Re: [PATCH v4] CPU hotplug: active_writer not woken up in some cases - deadlock

[Oleg Nesterov <oleg@redhat.com>]

On 12/12, David Hildenbrand wrote:
>
> > This is subjective, but how about
> >
> > 	static bool xxx(void)
> > 	{
> > 		mutex_lock(&cpu_hotplug.lock);
> > 		if (atomic_read(&cpu_hotplug.refcount) == 0)
> > 			return true;
> > 		mutex_unlock(&cpu_hotplug.lock);
> > 		return false;
> > 	}
> >
> > 	void cpu_hotplug_begin(void)
> > 	{
> > 		cpu_hotplug.active_writer = current;
> >
> > 		cpuhp_lock_acquire();
> > 		wait_event(&cpu_hotplug.wq, xxx());
> > 	}
> >
> > instead?
> >
> > Oleg.
> >
>
> [   50.662459] do not call blocking ops when !TASK_RUNNING; state=2 set at [<000000000017340e>] prepare_to_wait_event+0x7a/0x124
> [   50.662472] ------------[ cut here ]------------
> [   50.662475] WARNING: at kernel/sched/core.c:7301
> [   50.662477] Modules linked in:
> [   50.662482] CPU: 5 PID: 225 Comm: cpu_start_stop. Not tainted 3.18.0+ #59
> [   50.662485] task: 0000000001f94b20 ti: 0000000001ffc000 task.ti: 0000000001ffc000
> ...
>
> Looks like your suggestion won't work. We can only set the task to
> TASK_UNINTERRUPTIBLE after taking the lock.

Yeees, this warning (and wait_woken() helpers) was specially added
to catch/fix the problem like this, sorry for confusion.

Easy to fix, just

	-	mutex_lock(&cpu_hotplug.lock);
	+	if (!mutex_trylock(&cpu_hotplug.lock))
	+		return false;

If .lock is locked then it is hold by get_online_cpus(), and it is going
to increment the counter.

I would like to say that this is what I actually meant but now I can not
recall if this is true ;)

But please ignore. Your next version looks simple/clear enough.

Oleg.