mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
To: Paul Turner <pjt@google.com>
Cc: LKML <linux-kernel@vger.kernel.org>,
	cgroups <cgroups@vger.kernel.org>, Ingo Molnar <mingo@elte.hu>,
	Peter Zijlstra <a.p.zijlstra@chello.nl>,
	Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Subject: Re: [PATCH 1/3] sched: fix cgroup movement of newly created process
Date: Wed, 14 Dec 2011 10:04:38 +0900	[thread overview]
Message-ID: <20111214100438.6226d661.nishimura@mxp.nes.nec.co.jp> (raw)
In-Reply-To: <CAPM31RJzALSAcxEYELUbM0ZXwT=WoOHj7w-rDVC73VujnQVH1A@mail.gmail.com>

On Tue, 13 Dec 2011 04:01:21 -0800
Paul Turner <pjt@google.com> wrote:

> On 12/12/2011 10:57 PM, Daisuke Nishimura wrote:
> 
> > There is a small race between do_fork() and sched_move_task(), which is trying
> > to move the child.
> >
> >             do_fork()                 sched_move_task()
> > --------------------------------+---------------------------------
> >   copy_process()
> >     sched_fork()
> >       task_fork_fair()
> >         -> vruntime of the child is initialized
> >            based on that of the parent.
> 
> 
> Hmm, so here vruntime of child is actually initialized to vruntime - min_V
> 
> >   -> we can see the child in "tasks" file now.
> >                                     task_rq_lock()
> >                                     task_move_group_fair()
> 
> 
> So since on a regular fork we just copy and don't actually go through
> the attach muck I'm assuming this is an external actor who's seen the
> child in the tasks file and is moving it?
> 
Right.

> >                                       ->child.se.vruntime -= (old)cfs_rq->min_vruntime
> >                                       ->child.se.vruntime += (new)cfs_rq->min_vruntime
> 
> 
> This would then add delta min_V between the two cfs_rqs
> 
> >                                     task_rq_unlock()
> >   wake_up_new_task()
> >     ...
> >     enqueue_entity()
> >       child.se->vruntime += cfs_rq->min_vruntime
> 
> >
> > As a result, vruntime of the child becomes far bigger than min_vruntime,
> > if (new)cfs_rq->min_vruntime >> (old)cfs_rq->min_vruntime.
> >
> > This patch fixes this problem by just ignoring such process in task_move_group_fair(),
> > because the vruntime has already been normalized in task_fork_fair().
> >
> > Signed-off-by: Daisuke Nishimura <nishimura-YQH0OdQVrdy45+QrQBaojngSJqDPrsil@public.gmane.org>
> > ---This would need an explanatory
> >  kernel/sched_fair.c |    4 ++--
> >  1 files changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
> > index 5c9e679..df145a9 100644
> > --- a/kernel/sched_fair.c
> > +++ b/kernel/sched_fair.c
> > @@ -4922,10 +4922,10 @@ static void task_move_group_fair(struct task_struct *p, int on_rq)
> >  	 * to another cgroup's rq. This does somewhat interfere with the
> >  	 * fair sleeper stuff for the first placement, but who cares.
> >  	 */
> > -	if (!on_rq)
> > +	if (!on_rq && p->state != TASK_RUNNING)
> >  		p->se.vruntime -= cfs_rq_of(&p->se)->min_vruntime;
> 
> 
> We also have the choice of keying off something like
> !se.sum_exec_runtime here which is reset in sched_fork() which might
> be less fragile/make the problem interaction more obvious to.  Either

sum_exec_runtime can be used for a process that has just been created,
but IIUC it cannot be used for a process that is being woken up from
sleeping(in [3/3] case).
Or, do you mean using p->se.sum_exec_runtime for [1/3] and p->state for [3/3] ?
If so, I'll do it in the next post.

> way this is really a corner case deserving of an explanatory comment.
> This is a little icky but I don't have any wildly better ideas.
> 
I'll add comments.

> >  	set_task_rq(p, task_cpu(p));
> > -	if (!on_rq)
> > +	if (!on_rq && p->state != TASK_RUNNING)
> >  		p->se.vruntime += cfs_rq_of(&p->se)->min_vruntime;
> >  }
> >  #endif
> 
> 
> Acked-by: Paul Turner <pjt@google.com>

Thanks you for your reviews.

Daisuke Nishimura.

  reply	other threads:[~2011-12-14  1:16 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2011-12-13  6:57 [PATCH 0/3] sched: some fixes for vruntime calculation related to cgroup movement Daisuke Nishimura
2011-12-13  6:57 ` [PATCH 1/3] sched: fix cgroup movement of newly created process Daisuke Nishimura
2011-12-13 12:01   ` Paul Turner
2011-12-14  1:04     ` Daisuke Nishimura [this message]
2011-12-13 12:41   ` Peter Zijlstra
2011-12-14  1:05     ` Daisuke Nishimura
2011-12-13  6:58 ` [PATCH 2/3] sched: fix cgroup movement of forking process Daisuke Nishimura
2011-12-13 12:22   ` Paul Turner
2011-12-13  6:59 ` [PATCH 3/3] sched: fix cgroup movement of waking process Daisuke Nishimura
2011-12-13 12:24   ` Paul Turner
2011-12-15  5:35 ` [PATCH -tip 0/3] sched: some fixes for vruntime calculation related to cgroup movement(v2) Daisuke Nishimura
2011-12-15  5:36   ` [PATCH 1/3] sched: fix cgroup movement of newly created process Daisuke Nishimura
2011-12-21 11:45     ` [tip:sched/core] sched: Fix " tip-bot for Daisuke Nishimura
2011-12-15  5:36   ` [PATCH 2/3] sched: fix cgroup movement of forking process Daisuke Nishimura
2011-12-21 11:44     ` [tip:sched/core] sched: Fix " tip-bot for Daisuke Nishimura
2011-12-21 17:26       ` Tejun Heo
2011-12-21 17:37         ` Tejun Heo
2011-12-22  1:54           ` Frederic Weisbecker
2011-12-22  2:01             ` Tejun Heo
2011-12-15  5:37   ` [PATCH 3/3] sched: fix cgroup movement of waking process Daisuke Nishimura
2011-12-21 11:45     ` [tip:sched/core] sched: Fix " tip-bot for Daisuke Nishimura
2011-12-19  2:55   ` [PATCH -tip 0/3] sched: some fixes for vruntime calculation related to cgroup movement(v2) Daisuke Nishimura

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20111214100438.6226d661.nishimura@mxp.nes.nec.co.jp \
    --to=nishimura@mxp.nes.nec.co.jp \
    --cc=a.p.zijlstra@chello.nl \
    --cc=cgroups@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@elte.hu \
    --cc=pjt@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®