diff --git a/src/pkg/runtime/malloc.goc b/src/pkg/runtime/malloc.goc
index 2dff981fb40..babe4d2f4cc 100644
--- a/src/pkg/runtime/malloc.goc
+++ b/src/pkg/runtime/malloc.goc
@@ -76,7 +76,7 @@ runtime·mallocgc(uintptr size, uint32 flag, int32 dogc, int32 zeroed)
 	if (sizeof(void*) == 4 && c->local_total_alloc >= (1<<30)) {
 		// purge cache stats to prevent overflow
 		runtime·lock(&runtime·mheap);
-		runtime·purgecachedstats(m);
+		runtime·purgecachedstats(c);
 		runtime·unlock(&runtime·mheap);
 	}
 
@@ -181,7 +181,7 @@ runtime·mlookup(void *v, byte **base, uintptr *size, MSpan **sp)
 	if (sizeof(void*) == 4 && m->mcache->local_nlookup >= (1<<30)) {
 		// purge cache stats to prevent overflow
 		runtime·lock(&runtime·mheap);
-		runtime·purgecachedstats(m);
+		runtime·purgecachedstats(m->mcache);
 		runtime·unlock(&runtime·mheap);
 	}
 
@@ -234,6 +234,7 @@ runtime·allocmcache(void)
 	mstats.mcache_inuse = runtime·mheap.cachealloc.inuse;
 	mstats.mcache_sys = runtime·mheap.cachealloc.sys;
 	runtime·unlock(&runtime·mheap);
+	runtime·memclr((byte*)c, sizeof(*c));
 
 	// Set first allocation sample size.
 	rate = runtime·MemProfileRate;
@@ -246,12 +247,19 @@ runtime·allocmcache(void)
 }
 
 void
-runtime·purgecachedstats(M* m)
+runtime·freemcache(MCache *c)
 {
-	MCache *c;
+	runtime·MCache_ReleaseAll(c);
+	runtime·lock(&runtime·mheap);
+	runtime·purgecachedstats(c);
+	runtime·FixAlloc_Free(&runtime·mheap.cachealloc, c);
+	runtime·unlock(&runtime·mheap);
+}
 
+void
+runtime·purgecachedstats(MCache *c)
+{
 	// Protected by either heap or GC lock.
-	c = m->mcache;
 	mstats.heap_alloc += c->local_cachealloc;
 	c->local_cachealloc = 0;
 	mstats.heap_objects += c->local_objects;
diff --git a/src/pkg/runtime/malloc.h b/src/pkg/runtime/malloc.h
index f2408f18f2f..fee6e017890 100644
--- a/src/pkg/runtime/malloc.h
+++ b/src/pkg/runtime/malloc.h
@@ -401,7 +401,7 @@ void	runtime·markspan(void *v, uintptr size, uintptr n, bool leftover);
 void	runtime·unmarkspan(void *v, uintptr size);
 bool	runtime·blockspecial(void*);
 void	runtime·setblockspecial(void*, bool);
-void	runtime·purgecachedstats(M*);
+void	runtime·purgecachedstats(MCache*);
 
 enum
 {
diff --git a/src/pkg/runtime/mgc0.c b/src/pkg/runtime/mgc0.c
index 70d0a0f2e28..147c78ad80e 100644
--- a/src/pkg/runtime/mgc0.c
+++ b/src/pkg/runtime/mgc0.c
@@ -823,7 +823,8 @@ cachestats(GCStats *stats)
 	stacks_inuse = 0;
 	stacks_sys = 0;
 	for(m=runtime·allm; m; m=m->alllink) {
-		runtime·purgecachedstats(m);
+		c = m->mcache;
+		runtime·purgecachedstats(c);
 		stacks_inuse += m->stackalloc->inuse;
 		stacks_sys += m->stackalloc->sys;
 		if(stats) {
@@ -833,7 +834,6 @@ cachestats(GCStats *stats)
 				dst[i] += src[i];
 			runtime·memclr((byte*)&m->gcstats, sizeof(m->gcstats));
 		}
-		c = m->mcache;
 		for(i=0; i<nelem(c->local_by_size); i++) {
 			mstats.by_size[i].nmalloc += c->local_by_size[i].nmalloc;
 			c->local_by_size[i].nmalloc = 0;
diff --git a/src/pkg/runtime/mheap.c b/src/pkg/runtime/mheap.c
index a8a435b20e9..1ea74890495 100644
--- a/src/pkg/runtime/mheap.c
+++ b/src/pkg/runtime/mheap.c
@@ -71,7 +71,7 @@ runtime·MHeap_Alloc(MHeap *h, uintptr npage, int32 sizeclass, int32 acct, int32
 	MSpan *s;
 
 	runtime·lock(h);
-	runtime·purgecachedstats(m);
+	runtime·purgecachedstats(m->mcache);
 	s = MHeap_AllocLocked(h, npage, sizeclass);
 	if(s != nil) {
 		mstats.heap_inuse += npage<<PageShift;
@@ -271,7 +271,7 @@ void
 runtime·MHeap_Free(MHeap *h, MSpan *s, int32 acct)
 {
 	runtime·lock(h);
-	runtime·purgecachedstats(m);
+	runtime·purgecachedstats(m->mcache);
 	mstats.heap_inuse -= s->npages<<PageShift;
 	if(acct) {
 		mstats.heap_alloc -= s->npages<<PageShift;
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index cdd71726e5a..07c1585c778 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -590,6 +590,7 @@ int32	runtime·funcline(Func*, uintptr);
 void*	runtime·stackalloc(uint32);
 void	runtime·stackfree(void*, uintptr);
 MCache*	runtime·allocmcache(void);
+void	runtime·freemcache(MCache*);
 void	runtime·mallocinit(void);
 bool	runtime·ifaceeq_c(Iface, Iface);
 bool	runtime·efaceeq_c(Eface, Eface);