Linux Kernel Vulnerability CVE-2026-89775 Allows Guest VM to Access Host Memory on ARM64
A critical flaw in the Linux kernel's ARM64 KVM virtualization code, CVE-2026-89775, permits guest virtual machines to read and write host kernel memory when nested virtualization is enabled.
A newly disclosed vulnerability within the Linux kernel's KVM (Kernel-based Virtual Machine) component for ARM64 processors presents a significant security risk, potentially allowing guest virtual machines to escape their confines and access host memory. The flaw, tracked as CVE-2026-89775, specifically affects systems with nested virtualization enabled, a feature that permits a guest VM to host its own virtual machines.
The vulnerability arises from an issue in how KVM handles nested virtualization on ARM64 architectures. When a guest VM manipulates its memory in a specific manner, a size calculation error can occur, leading to the omission of a crucial TLB (Translation Lookaside Buffer) invalidation step. This oversight leaves a freed piece of host memory mapped and accessible to the guest.
According to researcher Hyunwoo Kim, who discovered and reported the flaw, this vulnerability can be exploited to achieve a guest-to-host escape. This would allow an attacker running within a compromised guest VM to execute arbitrary code on the underlying host machine, thereby compromising the entire system. While no exploit code has been publicly released, and there is no indication of active exploitation in the wild, the potential impact is severe.
The affected code was present in mainline Linux kernels starting from version 6.16, though the specific exploitable behavior is believed to have been introduced in version 6.17. The vulnerability has been addressed in upstream Linux kernel releases 6.18.51, 7.2.5, and 7.3-rc1. Various Linux distributions are in the process of incorporating these fixes into their respective stable releases.
Beyond the direct guest-to-host escape, the vulnerability also presents a risk for local privilege escalation on systems where the /dev/kvm device is accessible to unprivileged users. This is particularly relevant for distributions like Red Hat Enterprise Linux, where this device is open to all users by default. In such scenarios, a local attacker could potentially leverage the flaw to gain root privileges on the host, provided nested virtualization is enabled.
While the vulnerability requires nested virtualization to be enabled—a feature that is not enabled by default on ARM64 systems and requires specific hardware capabilities (Armv8.4 with FEAT_NV2)—its presence is a concern for environments that do utilize this configuration. Major cloud providers like AWS and Google Cloud do not offer nested virtualization for their standard ARM instances, limiting the direct applicability of this specific attack vector on their platforms.
This disclosure marks the fourth KVM guest-to-host escape vulnerability reported by Hyunwoo Kim this year, following two similar issues in the x86 KVM implementation and a prior ARM64 escape named ITScape. The consistent discovery of such flaws highlights ongoing challenges in securing complex virtualization technologies.
As of September 22, CVE-2026-89775 was not listed on CISA's Known Exploited Vulnerabilities (KEV) catalog, and its predicted exploitation score remained low. However, vendors have assigned high severity ratings to the flaw, ranging from 7.8 to 9.3 out of 10, underscoring the critical nature of patching affected systems promptly.
The newly disclosed article provides further technical details on the CVE-2026-89775 vulnerability, explaining how a type truncation issue in KVM/arm64 page-table walks can lead to a freed host memory page remaining mapped as writable. This allows a malicious guest to gain 64-bit read and write access to host memory without triggering a trap, potentially compromising cloud infrastructure and systems with nested virtualization enabled.