RE: [Linux-ia64] glibc bug / pthread bug???

From: Boehm, Hans <hans_boehm_at_hp.com>
Date: 2002-10-02 07:28:22
It's probably a feature.  The test program is incorrect.  However, this does seem to be a rather common bug, and a particularly nasty corner of the pthreads spec.

In a multithreaded application, you are only allowed to make async-signal-safe calls in the child.  AFAIK, neither timer_create nor perror are async-signal-safe.  Neither is exit().

This requirement makes some small amount of sense, since fork() creates a new address space containing only that thread.  Locks that happened to be held by other threads at the time of the fork may be held in the child process, and thus any attempt to acquire a (system?) lock or the like may deadlock the child process.  There may also be other linuxthreads specific issues, e.g. I'm not sure to what extent the manager thread is functional in the child.

Hans

> -----Original Message-----
> From: Jack Steiner [mailto:steiner@sgi.com]
> Sent: Tuesday, October 01, 2002 1:08 PM
> To: linux-ia64@linuxia64.org
> Subject: [Linux-ia64] glibc bug / pthread bug???
> 
> 
> Has anyone seen this problem or know enough about glibc/threads
> to have a good idea about what is busted.
> 
> 
> The test case listed below consistently hang using 2.4.18 & 
> glibc-2.2.4-19.3.
> 
> If only one instance (fork=1)of the test is run, it runs ok. 
> If multiple copies
> of the test are run, it hangs in:
> 
> 	 *      rt_sigsuspend
> 	 *      __sigsuspend
> 	 *      __pthread_wait_for_restart_signal
> 	 *      pthread_cond_wait
> 	 *      thread_func
> 	 *      pthread_start_thread
> 
> Has anyone seen this???
> 
> /*
>   * compile with
>   *      gcc -g -Wall -o test test.c -lrt
>   *
>   * execute 
>   *      test [<inner_loop_count> [<outer_loop_count>]]
>   *
>   *
>   * Test creates timer threads that hang in 
>   *      rt_sigsuspend
>   *      __sigsuspend
>   *      __pthread_wait_for_restart_signal
>   *      pthread_cond_wait
>   *      thread_func
>   *      pthread_start_thread
>   *
>   */ 
> 
>  #include <signal.h>
>  #include <time.h>
>  #include <stdio.h>
>  #include <sys/types.h>
>  #include <sys/time.h>
>  #include <sys/times.h>
>  #include <stdlib.h>
>  #include <unistd.h>
>  #include <sys/wait.h>
> 
>  int     forks=5;
>  int     count=5;
> 
>  void *
>  slave(void)
>  {
>          timer_t timerid;
>          pid_t   pid;
>          int     i, status;
> 
>          for (i=0; i<forks; i++) {
>                  if (fork() == 0) {
>                          if (timer_create(CLOCK_REALTIME, 
> NULL, &timerid) == -1) {
>                                  perror("timer_create");
>                                  exit(1);
>                          }
>                          execlp("/bin/date", "date",  (char *)0);
>                  }
> 
>                  pid = wait(&status);
>          }
> 
>          exit(0);
>  } 
> 
>  int
>  main (int argc, char *argv[])
>  {
>          int i;
> 
>          count = argc >= 2 ? atoi(argv[1]) : 5;
>          forks = argc >= 3 ? atoi(argv[2]) : 5;
> 
>          for (i=0; i<count; i++) {
>                  if (fork() == 0)
>                          slave();
>          }
> 
>          exit(0);
>  }
> 
> 
> 
> -- 
> Thanks
> 
> Jack Steiner    (651-683-5302)   (vnet 233-5302)      steiner@sgi.com
> 
> 
> _______________________________________________
> Linux-IA64 mailing list
> Linux-IA64@linuxia64.org
> http://lists.linuxia64.org/lists/listinfo/linux-ia64
> 
Received on Tue Oct 01 14:28:42 2002

This archive was generated by hypermail 2.1.8 : 2005-08-02 09:20:10 EST