Static analysis for detecting high-level races in RTOS kernels

We propose a static analysis based approach for detecting high-level races in RTOS kernels popularly used in safety-critical embedded software. High-Level races are indicators of atomicity violations and can lead to erroneous software behaviour with serious consequences. Hitherto techniques for detecting high-level races have relied on model-checking approaches, which are inefficient and apriori unsound. In contrast we propose a technique based on static analysis that is both efficient and sound. The technique is based on the notion of disjoint blocks recently introduced in Chopra et al. (In: Proceedings of 28th European symposium on programming (ESOP), Prague, Czech Republic. LNCS, vol 11423, pp 1–27. Springer, 2019). We evaluate our technique on four popular RTOS kernels and show that it is effective in detecting races, many of them harmful, with a high rate of precision.