HSFI: Accurate fault injection scalable to large code bases

When software fault injection is used, faults aretypically inserted at the binary or source level. The former isfast but provides poor fault accuracy while the latter cannotscale to large code bases because the program must be rebuiltfor each experiment. Alternatives that avoid rebuilding incurlarge run-time overheads by applying fault injection decisionsat run-time. HSFI, our new design, injects faults with all contextinformation from the source level and applies fault injectiondecisions efficiently on the binary. It places markers in theoriginal code that can be recognized after code generation. Weimplemented a tool according to the new design and evaluated thetime taken per fault injection experiment when using operatingsystems as targets. We can perform experiments more quicklythan other source-based approaches, achieving performance thatcome close to that of binary-level fault injection while retainingthe benefits of source-level fault injection.