In message <x766sjwqgv.fsf@capsicum.wsrcc.com>Wolfgang Rupprecht writes
>


>I'm concerned that most of the advantage of using a GPS with a few
>10's of nS of noise on it PPS line will be lost if the kernel proceeds
>to timestamp things with only 1uS of resolution.

It still helps. Dave Mills has nearly a decade of experience with Suns
using microsecond-resolution clocks. Unfortunatley the MI timekeeping
in NetBSD still uses microseconds.  I've suggested changes to that,
but there was a lack of consensus about formats and inkernel APIs:
other people focusing on different uses of timing proposed rather
different APIs.


>Anyone know how the other open source folks do this?  Do they grab the
>higher frequency information from the CPU's tick-counter register?

FreeBSD uses PHK's "timecounter" abstraction.  Linux 2.2.x officially
doesn't do NTP; Ulrich Windl distributes NTP timekeeping as a set of
patches which add nanosecond resolution. (That's from memory; i
apologize in advance if it's wrong.

I also have a slightly more robust ppstest test and diagnosis app than
the one I've offered several folks.  I'll try and get that committed
into the tree. Output from that app would be one way I can get closer
to the loop (till I get a GPS clock).