On Wed, Mar 25, 2015 at 04:33:20PM +0200, Irina Tirdea wrote:
> There are devices that need to handle block transactions
> regardless of the capabilities exported by the adapter.
> For performance reasons, they need to use i2c read blocks
> if available, otherwise emulate the block transaction with word
> or byte transactions.
> 
> Add support for a helper function that would read a data block
> using the best transfer available: I2C_FUNC_SMBUS_READ_I2C_BLOCK,
> I2C_FUNC_SMBUS_READ_WORD_DATA or I2C_FUNC_SMBUS_READ_BYTE_DATA.
> 
> Signed-off-by: Irina Tirdea <irina.tirdea@intel.com>
> ---
> 
> Hi,
> 
> This is a new API proposal to handle i2c block emulation in the
> core instead of the driver code.
> 
> This is needed for a set of iio sensor changes ([1], [2], [3])
> that would otherwise duplicate this code. There are also some
> usages of this functionality in the kernel (e.g. eeprom driver at24).
> 
> Please let me know what you think.

I am open to add something like this. One change I'd like to request is
to introduce a user, e.g. convert at24.

> 
> Thanks,
> Irina
> 
> [1] https://lkml.org/lkml/2015/2/16/408
> [2] https://lkml.org/lkml/2015/2/16/413
> [3] https://lkml.org/lkml/2015/2/16/402
> 
>  drivers/i2c/i2c-core.c | 62 ++++++++++++++++++++++++++++++++++++++++++++++++++
>  include/linux/i2c.h    |  3 +++
>  2 files changed, 65 insertions(+)
> 
> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> index fe80f85..2579f7d 100644
> --- a/drivers/i2c/i2c-core.c
> +++ b/drivers/i2c/i2c-core.c
> @@ -2907,6 +2907,68 @@ trace:
>  }
>  EXPORT_SYMBOL(i2c_smbus_xfer);
>  
> +/**
> + * i2c_smbus_read_i2c_block_data_or_emulated - read block or emulate
> + * @client: Handle to slave device
> + * @command: Byte interpreted by slave
> + * @length: Size of data block; SMBus allows at most 32 bytes
> + * @values: Byte array into which data will be read; big enough to hold
> + *	the data returned by the slave.  SMBus allows at most 32 bytes.

Sidenote: SMBus3 allows 255 byte, but we don't support that (yet), so
this is okay for now.

> + *
> + * This executes the SMBus "block read" protocol if supported by the adapter.
> + * If block read is not supported, it emulates it using either word or byte
> + * read protocols depending on availability.

Here I'd like to see a warning that people should double-check if their
I2C slave does support that. Sometimes one can't exchange a block
transfer with a byte transfer.

> + */
> +s32 i2c_smbus_read_i2c_block_data_or_emulated(const struct i2c_client *client,
> +					      u8 command, u8 length, u8 *values)
> +{
> +	u8 i;
> +	int status;
> +
> +	if (length > I2C_SMBUS_BLOCK_MAX)
> +		length = I2C_SMBUS_BLOCK_MAX;
> +
> +	if (i2c_check_functionality(client->adapter,
> +				    I2C_FUNC_SMBUS_READ_I2C_BLOCK)) {
> +		return i2c_smbus_read_i2c_block_data(client, command,
> +						     length, values);
> +	} else if (i2c_check_functionality(client->adapter,
> +					   I2C_FUNC_SMBUS_READ_WORD_DATA |
> +					   I2C_FUNC_SMBUS_READ_BYTE_DATA)) {

What about skipping the need for READ_BYTE_DATA and dump the byte which
was maybe read too much?

> +		for (i = 0; (i + 2) <= length; i += 2) {
> +			status = i2c_smbus_read_word_data(client, command + i);
> +			if (status < 0)
> +				return status;
> +			values[i] = status & 0xff;
> +			values[i+1] = status >> 8;
> +		}
> +		if (i < length) {
> +			status = i2c_smbus_read_byte_data(client, command + i);
> +			if (status < 0)
> +				return status;
> +			values[i] = status;
> +			i++;
> +		}
> +		return i;
> +	} else if (i2c_check_functionality(client->adapter,
> +					   I2C_FUNC_SMBUS_READ_BYTE_DATA)) {
> +		for (i = 0; i < length; i++) {
> +			status = i2c_smbus_read_byte_data(client, command + i);
> +			if (status < 0)
> +				return status;
> +			values[i] = status;
> +		}
> +		return i;
> +	}
> +
> +	dev_err(&client->adapter->dev, "Unsupported transactions: %d,%d,%d\n",
> +		I2C_SMBUS_I2C_BLOCK_DATA, I2C_SMBUS_WORD_DATA,
> +		I2C_SMBUS_BYTE_DATA);
> +
> +	return -EOPNOTSUPP;
> +}
> +EXPORT_SYMBOL(i2c_smbus_read_i2c_block_data_or_emulated);

Thanks,

   Wolfram