On 01/25/2013 12:57 AM, [email protected] wrote:
> Gilles Chanteperdrix <[email protected]> wrote on
> 01/24/2013 04:03:54 AM:
>
>> As said in the porting guide, it means that you have a timer issue. What
>> happens when you load a skin module is that Xenomai takes over the
>> control of the timer. So, that is where you should look. Do not assume
>> anything about the TWD code, it is not because it works on other
>> platforms that it works on this one.
>
> I modified the xttcpss code to support the ipipe_timer structure and got
> some interesting results:
>
> If I simply register this device with ipipe, there is largely no
> difference-- this is because the twd timer has a higher rating, and is
> registered by the time xenomai loads
>
> If I increase the rating of the xttcpss timer, the system is able to
> boot and run the xenomai sample applications
> CPU0 uses the xttcpss timer as it's timer
> CPU1 uses the twd timer as it's timer
>
> latency test is able to run on both cores (though CPU1 gets
> considerably better numbers-- likely because the twd timer runs at a
> much higher frequency)
I would say the difference is probably the timer programming latency.
>
> This seems to indicate there is some problem on this platform with the
> twd code (maybe the sharing of the twd interrupt between linux and
> xenomai?).
The twd interrupt is not shared between linux and xenomai, only xenomai
sees it, the linux timer interrupt then become a sort of soft interrupt
triggered by Xenomai.
>
> Of note-- when both cores are using the twd (the still broken case), if
> I turn on CONFIG_IPIPE_DEBUG_INTERNAL and enable the
> __ipipe_serial_debug message in twd_hrtimer_debug, I see far more
> interrupts on CPU1 than on CPU0. Is this to be expected?
>
> In general, do is there any advice for debugging where the twd_smp.c
> code is falling over on my platform?
No, unfortunately there is no magic recipe, you have to add traces
around to understand what happens.
--
Gilles.
_______________________________________________
Xenomai mailing list
[email protected]
http://www.xenomai.org/mailman/listinfo/xenomai
