aboutsummaryrefslogtreecommitdiffstats
path: root/net/ipv4/tcp_output.c
diff options
context:
space:
mode:
authorEric Dumazet <edumazet@google.com>2016-12-03 14:14:52 -0500
committerDavid S. Miller <davem@davemloft.net>2016-12-05 13:32:22 -0500
commitb223feb9de2a65c533ff95c08e834fa732906ea5 (patch)
tree0327e31d3fe77301791b56fd3a4082cb3229737e /net/ipv4/tcp_output.c
parent408f0a6c21e124cc4f6c7aa370b38aa47e55428d (diff)
tcp: tsq: add shortcut in tcp_tasklet_func()
Under high stress, I've seen tcp_tasklet_func() consuming ~700 usec, handling ~150 tcp sockets. By setting TCP_TSQ_DEFERRED in tcp_wfree(), we give a chance for other cpus/threads entering tcp_write_xmit() to grab it, allowing tcp_tasklet_func() to skip sockets that already did an xmit cycle. In the future, we might give to ACK processing an increased budget to reduce even more tcp_tasklet_func() amount of work. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/ipv4/tcp_output.c')
-rw-r--r--net/ipv4/tcp_output.c22
1 files changed, 12 insertions, 10 deletions
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 4adaf8e1bb63..fa23b688a6f3 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -767,19 +767,19 @@ static void tcp_tasklet_func(unsigned long data)
767 list_for_each_safe(q, n, &list) { 767 list_for_each_safe(q, n, &list) {
768 tp = list_entry(q, struct tcp_sock, tsq_node); 768 tp = list_entry(q, struct tcp_sock, tsq_node);
769 list_del(&tp->tsq_node); 769 list_del(&tp->tsq_node);
770 clear_bit(TSQ_QUEUED, &tp->tsq_flags);
770 771
771 sk = (struct sock *)tp; 772 sk = (struct sock *)tp;
772 bh_lock_sock(sk); 773 if (!sk->sk_lock.owned &&
773 774 test_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags)) {
774 if (!sock_owned_by_user(sk)) { 775 bh_lock_sock(sk);
775 tcp_tsq_handler(sk); 776 if (!sock_owned_by_user(sk)) {
776 } else { 777 clear_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
777 /* defer the work to tcp_release_cb() */ 778 tcp_tsq_handler(sk);
778 set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags); 779 }
780 bh_unlock_sock(sk);
779 } 781 }
780 bh_unlock_sock(sk);
781 782
782 clear_bit(TSQ_QUEUED, &tp->tsq_flags);
783 sk_free(sk); 783 sk_free(sk);
784 } 784 }
785} 785}
@@ -884,7 +884,7 @@ void tcp_wfree(struct sk_buff *skb)
884 if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED)) 884 if (!(oval & TSQF_THROTTLED) || (oval & TSQF_QUEUED))
885 goto out; 885 goto out;
886 886
887 nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED; 887 nval = (oval & ~TSQF_THROTTLED) | TSQF_QUEUED | TCPF_TSQ_DEFERRED;
888 nval = cmpxchg(&tp->tsq_flags, oval, nval); 888 nval = cmpxchg(&tp->tsq_flags, oval, nval);
889 if (nval != oval) 889 if (nval != oval)
890 continue; 890 continue;
@@ -2229,6 +2229,8 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
2229 unlikely(tso_fragment(sk, skb, limit, mss_now, gfp))) 2229 unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
2230 break; 2230 break;
2231 2231
2232 if (test_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags))
2233 clear_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
2232 if (tcp_small_queue_check(sk, skb, 0)) 2234 if (tcp_small_queue_check(sk, skb, 0))
2233 break; 2235 break;
2234 2236