Re: [Xenomai] __xn_sys_current_info failed: Operation not permitted

[Philippe Gerum <rpm@xenomai.org>]

On 07/19/2013 12:57 PM, Franz Engel wrote:
>
>
>     ------------------------------------------------------------------------
>     *Von:* Philippe Gerum <rpm@xenomai.org>
>     *An:* Franz Engel <franz_lambert_engel@yahoo.de>
>     *CC:* "xenomai@xenomai.org" <xenomai@xenomai.org>
>     *Gesendet:* 12:09 Freitag, 19.Juli 2013
>     *Betreff:* Re: [Xenomai] __xn_sys_current_info failed: Operation not
>     permitted
>
>     On 07/19/2013 11:49 AM, Franz Engel wrote:
>      > Philippe wrote:
>      >>> Franz Engel wrote:
>      >>> Hi,
>      >> >
>      >>> I get this error when I run a programm:
>      >>> __xn_sys_current_info failed: Operation not permitted
>      >> >
>      >>> I know that the error comes from the assert_nrt_inner(void)
>     function inside the assert_context.c file. But I have no idea what
>     is the reason. The Operation not permitted is the return value from
>     XENOMAI_SYSCALL1(__xn_sys_current_info, &info);. But I found no
>     documentation about this function.
>      >
>      >
>      >>This is reminiscent of a symptom observed for a bug fixed in the
>     -forge
>      >>tree, a while ago:
>      >>http://git.xenomai.org/?p=xenomai-forge.git;a=commit;h=014898add6a697e1a0d69f9e20b910ade74f1940
>      >
>      >>Do you happen to call fork() in your  application?
>      >
>      > Yes. I do? What I can do to solve the problem. Should I try to modify
>      > the xenomai files as discrept in the git-link?
>      >
>
>     I'm unsure, although this would be possible to backport portions of the
>     logic, the implementation has now diverged significantly between the
>     2.x
>     and 3.x code bases.
>
>     Could you try the patch below? It's untested, but should work around
>     the
>     root issue in your case:
>
>     diff --git a/include/nucleus/compiler.h b/include/nucleus/compiler.h
>     index 30f32c7..da4ea03 100644
>     --- a/include/nucleus/compiler.h
>     +++ b/include/nucleus/compiler.h
>     @@ -37,9 +37,11 @@
>
>        #ifdef __IN_XENO__
>        #if __GNUC__ == 4 && __GNUC_MINOR__ == 6
>     -#define __constructor__ __attribute__((constructor, noclone))
>     +#define __constructor__ __attribute__((constructor(200), noclone))
>     +#define __late_constructor__ __attribute__((constructor, noclone))
>        #else /* Gcc not 4.6 */
>     -#define __constructor__ __attribute__((constructor))
>     +#define __constructor__ __attribute__((constructor(200)))
>     +#define __late_constructor__ __attribute__((constructor))
>        #endif /* Gcc not 4.6 */
>        #endif /* In xenomai */
>
>     diff --git a/src/skins/common/rt_print.c b/src/skins/common/rt_print.c
>     index dfa8b33..e056cd6 100644
>     --- a/src/skins/common/rt_print.c
>     +++ b/src/skins/common/rt_print.c
>     @@ -738,7 +738,7 @@ static void forked_child_init(void)
>                spawn_printer_thread();
>     }
>
>     -static __constructor__ void __rt_print_init(void)
>     +static __late_constructor__ void __rt_print_init(void)
>        {
>            const char *value_str;
>            unsigned long long period;
>
>     --
>     Philippe.
>
>     I tried the patch. But I think it is not the correct version of
>     xenomai. In my compiler.h there is no line "#define __constructor__
>     __attribute__((constructor, noclone))". I use Xenomai 2.6.2.1.
>

Simply ignore the noclone spec when applying the patch manually. You 
should end up with the normal/high priority ctor at priority 200, the 
late one with an implicit, unspecified low priority.

-- 
Philippe.