openvswitch: Per NUMA node flow stats.

Keep kernel flow stats for each NUMA node rather than each (logical)
CPU.  This avoids using the per-CPU allocator and removes most of the
kernel-side OVS locking overhead otherwise on the top of perf reports
and allows OVS to scale better with higher number of threads.

With 9 handlers and 4 revalidators netperf TCP_CRR test flow setup
rate doubles on a server with two hyper-threaded physical CPUs (16
logical cores each) compared to the current OVS master.  Tested with
non-trivial flow table with a TCP port match rule forcing all new
connections with unique port numbers to OVS userspace.  The IP
addresses are still wildcarded, so the kernel flows are not considered
as exact match 5-tuple flows.  This type of flows can be expected to
appear in large numbers as the result of more effective wildcarding
made possible by improvements in OVS userspace flow classifier.

Perf results for this test (master):

Events: 305K cycles
+   8.43%     ovs-vswitchd  [kernel.kallsyms]   [k] mutex_spin_on_owner
+   5.64%     ovs-vswitchd  [kernel.kallsyms]   [k] __ticket_spin_lock
+   4.75%     ovs-vswitchd  ovs-vswitchd        [.] find_match_wc
+   3.32%     ovs-vswitchd  libpthread-2.15.so  [.] pthread_mutex_lock
+   2.61%     ovs-vswitchd  [kernel.kallsyms]   [k] pcpu_alloc_area
+   2.19%     ovs-vswitchd  ovs-vswitchd        [.] flow_hash_in_minimask_range
+   2.03%          swapper  [kernel.kallsyms]   [k] intel_idle
+   1.84%     ovs-vswitchd  libpthread-2.15.so  [.] pthread_mutex_unlock
+   1.64%     ovs-vswitchd  ovs-vswitchd        [.] classifier_lookup
+   1.58%     ovs-vswitchd  libc-2.15.so        [.] 0x7f4e6
+   1.07%     ovs-vswitchd  [kernel.kallsyms]   [k] memset
+   1.03%          netperf  [kernel.kallsyms]   [k] __ticket_spin_lock
+   0.92%          swapper  [kernel.kallsyms]   [k] __ticket_spin_lock
...

And after this patch:

Events: 356K cycles
+   6.85%     ovs-vswitchd  ovs-vswitchd        [.] find_match_wc
+   4.63%     ovs-vswitchd  libpthread-2.15.so  [.] pthread_mutex_lock
+   3.06%     ovs-vswitchd  [kernel.kallsyms]   [k] __ticket_spin_lock
+   2.81%     ovs-vswitchd  ovs-vswitchd        [.] flow_hash_in_minimask_range
+   2.51%     ovs-vswitchd  libpthread-2.15.so  [.] pthread_mutex_unlock
+   2.27%     ovs-vswitchd  ovs-vswitchd        [.] classifier_lookup
+   1.84%     ovs-vswitchd  libc-2.15.so        [.] 0x15d30f
+   1.74%     ovs-vswitchd  [kernel.kallsyms]   [k] mutex_spin_on_owner
+   1.47%          swapper  [kernel.kallsyms]   [k] intel_idle
+   1.34%     ovs-vswitchd  ovs-vswitchd        [.] flow_hash_in_minimask
+   1.33%     ovs-vswitchd  ovs-vswitchd        [.] rule_actions_unref
+   1.16%     ovs-vswitchd  ovs-vswitchd        [.] hindex_node_with_hash
+   1.16%     ovs-vswitchd  ovs-vswitchd        [.] do_xlate_actions
+   1.09%     ovs-vswitchd  ovs-vswitchd        [.] ofproto_rule_ref
+   1.01%          netperf  [kernel.kallsyms]   [k] __ticket_spin_lock
...

There is a small increase in kernel spinlock overhead due to the same
spinlock being shared between multiple cores of the same physical CPU,
but that is barely visible in the netperf TCP_CRR test performance
(maybe ~1% performance drop, hard to tell exactly due to variance in
the test results), when testing for kernel module throughput (with no
userspace activity, handful of kernel flows).

On flow setup, a single stats instance is allocated (for the NUMA node
0).  As CPUs from multiple NUMA nodes start updating stats, new
NUMA-node specific stats instances are allocated.  This allocation on
the packet processing code path is made to never block or look for
emergency memory pools, minimizing the allocation latency.  If the
allocation fails, the existing preallocated stats instance is used.
Also, if only CPUs from one NUMA-node are updating the preallocated
stats instance, no additional stats instances are allocated.  This
eliminates the need to pre-allocate stats instances that will not be
used, also relieving the stats reader from the burden of reading stats
that are never used.

Signed-off-by: Jarno Rajahalme <jrajahalme@nicira.com>
Acked-by: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Jesse Gross <jesse@nicira.com>

1 files changed, 77 insertions, 42 deletions

diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c
index aad7a8d..432f04d 100644
--- a/net/openvswitch/flow.c
+++ b/net/openvswitch/flow.c
@@ -65,8 +65,9 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
 {
 	struct flow_stats *stats;
 	__be16 tcp_flags = 0;
+	int node = numa_node_id();
 
-	stats = this_cpu_ptr(flow->stats);
+	stats = rcu_dereference(flow->stats[node]);
 
 	if ((flow->key.eth.type == htons(ETH_P_IP) ||
 	     flow->key.eth.type == htons(ETH_P_IPV6)) &&
@@ -76,68 +77,102 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
 		tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb));
 	}
 
-	spin_lock(&stats->lock);
+	/* Check if already have node-specific stats. */
+	if (likely(stats)) {
+		spin_lock(&stats->lock);
+		/* Mark if we write on the pre-allocated stats. */
+		if (node == 0 && unlikely(flow->stats_last_writer != node))
+			flow->stats_last_writer = node;
+	} else {
+		stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */
+		spin_lock(&stats->lock);
+
+		/* If the current NUMA-node is the only writer on the
+		 * pre-allocated stats keep using them.
+		 */
+		if (unlikely(flow->stats_last_writer != node)) {
+			/* A previous locker may have already allocated the
+			 * stats, so we need to check again.  If node-specific
+			 * stats were already allocated, we update the pre-
+			 * allocated stats as we have already locked them.
+			 */
+			if (likely(flow->stats_last_writer != NUMA_NO_NODE)
+			    && likely(!rcu_dereference(flow->stats[node]))) {
+				/* Try to allocate node-specific stats. */
+				struct flow_stats *new_stats;
+
+				new_stats =
+					kmem_cache_alloc_node(flow_stats_cache,
+							      GFP_THISNODE |
+							      __GFP_NOMEMALLOC,
+							      node);
+				if (likely(new_stats)) {
+					new_stats->used = jiffies;
+					new_stats->packet_count = 1;
+					new_stats->byte_count = skb->len;
+					new_stats->tcp_flags = tcp_flags;
+					spin_lock_init(&new_stats->lock);
+
+					rcu_assign_pointer(flow->stats[node],
+							   new_stats);
+					goto unlock;
+				}
+			}
+			flow->stats_last_writer = node;
+		}
+	}
+
 	stats->used = jiffies;
 	stats->packet_count++;
 	stats->byte_count += skb->len;
 	stats->tcp_flags |= tcp_flags;
-	spin_unlock(&stats->lock);
-}
-
-static void stats_read(struct flow_stats *stats,
-		       struct ovs_flow_stats *ovs_stats,
-		       unsigned long *used, __be16 *tcp_flags)
-{
-	spin_lock(&stats->lock);
-	if (!*used || time_after(stats->used, *used))
-		*used = stats->used;
-	*tcp_flags |= stats->tcp_flags;
-	ovs_stats->n_packets += stats->packet_count;
-	ovs_stats->n_bytes += stats->byte_count;
+unlock:
 	spin_unlock(&stats->lock);
 }
 
 void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
 			unsigned long *used, __be16 *tcp_flags)
 {
-	int cpu;
+	int node;
 
 	*used = 0;
 	*tcp_flags = 0;
 	memset(ovs_stats, 0, sizeof(*ovs_stats));
 
-	local_bh_disable();
-
-	for_each_possible_cpu(cpu) {
-		struct flow_stats *stats;
+	for_each_node(node) {
+		struct flow_stats *stats = rcu_dereference(flow->stats[node]);
 
-		stats = per_cpu_ptr(flow->stats.cpu_stats, cpu);
-		stats_read(stats, ovs_stats, used, tcp_flags);
+		if (stats) {
+			/* Local CPU may write on non-local stats, so we must
+			 * block bottom-halves here.
+			 */
+			spin_lock_bh(&stats->lock);
+			if (!*used || time_after(stats->used, *used))
+				*used = stats->used;
+			*tcp_flags |= stats->tcp_flags;
+			ovs_stats->n_packets += stats->packet_count;
+			ovs_stats->n_bytes += stats->byte_count;
+			spin_unlock_bh(&stats->lock);
+		}
 	}
-
-	local_bh_enable();
-}
-
-static void stats_reset(struct flow_stats *stats)
-{
-	spin_lock(&stats->lock);
-	stats->used = 0;
-	stats->packet_count = 0;
-	stats->byte_count = 0;
-	stats->tcp_flags = 0;
-	spin_unlock(&stats->lock);
 }
 
 void ovs_flow_stats_clear(struct sw_flow *flow)
 {
-	int cpu;
-
-	local_bh_disable();
-
-	for_each_possible_cpu(cpu)
-		stats_reset(per_cpu_ptr(flow->stats, cpu));
-
-	local_bh_enable();
+	int node;
+
+	for_each_node(node) {
+		struct flow_stats *stats = rcu_dereference(flow->stats[node]);
+
+		if (stats) {
+			spin_lock_bh(&stats->lock);
+			stats->used = 0;
+			stats->packet_count = 0;
+			stats->byte_count = 0;
+			stats->tcp_flags = 0;
+			spin_unlock_bh(&stats->lock);
+		}
+	}
 }
 
 static int check_header(struct sk_buff *skb, int len)