New Interrupt Injection Attack Compromises Spectre v2 Defenses on CPUs

Researchers from MIT CSAIL, Daniël Trujillo and Mengjia Yan, have unveiled a new attack technique named INTERRUPT INJECTION that can exploit vulnerabilities in Intel and AMD CPUs. This technique allows an unprivileged Linux program to time a hardware interrupt to occur during a critical moment, essentially re-poisoning the branch predictor following the execution of Spectre v2 mitigations.

The attack is viable on an AMD Zen 2 machine using Linux version 6.14, demonstrating that the new method can successfully bypass existing defenses. This capability highlights potential security gaps in the current mitigation strategies for Spectre v2 vulnerabilities.

This finding challenges the effectiveness of current protections against Spectre v2, raising concerns about branch predictor poisoning and the overall security of affected processors.