Problem with pthread_con

Hello Ulrich

I have a problem with pthread_cond_signal()  (nptl 0.58)

A very simple program can block. It is very rare.

I dont know yet if this is a kernel bug, a processor bug, or a bug from me !

Usually, doing a strace -p one_pid unblocks the program.
(the threads seems blocked to get the condvar.lock)

Tested on a bi athlon, kernel 2.6.0-test5

Thanks
Eric Dumazet

Sample program (without error checks for clarification) :

#include <pthread.h>

#define MAXWORK 8

struct {
        pthread_mutex_t lock ;
        pthread_cond_t  cond ;
        unsigned int unload ;
        unsigned int load ;
        unsigned long miss ;
        struct {
                int delai ;
                } tab[MAXWORK] ;
        } U ;


void *worker(void *arg)
{
int delai ;

for (;;) {

        pthread_mutex_lock(&U.lock) ;
        while (U.load == U.unload) {
                pthread_cond_wait(&U.cond, &U.lock) ;
                }
        delai = U.tab[U.unload%MAXWORK].delai ;
        U.unload++ ;
        pthread_mutex_unlock(&U.lock) ;

        usleep(delai) ; /* some work */
        }

}

void post_work()
{
int delai = rand() % 50000  ;
unsigned int nb ;
pthread_mutex_lock(&U.lock) ;
nb = U.load - U.unload ;
if (nb < MAXWORK) {
        U.tab[U.load%MAXWORK].delai = delai ;
        U.load++ ;
        pthread_mutex_unlock(&U.lock) ;/* so that awaken thread doesnt have
to wait for the lock */
        pthread_cond_signal(&U.cond) ;
        }
else {
        U.miss++ ;
        pthread_mutex_unlock(&U.lock) ;
        }
}


int main(int argc, char *argv[])
{
time_t t0 , t1 ;
pthread_t tid1 ;
pthread_t tid2 ;

pthread_create(&tid1, 0, worker, 0) ;
pthread_create(&tid2, 0, worker, 0) ;

t0 = time(0) ;
for (;;) {
        post_work() ;
        usleep(rand() % 50000) ;
        t1 = time(0) ;
        if (t1 != t0) {
                t0 = t1 ;
                write(1, "*", 1) ;
                }
        }
}