Re: lockdep splat in CPU hotplug

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

On Tue, Oct 21, 2014 at 05:21:21PM +0200, Jiri Kosina wrote:
> On Tue, 21 Oct 2014, Dave Jones wrote:
> 
> >  > I am seeing the lockdep report below when resuming from suspend-to-disk 
> >  > with current Linus' tree (c2661b80609).
> >  > 
> >  > The reason for CCing Ingo and Peter is that I can't make any sense of one 
> >  > of the stacktraces lockdep is providing.
> >  > 
> >  > Please have a look at the very first stacktrace in the dump, where lockdep 
> >  > is trying to explain where cpu_hotplug.lock#2 has been acquired. It seems 
> >  > to imply that cpuidle_pause() is taking cpu_hotplug.lock, but that's not 
> >  > the case at all.
> > 
> > Could inlining be confusing the trace here ?
> > 
> > You can get from cpuidle_pause to cpuidle_uninstall_idle_handler -> synchronize_rcu
> >  -> synchronize_sched -> synchronize_sched_expedited which
> > does a try_get_online_cpus which will take the cpu_hotplug.lock
> 
> Looks like this indeed is something that lockdep *should* report (*), 
> although I would be suprised that stack unwinder would be so confused by 
> this -- there is no way for synchronize_sched_expedited() to be inlined 
> all the way to cpuidle_pause().

I think that if synchronize_sched_expedited() was in fact called, it
had already returned by the time we hit this problem.  But I must confess
that I am not seeing how cpuidle_uninstall_idle_handler() gets to
synchronize_rcu().

> (*) there are multiple places where cpu_hotplug.lock -> cpuidle_lock lock 
>     dependency is assumed. The patch that Dave pointed out adds 
>     cpuidle_lock -> cpu_hotplug.lock dependency.
> 
> Still not clear whether this is what's happening here ... anyway, adding 
> Paul to CC.

Hmmm...

Both cpuidle_pause() and cpuidle_pause_and_lock() acquire cpuidle_lock,
and are at the top of both stacks.  Which was the original confusion.  ;-)

							Thanx, Paul