Re: [Xenomai-help] Multiple NIC trouble

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

[-- Attachment #1: Type: text/plain, Size: 2276 bytes --]

Eric Noulard wrote:
> 2007/2/9, Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>:
>> Maybe you should enable kernel debugging and xenomai debugging. If I
>> were you, I would reserve the console for error logging, I would not use
>> it to log in.
>>
>> From your logs, there is one thing you should try first: disable HPET
>> timer, since the box seems to have one and it is used by Linux, and
>> there is no support whatsoever of the HPET timer by Xenomai.
> 
> Bingo,
> 
> I disable HPET and xeno-test now runs just fine with no hang at all.
> 

Interesting.

>>>>>>>>>>
> == Sampling period: 100 us
> == Test mode: periodic user-mode task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (periodic user-mode task, 100 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat
> worst
> RTD|       3.346|       3.622|      13.130|       0|       3.346|     
> 13.130
> RTD|       3.352|       3.626|      12.365|       0|       3.346|     
> 13.130
> RTD|       3.352|       3.629|      12.533|       0|       3.346|     
> 13.130
> RTD|       3.342|       3.625|      11.598|       0|       3.342|     
> 13.130
>>>>>>>>>>>
> 
> Thanks to all of us for this bug hunt.
> 
> I've just added CONFIG_HPET and CONFIG_HPET_TIMER to
> "should be avoided?" kernel option in the Wiki FAQ:
> http://www.xenomai.org/index.php/FAQs#Installation_and_Setup
> 
> I don't know much about Kconfig syntax but I think it should be
> useful to add explicit warnings when HPET_xxx are enabled
> just as you do with APM, CPU_FREQ etc...
> 
> I located this in linux/init/Kconfig
> and would try something like:
> 
> comment "WARNING! You enabled HPET timer support "
>        depends on HPET || HPET_TIMER
> comment "option. These options are known to cause troubles with Xenomai."
>        depends on HPET || HPET_TIMER

ACK.

But can someone explain *why* there is a clash? Is it due to I-pipe
injecting tick events *aside* HPET-generated ticks? I think we should
address this clash when I-pipe has to be adopted to clock_event/dyntick
in 2.6.21.

> 
> I may try to re-enable MSI in order to see if its OK without HPET
> but with MSI.
> 
> Thanks again.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]