Re: [PATCH] vdpa: do not save failed dma maps in SVQ iova tree

[Eugenio Perez Martin]

On Thu, Aug 4, 2022 at 5:02 PM Eugenio Perez Martin <eperezma@redhat.com> wrote:
>
> On Thu, Aug 4, 2022 at 6:51 AM Jason Wang <jasowang@redhat.com> wrote:
> >
> >
> > 在 2022/8/3 16:12, Eugenio Perez Martin 写道:
> > > On Wed, Aug 3, 2022 at 10:09 AM Jason Wang <jasowang@redhat.com> wrote:
> > >> On Tue, Aug 2, 2022 at 10:39 PM Eugenio Pérez <eperezma@redhat.com> 
> > >> wrote:
> > >>> If a map fails for whatever reason, it must not be saved in the tree.
> > >>> Otherwise, qemu will try to unmap it in cleanup, leaving to more errors.
> > >>>
> > >>> Fixes: 34e3c94eda ("vdpa: Add custom IOTLB translations to SVQ")
> > >>> Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
> > >>> ---
> > >>>   hw/virtio/vhost-vdpa.c | 20 +++++++++++++-------
> > >>>   1 file changed, 13 insertions(+), 7 deletions(-)
> > >>>
> > >>> diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> > >>> index 3ff9ce3501..e44c23dce5 100644
> > >>> --- a/hw/virtio/vhost-vdpa.c
> > >>> +++ b/hw/virtio/vhost-vdpa.c
> > >>> @@ -176,6 +176,7 @@ static void 
> > >>> vhost_vdpa_listener_commit(MemoryListener *listener)
> > >>>   static void vhost_vdpa_listener_region_add(MemoryListener *listener,
> > >>>                                              MemoryRegionSection 
> > >>> *section)
> > >>>   {
> > >>> +    DMAMap mem_region = {};
> > >>>       struct vhost_vdpa *v = container_of(listener, struct vhost_vdpa, 
> > >>> listener);
> > >>>       hwaddr iova;
> > >>>       Int128 llend, llsize;
> > >>> @@ -212,13 +213,13 @@ static void 
> > >>> vhost_vdpa_listener_region_add(MemoryListener *listener,
> > >>>
> > >>>       llsize = int128_sub(llend, int128_make64(iova));
> > >>>       if (v->shadow_vqs_enabled) {
> > >>> -        DMAMap mem_region = {
> > >>> -            .translated_addr = (hwaddr)(uintptr_t)vaddr,
> > >>> -            .size = int128_get64(llsize) - 1,
> > >>> -            .perm = IOMMU_ACCESS_FLAG(true, section->readonly),
> > >>> -        };
> > >> Nit: can we keep this part unchanged?
> > >>
> > > We can, but that implies we should look for iova again at fail_map
> > > tag. If you are ok with that I'm fine to perform the search again.
> >
> >
> > I meant something like:
> >
> > diff --git a/hw/virtio/vhost-vdpa.c b/hw/virtio/vhost-vdpa.c
> > index 9a2daef7e3..edf40868e3 100644
> > --- a/hw/virtio/vhost-vdpa.c
> > +++ b/hw/virtio/vhost-vdpa.c
> > @@ -232,11 +232,15 @@ static void
> > vhost_vdpa_listener_region_add(MemoryListener *listener,
> >                                vaddr, section->readonly);
> >       if (ret) {
> >           error_report("vhost vdpa map fail!");
> > -        goto fail;
> > +        goto fail_unmap;
> >       }
> >
> >       return;
> >
> > +fail_unmap:
> > +    if (v->shadow_vqs_enabled) {
> > +        vhost_iova_tree_remove(v->iova_tree, &mem_region);
> > +    }
> >   fail:
> >       /*
> >        * On the initfn path, store the first error in the container so we
> >
>
> Sorry, still not following.
>
> mem_region does not exist in the error path, since it's declared in
> the if (v->shadow_vqs_enabled){} block. We can left first part
> unchanged if we do a lookup for the mem region, based on the
> translated addr.
>

Sending v2 expanding the fix without this comment, please let me know
ifI misunderstood something.

Thanks!