Re: [Xenomai-help] latency in kernel mode

[Wolfgang Grandegger <wg@domain.hid>]

Joachim Meyer wrote:
> Sorry, for my late answer.
> 
> I enabled the timerbench module, and installed it with the other modules (IRQ Benchmark and Context Switch).
> The Switchtest test works, so I think the modules are correct installed and the tests are able to load them.
> Does this mean the timerbench module isn't ok? Or are there other possible reasons, why the latency test still says:
> 
> bash-3.00# ./run -- -t 1
> *
> *
> * Type ^C to stop this application.
> *
> *
> == Sampling period: 100 us
> == Test mode: in-kernel periodic task
> == All results in microseconds
> latency: failed to start in-kernel timer benchmark, code -25
> ---|------------|------------|------------|--------|-------------------------
> RTS|  267928.028|  805352.732|      32.768|       0|    00:00:07/00:00:07
> bash-3.00#
> 

What does the following command show:

   # cat /proc/xenomai/rtdm/rttest*/information

You may need to use the option "-D 1".

> I just want to add that I'm using a ppc405 on an Xilinx Virtex2pro (XUP Board), Linux v2.6.23xlnx, a rootFS from eldk4.1 and Xenomai 2.4.2.
> 
> Are there any other importent options in compiling the kernel. I thought perhaps there are good default values, which will do it for me in the beginning. 
> My Real-time sub-system config: 
> [*] Xenomai
>  <*> Nucleus
>  [*] Pervasive real-time support in user-space
>  [ ] Interuppt shield support
>  [*] Priority coupling support
>  [*] optimize as pipeline head
>  (32) number of pipe devices
>  (512) Number of registry slots
>  (128) Size of the system heap (Kb)
>  [*] Statistics collection
>  [ ] Debug Support
>  [ ] Shared interuppts
>  Timing --> [ ] Enable periodic timing
>                   (0) Scheduling latency (ns)
>  Scalability --> [ ] 0(1) scheduler
>                        Timer indexing method (Linear) --> (X) Linear
>                                                                                ( ) Tree
>                                                                                ( ) Hash
> Machine --> No options!
> Interfaces --> Not in the mood to print all of this, but if it would help, i 'would do it.
> Drivers --> only the three testing drivers as modules

The default values are good to start with.

> Another Question I have is about the Output of my switchbench test:
> 
> bash-3.00# ./run -- -n 5000
> *
> *
> * Type ^C to stop this application.
> *
> *
> == Sampling period: 100 us
> == Do not interrupt this program
> RTH|       lat min|     lat avg|     lat max|        lost
> RTD|      -29560|        5810|       49723|        4352
> bash-3.00#
> 
> Is it normal that the lost number is that high? If not, what can i do?

A period of 100us is tough for your system. What worst case latencies do 
you measure with the latency program under load. It will be more than 
100us, I guess.

Wolfgang.