From 5c3032598344555e117f863f9c4227f5659ce3ab Mon Sep 17 00:00:00 2001
From: Hector Chu <hectorchu@gmail.com>
Date: Wed, 14 Sep 2011 20:23:21 -0400
Subject: [PATCH] runtime: eliminate handle churn when churning channels on
 Windows

The Windows implementation of the net package churns through a couple of channels for every read/write operation.  This translates into a lot of time spent in the kernel creating and deleting event objects.

R=rsc, dvyukov, alex.brainman, jp
CC=golang-dev
https://golang.org/cl/4997044
---
 src/pkg/runtime/mgc0.c           |  2 --
 src/pkg/runtime/runtime.h        |  9 ++++--
 src/pkg/runtime/windows/thread.c | 51 ++++++++++++++++++++++----------
 3 files changed, 42 insertions(+), 20 deletions(-)

diff --git a/src/pkg/runtime/mgc0.c b/src/pkg/runtime/mgc0.c
index e79317bf6f..03d6f7d629 100644
--- a/src/pkg/runtime/mgc0.c
+++ b/src/pkg/runtime/mgc0.c
@@ -603,8 +603,6 @@ runtime·gc(int32 force)
 
 	m->gcing = 1;
 	runtime·stoptheworld();
-	if(runtime·mheap.Lock.key != 0)
-		runtime·throw("runtime·mheap locked during gc");
 
 	cachestats();
 	heap0 = mstats.heap_alloc;
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index 8753842a01..25751b80e1 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -119,10 +119,10 @@ enum
  */
 struct	Lock
 {
-	uint32	key;
 #ifdef __WINDOWS__
-	void*	event;
+	M*	waitm;	// linked list of waiting M's
 #else
+	uint32	key;
 	uint32	sema;	// for OS X
 #endif
 };
@@ -251,6 +251,11 @@ struct	M
 	uint32	freglo[16];	// D[i] lsb and F[i]
 	uint32	freghi[16];	// D[i] msb and F[i+16]
 	uint32	fflag;		// floating point compare flags
+
+#ifdef __WINDOWS__
+	void*	event;		// event for signalling
+	M*	nextwaitm;	// next M waiting for lock
+#endif
 };
 
 struct	Stktop
diff --git a/src/pkg/runtime/windows/thread.c b/src/pkg/runtime/windows/thread.c
index fe8a24f1cd..33637f1d7a 100644
--- a/src/pkg/runtime/windows/thread.c
+++ b/src/pkg/runtime/windows/thread.c
@@ -40,14 +40,12 @@ extern void *runtime·WaitForSingleObject;
 extern void *runtime·WriteFile;
 
 static int64 timerfreq;
-static void destroylock(Lock *l);
 
 void
 runtime·osinit(void)
 {
 	runtime·stdcall(runtime·QueryPerformanceFrequency, 1, &timerfreq);
 	runtime·stdcall(runtime·SetConsoleCtrlHandler, 2, runtime·ctrlhandler, (uintptr)1);
-	runtime·destroylock = destroylock;
 }
 
 void
@@ -120,22 +118,50 @@ initevent(void **pevent)
 	}
 }
 
+#define LOCK_HELD ((M*)-1)
+
 static void
 eventlock(Lock *l)
 {
 	// Allocate event if needed.
-	if(l->event == 0)
-		initevent(&l->event);
+	if(m->event == nil)
+		initevent(&m->event);
+
+	for(;;) {
+		m->nextwaitm = runtime·atomicloadp(&l->waitm);
+		if(m->nextwaitm == nil) {
+			if(runtime·casp(&l->waitm, nil, LOCK_HELD))
+				return;
+		// Someone else has it.
+		// l->waitm points to a linked list of M's waiting
+		// for this lock, chained through m->nextwaitm.
+		// Queue this M.
+		} else if(runtime·casp(&l->waitm, m->nextwaitm, m))
+			break;
+	}
 
-	if(runtime·xadd(&l->key, 1) > 1)	// someone else has it; wait
-		runtime·stdcall(runtime·WaitForSingleObject, 2, l->event, (uintptr)-1);
+	// Wait.
+	runtime·stdcall(runtime·WaitForSingleObject, 2, m->event, (uintptr)-1);
 }
 
 static void
 eventunlock(Lock *l)
 {
-	if(runtime·xadd(&l->key, -1) > 0)	// someone else is waiting
-		runtime·stdcall(runtime·SetEvent, 1, l->event);
+	M *mp;
+
+	for(;;) {
+		mp = runtime·atomicloadp(&l->waitm);
+		if(mp == LOCK_HELD) {
+			if(runtime·casp(&l->waitm, LOCK_HELD, nil))
+				return;
+		// Other M's are waiting for the lock.
+		// Dequeue a M.
+		} else if(runtime·casp(&l->waitm, mp, mp->nextwaitm))
+			break;
+	}
+
+	// Wake that M.
+	runtime·stdcall(runtime·SetEvent, 1, mp->event);
 }
 
 void
@@ -156,17 +182,10 @@ runtime·unlock(Lock *l)
 	eventunlock(l);
 }
 
-static void
-destroylock(Lock *l)
-{
-	if(l->event != 0)
-		runtime·stdcall(runtime·CloseHandle, 1, l->event);
-}
-
 void
 runtime·noteclear(Note *n)
 {
-	n->lock.key = 0;	// memset(n, 0, sizeof *n)
+	n->lock.waitm = nil;
 	eventlock(&n->lock);
 }
 
-- 
2.48.1