> >Yep. Do you think it's possible to simply chop these devices out of the
> >after initialisation but before passing up to the generic PCI layer?
> >be excellent.
> That'd be OS-specific, no?
Yep. With the new PCI architecture with dom0 owning the PCI bus, this stuff
is going to need implementing in each dom0 OS. The same goes for any dom0
code but since 3.0 pushes more stuff up to dom0, there'll be more stuff to
It's a tradeoff, though: in return for this, we get better hardware support,
reduce code duplication and make Xen's codebase smaller.
> I mean, do we know that all kernels in this world have a clean ability to
> Support this? Despite that this is probably not what the kernel developers
> expected, thus their pci-codes may look strange after adding such feature,
> I think.
> (totally unsure because I actually never wrote kernel code yet ;))
No, it's a good point. It may be that some kernels will be better or worse
suited to this. However, for a portable OS it's probably possible to do some
tricks in the arch/ subdir. If any other code gets confused by this, it may
need fixing, however it's only the core PCI code that'll be affected rather
than all PCI drivers - it should be possible to audit the interface to check
it won't break.
As an alternative, if the ability to ignore certain PCI devices is useful
outside Xen (e.g. for excluding faulty devices), it could be added to the
core PCI code, then all systems would benefit from it.
Xen-devel mailing list