Re: [PATCH] Lightweight userspace semphores...

[Martin Wirth <Martin.Wirth@dlr.de>]

Hubertus Franke wrote:

> Again, the trick is to not
> sync the state of the kernel and the user level. It comes naturally 
> if you properly separate the duties.
> 

Your code for usema_down is simply:

int usema_down(ulock_t *ulock)
{
	if (!__ulock_down(ulock)) {
		return 0;
	}
	return ulock_wait(ulock,0);
}

This means you do not recheck if the usema is really available after you 
return form ulock_wait(), but you may have the following situtation:

Process 1             Process 2                        Process 3
  down

                       down
                         -> call usema_wait
                         but gets preempted shortly
                         before it
                         can enter kernel mode

  up
   (kernel sema is free)

                                                          down -> got it


                       resume execution,
                       thinks its also got it
                       because kernel sema is free!

Now you have two owners!!


Martin Wirth