NetBSD Problem Report #7602

Received: (qmail 5652 invoked by uid 1000); 18 May 1999 18:22:24 -0000
Message-Id: <19990518182224.5651.qmail@mail.netbsd.org>
Date: 18 May 1999 18:22:24 -0000
From: cgd@netbsd.org
Reply-To: cgd@netbsd.org
To: gnats-bugs@gnats.netbsd.org
Subject: 1.4 xntpd says "Previous time adjustment incomplete" on alpha
X-Send-Pr-Version: 3.92

>Number:         7602
>Category:       port-alpha
>Synopsis:       1.4 xntpd says "Previous time adjustment incomplete" on alpha
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    port-alpha-maintainer
>State:          closed
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue May 18 11:35:01 +0000 1999
>Closed-Date:    Thu Jan 24 00:38:48 +0000 2002
>Last-Modified:  Thu Jan 24 00:38:48 +0000 2002
>Originator:     Chris G. Demetriou
>Release:        
>Organization:
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.
>Environment:
System: NetBSD alpha 1.4 NetBSD 1.4 (GENERIC) #1: Sat May  8 11:37:01 PDT 1999     ross@netbsd.org:/usr/src/sys/arch/alpha/compile/GENERIC alpha

>Description:
	After upgrading one of my alphas from 1.3F to 1.4, xntpd started
	complaining about:

May 18 10:47:59 alpha xntpd[142]: 5.93c 3-NetBSD Wed Aug 12 10:05:25 EDT 1998 (1)
May 18 10:47:59 alpha xntpd[142]: tickadj = 40, tick = 976, tvu_maxslew = 40920, est. hz = 1024
May 18 10:47:59 alpha xntpd[142]: precision = 762 usec
May 18 10:52:49 alpha xntpd[142]: synchronized to 10.1.1.5, stratum=2
May 18 10:53:00 alpha xntpd[142]: time reset (step) 10.776500 s
May 18 10:53:00 alpha xntpd[142]: synchronisation lost
May 18 10:58:19 alpha xntpd[142]: synchronized to 10.1.1.5, stratum=2
May 18 10:58:24 alpha xntpd[142]: Previous time adjustment incomplete; residual -0.001640 sec
May 18 10:58:26 alpha xntpd[142]: Previous time adjustment incomplete; residual -0.001640 sec
May 18 10:58:28 alpha xntpd[142]: Previous time adjustment incomplete; residual -0.001560 sec
May 18 10:58:40 alpha last message repeated 6 times
May 18 10:58:42 alpha xntpd[142]: Previous time adjustment incomplete; residual -0.001600 sec
May 18 10:58:48 alpha last message repeated 3 times
May 18 10:58:50 alpha xntpd[142]: Previous time adjustment incomplete; residual -0.001560 sec
May 18 10:59:02 alpha last message repeated 6 times


	After those messages, it seems to have become quiet.

	My intuition says that while those messages aren't indicative of
	a fatal or "really really bad" condition, they indicate that adjtime()
	is doing _something_ that xntp doesn't expect.  I'm reporting this
	(it may be a non-bug) because i've never seen those messages before,
	on any platform, which makes me think something changed for
	the worse.

>How-To-Repeat:
	Run xntpd?  (It doesn't seem quite that easy, as the next
	time I booted the box, it did not report the same problem adjusting
	the time (though neither did it have to step the time at startup.)

>Fix:
	I think the right way to 'fix' this one is for somebody with an
	understanding of how adjtime is supposed to work to look through
	the clock code to make sure that the right things are happening on
	the alpha.
>Release-Note:
>Audit-Trail:

From: "Erik E. Fair" <fair@clock.org>
To: cgd@netbsd.org
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: port-alpha/7602: 1.4 xntpd says "Previous time adjustment
 incomplete" on alpha
Date: Wed, 19 May 1999 02:46:34 -0700

 The two most common problems which cause clock inaccuracy in computer
 systems are:

 1. computer manufacturers drive their interval interrupt timers with very
 cheap crystal oscillators that drift badly.

 2. systems programmers write interrupt handlers and other "critical
 sections" that disable the clock interrupt for "too long" (ideal would be
 "never").

 Presuming that your system hardware has not changed and this behavior is
 new, then we can probably point the finger at #2.

 It is time we all went on a hunt for code that can be taken out of such
 "critical sections" and thereby improve overall system interrupt response
 latency; NTP is merely the canary in the coal mine ("first to fall over
 when the atmosphere is less than perfect." - The Police).

 	Erik <fair@clock.org>

From: woods@most.weird.com (Greg A. Woods)
To: gnats-bugs@gnats.netbsd.org,
    netbsd-bugs@NetBSD.ORG (NetBSD Bugs and PR posting List) 
Cc:  Subject: Re: port-alpha/7602: 1.4 xntpd says "Previous time adjustment incomplete" on alpha
Date: Thu, 20 May 1999 12:32:27 -0400 (EDT)

 [ On , May 18, 1999 at 18:22:24 (-0000), cgd@netbsd.org wrote: ]
 > Subject: port-alpha/7602: 1.4 xntpd says "Previous time adjustment incomplete" on alpha
 >
 > 	After those messages, it seems to have become quiet.
 > 
 > 	My intuition says that while those messages aren't indicative of
 > 	a fatal or "really really bad" condition, they indicate that adjtime()
 > 	is doing _something_ that xntp doesn't expect.  I'm reporting this
 > 	(it may be a non-bug) because i've never seen those messages before,
 > 	on any platform, which makes me think something changed for
 > 	the worse.

 I've seen those messages on SunOS-4.1.x too (albiet with older versions
 of xntpd).

 I've found they are much more common on systems with conjested
 connections (eg. my 28.8kbps link, especially when I'm doing a long FTP
 or two).  I don't know if this is truly the cause, or not, but I thought
 I should report it as an observation.

 I've never seen those on my local clients that sync to my central NTP
 server over the LAN though....

 -- 
 							Greg A. Woods

 +1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
 Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>

From: Ross Harvey <ross@ghs.com>
To: gnats-bugs@gnats.netbsd.org, netbsd-bugs@netbsd.org, woods@most.weird.com
Cc:  Subject: Re: port-alpha/7602: 1.4 xntpd says "Previous time adjustment incomplete" on alpha
Date: Thu, 20 May 1999 09:47:03 -0700 (PDT)

 Regarding Eric's long discussion of clock accuracy and missed interrupts,
 I have to point out that the problem the PR refers to has absolutely nothing
 to do with clock accuracy of any kind. adjtime(2) just adds or subtracts
 a certain value over a certain number of ticks, and it's behavior for a given
 call would be the same if the clock was off by .00001% or 50%.

 Anyway, we found some MD problems in that system call...

 	Ross.Harvey@Computer.Org
State-Changed-From-To: open->feedback 
State-Changed-By: sommerfeld 
State-Changed-When: Sat Oct 9 12:42:50 PDT 1999 
State-Changed-Why:  
there was an obvious LP64 glitch in adj_systime() in libntp/systime.c, which  
i fixed; the result was that xntpd severely miscalculated the amount of  
slew to ask for when tweaking the local clock with adjtime. 

From: "Erik E. Fair" <fair@clock.org>
To: gnats-bugs@netbsd.org
Cc:  Subject: Re: port-alpha/7602
Date: Fri, 4 Feb 2000 16:18:54 -0800

 Were the MD fixes pulled up to the 1.4 branch? Because, if so, we 
 still have the problem - my PC164 does this, too.

 	Erik <fair@clock.org>
State-Changed-From-To: feedback->closed 
State-Changed-By: ross 
State-Changed-When: Wed Jan 23 16:36:57 PST 2002 
State-Changed-Why:  
originator approved close 
>Unformatted:

NetBSD Home
NetBSD PR Database Search

(Contact us) $NetBSD: query-full-pr,v 1.51 2026/08/10 02:28:17 riastradh Exp $
$NetBSD: gnats_config.sh,v 1.10 2026/05/13 22:00:09 riastradh Exp $
Copyright © 1994-2026 The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.