quoted

Asking linux-pm for help here: If we want to support RuntimePM for I2C
clients, do we need to enable RuntimePM on the logical I2C adapter
device (the bus master) which is already marked using
pm_runtime_no_callbacks?

In theory you don't need to.  But there are some advantages if you do: 
You get automatic runtime PM time accounting for the adapter device 
(how much time active and how much suspended), and suspend events will 
propagate from the I2C clients all the way up to the adapter's parent.

That's exactly what I want. Thank you very much!