Precise Dynamic Data Race Prediction for Interrupt-Driven Embedded Software
Tingting Yu, Chunpeng Jia, Rui Chen, Chao Li, Boxiang Wang, Yunsong Jiang · 2023
Data races are common in interrupt-driven embedded software and have caused serious safety accidents. Unfortunately, existing dynamic race detection techniques for these software suffer from high false positives and overhead, mainly due to the negligence of widely existing user-defined synchronizations and excessively simulating interrupt triggering during execution. This paper proposes intRace, a precise dynamic data race detection technique for interrupt-driven embedded software. Based on an industrial full-system simulation platform VTest, intRace can detect races on-the-fly during dynamic testing of embedded software. Guided by existing test cases with high coverage, intRace uses a pattern-based hybrid method to identify user-defined synchronizations and detect data races based on accurate partial order relationships precisely and efficiently.We evaluated intRace on 4 real-world interrupt-driven aerospace embedded software. The results showed that intRace identified all user-defined synchronizations with no false positives, and found 40 data races with an average FP rate of 11.1% at low overhead.