Dynamic analysis for concurrent modern C/C++ applications

Concurrent programs are executed by multiple threads that run simultaneously. While this allows programs to run more efficiently by utilising multiple processors, it brings with it numerous complications. For example, a program may behave unpredictably or erroneously when multiple threads modify the same memory location in an uncoordinated manner. Issues such as this are difficult to avoid, and when introduced, can break the program in unpredictable ways. Programmers will therefore often turn towards automated tools to aide in the detection of concurrency bugs. The work presented in this thesis aims to provide methods to aid in the creation of tools for the purpose of finding and explaining concurrency bugs. In particular, the following studies have been conducted: Dynamic Race Detection for C/C++11 With the introduction of a weak memory model in C++, many tools that provide dynamic race detection have become outdated, and are unable to adequately identify data races. This work updates an existing data race detection algorithm such that it can identify data races according to this new definition. A method for allowing programs to explore many of the weak behaviours that this new memory model permits is also provided. Record and Replay Much work has gone into record and replay, however, most of this work is focussed on whole system replay, whereby a tool will aim to record as much of the program execution as possible. Contrasting this, the work presented here aims to record as little as possible. This sparse approach has many interesting implications: some programs that were previously out of reach for record and reply become tractable, and vice versa. To back this up, controlled scheduling is introduced that is capable of applying different scheduling strategies, which combined with the record and replay is beneficial for helping to root out bugs. Tool Support Both of the above techniques have been implemented in a tool, tsan11rec, that builds on the tsan dynamic race detection tool. A large experimental evaluation is presented investigating the effectiveness of the enhanced data race detection algorithm when applied to the Firefox and Chromium web browsers, and of the novel approach to record and replay when applied to a diverse set of concurrent applications.Open Acces