Program verification under weak memory consistency using separation logic

Abstract

The semantics of concurrent programs is now defined by a weak memory model, determined either by the programming language (e.g., in the case of C/C++11 or Java) or by the hardware architecture (e.g., for assembly and legacy C code). Since most work in concurrent software verification has been developed prior to weak memory consistency, it is natural to ask how these models affect formal reasoning about concurrent programs. In this overview paper, we show that verification is indeed affected: for example, the standard Owicki-Gries method is unsound under weak memory. Further, based on concurrent separation logic, we develop a number of sound program logics for fragments of the C/C++11 memory model. We show that these logics are useful not only for verifying concurrent programs, but also for explaining the weak memory constructs of C/C++.