[lwip-users] ¿¿¿¿lwIP BUGin sys_mbox_post????

[Sergio PérezAlcañiz]

I'm reporting a need that I had when porting lwIP to RT-Linux. RT-Linux offers
semaphores, so mapping rtl_sem to sys_sem was a simple task. But when posting a
semaphore something "special" happened. The RT-Linux function that posts
semaphores has a call to the scheduler at the end of its implementation, so when
called, the task performing the call is ¿ejected? from the CPU. When
multithreading is possible as in RT-Linux, there is eventually a deadlock
between tasks calling sys_mbox_post. The way I solved that was implementing a
non ejecting post for semaphores that I called sys_sem_signal_pre. At least,
that call is needed in sys_mbox_post implementation. ¿¿Is this a bug?? I don't
think so, but in those contexts similar to the RT-Linux one may cause errors.


Regards. 
Sergio.


void
sys_mbox_post(struct sys_mbox *mbox, void *msg)
{
  u8_t first;
  unsigned int state;

  sys_sem_wait(mbox->mutex);
  
  mbox->msgs[mbox->last] = msg;

  if(mbox->last == mbox->first) {
    first = 1;
  } else {
    first = 0;
  }
  
  mbox->last++;
  if(mbox->last == SYS_MBOX_SIZE) {
    mbox->last = 0;
  }

  sys_stop_interrupts(&state);
  
  if(first){
    sys_sem_signal_pre(mbox->mail);
  }
  
  sys_sem_signal(mbox->mutex);
  
  sys_allow_interrupts(&state);

}