Re: system keeps freezing once every 24 hours / random apps crashing

[Mark v Wolher <trilight@ns666.com>]

Lee Revell wrote:
> On Fri, 2005-12-30 at 23:05 +0100, Mark v Wolher wrote:
> 
>>Lee Revell wrote:
>>
>>>On Fri, 2005-12-30 at 22:57 +0100, Mark v Wolher wrote:
>>>
>>>
>>>>Lee Revell wrote:
>>>>
>>>>
>>>>>On Fri, 2005-12-30 at 22:47 +0100, Mark v Wolher wrote:
>>>>>
>>>>>
>>>>>
>>>>>>Lee Revell wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>>On Fri, 2005-12-30 at 22:30 +0100, Mark v Wolher wrote:
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>Mark v Wolher wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>>Mark v Wolher wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>>Folkert van Heusden wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>>>Hmm, i disabled MSI in the kernel, irq-balancing is on in the kernel,
>>>>>>>>>>>>and after a restart with irqbalance i see the cpu's show numbers !
>>>>>>>>>>>>I guess MSI was preventing them ? But does that means because of MSI
>>>>>>>>>>>>that performance was lower in some way ?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>did you also restart with only irqbalance activated?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>Folkert van Heusden
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>Yes, when MSI was disabled i had irq-balancing in the kernel on, i
>>>>>>>>>>rebooted without the irqbalance daemon and it showed no reaction on the
>>>>>>>>>>cpu's. When i enabled the irqbalance daemon then i got finally reaction
>>>>>>>>>
>>>>>>>>>>from the cpu's.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>>I'm also curious if this will solve those random freezes...which somehow
>>>>>>>>>>i suspect have to do with the tvcard and maybe having MSI on.
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>:( just got a total freeze, number 2 today. This time i noticed the
>>>>>>>>>mouse started to go very slow and 2 seconds later all was frozen.
>>>>>>>>>
>>>>>>>>>Maybe it's because of vmware ... i will not use vmware and see how it goes.
>>>>>>>>>-
>>>>>>>>
>>>>>>>>
>>>>>>>>Some new info, i just noticed this in the logs:
>>>>>>>>
>>>>>>>>
>>>>>>>>Dec 30 22:21:24 localhost kernel: bttv0: OCERR @ 1fde0000,bits: HSYNC
>>>>>>>>OFLOW FBUS OCERR*
>>>>>>>>Dec 30 22:21:24 localhost last message repeated 5 times
>>>>>>>>Dec 30 22:21:24 localhost kernel: bttv0: timeout: drop=0
>>>>>>>>irq=41296/41296, risc=1fde001c, bits: HSYNC OFLOW
>>>>>>>>Dec 30 22:21:24 localhost kernel: bttv0: reset, reinitialize
>>>>>>>>Dec 30 22:21:24 localhost kernel: bttv0: PLL: 28636363 => 35468950 . ok
>>>>>>>>
>>>>>>>>
>>>>>>>>But vmware is not active at this moment, i'll not use vmware and see if
>>>>>>>>a freeze occurs, i'll test up to 24 hours.
>>>>>>>>
>>>>>>>>I can swear it might have to do with the tvcard with tv on and vmware at
>>>>>>>>the same time also active. Or maybe just 1 of them. I'm even considering
>>>>>>>>to buy tomorrow a new tvcard and see if it makes any difference.
>>>>>>>
>>>>>>>
>>>>>>>It does not matter whether VMWare is active at the moment.  Any bug
>>>>>>>report where a binary module has been loaded, active or not, is tainted.
>>>>>>>
>>>>>>>Can you reproduce with a 100% clean kernel?
>>>>>>>
>>>>>>>Lee
>>>>>>>
>>>>>>
>>>>>>Hi Lee,
>>>>>>
>>>>>>But unloading all vmware modules should be good enough ? And how about
>>>>>>the binary module of nvidia ? It's active ofcourse else i'd not be able
>>>>>>to use all the features i normally use.
>>>>>>
>>>>>>Eitherway, several kernels back also made no difference for this issue.
>>>>>>
>>>>>>So right now, i'm leaning on the experience i have with working with
>>>>>>this system and trying to isolate things i suspect, before i go to more
>>>>>>radical steps :)
>>>>>>
>>>>>>And again, under windows 2k server, xp pro, redhat enterprise and
>>>>>>freebsd i never had these issues.
>>>>>>
>>>>>
>>>>>
>>>>>No, it's not good enough to unload them and no, the nvidia module
>>>>>absolutely cannot have been loaded.  Binary modules can do ANYTHING and
>>>>>we have NO IDEA what they are up to, how could you possibly think such a
>>>>>system would be debuggable?
>>>>>
>>>>>Please, search the LKML archives, this has come up again and again and
>>>>>again and still people post bug reports with binary only modules and
>>>>>expect them to be debuggable.
>>>>>
>>>>>Lee
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>Well Lee, you might be right indeed, but i see only some one who has
>>>>something against binary modules and assuming they're always to blame.
>>>>I do not agree with this and i will proceed from simple to advanced in
>>>>solving this problem, not start thinking difficult while the problem
>>>>might be solved by a simple move. :)
>>>>
>>>
>>>
>>>No I'm not assuming they are to blame, the point is that without the
>>>source code we can't prove they aren't.
>>>
>>>What if the vmware or nvidia module did something wrong that just
>>>happened to work by chance in older kernels?  We have no way to prove
>>>that this is not happening.
>>>
>>>Lee
>>>
>>>
>>>
>>
>>Yes, "what if" .. that means anything could be the cause, even a
>>cockroach in the box causing circuit to fail when it moves hehe
>>So don't worry, first i'll do these very simple tests and see what
>>effect they have and later go to more radical moves and strip all binary
>>modules and so on. This is only the initial point where i start to look
>>for common problems and people can give their advise in general.
> 
> 
> Removing binary modules is not a radical step - it's the first thing you
> should try if you have a problem with the kernel.
> 
> Basically you are asking for help with an unsupported configuration.  In
> general people on LKML will be more helpful if you take the time to find
> out what the bug reporting guidelines are before posting.
> 
> Lee
> 
> 
> 


Thank you for your input, but sometimes thinking out of the box gives a
solution instead of hiding behind "guidelines".