Nowadays, microcontrollers used in critical real-time embedded systems use mostly one core, but are being replaced with more powerful hardware platforms that implement multicore systems. Among the latter, it is possible to identify in the space domain, for instance, the Cobham Gaisler NGMP developed for the European Space Agency (ESA), which is built with a SPARC quad-core processor that has a two-level cache hierarchy. For what concerns automotive and avionics environments, very flexible platforms like the Zynq UltraScale+ EG one has been regarded as a very powerful platform for these high-performance safety-critical systems. In fact, the aforementioned Zynq board implements two multicore clusters, namely an ARM dual-core Cortex R5 and an ARM quad-core Cortex A53, as well as a GPU and an FPGA. Due to the industrial trend towards the deployment of autonomous driving in the automotive domain and unmanned vehicles in the avionics domain, boards with such multicore systems are very promising. The use of multicores brings a concern related to contention (interference) in the access to shared hardware resources, which challenges timing verification needed to prove that all critical real-time tasks will execute by their respective deadlines. In particular, Worst-Case Execution Time (WCET) estimates for tasks need to account for the impact in execution time that contention in shared resources may have. While such analysis has been performed on relatively-simple multicores, like the NGMP, it needs to be carried out on the more powerful and complex Zynq UltraScale+ EG platform. In particular, it is required to analyze the different sources of interference for the multicore clusters and how tasks need to be consolidated so that resource sharing is performed efficiently across tasks, thus minimizing the impact on execution time for the most critical real-time tasks. In this Master thesis work, the measurement-based methodology developed at Barcelona Supercomputing Center (BSC) to quantify the interference that arises across cores due to contention in shared hardware resources, is ported from the (simple) NGMP platform to each of the computing clusters of the Zynq UltraScale+ EG platform. Such methodology consists in the use of small microbenchmarks that aim at stressing specific shared hardware resources to create very high contention. Hence, this thesis investigates how to produce high contention in the shared hardware resources of the Zynq UltraScale+ EG platform, thus integrating those concepts working on the SPARC V8 instruction set of the NGMP to the ARM v7 and ARM v8 instruction sets of the Zynq platform. This requires porting and adapting microbenchmarks written partly in assembly code, verifying the Performance Monitoring Unit, and analyzing the sources of contention. As final step, guidelines are devised to properly consolidate software to be implemented on the target platform in order to contain as much as possible interference on critical tasks.
