JP2000082052A - Multiprocessor system, its diagnostic method, and recording medium recorded with its control program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an error detection system which does not lower the reliability of diagnostic processes of individual processors in a multiprocessor system. SOLUTION: When an IOP(input/output processor) 1 performs a normal process and reaches a diagnostic point, the IOP 1 starts a process for diagnosis. Namely, the IOP 1 stores the last calculation contents and its current calculation result in a storage means and refers to flags in an idle flag function part showing that no other IOP is performing a process. The last calculation contents are set to, for example, an IOP 2 whose flat is set. The IOP 2 having received them performs calculation according to the received calculation contents and sends the calculation result back to the IOP 1. The IOP 1 stores the sent-back calculation result in the storage means 19 and compares it with the calculation result in the storage means. When both the calculation results are equal to each other, it is considered that neither processor is malfunctioning and the diagnosing process is ended.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multiprocessor system, an error detection method therefor, and a recording medium storing a control program therefor, and more particularly to an error detection method in a multiprocessor system in diagnostic processing.

[0002]

2. Description of the Related Art In a system having a plurality of general-purpose processors such as a supercomputer and a large general-purpose machine, there are mainly a multiplexing method and a patrol method as a technique for detecting an error of the processor itself. The former multiplexing method is to make a plurality of processors perform the same processing, collate the processing results as needed, and perform disconnection failure processing from the system if they do not match. In this method, the reliability increases, but the utilization efficiency of the processor is poor, and the amount of hardware increases.

In the latter patrol system, several diagnostic points are set in a program of a processor, a diagnosis result is reported to an upper processor, and collation is performed by the upper processor. In this method, the diagnosis processing is fixed, and the reliability of the diagnosis decreases. These two conventional methods have advantages and disadvantages in terms of the amount of hardware and reliability.

Japanese Patent Laid-Open Publication No. Hei 6-28328 proposes a multiprocessor self-diagnosis system as shown in FIG. In FIG. 5, the multiprocessor system includes a central processing unit (CPU) 1 which is a main component of the system.
1. Additional processing unit (ADP) 41, peripheral control unit (PC)
L) 42, 43, these devices (processors) 11, 4
Interface bus (BU) connecting between
S) 10.

The CPU 11 has an expected value register (DRR) in which an expected value of a diagnosis result for performing self-diagnosis is stored in advance.
56, a result holding register (RSR) for holding a diagnosis result
58, expected value stored in DRR 56 and RSR 58
And a comparison analysis circuit (CMP) 57 for comparing and analyzing the diagnosis result stored in the memory. Also, a writable microprogram control storage device (WOS) 60 for storing a normal microinstruction, a self-diagnosis microprogram storage device (DME) 61, a normal mode for executing a normal program, and a self-diagnosis microprogram are executed. Mode switching circuit (M
DS) 53.

Further, according to the instruction of the MDS 53, the WOS6
0, any one of the data of the DME 61 and a multiplexer (MPX) 54 for outputting to the data register (DTR) 51, a data register (DTR) 55 for temporarily storing the output of the MPX 54, and a data processing for executing the output data of the DTR 55 The device (DTP) 51 is provided. Furthermore, a microprogram controller (MPC) 52 for controlling the MDS 53 in accordance with an instruction from the DTP 51,
3. WOS60, DME6 by instruction from DTR55
1 is provided with a next address register (ADE) 62 for giving next address information, and a bus interface circuit (BIF) 59 for interfacing with the BUS 10.

At the time of executing the self-diagnosis, the MPC 52
Controls the MPX 54 by controlling the
The next address information from S53 is provided to the DME 61 via the ADR 62, and the self-diagnosis microprogram of the DME 61 is transmitted to the DTP 51 via the MPX 54 and the DTR 55.
And the DTP 51 executes the diagnostic instruction. DTP
When the diagnostic calculation is performed in the failure state of 51, the result information of the self-diagnosis is stored in the RSR 58, and the contents of the DRR 56 preset by the DTP 51 and the contents of the RSR 58 are compared and analyzed by the CMP 57.

[0008]

SUMMARY OF THE INVENTION FIG.
In the case of the technique proposed in Japanese Patent No. 28328, there is a problem that the reliability of the diagnostic processing is reduced. That is, in this proposal, the expected value of the diagnostic processing is fixed by the diagnostic microprogram stored in the DME.
Since the content of the diagnosis is the same every time, the diagnosis other than the portion used in the calculation at the time of the diagnosis is not performed.
In addition, there is a problem that the performance of the entire system is reduced. That is, when the diagnosis is performed, the normal mode of performing the original work of the processor to be diagnosed is completely stopped, and the diagnosis processing is performed by switching to the diagnosis mode by the mode switching circuit. No normal processing is performed from the start to the end of the diagnostic processing.

An object of the present invention is to provide a multiprocessor system which does not reduce the reliability of diagnostic processing, an error detection method thereof, and a recording medium on which a control program for the method is recorded.

[0010]

According to the present invention, there is provided a multiprocessor system comprising a plurality of processors interconnected via a bus and each performing a calculation process, wherein each of the processors is Free time display means for displaying the free time of the calculation processing, and storage means for storing the contents of the calculation processing immediately before the predetermined diagnosis point in the course of the calculation processing and the calculation result thereof Transmitting means for transmitting the calculation processing content of the storage means as a destination processor by referring to the free time display means of another processor when the diagnosis point is reached, and a transmission source processor Means for receiving the calculation processing content, which is the transmission content from the transmission means, and performing calculation processing according to the calculation processing content; Means for returning to the serial transmitting processor, and comparing means for comparing the calculation result of the calculation result and said storage means has been returned from the destination processor,
And a multiprocessor system characterized by having the following.

Each of the processors transmits the calculation processing contents to the transmission destination processor, and then shifts to a normal processing.
The processor further includes an upper-level diagnosis unit, and each of the processors reports the calculation processing content and the calculation result thereof and the calculation result from the transmission destination processor to the diagnosis processor when the comparison unit detects a mismatch. The diagnostic means is characterized in that the diagnostic means performs a calculation process in accordance with the content of the calculation process, and compares the calculation result with the calculation result of the processor.

Further, according to the present invention, there is provided a method for diagnosing a multiprocessor system comprising a plurality of processors interconnected via a bus and each performing a calculation process, wherein the idle time of the calculation process is displayed. A display step, a storage step of storing the content of its own calculation processing immediately before reaching a predetermined diagnosis point in the course of the calculation processing and the calculation result, and A transmitting step of transmitting the calculation processing content of the storing step with the processor indicating the vacancy by referring to the vacant time display of another processor as a destination processor, and the calculation being the transmission content of the transmitting step of the transmitting processor Receiving the processing content and performing a calculation process in accordance with the calculation content; and transmitting the calculation result to the transmission source. Returning to the processor;
A diagnostic method is provided, comprising: a comparison step of comparing the calculation result returned from the transmission destination processor with the calculation result of the storage step.

Further, according to the present invention, there is provided a recording medium storing a control program of a diagnostic method for a multiprocessor system which includes a plurality of processors which are interconnected via a bus and each perform a calculation process. The program stores a display step of displaying a free time of the calculation processing, and, when reaching a predetermined diagnostic point in the course of the calculation processing, the content of its own calculation processing immediately before the diagnosis point and the calculation result. A storage step, a transmission step of sending the calculation processing content of the storage step as a destination processor by referring to the idle time display of another processor when the diagnostic point is reached, The content of the calculation process, which is the content of the transmission by the transmission step of the processor, is received, and the content is calculated according to the content of the calculation process Performing a process, the recording medium is obtained which is characterized by having, a comparison step of comparing a calculation result of the calculation result and said storing step the calculation result sent back from the transmitting processor.

The operation of the present invention will be described. First, the problem that the reliability is reduced by fixing the expected value is solved by diverting the processing performed immediately before by the processor. That is, the processor to be subjected to the diagnosis processing usually performs some processing in a multiprocessor system. As the content of the question to be used for the diagnosis, for example, an address calculation or the like performed by the processor immediately before the diagnosis is used. The other processor in the multiprocessor system performs the calculation that the user has already performed, and diagnoses by comparing the calculation result of the user with the calculation result of the other processor. In order to enable this, instead of the expected value register and the result holding register of the conventional example shown in FIG. 5, a question content holding means for holding the immediately preceding calculation performed by the user and the self are included. It has means for holding the calculation results of each processor.

In order to further increase the reliability of diagnosis, the system has a diagnosis processor that specializes in diagnosis processing. If a malfunction is detected by the diagnosis of the processor itself to be diagnosed, the malfunctioning processor is identified by re-calculating the malfunction by the diagnosis processor.

Next, in order to avoid a decrease in performance due to the diagnosis processing, the diagnosis processing is performed using the idle time of the processor. In a multiprocessor system, each processor is not always working. For example, in the case of an input / output processor, the time during which data input / output between the upper system and the lower auxiliary storage device is not performed is usually idle time. Diagnosis processing is performed using such time,
Avoid degradation of system performance. In order to make this possible, an idle flag indicating a use state of each processor to be diagnosed is newly provided. When the idle flag is raised, it indicates that the processor is idle time waiting for the next instruction.

Each processor starts processing after lowering its own idle flag when trying to perform processing. To end the processing, the idle flag is raised and the processing ends. As a result, it is possible to know the use state of each processor by looking at the idle flag, and it is possible to know the processor that can perform the diagnostic processing that is the idle time.

In the conventional example shown in FIG. 5, each processor controls its own diagnosis and makes a diagnosis on the basis of its own expected value. The processors mutually control the diagnosis, and diagnose a plurality of processors in one diagnosis without fixing the expected value of the diagnosis. In addition, a diagnostic processor is provided as an upper processor in the system, and when a malfunction is detected, the diagnostic processor identifies the processor in which the malfunction has occurred, thereby improving the reliability of diagnosis.

[0019]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an embodiment of a multiprocessor system according to the present invention, and the same parts as those in FIG. 5 are denoted by the same reference numerals. In FIG. 1, a multiprocessor system according to the present invention includes a central processing unit (CPU) 11 for controlling the entire system, and a main storage device (MM) for storing system control programs and the like.
U) 12, a diagnostic processing device (DGP) 13 for failure processing,
Input / output processors (IOPs) 1 to n for controlling input / output operations with peripherals and terminal devices, etc .;
Bus line (BUS) 10 for connecting between 11 and 13
It is composed of

The processor to be subjected to error detection (diagnosis) in the multiprocessor system according to the present invention is:
Input / output processors (IOPs) 1 to n. IOP1
n controls input / output operations between the CPU 11, the MMU 12, and the like, a peripheral terminal device such as a disk device and a tape device, and an external network or the like (neither is shown). A DGP (DiaGnostic Processing unit) 13 performs fault processing for the entire system, and includes IOPs 1 to n.
Performs fault processing when detects a malfunction. These processors and devices 1 to n and 11 to 13 can perform interprocessor communication via the bus 10.

FIG. 2 is a detailed functional block diagram of input / output processors (IOPs) 1 to n to be actually diagnosed.
The normal processing function part 15 includes an input / output processor (IOP).
It controls the input / output of the system, which is the primary function of 1 to n, and performs arithmetic processing such as interprocessor communication with other processors and address calculation. In the multiprocessor system according to the present invention, the diagnostic processing is performed using the function of the normal processing function part 15.

The idle flag function part 16 indicates the use state of the processor. The idle flag portion 16 is set by the normal processing function portion 15. The other processor can read the idle flag portion 16 via the bus 10 to know the current use state of the processor. The question content storage unit 17 stores the content of the calculation performed by itself for use in the diagnostic processing as a diagnostic question, and is read and written by the normal processing function unit 15. The calculation result at the time of performing the calculation is stored in its own calculation result holding unit 18 by the normal processing function unit 15. The calculation result holding means 19 of the other processor stores the result obtained when the other processor performs the calculation stored by the question content storage means 17.
The data is stored by the normal processing function unit 15 via the inter-processor communication function of the bus 10 and the normal processing function unit 15.

The operation of the embodiment of the present invention will be described with reference to the flowcharts of FIGS. 3 and 4 are flowcharts showing the operation flow of the input / output processors 1 to n in the present embodiment. In the present embodiment, the input / output processors 1 to n to be subjected to error detection are controlled by a microprogram as a normal operation. In this embodiment, a plurality of diagnostic points are set in the microprogram, and a diagnosis is performed when the operation of the input / output processor reaches the diagnostic point. In this embodiment, a diagnostic point is set at the end of the normal operation.

The calculation performed immediately before the diagnosis point is stored as a diagnosis question by the question content storage unit 17, and the calculation result is stored by its own calculation result storage unit 18. Each of the input / output processors 1 to n waits for an input / output operation command by inter-processor communication from the CPU 11 of the system as a normal operation, and operates in accordance with its own microprogram.

In FIG. 3, when any operation is started by the microprogram, first, the normal processing function part 15 is used to lower the idle flag 16 to indicate that the user is no longer idle (step 31). Next, the normal processing function part 1 describes what operation is to be started.
5 is determined (step 32). When the operation to be started is the normal processing by the instruction from the CPU 11, the input / output processors 1 to n perform the processing received by the normal processing function part 15 (step 33).

The normal processing is performed, and the own micro program reaches the diagnosis point (YES in step 34).
In this case, the input / output processors 1 to n start a process for diagnosis (process A, see FIG. 4) (step 35). In FIG. 4, the processor that starts the diagnosis is, for example, IOP1, and the IOP1 stores the contents and results of the calculation performed in the immediately preceding normal processing by the question content storage unit 17 and its own calculation result storage unit 18. (Step 21).

Next, the other microprocessors IOP2
n idle flags are sequentially read out via the bus 10,
Search for a processor whose state is idle time (step 2
2). If the idle flags of all the other processors IOP2 to IOPn are not idle time (NO in step 22), the diagnostic processing is abandoned and terminated. For example, if the status of IOP2 is working, IOP1 is
Give up the diagnosis by 2. Next, for example, the idle time of IOP3 was checked, and the idle time was found (step 22).
If the answer is YES), the diagnosis process is started with the IOP3 as a diagnosis partner.

The processor IOP1 which starts the diagnosis notifies the processor IOP3 to be diagnosed of the start of the diagnosis via the bus 10. Then, the question content storage means 17
Is transmitted via the bus 10 (step 23). After telling IOP3 the content of the questions, I
OP1 terminates the diagnosis questioning process and suspends the diagnosis. Here, by temporarily interrupting the diagnosis processing, a decrease in the utilization efficiency of the IOP1 due to waiting for the reception of a response from the IOP3 is avoided. IOP3 which received the contents of the question performs the same calculation,
IOP1 which started the diagnosis via the bus 10 based on the calculation result
To answer.

Returning to FIG. 3, since the IOP 1 receives the communication between the processors from the IOP 3 and the content is the reception of the calculation result (step 32 is the reception of the calculation result), the normal operation function part 15 of the IOP 1 (Step 36).
The IOP 1 reads out the contents of its own calculation result storage means 18 and the contents of the calculation result storage means 19 of another processor by the normal function part 15 and performs a comparison operation (step 3).
7). As a result of the comparison, if both calculation results are the same (YES in step 37), it is diagnosed that no malfunction has occurred in both processors, the idle flag is raised (step 39), and the diagnosis process is terminated. .

If there is a difference between the two calculation results (NO in step 37), it can be determined that a malfunction has occurred in any of the processors. In this case, the diagnostic processor (DGP) 13 of the higher-level system performs the fault processing. Therefore, the occurrence of the fault is notified to the DGP 13 via the bus 10, the question content storage unit 17, and the calculation result of each processor is stored. The contents of the means 18 and 19 are also notified to the DGP 13 (step 38). The DGP 13 receives the diagnosis content and the calculation result of each processor only when a failure occurs, and performs the same calculation on the content.

The diagnosis by the DGP 13 is performed by comparing the received calculation result of each processor with the calculation result of the DGP 13,
It is determined that the matched processor is performing a normal operation, and the mismatched processor is determined to have a malfunction, thereby identifying the processor in which the malfunction has occurred. D for the malfunctioning processor
GP13 performs failure processing such as disconnecting from the system,
Avoid failures due to processor malfunctions affecting the entire system.

[0032]

As described above, in the present invention, an input / output processor (IOP) for controlling input / output to a system and a diagnostic processor (DGP) for controlling system diagnostic processing are provided separately from an arithmetic processor (CPU). In such a system, processors with dedicated functions perform processing according to the processing contents, and multiple processors with similar functions can be provided, so that multiple operations can be performed at the same time. This aims to speed up the entire system.

According to the present invention, there is an effect that it is possible to perform an efficient diagnosis using the idle time between a plurality of processors performing the same operation. Further, according to the present invention, in the diagnostic processing of the multiprocessor system, there is an effect that a load on the diagnostic processor is reduced and a malfunction due to a failure of the processor itself can be detected.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a detailed block diagram of an input / output processor according to the embodiment of the present invention.

FIG. 3 is an operation flowchart of the embodiment of the present invention.

FIG. 4 is an operation flowchart of the embodiment of the present invention.

FIG. 5 is a block diagram showing an example of a conventional self-diagnosis method of a multiprocessor system.

[Explanation of symbols]

 1 to n input / output processor 10 bus 11 CPU 12 main storage device 13 diagnostic processing device

Claims (9)

[Claims] 1. A multiprocessor system comprising a plurality of processors interconnected via a bus and each performing a calculation process, wherein each of said processors displays a free time of said calculation process. Display means, storage means for storing the contents of its own calculation processing immediately before reaching a predetermined diagnosis point in the course of the calculation processing and the calculation result, and when the diagnosis point is reached Transmitting means for transmitting the calculation processing contents of the storage means with the processor indicating the vacancy by referring to the vacant time display means of the other processor as the destination processor; and the transmission contents from the transmitting means of the transmission source processor. Means for receiving the calculation processing content and performing the calculation processing according to the calculation processing content; and returning the calculation result to the transmission source processor. And a comparing means for comparing the calculation result returned from the transmission destination processor with the calculation result of the storage means. 2. The multiprocessor system according to claim 1, wherein each of the processors shifts to a normal process after transmitting the calculation processing content to the transmission destination processor. 3. The diagnostic processor, wherein each of the processors includes, when the comparing means detects a mismatch, the content of the calculation processing, the calculation result thereof, and the calculation result from the transmission destination processor. 3. A multiprocessor system according to claim 1, further comprising means for reporting to the multiprocessor. 4. The multiprocessor system according to claim 3, wherein said diagnosis means performs calculation processing according to said calculation processing contents, and compares the calculation result with the calculation result of said processor. 5. A method for diagnosing a multiprocessor system comprising a plurality of processors interconnected via a bus and each performing a calculation process, wherein: a display step of displaying an idle time of the calculation process; A storage step of storing its own calculation processing content and its calculation result immediately before reaching a predetermined diagnosis point in the course of calculation processing; and A transmitting step of transmitting the calculation processing content of the storing step with the processor indicating the vacancy by referring to the time display as a transmission destination processor; and receiving the calculation processing content being the transmission content of the transmission step of the transmission source processor. Performing a calculation process in accordance with the content of the calculation process, and returning the calculation result to the transmission source processor. Step a diagnostic method characterized by having, a comparison step of comparing a calculation result of said storing step calculation results that have been returned from the destination processor that. 6. The diagnostic method according to claim 5, further comprising a step of, after transmitting the calculation processing content to the transmission destination processor, shifting to normal processing. 7. The method according to claim 1, further comprising the step of, when a mismatch is detected in said comparing step, reporting the calculation processing contents, the calculation result thereof, and the calculation result from said transmission destination processor to a diagnosis processor. Item 7. The diagnostic method according to item 5 or 6. 8. The diagnostic method according to claim 7, further comprising the step of the diagnostic processor performing a calculation process according to the content of the calculation process, and comparing the calculation result with the calculation result of the processor. 9. A recording medium storing a control program of a diagnostic method for a multiprocessor system, which is connected to each other via a bus and each performs a calculation process, wherein the program stores the control program. A display step of displaying a free time of processing; and a storage step of storing the content of its own calculation processing immediately before reaching a predetermined diagnostic point in the course of the calculation processing and the calculation result; A transmission step of sending the calculation processing content of the storing step as a destination processor by referring to the idle time display of another processor when the diagnosis point is reached, and a transmission step of the source processor. A step of receiving the calculation processing content that is the transmission content and performing a calculation process in accordance with the calculation processing content; Flop and, a recording medium, characterized in that the calculation result having a comparison step of comparing the calculation result of the storing step calculation results that have been returned from the transmitting processor.