Static Detection of Atomic-Set-Serializability Violations

Abstract

Vaziri et al. propose a data-centric approach to synchronization. The key underlying concept of their work is the atomic set, which specifies the existence of an invariant that holds on a set of fields of an object type. In addition, they formalize a set of eleven data-access scenarios that completely specify the set of non-serializable interleaving patterns that can lead to an atomic-set serializability violation of the expressed invariant.

We present an algorithm that uses state-space exploration techniques to statically detect atomic-set serializability violations. The key idea is that the data-access scenarios can be used as a property specification for a software model checker. We tested our technique on programs with known serialiability violations from the concurrency-testing benchmark created by Eytani et al. Of the ten programs analyzed, our tool reported eight atomic-set serializability violations, with seven of them being true bugs.