JP2001155001A - Method and device for controlling multiprocessor task - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent the competition of buses between processors even when each of processors accesses a kernel and the task information of an execution task. SOLUTION: The bus configuration of a system is made into two stages of a local bus 15 and a global bus 16, two kinds of memories of private memory 11 accessible only for a present processor and a shared memory 12 commonly accessible even from the other processors are loaded on the local bus 15 of each of processors 1-3, a kernel part is loaded in the private memory 11 and a task information part is loaded in the shared memory 12. Thus, even when each of processors 1-3 accesses the kernel and the task information of the execution task, the competition of buses between processors can be prevented. Further, by performing task scheduling while referring to a data transfer time table, the data transfer time of task information between processors can be shortened.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multiprocessor task control method and a task control apparatus for executing one application program (hereinafter, referred to as an application) including a plurality of tasks on a multiprocessor system having a plurality of processors. About.

[0002]

2. Description of the Related Art Conventionally, a multiprocessor task control method disclosed in, for example, Japanese Patent Application Laid-Open No. 8-36553 is known. This includes multiple processors, one shared memory commonly accessible by each processor,
It is composed of each processor and a system bus connecting the shared memory. Multitasking operating systems are implemented on shared memory. The scheduler on the shared memory performs multiprocessor task control by interrupt driving.

[0003]

However, in the above-described conventional multiprocessor task control method, each processor must always access a shared memory when executing a task. There is a bottleneck in capacity, and there is a problem that throughput does not increase as expected.

[0004] The present invention has been made in view of such a point, and even when each processor accesses task information of a kernel and an execution task, a multiprocessor task control that does not cause bus contention between processors. It is an object to provide a method and a task control device.

[0005]

A multiprocessor task control method according to the present invention realizes a multiprocessor system by mounting the same multitask operating system in local memories of a plurality of processors and realizes a multitask operating system. The system is divided into a kernel part and a task information part, the kernel part is mounted on a private memory accessible only by the own processor, and the task information part is mounted on a shared memory commonly accessible from other processors. I made it.

According to this method, even when each processor accesses the kernel, there is no bus contention between the processors, so that the throughput can be increased. Further, a multiprocessor system can be easily realized simply by changing an existing multitask operating system so that only information that needs to be shared between processors, such as task information, is stored in a specific memory (shared memory). be able to.

In the multiprocessor task control method according to the present invention, in the above multiprocessor task control method, if the task information of the task to be dispatched is stored in a shared memory of a different processor when the processor dispatches the task. Then, the task information on the shared memory is transferred to the shared memory of the own processor.

According to this method, even when each processor accesses the task information of the task being executed,
Since the bus does not compete between the processors, the throughput can be increased.

Further, in the multiprocessor task control method according to the present invention, in the multiprocessor task control method, an area for storing a storage processor identifier is provided in a task information portion as means for managing a processor in which task information is stored. .

According to this method, the processor in which the task information is stored can be centrally managed as one element of the task information, so that the management becomes easy.

In the multiprocessor task control method according to the present invention, in the multiprocessor task control method, a system manages one data transfer time table describing data transfer time between processors in a shared memory, Refers to the data transfer time table to determine the dispatch target task.

[0012] According to this method, the data transfer time of task information generated at the time of dispatch can be reduced, so that the throughput can be increased.

Also, in the multiprocessor task control method of the present invention, at the time of system startup, an application program is downloaded to the private memory of all processors, the master processor starts an entry task of the application program, and all the remaining processors execute Activate an idle task and wait for the task to be interrupted.

According to this method, even when each processor accesses the execution code (source) of the task, there is no bus contention between the processors, so that the throughput can be increased. Further, the application program developer does not need to be conscious of multiprocessor task control at all, and can perform multitask programming using a normal compiler, so that programming becomes easy.

In the multiprocessor task control method of the present invention, when a system call is issued, a context is saved and a system call process is performed. The dispatch target task is selected from the executable tasks in consideration of the data transfer time from the stored processor to the own processor, and if the selected dispatch target task is stored in another processor, the dispatch target task is selected. When the task information is transferred to the processor, dispatching is performed, and the status of another task transits due to system call processing.If there is a task in the executable state with a higher priority than the task in the execution state,
Select a combination of preemption target task and dispatch target task from task priority and data transfer time,
If the selected task to be dispatched is stored in another processor, preemption and dispatching are performed after transferring the task information to its own processor, and context return and system call return are performed. .

According to this method, even when each processor accesses the kernel and the task information, there is no bus contention between the processors and the data transfer time of the task information between the processors is reduced. Therefore, the throughput can be increased.

Further, in the multiprocessor task control method of the present invention, in the above multiprocessor task control method, the method of selecting a dispatch target task is such that all tasks having the highest priority among the tasks in the executable state are assigned to the tasks having the highest priority. The data transfer time between the processor storing the information and the dispatch target processor is calculated from the data transfer time table, and the task with the shortest data transfer time is selected as the dispatch target task.

According to this method, the data transfer time of the task information can be reduced in the case of the same priority while performing the task priority control, so that the throughput can be increased. This is particularly effective for an application in which a number of tasks having the same priority exist.

In the multiprocessor task control method according to the present invention, in the above multiprocessor task control method, the method for selecting a combination of a preemption target task and a dispatch target task can execute a task having a higher priority than a task priority in a certain execution state. The task in the state is selected as the dispatch candidate task, the task in the execution state having a lower priority than the task priority in a certain executable state is selected as the preemption candidate task, and all the combinations of the preemption candidate task and the dispatch candidate task are selected. The data transfer time is calculated from the data transfer time table, and the matching between the preemption target task and the dispatch target task is determined so that the data transfer time is reduced.

According to this method, the matching between the preemption candidate task and the dispatch candidate task can be freely determined while performing the task priority control, so that the data transfer time can be reduced and the throughput can be increased. . This is particularly effective for an application in which there are many controls for giving events to a plurality of tasks at the same time.

Further, in the multiprocessor task control method according to the present invention, in the above multiprocessor task control method, the method of determining the matching between the preemption target task and the dispatch target task is performed when the number of dispatch candidate tasks is larger. Compare the data transfer time of the dispatch candidate task to the candidate task and match the combination with the smallest data transfer time. If the number of preemption candidate tasks is larger, conversely, the data transfer time of the preemption candidate task to the dispatch candidate task If the number of preemption candidate tasks and the number of dispatch candidate tasks are different, if the unmatched task is a preemption candidate task, the task with the highest priority is assigned to the dispatch candidate task. The lowest priority of the task in the case of.

According to this method, the matching between the preemption target task and the dispatch target task can be easily determined without contradicting the task priority control.

In the multiprocessor task control method according to the present invention, in the multiprocessor task control method, when the dispatch target task is stored in its own processor, the dispatch target task is extracted from the executable queue. When connected to the execution queue and stored in another processor, the task to be dispatched is extracted from the executable queue, the task information is transferred to the dispatched processor and connected to the execution queue, while the other queue is The connection is also changed, and the storage processor identifier on the task control block is changed.

According to this method, the transfer of the task information between the processors can be realized by a simple method.
The access to the task information of the execution task of the processor can always be performed within the processor.

A multiprocessor task control device according to the present invention includes a plurality of processors each having a kernel portion of a multitask operating system mounted on a local bus, a private memory accessible only by the processor itself among the processors, A shared memory that implements the task information part of the multitasking operating system and is commonly accessible from other processors,
Is adopted.

According to this configuration, even when each processor accesses the kernel, there is no bus contention between the processors, so that the throughput can be increased.

Further, the multiprocessor task control device of the present invention comprises: a common input / output means for downloading an application program from an external device to each of a plurality of processors; a downloader existing in each of the plurality of processors; The master processor of the plurality of processors starts an entry task of the application program, and the slave processor has an initial program loader that starts an idle task.

According to this configuration, since the master processor starts the entry task and the slave processor starts the idle task, a multiprocessor system can be easily constructed as an application of the multitask system.

Further, the multiprocessor task control device of the present invention has a common input / output means for downloading an application program from an external device to each processor, and a common input / output means for each of the processors. Download the program and link with the multitasking operating system,
If the processor is a master, the application task starts an entry task of the application program.If the processor is a slave, the application loading means starts an idle task of the application program.The processor exists for each processor and saves and restores the context when a system call is issued. A task control block including a context preserving area that exists for each processor and includes a context storage area that exists for each processor;
A task information storage unit for storing a stack area as task information; a system call processing unit that exists for each processor and performs a required system call process; and a system call processing unit that exists for each processor and Task scheduling means for performing various queue controls in accordance with the task state transitioned by the above, and issuing an interrupt control request, a task information transfer request, and a preemption dispatch request based on the determined task switching information, and the task information storage means Executable task management means for managing queues of tasks in a more executable state, execution task management means for managing queues of tasks in an executable state from the task information storage means, and data transfer time between each processor and other processors When managing data transfer Storage means, and matching determination means for determining a task switching information by the present in each processor, said executable task management unit and the execution task management unit and the data transfer time storing means,
Interrupt generating means for each processor which transmits a preemption dispatch request to another processor in response to a request from the task scheduling means; and a preemption dispatch request for each processor which is present for each processor. Interrupt receiving means for receiving the task information and transmitting the task information of the task to be dispatched, which is present for each processor and is requested by the task scheduling means, to the dispatch target processor. And a transfer means.

According to this configuration, even when each processor accesses the kernel and the task information, no bus contention occurs between the processors and the data transfer time of the task information between the processors is reduced. Therefore, the throughput can be increased.

[0031]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The gist of the present invention is to implement a multiprocessor system by mounting the same multitask operating system in local memories of a plurality of processors, and to combine the multitask operating system with a kernel part and a task. Divided into information parts,
The kernel part is mounted on a private memory accessible only by the own processor, and the task information part is mounted on a shared memory commonly accessible from other processors.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a multiprocessor task control device 100 according to an embodiment of the present invention. In the present embodiment, the multiprocessor task control device 100 exemplifies a system including three processors (processor 1, processor 2, and processor 3) having a hardware configuration as shown in the block diagram of FIG. I have. Processors 1 to 3
The memory map viewed from the CPUs 10-1 to 10-3 is as shown in FIG. 4, and the software configuration is as shown in FIG.

In FIG. 2, processor 1 is a master processor, and the remaining processors 2 and 3 are slave processors. Each of the processors 1 to 3 is a CPU (the processor 1 is the CPU 10-1, the processor 2 is the CPU 10-2, and the processor 3 is the CPU 10-1).
-3), private memory 11, and shared memory 12
, An interrupt controller 13, and a bus bridge 14. The CPU, the private memory 11, the shared memory 12, and the interrupt controller 13 are connected to each other via a local bus 15. In addition, the processors 1 to 3 are connected to the respective bus bridges 1.
4 is connected to the global bus 16. A common I / O 4 commonly used by the processors 1 to 3 is connected to the global bus 16 via a bus bridge 14.

In FIG. 1, the multiprocessor task control device 100 includes an application loading unit 110
, Preempt dispatch means 120, task information storage means 140, system call processing means 130
Task scheduling means 150, executable task management means 141, execution task management means 142,
It comprises a data transfer time storage means 160, a matching determination means 170, an interrupt generation means 181, an interrupt acceptance means 182, and a task information transfer means 190. In this case, the executable task management means 14
1, the execution task management means 142, and the data transfer time storage means 160 exist only in the processor 1 which is the master processor, and the other means exist in common in all the processors 1 to 3.

The application loading means 110 downloads the application from the common I / O 4 and links it with the multitasking operating system. If the processor is a master, it starts an application entry task, and if it is a slave, it starts an idle task. I do. The dispatch / preempt means 120 saves and restores the context when issuing a system call. Here, as shown in FIG. 5, three task states are defined: a waiting state, an executable state, and an execution state, and the operation of transitioning the task state from the executable state to the execution state is dispatched, and the operation from the execution state is executable. The operation to transition to the state is defined as preemption.

The task information storage means 140 stores a task control block (TCB) including a context storage area as shown in FIG. 6 and a task stack area as task information. The system call processing means 130 performs a required system call process. Although details are omitted in the present embodiment, control of system resources (existing in each of the processors 1 to 3) and management (one in the system)
Are included in the system call processing means 130. Task scheduling means 150
Performs various queue controls according to the task state transitioned by the system call processing means 130, and performs an interrupt control request based on the determined task switching information;
Issues a task information transfer request and a dispatch preempt request.

The executable task management means 141 uses the task information stored in the task information storage means 140
Tasks in the executable state are queued as shown in FIG. Based on the task information stored in the task information storage unit 140, the execution task management unit 142 manages the queue of the task in the execution state as shown in FIG. The data transfer time storage means 160 manages the data transfer time between each processor and other processors as shown in FIG. The matching determination unit 170 creates a matching table as shown in FIG. 10 by the executable task management unit 141, the execution task management unit 142, and the data transfer time storage unit 160, and determines a preemption / dispatch target task.

The interrupt generating means 181 transmits a preempt dispatch request to another processor in response to a request from the task scheduling means 150. Conversely, the interrupt receiving unit 182 receives a preemption dispatch request from another processor and reports it to the preemption dispatching unit 120. The task information transfer unit 190 transfers task information of the task to be dispatched requested by the task scheduling unit 150 to the processor to be dispatched.

The CPUs 10-1 to 10-1 of the processors 1 to 3
As shown in FIG. 4, the memory map viewed from the CPU 10-3 is such that the private memory 11, the shared memory 12, and the interrupt controller 13 are allocated to the local bus space, and the processors 1 to 3 are allocated to the global bus space.
Memory 12 and interrupt controller 13
And a common I / O4. On the other hand, as shown in FIG. 3, the software configuration of the processor 1 includes a hardware layer including a private memory 11 and a shared memory 12 and an OS (operating system) including a kernel, task information, queue entries, and a data transfer time table. The processor 2 and the processor 3 include a hardware layer including a private memory 11 and a shared memory 12, an OS layer including kernel and task information, and an application layer including an application and a loader. Application layer.

The operation of the multiprocessor task control device 100 configured as described above will be described with reference to FIGS. The multiprocessor task control operation is roughly divided into a system activation operation shown in FIG.
It is composed of two system call operations shown in FIGS. The system call operation includes the main operation shown in FIG. 12, the preemption and dispatch target task selection operation shown in FIGS. 13 to 15, the preemption and dispatch operation shown in FIGS. 16 to 18, and the matching operation shown in FIG. Consisting of one. A multiprocessor task control device including three processors as shown in FIG. 2 will be described as an example.

<System Start Operation> FIG. 11 is a flowchart showing the system start operation. First, step 1
At 001, the application is downloaded from the common I / O 20 onto the private memory 11 of each of the processors 1 to 3, and then, at step 1002, the downloaded application is linked to the multitasking operating system. Next, in step 1003, it is determined whether or not the processor is a master processor. If the processor is a master, the process proceeds to step 1004 to start an application entry task. On the other hand, if the processor is a slave processor, the flow advances to step 1005 to start an idle task (a task that does nothing) and end the processing.

<System Call Operation> FIG. 12 is a flowchart showing the system call operation. First, after saving the context in step 201, step 202
Performs the processing of the requested system call. Then, in step 221, it is determined whether or not the state of the invoking task has transitioned by the system call processing. If the state of the invoking task has transitioned (from the execution state to the waiting state) by the system call processing, the process proceeds to step 222. Take out the invoking task from the execution queue. As a result, the processor becomes idle, so that a task to be dispatched is selected in step 231, and dispatching is performed in step 241. After dispatching, the context is restored in step 281 and the system call process ends.

On the other hand, if it is determined in step 221 that the state of the invoking task has not transitioned by the system call processing (it remains in the execution state), step 251 is executed.
To determine whether or not the state of another task has transitioned by the system call processing. If the state of another task has transitioned (from the waiting state to the execution state), step 252 is executed.
To connect the other tasks in the ready state to the ready queue. Then, in step 253, it is determined whether or not the task with the highest priority in the executable queue has a higher priority than the task with the lowest priority in the execution queue. If the highest priority task in the executable queue has a higher priority than the lowest priority task in the execution queue, step 2
Proceeding to 61, a preemption / dispatch target task is selected, and in step 271, preemption / dispatching is performed. Then, in step 281, the context is restored, and the system call process ends.

If it is determined in step 253 that the highest priority task in the executable queue has a lower priority than the lowest priority task in the execution queue, the context is immediately restored in step 281. Terminates the system call processing. On the other hand, if it is determined in step 251 that the state of the other task has not transitioned (remains in the waiting state), the context is immediately restored in step 281 and the system call processing ends.

<Dispatch target task selection operation> FIG.
3 is a flowchart showing a dispatch target task selecting operation. First, in step 301, the tasks of the highest priority in the executable queue (all tasks having the same priority, if any) are set as dispatch candidate tasks.
Then, in step 302, the data transfer time between the processor storing the dispatch candidate task and the dispatch target processor is calculated from the data transfer time table for all combinations. When the result of this calculation is obtained, in step 303, the task with the shortest data transfer time is selected as the task to be dispatched, and the process ends.

<Preemption / Dispatch Target Task Selection Operation> FIG. 14 is a flowchart showing the preemption / dispatch target task selection operation. First, in step 401, a preempt dispatch candidate task is selected. Next, in step 411, the data transfer time between the processor in which the dispatch candidate task is stored and the processor in which the preemption candidate task is executed is calculated from the data transfer time table for all combinations, as shown in FIG. Create a matching table. After creating the matching table, step 4
At 21, matching is performed between the preemption target task and the dispatch target task from the created matching table. Then, the combination determined by the matching in step 431 is selected as the dispatch target task and the preemption target task, and the processing is terminated.

<Preempt / Dispatch Candidate Task Selection Operation> FIG. 15 is a flowchart showing a preempt / dispatch candidate task selection operation. First, in step 501, an execution queue entry is selected, and a head task of the queue is selected from the executable queue entries. After selecting the execution queue entry and the head task of the queue, in step 502, the priorities of the tasks in the execution queue and the tasks in the executable queue are compared. If the task in the executable queue has a higher priority in the determination in step 503 based on the comparison result, step 504 is executed.
The task in the execution queue is set as a preemption candidate task, and the task in the executable queue is set as a dispatch candidate task.

Next, in step 505, the tasks in the execution queue and the executable queue are advanced by one. Next, if it is determined in step 506 that the task has reached the end of the execution queue or the end of the executable queue, the process proceeds to step 507 and a task on the execution queue having the same priority as the priority of the preemption candidate task is added to the preemption candidate task. Then, a task on the executable queue having the same priority as the priority of the dispatch candidate task is added to the dispatch candidate task, and the process ends.

On the other hand, if the priority of the task in the executable queue is lower than the priority of the task in the executable queue in step 503, the process immediately proceeds to step 507, and the task in the execution queue having the same priority as the priority of the preemption candidate task is executed. Is added to the preemption candidate task, a task on the executable queue having the same priority as the priority of the dispatch candidate task is added to the dispatch candidate task, and the process ends.

As a result of advancing the tasks of the execution queue and the executable queue by one, if it is determined in step 506 that the task has not reached the execution queue end or the executable queue end, the process returns to step 502 and the task of the execution queue is A process for comparing the priorities of the tasks in the executable queue is performed.

<Preemption / dispatch operation> FIG.
6 is a flowchart showing the preempt dispatch operation. First, if it is determined in step 601 that the preemption target task is being executed by the own processor, the process proceeds to step 611 to perform preemption processing. Then, after the dispatch processing is performed in step 621, in the determination in step 631, if all the selected tasks have been preempted and dispatched, the processing ends. If all of the selected tasks have not been preempted and dispatched yet, step Returning to 601, a determination process is performed to determine whether the preemption target task is being executed by its own processor.

On the other hand, if it is determined in step 601 that the preemption target task is not being executed by its own processor, the flow advances to step 641 to generate an interrupt indicating a preemption / dispatch request to the dispatch target processor. To determine whether all of the selected tasks have been preempted and dispatched.

In the processor in which the interrupt indicating the preempt dispatch request is generated, step 65 is executed.
In step 1, the context is saved, and then in step 661.
Performs preemption processing. Thereafter, in step 671, dispatch processing is performed, and then, in step 681, the context is saved, and the processing ends.

<Dispatch Operation> FIG. 17 is a flowchart showing the dispatch operation. First, step 7
In the determination of 01, if the task to be dispatched is on its own processor, the process proceeds to step 702, where the task selected from the executable queue is taken out, connected to the execution queue, and the process is terminated. If the task to be dispatched is not on its own processor (is on another processor), the process proceeds to step 703, where the task selected from the executable queue is taken out, and the task information at that time is transferred to the shared memory of the processor to be dispatched. , Connect it to the run queue. Further, the connection is also taken out and connected to other links of the queue (the front and rear links of the task control block, etc.). Finally, the storage processor identifier 20 in the task control block (TBC, see FIG. 6) is changed, and the process ends.

<Preemption Operation> FIG. 18 is a flowchart showing the preemption operation. Step 801
Then, the selected task is taken out of the execution queue, connected to the executable queue, and the processing is terminated.

<Matching Operation> FIG. 19 is a flowchart showing the matching operation. First, step 901
Then, the difference between the number of dispatch candidate tasks and the number of preemption candidate tasks is calculated as the minimum priority limit number. next,
If it is determined in step 902 that the number of dispatch candidate tasks is larger than the number of preemption candidate tasks,
Proceeding to step 911, a preemption candidate task is selected from the matching table. After selecting a preemption candidate task, at step 912, a dispatch candidate task having a minimum data transfer time for the selected preemption candidate task is selected.

If the number of dispatch candidate tasks is different from the number of preemption candidate tasks in step 913, it is determined in step S914 whether or not the lowest priority dispatch candidate tasks have been matched by the minimum priority limit number. If the matching is performed for the priority limit number,
In step 915, the lowest priority dispatch candidate task that has not been matched is deleted from the matching table.

If it is determined in step 916 that all preemption candidate tasks have been matched, the process ends. If all preemption candidate tasks have not been matched yet, the process returns to step 911 to return to the preemption candidate task. Select a task from the matching table. When the number of dispatch candidate tasks and the number of preemption candidate tasks are the same (step 913), when the number of dispatch candidate tasks is different from the number of preemption candidate tasks (step 913), and the lowest priority dispatch candidate tasks are matched by the minimum priority limit number If not (step 914), the process proceeds to step 916, and a determination process is performed to determine whether all preemption candidate tasks have been matched.

On the other hand, if it is determined in step 902 that the number of dispatch candidate tasks is smaller than the number of preemption candidate tasks, as in steps 921 to 926, the dispatch candidate tasks are replaced with preemption candidate tasks. Is replaced by a dispatch candidate task, and the same operations as those in steps 911 to 916 described above are performed.

Next, the manner of execution of a task according to the embodiment of the present invention will be described with reference to FIG. As will be described below, it can be seen that the present invention can minimize the data transfer time. The processor 1 is the master, and starts task_A which is an entry task of the application at time t0. Processor 2 and processor 3
Is a slave, starts an idle task idle at time t0,
Waiting for a task to be assigned (interrupted). At time t1, task task_A generates task task_B, processor 2 is selected as a dispatch target processor according to the priority control, processor 1 generates an interrupt indicating a preemption dispatch request to processor 2, and processor 2 issues task task_B Preempt
Dispatch. At this time, the task information of the task task_B is transferred from the processor 1 to the processor 2.

Thereafter, similarly, at time t2, the task task_A generates the task task_C, and the processor 3
Is selected as a dispatch target processor, the processor 1 generates an interrupt indicating a preempt dispatch request to the processor 3, and the processor 3
Preempt dispatch task_B. At this time, the task information of the task task_C is transferred from the processor 1 to the processor 3. At time t8, the task task_A transitions to the waiting state, and the processor 1 dispatches the highest priority task task_G among tasks executable at that time.

Thereafter, at time t9, the task task_A transitions from the wait state to the executable state, and the processor 1 having a short data transfer time is selected as the dispatch target processor according to the multiprocessor task control method of the present invention, and no data transfer processing is performed. (Because the task information of task task_A originally existed in processor 1, processor 1
Preempt dispatch task_A. At time t10, the task task_E transitions to the wait state, and the processor 3 selects the task task_G as the dispatch target task according to the priority control, and dispatches the data with data transfer.

At time t11, the task task_F shifts to the waiting state, and the processor 2 executes the task task according to the priority control.
task_D is selected as the task to be dispatched, and dispatching is performed without data transfer (data transfer did not occur by accident). At time t12, the task task_A transitions to the waiting state again, and the processor 1 executes the task task_B according to the priority control.
Is selected as a dispatch target task, and dispatch is performed with data transfer.

Then, at time t13, task task_A and task t
The ask_E transitions from the wait state to the executable state, matching is performed so that the data transfer time is reduced according to the multiprocessor task control method of the present invention, and the processor 1 dispatches the task task_A and the processor 2 dispatches the task task_E. And so on. The preemption dispatch for the task task_A of the processor 1 is performed without the data transfer processing (since the task information of the task task_A originally exists in the processor 1), and the preemption dispatch for the task task_E of the processor 2 is performed with the data transfer processing. .

As described above, according to the embodiment of the present invention, the system has a two-stage configuration of the local bus 15 and the global bus 25 via the bus bridge 14, and the local bus of each of the processors 1 to 3. 15 and a private memory 11 accessible only by its own processor.
In addition, two types of memories having different applications of the shared memory 12 that can be commonly accessed from other processors via the global bus 25 and the bus bridge 14 are mounted. Then, the multitasking operating system is divided into a kernel part that does not need to be shared by the entire system, and management information and task information parts of system resources (for example, semaphores and message queues) that need to be shared. Is implemented in the private memory 15 and the task information portion is implemented in the shared memory 12.

The shared memory 1 of each of the processors 1 to 3
The data transfer time between the two is stored, and the task scheduling is performed so that the data transfer time of the task information between the processors is reduced.

Therefore, even when each of the processors 1 to 3 accesses the kernel, there is no bus contention between the processors, and even when accessing the task information, the task information is not updated before executing each task. If it is stored on the processor, the task information data transfer processing must be performed, but there is no bus contention between the processors for subsequent access to the task information, so that the throughput can be increased. .

The task information data transfer process also includes a data transfer time storage unit 160 and a matching determination unit 1.
Since task scheduling is performed based on the stored data transfer time, the processing can be completed in a short time, and the throughput can be further increased by doing this.

Although the present invention relates to general multiprocessor task control, if the target application is considered as, for example, a call control system application, it can be applied to a mobile communication system (particularly, a base station device). In this case, for example, each task is a call control task for controlling each call.

[0070]

As described above, according to the present invention, the system has a two-stage configuration of a local bus and a global bus.
Two types of memory are implemented on the local bus of each processor: a private memory that can be accessed only by its own processor, and a shared memory that can be commonly accessed by other processors.
By mounting the task information portion in the shared memory, even when each processor accesses the task information of the kernel and the execution task, the bus does not compete between the processors, so that the throughput can be increased. The effect is obtained.

Further, by performing the task scheduling with reference to the data transfer time table, the data transfer time of the task information between the processors can be reduced, and the throughput can be further increased by doing this.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a multiprocessor task control device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 3 is a diagram showing a software configuration in the multiprocessor task control device according to the embodiment of the present invention;

FIG. 4 is a diagram showing an example of a memory map viewed from each CPU of the multiprocessor task control device according to the embodiment of the present invention;

FIG. 5 is a diagram for explaining a task state transition on a multitask operating system in the multiprocessor task control device according to the embodiment of the present invention;

FIG. 6 is a diagram showing task information in the multiprocessor task control device according to the embodiment of the present invention.

FIG. 7 is a diagram for explaining an executable queue in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 8 is a diagram for explaining an execution queue in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 9 is a diagram showing a data transfer time table in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 10 is a view showing a matching table in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 11 is a flowchart showing a system startup operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 12 is a flowchart showing a system call operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 13 is a flowchart showing a dispatch target task selecting operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 14 is a flowchart showing a preemption / dispatch operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 15 is a flowchart showing a preemption / dispatch candidate task selection operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 16 is a flowchart showing a preemption / dispatch operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

FIG. 17 is a flowchart showing a dispatch operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 18 is a flowchart showing a preemption operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 19 is a flowchart showing a matching operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention.

FIG. 20 is a diagram for explaining an execution task switching operation in a multiprocessor system using the multiprocessor task control device according to the embodiment of the present invention;

[Explanation of symbols]

 1, 2, 3 Processor 4 Common I / O 10-1, 10-2, 10-3 CPU 11 Private memory 12 Shared memory 13 Interrupt controller 14 Bus bridge 15 Local bus 16 Global bus 20 Saved processor identifier 100 Multiprocessor task control Apparatus 110 Application loading means 120 Preempt / dispatch means 130 System call processing means 140 Task information storage means 141, 142 Executable task management means 150 Task scheduling means 160 Data transfer time storage means 170 Matching determination means 181 Interrupt generation means 182 Interrupt acceptance Means 190 Task information transfer means

Claims (13)

[Claims] 1. A multi-processor system in which the same multi-task operating system is implemented in a local memory of each of a plurality of processors to realize a multi-processor system. The multi-task operating system is divided into a kernel part and a task information part. A multiprocessor task control method, wherein the task information portion is implemented in a shared memory that can be commonly accessed by other processors, with the portion being implemented in a private memory accessible only by the own processor. 2. When a processor dispatches a task, if task information of a task to be dispatched is stored in a shared memory of a different processor, the task information on the shared memory is transferred to the shared memory of the own processor. 2. The multiprocessor task control method according to claim 1, further comprising: 3. The multiprocessor task control according to claim 1, wherein an area for storing a saved processor identifier is provided in the task information portion as means for managing the processor in which the task information is stored. Method. 4. A system in which a data transfer time table describing data transfer time between processors is managed by a system on a shared memory, and each processor determines a task to be dispatched by referring to the data transfer time table. The multiprocessor task control method according to any one of claims 1 to 3, wherein: 5. When the system is started, the application programs are downloaded onto the private memories of all the processors, the master processor starts an entry task of the application program, and all the remaining processors start an idle task to interrupt the task. A multiprocessor task control method, comprising: 6. When a system call is issued, a context is saved and a system call process is performed. When the status of the invoking task is changed by the system call process, a task priority and task information are transferred from the processor to the invoking processor. The task to be dispatched is selected from the executable tasks in consideration of the data transfer time, and if the selected task to be dispatched is stored in another processor, the task information is transferred to its own processor before dispatching When the status of another task changes due to system call processing, if there is a task in the executable state with a higher priority than the task in the execution state, the task to be preempted is dispatched based on the task priority and the data transfer time. Select the target task combination and select the selected disk. If the task to be patched is stored in another processor, pre-emption and dispatching are performed after transferring the task information to the own processor, and context recovery and system call return are performed. Multiprocessor task control method. 7. A method for selecting a task to be dispatched includes:
For all the tasks with the highest priority among the tasks in the executable state, the data transfer time between the processor in which the task information is stored and the dispatch target processor is calculated from the data transfer time table, and the data transfer time is calculated. 7. The multiprocessor task control method according to claim 6, wherein the smallest task is selected as a dispatch target task. 8. A method for selecting a combination of a task to be preempted and a task to be dispatched, wherein a task in an executable state having a higher priority than a task priority in a certain execution state is selected as a dispatch candidate task, and a task priority in a certain executable state is selected. Tasks in the execution state with a lower priority than the task are selected as preemption candidate tasks, and the data transfer time is calculated from the data transfer time table for all combinations of the preemption candidate task and the dispatch candidate task, and the data transfer time is reduced. 7. The multiprocessor task control method according to claim 6, wherein the matching between the preemption target task and the dispatch target task is determined as described above. 9. A method for determining matching between a task to be preempted and a task to be dispatched, wherein when the number of dispatch candidate tasks is larger, the data transfer time of the dispatch candidate task with respect to the preemption candidate task is compared. If the smallest combination is matched and the number of preemption candidate tasks is larger, on the contrary, matching is performed based on the data transfer time of the preemption candidate task with respect to the dispatch candidate task,
In addition, if the number of preemption candidate tasks and the number of dispatch candidate tasks are different, the unmatched task must be the highest priority task if it is a preemption candidate task and the lowest priority task if it is a dispatch candidate task. 9. The multiprocessor task control method according to claim 8, wherein: 10. The dispatch method according to claim 1, wherein when the dispatch target task is stored in the own processor, the dispatch target task is taken out of the executable queue and connected to the execution queue, and when the task is stored in another processor, the execution is executed. The dispatch target task is taken out from the possible queue, the task information is transferred to the dispatch target processor and connected to the execution queue, while the other queues are switched and the storage processor identifier on the task control block is changed. 7. The multiprocessor task control method according to claim 6, wherein: 11. A processor having a kernel part of the multitask operating system mounted on a local bus, a private memory accessible only by the processor among the processors, and task information of the multitask operating system. A shared memory that implements a part and that can be commonly accessed by other processors. 12. A common input / output unit for downloading an application program from an external device to each of a plurality of processors, a downloader existing in each of the plurality of processors, and a master processor among the plurality of processors, Start the program entry task,
A multiprocessor task control device, comprising: an initial program loader that starts an idle task in a slave processor. 13. A multi-task operating system, comprising: a common input / output unit for downloading an application program from an external device to each processor; Application loading means for activating the application program entry task when the processor is a master, and activating the application program idle task when the processor is a slave; and Preempt dispatch means for saving and restoring; a task control block and a task stack area that are provided for each processor and include a context saving area that exists for each processor. Task information storage means for storing information as task information, system call processing means for each of the processors for performing a required system call processing, and system call processing means for each of the processors, wherein the system call processing means A task scheduling unit that performs various queue controls according to the transitioned task state and issues an interrupt control request, a task information transfer request, and a preemption dispatch request based on the determined task switching information; and Executable task management means for managing queues of tasks in the executable state; execution task management means for managing queues of tasks in the executable state from the task information storage means; and data transfer time between each processor and other processors. Data transfer time memory to manage And a matching determination unit that exists for each processor, and determines task switching information by the executable task management unit, the execution task management unit, and the data transfer time storage unit, and exists for each processor. Interrupt generating means for transmitting a preemption dispatch request to another processor in response to a request from the task scheduling means; and an interrupt generating means which exists for each of the processors and receives a preemption dispatch request from the other processor, and receives the preemption dispatch request. Interrupt accepting means for reporting to the means; and task information transfer means for each of the processors, for transferring task information of the task to be dispatched requested by the task scheduling means to the dispatch target processor. Multiprocessor control apparatus according to claim and.