During reconfiguration, the NoC supports built-in quiescing for the NoC paths where the NMU is in the dynamic region and the NSU is in the static region. This means that the startup and shutdown sequence of the NoC traffic is embedded within the partial PDI, and you do not need to instantiate IP, such as the DFX AXI Shutdown Manager IP or DFX Decoupler IP in the PL. However, you must oversee AXI transactions to and from the RP during reconfiguration to avoid NoC timeout errors. It is the designer's responsibility to pause traffic to and from a Reconfigurable Partition while it is being reconfigured.
The NoC facilitates dynamic ownership of its paths across DFX partitions. This allows subsequent RMs to incorporate additional NMUs in the dynamic region to service specific NSUs in the static region with higher QoS requirements. Consequently, the NoC compiler might offer solutions where physical channels of the NoC are shared between locked static paths and dynamically owned paths. This competition for NoC resources within the same physical channel between AXI transactions in the static region and NoC resources from the dynamic region can introduce unintended delays in static region functionality. You can prevent this scenario by applying the EXCLUSIVE_ROUTING_GROUP constraint to ensure that two NoC networks do not use the same physical channels.
In multiple-RP designs, AMD recommends that you maintain ownership of boundary NoC paths within the static region. This guarantees that the QoS parameters and physical paths of these boundary NoC topologies remain fixed and unaffected by changes from the dynamic region. Allowing dynamic ownership of boundary NoC paths in multiple-RP designs can lead to overlapping NoC solutions due to the absence of enforced routing exclusivity.