c++ - pthread_cond_wait : random segmentation fault

Question

更新3

最近，我注意到我的代码随机导致Segmentation Fault 错误。但我认为到目前为止我的代码非常简单，我无法弄清楚错误的来源。由于它是随机发生的，我假设存在某种竞争条件。我认为这是所有可能相关的代码，如果您需要更多，请告诉我:

namespace thread {
    pthread_t terminated_thread_id, /* and others */;
    pthread_mutex_t terminate_thread = PTHREAD_MUTEX_INITIALIZER;
    pthread_cond_t terminate_thread_signal = PTHREAD_COND_INITIALIZER;
    int total_thread_count = 0;
    int termination; // + sembufs

    inline void* Exit(void* value) {
    //  This must be unlocked after all join-related jobs are done
        semop(thread::termination, thread::termination_in_process, 2)
            pthread_mutex_lock(&thread::terminate_thread);
                thread::terminated_thread_id = pthread_self();
                pthread_cond_signal(&thread::terminate_thread_signal);
            pthread_mutex_unlock(&thread::terminate_thread);

        pthread_exit(value);
        return value;
    }
}
int main(int argc, const char** argv){
...
    pthread_mutex_lock(&thread::terminate_thread);
    if(0 != pthread_create(&thread::communication_handler_thread_id, NULL,    \
                           CommunicationHandler, NULL)){
        global::PrintDebug("pthread_create() failed", __FILE__, __LINE__);
    }
    /** 2 more pthread_create()-calls */       
    do{
        thread::terminated_thread_id = pthread_self();
        pthread_cond_wait(&thread::terminate_thread_signal,                   \
                          &thread::terminate_thread);
        if(!pthread_equal(thread::terminated_thread_id, pthread_self())){
            pthread_join(thread::terminated_thread_id, NULL);
    ...
            semop(thread::termination, thread::termination_done, 1)
        }
    }while(thread::total_thread_count > 0);

    pthread_mutex_unlock(&thread::terminate_thread);
    return 0;
}

terminate_thread_signal 信号仅在 thread::Exit() 函数中发出。该函数也仅在用于创建线程的函数末尾调用。

这是调试器为调用堆栈显示的内容:

#0 (    0xb7fe2424 in __kernel_vsyscall() (??:??)
#1 0xb7fbdfcf   __pthread_cond_wait(cond=0x80539c0, mutex=0x8053998) (pthread_cond_wait.c:153)
#2 0x804a094    main(argc=1, argv=0xbffff9c4) (/home/papergay/SeekYourCar/0.2/Server/main.cpp:121)

我已经知道的是，如果发生错误，则说明还没有线程调用 thread::Exit()。我还使用了一个未命名的命名空间和一些初始化(如果可能相关的话)。我使用 Code::Blocks 作为 IDE，使用 GCC 作为编译器。

Answer 1

pthread_cond_wait() 允许虚假唤醒，因此您必须在每次唤醒后重新测试条件本身。这可能会导致您的问题 - 如果主线程在设置 thread::terminated_thread_id 之前虚假地唤醒，它会将无效的线程 ID 传递给 pthread_join()。

您的代码中还有另一个问题 - 无法保证在互斥量解锁后，已发出信号的线程将是下一个被唤醒的线程，因此两个线程可能会调用 thread::Exit() 快速连续，主线程直到第二个退出线程解锁互斥锁后才运行。在这种情况下，您永远不会在第一个线程上调用 pthread_join()。

像这样应该可以解决这些问题:

namespace thread {
    int terminate_thread_set = 0;
    pthread_mutex_t terminate_thread = PTHREAD_MUTEX_INITIALIZER;
    pthread_cond_t terminate_thread_set_cond = PTHREAD_COND_INITIALIZER;
    pthread_cond_t terminate_thread_unset_cond = PTHREAD_COND_INITIALIZER;

    /* ... */

    inline void Exit(void* value)
    {
        pthread_mutex_lock(&thread::terminate_thread);
        while (thread::terminate_thread_set)
            pthread_cond_wait(&thread::terminate_thread_unset_cond);
        thread::terminated_thread_id = pthread_self();
        thread::terminate_thread_set = 1;
        pthread_cond_signal(&thread::terminate_thread_set_cond);
        pthread_mutex_unlock(&thread::terminate_thread);

        pthread_exit(value);
    }
}

在 main 中:

pthread_mutex_lock(&thread::terminate_thread);

/* ... */

while(thread::total_thread_count > 0) {
    while (!thread::terminate_thread_set)
        pthread_cond_wait(&thread::terminate_thread_set_cond, &thread::terminate_thread);
    thread::terminate_thread_set = 0;
    pthread_join(thread::terminated_thread_id, NULL);
    pthread_cond_signal(&thread::terminate_thread_unset_cond);
...
}
pthread_mutex_unlock(&thread::terminate_thread);

当然，这并不是说您没有其他问题。