Sequence of rdlock, rdunlock followed by wrlock,wrunlock in threads causes deadlock

[Ashish]

This little piece of code should keep running forever without any trouble. However, when I ran it at my end it results in deadlock situation.
I apparently can't see any reason for deadlock. Please can someone have a quick look to ensure I am not doing anything in wrong order here.
(This is on Windows XP. I am using libuv-v1.0.0. Also, I've set threadpool size to 31 before running the code)


#include <uv.h>

uv_work_t gWorkT[31];
uv_rwlock_t g_rwlLock;
void rd_wr_locks_in_thread(uv_work_t* work_t)
{
    static int value = 0;
    while(1)
    {
        uv_rwlock_rdlock(&g_rwlLock);
        BOOL isvalueodd = value % 2;
        uv_rwlock_rdunlock(&g_rwlLock);

        uv_rwlock_wrlock(&g_rwlLock);
        value ++;
        uv_rwlock_wrunlock(&g_rwlLock);
    }
}

void after_rd_wr_locks_in_thread(uv_work_t* work_t, int status)
{
}

int main()
{
    loop = uv_default_loop();

    int retval = uv_rwlock_init(&g_rwlLock);
    if (retval < 0)
    {
        printf("\nError initilizing lock");
        return -1;
    }

    for (int i=0; i<31; i++)
    {
        int RetVal = uv_queue_work(loop, &gWorkT[i], rd_wr_locks_in_thread, after_rd_wr_locks_in_thread);
    }
    return uv_run(loop, UV_RUN_DEFAULT);

}

[Ashish]

Any updates on this please? I investigated it little further and realized libuv relies on only Critical Section objects to synchronize threads.
I suspect there are problems with that approach.

Now if we use read and write locks as demonstrated in the code below, what happens is, a thread enters in a critical section then another thread is very likely to go ahead of it and call leave critical section.

As per Microsoft's documentation,
"If a thread calls LeaveCriticalSection when it does not have ownership of the specified critical section object, an error occurs that may cause another thread using EnterCriticalSection to wait indefinitely."

[Bert Belder]

You're right - the xp/2k3 "fallback" implementation of read-write locks seems broken. A patch would be welcome.