On Sat, 2010-02-06 at 09:31 +0000, Tom Parker wrote:
> On 2 February 2010 21:16, maximilian attems <max@stro.at> wrote:
> > Squeeze will release with 2.6.32 can anyone of you still
> > reproduce this?
> >
> > also please make sure to use latest powertop, aka
> > ii powertop 1.13~pre201001 Linux tool to find out what is using power
>
> palfrey@drone:[~] dpkg -l |grep powertop
> ii powertop 1.13~pre20100125-1
> Linux tool to find out what is using power
> o
> palfrey@drone:[~] uname -a
> Linux drone 2.6.32-1-686 #1 SMP Mon Feb 1 01:37:26 UTC 2010 i686 GNU/Linux
>
> Attached is powertop.log ('powertop -d') and top.log('top -b -n 1')
>
> As you can see, the major wakeup is still "Load balancing tick", which
> I'm guessing is a translated name for hrtimer_start_expires? I think
> the ACPI estimate of 1.1W is a little off though...
hrtimer_start_expires() is a generic function in the kernel to schedule
a wakeup by the high-resolution timer. In 2.6.32 it is an inline
function that calls hrtimer_start_range_ns(). Powertop recognises and
relabels some specific calling sequences, including this one.
If I understand correctly, these are scheduler ticks that periodically
interrupt the running task (preemption). However, on a kernel compiled
with the NOHZ option (as Debian's kernel images are) this should never
wake the system up - scheduler ticks are disabled when there are no
tasks ready to run and no other timers due to expire before the next
tick.
I can only suggest you report this upstream at
<http://bugzilla.kernel.org>. Use product 'Process Management',
component 'Scheduler'. Let us know the bug number so we can track it.
> So, doesn't look fixed. I'm currently at FOSDEM if anyone with more
> knowledge wants to borrow my laptop to do more debugging on this.
Sorry I didn't have the time to do this. This is not really my area of
expertise though.
Ben.
--
Ben Hutchings
Humour is the best antidote to reality.
Attachment:
signature.asc
Description: This is a digitally signed message part