c++ std::lock与std::scoped_lock底层实现原理源码剖析

@马云

已于 2023-10-21 19:30:09 修改

阅读量1.5k

点赞数 1

分类专栏： # c++并发编程文章标签： c++ 开发语言

于 2023-10-03 16:55:08 首次发布

本文链接：https://blog.csdn.net/weixin_50448879/article/details/133522609

版权

c++并发编程专栏收录该内容

2 篇文章 0 订阅

订阅专栏

本文详细介绍了C++17中的std::lock和std::scoped_lock，特别是后者作为std::lock的RAII封装，展示了如何避免死锁并优化代码执行。讨论了std::lock的实现原理，包括层次结构和尝试获取锁的逻辑。

摘要由CSDN通过智能技术生成

std::lock用于同时获取多把锁(两把及两把以上)

std::scoped_lock定义于c++ 17标准,是std::lock的RAII封装类

以下代码纯来自msvc实现

std::scoped_lock原型

在构造函数中调用std::lock,在析构函数中unlock

class _NODISCARD_LOCK scoped_lock { // class with destructor that unlocks mutexes
public:
    explicit scoped_lock(_Mutexes&... _Mtxes) : _MyMutexes(_Mtxes...) { // construct and lock
        _STD lock(_Mtxes...);
    }

    explicit scoped_lock(adopt_lock_t, _Mutexes&... _Mtxes) noexcept // strengthened
        : _MyMutexes(_Mtxes...) {} // construct but don't lock

    ~scoped_lock() noexcept {
        _STD apply([](_Mutexes&... _Mtxes) { (..., (void) _Mtxes.unlock()); }, _MyMutexes);
    }

    scoped_lock(const scoped_lock&)            = delete;
    scoped_lock& operator=(const scoped_lock&) = delete;

private:
    tuple<_Mutexes&...> _MyMutexes;
};

lock函数原型

这里我画了一个流程图,std::lock调用层次_Lock_nonmember1->_Lock_attempt_small

如果不能将所有的锁全部获取就会将已经获取的锁全部释放后再次尝试获取

在这里插入图片描述

void lock(_Lock0& _Lk0, _Lock1& _Lk1, _LockN&... _LkN) { // lock multiple locks, without deadlock
    _Lock_nonmember1(_Lk0, _Lk1, _LkN...);
}

template <class _Lock0, class _Lock1>
void _Lock_nonmember1(_Lock0& _Lk0, _Lock1& _Lk1) {
    // lock 2 locks, without deadlock, special case for better codegen and reduced metaprogramming for common case
    while (_Lock_attempt_small(_Lk0, _Lk1) && _Lock_attempt_small(_Lk1, _Lk0)) { // keep trying
    }
}

template <class _Lock0, class _Lock1>
bool _Lock_attempt_small(_Lock0& _Lk0, _Lock1& _Lk1) {
    // attempt to lock 2 locks, by first locking _Lk0, and then trying to lock _Lk1 returns whether to try again
    _Lk0.lock();
    {
        _Unlock_one_guard<_Lock0> _Guard{_Lk0};
        if (_Lk1.try_lock()) {
            _Guard._Lk_ptr = nullptr;
            return false;
        }
    }

    _STD this_thread::yield();
    return true;
}