(2.0) <kernel core> : clocksource_register (clocksource_watchdog)

[Alan Jenkins]

Context: I sent a message to the powertop discussion list a while back 
asking about the above output line.  I was hoping for a response since 
there'd just been an announcement on the list from your address, and I 
thought you'd be interested in - what seemed to me - unnecessary wakeups 
caused by clocksource code.

Alan Jenkins wrote:
> I'm now on linux v2.6.23 -hrt1 (amd64).  I'm a bit puzzled because I'd 
> not seen the above before; I've looked it up and I think I know what it 
> does - sanity checking clocks, TSC especially.
>
> Can I configure it away?  (My TSC seems OK).
>   
[answer: not without hacking C or Kconfig files]
> Also, would it be possible to move it to a deferred timer like the 
> ondemand cpu frequency governor?  My vague understanding of this TSC 
> issue is that it's triggered by some software action: cpu frequency 
> control or suspend/resume.  I don't see why you'd need to wake up just 
> to check the TSC.
>   
I finally got round to trying this, since it was a single line change 
(init_timer() -> init_timer_deferrable()).  As expected it gets rid of 
the clocksource_watchdog wakeups, and it does actually reduce my overall 
wakeups.  Unfortunately my hardware doesn't have any TSC issues, so I 
can't back up my reasoning with test results.

Am I making sense?   Does this sound like a bad idea?

Alan