Exposure of Sensitive Information caused by Shared Microarchitectural Predictor State that Influences Transient Execution

Many commodity processors have Instruction Set Architecture (ISA) features that protect software components from one another. These features can include memory segmentation, virtual memory, privilege rings, trusted execution environments, and virtual machines, among others. For example, virtual memory provides each process with its own address space, which prevents processes from accessing each other's private data. Many of these features can be used to form hardware-enforced security boundaries between software components.

When separate software components (for example, two processes) share microarchitectural predictor state across a hardware boundary, code in one component may be able to influence microarchitectural predictor behavior in another component. If the predictor can cause transient execution, the shared predictor state may allow an attacker to influence transient execution in a victim, and in a manner that could allow the attacker to infer private data from the victim by monitoring observable discrepancies (CWE-203) in a covert channel [REF-1400].

Predictor state may be shared when the processor transitions from one component to another (for example, when a process makes a system call to enter the kernel). Many commodity processors have features which prevent microarchitectural predictions that occur before a boundary from influencing predictions that occur after the boundary.

Predictor state may also be shared between hardware threads, for example, sibling hardware threads on a processor that supports simultaneous multithreading (SMT). This sharing may be benign if the hardware threads are simultaneously executing in the same software component, or it could expose a weakness if one sibling is a malicious software component, and the other sibling is a victim software component. Processors that share microarchitectural predictors between hardware threads may have features which prevent microarchitectural predictions that occur on one hardware thread from influencing predictions that occur on another hardware thread.

Features that restrict predictor state sharing across transitions or between hardware threads may be always-on, on by default, or may require opt-in from software.

Branch Target Injection (BTI) is a vulnerability that can allow an SMT hardware thread to maliciously train the indirect branch predictor state that is shared with its sibling hardware thread. A cross-thread BTI attack requires the attacker to find a vulnerable code sequence within the victim software. For example, the authors of [REF-1415] identified the following code sequence in the Windows library ntdll.dll:

To successfully exploit this code sequence to disclose the victim's private data, the attacker must also be able to find an indirect branch site within the victim, where the attacker controls the values in edi and ebx, and the attacker knows the value in edx as shown above at the indirect branch site.

A proof-of-concept cross-thread BTI attack might proceed as follows:

BTI can also allow software in one execution context to maliciously train branch predictor entries that can be used in another context. For example, on some processors user-mode software may be able to train predictor entries that can also be used after transitioning into kernel mode, such as after invoking a system call. This vulnerability does not necessarily require SMT and may instead be performed in synchronous steps, though it does require the attacker to find an exploitable code sequence in the victim's code, for example, in the kernel.