Interrupt Injection Attack Bypasses Spectre v2 CPU Defenses

Researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have identified a novel security vulnerability, dubbed INTERRUPT INJECTION, capable of circumventing Spectre v2 defenses on processors manufactured by both Intel and AMD. This attack exploits a timing window that occurs during the handling of hardware interrupts, allowing an unprivileged program to re-poison the processor's branch predictor after the system's security mitigations have already been applied. The researchers, Daniël Trujillo and Mengjia Yan, detailed their findings, demonstrating that this technique can effectively bypass existing protections designed to prevent speculative execution attacks.

The core mechanism of the INTERRUPT INJECTION attack involves an unprivileged Linux program carefully timing a hardware interrupt. This interrupt is engineered to arrive precisely in the narrow time gap that exists between when the processor sanitizes its branch predictor—a crucial step in the Spectre v2 mitigation process—and when the operating system's kernel subsequently utilizes that predictor. By injecting the interrupt at this specific moment, the malicious program can re-introduce malicious data into the branch predictor, effectively re-poisoning it after the defense has supposedly run. This re-poisoning allows the attacker to influence the processor's speculative execution path, potentially leading to the leakage of sensitive information.

In their experiments, Trujillo and Yan successfully demonstrated the attack's efficacy on an AMD Zen 2 machine running the Linux kernel version 6.14. Critically, this test was conducted with all default Spectre v2 mitigations enabled, highlighting the attack's ability to bypass even robust security configurations. The implications of this discovery are significant, as Spectre v2 and its variants have been a persistent concern for processor security for years. The original Spectre vulnerabilities, first disclosed in 2018, allowed attackers to trick processors into speculatively executing instructions that they would not normally execute, potentially reading sensitive data from memory. Spectre v2 specifically targets the processor's branch predictor, a component that guesses which path of execution a program will take to improve performance.

The INTERRUPT INJECTION attack represents a new dimension in the ongoing battle between hardware security researchers and processor manufacturers. While Spectre v2 mitigations aim to prevent the branch predictor from being misused, this new technique finds a way to subvert those very defenses by exploiting the interrupt handling mechanism. The researchers have not yet released specific details on the exact timing requirements or the full range of affected processors beyond the AMD Zen 2 architecture tested. However, the potential for this attack to affect a broad range of modern CPUs, including those from Intel, underscores the need for further investigation and the development of new, more resilient security measures to protect against speculative execution vulnerabilities.