Bug 104915

Summary: kernel spams log with "acpi-cpufreq: Transition failed"
Product: [openSUSE] SUSE LINUX 10.0 Reporter: Marcus Meissner <meissner>
Component: KernelAssignee: Thomas Renninger <trenn>
Status: RESOLVED FIXED QA Contact: E-mail List <qa-bugs>
Severity: Critical    
Priority: P5 - None CC: trenn
Version: Beta 1   
Target Milestone: ---   
Hardware: All   
OS: All   
Whiteboard:
Found By: Development Services Priority:
Business Priority: Blocker: ---
Marketing QA Status: --- IT Deployment: ---

Description Marcus Meissner 2005-08-16 12:27:36 UTC 
my "dmesg" buffer is spammed with thousands of "acpi-cpufreq: Transition 
failed". 
 
This should be at least ratelimited, if not supressed. 
 
(laptop: fu-si lifebook)
Comment 1 Thomas Renninger 2005-08-16 12:35:54 UTC 
This could get an ugly one.
Can I access the machine?
Comment 2 Marcus Meissner 2005-08-16 14:05:12 UTC 
its my private laptop, so please do not destroy data on /home 
 
currently online as f70.suse.de, standard rootpassword. 
 
i can also bring it to you if you need it ;)
Comment 3 Thomas Renninger 2005-08-18 17:11:32 UTC 
I come back to you soon, sorry I don't know what to do first at the moment.
Comment 4 Marcus Meissner 2005-08-29 10:55:36 UTC 
this message needs at least be rate limited, or printed only _once_
Comment 6 Thomas Renninger 2005-09-05 17:25:24 UTC 
If transitions failed, the modules should just not be loaded IMO.
/etc/sysconfig/powersave/cpufreq
CPUFREQD_MODULE="off"
Comment 7 Marcus Meissner 2005-09-06 07:25:57 UTC 
after upgrading to your kernel the message is gone. 
 
How can I check that it switches freq correctly? 
 
/proc/cpuinfo shows 500mhz for dynamic and 800mhz for "Leistung" already.
Comment 8 Thomas Renninger 2005-09-06 10:49:44 UTC 
Great!
Yoy should always use dynamic mode and enable cpufreq scaling.
Give load to your processor:
glxgears
or
cat /dev/zero >/dev/null

The cpufreq should go up and down if you start/stop the above commands.
Does it work? Then I will suggest this patch (will go into 2.6.14) for inclusion.
Comment 9 Marcus Meissner 2005-09-06 11:16:24 UTC 
yes, it does go up and down.
Comment 10 Thomas Renninger 2005-09-06 11:37:21 UTC 
I am still a bit afraid about suppressing the message (Theoretically could still
apear on broken machines). I could:

only print it three times and modify the message and add something like:
better do not load acpi-cpufreq module (as performance could go down a bit by
unnecessarly trying to assign cpufreqs and wait for success for 1ms - there I
could not walk in anymore which means if a machine runs 10 days and more than 3
failed transtions occur, the functionality of cpufreq is lost -> I don't like to
do that).

Or after it works for you now, just let it as it is, people must find and set:
/etc/sysconfig/powersave/cpufreq
CPUFREQD_MODULE="off"

then. Is the last one OK for you Markus? Shall we still ask aj?

olh commited the patch -> closing.
Verifing whether it really works on RC would be nice.
Comment 11 Marcus Meissner 2005-09-06 11:40:15 UTC 
the last case is ok (just let it as it is right now).