Re: [Xenomai-help] xenomai 2.4.10 - invalid opcode: 0000

[Jan Kiszka <jan.kiszka@domain.hid>]

Petr Cervenka wrote:
> 
> Jan Kiszka wrote:
>> Petr Cervenka wrote:
>>> Jan Kiszka wrote:
>>>> Petr Cervenka wrote:
>>>>> Hello,
>>>>> I tried to upgrade linux kernel and xenomai and now I have problems
>>>>> with
>>>>> freezing computer because of "invalid opcode: 0000".
>>>>> It happens without any real-time app. running. Most often when I run X
>>>>> with KDE and netbeans (almost 100% probability), but it can freeze
>>>>> also
>>>>> during the boot.
>>>>> It perhaps depends on system load (CPU, disk, IO, ...?).
>>>>>
>>>>> My configuration:
>>>>> SW:
>>>>> linux kernel 2.6.30.9, arch: x86_64
>>>>> xenomai-2.4.10, smp
>>>>>
>>>>> also:
>>>>> rtnet-0.9.11 - no driver loaded
>>>>> peak-linux-driver-6.11 - no device connected
>>>>>
>>>>> HW:
>>>>> CPU: Intel Core2Duo  E7400
>>>>> chipset: Intel 3210
>>>>> system partition is on CF (compact flash) mounted in CF-SATA
>>>>> converter.
>>>>>
>>>>> Do you have any advices or tips? Thanks in advice.
>>>>> I can provide you by any needed information (.config of linux kernel,
>>>>> full systemlog, ...), just ask me.
>>>> Yes, please. I assume your system is otherwise stable (without
>>>> Xenomai/I-pipe enabled).
>>>>
>>>> It would also be nice if you could retry over 2.6.31.x with latest
>>>> ipipe
>>>> patch. And try to switch on CONFIG_IPIPE_DEBUG_TRACE_MCOUNT. That may
>>>> give a function back-trace on oops, maybe providing more pointers where
>>>> we come from (but it may also make the issue disappear).
>>>>
>>>> Jan
>>>>
>>>
>>> 1) When I disable Xenomai and I-pipe, the system is more or less
>>> stable. At least I was not able to produce the error (any error).
>>> I also ran memtest during the night without any error.
>>>
>>> 2) It is repeatable also in 2.6.31.7. with
>>> adeos-ipipe-2.6.31.1-x86-2.4-08.patch
>>>
>>> 3) Switching CONFIG_IPIPE_DEBUG_TRACE_MCOUNT on doesn't make the
>>> error to dissapear. So maybe the catched logs could provide you more
>>> info.
>>
>> Not yet. Is CONFIG_IPIPE_DEBUG_TRACE_ENABLE on? Also
>> CONFIG_FRAME_POINTER should be on. And please provide your .config for
>> reference.
> 
> Oh, my fault. I overlooked the meaning of that switch. In attached file
> _syslog3.txt are 3 freezes with manually enabled tracing (by echo 1 >
> /proc/ipipe/trace/enable) and in _syslog4.txt are 3 freezes with
> CONFIG_IPIPE_DEBUG_TRACE_ENABLE switched on, usually frozen during boot.
> The freezes are repeatable also on some other machine (Athlon X2 64 ?),
> but there it is little bit more stable.
> 
>>
>>>
>>> I realized that the error could be different, but there is almost
>>> every time "__ipipe_handle_irq" call in the call stack
>>>
>>> Also a very strange thing happened: When I tried to build "stable"
>>> kernel with disabled Xenomai and I-pipe, I was booted in the
>>> "unstable" kernel. It corrupted the .config file in some time between
>>> "make xconfig" and freeze during building. When I checked the .config
>>> file next time (because it failed to continue the compile) big part
>>> from the beginning was missing and some other binary data were
>>> appended to the end.
>>
>> I don't get this yet: It happens when you build/install a kernel while
>> running an i-pipe enabled kernel?
>>
> Yes, it happened when I tried to build new kernel while I was running
> I-pipe enabled kernel. But I got another error of this type:
> I backuped (I mean copied) some file (/boot/config-..) and the backup
> copy had totally different (some half binary data) content. The date and
> size were the same as the original. I realized that after couple of
> reboots when I tried to compare some other config file with the backup
> copy.

Reproducible with your .config under kvm, will investigate later.

Thanks,
Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux