Core-Local Memory Assisted Protection

Abstract

This paper proposes a method for protecting data by leveraging core-local memory. Core-local (or software coherency managed) memory is a programmable memory which is equipped in a core of multicore processors. It is accessible from the core with low latency compared to a shared cache and a shared main memory. This is equipped in multicore processors in order to exploit locality of threads and to improve scalability. In addition to low latency, core-local memory is invisible and inaccessible from the other cores. We leverage this characteristic to provide a novel mechanism of protecting an OS kernel beside MMU based address space separation.