PERFORCE change 104043 for review

[John Baldwin]

On Tuesday 15 August 2006 11:44, Divacky Roman wrote:
> On Tue, Aug 15, 2006 at 11:30:16AM -0400, John Baldwin wrote:
> > On Tuesday 15 August 2006 11:23, Roman Divacky wrote:
> > > http://perforce.freebsd.org/chv.cgi?CH=104043
> > > 
> > > Change 104043 by rdivacky at rdivacky_witten on 2006/08/15 15:22:57
> > > 
> > > 	Grrrr. Do the locking/unlocking correctly this time.
> > > 
> > > Affected files ...
> > > 
> > 
> .. //depot/projects/soc2006/rdivacky_linuxolator/compat/linux/linux_futex.c#27 
> > edit
> > > 
> > > Differences ...
> > > 
> > > 
> > 
==== //depot/projects/soc2006/rdivacky_linuxolator/compat/linux/linux_futex.c#27 
> > (text+ko) ====
> > > 
> > > @@ -339,9 +339,13 @@
> > >  			return f;
> > >  		}
> > >  	}
> > > +	if (locked == FUTEX_UNLOCKED)
> > > +	   	FUTEX_UNLOCK;
> > >  
> > >  	/* Not found, create it */
> > >  	f = malloc(sizeof(*f), M_LINUX, M_WAITOK);
> > > +	if (locked == FUTEX_UNLOCKED)
> > > +	   	FUTEX_LOCK;
> > >  	f->f_uaddr = uaddr;
> > >  	f->f_refcount = 1;
> > >  	TAILQ_INIT(&f->f_waiting_proc);
> > 
> > This readds the race. :)  See my other e-mail on what you have to do to 
handle 
> > it.
> 
> what is wrong with unprotected malloc? the memory at which f points at is 
nowhere referenced
> nowhere added etc. until protected by the lock.
> 
> can you please explain me why is this wrong? I dont see any harm with two 
processes executing
> this code paralelly.

The problem is if both threads add a futex for the same address then you'll 
have two structures for the same futex.

-- 
John Baldwin