Go的栈空间管理源码剖析

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

void*runtime·stackalloc(uint32 n) {
	uint32 pos;
void*v;
	if(n == FixedStack || m->mallocing || m->gcing) {
		if(m->stackcachecnt == 0)
			stackcacherefill();
		pos = m->stackcachepos;
		pos = (pos - 1) % StackCacheSize;
		v = m->stackcache[pos];
		m->stackcachepos = pos;
		m->stackcachecnt--;
		m->stackinuse++;
		return v;
	}
	return runtime·mallocgc(n, 0, FlagNoProfiling|FlagNoGC|FlagNoZero|FlagNoInvokeGC);
}

1
2
3
4

type stack struct {
	lo uintptr
	hi uintptr
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15

type g struct {
    // Stack parameters.
    // stack describes the actual stack memory: [stack.lo, stack.hi).
    // stackguard0 is the stack pointer compared in the Go stack growth prologue.
    // It is stack.lo+StackGuard normally, but can be StackPreempt to trigger a preemption.
    // stackguard1 is the stack pointer compared in the C stack growth prologue.
    // It is stack.lo+StackGuard on g0 and gsignal stacks.
    // It is ~0 on other goroutine stacks, to trigger a call to morestackc (and crash).
    stack       stack   // offset known to runtime/cgo
    // stackguard0 用于跟函数的 sp 进行比较，通常等于 stack.lo+StackGuard，但是在需要触发抢占调度时，会被赋值为 StackPreempt
    stackguard0 uintptr // offset known to liblink
    // stackguard1 用于跟系统线程的 sp 进行比较，在 g0 和 gsignal 上等于 stack.lo+StackGuard，在其他 goroutine 上等于 ~0，用于触发栈扩容和崩溃
    stackguard1 uintptr // offset known to liblink
    ......
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29

var stackpool [_NumStackOrders]struct {
	item stackpoolItem
	_    [cpu.CacheLinePadSize - unsafe.Sizeof(stackpoolItem{})%cpu.CacheLinePadSize]byte
}

type stackpoolItem struct {
	mu   mutex
	span mSpanList
}

//大空间内存池
var stackLarge struct {
        //由于是全局共享变量，所以使用时会加锁
    lock mutex

    //heapAddrBits，
    //pageShift=13,go的内存管理中一页的大小为8kb，此值代表log2(8*1024) = 13
    //索引其实是从2开始才可能有空闲内存
    //2 => 全为2^2 * page大小的mspan   （注：page大小为8KB）
    //3 => 全为2^3 * page大小的mspan
    //...
    free [heapAddrBits - pageShift]mSpanList
}

//mspan结构中有三个指针next、prev、list(指向mSpanList)
type mSpanList struct {
    first *mspan // first span in list, or nil if none
    last  *mspan // last span in list, or nil if none
}

1
2
3
4
5
6
7
8

func stackinit() {
	for i := range stackpool {
		stackpool[i].item.span.init()
	}
	for i := range stackLarge.free {
		stackLarge.free[i].init()
	}
}

1
2
3
4
5
6
7
8

type mcache struct {
	stackcache [_NumStackOrders]stackfreelist
}

type stackfreelist struct {
	list gclinkptr
	size uintptr
}

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126

// stackalloc allocates an n byte stack.
//
// stackalloc must run on the system stack because it uses per-P
// resources and must not split the stack.
//
//go:systemstack
stackNoCache     = 0 // disable per-P small stack caches
func stackalloc(n uint32) stack {
	// Stackalloc must be called on scheduler stack, so that we
	// never try to grow the stack during the code that stackalloc runs.
	// Doing so would cause a deadlock (issue 1547).
	//必须在调度程序堆栈上调用Stackalloc，
	//以防止在正在执行此动作时发送栈扩导致的死锁
	// 因为是在 systemstack 上执行的，当前 g 必须是 g0
	thisg := getg()
	if thisg != thisg.m.g0 {
		throw("stackalloc not on scheduler stack")
	}
	if n&(n-1) != 0 {
		throw("stack size not a power of 2")
	}
	if stackDebug >= 1 {
		print("stackalloc ", n, "\n")
	}

	if debug.efence != 0 || stackFromSystem != 0 {
		n = uint32(alignUp(uintptr(n), physPageSize))
		v := sysAlloc(uintptr(n), &memstats.stacks_sys)
		if v == nil {
			throw("out of memory (stackalloc)")
		}
		return stack{uintptr(v), uintptr(v) + uintptr(n)}
	}

	// Small stacks are allocated with a fixed-size free-list allocator.
	// If we need a stack of a bigger size, we fall back on allocating
	// a dedicated span.
	//需要小堆栈在固定大小空闲列表中分配，大堆栈则到专用span中申请分配
    //_StackCacheSize 是一个常量32*1024 即32KB
	var v unsafe.Pointer
	//如果栈空间较小，使用全局栈缓存或者线程缓存上固定大小的空闲链表分配内存；
	if n < _FixedStack<<_NumStackOrders && n < _StackCacheSize {
		// 需要分配的栈空间比较小
        // order 用于索引对应大小的栈空间
		order := uint8(0)
		n2 := n
		//order 代表: n是_FixedStack的几倍
        // 0: _FixedStack大小，1：代表2倍_FixedStack大小；2代表四倍
		for n2 > _FixedStack {
			order++
			n2 >>= 1
		}
		var x gclinkptr
		// 如果
        // 1. 不使用 P 的空闲栈空间缓存
		// 2. 当前 m 没有关联的P,也就没有 mcache
        // 3. thisg.m.preemptoff 不为空
        // 则从全局的空闲栈空间池子中分配
		if stackNoCache != 0 || thisg.m.p == 0 || thisg.m.preemptoff != "" {
			// thisg.m.p == 0 can happen in the guts of exitsyscall
			// or procresize. Just get a stack from the global pool.
			// Also don't touch stackcache during gc
			// as it's flushed concurrently.
			lock(&stackpool[order].item.mu)
			x = stackpoolalloc(order)
			unlock(&stackpool[order].item.mu)
		} else {
			//本地内存缓存mcache.stackcache中获取可用内存
        	//如果本地缓存中存在，则直接获取，并更新链表
        	//如果不存在则 调用stackpoolalloc 向mheap去申请
			c := thisg.m.p.ptr().mcache
			x = c.stackcache[order].list
			if x.ptr() == nil {
				//stackcacherefill会调用stackpoolalloc然后把申请到的内存填充到c.stackcache 中
				stackcacherefill(c, order)
				x = c.stackcache[order].list
			}
			c.stackcache[order].list = x.ptr().next
			c.stackcache[order].size -= uintptr(n)
		}
		v = unsafe.Pointer(x)
	} else {
		//如果栈空间较大，从全局的大栈缓存 runtime.stackLarge 中获取内存空间；
		//如果是大空间(大于等于32KB)
        //则先到stackLarge的数组中获取，如果对应下标的数组为空则向mheap申请
		var s *mspan
		npage := uintptr(n) >> _PageShift
		// stacklog2 返回 log_2(n)
		log2npage := stacklog2(npage)

		// Try to get a stack from the large stack cache.
		//stackLarge数组所指向的空闲内存空间全部是通过栈回收来获得的。
		lock(&stackLarge.lock)
		if !stackLarge.free[log2npage].isEmpty() {
			s = stackLarge.free[log2npage].first
			stackLarge.free[log2npage].remove(s)
		}
		unlock(&stackLarge.lock)

		lockWithRankMayAcquire(&mheap_.lock, lockRankMheap)

		if s == nil {
			//如果栈空间较大并且 runtime.stackLarge 空间不足，在堆上申请一片大小足够内存空间；
			//向mheap申请npage大小的内存空间用于栈
			// Allocate a new stack from the heap.
			s = mheap_.allocManual(npage, &memstats.stacks_inuse)
			if s == nil {
				throw("out of memory")
			}
			osStackAlloc(s)
			s.elemsize = uintptr(n)
		}
		v = unsafe.Pointer(s.base())
	}

	if raceenabled {
		racemalloc(v, uintptr(n))
	}
	if msanenabled {
		msanmalloc(v, uintptr(n))
	}
	if stackDebug >= 1 {
		print("  allocated ", v, "\n")
	}
	return stack{uintptr(v), uintptr(v) + uintptr(n)}
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

// stackcacherefill/stackcacherelease implement a global pool of stack segments.
// The pool is required to prevent unlimited growth of per-thread caches.
//
//go:systemstack
func stackcacherefill(c *mcache, order uint8) {
	if stackDebug >= 1 {
		print("stackcacherefill order=", order, "\n")
	}

	// Grab some stacks from the global cache.
	// Grab half of the allowed capacity (to prevent thrashing).
	var list gclinkptr
	var size uintptr
	lock(&stackpool[order].item.mu)
	//这里有一个循环获取的方法，看起来是获取的大小要超过_StackCacheSize的一半，官方给出的解释是防止系统颠簸
	for size < _StackCacheSize/2 {
		x := stackpoolalloc(order)
		x.ptr().next = list
		list = x
		size += _FixedStack << order
	}
	unlock(&stackpool[order].item.mu)
	c.stackcache[order].list = list
	c.stackcache[order].size = size
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49

// Allocates a stack from the free pool. Must be called with
// stackpool[order].item.mu held.
func stackpoolalloc(order uint8) gclinkptr {
	//我们有一个全局的缓存池，stackpool，存储的就是不同规格的spanlist，首先我们获取对应规格的spanlist
	list := &stackpool[order].item.span
	//s为指向span的指针
	s := list.first
	lockWithRankMayAcquire(&mheap_.lock, lockRankMheap)
	//如果list是空的，我们就调用allocManual申请新的span
	if s == nil {
		// no free stacks. Allocate another span worth.
		//一次申请32KB内存即4页
		s = mheap_.allocManual(_StackCacheSize>>_PageShift, &memstats.stacks_inuse)
		if s == nil {
			throw("out of memory")
		}
		if s.allocCount != 0 {
			throw("bad allocCount")
		}
		if s.manualFreeList.ptr() != nil {
			throw("bad manualFreeList")
		}
		//对新申请的内存，在使用之前先初始化系统栈
		osStackAlloc(s)
		s.elemsize = _FixedStack << order
		for i := uintptr(0); i < _StackCacheSize; i += s.elemsize {
			//gclinkptr 也是一个指针类型
            //作用是屏蔽gc扫描
			x := gclinkptr(s.base() + i)
			//链表头插法
			x.ptr().next = s.manualFreeList
			s.manualFreeList = x
		}
		list.insert(s)
	}
	x := s.manualFreeList
	if x.ptr() == nil {
		throw("span has no free stacks")
	}
	//获取到list后，我们读取第一个span，然后分配manualFreeList的第一个单元给栈即可，如果manualFreeList用完了，则直接移除该span即可。
	s.manualFreeList = x.ptr().next
	s.allocCount++
	if s.manualFreeList.ptr() == nil {
		// all stacks in s are allocated.
		//所有内存已经分配完毕，删除节点s
		list.remove(s)
	}
	return x
}

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249

// allocManual allocates a manually-managed span of npage pages.
// allocManual returns nil if allocation fails.
//
// allocManual adds the bytes used to *stat, which should be a
// memstats in-use field. Unlike allocations in the GC'd heap, the
// allocation does *not* count toward heap_inuse or heap_sys.
//
// The memory backing the returned span may not be zeroed if
// span.needzero is set.
//
// allocManual must be called on the system stack because it may
// acquire the heap lock via allocSpan. See mheap for details.
//
//go:systemstack
func (h *mheap) allocManual(npages uintptr, stat *uint64) *mspan {
	return h.allocSpan(npages, true, 0, stat)
}
// allocSpan allocates an mspan which owns npages worth of memory.
//
// If manual == false, allocSpan allocates a heap span of class spanclass
// and updates heap accounting. If manual == true, allocSpan allocates a
// manually-managed span (spanclass is ignored), and the caller is
// responsible for any accounting related to its use of the span. Either
// way, allocSpan will atomically add the bytes in the newly allocated
// span to *sysStat.
//
// The returned span is fully initialized.
//
// h must not be locked.
//
// allocSpan must be called on the system stack both because it acquires
// the heap lock and because it must block GC transitions.
//
//go:systemstack
func (h *mheap) allocSpan(npages uintptr, manual bool, spanclass spanClass, sysStat *uint64) (s *mspan) {
	// Function-global state.
	gp := getg()
	base, scav := uintptr(0), uintptr(0)

	// If the allocation is small enough, try the page cache!
	pp := gp.m.p.ptr()
	if pp != nil && npages < pageCachePages/4 {
		c := &pp.pcache

		// If the cache is empty, refill it.
		if c.empty() {
			lock(&h.lock)
			*c = h.pages.allocToCache()
			unlock(&h.lock)
		}

		// Try to allocate from the cache.
		base, scav = c.alloc(npages)
		if base != 0 {
			s = h.tryAllocMSpan()

			if s != nil && gcBlackenEnabled == 0 && (manual || spanclass.sizeclass() != 0) {
				goto HaveSpan
			}
			// We're either running duing GC, failed to acquire a mspan,
			// or the allocation is for a large object. This means we
			// have to lock the heap and do a bunch of extra work,
			// so go down the HaveBaseLocked path.
			//
			// We must do this during GC to avoid skew with heap_scan
			// since we flush mcache stats whenever we lock.
			//
			// TODO(mknyszek): It would be nice to not have to
			// lock the heap if it's a large allocation, but
			// it's fine for now. The critical section here is
			// short and large object allocations are relatively
			// infrequent.
		}
	}

	// For one reason or another, we couldn't get the
	// whole job done without the heap lock.
	lock(&h.lock)

	if base == 0 {
		// Try to acquire a base address.
		base, scav = h.pages.alloc(npages)
		if base == 0 {
			if !h.grow(npages) {
				unlock(&h.lock)
				return nil
			}
			base, scav = h.pages.alloc(npages)
			if base == 0 {
				throw("grew heap, but no adequate free space found")
			}
		}
	}
	if s == nil {
		// We failed to get an mspan earlier, so grab
		// one now that we have the heap lock.
		s = h.allocMSpanLocked()
	}
	if !manual {
		// This is a heap span, so we should do some additional accounting
		// which may only be done with the heap locked.

		// Transfer stats from mcache to global.
		var c *mcache
		if gp.m.p != 0 {
			c = gp.m.p.ptr().mcache
		} else {
			// This case occurs while bootstrapping.
			// See the similar code in mallocgc.
			c = mcache0
			if c == nil {
				throw("mheap.allocSpan called with no P")
			}
		}
		memstats.heap_scan += uint64(c.local_scan)
		c.local_scan = 0
		memstats.tinyallocs += uint64(c.local_tinyallocs)
		c.local_tinyallocs = 0

		// Do some additional accounting if it's a large allocation.
		if spanclass.sizeclass() == 0 {
			mheap_.largealloc += uint64(npages * pageSize)
			mheap_.nlargealloc++
			atomic.Xadd64(&memstats.heap_live, int64(npages*pageSize))
		}

		// Either heap_live or heap_scan could have been updated.
		if gcBlackenEnabled != 0 {
			gcController.revise()
		}
	}
	unlock(&h.lock)

HaveSpan:
	// At this point, both s != nil and base != 0, and the heap
	// lock is no longer held. Initialize the span.
	s.init(base, npages)
	if h.allocNeedsZero(base, npages) {
		s.needzero = 1
	}
	nbytes := npages * pageSize
	if manual {
		s.manualFreeList = 0
		s.nelems = 0
		s.limit = s.base() + s.npages*pageSize
		// Manually managed memory doesn't count toward heap_sys.
		mSysStatDec(&memstats.heap_sys, s.npages*pageSize)
		s.state.set(mSpanManual)
	} else {
		// We must set span properties before the span is published anywhere
		// since we're not holding the heap lock.
		s.spanclass = spanclass
		if sizeclass := spanclass.sizeclass(); sizeclass == 0 {
			s.elemsize = nbytes
			s.nelems = 1

			s.divShift = 0
			s.divMul = 0
			s.divShift2 = 0
			s.baseMask = 0
		} else {
			s.elemsize = uintptr(class_to_size[sizeclass])
			s.nelems = nbytes / s.elemsize

			m := &class_to_divmagic[sizeclass]
			s.divShift = m.shift
			s.divMul = m.mul
			s.divShift2 = m.shift2
			s.baseMask = m.baseMask
		}

		// Initialize mark and allocation structures.
		s.freeindex = 0
		s.allocCache = ^uint64(0) // all 1s indicating all free.
		s.gcmarkBits = newMarkBits(s.nelems)
		s.allocBits = newAllocBits(s.nelems)

		// It's safe to access h.sweepgen without the heap lock because it's
		// only ever updated with the world stopped and we run on the
		// systemstack which blocks a STW transition.
		atomic.Store(&s.sweepgen, h.sweepgen)

		// Now that the span is filled in, set its state. This
		// is a publication barrier for the other fields in
		// the span. While valid pointers into this span
		// should never be visible until the span is returned,
		// if the garbage collector finds an invalid pointer,
		// access to the span may race with initialization of
		// the span. We resolve this race by atomically
		// setting the state after the span is fully
		// initialized, and atomically checking the state in
		// any situation where a pointer is suspect.
		s.state.set(mSpanInUse)
	}

	// Commit and account for any scavenged memory that the span now owns.
	if scav != 0 {
		// sysUsed all the pages that are actually available
		// in the span since some of them might be scavenged.
		sysUsed(unsafe.Pointer(base), nbytes)
		mSysStatDec(&memstats.heap_released, scav)
	}
	// Update stats.
	mSysStatInc(sysStat, nbytes)
	mSysStatDec(&memstats.heap_idle, nbytes)

	// Publish the span in various locations.

	// This is safe to call without the lock held because the slots
	// related to this span will only ever be read or modified by
	// this thread until pointers into the span are published (and
	// we execute a publication barrier at the end of this function
	// before that happens) or pageInUse is updated.
	h.setSpans(s.base(), npages, s)

	if !manual {
		if !go115NewMCentralImpl {
			// Add to swept in-use list.
			//
			// This publishes the span to root marking.
			//
			// h.sweepgen is guaranteed to only change during STW,
			// and preemption is disabled in the page allocator.
			h.sweepSpans[h.sweepgen/2%2].push(s)
		}

		// Mark in-use span in arena page bitmap.
		//
		// This publishes the span to the page sweeper, so
		// it's imperative that the span be completely initialized
		// prior to this line.
		arena, pageIdx, pageMask := pageIndexOf(s.base())
		atomic.Or8(&arena.pageInUse[pageIdx], pageMask)

		// Update related page sweeper stats.
		atomic.Xadd64(&h.pagesInUse, int64(npages))

		if trace.enabled {
			// Trace that a heap alloc occurred.
			traceHeapAlloc()
		}
	}

	// Make sure the newly allocated span will be observed
	// by the GC before pointers into the span are published.
	publicationBarrier()

	return s
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

// cmd/internal/obj/x86/obj6.go
func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
	...
	p := cursym.Func.Text
	autoffset := int32(p.To.Offset) // 栈帧大小
	// 一些额外的栈帧大小计算
	...
	if !cursym.Func.Text.From.Sym.NoSplit() {
		p = stacksplit(ctxt, cursym, p, newprog, autoffset, int32(textarg)) // 触发分段检查
	}
	...
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45

func stacksplit(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog, newprog obj.ProgAlloc, framesize int32, textarg int32) *obj.Prog {
	...

	var q1 *obj.Prog
	if framesize <= objabi.StackSmall {
		// 小栈: SP <= stackguard，直接比较 SP 和 stackguard
		//	CMPQ SP, stackguard
		...
	} else if framesize <= objabi.StackBig {
		// 大栈: SP-framesize <= stackguard-StackSmall
		//	LEAQ -xxx(SP), AX
		//	CMPQ AX, stackguard
		...
	} else {
		// 更大的栈需要防止 wraparound
		// 如果 SP 接近于零:
		//	SP-stackguard+StackGuard <= framesize + (StackGuard-StackSmall)
		// 两端的 +StackGuard 是为了保证左侧大于零。
		// SP 允许位于 stackguard 下面一点点
		//
		// 抢占设置了 stackguard 为 StackPreempt，一个大到能够打破上面的数学计算的值，
		// 因此必须显式的进行检查：
		//	MOVQ	stackguard, CX
		//	CMPQ	CX, $StackPreempt
		//	JEQ	label-of-call-to-morestack
		//	LEAQ	StackGuard(SP), AX
		//	SUBQ	CX, AX
		//	CMPQ	AX, $(framesize+(StackGuard-StackSmall))
		...
	}

	...
	// 函数的尾声
	morestack := "runtime.morestack"
	switch {
	case cursym.CFunc():
		morestack = "runtime.morestackc" // morestackc 会 panic，因为此时是系统栈上的 C 函数
	case !cursym.Func.Text.From.Sym.NeedCtxt():
		morestack = "runtime.morestack_noctxt"
	}
	call.To.Sym = ctxt.Lookup(morestack)
	...

	return jls
}

1
2
3

TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0
	MOVL	$0, DX
	JMP	runtime·morestack(SB)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44

TEXT runtime·morestack(SB),NOSPLIT,$0-0
	// 无法增长调度器的栈(m->g0)
	get_tls(CX)
	MOVQ	g(CX), BX
	MOVQ	g_m(BX), BX
	MOVQ	m_g0(BX), SI
	CMPQ	g(CX), SI
	JNE	3(PC)
	CALL	runtime·badmorestackg0(SB)
	CALL	runtime·abort(SB)

	// 无法增长信号栈 (m->gsignal)
	MOVQ	m_gsignal(BX), SI
	CMPQ	g(CX), SI
	JNE	3(PC)
	CALL	runtime·badmorestackgsignal(SB)
	CALL	runtime·abort(SB)

	// 从 f 调用
	// 将 m->morebuf 设置为 f 的调用方
	MOVQ	8(SP), AX	// f 的调用方 PC
	MOVQ	AX, (m_morebuf+gobuf_pc)(BX)
	LEAQ	16(SP), AX	// f 的调用方 SP
	MOVQ	AX, (m_morebuf+gobuf_sp)(BX)
	get_tls(CX)
	MOVQ	g(CX), SI
	MOVQ	SI, (m_morebuf+gobuf_g)(BX)

	// 将 g->sched 设置为 f 的 context
	MOVQ	0(SP), AX // f 的 PC
	MOVQ	AX, (g_sched+gobuf_pc)(SI)
	MOVQ	SI, (g_sched+gobuf_g)(SI)
	LEAQ	8(SP), AX // f 的 SP
	MOVQ	AX, (g_sched+gobuf_sp)(SI)
	MOVQ	BP, (g_sched+gobuf_bp)(SI)
	MOVQ	DX, (g_sched+gobuf_ctxt)(SI)

	// 在 m->g0 栈上调用 newstack.
	MOVQ	m_g0(BX), BX
	MOVQ	BX, g(CX)
	MOVQ	(g_sched+gobuf_sp)(BX), SP
	CALL	runtime·newstack(SB)
	CALL	runtime·abort(SB)	// 如果 newstack 返回则崩溃
	RET

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163

// Called from runtime·morestack when more stack is needed.
// Allocate larger stack and relocate to new stack.
// Stack growth is multiplicative, for constant amortized cost.
//
// g->atomicstatus will be Grunning or Gscanrunning upon entry.
// If the scheduler is trying to stop this g, then it will set preemptStop.
//
// This must be nowritebarrierrec because it can be called as part of
// stack growth from other nowritebarrierrec functions, but the
// compiler doesn't check this.
//
//go:nowritebarrierrec
func newstack() {
	//先做一些准备工作
	thisg := getg()
	// TODO: double check all gp. shouldn't be getg().
	if thisg.m.morebuf.g.ptr().stackguard0 == stackFork {
		throw("stack growth after fork")
	}
	if thisg.m.morebuf.g.ptr() != thisg.m.curg {
		print("runtime: newstack called from g=", hex(thisg.m.morebuf.g), "\n"+"\tm=", thisg.m, " m->curg=", thisg.m.curg, " m->g0=", thisg.m.g0, " m->gsignal=", thisg.m.gsignal, "\n")
		morebuf := thisg.m.morebuf
		traceback(morebuf.pc, morebuf.sp, morebuf.lr, morebuf.g.ptr())
		throw("runtime: wrong goroutine in newstack")
	}

	gp := thisg.m.curg

	if thisg.m.curg.throwsplit {
		// Update syscallsp, syscallpc in case traceback uses them.
		morebuf := thisg.m.morebuf
		gp.syscallsp = morebuf.sp
		gp.syscallpc = morebuf.pc
		pcname, pcoff := "(unknown)", uintptr(0)
		f := findfunc(gp.sched.pc)
		if f.valid() {
			pcname = funcname(f)
			pcoff = gp.sched.pc - f.entry
		}
		print("runtime: newstack at ", pcname, "+", hex(pcoff),
			" sp=", hex(gp.sched.sp), " stack=[", hex(gp.stack.lo), ", ", hex(gp.stack.hi), "]\n",
			"\tmorebuf={pc:", hex(morebuf.pc), " sp:", hex(morebuf.sp), " lr:", hex(morebuf.lr), "}\n",
			"\tsched={pc:", hex(gp.sched.pc), " sp:", hex(gp.sched.sp), " lr:", hex(gp.sched.lr), " ctxt:", gp.sched.ctxt, "}\n")

		thisg.m.traceback = 2 // Include runtime frames
		traceback(morebuf.pc, morebuf.sp, morebuf.lr, gp)
		throw("runtime: stack split at bad time")
	}

	morebuf := thisg.m.morebuf
	thisg.m.morebuf.pc = 0
	thisg.m.morebuf.lr = 0
	thisg.m.morebuf.sp = 0
	thisg.m.morebuf.g = 0

	// NOTE: stackguard0 may change underfoot, if another thread
	// is about to try to preempt gp. Read it just once and use that same
	// value now and below.
	preempt := atomic.Loaduintptr(&gp.stackguard0) == stackPreempt

	// Be conservative about where we preempt.
	// We are interested in preempting user Go code, not runtime code.
	// If we're holding locks, mallocing, or preemption is disabled, don't
	// preempt.
	// This check is very early in newstack so that even the status change
	// from Grunning to Gwaiting and back doesn't happen in this case.
	// That status change by itself can be viewed as a small preemption,
	// because the GC might change Gwaiting to Gscanwaiting, and then
	// this goroutine has to wait for the GC to finish before continuing.
	// If the GC is in some way dependent on this goroutine (for example,
	// it needs a lock held by the goroutine), that small preemption turns
	// into a real deadlock.
	//检查当前 Goroutine 是否发出了抢占请求，如果发出了抢占请求：
	if preempt {
		//当前线程可以被抢占时，直接调用 runtime.gogo 触发调度器的调度；
		if !canPreemptM(thisg.m) {
			// Let the goroutine keep running for now.
			// gp->preempt is set, so it will be preempted next time.
			gp.stackguard0 = gp.stack.lo + _StackGuard
			gogo(&gp.sched) // never return
		}
	}

	if gp.stack.lo == 0 {
		throw("missing stack in newstack")
	}
	sp := gp.sched.sp
	if sys.ArchFamily == sys.AMD64 || sys.ArchFamily == sys.I386 || sys.ArchFamily == sys.WASM {
		// The call to morestack cost a word.
		// 到 morestack 的调用会消耗一个字
		sp -= sys.PtrSize
	}
	if stackDebug >= 1 || sp < gp.stack.lo {
		print("runtime: newstack sp=", hex(sp), " stack=[", hex(gp.stack.lo), ", ", hex(gp.stack.hi), "]\n",
			"\tmorebuf={pc:", hex(morebuf.pc), " sp:", hex(morebuf.sp), " lr:", hex(morebuf.lr), "}\n",
			"\tsched={pc:", hex(gp.sched.pc), " sp:", hex(gp.sched.sp), " lr:", hex(gp.sched.lr), " ctxt:", gp.sched.ctxt, "}\n")
	}
	if sp < gp.stack.lo {
		print("runtime: gp=", gp, ", goid=", gp.goid, ", gp->status=", hex(readgstatus(gp)), "\n ")
		print("runtime: split stack overflow: ", hex(sp), " < ", hex(gp.stack.lo), "\n")
		throw("runtime: split stack overflow")
	}
	//检查当前 Goroutine 是否发出了抢占请求，如果发出了抢占请求：
	if preempt {
		if gp == thisg.m.g0 {
			throw("runtime: preempt g0")
		}
		if thisg.m.p == 0 && thisg.m.locks == 0 {
			throw("runtime: g is running but p is not")
		}
		//如果当前 Goroutine 在垃圾回收被 runtime.scanstack 函数标记成了需要收缩栈，调用 runtime.shrinkstack；
		if gp.preemptShrink {
			// We're at a synchronous safe point now, so
			// do the pending stack shrink.
			gp.preemptShrink = false
			shrinkstack(gp)
		}
		//如果当前 Goroutine 被 runtime.suspendG 函数挂起，调用 runtime.preemptPark 被动让出当前处理器的控制权并将 Goroutine 的状态修改至 _Gpreempted
		if gp.preemptStop {
			preemptPark(gp) // never returns
		}

		// Act like goroutine called runtime.Gosched.
		//调用 runtime.gopreempt_m 主动让出当前处理器的控制权；
		gopreempt_m(gp) // never return
	}
	//如果当前 Goroutine 不需要被抢占，也就意味着我们需要新的栈空间来支持函数调用和本地变量的初始化，运行时会先检查目标大小的栈是否会溢出：
	// Allocate a bigger segment and move the stack.
	// 分配一个更大的段，并对栈进行移动
	oldsize := gp.stack.hi - gp.stack.lo
	newsize := oldsize * 2	// 比原来大一倍

	// Make sure we grow at least as much as needed to fit the new frame.
	// (This is just an optimization - the caller of morestack will
	// recheck the bounds on return.)
	if f := findfunc(gp.sched.pc); f.valid() {
		max := uintptr(funcMaxSPDelta(f))
		for newsize-oldsize < max+_StackGuard {
			newsize *= 2
		}
	}
	// 需要的栈太大，直接溢出
	if newsize > maxstacksize {
		print("runtime: goroutine stack exceeds ", maxstacksize, "-byte limit\n")
		print("runtime: sp=", hex(sp), " stack=[", hex(gp.stack.lo), ", ", hex(gp.stack.hi), "]\n")
		throw("stack overflow")
	}
	//如果目标栈的大小没有超出程序的限制，我们会将 Goroutine 切换至 _Gcopystack 状态并调用 runtime.copystack 开始栈的拷贝
    //处于_Gcopystack 状态时 GC不会扫描栈空间
	// The goroutine must be executing in order to call newstack,
	// so it must be Grunning (or Gscanrunning).
	casgstatus(gp, _Grunning, _Gcopystack)

	// The concurrent GC will not scan the stack while we are doing the copy since
	// the gp is in a Gcopystack status.
	copystack(gp, newsize)	//修改协程状态
	if stackDebug >= 1 {
		print("stack grow done\n")
	}
	casgstatus(gp, _Gcopystack, _Grunning)
	gogo(&gp.sched)
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64

// Maybe shrink the stack being used by gp.
//
// gp must be stopped and we must own its stack. It may be in
// _Grunning, but only if this is our own user G.
func shrinkstack(gp *g) {
	if gp.stack.lo == 0 {
		throw("missing stack in shrinkstack")
	}
	if s := readgstatus(gp); s&_Gscan == 0 {
		// We don't own the stack via _Gscan. We could still
		// own it if this is our own user G and we're on the
		// system stack.
		if !(gp == getg().m.curg && getg() != getg().m.curg && s == _Grunning) {
			// We don't own the stack.
			throw("bad status in shrinkstack")
		}
	}
	if !isShrinkStackSafe(gp) {
		throw("shrinkstack at bad time")
	}
	// Check for self-shrinks while in a libcall. These may have
	// pointers into the stack disguised as uintptrs, but these
	// code paths should all be nosplit.
	if gp == getg().m.curg && gp.m.libcallsp != 0 {
		throw("shrinking stack in libcall")
	}

	if debug.gcshrinkstackoff > 0 {
		return
	}
	f := findfunc(gp.startpc)
	if f.valid() && f.funcID == funcID_gcBgMarkWorker {
		// We're not allowed to shrink the gcBgMarkWorker
		// stack (see gcBgMarkWorker for explanation).
		return
	}

	oldsize := gp.stack.hi - gp.stack.lo
	newsize := oldsize / 2
	// Don't shrink the allocation below the minimum-sized stack
	// allocation.
	if newsize < _FixedStack {
		return
	}
	// Compute how much of the stack is currently in use and only
	// shrink the stack if gp is using less than a quarter of its
	// current stack. The currently used stack includes everything
	// down to the SP plus the stack guard space that ensures
	// there's room for nosplit functions.
	//当所有的已使用的空间小于栈总空间的1/4时，栈收缩为原来的一半
	//gp.sched.sp 为栈顶指针kongjian
	//_StackLimit 在不扩容情况下执行被调函数所能用的最大空间
	avail := gp.stack.hi - gp.stack.lo
	//当已使用的栈占不到总栈的1/4 进行缩容
	if used := gp.stack.hi - gp.sched.sp + _StackLimit; used >= avail/4 {
		return
	}

	if stackDebug > 0 {
		print("shrinking stack ", oldsize, "->", newsize, "\n")
	}

	copystack(gp, newsize)
}

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148

// Copies gp's stack to a new stack of a different size.
// Caller must have changed gp status to Gcopystack.
func copystack(gp *g, newsize uintptr) {
	if gp.syscallsp != 0 {
		throw("stack growth not allowed in system call")
	}
	old := gp.stack
	if old.lo == 0 {
		throw("nil stackbase")
	}
	used := old.hi - gp.sched.sp

	// allocate new stack
	//在拷贝栈的内存之前，运行时会通过 runtime.stackalloc 函数分配新的栈空间
	new := stackalloc(uint32(newsize))
	if stackPoisonCopy != 0 {
		fillstack(new, 0xfd)
	}
	if stackDebug >= 1 {
		print("copystack gp=", gp, " [", hex(old.lo), " ", hex(old.hi-used), " ", hex(old.hi), "]", " -> [", hex(new.lo), " ", hex(new.hi-used), " ", hex(new.hi), "]/", newsize, "\n")
	}

	// Compute adjustment.
	// 这里主要调整g的一些调度相关参数
    // 如果它们存储在老的栈上面，需要将它们拷贝到新栈上
	var adjinfo adjustinfo
	adjinfo.old = old
	adjinfo.delta = new.hi - old.hi	// 计算新栈和旧栈之间内存地址差

	// Adjust sudogs, synchronizing with channel ops if necessary.
	ncopy := used
	if !gp.activeStackChans {
		if newsize < old.hi-old.lo && atomic.Load8(&gp.parkingOnChan) != 0 {
			// It's not safe for someone to shrink this stack while we're actively
			// parking on a channel, but it is safe to grow since we do that
			// ourselves and explicitly don't want to synchronize with channels
			// since we could self-deadlock.
			throw("racy sudog adjustment due to parking on channel")
		}
		adjustsudogs(gp, &adjinfo)
	} else {
		// sudogs may be pointing in to the stack and gp has
		// released channel locks, so other goroutines could
		// be writing to gp's stack. Find the highest such
		// pointer so we can handle everything there and below
		// carefully. (This shouldn't be far from the bottom
		// of the stack, so there's little cost in handling
		// everything below it carefully.)
		adjinfo.sghi = findsghi(gp, old)

		// Synchronize with channel ops and copy the part of
		// the stack they may interact with.
		ncopy -= syncadjustsudogs(gp, used, &adjinfo)
	}

	// Copy the stack (or the rest of it) to the new location
	//把旧栈数据复制到新栈
	memmove(unsafe.Pointer(new.hi-ncopy), unsafe.Pointer(old.hi-ncopy), ncopy)

	// Adjust remaining structures that have pointers into stacks.
	// We have to do most of these before we traceback the new
	// stack because gentraceback uses them.
	adjustctxt(gp, &adjinfo)
	adjustdefers(gp, &adjinfo)
	adjustpanics(gp, &adjinfo)
	if adjinfo.sghi != 0 {
		adjinfo.sghi += adjinfo.delta
	}

	// Swap out old stack for new one
	// 切换到新堆栈上工作
	gp.stack = new
	gp.stackguard0 = new.lo + _StackGuard // NOTE: might clobber a preempt request
	gp.sched.sp = new.hi - used
	gp.stktopsp += adjinfo.delta

	// Adjust pointers in the new stack.
	// 在新栈重调整指针
	gentraceback(^uintptr(0), ^uintptr(0), 0, gp, 0, nil, 0x7fffffff, adjustframe, noescape(unsafe.Pointer(&adjinfo)), 0)

	// free old stack
	// 释放老的堆栈
	if stackPoisonCopy != 0 {
		fillstack(old, 0xfc)
	}
	//调用runtime.stackfree销毁并回收旧栈的内存空间；
	stackfree(old)
}


func fillstack(stk stack, b byte) {
	for p := stk.lo; p < stk.hi; p++ {
		*(*byte)(unsafe.Pointer(p)) = b
	}
}
func findsghi(gp *g, stk stack) uintptr {
	var sghi uintptr
	for sg := gp.waiting; sg != nil; sg = sg.waitlink {
		p := uintptr(sg.elem) + uintptr(sg.c.elemsize)
		if stk.lo <= p && p < stk.hi && p > sghi {
			sghi = p
		}
	}
	return sghi
}
func syncadjustsudogs(gp *g, used uintptr, adjinfo *adjustinfo) uintptr {
	if gp.waiting == nil {
		return 0
	}

	// Lock channels to prevent concurrent send/receive.
	// It's important that we *only* do this for async
	// copystack; otherwise, gp may be in the middle of
	// putting itself on wait queues and this would
	// self-deadlock.
	var lastc *hchan
	for sg := gp.waiting; sg != nil; sg = sg.waitlink {
		if sg.c != lastc {
			lock(&sg.c.lock)
		}
		lastc = sg.c
	}

	// Adjust sudogs.
	adjustsudogs(gp, adjinfo)

	// Copy the part of the stack the sudogs point in to
	// while holding the lock to prevent races on
	// send/receive slots.
	var sgsize uintptr
	if adjinfo.sghi != 0 {
		oldBot := adjinfo.old.hi - used
		newBot := oldBot + adjinfo.delta
		sgsize = adjinfo.sghi - oldBot
		memmove(unsafe.Pointer(newBot), unsafe.Pointer(oldBot), sgsize)
	}

	// Unlock channels.
	lastc = nil
	for sg := gp.waiting; sg != nil; sg = sg.waitlink {
		if sg.c != lastc {
			unlock(&sg.c.lock)
		}
		lastc = sg.c
	}

	return sgsize
}