i2c: Get rid of the legacy binding model

As the last few remaining legacy i2c drivers are being converted, we
can finally get rid of the legacy binding model supporting code in
i2c-core. Here come 6 patches which do exactly that:

[1/6] i2c: Kill client_register and client_unregister methods
[2/6] i2c: Get rid of the legacy binding model
[3/6] i2c: Drop i2c_probe function
[4/6] i2c: Merge i2c_attach_client into i2c_new_device
[5/6] i2c: Kill is_newstyle_driver
[6/6] i2c: Kill the redundant client list

As one would expect, the size reduction is very significant:

 Documentation/feature-removal-schedule.txt |   10 -
 Documentation/i2c/writing-clients          |   16 -
 drivers/i2c/i2c-core.c                     |  366 +++--------------------------
 include/linux/i2c.h                        |   57 +----
 4 files changed, 48 insertions(+), 401 deletions(-)

From a binary perspective, "size" announces an 18% shrink.

There are two things which should still be investigated.

1* In i2c_device_uevent() there is the following piece of code:

	/* by definition, legacy drivers can't hotplug */
	if (dev->driver)
		return 0;

David, I presume this can go away?

2* Core locking should be checked. I am not completely sure what
core_lock was supposed to protect, nor what it should protect now. I
think it should protect at least i2c_adapter_idr, driver->detect()
calls (which may add a device to the driver->clients list) and
driver->attach_adapter() calls (for the same reason), although for the
latter two, a per-driver lock may be more appropriate.

At the moment, we also hold the lock for all of i2c_register_adapter(),
i2c_del_adapter() and i2c_del_driver(), but not i2c_register_driver(),
which seems a little inconsistent. David, do you know if we need to
serialize the calls to device_register() and device_unregister(), or if
the driver core takes care of this for us?

Anything else needs to be protected?

-- 
Jean Delvare