Re: [PATCH tip/core/rcu 01/11] mm: Place preemption point in do_mlockall() loop

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Wed, Sep 25, 2013 at 12:35:37PM -0700, Andrew Morton wrote:
> On Wed, 25 Sep 2013 06:48:04 -0700 "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:
> 
> > On Tue, Sep 24, 2013 at 09:10:47PM -0700, Andrew Morton wrote:
> > > On Tue, 24 Sep 2013 18:29:11 -0700 "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:
> > > 
> > > > --- a/mm/mlock.c
> > > > +++ b/mm/mlock.c
> > > > @@ -736,6 +736,7 @@ static int do_mlockall(int flags)
> > > >  
> > > >  		/* Ignore errors */
> > > >  		mlock_fixup(vma, &prev, vma->vm_start, vma->vm_end, newflags);
> > > > +		cond_resched();
> > > >  	}
> > > >  out:
> > > >  	return 0;
> > > 
> > > Might need one in munlock_vma_pages_range() as well - it's a matter of
> > > finding the right test case.  This will be neverending :(
> > 
> > Indeed...  I suspect that Trinity running on big-memory systems will
> > eventually find most of them via RCU CPU stall warnings, but as you say...
> > 
> > Would you like the corresponding change to munlock_vma_pages_range()
> > beforehand?
> 
> Can't decide.  If we went and poked holes in every place which looks
> like it loops for a long time, we'd be poking holes everywhere, some of
> them unnecessary.  otoh if we wait around for people to say "hey" then
> it will take a very long time to poke all the needed holes.

Yep.

> The best approach would be for someone to sit down, identify all the
> potential problem spots, attempt to craft a userspace exploit to verify
> that each one really is a problem, then fix it.  Nobody will bother
> doing this.

And if someone does bother doing this, there will no doubt be some debate
about whether or not the exploit is reasonable.

> So I dunno.  Stop asking difficult questions ;)

;-) ;-) ;-)

							Thanx, Paul