Re: [Xenomai-help] latency problems

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

Nadym Salem schrieb:
> On Mon, September 24, 2007 11:38, Jan Kiszka wrote:
>> Nadym Salem schrieb:
>>> On Wed, September 19, 2007 11:05, Jan Kiszka wrote:
>>>> Nadym Salem schrieb:
>>>>> On Thu, August 16, 2007 16:31, Gilles Chanteperdrix wrote:
>>>>>> On 8/16/07, Nadym Salem <Kaiman@domain.hid> wrote:
>>>>>>> On Thu, August 16, 2007 16:01, Gilles Chanteperdrix wrote:
>>>>>>>> Aha, this probably means that your chipset is not supporte, could
>>>>>>>> you
>>>>>>>> run lspci and send us the output ?
>>>>>>> it is attached.
>>>>>> Could you try to apply the attached patch to xenomai sources ? You
>>>>>> will then have to recompile your xenomai-patched kernel.
>>>>> Hmm, I have a new (quite old) machine now and installed Xenomai 2.3.3,
>>>>> having the same problem as before. The SMI workaround doesn't appear
>>>>> in
>>>>> the kernel log at all. Last time you gave me a "patch" for smi.c after
>>>>> which the system at least tried to start the smi workaround. Is there
>>>>> another possibility like that for my other machine ? I attached the
>>>>> output
>>>>> of lspci.
>>>>> tx in advance,
>>>>> greets, Nadym
>>>>> PS: Could it be, that this machine doesn't have SMIs at all ?
>>>> For sure, but there can also be other hardware-related latency killers.
>>>> So far the SMI workaround only deals with Intel chipsets (because there
>>>> are known knobs thanks to the chipset manual).
>>>> I can't join this party for the next days, but I would like to throw in
>>>> the question if you already tried to analyse the latency spot with the
>>>> I-pipe tracer (see Xenomai wiki). That may either underline that it is
>>>> outside the software's scope (arbitrary delays in unrelated kernel
>>>> functions) or point to a potential (but less likely) software issue.
>>> In advice of Gilles, I did a longterm test (from friday evening until
>>> this
>>> morning) on two machines. One being a slower, older one, the other being
>>> a
>>> core2duo. Both having the same latency problems. The Logfiles are
>>> attached.
>> Maybe due to my work e-mail interface, I'm having troubles opening those
>> tars. Do they now contain an I-pipe tracer output or only the testsuite
>> output like in the previous posting? In case of the latter please check
>> [1].
>>
>> Jan
>>
>> [1] http://www.xenomai.org/index.php/I-pipe:Tracer
> 
> 
> They are the same tests as before. What I was wondering about before is
> the documentation in the wiki.
> It's written, that the I-Pipe tracer as shown was activated by latency -f.

Did you enable the tracer in the kernel? That should improve things as 
well. :)

> The trace I did was also taken with latency -f... a mistake in the wiki ?
> 
> If necessary I can post the logs again, don't know what my tar did there..

Yes, please. Then including the properly configured output of 
/proc/ipipe/trace/frozen (again, check the wiki).

Jan

-- 
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux