timekeeping: Use cached ntp_tick_length when accumulating error
authorJohn Stultz <john.stultz@linaro.org>
Thu, 24 Apr 2014 03:53:29 +0000 (20:53 -0700)
committerJohn Stultz <john.stultz@linaro.org>
Wed, 23 Jul 2014 22:01:57 +0000 (15:01 -0700)
commit375f45b5b53a91dfa8f0c11328e0e044f82acbed
tree1f91a054d833c05bd95db78dd81342a88cf16ea4
parentdc491596f6394382fbc74ad331156207d619fa0a
timekeeping: Use cached ntp_tick_length when accumulating error

By caching the ntp_tick_length() when we correct the frequency error,
and then using that cached value to accumulate error, we avoid large
initial errors when the tick length is changed.

This makes convergence happen much faster in the simulator, since the
initial error doesn't have to be slowly whittled away.

This initially seems like an accounting error, but Miroslav pointed out
that ntp_tick_length() can change mid-tick, so when we apply it in the
error accumulation, we are applying any recent change to the entire tick.

This approach chooses to apply changes in the ntp_tick_length() only to
the next tick, which allows us to calculate the freq correction before
using the new tick length, which avoids accummulating error.

Credit to Miroslav for pointing this out and providing the original patch
this functionality has been pulled out from, along with the rational.

Cc: Miroslav Lichvar <mlichvar@redhat.com>
Cc: Richard Cochran <richardcochran@gmail.com>
Cc: Prarit Bhargava <prarit@redhat.com>
Reported-by: Miroslav Lichvar <mlichvar@redhat.com>
Signed-off-by: John Stultz <john.stultz@linaro.org>
include/linux/timekeeper_internal.h
kernel/time/timekeeping.c