Tuning Static Data Race Analysis for Automotive Control Software

Abstract

Implementation of concurrent software systems is difficult and error-prone. Race conditions can cause intermittent failures, which are rarely found during testing. In safety-critical applications, the absence of race conditions should be demonstrated before deployment of the system. Several static analysis techniques to show the absence of data races are known today. In this paper, we report on our experiences with a static data race detector. We define a basic analysis based on classical lockset analysis and present three enhancements to that algorithm. We evaluate and compare the effectiveness of the basic and enhanced analysis algorithms empirically for an automotive embedded system. We find that the number of warnings could be reduced by more than 40% and that the ratio of true positives per total number of warnings could be doubled.