JPH0855074A - Input/output processing system - Google Patents

Abstract

PURPOSE:To optimize the processing efficiency as the whole of the transfer processing by adjusting the storage start position of transfer data on a kernel space in accordance with the length or user data transferred to an input/output device. CONSTITUTION:In the case of data to be copied whose length is shorter than the value preliminarily determined by the computer system, true storage start position or transfer data on the kernel space is set so as to be on the boundary of 32 bytes because transfer in 32-byte units is most efficient with respect to the transfer capability or an input/output bus when data on the memory is transferred from the input device into the input/output device by the direct memory access system. In the case of the data length longer than the preliminarily determined value, the storage start position or transfer data on the kernel space is so set that the data storage start position of the copy source and that of the copy destination match with each other, thus realizing the most efficient data copy processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an input / output processing system for a computer system, and more particularly to a system for optimizing the processing efficiency of transferring user data to an input / output device.

[0002]

2. Description of the Related Art Conventional processing for transferring user data to an input / output device is described in "WRITING A UNIX DEVICE DRIVER", John
As described in "Wiley & Sons, 1988", methods for optimizing the process of copying data from the user space to the kernel space and the process of transferring data from the kernel space to the input / output device are shown. The former is a method of performing copy processing by using a data type that is as large as possible corresponding to a common boundary value at the data storage start positions on both the copy source and copy destination memories, and the latter is the data storage start position on the memory. This is a method of performing transfer processing in a transfer unit as large as possible according to the boundary value of. However, they are treated as independent processes, and there is no relation between them.

[0003]

In the conventional transfer processing method, the processing of copying data from the user space to the kernel space and the processing of transferring from the kernel space to the input / output device are handled as independent processing. Is not considered to be optimized as the processing of.

Therefore, for example, if the boundary values of the data storage start positions on the memory of the copy source and the copy destination are made coincident with each other in order to optimize the process of copying the data from the user space to the kernel space, the data is transferred from the kernel space. The boundary value of the storage start position on the memory becomes inappropriate for the transfer processing when transferring to the output device, and transfer is performed in small units until at least the appropriate boundary value is reached, especially when the data length is short, overhead Grows larger. Conversely, when the boundary value of the storage start position in the memory is set to an optimum value for the transfer processing when transferring from the kernel space to the I / O device, data is copied from the user space to the kernel space. The boundary values of the data storage head positions on the memory of the copy source and the copy destination do not always match, and the unit of copying becomes small, and the overhead becomes large especially when the data length is long.

As described above, depending on the transfer data length, there is a possibility that either of the above processes becomes a bottleneck and the transfer performance cannot be improved as a series of processes.

An object of the present invention is to provide a method for optimizing the processing efficiency of the entire transfer processing without depending on the transfer data length.

[0007]

The above object is to operate under an operating system that provides two process execution modes, a user mode and a kernel mode, each of which has an independent operation space. In a computer system that transfers data to and from a device by the direct memory access method, the data existing in the operation space (user space) of a process operating in the user mode is transferred to the input / output device. In the process of copying to the operating space (kernel space) in kernel mode, if the data length to be copied is shorter than the length predetermined by the computer system, the storage start position of the transfer data in the kernel space is transferred to the input / output device. The means to set the most efficient value for transfer, and the transfer method in kernel space otherwise. It is achieved by providing a means for setting a storage start position of the data from the user space in the most efficient value in order to copy the kernel space.

Further, in the above computer system, in the process of copying data from the user space to the kernel space, when the data length to be copied is longer than the length predetermined by the computer system, the transfer data of the user space is stored. This is achieved by providing a means for matching the head position and the head position for storing the transfer data in the user space, and for copying the data in a transfer unit corresponding to the maximum data type according to the boundary value.

In the above computer system, in the process of adding control information for controlling transfer to the data copied from the user space to the kernel space and transferring the data to the input / output device, an area for storing the control information is set. This is achieved by providing means for securing the transfer data separately from the transfer data, associating with the data, and storing the control information from the most efficient position for transfer to the input / output device.

[0010]

According to the configuration of the computer system, the processing efficiency of the entire transfer process is optimized by adjusting the storage start position of the transfer data in the kernel space according to the user data length transferred to the input / output device. be able to.

Further, according to the computer system, even when the control information for controlling the data transfer is added, by setting this in another area from an appropriate position, the processing efficiency as a whole can be optimized. Can be converted.

[0012]

【Example】

System Configuration: FIG. 2 illustrates the configuration of a computer system according to an embodiment of the present invention, taking input / output with a communication line as an example. The computer system is first divided into a software level and a hardware level, and the software level is divided into a user level and a kernel level.

The user level is a level at which a program created by a user of the computer system operates, and there are a user program 3 and a system call library 4. At the user level, a program operates in an execution unit called a process that has an independent operation space (this is called user space).

The kernel level is a level at which a program for controlling the movement of the entire system such as an operating system operates, and has one independent operation space (this is called a kernel space). At the kernel level there are several modules: The system call interface 5 is a module for a user program to use the function of the kernel. Communication protocol processing unit 6
Manages and identifies the message format and device driver program on the communication line. The device driver 8 is a module that controls the operation of the input / output device 10 such as a communication line. The process control subsystem 7 performs process generation / execution, scheduling, memory management, and the like. The hardware control mechanism 9 executes communication with hardware and interrupt processing.

Further, in the present invention, it is premised that there are two modes, a user mode (corresponding to the user level) and a kernel mode (corresponding to the kernel level), as an operation mode of a process which is a program execution unit. And In the user mode, it is possible to access only the user space data that exists in the process unit, and it is possible to operate in parallel with other processes by time sharing. In the kernel mode, the space of both the kernel and the user can be accessed, it does not operate in parallel with other user mode and kernel mode processes, and it runs until the series of processing ends.

Flow of data transfer processing: Next, an example of the flow of processing in which the user program transfers data in the user space to the communication line will be described with reference to FIG.

(1) A user program issues a system call for requesting data transfer to the kernel ().

(2) Switch from the user mode to the kernel mode and call the corresponding kernel-level processing entry (system call interface). At this time, the processing at the kernel level operates as the process of the user program that issued the system call.

(3) The system call interface copies data to be transferred to the input / output device from the user space (). Next, after the communication protocol processing unit performs processing such as creating a message format for passing the transfer request (), the device driver and the hardware control mechanism start the operation of the input / output device (), and then the transfer operation. Process is suspended (wa
it) (). At this time, control passes to another process.

(4) When the transfer operation is completed, the input / output device issues an interrupt to the software. This interrupt is used to call the interrupt routine of the hardware control mechanism ().

(5) The interrupt routine activates and restarts the calling process (post) ().

(6) The kernel mode is switched to the user mode, and control is returned to the user program with the transfer result ().

Details of data transfer processing method: An example of processing for copying the transfer data (1, 2) from the user space to the kernel space will be described with reference to FIG. When the user level is switched to the kernel level in response to the system call issued by the user program, the system call interface unit copies the transfer data from the user space to the kernel space according to the following procedure.

First, the length of data to be copied is referred (this is recorded in the buffer management table 11 for managing the substance of the buffer 12 as shown in FIG. 4, for example). When this data length is shorter than a value predetermined by the computer system, the storage start position of the transfer data in the kernel space is set to the most efficient value for transfer to the input / output device.

For example, in the case where the input / output device transfers data on the memory to the inside of the input / output device by the direct memory access method (hereinafter abbreviated as DMA), the transfer in units of 32 bytes depends on the transfer capability of the input / output bus. It is most efficient, and for that purpose, it is necessary to set the storage start position of the transfer data on a 32-byte boundary. In this case, the storage start position of the transfer data in the kernel space is set to be on a 32-byte boundary.

On the other hand, when the data length is longer than the value predetermined by the computer system, the most efficient value for copying the storage start position of the transfer data in the kernel space from the user space to the kernel space. Set to.

For example, in copying data in the memory of a computer system, longword (4 byte unit), word (2
Byte unit), byte (1 byte) unit,
In copying in each data unit, it is assumed that the data storage start position of the copy source and the data storage start position of the copy destination need to be arranged at the respective boundaries. In this case, the most efficient data copy process can be performed by setting the transfer data storage start position in the kernel space so that the boundary between the copy source data storage start position and the copy destination data storage start position matches. realizable.

The reason for this is that, as shown in FIG. 5, if the data storage start position of the copy source and the data storage start position of the copy destination are the same, a longword unit is used except for half-length until reaching the longword boundary. This is because copying is possible.

Next, an example of "a value predetermined by the computer system" which is a judgment condition of the data copying process will be shown. First, the total transfer time is calculated when the data copy from the user space to the kernel space is optimized. User space transfer data storage start position is 13-byte boundary, transfer data length x bytes, data copy once or D
Unit prices required for MA transfer are 1 μs / byte and 10 respectively.
μs / byte. Further, the data copying and transfer processing shall be carried out by the above-mentioned method. At this time,
The total transfer time is as shown in the following formula 1.

[0030]

[Formula 1] {(x-1) / 4 + 1} + {13+ (x-16) / 32} × 10 (Formula 1) Next, total transfer when transfer to the input / output device is optimized under the same conditions When the time is obtained, it becomes the following formula 2.

[0031]

X + {x / 32 × 10} (Equation 2) In both (Equation 1) and (Equation 2), the first term is the time required to copy data from the user space to the kernel space, and the second term is input / output. This is the transfer time to the device. These are shown graphically in FIG.
In this example, it can be seen that 168 bytes is the boundary value.

Control information addition method: When data is transferred to the input / output control device, control information is generally added to the transfer data in many cases. For example, in data transmission / reception with a communication line, a protocol header is added to transfer data. Control information is added at the kernel level. FIG. 7 shows an example of a method of adding control information. In this example, in the case of data transmission / reception with a communication line, a buffer in another area is secured for transfer data and a protocol header is added. The storage header position of the protocol header is the most efficient position for transfer to the input / output device. In the above example, it is a 32-byte boundary.

Data input method from input / output control device: As an example of a method of inputting data from the input / output control device,
There are the following methods. The user program prepares an input buffer in the user space and then issues a system call. On the other hand, the kernel-level program prepares an input buffer in the kernel space in advance and issues an input request to the input / output control device.

At this time, the transfer data storage start position of the input buffer in the kernel space is set to the most efficient position for transfer from the input / output device as in the case of output. In the above example, it is a 32-byte boundary. On the other hand, the data input from the input / output control device exists in the kernel space and needs to be transferred to the user space. At this time, an efficient copy process can be performed by matching the boundary of the transfer data storage position of the buffer prepared by the user program with the boundary of the transfer data storage position of the kernel space (longword boundary in the above example). .

[0035]

According to the present invention, the storage efficiency of the entire transfer process is optimized by adjusting the storage start position of the transfer data in the kernel space according to the user data length transferred to the input / output device. You can

Further, even if control information for controlling data input / output is added, by setting this in another area from an appropriate position, the processing efficiency as a whole of the transfer processing can be optimized.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an input / output processing system according to an embodiment of the present invention.

FIG. 2 is a diagram showing a configuration of a computer system targeted by an input / output processing system according to an embodiment of the present invention.

FIG. 3 is a diagram showing a flow of input / output processing in the input / output processing method according to the embodiment of the present invention.

FIG. 4 is a diagram showing an example of a buffer structure in an input / output processing system according to an embodiment of the present invention.

FIG. 5 is a diagram showing a data copying method on a memory in an input / output processing method according to an embodiment of the present invention.

FIG. 6 is a diagram showing an example of input / output processing time in the input / output processing method according to the embodiment of the present invention.

FIG. 7 is a diagram showing a method of adding control information in the input / output processing system according to the embodiment of the present invention.

[Explanation of symbols]

1 ... Transfer data, 2 ... Transfer data, 3 ... User program, 4 ... System call library, 5 ... System call interface, 6 ... Communication protocol processing unit, 7 ... Process control subsystem, 8 ... Device driver, 9 ...
Hardware control mechanism, 10 ... I / O device, 11 ... Buffer management table, 12 ... Buffer.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Osamu Takada 1099, Ozenji, Aso-ku, Kawasaki-shi, Kanagawa Ltd. System Development Laboratory, Hitachi, Ltd.

Claims (3)

[Claims] 1. A direct memory access system that operates under an operating system that provides two process execution modes, a user mode and a kernel mode, each of which has an independent operating space, and a central processing unit and an input / output unit. In a computer system that transfers data in the above-mentioned manner, in the process of transferring the data existing in the operation space (user space) of the process operating in the user mode to the input / output device, the data is temporarily operated in the kernel mode for the transfer. When copying to the space (kernel space), if the data length is shorter than the length predetermined by the computer system, the storage data start position in the kernel space is set to the most efficient value for transfer to the I / O device. If not set, the storage data start position in the kernel space from the user space Output processing method and setting the most efficient value in order to replicate the space. 2. In the process of copying data from the user space to the kernel space, when the data length to be copied is longer than the length predetermined by the computer system, the storage start position of the transfer data in the user space and the user. 2. The input / output processing method according to claim 1, wherein the storage start positions of the transfer data in the spaces are made to coincide with each other, and the data is copied in a transfer unit corresponding to the maximum data type according to the boundary value. 3. In a process of adding control information for controlling transfer to data copied from a user space to a kernel space and transferring the data to an input / output device, an area for storing the control information is separated from the transfer data. 3. The input / output processing method according to claim 1, wherein the control information is secured and associated with the data, and the control information is stored from the most efficient position for transfer to the input / output device.