Re: [Qemu-devel] [PATCH v2] pci: fix pci_requester_id()

[Peter Xu <peterx@redhat.com>]

On Mon, May 16, 2016 at 08:20:46PM +0300, Michael S. Tsirkin wrote:
[...]
> > > Actually I am curious about how generic PCI system delivers
> > > requester ID (if there is)... For PCIe, we have encoded TLP header,
> > > and requester ID is filled in the specific field of the header.
> > > However for legacy PCI system, all the data is transmitted via the
> > > parallel interface (no matter 32/64 bits) and I found no place that
> > > the requester ID can be included. I was assuming there is some way
> > > for the root complex to know it (when request comes, the root
> > > complex should be able to know where the request come from, or say,
> > > its connected BDF). Never digger into the details, or am I wrong?
> > 
> > There's no such thing as a requester ID on conventional PCI.  We should
> > probably be making use of pci_bus_is_express() to determine whether we
> > have a valid requester ID and error if we hit pci_bus_is_root() and we
> > still don't have an express bus.  And as MST says, testing for
> > bus number zero is not a valid test for the root bus.  Thanks,
> > 
> > Alex
> 
> Well MSI code sticks the requester ID in the MSI message
> unconditionally. It's typically later ignored by the the
> PC machine type though.

Yes, I can first detect root PCI bus type as mentioned by Alex, and
fill in SID_INVALID if it's a conventional PCI root complex.

-- peterx