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, 35 insertions, 11 deletions

diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c
index aa92da23053d..d8ef37b937bd 100644
--- a/net/openvswitch/flow_table.c
+++ b/net/openvswitch/flow_table.c
@@ -48,6 +48,7 @@
 #define REHASH_INTERVAL         (10 * 60 * HZ)
 
 static struct kmem_cache *flow_cache;
+struct kmem_cache *flow_stats_cache __read_mostly;
 
 static u16 range_n_bytes(const struct sw_flow_key_range *range)
 {
@@ -75,7 +76,8 @@ void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src,
 struct sw_flow *ovs_flow_alloc(void)
 {
         struct sw_flow *flow;
-        int cpu;
+        struct flow_stats *stats;
+        int node;
 
         flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
         if (!flow)
@@ -83,17 +85,22 @@ struct sw_flow *ovs_flow_alloc(void)
 
         flow->sf_acts = NULL;
         flow->mask = NULL;
+        flow->stats_last_writer = NUMA_NO_NODE;
 
-        flow->stats = alloc_percpu(struct flow_stats);
-        if (!flow->stats)
+        /* Initialize the default stat node. */
+        stats = kmem_cache_alloc_node(flow_stats_cache,
+                                      GFP_KERNEL | __GFP_ZERO, 0);
+        if (!stats)
                 goto err;
 
-        for_each_possible_cpu(cpu) {
-                struct flow_stats *cpu_stats;
+        spin_lock_init(&stats->lock);
+
+        RCU_INIT_POINTER(flow->stats[0], stats);
+
+        for_each_node(node)
+                if (node != 0)
+                        RCU_INIT_POINTER(flow->stats[node], NULL);
 
-                cpu_stats = per_cpu_ptr(flow->stats, cpu);
-                spin_lock_init(&cpu_stats->lock);
-        }
         return flow;
 err:
         kmem_cache_free(flow_cache, flow);
@@ -130,8 +137,13 @@ static struct flex_array *alloc_buckets(unsigned int n_buckets)
 
 static void flow_free(struct sw_flow *flow)
 {
+        int node;
+
         kfree((struct sf_flow_acts __force *)flow->sf_acts);
-        free_percpu(flow->stats);
+        for_each_node(node)
+                if (flow->stats[node])
+                        kmem_cache_free(flow_stats_cache,
+                                        (struct flow_stats __force *)flow->stats[node]);
         kmem_cache_free(flow_cache, flow);
 }
 
@@ -586,16 +598,28 @@ int ovs_flow_init(void)
         BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long));
         BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long));
 
-        flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0,
-                                        0, NULL);
+        flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow)
+                                       + (num_possible_nodes()
+                                          * sizeof(struct flow_stats *)),
+                                       0, 0, NULL);
         if (flow_cache == NULL)
                 return -ENOMEM;
 
+        flow_stats_cache
+                = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats),
+                                    0, SLAB_HWCACHE_ALIGN, NULL);
+        if (flow_stats_cache == NULL) {
+                kmem_cache_destroy(flow_cache);
+                flow_cache = NULL;
+                return -ENOMEM;
+        }
+
         return 0;
 }
 
 /* Uninitializes the flow module. */
 void ovs_flow_exit(void)
 {
+        kmem_cache_destroy(flow_stats_cache);
         kmem_cache_destroy(flow_cache);
 }