Re: [Qemu-devel] [PATCH] hw/pci: completed master-abort emulation

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Sep 25, 2013 at 08:36:09AM +0900, Peter Maydell wrote:
> On 25 September 2013 00:41, Michael S. Tsirkin <mst@redhat.com> wrote:
> > On Tue, Sep 24, 2013 at 08:21:50PM +0900, Peter Maydell wrote:
> > Right. You might be able to use MR hierarchy to find the last bridge to
> > claim transaction.  That should to be enough for PCI, for express you
> > then need to find all devices on the path between bus master and the
> > last bridge.  For this task, memory subsystem can't be used I think.
> 
> Mmm, if you need to do something for every device on the
> path then that's more awkward with MRs.
>
> > Whether the result will be cleaner than open-coding it all, I don't
> > really know.
> 
> Yes. I was just suggesting it was worth considering at least to
> the point of being able to make a decision about which design
> is likely to be cleaner. If PCI express needs the dynamic walk
> of the path then it's probably better to handle plain PCI that
> way too for consistency.
> 
> -- PMM

It would seem so. The express spec is kind of murky about this
but experiments with real hardware seem to confirm this.

-- 
MST