Merge branch 'tcp-rx-tx-cache'

Eric Dumazet says:

====================
tcp: add rx/tx cache to reduce lock contention

On hosts with many cpus we can observe a very serious contention
on spinlocks used in mm slab layer.

The following can happen quite often :

1) TX path
  sendmsg() allocates one (fclone) skb on CPU A, sends a clone.
  ACK is received on CPU B, and consumes the skb that was in the retransmit
  queue.

2) RX path
  network driver allocates skb on CPU C
  recvmsg() happens on CPU D, freeing the skb after it has been delivered
  to user space.

In both cases, we are hitting the asymetric alloc/free pattern
for which slab has to drain alien caches. At 8 Mpps per second,
this represents 16 Mpps alloc/free per second and has a huge penalty.

In an interesting experiment, I tried to use a single kmem_cache for all the skbs
(in skb_init() : skbuff_fclone_cache = skbuff_head_cache =
                  kmem_cache_create("skbuff_fclone_cache", sizeof(struct sk_buff_fclones),);
qnd most of the contention disappeared, since cpus could better use
their local slab per-cpu cache.

But we can do actually better, in the following patches.

TX : at ACK time, no longer free the skb but put it back in a tcp socket cache,
     so that next sendmsg() can reuse it immediately.

RX : at recvmsg() time, do not free the skb but put it in a tcp socket cache
   so that it can be freed by the cpu feeding the incoming packets in BH.

This increased the performance of small RPC benchmark by about 10 % on a host
with 112 hyperthreads.

v2 : - Solved a race condition : sk_stream_alloc_skb() to make sure the prior
       clone has been freed.
     - Really test rps_needed in sk_eat_skb() as claimed.
     - Fixed rps_needed use in drivers/net/tun.c

v3: Added a #ifdef CONFIG_RPS, to avoid compile error (kbuild robot)
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 27 insertions, 27 deletions

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 6baa6dc1b13b..29b94edf05f9 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -865,6 +865,21 @@ struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
 {
 	struct sk_buff *skb;
 
+	skb = sk->sk_tx_skb_cache;
+	if (skb && !size) {
+		const struct sk_buff_fclones *fclones;
+
+		fclones = container_of(skb, struct sk_buff_fclones, skb1);
+		if (refcount_read(&fclones->fclone_ref) == 1) {
+			sk->sk_wmem_queued -= skb->truesize;
+			sk_mem_uncharge(sk, skb->truesize);
+			skb->truesize -= skb->data_len;
+			sk->sk_tx_skb_cache = NULL;
+			pskb_trim(skb, 0);
+			INIT_LIST_HEAD(&skb->tcp_tsorted_anchor);
+			return skb;
+		}
+	}
 	/* The TCP header must be at least 32-bit aligned.  */
 	size = ALIGN(size, 4);
 
@@ -1098,30 +1113,6 @@ int tcp_sendpage(struct sock *sk, struct page *page, int offset,
 }
 EXPORT_SYMBOL(tcp_sendpage);
 
-/* Do not bother using a page frag for very small frames.
- * But use this heuristic only for the first skb in write queue.
- *
- * Having no payload in skb->head allows better SACK shifting
- * in tcp_shift_skb_data(), reducing sack/rack overhead, because
- * write queue has less skbs.
- * Each skb can hold up to MAX_SKB_FRAGS * 32Kbytes, or ~0.5 MB.
- * This also speeds up tso_fragment(), since it wont fallback
- * to tcp_fragment().
- */
-static int linear_payload_sz(bool first_skb)
-{
-	if (first_skb)
-		return SKB_WITH_OVERHEAD(2048 - MAX_TCP_HEADER);
-	return 0;
-}
-
-static int select_size(bool first_skb, bool zc)
-{
-	if (zc)
-		return 0;
-	return linear_payload_sz(first_skb);
-}
-
 void tcp_free_fastopen_req(struct tcp_sock *tp)
 {
 	if (tp->fastopen_req) {
@@ -1272,7 +1263,6 @@ restart:
 
 		if (copy <= 0 || !tcp_skb_can_collapse_to(skb)) {
 			bool first_skb;
-			int linear;
 
 new_segment:
 			if (!sk_stream_memory_free(sk))
@@ -1283,8 +1273,7 @@ new_segment:
 				goto restart;
 			}
 			first_skb = tcp_rtx_and_write_queues_empty(sk);
-			linear = select_size(first_skb, zc);
-			skb = sk_stream_alloc_skb(sk, linear, sk->sk_allocation,
+			skb = sk_stream_alloc_skb(sk, 0, sk->sk_allocation,
 						  first_skb);
 			if (!skb)
 				goto wait_for_memory;
@@ -2552,6 +2541,13 @@ void tcp_write_queue_purge(struct sock *sk)
 		sk_wmem_free_skb(sk, skb);
 	}
 	tcp_rtx_queue_purge(sk);
+	skb = sk->sk_tx_skb_cache;
+	if (skb) {
+		sk->sk_wmem_queued -= skb->truesize;
+		sk_mem_uncharge(sk, skb->truesize);
+		__kfree_skb(skb);
+		sk->sk_tx_skb_cache = NULL;
+	}
 	INIT_LIST_HEAD(&tcp_sk(sk)->tsorted_sent_queue);
 	sk_mem_reclaim(sk);
 	tcp_clear_all_retrans_hints(tcp_sk(sk));
@@ -2587,6 +2583,10 @@ int tcp_disconnect(struct sock *sk, int flags)
 
 	tcp_clear_xmit_timers(sk);
 	__skb_queue_purge(&sk->sk_receive_queue);
+	if (sk->sk_rx_skb_cache) {
+		__kfree_skb(sk->sk_rx_skb_cache);
+		sk->sk_rx_skb_cache = NULL;
+	}
 	tp->copied_seq = tp->rcv_nxt;
 	tp->urg_data = 0;
 	tcp_write_queue_purge(sk);