Bug 246525

Summary: CPU loses cpufreq link after offline/online transition
Product: [openSUSE] openSUSE 10.2 Reporter: Bernhard Bender <bernhard.bender>
Component: KernelAssignee: E-mail List <kernel-maintainers>
Status: RESOLVED INVALID QA Contact: E-mail List <qa-bugs>
Severity: Normal    
Priority: P5 - None CC: rjwysocki, trenn
Version: Final   
Target Milestone: ---   
Hardware: i686   
OS: Other   
Whiteboard:
Found By: Customer Services Priority:
Business Priority: Blocker: ---
Marketing QA Status: --- IT Deployment: ---
Attachments: losing cpufreq link on CPU
kernel config for 2.6.20-20070215134239

Description Bernhard Bender 2007-02-16 22:34:57 UTC
This bug entry is created to separately collect information previously handled in bug 239101.

Taking a secondary CPU offline and online again makes this CPU lose its /sys/devices/system/cpu/cpu1/cpufreq link.

This will cause kpowersave to show the CPU as deactivated.

The same will happend when the machine goes into suspend2disk and resumes again.

Hardware: hp compaq nx6325
Kernel: 2.6.20 KOTD
ftp://ftp.suse.com/pub/projects/kernel/kotd/i386/HEAD/kernel-default-2.6.20-20070215134239.i586.rpm
Comment 1 Bernhard Bender 2007-02-16 22:39:30 UTC
Created attachment 119765 [details]
losing cpufreq link on CPU

The file shows the loss of the cpufreq link for CPU1 after the offline/online transition
Comment 2 Rafael Wysocki 2007-02-16 23:58:02 UTC
I use 2.6.20 on the same machine type (nx6325) with the same distribution and I do _not_ observe the symptoms described.  Thus there must be a difference between our configurations that leads to this problem.

Can you please attach your kernel config?
Comment 3 Bernhard Bender 2007-02-17 01:04:33 UTC
Created attachment 119787 [details]
kernel config for 2.6.20-20070215134239

This is the config file installed with the 2.6.20-20070215134239 kernel for KOTD
Comment 4 Thomas Renninger 2007-02-17 08:48:27 UTC
This is really weird, I also cannot reproduce this with the same machine.
Maybe:
  a) you modified BIOS configs (rather unlikely that it's related)
  b) we have different userspace application versions (maybe userspace rips out
     some more modules, aditionally writes some /sys, /proc entries, have you
     added some workaround in userspace for s2disk?)

I suggest to wait for the next 10.3 Alpha2 version. If there still is the problem, we must be able to reproduce this and can work on the exactly same code base (hmm or do we have different BIOS versions, mine is F.02).
Comment 5 Rafael Wysocki 2007-02-17 09:57:20 UTC
Referring to Comment #3:

Ah, I see.  Your kernel is 32-bit, while I use a 64-bit one.  I bet the Thomas' kernel is also 64-bit and that would explain the difference.

If that is the case, CPU hotplug vs cpufreq may be broken for all AMD64 machines with 32-bit kernels, but that would need to be confirmed.

Thomas, is your kernel 64-bit?
Comment 6 Thomas Renninger 2007-02-17 10:07:06 UTC
Good catch!
If onlining the cpu works in general, but only the cpufreq directory is missing I expect the culprit is in drivers/cpufreq/cpufreq.c or in the powernow-k8 driver itself (a guess). However, I won't have time to look into this the next days. If I don't hear anything I try to start digging in the middle/end of next week...
Comment 7 Rafael Wysocki 2007-02-17 10:23:03 UTC
Well, there is a known issue with the CPU hotplug locking that affects cpufreq, but I haven't seen any user-visible symptoms of it yet.  It may or may not be related.

Anyway, I think we can safely create a kernel bugzilla entry for this issue.  Bernhard, could you please do so?  I think you can say that the 2.6.20 (vanilla) kernel is affected.
Comment 8 Greg Kroah-Hartman 2007-02-18 01:08:00 UTC
ok, closing this for now, as this really isn't a problem, right?
Comment 9 Bernhard Bender 2007-02-18 13:55:08 UTC
Posted to kernel.org

http://bugzilla.kernel.org/show_bug.cgi?id=8033
Comment 10 Rafael Wysocki 2007-02-18 14:48:53 UTC
OK, thanks.