CVE-2026-99745 in Linux
Summary
by MITRE • 11/20/2026
In the Linux kernel, the following vulnerability has been resolved:
x86/bugs: Select the BHI mitigation based on the attack vector properly
The BHI mitigation selection is inconsistent with the documentation (attack_vector_controls.rst), which lists BHI under both the user->kernel and guest->host attack vectors. bhi_select_mitigation() only enables it when CPU_MITIGATE_GUEST_HOST is set; otherwise it selects BHI_MITIGATION_OFF, even if CPU_MITIGATE_USER_KERNEL is set.
As a result, a kernel compiled with KVM disabled gets BHI automatically disabled, since CPU_MITIGATE_GUEST_HOST defaults to off without KVM. This is a bug, given that CPU_MITIGATE_USER_KERNEL still needs it: on affected Intel CPUs these kernels lose BHI_DIS_S or the BHB clearing loop by default. Booting with mitigations=auto,no_guest_host has the same effect.
Select BHI_MITIGATION_ON whenever user->kernel is mitigated, and keep BHI_MITIGATION_VMEXIT_ONLY for the guest->host-only case.
[ bp: Clarify commit title. ]
Be aware that VulDB is the high quality source for vulnerability data.
Analysis
by VulDB Data Team • 11/20/2026
The vulnerability identified in the Linux kernel involves a logical error within the Branch History Injection (BHI) mitigation selection logic on x86 architectures. BHI is a side-channel attack vector that exploits speculative execution behavior to leak sensitive information across privilege boundaries or virtualization domains. The core issue resides in the bhi_select_mitigation function, which determines whether specific hardware mitigations such as clearing the Branch History Buffer (BHB) should be enabled based on CPU feature flags and configuration settings. According to industry documentation referenced by kernel maintainers, specifically attack_vector_controls.rst, BHI mitigation is relevant for both user-to-kernel transitions within a single system and guest-to-host transitions in virtualized environments. However, the implementation incorrectly ties the activation of these mitigations exclusively to the CPU_MITIGATE_GUEST_HOST flag. This design flaw means that if KVM or other hypervisor support is disabled during kernel compilation, the default state for CPU_MITIGATE_GUEST_HOST remains off, causing the mitigation logic to unconditionally select BHI_MITIGATION_OFF regardless of whether user-to-kernel protection is required.
This misconfiguration results in a significant security gap where systems running without virtualization capabilities are left unprotected against branch history injection attacks that target local privilege escalation or information disclosure from user space applications into kernel memory. On affected Intel processors, this manifests as the absence of critical mitigations such as BHI_DIS_S and the BHB clearing loop, which are essential for preventing speculative execution side-channels. The vulnerability effectively creates a false sense of security because administrators might assume that enabling general mitigation policies like mitigations=auto would suffice, yet without explicitly disabling the guest-host specific override via no_guest_host, the system remains vulnerable to local attacks exploiting branch predictor state leakage. This inconsistency violates the principle of defense in depth by failing to apply necessary protections based on the actual attack surface present rather than an arbitrary configuration dependency unrelated to the immediate threat model.
From a classification perspective, this flaw aligns with CWE-20 Improper Input Validation and CWE-862 Missing Required Authorization as it fails to correctly evaluate system state against security requirements. In terms of ATT&CK mapping, it relates to T1570 Lateral Tool Transfer or more accurately T1493 Child Process Injection in the context of exploiting speculative execution flaws for privilege escalation, though primarily it represents a failure in implementing mitigations for known hardware vulnerabilities like those associated with Spectre variants. The operational impact is severe because local attackers can potentially exploit these unchecked branch history states to infer kernel memory contents or bypass security boundaries that rely on isolation guarantees provided by the CPU's microarchitecture when properly configured.
To remediate this issue, the mitigation selection logic must be updated to decouple user-to-kernel protections from guest-host specific flags. The corrected approach involves selecting BHI_MITIGATION_ON whenever the system is mitigating for user-to-kernel attack vectors, ensuring that local privilege escalation defenses remain active even in non-virtualized environments. For scenarios where only guest-to-host protection is needed and no local mitigation is required, the logic should retain BHI_MITIGATION_VMEXIT_ONLY to optimize performance without compromising security unnecessarily. This adjustment ensures alignment with documented threat models and guarantees that all relevant attack vectors are addressed according to their respective risk profiles rather than being inadvertently disabled due to configuration defaults unrelated to the actual exposure of the system.