Re: [RFC PATCHSET 0/6] timer/hrtimer/timer_stats: Improvement on timer_stats

[Yong Zhang]

On Sun, Aug 29, 2010 at 11:44 PM, Arjan van de Ven
<arjan@xxxxxxxxxxxxxxx> wrote:
> On 8/29/2010 8:05 AM, Yong Zhang wrote:
>>
>> Hi,
>>
>> This patchset focus on two problem:
>>
>> 1)timer_stats_timer_*_start_info() should belong to timer core only.
>> Â It should not be seen by outsider, such as workqueue. And should
>> Â not resident in timer.h.
>> Â Patch-0001~0003 first remove the only explicit out call, then move
>> Â timer_stats_timer_*_start_info() from timer.h to timer.c.
>>
>> 2)call site recorded by timer_stats sometimes show the timer core
>> functions
>> Â instead of the real call site.
>> Â Now /proc/timer_stats is like below on my machine:
>> ---now---
>>
>
>
> one question; have you tested this with powertop? (powertop being one of the
> prime consumers of this interface)
>

Now, vanilla kernel with powtop is
Top causes for wakeups:
  49.4% (250.3)       kworker/0:0 : hrtimer_start_range_ns (tick_sched_timer)
  49.4% (250.3)     <kernel core> : hrtimer_start_range_ns (tick_sched_timer)
   0.4% (  2.0)     <kernel core> : clocksource_watchdog (clocksource_watchdog)
   0.4% (  1.8)       <interrupt> : eth1
   0.2% (  1.0)       kworker/0:0 : add_timer (tg3_timer)
   0.1% (  0.3)     <kernel core> : dev_watchdog (dev_watchdog)
   0.0% (  0.2)          sendmail : hrtimer_start_range_ns (hrtimer_wakeup)
   0.0% (  0.2)              init : hrtimer_start_range_ns (hrtimer_wakeup)
   0.0% (  0.1)     <kernel core> : bdi_arm_supers_timer (sync_supers_timer_fn)
   0.0% (  0.1)        flush-0:13 : bdi_writeback_thread (process_timeout)
   0.0% (  0.1)     <kernel core> : __enqueue_rt_entity (sched_rt_period_timer)

And the patched kernel:
Top causes for wakeups:
  34.9% (  2.0)       kworker/0:0 : clocksource_watchdog (clocksource_watchdog)
  29.1% (  1.7)       <interrupt> : eth1
  17.4% (  1.0)       kworker/0:0 : tg3_timer (tg3_timer)
   3.5% (  0.2)          sendmail : schedule_hrtimeout_range_clock
(hrtimer_wakeup)
   3.5% (  0.2)     <kernel core> : dev_watchdog (dev_watchdog)
   3.5% (  0.2)              init : schedule_hrtimeout_range_clock
(hrtimer_wakeup)
   2.3% (  0.1)     <kernel core> : bdi_arm_supers_timer (sync_supers_timer_fn)
   2.3% (  0.1)        flush-0:13 : bdi_writeback_thread (process_timeout)
   1.2% (  0.1)     <kernel core> : __enqueue_rt_entity (sched_rt_period_timer)
   1.2% (  0.1)           syslogd : do_setitimer (it_real_fn)

The difference is tick_sched_timer() is gone in the patches version.

And after taking a look at powertop, I think you want to get rid of the tick
events, right?  Or maybe I'm missing something?

Thanks,
Yong
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/