Call to pthread_barrier_wait() fails-CodePudding

I have a multithreaded application in C, that does some computations on a matrix. I use barriers to synchronise work. I was getting a bunch of weird errors and non-deterministic behaviour, and I realised I forgot to check the return values of pthread_barrier_wait().

Here I declare some barriers globally:

pthread_barrier_t passa,passb,check;

I have a main function which does some initialization and then spawns workers:

double **compute (int p, double P, int n, double **a){

    int r1 = pthread_barrier_init(&passa,NULL,p);
    int r2 = pthread_barrier_init(&passb,NULL,p);
    int r3 = pthread_barrier_init(&check,NULL,p);
    if(!(r1==r2==r3==0)){printf("barrier init failed\n"); exit(1);}

    pthread_t *threads = malloc(sizeof(pthread_t)*p);
    //some admin stuff 
    //spawning threads in while loop 
    int err = pthread_create(&threads[i],NULL,&compute0,args);
        if(err){
            printf("Thread Creation Error, exiting..\n");
            exit(1);
        }
        else{ //etc

Then I have the worker thread function compute0():

void *compute0(void *argsv){
    //stuff
    while(1){
        b = pthread_barrier_wait(&check);
        if(b != PTHREAD_BARRIER_SERIAL_THREAD|| b!= 0){
            printf("b : %d\n",b);
            printf("barrier failed\n"); exit(1);
        }
        //some checks
        b = pthread_barrier_wait(&passa);
        if(b != PTHREAD_BARRIER_SERIAL_THREAD|| b!= 0){
            printf("barrier failed\n"); exit(1);
        }
        //First pass
        // work 
        b = pthread_barrier_wait(&passb);
        if(b != PTHREAD_BARRIER_SERIAL_THREAD || b!= 0){
            printf("barrier failed\n"); exit(1);
        }
        //second pass
        // more work
        }
    }
}

Now I never noticed this before, but the barrier waits are actually failing. I never checked the return value of this previously :

note: one thread is used for control, computations will be run on 2 threads
Thread Created with ID : 139740189513280
Thread Created with ID : 139740181120576
================================================================
b : -1
barrier failed
b : b : 0
make: *** [Makefile:3: all] Error 1

What could be causing this?

CodePudding user response：

You have the following:

b != PTHREAD_BARRIER_SERIAL_THREAD || b != 0

It should be either of the following:

!( b == PTHREAD_BARRIER_SERIAL_THREAD || b == 0 )

b != PTHREAD_BARRIER_SERIAL_THREAD && b != 0

So you probably aren't actually getting an error. But let's assume an error is actually reported despite the above error. This only lists EINVAL (indicating a bad argument) as a possible error. I wouldn't count on that being a complete list, though. You can verify by using the following:

errno = b;
perror("pthread_barrier_wait");
exit(1);