mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Peter Zijlstra <peterz@infradead.org>
To: xialiang <xiaiaxaxi@sjtu.edu.cn>
Cc: Robin Holt <holt@sgi.com>,
	Bart Van Assche <bart.vanassche@gmail.com>,
	linux-kernel@vger.kernel.org
Subject: Re: How can I migrate a currently running task?
Date: Sat, 28 Jun 2008 09:55:39 +0200	[thread overview]
Message-ID: <1214639739.12348.13.camel@twins> (raw)
In-Reply-To: <20080628120521.kjrvzpgpwwcc0g0c@webmail1.sjtu.edu.cn>

On Sat, 2008-06-28 at 12:05 +0800, xialiang wrote:
> Quoting Robin Holt <holt@sgi.com>:
> 
> > On Fri, Jun 27, 2008 at 09:38:05PM +0800, xialiang wrote:
> >> Quoting Bart Van Assche <bart.vanassche@gmail.com>:
> >>
> >>> On Fri, Jun 27, 2008 at 3:17 PM, 夏亮 <xiaiaxaxi@sjtu.edu.cn> wrote:
> >>>> I am working on Intel Duo Core with Linux OS 2.6.21, and I'd like to
> >>>> migrate task from one cpu to another cpu.
> >>>> In SMP systems, load_balance() function uses move_tasks() to move  
> >>>>  processes
> >>>> from source runqueue to local runqueue, but it does not move a currently
> >>>> running task. If I want to migrate a currently running task from source
> >>>> runqueue to local runqueue, how can I do? Any suggestion is preferred.
> >>>
> >>> Are you familiar with the glibc pthread_setaffinity_np() function
> >>> and/or the sched_setaffinity() system call ?
> >>>
> >>> Bart.
> >>>
> >>
> >> Yes. I know sched_setaffinity(), it sets cpu_mask of a task. I want to
> >> use it in a timer interrupt( scheduler_tick() ), can I?
> >
> > Don't use it from a timer handler.  You can use migrate_task() to move the
> > task, but your email from a couple days ago said you wanted to move the
> > task at the head of the runqueue due to cpu heat or something like that.
> > That is a very imprecise way to move the task as any unfortunate task
> > that happens to be running when the timer tick occurs could get migrated,
> > not the one generating the work.
> >
> > I think you really want to look at the other areas of the kernel which do
> > stuff like throttling the cpu.  Just moving a task off the cpu does not
> > prevent it from being used for another compute intesive load.  You could
> > take the cpu offline.  I guess to do that, I would use schedule_work()
> > or kthread_create() to get out of the timer context and into a regular
> > thread context then take the cpu offline from there.
> >
> > Good Luck,
> > Robin
> >
> 
>        I would move a task at the head of the runqueue, but I already  
> know which task creates the heat. I get the profile of tasks  
> beforehand. So if I find the temperature of a cpu is much higher than  
> that of the other, and coincidentally the currently running task of  
> the cpu is what I want to migrate. Can I migrate it in the  
> scheduler_tick()? 

No, you cannot migrate current from hardirq context. 

> Or where should I put the migration code? Thanks!

If you want to balance temperature, the best place would be the load
balancer.

Just don't wreck all the other balance conditions.


  reply	other threads:[~2008-06-28  7:55 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-06-27 13:17 夏亮
2008-06-27 13:25 ` Bart Van Assche
2008-06-27 13:38   ` xialiang
2008-06-27 18:48     ` Robin Holt
2008-06-28  4:05       ` xialiang
2008-06-28  7:55         ` Peter Zijlstra [this message]
2008-06-27 14:27 ` Dmitry Adamushko

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=1214639739.12348.13.camel@twins \
    --to=peterz@infradead.org \
    --cc=bart.vanassche@gmail.com \
    --cc=holt@sgi.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=xiaiaxaxi@sjtu.edu.cn \
    /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®