[RFC/PATCH] Set up PCI tree from OF on ppc64

[Paul Mackerras]

John Rose writes:

> Would we not want to use PROBE_DEVTREE for non-partitioned and legacy
> pSeries?

We could, once we've tested that it all works as expected on POWER3
and RS64 systems.  I was just being a bit conservative.

> +static void __devinit of_scan_bus(struct device_node *node,
> +				  struct pci_bus *bus)
> ...
> +static void __devinit of_scan_pci_bridge(struct device_node *node,
> +					 struct pci_dev *dev)
> 
> We should probably expose these for use by the PCI Hotplug/DLPAR
> drivers, along w/ scan_phb().

Sure.  I was thinking as I was writing this code that we should use it
for hotplug as well.

> Could pci_process_bridge_OF_ranges() and of_scan_pci_bridge() use
> common "ranges" parsing logic?  We're parsing the same property
> in two different ways in the same file.  This could be a "todo" I
> guess.

Yes, probably.  The size of each entry in the ranges property depends
on #address-cells for the parent, #address-cells for the child, and
#size-cells for the child.  Currently we have those values hard-coded
to 3, 3, and 2 for of_scan_pci_bridge (which is OK because those
values are mandated by the PCI binding to OF), and 2, 3, 2 in
pci_process_bridge_OF_ranges.  That's the main difference between
them.

Paul.