Re: [PATCH 2/5] introduce CLOCK_MONOTONIC_RAW

[john stultz <johnstul@us.ibm.com>]

On Mon, 2008-03-17 at 19:03 -0700, john stultz wrote:
> On Sat, 2008-03-15 at 05:50 +0100, Roman Zippel wrote:
> > > @@ -439,6 +475,7 @@ static void clocksource_adjust(s64 offset)
> > >  void update_wall_time(void)
> > >  {
> > >  	cycle_t offset;
> > > +	static u64 raw_snsec; /* shifted raw nanosecnds */
> > >  
> > >  	/* Make sure we're fully resumed: */
> > >  	if (unlikely(timekeeping_suspended))
> > 
> > IMO that's really a clock property, so this belongs in the clock 
> > structure.
> > (Some day we may want to have multiple active clocks for various purposes 
> > and thus export multiple raw clocks.)
> 
> I disagree. I think that crufts up the clocksource structure (which is
> ideally just a simple hw counter abstraction), with timekeeping state.

Bah. Ok, I've talked myself out of this one. 

I still think it crufts up the clocksource structure, but its more
consistent that we follow the established cruft (such as the
pre-calculated cycle_interval/xtime_interval/raw_interval combo) rather
then me trying to arbitrarily draw the line in the sand at this
variable.


> I'm still not sold on the multiple clocks with multiple notions of time
> idea you keep on bringing up. But if/when we cross that bridge, maybe it
> would be better to add a timekeeping_clock mid-layer abstraction that
> keeps the clocksource specific timekeeping state. That way we don't add
> lots of complexity for the clocksource driver writers to deal with and
> we allow the clocksources to be better re-purposed (for maybe more sane
> things like performance counters) without getting too bloated.

I still think pulling out all of the non-counter-abstraction bits out of
the clocksource and into a mid-level timekeeping_clock structure would
still be ideal here, but I'll save our time/energy on that one for
another day. :)

thanks
-john