> Briefly looking into ACPI tables we have and mechanisms that we can
> use in ACPI case, I doubt we may apply all the ideas, probably some of
> them, though I didn't get yet where to read about in details. What I
> could say now is that the patch provided by Bin Gao is definitely no
> go.

Laurent explained to me what V4L did and now does. It used to be the way
tha V4L drivers did register I2C slaves according to platform_data. Now,
with DT the slaves get instanciated earlier, so they now use notifiers
to know when the slaves are in place. Something like this should
probably be done here, too, instead of unregistering and re-registering.