Linux Kernel KVM EUVD-2026-32395

Exploitation requires: (1) local access to the host system or VM control plane with low privileges (PR:L per CVSS); (2) the host kernel running an affected Linux version prior to 6.18.27, 7.0.4, or 7.1-rc1; (3) AMD SVM hardware with LBR virtualization (LBRV) specifically enabled in the KVM/SVM module - this is the key non-default configuration gate; (4) a VM save/restore, live migration, or nested virtualization (L1/L2) scenario that triggers LBR MSR context switches. Exploitation is limited by: the need for local system access (not remotely exploitable), the requirement for LBRV to be active (non-default in many deployments), and the absence of a network-facing attack surface. The secondary Intel VMX finding (MSR_IA32_DEBUGCTLMSR reporting gap) is effectively mitigated by existing kernel code and does not present an independent exploitation path.

Real-world risk is moderate-to-low from an exploitation standpoint but meaningful for operators running high-availability KVM workloads on AMD hardware. The CVSS vector (AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H) indicates that a local, low-privileged attacker can trigger a high-availability impact without user interaction. EPSS is 0.02% (4th percentile), and there is no CISA KEV listing and no known public exploit - signals that this is not being actively targeted opportunistically. The primary real-world risk is operational: incorrect LBR state restoration during live migration or snapshot/restore of KVM guests on SVM hardware can cause VM instability or crashes, with highest impact on environments using LBR virtualization and nested guests. One notable discrepancy: the vulnerability tags include 'Information Disclosure,' yet the CVSS base vector carries C:N (no confidentiality impact). Improperly restored LBR MSRs could theoretically expose branch-history data across VM boundaries, constituting a side-channel - this potential information disclosure dimension is NOT confirmed by the CVSS score and should be verified with the vendor. The flaw does NOT affect Intel-only deployments at the primary impact level, though the MSR_IA32_DEBUGCTLMSR reporting gap on VMX is a secondary finding.

A local user with sufficient privilege to operate or interact with a KVM guest on an AMD SVM host with LBRV enabled triggers a live migration or snapshot/restore of an L1 guest that has an active L2 (nested) guest running. Because LBR MSRs are not correctly enumerated or restored, the VMCB01 save area is not populated with valid LBR state on L2 exit, causing the L1 guest to encounter corrupt machine state and crash or behave unpredictably. No public exploit code has been identified, and successful exploitation requires direct access to the virtualization control plane rather than a network-facing path.