On Mon, Sep 12, 2016 at 2:09 PM, Peter Xu <pet...@redhat.com> wrote:
> On Mon, Sep 12, 2016 at 01:08:05PM +0300, David Kiarie wrote:
>
> [...]
>
> > @@ -2252,14 +2250,17 @@ static MemTxResult vtd_mem_ir_write(void
> *opaque, hwaddr addr,
> > {
> > int ret = 0;
> > MSIMessage from = {}, to = {};
> > - uint16_t sid = X86_IOMMU_SID_INVALID;
> > + VTDAddressSpace *as = opaque;
> > + uint16_t sid = PCI_BUILD_BDF(pci_bus_num(as->bus), as->devfn);
>
> I remembered to have commented on this... PCI_BUILD_BDF() should be
> problematic. SID may not be built that way when with PCI bridges (or
> say, I think current code won't work with PCI bridges). Please see
> commit:
>
> commit 4a94b3aa6d97dfa67a20c7a0315c9773352f0e8e
> Author: Peter Xu <pet...@redhat.com>
> Date: Tue May 17 19:26:10 2016 +0800
>
> pci: fix pci_requester_id()
>
> That's why we explicitly differenciate BDF and SID.
>
> I would suggest to make it simpler: we just do not do this extra
> check, and pass attrs.requester_id to vtd_interrupt_remap_msi()
directly.
>
Yes, we did discuss this but didn't come up with a conclusion so I stuck
with the check.
I will remove the check in the next version.
> Thanks,
>
> -- peterx
>
