Re: [PATCH -mm] scheduler: fix the return of the first time_slice

[Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>]

At Mon, 16 Apr 2007 14:45:59 +0400,
Oleg Nesterov wrote:
> 
> On 04/16, Satoru Takeuchi wrote:
> >
> > --- linux-2.6.21-rc6-mm1.tsfix.orig/kernel/exit.c	2007-04-15 16:56:12.000000000 +0900
> > +++ linux-2.6.21-rc6-mm1.tsfix/kernel/exit.c	2007-04-15 16:56:17.000000000 +0900
> > @@ -680,10 +680,14 @@
> >  			reaper = child_reaper(father);
> >  			break;
> >  		}
> > +		if (reaper->time_slice_reaper == father)
> > +			reaper->time_slice_reaper = NULL;
> >  	} while (reaper->exit_state);
> >  
> >  	list_for_each_safe(_p, _n, &father->children) {
> >  		p = list_entry(_p, struct task_struct, sibling);
> > +		if (p->time_slice_reaper == father)
> > +			p->time_slice_reaper = NULL;
> >  		choose_new_parent(p, reaper);
> >  		reparent_thread(p, father);
> >  	}
> > Index: linux-2.6.21-rc6-mm1.tsfix/kernel/sched.c
> > ===================================================================
> > --- linux-2.6.21-rc6-mm1.tsfix.orig/kernel/sched.c	2007-04-15 16:56:12.000000000 +0900
> > +++ linux-2.6.21-rc6-mm1.tsfix/kernel/sched.c	2007-04-15 16:56:17.000000000 +0900
> > @@ -1693,7 +1693,7 @@
> >  	 * The remainder of the first timeslice might be recovered by
> >  	 * the parent if the child exits early enough.
> >  	 */
> > -	p->first_time_slice = 1;
> > +	p->time_slice_reaper = current;
> >  	p->timestamp = sched_clock();
> >  	local_irq_enable();
> 
> I am afraid this doesn't work for CLONE_THREAD. Suppose that some sub-thread (not
> main thread) T1 creates another sub-thread, T2.
> 
> 	T2->time_slice_reaper = T1.
> 
> then T1 exits, but forget_original_parent(T1) doesn't change T2->time_slice_reaper,
> because T2 is not on T1->children list. Now T2->time_slice_reaper points to an
> already freed task_struct.

I intended that CLONE_THREAD case is managed on the do-while loop seeking `reaper'.
However, anyway, my patch has a bug. On T1 exiting, if there are any living threads
ahead of T2, T2->time_slice_reaper isn't checked. What a easy mistake ... sorry.

I'll post fixed version soon.

Satoru