Multiprocessor/Multimemory Control System

This invention relates to regulating access by multiple processors to multiple memory elements in a data processing system.

In a multi-memory, multi-processor data processing system, access by the processors to the memory elements must be regulated to avoid conflicts.

The invention features apparatus for regulating access by each of a plurality of data processors to each of a plurality of memories, each of the processors being associated with one of the memories, each of the processors needing access to its associated memory and to the memory associated with each other of the processors, the apparatus comprising local bus circuitry connected to selectably permit each of the processors to have, or to prevent each of the processors from having, access to its associated memory, connecting bus circuitry connected to selectably permit each of the processors to have, or prevent each of said processors from having, access to the memory associated with each other of the processors, and control circuitry connected to give each of the processors making a request access over the connecting bus circuitry to the memory associated with one other of the processors, and to give each of the processors access over the local bus circuitry to its associated memory except while access to the associated memory is being given to another of the processors.

In preferred embodiments, each of the processors has an associated internal control clock, and the control circuitry comprises hold triggering circuitry connected to stop said internal control clock of the processor making the request until the request has been satisfied, disconnect triggering circuitry connected to cause the processor whose associated memory is the subject of the request to disconnect itself from the local bus circuitry and to continue to operate its internal control clock to provide timing control pulses to the control circuitry, and timing circuitry connected to activate the connecting bus circuitry at least during part of the time when the internal control clock of the processor making the request is stopped and the processor whose associated memory is the subject of said request is disconnected from the local bus circuitry; each of the processors has a direct memory access pin, and a hold pin, and the control circuitry is connected to provide, when the request is made, a signal to the direct memory access pin of the processor whose associated memory is the subject of the request to cause the processor to disconnect itself from the local bus circuitry, and a signal to the hold pin of the processor making the request to cause the processor to wait until the request has been satisfied; arbitrate circuitry is included and is connected to cause the control circuitry to satisfy simultaneously issued requests of more than one of the processors in a predetermined order; the arbitrate circuitry comprises sampling circuitry connected to provide a signal when the simultaneously issued requests occur, bus enabling circuitry responsive to the sampling circuitry for satisfying the request of one of the processors, and reporting circuitry for signaling the one processor when its request has been satisfied to cause its associated memory to be made available to satisfy the request of another one of the processors; and there are two of said processors and two of said memories.

The invention enables each processor to operate independently at full speed, except when access to its associated memory is requested by another processor; each processor can directly and quickly access any other processor's memory; stalemates between simultaneously requesting processors are resolved; and each processor can perform any task using any of the memories, including diagnosis of failures in another processor.

Other features and advantages of the invention will be apparent from the following description of the preferred embodiment, and from the claims.

We turn now to the structure and operation of the preferred embodiment, first briefly describing the drawings thereof.

FIG. 1 is a block diagram of a two-processor, two-memory data processing system.

FIG. 2 is a block diagram of the two processors of FIG. 1 with representative circuitry (of the bus controller of FIG. 1) used to satisfy one processor's request for access to the other's memory.

FIG. 3 is a timing chart of signals corresponding to the steps in the receipt and processing of a request by one processor for access to the other's memory.

Referring to FIG. 1, each of microprocessors A and B is connected to its corresponding memory A or B (and its corresponding input/output port A or B) through its corresponding local bus A or B and through local transceivers A.sub.L and B.sub.L, and to the local bus of the other microprocessor through connecting bus 10. Conventional bidirectional transceivers A.sub.C and B.sub.C, which together open and close connecting bus 10, are connected by control line 12 to bus controller 14. Conventional bidirectional transceivers A.sub.L and B.sub.L respectively open and close local bus A and local bus B under the control respectively of microprocessors A or B, or of arbitrate logic 60. Bus controller 14 (which controls the receipt and execution of requests by each microprocessor to use the other's memory) is connected, as shown, to microprocessors A and B, conventional decoders A and B, and memories A and B.

Referring to FIG. 2, microprocessor A (needing access to memory B) is connected, as shown, to microprocessor B through flip-flops, gates and other circuitry which are part of bus controller 14. Decoder A is connected to provide a request signal on line 16 whenever an address appearing on local address bus A specifies a location in memory B. NAND gate 22 is connected to signal microprocessor A (on its memory ready (i.e., hold) input (MRDY)) to "stop" its internal control clock A beginning when a request signal appears and until the request has been satisfied.

Conventional J-K flip-flop 24 is arranged to signal microprocessor B on line 26 that a request signal has been received. Conventional D flip-flops 28, 30, 32 and gates 34, 36, 38 are connected, as shown, to provide control signals on line 12 to transceiver A.sub.C and B.sub.C to open and close connecting bus 10; to gate 22 to signal microprocessor A to restart its internal clock A when the request has been satisfied; and to AND gate 42 which provides control signals to enable memory B to be read or loaded (synchronously with internal clock B of microprocessor B) when connecting bus 10 is open. Flip-flops 24, 28, 30, 32 have their clock inputs connected to internal clock B of microprocessor B (which synchronizes the satisfying of microprocessor A's request) and their reset inputs to internal clock A of microprocessor A.

A second set of identical circuitry (not shown) in bus controller 14 receives and satisfies requests from microprocessor B for access to memory A. Arbitrate logic 60 (containing conventional circuitry including a gate, a D-flip flop and a free running oscillator) for sampling the lines to the MRDY inputs of memory processors A and B is connected as shown to line 12, gate 22, XCVR A.sub.L and XCVR B.sub.L to cause a request signal from microprocessor A to be processed first when both microprocessors simultaneously make requests.

Microprocessors A and B normally operate independently and without interruption at different processing speeds using their respective memories A and B.

Referring to FIGS. 2 and 3, when microprocessor A needs to use memory B, it issues a memory B address on local address bus A. Decoder A, recognizing the memory B address, begins to provide a request signal at time t.sub.1 over line 16. NAND gate 22 (triggered by the request signal) promptly stops providing a signal (which it had previously been sending) to the memory ready (MRDY) input of microprocessor A, thereby telling microprocessor A to wait until the request can be satisfied. Microprocesor A waits by fixing its internal clock A (which normally switches between high and low values periodically) at a high value beginning the next time the high value is reached. The timing of the remaining steps in satisfying the request is controlled by the pulses of internal clock B applied to the clock inputs of flip-flops 24, 28, 30, 32.

The request signal on line 16 causes flip-flop 24 to switch to a high state at time t.sub.2 (i.e., promptly after the next transition to a high state occurs on clock B) thereby telling microprocessor B (via line 26 to the DMA pin of microprocessor B) to release local bus B at the end of the current cycle of clock B. At time t.sub.3 (promptly after the next transition to a low state of clock B), microprocessor B disconnects itself from local bus B (by causing its address and data bus drivers to go to a high impedance) and issues a grant signal over line 18, indicating to bus controller 14 that local bus B is available and causing XCVR B.sub.L to disconnect microprocessor B from local bus B. At time t.sub.4 (i.e., immediately after the next transition of clock B to a low state), flip flop 32 changes to a high state (because of a signal received from AND gate 34) and triggers transceivers A.sub.C and B.sub.C (via line 12) to open connecting bus 10 to the transfer of an address and a piece of data. Then, at time t.sub.5 (i.e., upon the next transition of clock B to a high state), a signal is provided from gate 42 activating memory B to accept or provide the data, and the request signal is removed from line 16 by the transition of flip-flop 24 to a low state (triggered from the output of flip-flop 32), thereby permitting transfer of only one piece of data per request.

At time t.sub.6 (when clock B returns to a low state), gate 42 ceases to provide a memory B enabling signal, and the memory ready signal (MRDY) to microprocessor B is reestablished by gate 22 (because gate 38, having determined from flip-flops 28 and 30 that the request has been satisfied, ceases to provide a signal to gate 22), indicating to microprocessor A that its request has been satisfied and it can restart its internal clock A.

Then microprocessor B, having had its DMA input deactivated (by the state change of flip-flop 24 at time t.sub.5), withdraws the grant signal from line 18, indicating that it is ready to reassert control of local bus B. At time t.sub.7, the operation of clock A is reestablished by a down transition which causes flip-flops 24, 28, 30 and 32 to be reset. Flip-flop 32 and line 12 promptly change to a low state, triggering transceivers A.sub.C and B.sub.C to close connecting bus 10, ending the processing of microprocessor A's request. Microprocessors A and B then continue to operate using their respective memories A and B, until, when either microprocessor A or B requests access to the other's memory, the request is satisfied in the same manner.

Arbitrate logic 60 constantly samples the MRDY inputs of microprocessors A and B, and, when both are deactivated (indicating that a stalemate has resulted from simultaneous access requests issued by both processors), logic 60 sends a signal on line 12 to open connecting bus 10 and a signal to XCVR B.sub.L to disconnect local bus B to enable the data to pass between microprocessor A and memory B, and sends a signal to gate 22 which then signals microprocessor A that the request has been satisfied. Microprocessor B then has its request satisfied.

Other embodiments are within the following claims.