CN104239248B - PCI E multiple buffer DMA data transfer methods - Google Patents
PCI E multiple buffer DMA data transfer methods Download PDFInfo
- Publication number
- CN104239248B CN104239248B CN201410472222.3A CN201410472222A CN104239248B CN 104239248 B CN104239248 B CN 104239248B CN 201410472222 A CN201410472222 A CN 201410472222A CN 104239248 B CN104239248 B CN 104239248B
- Authority
- CN
- China
- Prior art keywords
- data
- memory
- purpose system
- source systems
- buffer circle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Bus Control (AREA)
Abstract
The invention provides a kind of PCI E multiple buffers DMA data transfer method, it is related to PCI E technical field of data transmission.This method each distributes N block physical memories in data source systems and data purpose system, sets up available memory pool and working memory pond by doubly linked list to the N blocks physical memory in two systems respectively；The internal memory for writing data is obtained from available memory pool, the internal memory for writing data adds working memory pond；The physical start address and size of the internal memory to be operated are write in buffer circle.The present invention reduces memory copying number of times and system call number by available memory pool and using for working memory pond；Increase the message transmission rate of DMA by buffer circle.The present invention reduces the wait expense of system under the transmission of big data quantity, improves efficiency of transmission.
Description
Technical field
The present invention relates to PCI-E technical field of data transmission, the DMA data transfer method of specifically a kind of multiple buffer.
Background technology
Traditional PCI-E data transmission methods are using DMA (Direct Memory Access, direct memory access).
Specific method is to be both needed to distribute one piece of continuous physical memory in data source and data purpose system, in data source system before starting
User application is responsible for, in data copy to physical memory, being responsible for data from number by dma controller during DMA transfer in system
Transmitted into the physical memory of data purpose system according to source physical memory, DMA terminate after in data purpose system by application journey
Sequence is copied to user's space and uses, and then former 2 pieces of physical memories (physical memory of data source systems and data purpose system) can
It is used as data transfer next time.Fig. 1 is traditional data transmission method schematic diagram：
(1) data source systems：User application is in physical memory in data copy to kernel spacing；
(2) DMA transfer；
(3) data purpose system：Data, the physical memory from kernel spacing is copied to user's space to user application
Use.
In data transmission procedure in the system of data source and data purpose, only one of which is for data transfer for the method
Buffering area, DMA next time cannot be again carried out before a DMA is completed, and reduce efficiency of transmission.
The content of the invention
Data transmission efficiency the invention aims to improve PCI-E, mitigates CPU pressure, proposes that a kind of PCI-E is more
Buffering area DMA data transfer method.
The PCI-E multiple buffer DMA data transfer methods that the present invention is provided, comprise the following steps：
Step one, the respective distribution N block physical memories in data source systems and data purpose system, respectively to two systems
In N blocks physical memory available memory pool and working memory pond are set up by doubly linked list；Meanwhile, all set up in two systems
More than one buffer circle and its corresponding read-write pointer；
Step 2, the user application in data source systems are obtained using API from the available memory pool of data source systems
Obtain internal memory and insert data, the internal memory that will insert data is added in the working memory pond of data source systems, will insert data
The physical start address and size of internal memory write the buffer circle of data source systems；
Step 3, the user application in data source systems initiate data transfer request using API；Data purpose system
Kernel in system chooses free memory from the available memory pool of data purpose system, and by the physical start address of selected internal memory and
Size is inserted in the buffer circle of data purpose system, is updated the data the write pointer of the buffer circle of purpose system and is notified
Data source systems；
After the notice that the buffer circle that kernel in step 4, data source systems receives data purpose system updates, take
Obtaining data purpose system will write the internal memory physical start address and size of data, initiate DMA and complete data transfer；
After the completion of step 5, data transfer, data source systems update the data the buffer circle read pointer of origin system and lead to
Primary data purpose system；
The working memory pond that the internal memory of write-in data is put into data purpose system is supplied user by step 6, data purpose system
Application program is used.
Multiple buffer DMA data transfer method of the invention, deletes unnecessary memory copying, dramatically reduces
The pressure of CPU；The wait expense of system is reduced under the transmission of big data quantity, efficiency of transmission is improve.
Brief description of the drawings
Fig. 1 is traditional data transmission method schematic diagram；
Fig. 2 is PCI-E multiple buffers data transmission method schematic diagram of the invention；
Specific embodiment
Below in conjunction with drawings and Examples, the present invention is described in further detail.
Data transmission efficiency the invention aims to improve PCI-E, mitigates CPU pressure, proposes a kind of zero-copy
Data transmission method.
As shown in Fig. 2 the PCI-E multiple buffer DMA data transfer methods that the present invention is provided specifically include following steps：
Step one, the appropriately sized physical memory of respective distribution N blocks in data source systems and data purpose system, respectively
Available memory pool and working memory pond are set up to the N block physical memories in two systems.All set up single in two systems simultaneously
The read-write pointer of individual or multiple buffer circles and buffer circle.
The size of N as needed or practice, be set by the user.Available memory pool and working memory pond use doubly linked list
To represent.Data source systems and data purpose system are respectively to that should have an available memory pool and a working memory pond.From sky
Not busy memory pool obtains the internal memory for writing data, and the internal memory for writing data adds working memory pond.Write in buffer circle
Enter the physical start address and size of the internal memory to be operated.
As shown in Fig. 2 the buffer circle set up in data source systems and data purpose system, including at least two slow
Rush area.
Step 2, the user application in data source systems, use API (Application Programming
Interface, application programming interface) insert data from after the available memory pool acquirement memory headroom of data source systems.Fill out
The internal memory for entering data is added to the working memory pond of data source systems.Meanwhile, the physical start address of the internal memory of data will be inserted
And size writes the buffer circle of data source systems.
Step 3, the user application in data source systems initiate data transfer request using API.Data purpose system
After the physical memory block that kernel in system selects the free time from the available memory pool of data purpose system, in the ring of data purpose system
Physical start address and the size notice of selected internal memory are inserted in shape buffering area, and the buffer circle of data purpose system
Write pointer updates and notifies data source systems.
After the notice that the buffer circle that kernel in step 4, data source systems receives data purpose system updates, take
To write in data purpose system data internal memory physical start address and size, initiate DMA and complete data transfer.
Kernel in data source systems is according to the physical start address of internal memory in the buffer circle of data source systems and big
It is small, data are read from corresponding physical memory, and be transferred to data purpose system.The data that data purpose system will be received
The physical start address of internal memory in buffer circle according to data purpose system, corresponding physics in write-in data purpose system
In internal memory.
After the completion of step 5, data transfer, data source systems update buffer circle read pointer and notify data purpose system
System.
The memory block for having obtained data is put into work pool and is used for user program by step 6, purpose system.
In the inventive method, realize the data directly by user application in data source systems and be transferred to data mesh
System in used, reduce data copy number of times and system call number, meanwhile, be increased using buffer circle
The speed and size of DMA transfer data.PCI-E data transfers are made using the inventive method, with the existing biography in background technology
The contrast effect of system data transmission scheme is as shown in table 1.Test uses+1 piece of PCI-E accelerator card of 1 DELL2950 server, first
The data transfer of PC to accelerator card is carried out in units of 2K/8K/16K afterwards, respectively carry out it is average after 5 tests obtained by transfer rate.
The contrast table of the Zero-copy data transmission plan of the present invention of table 1 and traditional scheme
As shown in table 1, compared with scheme is spread out of, there is transmission rate higher using the inventive method, is more applicable for
The transmission of big data quantity, can effectively reduce the wait expense of system.
Claims (1)
1. a kind of PCI-E multiple buffers DMA data transfer method, it is characterised in that comprise the following steps：
Step one, the respective distribution N block physical memories in data source systems and data purpose system, respectively in two systems
N blocks physical memory sets up available memory pool and working memory pond by doubly linked list；Meanwhile, all set up multiple in two systems
Buffer circle and its corresponding read-write pointer, write the physical start address of the internal memory to be operated in buffer circle
And size；For PCI-E interface transmits data between described data source systems and data purpose system；
Step 2, the user application in data source systems obtain interior using API from the available memory pool of data source systems
Data are deposited and insert, the internal memory that will insert data is added in the working memory pond of data source systems, will insert the internal memory of data
Physical start address and size write data source systems buffer circle；
Step 3, the user application in data source systems initiate data transfer request using API；In data purpose system
Kernel choose free memory from the available memory pool of data purpose system, and by the physical start address and size of selected internal memory
Insert in the buffer circle of data purpose system, update the data the write pointer of the buffer circle of purpose system and notify data
Origin system；
After the notice that the buffer circle that kernel in step 4, data source systems receives data purpose system updates, number is obtained
The internal memory physical start address and size of data are write according to purpose system, DMA is initiated and is completed data transfer；
After the completion of step 5, data transfer, data source systems update the data the buffer circle read pointer of origin system and notify number
According to purpose system；
Step 6, data purpose system are applied the working memory pond that the internal memory of write-in data is put into data purpose system for user
Program is used.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410472222.3A CN104239248B (en) | 2014-09-16 | 2014-09-16 | PCI E multiple buffer DMA data transfer methods |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410472222.3A CN104239248B (en) | 2014-09-16 | 2014-09-16 | PCI E multiple buffer DMA data transfer methods |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104239248A CN104239248A (en) | 2014-12-24 |
| CN104239248B true CN104239248B (en) | 2017-06-06 |
Family
ID=52227355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410472222.3A Active CN104239248B (en) | 2014-09-16 | 2014-09-16 | PCI E multiple buffer DMA data transfer methods |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104239248B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101438B (en) * | 2018-07-25 | 2020-07-28 | 百度在线网络技术（北京）有限公司 | Method and apparatus for storing data |
| CN110046114B (en) * | 2019-03-06 | 2020-08-14 | 上海熠知电子科技有限公司 | DMA controller based on PCIE protocol and DMA data transmission method |
| CN112286688B (en) * | 2020-11-05 | 2024-01-05 | 北京深维科技有限公司 | Memory management and use method, device, equipment and medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467473A (en) * | 2010-11-03 | 2012-05-23 | Tcl集团股份有限公司 | Method and device for transmitting data between user space and kernel |
| CN103645994A (en) * | 2013-11-05 | 2014-03-19 | 华为技术有限公司 | Data processing method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7673105B2 (en) * | 2005-06-27 | 2010-03-02 | Ab Inition Technology LLC | Managing memory pages |
-
2014
- 2014-09-16 CN CN201410472222.3A patent/CN104239248B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102467473A (en) * | 2010-11-03 | 2012-05-23 | Tcl集团股份有限公司 | Method and device for transmitting data between user space and kernel |
| CN103645994A (en) * | 2013-11-05 | 2014-03-19 | 华为技术有限公司 | Data processing method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104239248A (en) | 2014-12-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9760486B2 (en) | Accelerating cache state transfer on a directory-based multicore architecture | |
| US9934160B1 (en) | Bit-mapped DMA and IOC transfer with dependency table comprising plurality of index fields in the cache for DMA transfer | |
| CN104239249B (en) | PCI E zero-copy DMA data transfer methods | |
| JP6009692B2 (en) | Multi-mode memory access technique for graphics processing unit based memory transfer operations | |
| CN109564545B (en) | Method and apparatus for compressing addresses | |
| CN102467473B (en) | Method and device for transmitting data between user space and kernel | |
| US10055150B1 (en) | Writing volatile scattered memory metadata to flash device | |
| CN104239248B (en) | PCI E multiple buffer DMA data transfer methods | |
| DE102011101202A1 (en) | Non-volatile storage for graphics hardware | |
| CN106339061A (en) | Mobile terminal and method of run application thereof | |
| US9881680B2 (en) | Multi-host power controller (MHPC) of a flash-memory-based storage device | |
| US20220159286A1 (en) | Low-latency consumption of an encoded video bitstream | |
| CN102314400B (en) | Method and device for dispersing converged DMA (Direct Memory Access) | |
| CN106775477B (en) | SSD (solid State disk) master control data transmission management device and method | |
| WO2018119738A1 (en) | Speculative read mechanism for distributed storage system | |
| CN105450986B (en) | Video processing method under Linux | |
| CN102291298A (en) | Efficient computer network communication method oriented to long message | |
| TW201640487A (en) | Display controller for reducing display noise and system including the same | |
| US9904622B2 (en) | Control method for non-volatile memory and associated computer system | |
| TW201351276A (en) | Scheduling and execution of compute tasks | |
| CN105843631B (en) | A method of improving system firmware burn writing speed | |
| CN107179929B (en) | WRITE _ SAME function optimization implementation method and device | |
| US9652560B1 (en) | Non-blocking memory management unit | |
| WO2019114044A1 (en) | Image processing method and device, electronic apparatus, and computer readable storage medium | |
| CN106155626A (en) | One Android Sparse form image download method fast and effectively |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |