Re: [RFC][patch 10/12] move NTP adjusted clock multiplier to struct timekeeper

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

On Fri, 2009-07-31 at 11:00 +0200, Martin Schwidefsky wrote:
> On Fri, 31 Jul 2009 01:12:59 -0700
> john stultz <johnstul@us.ibm.com> wrote:
> 
> > On Fri, 2009-07-31 at 09:52 +0200, Martin Schwidefsky wrote:
> > > On Thu, 30 Jul 2009 15:04:48 -0700
> > > john stultz <johnstul@us.ibm.com> wrote:
> > > 
> > > > On Wed, 2009-07-29 at 15:41 +0200, Martin Schwidefsky wrote:
> > > > > plain text document attachment (timekeeper-mult.diff)
> > > > > From: Martin Schwidefsky <schwidefsky@de.ibm.com>
> > > > > 
> > > > > The clocksource structure has two multipliers, the unmodified multiplier
> > > > > clock->mult_orig and the NTP corrected multiplier clock->mult. The NTP
> > > > > multiplier is misplaced in the struct clocksource, this is private
> > > > > information of the timekeeping code. Add the mult field to the struct
> > > > > timekeeper to contain the NTP corrected value, keep the unmodifed
> > > > > multiplier in clock->mult and remove clock->mult_orig. For consistency
> > > > > add the shift value associated with the NTP corrected mult value to
> > > > > struct timekeeper as well.
> > > > > 
> > > > > Cc: Ingo Molnar <mingo@elte.hu>
> > > > > Cc: Thomas Gleixner <tglx@linutronix.de>
> > > > > Cc: john stultz <johnstul@us.ibm.com>
> > > > > Cc: Daniel Walker <dwalker@fifo99.com>
> > > > > Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
> > > > > ---
> > > > >  arch/arm/plat-omap/common.c |    2 +-
> > > > >  include/linux/clocksource.h |    4 +---
> > > > >  kernel/time/timekeeping.c   |   43 +++++++++++++++++++++++++------------------
> > > > >  3 files changed, 27 insertions(+), 22 deletions(-)
> > > > > 
> > > > > Index: linux-2.6/include/linux/clocksource.h
> > > > > ===================================================================
> > > > > --- linux-2.6.orig/include/linux/clocksource.h
> > > > > +++ linux-2.6/include/linux/clocksource.h
> > > > > @@ -149,8 +149,7 @@ extern u64 timecounter_cyc2time(struct t
> > > > >   * @disable:		optional function to disable the clocksource
> > > > >   * @mask:		bitmask for two's complement
> > > > >   *			subtraction of non 64 bit counters
> > > > > - * @mult:		cycle to nanosecond multiplier (adjusted by NTP)
> > > > > - * @mult_orig:		cycle to nanosecond multiplier (unadjusted by NTP)
> > > > > + * @mult:		cycle to nanosecond multiplier
> > > > >   * @shift:		cycle to nanosecond divisor (power of two)
> > > > >   * @flags:		flags describing special properties
> > > > >   * @vread:		vsyscall based read
> > > > > @@ -168,7 +167,6 @@ struct clocksource {
> > > > >  	void (*disable)(struct clocksource *cs);
> > > > >  	cycle_t mask;
> > > > >  	u32 mult;
> > > > > -	u32 mult_orig;
> > > > >  	u32 shift;
> > > > >  	unsigned long flags;
> > > > >  	cycle_t (*vread)(void);
> > > > > Index: linux-2.6/kernel/time/timekeeping.c
> > > > > ===================================================================
> > > > > --- linux-2.6.orig/kernel/time/timekeeping.c
> > > > > +++ linux-2.6/kernel/time/timekeeping.c
> > > > > @@ -29,6 +29,8 @@ struct timekeeper {
> > > > >  	s64	ntp_error;
> > > > >  	int	xtime_shift;
> > > > >  	int	ntp_error_shift;
> > > > > +	u32	mult;			/* NTP adjusted clock multiplier */
> > > > > +	u32	shift;			/* NTP adjusted clock shift */
> > > > 
> > > > I'm a little confused here. NTP doesn't adjust the shift value. This
> > > > seems redundant with the xtime_shift value introduced in [patch 9/12].
> > > 
> > > True the NTP code does not adjust the shift value. But mult and shift
> > > build a pair, no? I think it is more readable if we use timekeeper.mult 
> > > and timekeeper.shift instead of timekeeper.mult and clock->shift.
> > > And the xtime_shift does not need to be the same value as the clock
> > > shift - at least in theory. It is just a shift to increase precision. 
> > > Therefore I would like to use the xtime_shift and ntp_error_shift values
> > > exclusively as the shifts for the respective xtime_nsec and ntp_error
> > > values. Makes sense ?
> > 
> > But can't we drop the xtime_shift if we have the clocksource shift kept
> > internally? Unless I'm missing something more subtle in your patches,
> > they should be the same at all times.
> 
> How about this definition for struct timekeeper?
> 
> /* Structure holding internal timekeeping values. */
> struct timekeeper {
>         /* Current clocksource used for timekeeping. */
>         struct clocksource *clock;
>         /* The shift value of the current clocksource. */
>         int     shift;
> 
>         /* Number of clock cycles in one NTP interval. */
>         cycle_t cycle_interval;
>         /* Number of clock shifted nano seconds in one NTP interval. */
>         u64     xtime_interval;
>         /* Raw nano seconds accumulated per NTP interval. */

I might just use the term "accumulation interval"  instead of NTP
interval. This isn't something that's strictly connected to NTP, other
then they both just need to have some chunk of time that we use as an
comparative interval.


>         u32     raw_interval;
> 
>         /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
>         u64     xtime_nsec;
>         /* Difference between accumulated time and NTP time in ntp
>          * shifted nano seconds. */
>         s64     ntp_error;
>         /* Shift conversion between clock shifted nano seconds and
>          * ntp shifted nano seconds. */
>         int     ntp_error_shift;
>         /* NTP adjusted clock multiplier */
>         u32     mult;
> };

Other then that it looks good to me!

thanks
-john