Abstract: A method and apparatus for a large-way, symmetric multiprocessor system using a bus-based cache-coherence protocol is provided. The distributed system structure contains an address switch, multiple memory subsystems, and multiple master devices, either processors, I/O agents, or coherent memory adapters, organized into a set of nodes supported by a node controller. Each of the processors may have multiple caches. The address switch connects to each of the node controllers and to each of the memory subsystems, and each of the memory subsystems connects to the address switch and to each of the node controllers. The node controller receives commands from a master device and queues commands received from a master device. The node controller has a deterministic delay between latching a snooped command broadcast by the address switch and presenting the command to the master devices on the node controller's master device buses.

Abstract: A method and apparatus in a data processing system having a plurality of node controllers and a memory unit for each of the node controllers. Each one of the node controllers including at least one processor having a cache. Each memory unit including a plurality of entries each having an exclusive bit, an address tag, and an inclusion field. Each bit of the inclusion field representing one of the node controllers. The method and apparatus allow error recovery for errors occurring within the entries without using the ECC implementation. Specifically, two parity bits are used for detecting errors within an entry and logic for flushing any cache lines represented by the entry in error. The method and apparatus also includes means for detecting persistent errors and for indicating whether the error is generated by either hardware or software.