On Wed, Jun 30, 2021 at 07:17:28PM +0200, H. Nikolaus Schaller wrote:

> Splitting into many regulators also needs to touch the device trees
> to have individual compatible entries which currently do not exist.

No, it doesn't.  There's absolutely no need for any specific mapping
between Linux devices and compatible strings or nodes in the DT, we can
create any number of Linux devices for any number of compatibles - just
look at MFDs (where we create multiple Linux devices for a single DT
compatible string) or system devices (where we create Linux devices with
potentially no node in the device tree).

> On the other hand, a theoretical system could have a real fixed regulator
> in between (maybe a power switch?) and should still work. Why should 
> driver core care about that case and not the core system it is using?

For deferred probe to be guaranteed to work we really should have one
regulator per Linux device but in practice that is overhead and effort
that almost never buys us anything in practical systems (I can't
emphasize strongly enough how unusual chains of more than two regulators
are) so we don't enforce doing that.