I could fix this in the driver by calling of_device_get_match_data() in
the probe function with dev->of_node is not NULL, but I feel this is
really an issue that should be handled by the framework. Has anyone ever
given it a thought ?

The of entries should also have .data entries and use that.

The driver could also be converted to probe_new to become independent of
the i2c_device_id.