Intermediary inclusive caches (IICs) translate between some number of processors using virtual addressing and a physically addressed bus. The IICs support at least one virtual bus (upper bus) connecting the IICs to central processor units (CPUs), and at least one physical bus (lower bus) connecting the IICs to a memory controller, input/output (I/O) devices and perhaps other IICs. Whenever a CPU makes a request of memory (on the upper bus), the request is looked up in an IIC. Should the data reside in the IIC, the data is provided to the CPU from the IIC through the upper bus (except in the case of coherency filters which do not cache data). If the request misses the IIC, the request is repeated on the lower bus. When the requested data comes back from the lower bus, the data is cached in the IIC and passed up to the requesting CPU through the upper bus. Whenever a snoop request comes in from the lower bus, the snooped (requested) data is looked up in the IIC. Should the snoop miss the IIC, that is the requested data is not in the IIC, the request need not be repeated on the upper bus. In the case of the snoop hit on the IIC, the snoop may be repeated on the upper bus if a coherency protocol requires.
