一个使用pthread_cond_wait出现死锁的问题

[vesontio]

在写一个练习，就是最典型的“生产者”（Producer）和“消费者”（Consumer）的问题，两者各用一种线程来代表，各个线程共享一个int类型的数值val。
Producer尝试将val置1，如果发现val已经为1，则wait；
Consumer则尝试将val置0，如果发现val已经为0，则wait；
任意一个线程修改val的值之后，就会呼叫pthread_cond_signal，知会其他线程。

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <pthread.h>

#define TH_POOL_SZ 10
#define TH_NB_LOOP 100

static pthread_mutex_t th_lock = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t th_cond = PTHREAD_COND_INITIALIZER;

typedef struct th_data {
    int val;
} th_data_t;

程序头就是如此，一个pthread_mutex_t，一个pthread_cond_t，下面的结构体是传递给线程的参数。程序会一共创建TH_POOL_SZ个“生产者”和“消费者”，每个线程尝试循环TH_NB_LOOP次，每次都试着改变val的值。

static void *cons_runnable(void *arg) {
    int i;

    for (i = 0; i < TH_NB_LOOP; i ++) {
        if (pthread_mutex_lock(&th_lock) != 0) {
            perror("Failed to lock mutex");
            pthread_exit(NULL);
        }
        printf("Consumer locked.\n");

        th_data_t *data = (th_data_t *)arg;
        while (data->val == 0) {
            printf("Value null, consumer is waiting ...\n");
            if (pthread_cond_wait(&th_cond, &th_lock) != 0) {
                perror("Failed to wait on mutex");
                pthread_exit(NULL);
            }
        }
        printf("Consumer waked up.\n");
        data->val --;
        printf("Value decremented.\n");

        if (pthread_mutex_unlock(&th_lock) != 0) {
            perror("Failed to unlock mutex");
            pthread_exit(NULL);
        }
        printf("Consumer unlocked.\n");

        if (pthread_cond_signal(&th_cond) != 0) {
            perror("Failed to signal other threads");
            pthread_exit(NULL);
        }
        printf("Consumer signaled other threads\n");

    }
    printf("Consumer terminated.\n");
    pthread_exit(NULL);
}

“消费者”（Consumer）的代码，没什么特别的，先加锁，发现val为0就wait，被**后就更新val的值，更新完解锁，并知会其他线程。

static void *prod_runnable(void *arg) {
    int i;

    for (i = 0; i < TH_NB_LOOP; i ++) {
        if (pthread_mutex_lock(&th_lock) != 0) {
            perror("Failed to lock mutex");
            pthread_exit(NULL);
        }
        printf("Producer locked.\n");

        th_data_t *data = (th_data_t *)arg;
        while (data->val == 1) {
            printf("Value set, producer is waiting ...\n");
            if (pthread_cond_wait(&th_cond, &th_lock) != 0) {
                perror("Failed to wait on mutex");
                pthread_exit(NULL);
            }
        }
        printf("Producer waked up.\n");
        data->val ++;
        printf("Value incremented.\n");

        if (pthread_mutex_unlock(&th_lock) != 0) {
            perror("Failed to unlock mutex");
            pthread_exit(NULL);
        }
        printf("Producer unlocked.\n");

        if (pthread_cond_signal(&th_cond) != 0) {
            perror("Failed to signal other threads");
            pthread_exit(NULL);
        }
        printf("Producer signaled other threads.\n");
    }
    printf("Producer terminated.\n");
    pthread_exit(NULL);
}

“生产者”（Producer）的代码基本就是和“消费者”（Consumer）反过来。

int main (void) {

    int i;
    th_data_t data;

    pthread_t th_pool_cons[TH_POOL_SZ];
    pthread_t th_pool_prod[TH_POOL_SZ];
    data.val = 0;

    for (i = 0; i < TH_POOL_SZ; i ++) {
        if (pthread_create(&th_pool_cons[i], NULL, cons_runnable, &data) != 0) {
            perror("Failed to create consumer thread");
            return -1;
        }
        if (pthread_create(&th_pool_prod[i], NULL, prod_runnable, &data) != 0) {
            perror("Failed to create producer thread");
            return -1;
        }
    }
    printf("%d threads created.\n", TH_POOL_SZ * 2);

    for (i = 0; i < TH_POOL_SZ; i ++) {
        if (pthread_join(th_pool_cons[i], NULL) != 0) {
            perror("Failed to join consumer thread");
            return -1;
        }
    }
    for (i = 0; i < TH_POOL_SZ; i ++) {
        if (pthread_join(th_pool_prod[i], NULL) != 0) {
            perror("Failed to join producer thread");
            return -1;
        }
    }
    printf("%d threads joined.\n", TH_POOL_SZ * 2);

    printf("Final value of data: %d.\n", data.val);
    printf("Fin.\n");
    return 0;
}

main函数的话基本就是创建几个线程，然后等着join他们。最后的结果出来，理论上val等于0。但是每次都是在某个“消费者”（Consumer）等待的时候就卡死不动了。
前前后后看了几遍也不知道问题出在神马地方。

希望各路大神们帮帮看看，谢谢！

[vesontio]

已经解决，应当用pthread_cond_broadcast()，而不是pthread_cond_signal()，后者如果没有叫醒正确的线程，会出现所有线程都在wait的情况。