ReDPro

Abstract

Process-level concurrency failures occur when multiple processes read and write a shared data item without proper synchronization. Process-level failures are difficult to detect because they are sensitive to the execution order of system calls. These failures can not be detected unless a specific system-call interleaving is exercised during testing. A common practice when testing or debugging concurrency failures is to execute many runs of a program with the hope of exercising interleaving that causes races known as stress testing. However, existing research prove that stress testing is neither efficient nor reproducible. In this paper, we presented an automated technique, RedPro that can detect process-level races and also regenerate the failures. We used a binary instrumentation tool named PIN for run-time monitoring and controlling the order of execution for potential race pairs. We evaluated the performance of RedPro by detecting and regenerating nine real-world bugs. The result demonstrates that our proposed technique is both effective and efficient.