CRITICAL Introduced in 6.16
kvm VncrTlb Overflow
CVE-2026-89918
CVSS:3.1/AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
KernelScan AI8.4HIGH
01Description
In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: Correctly handle end of VA space TLBI invalidation Our TLB invalidation by VA code is based on comparing two ranges, one defined by the TLB, and one defined by the TLBI instruction. Each range is defined by a start and a size. However, the way the comparison is done doesn't account for address rollover, as it compares an address with (base + size). This works nicely until this expression represent the last page/block in the TTBR1 VA space, as the result is a big fat 0. And a failed TLB invalidation. Rewrite the comparison in a way that is immune to the address rollover (making the end address inclusive instead of exclusive), and move this into a common helper that is used by both VA and IPA invalidations, as suggested by Hyunwoo Kim (although the IPA version didn't suffer from this particular problem, obviously).
02KernelScan AI Analysis
Risk summary
A guest VM running on an ARM64 KVM host with nested virtualization can cause TLB invalidation to be silently skipped by targeting the end of the virtual address space, leaving stale TLB entries in the VNCR pseudo-TLB. These stale entries may allow the guest to access host or other guest memory through mappings that should have been removed, potentially enabling a guest-to-host escape. Products running untrusted nested virtualization guests on ARM64 are at highest risk.
Vulnerability analysis
In the KVM arm64 nested virtualization TLB invalidation path, the code compares a TLB entry's address range against the invalidation scope using an exclusive end address computed as base plus size. When this sum represents the last page in the TTBR1 VA space, the addition wraps to zero, causing the range intersection check to incorrectly conclude that the ranges do not overlap. As a result, a TLB entry that should be invalidated is left in place, leaving a stale mapping that a guest could continue to use to access memory that has been repurposed or belongs to another context. The fix rewrites the comparison to use inclusive end addresses (base plus size minus one), which avoids the address rollover entirely, and consolidates the logic into a shared helper used by both VA and IPA invalidation paths. The vulnerability is reachable from inside a guest VM on an ARM64 host with KVM nested virtualization using the VNCR_EL2 (NV2_ONLY) feature; the guest needs no special privileges beyond running at EL1 within the VM.
Lifecycle
03Fix Versions
| Branch | Introduced | Fixed in | Patch commit |
|---|---|---|---|
| mainline | 6.16 | 7.3-rc1 | 34af2c3e31f9 |
| 7.2 | 6.16 | 7.2.5 | c624731ca6fe |
| 6.18 | 6.16 | 6.18.51 | ce447651a52e |