Re: [Qemu-devel] [PATCH 4/8] qdev/core: bus list

[Paul Brook <paul@codesourcery.com>]

On Tuesday 30 June 2009, Gerd Hoffmann wrote:
> On 06/30/09 21:49, Anthony Liguori wrote:
> > I think it's a perfectly valid suggestion that we should identify buses
> > based on the their location in the tree to users verses a number
> > generated based on some hashing algorithm.
> >
> > A tree location has meaning to a user. A random integer doesn't.
>
> Well.  Depends on the bus I think.  About PCI devices the usual user
> probably thinks in terms of "$bus:$slot.$function", which includes a bus
> number.
>
> Speaking of PCI: the PCI bus number (aka PCIBus->bus_num) has nothing to
> do with the more or less random bus number introduced by the (now
> dropped) patch (aka PCIBus->qdev.busnr).  Which indicates that it is
> probably less confusing to have the bus implementation handle the
> enumeration of busses.  If it makes sense for the bus in question of
> course.  sysbus probably doesn't care ;)

PCI bus numbers should be determined by the guest, and are not fixed values. 
The only reason we get away with it now is because most guests will honor a 
pre-existing bridge device configuration. A guest firmware could choose to 
enumerate the PCI busses in a different order.

Paul