Bug 231626

Is there more output from ntpd in /var/log/messages, when the daemon is running
for few hours?

Also, output from /usr/sbin/ntpq -pn would be helpful.

There is no other information in /var/log/messages.

[root@grant ~]# /usr/sbin/ntpq -pn
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
 193.45.254.143  212.94.162.33    3 u   30   64  377   89.962  4137499 316160.
 202.234.64.222  240.195.23.255   2 u   30   64  377  131.588  4137501 316611.
 217.172.188.162 193.120.142.71   2 u   18   64  377   72.697  4138816 317799.

I don't know what those numbers mean.

Since I filed the but it lost 12 hours.  I just reset it.

This really looks like a hw or kernel problem, probably lost interrupts.
Reassigning to kernel component.

Please attach the file /var/log/dmesg

Created attachment 149947 [details]
/var/log/dmesg

Further info:
The machine was fine with FC5.  When it was upgrades to FC6 it was fine for over
a month.  Autoupdate was turned off.  When I ran yum update is when it started
losing time.

Please test kernel 2.6.20-1.2925.fc6

Kernel updated, no change:

[root@grant ~]# /usr/sbin/ntpq -pn
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
 193.218.127.251 131.188.3.221    2 u   15   64  377  125.217  711734. 316200.
 193.125.143.140 192.36.143.150   2 u   12   64  377   74.017  715045. 316877.
 192.36.143.151  .PPS.            1 u    7   64  377   64.790  720534. 318072.
[root@grant ~]# uname -a
Linux grant.happykilroy.com 2.6.20-1.2925.fc6 #1 SMP Sat Mar 10 19:15:16 EST
2007 i686 i686 i386 GNU/Linux
[root@grant ~]# 

just a hunch, but can you try booting with hpet=off ?

default=0
timeout=5
splashimage=(hd0,0)/boot/grub/splash.xpm.gz
hiddenmenu
title Fedora Core (2.6.20-1.2925.fc6)
        root (hd0,0)
        kernel /boot/vmlinuz-2.6.20-1.2925.fc6 ro root=LABEL=/ rhgb quiet hpet=off
        initrd /boot/initrd-2.6.20-1.2925.fc6.img
title Fedora Core (2.6.19-1.2911.fc6)
        root (hd0,0)
        kernel /boot/vmlinuz-2.6.19-1.2911.fc6 ro root=LABEL=/ rhgb quiet
        initrd /boot/initrd-2.6.19-1.2911.fc6.img

I assume that is where the hpet=off goes?

[root@grant ~]# uname -a
Linux grant.happykilroy.com 2.6.20-1.2925.fc6 #1 SMP Sat Mar 10 19:15:16 EST
2007 i686 i686 i386 GNU/Linux

So I'm booting the kernel with that option.

Still a reference to hpet in dmesg:


Enabling APIC mode:  Flat.  Using 1 I/O APICs
ACPI: HPET id: 0x8086a201 base: 0xfed00000
Using ACPI (MADT) for SMP configuration information
Allocating PCI resources starting at c4000000 (gap: c0000000:30000000)
Detected 5864.670 MHz processor.
Built 1 zonelists.  Total pages: 776352
Kernel command line: ro root=LABEL=/ rhgb quiet hpet=off
mapped APIC to ffffd000 (fee00000)
mapped IOAPIC to ffffc000 (fec00000)


[root@grant ~]# ntpq -pn
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
 216.165.129.244 .GPS.            1 u   22   64    7    7.626  142967. 112662.
[root@grant ~]# 


Computer still can't tell time.  I seem to think it's the applet.  If I select
"adjust time" manually it updates it correctly.  Without doing that it leaks
time like a sieve.

Perhaps I'm wrong, I often am.

Does booting w/ "clocksource=acpi_pm" work around the issue?

Also do you know what was the last working version of the kernel you used?

IBM huh?  In this case that would be an abbreviation for "I bring mitigation." 
Removing the HPET entry and adding the acpi_pm fixed it.

As to Kernel I really can't tell exactly which it was.  I know that .19 and .20
were both bad.  I have yum retaining the cache and these are on the machine:

kernel-2.6.18-1.2849.fc6.i586.rpm
kernel-2.6.18-1.2849.fc6.i686.rpm
kernel-2.6.18-1.2869.fc6.i586.rpm
kernel-2.6.18-1.2869.fc6.i686.rpm
kernel-2.6.19-1.2895.fc6.i586.rpm
kernel-2.6.19-1.2911.6.4.fc6.i686.rpm
kernel-2.6.19-1.2911.fc6.i586.rpm
kernel-2.6.20-1.2925.fc6.i686.rpm

The i586 are "cruft" from the FC6 install problem I'd think.

.2911 and .2925 are the only two installed atm.  So 2869 or 2895 was the last
working kernel.  I tend to think it was the 2869 as there is also a PAE kernel
(2849) and I recall switching off that and it was still working for some time. 
So that would be 2869 working and 2895 being the most likely culprit.  2911 and
2925 are both known bad.

Ok. So it looks like your system is normally selecting the TSC for its
clocksource, however your TSC seems to be either changing frequency or halting
in C3 idle (which should normally disqualify the TSC and force the ACPI PM to be
used by default).

Quick test: does removing the "clocksource=acpi_pm" and adding "idle=poll" avoid
the issue?

Regarding the last kernel that functioned properly, 2.6.18 first introduced the
new timekeeping code, so if the problem was not in 2.6.18 but is in 2.6.19, that
means its some subtle change that I'm not aware of. Hmm.

Sorry for the delay.  I removed clocksource=acpi_pm and added idle=poll to grub.
 The time problem returned and it was immediate.  Lost 1/2 hour in the hour I
let it run.

Regards the bug, it's still there but using acpi_pm solves my issue.  Close the
bug?  Leave it open to track down the problem?  I'm willing to help of course.

Your call.

From the dmesg above:
   Detected 5863.607 MHz processor.

I doubt you have a ~6Ghz cpu. Sounds like the cpu_khz calibration is going
wonky, possibly due to the PIT mis-behaving.

Does booting w/ noapic (and not using the clocksource=acpi_pm option) change the
behavior?

Removed acpi_pm and added noapic.  Started bleeding time immediately.  

Apr  4 00:11:27 grant kernel: ACPI: Skipping IOAPIC probe due to 'noapic' option.
Apr  4 00:11:27 grant kernel: Kernel command line: ro root=LABEL=/ rhgb quiet noapic

I can attach the full dmesg if required.

Joseph: Does this timekeeping issue still exist with the current fedora kernel?

Good question.  I don't know.  The latest fedora kernel would be for Fedora 7. 
This machine, a bog standard Dell Dimension 5150, will not boot the Fedora 7
DVD.  I'll see if there is a respin.  I tried to install Fedora 7 right after
release but the machine would just hang on booting the DVD.  I suspect that has
something to do with the replacement of the IDE with SATA.  This machine has an
IDE DVD but SATA hard drives.

Fedora 6 and 7 have basically the same kernel, both are based on 2.6.22.9.

Created attachment 229311 [details]
DMesg

Dang it, adding the attachment ate my message...

Thank your for the kernel info.  I yummed it.

=====================

#boot=/dev/sda
default=0
timeout=5
splashimage=(hd0,0)/boot/grub/splash.xpm.gz
hiddenmenu
title Fedora Core (2.6.22.9-61.fc6)
        root (hd0,0)
        kernel /boot/vmlinuz-2.6.22.9-61.fc6 ro root=LABEL=/ rhgb quiet
        initrd /boot/initrd-2.6.22.9-61.fc6.img
title Fedora Core (2.6.20-1.2925.fc6)
        root (hd0,0)
        kernel /boot/vmlinuz-2.6.20-1.2925.fc6 ro root=LABEL=/ rhgb quiet
clocksource=acpi_pm
        initrd /boot/initrd-2.6.20-1.2925.fc6.img

===========================

[root@grant ~]# uname -a
Linux grant.happykilroy.com 2.6.22.9-61.fc6 #1 SMP Thu Sep 27 18:48:03 EDT 2007
i686 i686 i386 GNU/Linux

==========================

[root@grant ~]# uptime
 18:46:08 up 19 min,  2 users,  load average: 0.03, 0.10, 0.10

==========================

[root@grant ~]# uptime
 18:46:08 up 19 min,  2 users,  load average: 0.03, 0.10, 0.10
[root@grant ~]# /usr/sbin/ntpq -pn
     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
*216.165.129.244 .GPS.            1 u   14   64  377   17.495  243.945  60.926

=========================

The jitter is growning over time buy slowly.  The games play fine.  This kernel
seems to fix the problem.

Thanks.