Re: [Qemu-devel] [PATCHv3 5/7] memory: Allow replay of IOMMU mapping notifications

[David Gibson]

On Wed, Sep 30, 2015 at 10:59:52AM +0200, Laurent Vivier wrote:
> 
> 
> On 30/09/2015 04:13, David Gibson wrote:
> > When we have guest visible IOMMUs, we allow notifiers to be registered
> > which will be informed of all changes to IOMMU mappings.  This is used by
> > vfio to keep the host IOMMU mappings in sync with guest IOMMU mappings.
> > 
> > However, unlike with a memory region listener, an iommu notifier won't be
> > told about any mappings which already exist in the (guest) IOMMU at the
> > time it is registered.  This can cause problems if hotplugging a VFIO
> > device onto a guest bus which had existing guest IOMMU mappings, but didn't
> > previously have an VFIO devices (and hence no host IOMMU mappings).
> > 
> > This adds a memory_region_iommu_replay() function to handle this case.  It
> > replays any existing mappings in an IOMMU memory region to a specified
> > notifier.  Because the IOMMU memory region doesn't internally remember the
> > granularity of the guest IOMMU it has a small hack where the caller must
> > specify a granularity at which to replay mappings.
> > 
> > If there are finer mappings in the guest IOMMU these will be reported in
> > the iotlb structures passed to the notifier which it must handle (probably
> > causing it to flag an error).  This isn't new - the VFIO iommu notifier
> > must already handle notifications about guest IOMMU mappings too short
> > for it to represent in the host IOMMU.
> > 
> > Signed-off-by: David Gibson <address@hidden>
> > ---
> >  include/exec/memory.h | 13 +++++++++++++
> >  memory.c              | 20 ++++++++++++++++++++
> >  2 files changed, 33 insertions(+)
> > 
> > diff --git a/include/exec/memory.h b/include/exec/memory.h
> > index 5baaf48..0f07159 100644
> > --- a/include/exec/memory.h
> > +++ b/include/exec/memory.h
> > @@ -583,6 +583,19 @@ void memory_region_notify_iommu(MemoryRegion *mr,
> >  void memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n);
> >  
> >  /**
> > + * memory_region_iommu_replay: replay existing IOMMU translations to
> > + * a notifier
> > + *
> > + * @mr: the memory region to observe
> > + * @n: the notifier to which to replay iommu mappings
> > + * @granularity: Minimum page granularity to replay notifications for
> > + * @is_write: Whether to treat the replay as a translate "write"
> > + *     through the iommu
> > + */
> > +void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
> > +                                hwaddr granularity, bool is_write);
> > +
> > +/**
> >   * memory_region_unregister_iommu_notifier: unregister a notifier for
> >   * changes to IOMMU translation entries.
> >   *
> > diff --git a/memory.c b/memory.c
> > index ef87363..1b03d22 100644
> > --- a/memory.c
> > +++ b/memory.c
> > @@ -1403,6 +1403,26 @@ void 
> > memory_region_register_iommu_notifier(MemoryRegion *mr, Notifier *n)
> >      notifier_list_add(&mr->iommu_notify, n);
> >  }
> >  
> > +void memory_region_iommu_replay(MemoryRegion *mr, Notifier *n,
> > +                                hwaddr granularity, bool is_write)
> > +{
> > +    hwaddr addr;
> > +    IOMMUTLBEntry iotlb;
> > +
> > +    for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
> > +        iotlb = mr->iommu_ops->translate(mr, addr, is_write);
> 
> in iotlb, there is an "address_mask", on spapr, it is copied from
> "page_shift", which is SPAPR_TCE_PAGE_SHIFT (12 -> 4k).
> 
> At a first glance, we would like to use it to scan the memory region,
> but as granularity could be a greater value, I think it is a better choice.

Using address_mask doesn't quite work.  *If* you start with an
existing, valid translation, then you can use address mask to skip to
the end of it - that might be a useful optimization in future,
particularly if the guest IOMMU has variable page sizes.

But if you start on an address that doesn't have a current valid
translation in the IOMMU, then address_mask gets set to ~0, so it
doesn't give you any information on where to try next for a valid
mapping.  That's what the granularity parameter is needed for.

> But the question is: why the iotlb page_size is not equal to the
> granularity given by VFIO_IOMMU_GET_INFO _IO ?

Well, the iotlb page size is the page size from the *guest* iommu,
whereas VFIO_IOMMU_GET_INFO tells you the page size of the *host*
iommu.  In practice, they'll probably be the same, at least on setups
likely to work well with VFIO, but in theory they could be different.

> > +        if (iotlb.perm != IOMMU_NONE) {
> > +            n->notify(n, &iotlb);
> > +        }
> > +
> > +        /* if (2^64 - MR size) < granularity, it's possible to get an
> > +         * infinite loop here.  This should catch such a wraparound */
> > +        if ((addr + granularity) < addr) {
> > +            break;
> > +        }
> > +    }
> > +}
> > +
> >  void memory_region_unregister_iommu_notifier(Notifier *n)
> >  {
> >      notifier_remove(n);
> > 
> 
> As my question is not a bout this particular patch but on another
> existing part, I can say:
> 
> Reviewed-by: Laurent Vivier <address@hidden>
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

pgpDEpfdhJe32.pgp
Description: PGP signature