Performance Counters to Rescue: A Machine Learning based safeguard against Micro-architectural Side-Channel-Attacks

Micro-architectural side-channel-attacks are presently daunting threats to most mathematically elegant encryption algorithms. Even though there exist various defense mechanisms, most of them come with the extra overhead of implementation. Recent studies have prevented some particular categories of these attacks but fail to address the detection of other classes. This paper presents a generic machine learning based multi-layer detection approach targeting these micro-architectural side-channel-attacks, without concentrating on a single category. The proposed approach work by proling low-level hardware events using Linux perf event API and then by analyzing these data with some appropriate machine learning techniques. This paper also presents a novel approach, using time-series data, to correlate the execution trace of the adversary with the secret key of encryption for dealing with false-positives and unknown attacks. The experimental results and performance of the proposed approach suggest its superiority with high detection accuracy and low performance overhead