CVE-2026-89916 PUBLISHED

KVM: arm64: Make VNCR invalidation participate in MMU invalidation retry

Assigner: Linux
Reserved: 11.09.2026 Published: 16.09.2026 Updated: 16.09.2026

In the Linux kernel, the following vulnerability has been resolved:

KVM: arm64: Make VNCR invalidation participate in MMU invalidation retry

A VNCR TLB invalidation can occur on one vcpu while another vcpu is faulting in this same page. Without correctly handling this, we can end up with the following scenario:

  • vcpu A walks the PTs to translate VNCR
  • before vcpu A is able to grab the MMU lock to insert the TLB, vcpu B updates the S1 PTs with an invalid entry, and issues a TLBI S1E2 for this VA
  • vcpu A inserts the TLB for something that is now invalid

This isn't a new problem, and we manage S2 by having the MMU notifier to bump up mmu_invalidate_seq on invalidation so that the fault can be replayed.

We can perform something similar here, and extend invalidate_vncr_va() to update the same counter, clearly indicating that the context has changed under our feet. This is safe as the invalidation always happen while holding the MMU lock for write, and that we sample the sequence number before walking S1.

Metrics

CVSS Vector: CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
CVSS Score: 9.3

AV:L - An arm64 nested L1 hypervisor reaches the bug via KVM_RUN: a VNCR data abort (kvm_handle_vncr_abort → kvm_translate_vncr) races a trapped EL2 TLBI (handle_tlbi_el2 → kvm_handle_s1e2_tlbi → invalidate_vncr_va). This is local KVM ioctl/guest execution, not a network, adjacent-radio, or physical path. AC:L - The attacker owns both vCPUs of the same VM: one walks EL2 stage-1 for VNCR while another updates those S1 page tables and issues TLBI S1E2. A race the attacker drives on both sides is AC:L, and nested virt on NV-capable arm64 hosts is an attacker-selectable deployment. PR:N - Exploitation requires no host root or init-namespace capabilities beyond running hypervisor code in a tenant KVM guest on an arm64 host with nested virtualization enabled; the attacker operates entirely from within their assigned VM, consistent with other arm64 KVM nested-virt CVE scoring. UI:N - No victim user or administrator action is required beyond the attacker operating their own nested-virtualization workload; VNCR faults, S1 page-table updates, and TLBI S1E2 on a second vCPU are fully attacker-driven. S:C - The stale VNCR pseudo-TLB is installed in host KVM and mapped into the host per-CPU fixmap via __set_fixmap, so a guest-controlled walk/TLBI race corrupts host nested-MMU state and can retain a writable kernel mapping of a revoked PFN, crossing the guest-to-hypervisor security boundary. C:H - kvm_translate_vncr() stores a raw HPA and drops the page pin via kvm_release_faultin_page(). A walk that misses S1 TLBI retry still maps that HPA PAGE_KERNEL in the host VNCR fixmap, so NV2/sysreg emulation can read pages the guest has invalidated and the host may have reused, which is UAF-class disclosure. I:H - The same stale vncr_tlb is remapped writable (PAGE_KERNEL) in the host fixmap, so nested EL2 register writes and NV2 stores land in the old HPA after the guest's TLBI/BBM sequence. That is a host-kernel write to memory whose translation was revoked and may have been reused. A:H - Using a stale HPA through the host VNCR fixmap can oops or panic the host if the PFN was unmapped or recycled, and corrupting nested EL2 control state or reused host pages can hang or crash the hypervisor and co-resident VMs; the attacker can retrigger the walk/TLBI race.

Product Status

Vendor Linux
Product Linux
Versions Default: unaffected
  • affected from 4ffa72ad8f37e73bbb6c0baa88557bcb4fd39929 to d8580e7aa189b0c6c5bb4d94b8352728a7c35bbd (excl.)
  • affected from 4ffa72ad8f37e73bbb6c0baa88557bcb4fd39929 to 11ee3faab4f1755d126bead9f882e75a7a677860 (excl.)
  • affected from 4ffa72ad8f37e73bbb6c0baa88557bcb4fd39929 to 2b7324f3a0c1072b9d578b8d42f199506753f26e (excl.)
Vendor Linux
Product Linux
Versions Default: affected
  • Version 6.16 is affected
  • unaffected from 0 to 6.16 (excl.)
  • unaffected from 6.18.51 to 6.18.* (incl.)
  • unaffected from 7.2.5 to 7.2.* (incl.)
  • unaffected from 7.3-rc1 to * (incl.)

References