Fast and accurate static data-race detection for concurrent programs

Abstract

We present new techniques for fast, accurate and scalable static data race detection in concurrent programs. Focusing our analysis on Linux device drivers allowed us to identify the unique challenges posed by debugging largescale real-life code and also pinpointed drawbacks in existing race warning generation methods. This motivated the development of new techniques that helped us in improving both the scalability as well as the accuracy of each of the three main steps in a race warning generation system. The first and most crucial step is the automatic discovery of shared variables. Towards that end, we present a new, efficient dataflow algorithm for shared variable detection which is more effective than existing correlation-based techniques that failed to detect the shared variables responsible for data races in majority of the drivers in our benchmark suite. Secondly, accuracy of race warning generation strongly hinges on the precision of the pointer analysis used to compute aliases for lock pointers. We formulate a new scalable context sensitive alias analysis that effectively combines a divide and conquer strategy with function summarization and is demonstrably more efficient than existing BDD-based techniques. Finally, we provide a new warning reduction technique that leverages lock acquisition patterns to yield provably better warning reduction than existing lockset based methods. © Springer-Verlag Berlin Heidelberg 2007.