Re: Problem with pthread_con

Just got it now :

In fact the 3 threads looks to be stuck on the pthread_mutex 'U.lock' , not
the cond.lock

pid 1234 is the pid of the the 'main' thread,
pid 1235 and 1236 pids of the workers threads

0x809ac0 is the address of U.lock

# strace -p 1234
Process 1234 attached - interrupt to quit
futex(0x809ac0, FUTEX_WAIT, 8, NULL <unfinished ...>
Process 1234 detached

# strace -p 1234
Process 1234 attached - interrupt to quit
futex(0x809ac0, FUTEX_WAIT, 12, NULL <unfinished ...>
Process 1234 detached

# strace -p 1235
Process 1235 attached - interrupt to quit
futex(0x809ac0, FUTEX_WAIT, 16, NULL <unfinished ...>
Process 1235 detached

# strace -p 1236
Process 1236 attached - interrupt to quit
futex(0x809ac0, FUTEX_WAKE, 1)          = 1
futex(0x809ad8, FUTEX_WAKE, 1)          = 1
futex(0x809ae8, FUTEX_WAIT, 12322, NULL) = -1 EAGAIN (Resource temporarily
unavailable)
futex(0x809ac0, FUTEX_WAIT, 2, NULL)    = -1 EAGAIN (Resource temporarily
unavailable)
futex(0x809ac0, FUTEX_WAIT, 5, NULL)    = 0
futex(0x809ac0, FUTEX_WAKE, 1)          = 0
futex(0x809ae8, FUTEX_WAIT, 12323, NULL) = 0
futex(0x809ac0, FUTEX_WAKE, 1)          = 0
...

All threads are running now.

Next time it reblocks, I will use 'strace -p 1236' directly

Thanks

From: "dada1" <[email protected]>
> 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) ;
>                 }
>         }
> }
>
>
> -- 
> Phil-list mailing list
> [email protected]
> https://www.redhat.com/mailman/listinfo/phil-list
>