Ethernet is an asynchronous Carrier Sense Multiple Access with Collision Detect (CSMA/CD) protocol/interface. It is generally not well suited for low-power applications. However, with ubiquitous deployment, internet connectivity, high data rates, and limitless-range expandability, Ethernet can accommodate nearly all wired communications.
Normal IEEE 802.3 compliant Ethernet frames are between 64 and 1518 bytes in length. They are made up of five or six different fields: a destination MAC address (DA), a source MAC address (SA), a type/length field, a data payload, an optional padding field and a Cyclic Redundancy Check (CRC). Additionally, when transmitted on the Ethernet medium, a 7-byte preamble field and Start-of-Frame (SOF) delimiter byte are appended to the beginning of the Ethernet packet.
The Start-of-Frame Delimiter (SFD) is a binary sequence ``10101011`` (as seen on the physical medium). It is sometimes considered to be part of the preamble.
The destination address field contains a 6-byte length MAC address of the device that the packet is directed to. If the Least Significant bit in the first byte of the MAC address is set, the address is a multicast destination. For example, 01-00-00-00-F0-00 and 33-45-67-89-AB-CD are multi-cast addresses, while 00-00-00-00-F0-00 and 32-45-67-89-AB-CD are not.
Packets with multi-cast destination addresses are designed to arrive and be important to a selected group of Ethernet nodes. If the destination address field is the reserved multicast address, i.e. FF-FF-FF-FF-FF-FF, the packet is a broadcast packet and it will be directed to everyone sharing the network. If the Least Significant bit in the first byte of the MAC address is clear, the address is a unicast address and will be designed for usage by only the addressed node.
Normally the EMAC controller incorporates receive filters which can be used to discard or accept packets with multi-cast, broadcast and/or unicast destination addresses. When transmitting packets, the host controller is responsible for writing the desired destination address into the transmit buffer.
The source address field contains a 6-byte length MAC address of the node which created the Ethernet packet. Users of Ethernet must generate a unique MAC address for each controller used. MAC addresses consist of two portions. The first three bytes are known as the Organizationally Unique Identifier (OUI). OUIs are distributed by the IEEE. The last three bytes are address bytes at the discretion of the company that purchased the OUI. For more information about MAC Address used in ESP-IDF, please see :ref:`MAC Address Allocation <MAC-Address-Allocation>`.
The type/length field is a 2-byte field. If the value in this field is <= 1500 (decimal), it is considered a length field and it specifies the amount of non-padding data which follows in the data field. If the value is >= 1536, it represents the protocol the following packet data belongs to. The followings are the most common type values:
Users implementing proprietary networks may choose to treat this field as a length field, while applications implementing protocols such as the Internet Protocol (IP) or Address Resolution Protocol (ARP), should program this field with the appropriate type defined by the protocol’s specification when transmitting packets.
Payload
^^^^^^^
The payload field is a variable length field, anywhere from 0 to 1500 bytes. Larger data packets will violate Ethernet standards and will be dropped by most Ethernet nodes.
The padding field is a variable length field added to meet the IEEE 802.3 specification requirements when small data payloads are used.
The DA, SA, type, payload, and padding of an Ethernet packet must be no smaller than 60 bytes in total. If the required 4-byte FCS field is added, packets must be no smaller than 64 bytes. If the payload field is less than 46-byte long, a padding field is required.
The FCS field is a 4-byte field that contains an industry-standard 32-bit CRC calculated with the data from the DA, SA, type, payload, and padding fields. Given the complexity of calculating a CRC, the hardware normally will automatically generate a valid CRC and transmit it. Otherwise, the host controller must generate the CRC and place it in the transmit buffer.
Normally, the host controller does not need to concern itself with padding and the CRC which the hardware EMAC will also be able to automatically generate when transmitting and verify when receiving. However, the padding and CRC fields will be written into the receive buffer when packets arrive, so they may be evaluated by the host controller if needed.
Besides the basic data frame described above, there're two other common frame types in 10/100 Mbps Ethernet: control frames and VLAN-tagged frames. They're not supported in ESP-IDF.
The communication between MAC and PHY can have diverse choices: **MII** (Media Independent Interface), **RMII** (Reduced Media Independent Interface), etc.
One of the obvious differences between MII and RMII is signal consumption. MII usually costs up to 18 signals, while the RMII interface can reduce the consumption to 9.
In RMII mode, both the receiver and transmitter signals are referenced to the ``REF_CLK``. **REF_CLK must be stable during any access to PHY and MAC**. Generally, there are three ways to generate the ``REF_CLK`` depending on the PHY device in your design:
* Some PHY chips can derive the ``REF_CLK`` from its externally connected 25 MHz crystal oscillator (as seen the option *a* in the picture). In this case, you should select ``CONFIG_ETH_RMII_CLK_INPUT`` in :ref:`CONFIG_ETH_RMII_CLK_MODE`.
* Some PHY chip uses an externally connected 50MHz crystal oscillator or other clock sources, which can also be used as the ``REF_CLK`` for the MAC side (as seen the option *b* in the picture). In this case, you still need to select ``CONFIG_ETH_RMII_CLK_INPUT`` in :ref:`CONFIG_ETH_RMII_CLK_MODE`.
* Some EMAC controllers can generate the ``REF_CLK`` using an internal high-precision PLL (as seen the option *c* in the picture). In this case, you should select ``CONFIG_ETH_RMII_CLK_OUTPUT`` in :ref:`CONFIG_ETH_RMII_CLK_MODE`.
``REF_CLK`` is configured via Project Configuration as described above by default. However, it can be overwritten from user application code by appropriately setting :cpp:member:`eth_esp32_emac_config_t::interface` and :cpp:member:`eth_esp32_emac_config_t::clock_config` members. See :cpp:enum:`emac_rmii_clock_mode_t` and :cpp:enum:`emac_rmii_clock_gpio_t` for more details.
If the RMII clock mode is selected to ``CONFIG_ETH_RMII_CLK_OUTPUT``, then ``GPIO0`` can be used to output the ``REF_CLK`` signal. See :ref:`CONFIG_ETH_RMII_CLK_OUTPUT_GPIO0` for more information.
What's more, if you're not using PSRAM in your design, GPIO16 and GPIO17 are also available to output the reference clock. See :ref:`CONFIG_ETH_RMII_CLK_OUT_GPIO` for more information.
If the RMII clock mode is selected to ``CONFIG_ETH_RMII_CLK_INPUT``, then ``GPIO0`` is the only choice to input the ``REF_CLK`` signal. Please note that ``GPIO0`` is also an important strapping GPIO on ESP32. If GPIO0 samples a low level during power-up, ESP32 will go into download mode. The system will get halted until a manually reset. The workaround for this issue is disabling the ``REF_CLK`` in hardware by default so that the strapping pin won't be interfered by other signals in the boot stage. Then, re-enable the ``REF_CLK`` in the Ethernet driver installation stage.
* Force the PHY device to reset status (as the case *a* in the picture). **This could fail for some PHY device** (i.e. it still outputs signals to GPIO0 even in reset state).
**No matter which RMII clock mode you select, you really need to take care of the signal integrity of REF_CLK in your hardware design!** Keep the trace as short as possible. Keep it away from RF devices and inductor elements.
Signals used in the data plane are fixed to specific GPIOs via MUX, they can't be modified to other GPIOs. Signals used in the control plane can be routed to any free GPIOs via Matrix. Please refer to :doc:`ESP32-Ethernet-Kit <../../hw-reference/esp32/get-started-ethernet-kit>` for hardware design example.
You need to set up the necessary parameters for MAC and PHY respectively based on your Ethernet board design, and then combine the two together to complete the driver installation.
*:cpp:member:`eth_mac_config_t::sw_reset_timeout_ms`: software reset timeout value, in milliseconds. Typically, MAC reset should be finished within 100 ms.
*:cpp:member:`eth_mac_config_t::rx_task_stack_size` and :cpp:member:`eth_mac_config_t::rx_task_prio`: the MAC driver creates a dedicated task to process incoming packets. These two parameters are used to set the stack size and priority of the task.
*:cpp:member:`eth_mac_config_t::flags`: specifying extra features that the MAC driver should have, it could be useful in some special situations. The value of this field can be OR'd with macros prefixed with ``ETH_MAC_FLAG_``. For example, if the MAC driver should work when the cache is disabled, then you should configure this field with :c:macro:`ETH_MAC_FLAG_WORK_WITH_CACHE_DISABLE`.
:SOC_EMAC_SUPPORTED:* :cpp:member:`eth_esp32_emac_config_t::smi_mdc_gpio_num` and :cpp:member:`eth_esp32_emac_config_t::smi_mdio_gpio_num`: the GPIO number used to connect the SMI signals.
:SOC_EMAC_SUPPORTED:* :cpp:member:`eth_esp32_emac_config_t::clock_config`: configuration of EMAC Interface clock (``REF_CLK`` mode and GPIO number in case of RMII).
*:cpp:member:`eth_phy_config_t::phy_addr`: multiple PHY devices can share the same SMI bus, so each PHY needs a unique address. Usually, this address is configured during hardware design by pulling up/down some PHY strapping pins. You can set the value from 0 to 15 based on your Ethernet board. Especially, if the SMI bus is shared by only one PHY device, setting this value to -1 can enable the driver to detect the PHY address automatically.
*:cpp:member:`eth_phy_config_t::reset_timeout_ms`: reset timeout value, in milliseconds. Typically, PHY reset should be finished within 100 ms.
*:cpp:member:`eth_phy_config_t::autonego_timeout_ms`: auto-negotiation timeout value, in milliseconds. The Ethernet driver will start negotiation with the peer Ethernet node automatically, to determine to duplex and speed mode. This value usually depends on the ability of the PHY device on your board.
*:cpp:member:`eth_phy_config_t::reset_gpio_num`: if your board also connects the PHY reset pin to one of the GPIO, then set it here. Otherwise, set this field to -1.
* When creating MAC and PHY instances for SPI-Ethernet modules (e.g. DM9051), the constructor function must have the same suffix (e.g. `esp_eth_mac_new_dm9051` and `esp_eth_phy_new_dm9051`). This is because we don't have other choices but the integrated PHY.
* The SPI device configuration (i.e. `spi_device_interface_config_t`) may slightly differ for other Ethernet modules or to meet SPI timing on specific PCB. Please check out your module's specs and the examples in ESP-IDF.
To install the Ethernet driver, we need to combine the instance of MAC and PHY and set some additional high-level configurations (i.e. not specific to either MAC or PHY) in :cpp:class:`esp_eth_config_t`:
*:cpp:member:`esp_eth_config_t::check_link_period_ms`: Ethernet driver starts an OS timer to check the link status periodically, this field is used to set the interval, in milliseconds.
*:cpp:member:`esp_eth_config_t::stack_input`: In most Ethernet IoT applications, any Ethernet frame received by a driver should be passed to the upper layer (e.g. TCP/IP stack). This field is set to a function that is responsible to deal with the incoming frames. You can even update this field at runtime via function :cpp:func:`esp_eth_update_input_path` after driver installation.
*:cpp:member:`esp_eth_config_t::on_lowlevel_init_done` and :cpp:member:`esp_eth_config_t::on_lowlevel_deinit_done`: These two fields are used to specify the hooks which get invoked when low-level hardware has been initialized or de-initialized.
The Ethernet driver also includes an event-driven model, which will send useful and important events to user space. We need to initialize the event loop before installing the Ethernet driver. For more information about event-driven programming, please refer to :doc:`ESP Event <../system/esp_event>`.
esp_event_loop_create_default(); // create a default event loop that runs in the background
esp_event_handler_register(ETH_EVENT, ESP_EVENT_ANY_ID, ð_event_handler, NULL); // register Ethernet event handler (to deal with user-specific stuff when events like link up/down happened)
Up until now, we have installed the Ethernet driver. From the view of OSI (Open System Interconnection), we're still on level 2 (i.e. Data Link Layer). While we can detect link up and down events and gain MAC address in user space, it's infeasible to obtain the IP address, let alone send an HTTP request. The TCP/IP stack used in ESP-IDF is called LwIP. For more information about it, please refer to :doc:`LwIP <../../api-guides/lwip>`.
It is recommended to fully initialize the Ethernet driver and network interface before registering the user's Ethernet/IP event handlers, i.e. register the event handlers as the last thing prior to starting the Ethernet driver. Such an approach ensures that Ethernet/IP events get executed first by the Ethernet driver or network interface so the system is in the expected state when executing the user's handlers.
Ethernet on MCU usually has a limitation in the number of frames it can handle during network congestion, because of the limitation in RAM size. A sending station might be transmitting data faster than the peer end can accept it. The ethernet flow control mechanism allows the receiving node to signal the sender requesting the suspension of transmissions until the receiver catches up. The magic behind that is the pause frame, which was defined in IEEE 802.3x.
Pause frame is a special Ethernet frame used to carry the pause command, whose EtherType field is 0x8808, with the Control opcode set to 0x0001. Only stations configured for full-duplex operation may send pause frames. When a station wishes to pause the other end of a link, it sends a pause frame to the 48-bit reserved multicast address of 01-80-C2-00-00-01. The pause frame also includes the period of pause time being requested, in the form of a two-byte integer, ranging from 0 to 65535.
One thing that should be kept in mind is that the pause frame ability will be advertised to the peer end by PHY during auto-negotiation. The Ethernet driver sends a pause frame only when both sides of the link support it.
There are multiple PHY manufacturers with wide portfolios of chips available. The ESP-IDF already supports several PHY chips however one can easily get to a point where none of them satisfies the user's actual needs due to price, features, stock availability, etc.
Luckily, a management interface between EMAC and PHY is standardized by IEEE 802.3 in Section 22.2.4 Management Functions. It defines provisions of the so-called “MII Management Interface” to control the PHY and gather status from the PHY. A set of management registers is defined to control chip behavior, link properties, auto-negotiation configuration, etc. This basic management functionality is addressed by :component_file:`esp_eth/src/esp_eth_phy_802_3.c` in ESP-IDF and so it makes the creation of a new custom PHY chip driver quite a simple task.
Always consult with PHY datasheet since some PHY chips may not comply with IEEE 802.3, Section 22.2.4. It does not mean you are not able to create a custom PHY driver, it will just require more effort. You will have to define all PHY management functions.
The majority of PHY management functionality required by the ESP-IDF Ethernet driver is covered by the :component_file:`esp_eth/src/esp_eth_phy_802_3.c`. However, the following may require developing chip-specific management functions:
* optionally contain additional variables needed to support non-IEEE 802.3 or customized functionality. See :component_file:`esp_eth/src/esp_eth_phy_ksz80xx.c` as an example.
Once you finish the new custom PHY driver implementation, consider sharing it among other users via `IDF Component Registry <https://components.espressif.com/>`_.
