Virtual-Memory Multiprocessor System for Parallel Purge Operation

The present invention relates to a virtual memory control multiprocessor system having a plurality of processors.

In a conventional virtual memory control multiprocessor system of this type, a translation lookaside buffer (to be referred to as a TLB hereinafter) is arranged for each processor to perform address translation from a logical address to a real address at high speed. In a system of this type, purge (initialization) of TLBs of all processors must often be performed to equalize the contents of the TLBs.

As shown in FIG. 1, in a multiprocessor system having processors 10.sub.0 through 10.sub.3, assume that the processor 10.sub.0 supplies a TLB purge request to the processors 10.sub.1 through 10.sub.3. For this purpose, the processor 10.sub.0 supplies a purge request signal la to the processor 10.sub.3. The processor 10.sub.3 performs TLB purge processing in response to the signal 1a. In this case, the processor 10.sub.0 performs its own TLB purge processing while the processor 10.sub.3 performs TLB purge processing. When TLB purge processing of the processor 10.sub.3 is completed, the processor 10.sub.3 sends back a purge end signal 2a to the processor 10.sub.0. When the processor 10.sub.0 receives the signal 2a from the processor 10.sub.3, the processor 10.sub.0 sends a purge request signal 1b to the processor 10.sub.2. The processor 10.sub.2 performs TLB purge processing in response to the signal 1b and sends back a purge end signal 2b to the processor 10.sub.0. When the processor 10.sub.0 receives the signal 2b from the processor 10.sub.2, the processor 10.sub.0 then sends a purge request signal 1c to the next processor 10.sub.1. The processor 10.sub.1 performs TLB purge processing in response to the signal 1c and sends back a purge end signal 2c to the processor 10.sub.0. When the processor 10.sub.0 receives the signal 2c from the processor 10.sub.1, the processor 10.sub.0 determines that all TLB purge operations of the processors 10.sub.0 through 10.sub.3 are completed.

In the conventional multiprocessor system described above, the TLB purge operations of the processors are sequentially performed. Assume that the time required for a given processor to receive a purge request signal, perform purge processing in response to this signal, and send back a purge end signal to the processor which has supplied the purge request signal thereto is defined as T. In this case, in the system of FIG. 1, the total time for all the TLB purge operations of the four processors 10.sub.0 through 10.sub.3 is 3T, as shown in FIG. 2. In a multiprocessor for sequentially performing TLB purge operations, TLB purge time is linearly increased in proportion to the number of processors, resulting in inconvenience.

It is an object of the present invention to provide a multiprocessor system wherein the total purge time for translation lookaside buffers, with respect to the total processing time of the entire system, can be shortened.

According to the multiprocessor system of the present invention, all the processors including a purge request source processor can perform parallel purge operations. In addition, the end of the purge operation of each processor can be set by a corresponding flip-flop. By checking the statuses of the flip-flops, the end of the purge operations of all the processors can be detected. Therefore, the time for purge operations can be greatly shortened.

In order to achieve the above object of the present invention, there is provided a virtual memory control multiprocessor system having a plurality of processors each having a translation lookaside buffer, including:

purge request signal generating means for supplying a common purge request signal requesting purge operations of translation lookaside buffers of other processors excluding a source processor for generating the common purge request signal which is supplied in parallel to the other processors;

a plurality of flip-flops for storing, in units of processors, a purge end signal sent back from the other processors which complete purge operations in response to the common purge request signal generated from the purge request signal generating means; and

purge end detecting means for checking the statuses of the plurality of flip-flops and detecting the end of purge operations of all processors.

FIG. 1 is a schematic block diagram of a conventional multiprocessor system;

FIGS. 2A through 2B are respectively timing charts for explaining TLB purge operations of all processors shown in FIG. 1;

FIG. 3 is a schematic block diagram of a multiprocessor system to which the present invention is applied;

FIG. 4 is a circuit diagram of a circuit of the processor system (FIG. 3) which is directly associated with the present invention; and

FIGS. 5A through 5D are respectively timing charts for explaining TLB purge operations of the processors shown in FIG. 3.

FIG. 3 shows a virtual memory control multiprocessor system according to an embodiment of the present invention. Referring to FIG. 3, processors 20.sub.0 through 20.sub.3 have translation lookaside buffers (TLBs), respectively. The processors 20.sub.0 through 20.sub.3 are connected to each other through a system bus 21. The processors 20.sub.0 through 20.sub.3 receive and send a purge request signal a through a signal line 22 and a purge end signal b through a signal line 23. The lines 22 and 23 shown in FIG. 3 constitute part of the system bus but are separately illustrated for descriptive convenience.

FIG. 4 shows an arrangement of the processor 20.sub.i (i=0 through 3) of FIG. 3 which is directly associated with the present invention. A bus driver 31 sends the signal a onto the line 22. A bus receiver 32 receives the signal a from the line 22. A bus driver 33 sends the signal b onto the line 23. A bus receiver 34 receives the signal b from the line 23. A bus driver 35 sends a processor number including an address to an address bus of the bus 21. A bus receiver 36 receives address data from the address bus of the bus 21. The address data comprises, for example, a processor number, a command and auxiliary data. The signal b is included in the command.

A decoder 37 decodes a predetermined field (i.e., processor number) of the address fetched by the receiver 36. Decode signal lines 38.sub.0 through 38.sub.3 correspond to processor numbers of the processors 20.sub.0 through 20.sub.3, respectively. Each of the lines 38.sub.0 through 38.sub.3 is connected to one input terminal of a corresponding one of 2-input NAND gates 39.sub.0 through 39.sub.3. The signal b is supplied to the other input terminal of each of the NAND gates 39.sub.0 through 39.sub.3 through the receiver 34. Output signals from the NAND gates 39.sub.0 through 39.sub.3 are supplied to set input terminals S of R-S flip-flops 40.sub.0 through 40.sub.3, respectively. A signal 41 is supplied to a reset input terminal R of each flip-flop 40.sub.i (i=0 through 3) to reset it. Q outputs from the flip-flops 40.sub.i are supplied to 3-input OR gates 42.sub.0 through 42.sub.3, respectively. The OR gates 42.sub.0 through 42.sub.3 also receive signals 47.sub.0 through 47.sub.3 representing that the processors generating the signals a are the processors 20.sub.0 through 20.sub.3 and signals 48.sub.0 through 48.sub.3 representing that the processors 20.sub.0 through 20.sub.3 are not present. Outputs from the OR gates 42.sub.0 through 42.sub.3 are supplied to a 4-input AND gate 43. An output signal 45 from the AND gate 43 is used as the purge end signal which is received by the processor which has generated the signal a.

The operation of the multiprocessor system of this embodiment will be exemplified wherein the processor 20.sub.0 sends the TLB purge request signal a to the remaining processors 20.sub.1 through 20.sub.3, as shown in FIG. 3, with reference to the timing charts of FIGS. 5A through 5D.

The processor 20.sub.0 supplies the signal a onto the line 22 through the driver 31 in the processor 20.sub.0. The signal a on the line 22 is commonly supplied to the processors 20.sub.1 through 20.sub.3 and is fetched thereby through the corresponding bus receivers 32 in the processors 20.sub.1 through 20.sub.3. The TLB purge operations are simultaneously performed in response to the signal a commonly fetched by the processors 20.sub.1 through 20.sub.3, as shown in FIGS. 5A through 5D. The TLB purge operation of the source processor 20.sub.0 is also performed while the processors 20.sub.1 through 20.sub.3 perform the corresponding TLB purge operations after the processor 20.sub.0 sends the signal a thereto.

When the processors 20.sub.1 through 20.sub.3 complete the TLB purge operations, they send back the signals b onto the line 23 through the corresponding bus drivers 33 if the bus 21 is available. In this case, the processors 20.sub.1 through 20.sub.3 send their own processor numbers onto the address bus of the bus 21 through the corresponding bus drivers 35. The processor number data on the system bus 21 are supplied to the decoder 37 in the source processor 20.sub.0. The decoder 37 decodes the processor numbers supplied through the receivers 36. When the processor number data supplied to the decoder 37 of the processor 20.sub.0 represents that from the processor 20.sub.1, the decoder 37 sends an active signal of logic "1" onto the signal line 38.sub.1. This signal is supplied to one input terminal of the NAND gate 39.sub.1. The other input terminal of the NAND gate 39.sub.1 receives the signal b sent from the processor 20.sub.1 through the line 23 and the corresponding receiver 34. The NAND logic of the NAND gate 39.sub.1 in the processor 20.sub.0 is established, and an output signal therefrom goes from logic "1" to logic "0". This transition indicates that the signal b from the processor 20.sub.1 is detected by the NAND gate 39.sub.1.

The output signal from the NAND gate 39.sub.1 is supplied to the set terminal S of the flip-flop 40.sub.1. The flip-flop 40.sub.1 is set when the output signal from the NAND gate 39.sub.1 is inverted from logic "1" to logic "0" and the set input terminal of the flip-flop 40.sub.1 is inverted from logic "0" to logic "1". In other words, the flip-flop 40.sub.i stores the purge end signal sent from the processor 20.sub.1. The above operation can also be applied when the signals b and the corresponding processor numbers are sent from the processors 20.sub.2 and 20.sub.3. More particularly, when the processor 20.sub.2 supplies the signal b and its own processor number to the processor 20.sub.0 through the line 23 and the bus 21, the flip-flop 40.sub.2 in the source processor 20.sub.0 is set. Similarly, when the processor 20.sub.3 supplies the signal b and its own processor number through the line 23 and the bus 21 to the processor 20.sub.0, the flip-flop 40.sub. 3 in the source processor 20.sub.0 is set.

When the flip-flops 40.sub.1 through 40.sub.3 are set, their Q output signals 46.sub.1 through 46.sub.3 go from logic "0" to logic "1". The Q output signals 46.sub.1 through 46.sub.3 are supplied to the OR gates 42.sub.1 through 42.sub.3, respectively. As a result, the OR gates 42.sub.1 through 42.sub.3 are rendered conductive, and their output signals 44.sub.1 through 44.sub.3 go from logic "0" to logic "1", respectively.

Since the processor 20.sub.0 serves as the TLB purge request source processor, it does not generate the signal b and its own processor number. As a result, the flip-flop 40.sub.0 in the processor 20.sub.0 is kept reset, and the Q output signal of logic "0" therefrom is supplied to the OR gate 42.sub.0. Other terminals of the OR gate 42.sub.0 receive the signal 47.sub.0 representing that the processor generating the signal a is the processor 20.sub.0 and the signal 48.sub.0 representing that the processor 20.sub.0 is not present. In this embodiment, since the processor 20.sub.0 serves as the TLB purge request source processor, the signal 47.sub.0 of logic "1" is supplied to the OR gate 42.sub.0. As a result, the output signal 44.sub.0 from the OR gate 42.sub.0 is set at logic "1".

The output signals 44.sub.0 through 44.sub.3 from the OR gates 42.sub.0 through 42.sub.3 are supplied to the AND gate 43. The AND gate 43 generates a signal of logic "1" when all the signals 44.sub.0 through 44.sub.3 are set at logic "1". As a result, the TLB purge request source processor 20.sub.0 checks the output signal 45 from the AND gate 43. When this signal is set at logic "1", the processor 20.sub.0 determines that all the TLB purge operations of the processors 20.sub.0 through 20.sub.3 are completed, thereby initiating the next operation.

The above operation can be performed when all the processors 20.sub.0 through 20.sub.3 are present. However, when the processor 20.sub.3 is not present, the following operation is performed. The signal 48.sub.3 representing that the processor 20.sub.3 is not present is set at logic "1". The signal 48.sub.3 of logic "1" is supplied to the OR gate 42.sub.3. The OR gate 42.sub.3 supplies the dummy TLB purge end signal to the AND gate 43. If the processor 20.sub.3 is not present, the processor 20.sub.0 can detect that all the TLB purge operations of the processors 20.sub.0 through 20.sub.2 are completed.

In the above embodiment, the maximum number of processors is four. However, the present invention can be applied to a multiprocessor system having any number of processors.

In the above embodiment, the purge request signal transfer signal line 22 and the purge end signal transfer signal line 23 are arranged. However, these signal lines can be omitted. For example, a purge request code corresponding to the signal a and a purge end code corresponding to the signal b can be provided, and these code data can be transferred through a predetermined field of the address bus in the bus 21. In this case, each processor must have decoding means for decoding the purge request code and the purge end code.