On Tue, 2009-06-23 at 13:34 -0700, Jeremy Fitzhardinge wrote:
On reflection, I think this will work. We have a Xen pci passthrough
driver which gets told about the passed-through devices via xenbus,
and
does the appropriate setup. At first glance, there doesn't seem to be
any problem with that code just explicitly instantiate the devices at
the PCI level in the same way pci_scan_device does (ie,
alloc_pci_device, initalize the dev struct, pci_setup_device).
Is that what you mean?
IanC has looked at that code more closely, so perhaps he can confirm
that this will work on our side.
Yes, we do that in arch/powerpc/kernel/pci_64.c (and somewhere in
arch/sparc too). I'll merge those 2 implementations one of these days
and move them to drivers/of.
I think Willy was actually suggesting that you start using our
OF/device-tree stuff though which is a slightly more invasive deal but
might end up useful for other things in the long run :-)
Cheers,
Ben.