Efficient Fault Localization During Replay of Embedded Software

When testing embedded software in real world operation, remaining failures which have not been detected during unit or system testing can often be observed. The execution of these failures is traced during operation tests of the integrated embedded system to replay it for offline debugging in the laboratory. It is time-consuming to detect the root-causes of noncrashing failures in the replay manually. Dynamic analyses during replay assist the developers in fixing these failures. The implementation of dynamic analyses using standard debugging tools improves the portability to different embedded platforms. Unfortunately, fine-grained dynamic analyses with debugger tools are often not applicable in practice because they require too much runtime effort for monitoring. Our approach accelerates the analyses by monitoring in detail only the locations which are relevant for the failure. It starts with fast monitoring of method calls to detect the approximate location of the failure in the replayed execution. Afterwards, in additional replays, it detects the root-causes by monitoring in detail only the relevant executions of methods. This paper shows how our multi-level concept makes fault localization for embedded software using debugger tools applicable in practice.