Bug 426592

Summary: clocksource 'acpi_pm' looses time.
Product: [openSUSE] openSUSE 11.0 Reporter: Carlos Robinson <carlos.e.r>
Component: KernelAssignee: E-mail List <kernel-maintainers>
Status: RESOLVED WONTFIX QA Contact: E-mail List <qa-bugs>
Severity: Major    
Priority: P5 - None CC: jeffm
Version: Final   
Target Milestone: ---   
Hardware: i586   
OS: Other   
Whiteboard:
Found By: --- Services Priority:
Business Priority: Blocker: ---
Marketing QA Status: --- IT Deployment: ---

Description Carlos Robinson 2008-09-16 12:27:05 UTC 
My 11.0 system by default uses the 'acpi_pm' clocksource, and this one is
bad. It can run OK for hours. Then, I leave the PC alone for some hours,
come back, and the computer clock is off by 6 minutes from the wall clock
time! Yes, the ntp daemon was running but it couldn't cope. In fact, it
did not even complain:


13 Sep 21:16:51 ntpd[21753]: offset 0.007487 sec freq 14.405 ppm error 0.000662 poll 10
13 Sep 22:16:51 ntpd[21753]: offset 0.006803 sec freq 14.409 ppm error 0.000539 poll 10
13 Sep 23:12:36 ntpd[21753]: synchronized to 130.206.130.95, stratum 2
13 Sep 23:13:28 ntpd[21753]: synchronized to 212.101.3.211, stratum 2
13 Sep 23:14:30 ntpd[21753]: synchronized to 130.206.130.95, stratum 2
13 Sep 23:15:08 ntpd[21753]: synchronized to LOCAL(0), stratum 10
13 Sep 23:16:51 ntpd[21753]: offset 0.000000 sec freq 14.415 ppm error 0.000001 poll 10
13 Sep 23:30:10 ntpd[21753]: synchronized to 134.34.3.19, stratum 1
13 Sep 23:37:48 ntpd[21753]: ntpd exiting on signal 15


I restarted the daemon, it doesn't log the time error.


13 Sep 23:37:50 ntpd[12066]: system event 'event_restart' (0x01) status 'sync_alarm, sync_unspec, 1 event,
event_unspec' (0xc010)
13 Sep 23:37:50 ntpd[12066]: peer LOCAL(0) event 'event_reach' (0x84) status 'unreach, conf, 1 event, event_reach'
(0x8014)
13 Sep 23:37:50 ntpd[12066]: can't open /var/log/ntpstat/peerstats.20080913: Permission denied

I get many of those permissions denied, I have to check the config file
yet. That's not related to the current problem

13 Sep 23:37:59 ntpd[12066]: synchronized to 80.38.21.80, stratum 3
13 Sep 23:43:51 ntpd[12066]: time reset +351.529639 s



Solution (hack)?

In "/etc/init.d/boot.local" I force to another clocksource:

echo tsc > /sys/devices/system/clocksource/clocksource0/current_clocksource

This one remains stable.


This problem started with opensuse 10.3 (reported Bug 350981). 10.2 was correct. See there for the entire report (sp. where it is demonstrated it is software, not hardware error, 10.2 works fine). Ask for further tests or logs if necessary.
Comment 1 Greg Kroah-Hartman 2008-10-23 14:54:05 UTC 
So if you boot with "clock=tsc" added to the boot command line, everything works properly?
Comment 2 Carlos Robinson 2008-10-26 21:01:53 UTC 
Excuse the delay; as testing that requires a reboot, I used the chance to also download and install Beta 3. I've been testing during the weekend.

Yes, that parameter works; I tried, and system is using tsc:

nimrodel:~ # cat  /sys/devices/system/clocksource/clocksource0/current_clocksource
tsc

so it must be working, it is equivalent to the hack I used.

The kernel complains a bit:

<4>Warning! clock= boot option is deprecated. Use clocksource=xyz

so I just changed grub.



For your info, factory beta 3 selected the 'acpi_pm' clocksource, and at least during this time I saw no problem. I wasn't using ntpd. It needs a longer period of testing, but I can't do that now. I'll report on that when I install the final 11.1
Comment 3 Greg Kroah-Hartman 2009-01-23 04:38:18 UTC 
Please let us know how 11.1 works for you, or not.
Comment 4 Carlos Robinson 2009-01-23 13:30:29 UTC 
Sorry, I can't test: I'm waiting for "Bug #448007", machine crashes. Impossible to upgrade system to 11.1.
Comment 5 Carlos Robinson 2009-02-10 14:23:02 UTC 
Bug 448007 counts as fixed, but the fix means I have to use an experimental
kernel: kernel-debug-2.6.27.13-SL111_BRANCH_20090203134609_5784a3e1.i586.rpm

There is no info on whether this fix will be propagated to standard 11.1, which
means that, with the info I have, I'll have to skip upgrading to 11.1, even
though I have your complimentary DVD (and torch!) for my collaboration as
translator and tester.

So, if you want me to test this bug with the above experimental kernel (in my
test partition), I'll try. If not, we'll have to wait till 11.2.

And no, before you ask: I can not install factory, because that's the 11.1 test
partition.
Comment 6 Jeff Mahoney 2009-11-24 16:09:46 UTC 
NEEDINFO was never cleared on this bug so it didn't reappear on my RADAR. Since it's against 11.0, was fixed in 11.1, and now we've moved on to 11.2, I'm closing as WONTFIX.