JP2005108262A - Data processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To give flexibility to a replacement object of an address translation counterpart in an address translation mechanism. <P>SOLUTION: This data processor for supporting a virtual memory has a buffer memory (1) consisting of four-way set associative type cache memory provided with storage areas for storing correspondence information between logical page numbers (VPN) and physical page numbers (PPN), respectively, having a plurality of banks (11 to 14) having their index addresses shared, when storage information should be replaced from among the plurality of banks by the occurrence of cache mistakes, etc., a set to be replaced is optionally specified by execution of software by a central processor. For example, two-bit information for optionally specifying the bank is set in a register (MMUCR. RC). Signals (BSK1 to BSL4) for selecting one of the four banks are formed by decoding of a value set in the register. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a data processing apparatus having an address conversion mechanism, and more particularly to a data processing apparatus that uses a set associative cache memory as the address conversion mechanism, and further intends to diversify the address conversion method. For example, the present invention relates to a technique effective when applied to a microcomputer.

  In a field where an operating system (hereinafter also referred to as OS) performs memory management without the user being aware of real memory, the data processing device needs to support an address translation mechanism. The address conversion mechanism is a mechanism for converting a logical address formed by a central processing unit (CPU) into a physical address in order to realize virtual storage. In order to execute this address translation mechanism at high speed, an address translation buffer (translation lookaside buffer, hereinafter simply referred to as TLB) holding a translation pair between a logical address and a physical address is built in the data processing apparatus together with the central processing unit. Technology is adopted. The address translation buffer is configured, for example, as a buffer memory having an associative memory structure that holds a translation pair of a recently used logical address and physical address. A buffer memory with an associative memory structure achieves a relatively high hit rate using a fully associative memory consisting of CAM (Content Addressable Memory) with a circuit configuration for comparison in each memory cell and a general-purpose random access memory. A set associative type associative memory can be used. Examples of documents describing the address conversion buffer in the associative memory format include pages 287 and 288 issued from Japan.

“Ultra-high-speed MOS device”, Japan, Baifukan Co., Ltd., February 10, 1986, p. 287-288

    The present inventor has examined the point that such an address translation buffer can meet the user's required specifications and realizes good usability, and has found the following problems.

  (1) In an address translation mechanism that supports virtual memory, a logical address space is divided into units called logical pages, and address translation to physical addresses is performed in units of pages. On the other hand, when the total storage capacity of the actual memory (physical memory) actually mounted in the system is small, the logical page size should be made relatively small to improve the memory utilization efficiency by each process. There is a request. For example, when a task to be executed is configured by a program having a relatively small size, if the size of a logical page allocated to the task is relatively large, the physical for executing the task is correspondingly corresponding to this. The page size is also relatively large. Therefore, more storage space than necessary is allocated to the task, and the memory utilization efficiency is reduced. In particular, when the storage capacity of the real memory is relatively small, it is desired to reduce the size of the logical page to prevent the use efficiency of the real memory from decreasing. In order to be able to respond appropriately to such a request, it is desirable to make the size of the logical page variable. However, when the size of the logical page is made variable, the number of bits of information for defining the logical page in a certain logical space is changed. As a result, the number of bits and the bit position of comparison target information for associatively retrieving information from the address translation buffer must be changed according to the logical page size. In order to cope with this, it is conceivable to adopt a full associative format composed of CAM for the address translation buffer. In this way, since each memory cell includes a comparison circuit, the logical page can be changed relatively easily without requiring special consideration. However, since each memory cell is provided with a comparison circuit, there is a drawback that both the chip-occupied area and the power consumption are doubled as compared with a set-associative address translation buffer.

  (2) In a set associative address translation buffer, if the number of banks in a plurality of sets, that is, the number of ways is increased, the number of entries that can be held for one index address can be increased and the hit rate can be improved. . For example, in the case of the 4-way set associative format, a maximum of 4 entries can be held for one index address. However, when each of a plurality of processes has address translation information over the entire logical space and performs multiple virtual storage in which a logical address is modified or expanded by a process number, a relatively large number of processes are started in parallel. Each process uses the same logical page more frequently. Since the processes have different process numbers, the processes are held in different entries indicated by one index address. Therefore, when the number of processes exceeds the number of ways, the hit rate is relatively lowered even in the set associative format. Therefore, in a usage mode in which many processes are started in parallel in multiple virtual memory, it is necessary to be able to select an indexing method according to the usage mode so as to suppress a decrease in the hit rate. Was found.

  (3) If the target translation pair is not stored in the address translation buffer, the translation pair related to the cache miss (desired translation pair) is added to the address translation buffer as a new entry. At this time, if all of the conversion pairs in the indexed entry are valid conversion pairs, the conversion pairs are replaced. As the replacement algorithm (replacement algorithm), there are random, first-load FIFO to be replaced, or last-referenced LRU (Least Recentry Used). However, when the replacement algorithm is fixed, for the sake of data processing, a request to always store a specific translation pair as an entry in the address translation buffer or a specific address translation pair is not desired to be replaced. I can't answer any request.

  (4) In addition to the above-described multiple virtual memory, there is a single virtual memory in which a part of the logical address space is exclusively allocated to a plurality of processes. It is desirable to be able to support both of these in terms of improving usability.

  An object of the present invention is to realize a data processing apparatus having an address translation mechanism that can meet user's required specifications and is easy to use.

  The object of the present invention will be described in further detail as follows. To provide a data processing apparatus having an address conversion mechanism capable of supporting a plurality of page sizes without increasing a chip exclusive area and power consumption.

  A data processing apparatus including an address translation mechanism that supports a plurality of index methods and from which an index method can be selected is provided.

  To provide a data processing device including an address conversion mechanism that can give a degree of freedom to a replacement target of an address conversion pair.

  In addition, the present invention provides a data processing apparatus having an address translation mechanism that can give a degree of freedom to a virtual storage format that can be supported.

  The above and other objects and novel features of the present invention will be apparent from the description of this specification and the accompanying drawings.

  The following is a brief description of an outline of typical inventions disclosed in the present application.

  << Supporting Multiple Page Sizes >> A data processing apparatus that supports virtual storage divides a logical address space into units called logical pages, and converts logical addresses into physical addresses (address conversion) in units of pages. In this data processing apparatus, as illustrated in FIG. 1, the size of the logical page is variable for each page. A buffer memory 1 such as an address translation buffer is commonly used for a plurality of logical pages each having a variable size. The buffer memory 1 is provided with a plurality of banks 11 each provided with a storage area for storing correspondence information including a logical page number VPN and a corresponding physical page number PPN, and accessed by a common index address 2. A set associative cache memory with ~ 14 is used.

  The maximum size of the logical page to be supported is a power of 2 times the minimum size (2 to the power of N), and the number of banks of the set associative buffer memory is greater than the power of 2 (the power of 2 to the N power). Is done. More specifically, as illustrated in FIG. 1, in a logical address space (specified by a logical address of bits 0 to 31) of 4 GB (gigabytes) as a whole, the size of the logical page is 4 KB (kilobytes). ) And 1 KB. In this case, the number of banks of the buffer memory 1 is four, and the buffer memory 1 is configured as a 4-way set associative cache memory. When the logical page size is 4 KB, the logical address offset is bits 0 to 11 of the logical address, and the logical page number vpn is bits 12 to 31 of the logical address. When the logical page size is 1 KB, the logical address offset is bits 0 to 9 of the logical address, and the logical page number vpn is bits 10 to 31 of the logical address. In this specification, vpn, ppn, and asid indicated by lowercase letters are logical page numbers, physical page numbers, and process numbers used for access, and entries or pages of a buffer memory (address translation buffer) indicated by uppercase letters. It is distinguished from the logical page number VPN, physical page number PPN, and process number ASID as table entries.

  In the address conversion buffer, the index address designation method for the buffer memory 1 is such that the logical page size is common to both 4 KB and 1 KB, and a predetermined plurality of bits from the least significant logical page address of the maximum page size are index addresses. As given. According to the example of FIG. 1, the buffer memory 1 is indexed using 5 bits of bits 12 to 16 in a logical address of 32 bits in total. Since the index address is 5 bits, a maximum of 32 entries can be held per bank. In the case of having four banks as in the example of FIG. 1, a maximum of four entries can be held for one index address. When the page size is 4 KB, the index address is 5 bits (bits 12 to 16) from the lowest order of the logical page number vpn. Therefore, when the page size is 4 KB, an entry of an arbitrary logical page number is used. Can be stored in each bank at a maximum of 32 (128 entries in total). On the other hand, when the page size is 1 KB, the least significant 2 bits (bits 10 and 11) of the logical page number vpn are not used for the index. That is, when the logical page size is 1 KB and only the index for one bank is considered, one index is to be selected from 2 bits (bits 10 and 11) that are not used for the index. One of the four logical page numbers. Therefore, when there is only one bank, only one of four consecutive logical pages can be held as an entry. In this respect, the maximum size (4 KB) of the logical page is set to the square of 2 of the minimum size (1 KB), and the number of banks is also set to the square of 2 (4), thereby providing buffer memory (address conversion). As a whole, the buffer) can hold a total of 128 entries having an arbitrary logical page number in the same manner as the 4 KB page size. However, the logical page number of an entry that can be held in one bank is restricted by every 4 KB. This restriction has no effect on the hit (TLB hit) rate if the logical page numbers of entries held in the address translation buffer are continuous. In the case of decentralization, the hit rate is affected to some extent. Even in this case, if 1KB logical page is address-mapped for every 2KB, the influence can be reduced, and if address mapping is made for every 4KB, it can be made not affected at all.

  In order to determine a TLB hit / miss, the number of address comparison bits to be reflected in the hit determination in the comparison unit 15 provided in each bank 11 to 14 must be changed according to the logical page size. According to the example of FIG. 1, when the logical page size is 1 KB, bits 10 and 11 of the logical address must be compared as compared with the case where the logical page size is 4 KB. The buffer memory 1 is set with the number of bits in the area for storing the correspondence information (conversion pair) between the logical page number and the physical page number so as to cover all the number of bits to be compared. In the buffer memory 1, a field indicating the logical page size supported by each conversion pair is provided, and the number of bits for address comparison for hit determination changes according to the value (size bit) SZ of the field. Is done. The object whose comparison bit number is changed is both the logical page number of the logical address and the logical page number included in the conversion pair in the buffer memory. Hit 1 to hit 4 are hit signals for notifying the hit determination result in each of the banks 11 to 14, and a TLB hit / miss is determined based on them.

  A data processing apparatus that selectively supports multiple page sizes can reduce the size of logical pages when the total storage capacity of real memory installed in the system is small, and the memory usage efficiency by each process It is possible to easily meet the demand to improve the quality. At this time, the maximum logical page size is set to a power of 2 times the minimum size, and the number of banks of the set associative buffer memory is set to a power of 2 or more. Regardless of the size of the selected logical page, in principle, the same number of entries can be held in the buffer memory for any logical page number entry. Changing the number of bits of the address comparison to be reflected in the hit determination using information for indicating the size of the logical page means changing the number of bits and the bit position of the comparison target information for associatively searching the buffer memory. It can be easily changed according to the size of the logical page. Realizing a buffer memory that supports a plurality of page sizes with a set-associative cache memory halves both the chip-occupied area and the power consumption compared to the case where it is configured with a CAM.

  << Supporting multiple index methods >> Data processing devices that support virtual storage each have storage areas for storing correspondence information between logical page numbers and physical page numbers, and each is accessed by a common index address. A buffer memory composed of a set associative cache memory having a plurality of banks. And it has a means to make variable the index address generation method for the buffer memory. For example, in the case where there are a plurality of processes, each process has respective address translation information, and supports multiple virtual memories in which each process is distinguished by a process number, an address for indexing an entry from the buffer memory As a designation method, a method of decoding only a part of the logical address (bits 12 to 16) as illustrated in FIG. 2 and a part of the logical address (bits 12 to 16) as illustrated in FIG. A method of decoding the result of modification (XOR = exclusive OR) by a part of the current process number (asid), and register MMUCR. The instruction can be made according to the logical value of IX. 3 and 2, the information of the logical page number VPN included in the conversion pair in the buffer memory 1 is excluded from the information corresponding to vpn (16-12) used for the index, and VPN (31 -17), VPN (11-10). Here, the notation of vpn (16-12) means the information of the logical page number including bits 12 to 16 of the logical address. The notation of VPN (31-17) means the information of the logical page number as the buffer memory entry corresponding to bit 17 to bit 31 of the logical address.

  According to the means for supporting a plurality of index methods, the index method can be set according to the usage mode so that a decrease in hit rate can be suppressed in a usage mode in which many processes are started in parallel in multiple virtual memory. Can be selectable. A hit in a usage mode in which many processes are started in parallel in multiple virtual memory by qualifying a part of the logical address with a process number (asid) using the logical address and indexing the buffer memory with this. It is possible to suppress a decrease in rate.

  << Degree of Freedom of Replacement >> As illustrated in FIG. 4, each data processing apparatus that supports virtual storage provides storage areas for storing correspondence information between logical page numbers VPN and physical page numbers PPN, respectively. The buffer memory 1 includes a 4-way set associative cache memory having a plurality of banks 11 to 14 with a common index address. When a cache miss occurs, stored information is stored in the plurality of banks. In the case of replacement, the set to be replaced can be arbitrarily specified by execution of software by the central processing unit. For example, MMUCR. RC is a register in which 2-bit information for arbitrarily designating the bank is set. A value (BSK1 to BSL4) for selecting one of the four banks 11 to 14 is formed by the decoder 17 decoding the set value. Thus, one of the four banks 11 to 14 designated by the selection signal and one entry designated by the index address 2 is a replacement target. Bits MMUCR. When setting of a value for RC is performed by hardware means, designation of a bank by the hardware means can be arbitrarily changed by execution of software by the central processing unit (CPU).

  As illustrated in FIG. 4, the hardware means includes MMUCR. RC is provided as a counting means such as a random counter, and the power of 2 of the number of bits of the counting means matches the number of banks of the buffer memory 1 (= 4). The control circuit CTRL included in the hardware means increments the counting means by 1 when replacement of the stored information for the buffer memory 1 is necessary, and is effective for any indexed storage area of any bank. Is held (each entry of each indexed bank has a valid bit V of “1” indicating validity), the incremented result is set as a bank number to be replaced. On the other hand, valid data is not held in the indexed storage area in any bank (any indexed entry has a valid bit V of “0” indicating invalidity). In this case, the bank number including the entry that does not hold valid data is set in the counting means, and the bank having the set number is set as the bank to be replaced. The bank to be replaced is designated according to such a certain rule. At this time, each bit of the counting means is a target that can be changed to an arbitrary value by software executed by the central processing unit.

  According to the means for improving the degree of freedom of replacement, it is possible to arbitrarily determine the bank to be replaced by software without fixing the replacement algorithm for replacing the conversion pair of the buffer memory. As a result, it is possible to easily respond to a request to always store a specific translation pair as an entry in the address translation buffer for convenience of data processing or a request not to make a specific address translation pair to be replaced. . Having a register for designating a replacement target by software can guarantee the degree of freedom of replacement according to random, FIFO, LRU, or the like, depending on the algorithm of the software.

  << Support for Single Virtual Storage and Multiple Virtual Storage >> A data processing device that supports virtual storage divides the logical address space into units called logical pages, and converts the logical addresses into physical addresses for each page. It has an address translation mechanism to do it. The address translation mechanism includes a multiple virtual memory that modifies or expands a logical address by a process number (asid) when each of a plurality of processes has address translation information over the entire logical space, and a logical address space for a plurality of processes. Having a single virtual memory that does not qualify or extend a logical address by a process number (asid) when a portion is allocated exclusively and each process has address translation information for the logical address space assigned to it, The multiple virtual memory and the single virtual memory can be selected. Therefore, as illustrated in FIG. 5, each entry in the buffer memory is provided with a process number ASID field in addition to a conversion pair of a logical page number VPN and a physical page number PPN. The contents of this field differ depending on whether it is a single virtual memory or multiple virtual memories. When a logical page corresponding to a process number included in a certain TLB entry cannot be shared with other processes, the contents of the field of the process number ASID are the hit / Used to determine a miss (TLB hit / TLB miss). That is, in this case, the information on the logical page number VPN of the entry stored in the buffer memory matches the information on the logical page address vpn, and the process number ASID of the entry matches the current process number asid. Otherwise, it will not be a TLB hit. In the single virtual memory, the contents of the field of the process number ASID are used as memory protection information. That is, in this case, when a TLB miss occurs due to a difference in the process number, the software determines whether the cause of the TLB miss is due to a difference in the process number or a difference in the logical page address. In case of, it is processed as a protection error. Whether the single virtual memory or the multiple virtual memory is the bit MMUCR. Of the register MMUCR illustrated in FIG. Indicated by the value of SV. This bit MMUCR. The value of SV can be arbitrarily set by software executed by the central processing unit.

  According to the means for supporting single virtual memory and multiple virtual memory, the usability of the address translation mechanism can be improved. In particular, it is possible to select which one to use via software, thereby further improving usability. By using the process number (ASID) in the multiple virtual memory as the memory protection information in the single virtual memory, it is possible to easily realize the complete memory protection when the single virtual memory is selected.

  The effects obtained by the representative ones of the inventions disclosed in the present application will be briefly described as follows.

  That is, according to one representative invention, it is possible to realize a data processing apparatus capable of supporting a plurality of page sizes without increasing the chip exclusive area and power consumption.

  According to another typical invention, an address translation mechanism that can select an index method can be realized.

  According to another representative invention, the replacement target of the address translation pair can have a degree of freedom.

  According to yet another representative invention, the virtual storage format that can be supported can also have a degree of freedom.

  As a result, it is possible to realize a data processing apparatus that can meet the user's required specifications and has an easy-to-use address translation mechanism.

  << Microcomputer >> FIG. 7 shows a main part of a microcomputer as an embodiment of the data processing apparatus according to the present invention. The microcomputer of the present embodiment is not particularly limited, but is formed on a single semiconductor substrate such as single crystal silicon by a known semiconductor integrated circuit manufacturing technique. The figure shows a logical address bus VABUS, a physical address bus PABUS, a data bus DBUS, a central processing unit (CPU) 3, a cache memory (CACHE) 4, an address translation buffer (TLB) 1 as a buffer memory, and a TLB controller ( TLBC) 5 is shown as a representative circuit block. The CPU 3 fetches an instruction from a program memory (not shown), performs data processing according to the instruction description, outputs a logical address to the logical address bus VABUS when external access is required, and inputs data via the data bus DBUS. Output. The cache memory 4 is not particularly limited, but has a 4-way set associative format. The bank is indexed using a part of the logical address supplied from the logical address bus VABUS, and a physical address is included in the tag portion of the entry. And the indexed tag portion is compared with a physical address whose logical address is converted by the address conversion buffer 1 and output to the physical address bus PABUS, and a cache miss / hit is determined according to the comparison result.

  The microcomputer according to the present embodiment divides a logical address space into units called logical pages, and supports virtual storage for performing address conversion from logical addresses to physical addresses in units of pages. The address translation buffer 1 stores a translation pair associated with the logical page number VPN and the corresponding physical page number PPN as a TLB entry, and the TLB controller 5 receives the logical address output from the central processing unit 3 as the address translation buffer 1. Etc. to convert to a physical address. The address conversion buffer 1 is constituted by a 4-way set associative cache memory having four banks each having a common index address. In order to avoid the complexity of the drawing, FIG. 7 typically shows only one bank 11, but actually, another bank 12-similar to the bank 11 as shown in FIG. 14 is arranged in the front and back direction of the paper. The address translation buffer 1, TLB controller 5, CPU 3, and system software or operating system for address translation and memory protection constitute an address translation mechanism that supports the virtual memory.

  Here, the address space of the microcomputer of this embodiment will be described in advance. Since the microcomputer of this embodiment supports a 4 GB logical address space, the central processing unit 3 forms a 32-bit logical address. The logical address can be expanded by a process number (also referred to as a space number) asid described later. 8A and 8B show address mapping of the supported logical address space. In these drawings, an area indicated as “Mapped” is an area to be subjected to address translation using the address translation buffer (TLB) 1. An area from H′FFFFFFFF (H ′ means a hexadecimal number) to H′80000000 is an accessible area in the privileged mode, and an access in the user mode is an address error. The P4 area is a control space to which peripheral control registers and the like are mapped. The P1 and P2 areas are areas where the corresponding physical addresses are fixed, and are not subject to address translation using the address translation buffer 1. The logical addresses of these areas P1 and P2 are converted into physical addresses by adding or subtracting constant constants. Therefore, in the privileged mode, when accessing the areas P1 and P2, no exception related to address translation such as a TLB miss occurs. In particular, the P2 area is not subject to caching by the cache memory CACHE, and the P1 area is subject to caching.

  As described above, the microcomputer according to the present embodiment has a privileged mode and a user mode, and the status included in the central processing unit 3 indicates whether it is operating in the privileged mode or the user mode. It is indicated by the value of mode bit MD of register SR. MD = 0 represents operating in the user mode, and MD = 1 represents operating in the privileged mode. As shown in FIGS. 8A and 8B, the privileged state (the state in which the privileged mode is set) is an address error in the access in the user state (the state in which the user mode is set). Is different from the user state in that an address space (P1 to P4) can be accessed and privileged instructions or system control instructions that cannot be executed in the user state can be executed.

  The 4 GB logical address space shown in FIGS. 8A and 8B is not particularly limited, but the P0 area and the P3 area are divided into a plurality of units called logical pages and divided page units. Thus, the logical address is converted into a physical address. There are two types of logical page sizes supported by the microcomputer of this embodiment: 4 KB and 1 KB. This is because the maximum logical page size (= 4 KB) supported is a power of 2 times the minimum size (= 1 KB), and the number of banks of the address translation buffer 1 (= 4) is greater than the power of 2 Satisfy the relationship of As shown in FIG. 5, when the logical page size is 1 KB, bits 0 to 9 are among the logical addresses (32-bit addresses from bit 0 to bit 31) formed by the central processing unit. An offset is used, and bits 10 to 31 are logical page numbers (logical page addresses). When the logical page size is 4 KB, bits 0 to 11 of the logical address are offset, and bits 12 to 31 are a logical page number (logical page address). The index for the address translation buffer 1 includes information on the bit position of the lower 5 bits of the logical page address of the 4 KB page size, in other words, the 32-bit logical address, regardless of whether the page size is 4 KB or 1 KB. Bits 12 to 16 are used. When converting a logical address to a physical address, the physical page number is obtained from the entry corresponding to the logical page number of the logical address, and the offset information of the logical address is added to the lower side of the physical page number to obtain the physical address. It is done.

  When a TLB miss or the like occurs, information (page table entry) taken into an entry in the address translation buffer 1 is formed in advance by software, and is not stored in FIG. RAM). That is, the page table entry for virtual storage includes a description of conversion information indicating the correspondence between the logical page number VPN and the physical page number PPN, access attributes, and the like, and is a page table formed in the external memory. Stored in Some of the entries stored in the page table are stored in the banks 11 to 14 in the TLB 1 so that high-speed address translation can be performed. The address of the page table in the external memory is set in advance in a conversion table register (TTB) 53 described later with reference to FIG. When a TLB miss or the like occurs, the CPU 3 accesses the page table using the base address set in the conversion table register 53, the logical page number at the time of the miss, and the logic at that time. An entry including a physical page number or the like corresponding to the page number is obtained, and the obtained entry is written to a bank in TLB1, for example. Thereby, a physical address corresponding to the logical address at that time is formed.

  The entries stored in each bank in the TLB 1 are divided into an address part and a data part for convenience, as shown in detail in FIG. The address part includes logical page number information VPN (31-17), VPN (11-10), a valid bit V (1 bit) indicating that the entry is valid, a space number ASID (8 bits), a size bit SZ ( 1 bit) and a sharing status SH (1 bit). The data part includes a protection PR (2 bits) for storage protection, a physical page number PPN (22 bits) corresponding to the logical page number, a dirty bit D (1 bit), and a cacheable bit C (1 bit). Possess. The information of the logical page number held by the address part is composed of bits 10 to 11 and bits 17 to 31 excluding bits used for the index in the bit format of the 32-bit logical address. The former is expressed as VPN (11-10), and the latter is expressed as VPN (31-17). The sharing status SH indicates whether or not the page is shared among a plurality of processes. SH = 0 means non-shared and SH = 1 means shared. The space number ASID is used to define a logical page as belonging to a specific process, and is also referred to as a process number. The protection PR is data encoded to define the access right for the page, and the access right is defined in a manner shown in FIG. 9 by a combination of the values. The size bit SZ is a bit for designating a logical page size. A logical value 1 designates a 4 KB page size, and a logical value 0 designates a 1 KB page size.

  Here, the concept of single virtual memory and multiple virtual memory will be described with reference to FIGS. Multiple virtual storage means that when each of a plurality of processes has address translation information over the entire logical space, the logical address is modified or expanded by the process number asid. On the other hand, in the single virtual memory, a part of the logical address space is exclusively allocated to a plurality of processes. In other words, in the single virtual memory, when each process has the address translation information of the logical address space assigned to it, the logical address is not modified or expanded by the process number asid. In this way, in a single virtual memory, a logical address space is exclusively allocated among a plurality of processes. Therefore, as shown in FIG. 20B, for each logical address space allocated to a process. There is unique address translation information. Therefore, there is only one address conversion table. If the one address conversion table is used, a certain logical address A is uniquely converted into a physical address D corresponding thereto. On the other hand, in the multiple virtual memory, logical address spaces are allocated overlappingly among a plurality of processes. For this reason, the address translation information for each process must be included in different address translation tables. Accordingly, a certain logical address A is converted into different physical addresses B and C through different address conversion tables i and j. At this time, the process number identifies the process to which the logical address A related to the conversion belongs. According to FIG. 20A, in the multiple virtual memory, the address conversion table i corresponds to the process number i, and the address conversion table j corresponds to the process number j. The process number can be regarded as an identification number in a plurality of processes that use (access) the same logical address space.

  FIG. 19 shows the difference between single virtual storage and multiple virtual storage regarding logical space, conversion information, and protection. In this drawing, PR is the protection information shown in FIG.

  In FIG. 7, the TLB controller 5 includes a control circuit (CTRL) 50, a page table entry upper (PTEH) register 51, a page table entry lower (PTEL) register 52, a conversion table base (TTB) register 53, a TLB exception address (TEA). A register 54 and a control (MMUCR) register 55 are provided. As will be described later with reference to FIG. 22, these registers are coupled to the CPU 3, and at least the TEA register 54 and the MMUCR register 55 are directly readable / writable by the CPU 3. The other registers 51 to 53 are also directly accessible by the CPU 3.

  As shown in FIG. 6, the PTEH register 51 and the PTEL register 52 are provided with a field that can hold a page table entry for updating or adding a TLB1 entry in a TLB miss or the like. The PTEH register 51 is set with the space number asid of the current process from the central processing unit 3 by execution of software. The PTEH register 51 also has a function of holding the logical page address vpn of the TLB miss output by the CPU 3 when a TLB miss occurs. The logical address offset output by the CPU 3 is held in a latch circuit (not shown) inside the CTRL 50. The TTB register 53 holds the base address of the current page table. When an exception related to TLB or an address error exception occurs, the TEA register 54 holds a logical address at that time. The MMUCR register 55 is a bit AT (logical value 1 = valid, logical value 0 = invalid) that indicates whether to enable or disable address translation, a bit TF that indicates flushing of TLB1, and a 2-bit random counter field RC, index mode bit IX for designating index mode, single virtual bit SV for indicating whether to select single virtual memory or multiple virtual memory (single virtual memory is selected with SV = 1, SV = 0 Select multiple virtual memory). When the index mode bit IX is 1, asid (4-0), that is, bits 0 to 4 of the current space number asid and the output from the central processing unit 3, as shown in FIG. An index method using bits 12 to 16 of the logical page number vpn is selected. On the other hand, when IX is 0, as shown in FIG. 2, an index method using bits 12 to 16 of the logical page number vpn output from the central processing unit 3 is selected. The More specifically, as shown in FIG. 7, the index address in the case of IX = 1 includes the bit 0 to bit 4 of the space number asid held by the PTEH register 51 and the logical page number output from the CPU 3. It is formed by an exclusive OR gate XOR which inputs bits 12 to 16 of vpn, that is, vpn (16-12). The selector 18 in FIG. 7 selects either the output hashed by the exclusive OR gate XOR or vpn (16-12) as an index address. The selection control is determined according to the value of the index mode bit IX, and is performed by a selection signal 550 output from the CTRL 50.

  In FIG. 7, of the logical page number vpn (31-10) of the logical address output from the CPU 3, vpn (16-12) corresponding to bit 12 to bit 16 is the four banks 11 to 11 in the TLB. 14 is used to generate an index address 2 common to 14. As described above, since the logical page number is represented by 5 bits, one is selected from each of 32 entries in each bank by the common index address 2, and is read out. Each entry selected includes VPN (31-17), VPN (11-10), ASID, SH, SZ, V, PPN (31-10), PR, C, and D as information. Of the information in the selected and read entry, the read VPN (31-17) is compared with the logical address vpn (31-17) output from the central processing unit 3 by the comparator 151. The VPN (11-10) is compared with the logical address vpn (11-10) output from the central processing unit 3 by the comparator 152, and the read ASID is the current space number held in the PTEH register 51. Asid is compared with the comparator 153. For the comparison result, the control logic 154 determines the miss / hit of the bank 11 in consideration of the values of the shared status SH, the size bit SZ, the single virtual bit SV, and the mode bit MD. hit 1 is a hit signal as a result of determination of a miss / hit in the bank 11. The comparators 151 to 153 and the control logic 154 are provided in the banks 11 to 14, respectively. In FIG. 7, the hit signals hit2 to hit4 are representatively shown for the banks 12 to 14. According to this embodiment, the outputs of the respective comparators 151 to 153 are set to the coincidence level. The hit signals hit1 to hit4 are also set to the high level. The control circuit 50 determines a TLB hit when any one of the hit signals hit1 to hit4 is set to the hit level. Reference numeral 501 in FIG. 7 denotes a TLB hit signal for notifying the CPU 3 of the TLB hit / miss determination result. When a TLB miss occurs, a process for replacing an entry in the address translation buffer 1 is performed. In this replacement process, the contents of the entry to be replaced are held in the PTEH register 51 and the PTEL register 52, and information to be replaced from the PTEH register 51 and the PTEL register 52 to the TLB1 is supplied and stored in the TLB1. The index method for selecting an entry to be replaced is the same as the index method at the time of reading described above, but the bank (set) in which the entry is stored depends on the RC value of the MMCCR register 55 (MMUCR. RC).

  FIG. 21 shows an example of the logical configuration of the control logic 154. Reference numeral 1541 denotes a 3-input AND gate, 1542 denotes a 2-input OR gate, and 1543 denotes a 2-input OR gate. The OR gate 1542 is a comparator that is a comparison result between the VPN (11-10) included in the indexed TLB entry (the entry selected and read by the index address) and vpn (11-10) included in the logical address. The output of 152 and the size bit SZ from the indexed TLB entry are input. When the size bit SZ is set to 1 and a logical page size of 4 KB is indicated, in the address comparison for the TLB miss / hit determination, the VPN (11-10) in the TLB entry and the vpn (CPU 3) It is not necessary to make a comparison with 11-10). Therefore, when the size bit SZ = 1, the comparison result of the address bits 11 and 10 is not reflected in the output signal hit1 of the AND gate 1541. When the size bit SZ is set to 0 and a logical page size of 1 KB is instructed, the comparison operation of the address bits 11 and 10 is required for the TLB miss / hit determination, and the comparison result is added to the signal hit1. Reflected.

  The OR gate 1543 includes the output of the comparator 153, which is a comparison result between the process number ASID included in the indexed TLB entry and the current process number (process number output from the PTEH register 51) asid, and the indexed The shared bit SH from the TLB entry is input. If the shared bit SH is set to 1, the state of “shared logical page between processes” is instructed. Therefore, in this state, the comparison between the process numbers is not required in the address comparison for the TLB miss / hit determination. Therefore, in the state where the shared bit SH = 1, the output signal hit1 of the AND gate 1541 includes between the process numbers. Comparison results are not reflected. When the shared bit SH in the index is set to 0 and the state of “unshared logical page between processes” is instructed, comparison between process numbers is performed in the address comparison for TLB miss / hit determination. Therefore, when the shared bit SH = 0, the output signal hit1 of the AND gate 1541 reflects the comparison result between the process numbers.

  In the single virtual memory, the process number ASID held by the TLB entry is used as memory protection information (domain number). Sharing or non-sharing is instructed by the shared bit SH in both single virtual memory and multiple virtual memory. When non-sharing is instructed, in the multiple virtual memory, a mismatch between the current process number asid and the process number ASID included in the TLB entry is regarded as a TLB miss. On the other hand, when non-sharing is instructed, in the single virtual memory, the mismatch between the process numbers asid and ASID is used for detecting a TLB protection violation exception. In order to realize this, the control circuit 50 has an AND gate 502 shown in FIG. 21 as one logic for checking the access right. The AND gate 502 receives the inverted output of the OR gate 1543, the single virtual bit SV from the MMUCR register, and the inverted signal of the mode bit MD in the status register in the central processing unit 3, and detects a TLB protection violation exception detection signal. 503 is formed. The detection signal 503 has a high level as a detection level of a TLB protection violation exception. The TLB protection violation exception is detected by the signal 503 when the process numbers do not match and are not shared (the output of the OR gate 1543 is low level), single virtual memory (SV = 1), user mode (MD = This is when the condition 0) is satisfied. That is, in the single virtual memory, when the process numbers do not match and are not shared, a TLB miss is substantially caused. This state is regarded as a TLB protection violation exception for memory protection. In the privileged state indicated by mode bit MD = 1, it is desirable to be able to access a logical page assigned to another process, so that a TLB protection violation exception is not detected in the privileged state of MD = 1. I have to.

  FIG. 10 is a main flowchart showing control of the address conversion operation by the control circuit 50. This control is roughly divided into index processing L1 of the address translation buffer 1, address comparison and V-bit check processing L2, access right check processing L3, and physical address generation processing L4. These processes are controlled by the CPU 3 and the controller 5.

  In the index processing L1 of the address translation buffer 1, the logical address used for it is set to vpn (16-12) regardless of the size of the logical page. In this index processing, whether or not to use a logical address to be used as an index address is a hashed part of the space number asid (asid (4-0)) using an exclusive OR gate XOR. It is determined by the value of IX (MMUCR.IX) in the MMUCR register 55. As shown in FIG. When IX is 1, the logical address to be used is hashed by asid (4-0) to be used as an index address. In contrast, MMUCR. When IX is 0, vpn (16-12) is used as an index address as it is. The former index method is shown in FIG. 3, and the latter index method is shown in FIG. When TLB1 is indexed, one of 32 entries is selected and read in each bank 11-14. Each entry selected includes VPN (31-12), VPN (11-10), ASID, SH, SZ, V, PPN (31-10), PR, C, and D as information.

  An example of the address comparison procedure executed in the address comparison and V-bit check processing L2 is shown in FIG. The procedure shown here is based on the logic of the control logic 154 shown in FIG. 21, but the comparison of the space number ASID is shown in consideration of the detection of the TLB protection violation exception. How to compare the addresses to be reflected in the hit signals hit1 to hit4 is determined as follows. First, whether or not to consider a space number as an address comparison target is broadly determined depending on whether or not SH is 1, and VPN (11-10) is considered as an address comparison target depending on whether or not SZ is 0. It is determined whether or not to do so. In particular, in the single virtual memory (SV = 1), the data in the field of the space number ASID is used as the memory protection information. As described above, in the privileged mode, the logical page assigned to another process is also the current process. In order to be able to access from, or in other words not to detect a TLB protection violation exception, when in single virtual memory and in privileged mode (SV = 1 and MD = 1) The field of the space number ASID is excluded from the comparison target.

  Based on the index, a TLB entry is read from TLB1, and based on the shared status SH in the read TLB entry, it is determined whether or not to consider the space number ASID in the address comparison. When SH = 1 (shared), the space number ASID is not considered as an address comparison target, and when SH = 0 (non-shared), the space number ASID is considered as an address comparison target. Further, the SV value (MMUCR.SV) of the MMUCR register 55 is set to 1, and a single virtual memory is set, and the mode bit SR. When MD is 1 (when the central processing unit is operating in the privileged mode), the space number ASID is not considered in detecting the TLB protection violation exception. This is because a logical page allocated to another process can be accessed from the current process due to the nature of the privileged mode.

  In the present embodiment, the size of 1 KB or 4 KB is selected as the size of the logical page according to the value of the size bit SZ in the TLB entry. When the size of the logical page is 1 KB, the information VPN (31-17) and VPN (11-10) in each TLB entry indexed in each bank are the corresponding bits vpn (31-17) of the logical address. , Vpn (11-10). When the logical page size is 4 KB, the VPN (11-10) and the corresponding logical address vpn (11-10) in the information VPN in each TLB entry indexed in each bank are: It is excluded from the target of comparison judgment.

  As described above, comparison targets (VPN (31-17, 11-10), vpn (31-17, 11-10), ASID, asid) are determined, and comparison is performed. As a result of this comparison, if there is a match in any bank, a hit signal is output from the matched bank. The hits in each bank are output as hit signals hit1 to hit4 and are TLB hits. On the other hand, when no hit signal is output from any bank, that is, when there is a mismatch, a TLB miss exception is detected. The result is also notified to the CPU 3 by a signal 501, and the TLB miss exception processing described later is performed by the CPU 3.

  In addition, for each indexed entry in the bank, the V bit in the entry is also checked. That is, a check is performed on the V bit in the read entry by the index. In the case of a TLB hit, if the V bit in the entry related to the hit is 0 (invalid), a TLB invalid exception is detected and notified to the CPU 3. The contents of this exception processing will be described later. The determination result of the V bit in the case of the TLB miss is used in the entry replacement described later related to the TLB miss.

  In the access right check processing L3, the contents of the information PR in the TLB entry read by the index and the bits MMUCR. The access right is checked according to the contents of the SV. For example, as shown in FIG. 13, first, the bits MMUCR. It is determined whether the SV is 1 (single virtual memory) or 0 (multiple virtual memory). In the case of multiple virtual storage (SV = 0), the address space represented by the information in the TLB entry is protected according to the contents of the PR shown in FIG. When in single virtual memory and in privileged mode, ie MMUCR. When SV = 1 (single virtual memory), bits SR. When MD = 1 (privileged mode), the address space represented by information in the TLB entry can be accessed unconditionally. On the other hand, even in the case of a single virtual memory, in the case of the user mode (SR.MD = 0: user mode), the space number ASID and the shared status SH are considered when checking the access right. That is, when the space number asid of the PTEH register 51 at the time of access matches the space number ASID read from TLB1, or when SH = 1 (shared), it is represented by the information in the TLB entry according to the PR. Address space protection is performed. In contrast, if the process numbers do not match and are not shared, a TLB protection violation exception is detected. The contents of the exception will be described later. Further, in the access right check process L3, an access type determination is made as to whether the access is for reading or writing, and the D bit of the entry read from the TLB 1 is determined. If the access is the first write (for example, the first write access after power-on or reset), a TLB initial page write exception is detected. That is, the TLB initial page write exception is a condition in which the comparison result between the logical address and the indexed TLB entry is a TLB hit, the dirty bit D in the TLB entry is 0, and the access at that time is a write access. Detected by. The contents of this TLB initial page write exception process will be described later.

  In the physical address generation process L4, a physical address is generated as shown in FIG. 14 in accordance with the size bit SZ of the indexed TLB entry. The offset va (9-0) of the logical address va used to form the physical address pa is not shown in the figure, but is held in a latch circuit in the control circuit (CTRL) 50. When SZ = 0 (logical page size is 1 KB), the offset va (9-0) of the logical address va output from the CPU 3 is set as the offset pa (9-0) of the physical address pa. That is, all bits PPN (31-10) of the physical page number PPN included in the data part of the hit TLB entry are set as the physical page address pa (31-10), and the address (9-0) is subordinate to this as an offset. The physical address pa is generated. When SZ = 1 (logical page size is 4 KB), the offset va (11-0) of the logical address va output from the CPU 3 is set as the offset pa (11-0) of the physical address pa. Of the physical page number PPN included in the data part in the hit TLB entry, PPN (31-12) in which the lower 2 bits are ignored is set as the physical page address pa (31-12), and an address (11 as an offset) -0) is added to the lower side, and the physical address pa is generated.

  FIG. 15 shows a hardware designation method of the replacement target bank in the TLB miss exception detected in the address comparison and V-bit check processing L2. This control is uniquely performed by the control circuit 50 according to the logical configuration, and the bits MMUCR. RC is used as a counting means such as a random counter. Here, MMUCR. The power of 2 of the number of bits of RC is matched with the number of banks of TLB1 (= 4). In response to the fact that none of the hit signals hit1 to hit4 indicates a hit state, the control circuit 50 determines that it is necessary to replace the entry for TLB1 due to a TLB miss. If it judges in this way, the control circuit 50 will carry out MMUCR. RC is incremented by 1 (+1), and it is checked whether or not there is an invalid entry in the indexed four entries. This is accomplished by examining the V bit in each of the four indexed entries. As a result of examining the V bit, if there is no invalid entry (all the entries in each indexed bank hold valid data “V = 1”), the incremented result is replaced. Power bank number and MMUCR. No operation for RC. When an invalid entry exists (any entry indexed in each bank indicates V = 0 and does not hold valid data), the bank number of the bank having the invalid entry is set to MMUCR. Set to RC, and the bank with the set number is the bank to be replaced. In addition, the above MMUCR. The value of each bit of RC can be arbitrarily changed by execution of software by the CPU 3. Therefore, as described above, the bank number set in this register can be further changed by software. Therefore, any bank can be a target for replacement.

  FIG. 16 shows a processing procedure by the TLB miss handler for updating the TLB entry in order to deal with the TLB miss exception. The load TLB instruction is used to update the TLB1 entry. The load TLB instruction (also referred to as LDTLB) is executed by the CPU 3, whereby the following processing is performed. That is, processing for writing the values of the registers 51 and 52 of PTEH and PTEL into the entry of TLB1 is performed. In this case, the entry to be written is not particularly limited, but MMUCR. This is an entry in the bank indicated by the bank number set in RC, and is indicated by using the logical address (bit 12 to bit 16) held in the PTEH register 51 as an index address. When the TLB miss exception is detected, a part of the logical address (bit 10 to bit 31) output from the CPU 3 is held in the PTEH register 51. Thereby, the entry at the time of replacement is instructed by the index address having the same value as the index address when the TLB miss exception occurs. However, the bank used for replacement is MMUCR. It is determined by the bank number set in RC.

  In order to cope with the TLB miss exception, a page table is formed in advance in the external memory RAM as shown in FIG. In the page table, although not particularly limited, a plurality of pieces of conversion information (page table entries) respectively corresponding to a plurality of logical addresses are stored according to a predetermined rule. The page table address, for example, its start address is stored in advance in the register TTB 53 as a base address. The page table is not particularly limited, but based on the base address and logical address as the start address, a page table entry corresponding to the logical address (corresponding physical page number ppn, valid bit v, protection bit pr, A plurality of page table entries respectively corresponding to a plurality of logical addresses are arranged in accordance with a rule such that a size bit sz, a cacheable bit c, a dirty bit d, and a status sh can be searched.

  The TLB miss handler is described by the user. When a TLB miss exception is detected, this TLB miss handler is activated by the CPU 3. As a result, the logical address information vpn (10-31) when the TLB miss occurs is stored in the PTEH register 51. At this time, the PTEH register 51 may also store the space number asid when the TLB miss occurs. Further, the CPU 3 searches the base table in the external memory using the base address stored in the register TTB 53 and the logical address when the TLB miss occurs. When a page table entry corresponding to the logical address at the time of occurrence of the TLB miss is found by this search, the contents of the found page table entry are loaded into the PTEL register 52. Next, a load TLB instruction is issued, and the entry of TLB 1 is updated with the values of the registers 51 and 52 of PTEH and PTEL. Accordingly, the logical address information vpn and asid held in the PTEH register 51 at the time of occurrence of the TLB miss is adopted as a part of VPN and ASID of the TLB entry. At this time, the entry to be replaced is MMUCR. This is an entry in the bank indicated by the bank number set in RC and indicated by the same index address as the index address when a TLB miss occurs.

  The TLB invalid exception occurs in the case of a page fault in a TLB hit. For this exception, for example, first, the page table entry in the external memory is recovered, and the V bit in the page table entry is set to the logical value 1. Thereafter, the page table entry is loaded from the external memory into the PTEL register 52, the above-described load TLB instruction is issued, and the corresponding entry of TLB1 is updated with the values of the registers 51 and 52 of PTEH and PTEL.

  When the TLB initial page write exception is detected, the D bit of the corresponding page table entry on the external memory is set to the logical value 1, and the page table entry is loaded from the external memory to the PTEL register 52. The above-described load TLB instruction is issued, and the corresponding entry of TLB 1 is updated with the values of the registers 51 and 52 of PTEH and PTEL. Note that the logical address information vpn and asid that are the exception factors are held in the PTEH register 51. The situation to be set to D = 1 occurs when the physical page area on the main memory is first written. In the virtual storage, in order to match data between the auxiliary storage device and the main memory (for example, the external memory in FIG. 22) at the time of page replacement, the contents of the page to be replaced in the main memory are copied back to the auxiliary storage device. Whether or not is required. The dirty bit D is used for this determination.

  When the TLB protection violation exception is detected, in order to resolve the protection violation after the logical page number vpn of the logical address as the exception cause is written in the PTEH register 51 and the logical address is written in the TEA register 54. The handler is started.

  As mentioned above, MMUCR. When RC is used as a counter, it is desirable to increment as described above so that an entry can be registered in a new bank when the TLB miss exception is taken. On the other hand, in the countermeasure against the TLB invalid exception, the TLB initial page write exception, and the TLB protection violation exception, the MMUCR. It is desirable not to increment RC. In these measures, there is a case where only the D bit or V bit needs to be changed, so that it is possible to use the TLB more effectively by registering in the original bank without registering in the new bank. . Of course, the present invention is not limited to this.

  17 and 18 show the bits MMUCR. The entire exception detection flow related to TLB1 when SV is set to 0 and multiple virtual storage is instructed is shown. According to FIG. 17, the index of TLB1 is performed by a predetermined method according to the logical address vpn and the current space number asid (S1). If the indexed entry includes SH = 0 (non-shared), if the ASID or VPN does not match (S3), a TLB miss exception (EX1) is detected. If the indexed entry contains SH = 1 (shared), the ASID is not compared and if the VPNs do not match (S4), a TLB miss exception (EX1) is detected. In the case of a TLB hit (YES in S3 and S4), it is determined whether or not V = 1 (S5). If V = 0 (invalid), a TLB invalid exception (EX2) is detected. When V = 1 (valid), as shown in FIG. It is determined from the MD whether the user mode (User) or the privileged mode (Privileged) (S6). Information indicating that the CPU 3 is operating in the user mode and that the entry read by the access by the CPU 3 operating in this mode permits access in the privileged mode (access prohibition in the user mode). If it has PR (PR = 00 or 01), a TLB protection violation exception (EX3) is detected. When it is determined that the information PR in the read entry is 10 by access in the user mode, it is further determined whether this access is a read access type or a write access type. As shown in FIG. 9, when the information PR is 10, only the read access type is allowed for user access. For this reason, when the access is the write access type, the access type is different (write in S7), and therefore the TLB protection violation exception (EX3) is detected.

  If PR = 00 or 10 is also determined in the access in the privileged mode, a TLB protection violation exception (EX4) is detected if the read / write access type is contrary to the PR contents (write in S8). Is done. That is, the access right permitted by the PR differs depending on whether the CPU 3 is operating in the user mode or the privileged mode, but in any case, access other than that permitted by the PR is allowed. When accessing by type, a TLB protection violation exception (EX3, 4) is detected. When the access type is a write permitted by PR (write of S9 and S10), if the information D in the entry is 0 (unwritten page), a TLB initial write exception (EX5) is detected. When the information D in the entry is 1, if the information C in the entry is 1, the cache memory 4 is accessed, and if C = 0, the main memory (for example, the external memory RAM or ROM in FIG. 22). Will be accessed. When the access type is a read permitted by PR (read of S7 to S10), if C = 1, the cache memory 4 is accessed, and if C = 0, the main memory is accessed. .

  FIG. 22 mainly shows the connection relationship between each register shown in FIG. 7 and the central processing unit CPU. Each register is assigned a unique address. The logical address formed by the central processing unit CPU3 is supplied to the selection circuit in the control circuit (TLBC) 50 via the internal logical address bus VABUS. The selection circuit decodes the logical address, and when the logical address is a unique address assigned to the register, generates a selection signal for selecting the register. For example, when the logical address is an address assigned to the PTEL register 52, the selection circuit forms the selection signal C4 and selects the register. Similarly, other registers (PTEH, MMUCR, TEA, TTB) are also selected. In other words, these registers are address mapped. The read / write from the central processing unit to the selected register is instructed by a read / write control signal supplied from the central processing unit to each register via an internal control bus (not shown). Of course, as shown in this drawing, each register and the central processing unit CPU3 are connected to each other via an internal data bus DBUS. Each register is also connected to the control circuit 50 and TLB 1 shown in FIG. 7, but is not shown in FIG. 22 to avoid complicating the drawing. The central processing unit can write data to the register by executing software. That is, by executing the software, the central processing unit outputs the logical address assigned to the register to the bus VABUS, outputs the data to the bus DBUS, and instructs the write by the read / write control signal. Data can be written. Similarly, the central processing unit 3 can read data from the register by executing the software. As described above, by executing the software, the logical space number and the logical address at the time of TLB miss can be written to the PTEH register 51, and the table entry at the time of replacement is written to the PTEL register 52. The base address can be written to the TTB register 53, and the logical address can be written to the TEA register 54 when a protection violation exception occurs. Various control data shown in FIG. 6 can be written into the MMUCR register 55, and specific bits can be used like a counter.

  The internal logical address bus VABUS is also connected to the control circuit (CTRL) 50 and TLB1. The control circuit (CTRL) 50 is provided with a latch circuit for holding the offset of the logical address as described above, and holds the offset of the logical address from the internal logical address bus VABUS. Further, a logical address is supplied from the CPU 3 to the TLB 1 via the internal logical address bus VABUS, and is used as an index address and a search address. Of course, the latch circuit provided in the control circuit (CTRL) 50 may hold not only the offset address but all the logical addresses.

  In the drawing, only the connection between the PTEH register 51 and the TLB 1 is explicitly shown. The space number for determining the TLB1 miss / hit is set in this register and supplied to the TLB1 from this register. Also, as a countermeasure against the TLB1 miss exception, a logical address or the like is supplied from the register to the TLB1 as described above.

  The internal data bus DBUS and the internal physical address bus PABUS are connected to the external data bus DBUS and the external address bus ABUS via external terminals TD and TA provided in the data processing device. For example, as shown in the figure, external memory RAM and ROM are connected to these external buses. Although not particularly limited, the external memory RAM is a volatile memory, and various tables as described above are formed. The external memory ROM is a non-volatile memory, and stores various programs (for example, software such as the above-described handler).

  In the following, the effect of virtual memory in the microcomputer of this embodiment will be described for each feature point.

  << Support of Multiple Page Sizes >> As shown in FIGS. 5 and 6, the page table entry and the TLB entry have the size bit SZ, and the size of the logical page can be set for each page. The 4-way set associative address translation buffer 1 is commonly used for a plurality of logical pages whose size can be set variably. In this embodiment, there are two types of logical page sizes supported by the microcomputer: 1 KB and 4 KB. The index address designation method for TLB 1 is common to both 4 KB and 1 KB. In this embodiment, as shown in FIG. 2 and FIG. 3, the total number of bits formed by the central processing unit 3 is 32 bits. Of these logical addresses, bits 12 to 16, that is, vpn (16-12) are used for the index of TLB 1. Since the index address is 5 bits, a maximum of 32 entries can be held per bank (way). Since the TLB 1 has four banks 11 to 14, a maximum of four entries can be held for one index address. When the page size is 4 KB, the index address is 5 bits vpn (16-12) from the lowest order of the logical page number vpn. Therefore, when the page size is 4 KB, an entry of an arbitrary logical page number is set. Each bank can have a maximum of 32 entries (128 entries in total). On the other hand, when the page size is 1 KB, the least significant 2 bits vpn (11-10) of the logical page number vpn are not used for the index. Therefore, considering an index in the case where the logical page size is 1 KB and the number of banks is one, the indexed entry has four logical page numbers selected by a 5-bit index address (each is a page of 1 KB). Size). Two bits (bits 10 and 11) not used for the index are used to indicate any one of the four selected logical page numbers. Therefore, when there is only one bank, only one entry is assigned to four consecutive logical pages. In this embodiment, the maximum size of the logical page is set to 2 N times the minimum size, and the number of banks is also set to 2 N powers. That is, the minimum size of the logical page size is 1 KB, the maximum size is 4 KB, which is a squared multiple of 2, and the number of banks is a number that is the square of 2 (four). By doing so, almost as in the case of the 4 KB page size, the address translation buffer 1 can hold 128 entries of arbitrary logical page numbers having a 1 KB page size as a whole. However, the logical page number of an entry that can be held in one bank is restricted by every 4 KB. This restriction has no effect on the hit rate of the TLB 1 if the logical page numbers of entries held in the address translation buffer 1 are continuous. In the case of decentralization, the hit rate is affected to some extent. Even in this case, if 1KB logical page is address-mapped for every 2KB, the influence can be reduced, and if address mapping is made for every 4KB, it can be made not affected at all. For example, when 1 KB page size conversion information is set (written) to the address conversion buffer 1, four entries indicated by a 5-bit index address from each of the four banks 11, 12, 13, and 14 Conversion information when the 2-bit vpn (11-10) is “00”, conversion information when the 2-bit vpn (11-10) is “01”, and 2-bit vpn (11-10) is “10” Conversion information when the 2-bit vpn (11-10) is “11”, respectively. In this way, it is possible to prevent a decrease in hit rate.

  The number of bits of address comparison for hit determination must be changed according to the logical page size as described with reference to FIG. When the logical page size is 1 KB, vpn (11-10) must be compared with VPN (11-10) compared to the case of 4 KB. The TLB 1 has storage areas for VPN (31-17) and VPN (11-10) so as to cover all the numbers of bits to be compared, and a 22-bit storage area for the physical page number PPN. It has. The TLB 1 has a size bit SZ indicating the logical page size supported by the data portion of each entry, and the number of bits for address comparison for hit determination is changed according to the value. As described in FIG. 12, when the size bit SZ = 1 (logical page size = 4 KB), the VPN (31-17) is a comparison target with the corresponding bit of the logical address, and the size bit SZ = 0 ( In the case of logical page size = 1 KB), VPN (31-17) and VPN (11-10) are to be compared with the corresponding bits of the logical address.

  In this way, a macro computer that selectively supports multiple page sizes can reduce the size of logical pages and reduce the memory usage efficiency of each process when the total storage capacity of the real memory installed in the system is small. It is possible to easily meet the demand to improve the quality. At this time, the maximum logical page size is set to a power of 2 times the minimum size, and the number of banks of the TLB1 of the set associative method is set to a power of 2 or more, whereby the index address designation method is set to the maximum logical page Regardless of whether the size is unified or the selected logical page size is the maximum or minimum, in principle, any logical page number entry can be held in the address translation buffer 1. Can do. Corresponding to the logical page number, information for indicating the size of the logical page is provided, and by using this size, the address comparison buffer 1 is associatively searched by changing the number of bits for address comparison for hit determination. Therefore, it is possible to easily change the number of bits and the bit position of the information to be compared according to the size of the logical page. Realizing the address translation buffer 1 that supports a plurality of page sizes with a set-associative cache memory can halve both the chip-occupied area and the power consumption as compared with the case where it is configured with a CAM.

  << Supporting multiple index methods >> When there are multiple processes, each process has its own address translation information, and supports multiple virtual storage where each process is distinguished by the process number asid, As a method of designating the index address of TLB1, a method of decoding only a part of the logical address (index address) as shown in FIG. 2 and its logical address by exclusive OR gate XOR as shown in FIG. And a method of decoding a result obtained by modifying a part (index address) of the current process number asid with a bit MMUCR. It can be indicated according to the logical value of IX. According to this, it is possible to select an index method according to the usage mode so that a decrease in hit rate can be suppressed in a usage mode in which many processes are started in parallel in multiple virtual memory. . In addition, a part of the logical address is modified by the number asid of the process that uses the logical address, and the buffer memory is indexed by this. The decrease in rate can be suppressed.

  << Freedom of Replacement >> As described above, the TLB 1 is configured as a 4-way set associative cache memory having a plurality of banks 11 to 14 each having a common index address. When the stored information is to be replaced from among the plurality of banks (way) due to a cache miss or the like, the bank to be replaced can be arbitrarily designated by execution of software by the central processing unit 3. Of the register MMUCR shown in FIG. RC is an area in which 2-bit information for arbitrarily designating the bank is set. A value (BSL1 to BSL4) for selecting one of the four banks 11 to 14 is formed by decoding the set value by the decoder 17 of FIG. As a result, one of the entries in the four banks 11 to 14 specified by the index address 2 is selected by the selection signal (BSL1 to BSL4) and is set as a replacement target. Bit MMUCR. RC can also be used as a counting means such as a random counter. MMUCR. The power of 2 of the number of bits of RC matches the number of banks (= 4) of the buffer memory 1. As described with reference to FIG. 15, when the control information (CTRL) 50 needs to replace stored information for TLB 1 (TLB miss), the MMUCR. RC is incremented by 1 and if any bank holds valid data in the indexed storage area (conversion information V in each indexed entry has 1), the incremented result is replaced. If there is a bank that does not hold valid data in the indexed storage area (any indexed entry has V = 0), the bank number is set to MMUCR. A bank to be replaced is designated in accordance with a certain rule that is set to RC and the bank of the set number is a bank to be replaced. At this time, MMUCR. Each bit of RC is a target whose value can be arbitrarily changed by execution of software by the central processing unit 3. That is, MMUCR. RC is not limited to the operation incremented by the occurrence of a TLB miss. To exclude specific values, MMUCR. The RC may be updated. Furthermore, various substitutions are possible depending on the software algorithm executed by the CPU 3. For example, replacement such as random, first-loaded FIFO to be replaced, or last-replaced LRU (Least Recentry Used) is replaced with MMUCR. This can be easily realized by changing the RC update method. Therefore, the degree of freedom of replacement can be guaranteed.

  In this way, the replacement algorithm for replacing the entry of TLB 1 is not fixed, and the bank to be replaced can be arbitrarily determined by software, so that a specific conversion pair is always stored in the address conversion buffer 1 for the convenience of data processing. It becomes possible to easily respond to a request to store as an entry or a request not to make a specific entry a replacement target.

  << Support for Single Virtual Memory and Multiple Virtual Memory >> In the microcomputer of this embodiment, when each of a plurality of processes to be executed has address translation information over the entire logical space, the logical number is indicated by the process number asid. Multiple virtual memories that modify or expand addresses, and a part of the logical address space is exclusively allocated to a plurality of processes to be executed, and address conversion information of the logical address space to which each process is allocated If so, it is possible to select a single virtual memory that does not qualify or extend the logical address by the process number asid. Such control for virtual memory is controlled by the MMUCR.exe illustrated in FIG. Indicated by the value of SV. MMUCR. The value of SV is arbitrarily set by the central processing unit 3 executing software. The conceptual difference between the single virtual memory and the multiple virtual memory is as described with reference to FIG. 20, and other typical differences are illustrated in FIG. As shown in FIG. 5, the TLB 1 includes a process number ASID field together with a logical page number VPN and a physical page PPN number. The processing content of the value of this field differs depending on whether it is a single virtual memory or a multiple virtual memory. When a logical page corresponding to a process number included in a certain TLB entry is set to be non-shareable with other processes, the process number ASID in the multiple virtual memory is as described with reference to FIG. Used to determine TLB1 TLB hit / miss. Therefore, if the logical page number VPN of the entry stored in TLB1 matches the logical page address vpn and the process number ASID of the entry does not match the current process number asid, a TLB hit is not made. In a single virtual memory, the process number ASID is used as memory protection information (domain number). In the user mode, access to a non-shared page by another process is processed by software as a TLB protection violation exception.

  MMUCR. By setting the SV value via the CPU 3, it is possible to selectively support both single virtual memory and multiple virtual memory, thereby improving the usability of the address translation mechanism. By using the process number ASID in the multiple virtual memory as the memory protection information in the single virtual memory, complete memory protection in that case can be easily realized.

  Although the invention made by the present inventor has been specifically described based on the embodiments, it is needless to say that the present invention is not limited thereto and can be variously modified without departing from the gist thereof.

  For example, in the above embodiment, the number of TLB ways, that is, the number of banks can be set to 5 or more, for example, 8. For example, when the page size is 1 KB and 8 KB, the number of LTB ways (the number of banks) may be 8 or more. When the page size is 4 KB and 16 KB, the number of LTB ways (the number of banks) may be four or more. The effect similar to the said Example can be acquired also by them. In short, the maximum size of the supported page is set to a power of 2 times the minimum size, and the number of sets of the buffer memory of the set associative method may be equal to or greater than the power of 2. The size of the address space supported by the microcomputer is not limited to 4 GB, and the number of bits of the logical address is not limited accordingly. The supported page size can be changed to an appropriate size, and the number of supported types can be changed as appropriate. Further, the logical page number VPN held as the TLB entry is not limited to the configuration excluding the corresponding bits used for the index as in the above embodiment, and all bits of the logical page number are held as the TLB entry. You can also.

  In the above description, the case where the invention made mainly by the present inventor is used for the microcomputer which is the field of use behind the present invention has been described. However, the present invention is not limited to this. It can be widely applied to controller chips and the like.

It is explanatory drawing of the structure which supports multiple page size in the microcomputer based on one Example of this invention. It is explanatory drawing of the method of indexing TLB using a part of logical page address as it is. It is explanatory drawing of the method of indexing TLB using a part of logical page address and a part of process number. It is explanatory drawing of the structure which can determine arbitrarily the replacement object bank of a TLB entry by software. It is explanatory drawing for demonstrating an example of the format of the logical address from which page sizes differ, and the TLB entry for supporting them. It is explanatory drawing for demonstrating an example of the various registers | resistors for TLB. It is a whole block diagram which shows the principal part of the microcomputer based on one Example of this invention. It is explanatory drawing which shows an example of the address map which the microcomputer of a present Example supports in (A) and (B). It is explanatory drawing of the access right prescribed | regulated by the protection bit PR utilized for storage protection. It is a control main flowchart of address conversion. It is a control flowchart for the index method selection of TLB. It is a control flowchart of an address comparison between a logical address and a tag indexed thereby. It is a partial flowchart of protection control including control which uses ASID for memory protection in a single virtual memory. It is a flowchart which shows the physical address production | generation procedure according to page size. It is a control flowchart for designating the replacement target bank in hardware. It is a flowchart which shows an example of the process by the TLB miss handler for updating the TLB entry started by the exception which arises by TLB miss. It is a flowchart which shows the first half of the exception detection process regarding TLB in a single virtual memory. It is a flowchart which shows the second half of the exception detection process regarding TLB in a single virtual memory. It is explanatory drawing which shows the whole difference of single virtual memory and multiple virtual memory. It is explanatory drawing which shows the concept of single virtual memory and multiple virtual memory in (A) and (B). It is a logic circuit diagram showing an example of control logic for controlling an address comparison result to be reflected in a hit signal in each bank of TLB. FIG. 8 is a block diagram showing a part of FIG. 7 in more detail.

Explanation of symbols

1 Address translation buffer (buffer memory)
VPN logical page number PPP physical page number ASID, asid space number (process number)
vpn logical page address (logical page number)
ppn physical page address (physical page number)
PR protection key 11-14 bank 15 comparison means 151-153 comparator 154 control logic 2 index address XOR exclusive OR gate 3 central processing unit 4 cache memory 5 TLB controller 50 control circuit 51 PTEH register 52 PTEL register 53 TTB register 54 TEA register 55 MMUCR register SV Single virtual bit SZ Size bit IX Index mode bit RC Random counter MD Mode bit

Claims (14)

A virtual storage function that includes a central processing unit, divides a logical address space into a plurality of logical pages, receives a logical address from the central processing unit, and converts a logical address allocated on the logical page into a physical address on a physical page A cache memory, a register,
The cache memory has a plurality of entries, and each of the plurality of entries has a logical page number for specifying a logical page and a physical page number for specifying a physical page corresponding to the logical page number in each entry. ,
The register is connected to the central processing unit and a cache memory, and data for identifying one of the plurality of entries as a replacement target is set by the central processing unit. apparatus. A central processing unit, a main storage, a cache memory, and a control register;
The main storage device is connected to the central processing unit and is used for storing data,
The cache memory is connected to the central processing unit and the main storage device, and has a plurality of entries, each entry selectively storing data stored in the main storage device,
The control register is connected to the central processing unit and a cache memory, and stores a value supplied by the central processing unit, and the stored value is to be replaced with data from the main storage device next. A data processing device used for specifying an entry in a cache memory. A central processing unit, a main storage, a cache memory, and a control register;
The main storage device is connected to the central processing unit and is used for storing data,
The cache memory is connected to the central processing unit and the main storage device, and has a plurality of entries, each entry selectively storing data stored in the main storage device,
The control register is connected to the central processing unit and a cache memory, and stores a value set by the central processing unit, and the stored value is to be replaced with data from the main storage device next. A data processing apparatus characterized by being used for replacement control of entries. A central processing unit, a cache memory and a control register;
The cache memory is connected to the central processing unit and has a plurality of entries, each entry selectively storing data supplied from the central processing unit,
The control register is connected to the central processing unit and a cache memory and stores a value supplied from the central processing unit, and the stored value is a cache to be replaced with data supplied from the central processing unit next. A data processing apparatus used for specifying an entry in a memory. A central processing unit, a cache memory and a control register;
The cache memory is connected to the central processing unit and has a plurality of entries, each entry selectively storing data supplied from the central processing unit,
The control register is connected to the central processing unit and a cache memory, stores a value supplied from the central processing unit, and the stored value is a cache to be replaced with data supplied from the central processing unit next. A data processing apparatus used for replacing and controlling a memory entry. A buffer memory, a register and a central processing unit;
The buffer memory has a plurality of entries, and each entry has address information including a logical address tag and physical address information,
The register stores data for specifying one of the plurality of entries,
The central processing unit outputs a logical address,
If the logical address tag stored in the entry selected by the logical address matches the address data included in the logical address output by the central processing unit, the central processing unit stores the physical address information stored in the entry. Use
When the logical address tag included in the selected entry does not match the address data included in the logical address, the address information of the entry specified by the data stored in the register is replaced with new address information. A data processing device. The logical address space is divided into a plurality of virtual pages, and has a virtual storage function in which logical addresses are converted into physical addresses in units of virtual pages, and includes a processing device, a bank specifying circuit, and an address conversion buffer.
The processing device generates a logical address having an index address, a comparison address, and an offset address,
The bank designating circuit can be set by the processing device,
The address conversion buffer is used to convert a logical address generated by the processing device into a physical address, and includes a plurality of banks, a selection circuit, a comparison circuit, and a bank selector.
Each of the plurality of banks has a plurality of entries each having a virtual page number and a physical page number;
The selection circuit selects an entry specified by the index address from each of the plurality of banks,
The comparator compares the virtual page number of the selected entry with a comparison address,
When the entry selected by the comparator indicates that the virtual page number corresponding to the comparison address is not stored, the bank selector replaces the bank designated by the bank designation circuit. Specify for and
A data processing apparatus, wherein the physical address is generated from a physical page number and an offset address stored in the entry. The logical address space is divided into a plurality of virtual pages, and has a virtual memory function in which logical addresses are converted into physical addresses in units of virtual pages, and a set associative cache memory, a control signal output unit, and an index address generation circuit are provided. The set associative cache memory has a plurality of memory banks, each of the plurality of memory banks has a plurality of memory fields, and each memory field has a virtual page number for designating a virtual page and The physical page number corresponding to the virtual page number can be stored,
The index address generation circuit receives at least a part of a logical address and a control signal, and generates an index address for selecting a memory field from each of a plurality of banks. In this case, a different index depends on the control signal. A data processing apparatus characterized by generating an address. The logical address space is divided into a plurality of virtual pages, and has a virtual storage function in which logical addresses are converted into physical addresses in units of virtual pages, a central processing unit, a set associative cache memory, and a designation circuit,
The set associative cache memory is divided into a plurality of banks, and each bank has a plurality of memory fields that can store a virtual page number for designating a virtual page and a physical page number corresponding to the virtual page number. Have
The designating circuit designates a bank that includes a memory field that is accessed by the central processing unit and whose information is to be replaced;
A data processing apparatus characterized in that designation of a bank including information to be replaced is determined by software executed by the central processing unit. The logical address space is divided into a plurality of virtual pages, has a virtual storage function in which logical addresses are converted into physical addresses in units of virtual pages, has a set associative configuration cache memory, a designation circuit, and a central processing unit,
The set associative cache memory is divided into a plurality of banks, each bank having a plurality of memory fields capable of storing a virtual page number indicating a virtual page and a physical page number corresponding to the virtual page number,
The designation circuit designates, from a plurality of banks, a bank including a memory field whose stored information is to be replaced according to a predetermined rule.
The central processing unit is connected to the designating circuit, and a bank designated according to software being executed in the central processing unit is changed. The logical address space is divided into a plurality of virtual pages, and has a virtual memory function in which a logical address is converted into a physical address in units of virtual pages. A plurality of processes are executed using the virtual memory function, and a single virtual Having storage mode and multiple virtual storage mode,
The multiple virtual memory mode is changed according to a process in which a logical address is executed, and each of the plurality of processes can use an address over the entire logical address space.
In the single virtual memory mode, the logical address is not changed according to the process being executed, and each of the plurality of processes can use the logical address of the address space allocated in the logical address space. , A portion of the address space is allocated exclusively among the processes that are executed,
A data processing apparatus having a selector for selecting the multiple storage mode and the single storage mode. The logical address space of the central processing unit is divided into a plurality of virtual pages, and has a virtual storage function in which logical addresses are converted into physical addresses in units of virtual pages, and a plurality of processes are executed using the virtual storage function. , Mode information designating the operation mode of the central processing unit, control information designating the virtual storage mode, an address translation buffer, and a control circuit,
The mode information specifies whether the central processing unit is operating in a privileged mode or a user mode,
The control information specifies either a multiple virtual storage mode or a single virtual storage mode, and in the multiple virtual storage mode, each of the processes can use the address of the entire logical address space. Now, change the logical address according to the process that uses the logical address, each process will be allocated a part of the logical address space exclusively, and each process will be changed after the process The logical address of a part of the allocated address space is made available,
The address translation buffer is accessed by using at least a part of the logical address as an index address, and has a plurality of fields. Each field is assigned to process information for each process and a virtual page for each process. Including virtual page information, physical page information corresponding to the virtual page information, and shared information indicating whether a virtual page used by a process is shared with other processes,
When the control circuit is used for address translation and indicates that the shared information is non-shared, the process information is used for hit judgment in the address translation buffer, and the mode information is stored by the central processing unit. The process information is used as storage protection information when the user mode is indicated, the control information is in a single virtual storage mode, and the shared information is non-shared. A data processing device. The logical address space is divided into a plurality of virtual pages, and has a virtual memory function that converts the logical address into a physical address in units of virtual pages. The virtual memory function enables a plurality of processes to be executed, and outputs to the processing device. A device, a setting device, and an address translation buffer;
The processing device generates a logical address including an index address, a comparison address, and an offset address,
The output device outputs a process number for designating one of the plurality of processes,
The setting circuit is set by the processing device and outputs information for designating an index.
The address conversion buffer is used to convert a logical address generated by the processing device into a physical address, and includes a plurality of banks, a selection circuit, and a comparator.
Each of the plurality of banks has a plurality of entries each including a virtual page number and a physical page number;
The selection circuit selects either the index address of the logical address or the index address modified by the process number based on the index information, and selects an entry determined by the index address selected from each of the plurality of banks. Selected,
The comparator compares the virtual address of the selected entry with the comparison address,
A data processing apparatus, wherein the physical address is generated from the physical page number and offset address of the entry. The logical address space is divided into a plurality of virtual pages, and has a virtual storage function that converts the logical address into a physical address in units of virtual pages. A plurality of processes are executed, and a processing device, a register, and an address conversion buffer are provided. And
The processing device generates a logical address composed of an index address, a comparison address, and an offset address,
The register can set a process number for specifying one process from a plurality of processes by the processing device,
The address conversion buffer is used to convert a logical address generated by the processing device into a physical address, and includes a plurality of entries, a selection circuit, and a comparator.
The plurality of entries are common to each of the virtual page number, the physical page number corresponding to the virtual page number, the process number for specifying the process, and whether the virtual page is shared by a plurality of processes. Have information,
The selection circuit selects an entry specified by the index address,
The comparator compares the virtual page number of the selected entry with the comparison address, and determines whether to compare the process number stored in the register with the process number stored in the selected entry based on the common information. Determined,
The data processing apparatus, wherein the physical address is generated by a physical page number and an offset address of the entry.

Images