Re: Bootup time regression from 2.6.27 to 2.6.28-rc3+

[Frans Pop <elendil@planet.nl>]

On Sunday 09 November 2008, Thomas Gleixner wrote:
> Find below the lineup of the timers-fixes-for-linus branch of the tip
> tree (the same as Arjan posted minus the irq fixes)

I've tested with Thomas' patch set on top of v2.6.28-rc3-261-g8b805ef.

The notebook boots normally, but spends way too much time in C0.

Cheers,
FJP


POWERTOP BEFORE
===============
This is somewhere around -rc3-200 with peek_ahead disabled with a return
in the function. System is on battery after having been suspended to RAM
a few times

Note the double C0 and missing P-states.

Cn                Avg residency       P-states (frequencies)
C0 (cpu running)        ( 3.2%)
C0                0.0ms ( 0.0%)
C1                0.0ms ( 0.0%)
C2                0.9ms ( 0.1%)
C4                3.4ms (96.7%)

Wakeups-from-idle per second : 289.1    interval: 10.0s
Power usage (ACPI estimate): 9.8W (4.5 hours)

Top causes for wakeups:
  42.8% (136.5)       <interrupt> : extra timer interrupt
  30.5% ( 97.2)       <interrupt> : PS/2 keyboard/mouse/touchpad
  12.4% ( 39.5)      <kernel IPI> : Rescheduling interrupts
   3.6% ( 11.6)       <interrupt> : iwlagn
   2.5% (  8.0)   <kernel module> : mod_timer (rh_timer_func)
   1.7% (  5.4)             artsd : schedule_hrtimeout_range (hrtimer_wakeup)
   1.1% (  3.6)     <kernel core> : hrtimer_start (tick_sched_timer)
   1.0% (  3.1)       <interrupt> : acpi
   0.6% (  2.0)            kicker : schedule_hrtimeout_range (hrtimer_wakeup)
   0.4% (  1.2)           klipper : schedule_hrtimeout_range (hrtimer_wakeup)
   0.3% (  1.1)          kdesktop : schedule_hrtimeout_range (hrtimer_wakeup)
   0.3% (  1.0)      kwifimanager : schedule_hrtimeout_range (hrtimer_wakeup)
   0.3% (  1.0)     <kernel core> : enqueue_task_rt (sched_rt_period_timer)
   0.3% (  1.0)             artsd : hrtimer_start (it_real_fn)
   0.3% (  1.0)          kwrapper : do_nanosleep (hrtimer_wakeup)
   0.3% (  1.0)              kwin : schedule_hrtimeout_range (hrtimer_wakeup)
   0.3% (  0.9)              Xorg : schedule_hrtimeout_range (hrtimer_wakeup)
   0.2% (  0.5)            iwlagn : mod_timer (ieee80211_sta_timer)
   0.2% (  0.5)         gpg-agent : schedule_hrtimeout_range (hrtimer_wakeup)
   0.2% (  0.5)   hald-addon-stor : schedule_hrtimeout_range (hrtimer_wakeup)


POWERTOP AFTER
==============
Again on battery; P-states are back and the second C0 is now "polling"
(both same as with 2.6.27.4).

System permanently remains very high in C0!
Wakeups go down to ~30 without user input (keyboard/mouse activity), which
is a bit lower than I get with .27.4.

Cn                Avg residency       P-states (frequencies)
C0 (cpu running)        (49.7%)         1333 Mhz     0.0%
polling           0.0ms ( 0.0%)         1200 Mhz     0.0%
C1                0.0ms ( 0.0%)         1067 Mhz     0.0%
C2                7.5ms ( 1.9%)          800 Mhz   100.0%
C4                8.5ms (48.4%)

Wakeups-from-idle per second : 59.6     interval: 15.0s
Power usage (ACPI estimate): 9.7W (4.4 hours)

Top causes for wakeups:
  34.6% ( 36.3)       <interrupt> : PS/2 keyboard/mouse/touchpad
  16.0% ( 16.7)      <kernel IPI> : Rescheduling interrupts
   8.9% (  9.3)       <interrupt> : iwlagn
   8.4% (  8.8)       <interrupt> : extra timer interrupt
   7.8% (  8.1)   <kernel module> : mod_timer (rh_timer_func)
   5.6% (  5.9)     <kernel core> : hrtimer_start (tick_sched_timer)
   5.2% (  5.4)             artsd : schedule_hrtimeout_range (hrtimer_wakeup)
   2.9% (  3.0)            kicker : schedule_hrtimeout_range (hrtimer_wakeup)
   1.1% (  1.2)          kdesktop : schedule_hrtimeout_range (hrtimer_wakeup)
   1.0% (  1.1)              Xorg : schedule_hrtimeout_range (hrtimer_wakeup)
   1.0% (  1.0)      kwifimanager : schedule_hrtimeout_range (hrtimer_wakeup)
   1.0% (  1.0)             artsd : hrtimer_start (it_real_fn)
   1.0% (  1.0)          kwrapper : do_nanosleep (hrtimer_wakeup)
   1.0% (  1.0)              kwin : schedule_hrtimeout_range (hrtimer_wakeup)
   1.0% (  1.0)           klipper : schedule_hrtimeout_range (hrtimer_wakeup)
   0.5% (  0.5)   hald-addon-stor : schedule_hrtimeout_range (hrtimer_wakeup)
   0.5% (  0.5)            iwlagn : mod_timer (ieee80211_sta_timer)
   0.5% (  0.5)     <kernel core> : __mod_timer (neigh_periodic_timer)
   0.4% (  0.5)         gpg-agent : schedule_hrtimeout_range (hrtimer_wakeup)
   0.4% (  0.4)           knotify : schedule_hrtimeout_range (hrtimer_wakeup)

In both cases there's no C3, but I think that's because C3 changes to C4
when on battery?

Cheers,
FJP