Re: [PATCH 1/1] sched: fix cpu_down deadlock

[Oleg Nesterov <oleg@redhat.com>]

On 09/09, Jiri Slaby wrote:
>
> On 09/09/2009 01:53 PM, Peter Zijlstra wrote:
> > On Wed, 2009-09-09 at 13:41 +0200, Jiri Slaby wrote:
> >> Jiri Slaby wrote:
> >>> Thanks, in the end I found it manually. Goddammit! It's an -mm thing:
> >>> cpu_hotplug-dont-affect-current-tasks-affinity.patch
> >
> > Is there a git tree with -mm in some place? I can't seem to find that
> > patch in my inbox.
> >
> > All I can find is some comments from Oleg that the patch looks funny.
>
> Yes, here:
> git://git.zen-sources.org/zen/mmotm.git
>
> Actually I found Oleg came up with better solution to add
> move_task_off_dead_cpu to take_cpu_down.
>
> A discussion regarding this is at:
> http://lkml.indiana.edu/hypermail/linux/kernel/0907.3/02278.html
>
> So what's the status of the patches, please?

This patch depends on another one, please see

	"[PATCH] cpusets: rework guarantee_online_cpus() to fix deadlock with cpu_down()"
	http://marc.info/?t=124910242400002

	(as the changelog says, the patch is not complete: we need
	 ->cpumask_lock every time we update cs->allowed, but this
	 should be trivial)

In short: cpuset_lock() is buggy. But more importantly it is afaics unneeded,
and imho should die. I seem to answer all Lai's questions, but the patch was
ignored by maintainers.

I noticed another race in update_cpumask() which I was going to fix, but since
maintainers ignore me I lost the motivtion ;) Besides, currently I dont have
the time anyway.

So I think the original patch which creates the kthread is the best option.

Oleg.