Re: 2.6.32-rc5-mmotm1101 - lockdep whinge during early boot

[Oleg Nesterov <oleg@redhat.com>]

On 11/05, Thomas Gleixner wrote:
>
> On Thu, 5 Nov 2009, Oleg Nesterov wrote:
> > > Oleg (CC'd) made workqueues use cpu_maps_update_begin() instead of the
> > > more obvious get_online_cpus() in 3da1c84c00c7e5f.  Reverting that seems like
> > > a bad idea.
> >
> > Even if create_workueue() used get_online_cpus() instead of cpu_add_remove_lock,
> > we have the same problem: _cpu_up() takes cpu_hotplug.lock which is needed for
> > get_online_cpus(). The dependency above becomes:
> >
> > 	cpu_up()->clocksource_change_rating() takes clocksource_mutex under
> > 	cpu_hotplug.lock (cpu_hotplug_begin)
> >
> > 	clocksource_done_booting()->create_workueue() takes cpu_hotplug.lock
> > 	(get_online_cpus) under clocksource_mutex
>
> Hmm, we fixed all this lock madness in mainline already.
> clocksource_done_booting() does not longer call
> create_workqueue(). How got this code reverted in motm ?

I don't understand this code at all. But this is what I see
in Linus's tree:

	clocksource_done_booting
	    clocksource_select
	        timekeeping_notify
	            stop_machine
	                 stop_machine_create


OTOH, I don't see where native_cpu_up() path calls clocksource_change_rating(),
perhaps this was changed in -mm.

Oleg.