Software Model Checking with Explicit Scheduler and Symbolic Threads

Abstract

In many practical application domains, the software is organized into a set of threads, whose activation is exclusive and controlled by a cooperative scheduling policy: threads execute, without any interruption, until they either terminate or yield the control explicitly to the scheduler. The formal verification of such software poses significant challenges. On the one side, each thread may have infinite state space, and might call for abstraction. On the other side, the scheduling policy is often important for correctness, and an approach based on abstracting the scheduler may result in loss of precision and false positives. Unfortunately, the translation of the problem into a purely sequential software model checking problem turns out to be highly inefficient for the available technologies. We propose a software model checking technique that exploits the intrinsic structure of these programs. Each thread is translated into a separate sequential program and explored symbolically with lazy abstraction, while the overall verification is orchestrated by the direct execution of the scheduler. The approach is optimized by filtering the exploration of the scheduler with the integration of partial-order reduction. The technique, called ESST (Explicit Scheduler, Symbolic Threads) has been implemented and experimentally evaluated on a significant set of benchmarks. The results demonstrate that ESST technique is way more effective than software model checking applied to the sequentialized programs, and that partial-order reduction can lead to further performance improvements.