Talk to an expert
ProductsSolutionsTrustPricingNewsTalk to an expert
Back to Newsroom
Threat Intel

ARM64 KVM Flaw Lets Virtual Machines Read and Write Host Memory

A newly disclosed Linux kernel bug lets a guest virtual machine escape onto its ARM64 host by exploiting a skipped memory-cache cleanup step. Here's what's affected and how to respond.

4T4Tify Team4Tify
Sep 22, 2026·5 min read
ARM64 KVM Flaw Lets Virtual Machines Read and Write Host Memory

A newly disclosed Linux kernel vulnerability, tracked as CVE-2026-89775, allows a guest virtual machine running on an ARM64 host to read and write memory that belongs to the host itself — and, in the right configuration, break out of the guest entirely.

What happened

The bug lives in the part of the Linux KVM hypervisor that handles nested virtualization on ARM64 — the ability for a guest VM to run its own hypervisor and host virtual machines of its own. When a guest sets up its memory in a specific way, a size calculation inside the kernel comes out as zero, and a step that should flush stale entries from the processor's address-translation cache gets skipped. The result: a page of host memory that has already been freed stays mapped and writable inside the guest, which can read and write it 64 bits at a time with no hardware trap alerting the host.

The researcher who found the flaw says it can be used to escape the guest and execute code directly on the host. A second path exists on systems where any local user can open /dev/kvm — the device used to create virtual machines — letting an unprivileged user build a guest and use the same bug to gain root. Red Hat Enterprise Linux ships with that device open to all users by default.

No proof-of-concept has been published, and there is no evidence of exploitation in the wild. As of publication, the flaw is not in CISA's Known Exploited Vulnerabilities catalog and carries a predicted exploitation score below 1%.

Why it matters

Nested virtualization is not enabled by default on ARM64 — it requires an experimental boot-time mode and Armv8.4-class hardware — so most standard deployments are not exposed out of the box. Vendors rate the impact between 7.8 and 9.3 out of 10, agreeing the severity is high while disagreeing on how easy it is to exploit; the attack requires local access and cannot be launched over the network.

The fix landed upstream in Linux 6.18.51, 7.2.5, and 7.3-rc1. Coverage varies by distribution: Red Hat's version 10 kernel is affected (versions 6 through 9 are not), Ubuntu 26.04 and its cloud kernels are vulnerable while 24.04 LTS's general kernel is not, and Debian's fix is already in sid with forky still open. On the largest cloud platforms, the configuration this flaw needs isn't on offer by default — AWS limits nested virtualization to Intel-based instances and Google Cloud excludes its ARM VMs from it — which narrows, but doesn't eliminate, the exposure for tenants running self-managed ARM64 hypervisors.

What to do

  • Inventory any ARM64 hosts running nested virtualization, especially multi-tenant or hosting environments.
  • Patch to Linux 6.18.51, 7.2.5, 7.3-rc1, or your distribution's backported fix as soon as it's available.
  • Restrict /dev/kvm access to trusted users only where nested virtualization is enabled, closing the local privilege-escalation path.
  • Disable nested virtualization on ARM64 where it isn't actively needed — it remains the single most effective mitigation until patches are fully rolled out.
SHARE