Reverse Engineering the Stream Prefetcher for Profit

Abstract

Micro-architectural attacks exploit timing channels at different micro-architecture units. Some of the micro-architecture units like cache automatically provide the timing difference (the difference between a hit and a miss). However, there are other units that are not documented, and their influence on the timing difference is not fully understood. One such micro-architecture unit is an L2 hardware prefetcher named Streamer. In this paper, we reverse-engineer the Stream prefetcher, which is commercially available in the Intel machines. We perform a set of experiments and provide our observations and insights. Further, we use these observations to construct a cross-thread covert channel using the Stream prefetcher, with an accuracy of 91.3% and a bandwidth of 54.44 KBps.