Re: [PATCH] PCI: remove pcibios_scan_all_fns()

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Tue, 2009-06-23 at 15:53 -0700, Jeremy Fitzhardinge wrote:
> On 06/23/09 15:41, Benjamin Herrenschmidt wrote:
> >> Do you have any other cases in mind where it would be helpful?
> >>     
> >
> > Well, it might be for virtual device discovery etc... but don't bother
> > now. We might talk about it at KS for those interested. It's more
> > something we see as useful for embedded archs at the moment but in the
> > long run it might make sense for hypervisors as well.
> >   
> 
> Perhaps.  We have Xenbus - which is  a little bit like OF in that it has
> data in a hierarchical namespace - and I guess it might be possible to
> find a mapping onto some generic OF-like interface.

Which is sort-of what we did. IE. We disconnected the device-tree itself
from the underlying firmware, using the device-tree and OF-style
bindings (in some case simplified) as a basis for representing devices
but without the need for an actual open firmware underneath.

> However, Xenbus is an active communication channel between virtual machines
> rather than a
> static representation of a machine configuration (for example, you can
> put a watch on a particular path to get a notification of when someone
> else changes it).

On ppc64 too, the HV can feed us with new tree nodes or remove some, it
doesn't have to be static. Though we mostly use it as a static tree on
embedded.

> But, yes, this is a good KS hallway track subject.

Cheers,
Ben.