At Thu, 04 Feb 2010 16:31:54 +0200,
Alexe Fisher wrote:
>
> Am Donnerstag, den 04.02.2010, 12:30 +0100 schrieb Adam Sloboda:
> > Hi all,
> >
> > In the past week I've been searching for the source of "<kernel IPI> :
> > Rescheduling interrupts" -- I have more than 500 wakeups on idle
> > system (ThinkPad X200).
> >
> > Today I tried to set affinity of processes to one core and then
> > affinity of X process to the other one:
> >
> > for pid in `ps aux | awk '{print $2}'`; do taskset -c -p 0 $pid; done
> >
> > taskset -c -p 1 `pidof X`
> >
> > # You can restore it back (on dual-core):
> > #for pid in `ps aux | awk '{print $2}'`; do taskset -c -p 0,1 $pid; done
> >
> > First command reduced Rescheduling interrupts rapidly and the second
> > one caused it skyrocket back to hundreds. I noticed this with 2.6.30
> > kernel but have no idea when this started. I'm now on 2.6.32.7, I
> > tried minimal kernel config but no luck. I'm pretty sure this is not
> > happening on 2.6.26 distribution kernel (Debian squeeze).
> >
> > I'm also getting a lot of "extra timer interrupt" wakeups (over 100)
> > and a little less if all processes run on the same core. This might
> > be also connected with the issue. So I still have pretty poor
> > residency in sleep state but it's better.
> >
> > So it seems it's not any hardware driver but userspace process
> > interaction (or X processes). This is dead end for me. As the
> > result, I have 12-13 W power consumption on 12.1" centrino 2 laptop
> > (Vista can get it down to somewhere around 6 when idle and disk turned
> > off, wifi on).
> >
> > I will appreciate any ideas.
> >
> > Regards,
> > Adam
>
> I'm not 100% sure but your laptop get 8wh minimal. Please not post some
> thing like "Vista can get 6W" haw did you get this info?
This is what I get from that ThinkVantage tools. And it seems this
can be done on Linux too but I never got anywhere near those numbers:
http://itgen.blogspot.com/2009/03/energy-management-in-linux.html
>
> 4-cell battery, 29 Wh (14.4 V, 2.0 Ah). up to 3.3 hr
> 29 / 3.3 = 8,48W
>
> Wifi is about 1-2W, hdd 0.5W, 1-2W display.
Yeah, and I should be getting up to 6.5 hr with 6-cell. But 6 hours
is impossible, more like 3-4 max. I was getting that with previous
generation 15.4" laptop (patched 2.6.22 kernel).
>
> IPI bug is know. The count of IPI wakeups increase with count of CPU
> usage.
Where can I get more info about this problem? I wasn't able to find
anything relevant.
>
> here is my dump with used system (ubuntu 9.10) and kernel
> 2.6.32-04066-g2b876f9-dirty. Here you can see IPI is not the problem.
>
> Cn Avg residency
> C0 (cpu running) ( 4.4%)
> C0 0.0ms ( 0.0%)
> C1 mwait 0.0ms ( 0.0%)
> C2 mwait 0.3ms ( 0.1%)
> C4 mwait 3.6ms (95.5%)
> P-states (frequencies)
> 1.67 Ghz 1.3%
> 1333 Mhz 0.0%
> 1000 Mhz 98.7%
> Wakeups-from-idle per second : 269.4 interval: 15.0s
> Power usage (ACPI estimate): 10.2W (5.3 hours)
> Top causes for wakeups:
> 23.0% ( 49.2) <kernel core> : hrtimer_start_range_ns
> (tick_sched_timer)
> 17.4% ( 37.3) <interrupt> : extra timer interrupt
> 13.6% ( 29.1) firefox : hrtimer_start_range_ns
> (hrtimer_wakeup)
> 7.0% ( 15.0) <interrupt> : ehci_hcd:usb1, uhci_hcd:usb2
> 6.5% ( 13.9) skype : hrtimer_start_range_ns
> (hrtimer_wakeup)
> 6.5% ( 13.9) USB device 1-1 : HUAWEI Mobile (HUAWEI Technology)
> 5.8% ( 12.4) <kernel IPI> : Rescheduling interrupts
> 4.6% ( 9.9) desktopcouch-se : hrtimer_start_range_ns
> (hrtimer_wakeup)
> 2.8% ( 6.0) <kernel core> : hrtimer_start (tick_sched_timer)
>
>
I am sorry, are you trying to convince me that my problem is not real?
You can compare to my completely idle system (I even closed the
browser but with these residency values it doesn't really matter
anymore in terms of power saving):
Cn Avg residency
C0 (cpu running) ( 4.7%)
C0 0.1ms ( 0.0%)
C1 mwait 0.0ms ( 0.0%)
C2 mwait 0.1ms ( 0.5%)
C6 mwait 2.9ms (94.8%)
P-states (frequencies)
2.41 Ghz 2.3%
2.40 Ghz 0.4%
1.60 Ghz 0.9%
800 Mhz 96.5%
Wakeups-from-idle per second : 369.8 interval: 15.0s
Power usage (ACPI estimate): 12.2W (3.0 hours)
Top causes for wakeups:
73.0% (363.4) <kernel IPI> : Rescheduling interrupts
9.2% ( 45.7) <interrupt> : extra timer interrupt
6.1% ( 30.3) <interrupt> : iwlagn
4.0% ( 19.8) <interrupt> : acpi
2.1% ( 10.5) gkrellm : schedule_hrtimeout_range (hrtimer_wakeup)
2.0% ( 10.1) xfwm4 : schedule_hrtimeout_range (hrtimer_wakeup)
_______________________________________________
Power mailing list
[email protected]
http://www.bughost.org/mailman/listinfo/power
