openvswitch: Use skb_zerocopy() for upcall

Use of skb_zerocopy() can avoid the expensive call to memcpy()
when copying the packet data into the Netlink skb. Completes
checksum through skb_checksum_help() if not already done in
GSO segmentation.

Zerocopy is only performed if user space supported unaligned
Netlink messages. memory mapped netlink i/o is preferred over
zerocopy if it is set up.

Cost of upcall is significantly reduced from:
+   7.48%       vhost-8471  [k] memcpy
+   5.57%     ovs-vswitchd  [k] memcpy
+   2.81%       vhost-8471  [k] csum_partial_copy_generic

to:
+   5.72%     ovs-vswitchd  [k] memcpy
+   3.32%       vhost-5153  [k] memcpy
+   0.68%       vhost-5153  [k] skb_zerocopy

(megaflows disabled)

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: Jesse Gross <jesse@nicira.com>

diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c
index 235acaeaedc7979ad8371c74e2766b94e3e811dc..85578342d5664c53ac69911cfb59ac7dff224710 100644 (file)
--- a/net/openvswitch/datapath.c
+++ b/net/openvswitch/datapath.c
@@ -371,11 +371,11 @@ static size_t key_attr_size(void)
                + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 }
 
-static size_t upcall_msg_size(const struct sk_buff *skb,
-                             const struct nlattr *userdata)
+static size_t upcall_msg_size(const struct nlattr *userdata,
+                             unsigned int hdrlen)
 {
        size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
-               + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
+               + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
                + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 
        /* OVS_PACKET_ATTR_USERDATA */
@@ -397,6 +397,7 @@ static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
                .snd_portid = upcall_info->portid,
        };
        size_t len;
+       unsigned int hlen;
        int err, dp_ifindex;
 
        dp_ifindex = get_dpifindex(dp);
@@ -421,7 +422,21 @@ static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
                goto out;
        }
 
-       len = upcall_msg_size(skb, upcall_info->userdata);
+       /* Complete checksum if needed */
+       if (skb->ip_summed == CHECKSUM_PARTIAL &&
+           (err = skb_checksum_help(skb)))
+               goto out;
+
+       /* Older versions of OVS user space enforce alignment of the last
+        * Netlink attribute to NLA_ALIGNTO which would require extensive
+        * padding logic. Only perform zerocopy if padding is not required.
+        */
+       if (dp->user_features & OVS_DP_F_UNALIGNED)
+               hlen = skb_zerocopy_headlen(skb);
+       else
+               hlen = skb->len;
+
+       len = upcall_msg_size(upcall_info->userdata, hlen);
        user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
        if (!user_skb) {
                err = -ENOMEM;
@@ -441,13 +456,19 @@ static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
                          nla_len(upcall_info->userdata),
                          nla_data(upcall_info->userdata));
 
-       nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
+       /* Only reserve room for attribute header, packet data is added
+        * in skb_zerocopy() */
+       if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
+               err = -ENOBUFS;
+               goto out;
+       }
+       nla->nla_len = nla_attr_size(skb->len);
 
-       skb_copy_and_csum_dev(skb, nla_data(nla));
+       skb_zerocopy(user_skb, skb, skb->len, hlen);
 
-       genlmsg_end(user_skb, upcall);
-       err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
+       ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 
+       err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 out:
        kfree_skb(nskb);
        return err;