On Thu, Nov 01, 2018 at 11:40:01AM +0000, Richard Fitzgerald wrote:

On 01/11/18 10:28, Charles Keepax wrote:

quoted

So pulling this out from earlier discussions in this thread,
it seems I can happily move all the child device registration
into device tree. I will also try this for the next version of
the patch, unless anyone wants to object? But it does change
the DT binding quite a lot as the individual sub drivers now
each require their own node rather than one single unified
Lochnagar node.

We went through this discussion with the Madera MFD patches. I had
originally implemented it using DT to register the child drivers and
it was nice in some ways each driver having its own node. But Mark
and Rob didn't like it so I went back to non-DT child registration with
all sharing the parent MFD node. It would be nice if we could stick to
one way of doing it so that Cirrus drivers don't flip-flop between
different styles of DT binding.

The basic concern I have is encoding the current Linux idea of how to
split the subfunctions up into drivers into an ABI - the clocks in CODEC
drivers is the obvious example, they might want to be in the clock API
in future.  If there's a very direct mapping onto individual hardware
blocks that worries me a lot less since it's more obviously reflecting
how the hardware is designed.