Re: [Qemu-devel] [PATCH] pci: fix requester id with PCI bridges

[Peter Xu <peterx@redhat.com>]

On Thu, May 12, 2016 at 10:11:31AM +0300, Michael S. Tsirkin wrote:
> On Wed, May 11, 2016 at 02:53:18PM +0800, Peter Xu wrote:
> > On Wed, May 11, 2016 at 02:40:30PM +0800, Peter Xu wrote:
> > > Recently I encountered issue when debugging Intel IOMMU IR codes,
> > > that interrupts are not working correctly with PCI bridges (reported
> > > by Radim). This patch fixes the problem. I assume requester ID
> > > should be the devfn on root PCI bus (that's how I understand it
> > > before, and also in guest kernel, IRTE entry SID is filled in that
> > > way), however I failed to find any good document to confirm
> > > this. Please let me know if this is correct (or I made any
> > > mistake).
> > 
> > One thing to mention is that, this patch does *not* fix the problem
> > if directly applied, because IR patchset introduced another patch
> > that also need a similar fix. In case if there is someone (Radim?)
> > who would like to try this patch, we need to apply this patch as
> > well:
> > 
> > ---8<---
> > 
> > diff --git a/target-i386/kvm.c b/target-i386/kvm.c
> > index 80b3251..0876a1c 100644
> > --- a/target-i386/kvm.c
> > +++ b/target-i386/kvm.c
> > @@ -3342,7 +3342,8 @@ int kvm_arch_fixup_msi_route(struct kvm_irq_routing_entry *route,
> >          src.data = route->u.msi.data;
> > 
> >          ret = class->int_remap(iommu, &src, &dst, dev ? \
> > -                               pci_requester_id(dev) : X86_IOMMU_SID_INVALID);
> > +                               pci_requester_id_recursive(dev) : \
> > +                               X86_IOMMU_SID_INVALID);
> >          if (ret) {
> >              trace_kvm_x86_fixup_msi_error(route->gsi);
> >              return 1;
> > 
> > --->8---
> > 
> > This should be able to be applied directly onto IR v6 patchset as
> > well.
> > 
> > Thanks,
> > 
> > -- peterx
> 
> I really dislike the name pci_requester_id_recursive.

Do you have better suggestion on this? :)

> Are there cases where we need the original requester id value?
> I am guessing not, and if I'm right we should just change the implementation
> of pci_requester_id to DTRT.

There are two other callers for it (besides the MSI one):

- assigned_device_pci_cap_init()
- do_pcie_aer_inject_error()

For both the cases, I am not sure whether we can do the replacement.

-- peterx