摘要:
In a multiprocessor system having a plurality of bus devices coupled to a storage device via a bus, wherein the plurality of bus devices have a snoop capability, and wherein the plurality of bus devices have first and second caches, and wherein the plurality of bus devices utilize a modified MESI data coherency protocol, the system provides for reading of a data portion from the storage device into one of the plurality of bus devices, wherein the first cache associated with the one of the plurality of bus devices associates a special exclusive state with the data portion, and wherein the second cache associated with the one of the plurality of bus devices associates an exclusive state with the data portion, initiating, by the one of the plurality of bus devices, a write-back operation with respect to the data portion, determining if there are any pending snoops in the second cache, and changing the special exclusive state to a modified state if there are no pending snoops in the second cache. If there is a pending snoop in the second cache, a comparing of addresses of the pending snoop and the data portion is performed. The special exclusive state is changed to a modified state if the addresses are different. The special exclusive state indicates that a data portion is held in the primary cache in a shared state and that the data portion is held in the secondary in an exclusive state.
摘要:
A data processor has a block-multiplexed system channel coupled to a processing engine and a byte-multiplexed bus coupled to multiple I/O devices. A multi-buffer adapter transfers data by cycle-steal (direct memory access) operations between the channel and the bus. The adapter has multiple buffers switchable to the channel in a burst mode by a channel interface and to the bus in a byte mode by a device-level interface.
摘要:
In a multiprocessor system, when a store request has stalled, a signal is generated and sent to all processors indicating such a stalled store situation. In response, all processors will postpone the sending of load, or read, requests to memory until the stalled store request has completed.