MARTe2 on ARM platforms integration challenges: An asymmetric multiprocessing approach for the ITER magnetics diagnostics

The International Thermonuclear Experimental Reactor (ITER) magnetics diagnostic system will exploit high performance signal processing, network streaming and telemetry capabilities. Under this perspective, the adoption of the Multithreaded Application Real-Time Executor (MARTe2) control framework appeared as the most performing solution. To this aim, MARTe2 must be deployed over specific System-on-Module (SoM) platforms. In this paper, we focus on the porting of MARTe2 for the ARM® processor architecture, since it allows MARTe2 to run on a vast range of embedded platform. A series of further integration steps with respect to the state-of-the-art must be undertaken to deploy and run the complete solution. The paper will present the implementation and integration steps followed, giving also context about the application and motivations on the choice. As the target platform uses a quite standardized reference design, the described procedure can be adapted to a large number of System-on-Module based embedded boards. Specific platform adaptations, achieved using custom-developed components, offer another foundation for the further integration of the framework. Moreover, the ITER magnetics diagnostic system is required to run in real-time, therefore the implementation leverages the Asymmetric Multi-Processing (AMP) approach. AMP, in contrast with Symmetrical Multi-Processing (SMP), enables the central processing unit to run each core isolated from the others, without any operating system or orchestration layer in-between. Considering the strict quality assurance and continuous integration constraints required by the MARTe2 framework development process, the additional build system will be fully described. The described approach is based on the generic MARTe2 framework, ported on widespread ARM® architecture, and it could be easily repurposed for all scenarios where similar real-time signal acquisition, processing and streaming are required, thus enabling the rapid deployment of real-time control platforms with off-the-shelf components.