Re: [patch 2/2] x86_64: Collect host bridge resources

[Rajesh Shah <rajesh.shah@intel.com>]

On Tue, May 24, 2005 at 06:58:56PM +0400, Ivan Kokshaysky wrote:
> On Tue, May 24, 2005 at 02:05:27PM +0200, Andi Kleen wrote:
> > How about you allocate an extended structure with kmalloc in this case?
> 
> This would lead to quite a few changes in the PCI subsystem.
> Looks good as a long-term solution though.
> 
Yes, I did look at that and this would be a big change that would
affect almost all architectures. I was thinking something like
this would be more appropriate as part of the PCI rewrite that
Adam Belay had proposed.

> > Or if it is only 6 ranges max (it is not, is it?) you could extend
> > the array.
> > 
No, 6 is not guaranteed but would cover a larger percentage of
systems. 8 would probably cover all but a few special cases.

> > I doubt this information will need *that* much memory, so it should
> > be reasonable to just teach the PCI subsystem about it.
> 
> Agreed. As a bonus, extending the PCI_BUS_NUM_RESOURCES to 6 would
> cleanly resolve problems with "transparent" PCI bridges - the bus
> might have 3 "native" + 3 parent bus ranges in that case.
> 
The concern here isn't just increasing the size of pci_bus. The
resource pointers in pci_bus point to resource structures in the
corresponding pci_dev structure for p2p bridges. If we want to
maintain this scheme, we'd have to increase the number of resources
in the pci_dev structure too, which increases it for every single
pci device in the system. Probably ok for big server machines, but
would others (e.g. embedded folks) complain?

I just realized that I did not explicitly CC Greg in my original
post. Doing that now, to see what he thinks.

Rajesh