A newly documented vulnerability allows unprivileged software to circumvent established Spectre v2 security defenses on both Intel and AMD processors. According to The Hacker News, researchers at MIT CSAIL, specifically Daniël Trujillo and Mengjia Yan, have developed a technique referred to as Interrupt Injection that effectively re-poisons branch predictors even when standard kernel-level mitigations are active.

The mechanism exploits a narrow window of execution timing. In a typical secure environment, a processor sanitizes its branch predictor before the kernel performs operations. The Interrupt Injection attack utilizes the precise timing of a hardware interrupt to land exactly in the gap between the sanitization process and the kernel's subsequent use of the predictor. By forcing the processor to execute this sequence, an attacker can re-introduce malicious data into the predictor, neutralizing the intended security isolation.

Evidence of this vulnerability was demonstrated on an AMD Zen 2 machine running the Linux 6.14 kernel. Despite having all default Spectre v2 mitigation protocols enabled, the system remained susceptible to this targeted timing attack. This discovery highlights a potential flaw in how current hardware handles state-clearing operations during interrupt handling.

Why It Matters

The discovery of Interrupt Injection indicates that microarchitectural defenses are increasingly susceptible to sophisticated timing-based exploits. While Spectre v2 mitigations were considered a settled matter for many server environments, this research suggests that software-based patches may be insufficient when hardware state-clearing can be interrupted. This shift forces hardware manufacturers and kernel developers to reconsider how branch predictor states are isolated from external interrupts, potentially leading to performance overheads if stricter, more complex sanitization cycles are mandated to prevent this specific attack vector in the future.