Re: [BUG 2.6.25-rc3] scheduler/hotplug: some processes are dealocked when cpu is set to offline

[Gautham R Shenoy <ego@in.ibm.com>]

On Fri, Mar 07, 2008 at 05:54:51AM +0300, Oleg Nesterov wrote:
> On 03/06, Gautham R Shenoy wrote:
> >
> > On Tue, Mar 04, 2008 at 06:01:07PM +0300, Oleg Nesterov wrote:
> > > +static void check_running_task(struct task_struct *t, unsigned long now)
> > > +{
> > > +	if (!sysctl_hung_task_timeout_secs)
> > > +		return;
> > > +
> > 
> > This function gets called only when t->xxx == 0,
> > so the if below doesn't mean much, does it? :)
> > 
> > > +	if (time_before(now, t->xxx + HZ * sysctl_hung_task_timeout_secs)
> > > +		return;
> > .......
> >
> > > @@ -192,15 +214,17 @@ static void check_hung_uninterruptible_t
> > >  	if ((tainted & TAINT_DIE) || did_panic)
> > >  		return;
> > > 
> > > -	read_lock(&tasklist_lock);
> > > +	rcu_read_lock();
> > >  	do_each_thread(g, t) {
> > >  		if (!--max_count)
> > >  			goto unlock;
> > >  		if (t->state & TASK_UNINTERRUPTIBLE)
> > >  			check_hung_task(t, now);
> > > +		if (!t->xxx)
> > > +			check_running_task(t, jiff);
> 
> Of course, the check above should be
> 
> 		if (1t->xxx)
> 			check_running_task(t, jiff);
> 
> Thanks!
> 
> >From another message,
> >
> > Me too. With your patch applied there were quite a few tasks in the
> > running state which didn't get the cpu for more than 120 seconds.
> 
> (I assume you fixed the patch before using it ;)
No! Conversely, I fixed the patch because I found this behaviour a bit
odd. Couldn't run the tests again as it was a tad bit late.

> 
> Just to be sure, there were no "bad ->cpu..." messages, yes?

Hopefully should be able to catch them now. If yes, it's a problem in
the way we do migration after cpu-hotplug as Yi suggested in an earlier
mail.

http://lkml.org/lkml/2008/3/6/437

This mail from akpm says the same thing.

> 
> Oleg.

-- 
Thanks and Regards
gautham