CVE-2026-46032: Linux Kernel Privilege Escalation Flaw

CVE-2026-46032 Overview

CVE-2026-46032 is a Linux kernel vulnerability in the KVM nested SVM (nSVM) virtualization subsystem. When loading L1's CR3 register fails during a nested #VMEXIT, nested_svm_vmexit() returns an error code that most callers ignore. The hypervisor then continues running L1 with corrupted state instead of performing a clean shutdown as the AMD Architecture Programmer's Manual (APM) requires. The fix injects a triple fault and continues cleaning up vCPU state to handle the failure gracefully.

Critical Impact
A nested guest can be left executing on a corrupted CPU state path after a failed host CR3 reload, deviating from documented AMD CPU behavior and producing undefined hypervisor behavior.

Affected Products

Linux kernel KVM subsystem (arch/x86/kvm/svm/nested.c)
Systems running AMD SVM with nested virtualization enabled
Stable kernel branches receiving commits 5d291ef0585e and 9a738cf170a4

Discovery Timeline

2026-05-27 - CVE-2026-46032 published to NVD
2026-05-27 - Last updated in NVD database

Technical Details for CVE-2026-46032

Vulnerability Analysis

The vulnerability resides in the KVM nested SVM exit handler nested_svm_vmexit(). During a nested #VMEXIT, the hypervisor must restore the L1 host execution context, including the host CR3 page table base register. If that restore fails because of an illegal host state value, the function returns an error code.

Most callers of nested_svm_vmexit() do not check this return value. Execution continues with L1 in a partially restored, corrupted state. According to the AMD APM, the documented hardware behavior on illegal host state or an exception while loading host state is shutdown. The KVM implementation diverges from this contract.

The resolution removes the largely unchecked return value, injects a triple fault to model the documented shutdown behavior, and continues cleanup of vCPU state such as clearing pending exceptions. This avoids further execution on inconsistent state.

Root Cause

The root cause is an unchecked error path in the nested #VMEXIT handler combined with divergence from documented CPU behavior. The function signaled failure through a return code that callers ignored, producing a hypervisor state inconsistency rather than the architecturally required shutdown.

Attack Vector

Triggering the bug requires nested virtualization on AMD SVM hardware and a state where restoring L1's CR3 on #VMEXIT fails. A privileged actor inside an L1 guest controlling nested VM execution could potentially provoke the failure path, leading to a corrupted L1 state under the host KVM. Reachability from unprivileged guest contexts is not established in the published commit data.

No public proof-of-concept exploit, CISA KEV listing, or in-the-wild exploitation has been reported for this issue. The fix is described in the upstream commits referenced below; see the kernel commit 5d291ef0585e and kernel commit 9a738cf170a4 for the exact code changes.

Detection Methods for CVE-2026-46032

Indicators of Compromise

Unexpected guest shutdown events or triple-fault injections recorded in KVM tracepoints on AMD SVM hosts running nested workloads.
Kernel log entries from arch/x86/kvm/svm/nested.c indicating failed host state restoration during nested #VMEXIT.
L1 hypervisor instability or crashes correlated with nested guest VMRUN/VMEXIT cycles on unpatched kernels.

Detection Strategies

Inventory hosts running AMD CPUs with kvm_amd loaded and nested=1 enabled, and compare running kernel versions against patched stable branches.
Enable KVM tracepoints (kvm:kvm_nested_vmexit, kvm:kvm_inj_exception) to identify abnormal triple-fault injections following nested exits.
Audit virtualization hosts for nested virtualization usage that is not operationally required, reducing exposure surface.

Monitoring Recommendations

Forward dmesg and journald kernel logs from KVM hosts to a centralized log platform and alert on KVM-related warnings on AMD systems.
Track stable kernel package versions across the virtualization fleet to confirm uptake of the fix commits.
Monitor guest crash rates and unexplained shutdowns on AMD nested virtualization workloads as a behavioral signal.

How to Mitigate CVE-2026-46032

Immediate Actions Required

Update affected Linux kernels to a stable release containing commits 5d291ef0585e and 9a738cf170a4.
Identify all AMD SVM hosts with nested virtualization enabled and prioritize them for patching.
Restrict creation of nested guests to trusted administrators until patched kernels are deployed.

Patch Information

The upstream fix removes the return value of nested_svm_vmexit(), injects a triple fault when host state restore fails, and continues cleaning up vCPU state including pending exceptions. Apply the kernel updates from the stable kernel commit 5d291ef0585e and stable kernel commit 9a738cf170a4, or use distribution-provided kernel packages once available.

Workarounds

Disable nested virtualization on AMD hosts where it is not required by unloading or reloading kvm_amd with nested=0.
Limit guest privileges and prevent untrusted workloads from running their own hypervisors on shared hosts.
Migrate nested virtualization workloads to patched hosts as kernel updates land in the fleet.

bash

# Configuration example: disable nested virtualization for kvm_amd
sudo modprobe -r kvm_amd
echo "options kvm_amd nested=0" | sudo tee /etc/modprobe.d/kvm-amd-nested.conf
sudo modprobe kvm_amd
cat /sys/module/kvm_amd/parameters/nested  # expect: 0 or N