Re: [Qemu-devel] [RFC PATCH] pci: Use PCI aliases when determining device IOMMU address space

[Michael S. Tsirkin]

On Thu, Jul 25, 2019 at 08:00:23AM -0600, Alex Williamson wrote:
> On Thu, 25 Jul 2019 06:43:04 -0400
> "Michael S. Tsirkin" <address@hidden> wrote:
> 
> > On Thu, Jul 25, 2019 at 02:37:33PM +0800, Peter Xu wrote:
> > > After re-read Michael's reply, I feel like what Michael suggested is
> > > that we can simply ignore the bus-number-change case by the guest OS
> > > for now, but I might be wrong.  
> > That's what I suggested, yes.
> 
> "by the guest OS", yes, that's the part that's beyond the scope of this
> effort.  If we deem it necessary to support IVHD DMA alias though, it's
> the guest firmware that determines the initial bus numbers, which is
> part of the DeviceID used to defined IVHD entries and we would not be
> able to ignore that initial programming.  Everything related to the
> guest OS renumber PCI buses is out of scope, the guest firmware
> programming initial bus numbers is not.  

Right. That's par for the course, we have same issues with MCFG
and others. bios programs it and then we generate acpi based on
that.

>Thanks,
> 
> Alex