SPI Slave Half Duplex


The half duplex (HD) mode is a special mode provided by ESP SPI Slave peripheral. Under this mode, the hardware provides more services than the full duplex (FD) mode (the mode for general purpose SPI transactions, see SPI Slave Driver). These services reduce the CPU load and the response time of SPI Slave, but the communication format is determined by the hardware. The communication format is always half duplex, so comes the name of Half Duplex Mode.

There are several different types of transactions, determined by the command phase of the transaction. Each transaction may consist of the following phases: command, address, dummy, data. The command phase is mandatory, while the other fields may be determined by the command field. During the command, address, dummy phases, the bus is always controlled by the master, while the direction of the data phase depends on the command. The data phase can be either an in phase, for the master to write data to the slave; or an out phase, for the master to read data from the slave.

About the details of how master should communicate with the SPI Slave, see ESP SPI Slave HD (Half Duplex) Mode Protocol.

By these different transactions, the slave provide these services to the master:

  • A DMA channel for the master to write a great amount of data to the slave.

  • A DMA channel for the master to read a great amount of data from the slave.

  • Several general purpose registers, shard between the master and the slave.

  • Several general purpose interrupts, for the master to interrupt the SW of slave.


  • Transaction

  • Channel

  • Sending

  • Receiving

  • Data Descriptor

Driver Feature

  • Transaction read/write by master in segments

  • Queues for data to send and received

Driver usage

Slave initialization

Call spi_slave_hd_init() to initialize the SPI bus as well as the peripheral and the driver. The SPI slave will exclusively use the SPI peripheral, pins of the bus before it’s deinitialized. Most configurations of the slave should be done as soon as the slave is being initialized.

The spi_bus_config_t specifies how the bus should be initialized, while spi_slave_hd_slot_config_t specifies how the SPI Slave driver should work.

Deinitialization (optional)

Call spi_slave_hd_deinit() to uninstall the driver. The resources, including the pins, SPI peripheral, internal memory used by the driver, interrupt sources, will be released by the deinit function.

Send/Receive Data by DMA Channels

To send data to the master through the sending DMA channel, the application should properly wrap the data to send by a spi_slave_hd_data_t descriptor structure before calling spi_slave_hd_queue_trans() with the data descriptor, and the channel argument of SPI_SLAVE_CHAN_TX. The pointers to descriptors are stored in the queue, and the data will be send to the master upon master’s RDDMA command in the same order they are put into the queue by spi_slave_hd_queue_trans().

The application should check the result of data sending by calling spi_slave_hd_get_trans_res() with the channel set as SPI_SLAVE_CHAN_TX. This function will block until the transaction with command RDDMA from master successfully completes (or timeout). The out_trans argument of the function will output the pointer of the data descriptor which is just finished.

Receiving data from the master through the receiving DMA channel is quite similar. The application calls spi_slave_hd_queue_trans() with proper data descriptor and the channel argument of SPI_SLAVE_CHAN_RX. And the application calls the spi_slave_hd_get_trans_res() later to get the descriptor to the receiving buffer, before it handles the data in the receiving buffer.


This driver itself doesn’t have internal buffer for the data to send, or just received. The application should provide data descriptors for the data buffer to send to master, or to receive data from the master.

The application will have to properly keep the data descriptor as well as the buffer it points to, after the descriptor is successfully sent into the driver internal queue by spi_slave_hd_queue_trans(), and before returned by spi_slave_hd_get_trans_res(). During this period, the hardware as well as the driver may read or write to the buffer and the descriptor when required at any time.

Please note that the buffer doesn’t have to be fully sent or filled before it’s terminated. For example, in the segment transaction mode, the master has to send CMD7 to terminate a WRDMA transaction, or send CMD8 to terminate a RDDMA transaction (in segments), no matter the send (receive) buffer is used up (full) or not.

Using Data Arguments

Sometimes you may have initiator (sending data descriptor) and closure (handling returned descriptors) functions in different places. When you get the returned data descriptor in the closure, you may need some extra information when handle the finished data descriptor. For example, you may want to know which round it is for the returned descriptor, when you send the same piece of data for several times.

Set the arg member in the data descriptor to an variable indicating the transaction (by force casting), or point it to a a structure which wraps all the information you may need when handling the sending/receiving data. Then you can get what you need in your closure.

Using callbacks


These callbacks are called in the ISR, so that they are fast enough. However, you may need to be very careful to write the code in the ISR. The callback should return as soon as possible. No delay or blocking operations are allowed.

The spi_slave_hd_intr_config_t member in the spi_slave_hd_slot_config_t configuration structure passed when initialize the SPI Slave HD driver, allows you having callbacks for each events you may concern.

The corresponding interrupt for each callbacks that is not NULL will enabled, so that the callbacks can be called immediately when the events happen. You don’t need to provide callbacks for the unconcerned events.

The arg member in the configuration structure can help you pass some context to the callback, or indicate which SPI Slave instance when you are using the same callbacks for several SPI Slave peripherals. Set the arg member to an variable indicating the SPI Slave instance (by force casting), or point it to a context structure. All the callbacks will be called with this arg argument you set when the callbacks are initialized.

There are two other arguments: the event and the awoken. The event passes the information of the current event to the callback. The spi_slave_hd_event_t type contains the information of the event, for example, event type, the data descriptor just finished (The data argument will be very useful in this case!). The awoken argument is an output one, telling the ISR there are tasks are awoken after this callback, and the ISR should call portYIELD_FROM_ISR() to do task scheduling. Just pass the awoken argument to all FreeRTOS APIs which may unblock tasks, and the awoken will be returned to the ISR.

Writing/Reading Shared Registers

Call spi_slave_hd_write_buffer() to write the shared buffer, and spi_slave_hd_read_buffer() to read the shared buffer.


On ESP32-S2, the shared registers are read/written in words by the application, but read/written in bytes by the master. There’s no guarantee four continuous bytes read from the master are from the same word written by the slave’s application. It’s also possible that if the slave reads a word while the master is writing bytes of the word, the slave may get one word with half of them just written by the master, and the other half hasn’t been written into.

The master can confirm that the word is not in transition by reading the word twice and comparing the values.

For the slave, it will be more difficult to ensure the word is not in transition because the process of master writing four bytes can be very long (32 SPI clocks). You can put some CRC in the last (largest address) byte of a word so that when the byte is written, the word is sure to be all written.

Due to the conflicts there may be among read/write from SW (worse if there are multi cores) and master, it is suggested that a word is only used in one direction (only written by master or only written by the slave).

Receiving General Purpose Interrupts From the Master

When the master sends CMD 0x08, 0x09 or 0x0A, the slave corresponding will be triggered. Currently the CMD8 is permanently used to indicate the termination of RDDMA segments. To receiving general purpose interrupts, register callbacks for CMD 0x09 and 0x0A when the slave is initialized, see Using callbacks.

API reference


esp_err_t spi_slave_hd_init(spi_host_device_t host_id, const spi_bus_config_t *bus_config, const spi_slave_hd_slot_config_t *config)

Initialize the SPI Slave HD driver.


  • ESP_OK: on success

  • ESP_ERR_INVALID_ARG: invalid argument given

  • ESP_ERR_INVALID_STATE: function called in invalid state, may be some resources are already in use

  • ESP_ERR_NO_MEM: memory allocation failed

  • or other return value from esp_intr_alloc

  • host_id: The host to use

  • bus_config: Bus configuration for the bus used

  • config: Configuration for the SPI Slave HD driver

esp_err_t spi_slave_hd_deinit(spi_host_device_t host_id)

Deinitialize the SPI Slave HD driver.


  • ESP_OK: on success

  • ESP_ERR_INVALID_ARG: if the host_id is not correct

  • host_id: The host to deinitialize the driver

esp_err_t spi_slave_hd_queue_trans(spi_host_device_t host_id, spi_slave_chan_t chan, spi_slave_hd_data_t *trans, TickType_t timeout)

Queue data transaction.


  • ESP_OK: on success

  • ESP_ERR_INVALID_ARG: The input argument is invalid. Can be the following reason:

    • The buffer given is not DMA capable

    • The length of data is invalid (not larger than 0, or exceed the max transfer length)

    • The function is invalid

  • ESP_ERR_TIMEOUT: Cannot queue the data before timeout. This is quite possible if the master doesn’t read/write the slave on time.

  • host_id: Host to queue the transaction

  • chan: Channel to queue the data, SPI_SLAVE_CHAN_TX or SPI_SLAVE_CHAN_RX

  • trans: Descriptor of data to queue

  • timeout: Timeout before the data is queued

esp_err_t spi_slave_hd_get_trans_res(spi_host_device_t host_id, spi_slave_chan_t chan, spi_slave_hd_data_t **out_trans, TickType_t timeout)

Get the result of a data transaction.


  • ESP_OK: on success

  • ESP_ERR_INVALID_ARG: Function is not valid

  • ESP_ERR_TIMEOUT: There’s no transaction done before timeout

  • host_id: Host to queue the transaction

  • chan: Channel to get the result, SPI_SLAVE_CHAN_TX or SPI_SLAVE_CHAN_RX

  • [out] out_trans: Output descriptor of the returned transaction

  • timeout: Timeout before the result is got

void spi_slave_hd_read_buffer(spi_host_device_t host_id, int addr, uint8_t *out_data, size_t len)

Read the shared registers.

  • host_id: Host to read the shared registers

  • addr: Address of register to read, 0 to SOC_SPI_MAXIMUM_BUFFER_SIZE-1

  • [out] out_data: Output buffer to store the read data

  • len: Length to read, not larger than SOC_SPI_MAXIMUM_BUFFER_SIZE-addr

void spi_slave_hd_write_buffer(spi_host_device_t host_id, int addr, uint8_t *data, size_t len)

Write the shared registers.

  • host_id: Host to write the shared registers

  • addr: Address of register to write, 0 to SOC_SPI_MAXIMUM_BUFFER_SIZE-1

  • data: Buffer holding the data to write

  • len: Length to write, SOC_SPI_MAXIMUM_BUFFER_SIZE-addr


struct spi_slave_hd_data_t

Descriptor of data to send/receive.

Public Members

uint8_t *data

Buffer to send, must be DMA capable.

size_t len

Len of data to send/receive. For receiving the buffer length should be multiples of 4 bytes, otherwise the extra part will be truncated.

size_t trans_len

Data actually received.

void *arg

Extra argument indiciating this data.

struct spi_slave_hd_event_t

Information of SPI Slave HD event.

Public Members

spi_event_t event

Event type.

spi_slave_hd_data_t *trans

Corresponding transaction for SPI_EV_SEND and SPI_EV_RECV events.

struct spi_slave_hd_callback_config_t

Callback configuration structure for SPI Slave HD.

Public Members

slave_cb_t cb_recv

Callback when receive data.

slave_cb_t cb_sent

Callback when data sent.

slave_cb_t cb_buffer_tx

Callback when master reads from shared buffer.

slave_cb_t cb_buffer_rx

Callback when master writes to shared buffer.

slave_cb_t cb_cmd9

Callback when CMD9 received.

slave_cb_t cb_cmdA

Callback when CMDA received.

void *arg

Argument indicating this SPI Slave HD peripheral instance.

struct spi_slave_hd_slot_config_t

Configuration structure for the SPI Slave HD driver.

Public Members

int spics_io_num

CS GPIO pin for this device.

uint32_t flags

Bitwise OR of SPI_SLAVE_HD_* flags.

uint8_t mode

SPI mode (0-3)

int command_bits

command field bits, multiples of 8 and at least 8.

int address_bits

address field bits, multiples of 8 and at least 8.

int dummy_bits

dummy field bits, multiples of 8 and at least 8.

int queue_size

Transaction queue size. This sets how many transactions can be ‘in the air’ (queued using spi_slave_hd_queue_trans but not yet finished using spi_slave_hd_get_trans_result) at the same time.

int dma_chan

DMA channel used.

spi_slave_hd_callback_config_t cb_config

Callback configuration.



Transmit command/address/data LSB first instead of the default MSB first.


Receive data LSB first instead of the default MSB first.


Transmit and receive LSB first.

Type Definitions

typedef bool (*slave_cb_t)(void *arg, spi_slave_hd_event_t *event, BaseType_t *awoken)

Callback for SPI Slave HD.


enum spi_slave_chan_t

Channel of SPI Slave HD to do data transaction.



The output channel (RDDMA)


The input channel (WRDMA)