JP2007207024A - Resource management device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To avoid concentration of access permission to the same master to reduce worst latency, and to improve access efficiency according to the kind of a shared resource. <P>SOLUTION: A bus arbitration part 104 and a resource control part 110 are interposed between a plurality of bus masters 101, 102, 103 and the shared resource 111, and access requirement is arbitrated on the basis of first bus arbitration information 108 showing a minimum frequency allowing reception of the access permission within a prescribed time in each bus master. An arbitration information operation part 105 operates the first bus arbitration information 108 such that the access permission to the same bus master is distributed on the basis of second bus arbitration information 106, and operates the first bus arbitration information 108 such that access interference between the bus masters is reduced in conformity with to the kind of the shared resource 111 or operates the first bus arbitration information 108 according to a band width acquisition situation of the bus master. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a resource management apparatus for arbitrating access requests to a shared resource from each master in a system in which a plurality of masters respectively issue access requests to at least one shared resource.

  In the system LSI, a plurality of masters such as a microprocessor, a DSP (Digital Signal Processor), and a DMA (Direct Memory Access) controller access a shared resource such as a memory and a peripheral I / O (input / output) controller. Here, there is a need for a resource management device for efficiently arbitrating access requests from each master to shared resources.

According to a certain prior art, the resource management device arbitrates access requests at regular intervals based on the priority order information of each master that is previously stored in a table format. The priority order information of a plurality of patterns is stored in the table, and the patterns are switched in order at regular time intervals for arbitration. Each master can receive an access permission for the number of times set to the highest priority in the priority pattern, and can guarantee the minimum access bandwidth of each master (see Patent Document 1).
JP 2004-246862 A

  In the system in which the master having the highest priority at each arbitration time is determined in advance as described above, a problem arises when there is a time bias in the timing of access request generation. For example, when a master that is equally assigned the highest priority at each arbitration time in one cycle makes an access request in the first half of the cycle, it cannot guarantee the desired access bandwidth to the master. As a result, the access efficiency is lowered. This is because the arbitration time to which the highest priority is assigned does not necessarily match the access request generation timing.

  If the timing at which each master makes an access request is not constant, it is difficult to determine in advance the master having the highest priority at each arbitration time so as to avoid a decrease in access efficiency.

  In addition, when the highest priority is given to a specific master, there is a problem in that the order of the masters receiving access permission is biased, and the access permission for the same master is concentrated and the worst latency to the shared resource deteriorates. there were.

  In addition, when the shared resource is SDRAM (Synchronous Dynamic Random Access Memory), the SDRAM requires a switching process between read (read) access and write (write) access. When the number of times that write access occurs alternately increases, there is a problem that the throughput decreases. Also, when there are a plurality of shared resources and each of them can be accessed in parallel, or when the shared resource is an SDRAM composed of a plurality of banks, if the same shared resource or the same bank of the SDRAM is continuously granted access permission There was a problem that access efficiency deteriorated.

  An object of the present invention is to provide a resource management device that can avoid concentration of access permission for the same master, reduce the worst latency, and improve the guarantee accuracy of the access bandwidth.

  Another object of the present invention is to provide a resource management apparatus capable of improving access efficiency by giving access permission in the order of improving access efficiency to shared resources.

  It is still another object of the present invention to provide a resource management apparatus that can improve access efficiency according to the type of shared resource while avoiding concentration of access permission for the same master.

  In order to solve the above-described problem, the present invention provides an arbitration information operation unit for operating arbitration information that is a set value for each master.

  When avoiding concentration of access permission for the same master, the mediation information operation unit generates a plurality of partial mediation information from one mediation information and operates the mediation information of the mediation information management unit based on the partial mediation information. .

  In the case of improving throughput by performing arbitration suitable for the type of shared resource, the arbitration information operation unit generates a plurality of pieces of partial arbitration information so as to continuously grant access to masters that do not interfere with access. .

  When the master does not acquire the bandwidth according to the set value and fails, the arbitration information operation unit operates the arbitration information so that the bandwidth approaches the set value according to the bandwidth acquisition status.

  According to the present invention, since access permission concentration of the same master can be avoided, it is possible to reduce the worst latency to the shared resource, and it is possible to avoid the deterioration of the guaranteed accuracy of the access bandwidth. Furthermore, it is possible to improve throughput by granting access permission to each master in the order in which the access efficiency to shared resources improves, and by operating arbitration information according to the bandwidth acquisition status, bandwidth guarantee accuracy Can be improved.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following description, each bus master has a function of actively transmitting an address signal, a control signal, and the like for performing data access. In general, one functional block has a plurality of interfaces for bus connection, some function as a bus master, and some function as a bus slave. For example, a Harvard architecture processor includes an instruction bus master and a data bus master.

  FIG. 1 shows a configuration example of a resource management apparatus according to the present invention. The resource management device 109 shown in FIG. 1 is interposed between the three bus masters 101, 102, 103 and the shared resource 111, and includes a bus arbitration unit 104, a bus arbitration information operation unit 105, and a bus arbitration information management. Unit 107 and resource control unit 110. Reference numeral 108 denotes first bus arbitration information managed by the bus arbitration information management unit 107, and 106 denotes second bus arbitration information managed by the bus arbitration information operation unit 105. The first, second, and third bus masters 101, 102, and 103 are connected to the bus arbitration unit 104 via the buses 112, 113, and 114, respectively. The first to third bus masters 101 to 103 and the resource management device 109 constitute one data processing system 100.

  The bus arbitration information management unit 107 includes a first set value that is the minimum number of times each of the first to third bus masters 101 to 103 can receive access permission within a predetermined time, and the bus masters 101 to 103. The first bus arbitration information 108 is managed. On the other hand, each of the first to third bus masters 101 to 103 issues an access request to the shared resource 111 at an arbitrary time. The bus arbitration unit 104 arbitrates access requests from the first to third bus masters 101 to 103 based on the first bus arbitration information 108, and sends one access request selected by the arbitration to the resource control unit 110. To tell. At this time, when two or more bus masters are making access requests at the same time, the bus arbitration unit 104 determines the first bus arbitration information as the number of times access permission is received within a predetermined time among the two or more access request masters. An access permission is preferentially given to a bus master that does not satisfy the first set value in 108. In addition, when there are a plurality of bus masters whose access permission is received within a predetermined time among the two or more access request masters, the bus masters 101 to 101 managed by the bus arbitration information management unit 107 are present. The bus arbitration unit 104 arbitrates the access request according to the priority rule based on the priority order information between 103. The resource control unit 110 performs data transfer control between the bus master selected by the bus arbitration unit 104 and the shared resource 111. The bus arbitration information operation unit 105 uses, as second bus arbitration information 106, the second set value relating to the minimum number of times each of the first to third bus masters 101 to 103 can receive access permission within a predetermined time. One or more are managed, and the first bus arbitration information 108 is operated based on the second bus arbitration information 106 according to a method described later as the first to fifth embodiments.

  A basic operation when the access request to the shared resource 111 from the first to third bus masters 101, 102, and 103 in FIG. 1 is arbitrated will be described. Here, it is assumed that there is a demand to give the first, second, and third bus masters 101, 102, and 103 access permission of 9, 9, and 2 times out of 20 arbitration operations. Assuming that a certain amount of access to the shared resource 111 can be made for each access permission, 45%, 45%, and 10% for the first, second, and third bus masters 101, 102, and 103, respectively. Access bandwidth is guaranteed.

  FIG. 2 shows one example of the first bus arbitration information 108 in FIG. A first period (t) 220 in FIG. 2 indicates the number of times that the bus arbitration unit 104 performs arbitration within a predetermined time. Here, since “20” is set as the first period 220, the 20 arbitration interval corresponds to the first period 220. In FIG. 2, 221 is a first count upper limit value, 222 is a first access permission count, 223 is first priority information, and 224 is second priority information. The first priority information 223 indicates that the higher the remaining priority permission count determined from the first count upper limit value 221 and the first access permission count 222 is, the higher the priority is. The remaining priority permission count is calculated by subtracting the first access permission count 222 from the first count upper limit value 221. The second priority information 224 indicates that the first bus master 101, the second bus master 102, and the third bus master 103 have a fixed priority in the order when the remaining priority permission counts are equal. In other words, here, the fixed priority is higher as the set numerical value of the second priority information 224 is smaller.

  FIG. 3 shows an example of arbitration based on the first bus arbitration information 108 of FIG. In the example of FIG. 3, all of the first to third bus masters 101 to 103 always make access requests. The arrows in FIG. 3 indicate the period during which an access request is issued. A white star indicates that a certain bus master has received access permission at a certain arbitration time. The numerical value below the arrow indicates the first access permission count 222 of the first to third bus masters 101 to 103 at each arbitration interval. The numerical values in parentheses indicate the remaining priority permission counts of the first to third bus masters 101 to 103. The black star mark indicates that the access attribute of access to the shared resource 111 changes. In the example of FIG. 3, the access attributes of the first bus master 101 and the third bus master 103 are the same, and the access attributes of the second bus master 102 are different. The access attribute is, for example, an attribute related to read or write that is an access direction or an attribute related to a range of a plurality of access addresses.

  According to FIG. 3, at the arbitration time 0, the first access permission count 222 of all the bus masters is 0, which is less than the first count upper limit value 221, and therefore, the first priority order for all the bus masters. Arbitration is performed based on the information 223. In this case, the remaining priority permission counts of the first bus master 101 and the second bus master 102 are the largest. As a result, since the remaining priority permission counts are the same for both, the first bus master 101 obtains access permission according to the second priority information 224. When the bus arbitration unit 104 grants access permission to the first bus master 101, the bus arbitration information management unit 107 increments the first access permission count 222 corresponding to the first bus master 101 by one. Next, at the arbitration time 1, all of the first to third bus masters 101 to 103 make an access request. At the arbitration time 1, since the first access permission count 222 of all bus masters is less than the first count upper limit value 221, arbitration is performed for all bus masters based on the first priority information 223. Is called. As a result, the second bus master 102 having the largest remaining priority permission count obtains access permission. If there is no access from a bus master whose first access permission count 222 is less than the first count upper limit value 221, an access request is made only from the bus master whose first access permission count 222 has reached the first count upper limit value 221. If it has occurred, the access request bus master with the higher priority obtains access permission as judged from the second priority information 224. Thereafter, arbitration is performed in the same manner, and when the arbitration at the arbitration time 19 is completed, the first access permission count 222 is reset to 0, and the arbitration in the first period 220 is completed. Thereafter, the same arbitration is performed periodically.

  Next, five variations when the bus arbitration information operation unit 105 operates the first bus arbitration information 108 will be described as first to fifth embodiments. Here, it is assumed that there is a demand to give the first, second, and third bus masters 101, 102, and 103 access permission of 9, 9, and 2 times out of 20 arbitration operations. Assuming that a certain amount of access to the shared resource 111 can be made for each access permission, 45%, 45%, and 10% for the first, second, and third bus masters 101, 102, and 103, respectively. Access bandwidth is guaranteed.

<< First Embodiment >>
FIG. 4 shows a first example of the second bus arbitration information 106 in FIG. 420 is a second period (T), and 421 is a second count upper limit set for each bus master. A second period 420 in FIG. 4 indicates the number of times that the bus arbitration unit 104 performs arbitration within a predetermined time. Here, since “20” is set as the second period 420, the 20 arbitration interval corresponds to the second period 420. The second count upper limit value 421 represents the minimum number of times each of the first to third bus masters 101 to 103 receives access permission within the second period 420, and is the sum of the second count upper limit values 421. Is set in advance to be equal to the set value of the second period 420. Here, the second count upper limit value 421 is 9, 9, 2 for the first, second, and third bus masters 101, 102, 103, respectively.

  FIG. 5 is an example in which the first period group (t1 to t4) 520 and the first count upper limit group 521 are generated based on the second count upper limit value 421 of FIG. The group 520 has four first periods, and the first count upper limit group 521 has four first count upper limits. Each of the first period groups 520 in FIG. 5 is the sum of all the bus masters for each column of the first count upper limit value group 521. The sum of the first period group 520 in FIG. 5 coincides with the second period 420 in FIG. 4, and the sum of the first count upper limit value group 521 for each bus master in FIG. 5 is the second sum for each bus master in FIG. To the count upper limit value 421. In addition, the absolute value of the difference between adjacent column values of each bus master in the first count upper limit group 521 of FIG. 5 is 0 or 1, and the second count upper limit value 421 is averagely distributed.

FIG. 6 is a diagram showing one algorithm for generating the first count upper limit value group 521 from the second count upper limit value 421 when the number of divisions (= N) is a power of two. In FIG. 6, the process of step S61 is initialization. Here, “/” indicates division, and “%” indicates an operation for obtaining the remainder of division. 6 is sequentially executed on the second count upper limit value 421 of each bus master in FIG. 4, whereby the first count upper limit group 521 of each bus master in FIG. Generate columns sequentially. In FIG. 6, “division remainder” and “adjustment value” are integers of the number of bits of the value of log ₂ N, and the calculation related to “division remainder” and “adjustment value” is the calculation of the number of bits of the value of log ₂ N. . That is, if the number of divisions (= N) is 4, “division remainder” and “adjustment value” are 2-bit integers, and operations relating to “division remainder” and “adjustment value” are 2-bit operations.

  A process of generating the first count upper limit group 521 of FIG. 5 from the second count upper limit value 421 of FIG. 4 using the algorithm shown in FIG. 6 will be described for the first bus master 101. Since the second count upper limit value 421 is divided into four, the “number of divisions” (= N) in FIG. 6 is 4, and the calculation related to the “adjustment value” is a 2-bit calculation. The execution result of step S61 is “number of divisions” = N = 4, “division value” = 9/4 = 2, “remainder of division” = 9% 4 = 1, “adjustment value” = 0, “number of processing times” = 0. Next, when step S62 is executed, “adjustment value” = “adjustment value” + “remainder of division” = 0 + 1 = 1 (no overflow), “adjustment value_over_flow” = 0, and “number of processing times” = 1. Next, when step S63 is executed, the first first upper limit value = “divided value” + “adjustment value_over_flow” = 2 + 0 = 2, and one first upper count value is generated. In step S64, the “number of processes” (= 1) is not equal to the “number of divisions” (= 4), so the process returns to step S62. Next, when step S62 is executed, “adjustment value” = 1 + 1 = 2 (no overflow), “adjustment value_over_flow” = 0, and “number of processing times” = 2. Next, when step S63 is executed, the second first count upper limit value = 2 + 0 = 2, and the second first count upper limit value is generated. In step S64, since the “number of processing times” (= 2) is not equal to the “number of divisions” (= 4), the process returns to step S62. Next, when step S62 is executed, “adjustment value” = 2 + 1 = 3 (no overflow), “adjustment value_over_flow” = 0, and “number of processing times” = 3. Next, when step S63 is executed, the third first count upper limit value = 2 + 0 = 2 is obtained, and the third first count upper limit value is generated. In step S64, the “number of processes” (= 3) is not equal to the “number of divisions” (= 4), so the process returns to step S62. Next, when step S62 is executed, “adjustment value” = 3 + 1 = 0 (with overflow), “adjustment value_over_flow” = 1, and “number of processing times” = 4. Next, when step S63 is executed, the fourth first count upper limit value = 2 + 1 = 3 is generated, and the fourth first count upper limit value is generated. In step S64, since the “number of processes” (= 4) is equal to the “number of divisions” (= 4), the process exits from the loop. As described above, the first count upper limit value group 521 of the first bus master 101 is generated as 2, 2, 2, 3.

  In FIG. 7, the bus arbitration information operation unit 105 generates the first period group 520 and the first count upper limit group 521 in FIG. 5 based on the second bus arbitration information 106 in FIG. In addition, an example is shown in which each of the first period group 520 and each of the first count upper limit value group 521 are sequentially set in the first bus arbitration information 108 managed by the bus arbitration information management unit 107. The timing at which the bus arbitration information operation unit 105 operates the first bus arbitration information 108 managed by the bus arbitration information management unit 107 passes after the first period 220 of the first bus arbitration information 108 has elapsed. The timing is the same as when the first access permission count 222 of the arbitration information 108 is reset to zero.

  In FIG. 7, the horizontal direction indicates time, and the bus arbitration unit 104 performs arbitration at each time indicated by a vertical line. These times are called “arbitration times”. An arrow in FIG. 7 indicates a period during which an access request is issued, a white star in FIG. 7 indicates that access is permitted, and a numerical value written below the arrow in FIG. The first access permission count 222 is shown, and the numerical value in parentheses is the remaining priority that is a value obtained by subtracting the first access permission count 222 from the first count upper limit value 221 of the first bus arbitration information 108. Indicates the permission count. The higher the remaining priority permission count, the higher the priority.

  FIG. 7 will be specifically described. In FIG. 7, it is assumed that the first to third bus masters 101 to 103 are always requesting access, and the first period (t1) 220 shown in FIG. The bus arbitration information operating unit 105 sets the leftmost value of the period group 520 and the leftmost column of the first count upper limit value group 521 as the bus arbitration information 108 in the bus arbitration information management unit 107. Shall. Since the remaining priority permission counts of the first bus master 101 and the second bus master 102 are equal at the arbitration time 0 in the first period 220 at the leftmost end in FIG. 7, the second priority order information 224 is used. Mediation takes place. As a result, the first bus master 101 having the highest priority obtains access permission. When the bus arbitration unit 104 grants access permission to the first bus master 101, the bus arbitration information management unit 107 increments the first access permission count 222 corresponding to the first bus master 101 by one. Next, at the arbitration time 1 in the first period 220 at the leftmost end of FIG. 7, the master having the largest remaining priority permission count is the second bus master 102, so the second bus master 102 obtains access permission. Arbitration is also performed in the same manner at the arbitration times 2 and 3 in the first period 220 at the leftmost end of FIG. When the mediation at the mediation time 3 in the first period 220 at the leftmost end in FIG. 7 is completed, the mediation in the first period 220 at the leftmost end is completed. When the mediation of the first period 220 at the leftmost end in FIG. 7 is completed, the second value from the left in the first period group 520 in FIG. 5 and the second value from the left in the first count upper limit group 521 in FIG. The bus arbitration information operation unit 105 sets the value in the second column as the first bus arbitration information 108 in the bus arbitration information management unit 107. Thereafter, the arbitration and the operation of the first bus arbitration information 108 are similarly repeated.

  As described above, according to FIGS. 4 to 7, the bus arbitration information operation unit 105 causes the first count upper limit values 221 for the respective bus masters to be dispersed on the average based on the second bus arbitration information 106. The first period group 520 and the first count upper limit value group 521 are generated, and one set among them is set as the first bus arbitration information 108 managed by the bus arbitration information management unit 107, so that the second Even when the count upper limit value 421 is biased, the concentration of access permission can be relaxed, and the maximum value of the access waiting period (worst latency) can be reduced.

  Note that two or more or all of the first period group 520 and the first count upper limit value group 521 as shown in FIG. 5 may be generated in advance, or the first period 220 elapses. The first period 220 and the first count upper limit 222 may be generated one by one. When the first period group 520 and the first count upper limit group 521 as shown in FIG. 5 are generated from the second bus arbitration information 106, the second of each master of the second bus arbitration information 106 is displayed. The first period group 520 and the first count upper limit group 521 may be generated after the count upper limit 421 is changed to twice or more. FIG. 5 shows an example in which the absolute value of the difference between the adjacent column values of each bus master in the first count upper limit value group 521 is 0 or 1 and is dispersed on average. The absolute value of the difference between adjacent columns of each bus master in the count upper limit value group 521 may be 2 or more, and the algorithm shown in FIG. The number of divisions between the first period group 520 and the first count upper limit group 521 is not limited to four and may be two or more.

<< Second Embodiment >>
FIG. 8 shows a second example of the second bus arbitration information 106 in FIG. 820 is a second period (T), 821 is a second upper limit value set for each bus master, and 822 is group information for grouping the bus masters. A second period 820 in FIG. 8 indicates the number of times that the bus arbitration unit 104 performs arbitration within a predetermined time. Here, since “20” is set as the second period 820, the 20 arbitration interval corresponds to the second period 820. The second count upper limit value 821 in FIG. 8 represents the minimum number of times each of the first to third bus masters 101 to 103 receives access permission within the second period 820, and the second count upper limit value The total of 821 is set in advance so as to be equal to the set value of the second period 820. Here, the second count upper limit value 821 is 9, 9, 2 for the first, second, and third bus masters 101, 102, 103, respectively. The group information 822 in FIG. 8 includes two access attributes, that is, a read attribute and a write attribute, which are access directions of the bus master. The write attribute indicates that the bus master access is a bus master with more writes than read, and the read attribute indicates that the bus master access is a bus master with more reads than write.

  FIG. 9 shows an example in which the first period group (t1 to t2) 920 and the first count upper limit group 921 are generated based on the second bus arbitration information 106 of FIG. Each of the first period groups 920 in FIG. 9 is a sum for each column of the first count upper limit group 921. The sum total of the first period group 920 in FIG. 9 coincides with the second period 820 in FIG. 8, and the sum total of the first count upper limit group 921 for each bus master in FIG. 9 is the second sum for each bus master in FIG. It is made to coincide with the count upper limit value 821. Each of the generated first count upper limit groups 921 is divided into two groups, that is, a count upper limit group of bus masters having a read attribute and a count upper limit group of bus masters having a write attribute. A count upper limit value group 921 is generated.

  10, the bus arbitration information operation unit 105 generates the first period group 920 and the first count upper limit group 921 in FIG. 9 based on the second bus arbitration information 106 in FIG. An example of arbitration when each of the first period group 920 and each of the first count upper limit group 921 is sequentially set in the first bus arbitration information 108 managed by the bus arbitration information management unit 107 is shown. Yes. The timing at which the bus arbitration information operation unit 105 operates the first bus arbitration information 108 managed by the bus arbitration information management unit 107 passes after the first period 220 of the first bus arbitration information 108 has elapsed. The timing is the same as when the first access permission count 222 of the arbitration information 108 is reset to zero. A black star in FIG. 10 indicates that the access attribute of access to the shared resource 111 changes.

  8 to 10, it is possible to reduce the number of changes in access attributes of access to the shared resource 111. When the shared resource 111 is an SDRAM, a switching process is required when switching between read and write, and access efficiency deteriorates. Therefore, reducing the number of times of switching between read and write improves the access efficiency to the shared resource 111. It leads to things.

  Note that the timing of generating each of the first period group 920 and each of the first count upper limit group 921 of FIG. 9 based on the second bus arbitration information 106 of FIG. May be generated one for each timing of operating the first bus arbitration information 108 managed by the bus arbitration information management unit 107, or the first period group 920 and the first All of the count upper limit group 921 may be generated.

  FIG. 11 shows a first modification of FIG. Reference numeral 1120 denotes a second period (T), reference numeral 1121 denotes a second upper limit value set for each bus master, and reference numeral 1122 denotes group information for grouping the bus masters. The group information 1122 in FIG. 11 includes address attributes representing the address range of bus master access. In FIG. 11, there are two address attributes: a first address attribute and a second address attribute. There are chip, bank, row, and column addresses as addresses used when accessing the SDRAM. As a feature of the SDRAM, when a row address is switched for the same bank, a switching process is required, resulting in a decrease in throughput. Therefore, the switching process can be reduced by reducing the continuous access to the same bank. When two or more shared resources 111 are configured and each of them can be accessed in parallel, access permission is alternately given to different shared resources instead of continuously granting access permission to the same shared resource. As a result, access contention can be reduced and access efficiency can be improved.

  The first address attribute and the second address attribute shown in FIG. 11 are a set of bus masters that frequently access different bank addresses. The first address attribute and the second address attribute may be a set of bus masters that are frequently accessed for different shared resources, or an address range may be arbitrarily designated for each attribute.

  According to FIG. 11, it is possible to reduce the number of changes in the access attribute of access to the shared resource 111. As a result, it is possible to reduce row address switching processing in the same bank and improve throughput.

  FIG. 12 shows a second modification of FIG. 1220 is a second period (T), 1221 is a second upper limit value set for each bus master, and 1222 is group information for grouping the bus masters. The group information 1222 in FIG. 12 includes two master attributes for grouping bus masters, that is, a first master attribute and a second master attribute. The bus masters to be arbitrated within the same first period 220 are set to the same master attribute. Here, the first and third bus masters 101 and 103 are set as the first master attribute, and the second bus master 102 is set as the second master attribute.

  According to FIG. 12, it is possible to create a period in which arbitration is performed preferentially for a specific bus master 102. As a result, it is possible to improve the latency of a specific bus master during a period of preferential arbitration while guaranteeing the access bandwidth allocated to each bus master. The number of master attributes may be three or more, and the number of bus masters belonging to one master attribute may be two or more.

  FIG. 13 is another example in which the first period group (t1 to t8) 1321 and the first count upper limit group 1322 are generated based on the second count upper limit value 821 of FIG. First, a first count upper limit group 1320 similar to the first count upper limit group 521 in FIG. 5 is generated based on the second count upper limit value 821 in FIG. The sum of the first count upper limit values 1320 for each bus master in FIG. 13 is made to match the second count upper limit value 821 for each bus master in FIG. Also, the absolute value of the difference between the adjacent column values of each bus master in the first count upper limit group 1320 in FIG. 13 is 0 or 1. Next, each column of the first count upper limit group 1320 is divided for each group attribute to generate a first count upper limit group 1322. Each of the first period group 1321 is a sum of each column of the first counter upper limit value group 1322. As a result, the sum of the first period group 1321 in FIG. 13 coincides with the second period 820 in FIG. 8, and the sum for each master of the first count upper limit group 1322 in FIG. 13 is the second sum in FIG. It corresponds to the count upper limit value 821.

  14, the bus arbitration information operation unit 105 generates the first period group 1321 and the first count upper limit group 1322 of FIG. 13 based on the second bus arbitration information 106 of FIG. An example of arbitration when each of the first period group 1321 and each of the first count upper limit group 1322 is sequentially set in the first bus arbitration information 108 managed by the bus arbitration information management unit 107 is shown. Yes.

  According to FIG. 14, the access concentration is reduced as a whole, and the latency can be reduced. Further, by arbitrating preferentially bus masters having partially the same access attribute within a partial period, it is possible to reduce the number of access attribute changes for access to the shared resource 111. Therefore, when the shared resource 111 is SDRAM, access efficiency can be improved.

  The timing for generating each of the first period group 1321 and each of the first count upper limit group 1322 of FIG. 13 based on the second bus arbitration information 106 of FIG. May be generated one for each timing at which the first bus arbitration information 108 managed by the bus arbitration information management unit 107 is operated, or the first period group 1321 as shown in FIG. All the count upper limit groups 1321 may be generated.

  The method shown in FIG. 13 can be similarly applied to the example of FIG. 11 using the address attribute as the group information 1122 and the example of FIG. 12 using the master attribute as the group information 1222.

<< Third Embodiment >>
FIG. 15 shows a third example of the second bus arbitration information 106 in FIG. 1520 is a second period (T), 1521 is a second upper limit value set for each bus master, 1522 is group information, and 1523 is access history information. The group information 1522 in FIG. 15 includes two access attributes, that is, a read attribute and a write attribute, which are directions of access by the bus master. The write attribute indicates that the bus master access is a bus master with more writes than read, and the read attribute indicates that the bus master access is a bus master with more reads than write. The access history information 1523 in FIG. 15 has a value obtained by counting the number of read accesses and the number of write accesses for each bus master with respect to accesses performed in the second period 1520. These values are cleared to 0 just before starting the next second period 1520. Based on the access history information 1523, group information 1522 is set for each bus master. In the access history information 1523, a read attribute is set in the group information 1522 for a bus master having a read count larger than the write count and a bus master having a write count larger than the read count. The timing set in the group information 1522 is immediately before the start of the next second period 1520 and immediately before the count value of the access history information 1523 is reset to zero.

  According to FIG. 15, group information 1522 including a read attribute and a write attribute can be generated according to the access status of the bus master. In SDRAM, the more switching between read and write, the lower the access efficiency. Therefore, when the shared resource 111 is SDRAM, access efficiency can be improved by giving access permission to the bus master in such an order that the number of times of switching between read and write decreases.

  It should be noted that the timing for resetting the count of the access history information 1523 to 0 may be after the second period 1520 has passed twice or more, and the timing for setting the read attribute or write attribute in the group information 1522 is also the second. It may be after two or more periods 1520 have elapsed.

  FIG. 16 shows a modification of FIG. 1620 is a second period (T), 1621 is a second upper limit value set for each bus master, 1622 is group information, and 1623 is access history information. The group information 1622 in FIG. 16 includes address attributes representing the address range of bus master access. In FIG. 16, there are two address attributes: a first address attribute and a second address attribute. The access history information 1623 in FIG. 16 stores the result of counting the number of accesses by the bus master for each address range. In FIG. 16, there are two address ranges, a first address range and a second address range. The access history information 1623 in FIG. 16 has a value obtained by counting the number of accesses to the first address range and the number of accesses to the second address range for each bus master with respect to accesses performed in the second period 1620. . These values are cleared to 0 before starting the next second period 1620. Based on the access history information 1623, group information 1622 of each bus master is set.

  According to FIG. 16, group information 1622 having address attributes can be generated according to the access status of the bus master. By preferentially arbitrating bus masters with the same access attribute within a partial period, the number of access attribute changes for access to the shared resource 111 can be reduced according to the access status.

<< Fourth Embodiment >>
FIG. 17 shows a fourth example of the second bus arbitration information 106 in FIG. 1720 is a second period (T), 1721 is a second count upper limit value, and 1722 is access history information. The access history information 1722 in FIG. 17 is a remaining priority permission count that is a value obtained by subtracting the first access permission count 222 from the first count upper limit value 221. Each time the first period 220 elapses, the remaining priority permission count is checked, and the second priority information 224 is set so that the priority becomes higher in order of the bus master having the largest remaining priority permission count. Then, after setting the second priority information 224, the first count upper limit value 221 is set in the access history information 1722. Thereafter, the process is repeated every time the first period 220 elapses.

  According to FIG. 17, the second priority information 224 is set so that the priority order of the bus masters becomes higher in the order of the bus master having the largest remaining priority permission count according to the access status. This increases the possibility that a bus master that has not been able to issue an access request in the first period 220 one time before may obtain access permission in the next first period 220, and the first count upper limit value 222 The accuracy of bandwidth guarantee by setting can be increased.

  Note that the second priority information 224 may be set every time the first period 220 passes two or more times. In this case, the reset timing of the access history information 1722 is also adjusted. The value of the remaining priority permission count at the time of reset is a value obtained by multiplying the value of the first count upper limit value 221 by the value indicated by the number of times the first period 220 has elapsed.

  It is also possible to combine the first to third embodiments and the fourth embodiment.

<< Fifth Embodiment >>
FIG. 18 shows a fifth example of the second bus arbitration information 106 in FIG. 1820 is a first period group (t1 to t8), and 1821 is a first count upper limit group set for each bus master. Here, the bus arbitration information operation unit 105 manages the first period group 1820 and the first count upper limit group 1821 set in advance, and operates the first bus arbitration information 108 based on these. The example shown in FIG. 18 corresponds to the example of FIG. 13 described above, and each time the first period 220 elapses, the first period group 1820 and the first count upper limit value group 1821 The columns are sequentially set in the first bus arbitration information 108 of the bus arbitration information management unit 107 from the left. In the example of FIG. 18, the sum total of each of the first period groups 1820 is 20, the sum of the first count upper limit group 1821 of the first bus master 101 is 9, and the first count upper limit of the second bus master 102 The sum of the value group 1821 is 9, and the sum of the first count upper limit value group 1821 of the third bus master 103 is 2. Therefore, the first to third bus masters 101 to 103 can receive access permission at a rate of 45%, 45%, and 10%, respectively.

  The first to fifth embodiments have been described above. In the above description, the number of bus masters is 3. However, the number of bus masters is actually arbitrary, and the bus form can take various forms such as a multi-layer bus. In a form in which a plurality of bus masters are connected to one bus, the present invention can also be applied to arbitration between these bus masters.

  The arbitration rule may be a rule that can guarantee a desired access permission ratio to each bus master using the first period 220, the first count upper limit value 221 and the first access permission count 222. The first example is an arbitration rule based on a priority order that is fixed between bus masters in which the remaining priority permission count obtained by subtracting the first access permission count 222 from the first count upper limit value 221 is greater than zero. The second example is a round-robin arbitration rule that sequentially changes the priority between bus masters having a remaining priority permission count greater than zero. The third example is an arbitration rule in which the read access request has a higher priority than the write access request, or the write access request has a higher priority than the read access request, between the bus masters whose remaining priority permission count is greater than 0. It is a mediation rule. The fourth example is arbitration according to a so-called LRU (Least Recently Used) algorithm that indicates a higher priority among the masters that have received access permission in the past among bus masters with a remaining priority permission count greater than zero. It is a rule. The fifth example is an arbitration rule having a higher priority as the product of the access period and the remaining priority permission count is larger between bus masters. The sixth example is a round-robin arbitration rule that sequentially changes the priority order so that the master having the largest remaining priority permission count has the highest priority. To the arbitration rule as described above, a rule may be added in which a master that has received access permission cannot subsequently be granted access permission for a certain period.

  As described above, the resource management device according to the present invention can guarantee the minimum access bandwidth of each master, further reduce the worst latency, improve the access efficiency, and increase the guaranteed bandwidth accuracy. Therefore, it can be suitably used for a system LSI or the like.

It is a block diagram which shows the structural example of the resource management apparatus which concerns on this invention. It is a conceptual diagram which shows one example of the 1st bus arbitration information in FIG. FIG. 3 is a timing diagram showing an example of arbitration based on the first bus arbitration information of FIG. 2. It is a conceptual diagram which shows the 1st example of the 2nd bus arbitration information in FIG. It is a conceptual diagram which shows the example of the 1st bus arbitration information group produced | generated based on the 2nd bus arbitration information of FIG. FIG. 6 is a flowchart showing one example of an algorithm used when generating the first bus arbitration information group in FIG. 5. FIG. 6 is a timing chart showing an example of arbitration based on the first bus arbitration information group in FIG. 5. It is a conceptual diagram which shows the 2nd example of the 2nd bus arbitration information in FIG. It is a conceptual diagram which shows the 1st example of the 1st bus arbitration information group produced | generated based on the 2nd bus arbitration information of FIG. FIG. 10 is a timing chart showing an example of arbitration based on the first bus arbitration information group in FIG. 9. It is a conceptual diagram which shows the 1st modification of FIG. It is a conceptual diagram which shows the 2nd modification of FIG. It is a conceptual diagram which shows the 2nd example of the 1st bus arbitration information group produced | generated based on the 2nd bus arbitration information of FIG. FIG. 14 is a timing chart showing an example of arbitration based on the first bus arbitration information group in FIG. 13. It is a conceptual diagram which shows the 3rd example of the 2nd bus arbitration information in FIG. It is a conceptual diagram which shows the modification of FIG. It is a conceptual diagram which shows the 4th example of the 2nd bus arbitration information in FIG. It is a conceptual diagram which shows the 5th example of the 2nd bus arbitration information in FIG.

Explanation of symbols

100 data processing systems 101 to 103 bus master 104 bus arbitration unit 105 bus arbitration information operation unit 106 second bus arbitration information 107 bus arbitration information management unit 108 first bus arbitration information 109 resource management device 110 resource control unit 111 shared resource 112 -113 Bus 220 First period 221 First count upper limit 222 First access permission count 223 First priority information 224 Second priority information 420, 820, 1120, 1220, 1520, 1620, 1720 First 2 periods 421, 821, 1121, 1221, 1521, 1621, 1721 second count upper limit values 520, 920, 1321, 1820 first period groups 521, 921, 1320, 1322, 1821 first count upper limit value groups 822, 1122, 12 2,1522,1622 group information 1523,1623,1722 access history information

Claims (15)

A resource management device interposed between a plurality of masters and at least one shared resource,
A bus arbitration information management unit that manages a first set value for each of the plurality of masters as first bus arbitration information;
A bus arbitration information operation unit for operating the first bus arbitration information managed by the bus arbitration information management unit;
A bus arbitration unit that arbitrates an access request from each master to the shared resource based on the first bus arbitration information;
A resource control unit that controls data transfer between the shared resource and the master that is permitted to access by the bus arbitration unit;
When two or more masters among the plurality of masters are simultaneously requesting access, the bus arbitration unit determines the number of times access permission has been received within a predetermined time among the two or more access request masters. A resource management apparatus characterized in that an access permission is preferentially given to an access request master less than one set value. The resource management device according to claim 1, wherein
The bus arbitration information management unit has first priority information related to priority among the plurality of masters,
The bus arbitration unit, when there are a plurality of access request masters in which the number of times access permission is received within the predetermined time among the two or more access request masters does not satisfy the first set value, A resource management device characterized in that an access request is arbitrated according to a first priority rule based on priority order information. The resource management apparatus according to claim 2, wherein
The bus arbitration information operation unit manages a second set value relating to each of the plurality of masters as second bus arbitration information, and in accordance with a bus arbitration information operation rule, based on the second bus arbitration information, A resource management device that operates one bus arbitration information. The resource management device according to claim 3, wherein
The bus arbitration information operation unit generates a first bus arbitration information group based on the second bus arbitration information, and follows the bus arbitration information operation rule based on the first bus arbitration information group. A resource management device that operates the first bus arbitration information. The resource management device according to claim 3, wherein
The bus arbitration information operation rule is a difference between the first set value before operating the first bus arbitration information and the first set value after operating the first bus arbitration information. A resource management device, wherein the absolute value of the rule is 0 or 1. The resource management device according to claim 3, wherein
The bus arbitration information operation unit manages, as group information, information for dividing the second bus arbitration information into a plurality of groups according to the access characteristics of the plurality of masters.
The resource management apparatus according to claim 1, wherein the bus arbitration information operation rule is a rule for operating the first bus arbitration information based on the group information. The resource management device according to claim 6, wherein
The resource management apparatus according to claim 1, wherein the group information includes two attribute information: read attribute information indicating that there are more writes than reads of the shared resource, and write attribute information indicating that there are more reads than writes. The resource management device according to claim 6, wherein
The resource management apparatus, wherein the group information includes information regarding a range of a plurality of access addresses to the shared resource. The resource management device according to claim 6, wherein
The resource management apparatus, wherein the group information includes information for specifying individual masters among the plurality of masters. The resource management device according to claim 6, wherein
The bus arbitration information operation unit manages, as access history information, access information of a master permitted by the bus arbitration unit to access the shared resource, and generates the group information based on the access history information. A featured resource management device. The resource management device according to claim 10, wherein
The resource management apparatus, wherein the access history information includes a read access count and a write access count to the shared resource. The resource management device according to claim 10, wherein
The resource management apparatus, wherein the access history information includes information related to an address range of access to the shared resource. The resource management apparatus according to claim 2, wherein
The bus arbitration information management unit has second priority information related to priority among the plurality of masters,
The bus arbitration unit has a maximum of the two or more access request masters when the remaining priority permission count is a count obtained by subtracting the number of times access permission is received within the predetermined time from the first set value. A resource management device characterized in that when there are a plurality of access request masters having equal remaining priority permission counts, an access request is arbitrated according to a second priority rule based on the second priority information. The resource management device according to claim 13, wherein
The bus arbitration information operation unit manages information related to the remaining priority permission count for each of the plurality of masters as access history information, and operates the second priority information based on the access history information. A featured resource management device. The resource management apparatus according to claim 2, wherein
The bus arbitration information operation unit manages a set first bus arbitration information group and operates the first bus arbitration information based on the first bus arbitration information group. Management device.

Images