Stream-based dynamic data race detection

Abstract

Detecting data races in modern code executing on multicore processors is challenging. Instrumentation-based techniques for race detection not only have a high performance impact, but also are not likely to be certified for safety-critical systems. This paper presents a data race detector based on the well-known lockset algorithm in the runtime verification language TeSSLa, which is a stream-based specification using dynamic data structures to record lock operations and memory accesses. Such a specification can then be instantiated with particular parameters to make it suitable for the more limited planned monitoring using field- programmable gate arrays.