CN103391256B - A kind of base station user face data processing optimization method based on linux system - Google Patents

Abstract

The present invention proposes a kind of base station user face data processing optimization method based on linux system, comprise and adopt bag process accelerating module to complete bag classification, utilize and interrupt poll intelligence switching mode minimizing packet receiving interruption, kernel-space-user-space memory mapping technique is utilized to avoid memory copying, lock-free queue technology is utilized to reduce the contextual processing of kernel state and User space, thus the significant performance improving base station user face data processing.

Description

A Linux-based system user plane data processing optimization method for base stations

technical field

The invention relates to the technical field of wireless communication, in particular to a Linux system-based base station user plane data processing optimization method.

Background technique

With the advancement of wireless communication technology, the requirements for user plane data processing of LTE (3GPP Long Term Evolution) base stations, especially in the downlink direction, are getting higher and higher. LTE base station user plane data is mainly GTPU (Tunneling Protocol) service carried on UDP (User Datagram Protocol). The traditional processing method is to use the network card to receive the message, and the network card notifies the Linux kernel through an interrupt, and the Linux kernel passes through the network The layer-by-layer processing of the protocol stack, for GTPU business, includes the Ethernet layer, IP (Internetwork Interconnection Protocol) layer, UDP layer, Socket (network socket) layer, wakes up the user mode program and copies the GTPU message to the user space.

Drawbacks of traditional methods include:

Too many copies: Each GTPU packet needs to be copied from the kernel space to the user space after being processed by the Linux kernel.

Excessive context switching: The GTPU packet receiving socket adopts a blocking input method. Every time the receiving function is called to receive a UDP packet, there is a system call process, that is, the process of context switching between the kernel mode and the user mode.

Too many interruptions: In the case of high network load, the network card receives packets and receives too many interruptions. Too many interrupts will seriously reduce the real-time performance of the Linux system.

Contents of the invention

The present invention proposes a Linux system-based base station user plane data processing optimization method, and its purpose is to reduce context switching, interruption and copying in the process of LTE base station user plane data processing.

The technical solution of the present invention is a Linux-based base station user plane data processing optimization method, using a packet processing acceleration module to process user plane data, and the packet processing acceleration module includes a buffer management unit, a network data frame management unit and a queue Management unit, the network data frame management unit includes a packet classifier, including an initialization process and a data transmission process,

The initialization process includes the following sub-steps,

Step 1.1, defining the packet classification rules of the packet classifier of the packet processing acceleration module, the packet classification rules are the rules for classifying and distinguishing the user plane data of the base station, the user plane data is a GTPU message, and GTPU represents a tunneling protocol;

Step 1.2, establish the buffer of the buffer management unit, including when the kernel is initialized, reserve a memory block as the buffer of the buffer management unit, divide a memory block into multiple grids of equal size, and divide each grid The physical address and size are notified to the buffer management unit;

Step 1.3, in the device tree file of the kernel, associate the network data frame management unit port of the LTE base station connected to the core network with the buffer zone of the buffer management unit established in step 1.2;

Step 1.4, establish a ring buffer, including reserving a memory block when the driver is initialized, and dividing the memory block into a plurality of equal-sized grids for storing message descriptors, and the information in the message descriptors Comprise the address offset and the length of the data packet of GTPU message; Save the read pointer and the write pointer of control ring buffer in the head structure of memory block, described write pointer is the numbering of the corresponding grid that kernel fills in data to ring buffer, The read pointer is the serial number of the corresponding grid where the user state packet receiving process reads data from the ring buffer; the user state packet receiving process directly accesses the buffer management according to the information in the message descriptor read from the ring buffer The buffer of the unit, the data part of the read GTPU message is composed of a message and submitted to other business modules for processing;

Step 1.5, when the user state packet receiving process is initialized, the buffer of the buffer management unit established in step 1.2 and the physical address space where the ring buffer established in step 1.4 is located are mapped to user space;

The data transmission process includes the following sub-steps,

Step 2.1, the packet classifier of the packet processing acceleration module distinguishes the user plane data of the base station, stores the user plane data of the base station in the buffer of the buffer management unit, and enqueues the corresponding message descriptor to the network data frame management unit The corresponding queue generates an interrupt to notify the kernel to receive the packet;

Step 2.2, the kernel receives packets, which is completed by the packet receiving interrupt processing callback function of QMAN. The packet receiving interrupt processing callback function first closes the packet receiving interrupt, enters the polling state, and fills the information in the message descriptor into the ring buffer In the area, increment the write pointer of the ring buffer to wake up the process of receiving packets in the user mode; the packet receiving interrupt processing callback function counts the number of tunnel protocol data packets received in this polling in each polling, if If the number is less than the preset threshold, the polling state will end, and the packet receiving interrupt will be restarted; when the packet receiving is interrupted, the user state packet receiving process will sleep on the waiting queue defined by the ring buffer driver;

Step 2.3, after the user state packet receiving process sleeping on the waiting queue defined by the ring buffer driver is woken up, it directly accesses the buffer of the buffer management unit according to the information in the message descriptor, skips other information and sends the GTPU message The data part of the message is then composed and submitted to other business modules for processing.

Moreover, in step 2.3, the user state packet receiving process assembles the data part in the GTPU message into a message and submits it to other service modules by stripping off the Ethernet header, VLAN header, IP header and UDP header information.

Also, whether the grid is used or not is marked in each grid of the ring buffer.

Moreover, for GTPU messages that are not processed by the protocol stack, when the network port enters the promiscuous mode, a copy of the data packet is copied to the protocol stack, and a corresponding mark is made in the message to avoid repeated processing by the upper layer.

Moreover, in step 2.2, if the ring buffer is full, the newly received message descriptor is discarded, and the cells in the buffer of the buffer management unit corresponding to the message descriptor are released.

Moreover, in step 1.2, when establishing the buffer of the buffer management unit, more than one memory block is reserved as the buffer of the buffer management unit, and each memory block is divided into grids according to different sizes.

Compared with the traditional LTE user plane data processing technology, the present invention has the following innovations:

1. Use the packet classification technology to distinguish the LTE user plane data that can be omitted from the standard protocol stack processing, that is, the LTE user plane data that needs to be optimized, and the packets that need to be processed by the standard protocol stack, such as ICMP (Internet Control Message Protocol), ARP (Address Parsing protocol) and other messages, which are small in number and do not require high performance, are still submitted to the Linux protocol stack for processing, thus avoiding the "one size fits all" approach that general optimization technologies must implement the entire user mode protocol stack.

2. Using the interrupt polling intelligent switching technology, the number of hardware interrupts is greatly reduced, avoiding the real-time nature of real-time business tasks caused by too many hard interrupts; the interrupt polling intelligent switching technology intelligently switches between the interrupt mode and polling according to the rate of message entry Switching between modes, on the one hand, can ensure the processing efficiency when high-rate packets enter, and on the other hand, can ensure low delay when low-rate packets enter.

3. Use lock-free queue technology to complete the synchronization of single producer and single consumer, avoiding frequent system calls.

4. Use kernel-user space memory mapping technology to avoid memory copying.

Description of drawings

FIG. 1 is a structural diagram of a BMAN buffer in an embodiment of the present invention.

Fig. 2 is a schematic diagram of the operating principle of the ring buffer in the embodiment of the present invention.

FIG. 3 is a schematic diagram of an empty ring buffer in an embodiment of the present invention.

FIG. 4 is a schematic diagram of a ring buffer being full in an embodiment of the present invention.

FIG. 5 is an interaction diagram between a BMAN buffer and a ring buffer in an embodiment of the present invention.

Fig. 6 is a classification logic diagram of the packet classifier in the embodiment of the present invention.

detailed description

The present invention is mainly aimed at the optimization of user plane data processing of LTE base stations, and is applicable to but not limited to LTE base stations. This solution is also applicable to other systems that need to realize efficient user plane data processing in an embedded Linux system. This solution can meet the needs of high-speed data transmission in the construction of wireless communication base stations, and effectively reduce the occupation of system resources. This design makes full use of a series of cutting-edge technologies such as packet classification technology, interrupt polling intelligent switching technology, lock-free queue technology and memory mapping technology. , can effectively reduce the number of interrupts, reduce process context switching and completely avoid data copying.

The technical solution of the present invention will be described in detail below in conjunction with the drawings and embodiments.

The initialization process includes the following sub-steps in sequence:

Step 1.1, define the packet classification rules of the packet classifier of the packet processing acceleration module, and distinguish the user plane data of the base station (that is, the GTPU message, characterized by the UDP destination port number being 2152).

During specific implementation, this step can be implemented according to the coprocessor in the CPU used by the packet processing acceleration module. In the embodiment, Freescale PowerPC in the prior art is adopted, and Freescale PowerPC adopts DPAA (Data Path Acceleration Architecture) to realize the packet processing acceleration module, wherein a network data frame management unit, a buffer management unit, and a queue management unit are provided , generally referred to as FMAN, BMAN, and QMAN respectively, and a packet classifier is set in the network data frame management unit. Define the classification rules of Freescale PowerPC's packet classifier and configure them in the hardware.

The purpose of this step is to distinguish the packets that need to be optimized. Packets that need to be optimized will no longer be processed by the Linux network protocol stack. For LTE user plane data, such packets are GTPU packets (UDP destination port number is 2152); Packets are packets that need to be processed by the TCP/IP protocol stack, such as ICMP packets and ARP packets. The packet classification rules can be set according to the characteristics of the GTPU message that the UDP destination port number is 2152, and the subsequent data transmission process will enqueue the two types of messages into different queues according to the packet classification rules.

Step 1.2, establish the BMAN buffer, including dividing a memory block into multiple equal-sized grids by allocating kernel memory reserved memory blocks during kernel initialization, and notifying DPAA of the physical address and size of each grid The BMAN (buffer management) unit of the module. In order to meet the Ethernet standard MTU (Maximum Transmission Unit) requirements, the grid size is generally set to about 2K bytes.

In the embodiment, when the kernel is initialized, a memory block is reserved as a BMAN buffer. As shown in Figure 1, the memory block is divided into multiple cells of equal size, each with a size of 2112 bytes, and a 4 megabyte memory block can be divided into 1985 cells, as shown in the figure 0, 2112 , 2112*2...2112*1984, in the computer field, * is generally used to represent multiplication by ×. 2112 bytes is an odd multiple of 64 bytes. This size selection can accommodate a standard MTU of 1500 bytes, and allows the hardware to add some additional information, and can evenly use the two L3 caches of PowerPC4080. After the division is completed, the The physical address and size of each grid is notified to the BMAN unit of the DPAA module. In order to avoid wasting physical memory, multiple memory blocks can also be reserved, each memory block is used as a different BMAN buffer, and the cells in each memory block have different sizes, such as 64 bytes, 172 bytes, 320 bytes, etc. , the DPAA hardware module will intelligently select the most appropriate size according to the size of the network message frame, such as 173 bytes will select a 320-byte BMAN buffer, and 100 bytes will select a 172-byte BMAN buffer.

In step 1.3, the association between the FMAN (network data frame management unit) port and the BMAN buffer requires the FMAN port connecting the LTE base station to the core network and the BMAN buffer established in step 1.2 in the DTS (device tree) file of the kernel Areas are associated together. If multiple BMAN buffers are allocated, multiple BMAN buffers can also be bound to this port.

Step 1.4, establish a ring buffer, including when the driver is initialized, reserve a memory block by allocating kernel memory, divide the memory block into multiple grids of equal size, and use it to store the address offset of the data packet of the GTPU message and length, all grids share a header structure, and the header structure stores the read and write pointers that control the ring buffer to complete the synchronization between the kernel (producer) and the user state packet receiving process (consumer). The write pointer in the header structure represents the number of the grid where the producer fills in the data, and the read pointer represents the number of the grid where the consumer reads the data. During specific implementation, the size of the ring buffer is defined as a power of 2, which is beneficial to improve the efficiency of reading and writing pointer operations.

In the embodiment, the synchronous operation schematic diagram of the ring buffer is shown in FIG. 2 , and the traditional mutual exclusion lock structure is no longer needed by using the ring buffer. Initially, the ring buffer read pointer and write pointer are both zero. Every time the producer produces a piece of data, that is, after the kernel driver receives a GTPU message, it needs to increment the write pointer. When the increase of the write pointer exceeds the total number of grids in the ring buffer, it needs to take the remainder of the incremented write pointer by the total number of grids . For example, assume that the total number of grids in the ring buffer is 256, numbered from 0 to 255. The write pointer is 255 before incrementing, and it is 256 after incrementing. The remainder of 256 becomes 0, that is, the write pointer wraps around to 0 and points to the first grid in the ring buffer again. Express the above operation with NEXT, so NEXT(X)=(X+1)%N, where X is the write pointer, and N is the number of grids in the ring buffer. Similarly, when the consumer fetches a piece of data, that is, after the user-mode packet receiving process finishes processing a GTPU packet, it also needs to use the above method to rewrite the read pointer. As shown in Figure 3, when the read pointer is equal to the write pointer, it means that the data produced by the producer has been taken away by the consumer, and the producer has not yet produced new data. At this time, the user mode program needs to sleep to wait Producers produce more data. As shown in Figure 4, when the result of the write pointer is equal to the read pointer after the above NEXT operation, it means that the buffer is full, and the producer needs to actively discard subsequent received GTPU packets. The number of grids in the ring buffer represents The degree of difference in producer and consumer processing rates that the system can tolerate. The driver also needs to provide the implementation of the memory mapping function, as a bridge to complete the mapping between the BMAN buffer memory and the virtual address space of the user space.

The embodiment provides a user state packet receiving process, which is used to directly access the BMAN memory according to the information in the message descriptor read from the ring buffer, and recompose the GTPU message to other business modules for processing.

In step 1.5, when the user mode packet receiving process is initialized, the kernel memory allocated in step 1.2 and step 1.4 is mapped to the user space.

In the embodiment, when the user state packet receiving process is initialized, the physical address space where the BMAN buffer and the ring buffer established in step 1.4 are established in step 1.2 are mapped to user space, that is, a section of kernel space is mapped to a section of user space , so as to avoid the copying of network data from kernel mode to user mode.

The use process includes the following sub-steps in sequence:

Step 2.1, the DPAA packet classifier distinguishes the user plane data of the base station, stores it in the BMAN buffer of DPAA, and enqueues the corresponding message descriptor (describing the storage address and size of the message) to the QMAN of DPAA ( The corresponding queue in the queue management) unit, the QMAN unit generates an interrupt to notify the kernel to receive the packet.

In the embodiment, after the network interface of the LTE base station receives the message sent from the core network side, the hardware packet classifier in the DPAA module performs packet classification on the message according to the packet classification grammar defined in step 1.1 during initialization, and the UDP destination port The GTPU message descriptor with the number 2152 is enqueued to one queue, and all other message descriptors are enqueued to another queue. The specific implementation can use the preset queue number, which can be arbitrarily selected between 0x1 and 0xFFFF value. The packet classification logic is shown in Figure 6: the packet enters the packet classifier, and the packet classifier judges whether it is an IP fragment. If it is, it enters the queue at 0x2000; otherwise, it continues to judge whether the destination port is 2152. If it is 2152, it enters the queue at 0x2001 queue, otherwise it will still enter the queue at 0x2000, and the packet classification ends. The information recorded in the message descriptor includes the initial physical address of each cell in the BMAN buffer and the size of the message. After the hardware enqueues the message descriptor, a corresponding hardware interrupt will be generated.

In step 2.2, the kernel receives the packet, and the kernel receives the packet by the packet receiving interrupt processing callback function of QMAN to complete. When packet receiving is interrupted, the user mode packet receiving process sleeps on the waiting queue defined by the ring buffer driver. The content of the packet receiving interrupt processing callback function includes first closing the packet receiving interrupt, entering the polling mode, filling the information in the message descriptor into the ring buffer, incrementing the write pointer of the ring buffer, and waking up the user state to receive the packet process. In a polling, the packet receiving interrupt processing callback function will count the number of GTPU packets received in this polling, if the number is less than the preset threshold (the preset threshold can be pre-defined by those skilled in the art, such as Setting it to 64) will re-enable the interrupt, that is, the polling mode has returned to the interrupt mode. If the number of GTPU packets received in one poll is not less than the budget, the interrupt will not be re-opened, that is, it will remain in the polling state .

In the embodiment, the packet receiving by the kernel is completed by the packet receiving interrupt processing callback function of QMAN. First close the packet receiving interrupt and enter the polling state. Since the GTPU message and the non-GTPU message have been enqueued into different queues in step 2.1, different processing may be performed according to different queue numbers. If it is a GTPU message, fill in the information in the message descriptor into the ring buffer. The interaction between the BMAN buffer and the ring buffer is shown in Figure 5. In the figure, the kernel-user space mapped memory is allocated by the kernel. In the BMAN buffer, X represents the start address of this area, and X+2112, X+2112×2, etc. represent the start address of each grid in the buffer, and it can be seen that the length of each grid is 2112 bytes. The user state packet receiving process maps this segment address to its own virtual address space. When a GTPU message comes in, the DPAA hardware module will place the message in a cell in the BMAN buffer, assuming it is placed in the physical address space from X+2112×4 to X+2112×5, the kernel driver The address offset of this cell in the kernel-user space mapped memory is 2112×4 and the actual length of the GTPU message is filled in the ring buffer cell pointed to by the current write pointer in the ring buffer, and the write value of the ring buffer is incremented. Pointer, and then wake up the user mode receiving process waiting on the waiting queue defined by the ring buffer. When the kernel driver receives packets, if the ring buffer is full, the message descriptor is discarded, and the cell of the BMAN buffer memory corresponding to the message descriptor is released. After incrementing the write pointer, the BMAN buffer cell pointed to by the write pointer can actually be released, but because the hardware operation corresponding to the cell release in the BMAN buffer can release up to 8 BMAN buffer cells, and a hardware The operation call overhead is high, so in order to improve efficiency, the cell address pointed to by the write pointer can be cached, and after accumulating to 8 cells, the corresponding BMAN buffer memory will be released at a time. Since it is necessary to extract the cell address and size information stored in the ring buffer when releasing, it is necessary to set a variable to determine whether the cell has been used once, that is, the address of the BMAN buffer cell that has been filled once . You can mark whether the cell is used (key value) in each cell of the ring buffer to avoid freeing the buffer that has not been filled.

In addition, when debugging network problems, the most common method is to capture packets through TCPDUMP (network capture tool), and because GTPU packets bypass the Linux network protocol stack at this time, when the network interface enters the promiscuous mode, that is, when TCPDUMP is enabled, you can copy the GTPU message to sk_buff (the data structure describing the message in the Linux network protocol stack), and make a special mark in the sk_buff structure, and then submit it to the protocol stack. On the one hand, it satisfies For debugging needs, on the one hand, the special mark can prevent the protocol stack from repeatedly submitting such messages to the upper-layer user mode processing program. If it is not a GTPU message, it means that this type of message needs to be processed by the Linux network protocol stack. Since the network protocol stack must use the sk_buff structure, the callback function copies the message described in the message descriptor to the kernel memory. In the sk_buff dynamically allocated by the management allocator and its auxiliary structure, the call-related functions are handed over to the protocol stack. Since they have been copied, the BMAN buffer cells can be released immediately, and the memory related to sk_buff is handed over to the network protocol stack. freed.

In one round of polling, if the number of processed packets is less than the budget, it means that the network packets enter the system slowly at this time. It may be that there is no data in the BMAN buffer after the packet receiving callback has been processed many times, then Re-enable the packet receiving interrupt, that is, the polling mode returns to the interrupt mode. It can be seen that the packet receiving process can be intelligently switched between the interrupt and polling methods according to the speed of network packets entering the system.

Step 2.3, after the user-mode packet receiving process sleeping on the waiting queue defined by the ring buffer driver is awakened by step 2.2, it directly accesses the BMAN memory according to the information in the message descriptor, and transfers the Ethernet header, VLAN header, IP Information such as the header and UDP header is skipped, and the GTPU message is recomposed into a message and submitted to other business modules for processing.

After the user state packet receiving process is woken up, according to the packet address offset and packet size filled in the ring buffer cell pointed by the current read pointer in the ring buffer, you can access the corresponding data in the BMAN buffer without copying. The GTPU message has been sent. In the embodiment, after the user state packet receiving process starts, it will sleep on the waiting queue defined by the ring buffer driver. After being woken up by step 2.2, since the BMAN buffer has been mapped to its own user address space, it can be processed according to the message. The information in the descriptor (the descriptor contains information such as the physical address offset of the message and the size of the message) directly accesses the BMAN buffer. Depending on the networking model, such as whether there is a VLAN tag, the IP version is IPv4 or IPv6, The user program needs to skip the Ethernet header, VLAN header, IP header, UDP header and other information, and then recompose the GTPU message and submit it to other business modules for processing.

Due to the limitation of the packet classifier of PowerPC4080, when the GTPU message sent by the message sender is an IP fragment, only the first IP fragment contains the UDP header, and the classification rule of the classifier is actually to simply judge an IP The UDP destination port number of the message, so other IP fragments will be submitted to the Linux network protocol stack for processing, but because the first fragment is not submitted to the protocol stack, the IP reassembly of the protocol stack will fail. In implementation, in order to avoid such problems, such packets can be classified as non-GTPU packets, and the packet classifiers in higher-end CPUs of the PowerPC series do not have such problems.

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention belongs can make various modifications or supplements to the described specific embodiments or adopt similar methods to replace them, but they will not deviate from the spirit of the present invention or go beyond the definition of the appended claims range.

Claims (6)

1. the base station user face data processing optimization method based on linux system, bag process accelerating module is adopted to carry out user face data process, described bag process accelerating module comprises buffer management unit, network data frame administrative unit and queue management unit, network data frame administrative unit comprises packet classifier, it is characterized in that: comprise initialization procedure and data transmission procedure

Described initialization procedure comprises following sub-step,

Step 1.1, the bag classifying rules of the packet classifier of definition bag process accelerating module, described bag classifying rules is the rule that the user face data of base station is distinguished in classification, and described user face data is GTPU message, and GTPU represents tunnel protocol;

Step 1.2, set up the buffering area of buffer management unit, when being included in kernel initialization, reserved memory block is as the buffering area of buffer management unit, be multiple equal-sized grid by a memory block cutting, the physical address of each grid and size are informed to buffer management unit;

Step 1.3, in the device tree file of kernel, in the network data frame administrative unit port that LTE base station is connected with core net and step 1.2 set up buffer management unit buffering area be associated together;

Step 1.4, sets up buffer circle, when being included in driving initialization, a reserved memory block, be multiple equal-sized grid by memory block cutting, be used for stored messages descriptor, the information in described message descriptor comprises address offset and the length of the packet of GTPU message; The read pointer controlling buffer circle and write pointer is preserved in the header structure of memory block, described write pointer is kernel fills in numbering from the corresponding grid of data to buffer circle, and read pointer is User space packet receiving process reads the corresponding grid of data numbering from buffer circle; Described User space packet receiving process is the buffering area according to the information direct access buffer district administrative unit the message descriptor read from buffer circle, gives the process of other business module process by data division recomposition message delivery in the GTPU message of reading;

Step 1.5, during User space packet receiving process initialization, by step 1.2 set up the buffering area of buffer management unit and step 1.4 set up place, buffer circle physical address space be mapped to user's space;

Described data transmission procedure comprises following sub-step,

Step 2.1, the packet classifier of bag process accelerating module distinguishes base station user face data, by base station user face deposit data in the buffering area of buffer management unit, and the respective queue of network data frame administrative unit that corresponding message descriptor is joined the team to, produce the packet receiving of interrupt notification kernel;

Step 2.2, carry out kernel packet receiving, completed by the packet receiving interrupt processing call back function of QMAN, described QMAN is queue management unit, first described packet receiving interrupt processing call back function is closed packet receiving and is interrupted, and enters polling status, by the information solicitation in message descriptor in buffer circle, increase progressively the write pointer of buffer circle, wake User space packet receiving process up; Packet receiving interrupt processing call back function, in poll each time, is added up the message number of the tunneling protocol data received in this poll, if number is less than predetermined threshold value, is terminated polling status, reopens packet receiving and interrupts; When packet receiving is interrupted, the sleep of User space packet receiving process drives in buffer circle on the waiting list of definition;

Step 2.3, after the User space packet receiving process of sleeping on the waiting list that buffer circle drives definition is waken up, according to the buffering area of the direct access buffer district of the information in message descriptor administrative unit, skip over other information by the data division recomposition message delivery of GTPU message to other business module process.

2. according to claim 1 based on the base station user face data processing optimization method of linux system, it is characterized in that: in step 2.3, User space packet receiving process is by peelling off Ethernet head, VLAN head, IP head and UDP header information, by the data division in GTPU message, be assembled into message delivery to other business modules.

3. according to claim 1 based on the base station user face data processing optimization method of linux system, it is characterized in that: in each grid of buffer circle, whether grid is used and marked.

4. according to claim 1 based on the base station user face data processing optimization method of linux system, it is characterized in that: for the GTPU message without protocol stack process, when network interface enters promiscuous mode, copy a packet to protocol stack, and respective markers is done in message, avoid upper strata reprocessing.

5. according to claim 1 based on the base station user face data processing optimization method of linux system, it is characterized in that: in step 2.2, if buffer circle is full, then the message descriptor newly received is abandoned, and cell in the buffering area discharging buffer management unit corresponding to this message descriptor.

6. according to claim 1 or 2 or 3 or 4 or 5 based on the base station user face data processing optimization method of linux system, it is characterized in that: in step 1.2, when setting up the buffering area of buffer management unit, reserved more than one piece memory block is as the buffering area of buffer management unit, and each memory block is by different size cutting grid.