Re: mm: deadlock between get_online_cpus/pcpu_alloc

From: Thomas Gleixner
Date: Thu Feb 09 2017 - 11:14:24 EST

On Thu, 9 Feb 2017, Christoph Lameter wrote:
> On Thu, 9 Feb 2017, Thomas Gleixner wrote:
> > > The stop_machine would need to ensure that all cpus cease processing
> > > before proceeding.
> >
> > Ok. I try again:
> >
> > CPU 0 CPU 1
> > for_each_online_cpu(cpu)
> > ==> cpu = 1
> > stop_machine()
> >
> > Stops processing on all CPUs by preempting the current execution and
> > forcing them into a high priority busy loop with interrupts disabled.
> Exactly that means we are outside of the sections marked with
> get_online_cpous().
> > It does exactly what you describe. It stops processing on all other cpus
> > until release, but that does not invalidate any data on those cpus.
> Why would it need to invalidate any data? The change of the cpu masks
> would need to be done when the machine is stopped. This sounds exactly
> like what we need and much of it is already there.

You are just not getting it, really.

The problem is that this for_each_online_cpu() is racy against a concurrent
hot unplug and therefor can queue stuff for a not longer online cpu. That's
what the mm folks tried to avoid by preventing a CPU hotplug operation
before entering that loop.

> Lets get rid of get_online_cpus() etc.

And that solves what?

Can you please start to understand the scope of the whole hotplug machinery
including the requirements for get_online_cpus() before you waste
everybodys time with your uninformed and halfbaken proposals?