Merge branch 'optimize-openvswitch-flow-looking-up'

Tonghao Zhang says:

====================
optimize openvswitch flow looking up

This series patch optimize openvswitch for performance or simplify
codes.

Patch 1, 2, 4: Port Pravin B Shelar patches to
linux upstream with little changes.

Patch 5, 6, 7: Optimize the flow looking up and
simplify the flow hash.

Patch 8, 9: are bugfix.

The performance test is on Intel Xeon E5-2630 v4.
The test topology is show as below:

+-----------------------------------+
|   +---------------------------+   |
|   | eth0   ovs-switch    eth1 |   | Host0
|   +---------------------------+   |
+-----------------------------------+
      ^                       |
      |                       |
      |                       |
      |                       |
      |                       v
+-----+----+             +----+-----+
| netperf  | Host1       | netserver| Host2
+----------+             +----------+

We use netperf send the 64B packets, and insert 255+ flow-mask:
$ ovs-dpctl add-flow ovs-switch "in_port(1),eth(dst=00:01:00:00:00:00/ff:ff:ff:ff:ff:01),eth_type(0x0800),ipv4(frag=no)" 2
...
$ ovs-dpctl add-flow ovs-switch "in_port(1),eth(dst=00:ff:00:00:00:00/ff:ff:ff:ff:ff:ff),eth_type(0x0800),ipv4(frag=no)" 2
$
$ netperf -t UDP_STREAM -H 2.2.2.200 -l 40 -- -m 18

* Without series patch, throughput 8.28Mbps
* With series patch, throughput 46.05Mbps

v6:
some coding style fixes

v5:
rewrite patch 8, release flow-mask when freeing flow

v4:
access ma->count with READ_ONCE/WRITE_ONCE API. More information,
see patch 5 comments.

v3:
update ma point when realloc mask_array in patch 5

v2:
simplify codes. e.g. use kfree_rcu instead of call_rcu
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

net/openvswitch/datapath.c

@@ -227,7 +227,8 @@ void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 	stats = this_cpu_ptr(dp->stats_percpu);
 
 	/* Look up flow. */
-	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
+	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
+					 &n_mask_hit);
 	if (unlikely(!flow)) {
 		struct dp_upcall_info upcall;
 
@@ -1575,6 +1576,31 @@ static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
 	return 0;
 }
 
+static int ovs_dp_stats_init(struct datapath *dp)
+{
+	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
+	if (!dp->stats_percpu)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static int ovs_dp_vport_init(struct datapath *dp)
+{
+	int i;
+
+	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
+				  sizeof(struct hlist_head),
+				  GFP_KERNEL);
+	if (!dp->ports)
+		return -ENOMEM;
+
+	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(&dp->ports[i]);
+
+	return 0;
+}
+
 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 {
 	struct nlattr **a = info->attrs;
@@ -1583,7 +1609,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 	struct datapath *dp;
 	struct vport *vport;
 	struct ovs_net *ovs_net;
-	int err, i;
+	int err;
 
 	err = -EINVAL;
 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
@@ -1596,35 +1622,26 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 	err = -ENOMEM;
 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
 	if (dp == NULL)
-		goto err_free_reply;
+		goto err_destroy_reply;
 
 	ovs_dp_set_net(dp, sock_net(skb->sk));
 
 	/* Allocate table. */
 	err = ovs_flow_tbl_init(&dp->table);
 	if (err)
-		goto err_free_dp;
+		goto err_destroy_dp;
 
-	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
-	if (!dp->stats_percpu) {
-		err = -ENOMEM;
+	err = ovs_dp_stats_init(dp);
+	if (err)
 		goto err_destroy_table;
-	}
-
-	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
-				  sizeof(struct hlist_head),
-				  GFP_KERNEL);
-	if (!dp->ports) {
-		err = -ENOMEM;
-		goto err_destroy_percpu;
-	}
 
-	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
-		INIT_HLIST_HEAD(&dp->ports[i]);
+	err = ovs_dp_vport_init(dp);
+	if (err)
+		goto err_destroy_stats;
 
 	err = ovs_meters_init(dp);
 	if (err)
-		goto err_destroy_ports_array;
+		goto err_destroy_ports;
 
 	/* Set up our datapath device. */
 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
@@ -1656,6 +1673,7 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 				ovs_dp_reset_user_features(skb, info);
 		}
 
+		ovs_unlock();
 		goto err_destroy_meters;
 	}
 
@@ -1672,17 +1690,16 @@ static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
 	return 0;
 
 err_destroy_meters:
-	ovs_unlock();
 	ovs_meters_exit(dp);
-err_destroy_ports_array:
+err_destroy_ports:
 	kfree(dp->ports);
-err_destroy_percpu:
+err_destroy_stats:
 	free_percpu(dp->stats_percpu);
 err_destroy_table:
 	ovs_flow_tbl_destroy(&dp->table);
-err_free_dp:
+err_destroy_dp:
 	kfree(dp);
-err_free_reply:
+err_destroy_reply:
 	kfree_skb(reply);
 err:
 	return err;

net/openvswitch/flow.h

@@ -166,7 +166,6 @@ struct sw_flow_key_range {
 struct sw_flow_mask {
 	int ref_count;
 	struct rcu_head rcu;
-	struct list_head list;
 	struct sw_flow_key_range range;
 	struct sw_flow_key key;
 };

net/openvswitch/flow_table.h

@@ -22,6 +22,17 @@
 
 #include "flow.h"
 
+struct mask_cache_entry {
+	u32 skb_hash;
+	u32 mask_index;
+};
+
+struct mask_array {
+	struct rcu_head rcu;
+	int count, max;
+	struct sw_flow_mask __rcu *masks[];
+};
+
 struct table_instance {
 	struct hlist_head *buckets;
 	unsigned int n_buckets;
@@ -34,7 +45,8 @@ struct table_instance {
 struct flow_table {
 	struct table_instance __rcu *ti;
 	struct table_instance __rcu *ufid_ti;
-	struct list_head mask_list;
+	struct mask_cache_entry __percpu *mask_cache;
+	struct mask_array __rcu *mask_array;
 	unsigned long last_rehash;
 	unsigned int count;
 	unsigned int ufid_count;
@@ -61,6 +73,7 @@ struct sw_flow *ovs_flow_tbl_dump_next(struct table_instance *table,
 				       u32 *bucket, u32 *idx);
 struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *,
 					  const struct sw_flow_key *,
+					  u32 skb_hash,
 					  u32 *n_mask_hit);
 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *,
 				    const struct sw_flow_key *);