[gccsdk] Dynamic libraries vs pthreads

Peter Naulls peter at chocky.org
Tue Sep 2 21:26:31 PDT 2008

Lee Noar wrote:

> Yes, does work on my machine, and I see that Firefox 
> uses pthread_yield() quite extensively. Whatever triggers this bug must 
> be fairly rare as it doesn't always fail. I don't see how it could be 
> specific to shared library code as it really isn't any different to 
> non-PIC code as far as the CPU is concerned

Well, more to the mystery.  I'm pretty sure I have your change - 
although TBH it's sometimes a bit hard to tell with our build
setup :-(

I'm looking for problems with pthread conditions, which is
afflicting nspr.  Anyway, using the program below, it dies
in __pthread_yield_return, looks like after the sleep call returns:


Fatal signal received: Quit

Stack backtrace:

Running thread 0x3e980
   (   a7f4c) pc: 5c860c90 lr: 5c86123c sp:    a7f50  ?()
   (   a7ff4) pc: 5c860d94 lr: 5c861834 sp:    a7ff8  ?()

Thread 0xdb18
   (   a7f28) pc: 5c7fcbb8 lr: 5c7fa714 sp:    a6f58 
   (   a6f68) pc: 5c7fcb18 lr: 5c7fa714 sp:    a6f6c  pthread_yield()
   (   a6fa0) pc: 5c7fa5d4 lr:     877c sp:    a6fa4  ?()
   (   a6fc8) pc:     8704 lr:     87e0 sp:    a6fcc  ?()
   (   a6fe0) pc:     87b8 lr: 5c7eef04 sp:    a6fe4  ?()
   (   a6ff4) pc: 5c7eeeac lr: 5c8ac0b0 sp:    a6ff8  ?()

But only outside the desktop, not in a taskwindow, and it goes away
if it's static.  This is RISC OS 5 of course.


#include <stdio.h>
#include <pthread.h>
#include <sys/time.h>
#include <unistd.h>

pthread_mutex_t mut = PTHREAD_MUTEX_INITIALIZER;
pthread_cond_t cond = PTHREAD_COND_INITIALIZER;

static void *thread_loop(void *unused)
   while (1) {

   return NULL;

void waittest(void) {
   struct timeval now;
   struct timespec timeout;
   int retcode;

   gettimeofday(&now, NULL);
   timeout.tv_sec = now.tv_sec + 5;
   timeout.tv_nsec = now.tv_usec * 1000;
   retcode = 0;

   retcode = pthread_cond_timedwait(&cond, &mut, &timeout);
   printf("done: %d\n", retcode);


int main(void) {
  pthread_t thread;

   pthread_create(&thread, NULL, thread_loop, NULL);

   return 0;


More information about the gcc mailing list