Re: [PATCH] MFD: Add U300 AB3100 core support v1

[Jean Delvare <khali-PUYAD+kWke1g9hUCZPvPmw@public.gmane.org>]

Hi Linus,

On Mon, 18 May 2009 16:22:36 +0200, Linus Walleij wrote:
> Here we have a problem. See above:
> msgs[0].addr = ab3100_i2c_client->addr + 1;
> 
> So this chip actually has two addresses. A "special" address
> to deal with test registers, one address up. The I2C framework
> assume all devices have one and one address only (which is
> of course mostly true).

No, the I2C framework doesn't assume this. All the I2C framework
assumes is that only one device can use a given address on any one I2C
segment (which seems reasonable.)

> Here is a special case. When the first device has registered,
> you know that the other address is available as well.
> 
> You could think of several ugly solutions:
> 
> * Keep using i2c_transfer() directly as we do now.
> 
> * Make a raw copy of the i2c_device with something like
>   memcpy(mock_client, i2c_client, sizeof(i2c_client);
>   mock_client->addr++;
>   then use i2c_master_send()
> 
> * Register a new i2c_device in board_info for the other
>   address while strictly speaking it is the same device, and
>   this will yield a lot of probing and synchronization code,
>   because writing the test registers is used when probing the
>   first device, so we have to wait for that device to be probed
>   before we can probe the other one etc.
> 
> Right now I lean toward the first alternative.

Neither is correct. Simply use i2c_new_dummy() on the second I2C
address, and keep a pointer to the instantiated i2c_client for future
use. Don't forget to call i2c_unregister_device() in your .remove()
method.

-- 
Jean Delvare