Operating System Kernel Coprocessor for Embedded Applications

Abstract

The silicon evolution yields advances in contemporary processor architecture. As a result of the ever-increasing number of components in a chip, multi-core solutions have emerged. In general computing systems, their goal is to accommodate the parallel execution of processes, tasks or threads. Apart from general computing, the parallel execution of tasks is characteristic of asynchronous and dynamic embedded applications like automotive systems, process control, multimedia processing, security systems, etc., for which, in recent times, multi-core architecture has also raised interest [Lee (2010)]. However, in the case of processors for embedded systems and their ultimate requirement being predictability of temporal behaviour, the implementation of traditional multiprocessing is not straightforward. Their advanced architecture features (pipelines, cache, etc.), which are devised to improve average computing speed, may introduce severe sources of non determinism and unpredictability. Instead of symmetrical multiprocessing, it is more adequate to employ multi-core processors for specialised operations. One of these is the execution of operating system services with a goal to deal with the nondeterministic and unpredictable time delays caused by the very nature of asynchronous events by separating the execution of process control tasks from real time operating system (RTOS) kernel routines. This approach is similar to the idea of math coprocessors, graphical accelerators, intelligent peripherals, etc. These specialized units are able to perform operations much faster than general processors that implement them as software programs. The idea of migrating scheduling out of the main processor is already old [Halang (1988); Cooling (1993); Lindh et. all. (1998), etc.] However, with the advent of multi-core processors on one hand, and programmable hardware devices for prototyping on the other, its implementation has become much more feasible and realistic. In this contribution we are presenting a prototype of a separate Application specific integrated circuits (ASIC) implemented coprocessor performing operating system kernel functionalities. First, some background regarding the real-time properties of embedded systems is given, and some of the most characteristic solutions of real-time operating systems which jeopardise predictability are pointed out. Then, an architectural solution to the problem is proposed and validated with the prototype.