Assessing Software Abstractions in WCET Analysis of Reactive Programs

The estimation of the worst case execution time (WCET) of a reactive system on a given architecture is an important goal for time-critical systems. However, it cannot be achieved exactly, because of the complexity of modern architectures, the undecidability of most program analysis problems, and the need of taking into account the actual environment in which the system is intended to work. As a consequence, two approaches are possible: extensively testing the system with realistic input scenarios (dynamic method) provides an under-approximation of the WCET, while a guaranteed over-approximation can be obtained by applying static analysis of software and hardware. Comparing the results of both approaches and reducing the gap between them is interesting to assess the quality of the static analysis, and to decide when further refinements are useless. In this paper, we propose a methodology and a combination of tools to assess the result of software static analysis in the case of reactive programs. In order to permit a meaningful comparison, we perform a dynamic analysis using a cycle accurate simulator based on the same hardware model as the one used for static analysis. Moreover, we use an existing quite sophisticated framework to conduct the generation of reactive input scenarios, in order to track the worst case. This methodology and the use of associated tools is illustrated on a small but realistic example.