malloc.goc

// 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 "runtime.h"
#include "arch_GOARCH.h"
#include "malloc.h"
#include "type.h"
#include "typekind.h"
#include "race.h"
#include "stack.h"
#include "../../cmd/ld/textflag.h"

// Mark mheap as 'no pointers', it does not contain interesting pointers but occupies ~45K.
#pragma dataflag NOPTR
MHeap runtime·mheap;
MStats mstats;

int32	runtime·checking;

extern MStats mstats;	// defined in zruntime_def_$GOOS_$GOARCH.go

extern volatile intgo runtime·MemProfileRate;

static void* largealloc(uint32, uintptr*);

// 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.
// If the block will be freed with runtime·free(), typ must be 0.
void*
runtime·mallocgc(uintptr size, uintptr typ, uint32 flag)
{
	int32 sizeclass;
	uintptr tinysize, size1;
	intgo rate;
	MCache *c;
	MCacheList *l;
	MLink *v, *next;
	byte *tiny;

	if(size == 0) {
		// All 0-length allocations use this pointer.
		// The language does not require the allocations to
		// have distinct values.
		return &runtime·zerobase;
	}
	if(m->mallocing)
		runtime·throw("malloc/free - deadlock");
	// Disable preemption during settype_flush.
	// We can not use m->mallocing for this, because settype_flush calls mallocgc.
	m->locks++;
	m->mallocing = 1;

	if(DebugTypeAtBlockEnd)
		size += sizeof(uintptr);

	c = m->mcache;
	if(!runtime·debug.efence && size <= MaxSmallSize) {
		if((flag&(FlagNoScan|FlagNoGC)) == FlagNoScan && size < TinySize) {
			// Tiny allocator.
			//
			// Tiny allocator combines several tiny allocation requests
			// into a single memory block. The resulting memory block
			// is freed when all subobjects are unreachable. The subobjects
			// must be FlagNoScan (don't have pointers), this ensures that
			// the amount of potentially wasted memory is bounded.
			//
			// Size of the memory block used for combining (TinySize) is tunable.
			// Current setting is 16 bytes, which relates to 2x worst case memory
			// wastage (when all but one subobjects are unreachable).
			// 8 bytes would result in no wastage at all, but provides less
			// opportunities for combining.
			// 32 bytes provides more opportunities for combining,
			// but can lead to 4x worst case wastage.
			// The best case winning is 8x regardless of block size.
			//
			// Objects obtained from tiny allocator must not be freed explicitly.
			// So when an object will be freed explicitly, we ensure that
			// its size >= TinySize.
			//
			// SetFinalizer has a special case for objects potentially coming
			// from tiny allocator, it such case it allows to set finalizers
			// for an inner byte of a memory block.
			//
			// The main targets of tiny allocator are small strings and
			// standalone escaping variables. On a json benchmark
			// the allocator reduces number of allocations by ~12% and
			// reduces heap size by ~20%.

			tinysize = c->tinysize;
			if(size <= tinysize) {
				tiny = c->tiny;
				// Align tiny pointer for required (conservative) alignment.
				if((size&7) == 0)
					tiny = (byte*)ROUND((uintptr)tiny, 8);
				else if((size&3) == 0)
					tiny = (byte*)ROUND((uintptr)tiny, 4);
				else if((size&1) == 0)
					tiny = (byte*)ROUND((uintptr)tiny, 2);
				size1 = size + (tiny - c->tiny);
				if(size1 <= tinysize) {
					// The object fits into existing tiny block.
					v = (MLink*)tiny;
					c->tiny += size1;
					c->tinysize -= size1;
					m->mallocing = 0;
					m->locks--;
					if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
						g->stackguard0 = StackPreempt;
					return v;
				}
			}
			// Allocate a new TinySize block.
			l = &c->list[TinySizeClass];
			if(l->list == nil)
				runtime·MCache_Refill(c, TinySizeClass);
			v = l->list;
			next = v->next;
			if(next != nil)  // prefetching nil leads to a DTLB miss
				PREFETCH(next);
			l->list = next;
			l->nlist--;
			((uint64*)v)[0] = 0;
			((uint64*)v)[1] = 0;
			// See if we need to replace the existing tiny block with the new one
			// based on amount of remaining free space.
			if(TinySize-size > tinysize) {
				c->tiny = (byte*)v + size;
				c->tinysize = TinySize - size;
			}
			size = TinySize;
			goto done;
		}
		// Allocate from mcache free lists.
		// Inlined version of SizeToClass().
		if(size <= 1024-8)
			sizeclass = runtime·size_to_class8[(size+7)>>3];
		else
			sizeclass = runtime·size_to_class128[(size-1024+127) >> 7];
		size = runtime·class_to_size[sizeclass];
		l = &c->list[sizeclass];
		if(l->list == nil)
			runtime·MCache_Refill(c, sizeclass);
		v = l->list;
		next = v->next;
		if(next != nil)  // prefetching nil leads to a DTLB miss
			PREFETCH(next);
		l->list = next;
		l->nlist--;
		if(!(flag & FlagNoZero)) {
			v->next = nil;
			// block is zeroed iff second word is zero ...
			if(size > 2*sizeof(uintptr) && ((uintptr*)v)[1] != 0)
				runtime·memclr((byte*)v, size);
		}
	done:
		c->local_cachealloc += size;
	} else {
		// Allocate directly from heap.
		v = largealloc(flag, &size);
	}

	if(flag & FlagNoGC)
		runtime·marknogc(v);
	else if(!(flag & FlagNoScan))
		runtime·markscan(v);

	if(DebugTypeAtBlockEnd)
		*(uintptr*)((uintptr)v+size-sizeof(uintptr)) = typ;

	// TODO: save type even if FlagNoScan?  Potentially expensive but might help
	// heap profiling/tracing.
	if(UseSpanType && !(flag & FlagNoScan) && typ != 0) {
		uintptr *buf, i;

		buf = m->settype_buf;
		i = m->settype_bufsize;
		buf[i++] = (uintptr)v;
		buf[i++] = typ;
		m->settype_bufsize = i;
	}

	m->mallocing = 0;
	if(UseSpanType && !(flag & FlagNoScan) && typ != 0 && m->settype_bufsize == nelem(m->settype_buf))
		runtime·settype_flush(m);
	if(raceenabled)
		runtime·racemalloc(v, size);
	m->locks--;
	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
		g->stackguard0 = StackPreempt;

	if(runtime·debug.allocfreetrace)
		goto profile;

	if(!(flag & FlagNoProfiling) && (rate = runtime·MemProfileRate) > 0) {
		if(size >= rate)
			goto profile;
		if(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·MProf_Malloc(v, size, typ);
		}
	}

	if(!(flag & FlagNoInvokeGC) && mstats.heap_alloc >= mstats.next_gc)
		runtime·gc(0);

	return v;
}

static void*
largealloc(uint32 flag, uintptr *sizep)
{
	uintptr npages, size;
	MSpan *s;
	void *v;

	// Allocate directly from heap.
	size = *sizep;
	if(size + PageSize < size)
		runtime·throw("out of memory");
	npages = size >> PageShift;
	if((size & PageMask) != 0)
		npages++;
	s = runtime·MHeap_Alloc(&runtime·mheap, npages, 0, 1, !(flag & FlagNoZero));
	if(s == nil)
		runtime·throw("out of memory");
	s->limit = (byte*)(s->start<<PageShift) + size;
	*sizep = npages<<PageShift;
	v = (void*)(s->start << PageShift);
	// setup for mark sweep
	runtime·markspan(v, 0, 0, true);
	return v;
}

void*
runtime·malloc(uintptr size)
{
	return runtime·mallocgc(size, 0, FlagNoInvokeGC);
}

// Free the object whose base pointer is v.
void
runtime·free(void *v)
{
	int32 sizeclass;
	MSpan *s;
	MCache *c;
	uintptr size;

	if(v == nil)
		return;
	
	// If you change this also change mgc0.c:/^sweep,
	// which has a copy of the guts of free.

	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");
	}
	size = s->elemsize;
	sizeclass = s->sizeclass;
	// Objects that are smaller than TinySize can be allocated using tiny alloc,
	// if then such object is combined with an object with finalizer, we will crash.
	if(size < TinySize)
		runtime·throw("freeing too small block");

	if(raceenabled)
		runtime·racefree(v);

	if(s->specials != nil)
		runtime·freeallspecials(s, v, size);

	c = m->mcache;
	if(sizeclass == 0) {
		// Large object.
		*(uintptr*)(s->start<<PageShift) = (uintptr)0xfeedfeedfeedfeedll;	// 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);
		if(runtime·debug.efence)
			runtime·SysFree((void*)(s->start<<PageShift), size, &mstats.heap_sys);
		else
			runtime·MHeap_Free(&runtime·mheap, s, 1);
		c->local_nlargefree++;
		c->local_largefree += size;
	} else {
		// Small object.
		if(size > 2*sizeof(uintptr))
			((uintptr*)v)[1] = (uintptr)0xfeedfeedfeedfeedll;	// mark as "needs to be zeroed"
		else if(size > sizeof(uintptr))
			((uintptr*)v)[1] = 0;
		// 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_nsmallfree[sizeclass]++;
		runtime·MCache_Free(c, v, sizeclass, size);
	}
	m->mallocing = 0;
}

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

	m->mcache->local_nlookup++;
	if (sizeof(void*) == 4 && m->mcache->local_nlookup >= (1<<30)) {
		// purge cache stats to prevent overflow
		runtime·lock(&runtime·mheap);
		runtime·purgecachedstats(m->mcache);
		runtime·unlock(&runtime·mheap);
	}

	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;
	}

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

	return 1;
}

MCache*
runtime·allocmcache(void)
{
	intgo rate;
	MCache *c;

	runtime·lock(&runtime·mheap);
	c = runtime·FixAlloc_Alloc(&runtime·mheap.cachealloc);
	runtime·unlock(&runtime·mheap);
	runtime·memclr((byte*)c, sizeof(*c));

	// 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·freemcache(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)
{
	MHeap *h;
	int32 i;

	// Protected by either heap or GC lock.
	h = &runtime·mheap;
	mstats.heap_alloc += c->local_cachealloc;
	c->local_cachealloc = 0;
	mstats.nlookup += c->local_nlookup;
	c->local_nlookup = 0;
	h->largefree += c->local_largefree;
	c->local_largefree = 0;
	h->nlargefree += c->local_nlargefree;
	c->local_nlargefree = 0;
	for(i=0; i<nelem(c->local_nsmallfree); i++) {
		h->nsmallfree[i] += c->local_nsmallfree[i];
		c->local_nsmallfree[i] = 0;
	}
}

// Size of the trailing by_size array differs between Go and C,
// NumSizeClasses was changed, but we can not change Go struct because of backward compatibility.
// sizeof_C_MStats is what C thinks about size of Go struct.
uintptr runtime·sizeof_C_MStats = sizeof(MStats) - (NumSizeClasses - 61) * sizeof(mstats.by_size[0]);

#define MaxArena32 (2U<<30)

void
runtime·mallocinit(void)
{
	byte *p;
	uintptr arena_size, bitmap_size, spans_size;
	extern byte end[];
	uintptr limit;
	uint64 i;

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

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

	runtime·InitSizes();

	if(runtime·class_to_size[TinySizeClass] != TinySize)
		runtime·throw("bad TinySizeClass");

	// limit = runtime·memlimit();
	// See https://code.google.com/p/go/issues/detail?id=5049
	// TODO(rsc): Fix after 1.1.
	limit = 0;

	// 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.
		// 128 GB (MaxMem) should be big enough for now.
		//
		// The code will work with the reservation at any address, but ask
		// SysReserve to use 0x0000XXc000000000 if possible (XX=00...7f).
		// Allocating a 128 GB region takes away 37 bits, and the amd64
		// doesn't let us choose the top 17 bits, so that leaves the 11 bits
		// in the middle of 0x00c0 for us to choose.  Choosing 0x00c0 means
		// that the valid memory addresses will begin 0x00c0, 0x00c1, ..., 0x00df.
		// In little-endian, that's c0 00, c1 00, ..., df 00. None of those are valid
		// UTF-8 sequences, and they are otherwise as far away from 
		// ff (likely a common byte) as possible.  If that fails, we try other 0xXXc0
		// addresses.  An earlier attempt to use 0x11f8 caused out of memory errors
		// on OS X during thread allocations.  0x00c0 causes conflicts with
		// AddressSanitizer which reserves all memory up to 0x0100.
		// 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 136 GB (because the bitmap ends up being 8 GB)
		// but it hardly matters: e0 00 is not valid UTF-8 either.
		//
		// If this fails we fall back to the 32 bit memory mechanism
		arena_size = MaxMem;
		bitmap_size = arena_size / (sizeof(void*)*8/4);
		spans_size = arena_size / PageSize * sizeof(runtime·mheap.spans[0]);
		spans_size = ROUND(spans_size, PageSize);
		for(i = 0; i <= 0x7f; i++) {
			p = (void*)(i<<40 | 0x00c0ULL<<32);
			p = runtime·SysReserve(p, bitmap_size + spans_size + arena_size + PageSize);
			if(p != nil)
				break;
		}
	}
	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;
		spans_size = MaxArena32 / PageSize * sizeof(runtime·mheap.spans[0]);
		if(limit > 0 && arena_size+bitmap_size+spans_size > limit) {
			bitmap_size = (limit / 9) & ~((1<<PageShift) - 1);
			arena_size = bitmap_size * 8;
			spans_size = arena_size / PageSize * sizeof(runtime·mheap.spans[0]);
		}
		spans_size = ROUND(spans_size, PageSize);

		// 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.
		p = (byte*)ROUND((uintptr)end + (1<<18), 1<<20);
		p = runtime·SysReserve(p, bitmap_size + spans_size + arena_size + PageSize);
		if(p == nil)
			runtime·throw("runtime: cannot reserve arena virtual address space");
	}

	// PageSize can be larger than OS definition of page size,
	// so SysReserve can give us a PageSize-unaligned pointer.
	// To overcome this we ask for PageSize more and round up the pointer.
	p = (byte*)ROUND((uintptr)p, PageSize);

	runtime·mheap.spans = (MSpan**)p;
	runtime·mheap.bitmap = p + spans_size;
	runtime·mheap.arena_start = p + spans_size + 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);
	m->mcache = runtime·allocmcache();

	// See if it works.
	runtime·free(runtime·malloc(TinySize));
}

void*
runtime·MHeap_SysAlloc(MHeap *h, uintptr n)
{
	byte *p;

	if(n > 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;
		needed = ROUND(needed, 256<<20);
		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 <= h->arena_end - h->arena_used) {
		// Keep taking from our reservation.
		p = h->arena_used;
		runtime·SysMap(p, n, &mstats.heap_sys);
		h->arena_used += n;
		runtime·MHeap_MapBits(h);
		runtime·MHeap_MapSpans(h);
		if(raceenabled)
			runtime·racemapshadow(p, n);
		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, &mstats.heap_sys);
	if(p == nil)
		return nil;

	if(p < h->arena_start || 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, &mstats.heap_sys);
		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);
		runtime·MHeap_MapSpans(h);
		if(raceenabled)
			runtime·racemapshadow(p, n);
	}
	
	return p;
}

static struct
{
	Lock;
	byte*	pos;
	byte*	end;
} persistent;

enum
{
	PersistentAllocChunk	= 256<<10,
	PersistentAllocMaxBlock	= 64<<10,  // VM reservation granularity is 64K on windows
};

// Wrapper around SysAlloc that can allocate small chunks.
// There is no associated free operation.
// Intended for things like function/type/debug-related persistent data.
// If align is 0, uses default align (currently 8).
void*
runtime·persistentalloc(uintptr size, uintptr align, uint64 *stat)
{
	byte *p;

	if(align != 0) {
		if(align&(align-1))
			runtime·throw("persistentalloc: align is now a power of 2");
		if(align > PageSize)
			runtime·throw("persistentalloc: align is too large");
	} else
		align = 8;
	if(size >= PersistentAllocMaxBlock)
		return runtime·SysAlloc(size, stat);
	runtime·lock(&persistent);
	persistent.pos = (byte*)ROUND((uintptr)persistent.pos, align);
	if(persistent.pos + size > persistent.end) {
		persistent.pos = runtime·SysAlloc(PersistentAllocChunk, &mstats.other_sys);
		if(persistent.pos == nil) {
			runtime·unlock(&persistent);
			runtime·throw("runtime: cannot allocate memory");
		}
		persistent.end = persistent.pos + PersistentAllocChunk;
	}
	p = persistent.pos;
	persistent.pos += size;
	runtime·unlock(&persistent);
	if(stat != &mstats.other_sys) {
		// reaccount the allocation against provided stat
		runtime·xadd64(stat, size);
		runtime·xadd64(&mstats.other_sys, -(uint64)size);
	}
	return p;
}

static Lock settype_lock;

void
runtime·settype_flush(M *mp)
{
	uintptr *buf, *endbuf;
	uintptr size, ofs, j, t;
	uintptr ntypes, nbytes2, nbytes3;
	uintptr *data2;
	byte *data3;
	void *v;
	uintptr typ, p;
	MSpan *s;

	buf = mp->settype_buf;
	endbuf = buf + mp->settype_bufsize;

	runtime·lock(&settype_lock);
	while(buf < endbuf) {
		v = (void*)*buf;
		*buf = 0;
		buf++;
		typ = *buf;
		buf++;

		// (Manually inlined copy of runtime·MHeap_Lookup)
		p = (uintptr)v>>PageShift;
		p -= (uintptr)runtime·mheap.arena_start >> PageShift;
		s = runtime·mheap.spans[p];

		if(s->sizeclass == 0) {
			s->types.compression = MTypes_Single;
			s->types.data = typ;
			continue;
		}

		size = s->elemsize;
		ofs = ((uintptr)v - (s->start<<PageShift)) / size;

		switch(s->types.compression) {
		case MTypes_Empty:
			ntypes = (s->npages << PageShift) / size;
			nbytes3 = 8*sizeof(uintptr) + 1*ntypes;
			data3 = runtime·mallocgc(nbytes3, 0, FlagNoProfiling|FlagNoScan|FlagNoInvokeGC);
			s->types.compression = MTypes_Bytes;
			s->types.data = (uintptr)data3;
			((uintptr*)data3)[1] = typ;
			data3[8*sizeof(uintptr) + ofs] = 1;
			break;

		case MTypes_Words:
			((uintptr*)s->types.data)[ofs] = typ;
			break;

		case MTypes_Bytes:
			data3 = (byte*)s->types.data;
			for(j=1; j<8; j++) {
				if(((uintptr*)data3)[j] == typ) {
					break;
				}
				if(((uintptr*)data3)[j] == 0) {
					((uintptr*)data3)[j] = typ;
					break;
				}
			}
			if(j < 8) {
				data3[8*sizeof(uintptr) + ofs] = j;
			} else {
				ntypes = (s->npages << PageShift) / size;
				nbytes2 = ntypes * sizeof(uintptr);
				data2 = runtime·mallocgc(nbytes2, 0, FlagNoProfiling|FlagNoScan|FlagNoInvokeGC);
				s->types.compression = MTypes_Words;
				s->types.data = (uintptr)data2;

				// Move the contents of data3 to data2. Then deallocate data3.
				for(j=0; j<ntypes; j++) {
					t = data3[8*sizeof(uintptr) + j];
					t = ((uintptr*)data3)[t];
					data2[j] = t;
				}
				data2[ofs] = typ;
			}
			break;
		}
	}
	runtime·unlock(&settype_lock);

	mp->settype_bufsize = 0;
}

uintptr
runtime·gettype(void *v)
{
	MSpan *s;
	uintptr t, ofs;
	byte *data;

	s = runtime·MHeap_LookupMaybe(&runtime·mheap, v);
	if(s != nil) {
		t = 0;
		switch(s->types.compression) {
		case MTypes_Empty:
			break;
		case MTypes_Single:
			t = s->types.data;
			break;
		case MTypes_Words:
			ofs = (uintptr)v - (s->start<<PageShift);
			t = ((uintptr*)s->types.data)[ofs/s->elemsize];
			break;
		case MTypes_Bytes:
			ofs = (uintptr)v - (s->start<<PageShift);
			data = (byte*)s->types.data;
			t = data[8*sizeof(uintptr) + ofs/s->elemsize];
			t = ((uintptr*)data)[t];
			break;
		default:
			runtime·throw("runtime·gettype: invalid compression kind");
		}
		if(0) {
			runtime·lock(&settype_lock);
			runtime·printf("%p -> %d,%X\n", v, (int32)s->types.compression, (int64)t);
			runtime·unlock(&settype_lock);
		}
		return t;
	}
	return 0;
}

// Runtime stubs.

void*
runtime·mal(uintptr n)
{
	return runtime·mallocgc(n, 0, 0);
}

#pragma textflag NOSPLIT
void
runtime·new(Type *typ, uint8 *ret)
{
	ret = runtime·mallocgc(typ->size, (uintptr)typ | TypeInfo_SingleObject, typ->kind&KindNoPointers ? FlagNoScan : 0);
	FLUSH(&ret);
}

static void*
cnew(Type *typ, intgo n, int32 objtyp)
{
	if((objtyp&(PtrSize-1)) != objtyp)
		runtime·throw("runtime: invalid objtyp");
	if(n < 0 || (typ->size > 0 && n > MaxMem/typ->size))
		runtime·panicstring("runtime: allocation size out of range");
	return runtime·mallocgc(typ->size*n, (uintptr)typ | objtyp, typ->kind&KindNoPointers ? FlagNoScan : 0);
}

// same as runtime·new, but callable from C
void*
runtime·cnew(Type *typ)
{
	return cnew(typ, 1, TypeInfo_SingleObject);
}

void*
runtime·cnewarray(Type *typ, intgo n)
{
	return cnew(typ, n, TypeInfo_Array);
}

func GC() {
	runtime·gc(1);
}

func SetFinalizer(obj Eface, finalizer Eface) {
	byte *base;
	uintptr size;
	FuncType *ft;
	int32 i;
	uintptr nret;
	Type *t;
	Type *fint;
	PtrType *ot;
	Iface iface;

	if(obj.type == nil) {
		runtime·printf("runtime.SetFinalizer: first argument is nil interface\n");
		goto throw;
	}
	if(obj.type->kind != KindPtr) {
		runtime·printf("runtime.SetFinalizer: first argument is %S, not pointer\n", *obj.type->string);
		goto throw;
	}
	ot = (PtrType*)obj.type;
	// As an implementation detail we do not run finalizers for zero-sized objects,
	// because we use &runtime·zerobase for all such allocations.
	if(ot->elem != nil && ot->elem->size == 0)
		return;
	if(!runtime·mlookup(obj.data, &base, &size, nil) || obj.data != base) {
		// As an implementation detail we allow to set finalizers for an inner byte
		// of an object if it could come from tiny alloc (see mallocgc for details).
		if(ot->elem == nil || (ot->elem->kind&KindNoPointers) == 0 || ot->elem->size >= TinySize) {
			runtime·printf("runtime.SetFinalizer: pointer not at beginning of allocated block\n");
			goto throw;
		}
	}
	if(finalizer.type != nil) {
		if(finalizer.type->kind != KindFunc)
			goto badfunc;
		ft = (FuncType*)finalizer.type;
		if(ft->dotdotdot || ft->in.len != 1)
			goto badfunc;
		fint = *(Type**)ft->in.array;
		if(fint == obj.type) {
			// ok - same type
		} else if(fint->kind == KindPtr && (fint->x == nil || fint->x->name == nil || obj.type->x == nil || obj.type->x->name == nil) && ((PtrType*)fint)->elem == ((PtrType*)obj.type)->elem) {
			// ok - not same type, but both pointers,
			// one or the other is unnamed, and same element type, so assignable.
		} else if(fint->kind == KindInterface && ((InterfaceType*)fint)->mhdr.len == 0) {
			// ok - satisfies empty interface
		} else if(fint->kind == KindInterface && runtime·ifaceE2I2((InterfaceType*)fint, obj, &iface)) {
			// ok - satisfies non-empty interface
		} else
			goto badfunc;

		// compute size needed for return parameters
		nret = 0;
		for(i=0; i<ft->out.len; i++) {
			t = ((Type**)ft->out.array)[i];
			nret = ROUND(nret, t->align) + t->size;
		}
		nret = ROUND(nret, sizeof(void*));
		ot = (PtrType*)obj.type;
		if(!runtime·addfinalizer(obj.data, finalizer.data, nret, fint, ot)) {
			runtime·printf("runtime.SetFinalizer: finalizer already set\n");
			goto throw;
		}
	} else {
		// NOTE: asking to remove a finalizer when there currently isn't one set is OK.
		runtime·removefinalizer(obj.data);
	}
	return;

badfunc:
	runtime·printf("runtime.SetFinalizer: cannot pass %S to finalizer %S\n", *obj.type->string, *finalizer.type->string);
throw:
	runtime·throw("runtime.SetFinalizer");
}