openvswitch: do not propagate headroom updates to internal port
authorPaolo Abeni <pabeni@redhat.com>
Thu, 30 Nov 2017 14:35:33 +0000 (15:35 +0100)
committerDavid S. Miller <davem@davemloft.net>
Sun, 3 Dec 2017 02:14:59 +0000 (21:14 -0500)
commit183dea5818315c0a172d21ecbcd2554894bf01e3
treeef7a0a843aea2434716659d40fb141630022baf4
parentc5f66a85899705835f61d687a38f62d5a1ec4eb9
openvswitch: do not propagate headroom updates to internal port

After commit 3a927bc7cf9d ("ovs: propagate per dp max headroom to
all vports") the need_headroom for the internal vport is updated
accordingly to the max needed headroom in its datapath.

That avoids the pskb_expand_head() costs when sending/forwarding
packets towards tunnel devices, at least for some scenarios.

We still require such copy when using the ovs-preferred configuration
for vxlan tunnels:

    br_int
  /       \
tap      vxlan
           (remote_ip:X)

br_phy
     \
    NIC

where the route towards the IP 'X' is via 'br_phy'.

When forwarding traffic from the tap towards the vxlan device, we
will call pskb_expand_head() in vxlan_build_skb() because
br-phy->needed_headroom is equal to tun->needed_headroom.

With this change we avoid updating the internal vport needed_headroom,
so that in the above scenario no head copy is needed, giving 5%
performance improvement in UDP throughput test.

As a trade-off, packets sent from the internal port towards a tunnel
device will now experience the head copy overhead. The rationale is
that the latter use-case is less relevant performance-wise.

Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/openvswitch/vport-internal_dev.c