void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb)
{
- struct sw_flow_stats *stats = &flow->stats[smp_processor_id()];
+ struct flow_stats *stats;
__be16 tcp_flags = 0;
+ int node = numa_node_id();
+
+ stats = rcu_dereference(flow->stats[node]);
if ((flow->key.eth.type == htons(ETH_P_IP) ||
flow->key.eth.type == htons(ETH_P_IPV6)) &&
+ flow->key.ip.frag != OVS_FRAG_TYPE_LATER &&
flow->key.ip.proto == IPPROTO_TCP &&
likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) {
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;
+unlock:
spin_unlock(&stats->lock);
}
-void ovs_flow_stats_get(struct sw_flow *flow, struct sw_flow_stats *res)
+void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats,
+ unsigned long *used, __be16 *tcp_flags)
{
- int cpu, cur_cpu;
-
- memset(res, 0, sizeof(*res));
+ int node;
- cur_cpu = get_cpu();
- for_each_possible_cpu(cpu) {
- struct sw_flow_stats *stats = &flow->stats[cpu];
+ *used = 0;
+ *tcp_flags = 0;
+ memset(ovs_stats, 0, sizeof(*ovs_stats));
- if (cpu == cur_cpu)
- local_bh_disable();
-
- spin_lock(&stats->lock);
- if (time_after(stats->used, res->used))
- res->used = stats->used;
- res->packet_count += stats->packet_count;
- res->byte_count += stats->byte_count;
- res->tcp_flags |= stats->tcp_flags;
- spin_unlock(&stats->lock);
-
- if (cpu == cur_cpu)
- local_bh_enable();
+ for_each_node(node) {
+ struct flow_stats *stats = rcu_dereference(flow->stats[node]);
+ if (stats) {
+ /* Local CPU may write on non-local stats, so we must
+ * block bottom-halves here.
+ */
+ spin_lock_bh(&stats->lock);
+ if (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);
+ }
}
- put_cpu();
}
void ovs_flow_stats_clear(struct sw_flow *flow)
{
- int cpu, cur_cpu;
-
- cur_cpu = get_cpu();
- for_each_possible_cpu(cpu) {
- struct sw_flow_stats *stats = &flow->stats[cpu];
-
- if (cpu == cur_cpu)
- local_bh_disable();
-
- spin_lock(&stats->lock);
- stats->used = 0;
- stats->packet_count = 0;
- stats->byte_count = 0;
- stats->tcp_flags = 0;
- spin_unlock(&stats->lock);
-
- if (cpu == cur_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);
+ }
}
- put_cpu();
}
static int check_header(struct sk_buff *skb, int len)