MABFuzz: Multi-Armed Bandit Algorithms for Fuzzing Processors

• DOI: 10.48550/arxiv.2311.14594
• Publication Date: 2023-01-01
• Authors: Vasudev Gohil; Rahul Kande; Chen Chen; Ahmad‐Reza Sadeghi; Jeyavijayan Rajen...
• Keywords: Fuzz testing; Speedup

As the complexities of processors keep increasing, task effectively verifying their integrity and security becomes ever more daunting. The intricate web instructions, microarchitectural features, interdependencies woven into modern pose a formidable challenge for even most diligent verification engineers. To tackle this growing concern, recently, researchers have developed fuzzing techniques explicitly tailored hardware processors. However, prevailing issue with these fuzzers is heavy reliance on static strategies to make decisions in algorithms. address problem, we develop novel dynamic adaptive decision-making framework, MABFuzz, that uses multi-armed bandit (MAB) algorithms fuzz MABFuzz agnostic to, hence, applicable any existing fuzzer. In process designing encounter challenges related compatibility MAB maximizing efficacy fuzzing. We overcome by modifying tailoring accommodate special requirements integrate three widely used state-of-the-art fuzzer evaluate them popular RISC-V-based Experimental results demonstrate ability cover broader spectrum processors' landscapes doing so remarkable efficiency. particular, achieves up 308x speedup detecting vulnerabilities 5x achieving coverage compared technique.