On Tue, Sep 29, 2020 at 3:32 PM Willem de Bruijn
[off-list ref] wrote:

On Tue, Sep 29, 2020 at 4:00 AM [off-list ref] wrote:

quoted

From: Tonghao Zhang <redacted>

Open vSwitch and Linux bridge will disable LRO of the interface
when this interface added to them. Now when disable the LRO, the
virtio-net csum is disable too. That drops the forwarding performance.

I had focused on the code previously.

The s/w checksum verification cost is significant in a VM with traffic
to local destinations. A bridge does not verify transport layer
checksums OTOH?

Hi Willem.
No, think about GRO(In the GRO we don't know packets will be forwarded
to other ports or to local). The call tree as below:
   + 5.41% secondary_startup_64
   - 1.22% ret_from_fork
....
        net_rx_action
        napi_poll
        virtnet_poll
        virtnet_receive
        napi_gro_receive
        dev_gro_receive
        inet_gro_receive
        tcp4_gro_receive
        __skb_gro_checksum_complete
        skb_checksum
        __skb_checksum
        csum_partial
        do_csum
   - 1.13% do_csum

$ brctl show
bridge name bridge id STP enabled interfaces
br0 8000.001122330001 no eth1
eth2

quoted

Fixes: a02e8964eaf9 ("virtio-net: ethtool configurable LRO")
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: Tonghao Zhang <redacted>
---
v2:
* change the fix-tag
---
 drivers/net/virtio_net.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 7145c83c6c8c..21b71148c532 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -63,6 +63,11 @@ static const unsigned long guest_offloads[] = {
        VIRTIO_NET_F_GUEST_CSUM
 };

+#define GUEST_OFFLOAD_LRO_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
+                               (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
+                               (1ULL << VIRTIO_NET_F_GUEST_ECN)  | \
+                               (1ULL << VIRTIO_NET_F_GUEST_UFO))
+
 struct virtnet_stat_desc {
        char desc[ETH_GSTRING_LEN];
        size_t offset;

@@ -2531,7 +2536,8 @@ static int virtnet_set_features(struct net_device *dev,
                if (features & NETIF_F_LRO)
                        offloads = vi->guest_offloads_capable;
                else
-                       offloads = 0;
+                       offloads = vi->guest_offloads_capable &
+                                  ~GUEST_OFFLOAD_LRO_MASK;

                err = virtnet_set_guest_offloads(vi, offloads);
                if (err)
--

2.23.0

-- 
Best regards, Tonghao