Re: [RFC PATCH v1 1/4] arm/pci: PCI setup and PCI host bridge discovery within XEN on ARM.

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Fri, 24 Jul 2020, Julien Grall wrote:
> > > > +    list_add_tail(&bridge->node, &pci_host_bridges);
> > > It looks like &pci_host_bridges should be an ordered list, ordered by
> > > segment number?
> > 
> > Why? Do you expect bridge access in some specific order so ordered
> > 
> > list will make it faster?
> 
> Access to the configure space will be pretty random. So I don't think ordering
> the list will make anything better.
> 
> However, looking up for the bridge for every config spec access is pretty
> slow. When I was working on the PCI passthrough, I wanted to look whether it
> would be possible to have a pointer to the PCI host bridge passed in argument
> to the helpers (rather than the segment).

I was suggesting ordering the list because it is a rather cheap
optimization: it is easy to implement and it typically leads to decent
improvements (although it varies on a case by case basis). Something
more sophisticated as you mention here would be even better.