Diverse Compiling for Software-Based Recovery of Permanent Faults in COTS Processors

Digital systems used in critical infrastructures have to fulfill ever higher demands on performance and cost efficiency. Thus, there is the trend to commercial off-the-shelf processors. To ensure a correct functioning of such devices, even after a long time of operation, mechanisms to recover from permanent hardware faults (e.g. due to wear-out effects) are needed. However, there is a lack of flexible low-cost software-based fault mitigation approaches that do not base on a costly exhaustive redundancy. To address this challenge, we show how to adapt the software execution such that the faulty hardware resource is no longer used. We propose to update the embedded device with an adapted binary that is generated on a remote server with diverse compiling. Our experiments demonstrate that this approach allows recovering from 99% of internal memory and 52% of register faults.