From dbcbe5d3b1e3eef64c21a3cf44027d4514b84f11 Mon Sep 17 00:00:00 2001
From: Matthew Dillon <dillon@apollo.backplane.com>
Date: Wed, 2 Sep 2009 22:15:12 -0700
Subject: [PATCH] TCP - Reduce context switching when handling large send*() or
 write()'s

* Large sends or writes would cause sosend() to sleep and wakeup for every
  tcp segment acknowledged, resulting in massive inefficiencies.  To deal
  with this, if no LOWAT is set we automatically set the LOWAT to 1/2 the
  send buffer size and adjust the wakeup code.
---
 sys/kern/uipc_socket.c  |  2 ++
 sys/kern/uipc_socket2.c | 32 ++++++++++++++++++++++++++++----
 sys/sys/socketvar.h     |  1 +
 3 files changed, 31 insertions(+), 4 deletions(-)

diff --git a/sys/kern/uipc_socket.c b/sys/kern/uipc_socket.c
index f6091f3fa1..e6f2ca2260 100644
--- a/sys/kern/uipc_socket.c
+++ b/sys/kern/uipc_socket.c
@@ -1392,11 +1392,13 @@ sosetopt(struct socket *so, struct sockopt *sopt)
 				so->so_snd.ssb_lowat =
 				    (optval > so->so_snd.ssb_hiwat) ?
 				    so->so_snd.ssb_hiwat : optval;
+				so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT;
 				break;
 			case SO_RCVLOWAT:
 				so->so_rcv.ssb_lowat =
 				    (optval > so->so_rcv.ssb_hiwat) ?
 				    so->so_rcv.ssb_hiwat : optval;
+				so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT;
 				break;
 			}
 			break;
diff --git a/sys/kern/uipc_socket2.c b/sys/kern/uipc_socket2.c
index 094e693b01..c2020b2ec4 100644
--- a/sys/kern/uipc_socket2.c
+++ b/sys/kern/uipc_socket2.c
@@ -259,8 +259,10 @@ sonewconn(struct socket *head, int connstatus)
 	so->so_snd.ssb_lowat = head->so_snd.ssb_lowat;
 	so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo;
 	so->so_snd.ssb_timeo = head->so_snd.ssb_timeo;
-	so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags & SSB_AUTOSIZE;
-	so->so_snd.ssb_flags |= head->so_snd.ssb_flags & SSB_AUTOSIZE;
+	so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags &
+				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
+	so->so_snd.ssb_flags |= head->so_snd.ssb_flags &
+				(SSB_AUTOSIZE | SSB_AUTOLOWAT);
 	if (connstatus) {
 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
 		so->so_state |= SS_COMP;
@@ -312,6 +314,10 @@ socantrcvmore(struct socket *so)
 /*
  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
  * via SIGIO if the socket has the SS_ASYNC flag set.
+ *
+ * For users waiting on send/recv try to avoid unnecessary context switch
+ * thrashing.  Particularly for senders of large buffers (needs to be
+ * extended to sel and aio? XXX)
  */
 void
 sowakeup(struct socket *so, struct signalsockbuf *ssb)
@@ -321,8 +327,14 @@ sowakeup(struct socket *so, struct signalsockbuf *ssb)
 	selwakeup(selinfo);
 	ssb->ssb_flags &= ~SSB_SEL;
 	if (ssb->ssb_flags & SSB_WAIT) {
-		ssb->ssb_flags &= ~SSB_WAIT;
-		wakeup((caddr_t)&ssb->ssb_cc);
+		if ((ssb == &so->so_snd && ssb_space(ssb) >= ssb->ssb_lowat) ||
+		    (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) ||
+		    (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) ||
+		    (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE))
+		) {
+			ssb->ssb_flags &= ~SSB_WAIT;
+			wakeup((caddr_t)&ssb->ssb_cc);
+		}
 	}
 	if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
 		pgsigio(so->so_sigio, SIGIO, 0);
@@ -380,6 +392,8 @@ sowakeup(struct socket *so, struct signalsockbuf *ssb)
 int
 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl)
 {
+	if (so->so_snd.ssb_lowat == 0)
+		so->so_snd.ssb_flags |= SSB_AUTOLOWAT;
 	if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0)
 		goto bad;
 	if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0)
@@ -446,6 +460,16 @@ ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so,
 		ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max);
 	else
 		ssb->ssb_mbmax = cc * sb_efficiency;
+
+	/*
+	 * AUTOLOWAT is set on send buffers and prevents large writes
+	 * from generating a huge number of context switches.
+	 */
+	if (ssb->ssb_flags & SSB_AUTOLOWAT) {
+		ssb->ssb_lowat = ssb->ssb_hiwat / 2;
+		if (ssb->ssb_lowat < MCLBYTES)
+			ssb->ssb_lowat = MCLBYTES;
+	}
 	if (ssb->ssb_lowat > ssb->ssb_hiwat)
 		ssb->ssb_lowat = ssb->ssb_hiwat;
 	return (1);
diff --git a/sys/sys/socketvar.h b/sys/sys/socketvar.h
index aeb97afcbf..d7190bd6d9 100644
--- a/sys/sys/socketvar.h
+++ b/sys/sys/socketvar.h
@@ -85,6 +85,7 @@ struct signalsockbuf {
 #define SSB_MEVENT	0x200		/* need message event notification */
 #define SSB_STOP	0x400		/* backpressure indicator */
 #define	SSB_AUTOSIZE	0x800		/* automatically size socket buffer */
+#define SSB_AUTOLOWAT	0x1000		/* automatically scale lowat */
 
 /*
  * Per-socket kernel structure.  Contains universal send and receive queues,
-- 
2.11.4.GIT