Hi Bruce,

On Mon, 2008-10-20 at 16:09 -0400, Bruce Taylor wrote:

Try a 2.4-kernel, see below.

LinuxPPS is very much work in progress. You are on the bleeding edge!
I am not sure the dust has settled enough for stable performance.

If your LAB have the option -- and some hair left -- try the latest 2.4
kernel with a working PPSKit-patch.

http://www.kernel.org/pub/linux/daemons/ntp/PPS/

Do not know if there are later ones than 2.1.7 somewhere else.

I have 2.4.3X-something kernel running on a lab laptop with a standard
NMEA serial stream with PPS on DCD, that converges down to sub 10us
offset within a minute or two. (There might be reasons not to trust the
NTP stats down at this level, but...)

Most serial ports I have tested will tolerate a standard PPS-pulse even
if its voltage levels are (way) noncompliant with RS232. YMMV!

To see if you get the pulses through the serial port, you might run
the radioclkd in test mode.

 http://www.buzzard.me.uk/jonathan/downloads/radioclk-1.0.tar.gz

That will show you both edges on signals attached to DCD, CTS or DSR.
This approach could also be used to time your signals from a user space
program.

Good luck!

Bruce, unfortunately Linux doesn't play too nicely with PPS -- long
story, but you need to do patches to get even minimal functionality with
current kernels.

An option is to use a shared memory driver called shmpps (available from
http://time.qnan.org/).  I've done this in the past, but had a little
trouble keeping the driver running reliably; it would sometimes die
silently, causing loss of the PPS source.  However, I think I was using
an earlier version than the one that's on that web page.

Using this approach will let you use the PPS signal, but won't give you
coarse time via the serial data stream.  To get that, you can use the HP
driver in ntpd which supports the HP Z3801A/58503A command set.  I think
that the Jackson Fury emulates these commands, though I haven't tried it
with this driver.

But if you can, a FreeBSD system makes a much better ntp server.  I am
a Linux guy, but my primary ntp boxes all are slightly modified Soekris
4501A single-board computers running a stripped down FreeBSD.  They run
rings around anything else.

Hope this helps a bit.

John

Bruce Taylor said the following on 10/20/2008 04:09 PM:

I have it working under FreeBSD w/ a custom ntp refclock.  It does
not work with the hpgps refclock.  PTIME:TCODE is not on-time.
The 1-PPS output is a couple 100ms wide, so there is no need for a
PPS stretcher.  I'm using a soekris net4501 w/ the tmrin mod
that is documented on febo.com.  Attached is a patch
which should work on both linux and freebsd, but for
kernel pps support, you will need to apply one of the various
kernel patches... Older 2.4kernels are better supported.  FreeBSD
has it natively.

Here's a sample ntp.conf

server 127.127.45.0 prefer minpoll 4 maxpoll 4 iburst
fudge 127.127.45.0 time1 0.012 #115200
fudge 127.127.45.0 flag3 1
server 127.127.22.0 minpoll 4 maxpoll 4
tos mindist 0.050

driftfile /media/log/ntpd.drift

statsdir /media/log/
statistics loopstats peerstats clockstats
filegen loopstats file loopstats type day enable
filegen peerstats file peerstats type day enable
filegen clockstats file clockstats type day enable

sysctl.conf

machdep.elan_gpio_config=-----P...--..--------..---------

also I have to modify the sio driver in the kernel:
/sys/dev/sio/sio.c

diff -u sio.c.orig sio.c
--- sio.c.orig  2008-06-27 15:11:52.000000000 -0500
+++ sio.c      2008-10-04 10:32:57.000000000 -0500
@@ -1928,7 +1928,7 @@
* (about 3 ticks if input flow control is not used or not honoured,
* but a bit less for CS5-CS7 modes).
*/

•   cp4ticks = speed / 10 / hz * 4;
  

•   cp4ticks = speed / 10 / hz * 40;
     for (ibufsize = 128; ibufsize < cp4ticks;)
             ibufsize <<= 1;
     if (ibufsize == com->ibufsize) {
  

  Scott

Bruce Taylor wrote: