Control Register (VNCR) pseudo-Translation Lookaside Buffer (TLB) invalidation path incorrectly interprets this zero value as a valid range size. Consequently, the invalidation operation results in an empty address interval and is skipped entirely.

This stale mapping condition creates the core security risk: a host page that has been freed may remain mapped into the attacker-controlled guest environment with writable permissions.

According to the disclosure, the guest can then perform 64-bit reads and writes against the affected host page at a predictable fixed address. Since these operations occur without triggering a trap or VM exit, the attacker can manipulate host memory directly instead of relying on traditional hypervisor escape techniques.

CVE-2026-89775 is particularly concerning for multi-tenant ARM64 cloud environments that enable nested virtualization. An attacker who can create or control a nested-virtualization-capable guest instance might exploit this flaw to move from the guest operating system into the underlying host.

Additionally, the issue poses a local privilege-escalation risk on vulnerable Linux distributions. The advisory notes that some distributions, including Red Hat Enterprise Linux configurations, expose /dev/kvm with world-writable 0666 permissions.

If nested virtualization is enabled, an unprivileged local user could access KVM and potentially exploit the vulnerability to gain root-level privileges on the host.

Actual exposure depends on several factors, including ARM64 hardware, the kernel version in use, KVM availability, and whether nested virtualization is enabled. Systems that do not use KVM/arm64 nested virtualization are not expected to be affected through the described attack vector.

The vulnerable code was introduced in Linux commit 7270cc9157f47, dated May 14, 2025, and was present until the upstream fix in commit 8053393680d4, dated August 6, 2026.

Administrators should update to a kernel release that includes the upstream patch or obtain a vendor-provided fixed kernel package. Cloud providers operating ARM64 infrastructure need to verify whether nested virtualization is exposed to tenants and prioritize patch deployment on shared hosts.

As a temporary defense, organizations should turn off nested virtualization if it is not essential for operations and restrict access to /dev/kvm to trusted users and services.

Security teams should also inventory ARM64 virtualization hosts, review kernel versions, and monitor vendor advisories for backported fixes related to CVE-2026-89775.

Cut every SOC alert investigation by 21 min. Power your SOC with instant IOC context for immediate response: Integrate TI Lookup in your SOC