Fault Localization in Embedded Software

Abstract

Residual faults in embedded software are a major threat to systems dependability, and automated diagnosis techniques help to counter this threat in two ways: as a means to improve the efficiency of the debugging process they reduce the fault density, and as an integral part of fault detection, isolation, and recovery mechanisms they help harnessing runtime errors caused by residual faults. In this chapter we introduce spectrum-based fault localization (SFL), a diagnosis technique based on statistical analysis of execution profiles. SFL requires no additional modeling effort and has a low CPU and memory overhead, which makes it well suited for application in existing development environments and for the embedded systems domain. For deployment-time application, we complement SFL with light-weight generic error detection mechanisms based on program invariants, thus giving rise to systems that can autonomously identify potentially erroneous states, and locate the components that are likely to be involved. Using a probabilistic model of a software system, we show that the expected diagnostic accuracy of SFL approaches that of more expensive alternatives based on automated reasoning.