> 3) Thinking about Mainline: To reach the same target - no I2C
> detection - and taking
> into account above assumption "No changes in default behavior"
> the following will need to be done:
> - change i2c-omap/i2c-gpio DT bindings and add parameter which will
> allow to change
>   .class value for adapter. Not sure, it's possible because this parameter
>   will be Linux and not HW specific (smth. like "i2c_disable_detection")
> - update drivers i2c-omap/i2c-gpio to use "i2c_disable_detection"
> - update OMAP4/5 DTS files
> 
> So, It seemed a good solution for me to add 6 lines of code in i2c-core.c
> instead of doing all that stuff.

Well... I understand the "default behaviour" issue, yet I still think
that setting class to 0 is the right thing to do. OMAP is an embedded
SoC which always had i2c_board_info or devictree which are the preferred
ways of instantiating. Given that, I would accept a patch setting it to
0. The more user friendly way might be to introduce a new class which
makes users aware of the issue. Proof of concept follows, only compile
tested.