Re: [Xenomai] Xenomai/cobalt: low_init(): binding failed: Function not implemented error issued when trting to run latency app on xeno 3.x-rc3

[Philippe Gerum <rpm@xenomai.org>]

On 03/04/2015 04:26 PM, Helder Daniel wrote:
> I reinstall xenomai enabling debug support:
> 
> /usr/xenomai/sbin/version -a
> Xenomai/mercury v3.0-rc3 -- 
> Target: i686-pc-linux-gnu
> Compiler: gcc version 4.7.2 (Debian 4.7.2-5) 
> Build args:  '--with-core=mercury' '--enable-registry'
> '--enable-pshared' '--prefix=/usr/xenomai' '--enable-debug'
>

Looks good.

> Then I removed all the code from the second process (binder.c), now
> main() its empty.
> 
> Running binder.c while creator.c is running, still gives a segmentation
> fault.
> 
> Since now there is no code in binder.c I compile it with no Xenomai
> support, just gcc binder.c -o binder.
> This way it runs (of course does nothing) without segmentation fault.
> 
> So it must be something in the initialization of the Xenomai app.

Ack.

> And only happens when another Xenomai app (creator.c) is running.
>

In shared mode, it seems.

> I did not backtrace it with gdbg yet, but I recompile it (main() still
> empty) with Xenomai support and make an strace -f (while creator.c is
> still running).

gdb will pinpoint the exact location of this fault.

> It seems that the segmentation fault happens when trying to lock memory
> pages in RAM (called by the initialization code):
> 
> [pid  3907] mlockall(MCL_CURRENT|MCL_FUTURE) = 0
> [pid  3907] --- SIGSEGV (Segmentation fault) @ 0 (0) ---
> [pid  3908] +++ killed by SIGSEGV +++
> [pid  3909] +++ killed by SIGSEGV +++

strace only reports system calls, with mlockall returning a success
status, so the segfault happens later.

I'll build your test case and dig this issue tomorrow. Thanks for your
patient reports, this will help fixing up the rough edges.

-- 
Philippe.