new kernel ... / meaning of stable-tracking

[Werner Almesberger]

Andy Green wrote:
> It makes sense, Harald did this unto Glamo too.

Yup. That's actually the kind of structure I had in mind. There are
two styles:

- keep core and subsystems all in drivers/mfd/

- keep only the core there and put the subsystems elsewhere

Only SM501 seems to be of the latter type in our current mainline,
but upstream has some more examples of such drivers.

I think the "everything under drivers/mfd/" approach is clearer for
things like Glamo or PMU, where the subsystem drivers are specialized
anyway and there are many internal interfaces that are of no use to
anything else. But I can see the benefits of the other approach as
well.

I'll inquire what's best in this case.

> Werner, how do you feel about guiding and helping Balaji with that?

Sounds good.

> It's not simple but it is largely rearrangement and adaptation of known
> working stuff (so it is testable stage by stage),

There's one problem, though: this change, while modifying very little
code, makes the driver completely different from our mainline as far
as git, diff, etc. see it. So moving later changes from one branch to
the other would always require manual intervention and when the whole
thing eventually comes back from upstream, it would be difficult to
tell the restructuring from any other changes that have happened after.

So I'd suggest to do only the restructuring now, without any other
beautification or fixing, and then merging the restructured pcf50633
driver back into our mainline. That way, it's easy to review the
changes, because one can just cat the parts and diff that against the
current monolithic driver.

- Werner