On Tegra, we should always use the "new" I2C slave controller, to avoid
issues with the old controller. This was implemented in commit 65a1a0ac
"i2c: tegra: Enable new slave mode."

There is currently no driver for the Tegra I2C slave controller upstream.
Additionally, the controller cannot be completely disabled. Instead, we
need to:

a) Set I2C_SL_CNFG_NACK to make the controller automatically NACK any
incoming transactions.

b) The controller's definition of NACK isn't identical to the I2C
protocol's definition. Specifically, it will perform a standard NACK, but
*also* continue to hold the clock line low in expectation of receiving
more data. This can hang the bus, or at least cause transaction timeouts,
if something starts a transaction that matches the controller's slave
address. Since the default address is 0x00, the general call address,
this does occur in practice.

To avoid this, we explicitly program a slave address that is reserved for
future expansion. For current boards, this guarantees the address will
never be used. If a future board ever needs to use this address, we can
add platform data to determine a board-specific safe address. 0xfc is
picked by this patch.

This patch is based on a change previously posted by: Wei Ni <wni@nvidia.com>
http://www.spinics.net/lists/linux-i2c/msg05437.html
In turned based on internal changes by: Bharat Nihalani <bnihalani@nvidia.com>

A semantically equivalent change has been contained in the various
ChromeOS kernels for a while.

I tested this change on top of 3.0-rc2 on Harmony, and interacted with
the WM8903 I2C-based audio codec.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Ben Dooks <ben-linux@fluff.org>