On Tue, Nov 10, 2015 at 01:40:38PM -0600, Andrew F. Davis wrote:

On 11/10/2015 12:44 PM, Mark Brown wrote:

quoted

There's also the third option where we don't have any compatible strings
in the subnodes at all.

Ok, two, but would you really want to go that way? Matching by node name costs
us all of the flexibility of DT sub-device selection. Still don't see an upside
as we would now be locked to node names instead of compatible strings to declare
component type compatibility (what they are for).

Yes, we should go that way.  No, there is nothing meaningful being lost
- the fact that there is zero paramterisation in the bindings and each
subfunction has the full device name as a compatible string ought to be
a big red flag here.