Re: [PATCH] sched: Fix __load_balance_iterator() for cfq with only one task

[Peter Zijlstra <peterz@infradead.org>]

On Fri, 2008-09-12 at 12:05 +0530, Gautham R Shenoy wrote:
> On Fri, Sep 05, 2008 at 07:23:44PM +0200, Peter Zijlstra wrote:
> > On Fri, 2008-09-05 at 17:13 +0200, Peter Zijlstra wrote:
> > > On Fri, 2008-09-05 at 18:00 +0530, Gautham R Shenoy wrote:
> > > > sched: Fix __load_balance_iterator() for cfq with only one task.
> > > > 
> > > > From: Gautham R Shenoy <ego@in.ibm.com>
> > > > 
> > > > The __load_balance_iterator() returns a NULL when there's only one
> > > > sched_entity which is a task. It is caused by the following code-path.
> > > > 
> > > > 
> > > > 	/* Skip over entities that are not tasks */
> > > > 	do {
> > > > 		se = list_entry(next, struct sched_entity, group_node);
> > > > 		next = next->next;
> > > > 	} while (next != &cfs_rq->tasks && !entity_is_task(se));
> > > > 
> > > > 	if (next == &cfs_rq->tasks)
> > > > 		return NULL;
> > > > 	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > > >       This will return NULL even when se is a task.
> > > > 
> > > > As a side-effect, there was a regression in sched_mc behavior since 2.6.25,
> > > > since iter_move_one_task() when it calls load_balance_start_fair(),
> > > > would not get any tasks to move!
> > > > 
> > > > Fix this by checking if the last entity was a task or not.
> > > 
> > > Gregory did a similar fix a while ago, but that caused grief of some
> > > kind..
> > > 
> > > Greg, can you recollect why we pulled it? I can't seem to find it.
> > 
> > Gregory pointed me to this thread:
> > 
> >   http://lkml.org/lkml/2008/8/11/81
> > 
> > ego, can you run sysbench to confirm?
> 
> Am planning to run it today.
> 
> Mike, with what --oltp-* mode did you run the sysbench test?
> 
> That aside, if Mike's analysis is correct regarding the client/server
> pairs not running on the same CPU as buddies, shouldn't this be fixed in a
> higher level routine rather than have this anomaly in
> __load_balancer_iterator(), which is supposed to return the runnable
> tasks in the cfs_rq ?
> 
> It's current behavior is that __load_balancer_iterator() will
> return NULL even if the last entity in the list is a runnable task.
> 
> This behavior clearly hinders sched_mc powersavings from migrating
> a sole remaining task from a powersavings-sched_domain in-order
> to evacuate that domain and put all the CPUs of the domain into a
> low-power state.

Sure - there is buddy_hot in task_hot() to avoid moving buddies, and I
think we should do something like this:

@@ -590,7 +602,7 @@ account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
                add_cfs_task_weight(cfs_rq, se->load.weight);
        cfs_rq->nr_running++;
        se->on_rq = 1;
-       list_add(&se->group_node, &cfs_rq->tasks);
+       list_add_tail(&se->group_node, &cfs_rq->tasks);
 }

 static void

(most likely whitespace damaged)