Hypervisor for consolidating real-time automotive control units: Its procedure, implications and hidden pitfalls

Abstract

Abstract The future of automotive industry has a major focus on computing innovations such as autonomous driving, connectivity and mobility. These complex innovations integrate hardware and software components of different automotive safety integrity levels. Although standards like AUTOSAR provides flexibility in integrating different software components, it compromises on isolation from software interference. Virtualization is a technique, used for several decades in server domains, ensuring integration of heterogeneous subsystems without interference. Until recent years, inheriting virtualization onto real-time systems has been avoided due to its overhead and impact in-terms of timing behavior. With the usage of multi- and many-cores in the automotive industry, virtualization is gaining more importance to handle the upcoming innovations with incredible increased complexity and the needed computation power. This paper describes the technique for bringing virtualization into an automotive state-of-the-art multicore controller, which does not support virtualization. We have used two automotive state-of-the-art heterogeneous applications: engine electronic control unit and vehicle control unit, and a prototype version of hypervisor from ETAS-GmbH. Important aspects in virtualization like system-startup, handling of interrupts and traps, facilitating input–output access are discussed. Finally, the performance of a virtualized system in terms of core loading, interrupt and task timing parameters are evaluated and compared.