Re: [PATCH 1/7] lockdep: Print a nice description of an irq locking issue

[Yong Zhang <yong.zhang0@gmail.com>]

On Thu, Apr 21, 2011 at 07:40:29AM -0400, Steven Rostedt wrote:
> On Thu, 2011-04-21 at 15:02 +0800, Yong Zhang wrote:
> > On Thu, Apr 21, 2011 at 9:41 AM, Steven Rostedt <rostedt@goodmis.org> wrote:
> > >  ---
> > >
> > > The above is the case when the unsafe lock is taken while holding
> > > a lock taken in irq context. But when a lock is taken that also
> > > grabs a unsafe lock, the call chain is shown:
> > >
> > >  ---
> > > other info that might help us debug this:
> > >
> > > Chain exists of:
> > >  &rq->lock --> lockA --> lockC
> > >
> > >  Possible interrupt unsafe locking scenario:
> > >
> > >       CPU0                    CPU1
> > >       ----                    ----
> > >  lock(lockC);
> > >                               local_irq_disable();
> > >                               lock(&rq->lock);
> > >                               lock(lockA);
> > >  <Interrupt>
> > >    lock(&rq->lock);
> > >
> > >  *** DEADLOCK ***
> > 
> > Or we could show this:
> > Chain exists of:
> > &rq->lock --> lockA --> lockC
> > 
> >  Possible interrupt unsafe locking scenario:
> > 
> >       CPU0                    CPU1                           CPU2
> >       ----                    ----                                     ----
> >  lock(lockC);
> >                               local_irq_disable();
> >                               lock(&rq->lock);            lock(lockA);
> >                               lock(lockA);                   lock(lockC);
> >  <Interrupt>
> >    lock(&rq->lock);
> > 
> >  *** DEADLOCK ***
> 
> 
> We could but I prefer not to ;) We have some chains that are 8 locks
> deep. I really don't want to scatter that entirely across the screen.
> Hence my "Chain exists.." statement, following an example that any
> kernel developer can (with a little thinking) see is a possible
> deadlock.

Yup :)

> 
> In fact, this code doesn't even look at the full chain, it only examines
> 3 locks in the chain, and lets the developer figure out the rest.

OK, fair enough.

> I
> could show the entire chain too.

Sure :)

Thanks,
Yong