libgo/runtime/malloc.goc - gcc - Git at Google

 // Copyright 2009 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

 // See malloc.h for overview.
 //
 // TODO(rsc): double-check stats.

 package runtime
 #include <stddef.h>
 #include <errno.h>
 #include <stdlib.h>
 #include "go-alloc.h"
 #include "runtime.h"
 #include "arch.h"
 #include "malloc.h"
 #include "go-string.h"
 #include "interface.h"
 #include "go-type.h"

 MHeap runtime_mheap;

 extern MStats mstats;	// defined in extern.go

 extern volatile int32 runtime_MemProfileRate
   __asm__ ("runtime.MemProfileRate");

 // 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.
 void*
 runtime_mallocgc(uintptr size, uint32 flag, int32 dogc, int32 zeroed)
 {
 	M *m;
 	G *g;
 	int32 sizeclass, rate;
 	MCache *c;
 	uintptr npages;
 	MSpan *s;
 	void *v;

 	m = runtime_m();
 	g = runtime_g();
 	if(g->status == Gsyscall)
 		dogc = 0;
 	if(runtime_gcwaiting && g != m->g0 && m->locks == 0 && g->status != Gsyscall) {
 		runtime_gosched();
 		m = runtime_m();
 	}
 	if(m->mallocing)
 		runtime_throw("malloc/free - deadlock");
 	m->mallocing = 1;
 	if(size == 0)
 		size = 1;

 	c = m->mcache;
 	c->local_nmalloc++;
 	if(size <= MaxSmallSize) {
 		// Allocate from mcache free lists.
 		sizeclass = runtime_SizeToClass(size);
 		size = runtime_class_to_size[sizeclass];
 		v = runtime_MCache_Alloc(c, sizeclass, size, zeroed);
 		if(v == nil)
 			runtime_throw("out of memory");
 		c->local_alloc += size;
 		c->local_total_alloc += size;
 		c->local_by_size[sizeclass].nmalloc++;
 	} else {
 		// TODO(rsc): Report tracebacks for very large allocations.

 		// Allocate directly from heap.
 		npages = size >> PageShift;
 		if((size & PageMask) != 0)
 			npages++;
 		s = runtime_MHeap_Alloc(&runtime_mheap, npages, 0, 1);
 		if(s == nil)
 			runtime_throw("out of memory");
 		size = npages<<PageShift;
 		c->local_alloc += size;
 		c->local_total_alloc += size;
 		v = (void*)(s->start << PageShift);

 		// setup for mark sweep
 		runtime_markspan(v, 0, 0, true);
 	}
 	if(!(flag & FlagNoGC))
 		runtime_markallocated(v, size, (flag&FlagNoPointers) != 0);

 	m->mallocing = 0;

 	if(!(flag & FlagNoProfiling) && (rate = runtime_MemProfileRate) > 0) {
 		if(size >= (uint32) rate)
 			goto profile;
 		if((uint32) m->mcache->next_sample > size)
 			m->mcache->next_sample -= size;
 		else {
 			// pick next profile time
 			// If you change this, also change allocmcache.
 			if(rate > 0x3fffffff)	// make 2*rate not overflow
 				rate = 0x3fffffff;
 			m->mcache->next_sample = runtime_fastrand1() % (2*rate);
 		profile:
 			runtime_setblockspecial(v, true);
 			runtime_MProf_Malloc(v, size);
 		}
 	}

 	if(dogc && mstats.heap_alloc >= mstats.next_gc)
 		runtime_gc(0);
 	return v;
 }

 void*
 __go_alloc(uintptr size)
 {
 	return runtime_mallocgc(size, 0, 0, 1);
 }

 // Free the object whose base pointer is v.
 void
 __go_free(void *v)
 {
 	M *m;
 	int32 sizeclass;
 	MSpan *s;
 	MCache *c;
 	uint32 prof;
 	uintptr size;

 	if(v == nil)
 		return;

 	// If you change this also change mgc0.c:/^sweep,
 	// which has a copy of the guts of free.

 	m = runtime_m();
 	if(m->mallocing)
 		runtime_throw("malloc/free - deadlock");
 	m->mallocing = 1;

 	if(!runtime_mlookup(v, nil, nil, &s)) {
 		runtime_printf("free %p: not an allocated block\n", v);
 		runtime_throw("free runtime_mlookup");
 	}
 	prof = runtime_blockspecial(v);

 	// Find size class for v.
 	sizeclass = s->sizeclass;
 	c = m->mcache;
 	if(sizeclass == 0) {
 		// Large object.
 		size = s->npages<<PageShift;
 		*(uintptr*)(s->start<<PageShift) = 1;	// mark as "needs to be zeroed"
 		// Must mark v freed before calling unmarkspan and MHeap_Free:
 		// they might coalesce v into other spans and change the bitmap further.
 		runtime_markfreed(v, size);
 		runtime_unmarkspan(v, 1<<PageShift);
 		runtime_MHeap_Free(&runtime_mheap, s, 1);
 	} else {
 		// Small object.
 		size = runtime_class_to_size[sizeclass];
 		if(size > sizeof(uintptr))
 			((uintptr*)v)[1] = 1;	// mark as "needs to be zeroed"
 		// Must mark v freed before calling MCache_Free:
 		// it might coalesce v and other blocks into a bigger span
 		// and change the bitmap further.
 		runtime_markfreed(v, size);
 		c->local_by_size[sizeclass].nfree++;
 		runtime_MCache_Free(c, v, sizeclass, size);
 	}
 	c->local_nfree++;
 	c->local_alloc -= size;
 	if(prof)
 		runtime_MProf_Free(v, size);
 	m->mallocing = 0;
 }

 int32
 runtime_mlookup(void *v, byte **base, uintptr *size, MSpan **sp)
 {
 	uintptr n, i;
 	byte *p;
 	MSpan *s;

 	runtime_m()->mcache->local_nlookup++;
 	s = runtime_MHeap_LookupMaybe(&runtime_mheap, v);
 	if(sp)
 		*sp = s;
 	if(s == nil) {
 		runtime_checkfreed(v, 1);
 		if(base)
 			*base = nil;
 		if(size)
 			*size = 0;
 		return 0;
 	}

 	p = (byte*)((uintptr)s->start<<PageShift);
 	if(s->sizeclass == 0) {
 		// Large object.
 		if(base)
 			*base = p;
 		if(size)
 			*size = s->npages<<PageShift;
 		return 1;
 	}

 	if((byte*)v >= (byte*)s->limit) {
 		// pointers past the last block do not count as pointers.
 		return 0;
 	}

 	n = runtime_class_to_size[s->sizeclass];
 	if(base) {
 		i = ((byte*)v - p)/n;
 		*base = p + i*n;
 	}
 	if(size)
 		*size = n;

 	return 1;
 }

 MCache*
 runtime_allocmcache(void)
 {
 	int32 rate;
 	MCache *c;

 	runtime_lock(&runtime_mheap);
 	c = runtime_FixAlloc_Alloc(&runtime_mheap.cachealloc);
 	mstats.mcache_inuse = runtime_mheap.cachealloc.inuse;
 	mstats.mcache_sys = runtime_mheap.cachealloc.sys;
 	runtime_unlock(&runtime_mheap);

 	// Set first allocation sample size.
 	rate = runtime_MemProfileRate;
 	if(rate > 0x3fffffff)	// make 2*rate not overflow
 		rate = 0x3fffffff;
 	if(rate != 0)
 		c->next_sample = runtime_fastrand1() % (2*rate);

 	return c;
 }

 void
 runtime_purgecachedstats(M* m)
 {
 	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;
 	c->local_objects = 0;
 	mstats.nmalloc += c->local_nmalloc;
 	c->local_nmalloc = 0;
 	mstats.nfree += c->local_nfree;
 	c->local_nfree = 0;
 	mstats.nlookup += c->local_nlookup;
 	c->local_nlookup = 0;
 	mstats.alloc += c->local_alloc;
 	c->local_alloc= 0;
 	mstats.total_alloc += c->local_total_alloc;
 	c->local_total_alloc= 0;
 }

 extern uintptr runtime_sizeof_C_MStats
   __asm__ ("runtime.Sizeof_C_MStats");

 #define MaxArena32 (2U<<30)

 void
 runtime_mallocinit(void)
 {
 	byte *p;
 	uintptr arena_size, bitmap_size;
 	extern byte end[];
 	byte *want;
 	uintptr limit;

 	runtime_sizeof_C_MStats = sizeof(MStats);

 	p = nil;
 	arena_size = 0;
 	bitmap_size = 0;

 	// for 64-bit build
 	USED(p);
 	USED(arena_size);
 	USED(bitmap_size);

 	runtime_InitSizes();

 	limit = runtime_memlimit();

 	// Set up the allocation arena, a contiguous area of memory where
 	// allocated data will be found.  The arena begins with a bitmap large
 	// enough to hold 4 bits per allocated word.
 	if(sizeof(void*) == 8 && (limit == 0 || limit > (1<<30))) {
 		// On a 64-bit machine, allocate from a single contiguous reservation.
 		// 16 GB should be big enough for now.
 		//
 		// The code will work with the reservation at any address, but ask
 		// SysReserve to use 0x000000f800000000 if possible.
 		// Allocating a 16 GB region takes away 36 bits, and the amd64
 		// doesn't let us choose the top 17 bits, so that leaves the 11 bits
 		// in the middle of 0x00f8 for us to choose.  Choosing 0x00f8 means
 		// that the valid memory addresses will begin 0x00f8, 0x00f9, 0x00fa, 0x00fb.
 		// None of the bytes f8 f9 fa fb can appear in valid UTF-8, and
 		// they are otherwise as far from ff (likely a common byte) as possible.
 		// Choosing 0x00 for the leading 6 bits was more arbitrary, but it
 		// is not a common ASCII code point either.  Using 0x11f8 instead
 		// caused out of memory errors on OS X during thread allocations.
 		// These choices are both for debuggability and to reduce the
 		// odds of the conservative garbage collector not collecting memory
 		// because some non-pointer block of memory had a bit pattern
 		// that matched a memory address.
 		//
 		// Actually we reserve 17 GB (because the bitmap ends up being 1 GB)
 		// but it hardly matters: fc is not valid UTF-8 either, and we have to
 		// allocate 15 GB before we get that far.
 		//
 		// If this fails we fall back to the 32 bit memory mechanism
 		arena_size = (uintptr)(16LL<<30);
 		bitmap_size = arena_size / (sizeof(void*)*8/4);
 		p = runtime_SysReserve((void*)(0x00f8ULL<<32), bitmap_size + arena_size);
 	}
 	if (p == nil) {
 		// On a 32-bit machine, we can't typically get away
 		// with a giant virtual address space reservation.
 		// Instead we map the memory information bitmap
 		// immediately after the data segment, large enough
 		// to handle another 2GB of mappings (256 MB),
 		// along with a reservation for another 512 MB of memory.
 		// When that gets used up, we'll start asking the kernel
 		// for any memory anywhere and hope it's in the 2GB
 		// following the bitmap (presumably the executable begins
 		// near the bottom of memory, so we'll have to use up
 		// most of memory before the kernel resorts to giving out
 		// memory before the beginning of the text segment).
 		//
 		// Alternatively we could reserve 512 MB bitmap, enough
 		// for 4GB of mappings, and then accept any memory the
 		// kernel threw at us, but normally that's a waste of 512 MB
 		// of address space, which is probably too much in a 32-bit world.
 		bitmap_size = MaxArena32 / (sizeof(void*)*8/4);
 		arena_size = 512<<20;
 		if(limit > 0 && arena_size+bitmap_size > limit) {
 			bitmap_size = (limit / 9) & ~((1<<PageShift) - 1);
 			arena_size = bitmap_size * 8;
 		}

 		// SysReserve treats the address we ask for, end, as a hint,
 		// not as an absolute requirement.  If we ask for the end
 		// of the data segment but the operating system requires
 		// a little more space before we can start allocating, it will
 		// give out a slightly higher pointer.  Except QEMU, which
 		// is buggy, as usual: it won't adjust the pointer upward.
 		// So adjust it upward a little bit ourselves: 1/4 MB to get
 		// away from the running binary image and then round up
 		// to a MB boundary.
 		want = (byte*)(((uintptr)end + (1<<18) + (1<<20) - 1)&~((1<<20)-1));
 		if(0xffffffff - (uintptr)want <= bitmap_size + arena_size)
 		  want = 0;
 		p = runtime_SysReserve(want, bitmap_size + arena_size);
 		if(p == nil)
 			runtime_throw("runtime: cannot reserve arena virtual address space");
 		if((uintptr)p & (((uintptr)1<<PageShift)-1))
 			runtime_printf("runtime: SysReserve returned unaligned address %p; asked for %p", p, bitmap_size+arena_size);
 	}
 	if((uintptr)p & (((uintptr)1<<PageShift)-1))
 		runtime_throw("runtime: SysReserve returned unaligned address");

 	runtime_mheap.bitmap = p;
 	runtime_mheap.arena_start = p + bitmap_size;
 	runtime_mheap.arena_used = runtime_mheap.arena_start;
 	runtime_mheap.arena_end = runtime_mheap.arena_start + arena_size;

 	// Initialize the rest of the allocator.
 	runtime_MHeap_Init(&runtime_mheap, runtime_SysAlloc);
 	runtime_m()->mcache = runtime_allocmcache();

 	// See if it works.
 	runtime_free(runtime_malloc(1));
 }

 void*
 runtime_MHeap_SysAlloc(MHeap *h, uintptr n)
 {
 	byte *p;


 	if(n > (uintptr)(h->arena_end - h->arena_used)) {
 		// We are in 32-bit mode, maybe we didn't use all possible address space yet.
 		// Reserve some more space.
 		byte *new_end;
 		uintptr needed;

 		needed = (uintptr)h->arena_used + n - (uintptr)h->arena_end;
 		// Round wanted arena size to a multiple of 256MB.
 		needed = (needed + (256<<20) - 1) & ~((256<<20)-1);
 		new_end = h->arena_end + needed;
 		if(new_end <= h->arena_start + MaxArena32) {
 			p = runtime_SysReserve(h->arena_end, new_end - h->arena_end);
 			if(p == h->arena_end)
 				h->arena_end = new_end;
 		}
 	}
 	if(n <= (uintptr)(h->arena_end - h->arena_used)) {
 		// Keep taking from our reservation.
 		p = h->arena_used;
 		runtime_SysMap(p, n);
 		h->arena_used += n;
 		runtime_MHeap_MapBits(h);
 		return p;
 	}

 	// If using 64-bit, our reservation is all we have.
 	if(sizeof(void*) == 8 && (uintptr)h->bitmap >= 0xffffffffU)
 		return nil;

 	// On 32-bit, once the reservation is gone we can
 	// try to get memory at a location chosen by the OS
 	// and hope that it is in the range we allocated bitmap for.
 	p = runtime_SysAlloc(n);
 	if(p == nil)
 		return nil;

 	if(p < h->arena_start || (uintptr)(p+n - h->arena_start) >= MaxArena32) {
 		runtime_printf("runtime: memory allocated by OS (%p) not in usable range [%p,%p)\n",
 			p, h->arena_start, h->arena_start+MaxArena32);
 		runtime_SysFree(p, n);
 		return nil;
 	}

 	if(p+n > h->arena_used) {
 		h->arena_used = p+n;
 		if(h->arena_used > h->arena_end)
 			h->arena_end = h->arena_used;
 		runtime_MHeap_MapBits(h);
 	}

 	return p;
 }

 // Runtime stubs.

 void*
 runtime_mal(uintptr n)
 {
 	return runtime_mallocgc(n, 0, 1, 1);
 }

 func new(typ *Type) (ret *uint8) {
 	uint32 flag = typ->__code&GO_NO_POINTERS ? FlagNoPointers : 0;
 	ret = runtime_mallocgc(typ->__size, flag, 1, 1);
 }

 func GC() {
 	runtime_gc(1);
 }

 func SetFinalizer(obj Eface, finalizer Eface) {
 	byte *base;
 	uintptr size;
 	const FuncType *ft;

 	if(obj.__type_descriptor == nil) {
 		runtime_printf("runtime.SetFinalizer: first argument is nil interface\n");
 		goto throw;
 	}
 	if(obj.__type_descriptor->__code != GO_PTR) {
 		runtime_printf("runtime.SetFinalizer: first argument is %S, not pointer\n", *obj.__type_descriptor->__reflection);
 		goto throw;
 	}
 	if(!runtime_mlookup(obj.__object, &base, &size, nil) || obj.__object != base) {
 		runtime_printf("runtime.SetFinalizer: pointer not at beginning of allocated block\n");
 		goto throw;
 	}
 	ft = nil;
 	if(finalizer.__type_descriptor != nil) {
 		if(finalizer.__type_descriptor->__code != GO_FUNC)
 			goto badfunc;
 		ft = (const FuncType*)finalizer.__type_descriptor;
 		if(ft->__dotdotdot || ft->__in.__count != 1 || !__go_type_descriptors_equal(*(Type**)ft->__in.__values, obj.__type_descriptor))
 			goto badfunc;
 	}

 	if(!runtime_addfinalizer(obj.__object, finalizer.__type_descriptor != nil ? *(void**)finalizer.__object : nil, ft)) {
 		runtime_printf("runtime.SetFinalizer: finalizer already set\n");
 		goto throw;
 	}
 	return;

 badfunc:
 	runtime_printf("runtime.SetFinalizer: second argument is %S, not func(%S)\n", *finalizer.__type_descriptor->__reflection, *obj.__type_descriptor->__reflection);
 throw:
 	runtime_throw("runtime.SetFinalizer");
 }
	// Copyright 2009 The Go Authors. All rights reserved.
	// Use of this source code is governed by a BSD-style
	// license that can be found in the LICENSE file.

	// See malloc.h for overview.
	//
	// TODO(rsc): double-check stats.

	package runtime
	#include <stddef.h>
	#include <errno.h>
	#include <stdlib.h>
	#include "go-alloc.h"
	#include "runtime.h"
	#include "arch.h"
	#include "malloc.h"
	#include "go-string.h"
	#include "interface.h"
	#include "go-type.h"

	MHeap runtime_mheap;

	extern MStats mstats; // defined in extern.go

	extern volatile int32 runtime_MemProfileRate
	__asm__ ("runtime.MemProfileRate");

	// 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.
	void*
	runtime_mallocgc(uintptr size, uint32 flag, int32 dogc, int32 zeroed)
	{
	M *m;
	G *g;
	int32 sizeclass, rate;
	MCache *c;
	uintptr npages;
	MSpan *s;
	void *v;

	m = runtime_m();
	g = runtime_g();
	if(g->status == Gsyscall)
	dogc = 0;
	if(runtime_gcwaiting && g != m->g0 && m->locks == 0 && g->status != Gsyscall) {
	runtime_gosched();
	m = runtime_m();
	}
	if(m->mallocing)
	runtime_throw("malloc/free - deadlock");
	m->mallocing = 1;
	if(size == 0)
	size = 1;

	c = m->mcache;
	c->local_nmalloc++;
	if(size <= MaxSmallSize) {
	// Allocate from mcache free lists.
	sizeclass = runtime_SizeToClass(size);
	size = runtime_class_to_size[sizeclass];
	v = runtime_MCache_Alloc(c, sizeclass, size, zeroed);
	if(v == nil)
	runtime_throw("out of memory");
	c->local_alloc += size;
	c->local_total_alloc += size;
	c->local_by_size[sizeclass].nmalloc++;
	} else {
	// TODO(rsc): Report tracebacks for very large allocations.

	// Allocate directly from heap.
	npages = size >> PageShift;
	if((size & PageMask) != 0)
	npages++;
	s = runtime_MHeap_Alloc(&runtime_mheap, npages, 0, 1);
	if(s == nil)
	runtime_throw("out of memory");
	size = npages<<PageShift;
	c->local_alloc += size;
	c->local_total_alloc += size;
	v = (void*)(s->start << PageShift);

	// setup for mark sweep
	runtime_markspan(v, 0, 0, true);
	}
	if(!(flag & FlagNoGC))
	runtime_markallocated(v, size, (flag&FlagNoPointers) != 0);

	m->mallocing = 0;

	if(!(flag & FlagNoProfiling) && (rate = runtime_MemProfileRate) > 0) {
	if(size >= (uint32) rate)
	goto profile;
	if((uint32) m->mcache->next_sample > size)
	m->mcache->next_sample -= size;
	else {
	// pick next profile time
	// If you change this, also change allocmcache.
	if(rate > 0x3fffffff) // make 2*rate not overflow
	rate = 0x3fffffff;
	m->mcache->next_sample = runtime_fastrand1() % (2*rate);
	profile:
	runtime_setblockspecial(v, true);
	runtime_MProf_Malloc(v, size);
	}
	}

	if(dogc && mstats.heap_alloc >= mstats.next_gc)
	runtime_gc(0);
	return v;
	}

	void*
	__go_alloc(uintptr size)
	{
	return runtime_mallocgc(size, 0, 0, 1);
	}

	// Free the object whose base pointer is v.
	void
	__go_free(void *v)
	{
	M *m;
	int32 sizeclass;
	MSpan *s;
	MCache *c;
	uint32 prof;
	uintptr size;

	if(v == nil)
	return;

	// If you change this also change mgc0.c:/^sweep,
	// which has a copy of the guts of free.

	m = runtime_m();
	if(m->mallocing)
	runtime_throw("malloc/free - deadlock");
	m->mallocing = 1;

	if(!runtime_mlookup(v, nil, nil, &s)) {
	runtime_printf("free %p: not an allocated block\n", v);
	runtime_throw("free runtime_mlookup");
	}
	prof = runtime_blockspecial(v);

	// Find size class for v.
	sizeclass = s->sizeclass;
	c = m->mcache;
	if(sizeclass == 0) {
	// Large object.
	size = s->npages<<PageShift;
	(uintptr)(s->start<<PageShift) = 1; // mark as "needs to be zeroed"
	// Must mark v freed before calling unmarkspan and MHeap_Free:
	// they might coalesce v into other spans and change the bitmap further.
	runtime_markfreed(v, size);
	runtime_unmarkspan(v, 1<<PageShift);
	runtime_MHeap_Free(&runtime_mheap, s, 1);
	} else {
	// Small object.
	size = runtime_class_to_size[sizeclass];
	if(size > sizeof(uintptr))
	((uintptr*)v)[1] = 1; // mark as "needs to be zeroed"
	// Must mark v freed before calling MCache_Free:
	// it might coalesce v and other blocks into a bigger span
	// and change the bitmap further.
	runtime_markfreed(v, size);
	c->local_by_size[sizeclass].nfree++;
	runtime_MCache_Free(c, v, sizeclass, size);
	}
	c->local_nfree++;
	c->local_alloc -= size;
	if(prof)
	runtime_MProf_Free(v, size);
	m->mallocing = 0;
	}

	int32
	runtime_mlookup(void v, byte base, uintptr size, MSpan **sp)
	{
	uintptr n, i;
	byte *p;
	MSpan *s;

	runtime_m()->mcache->local_nlookup++;
	s = runtime_MHeap_LookupMaybe(&runtime_mheap, v);
	if(sp)
	*sp = s;
	if(s == nil) {
	runtime_checkfreed(v, 1);
	if(base)
	*base = nil;
	if(size)
	*size = 0;
	return 0;
	}

	p = (byte*)((uintptr)s->start<<PageShift);
	if(s->sizeclass == 0) {
	// Large object.
	if(base)
	*base = p;
	if(size)
	*size = s->npages<<PageShift;
	return 1;
	}

	if((byte)v >= (byte)s->limit) {
	// pointers past the last block do not count as pointers.
	return 0;
	}

	n = runtime_class_to_size[s->sizeclass];
	if(base) {
	i = ((byte*)v - p)/n;
	base = p + in;
	}
	if(size)
	*size = n;

	return 1;
	}

	MCache*
	runtime_allocmcache(void)
	{
	int32 rate;
	MCache *c;

	runtime_lock(&runtime_mheap);
	c = runtime_FixAlloc_Alloc(&runtime_mheap.cachealloc);
	mstats.mcache_inuse = runtime_mheap.cachealloc.inuse;
	mstats.mcache_sys = runtime_mheap.cachealloc.sys;
	runtime_unlock(&runtime_mheap);

	// Set first allocation sample size.
	rate = runtime_MemProfileRate;
	if(rate > 0x3fffffff) // make 2*rate not overflow
	rate = 0x3fffffff;
	if(rate != 0)
	c->next_sample = runtime_fastrand1() % (2*rate);

	return c;
	}

	void
	runtime_purgecachedstats(M* m)
	{
	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;
	c->local_objects = 0;
	mstats.nmalloc += c->local_nmalloc;
	c->local_nmalloc = 0;
	mstats.nfree += c->local_nfree;
	c->local_nfree = 0;
	mstats.nlookup += c->local_nlookup;
	c->local_nlookup = 0;
	mstats.alloc += c->local_alloc;
	c->local_alloc= 0;
	mstats.total_alloc += c->local_total_alloc;
	c->local_total_alloc= 0;
	}

	extern uintptr runtime_sizeof_C_MStats
	__asm__ ("runtime.Sizeof_C_MStats");

	#define MaxArena32 (2U<<30)

	void
	runtime_mallocinit(void)
	{
	byte *p;
	uintptr arena_size, bitmap_size;
	extern byte end[];
	byte *want;
	uintptr limit;

	runtime_sizeof_C_MStats = sizeof(MStats);

	p = nil;
	arena_size = 0;
	bitmap_size = 0;

	// for 64-bit build
	USED(p);
	USED(arena_size);
	USED(bitmap_size);

	runtime_InitSizes();

	limit = runtime_memlimit();

	// Set up the allocation arena, a contiguous area of memory where
	// allocated data will be found. The arena begins with a bitmap large
	// enough to hold 4 bits per allocated word.
	if(sizeof(void*) == 8 && (limit == 0 \|\| limit > (1<<30))) {
	// On a 64-bit machine, allocate from a single contiguous reservation.
	// 16 GB should be big enough for now.
	//
	// The code will work with the reservation at any address, but ask
	// SysReserve to use 0x000000f800000000 if possible.
	// Allocating a 16 GB region takes away 36 bits, and the amd64
	// doesn't let us choose the top 17 bits, so that leaves the 11 bits
	// in the middle of 0x00f8 for us to choose. Choosing 0x00f8 means
	// that the valid memory addresses will begin 0x00f8, 0x00f9, 0x00fa, 0x00fb.
	// None of the bytes f8 f9 fa fb can appear in valid UTF-8, and
	// they are otherwise as far from ff (likely a common byte) as possible.
	// Choosing 0x00 for the leading 6 bits was more arbitrary, but it
	// is not a common ASCII code point either. Using 0x11f8 instead
	// caused out of memory errors on OS X during thread allocations.
	// These choices are both for debuggability and to reduce the
	// odds of the conservative garbage collector not collecting memory
	// because some non-pointer block of memory had a bit pattern
	// that matched a memory address.
	//
	// Actually we reserve 17 GB (because the bitmap ends up being 1 GB)
	// but it hardly matters: fc is not valid UTF-8 either, and we have to
	// allocate 15 GB before we get that far.
	//
	// If this fails we fall back to the 32 bit memory mechanism
	arena_size = (uintptr)(16LL<<30);
	bitmap_size = arena_size / (sizeof(void)8/4);
	p = runtime_SysReserve((void*)(0x00f8ULL<<32), bitmap_size + arena_size);
	}
	if (p == nil) {
	// On a 32-bit machine, we can't typically get away
	// with a giant virtual address space reservation.
	// Instead we map the memory information bitmap
	// immediately after the data segment, large enough
	// to handle another 2GB of mappings (256 MB),
	// along with a reservation for another 512 MB of memory.
	// When that gets used up, we'll start asking the kernel
	// for any memory anywhere and hope it's in the 2GB
	// following the bitmap (presumably the executable begins
	// near the bottom of memory, so we'll have to use up
	// most of memory before the kernel resorts to giving out
	// memory before the beginning of the text segment).
	//
	// Alternatively we could reserve 512 MB bitmap, enough
	// for 4GB of mappings, and then accept any memory the
	// kernel threw at us, but normally that's a waste of 512 MB
	// of address space, which is probably too much in a 32-bit world.
	bitmap_size = MaxArena32 / (sizeof(void)8/4);
	arena_size = 512<<20;
	if(limit > 0 && arena_size+bitmap_size > limit) {
	bitmap_size = (limit / 9) & ~((1<<PageShift) - 1);
	arena_size = bitmap_size * 8;
	}

	// SysReserve treats the address we ask for, end, as a hint,
	// not as an absolute requirement. If we ask for the end
	// of the data segment but the operating system requires
	// a little more space before we can start allocating, it will
	// give out a slightly higher pointer. Except QEMU, which
	// is buggy, as usual: it won't adjust the pointer upward.
	// So adjust it upward a little bit ourselves: 1/4 MB to get
	// away from the running binary image and then round up
	// to a MB boundary.
	want = (byte*)(((uintptr)end + (1<<18) + (1<<20) - 1)&~((1<<20)-1));
	if(0xffffffff - (uintptr)want <= bitmap_size + arena_size)
	want = 0;
	p = runtime_SysReserve(want, bitmap_size + arena_size);
	if(p == nil)
	runtime_throw("runtime: cannot reserve arena virtual address space");
	if((uintptr)p & (((uintptr)1<<PageShift)-1))
	runtime_printf("runtime: SysReserve returned unaligned address %p; asked for %p", p, bitmap_size+arena_size);
	}
	if((uintptr)p & (((uintptr)1<<PageShift)-1))
	runtime_throw("runtime: SysReserve returned unaligned address");

	runtime_mheap.bitmap = p;
	runtime_mheap.arena_start = p + bitmap_size;
	runtime_mheap.arena_used = runtime_mheap.arena_start;
	runtime_mheap.arena_end = runtime_mheap.arena_start + arena_size;

	// Initialize the rest of the allocator.
	runtime_MHeap_Init(&runtime_mheap, runtime_SysAlloc);
	runtime_m()->mcache = runtime_allocmcache();

	// See if it works.
	runtime_free(runtime_malloc(1));
	}

	void*
	runtime_MHeap_SysAlloc(MHeap *h, uintptr n)
	{
	byte *p;


	if(n > (uintptr)(h->arena_end - h->arena_used)) {
	// We are in 32-bit mode, maybe we didn't use all possible address space yet.
	// Reserve some more space.
	byte *new_end;
	uintptr needed;

	needed = (uintptr)h->arena_used + n - (uintptr)h->arena_end;
	// Round wanted arena size to a multiple of 256MB.
	needed = (needed + (256<<20) - 1) & ~((256<<20)-1);
	new_end = h->arena_end + needed;
	if(new_end <= h->arena_start + MaxArena32) {
	p = runtime_SysReserve(h->arena_end, new_end - h->arena_end);
	if(p == h->arena_end)
	h->arena_end = new_end;
	}
	}
	if(n <= (uintptr)(h->arena_end - h->arena_used)) {
	// Keep taking from our reservation.
	p = h->arena_used;
	runtime_SysMap(p, n);
	h->arena_used += n;
	runtime_MHeap_MapBits(h);
	return p;
	}

	// If using 64-bit, our reservation is all we have.
	if(sizeof(void*) == 8 && (uintptr)h->bitmap >= 0xffffffffU)
	return nil;

	// On 32-bit, once the reservation is gone we can
	// try to get memory at a location chosen by the OS
	// and hope that it is in the range we allocated bitmap for.
	p = runtime_SysAlloc(n);
	if(p == nil)
	return nil;

	if(p < h->arena_start \|\| (uintptr)(p+n - h->arena_start) >= MaxArena32) {
	runtime_printf("runtime: memory allocated by OS (%p) not in usable range [%p,%p)\n",
	p, h->arena_start, h->arena_start+MaxArena32);
	runtime_SysFree(p, n);
	return nil;
	}

	if(p+n > h->arena_used) {
	h->arena_used = p+n;
	if(h->arena_used > h->arena_end)
	h->arena_end = h->arena_used;
	runtime_MHeap_MapBits(h);
	}

	return p;
	}

	// Runtime stubs.

	void*
	runtime_mal(uintptr n)
	{
	return runtime_mallocgc(n, 0, 1, 1);
	}

	func new(typ Type) (ret uint8) {
	uint32 flag = typ->__code&GO_NO_POINTERS ? FlagNoPointers : 0;
	ret = runtime_mallocgc(typ->__size, flag, 1, 1);
	}

	func GC() {
	runtime_gc(1);
	}

	func SetFinalizer(obj Eface, finalizer Eface) {
	byte *base;
	uintptr size;
	const FuncType *ft;

	if(obj.__type_descriptor == nil) {
	runtime_printf("runtime.SetFinalizer: first argument is nil interface\n");
	goto throw;
	}
	if(obj.__type_descriptor->__code != GO_PTR) {
	runtime_printf("runtime.SetFinalizer: first argument is %S, not pointer\n", *obj.__type_descriptor->__reflection);
	goto throw;
	}
	if(!runtime_mlookup(obj.__object, &base, &size, nil) \|\| obj.__object != base) {
	runtime_printf("runtime.SetFinalizer: pointer not at beginning of allocated block\n");
	goto throw;
	}
	ft = nil;
	if(finalizer.__type_descriptor != nil) {
	if(finalizer.__type_descriptor->__code != GO_FUNC)
	goto badfunc;
	ft = (const FuncType*)finalizer.__type_descriptor;
	if(ft->__dotdotdot \|\| ft->__in.__count != 1 \|\| !__go_type_descriptors_equal((Type*)ft->__in.__values, obj.__type_descriptor))
	goto badfunc;
	}

	if(!runtime_addfinalizer(obj.__object, finalizer.__type_descriptor != nil ? (void*)finalizer.__object : nil, ft)) {
	runtime_printf("runtime.SetFinalizer: finalizer already set\n");
	goto throw;
	}
	return;

	badfunc:
	runtime_printf("runtime.SetFinalizer: second argument is %S, not func(%S)\n", finalizer.__type_descriptor->__reflection, obj.__type_descriptor->__reflection);
	throw:
	runtime_throw("runtime.SetFinalizer");
	}