Admit Your Weakness: Verifying Correctness on TSO Architectures

Abstract

Linearizability has become the standard correctness criterion for fine-grained non-atomic concurrent algorithms, however, most approaches assume a sequentially consistent memory model, which is not always realised in practice. In this paper we study the correctness of concurrent algorithms on a weak memory model: the TSO (Total Store Order) memory model, which is commonly implemented by multicore architectures. Here, linearizability is often too strict, and hence, we prove a weaker criterion, quiescent consistency instead. Like linearizability, quiescent consistency is compositional making it an ideal correctness criterion in a component-based context. We demonstrate how to model a typical concurrent algorithm, seqlock, and prove it quiescent consistent using a simulation-based approach. Previous approaches to proving correctness on TSO architectures have been based on linearizabilty which makes it necessary to modify the algorithm’s high-level requirements. Our approach is the first, to our knowledge, for proving correctness without the need for such a modification.