On Wed, Jan 18, 2017 at 10:42:48AM +0800, Jason Wang wrote: > > > On 2017年01月17日 22:44, Michael S. Tsirkin wrote: > > On Tue, Jan 17, 2017 at 12:01:00PM +0800, Jason Wang wrote: > > > We allow vhost to clear VIRITO_F_IOMMU_PLATFORM which is wrong since > > > VIRTIO_F_IOMMU_PLATFORM is mandatory for security. Fixing this by > > > enforce it after vdc->get_features(). > > > > > > Signed-off-by: Jason Wang <jasow...@redhat.com> > > > --- > > > hw/virtio/virtio-bus.c | 5 +++-- > > > 1 file changed, 3 insertions(+), 2 deletions(-) > > > > > > diff --git a/hw/virtio/virtio-bus.c b/hw/virtio/virtio-bus.c > > > index d31cc00..a886011 100644 > > > --- a/hw/virtio/virtio-bus.c > > > +++ b/hw/virtio/virtio-bus.c > > > @@ -47,6 +47,7 @@ void virtio_bus_device_plugged(VirtIODevice *vdev, > > > Error **errp) > > > VirtioBusState *bus = VIRTIO_BUS(qbus); > > > VirtioBusClass *klass = VIRTIO_BUS_GET_CLASS(bus); > > > VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev); > > > + bool has_iommu = virtio_host_has_feature(vdev, > > > VIRTIO_F_IOMMU_PLATFORM); > > > DPRINTF("%s: plug device.\n", qbus->name); > > > @@ -63,8 +64,8 @@ void virtio_bus_device_plugged(VirtIODevice *vdev, > > > Error **errp) > > > klass->device_plugged(qbus->parent, errp); > > > } > > > - if (klass->get_dma_as != NULL && > > > - virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) { > > > + if (klass->get_dma_as != NULL && has_iommu) { > > > + virtio_add_feature(&vdev->host_features, > > > VIRTIO_F_IOMMU_PLATFORM); > > > vdev->dma_as = klass->get_dma_as(qbus->parent); > > > } else { > > > vdev->dma_as = &address_space_memory; > > I suspect that's not enough, we must also fail or disable vhost > > (depending on the options), otherwise things won't work. > > Looks like with the patch, VIRTIO_F_IOMMU_PLATFORM will be passed to > vhost_set_features().
Okay then. Could you please test on an old host kernel and confirm what happens? Pls mention this in commit log. > So if vhost backend does not support it, it will fall > back to userspace (This may not work for vhost-user, but it's a bug existed > even before this patch). > > Thanks I guess this is true but this suddenly makes it important to fix this. So I prefer a patchset where patch 2 addresses the fallback bug. In which way? Do you think we should just fail if vhost doesn't work? Does this apply to both userspace and kernel? -- MST