没有用户代码的死锁

Deadlock with no user code

我在使用 std::thread、std::mutex、std::condition_variable 等的 C++ 程序中出现死锁

这本身并没有什么奇怪的,直到我查看进程中每个线程的堆栈:

8532    0   Main Thread Main Thread msvcr120.dll!Concurrency::details::ExternalContextBase::Block   Normal
                        ntdll.dll!_ZwWaitForSingleObject@12()    
                        KernelBase.dll!_WaitForSingleObjectEx@12()   
                        kernel32.dll!_WaitForSingleObjectExImplementation@12()   
                        msvcr120.dll!Concurrency::details::ExternalContextBase::Block() Line 145     
                        ntdll.dll!_ZwQueryVirtualMemory@24()     
                        kernel32.dll!_BasepFillUEFInfo@8()   
                        ntdll.dll!_ZwQueryInformationProcess@20()    
                        msvcr120.dll!_initterm(void (void) * * pfbegin, void (void) * * pfend) Line 954  

-

6484    0   Worker Thread   ntdll.dll!_TppWaiterpThread@4() ntdll.dll!_NtWaitForMultipleObjects@20  Normal
                        ntdll.dll!_NtWaitForMultipleObjects@20()     
                        ntdll.dll!_TppWaiterpThread@4()  
                        kernel32.dll!@BaseThreadInitThunk@12()   
                        ntdll.dll!___RtlUserThreadStart@8()  
                        ntdll.dll!__RtlUserThreadStart@8()   

-

6296    0   Worker Thread   msvcr120.dll!_threadstartex msvcr120.dll!Concurrency::details::ExternalContextBase::Block   Normal
                        ntdll.dll!_ZwWaitForSingleObject@12()    
                        KernelBase.dll!_WaitForSingleObjectEx@12()   
                        kernel32.dll!_WaitForSingleObjectExImplementation@12()   
                        msvcr120.dll!Concurrency::details::ExternalContextBase::Block() Line 145     
                        msvcp120.dll!std::_Thrd_startX(struct _Thrd_imp_t *,unsigned int (*)(void *),void *)     
                        msvcr120.dll!_callthreadstartex() Line 376   
                        msvcr120.dll!_threadstartex(void * ptd) Line 354     
                        kernel32.dll!@BaseThreadInitThunk@12()   
                        ntdll.dll!___RtlUserThreadStart@8()  
                        ntdll.dll!__RtlUserThreadStart@8()   

None 个线程似乎正在执行我的代码,我知道我们已经进入了 main,因为程序在挂起之前已经做了一些事情。

我正在使用以下 class 与我的 std::thread 通信,以防我在那里犯了一些错误:

template <typename T>
class BlockingQueue
{
public:
    BlockingQueue() : _active(true) {}

    bool Get(T& out)
    {
        std::unique_lock<std::mutex> lock(_mutex);

        _cv.wait(lock, [&](){ return !_queue.empty() || !_active; });

        if (_queue.empty())
        {
            assert(!_active);

            return false;
        }

        out = std::move(_queue.front());
        _queue.pop();

        return true;
    }

    void Put(const T& in)
    {
        {
            std::unique_lock<std::mutex> lock(_mutex);

            _queue.push(in);
        }

        _cv.notify_one();
    }

    void Put(T&& in)
    {
        {
            std::unique_lock<std::mutex> lock(_mutex);

            _queue.push(std::move(in));
        }

        _cv.notify_one();
    }

    void Finish()
    {
        {
            std::unique_lock<std::mutex> lock(_mutex);

            _active = false;
        }

        _cv.notify_all();
    }

private:
    bool _active;
    std::mutex _mutex;
    std::condition_variable _cv;
    std::queue<T> _queue;
};

我现在有两个想法:

  1. Main 由于某种原因已经退出。这是一个 PoC,所以当出现错误时,我们会记录到 stdout 并调用 exit()(是的,我知道,这不是最好的,这是从另一个用 C++ 编写的 C 风格程序改编而来的)。我没有看到任何内容被记录到终端,但我想输出可能正在缓冲并且尚未写出?
  2. 调试器在某种程度上对我撒谎。通常它会在执行此操作时将 [frames below may be missing/incorrect] 放入堆栈跟踪中,但也许没有它也会发生。

原来我未能替换队列中的项目,导致我的线程在从队列中检索时死锁,这意味着调试器在骗我。 :(