Re: [PATCH net-next v2] virtio-net: ethtool configurable RXCSUM

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

On Fri, Oct 9, 2020 at 10:10 PM Tonghao Zhang <xiangxia.m.yue@gmail.com> wrote:
>
> On Fri, Oct 9, 2020 at 9:49 PM Willem de Bruijn
> <willemdebruijn.kernel@gmail.com> wrote:
> >
> > On Thu, Oct 8, 2020 at 9:19 PM Tonghao Zhang <xiangxia.m.yue@gmail.com> wrote:
> > >
> > > On Wed, Sep 30, 2020 at 6:05 PM Willem de Bruijn
> > > <willemdebruijn.kernel@gmail.com> wrote:
> > > >
> > > > On Wed, Sep 30, 2020 at 4:05 AM <xiangxia.m.yue@gmail.com> wrote:
> > > > >
> > > > > From: Tonghao Zhang <xiangxia.m.yue@gmail.com>
> > > > >
> > > > > Allow user configuring RXCSUM separately with ethtool -K,
> > > > > reusing the existing virtnet_set_guest_offloads helper
> > > > > that configures RXCSUM for XDP. This is conditional on
> > > > > VIRTIO_NET_F_CTRL_GUEST_OFFLOADS.
> > > > >
> > > > > If Rx checksum is disabled, LRO should also be disabled.
> > > > >
> > > > > Cc: Michael S. Tsirkin <mst@redhat.com>
> > > > > Cc: Jason Wang <jasowang@redhat.com>
> > > > > Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
> > > >
> > > > The first patch was merged into net.
> > > >
> > > > Please wait until that is merged into net-next, as this depends on the
> > > > other patch.
> > > >
> > > > > ---
> > > > > v2:
> > > > > * LRO depends the rx csum
> > > > > * remove the unnecessary check
> > > > > ---
> > > > >  drivers/net/virtio_net.c | 49 ++++++++++++++++++++++++++++++----------
> > > > >  1 file changed, 37 insertions(+), 12 deletions(-)
> > > > >
> > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > > > index 21b71148c532..5407a0106771 100644
> > > > > --- a/drivers/net/virtio_net.c
> > > > > +++ b/drivers/net/virtio_net.c
> > > > > @@ -68,6 +68,8 @@ static const unsigned long guest_offloads[] = {
> > > > >                                 (1ULL << VIRTIO_NET_F_GUEST_ECN)  | \
> > > > >                                 (1ULL << VIRTIO_NET_F_GUEST_UFO))
> > > > >
> > > > > +#define GUEST_OFFLOAD_CSUM_MASK (1ULL << VIRTIO_NET_F_GUEST_CSUM)
> > > > > +
> > > > >  struct virtnet_stat_desc {
> > > > >         char desc[ETH_GSTRING_LEN];
> > > > >         size_t offset;
> > > > > @@ -2522,29 +2524,49 @@ static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
> > > > >         return 0;
> > > > >  }
> > > > >
> > > > > +static netdev_features_t virtnet_fix_features(struct net_device *netdev,
> > > > > +                                             netdev_features_t features)
> > > > > +{
> > > > > +       /* If Rx checksum is disabled, LRO should also be disabled.
> > > > > +        * That is life. :)
> > > >
> > > > Please remove this second line.
> > > OK
> > > > > +        */
> > > > > +       if (!(features & NETIF_F_RXCSUM))
> > > > > +               features &= ~NETIF_F_LRO;
> > > > > +
> > > > > +       return features;
> > > > > +}
> > > > > +
> > > > >  static int virtnet_set_features(struct net_device *dev,
> > > > >                                 netdev_features_t features)
> > > > >  {
> > > > >         struct virtnet_info *vi = netdev_priv(dev);
> > > > > -       u64 offloads;
> > > > > +       u64 offloads = vi->guest_offloads &
> > > > > +                      vi->guest_offloads_capable;
> > > >
> > > > When can vi->guest_offloads have bits set outside the mask of
> > > > vi->guest_offloads_capable?
> > > In this case, we can use only vi->guest_offloads. and
> > > guest_offloads_capable will not be used any more.
> > > so should we remove guest_offloads_capable ?
> >
> > That bitmap stores the capabilities of the device as learned at
> > initial device probe. I don't see how it can be removed.
> Hi Willem
> guest_offloads_capable was introduced by
> a02e8964eaf9 ("virtio-net: ethtool configurable LRO")
> and used only for LRO. Now we don't use it anymore, right ?
> because this patch (virtio-net: ethtool configurable RXCSUM)
> doesn't use it.

It is needed, because it serves as an upper bound on configurable options.

virtnet_set_features cannot enable LRO unless the LRO flags are
captured by the feature negotiation at probe time.

I think on enable you need something like

-                       offloads = vi->guest_offloads_capable;
+                       offloads |= vi->guest_offloads_capable &
GUEST_OFFLOAD_LRO_MASK;

instead of unconditional

+                       offloads |= GUEST_OFFLOAD_LRO_MASK;