MARF: A Memory-Aware CLFLUSH-Based Intra- and Inter-CPU Side-Channel Attack

In this work, we conduct in-depth characterization to quantify the impact of DRAM refresh, the location of the target memory object within a non-uniform memory access (NUMA) node, and task and page placement across NUMA nodes and identify a set of the patterns in the clflush latency data. Based on characterization results, we propose MARF, a novel memory-aware clflush-based intra- and inter-CPU side-channel attack on NUMA systems. Our case studies on three real NUMA systems demonstrate that MARF can robustly be used to attack applications that use widely-used cryptographic and user-interface libraries. We also present potential countermeasures against MARF.