The present invention relates to an error detection device and method and in particular, but not exclusively to redundant arrays of inexpensive disks and in particular to a distributed exclusive-OR function for target devices to perform a read-modify-write operation.

Background of the Invention The integrity of data stored in storage systems is typically protected by error detection code methods. One such method is the Reed-Solomon Cyclic Redundancy Check (R-S CRC) . This method generates a code word for each data block stored in the storage system. The first k bits of an n bit code word represent the data block and the last n-k bits are the CRC bits. The CRC bits are created from a modulo-2 function of the data block k bits and a generator polynomial .

When the data is retrieved from the storage system, the data is checked for any errors . The check can be performed by executing the modulo-2 function of the data block k bits and the generator polynomial again. The resulting CRC bits are then compared to the previous CRC bits. If they are not equal, an error in either the data or its transmission occurred. Appropriate error correction techniques are then implemented.

Redundant arrays of inexpensive disks (RAID) are an example of a data storage system. RAIDs include the capability of correcting or recreating erroneous data and for remedying a complete failure of a disk in an array. Of the five RAID architectures 1-5, RAID 5 is presently the

most popular since it provides high data reliability with a low overhead cost for redundancy, good data read performance and satisfactory data write performance.

RAID 5 utilizes a known error detection concept based on an exclusive-OR function and distributes the calculated parity bits, as well as the data, among all the disks in the array. The error detection is performed by an array controller, which also oversees other operations of the disk arrays. The array controller typically incorporates multiple small computer system interface (SCSI) buses.

The array controller provides and receives data from the disks of the array through associated disk controllers. Each of these disk controllers can have a high bandwidth, such as 160 Mbytes/s. However, the bandwidth between the disk medium and the disk controller can be much less, such as 20 Mbytes/s. As a result, the disk controller has a considerable amount of bandwidth remaining that can be employed for other functions.

Transferring the error detection operation from the array controller to each disk controller has been proposed to take advantage of the remaining disk controller bandwidth. This operation transfer allows the use of a standard SCSI controller as an array controller for the RAID 5 architecture . The transfer would also reduce the amount of hardware in the array controller required to support the RAID 5 architecture and would free up bandwidth of the array controller. Moreover, the total overhead of the error detection operation for each disk controller would be less than the overhead for that operation in the array controller.

To transfer the error detection operation from the

array controller to each disk controller, a Read-Modify- Write operation must be supported by each disk controller. The Read-Modify-Write operation is used any time a write operation needs to be performed on the array, such as the writing of data from a host.

There are two architectures that an array controller must support: a multiple interface striping configuration and single interface striping configuration. Briefly, striping is storing of data constituents, anywhere from a bit to a disk sector, onto separate disk drives of the array at related addresses. A multiple interface configuration has the disk drives of the array on separate buses or loops . A single interface configuration has the disk drives on the same bus or loop. A disk drive of any array can communicate directly with another disk drive in the single interface configuration, but must communicate with the other disk drive through the array controller in a multiple interface configuration.

For both of these architectures, the disk controller of each disk drive should have the capability to store new data, perform a logic function (such as an exclusive-OR (XOR) function) on that new data with old data from the logical address of the new data, and output the result of that XOR function for use by a parity drive. The disk controller should also have the capability to receive from another drive in the array parity bits from an XOR function of the old and new data, XOR that result with the old parity bits corresponding to the old data stored on the other drive, and store the resulting new parity bits to the address of the old parity bits. Further detail is available in "RAID 5 Support on SCSI Disk Drives", Rev. 1.5, Seagate

Technology, which is hereby incorporated by reference.

An example of these capabilities is shown in FIGURE 1, which illustrates a multiple interface striping configuration with a distributed XOR function. An array controller 10 receives new data from a host (not shown) via a lead 12 and stores the new data in a new data buffer 14. New data from new data buffer 14 is supplied to a port 16 via a lead 18. From port 16, the new data is supplied to distributed XOR device 20 and disk medium 22 of data disk drive 24 via lead 26. XOR device 20 performs an exclusive- OR function on the new data and the old data supplied from disk medium 22 via lead 21. The old data is retrieved from the old data address on disk medium 22. A CRC check of the data is also performed. The result from XOR device 20 is supplied via lead 27 to buffer 28. The host then reads the XOR result from buffer 28 to port 16 over a lead 30.

Port 16 provides the result to XOR buffer 32 via lead 34. The result is then output from port 36 via lead 38 to parity drive 40 via lead 42. Distributed XOR device 44 of parity drive 40 receives the result and performs an exclusive-OR function on the result and the old parity bits from disk medium 46. For a single interface striping configuration, the output from buffer 28 of data disk drive 24 would be supplied directly to XOR device 44 of parity drive 40. The old parity bits corresponds to the old data of data disk drive 24. The resulting new parity bits are written to the disk medium 46.

To support the distributed XOR function, distributed XOR device 20 of data disk drive 24 in FIGURE 1 contains XOR buffering. A traditional arrangement would utilize two separate buffers. To illustrate, FIGURE 2 shows a

distributed XOR device 20 having data buffers 50 and 60, along with temporary buffer 70 and XOR logic 80.

Still referring to FIGURE 2, old data is written to data buffer 50 in response to a write command from the host (not shown) and stored in temporary buffer 70. New data from the host is stored in a data buffer 60. XOR logic 80 receives the data from both buffers 60 and 70, and performs an exclusive-OR function on both data. The result of the exclusive-OR function is then stored in temporary buffer 70. Since temporary buffer 70 is usually not large enough to store the entire contents of the result, the result is then stored in buffer 28. Buffer 28 can then supply data to either array controller 10 or parity disk drive 40 (FIGURE 1) , depending on whether a multiple or single interface striping arrangement is used.

Disadvantages of the traditional arrangement described above are its costs. If data buffers 50 and 60 are large, then temporary buffer 70 must be correspondingly large, which is expensive. If temporary buffer 70 is not large enough to store the result of the exclusive-OR function, additional buffer bandwidth must be used to write the result to buffer 28. If temporary buffer 70 is too small, then the old and new data cannot not be read in long bursts from data buffers 50, 60. This exacts a time cost since data will take longer to be read from data buffers 50, 60. Buffer 28 also imparts a time delay.

In addition, a read from and a write to temporary buffer 70 occurs for every data buffer 60 read. Consequently, temporary buffer 70 must maintain twice the bandwidth required for reading data buffer 60. The bandwidth of temporary buffer 70 will either put a limit on

the bandwidth of distributed XOR device 20, or will be expensive to implement in hardware.

Summary of the Invention According to one aspect of the present invention there is provided a logic function engine circuit comprising a data logic function circuit coupled to receive data to perform a logic function on the data to generate a plurality of results and an error detection circuit coupled to the data logic function circuit to determine from the plurality of results if there is an error.

The invention also provides for an error detection device to perform error detection on combined data.

The device may be coupled to a data logic function circuit that is coupled to receive data to generate the combined data. Also, the error detection can be coupled to the data logic function circuit as a direct connection.

In particular, the received data can be interleaved, and the device may further comprise at least one storage device coupled to the data logic function circuit to provide the interleaved data.

Preferably, the error detection device generates CRC bits in response to the combined data, encodes the CRC bits with a constant, and compares a one of the combined data to the encoded CRC bits.

Further, the error detection device generates an error signal if the one of the combined data and the encoded CRC bits are not equal. The error detection device can be arranged to perform error detection on a block of data. According to another aspect of the invention there is provided a distributed XOR device for a disk drive included

in an array of disk drives comprising a data buffer for storing a plurality of data words at least two of which include CRC bits, a data XOR circuit for receiving the data words in an interleaved manner and combining pairs of the data words to generate results, and an error detection circuit for generating from a plurality of the results other CRC bits, encoding the other CRC bits with a constant and comparing the encoded other CRC bits with a result corresponding to the combination of the at least two data words including CRC bits, wherein an error signal is generated if the encoded other CRC bits and the combination of the at least two data words including CRC bits are not equal .

According to yet another aspect of the present invention there is provided a logic function engine circuit comprising a data logic function circuit coupled to receive data and capable of at least one of performing a logic function on the data to generate a plurality of results and outputting the data without performing the logic function, and an error detection circuit coupled to the data logic function circuit to determine if the data is erroneous.

According to a further aspect of the present invention there is provided a method of data error detection comprising the steps of receiving at least two interleaved data blocks and performing a CRC check on the interleaved data blocks.

The invention also provides for an error detection method comprising the step of performing error detection on combined data. The error detection method can further comprise the step of combining data, and the step of combining data can

include combining interleaved data.

Also, the interleaved data can include at least two data blocks, and the data blocks can include data words and the step of combining data can include combining pairs of the data words.

Such a method may further comprise the step of performing error detection on a data block, and yet further comprise the step of receiving data through a direct connection. Still further, the invention can provide for an error detection method comprising the steps of outputting at least one of combined data or a data block and performing error detection on at least one of the combined data and the data block. It will thus be appreciated that the invention can provide for a distributed XOR device that preferably includes a data buffer which preferably stores at least two data blocks in an interleaved manner. The data blocks contain data words, and a specific data word for each data block contains CRC bits. In response to certain addresses, the data words of the data blocks are output from the data buffer in an interleaved manner. An XOR engine circuit receives the interleaved data words and preferably includes a data XOR circuit and an error detection circuit. The data XOR circuit preferably performs an exclusive-OR function on pairs of data words, where one data word is from a one data block and the other data word in from the other data block. The generated combinations or results are output to the error detection circuit. The error detection circuit generates CRC bits from the generated combinations or results. Preferably, the CRC bits are encoded with a

constant ID _CRC . These encoded CRC bits are compared to the result of the exclusive-OR function on the CRC bits of the data blocks. If these bits are not equal, an error signal is output. The comparison therefore checks whether the XOR engine circuit or the data blocks are in error.

The present invention provides a distributed logic function device, preferably including an XOR function, that includes a storage device, such as a data buffer. The data buffer stores at least two data blocks in a preferred interleaved manner. The data blocks contain data words, and a specific data word for each data block contains CRC bits . In response to certain addresses, the data words of the data blocks are output from the data buffer in an interleaved manner. A logic function engine circuit, such as an XOR engine circuit, receives the interleaved data words. The XOR engine circuit preferably includes a data logic function circuit, like a data XOR circuit, and an error detection circuit. The data XOR circuit combines pairs of data words, preferably by performing an exclusive-OR function, where one data word is from one data block and the other data word in from the other data block. The generated results are output to the error detection circuit.

The error detection circuit generates CRC bits from the results. Preferably, the CRC bits are encoded with a constant ID _CRC . These encoded CRC bits are compared to the result of the exclusive-OR function on the CRC bits of the data blocks. If these bits are not equal, an error signal is output. The comparison therefore checks whether the XOR engine circuit or the data blocks are in error.

The present invention also provides a method of data

error detection. This method preferably includes the steps of receiving at least two combined data blocks and performing a CRC check on the combined data blocks. If the CRC check determines that there is an error, an error signal is output .

Numerous other advantages and features of the present invention will become readily apparent from the following detailed description of the invention and the embodiments thereof, from the claims and from the accompanying drawings in which details of the invention are fully and completely disclosed as a part of this specification.

The invention is described further hereinafter, by way of example only, with reference to the accompanying drawings in which: Fig. 1 is a block diagram of a RAID multiple interface striping architecture having a distributed Read-Modify-Wπte operation according to a related art;

Fig. 2 is a block diagram of the distributed XOR device and buffer according to the related art of Fig. 1; and Fig. 3 is a block diagram of a distributed XOR device embodying the present invention.

Detailed Description of the Preferred Embodiment

Fig. 3 illustrates a distributed XOR device 120 according to an embodiment of the present invention. Distributed XOR device 120 includes a data buffer 140 coupled to receive addresses over an address lead 145. Data buffer 140 is also coupled to receive data from a MUX 150 over a lead 155. MUX 150 receives data from data leads 160, 165. Data buffer 140 is coupled to an XOR engine circuit 170 via a lead 180.

XOR engine circuit 170 includes a data XOR circuit 200, an error detection circuit 220 and a MUX 260. Data XOR circuit 200 includes XOR logic 205 that is coupled to data buffer 140 via lead 180. An XOR MUX 210 is coupled to XOR logic 205 via a lead 208 and is coupled to data buffer 140 via lead 180. An XOR register 215 is coupled to XOR MUX 210 via a lead 212 and is coupled to XOR logic 205 via a lead 216.

Data XOR circuit 200 is coupled to error detection circuit 220 and MUX 260 via a lead 218. Error detection circuit 220 includes a CRC generator 225 coupled to data XOR circuit 200 via lead 218. An error detection XOR logic 230 is coupled to CRC generator 225 via a lead 226. An ID counter 235 is coupled to error detection XOR logic 230 via a lead 232. Error detection XOR logic 230 is coupled to a comparator 250 and MUX 260 via a lead 235. Comparator 250 is coupled to data XOR circuit 200 via lead 218. Comparator 250 and MUX 260 provide output signals on respective leads 255 and 265. The operation of distributed XOR device 120 will be explained with reference to Fig. 3. An array controller sends a write command to a buffer manager (not shown) of the disk controller that includes distributed XOR device 120. New (written) and old (stored) data blocks are provided to data buffer 140 by MUX 150 on data leads 160 and 165. It is preferred that one of data leads 160, 165 is coupled to a disk medium (not shown) and the other lead is coupled to the array controller.

Addresses are provided over address lead 145 to data buffer 140 to load the multiplexed data blocks. It is preferred that lead 145 is coupled to the buffer manager.

The buffer manager preferably provides addresses that are selectively incremented by one or two. Providing the two- increment addresses allows data to be read from or written to all even or all odd addresses. Distributed XOR device 120 loads or stores a data block in data buffer memory 140 in response to a command from the host. Specifically, MUX 150 is controlled to provide over lead 155 to data buffer 140 an old data block received over one of data leads 160, 165. Coincidentally, addresses incremented by two are provided via lead 145 to store the old data block in alternating address locations, for example odd address locations. This data storage in data buffer 140 is represented by data words Al, A2, ..., A__CRC in FIGURE 3.

Distributed XOR device 120 then loads a new data block in data buffer 140. MUX 150 is switched to provide over lead 155 to data buffer 140 the new data block received over the other one of data leads 160, 165. Addresses incremented by two are concurrently provided via lead 145 to store the new data block in alternating, even address locations. This data storage in data buffer 140 is represented by data words Bl, B2 , ..., B_CRC in Fig. 3. The new and old data blocks are stored in data buffer 140 in a preferred interleaved manner as illustrated.

Each stored data block is a code word that was generated by a conventional CRC technique. For example, old data block Al, A2, ..., A_CRC is a code word where all the data words Al, A2, ... except A_CRC are the actual data. Data word A_CRC includes CRC bits that were generated by the conventional CRC technique. Data word A_CRC is encoded with a logical sector address (LSA) of the disk medium. The LSA represents the starting address of the disk medium that

stores the old data block. The LSA is included in data word A_CRC to confirm that the read data block is the correct data block that was addressed. New data block Bl, B2, ... , B_CRC is a similar code word with an encoded LSA that is the same as the LSA encoded for the old data block.

Once data buffer 140 is loaded, the buffer manager provides addresses successively incremented by one to data buffer 140. This causes data blocks to be read out in an interleaved series, e.g., Al, Bl, A2, B2, ..., A_CRC, B_CRC, and are provided to XOR engine circuit 170. Briefly, XOR engine circuit 170 will perform a preferred exclusive-OR function to combine corresponding data words (e.g. Al, Bl) from the data blocks. XOR engine circuit 170 will detect any error in the combined data words . Combining data includes arithmetic and Boolean functions.

In particular, an address associated to data word Al is provided to data buffer 140. Data word Al is then provided over lead 180 to MUX 210 which is selected to output data word Al to register 215. At about the same time, an address associated to data word Bl is provided to data buffer 140. Data word Bl is then provided to XOR logic 205 and is exclusive-ORed (combined) with data word Al from XOR register 215 via lead 216. The result is latched in XOR register 215 via MUX 210. The exclusive-OR result (combined data) of data words Al and Bl, temporarily stored in XOR register 215, is then provided to both error detection circuit 220 and MUX 260 via lead 218. MUX 260 is selected to pass this result over lead 265 to another buffer, such as buffer 28 shown in Fig. 1. Alternatively, lead 265 can be multiplexed onto one of leads 160, 165 in response to a write command.

Subsequent sequential even/odd addresses are provided to data buffer 140 to read out all the data words. The operation discussed above is performed for each pair of corresponding data words of the new and old data blocks. The exclusive-OR result (combined data) for each pair of data words is output from XOR register 215 to CRC generator 225 of error detection circuit 220. CRC generator 225 uses all these results or combined data to generate CRC bits for the exclusive-OR (combined) results of the data word pairs. Error detection XOR logic 230 performs a preferred exclusive-OR function on the CRC bits from CRC generator 225 and a constant lO^ (explained below) output from ID counter 235.

MUX 260 is selected to provide the exclusive-OR result of the CRC bits and the constant ID _CRC from error detection XOR logic 230. Comparator 250 compares the exclusive-OR result of data words A_CRC and B_CRC from XOR register 215 to the exclusive-OR result of the CRC bits and the constant ID _CRC . If the results are not equal, an error signal is transmitted over lead 255 to preferably either the array controller or a parity (CRC) disk drive.

As discussed above, the data words A_CRC and B_CRC are preferably encoded with an LSA. This "end loading" encoding is performed with an exclusive-OR function by preference. When the data words A_CRC and B_CRC are exclusive-ORed together by XOR logic 205, the LSA encoding is removed.

However, the result of the exclusive-OR of the data words A_CRC and B_CRC does not represent the correct CRC bits for the exclusive-OR of the previous data words Al, A2, .... and Bl, B2, ... Therefore, the result of the exclusive-OR of the data words A_CRC and B_CRC is preferably

encoded with the constant ID _Q ^. The constant ID^ represents the CRC bits generated by the conventional CRC technique of an all zero data block.

When a data read command is requested by the host, MUX 150 is controlled to provide over lead 155 to data buffer 140 an old data block received from the disk medium over one of data leads 160, 165. Addresses are provided over address lead 145 to data buffer 140 to load the old data blocks. The old data blocks are output to XOR engine circuit 170. MUX 210 is selected so that the old data blocks bypass XOR logic 205 via lead 180. The old data blocks are passed through register 215 to MUX 260. MUX 260 is selected to output the old data blocks from lead 218 to lead 265.

While the old data block is output on lead 265, the old data block is also provided to error detection circuit 220. CRC generator 225 generates CRC bits for the data words of the old data blocks. These CRC bits are encoded with the corresponding LSA stored in ID counter 235. ID counter 235 is incremented for the next data block. When the CRC bits of the old data block are available on lead 218, MUX 260 is selected to output the generated CRC bits that are encoded with the corresponding LSA. Concurrently, comparator 250 compares the CRC bits of the old data block to the output of the generated CRC bits that are encoded with the corresponding LSA. If the two CRC bits are not equal, an error signal is provided on lead 255.

The preferred embodiment of the present invention provides numerous advantages. First, the use of data buffer 140 that outputs data in an interleaved manner allows the XOR engine circuit 170 to perform the data word combination and the error detection on-the-fly while still

maintaining burst access to data buffer 140. This approach halves the necessary available bandwidth of the storage device, e.g. data buffer 140. If two storage devices are used instead of the single storage device, the maximum burst length is reduced.

Second, end loading has the advantage that constant ID^

(representing an all zero data block) is calculated once for all the data blocks with different LSAs . If the LSA were part of the seed data of the CRC generator, the ID _CRC would have to be recalculated for each data block. This recalculation would require another CRC generator. The presently preferred XOR engine circuit, therefore, has less hardware, particularly another CRC generator used to recalculate the constant ID _CRC for each data block. In addition, the circuitry of XOR engine circuit 170 not only checks data integrity as it is read and written, but also checks the XOR engine circuit 170 hardware integrity.

Numerous variations and modifications of the embodiment described above may be effected without departing from the spirit and scope of the novel features of the invention. For example, the preferred embodiment illustrated in Figure 3 includes the illustrated leads that are preferably buses. These buses allow the distributed XOR device 120 to process data in parallel. However, serial leads can be used without parting from the scope of the present invention.

Another example is that the preferred embodiment of the present invention as illustrated in Fig. 3 shows error detection circuit 220 coupled to data XOR circuit 200 as a direct connection, which is not required for the present invention. Although not preferred, error detection circuit 220 may be coupled to data XOR circuit 200 through another

device or circuit, such as a buffer. The other device or circuit would cause a time delay, and thus prevent data combination and error detection on-the-fly.

It is to be understood that no limitations with respect to the specific device illustrated herein are intended or should be inferred and that the invention is not restricted to the details of the foregoing embodiment .