Re: [RFC v4 2/2] memory: Skip bad range assertion if notifier is DEVIOTLB type

[Peter Xu]

On Wed, Aug 19, 2020 at 07:47:32PM +0200, Eugenio Perez Martin wrote:
> On Wed, Aug 19, 2020 at 6:40 PM Peter Xu <peterx@redhat.com> wrote:
> >
> > On Tue, Aug 18, 2020 at 03:01:51PM +0200, Eugenio Pérez wrote:
> > > Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
> >
> > The changes on the callers of memory_region_notify_one_iommu() seems to be
> > still missing (and, to embed the type into the notification process)..
> >
> 
> Hi Peter.
> 
> I thought that these were left for a future patch series (the main
> motivation was to avoid for guest code to hit the assertion).
> 
> Do you want me to put them in this series?

Imho it would be good to do that altogether.

For example, you have defined:

   /* Notify changes on device IOTLB entries */
   IOMMU_NOTIFIER_DEVIOTLB = 0x04,

The comment says we'll only notify on device iotlbs, however with the code
change we'll still notify even with normal iotlb invalidations (but always with
the type IOMMU_NOTIFIER_DEVIOTLB ).  IOW vhost with ats=on will still receive
two invalidations for the same range just like before.

-- 
Peter Xu