runtime: simplify MCache

The nlistmin/size thresholds are copied from tcmalloc,
but are unnecesary for Go malloc. We do not do explicit
frees into MCache. For sparse cases when we do (mainly hashmap),
simpler logic will do.

R=rsc, dave, iant
CC=gobot, golang-dev, r, remyoudompheng
https://golang.org/cl/9373043

src/pkg/runtime/malloc.h

@@ -109,8 +109,6 @@ enum
 	MaxSmallSize = 32<<10,
 
 	FixAllocChunk = 128<<10,	// Chunk size for FixAlloc
-	MaxMCacheListLen = 256,		// Maximum objects on MCacheList
-	MaxMCacheSize = 2<<20,		// Maximum bytes in one MCache
 	MaxMHeapList = 1<<(20 - PageShift),	// Maximum page length for fixed-size list in MHeap.
 	HeapAllocChunk = 1<<20,		// Chunk size for heap growth
 
@@ -283,13 +281,11 @@ struct MCacheList
 {
 	MLink *list;
 	uint32 nlist;
-	uint32 nlistmin;
 };
 
 struct MCache
 {
 	MCacheList list[NumSizeClasses];
-	uintptr size;
 	intptr local_cachealloc;	// bytes allocated (or freed) from cache since last lock of heap
 	intptr local_objects;	// objects allocated (or freed) from cache since last lock of heap
 	intptr local_alloc;	// bytes allocated (or freed) since last lock of heap
@@ -396,7 +392,7 @@ struct MCentral
 
 void	runtime·MCentral_Init(MCentral *c, int32 sizeclass);
 int32	runtime·MCentral_AllocList(MCentral *c, MLink **first);
-void	runtime·MCentral_FreeList(MCentral *c, int32 n, MLink *first);
+void	runtime·MCentral_FreeList(MCentral *c, MLink *first);
 void	runtime·MCentral_FreeSpan(MCentral *c, MSpan *s, int32 n, MLink *start, MLink *end);
 
 // Main malloc heap.

src/pkg/runtime/mcache.c

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// Per-thread (in Go, per-M) malloc cache for small objects.
+// Per-P malloc cache for small objects.
 //
 // See malloc.h for an overview.
 
@@ -14,26 +14,19 @@ void*
 runtime·MCache_Alloc(MCache *c, int32 sizeclass, uintptr size, int32 zeroed)
 {
 	MCacheList *l;
-	MLink *first, *v;
-	int32 n;
+	MLink *v;
 
 	// Allocate from list.
 	l = &c->list[sizeclass];
 	if(l->list == nil) {
 		// Replenish using central lists.
-		n = runtime·MCentral_AllocList(&runtime·mheap->central[sizeclass], &first);
-		if(n == 0)
+		l->nlist = runtime·MCentral_AllocList(&runtime·mheap->central[sizeclass], &l->list);
+		if(l->list == nil)
 			runtime·throw("out of memory");
-		l->list = first;
-		l->nlist = n;
-		c->size += n*size;
 	}
 	v = l->list;
 	l->list = v->next;
 	l->nlist--;
-	if(l->nlist < l->nlistmin)
-		l->nlistmin = l->nlist;
-	c->size -= size;
 
 	// v is zeroed except for the link pointer
 	// that we used above; zero that.
@@ -50,7 +43,7 @@ runtime·MCache_Alloc(MCache *c, int32 sizeclass, uintptr size, int32 zeroed)
 
 // Take n elements off l and return them to the central free list.
 static void
-ReleaseN(MCache *c, MCacheList *l, int32 n, int32 sizeclass)
+ReleaseN(MCacheList *l, int32 n, int32 sizeclass)
 {
 	MLink *first, **lp;
 	int32 i;
@@ -63,18 +56,14 @@ ReleaseN(MCache *c, MCacheList *l, int32 n, int32 sizeclass)
 	l->list = *lp;
 	*lp = nil;
 	l->nlist -= n;
-	if(l->nlist < l->nlistmin)
-		l->nlistmin = l->nlist;
-	c->size -= n*runtime·class_to_size[sizeclass];
 
 	// Return them to central free list.
-	runtime·MCentral_FreeList(&runtime·mheap->central[sizeclass], n, first);
+	runtime·MCentral_FreeList(&runtime·mheap->central[sizeclass], first);
 }
 
 void
 runtime·MCache_Free(MCache *c, void *v, int32 sizeclass, uintptr size)
 {
-	int32 i, n;
 	MCacheList *l;
 	MLink *p;
 
@@ -84,34 +73,13 @@ runtime·MCache_Free(MCache *c, void *v, int32 sizeclass, uintptr size)
 	p->next = l->list;
 	l->list = p;
 	l->nlist++;
-	c->size += size;
 	c->local_cachealloc -= size;
 	c->local_objects--;
 
-	if(l->nlist >= MaxMCacheListLen) {
-		// Release a chunk back.
-		ReleaseN(c, l, l->nlist/2, sizeclass);
-	}
-
-	if(c->size >= MaxMCacheSize) {
-		// Scavenge.
-		for(i=0; i<NumSizeClasses; i++) {
-			l = &c->list[i];
-			n = l->nlistmin;
-
-			// n is the minimum number of elements we've seen on
-			// the list since the last scavenge.  If n > 0, it means that
-			// we could have gotten by with n fewer elements
-			// without needing to consult the central free list.
-			// Move toward that situation by releasing n/2 of them.
-			if(n > 0) {
-				if(n > 1)
-					n /= 2;
-				ReleaseN(c, l, n, i);
-			}
-			l->nlistmin = l->nlist;
-		}
-	}
+	// We transfer span at a time from MCentral to MCache,
+	// if we have 2 times more than that, release a half back.
+	if(l->nlist >= 2*(runtime·class_to_allocnpages[sizeclass]<<PageShift)/size)
+		ReleaseN(l, l->nlist/2, sizeclass);
 }
 
 void
@@ -122,7 +90,10 @@ runtime·MCache_ReleaseAll(MCache *c)
 
 	for(i=0; i<NumSizeClasses; i++) {
 		l = &c->list[i];
-		ReleaseN(c, l, l->nlist, i);
-		l->nlistmin = 0;
+		if(l->list) {
+			runtime·MCentral_FreeList(&runtime·mheap->central[i], l->list);
+			l->list = nil;
+			l->nlist = 0;
+		}
 	}
 }

src/pkg/runtime/mcentral.c

@@ -62,21 +62,16 @@ runtime·MCentral_AllocList(MCentral *c, MLink **pfirst)
 	return n;
 }
 
-// Free n objects back into the central free list.
+// Free the list of objects back into the central free list.
 void
-runtime·MCentral_FreeList(MCentral *c, int32 n, MLink *start)
+runtime·MCentral_FreeList(MCentral *c, MLink *start)
 {
-	MLink *v, *next;
-
-	// Assume next == nil marks end of list.
-	// n and end would be useful if we implemented
-	// the transfer cache optimization in the TODO above.
-	USED(n);
+	MLink *next;
 
 	runtime·lock(c);
-	for(v=start; v; v=next) {
-		next = v->next;
-		MCentral_Free(c, v);
+	for(; start != nil; start = next) {
+		next = start->next;
+		MCentral_Free(c, start);
 	}
 	runtime·unlock(c);
 }