netbsd-bugs@netbsd.org
[Top] [All Lists]

bin/29685: refclock_palisade.c under NTP 4.2.0 from ntp.org is broken, f

Subject: bin/29685: refclock_palisade.c under NTP 4.2.0 from ntp.org is broken, fixed under NTP bug 250
From: Brian Buhrow
Date: Sun, 13 Mar 2005 04:57:00 +0000 UTC
        Note: There was a bad value `' for the field `Confidential'.
        It was set to the default value of `yes'.

>Number:         29685
>Category:       bin
>Synopsis:       refclock_palisade.c collects data from the GPS in 
>microseconds, rather than nanoseconds.
>Confidential:   yes
>Severity:       serious
>Priority:       medium
>Responsible:    bin-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Mar 13 04:57:00 +0000 2005
>Originator:     Brian Buhrow
>Release:        NetBSD 2.0 and later
>Organization:
        
>Environment:
        
        
System: NetBSD 2.0 NetBSD 2.0 (NFBNETBSD) #0: Tue Jan 4 10:03:02 PST 2005 
buhrow@xxxxxxxxxxxxxxxxxxxxx:/usr/src/sys/arch/i386/compile/NFBNETBSD i386

Architecture: i386
Machine: i386
>Description:
        
        While trying to setup our stratum 1 time server under NetBSD-2.0 using
a Trimble GPS time receiver, I found that I could not get accurate time
readings from the GPS receiver even though it is reputed to be an extremely
accurate device.  Searching to see if there were any known issues with the
Trimble Palisade receiver and NTP, I found the following data:

http://bugzilla.ntp.org/show_bug.cgi?id=250

    NTP: Bug 250 - Incorrect offset when using an Acutime 2000 / ntpd 4.2.0

   Skip Navigation Links

   Bugs Home | Documentation | New | Query | Reports | New Account |
   Log In

   Bugzilla Bug 250
     Incorrect offset when using an Acutime 2000 / ntpd 4.2.0
        Query page      Enter new bug
     _________________________________________________________________

   Bug#: 250   Platform: [PC.......]   Reporter:
   neil.burn@xxxxxxxxxxxxxxxxxx (Neil Burn)
   Product: [ntp.............]   OS: [Windows XP..]   Add CC:
   ______________________________
   Component: [refclock - palisade..]   Version: [4.2.0......]   CC:
   [_] bugs@xxxxxxx
   [_] Remove selected CCs
   Status: VERIFIED   Priority: [P3]
   Resolution: FIXED   Severity: [normal.....]
   Assigned To: stenn@xxxxxxx (Harlan Stenn)   Target Milestone: [4.2.1]

   URL: ____________________________________________________________
   Summary: ____________________________________________________________
   Attachment Type Modified Status                          Actions
   Create a New Attachment (proposed patch, testcase, etc.) View All
   Bug 250 depends on: ____________________ Show dependency tree
   Show dependency graph
   Bug 250 blocks: ____________________
     _________________________________________________________________

Using an Acutime 2000 connected to a Windows XP PC running ntp 4.2.0.  "ntpq -
c pe" reports that the offset tends to zero (as expected) but monitoring the
offset between the PC and the acutime using an oscilloscope shows that the
offset is anything up to a few hundred milliseconds out.  Running "ntpd -q -d"
repeatedly causes the clock to be stepped by a few hundred milliseconds every
time (rather than a step the first time followed by slews).

The problem seems to be caused by changes that have been made to
the "refclock_palisade.c" file between versions 4.1.1 and
4.2.0.  Specifically there has been a change to use nsecs instead of
usecs and line 497 of the new version does not take this into account.

To fix this I made the following change to "refclock_palisade.c":

497 <! pp->nsec = (long) (getdbl((u_char *) &mb(3)) * 1000000);
497 !> pp->nsec = (long) (getdbl((u_char *) &mb(3)) * 1000000000);

With this patch the offset as reported by ntpq -c pe and the oscilloscope both
tend to zero.

   ------- Additional Comment #1 From Harlan Stenn 2003-11-25 09:32
   ------- 
I've applied your patch to the bk repos.

Please mark this bug as VERIFIED or REOPENED, as appropriate.

Thanks!

   Additional Comments:

   (_) Leave as VERIFIED FIXED
   (_) Reopen bug
   (_) Mark bug as CLOSED
   Commit

   View Bug Activity   |   Format For Printing 
   Description: Opened: 2003-11-24 10:17
     _________________________________________________________________

        Query page      Enter new bug

   Enter a report # or some search terms:
   ____________________ Show [Help]

   This is Bugzilla: the Mozilla bug system. For more information about
   Bugzilla see bugzilla.org.
   New Account | Log In
>How-To-Repeat:

        To repeat this problem, install a stock NetBSD 2.0 system, or
NetBSD-current as of March 12 2005 or until this problem is resolved and
configure a Palisade GPS receiver on your system as follows:

ln -s /dev/tty00 /dev/palisade0


[/etc/ntp.conf example]
statsdir /var/ntpstats/     # directory for statistics files
filegen peerstats file peerstats type day enable
filegen loopstats file loopstats type day enable
filegen clockstats file clockstats type day enable
driftfile /etc/ntp.drift
pidfile /var/run/ntpd.pid
enable pps
#enable calibrate

server 127.127.29.0 prefer
#fudge 127.127.29.0 time1 0.020
server time.sdsc.edu
server time.nist.gov
server ntp-nasa.arc.nasa.gov

Enable  ntpd in /etc/rc.conf and start ntpd.
Watch the time jump all over the place, and watch the ntpd daemon jump from
server to server, trying to synchronize the clock.

        
>Fix:
        

        To fix, import the new version of ntp when it is released.  in the mean
time, the following patch works excellently on our stratum 1 server.


Index: refclock_palisade.c
===================================================================
RCS file: /cvsroot/src/dist/ntp/ntpd/refclock_palisade.c,v
retrieving revision 1.2
diff -u -r1.2 refclock_palisade.c
--- refclock_palisade.c 4 Dec 2003 16:23:37 -0000       1.2
+++ refclock_palisade.c 13 Mar 2005 04:36:25 -0000
@@ -496,7 +496,7 @@
                        return 0;
                }
 
-               pp->nsec = (long) (getdbl((u_char *) &mb(3)) * 1000000);
+               pp->nsec = (long) (getdbl((u_char *) &mb(3)) * 1000000000);
 
                if ((pp->day = day_of_year(&mb(14))) < 0) 
                        break;

>Unformatted:
        
        

<Prev in Thread] Current Thread [Next in Thread>
  • bin/29685: refclock_palisade.c under NTP 4.2.0 from ntp.org is broken, fixed under NTP bug 250, Brian Buhrow <=