net: openvswitch: add flow-mask cache for performance

The idea of this optimization comes from a patch which
is committed in 2014, openvswitch community. The author
is Pravin B Shelar. In order to get high performance, I
implement it again. Later patches will use it.

Pravin B Shelar, says:
| On every packet OVS needs to lookup flow-table with every
| mask until it finds a match. The packet flow-key is first
| masked with mask in the list and then the masked key is
| looked up in flow-table. Therefore number of masks can
| affect packet processing performance.

Link: https://github.com/openvswitch/ovs/commit/5604935e4e1cbc16611d2d97f50b717aa31e8ec5
Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Tested-by: Greg Rose <gvrose8192@gmail.com>
Acked-by: William Tu <u9012063@gmail.com>
Signed-off-by: Pravin B Shelar <pshelar@ovn.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 96 insertions, 13 deletions

diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c
index cf3582c5ed70..3d515c072eb8 100644
--- a/net/openvswitch/flow_table.c
+++ b/net/openvswitch/flow_table.c
@@ -36,6 +36,10 @@
 #define TBL_MIN_BUCKETS		1024
 #define REHASH_INTERVAL		(10 * 60 * HZ)
 
+#define MC_HASH_SHIFT		8
+#define MC_HASH_ENTRIES		(1u << MC_HASH_SHIFT)
+#define MC_HASH_SEGS		((sizeof(uint32_t) * 8) / MC_HASH_SHIFT)
+
 static struct kmem_cache *flow_cache;
 struct kmem_cache *flow_stats_cache __read_mostly;
 
@@ -168,10 +172,15 @@ int ovs_flow_tbl_init(struct flow_table *table)
 {
 	struct table_instance *ti, *ufid_ti;
 
-	ti = table_instance_alloc(TBL_MIN_BUCKETS);
+	table->mask_cache = __alloc_percpu(sizeof(struct mask_cache_entry) *
+					   MC_HASH_ENTRIES,
+					   __alignof__(struct mask_cache_entry));
+	if (!table->mask_cache)
+		return -ENOMEM;
 
+	ti = table_instance_alloc(TBL_MIN_BUCKETS);
 	if (!ti)
-		return -ENOMEM;
+		goto free_mask_cache;
 
 	ufid_ti = table_instance_alloc(TBL_MIN_BUCKETS);
 	if (!ufid_ti)
@@ -187,6 +196,8 @@ int ovs_flow_tbl_init(struct flow_table *table)
 
 free_ti:
 	__table_instance_destroy(ti);
+free_mask_cache:
+	free_percpu(table->mask_cache);
 	return -ENOMEM;
 }
 
@@ -243,6 +254,7 @@ void ovs_flow_tbl_destroy(struct flow_table *table)
 	struct table_instance *ti = rcu_dereference_raw(table->ti);
 	struct table_instance *ufid_ti = rcu_dereference_raw(table->ufid_ti);
 
+	free_percpu(table->mask_cache);
 	table_instance_destroy(ti, ufid_ti, false);
 }
 
@@ -425,7 +437,8 @@ static bool ovs_flow_cmp_unmasked_key(const struct sw_flow *flow,
 
 static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
 					  const struct sw_flow_key *unmasked,
-					  const struct sw_flow_mask *mask)
+					  const struct sw_flow_mask *mask,
+					  u32 *n_mask_hit)
 {
 	struct sw_flow *flow;
 	struct hlist_head *head;
@@ -435,6 +448,8 @@ static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
 	ovs_flow_mask_key(&masked_key, unmasked, false, mask);
 	hash = flow_hash(&masked_key, &mask->range);
 	head = find_bucket(ti, hash);
+	(*n_mask_hit)++;
+
 	hlist_for_each_entry_rcu(flow, head, flow_table.node[ti->node_ver]) {
 		if (flow->mask == mask && flow->flow_table.hash == hash &&
 		    flow_cmp_masked_key(flow, &masked_key, &mask->range))
@@ -443,30 +458,97 @@ static struct sw_flow *masked_flow_lookup(struct table_instance *ti,
 	return NULL;
 }
 
-struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
-				    const struct sw_flow_key *key,
-				    u32 *n_mask_hit)
+static struct sw_flow *flow_lookup(struct flow_table *tbl,
+				   struct table_instance *ti,
+				   const struct sw_flow_key *key,
+				   u32 *n_mask_hit)
 {
-	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 	struct sw_flow_mask *mask;
 	struct sw_flow *flow;
 
-	*n_mask_hit = 0;
 	list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
-		(*n_mask_hit)++;
-		flow = masked_flow_lookup(ti, key, mask);
+		flow = masked_flow_lookup(ti, key, mask, n_mask_hit);
 		if (flow)  /* Found */
 			return flow;
 	}
 	return NULL;
 }
 
+/*
+ * mask_cache maps flow to probable mask. This cache is not tightly
+ * coupled cache, It means updates to  mask list can result in inconsistent
+ * cache entry in mask cache.
+ * This is per cpu cache and is divided in MC_HASH_SEGS segments.
+ * In case of a hash collision the entry is hashed in next segment.
+ * */
+struct sw_flow *ovs_flow_tbl_lookup_stats(struct flow_table *tbl,
+					  const struct sw_flow_key *key,
+					  u32 skb_hash,
+					  u32 *n_mask_hit)
+{
+	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
+	struct mask_cache_entry  *entries, *ce, *del;
+	struct sw_flow *flow;
+	u32 hash = skb_hash;
+	int seg;
+
+	*n_mask_hit = 0;
+	if (unlikely(!skb_hash))
+		return flow_lookup(tbl, ti, key, n_mask_hit);
+
+	del = NULL;
+	entries = this_cpu_ptr(tbl->mask_cache);
+
+	for (seg = 0; seg < MC_HASH_SEGS; seg++) {
+		int index;
+
+		index = hash & (MC_HASH_ENTRIES - 1);
+		ce = &entries[index];
+
+		if (ce->skb_hash == skb_hash) {
+			struct sw_flow_mask *mask;
+			int i;
+
+			i = 0;
+			list_for_each_entry_rcu(mask, &tbl->mask_list, list) {
+				if (ce->mask_index == i++) {
+					flow = masked_flow_lookup(ti, key, mask,
+								  n_mask_hit);
+					if (flow)  /* Found */
+						return flow;
+
+					break;
+				}
+			}
+
+			del = ce;
+			break;
+		}
+
+		if (!del || (del->skb_hash && !ce->skb_hash)) {
+			del = ce;
+		}
+
+		hash >>= MC_HASH_SHIFT;
+	}
+
+	flow = flow_lookup(tbl, ti, key, n_mask_hit);
+
+	if (flow) {
+		del->skb_hash = skb_hash;
+		del->mask_index = (*n_mask_hit - 1);
+	}
+
+	return flow;
+}
+
 struct sw_flow *ovs_flow_tbl_lookup(struct flow_table *tbl,
 				    const struct sw_flow_key *key)
 {
+	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 	u32 __always_unused n_mask_hit;
 
-	return ovs_flow_tbl_lookup_stats(tbl, key, &n_mask_hit);
+	return flow_lookup(tbl, ti, key, &n_mask_hit);
 }
 
 struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
@@ -475,10 +557,11 @@ struct sw_flow *ovs_flow_tbl_lookup_exact(struct flow_table *tbl,
 	struct table_instance *ti = rcu_dereference_ovsl(tbl->ti);
 	struct sw_flow_mask *mask;
 	struct sw_flow *flow;
+	u32 __always_unused n_mask_hit;
 
 	/* Always called under ovs-mutex. */
 	list_for_each_entry(mask, &tbl->mask_list, list) {
-		flow = masked_flow_lookup(ti, match->key, mask);
+		flow = masked_flow_lookup(ti, match->key, mask, &n_mask_hit);
 		if (flow && ovs_identifier_is_key(&flow->id) &&
 		    ovs_flow_cmp_unmasked_key(flow, match))
 			return flow;
@@ -631,7 +714,7 @@ static int flow_mask_insert(struct flow_table *tbl, struct sw_flow *flow,
 			return -ENOMEM;
 		mask->key = new->key;
 		mask->range = new->range;
-		list_add_rcu(&mask->list, &tbl->mask_list);
+		list_add_tail_rcu(&mask->list, &tbl->mask_list);
 	} else {
 		BUG_ON(!mask->ref_count);
 		mask->ref_count++;