On-Chip Monitoring for Non-Intrusive Hardware/Software Observability

Sammanfattning: The increased integration of hardware and software components intoday's state-of-the-art computer systems make them complex and hardto analyse, test, and debug. Moreover, the advances in hardwaretechnology give system designers enormous possibilities to explorehardware as a means to implement performance demandingfunctionality. We see examples of this trend in novel microprocessors,and Systems-on-Chip, that comprise reconfigurable logic allowing forhardware/software co-design. To succeed in developing computer systemsbased on these premises, it is paramount to have efficient designtools and methods.An important aspect in the development process is observability, i.e.,the ability to observe the system's behaviour at various levels ofdetail. These observations are required for many applications: whenlooking for design errors, during debugging, during performanceassessments and fine-tuning of algorithms, for extraction of designdata, and a lot more. In real-time systems, and computers that allowfor concurrent process execution, the observability must be obtainedwithout compromising the system's functional and timing behaviour.In this thesis we propose a monitoring system that can be applied fornon-intrusive run-time observations of real-time and concurrentcomputer systems. The monitoring system, designatedMultipurpose/Multiprocessor Application Monitor (MAMon), is based on ahardware probe unit (IPU) which is integrated with the observedsystem's hardware. The IPU collects process-level events from ahardware-implemented Real-Time Kernel (RTK), without perturbing thesystem, and transfers the events to an external computer for analysis,debugging, and visualisation. Moreover, the MAMon concept alsofeatures hybrid monitoring for collection of more fine-grainedinformation, such as program instructions and data flows. We describe MAMon's architecture, the implementation of two hardwareprototypes, and validation of the prototypes in differentcase-studies. The main conclusion is that process level events can betraced non-intrusively by integrating the IPU with a hardware RTK. Asubsidiary conclusion, but yet relevant, is that the IPU's smallfootprint makes it attractive for SoC designs, as it providesincreased system observability for a low hardware cost.