Re: Oprofile not working on Xeon E5420 CPU in NMI mode - works in Timer mode.

[Maynard J. <may...@us...>]

Kristian Kostecky wrote:
> Hi Maynard,
> 
> I'm running 2.6.31.12 mainline with some patches to the sunrpc subsystem. I can't go higher than this because their was a major change to the kernel that completely rewrote the way that it writes dirty pages out to disk and it's buggy as hell [pdflush->flush-*:*]. After a few days of standard workload the flush-*:* processes consume a lot of CPU. Everything up to 2.6.36.2 is affected, so I'm avoiding it. In other words I've run the later kernels, but I have not tried running oprofile on them. Is there something I can do to troubleshoot this one as I can't upgrade anyhow? Any pointers would be appreciated.

Kris,
One thing you might consider doing is to rebuild the kernel and make oprofile a module.  Then you can add instrumentation to the module and easily build it and reload it without having to reboot.  Others (like John and Andi on cc) have Intel experience to help you with debugging the Intel oprofile kernel code.

Good luck.

-Maynard
> 
> Thanks again,
> Kris.
> 
> PGP Key: 4CC63A18
> PGP Server: pool.sks-keyservers.net
> 
> On 2010-12-17, at 4:48 PM, Maynard Johnson wrote:
> 
>> Kristian Kostecky wrote:
>>> Hi John,
>>>
>>> Thanks for responding. I've actually got it compiled statically in to the kernel (no module support), however, that's probably irrelevant.
>>>
>>> The relevant data is when cat'ing the special file is:
>>> i386/core_2
>>
>> Kris,
>> What kernel version are you running?  And is it a distro kernel or mainline? Have you tried a different (i.e., newer) kernel version?
>>
>> -Maynard
>>
>>>
>>> Kris.
>>>
>>> PGP Key: 4CC63A18
>>> PGP Server: pool.sks-keyservers.net
>>>
>>> On 2010-12-17, at 12:00 PM, John Villalovos wrote:
>>>
>>>> kr...@th... said the following on 12/17/10 08:03:
>>>>> Hi,
>>>>>
>>>>> I'd appreciate any help to get these Xenon machines working in NMI mode.
>>>>> Here is my configuration. I have this problem where 0 samples are
>>>>> generated in oprofile's default configuration. I have another machine with
>>>>> the same kernel that works in NMI mode with a slightly different model of
>>>>> Xeon chip, so I am familiar with some debugging and oprofile operation as
>>>>> I've used it successfully on this slightly older HP blade machine. The new
>>>>> one outlined below gets 0 NMIs and therefore 0 samples. Not sure why. If I
>>>>> pass the oprofile.timer=1 option on boot up (oprofile support is compiled
>>>>> in) I can get timer-mode profiling working. I'd like NMI mode to work on
>>>>> the new machine though. Thanks so much.
>>>>
>>>> What does /dev/oprofile/cpu_type display, once the oprofile kernel
>>>> module is loaded?
>>>>
>>>>
>>>>
>>>> -- 
>>>> John Villalovos
>>>> Intel Corporation on-site partner engineer at Red Hat, Inc.
>>>>
>>>> http://tabasco.usersys.redhat.com/
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> Lotusphere 2011
>>> Register now for Lotusphere 2011 and learn how
>>> to connect the dots, take your collaborative environment
>>> to the next level, and enter the era of Social Business.
>>> http://p.sf.net/sfu/lotusphere-d2d
>>> _______________________________________________
>>> oprofile-list mailing list
>>> opr...@li...
>>> https://lists.sourceforge.net/lists/listinfo/oprofile-list
>>>
>>
>