Debugging Distributed-Shared-Memory Communication at Multiple Granularities in Networks on Chip

We present a methodology to debug a SOC by concentrating on its communication. Our extended communication model includes a) multiple signal groups per interface protocol at each IP port, b) the handshakes per signal group (e.g. for command), and c) the handshakes within a signal group (e.g. for write and read data elements). As a result, our debug methodology is the first to offer debug control at three communication granularities: individual data elements in a message, messages (i.e. requests or responses), and entire transactions. Communication to distributed shared memories is supported in networks on chip (NOC) by transparently (demultiplexing different master-slave channels based on the memory address, also called narrowcast. In this paper, we extend previous work on NOC debug that allowed per-connection debug (i.e. a master without differentiating between its slaves) to also support per-channel (i.e. per master-slave pair) debugging, also for narrowcast connections. This enables essential fine-grained debug control for multi-processor SOCs that use distributed-shared-memory communication. The debug infrastructure consists of hardware components, and a software API and library. We define the hardware infrastructure and the required changes to a NOC. Our architecture cleanly separates the monitoring and distribution of events from how they are interpreted and used, in terms of hardware and programming. We define a high-level software API for run-time user control. The debug methodology offers run-time programmable breakpoints, stopping, continuing, and single-stepping of distributed- shared memory communication at three granularities, at the cost of 2.5% NOC area increase and no speed penalty.