Re: [PATCH 2/2] Allow VIRTIO_F_IN_ORDER to be negotiated for vdpa devices

[Eugenio Perez Martin <eperezma@redhat.com>]

On Thu, Feb 17, 2022 at 8:16 AM Jason Wang <jasowang@redhat.com> wrote:
>
> On Tue, Feb 15, 2022 at 3:23 PM Gautam Dawar <gautam.dawar@xilinx.com> wrote:
> >
> > This patch adds the ability to negotiate VIRTIO_F_IN_ORDER bit
> > for vhost-vdpa backend when the underlying device supports this
> > feature.
> > This would aid in reaping performance benefits with HW devices
> > that implement this feature. At the same time, it shouldn't have
> > any negative impact as vhost-vdpa backend doesn't involve any
> > userspace virtqueue operations.
> >
> > Signed-off-by: Gautam Dawar <gdawar@xilinx.com>
> > ---
> >  hw/net/virtio-net.c | 10 ++++++++++
> >  net/vhost-vdpa.c    |  1 +
> >  2 files changed, 11 insertions(+)
> >
> > diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> > index cf8ab0f8af..a1089d06f6 100644
> > --- a/hw/net/virtio-net.c
> > +++ b/hw/net/virtio-net.c
> > @@ -3507,11 +3507,21 @@ static void virtio_net_device_realize(DeviceState *dev, Error **errp)
> >      nc->rxfilter_notify_enabled = 1;
> >
> >     if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) {
> > +        uint64_t features = BIT_ULL(VIRTIO_F_IN_ORDER);
> >          struct virtio_net_config netcfg = {};
> > +
> >          memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN);
> >          vhost_net_set_config(get_vhost_net(nc->peer),
> >              (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER);
> > +
> > +       /*
> > +         * For vhost-vdpa, if underlying device supports IN_ORDER feature,
> > +         * make it available for negotiation.
> > +         */
> > +       features = vhost_net_get_features(get_vhost_net(nc->peer), features);
> > +       n->host_features |= features;
>
> This looks like a hack, considering we will finally support in_order.
> I wonder if it's better to
>
> 1) introduce command line parameters "in_order"
> 2) fail without vhost-vdpa
>
> ?
>

Do you mean this steps?:
- Add the cmdline parameter, defaulting it to false.
- Even if it's set to true in cmdline, set to false as long as the
backend is different from vDPA. Since we're only scoping for net
devices, this means to add this feature bit check at this same place
at virtio_net_device_realize only if device property has been set to
true, actually.

Have I understood correctly?

Thanks!


> Thanks
>
> >      }
> > +
> >      QTAILQ_INIT(&n->rsc_chains);
> >      n->qdev = dev;
> >
> > diff --git a/net/vhost-vdpa.c b/net/vhost-vdpa.c
> > index 25dd6dd975..2886cba5ec 100644
> > --- a/net/vhost-vdpa.c
> > +++ b/net/vhost-vdpa.c
> > @@ -62,6 +62,7 @@ const int vdpa_feature_bits[] = {
> >      VIRTIO_NET_F_CTRL_VQ,
> >      VIRTIO_F_IOMMU_PLATFORM,
> >      VIRTIO_F_RING_PACKED,
> > +    VIRTIO_F_IN_ORDER,
> >      VIRTIO_NET_F_RSS,
> >      VIRTIO_NET_F_HASH_REPORT,
> >      VIRTIO_NET_F_GUEST_ANNOUNCE,
> > --
> > 2.30.1
> >
>