JPH09134314A - Memory access controller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress the extension of a logical scale to the minimum without reducing the performance of the whole system and to efficiently execute storage module reading processing and data error correcting processing. SOLUTION: A processor A102 directly receives data read out from a storage module 101 by a latch 105 and executes processing by using the data. An error detecting circuit 110 simultaneously receives the data and executes error check by the use of an ECC code 114 and an error correcting circuit 109 corrects an error and stores corrected data in a latch 106. When the error is detected, the processor A102 aborts current processing by interruption, switches a selector 107, receives the corrected data from the latch 106 by the latch 105 and retries the processing. On the other hand, a processor B103 receives only the data corrected by the circuit 109 and stored in the latch 106 by its latch 104 without directly receiving the data read out from the module 101 and executes processing by the use of the corrected data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a memory access control device for storing data and an error detection / correction code for the data in a storage module and performing error detection / correction at the time of reading data, and particularly to a high speed memory access control device. The present invention relates to a memory access control device capable of performing optimum correction processing when read data is erroneous, for each of the processing that requires various data reading and the processing that does not require high-speed data reading.

[0002]

2. Description of the Related Art In recent years, as the amount of information processing handled by computer systems has increased, storage modules (main memory, cache memory, etc.) have also been required to have a large capacity. For this reason, computer systems, especially one-chip microprocessors, are often provided with these storage modules configured as chips separate from the central processing unit. Such a computer system is required to have high-speed processing and high reliability, but a system having a large-capacity storage module that is made into a separate chip as described above takes time to transfer data to the central processing unit. In addition, there is a high risk of data errors due to failures due to α-rays, failures of the memory chip itself, etc. In order to avoid this, it is essential to have a mechanism for rapid detection / correction of data errors. Has become.

This data error detection / correction is generally performed using an error detection / correction code (ECC code). The ECC code is generated at the time of writing the data and is held in the storage module, and is simultaneously read at the time of reading the data, and the data error is detected / corrected by using the read data and the ECC code. Is done. It takes some time to perform this error detection / correction.

The data read process generally requires some time until the data is read from the cache, such as the tag hit check of the cache memory and the read of the cache data. Therefore, the central processing unit stops the operation to wait for the data. Therefore, the data reading process requires extremely quick processing. Then, when extremely high-speed processing is required, the central processing unit performs processing using the data that has been read out without waiting for error detection of the data, and if an error is detected in the data, A method of interrupting the processing being executed to interrupt the processing of the central processing unit is used.

As a memory access control method for performing this type of control, for example, a technique described in Japanese Patent Laid-Open No. 304653/1990 is known.

FIG. 4 is a block diagram showing a configuration example of a memory access control device according to a conventional technique. In FIG.
Reference numeral 101 is a storage module, 102 and 103 are processors A and B, 104 and 105 are latches, 109 is an error correction circuit, 110 is an error detection circuit, and 112 and 113 are error interrupt control circuits.

In the prior art shown in FIG. 4, the processing device A
The processing device 102 and the processing device B 103 respectively perform processing using the data read from the storage module 101. Now, assuming that the processing device A 102 performs processing, its operation will be described below.

The processing device A 102 includes a storage module 10
The data read from 1 is the data line 115, the data line 1
Latch 10 through selector 107 selecting 15
5, and uses the data to start processing.
The data read from the storage module 101 is also sent to the error detection circuit 110 at the same time. Error detection circuit 11
0 performs an error check using this data and the ECC code on the signal line 114 read out together with the data from the storage module 101, and when an error is detected,
The processor A 102 is interrupted through the signal line 111.

When the error interrupt control circuit 112 detects an error interrupt, the processor A102 interrupts the current process which was being performed using the read data including the error, and reads the storage module 101 again. At this time, the error interrupt control circuit 112 switches the selector 107 to the error correction circuit 109 side.
The read data is irrespective of whether or not there is an error.
The data is corrected and output by the error correction circuit 109, and the data is received by the latch 105 again. Processor A
102 uses this data to redo the process. The operation of the processing device B103 is performed in the same manner as described above.

FIG. 5 shows the above-mentioned conventional technique in which a primary and a secondary are used.
6 is a flowchart for explaining a partial write process when applied to a system having a multi-stage cache memory, which will be described below.

The partial write is a process of writing data having a data length shorter than the unit write data length to the cache memory. After the data is once read from the cache memory, the write data is merged therein and the merged data is written. Is written in the cache, and the data read time is not important for system performance. In the case of the conventional technique, this partial process is performed according to the flow shown in FIG.

(1) When the partial write process is started, first, data is read from the secondary cache (step 501).

(2) The read data is received and merged with the write data (steps 502 and 503).

(3) The data read in step 501 is checked for data error by an error detection circuit in parallel with the merge processing, and if an error is detected in the data, the error detection circuit outputs an error. A signal is transmitted (steps 506-508).

(4) In step 503, when the merge processing is completed, it is checked whether an error signal from the error detection circuit is received. If no error signal is received, the merged data is secondarily cached. Is written in and the processing ends (steps 504 and 505).

(5) If the error signal is received in step 504, the data is read from the secondary cache again. At this point, the data merging process is not performed, and the read data is input to the error correction circuit and subjected to the error correction process (steps 509 and 510).

(6) The data after error correction is received,
The merge data performed in step 503 is eliminated, the merge process with the write data is performed again, and writing to the secondary cache is performed (steps 511 and 512).

[0018]

DISCLOSURE OF THE INVENTION Problems to be Solved by the Invention
As described with reference to FIG. 4, when applied to a system having a plurality of processing devices, the error signal from the error detection circuit needs to be distributed to each processing device, resulting in a large error signal load and delay. The problem is that it becomes severe. The error signal needs to be transmitted as soon as possible in order to start the cancel operation in the processing device as soon as possible, but since it is an output signal that has passed through the error detection circuit, the delay is large in the first place, and especially in the case of ultra-compact and highly integrated circuits. In a microprocessor or the like, a bit width obtained by combining data and an ECC code is larger than a chip size (the number of pins), so that data from a storage module is read from a plurality of sides on the chip. Those 1
Since the error detection / correction must be performed after the data is collected in some place, it takes a long time to complete the error detection / correction, and the increase of the delay of the error signal due to the increase of the load is a serious problem.

Further, in the above-mentioned conventional technique, as described in the example of FIG. 5, high-speed processing can be performed when no error occurs, but once an error occurs, the processing device operates. There is a problem that the processing device becomes very complicated because it is necessary to stop the internal processing, return to the state before receiving the data, and receive the corrected data again. Further, as the processing device becomes complicated, the logical scale also increases accordingly. Therefore, if this mechanism is added to each processing device, the logical scale also increases as a whole device, and a microprocessor of a one-chip configuration with a limited number of gates, etc. Then, there arises a problem that it is difficult to deal with.

An object of the present invention is to solve the above-mentioned problems of the prior art, reduce the load of error report signals with severe delays, and increase the logical scale without degrading the performance of the entire system. It is an object of the present invention to provide a memory access control device capable of performing reading of a storage module and error processing of data while keeping the number to a minimum.

[0021]

SUMMARY OF THE INVENTION In order to achieve the above-mentioned object, the present invention relates to a process in which a data read time from a storage module is important for system performance, and a process from a storage module. This was done by paying attention to the fact that the read time is not important for the system performance. For example, when the storage module is a tag of the cache memory, the tag read time is very important for system performance because the processor waits for the data and stops the operation until the data is read from the cache memory. In the case of hit check when performing the memory purging process (invalidating a certain entry), the frequency of the purging process is usually not so high, so the tag read time is not so important.

According to the present invention, from the above-mentioned viewpoint, the above object is provided with a storage module for storing data and an error detection / correction code, and an error detection / correction circuit for data read from the storage module. The memory access control device has a mode in which the read data is directly transferred to the processing device and a mode in which the corrected data is transferred to the processing device by the error detection / correction circuit, and the data read time from the storage module is system performance. The important processing is performed by directly transferring the read data to the processing device, and the processing in which the data read time from the storage module is not important is transferred to the processing device by the error detection / correction circuit. It is achieved by doing.

Further, the above-mentioned object is provided with a mechanism for holding the data corrected by the error detection / correction circuit in the above description, and the error detection / correction circuit is provided with the error detection / correction circuit during the processing in the mode in which the read data is directly transferred to the processing device. When an error in the read data is detected, the error occurrence is notified to the processing device, and the retained corrected data is transferred to the processing device.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a memory access control device according to the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a memory access control apparatus according to the first embodiment of the present invention, and the reference numerals in the figure are the same as those in FIG.

The first embodiment of the present invention shown in FIG. 1 has a configuration suitable for explaining the principle of the present invention, and is a storage module for two processing devices A and B. The read data is given and error detection / correction of the data is performed. In the example shown in FIG. 1, the processing device A 102 performs a process in which the data read time from the storage module 101 is important for the system performance, and the processing device B 103 performs a process in which the data read time is not important for the system performance. I am supposed to do it.

The processing device A 102 includes the storage module 10
The data read from 1 is the data line 115, the data line 1
15 is directly received by the latch 105 through the selector 107 that selects 15 and this data is used for processing.
On the other hand, the data read from the storage module 101 is also input to the error detection circuit 110 at the same time. The error detection circuit 110 performs an error check using the read data and the ECC code 114 read from the storage module 101 together with this data, and the error correction circuit 109 performs error correction to correct the error. Later data is held in the latch 106.

When the error detection circuit 110 detects an error in the read data, it interrupts the processing unit A 102 through the signal line 111. When the error interrupt control circuit 112 of the processing device A 102 detects the error interrupt, it interrupts the current processing in the processing device A 102, switches the selector 107 to receive the corrected data from the latch 106 to the latch 105 again, and the processing device A 102 receives the corrected data. Cause the process to be redone.

On the other hand, the processing device B103 can receive only the data corrected by the error correction circuit 109 and held in the latch 106 in the latch 105 without directly receiving the data read from the storage module 101. Use to process. Processor B10
3, the reception of read data from the storage module 101 is slower than that of the processing apparatus A 102, but the interruption of the processing by the interrupt after the error detection and the re-execution processing are not necessary, so the logical scale should be reduced. You can

According to the first embodiment of the present invention described above, an error processing mechanism for read data of a storage module can be configured without degrading the performance of the entire system and minimizing the increase in logical scale. Is possible.

Further, according to the first embodiment of the present invention,
Since the error signal 111 only needs to be connected to the processing device A102, the load of the error signal can be reduced and the delay can be improved as compared with the case of the conventional technique described with reference to FIG.

FIG. 2 is a block diagram showing the configuration of a memory access control device according to the second embodiment of the present invention. In FIG. 2, 120 is a secondary cache access control circuit, 121 is a secondary cache memory tag unit, 122 is a secondary cache memory data unit, 123 is a register file, 124 is a primary cache memory, and 125 is a control unit. Reference numeral 126 denotes a data read control unit (BT control unit), 1
27 is a partial write control unit, 128 is a hit check circuit, 129 and 130 are error detection circuits, 131 and 13
2 is an error correction circuit 133, 134, 139, 140,
Reference numerals 150 to 155 are latches, 135 is an ECC code generation circuit, 136 is a merge circuit, 137 and 138 are 3-state buffers, and 141, 142 and 149 are selectors.

The second embodiment of the present invention shown in FIG. 2 is an application of the present invention to a system having a two-stage cache memory of primary and secondary. Then, in the following, processing in which the data read time is important for system performance is
It is assumed that the read process of the next cache memory is performed, and the process in which the data read time is not important for system performance is the partial write process.

The second embodiment of the invention shown in FIG.
Next cache memory (tag section 121, data section 12
2) The secondary cache access control circuit 120, the register file 123, and the primary cache memory 124. The control unit 125, which controls the entire secondary cache access control circuit 120, includes control units for various processes for accessing the secondary cache memory. In FIG. Part 1
26, a partial write control unit 127 is shown.

Next, the operation of the second embodiment of the present invention shown in FIG. 2 will be described for the case of the operation of the secondary cache data read processing (BT processing) due to the primary cache miss.

Normally, during the BT process, the register file 12
3. Since the arithmetic processing in the processing device (not shown) including the primary cache memory 124 is often suspended (pipeline lock) for waiting for the secondary cache read data, the BT processing should be shortened even by one cycle. Is important for system performance. Therefore, the error process of the BT process is performed in the mode using the interrupt by the error signal. Therefore, in FIG. 2, the selector 1
41 selects the data line 143 side, and selector 142 selects the data line 144 side.

In this state, when the secondary cache read address is set in the latch 139 by the register file 123, the tag section 121 and the data section 122 of the secondary cache memory are simultaneously accessed by this address. The tag and the data are read simultaneously, and the tag is input to the hit check circuit 128 from the data line 143 via the latch 151. Hit check circuit 1
If it is determined in 28 that a cache hit has occurred, the control unit 125 that has received this cache hit report outputs a write signal 145 to the register file 123 and the primary cache memory 124. Thus, the data read from the data section 122 of the secondary cache memory at the same time as the tag is written in the register file 123 and the primary cache memory 124 via the latch 155, the data line 144, the selector 142, and the data line 147. .

The ECC code for the tag read at the same time as the tag is input to the error detection circuit 129 together with the tag via the latch 150, and the ECC code for the data read at the same time as the data is latched 154.
And is input to the error detection circuit 130 together with the data. When the error detection circuit 129 or 130 detects an error, the error detection circuit 129 or 130 reports the occurrence of the error to the error processing unit of the data read control unit 126 of the control unit 125 by the error signals 148 and 149.

The data read control unit 1 receiving this report
The error processing unit 26 starts error correction processing, outputs a write cancel signal 146 to the register file 123 and the primary cache memory 124, invalidates the above-described data writing, and selects 141, 142.
The error correction circuit side is selected. At this point in time, since the tags and data corrected by the error correction circuits 131 and 132 are held in the latches 133 and 134, the hit check operation, the register file 123, and the primary cache memory 124 are again performed using these data. The data write operation is performed.

FIG. 3 is a flow chart for explaining the operation of the partial write process of the second embodiment of the present invention shown in FIG. 2. Next, referring to FIG. 3, the second embodiment of the present invention shown in FIG. The operation of the above embodiment will be described for the case of the partial write operation to the secondary cache memory.

As described above, the partial write is a process of writing data having a data length shorter than the unit write data length to the cache memory, and once the data is read from the cache memory, the write data is merged therein. However, it is necessary to write the merged data in the cache. Since the partial write is such an inefficient process, it is usually used infrequently. Therefore, the read processing time of the cache memory in the partial write is not so important in terms of system performance, and thus the mode using the data after error correction is suitable as the error processing mode. For this reason, in FIG.
The side is selected.

(1) Register file 1 in the above state
From 23, when the write address is set in the latch 139, the secondary cache memory is referred to by this address, and the read data and its ECC code are
It is set in the latch 155 and the latch 154 (step 301).

(2) The read data set in the latch 155 and the latch 154 and the ECC code thereof are given to the error detection circuit 130 and the error correction circuit 132, and the error correction circuit 132 detects the error in the error detection circuit 130. Error correction is performed regardless of the presence or absence, and the corrected data is held in the latch 134 (steps 302 and 303).

(3) On the other hand, the write data of partial write is set in the latch 140 by the register file 123. The merge circuit 136 includes a latch 134.
The write data set in the latch 140 is merged with the error-corrected data from the secondary cache memory set in (step 304).

(4) The selector 149 is controlled by the partial write control unit 127 so as to select the merge circuit 136 side, and the merged data is stored in the latch 1
It is set to 52 and held, and is also given to the ECC generation circuit 135. The ECC generation circuit 135 generates an ECC code for the given data and
The C code is set in the latch 153. The data held in the latch 152 and the ECC code set in the latch 153 are then transferred to the 3-state buffer 137.
And 138 through the data section 1 of the secondary cache memory
22 is written (step 305).

The above-described series of partial write processing requires time until data after error correction is obtained, and the processing time becomes longer by that much, but interrupt time at the time of data error detection is not necessary. Therefore, the second embodiment of the present invention does not require a processing device relating to error processing in the partial write processing, and can reduce the logical scale. Furthermore, since it is not necessary to connect the error signal 156 to the partial write control unit 127,
The load of the error signal 156 can be reduced and the delay can be reduced.

[0047]

As described above, according to the present invention, a mode in which the read data is directly transferred to the processing device for the processing in which the data read time from the storage module is important for the system performance is also stored. To reduce the load of error report signals with severe delays by using the mode in which the data after error correction by the error detection / correction circuit is transferred to the processing device for the processing where the data read time from the module is not important. You can Also,
According to the present invention, without degrading the performance of the entire system,
It is possible to perform error processing on the read data from the storage module while minimizing the increase in the logical scale.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a memory access control device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a memory access control device according to a second embodiment of the present invention.

FIG. 3 is a flowchart illustrating an operation of a partial write process according to the second embodiment of the present invention shown in FIG.

FIG. 4 is a block diagram showing a configuration example of a memory access control device according to a conventional technique.

FIG. 5 is a flowchart illustrating a partial write process when the conventional technique is applied to a system having two-stage primary and secondary cache memories.

[Explanation of symbols]

101 Storage Module 102 Processing Device A 103 Processing Device B 104, 105, 133, 134, 139, 140, 1
50 to 155 Latch 107, 108, 141, 142, 149 Selector 109, 131, 132 Error correction circuit 110, 129, 130 Error detection circuit 112, 113 Error interrupt control circuit 120 Secondary cache access control circuit 121 Secondary cache memory tag Part 122 Secondary cache memory data part 123 Register file 124 Primary cache memory 125 Control part 126 Data read control part (BT control part) 127 Partial write control part 128 Hit check circuit 135 ECC code generation circuit 136 Merge circuit 137, 138 3 State buffer

Claims (2)

[Claims] 1. A read access data is directly processed in a memory access control device including a storage module for storing data and an error detection / correction code, and an error detection / correction circuit for data read from the storage module. Equipped with a mode to transfer data to the device and a mode to transfer the corrected data to the processing device by the error detection / correction circuit, and directly processes the read data for the process where the data read time from the storage module is important for system performance. A memory access control device, wherein the memory access control device transfers the data after correction by an error detection / correction circuit to a processing device for processing in which the data reading time from the storage module is not important. 2. A mechanism for holding the data corrected by the error detection / correction circuit, wherein an error of the data read by the error detection / correction circuit is eliminated during processing in a mode in which the read data is directly transferred to the processing device. 2. The memory access control device according to claim 1, wherein when detected, an error occurrence is notified to the processing device, and the held corrected data is transferred to the processing device.