runtime: eliminate false sharing on random number generators

Use machine-local random number generator instead of
racy global ones.

R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/4674049

src/pkg/runtime/chan.c

@@ -1221,17 +1221,6 @@ freesg(Hchan *c, SudoG *sg)
 	}
 }
 
-static uint32
-fastrand1(void)
-{
-	static uint32 x = 0x49f6428aUL;
-
-	x += x;
-	if(x & 0x80000000L)
-		x ^= 0x88888eefUL;
-	return x;
-}
-
 static uint32
 fastrandn(uint32 n)
 {
@@ -1240,12 +1229,12 @@ fastrandn(uint32 n)
 	if(n <= 1)
 		return 0;
 
-	r = fastrand1();
+	r = runtime·fastrand1();
 	if(r < (1ULL<<31)-n)  // avoid computing max in common case
 		return r%n;
 
 	max = (1ULL<<31)/n * n;
 	while(r >= max)
-		r = fastrand1();
+		r = runtime·fastrand1();
 	return r%n;
 }

src/pkg/runtime/malloc.goc

@@ -18,21 +18,6 @@ extern MStats mstats;	// defined in extern.go
 
 extern volatile int32 runtime·MemProfileRate;
 
-// Same algorithm from chan.c, but a different
-// instance of the static uint32 x.
-// Not protected by a lock - let the threads use
-// the same random number if they like.
-static uint32
-fastrand1(void)
-{
-	static uint32 x = 0x49f6428aUL;
-
-	x += x;
-	if(x & 0x80000000L)
-		x ^= 0x88888eefUL;
-	return x;
-}
-
 // Allocate an object of at least size bytes.
 // Small objects are allocated from the per-thread cache's free lists.
 // Large objects (> 32 kB) are allocated straight from the heap.
@@ -97,7 +82,7 @@ runtime·mallocgc(uintptr size, uint32 flag, int32 dogc, int32 zeroed)
 			// pick next profile time
 			if(rate > 0x3fffffff)	// make 2*rate not overflow
 				rate = 0x3fffffff;
-			m->mcache->next_sample = fastrand1() % (2*rate);
+			m->mcache->next_sample = runtime·fastrand1() % (2*rate);
 		profile:
 			runtime·setblockspecial(v);
 			runtime·MProf_Malloc(v, size);

src/pkg/runtime/proc.c

@@ -117,6 +117,7 @@ runtime·schedinit(void)
 
 	runtime·allm = m;
 	m->nomemprof++;
+	m->fastrand = 0x49f6428aUL + m->id;
 
 	runtime·mallocinit();
 	runtime·goargs();
@@ -495,6 +496,7 @@ matchmg(void)
 			m->alllink = runtime·allm;
 			runtime·allm = m;
 			m->id = runtime·sched.mcount++;
+			m->fastrand = 0x49f6428aUL + m->id;
 
 			if(runtime·iscgo) {
 				CgoThreadStart ts;

src/pkg/runtime/runtime.c

@@ -588,3 +588,16 @@ runtime·FuncForPC(uintptr pc, void *retf)
 	retf = runtime·findfunc(pc);
 	FLUSH(&retf);
 }
+
+uint32
+runtime·fastrand1(void)
+{
+	uint32 x;
+
+	x = m->fastrand;
+	x += x;
+	if(x & 0x80000000L)
+		x ^= 0x88888eefUL;
+	m->fastrand = x;
+	return x;
+}

src/pkg/runtime/runtime.h

@@ -229,6 +229,7 @@ struct	M
 	int32	waitnextg;
 	int32	dying;
 	int32	profilehz;
+	uint32	fastrand;
 	Note	havenextg;
 	G*	nextg;
 	M*	alllink;	// on allm
@@ -454,6 +455,7 @@ void	runtime·runpanic(Panic*);
 void*	runtime·getcallersp(void*);
 int32	runtime·mcount(void);
 void	runtime·mcall(void(*)(G*));
+uint32	runtime·fastrand1(void);
 
 void	runtime·exit(int32);
 void	runtime·breakpoint(void);