iOS探索多线程之GCD底层分析

时间 2020-04-20

标签 ios 探索多线程 gcd 底层分析栏目 iOS 繁體版

原文原文链接

欢迎阅读iOS探索系列（按序阅读食用效果更加）c++

iOS探索 alloc流程

iOS探索内存对齐&malloc源码

iOS探索 isa初始化&指向分析

iOS探索类的结构分析

iOS探索 cache_t分析

iOS探索方法的本质和方法查找流程

iOS探索动态方法解析和消息转发机制

iOS探索浅尝辄止dyld加载流程

iOS探索类的加载过程

iOS探索分类、类拓展的加载过程

iOS探索 isa面试题分析

iOS探索 runtime面试题分析

iOS探索 KVC原理及自定义

iOS探索 KVO原理及自定义

iOS探索多线程原理

iOS探索多线程之GCD应用

iOS探索多线程之GCD底层分析

iOS探索多线程面试题分析

写在前面

因为源码的篇幅较大、逻辑分支、宏定义较多，使得源码变得晦涩难懂，让开发者们望而却步。但若是带着疑问、有目的性的去看源码，就能减小难度，忽略无关的代码。首先提出本文分析的几个问题：面试

底层队列是如何建立的
死锁的产生
dispatch_block任务的执行
同步函数
异步函数
信号量的原理
调度组的原理
单例的原理

本文篇幅会比较大，函数之间的跳转也比较多，但只对核心流程代码作了研究，相信看下来应该会有所收获算法

源码的选择判断

分析源码首先得获取到GCD源码，以前已经分析过objc、malloc、dyld源码，那么GCD内容是在哪份源码中呢？数组

这里分享一个小技巧，因为已知要研究GCD，因此有如下几种选择源码的方法sass

Baidu/Google
下符号断点dispatch_queue_create
仅使用Debug->Debug Workflow->Always show Disassembly查看汇编也能看到

这样子就找到了咱们须要的libdispatch源码bash

1、底层队列是如何建立的

上层使用dispatch_queue_create，全局进行搜索。可是会出现搜索结果众多的状况（66 results in 17 files），这时候就考验一个开发者阅读源码的经验了多线程

新手会一个个找过去，宁肯错杀一千不可放过一个
老司机则会根据上层使用修改搜索条件
- 因为建立队列代码为dispatch_queue_create("", NULL)，因此搜索dispatch_queue_create(——将筛选结果降至（21 results in 6 files）
- 因为第一个参数为字符串，在c语言中用const修饰，因此搜索dispatch_queue_create(const——将筛选结果降至（2 results in 2 files）

1.dispatch_queue_create

常规中间层封装——便于代码迭代不改变上层使用并发

dispatch_queue_t
dispatch_queue_create(const char *label, dispatch_queue_attr_t attr)
{
	return _dispatch_lane_create_with_target(label, attr,
			DISPATCH_TARGET_QUEUE_DEFAULT, true);
}
复制代码

有时候也须要注意下源码中函数中的传参：app

此时label是上层的逆序全程域名，主要用在崩溃调试
attr是NULL/DISPATCH_QUEUE_SERIAL、DISPATCH_QUEUE_CONCURRENT，用于区分队列是异步仍是同步的

#define DISPATCH_QUEUE_SERIAL NULL 串行队列的宏定义实际上是个NULLless

2._dispatch_lane_create_with_target

DISPATCH_NOINLINE
static dispatch_queue_t
_dispatch_lane_create_with_target(const char *label, dispatch_queue_attr_t dqa,
		dispatch_queue_t tq, bool legacy)
{
	
	dispatch_queue_attr_info_t dqai = _dispatch_queue_attr_to_info(dqa);
	...
}
复制代码

dqa是这个函数中的第二个参数，即dispatch_queue_create中的attr

用到了串行/并发的区分符，咱们就跟进去瞧瞧

3._dispatch_queue_attr_to_info

dispatch_queue_attr_info_t
_dispatch_queue_attr_to_info(dispatch_queue_attr_t dqa)
{
	dispatch_queue_attr_info_t dqai = { };

	if (!dqa) return dqai;

#if DISPATCH_VARIANT_STATIC
	if (dqa == &_dispatch_queue_attr_concurrent) {
		dqai.dqai_concurrent = true;
		return dqai;
	}
#endif

	if (dqa < _dispatch_queue_attrs ||
			dqa >= &_dispatch_queue_attrs[DISPATCH_QUEUE_ATTR_COUNT]) {
		DISPATCH_CLIENT_CRASH(dqa->do_vtable, "Invalid queue attribute");
	}

	size_t idx = (size_t)(dqa - _dispatch_queue_attrs);
	
	dqai.dqai_inactive = (idx % DISPATCH_QUEUE_ATTR_INACTIVE_COUNT);
	idx /= DISPATCH_QUEUE_ATTR_INACTIVE_COUNT;

	dqai.dqai_concurrent = !(idx % DISPATCH_QUEUE_ATTR_CONCURRENCY_COUNT);
	idx /= DISPATCH_QUEUE_ATTR_CONCURRENCY_COUNT;

	dqai.dqai_relpri = -(idx % DISPATCH_QUEUE_ATTR_PRIO_COUNT);
	idx /= DISPATCH_QUEUE_ATTR_PRIO_COUNT;

	dqai.dqai_qos = idx % DISPATCH_QUEUE_ATTR_QOS_COUNT;
	idx /= DISPATCH_QUEUE_ATTR_QOS_COUNT;

	dqai.dqai_autorelease_frequency =
			idx % DISPATCH_QUEUE_ATTR_AUTORELEASE_FREQUENCY_COUNT;
	idx /= DISPATCH_QUEUE_ATTR_AUTORELEASE_FREQUENCY_COUNT;

	dqai.dqai_overcommit = idx % DISPATCH_QUEUE_ATTR_OVERCOMMIT_COUNT;
	idx /= DISPATCH_QUEUE_ATTR_OVERCOMMIT_COUNT;

	return dqai;
}
复制代码

dispatch_queue_attr_info_t dqai = { };进行初始化
- dispatch_queue_attr_info_t与isa同样，是个位域结构

typedef struct dispatch_queue_attr_info_s {
	dispatch_qos_t dqai_qos : 8;
	int      dqai_relpri : 8;
	uint16_t dqai_overcommit:2;
	uint16_t dqai_autorelease_frequency:2;
	uint16_t dqai_concurrent:1;
	uint16_t dqai_inactive:1;
} dispatch_queue_attr_info_t;
复制代码

接下来把目光放到这句代码if (!dqa) return dqai;
- 串行队列的dqa为NULL，直接返回NULL
- 异步队列往下继续走
size_t idx = (size_t)(dqa - _dispatch_queue_attrs);
- 使用DISPATCH_QUEUE_CONCURRENT的宏定义来进行位运算
- 并发队列的并发数dqai.dqai_concurrent与串行队列不一样

#define DISPATCH_QUEUE_CONCURRENT \ DISPATCH_GLOBAL_OBJECT(dispatch_queue_attr_t, \ _dispatch_queue_attr_concurrent)
复制代码

4.回到_dispatch_lane_create_with_target

咱们要研究的是队列的建立，因此能够忽略源码中的细节，专一查找alloc之类的代码

DISPATCH_NOINLINE
static dispatch_queue_t
_dispatch_lane_create_with_target(const char *label, dispatch_queue_attr_t dqa,
		dispatch_queue_t tq, bool legacy)
{
	
	dispatch_queue_attr_info_t dqai = _dispatch_queue_attr_to_info(dqa);
	
	dispatch_qos_t qos = dqai.dqai_qos;
	...
	
	// 这是部分的逻辑分支
	if (tq->dq_priority & DISPATCH_PRIORITY_FLAG_OVERCOMMIT) {
		overcommit = _dispatch_queue_attr_overcommit_enabled;
	} else {
		overcommit = _dispatch_queue_attr_overcommit_disabled;
	}
	
	if (!tq) {
		tq = _dispatch_get_root_queue(
				qos == DISPATCH_QOS_UNSPECIFIED ? DISPATCH_QOS_DEFAULT : qos, // 4
				overcommit == _dispatch_queue_attr_overcommit_enabled)->_as_dq; // 0 1
		if (unlikely(!tq)) {
			DISPATCH_CLIENT_CRASH(qos, "Invalid queue attribute");
		}
	}
	
	...
    
        // 开辟内存 - 生成响应的对象 queue
	dispatch_lane_t dq = _dispatch_object_alloc(vtable,
			sizeof(struct dispatch_lane_s));
	
	// 构造方法
	_dispatch_queue_init(dq, dqf, dqai.dqai_concurrent ?
			DISPATCH_QUEUE_WIDTH_MAX : 1, DISPATCH_QUEUE_ROLE_INNER |
			(dqai.dqai_inactive ? DISPATCH_QUEUE_INACTIVE : 0));
	// 标签
	dq->dq_label = label;
	// 优先级
	dq->dq_priority = _dispatch_priority_make((dispatch_qos_t)dqai.dqai_qos,
			dqai.dqai_relpri);
	if (overcommit == _dispatch_queue_attr_overcommit_enabled) {
		dq->dq_priority |= DISPATCH_PRIORITY_FLAG_OVERCOMMIT;
	}
	if (!dqai.dqai_inactive) {
		_dispatch_queue_priority_inherit_from_target(dq, tq);
		_dispatch_lane_inherit_wlh_from_target(dq, tq);
	}
	_dispatch_retain(tq);
	dq->do_targetq = tq;
	_dispatch_object_debug(dq, "%s", __func__);
	return _dispatch_trace_queue_create(dq)._dq;
}
复制代码

4.1 _dispatch_get_root_queue建立

tq是当前函数_dispatch_lane_create_with_target中的参数，该函数被调用时传了DISPATCH_TARGET_QUEUE_DEFAULT，因此if (!tq)必定为真

#define DISPATCH_TARGET_QUEUE_DEFAULT NULL
复制代码

_dispatch_get_root_queue两个传参分别为4和0/1
- qos == DISPATCH_QOS_UNSPECIFIED ? DISPATCH_QOS_DEFAULT : qos因为没有对qos进行过赋值，因此默认为0
- overcommit == _dispatch_queue_attr_overcommit_enabled)->_as_dq;在代码区有备注——“这是部分的逻辑分支”，串行队列为_dispatch_queue_attr_overcommit_disabled，并发队列是_dispatch_queue_attr_overcommit_enabled

#define DISPATCH_QOS_UNSPECIFIED ((dispatch_qos_t)0)
#define DISPATCH_QOS_DEFAULT ((dispatch_qos_t)4)

DISPATCH_ALWAYS_INLINE DISPATCH_CONST
static inline dispatch_queue_global_t
_dispatch_get_root_queue(dispatch_qos_t qos, bool overcommit)
{
	if (unlikely(qos < DISPATCH_QOS_MIN || qos > DISPATCH_QOS_MAX)) {
		DISPATCH_CLIENT_CRASH(qos, "Corrupted priority");
	}
	return &_dispatch_root_queues[2 * (qos - 1) + overcommit];
}
复制代码

串行队列、并发队列分别renturn &_dispatch_root_queues[6]、&_dispatch_root_queues[7]

全局搜索_dispatch_root_queues[](由于是个数组)能够看到数组中第7个和第8个分别是串行队列和并发队列

猜测：自定义队列是从_dispatch_root_queues模板中拿出来建立的

4.2 _dispatch_object_alloc开辟内存

其实GCD对象多为dispatch_object_t建立而来：苹果源码注释中有提到——它是全部分派对象的抽象基类型；dispatch_object_t做为一个联合体，只要一建立，就可使用你想要的类型（联合体的“有我没他”属性）

联合体的思想属于多态，跟objc_object的继承思想略有不一样

/*! * @typedef dispatch_object_t * * @abstract * Abstract base type for all dispatch objects. * The details of the type definition are language-specific. * * @discussion * Dispatch objects are reference counted via calls to dispatch_retain() and * dispatch_release(). */
typedef union {
	struct _os_object_s *_os_obj;
	struct dispatch_object_s *_do;
	struct dispatch_queue_s *_dq;
	struct dispatch_queue_attr_s *_dqa;
	struct dispatch_group_s *_dg;
	struct dispatch_source_s *_ds;
	struct dispatch_mach_s *_dm;
	struct dispatch_mach_msg_s *_dmsg;
	struct dispatch_semaphore_s *_dsema;
	struct dispatch_data_s *_ddata;
	struct dispatch_io_s *_dchannel;
} dispatch_object_t DISPATCH_TRANSPARENT_UNION;
复制代码

4.3 _dispatch_queue_init进行构造

前文中提到了并发队列的dqai.dqai_concurrent进行了设置，因此用dqai.dqai_concurrent进行区分并发队列和串行队列
串行队列的width为1，并发队列的width为DISPATCH_QUEUE_WIDTH_MAX
咔咔一顿操做以后返回dispatch_queue_class_t，即对传参dqu进行了赋值修改等操做

4.4 返回dispatch_queue_t

_dispatch_retain(tq);
dq->do_targetq = tq;
_dispatch_object_debug(dq, "%s", __func__);
return _dispatch_trace_queue_create(dq)._dq;
复制代码

dq是dispatch_lane_t类型
tq是dispatch_queue_t类型
_dispatch_trace_queue_create(dq)返回类型为dispatch_queue_class_t，它是个联合体
- dispatch_queue_class_t中的dq就是最终返回的dispatch_queue_t

typedef struct dispatch_queue_s *dispatch_queue_t;
typedef union {
	struct dispatch_queue_s *_dq;
	struct dispatch_workloop_s *_dwl;
	struct dispatch_lane_s *_dl;
	struct dispatch_queue_static_s *_dsq;
	struct dispatch_queue_global_s *_dgq;
	struct dispatch_queue_pthread_root_s *_dpq;
	struct dispatch_source_s *_ds;
	struct dispatch_mach_s *_dm;
	dispatch_lane_class_t _dlu;
#ifdef __OBJC__
	id<OS_dispatch_queue> _objc_dq;
#endif
} dispatch_queue_class_t DISPATCH_TRANSPARENT_UNION;
复制代码

5.验证猜测

NSLog会调用对象的describtion方法，而LLDB能够打印底层的指针

能够看到串行队列、并发队列的target与模板中对应的如出一辙
一样的，串行队列、并发队列、主队列、全局队列的width各不相同（width表示队列中能调度的最大任务数）
- 串行队列和主队列的width没有疑问
- 并发队列的width为DISPATCH_QUEUE_WIDTH_MAX是满值-2
- 全局队列的width为DISPATCH_QUEUE_WIDTH_POOL是满值-1

#define DISPATCH_QUEUE_WIDTH_FULL_BIT 0x0020000000000000ull
#define DISPATCH_QUEUE_WIDTH_FULL 0x1000ull
#define DISPATCH_QUEUE_WIDTH_POOL (DISPATCH_QUEUE_WIDTH_FULL - 1)
#define DISPATCH_QUEUE_WIDTH_MAX (DISPATCH_QUEUE_WIDTH_FULL - 2)

struct dispatch_queue_static_s _dispatch_main_q = {
	DISPATCH_GLOBAL_OBJECT_HEADER(queue_main),
#if !DISPATCH_USE_RESOLVERS
	.do_targetq = _dispatch_get_default_queue(true),
#endif
	.dq_state = DISPATCH_QUEUE_STATE_INIT_VALUE(1) |
			DISPATCH_QUEUE_ROLE_BASE_ANON,
	.dq_label = "com.apple.main-thread",
	.dq_atomic_flags = DQF_THREAD_BOUND | DQF_WIDTH(1),
	.dq_serialnum = 1,
};

struct dispatch_queue_global_s _dispatch_mgr_root_queue = {
	DISPATCH_GLOBAL_OBJECT_HEADER(queue_global),
	.dq_state = DISPATCH_ROOT_QUEUE_STATE_INIT_VALUE,
	.do_ctxt = &_dispatch_mgr_root_queue_pthread_context,
	.dq_label = "com.apple.root.libdispatch-manager",
	.dq_atomic_flags = DQF_WIDTH(DISPATCH_QUEUE_WIDTH_POOL),
	.dq_priority = DISPATCH_PRIORITY_FLAG_MANAGER |
			DISPATCH_PRIORITY_SATURATED_OVERRIDE,
	.dq_serialnum = 3,
	.dgq_thread_pool_size = 1,
};
复制代码

解决了自定义队列的建立流程，那么问题又来了，_dispatch_root_queues的建立又是怎么建立的呢？

6._dispatch_root_queues的建立

除了dispatch_get_main_queue，其余队列都是经过_dispatch_root_queues建立的

libdispatch_init以后调用_dispatch_introspection_init，经过 for 循环，调用_dispatch_trace_queue_create，再取出_dispatch_root_queues里的地址指针一个个建立出来的

7.一图看懂自定义队列建立流程

2、死锁的产生

死锁的产生是因为任务的相互等待，那么底层又是怎么实现死锁的呢？

1.dispatch_sync

全局搜索dispatch_sync(dispatch，忽略unlikely小几率事件

DISPATCH_NOINLINE void dispatch_sync(dispatch_queue_t dq, dispatch_block_t work) {
	uintptr_t dc_flags = DC_FLAG_BLOCK;
	if (unlikely(_dispatch_block_has_private_data(work))) {
		return _dispatch_sync_block_with_privdata(dq, work, dc_flags);
	}
	_dispatch_sync_f(dq, work, _dispatch_Block_invoke(work), dc_flags);
}
#endif // __BLOCKS__
复制代码

2._dispatch_sync_f

仍是常规的中间层封装

DISPATCH_NOINLINE
static void
_dispatch_sync_f(dispatch_queue_t dq, void *ctxt, dispatch_function_t func,
		uintptr_t dc_flags)
{
	_dispatch_sync_f_inline(dq, ctxt, func, dc_flags);
}
复制代码

3._dispatch_sync_f_inline

已知串行队列的width是1，因此串行队列知足dq->dq_width == 1
- return _dispatch_barrier_sync_f(dq, ctxt, func, dc_flags);
并发队列会继续往下走

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_sync_f_inline(dispatch_queue_t dq, void *ctxt,
		dispatch_function_t func, uintptr_t dc_flags)
{
	if (likely(dq->dq_width == 1)) {
		return _dispatch_barrier_sync_f(dq, ctxt, func, dc_flags);
	}

	if (unlikely(dx_metatype(dq) != _DISPATCH_LANE_TYPE)) {
		DISPATCH_CLIENT_CRASH(0, "Queue type doesn't support dispatch_sync");
	}

	dispatch_lane_t dl = upcast(dq)._dl;
	// Global concurrent queues and queues bound to non-dispatch threads
	// always fall into the slow case, see DISPATCH_ROOT_QUEUE_STATE_INIT_VALUE
	if (unlikely(!_dispatch_queue_try_reserve_sync_width(dl))) {
		return _dispatch_sync_f_slow(dl, ctxt, func, 0, dl, dc_flags);
	}

	if (unlikely(dq->do_targetq->do_targetq)) {
		return _dispatch_sync_recurse(dl, ctxt, func, dc_flags);
	}
	_dispatch_introspection_sync_begin(dl);
	_dispatch_sync_invoke_and_complete(dl, ctxt, func DISPATCH_TRACE_ARG(
			_dispatch_trace_item_sync_push_pop(dq, ctxt, func, dc_flags)));
}
复制代码

4._dispatch_barrier_sync_f

串行队列和栅栏函数的比较类似，因此跳转到这里，仍是中间层封装

DISPATCH_NOINLINE
static void
_dispatch_barrier_sync_f(dispatch_queue_t dq, void *ctxt,
		dispatch_function_t func, uintptr_t dc_flags)
{
	_dispatch_barrier_sync_f_inline(dq, ctxt, func, dc_flags);
}
复制代码

5._dispatch_barrier_sync_f_inline

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_barrier_sync_f_inline(dispatch_queue_t dq, void *ctxt,
		dispatch_function_t func, uintptr_t dc_flags)
{
	dispatch_tid tid = _dispatch_tid_self();

	if (unlikely(dx_metatype(dq) != _DISPATCH_LANE_TYPE)) {
		DISPATCH_CLIENT_CRASH(0, "Queue type doesn't support dispatch_sync");
	}

	dispatch_lane_t dl = upcast(dq)._dl;
	// The more correct thing to do would be to merge the qos of the thread
	// that just acquired the barrier lock into the queue state.
	//
	// However this is too expensive for the fast path, so skip doing it.
	// The chosen tradeoff is that if an enqueue on a lower priority thread
	// contends with this fast path, this thread may receive a useless override.
	//
	// Global concurrent queues and queues bound to non-dispatch threads
	// always fall into the slow case, see DISPATCH_ROOT_QUEUE_STATE_INIT_VALUE
	
	// 死锁
	if (unlikely(!_dispatch_queue_try_acquire_barrier_sync(dl, tid))) {
		return _dispatch_sync_f_slow(dl, ctxt, func, DC_FLAG_BARRIER, dl,
				DC_FLAG_BARRIER | dc_flags);
	}

	if (unlikely(dl->do_targetq->do_targetq)) {
		return _dispatch_sync_recurse(dl, ctxt, func,
				DC_FLAG_BARRIER | dc_flags);
	}
	_dispatch_introspection_sync_begin(dl);
	_dispatch_lane_barrier_sync_invoke_and_complete(dl, ctxt, func
			DISPATCH_TRACE_ARG(_dispatch_trace_item_sync_push_pop(
					dq, ctxt, func, dc_flags | DC_FLAG_BARRIER)));
}
复制代码

5.1 _dispatch_tid_self

_dispatch_tid_self是个宏定义，最终调用_dispatch_thread_getspecific来获取当前线程id（线程是以key-value的形式存在的）

#define _dispatch_tid_self() ((dispatch_tid)_dispatch_thread_port())

#if TARGET_OS_WIN32
#define _dispatch_thread_port() ((mach_port_t)0)
#elif !DISPATCH_USE_THREAD_LOCAL_STORAGE
#if DISPATCH_USE_DIRECT_TSD
#define _dispatch_thread_port() ((mach_port_t)(uintptr_t)\
		_dispatch_thread_getspecific(_PTHREAD_TSD_SLOT_MACH_THREAD_SELF))
#else
#define _dispatch_thread_port() pthread_mach_thread_np(_dispatch_thread_self())
#endif
#endif
复制代码

是时候表演真正的技术了！接下来就是死锁的核心分析！

5.2 _dispatch_queue_try_acquire_barrier_sync

_dispatch_queue_try_acquire_barrier_sync会从os底层获取到一个状态

DISPATCH_ALWAYS_INLINE DISPATCH_WARN_RESULT
static inline bool
_dispatch_queue_try_acquire_barrier_sync(dispatch_queue_class_t dq, uint32_t tid)
{
	return _dispatch_queue_try_acquire_barrier_sync_and_suspend(dq._dl, tid, 0);
}

DISPATCH_ALWAYS_INLINE DISPATCH_WARN_RESULT
static inline bool
_dispatch_queue_try_acquire_barrier_sync_and_suspend(dispatch_lane_t dq,
		uint32_t tid, uint64_t suspend_count)
{
	uint64_t init  = DISPATCH_QUEUE_STATE_INIT_VALUE(dq->dq_width);
	uint64_t value = DISPATCH_QUEUE_WIDTH_FULL_BIT | DISPATCH_QUEUE_IN_BARRIER |
			_dispatch_lock_value_from_tid(tid) |
			(suspend_count * DISPATCH_QUEUE_SUSPEND_INTERVAL);
	uint64_t old_state, new_state;
	// 从底层获取信息 -- 状态信息 - 当前队列 - 线程
	return os_atomic_rmw_loop2o(dq, dq_state, old_state, new_state, acquire, {
		uint64_t role = old_state & DISPATCH_QUEUE_ROLE_MASK;
		if (old_state != (init | role)) {
			os_atomic_rmw_loop_give_up(break);
		}
		new_state = value | role;
	});
}
复制代码

5.3 _dispatch_sync_f_slow

在5.2获取到new_state就会来到这里（死锁崩溃时的调用栈中就有这个函数）

DISPATCH_NOINLINE
static void
_dispatch_sync_f_slow(dispatch_queue_class_t top_dqu, void *ctxt,
		dispatch_function_t func, uintptr_t top_dc_flags,
		dispatch_queue_class_t dqu, uintptr_t dc_flags)
{
	dispatch_queue_t top_dq = top_dqu._dq;
	dispatch_queue_t dq = dqu._dq;
	if (unlikely(!dq->do_targetq)) {
		return _dispatch_sync_function_invoke(dq, ctxt, func);
	}

	pthread_priority_t pp = _dispatch_get_priority();
	struct dispatch_sync_context_s dsc = {
		.dc_flags    = DC_FLAG_SYNC_WAITER | dc_flags,
		.dc_func     = _dispatch_async_and_wait_invoke,
		.dc_ctxt     = &dsc,
		.dc_other    = top_dq,
		.dc_priority = pp | _PTHREAD_PRIORITY_ENFORCE_FLAG,
		.dc_voucher  = _voucher_get(),
		.dsc_func    = func,
		.dsc_ctxt    = ctxt,
		.dsc_waiter  = _dispatch_tid_self(),
	};
	

	_dispatch_trace_item_push(top_dq, &dsc);
	__DISPATCH_WAIT_FOR_QUEUE__(&dsc, dq);

	if (dsc.dsc_func == NULL) {
		dispatch_queue_t stop_dq = dsc.dc_other;
		return _dispatch_sync_complete_recurse(top_dq, stop_dq, top_dc_flags);
	}

	_dispatch_introspection_sync_begin(top_dq);
	_dispatch_trace_item_pop(top_dq, &dsc);
	_dispatch_sync_invoke_and_complete_recurse(top_dq, ctxt, func,top_dc_flags
			DISPATCH_TRACE_ARG(&dsc));
}
复制代码

_dispatch_trace_item_push压栈操做，将执行任务push到队列中，按照FIFO执行
__DISPATCH_WAIT_FOR_QUEUE__是崩溃栈的最后一个函数

DISPATCH_NOINLINE
static void
__DISPATCH_WAIT_FOR_QUEUE__(dispatch_sync_context_t dsc, dispatch_queue_t dq)
{
	// 获取队列的状态，看是不是处于等待状态
    uint64_t dq_state = _dispatch_wait_prepare(dq);
    if (unlikely(_dq_state_drain_locked_by(dq_state, dsc->dsc_waiter))) {
    	DISPATCH_CLIENT_CRASH((uintptr_t)dq_state,
    			"dispatch_sync called on queue "
    			"already owned by current thread");
    }
	...
}
复制代码

5.4 _dq_state_drain_locked_by

比较当前等待的value和线程tid，若是为YES就返回回去进行报错处理

DISPATCH_ALWAYS_INLINE
static inline bool
_dq_state_drain_locked_by(uint64_t dq_state, dispatch_tid tid)
{
	return _dispatch_lock_is_locked_by((dispatch_lock)dq_state, tid);
}

DISPATCH_ALWAYS_INLINE
static inline bool
_dispatch_lock_is_locked_by(dispatch_lock lock_value, dispatch_tid tid)
{
	// equivalent to _dispatch_lock_owner(lock_value) == tid
	// ^ (异或运算法) 两个相同就会出现 0 不然为1
	return ((lock_value ^ tid) & DLOCK_OWNER_MASK) == 0;
}
复制代码

6.一图看懂死锁流程

死锁的产生是线程tid和当前等待状态转换后的value做比较
同步执行dispatch_sync会进行压栈操做，按照FIFO去执行
栅栏函数和同步执行是差很少的

3、dispatch_block任务的执行

在dispatch_block处打下断点，LLDB调试打印出函数调用栈

1._dispatch_lane_barrier_sync_invoke_and_complete

这里也采用了相似于上文中的os底层回调，至于为何用回调——任务的执行依赖于线程的状态，若是线程状态不够良好的话任务不会执行

DISPATCH_NOINLINE
static void
_dispatch_lane_barrier_sync_invoke_and_complete(dispatch_lane_t dq,
		void *ctxt, dispatch_function_t func DISPATCH_TRACE_ARG(void *dc))
{
        ...
	// similar to _dispatch_queue_drain_try_unlock
	os_atomic_rmw_loop2o(dq, dq_state, old_state, new_state, release, {
		new_state  = old_state - DISPATCH_QUEUE_SERIAL_DRAIN_OWNED;
		new_state &= ~DISPATCH_QUEUE_DRAIN_UNLOCK_MASK;
		new_state &= ~DISPATCH_QUEUE_MAX_QOS_MASK;
		if (unlikely(old_state & fail_unlock_mask)) {
			os_atomic_rmw_loop_give_up({
				return _dispatch_lane_barrier_complete(dq, 0, 0);
			});
		}
	});
	if (_dq_state_is_base_wlh(old_state)) {
		_dispatch_event_loop_assert_not_owned((dispatch_wlh_t)dq);
	}
}
复制代码

_dispatch_lane_barrier_complete

直接跟到_dispatch_lane_class_barrier_complete

DISPATCH_NOINLINE
static void
_dispatch_lane_barrier_complete(dispatch_lane_class_t dqu, dispatch_qos_t qos,
		dispatch_wakeup_flags_t flags)
{
        ...
	uint64_t owned = DISPATCH_QUEUE_IN_BARRIER +
			dq->dq_width * DISPATCH_QUEUE_WIDTH_INTERVAL;
	return _dispatch_lane_class_barrier_complete(dq, qos, flags, target, owned);
}
复制代码

_dispatch_lane_class_barrier_complete

跟进_dispatch_queue_push_queue

DISPATCH_NOINLINE
static void
_dispatch_lane_class_barrier_complete(dispatch_lane_t dq, dispatch_qos_t qos,
		dispatch_wakeup_flags_t flags, dispatch_queue_wakeup_target_t target,
		uint64_t owned)
{
    ...
    if (tq) {
		if (likely((old_state ^ new_state) & enqueue)) {
			dispatch_assert(_dq_state_is_enqueued(new_state));
			dispatch_assert(flags & DISPATCH_WAKEUP_CONSUME_2);
			return _dispatch_queue_push_queue(tq, dq, new_state);
		}
		...
	}
}
复制代码

_dispatch_queue_push_queue

而其中的dx_push是个宏定义

#define dx_push(x, y, z) dx_vtable(x)->dq_push(x, y, z)

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_queue_push_queue(dispatch_queue_t tq, dispatch_queue_class_t dq,
		uint64_t dq_state)
{
	_dispatch_trace_item_push(tq, dq);
	return dx_push(tq, dq, _dq_state_max_qos(dq_state));
}
复制代码

dq_push 全局搜索来到dq_push，选择_dispatch_root_queue_push继续走下去
_dispatch_root_queue_push

大几率会走_dispatch_root_queue_push_inline

DISPATCH_NOINLINE
void
_dispatch_root_queue_push(dispatch_queue_global_t rq, dispatch_object_t dou,
		dispatch_qos_t qos)
{   
        ...
	#if HAVE_PTHREAD_WORKQUEUE_QOS
	if (_dispatch_root_queue_push_needs_override(rq, qos)) {
		return _dispatch_root_queue_push_override(rq, dou, qos);
	}
#else
	(void)qos;
#endif
	_dispatch_root_queue_push_inline(rq, dou, dou, 1);
}
复制代码

_dispatch_root_queue_push_inline

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_root_queue_push_inline(dispatch_queue_global_t dq,
		dispatch_object_t _head, dispatch_object_t _tail, int n)
{
	struct dispatch_object_s *hd = _head._do, *tl = _tail._do;
	if (unlikely(os_mpsc_push_list(os_mpsc(dq, dq_items), hd, tl, do_next))) {
		return _dispatch_root_queue_poke(dq, n, 0);
	}
}
复制代码

_dispatch_root_queue_poke

DISPATCH_NOINLINE
void
_dispatch_root_queue_poke(dispatch_queue_global_t dq, int n, int floor)
{
	...
	return _dispatch_root_queue_poke_slow(dq, n, floor);
}
复制代码

_dispatch_root_queue_poke_slow

DISPATCH_NOINLINE
static void
_dispatch_root_queue_poke_slow(dispatch_queue_global_t dq, int n, int floor)
{
	int remaining = n;
	int r = ENOSYS;

	_dispatch_root_queues_init();
	_dispatch_debug_root_queue(dq, __func__);
	_dispatch_trace_runtime_event(worker_request, dq, (uint64_t)n);
	...
}
复制代码

_dispatch_root_queues_init

跟到核心方法dispatch_once_f

static inline void
_dispatch_root_queues_init(void)
{
	dispatch_once_f(&_dispatch_root_queues_pred, NULL,
			_dispatch_root_queues_init_once);
}
复制代码

dispatch_once_f 当你看到_dispatch_once_callout函数就离成功不远了

DISPATCH_NOINLINE void dispatch_once_f(dispatch_once_t *val, void *ctxt, dispatch_function_t func) {
	// 若是你来过一次 -- 下次就不来
	dispatch_once_gate_t l = (dispatch_once_gate_t)val;
	//DLOCK_ONCE_DONE
#if !DISPATCH_ONCE_INLINE_FASTPATH || DISPATCH_ONCE_USE_QUIESCENT_COUNTER
	uintptr_t v = os_atomic_load(&l->dgo_once, acquire);
	if (likely(v == DLOCK_ONCE_DONE)) {
		return;
	}
#if DISPATCH_ONCE_USE_QUIESCENT_COUNTER
	if (likely(DISPATCH_ONCE_IS_GEN(v))) {
		return _dispatch_once_mark_done_if_quiesced(l, v);
	}
#endif
#endif

	// 知足条件 -- 试图进去
	if (_dispatch_once_gate_tryenter(l)) {
		// 单利调用 -- v->DLOCK_ONCE_DONE
		return _dispatch_once_callout(l, ctxt, func);
	}
	return _dispatch_once_wait(l);
}
复制代码

_dispatch_once_callout

DISPATCH_NOINLINE
static void
_dispatch_once_callout(dispatch_once_gate_t l, void *ctxt,
		dispatch_function_t func)
{
	_dispatch_client_callout(ctxt, func);
	_dispatch_once_gate_broadcast(l);
}
复制代码

2._dispatch_client_callout

f(ctxt)调用执行dispatch_function_t——dispatch_block的执行点

DISPATCH_NOINLINE
void
_dispatch_client_callout(void *ctxt, dispatch_function_t f)
{
	_dispatch_get_tsd_base();
	void *u = _dispatch_get_unwind_tsd();
	if (likely(!u)) return f(ctxt);
	_dispatch_set_unwind_tsd(NULL);
	f(ctxt);
	_dispatch_free_unwind_tsd();
	_dispatch_set_unwind_tsd(u);
}
复制代码

3.一图看懂任务保存流程

最后的最后咱们找到了任务执行点，但没有找到任务的保存点，接下来就要从同步函数和异步函数提及了

4、同步函数

前文中已经跟过dispatch_sync的实现了，这里上一张图再捋一捋（特别注意work和func的调用）

串行队列走dq->dq_width == 1分支
- _dispatch_barrier_sync_f -> _dispatch_barrier_sync_f_inline -> _dispatch_lane_barrier_sync_invoke_and_complete
- 而后就是3、dispatch_block任务的执行中的流程
其余状况大几率走_dispatch_sync_invoke_and_complete

1._dispatch_sync_invoke_and_complete

保存func并调用_dispatch_sync_function_invoke_inline

DISPATCH_NOINLINE
static void
_dispatch_sync_invoke_and_complete(dispatch_lane_t dq, void *ctxt,
		dispatch_function_t func DISPATCH_TRACE_ARG(void *dc))
{
	_dispatch_sync_function_invoke_inline(dq, ctxt, func);
	_dispatch_trace_item_complete(dc);
	_dispatch_lane_non_barrier_complete(dq, 0);
}
复制代码

2._dispatch_sync_function_invoke_inline

直接调用_dispatch_client_callout与3、dispatch_block任务的执行遥相呼应

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_sync_function_invoke_inline(dispatch_queue_class_t dq, void *ctxt,
		dispatch_function_t func)
{
	dispatch_thread_frame_s dtf;
	_dispatch_thread_frame_push(&dtf, dq);
	// f(ctxt) -- func(ctxt)
	_dispatch_client_callout(ctxt, func);
	_dispatch_perfmon_workitem_inc();
	_dispatch_thread_frame_pop(&dtf);
}
复制代码

3.一图看懂同步函数执行的部分流程

5、异步函数

1. 任务的保存

仍是同样的思路，跟到dispatch_async的源码实现中，关注dispatch_block_t

1.1 dispatch_async

void dispatch_async(dispatch_queue_t dq, dispatch_block_t work) {
	dispatch_continuation_t dc = _dispatch_continuation_alloc();
	uintptr_t dc_flags = DC_FLAG_CONSUME;
	dispatch_qos_t qos;

	qos = _dispatch_continuation_init(dc, dq, work, 0, dc_flags);
	_dispatch_continuation_async(dq, dc, qos, dc->dc_flags);
}
复制代码

1.2 _dispatch_continuation_init

_dispatch_Block_invoke将任务进行统一格式化

DISPATCH_ALWAYS_INLINE
static inline dispatch_qos_t
_dispatch_continuation_init(dispatch_continuation_t dc,
		dispatch_queue_class_t dqu, dispatch_block_t work,
		dispatch_block_flags_t flags, uintptr_t dc_flags)
{
	void *ctxt = _dispatch_Block_copy(work);

	dc_flags |= DC_FLAG_BLOCK | DC_FLAG_ALLOCATED;
	if (unlikely(_dispatch_block_has_private_data(work))) {
		dc->dc_flags = dc_flags;
		dc->dc_ctxt = ctxt;
		// will initialize all fields but requires dc_flags & dc_ctxt to be set
		return _dispatch_continuation_init_slow(dc, dqu, flags);
	}

	dispatch_function_t func = _dispatch_Block_invoke(work);
	if (dc_flags & DC_FLAG_CONSUME) {
		func = _dispatch_call_block_and_release;
	}
	return _dispatch_continuation_init_f(dc, dqu, ctxt, func, flags, dc_flags);
}
复制代码

1.3 _dispatch_continuation_init_f

dc->dc_func = f将block任务保存下来

DISPATCH_ALWAYS_INLINE
static inline dispatch_qos_t
_dispatch_continuation_init_f(dispatch_continuation_t dc,
		dispatch_queue_class_t dqu, void *ctxt, dispatch_function_t f,
		dispatch_block_flags_t flags, uintptr_t dc_flags)
{
	pthread_priority_t pp = 0;
	dc->dc_flags = dc_flags | DC_FLAG_ALLOCATED;
	dc->dc_func = f;
	dc->dc_ctxt = ctxt;
	// in this context DISPATCH_BLOCK_HAS_PRIORITY means that the priority
	// should not be propagated, only taken from the handler if it has one
	if (!(flags & DISPATCH_BLOCK_HAS_PRIORITY)) {
		pp = _dispatch_priority_propagate();
	}
	_dispatch_continuation_voucher_set(dc, flags);
	return _dispatch_continuation_priority_set(dc, dqu, pp, flags);
}
复制代码

异步函数的任务保存找到了，那它的任务又是什么时候执行的呢？以及线程是什么时候建立的？

2. 线程的建立

2.1 _dispatch_continuation_async

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_continuation_async(dispatch_queue_class_t dqu,
		dispatch_continuation_t dc, dispatch_qos_t qos, uintptr_t dc_flags)
{
#if DISPATCH_INTROSPECTION
	if (!(dc_flags & DC_FLAG_NO_INTROSPECTION)) {
		_dispatch_trace_item_push(dqu, dc);
	}
#else
	(void)dc_flags;
#endif
	return dx_push(dqu._dq, dc, qos);
}
复制代码

2.1 dx_push...

以前已经分析过了，至于为何要用_dispatch_root_queue_push研究——由于它最基本，省去了旁枝末节

dx_push->dq_push->_dispatch_root_queue_push->_dispatch_root_queue_push_inline->_dispatch_root_queue_poke->_dispatch_root_queue_poke_slow

2.2 _dispatch_root_queue_poke_slow

static void
_dispatch_root_queue_poke_slow(dispatch_queue_global_t dq, int n, int floor)
{
    ...
    // floor 为 0，remaining 是根据队列任务的状况处理的
    int can_request, t_count;
    // 获取线程池的大小
    t_count = os_atomic_load2o(dq, dgq_thread_pool_size, ordered);
    do {
        // 计算能够请求的数量
        can_request = t_count < floor ? 0 : t_count - floor;
        if (remaining > can_request) {
    	    _dispatch_root_queue_debug("pthread pool reducing request from %d to %d",
    	    		remaining, can_request);
    	    os_atomic_sub2o(dq, dgq_pending, remaining - can_request, relaxed);
    	    remaining = can_request;
        }
        if (remaining == 0) {
            // 线程池满了，就会报出异常的状况
            _dispatch_root_queue_debug("pthread pool is full for root queue: "
    				"%p", dq);
                return;
    	}
    } while (!os_atomic_cmpxchgvw2o(dq, dgq_thread_pool_size, t_count,
    		t_count - remaining, &t_count, acquire));
    
    pthread_attr_t *attr = &pqc->dpq_thread_attr;
    pthread_t tid, *pthr = &tid;
    #if DISPATCH_USE_MGR_THREAD && DISPATCH_USE_PTHREAD_ROOT_QUEUES
    if (unlikely(dq == &_dispatch_mgr_root_queue)) {
    	pthr = _dispatch_mgr_root_queue_init();
    }
    #endif
    do {
    	_dispatch_retain(dq); 
    	// 开辟线程
    	while ((r = pthread_create(pthr, attr, _dispatch_worker_thread, dq))) {
    		if (r != EAGAIN) {
    			(void)dispatch_assume_zero(r);
    		}
    		_dispatch_temporary_resource_shortage();
    	}
    } while (--remaining);
    #else
    (void)floor;
    #endif // DISPATCH_USE_PTHREAD_POOL
}
复制代码

第一个do-while是对核心线程数的判断、操做等等
第二个do-while调用pthread_create建立线程（底层仍是用了pthread）

3.任务的执行

任务的执行其实刚才已经讲过了

_dispatch_root_queues_init->dispatch_once_f->_dispatch_once_callout->_dispatch_client_callout

只不过任务在等待线程的状态，而线程怎么执行任务就不得而知了

4.一图看懂异步函数的执行流程

6、信号量的原理

信号量的基本使用是这样的，底层又是怎么个原理呢？

dispatch_semaphore_t sem = dispatch_semaphore_create(0);
dispatch_semaphore_wait(sem, DISPATCH_TIME_FOREVER);
dispatch_semaphore_signal(sem);
复制代码

1.dispatch_semaphore_create

只是初始化dispatch_semaphore_t，内部进行传值保存（value必须大于0）

dispatch_semaphore_t
dispatch_semaphore_create(long value)
{
	dispatch_semaphore_t dsema;

	// If the internal value is negative, then the absolute of the value is
	// equal to the number of waiting threads. Therefore it is bogus to
	// initialize the semaphore with a negative value.
	if (value < 0) {
		return DISPATCH_BAD_INPUT;
	}

	dsema = _dispatch_object_alloc(DISPATCH_VTABLE(semaphore),
			sizeof(struct dispatch_semaphore_s));
	dsema->do_next = DISPATCH_OBJECT_LISTLESS;
	dsema->do_targetq = _dispatch_get_default_queue(false);
	dsema->dsema_value = value;
	_dispatch_sema4_init(&dsema->dsema_sema, _DSEMA4_POLICY_FIFO);
	dsema->dsema_orig = value;
	return dsema;
}
复制代码

2.dispatch_semaphore_signal

相似KVC形式从底层取得当前信号量的value值，而且这个函数是有返回值的

long dispatch_semaphore_signal(dispatch_semaphore_t dsema) {
	long value = os_atomic_inc2o(dsema, dsema_value, release);
	if (likely(value > 0)) {
		return 0;
	}
	if (unlikely(value == LONG_MIN)) {
		DISPATCH_CLIENT_CRASH(value,
				"Unbalanced call to dispatch_semaphore_signal()");
	}
	return _dispatch_semaphore_signal_slow(dsema);
}

DISPATCH_NOINLINE
long
_dispatch_semaphore_signal_slow(dispatch_semaphore_t dsema)
{
	_dispatch_sema4_create(&dsema->dsema_sema, _DSEMA4_POLICY_FIFO);
	_dispatch_sema4_signal(&dsema->dsema_sema, 1);
	return 1;
}
复制代码

其实最核心的点在于os_atomic_inc2o进行了++操做

#define os_atomic_inc2o(p, f, m) \ os_atomic_add2o(p, f, 1, m)
#define os_atomic_add2o(p, f, v, m) \ os_atomic_add(&(p)->f, (v), m)
#define os_atomic_add(p, v, m) \ _os_atomic_c11_op((p), (v), m, add, +)
复制代码

3.dispatch_semaphore_wait

同理dispatch_semaphore_wait也是取value值，并返回对应结果

value>=0就马上返回
value<0根据等待时间timeout做出不一样操做
- DISPATCH_TIME_NOW将value加一(也就是变为0)——为了抵消 wait 函数一开始的减一操做，并返回KERN_OPERATION_TIMED_OUT表示因为等待时间超时
- DISPATCH_TIME_FOREVER调用系统的semaphore_wait方法继续等待，直到收到signal调用
- 默认状况与DISPATCH_TIME_FOREVER相似，不过须要指定一个等待时间

long dispatch_semaphore_wait(dispatch_semaphore_t dsema, dispatch_time_t timeout) {
	long value = os_atomic_dec2o(dsema, dsema_value, acquire);
	if (likely(value >= 0)) {
		return 0;
	}
	return _dispatch_semaphore_wait_slow(dsema, timeout);
}

DISPATCH_NOINLINE
static long
_dispatch_semaphore_wait_slow(dispatch_semaphore_t dsema,
		dispatch_time_t timeout)
{
	long orig;

	_dispatch_sema4_create(&dsema->dsema_sema, _DSEMA4_POLICY_FIFO);
	switch (timeout) {
	default:
		if (!_dispatch_sema4_timedwait(&dsema->dsema_sema, timeout)) {
			break;
		}
		// Fall through and try to undo what the fast path did to
		// dsema->dsema_value
	case DISPATCH_TIME_NOW:
		orig = dsema->dsema_value;
		while (orig < 0) {
			if (os_atomic_cmpxchgvw2o(dsema, dsema_value, orig, orig + 1,
					&orig, relaxed)) {
				return _DSEMA4_TIMEOUT();
			}
		}
		// Another thread called semaphore_signal().
		// Fall through and drain the wakeup.
	case DISPATCH_TIME_FOREVER:
		_dispatch_sema4_wait(&dsema->dsema_sema);
		break;
	}
	return 0;
}
复制代码

os_atomic_dec2o进行了--操做

#define os_atomic_dec2o(p, f, m) \ os_atomic_sub2o(p, f, 1, m)
#define os_atomic_sub2o(p, f, v, m) \ os_atomic_sub(&(p)->f, (v), m)
#define os_atomic_sub(p, v, m) \ _os_atomic_c11_op((p), (v), m, sub, -)
复制代码

7、调度组的原理

调度组的基本使用以下

dispatch_group_t group = dispatch_group_create();
dispatch_group_enter(group);
dispatch_group_leave(group);
dispatch_group_async(group, queue, ^{});
dispatch_group_notify(group, queue, ^{});
复制代码

1.dispatch_group_create

跟其余GCD对象同样使用_dispatch_object_alloc生成dispatch_group_t

从os_atomic_store2o能够看出group底层也维护了一个value值

dispatch_group_t
dispatch_group_create(void)
{
	return _dispatch_group_create_with_count(0);
}

DISPATCH_ALWAYS_INLINE
static inline dispatch_group_t
_dispatch_group_create_with_count(uint32_t n)
{
	dispatch_group_t dg = _dispatch_object_alloc(DISPATCH_VTABLE(group),
			sizeof(struct dispatch_group_s));
	dg->do_next = DISPATCH_OBJECT_LISTLESS;
	dg->do_targetq = _dispatch_get_default_queue(false);
	if (n) {
		os_atomic_store2o(dg, dg_bits,
				-n * DISPATCH_GROUP_VALUE_INTERVAL, relaxed);
		os_atomic_store2o(dg, do_ref_cnt, 1, relaxed); // <rdar://22318411>
	}
	return dg;
}
复制代码

2.dispatch_group_enter & dispatch_group_leave

这两个API与信号量的使用大同小异，os_atomic_sub_orig2o、os_atomic_add_orig2o负责--、++操做，若是不成对使用则会出错

dispatch_group_leave出组会对state进行更新
所有出组了会调用_dispatch_group_wake

void dispatch_group_enter(dispatch_group_t dg) {
	// The value is decremented on a 32bits wide atomic so that the carry
	// for the 0 -> -1 transition is not propagated to the upper 32bits.
	uint32_t old_bits = os_atomic_sub_orig2o(dg, dg_bits,
			DISPATCH_GROUP_VALUE_INTERVAL, acquire);
	uint32_t old_value = old_bits & DISPATCH_GROUP_VALUE_MASK;
	if (unlikely(old_value == 0)) {
		_dispatch_retain(dg); // <rdar://problem/22318411>
	}
	if (unlikely(old_value == DISPATCH_GROUP_VALUE_MAX)) {
		DISPATCH_CLIENT_CRASH(old_bits,
				"Too many nested calls to dispatch_group_enter()");
	}
}

void dispatch_group_leave(dispatch_group_t dg) {
	// The value is incremented on a 64bits wide atomic so that the carry for
	// the -1 -> 0 transition increments the generation atomically.
	uint64_t new_state, old_state = os_atomic_add_orig2o(dg, dg_state,
			DISPATCH_GROUP_VALUE_INTERVAL, release);
	uint32_t old_value = (uint32_t)(old_state & DISPATCH_GROUP_VALUE_MASK);

	if (unlikely(old_value == DISPATCH_GROUP_VALUE_1)) {
		old_state += DISPATCH_GROUP_VALUE_INTERVAL;
		
		do {
			new_state = old_state;
			if ((old_state & DISPATCH_GROUP_VALUE_MASK) == 0) {
				new_state &= ~DISPATCH_GROUP_HAS_WAITERS;
				new_state &= ~DISPATCH_GROUP_HAS_NOTIFS;
			} else {
				// If the group was entered again since the atomic_add above,
				// we can't clear the waiters bit anymore as we don't know for
				// which generation the waiters are for
				new_state &= ~DISPATCH_GROUP_HAS_NOTIFS;
			}
			if (old_state == new_state) break;
		} while (unlikely(!os_atomic_cmpxchgv2o(dg, dg_state,
				old_state, new_state, &old_state, relaxed)));
		
		return _dispatch_group_wake(dg, old_state, true);
	}

	if (unlikely(old_value == 0)) {
		DISPATCH_CLIENT_CRASH((uintptr_t)old_value,
				"Unbalanced call to dispatch_group_leave()");
	}
}
复制代码

3.dispatch_group_async

_dispatch_continuation_init_f保存任务（相似异步函数）
调用_dispatch_continuation_group_async

void dispatch_group_async_f(dispatch_group_t dg, dispatch_queue_t dq, void *ctxt, dispatch_function_t func) {
	dispatch_continuation_t dc = _dispatch_continuation_alloc();
	uintptr_t dc_flags = DC_FLAG_CONSUME | DC_FLAG_GROUP_ASYNC;
	dispatch_qos_t qos;

	qos = _dispatch_continuation_init_f(dc, dq, ctxt, func, 0, dc_flags);
	_dispatch_continuation_group_async(dg, dq, dc, qos);
}
复制代码

调用dispatch_group_enter进组

static inline void
_dispatch_continuation_group_async(dispatch_group_t dg, dispatch_queue_t dq,
		dispatch_continuation_t dc, dispatch_qos_t qos)
{
	dispatch_group_enter(dg);
	dc->dc_data = dg;
	_dispatch_continuation_async(dq, dc, qos, dc->dc_flags);
}
复制代码

进组了以后须要调用出组，也就是执行完任务会出组

在_dispatch_continuation_invoke_inline若是是group形式就会调用_dispatch_continuation_with_group_invoke来出组

4.dispatch_group_wait

dispatch_group_wait与信号量也是殊途同归

dispatch_group_create建立调度组的时候保存了一个value

若是当前value和原始value相同，代表任务已经所有完成，直接返回0
若是timeout为 0 也会马上返回，不然调用 _dispatch_group_wait_slow
- 在_dispatch_group_wait_slow会一直等到任务完成返回 0
- 一直没有完成会返回timeout

long dispatch_group_wait(dispatch_group_t dg, dispatch_time_t timeout) {
    uint64_t old_state, new_state;
    
    os_atomic_rmw_loop2o(dg, dg_state, old_state, new_state, relaxed, {
    	if ((old_state & DISPATCH_GROUP_VALUE_MASK) == 0) {
    		os_atomic_rmw_loop_give_up_with_fence(acquire, return 0);
    	}
    	if (unlikely(timeout == 0)) {
    		os_atomic_rmw_loop_give_up(return _DSEMA4_TIMEOUT());
    	}
    	new_state = old_state | DISPATCH_GROUP_HAS_WAITERS;
    	if (unlikely(old_state & DISPATCH_GROUP_HAS_WAITERS)) {
    		os_atomic_rmw_loop_give_up(break);
    	}
    });
    
    return _dispatch_group_wait_slow(dg, _dg_state_gen(new_state), timeout);
}

复制代码

5.dispatch_group_notify

等待_dispatch_group_wake回调（所有出组会调用）

DISPATCH_ALWAYS_INLINE
static inline void
_dispatch_group_notify(dispatch_group_t dg, dispatch_queue_t dq,
		dispatch_continuation_t dsn)
{
	uint64_t old_state, new_state;
	dispatch_continuation_t prev;

	dsn->dc_data = dq;
	_dispatch_retain(dq);

	prev = os_mpsc_push_update_tail(os_mpsc(dg, dg_notify), dsn, do_next);
	if (os_mpsc_push_was_empty(prev)) _dispatch_retain(dg);
	os_mpsc_push_update_prev(os_mpsc(dg, dg_notify), prev, dsn, do_next);
	if (os_mpsc_push_was_empty(prev)) {
		os_atomic_rmw_loop2o(dg, dg_state, old_state, new_state, release, {
			new_state = old_state | DISPATCH_GROUP_HAS_NOTIFS;
			if ((uint32_t)old_state == 0) {
				os_atomic_rmw_loop_give_up({
					return _dispatch_group_wake(dg, new_state, false);
				});
			}
		});
	}
}

static void
_dispatch_group_wake(dispatch_group_t dg, uint64_t dg_state, bool needs_release)
{
    uint16_t refs = needs_release ? 1 : 0; // <rdar://problem/22318411>
    
    if (dg_state & DISPATCH_GROUP_HAS_NOTIFS) {
    	dispatch_continuation_t dc, next_dc, tail;
    
    	// Snapshot before anything is notified/woken
    	dc = os_mpsc_capture_snapshot(os_mpsc(dg, dg_notify), &tail);
    	do {
    		dispatch_queue_t dsn_queue = (dispatch_queue_t)dc->dc_data;
    		next_dc = os_mpsc_pop_snapshot_head(dc, tail, do_next);
    		_dispatch_continuation_async(dsn_queue, dc,
    				_dispatch_qos_from_pp(dc->dc_priority), dc->dc_flags);
    		_dispatch_release(dsn_queue);
    	} while ((dc = next_dc));
    
    	refs++;
    }
    if (dg_state & DISPATCH_GROUP_HAS_WAITERS) {
    	_dispatch_wake_by_address(&dg->dg_gen);
    }
    if (refs) _dispatch_release_n(dg, refs);
}
复制代码

7、单例的原理

#define DLOCK_ONCE_UNLOCKED ((uintptr_t)0)
void dispatch_once_f(dispatch_once_t *val, void *ctxt, dispatch_function_t func) {
    dispatch_once_gate_t l = (dispatch_once_gate_t)val;
    #if !DISPATCH_ONCE_INLINE_FASTPATH || DISPATCH_ONCE_USE_QUIESCENT_COUNTER
    uintptr_t v = os_atomic_load(&l->dgo_once, acquire);
    if (likely(v == DLOCK_ONCE_DONE)) {
    	return;
    }
    #if DISPATCH_ONCE_USE_QUIESCENT_COUNTER
    if (likely(DISPATCH_ONCE_IS_GEN(v))) {
    	return _dispatch_once_mark_done_if_quiesced(l, v);
    }
    #endif
    #endif
    if (_dispatch_once_gate_tryenter(l)) {
    	return _dispatch_once_callout(l, ctxt, func);
    }
    return _dispatch_once_wait(l);
}

DISPATCH_ALWAYS_INLINE
static inline bool
_dispatch_once_gate_tryenter(dispatch_once_gate_t l)
{
	return os_atomic_cmpxchg(&l->dgo_once, DLOCK_ONCE_UNLOCKED,
			(uintptr_t)_dispatch_lock_value_for_self(), relaxed);
}
复制代码

第一次调用时外部传进来的onceToken为空，因此val为 NULL
- _dispatch_once_gate_tryenter(l)判断l->dgo_once是否标记为DLOCK_ONCE_UNLOCKED（是否存储过）
- DLOCK_ONCE_UNLOCKED=0，因此if 判断是成立的，就会进行block回调
- 再经过_dispatch_once_gate_broadcast将l->dgo_once标记为DLOCK_ONCE_DONE
第二次进来就会直接返回，保证代码只执行一次

写在后面

关于上一篇文章中的dispatch_barrier_async为何使用全局队列无效能够看深刻浅出 GCD 之 dispatch_queue

GCD源码还真不是通常的晦涩难懂，笔者水平有限，如有错误之处烦请指出

iOS探索 多线程之GCD底层分析