// 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 "malloc.h"
#include "go-string.h"
#include "interface.h"
#include "go-type.h"
typedef struct __go_empty_interface Eface;
typedef struct __go_type_descriptor Type;
typedef struct __go_func_type FuncType;

MHeap runtime_mheap;
extern MStats mstats;	// defined in extern.go

extern volatile int32 runtime_MemProfileRate
  __asm__ ("libgo_runtime.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.
void*
runtime_mallocgc(uintptr size, uint32 refflag, int32 dogc, int32 zeroed)
{
	int32 sizeclass, rate;
	MCache *c;
	uintptr npages;
	MSpan *s;
	void *v;
	uint32 *ref;

	if(!__sync_bool_compare_and_swap(&m->mallocing, 0, 1))
		runtime_throw("malloc/free - deadlock");
	if(size == 0)
		size = 1;

	mstats.nmalloc++;
	if(size <= MaxSmallSize) {
		// Allocate from mcache free lists.
		sizeclass = runtime_SizeToClass(size);
		size = runtime_class_to_size[sizeclass];
		c = m->mcache;
		v = runtime_MCache_Alloc(c, sizeclass, size, zeroed);
		if(v == nil)
			runtime_throw("out of memory");
		mstats.alloc += size;
		mstats.total_alloc += size;
		mstats.by_size[sizeclass].nmalloc++;

		if(!runtime_mlookup(v, nil, nil, nil, &ref)) {
			// runtime_printf("malloc %D; runtime_mlookup failed\n", (uint64)size);
			runtime_throw("malloc runtime_mlookup");
		}
		*ref = RefNone | refflag;
	} 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;
		mstats.alloc += size;
		mstats.total_alloc += size;
		v = (void*)(s->start << PageShift);

		// setup for mark sweep
		s->gcref0 = RefNone | refflag;
		ref = &s->gcref0;
	}

	__sync_bool_compare_and_swap(&m->mallocing, 1, 0);

	if(__sync_bool_compare_and_swap(&m->gcing, 1, 0)) {
		if(!(refflag & RefNoProfiling))
			__go_run_goroutine_gc(0);
		else {
			// We are being called from the profiler.  Tell it
			// to invoke the garbage collector when it is
			// done.  No need to use a sync function here.
			m->gcing_for_prof = 1;
		}
	}

	if(!(refflag & RefNoProfiling) && (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(rate > 0x3fffffff)	// make 2*rate not overflow
				rate = 0x3fffffff;
			m->mcache->next_sample = fastrand1() % (2*rate);
		profile:
			*ref |= RefProfiled;
			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)
{
	int32 sizeclass, size;
	MSpan *s;
	MCache *c;
	uint32 prof, *ref;

	if(v == nil)
		return;

	if(!__sync_bool_compare_and_swap(&m->mallocing, 0, 1))
		runtime_throw("malloc/free - deadlock");

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

	// Find size class for v.
	sizeclass = s->sizeclass;
	if(sizeclass == 0) {
		// Large object.
		if(prof)
			runtime_MProf_Free(v, s->npages<<PageShift);
		mstats.alloc -= s->npages<<PageShift;
		runtime_memclr(v, s->npages<<PageShift);
		runtime_MHeap_Free(&runtime_mheap, s, 1);
	} else {
		// Small object.
		c = m->mcache;
		size = runtime_class_to_size[sizeclass];
		if(size > (int32)sizeof(uintptr))
			((uintptr*)v)[1] = 1;	// mark as "needs to be zeroed"
		if(prof)
			runtime_MProf_Free(v, size);
		mstats.alloc -= size;
		mstats.by_size[sizeclass].nfree++;
		runtime_MCache_Free(c, v, sizeclass, size);
	}
	__sync_bool_compare_and_swap(&m->mallocing, 1, 0);

	if(__sync_bool_compare_and_swap(&m->gcing, 1, 0))
		__go_run_goroutine_gc(1);
}

int32
runtime_mlookup(void *v, byte **base, uintptr *size, MSpan **sp, uint32 **ref)
{
	uintptr n, nobj, i;
	byte *p;
	MSpan *s;

	mstats.nlookup++;
	s = runtime_MHeap_LookupMaybe(&runtime_mheap, (uintptr)v>>PageShift);
	if(sp)
		*sp = s;
	if(s == nil) {
		if(base)
			*base = nil;
		if(size)
			*size = 0;
		if(ref)
			*ref = 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;
		if(ref)
			*ref = &s->gcref0;
		return 1;
	}

	if((byte*)v >= (byte*)s->gcref) {
		// pointers into the gc ref counts
		// do not count as pointers.
		return 0;
	}

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

	// good for error checking, but expensive
	if(0) {
		nobj = (s->npages << PageShift) / (n + RefcountOverhead);
		if((byte*)s->gcref < p || (byte*)(s->gcref+nobj) > p+(s->npages<<PageShift)) {
			// runtime_printf("odd span state=%d span=%p base=%p sizeclass=%d n=%D size=%D npages=%D\n",
			//	s->state, s, p, s->sizeclass, (uint64)nobj, (uint64)n, (uint64)s->npages);
			// runtime_printf("s->base sizeclass %d v=%p base=%p gcref=%p blocksize=%D nobj=%D size=%D end=%p end=%p\n",
			//	s->sizeclass, v, p, s->gcref, (uint64)s->npages<<PageShift,
			//	(uint64)nobj, (uint64)n, s->gcref + nobj, p+(s->npages<<PageShift));
			runtime_throw("bad gcref");
		}
	}
	if(ref)
		*ref = &s->gcref[i];

	return 1;
}

MCache*
runtime_allocmcache(void)
{
	MCache *c;

	if(!__sync_bool_compare_and_swap(&m->mallocing, 0, 1))
		runtime_throw("allocmcache - deadlock");

	runtime_lock(&runtime_mheap);
	c = runtime_FixAlloc_Alloc(&runtime_mheap.cachealloc);

	// Clear the free list used by FixAlloc; assume the rest is zeroed.
	c->list[0].list = nil;

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

	__sync_bool_compare_and_swap(&m->mallocing, 1, 0);
	if(__sync_bool_compare_and_swap(&m->gcing, 1, 0))
		__go_run_goroutine_gc(2);

	return c;
}

extern int32 runtime_sizeof_C_MStats
  __asm__ ("libgo_runtime.runtime.Sizeof_C_MStats");

void
runtime_mallocinit(void)
{
	runtime_sizeof_C_MStats = sizeof(MStats);

	runtime_initfintab();
	runtime_Mprof_Init();

	runtime_SysMemInit();
	runtime_InitSizes();
	runtime_MHeap_Init(&runtime_mheap, runtime_SysAlloc);
	m->mcache = runtime_allocmcache();

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

// Runtime stubs.

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

func Alloc(n uintptr) (p *byte) {
	p = runtime_malloc(n);
}

func Free(p *byte) {
	runtime_free(p);
}

func Lookup(p *byte) (base *byte, size uintptr) {
	runtime_mlookup(p, &base, &size, nil, nil);
}

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");
	throw:
		runtime_throw("runtime.SetFinalizer");
	}
	if(obj.__type_descriptor->__code != GO_PTR) {
		// runtime_printf("runtime.SetFinalizer: first argument is %S, not pointer\n", *obj.type->string);
		goto throw;
	}
	if(!runtime_mlookup(obj.__object, &base, &size, nil, 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) {
		badfunc:
			// runtime_printf("runtime.SetFinalizer: second argument is %S, not func(%S)\n", *finalizer.type->string, *obj.type->string);
			goto throw;
		}
		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_getfinalizer(obj.__object, 0)) {
			// runtime_printf("runtime.SetFinalizer: finalizer already set");
			goto throw;
		}
	}
	runtime_addfinalizer(obj.__object, finalizer.__type_descriptor != nil ? *(void**)finalizer.__object : nil, ft);
}