Program Conversion for Detecting Data Races in Concurrent Interrupt Handlers

Abstract

Data races are one of the most notorious concurrency bugs in explicitly shared-memory programs including concurrent interrupt handlers, because these bugs are hard to reproduce and lead to unintended nondeterministic executions of the program. The previous tool for detecting races in concurrent interrupt handlers converts each original handler into a corresponding thread to use existing techniques that detect races in multi-threaded programs. Unfortunately, this tool reports too many false positives, because it uses a static technique for detecting races. This paper presents a program conversion tool that translates the program to be debugged into a semantically equivalent multi-threaded program considering real-time scheduling policies and interrupt priorities of processor. And then, we detect races in the converted programs using a dynamic tool which detects races in multi-threaded programs. To evaluate this tool, we used two flight control programs for unmanned aerial vehicle. The previous approach reported two and three false positives in these programs, respectively, while our approach did not report any false positive.