I²C NACK on write/address byte does not error using i2c_master (IDFGH-12073)

[MatthiasKunnen]

^{keywords: I²C NACK no error, i2c_master, i2c master, new driver}

IDF version: 5454d37 (current master)

Problem/Current behavior

When using the new I²C driver (i2c_master) functions, a NACK is perceived as an ACK on both slave
addresses and write bytes.

When the address byte results in a NACK (no device with this address is on the bus), the driver
continues as if an ACK was received.
When a write byte results in a NACK, the driver continues as if an ACK was received.

This affects the following functions:

• i2c_master_receive
  • slave address ACK not verified
• i2c_master_transmit
  • slave address ACK not verified
  • write byte ACK not verified
• i2c_master_transmit_receive
  • slave address ACK not verified
  • write byte ACK not verified

A non-exhaustive list of things that could go wrong because of this:

• i2c_master_transmit on a missing slave appears to succeed.
• i2c_master_transmit on an existing but busy slave appears to succeed.
• i2c_master_receive and i2c_master_transmit_receive on a missing slave results in fake FF bytes which is not distinguishable from an existing slave and valid FF bytes.

This is a regression from the old driver which verifies both the slave address ACK and the write byte's ACK, e.g.:

• i2c_master_write_read_device, address byte
  

  esp-idf/components/driver/i2c/i2c.c

  Line 1077 in 5454d37

  err = i2c_master_write_byte(handle, device_address << 1 | I2C_MASTER_WRITE, true);

• i2c_master_write_read_device, registry address byte
  

  esp-idf/components/driver/i2c/i2c.c

  Line 1082 in 5454d37

  err = i2c_master_write(handle, write_buffer, write_size, true);

I²C Read Byte example using i2c_master_transmit_receive resulting in ESP_OK and FF output.

Code that generated this signal

i2c_device_config_t dev_cfg = {
    .dev_addr_length = I2C_ADDR_BIT_LEN_7,
    .device_address = 0x2F,
    .scl_speed_hz = 50000,
};

i2c_master_dev_handle_t dev_handle;
ESP_ERROR_CHECK(i2c_master_bus_add_device(bus_handle, &dev_cfg, &dev_handle));

uint8_t readRegister = 0x20;
uint8_t receive[1]; // Will be 0xFF

 // Result will be 0, should be non-zero, e.g. 259
esp_err_t result = i2c_master_transmit_receive(dev_handle, &readRegister, 1, receive, 1, -1);

Expected behavior

To be compliant with I²C and/or SMBus, the controller upon detection of the NACK condition must generate a STOP condition to abort the transfer.

Solution

• Enable ACK verification on every write byte.
• Enable ACK verification on the address byte.

I will be making a PR that attempts to address both.

[mythbuster5]

#13136 (comment)

[jrahlf]

Can you clarify which IDF versions "new I2C driver" refers to?

[KaeLL]

v5.2 onwards.

[mythbuster5]

Solved