[pkg-ntp-maintainers] Bug#396937: Backgrounded ntpdate from ifup races with hwclock

[Andre Beck]

Re Kurt,

On Tue, Feb 27, 2007 at 07:40:18PM +0100, Kurt Roeckx wrote:
> 
> Do you still have a problem with this, or can we close the bug?

It depends.

When hwclock was moved to S8 (and later to S11) in rcS.d, the problem
disappeared for me. I wasn't aware for some time that this, though, also
relates to my kernel. I'm running a self compiled kernel that boots in
"old school" manner, in that it contains everything needed for the boot
built into the image and so no extra roundtrip through some initrd is
required.

For purposes of testing suspend to RAM/disk I've installed an additional
kernel, this time the unmodified distribution kernel (currently beeing
linux-image-2.6.18-4-686 2.6.18.dfsg.1-11). Now the interesting result is
that booting my initrd-free 2.6.17, I'm ending up with correct time, while
booting the distribution kernel (which includes an initrd) I end up one
hour into the future again. This is accompanied with the following log
messages:

 Setting the system clock..
 select() to /dev/rtc to wait for clock tick timed out
 System Clock set. Local time: Wed Feb 28 21:22:35 CET 2007.

Now that I found out what "iburst" really means I've got a solution that
works around the problem as long as I have network connectivity, but it's
still something broken here - though I don't think it's ntpdate or ntpd
any longer. It clearly seems to be a glitch in hwclock handling during
boot in conjunction with a distribution kernel. I just wonder why this
isn't happening everywhere else (running the distribution kernel should
be quite common). Others seem to have tripped about this before, see
for instance

 http://www.mail-archive.com/debian-devel@lists.debian.org/msg236510.html

I'll test to deinstall ntpdate just to make sure it isn't interfering,
though I dunno how it should, as what breaks is likely the distribution
kernel's RTC module and hwclock interaction with that.

I'd like to keep ntpdate on my system as a pure test tool (for checking
whether supposed NTP servers really do answer), but I don't need any
integration into ifupdown now that ntpd iburst does the same thing better.
IMHO it would be a nice idea to provide a NO-OP option for ntpdate-debian
in /etc/default/ntpdate (other than adding "exit 0" to the top, which works
but isn't wellformed ;) so network admins can leave it installed without
having it auto-integrate with ifupdown.

The bug, however, should probably be closed, as what happens seems no
longer related to ntpdate in any way.

Thanks,
Andre.
-- 
                  The _S_anta _C_laus _O_peration
  or "how to turn a complete illusion into a neverending money source"

-> Andre Beck    +++ ABP-RIPE +++    IBH Prof. Dr. Horn GmbH, Dresden <-