Fault injection techniques for real-time operating systems

When an electronic system stops working properly, the causes of such malfunctioning could be due to a human factor or to the external environment. If the former ones can be discarded, then it is highly probable that the system encountered an error as consequence of high energy particles which stroke the hardware causing a permanent or transient damage: in the former case, the whole system is definitively harmed and only a physical substitution of the broken circuit can solve the problem; in the latter case, instead, power cycling would be a sufficient solution. In both circumstances, if the system must be always active and respond respecting well defined deadlines, such misbehaviors can lead to catastrophic consequences; in the worst case, the element to be substituted cannot be even accessed: this is the case of automotive, avionic and aerospace applications. In order to avoid these scenarios, the system must be extensively tested and then strengthen where it showed high sensitivity to random variation in signals and data: fault injection is exactly that technique which allows to spot vulnerabilities in a system, highlighting those parts which need to be hardened. This work aims to investigate the effects of Single Event Upset (SEU), caused usually by high energy particles, in Real-Time Operating Systems (RTOS) specifically developed for embedded solutions, analyzing the consequences of faults injected on most relevant data of the operating system itself. SEU effects are simulated using a prototyping board designed by STMicroelectronics running FreeRTOS as embedded OS; parameters of the injection are sent to the hardware from a host computer, which automatizes the process. Fault injection campaigns are performed on various parts of the OS and information about the status of the system are extracted after the injection.