4.6. ENET

4.6.1. Introduction

Enet LLD is an unified Ethernet driver that support Ethernet peripherals found in TI SoCs, such as CPSW and ICSSG. Enet LLD supports CPSW_2G and CPSW_9G in J721E devices.

The diagram below shows the overall software architecture of the Enet low-level driver. A top-level driver layer provides the interface that the applications can use to configure the switch and to send/receive Ethernet frames.

For instance, the CPSW support in the Enet driver consists of several software submodules that mirror those of the CPSW hardware, like DMA, ALE, MAC port, host port, MDIO, etc. Additionally, the Enet driver also includes PHY driver support as well as a resource manager to administrate the CPSW resources.

Enet LLD relies on other PDK drivers like UDMA for data transfer to/from the Ethernet peripheral’s host port to the other processing cores inside the TI SoC devices. For the lower level access to the hardware registers, Enet LLD relies on the Chip Support Library (CSL).

../_images/EnetLLD_Diagram.png

4.6.2. User Interface

For details about individual fields of this library structure, see the PDK doxygen documentation

4.6.2.1. APIs

The Enet LLD APIs can be broadly divided into two categories: control and data path. The control APIs can be used to configure all Ethernet hardware submodules like FDB, MAC port, host port, MDIO, statistics, as well as PHY drivers and resource management. The data path APIs are exclusive for the DMA-based data transfers between the TI SoC processing cores and the Ethernet peripheral.

API reference for application:

#include <ti/drv/enet/enet.h>

The main APIs of the Enet LLD are the following:

  • Enet_open()
  • Enet_close()
  • Enet_ioctl()
  • Enet_poll()
  • Enet_periodicTick()

4.6.2.1.1. Data Path APIs

The main Enet LLD functions used to send and receive packets are:

  • EnetDma_openRxCh()
  • EnetDma_closeRxCh()
  • EnetDma_openTxCh()
  • EnetDma_closeTxCh()
  • EnetDma_retrieveRxPktQ()
  • EnetDma_submitRxPktQ()
  • EnetDma_retrieveTxPktQ()
  • EnetDma_submitTxPktQ()

It’s worth noting that the control path APIs are mainly IOCTL-based, and the data path APIs are direct functions in order to avoid any additional overhead associated with IOCTL calls as DMA data operations occur highly frequently.

4.6.2.1.2. Scatter-Gather

Starting in SDK 8.5, Enet LLD provides support for UDMA scatter-gather feature for packet transmission only. Scatter-gather is currently not supported for packet reception.

There are Enet LLD API changes introduced for scatter-gather which break compatibility with previous SDKs. The main changes in the parameters used by the application to pass the Ethernet frame buffer and the buffer length to the driver.

For the sake of comparison with previous API, let’s consider the case of a single continuous buffer (i.e. number of scatter segments of 1).

Parameter SDK 8.4 or older SDK 8.5+
Buffer pointer EnetDma_PktInfo::bufPtr EnetDma_PktInfo::sgList.list[0].bufPtr
Original buffer length EnetDma_PktInfo::orgBufLen EnetDma_PktInfo::sgList.list[0].segmentAllocLen
Filled buffer length EnetDma_PktInfo::userBufLen EnetDma_PktInfo::sgList.list[0].segmentFilledLen

The application can pass up to four segments for packet transmission as defined by ENET_UDMA_CPSW_MAX_SG_LIST. It’s worth noting that Enet LLD uses the same EnetDma_PktInfo type for packet reception but only a single segment is enforced as scatter-gather for packet reception is currently not enabled.

The scatter-gather list information is provided by the application to the driver via EnetUdma_PktInfo::sgList parameter of the packet info structure. Application must set the number of segments (EnetUdma_PktInfo::sgList.numScatterSegments), and the buffer pointer and length of each segment in EnetUdma_PktInfo::sgList.list[] array.

The relevant structures are shown below:

/*! Scatter gather list entry */
typedef struct EnetUdma_SGListEntry_s
{
    /*! Pointer to scatter fragment */
    uint8_t *bufPtr;
    /*! Length of valid data in the scatter fragment */
    uint32_t segmentFilledLen;
    /*! Length of allocated buffer for scatter fragment */
    uint32_t segmentAllocLen;
} EnetUdma_SGListEntry;

/*! Packet scatter list info */
typedef struct EnetUdma_SGList_s
{
    /*! Number of valid scatter segments in the packet */
    uint32_t numScatterSegments;
    /*! Array of scatterList having info on each individual scatter segement */
    EnetUdma_SGListEntry list[ENET_UDMA_CPSW_MAX_SG_LIST];
} EnetUdma_SGList;

/*! Packet data structure */
typedef struct EnetUdma_PktInfo_s
{
    ...
    /*! Scatter Gather list information for packets to be transmitted. */
    EnetUdma_SGList sgList;
} EnetUdma_PktInfo;

/*! Opaque handle that represents a DMA packet */
typedef struct EnetUdma_PktInfo_s EnetDma_Pkt;

Scatter-gather for packet transmission is also enabled in Enet LLD integration with lwIP stack. Note that LWIP_NETIF_TX_SINGLE_PBUF must be disabled for lwIP to pass multiple segments to the adaptation layer and to the Enet driver.

Scatter-gather feature is enabled by default in Enet LLD loopback test and lwIP example applications.

4.6.2.1.3. IOCTL Interface

IOCTLs are system calls that take an argument specifying the command code and can take none or additional parameters via Enet_IoctlPrms argument. IOCTL are used by all Enet submodules except for DMA.

The Enet_IoctlPrms parameter structure consists of input and output argument pointers and their corresponding size. The following helper macros are provided to help construct the IOCTL params:

  • ENET_IOCTL_SET_NO_ARGS(prms). Used for IOCTL commands that take no parameters.
  • ENET_IOCTL_SET_IN_ARGS(prms, in). Used for IOCTL commands that take input parameters but don’t output any parameter.
  • ENET_IOCTL_SET_OUT_ARGS(prms, out). Used for IOCTL commands that don’t take input parameters but return output parameters.
  • ENET_IOCTL_SET_INOUT_ARGS(prms, in, out). Used for IOCTL commands that take input parameters and also return output parameters.

where prms in a pointer to Enet_IoctlPrms variable, in is the pointer to IOCTL input argument and out is the pointer to IOCTL output argument.

It’s recommended that the application doesn’t set the Enet_IoctlPrms members individually, but only through the helper macros listed above.

Please refer to the individual IOCTL command to find out if it requires input and/or output parameters.

4.6.3. lwIP Integration

See lwIP User’s Guide for further information about lwIP integration into PDK using Enet LLD, and NDK-to-lwIP migration guide.

4.6.4. Application

4.6.4.1. Examples

Enet LLD comes with a set of examples demonstrating the usage of driver APIs. The examples are:

  • enet_loopback: Internal (MAC port) or external loopback test.
  • enet_lwip_example: TCP/IP stack integration using lwIP.
Name Description Expected Results Cores Supported Peripherals
Loopback Enet Loopback example demonstrates basic packet send and receive on an Ethernet peripheral configured in MAC loopback or PHY loopback. All packets sent from the example application shall be received back after being looped in MAC or PHY. mcu1_0 CPSW_2G
mcu2_0 CPSW_9G
mcu2_1 CPSW_2G
mpu1_0 CPSW_2G
lwIP Enet lwIP example demonstrates Enet driver integration with open source lwIP TCP/IP stack. The example enables DHCP client, it can get an IP address when connected to a network. lwIP example application shall be able to get an IP address when connected to a network. User can test ‘ping’, ‘echo’ and ‘iperf’. mcu1_0 CPSW_2G
mcu2_0 CPSW_9G
mcu2_1 CPSW_2G
mpu1_0 CPSW_2G

4.6.4.1.1. Enet loopback

This example exercises the MAC loopback functionality of the hardware. The example is developed and tested on both bare metal and TI RTOS code base. The Ethernet peripheral is opened with default initialization parameters and the MAC loopback is enabled.

A Tx channel and a Rx flow are opened to enable data transfers. Packets are transmitted from the Switch R5F (Main R5F0_0) to the host port using the Tx channel. These packets are routed back to the host port by the switch hardware as the internal loopback feature is enabled. These packets are then transmitted to the Switch R5F by the Rx flow and the application is notified.

The Tx and Rx functions in the example are set to transmit and receive 10000 packets. After reaching the count of 10000, the application closes the Tx channel, Rx flow, peripheral (i.e. CPSW) and restarts the application for a configurable number of times. Restarting the loopback test application ensures that there aren’t any memory leaks, and the hardware is closed properly and can be reopened any time.

4.6.4.1.2. Enet LWIP

Building LWIP example

  • Enabling specific features in the test

    To configure what gets built in the example, navigate to the example directory pdk/packages/ti/drv/enet/examples/enet_lwip_example and find the lwipcfg.h file. Inside, change the macros corresponding to the program that you wish to build to 1.

    For example, if udpecho is required, set:

    #define LWIP_UDPECHO_APP        1
    

    Note

    The heap memory may need to be increased in the FreeRTOSConfig.h file to accomodate multiple apps.

  • To build the lwip example, issue the following command:

    make -s enet_lwip_example_freertos
    

Running LWIP example

  • Load and run the enet_lwip_example image from the corresponding binaries folder.

  • The setup requires a LAN cable to be connected between the device and a multiport-router for the link status to be UP and aquire an IP address.

  • Connect the LAN cable between the same router and Laptop/PC from where the ping/echo tests are run.

  • Wait until the local interface IP is assigned and printed on the console like below.

    Starting lwIP, local interface IP is dhcp-enabled
    CPSW_2G Test on MCU NAVSS
    EnetPhy_bindDriver: PHY 0: OUI:080028 Model:23 Ver:01 <-> 'dp83867' : OK
    PHY 0 is alive
    Host MAC address: 70:ff:76:01:02:03
    [LWIPIF_LWIP] CPSW has been started successfully
    [LWIPIF_LWIP] NETIF INIT SUCCESS
    status_callback==UP, local interface IP is 0.0.0.0
    Cpsw_handleLinkUp: Port 1: Link up: 100-Mbps Full-Duplex
    MAC Port 1: link up
    link_callback==UP
    status_callback==UP, local interface IP is 192.168.0.9
    Initializing apps
    UDP server listening on port 5001
    

Testing LWIP example

Prerequisites: LAN cable setup and device link is UP and Local interface IP is successfully aquired.

  • Ping test

    ping -t <ip_addr>
    ping -t 192.168.0.9
    
  • UDP/ TCP echo test

    Download echotool from this website and execute the commands to see the response.

    • UDP echo command:

      echotool.exe <ip_addr> /p udp /r 7 /n 0
      echotool.exe 192.168.0.9 /p udp /r 7 /n 0
      
    • TCP echo command:

      echotool.exe <ip_addr> /p tcp /r 7 /n 0
      echotool.exe 192.168.0.9 /p tcp /r 7 /n 0
      
  • iperf test

    Download iperf2 from iperf website and execute the commands to see the response.

    • iperf TCP test command:

      iperf.exe -c <ip_addr> -r
      iperf.exe -c 192.168.0.9 -r
      
    • iperf UDP test command:

      iperf.exe -c <ip_addr> -r -u
      iperf.exe -c 192.168.0.9 -r -u
      

Static IP

The Enet lwIP example uses DHCP by default. If static IP needs to be tested, the following two config options need to be disabled in examples/enet_lwip_example/lwipcfg.h.

#define USE_DHCP    0
#define USE_AUTOIP  0

The IP address, gateway and netmask can be set in the same lwipcfg.h file.

#define LWIP_PORT_INIT_IPADDR(addr)   IP4_ADDR((addr), 192,168,1,200)
#define LWIP_PORT_INIT_GW(addr)       IP4_ADDR((addr), 192,168,1,1)
#define LWIP_PORT_INIT_NETMASK(addr)  IP4_ADDR((addr), 255,255,255,0)

4.6.4.1.3. Enet Multiport

Not supported for this SoC.

4.6.4.1.4. Enet TAS

Not supported for this SoC.