FSFI: A Full System Simulator-Based Fault Injection Tool

With VLSI technology advances and the increasing popularity of COTS components and multi-core processor in space, aviation, and military harsh environment, dependability becomes more attractive to overcome the increasing susceptibility to transient, permanent, and wear-out induced intermittent fault. Fault injection is widely used in dependability evaluation and fault emulation. As compared to physical- and software based fault-injection tool, this paper presents a simulation based Fault Injection tool, namely FSFI, to study high-level propagations of faults and system-level manifestations, especially the underlying causes of fault manifestations, namely Symptoms in this paper. The primary contributions of this paper are as follows. First and foremost, FSFI - A Full System simulator based Fault Injection tool is designed and described in great details. FSFI is based on an open source full system simulator (SAM) and it is heavily modified to support different processor component such as ALU, decoder, integer register files, and AGEN (Address Generation Unit). Additional four modules are added to the FSFI-simulator: Fault Injector, Monitor, Analyzer, and Controller. Second, transient faults are injected into different SPARC processor components, such as ALU, decoder, integer register files, and AGEN to deliberately study fault high level manifestations and propagations from processor component, through SPARC architecture level, hyper visor, OS, to application. Third, the underlying causes of fault manifestations, namely Symptoms, such as fatal trap, high OS, high hyper, and hangs, are captured by FSFI Monitor. The distributions of Symptoms for different components against typical benchmarks are investigated, and the underlying reasons are analyzed in detail. Preliminary experiment results show that FSFI is effective for fault high level propagation and Symptom research, and some of our future work is prospected.