Implementing I2C device drivers

This is a small guide for those who want to write kernel drivers for I2C or SMBus devices, using Linux as the protocol host/master (not slave).

To set up a driver, you need to do several things. Some are optional, and some things can be done slightly or completely different. Use this as a guide, not as a rule book!

General remarks

Try to keep the kernel namespace as clean as possible. The best way to do this is to use a unique prefix for all global symbols. This is especially important for exported symbols, but it is a good idea to do it for non-exported symbols too. We will use the prefix foo_ in this tutorial.

The driver structure

Usually, you will implement a single driver structure, and instantiate all clients from it. Remember, a driver structure contains general access routines, and should be zero-initialized except for fields with data you provide. A client structure holds device-specific information like the driver model device node, and its I2C address.

static struct i2c_device_id foo_idtable[] = {
      { "foo", my_id_for_foo },
      { "bar", my_id_for_bar },
      { }

MODULE_DEVICE_TABLE(i2c, foo_idtable);

static struct i2c_driver foo_driver = {
      .driver = {
              .name   = "foo",
              .pm     = &foo_pm_ops,  /* optional */

      .id_table       = foo_idtable,
      .probe          = foo_probe,
      .remove         = foo_remove,

      .shutdown       = foo_shutdown, /* optional */
      .command        = foo_command,  /* optional, deprecated */

The name field is the driver name, and must not contain spaces. It should match the module name (if the driver can be compiled as a module), although you can use MODULE_ALIAS (passing “foo” in this example) to add another name for the module. If the driver name doesn’t match the module name, the module won’t be automatically loaded (hotplug/coldplug).

All other fields are for call-back functions which will be explained below.

Extra client data

Each client structure has a special data field that can point to any structure at all. You should use this to keep device-specific data.

/* store the value */
void i2c_set_clientdata(struct i2c_client *client, void *data);

/* retrieve the value */
void *i2c_get_clientdata(const struct i2c_client *client);

Note that starting with kernel 2.6.34, you don’t have to set the data field to NULL in remove() or if probe() failed anymore. The i2c-core does this automatically on these occasions. Those are also the only times the core will touch this field.

Accessing the client

Let’s say we have a valid client structure. At some time, we will need to gather information from the client, or write new information to the client.

I have found it useful to define foo_read and foo_write functions for this. For some cases, it will be easier to call the I2C functions directly, but many chips have some kind of register-value idea that can easily be encapsulated.

The below functions are simple examples, and should not be copied literally:

int foo_read_value(struct i2c_client *client, u8 reg)
      if (reg < 0x10) /* byte-sized register */
              return i2c_smbus_read_byte_data(client, reg);
      else            /* word-sized register */
              return i2c_smbus_read_word_data(client, reg);

int foo_write_value(struct i2c_client *client, u8 reg, u16 value)
      if (reg == 0x10)        /* Impossible to write - driver error! */
              return -EINVAL;
      else if (reg < 0x10)    /* byte-sized register */
              return i2c_smbus_write_byte_data(client, reg, value);
      else                    /* word-sized register */
              return i2c_smbus_write_word_data(client, reg, value);

Probing and attaching

The Linux I2C stack was originally written to support access to hardware monitoring chips on PC motherboards, and thus used to embed some assumptions that were more appropriate to SMBus (and PCs) than to I2C. One of these assumptions was that most adapters and devices drivers support the SMBUS_QUICK protocol to probe device presence. Another was that devices and their drivers can be sufficiently configured using only such probe primitives.

As Linux and its I2C stack became more widely used in embedded systems and complex components such as DVB adapters, those assumptions became more problematic. Drivers for I2C devices that issue interrupts need more (and different) configuration information, as do drivers handling chip variants that can’t be distinguished by protocol probing, or which need some board specific information to operate correctly.

Device/Driver Binding

System infrastructure, typically board-specific initialization code or boot firmware, reports what I2C devices exist. For example, there may be a table, in the kernel or from the boot loader, identifying I2C devices and linking them to board-specific configuration information about IRQs and other wiring artifacts, chip type, and so on. That could be used to create i2c_client objects for each I2C device.

I2C device drivers using this binding model work just like any other kind of driver in Linux: they provide a probe() method to bind to those devices, and a remove() method to unbind.

static int foo_probe(struct i2c_client *client);
static void foo_remove(struct i2c_client *client);

Remember that the i2c_driver does not create those client handles. The handle may be used during foo_probe(). If foo_probe() reports success (zero not a negative status code) it may save the handle and use it until foo_remove() returns. That binding model is used by most Linux drivers.

The probe function is called when an entry in the id_table name field matches the device’s name. If the probe function needs that entry, it can retrieve it using

const struct i2c_device_id *id = i2c_match_id(foo_idtable, client);

Device Creation

If you know for a fact that an I2C device is connected to a given I2C bus, you can instantiate that device by simply filling an i2c_board_info structure with the device address and driver name, and calling i2c_new_client_device(). This will create the device, then the driver core will take care of finding the right driver and will call its probe() method. If a driver supports different device types, you can specify the type you want using the type field. You can also specify an IRQ and platform data if needed.

Sometimes you know that a device is connected to a given I2C bus, but you don’t know the exact address it uses. This happens on TV adapters for example, where the same driver supports dozens of slightly different models, and I2C device addresses change from one model to the next. In that case, you can use the i2c_new_scanned_device() variant, which is similar to i2c_new_client_device(), except that it takes an additional list of possible I2C addresses to probe. A device is created for the first responsive address in the list. If you expect more than one device to be present in the address range, simply call i2c_new_scanned_device() that many times.

The call to i2c_new_client_device() or i2c_new_scanned_device() typically happens in the I2C bus driver. You may want to save the returned i2c_client reference for later use.

Device Detection

The device detection mechanism comes with a number of disadvantages. You need some reliable way to identify the supported devices (typically using device-specific, dedicated identification registers), otherwise misdetections are likely to occur and things can get wrong quickly. Keep in mind that the I2C protocol doesn’t include any standard way to detect the presence of a chip at a given address, let alone a standard way to identify devices. Even worse is the lack of semantics associated to bus transfers, which means that the same transfer can be seen as a read operation by a chip and as a write operation by another chip. For these reasons, device detection is considered a legacy mechanism and shouldn’t be used in new code.

Device Deletion

Each I2C device which has been created using i2c_new_client_device() or i2c_new_scanned_device() can be unregistered by calling i2c_unregister_device(). If you don’t call it explicitly, it will be called automatically before the underlying I2C bus itself is removed, as a device can’t survive its parent in the device driver model.

Initializing the driver

When the kernel is booted, or when your foo driver module is inserted, you have to do some initializing. Fortunately, just registering the driver module is usually enough.

static int __init foo_init(void)
      return i2c_add_driver(&foo_driver);

static void __exit foo_cleanup(void)

The module_i2c_driver() macro can be used to reduce above code.


Note that some functions are marked by __init. These functions can be removed after kernel booting (or module loading) is completed. Likewise, functions marked by __exit are dropped by the compiler when the code is built into the kernel, as they would never be called.

Driver Information

/* Substitute your own name and email address */
MODULE_AUTHOR("Frodo Looijaard <>"
MODULE_DESCRIPTION("Driver for Barf Inc. Foo I2C devices");

/* a few non-GPL license types are also allowed */

Power Management

If your I2C device needs special handling when entering a system low power state -- like putting a transceiver into a low power mode, or activating a system wakeup mechanism -- do that by implementing the appropriate callbacks for the dev_pm_ops of the driver (like suspend and resume).

These are standard driver model calls, and they work just like they would for any other driver stack. The calls can sleep, and can use I2C messaging to the device being suspended or resumed (since their parent I2C adapter is active when these calls are issued, and IRQs are still enabled).

System Shutdown

If your I2C device needs special handling when the system shuts down or reboots (including kexec) -- like turning something off -- use a shutdown() method.

Again, this is a standard driver model call, working just like it would for any other driver stack: the calls can sleep, and can use I2C messaging.

Command function

A generic ioctl-like function call back is supported. You will seldom need this, and its use is deprecated anyway, so newer design should not use it.

Sending and receiving

If you want to communicate with your device, there are several functions to do this. You can find all of them in <linux/i2c.h>.

If you can choose between plain I2C communication and SMBus level communication, please use the latter. All adapters understand SMBus level commands, but only some of them understand plain I2C!

Plain I2C communication

int i2c_master_send(struct i2c_client *client, const char *buf,
                    int count);
int i2c_master_recv(struct i2c_client *client, char *buf, int count);

These routines read and write some bytes from/to a client. The client contains the I2C address, so you do not have to include it. The second parameter contains the bytes to read/write, the third the number of bytes to read/write (must be less than the length of the buffer, also should be less than 64k since msg.len is u16.) Returned is the actual number of bytes read/written.

int i2c_transfer(struct i2c_adapter *adap, struct i2c_msg *msg,
                 int num);

This sends a series of messages. Each message can be a read or write, and they can be mixed in any way. The transactions are combined: no stop condition is issued between transaction. The i2c_msg structure contains for each message the client address, the number of bytes of the message and the message data itself.

You can read the file The I2C Protocol for more information about the actual I2C protocol.

SMBus communication

s32 i2c_smbus_xfer(struct i2c_adapter *adapter, u16 addr,
                   unsigned short flags, char read_write, u8 command,
                   int size, union i2c_smbus_data *data);

This is the generic SMBus function. All functions below are implemented in terms of it. Never use this function directly!

s32 i2c_smbus_read_byte(struct i2c_client *client);
s32 i2c_smbus_write_byte(struct i2c_client *client, u8 value);
s32 i2c_smbus_read_byte_data(struct i2c_client *client, u8 command);
s32 i2c_smbus_write_byte_data(struct i2c_client *client,
                              u8 command, u8 value);
s32 i2c_smbus_read_word_data(struct i2c_client *client, u8 command);
s32 i2c_smbus_write_word_data(struct i2c_client *client,
                              u8 command, u16 value);
s32 i2c_smbus_read_block_data(struct i2c_client *client,
                              u8 command, u8 *values);
s32 i2c_smbus_write_block_data(struct i2c_client *client,
                               u8 command, u8 length, const u8 *values);
s32 i2c_smbus_read_i2c_block_data(struct i2c_client *client,
                                  u8 command, u8 length, u8 *values);
s32 i2c_smbus_write_i2c_block_data(struct i2c_client *client,
                                   u8 command, u8 length,
                                   const u8 *values);

These ones were removed from i2c-core because they had no users, but could be added back later if needed:

s32 i2c_smbus_write_quick(struct i2c_client *client, u8 value);
s32 i2c_smbus_process_call(struct i2c_client *client,
                           u8 command, u16 value);
s32 i2c_smbus_block_process_call(struct i2c_client *client,
                                 u8 command, u8 length, u8 *values);

All these transactions return a negative errno value on failure. The ‘write’ transactions return 0 on success; the ‘read’ transactions return the read value, except for block transactions, which return the number of values read. The block buffers need not be longer than 32 bytes.

You can read the file The SMBus Protocol for more information about the actual SMBus protocol.

General purpose routines

Below all general purpose routines are listed, that were not mentioned before:

/* Return the adapter number for a specific adapter */
int i2c_adapter_id(struct i2c_adapter *adap);