On Tue, 2007-01-09 at 23:17 +0100, Segher Boessenkool wrote:

quoted

I don't see the need of having some weird "soc" node that doesn't quite
mean anything (especially on axon) that has phandles to every sub 
device
in there :-)

Well simply, if I understood you correctly, you have some
register where some bits control emac #0 and some control
emac #1.  This register can't belong to either of those
devices because it can't belong to both, so it has to
belong to some "control" / "power management" / whatever
device.

quoted

On those ASICs, every device almost needs to know what is it's "cell
index" because of little details here or there.

Yes.

quoted

I really see that as an
attribute of the device and thus should be a property of the node.

You still need to describe that register somewhere in the
device tree.

I will not describe every single weird clock control or other magic DCRs
in the device-tree. That is simply over-bloat. We don't describe the U3
clock control or PM registers on js2x's for example, nor do we describe
the various individual registers controlling the HT link...

Ben.