Subj : Semaphores and pthreads
To   : comp.programming.threads
From : Paolo
Date : Sun Jan 09 2005 04:06 pm

Hello!
First of all: Happy New Year =)

Then my question:
I have a program that can be imagined as a "producers consumer" program.
it is written in C, and it uses pthreads.
2 threads produce resources and one cosumes them. to syncronize them i
used a semaphore (sem_t) and a mutex

imagine that the life of a producer is:

foo=produce();
ptrehad_mutex_lock(&mutex);
enqueue(foo);
pthread_mutex_unlock(&mutex);
sem_post(&semaphore);


and the life of the consumer is:

sem_wait(&semaphore);
ptrehad_mutex_lock(&mutex);
bar=dequeue();
check_if_the_queue_is_empty(&bar);
pthread_mutex_unlock(&mutex);
process_the_resource(bar);


now, i have the problem that seldom it happens that inside the consumer
i wake up from the sem_wait(&semaphore), but i found the queue as empty.

So i tried to define these macros, with the aid of a 2nd mutex, namely
mutex2:

MY_P() {
 sem_wait(&semaphore);\
 pthread_mutex_lock(&mutex2);\
 semcounter--;\
 pthread_mutex_unlock(&mutex2);\
}

MY_V() {
 pthread_mutex_lock(&mutex2);\
 sem_post(&semaphore);\
 semcounter++;\
 pthread_mutex_unlock(&mutex2);\
}

I use them in place of sem_wait(&semaphore) and sem_post(&semaphore).
When the disaster happen (that is i wake up but have empty queue), i
said to
the function check_if_the_queue_is_empty() to do:

int temp,val;
[...]
pthread_mutex_lock(&mutex2);
val=sem_getvalue(&semaphore,&temp);
val=semcounter;
printf("SEMCOUNTER:%d SEM_VALUE:%d\n",val,temp);
pthread_mutex_unlock(&semaforo);
[...]

and i obtain that the SEM_VALUE is correctly 0 (else i can't resume from
sem_wait()),
but the SEMCOUNTER is -1!!!!!  that is i have a deficit in the P()/V()
balance.

this means that either someone other than my threads do a V() or that
when i do
sem_wait() and the sem_value is 0, the function does not decrement the
semaphore counter..

I honestly find this a pretty strange behaviour. and i don't know how to
fix it, becouse
i'm sure that i do sem_wait() and sem_post() when i should, since my own
counter (semcounter)
cointains the right #V() - #P() value.

Any suggestion will be appreciated

Thank you and excuse me for this long message :-)
Paolo

.