[U-Boot] [PATCH] RFC: tegra: Avoid using I2C prior to relocation

[Simon Glass]

Hi Stephen,

On Wed, Aug 7, 2013 at 10:20 AM, Stephen Warren <swarren at wwwdotorg.org>wrote:

> On 08/06/2013 11:52 PM, Simon Glass wrote:
> > Tegra recently moved to the new I2C framework, which sets up I2C prior to
> > relocation, and prior to calling i2c_init_board(). This causes a crash on
> > Tegra boards.
> >
> > note:
> >
> > There are many ways to fix this. I believe this is one. It disables
> i2c_init()
> > until relocation is complete. I have been unable to test it so far due to
> > problems getting my Seaboard to work. I will try another Tegra board, but
> > send this for comment in the meantime.
>
> Tested-by: Stephen Warren <swarren at nvidia.com>
>
> (On Beaver and Dalmore, tested booting to U-Boot command prompt followed
> by "i2c dev 0; i2c probe")
>
> Note: I believe this is an enormous hack that hacks around the problem
> of dynamic device initialization just not being well thought out
> relative to the restrictions of U-Boot's various boot stages. I'd still
> prefer an outright revert of the broken code.
>
> In other words, tegra_i2c_init() simply shouldn't be called at the wrong
> time; it shouldn't have to handle being called at the wrong time and
> null itself out when that happens.
>
> However, if this is what it takes to get U-Boot working again, then
> let's apply it ASAP.
>

Fair enough.

Restricting my comments to I2C only, I think that we could have (at least
until driver model happens) an I2C init for each driver (in this case it is
i2c_init_board() that is called by the I2C subsystem twice during init
(once from board_init_f() and once from board_init_r()). The i2c_init()
call is odd because it does an init of a single I2C bus, which isn't really
necessary for most hardware. But I'm not sure it is worth worrying about
this until we have driver model.

Regards,
Simon