Extending Model Checking with Dynamic Analysis

Abstract

In model-driven verification a model checker executes a program by embedding it within a test harness, thus admitting program verification without the need to translate the program, which runs as native code. Model checking techniques in which code is actually executed have recently gained popularity due to their ability to handle the full semantics of actual implementation languages and to support verification of rich properties. In this paper, we show that combination with dynamic analysis can, with relatively low overhead, considerably extend the capabilities of this style of model checking. In particular, we show how to use the CIL framework to instrument code in order to allow the SPIN model checker, when verifying C programs, to check additional properties, simulate system resets, and use local coverage information to guide the model checking search. An additional benefit of our approach is that instrumentations developed for model checking may be used without modification in testing or monitoring code. We are motivated by experience in applying model-driven verification to JPL-developed flight software modules, from which we take our example applications. We believe this is the first investigation in which an independent instrumentation for dynamic analysis has been integrated with model checking.

The work described in this paper was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. Funding was also provided by NASA ESAS 6G.