Title	
Synthesis of Optimized AUTOSAR Embedded Systems: Automated System-Call Specialization and Lock Elision on Multicore Applications as a Whole-System Approach

Authors	
Andreas Kässens

E-Mail	
kaessens@sra.uni-hannover.de

Affiliation	
Leibniz Universität Hannover

Abstract	

Embedded real-time systems in safety-critical areas, such as in the automotive and aerospace industries, have particularly high software requirements regarding safety and reliability. In the automotive sector, the operating system standard AUTomotive Open System ARchitecture (AUTOSAR) has been established as a common basis among manufacturers and suppliers to meet these criteria. With priority-based real-time scheduling and static configuration, AUTOSAR- compliant operating systems are well suited for deployment in real-time systems. However, there is an aim to further optimize these software systems to improve their non-functional characteristics, including dependability, memory consumption, and runtime delays. In this thesis, I present the development of the Multicore AUTOSAR Compatible Application- specific Whole-system-optimizer (MACAW) system generator that supports the key interfaces of the AUTOSAR operating system specification. Using the Automated Real-time system Analyzer (ARA) framework, previous work developed the MultiSSE, a static code analysis to obtain a graph enumerating and connecting all possible system states. By automatically detecting certain interaction patterns between operating system objects in this graph, the performance of operating system calls can be optimized. Costly actions such as Inter-Processor Interrupts (IPIs) or spinlock operations can be eliminated if they do not affect subsequent system states. Spinlock operations that are never executed by two processor cores simultaneously can be omitted without affecting functionality, and IPIs that do not affect scheduling on the target core can be skipped. MACAW is implemented as the synthesis step within the ARA framework for real-time appli- cations and currently supports the POSIX platform. At compile time, the detected optimizations will be automatically applied to a specialized system call variant for each call site. Evaluation of available test applications shows that about 28 % of the affected cross-core system calls can be optimized. In particular, the avoidance of IPIs leads to a measurable reduction in delays and jitter.

Language of the Presentation	
German