Bug 552112

Summary: slow clock on Amilo Pro 2030 with NO_HZ=y
Product: [openSUSE] openSUSE 11.2 Reporter: Bernhard Wiedemann <novellbmw>
Component: KernelAssignee: Rafael Wysocki <rjw>
Status: RESOLVED WONTFIX QA Contact: E-mail List <qa-bugs>
Severity: Normal    
Priority: P3 - Medium CC: bwiedemann, jeffm
Version: RC 2   
Target Milestone: ---   
Hardware: Other   
OS: Other   
Whiteboard:
Found By: --- Services Priority:
Business Priority: Blocker: ---
Marketing QA Status: --- IT Deployment: ---

Description Bernhard Wiedemann 2009-11-03 10:33:24 UTC
User-Agent:       Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.8.1.23) Gecko/20090410 SUSE/1.1.18-0.1 SeaMonkey/1.1.18

reported upstream a while ago at http://bugzilla.kernel.org/show_bug.cgi?id=14280

hardware is detailed at
http://www.smolts.org/show?uuid=pub_c5345625-52df-4d29-9f31-acb6df810402

only nohz=off really helped. as TSC clocksource was also marked unstable with NOHZ

Reproducible: Always

Steps to Reproduce:
1. boot 2.6.31
2. sntp -v -P no -r pool.ntp.org

Actual Results:  
time drifts by 4% 

Expected Results:  
time should remain accurate as with nohz=off
Comment 1 Rafael Wysocki 2009-11-25 21:08:15 UTC
I'm afraid this issue will have to be resolved upstream first.
Comment 6 Rafael Wysocki 2010-09-21 23:16:20 UTC
Have you tried openSUSE 11.3?  If so, does it have that problem too?
Comment 7 Bernhard Wiedemann 2011-02-17 10:58:26 UTC
I went to using 11.4 milestones on this machine and at some point with nohz it started to not schedule any process anymore, unless I caused interrupts (i.e. by pressing keys). Luckily, nohz=off still makes it work.
Comment 8 Rafael Wysocki 2011-02-18 21:03:51 UTC
Let's use http://bugzilla.kernel.org/show_bug.cgi?id=14280 for further
tracking of this issue.
Comment 9 Rafael Wysocki 2011-03-12 21:40:12 UTC
It turns out that the hardware/BIOS of the machine in question is not
functioning as expected (specifically, the ACPI PM timer cannot cope with
NO_HZ, which it should).

hpet=force in the kernel command line is recommended as a permanent workaround.