Re: [Xenomai-help] Segmentation Fault when writing in registers (AT91SAM - ARM9)

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

at91_enthus wrote:
> On Mon, Jan 24, 2011 at 10:28 PM, Gilles Chanteperdrix <
> gilles.chanteperdrix@xenomai.org> wrote:
> 
>> at91_enthus wrote:
>>> On Mon, Jan 24, 2011 at 10:16 PM, Gilles Chanteperdrix <
>>> gilles.chanteperdrix@xenomai.org> wrote:
>>>
>>>> at91_enthus wrote:
>>>>> On Mon, Jan 24, 2011 at 9:25 PM, Gilles Chanteperdrix <
>>>>> gilles.chanteperdrix@xenomai.org> wrote:
>>>>>
>>>>>> at91_enthus wrote:
>>>>>>> On Mon, Jan 24, 2011 at 8:41 PM, Gilles Chanteperdrix <
>>>>>>> gilles.chanteperdrix@xenomai.org> wrote:
>>>>>>>
>>>>>>>> at91_enthus wrote:
>>>>>>>>> Hi.
>>>>>>>>>
>>>>>>>>> I  was testing Xenomai on my AT91SAM9G20-based board and tried to
>>>> adapt
>>>>>>>> the
>>>>>>>>> code found here:
>>>>>>>>> http://www.armadeus.com/wiki/index.php?title=Xenomai:Blinking_LEDs.
>>>>>>>> By the way, this code is ridiculous. It creates a real-time task to
>>>>>>>> write to some non real-time device, which makes the real-time task
>>>>>> useless.
>>>>>>> I know.
>>>>>> Yes, but this example is bad in many ways:
>>>>>> - first it uses xenomai api to create a real-time task, and uses write
>>>>>> which causes the real-time task to switch to secondary domain;
>>>>>> - the Makefile does not use xeno-config, which have been the
>> recommended
>>>>>> way of getting the compiler flags since day 1;
>>>>>> - it nullifies SIGTERM and SIGINT, which is a dubious decision, and
>>>>>> could lead people to believe that they should do the same when using
>>>>>> Xenomai, whereas there is no such requirement.
>>>>>> - the pointers point to outdated version of the documentation.
>>>>>> - it uses runinfo/xeno-load, which are kind of outdated too, make
>> things
>>>>>> uselessly complicated, and in any case, tell you to type Ctrl-C to
>>>>>> interrupt the program, which is bound to fail if anything goes wrong,
>>>>>> since SIGINT is neutralized in the program.
>>>>>>
>>>>>>> That's why I went to mmap().
>>>>>> That is not really the right solution either. I admit that for driving
>>>>>> leds, it may be a bit overkill, but anyway, what you should do is
>> write
>>>>>> an RTDM driver for driving the leds. Keeping the driver separated from
>>>>>> the application will allow the same program to run on other platforms
>>>>>> where driving the LEDS is done differently without changing much of
>> the
>>>>>> application code. In other word, the application/driver separation is
>> a
>>>>>> sane one.
>>>>>>
>>>>>> Anyway, mmap should work.
>>>>>>
>>>>>>  (I chose that sample code because, for
>>>>>>> now, I am only learning to use basic functions)
>>>>>>>> Wrong test here. Should be "if (aic_base == MAP_FAILED)"
>>>>>>> Indeed, my program fails miserably when I check the mapping of AIC,
>> not
>>>>>>> PIOB's. And yet, I just can't figure why in the non-Xenomai code, AIC
>>>>>>> mmap-ing works  fine.
>>>>>> Well errno might give you a clue. Do you have FCSE enabled?
>>>>>>
>>>>>> Regards.
>>>>>>
>>>>>> --
>>>>>>
>> Gilles.
>>>>> FCSE is enabled. errno returns  22 (invalid argument).
>>>> best effort or guaranteed?
>>>
>>> Best effort
>>>
>>>
>>>> Please activate FCSE messages.
>>>
>>> ... compiling new kernel
>>>
>>>
>>>> How big is
>>>> MAP_SIZE?
>>>>
>>>>
>>> 4096
>> Ok. Could you post your full kernel configuration, as well as the full
>> program?
>>
> 
> I put the .config and the code in the attachments.

If you remove all the native skin stuff, only keep main and
open_controller, do not link with the native library, you can confirm
that this same program runs without trouble?

-- 
                                                                Gilles.