1. Technical Field
The present invention relates in general to data processing systems, and more particularly, to an improved multi-processor data processing system. Still more particularly, the present invention relates to improved cache operation within multi-processor data processing systems.
2. Description of the Related Art
A conventional multi-processor data processing system (referred to hereinafter as an MP), typically includes a system memory, input/output (I/O) devices, multiple processing elements that each include a processor and one or more levels of high-speed cache memory, and a system bus coupling the processing elements to each other and to the system memory and I/O devices. The processors all utilize common instruction sets and communication protocols, have similar hardware architectures, and are generally provided with similar memory hierarchies.
Caches are commonly utilized to temporarily store values that might be accessed by a processor in order to speed up processing by reducing access latency as compared to loading needed values from memory. Each cache includes a cache array and a cache directory. An associated cache controller manages the transfer of data and instructions between the processor core or system memory and the cache. Typically, the cache directory also contains a series of bits utilized to track the coherency states of the data in the cache.
With multiple caches within the memory hierarchy, a coherent structure is required for valid execution results in the MP. This coherent structure provides a single view of the contents of the memory to all of the processors and other memory access devices (e.g., I/O devices). A coherent memory hierarchy is maintained through the utilization of a coherency protocol, such as the MESI protocol. In the MESI protocol, an indication of a coherency state is stored in association with each coherency granule (e.g., a cache line or sector) of one or more levels of cache memories. Each coherency granule can have one of the four MESI states, which is indicated by bits in the cache directory.
The MESI protocol allows a cache line of data to be tagged with one of four states: “M” (modified), “E” (exclusive), “S” (shared), or “I” (invalid). The Modified state indicates that a coherency granule is valid only in the cache storing the modified coherency granule and that the value of the modified coherency granule has not been written to system memory. When a coherency granule is indicated as Exclusive, then, of all caches at that level of the memory hierarchy, only that cache holds the coherency. The data in the Exclusive state is consistent with system memory, however. If a coherency granule is marked as Shared in a cache directory, the coherency granule is resident in the associated cache and in at least one other cache at the same level of the memory hierarchy, and all of the copies of the coherency granule are consistent with system memory. Finally, the Invalid state indicates that the data and address tag associated with a coherency granule are both invalid.
The state to which each coherency granule (e.g., cache line or sector) is set is dependent upon both a previous state of the data within the cache line and the type of memory access request received from a requesting device (e.g., the processor). Accordingly, maintaining memory coherency in the MP requires that the processors communicate messages across the system bus indicating their intention to read or write to memory locations. For example, when a processor desires to write data to a memory location, the processor must first inform all other processing elements of its intention to write data to the memory location and receive permission from all other processing elements to carry out the write operation. The permission messages received by the requesting processor indicate that all other cached copies of the contents of the memory location have been invalidated, thereby guaranteeing that the other processors will not access their stale local data.
In some MP systems, the cache hierarchy includes at least two levels. The level one (L1), or upper-level cache is usually a private cache associated with a particular processor core in an MP system. The processor core first looks for a data in the upper-level cache. If the requested data is not found in the upper-level cache, the processor core then access lower-level caches (e.g., level two (L2) or level three (L3) caches) for the requested data. The lowest level cache (e.g., L3) is often shared among several processor cores.
Typically, when a congruence class of one of the upper-level caches becomes full, data lines are “evicted” or written to the lower-level cache for storage. However, in any memory hierarchy, there may be several copies of the same data residing in the memory hierarchy at the same time. The policy of evicting lines to provide for more space in the upper-level cache may result in unnecessary writes to lower-level caches, which results in increased bandwidth demands.
Therefore, there is a need for a more intelligent system and method for managing a multi-level memory hierarchy to reduce unnecessary inter-cache communication.
A system and method for cache management in a data processing system are disclosed. The data processing system includes a processor and a memory hierarchy. The memory hierarchy includes at least an upper memory cache, at least a lower memory cache, and a write-back data structure. In response to replacing data from the upper memory cache, the upper memory cache examines the write-back data structure to determine whether or not the data is present in the lower memory cache. If the data is present in the lower memory cache, the data is replaced in the upper memory cache without casting out the data to the lower memory cache.
The above-mentioned features, as well as additional objectives, features, and advantages of the present invention will become apparent in the following detailed written description.
The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself, however, as well as a preferred mode of use, further objects and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings, wherein:
With reference now to the figures, and in particular, with reference to
Those skilled in the art will appreciate that multi-processor data processing system 100 can include many additional components not specifically illustrated in
Referring now to
With reference now to
However, if L2 cache controller 212 determines that there is a request for data missed in L2 cache 204, the process moves to step 306, which illustrates L2 cache controller 212 determining whether or not write-back history table information is utilized in replacement policy. The replacement policy may be varied by code running in multi-processor data processing system 100, a hardware switch that is physically toggled by a user, or another method. If L2 cache controller 212 determines that write-back history table information is utilized in the replacement policy, the process moves to step 308, which depicts processor core 200 determining whether or not write-back history table 206 includes an entry that indicates that there is a cache line in L2 cache 204 that is also included in L3 cache 108. If processor core 200 determines that there is a cache line in L2 cache 204 that is also included in L3 cache 108 by accessing write-back history table 206, the process moves to step 312, which illustrates a replacement of the cache line in L2 cache 204 that is determined to be also included in L3 cache 108. The process then returns to step 302, and proceeds in an iterative fashion.
Returning to step 308, if L2 cache controller 212 determines that there is not a cache line in congruence class L2 cache 204 that is also included in L3 cache 108 by accessing write-back history table 206, the process moves to step 310, which depicts L2 cache controller 212 utilizing a least-recently used (LRU) algorithm or another replacement algorithm. Returning to step 306, if L2 cache controller 212 determines that write-back history table information is not utilized in the replacement policy, L2 cache controller 212 also selects a victim cache line, as depicted in step 310, utilizing a default replacement policy.
After step 310, the process then continues to step 314, which illustrates L2 cache controller 212 placing the selected victim line into write-back queue 205. The process then proceeds to step 316, which depicts L2 cache controller 212 determining whether or not the selected victim line is dirty. Many data processing systems, including exemplary multi-processor data processing system 100, preferably utilize a coherency protocol, like the MESI protocol. For example, a dirty cache line is considered a Modified granule that is valid only in the cache storing the modified coherency and that the value of the modified coherency granule has not been written to memory 104 or any other type of data storage device (e.g., CD-ROM, hard disk drive, floppy diskette drive or others). A clean line may be marked with an Exclusive or Shared tag. If L2 cache controller 212 determines the selected victim line is dirty, the process continues to step 320, which illustrates L2 cache 204 issuing a write-back request to L3 cache 108. However, if L2 cache controller 212 determines the selected victim line is not dirty, the process continues to step 318, which depicts L2 cache controller 212 examining the contents of write-back history table 206 for an entry that indicates the selected victim line is also present in L3 cache 108. If L2 cache controller 212 determines the selected victim line is also present in L3 cache 108, the process then proceeds to step 312, which illustrates L2 cache controller 212 replacing the selected victim line without first writing the selected victim line to L3 cache 108 or memory 104. The process then returns to step 302 and continues in an iterative fashion.
Returning to step 318, if processor core 200 determines by reference to write-back history table 206 that the selected victim is not present in L3 cache 108, the process continues to step 320. The process continues to step 322, which depicts L3 cache controller 110 snooping a write-back request issued from L2 cache 204. The process then continues to step 324, which illustrates L3 cache controller 110 determining whether or not the selected victim line had been modified while present in L2 cache 204. If L3 cache 108 determines that the selected victim line had been modified while present in L2 cache 204, the process continues to step 326, which depicts L3 cache 108 accepting the selected victim line from L2 cache 204. If the selected victim line was modified while present in L2 cache 204, L3 cache 108 must accept and cache the selected victim line in order to preserve the changes made to the selected victim line while present in L2 cache 204. The process then continues to step 327, which illustrates L3 cache 108 performing a castout of a data line from L3 cache 108 according an algorithm such as a least recently used (LRU) algorithm. The process then returns to step 302 and proceeds in an iterative fashion.
The purpose of the present invention is to reduce unnecessary inter-cache operations. When a selected victim line has not been modified while present in L2 cache 204, L2 cache 204 first examines the contents of a write-back history table 206 to determine if the selected victim line is already present in L3 cache 108.
It is possible that when processor core 200 initially requests and processes a cache line, the line is written to only L1 cache 202 and L2 cache 204. The line is not written to look aside L3 cache 108 until L2 cache 204 casts out a line to make room for a new data line requested by processor core 200. Therefore, each time a clean (not modified) line is selected to be cast out of L2 cache 202, L2 cache 202 must examine write-back history table 206 to determine whether or not a copy of the selected victim line is already present in L3 cache 108. If an entry in write-back history table 206 indicates that a copy of the selected victim line is already present in L3 cache 108, which serves as a castout or victim cache, casting out the selected victim line to L3 cache 108 would be an unnecessary function. The present invention inhibits these unnecessary functions.
Therefore, returning to step 324, if L3 cache 108 determines that the selected victim line has not been modified while present in L2 cache 204, the process continues to step 328, which illustrates L3 cache 108 determining whether or not the selected victim line is already present in L3 cache 108. If L3 cache 108 determines that the selected victim line is not already present in L3 cache 108, the process continues to step 326, which depicts L3 cache 108 accepting the selected victim line received from L2 cache 204. This process is not necessary to preserve a copy of the selected victim line since the data in the selected victim line in L2 cache 204 also resides in memory 104, but is advantageous for latency purposes. Then, if L3 cache 204 is filled with castout lines from L2 cache 204, a data line from L3 cache 204 must be cast out, preferably to memory 104, to accommodate the selected victim line from L2 cache 204. Therefore, the process then continues to step 327, which illustrates L3 cache 108 casting out a data line to accommodate the selected victim line from L2 cache 204, if necessary. The process then returns to step 302 and continues in an iterative fashion.
Returning to step 328, if L3 cache 108 determines that the selected victim line is valid in L3 cache 108, the process continues to step 330, which illustrates L3 cache 108 setting an inhibit bit in the response to the write-back request from L2 cache 204. Setting the inhibit bit indicates to L2 cache 204 that the selected victim line is already present in the L3 cache 108 and that L2 cache 204 may replace the selected victim line with newly-requested data without casting out the selected victim line to L3 cache 108. Persons with ordinary skill in this art will appreciate that an actual inhibit bit is not required to be set by L3 cache 108. L3 cache 108 may achieve the same result by sending an alternate bus response that indicates the validity of the selected victim line in L3 cache 108.
The process then continues to step 332, which depicts L3 cache 108 sending the response with the inhibit bit set out on intra-chip interconnect 208. The process proceeds to step 334, which illustrates L2 cache 204 receiving the response from L3 cache 108. The process then continues to step 336, which depicts L2 cache 204 determining whether or not the response includes an inhibit bit set by L3 cache 108. If L2 cache 204 determines that the response includes an inhibit bit set by L3 cache 108. The process then moves to step 338, which illustrates L2 cache 204 allocating an entry in write-back history table 206 indicating that the selected victim line is already present and valid in L3 cache 108. If L2 cache 204 determines that write-back history table 206 is full, least recently accessed entries in write-back history table 206 are merely overwritten when L2 cache 204 allocates a new entry in step 338. On the next cycle of the process, if the selected victim line is considered as a candidate for cast out from L2 cache 204, L2 cache 204 will determine that the line is valid and present in L3 cache 108 by locating the entry in write-back history table 206 and will evict the selected victim line without attempting to write-back the selected victim line to L3 cache 108 or memory 104. The process then returns to step 302 and proceeds in an iterative fashion. However, returning to step 336, if L2 cache 204 determines that the response does not include an inhibit bit set by L3 cache 108, the process continues to step 340, which illustrates L2 cache 204 ignoring the response sent by L3 cache 108. The process then returns to step 302 and continues in an iterative fashion.
As has been described, the present invention is a system and method of reducing unnecessary cache operations within a data processing system. When a L2 cache controller detects a cache miss in an L2 cache, the L2 cache controller core selects a line (e.g., a victim line) to be cast out of the L2 cache to make room for a newly-requested cache line. The L2 cache controller determines the state of the victim line with respect to the L3 cache and memory by examining the contents of a write-back history table. The write-back history table includes entries that indicate the lines that have been recently evicted from the L2 cache and written to the L3 cache. If an entry in the write-back history table indicates that the victim line has been recently evicted from the L2 cache and that line is also valid in the L3 cache, the victim line is replaced by the newly-requested line without first writing the victim line to the L3 cache or memory. If, however, the victim line is not characterized by an entry in the write-back history table to be recently evicted from the L2 cache and valid in the L3 cache, an entry is made in the write-back history table to reflect such status. This system and method inhibits the redundant writing of cache lines to a L3 cache or memory if the cache lines are determined to be both present in the L3 cache or memory and valid (e.g. unmodified). Therefore, this system and method reduces wasted bandwidth normally utilized by such redundant writing of cache lines.
Of course, persons having ordinary skill in this art are aware that while this preferred embodiment of the present invention reduces the unnecessary writing of cache lines between an L2 cache and L3 cache and/or memory, the present invention may be implemented to reduce the unnecessary of writing any type of data among different levels of any memory hierarchy. Examples of such implementations include: the utilization of random access memory, writeable storage media (e.g., floppy diskette, hard disk drive, read/write CD-ROM, optical media), and flash memory among the different levels of the memory hierarchy. Also, the present invention may be implemented among different levels of data processing systems, where a client might suppress the writing of data to a server if the data is already present and valid on the server.
It should be understood that at least some aspects of the present invention may alternatively be implemented in a program product. Program defining functions on the present invention can be delivered to a data storage system or a computer system via a variety of signal-bearing media, which include, without limitation, non-writable storage media (e.g., CD-ROM), writable storage media (e.g., a floppy diskette, hard disk drive, read/write CD-ROM, optical media), and communication media, such as computer and telephone networks including Ethernet. It should be understood, therefore in such signal-bearing media when carrying or encoding computer readable instructions that direct method functions in the present invention, represent alternative embodiments of the present invention. Further, it is understood that the present invention may be implemented by a system having means in the form of hardware, software, or a combination of software and hardware as described herein or their equivalent.
While the invention has been particularly shown and described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.
The present application is a divisional of U.S. patent application Ser. No. 11/032,875, filed on Jan. 11, 2005. Applicants claim benefit of priority under 35 U.S.C. §120 to U.S. patent application Ser. No. 11/032,875, which is incorporated by reference herein in its entirety and for all purposes.
Number | Name | Date | Kind |
---|---|---|---|
5317720 | Stamm et al. | May 1994 | A |
5564035 | Lai | Oct 1996 | A |
5787478 | Hicks et al. | Jul 1998 | A |
5829038 | Merrell et al. | Oct 1998 | A |
6134634 | Marshall et al. | Oct 2000 | A |
6282615 | Arimilli et al. | Aug 2001 | B1 |
6349367 | Arimilli et al. | Feb 2002 | B1 |
6629210 | Arimilli et al. | Sep 2003 | B1 |
20040236912 | Glasco | Nov 2004 | A1 |
Number | Date | Country | |
---|---|---|---|
20070136535 A1 | Jun 2007 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 11032875 | Jan 2005 | US |
Child | 11674960 | US |