Fault tolerant multiprocessor computer system

Claim 1

In a fault tolerant multiprocessor computer system including: A) a plurality of central processing units with each central processing unit (CPU) including a store-into cache unit; B) a main memory; C) a system control unit, D) a system bus coupling the plurality of CPU's to each other and to the system control unit; and E) a memory bus coupling said system control unit to the main memory unit; said system control unit incorporating error detection and correction means for correcting a parity error in a block of data written into main memory from a store-into cache of a CPU transmitted therethrough; wherein the improvements comprise: F) each of the CPUs including means for issuing a siphon command to another CPU, the CPU issuing a siphon command, hereafter a requesting CPU, and the CPU to which the siphon command is issued, hereafter a sending CPU; a siphon command received by a sending CPU causing the sending CPU to read an identified block of data from the store-into cache of the sending CPU, said identified block of data being written into the store-into cache of the requesting CPU; G) providing the store-into cache unit of each of the CPUs with cache parity error detection means for detecting a parity error in a block of data read out of the store-into cache of a sending CPU, for producing a parity error signal if a parity error is detected in a block of data read out of the store-into cache unit of a sending CPU, for transmitting said identified block of data to the requesting CPU via the system bus, and for transmitting a parity error signal when produced via the system bus to the requesting CPU with said identified block of data; the requesting CPU upon the receipt of a parity error signal from the sending CPU suspending operation; and H) error recovery control means responsive to a parity error signal produced by the cache parity error detection means of the store-into cache unit of the sending CPU in response to the issuance of a siphon command by the requesting CPU for issuing to the sending CPU a swap command, the swap command causing the sending CPU to write the block of data identified by the siphon command which contains a parity error into main memory through the system control unit, said system control unit correcting the parity error in the identified block of data transmitted by the system control unit to main memory, hereafter a corrected block of data, prior to the corrected block of data being written into main memory, for reading the corrected block of data from main memory, for writing the corrected block of data into the store-into cache unit of the requesting CPU, and restarting the requesting CPU.