> As I said in the reply to v1, I think we should reserve addresses also
> when there is a compatible string but no matching driver, but this is
> another story and can be handled separately.

Unless I misunderstand you, I think they do already. Note that
only 'i2cdetect' shows a device as busy *IFF* there is a driver bound to
it. The internal 'i2c_check_addr_busy' does not care about a driver
being bound. You can check this by trying to use
i2c_new_ancillary_device() with an address which is already described in
DT but which driver is disabled.