[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] x86/msi: always propagate MSI writes when not in active system mode



On Tue, Mar 18, 2025 at 09:36:37AM +0100, Jan Beulich wrote:
> On 18.03.2025 09:29, Roger Pau Monne wrote:
> > --- a/xen/drivers/passthrough/amd/iommu_intr.c
> > +++ b/xen/drivers/passthrough/amd/iommu_intr.c
> > @@ -546,7 +546,7 @@ int cf_check amd_iommu_msi_msg_update_ire(
> >      rc = update_intremap_entry_from_msi_msg(iommu, bdf, nr,
> >                                              &msi_desc->remap_index,
> >                                              msg, &data);
> > -    if ( rc > 0 )
> > +    if ( rc >= 0 )
> >      {
> >          for ( i = 1; i < nr; ++i )
> >              msi_desc[i].remap_index = msi_desc->remap_index + i;
> 
> I understand that Marek's testing has made clear that this change is needed,
> yet I don't understand it. If we didn't allocate a new index, why would we
> need to update in-memory state, when memory is preserved across S3?

Is this always the case for device memory? (iow: contents of the BARs
and possibly the PCI config space?)

> (This
> lack of understanding on my part is why I didn't associate the last
> paragraph of the description with this extra change, when you first sent it
> in this shape on the original thread.)

At least for the AMD IOMMU driver it seems to be expected.  See how
amd_iommu_resume() performs a pair of disable_iommu() and
enable_iommu() calls, and in the enable_iommu() function there's a
call to set_{msi,x2apic}_affinity() that's expected to (re)set the
interrupts.  Or at least that would be my understanding.

This change reverts the behavior to what it used to be prior to
8e60d47cf011 for the suspend and resume paths.  I'm afraid I don't
have a sensible way to test changes in that area, so I cannot
investigate much.

Thanks, Roger.



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.