Possible circular locking dependency detected between cpu_hotplug_lock.rw_sem and wfc.work

[Thomas Gleixner]

On Thu, 27 Jul 2017, Michael Ellerman wrote:
> Thomas Gleixner <tglx at linutronix.de> writes:
> 
> > On Wed, 26 Jul 2017, Michael Ellerman wrote:
> >
> >> Hi Thomas,
> >> 
> >> I'm seeing the lockdep barf below on some bare metal Power8 machines.
> >> 
> >> This seems to be caused by our smp_cpus_done(), which does:
> >> 
> >>   void __init smp_cpus_done(unsigned int max_cpus)
> >>   {
> >>   	/*
> >>   	 * We want the setup_cpu() here to be called on the boot CPU, but
> >>   	 * init might run on any CPU, so make sure it's invoked on the boot
> >>   	 * CPU.
> >>   	 */
> >>   	if (smp_ops && smp_ops->setup_cpu)
> >>   		work_on_cpu_safe(boot_cpuid, smp_setup_cpu_workfn, NULL);
> >> 
> >> 
> >> I don't think CPU hotplug can happen at this point, so I don't think
> >> there's really a bug.
> >> 
> >> But it looks like the work_on_cpu_safe() call could just go away, since
> >> you pinned init to the boot CPU in 8fb12156b8db ("init: Pin init task to
> >> the boot CPU, initially"). Though I can't see where init is unpinned, so
> >> maybe we do still need to do it?
> >
> > It's undone in sched_init_smp(). So it looks safe. The call order is:
> >
> >      smp_init()
> > 	...
> > 	smp_cpus_done()
> >
> >      sched_init_smp()
> 
> Great thanks.
> 
> Patch on the way.

Hmm. Second thoughts. The issue is the stability of the CPUs. Surely the
boot CPU can't go away at that point, but the debug stuff does not know
about it. maybe I'm missing something ....

Thanks,

	tglx