Translation Lookaside Buffer (TLB) Arrangement Wherein the TLB Contents Are Retained From Task When It Is Swapped Out and Reloaded When the Task Is Rescheduled

1. Field of the Invention

The present invention relates generally to the field of computer systems and in particular, to translation lookaside buffers as used in memory paging subsystems.

2. Prior Art

Known prior art processors utilize a technique known as paging in order to manage memory resources. Paging is a memory management technique wherein a program is divided into uniform sized blocks called pages. Paging is used in support of various computer system features such as multi-tasking. An example of a processor utilizing paging is the Intel486.TM. family of processors, available from Intel Corporation of Santa Clara, Calif.

In a paging system data is moved to and from system memory by pages. A key aspect of paging is the translation of an address provided by a program, termed a linear address, to a physical address. The physical address is the real address of the page of data in storage. Various computer architectures utilize different address schemes. The address translation scheme as utilized by the Intel486 product is described with reference to FIG. 1. Referring to FIG. 1, a linear address is provided to a paging unit. Note that a linear address 101 is provided to the paging unit when servicing a page fault. A page fault occurs when a page accessed by an executing program, is not in memory. The linear address is first compared to entries in a Translation Lookaside Buffer (TLB) 102. The TLB 102 is a cache of the thirty-two most commonly referenced page table entries of a currently executing task. The page table entries contain the physical address for the page in a storage medium. If the linear address is found, a TLB hit has occurred. Thus the desired physical address is found directly in the TLB. This is desirable since it avoids subsequent processing by the paging unit and results in an increased in speed in the translation of a linear address to physical address.

If the linear address is not found in the TLB 102, then the linear address must be translated. The Intel486 utilizes a two level translation scheme. A first, portion of the linear address is utilized to index to an entry in Page Directory 104. The Page Directory 104 is a table of indices into Page Table 105. In the Intel486 the upper ten bits of the linear address are used as an index to the Page Directory 104.

A second portion of the linear address provides an offset to the Page Table index retrieved from the Page Directory 104 to create an index to one of the Page Table entries. Each Page Table entry in Page Table 105 contains the starting address of the page frame as well as statistical information about the page. This starting address is the desired physical address for the page.

Processing in this manner continues until a task switch occurs. A task switch may occur as a result of the expiration of allotted time for a task, or as a result of an interrupt. In any event, whenever a task switch occurs, the page directory and page table for the new task are loaded, and the TLB must be flushed. By flushed it is meant that the TLB's contents are cleared. The contents of the TLB are entered as page faults occur in the executing task. As tasks are continually swapped in and out, the TLB is continually being flushed. This has the effect of wasting the effort of building entries in the TLB in a prior execution of the task.

Thus, it is desirable to improve utilization of a Translation Lookaside Buffer by eliminating the need to reload the TLB through page faults when a task switch occurs.

A method and apparatus for utilizing a Translation Lookaside Buffer (TLB) in a paging unit on a computer system, is disclosed. In the present invention the contents of a TLB are saved when a task switch occurs. The TLB contents are associated with the task being executed. When the task is again scheduled for execution, the old TLB contents are re-loaded back into the TLB. Two storage areas are embodied in the TLB structure. One storage area is for storing page table entries for an incoming task and the second storage area is for storing page table entries for an outgoing task. On a program task switch, the contents of the TLB are moved to the outgoing TLB storage area and the TLB is loaded with the contents of the incoming TLB storage area. As a current task is executing, the next scheduled task is identified. When spare memory cycles are available, the TLB for the incoming task is moved from memory and into the incoming TLB storage area. Similarly, the outgoing TLB storage area is moved to memory during spare memory cycles while the new current task is executed.

FIG. 1 is a block diagram of a prior art paging unit.

FIG. 2a illustrates a computer system as may be utilized by the currently preferred embodiment of the present invention.

FIG. 2b is a simplified block diagram of a processor architecture having a paging unit as may be embodied by the currently preferred embodiment of the present invention.

FIG. 3a is a block diagram illustrating the address translation scheme of the currently preferred embodiment of the present invention.

FIG. 3b is a flow chart illustrating the steps taken for management of the contents of a Translation Lookaside Buffer during a task switch as may be performed in the currently preferred embodiment of the present invention.

FIG. 4 is a block diagram of Translation Lookaside Buffer arrangement as may be utilized in the currently preferred embodiment of the present invention.

FIG. 5 is a state diagram of a Translation Lookaside Buffer controller as may be utilized in the currently preferred embodiment of the present invention.

A translation lookaside buffer arrangement as used in paging units on a computer system, is described. In the following description, numerous specific details, e.g. page directory and page table structures, are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced without such specific details. In other instances, specific implementation details, such as timing diagrams for memory cycle transfers, have not been shown in detail in order not to unnecessarily obscure the present invention.

The computer system of the preferred embodiment is described with reference to FIG. 2a. The present invention may be implemented on a general purpose microcomputer, such as one of the members of the IBM compatible Personal Computer family, or one of several work-station or graphics computer devices which are presently commercially available. A computer system as may be utilized by the preferred embodiment generally comprises a bus or other communication means 201 for communicating information, a processor means 202 coupled with said bus 201 for processing information, a random access memory (RAM) or other storage device 203 (commonly referred to as a main memory) coupled with said bus 201 for storing information and instructions for said processor 202, a read only memory (ROM) or other static storage device 204 coupled with said bus 201 for storing static information and instructions for said processor 202, a data storage device 205, such as a magnetic disk and disk drive, coupled with said bus 201 for storing information, instructions and task data, an alphanumeric input device 206 including alphanumeric and other keys coupled to said bus 201 for communicating information and command selections to said processor 202, a cursor control device 207, such as a mouse, track-ball, cursor control keys, etc., coupled to said bus 201 for communicating information and command selections to said processor 202 and for controlling cursor movement and a display device 208 coupled to said bus 291 for providing visual information to a user.

The currently preferred embodiment of the present invention is implemented on a computer system capable of supporting multi-tasking and which utilizes a paging unit, or it's equivalent, for performing address translations. A multi-tasking system allows for a plurality of tasks to be in an execution state. The computer system would allocate processor time slices for each of the tasks. This provides an appearance of the multiple tasks being executed concurrently. Specifically, a paging unit would include a cache structure, commonly referred to as a Translation Lookaside Buffer (TLB), to facilitate address translations from a provided linear address to a physical address.

FIG. 2b is a simplified block diagram of a processor architecture with a paging unit as may be utilized by the currently preferred embodiment of the present invention. The architecture includes an ALU 220 for performing arithmetic computations, registers 221 for storing information and control data and instruction decode module 225 for decoding instructions. In order to load data into a system memory, segmentation unit 222 and paging unit 223 perform address translations to identify a physical address in storage. The TLB arrangement is preferably embodied in the paging unit 223. The components are all coupled to system bus interface 224. The system bus interface 224 provides for coupling to, a bus structure for example, Bus 201 of FIG. 2a.

As would be apparent to one skilled in the art, the processor architecture of FIG. 2b may be implemented as discrete components, or as a single integrated circuit, e.g. a microprocessor device. Accordingly, the present invention may be practiced on either a discrete computer system, or incorporated into a microprocessor device.

The present invention is practiced in systems which support paging. A Translation Lookaside Buffer (TLB) is a cache of frequently accessed page table entries. Each page table entry contains a physical address to a page in memory. The purpose of the TLB is to increase the speed by which an address translation occurs.

Generally, paging is a memory management technique where programs and addressable memory are divided onto uniform sizes, e.g. a page. A second memory management technique is termed segmentation. Segmentation is used to group regions of memory which have common attributes. Segmentation is used as a basis for memory protection schemes. For example, all of the code of a given program could be contained in a segment. This segment could then be protected to prevent unwanted modification of instructions.

The address translation scheme of the currently preferred embodiment incorporates segmentation and paging aspects. This scheme is utilized in the Intel486 family of products and is illustrated in FIG. 3a. Referring to FIG. 3a, an effective address 320 is derived from the contents and format of an instruction. The components of an effective address are a base address 321 plus a displacement value 322 plus in some instances a scaled index value 323. The effective address is then provided to a segmentation unit 324. The segmentation unit adds a segment component to the effective address 320 to create a linear address 325. The linear address 325 is then translated into a physical address 327 by a paging unit 326. The physical address 327 is then used to access physical memory 328.

The paging unit 326 translates the linear address 325 through a series of table look-ups using various portions of the linear address 325. A first portion is used to access an entry in a page directory (not illustrated). The entries in a page directories are base indices into a page table (not illustrated). A second portion of the linear address is then added to the obtained page directory entry to create an index into a page table. The entry in the page table contains the desired physical address. Each page directory and page table has a direct relationship to the task being executed.

The use of a Translation Lookaside Buffer (TLB) avoids the Table Look-ups. The TLB it is a cache of frequently accessed page table entries. By avoiding table Look-ups, processing time is saved.

A complication in the operation of the TLB cache is the switching of tasks. The present invention is used in systems that have multi-tasking capability. By multi-tasking capability it is meant that the application can concurrently execute multiple tasks according to some time sharing scheme. The complication introduced is that the page directory and page table are unique to a particular program, i.e. a task. Thus, whenever a new task is scheduled for execution, new page directory and table entries must be loaded. Further, the Translation Lookaside Buffer must be flushed so as to avoid generating incorrect physical addresses.

In the present invention, the process for preparing a paging unit for executing a different task, the TLB is loaded with the page table entries from the last time the task was executed. This has the advantage of not requiring the dynamic re-loading of the TLB or the flushing of the TLB.

FIG. 3b is a flow chart illustrating the steps taken for reloading the TLB with entries from the last time the task was executed in response to a task switch. First, a task switch is detected, step 301. Detection of a task switch within the paging unit is via signals sent to a TLB controller. The TLB controller will be described in greater detail below. A task switch may be generated as a result of a program interrupt or by the current task exhausting it's allotted processing time. In any event, once a task switch is detected, the current TLB contents are loaded into an outgoing TLB storage area, step 302. This has the effect of saving the contents of the TLB for the outgoing task. The current, incoming and outgoing TLB storage areas are preferably sets of registers having identical structures. A check is then made to determine if an interrupt caused the program switch, step 303. An interrupt is not a scheduled task, so special handling is performed. Assuming an interrupt is the reason for the task switch, the interrupt is serviced, step 304. Servicing an interrupt is a term of art which refers to an interrupt handler, which may be part of the computer operating system, responding to and correcting for the conditions which caused the interrupt to occur. Once the interrupt is serviced, the contents of the outgoing TLB storage area are loaded back into the current TLB storage area, step 305.

In the currently preferred embodiment, the originally executing task is restored after the interrupt is serviced. Once this is completed, the task resumes execution until the next task switch. Other embodiments may completely swap out the task and the next scheduled task is executed. Such other embodiments would of course not perform the step 305 described above. Such embodiments would not depart from the spirit and scope of the present invention.

If the task switch is not caused by an interrupt, two sequences of events may then occur concurrently. First, the contents of an incoming TLB storage area is loaded into the current TLB area, step 306. At this point, the current TLB contains TLB entries for the task that is scheduled to be executed. The execution of the next scheduled task may now begin, step 307. The contents of the outgoing TLB register may then be spilled to temporary storage, step 308. By spilling, it is meant that spare memory cycles, are used to make this transfer. Note, that the temporary storage may be associated with the task by adding the task ID to the stored TLB contents.

Concurrently with steps 306-308, the next scheduled task is identified, step 309. Identification of the next scheduled task depends on the task scheduling technique used. Generally, it can be accomplished by querying the task runlist managed by the scheduler. The incoming TLB storage area is filled with the TLB data for the next scheduled task, step 310. By filling it is meant that spare memory cycles are utilized to perform the transfer of data from the temporary storage to the incoming TLB registers.

Spare memory cycles is a term of art that refers to periods of time that are unused for task execution. For example, spare memory cycles occur when the processor has no reads pending from memory, nor are there any writes queued to memory. Note that the temporary storage may be a portion of the computer system's main memory, a cache structure, external storage or even additional provided memory. The choice of how to embody the temporary storage is one that would depend on available system resource, time given for task switch, desired cost of the system and other such variables.

The present invention may also take other embodiments. For example, the spill and fill of incoming and outgoing Translation Lookaside Buffer (TLBs) storage areas may not need to occur to an external memory. It would be apparent to one skilled in the art that the outgoing TLB may be transferred out in conjunction with the structure used for storing task information (e.g. a Task State Segment or TSS). Further, when the task is scheduled to be executed, the TLB from the TSS could be inserted into the TLB.

FIG. 4 is a block diagram of the TLB arrangement of the currently preferred embodiment of the present invention. The TLB arrangement is controlled by the TLB controller 407. The TLB controller manages not only the transfer of the TLB data between the incoming registers 404, TLB registers 405 and outgoing registers 406, but it also controls the scanning of the TLB registers 405 looking for matches. In the currently preferred embodiment, the TLB controller 407 may be a logic device such as a Programmable Logic Array (PLA) or a gate array. The TLB controller 407 is coupled to incoming task register 401 and outgoing register 403. Note that coupled between incoming task 401 and outgoing task register 403 is task register 402. The task register 402 contains the task information for the currently executing task. Note that the data from task register 402 will propagate to outgoing task register 403 and the data from incoming task register 401 will propagate to task register 402 as the task switch occurs. As described above, the incoming task register receives its information identifying the incoming task from the scheduler. In any event, the incoming task register is coupled to the TLB controller and is used to identify the data in memory from which the incoming task registers 494 will be filled. In a similar manner, the outgoing task register 403 is coupled to the TLB controller 407 for identifying the outgoing TLB information with the outgoing task. The TLB controller 407 is also used to determine whether a fill to the incoming task storage area or a spill to memory will occur. The TLB controller 497 will allocate spare memory cycles between a fill and spill operations according to the page unit operating scheme. For example, the fill of the incoming task storage area from memory may be performed before any spill of the outgoing task storage area to memory. It would be apparent to one skilled in the art to use any such scheme to accomplish the spills and fills. Finally, the memory controller 408 is used to access the system memory for fill and spill operations.

The present invention may also take on other embodiments. As the currently preferred embodiment was designed for use in computer system with processor architectures like the Intel486DX, certain variations on the implementations may be required on support of other architectures. Such variations would not depart from the spirit and scope of the present invention.

TLB Controller Operation

FIG. 5 is a state diagram showing the operation of TLB controller 407 of FIG. 4 with respect to the TLB management of the present invention. The TLB controller starts in an idle state 501. When in this idle state, no data is being transferred. Upon initiation of a task switch the TLB controller enters a TLB transfer state 502 where the contents of the current TLB is transferred to the outgoing TLB and the incoming TLB is transferred to the current TLB. When the transfer of the TLB registers is complete the fill/spill state 503 is entered. During the fill/spill state the outgoing TLB data is spilled to external storage and the incoming TLB is filled from external storage. As described above, this occurs when spare memory cycles are available. During this state if another task switch occurs, the TLB transfer state is entered. When this fill/spill is completed TLB controller returns to the idle state 501.

The currently preferred embodiment of the present invention would have particular application in systems with relatively small task-switch quanta, or systems which need predictable task switch times. However, the present invention may be practiced on systems with different operating characteristics.

Thus, a method and apparatus for optimizing use of Translation Lookaside Buffers, is disclosed.