No Leakage Without State Change: Repurposing Configurable CPU Exceptions to Prevent Microarchitectural Attacks

Daniel Weber, Leonard Niemann, Lukas Gerlach, Jan Reineke, Michael Schwarz

Microarchitectural side-channel attacks have become significant threats to computer system security. While writing side-channel-resistant code can mitigate these attacks, it is time-consuming and error-prone. Detection approaches provide an alternative by monitoring the system for signs of ongoing attacks. However, distinguishing between malicious and benign processes is complex, error prone, and ineffective against sophisticated attacks. In this paper, we propose a novel approach, IRQGuard, which shifts the focus to proactive mitigation. IRQGuard enables the victim to monitor its own microarchitectural events resulting from microarchitectural state changes. Leveraging existing CPU features, IRQGuard uses interrupt requests (IRQs) triggered by victim-specific microarchitectural state changes within predefined code regions. This self-monitoring eliminates noise of unrelated applications, enabling immediate detection and response to potential attacks. Our proof-of-concept implementation demonstrates that IRQGuard stops information leakage in under 200 CPU cycles, outperforming current methods significantly. We evaluate IRQGuard on both cryptographic (OpenSSL) and non-cryptographic (toilet command-line utility) applications. We demonstrate IRQGuard's real-world viability by protecting an OpenSSH server from cache attacks. IRQGuard offers a practical, low-overhead solution for mitigating a wide range of microarchitectural attacks on Intel, AMD, and Arm CPUs.