[Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[alex alex]

Hi all,

The attached program shows an interraction bug between malloc and nanosleep.
The use of these two functions in a rt thread leads to a page fault.
The dmesg output is:

Xenomai: Switching test to secondary mode after exception #14 from
user-space at 0xb760393e (pid 30730)

I need to remove this switch because it prevents me to used the
rtdm_sem_down function ( which return -1) in my RTDM driver and I guess it
must also deteriorate the real time performance of my application.

Thanks for your help.

Alex.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Bug_switch.bz2
Type: application/x-bzip2
Size: 1208 bytes
Desc: not available
URL: <http://www.xenomai.org/pipermail/xenomai/attachments/20121218/66ca8915/attachment.bin>

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Gilles Chanteperdrix]

On 12/18/2012 03:35 PM, alex alex wrote:
> Hi all,
>
> The attached program shows an interraction bug between malloc and nanosleep.

You can not use malloc in an rt thread.

-- 
					    Gilles.

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Gilles Chanteperdrix]

On 12/18/2012 03:35 PM, alex alex wrote:
> Hi all,
>
> The attached program shows an interraction bug between malloc and nanosleep.
> The use of these two functions in a rt thread leads to a page fault.
> The dmesg output is:
>
> Xenomai: Switching test to secondary mode after exception #14 from
> user-space at 0xb760393e (pid 30730)
>
> I need to remove this switch because it prevents me to used the
> rtdm_sem_down function ( which return -1) in my RTDM driver and I guess it
> must also deteriorate the real time performance of my application.
>
> Thanks for your help.

If you are compiling with the Makefile in the attachment, your rt thread 
is not in fact using xenomai.

-- 
					    Gilles.

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Gilles Chanteperdrix]

On 12/18/2012 04:27 PM, Gilles Chanteperdrix wrote:
> On 12/18/2012 03:35 PM, alex alex wrote:
>> Hi all,
>>
>> The attached program shows an interraction bug between malloc and nanosleep.
>> The use of these two functions in a rt thread leads to a page fault.
>> The dmesg output is:
>>
>> Xenomai: Switching test to secondary mode after exception #14 from
>> user-space at 0xb760393e (pid 30730)
>>
>> I need to remove this switch because it prevents me to used the
>> rtdm_sem_down function ( which return -1) in my RTDM driver and I guess it
>> must also deteriorate the real time performance of my application.
>>
>> Thanks for your help.
>
> If you are compiling with the Makefile in the attachment, your rt thread
> is not in fact using xenomai.
>
Sorry, I mixed LDFLAGS and CFLAGS. Anyway, please try adding 
-Wl,--wrap,malloc to the LDFLAGS to see if you get the same issue.

Also, if there is really an issue, I would expect this issue to be 
dependent on the architecture and I-pipe patch used, so, please give us 
this information.

-- 
					    Gilles.

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[alex alex]

My LDFLAGS is :
-Wl,@/usr/lib/posix.wrappers -L/usr/lib -lpthread_rt -lxenomai -lpthread
-lrt

and I use :
xenomai-2.6.1/ksrc/arch/x86/patches/adeos-ipipe-2.6.38.8-x86-2.11-01.patch

I guess "--wrap,malloc" is similar to @/usr/lib/posix.wrappers.
Anyway, malloc, nanosleep and pthread_create are well wrapped.

nm -g test | grep wrap
         U __wrap_malloc
         U __wrap_nanosleep
         U __wrap_pthread_create
         U __wrap_pthread_setschedparam
         U __wrap_puts


Actually, I realize a porting of a robotics software platform to
xenomai and allocations
are made in an initial phase where the real-time performance is not
important,
however I need to open my RTDM driver with blocking open during this
phase,I usenativeskin
semaphore for it. This page fault causes the blocking malfunction of the
semaphores (rt_sem_p returns EPERM).

2012/12/18 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>

> On 12/18/2012 04:27 PM, Gilles Chanteperdrix wrote:
>
>> On 12/18/2012 03:35 PM, alex alex wrote:
>>
>>> Hi all,
>>>
>>> The attached program shows an interraction bug between malloc and
>>> nanosleep.
>>> The use of these two functions in a rt thread leads to a page fault.
>>> The dmesg output is:
>>>
>>> Xenomai: Switching test to secondary mode after exception #14 from
>>> user-space at 0xb760393e (pid 30730)
>>>
>>> I need to remove this switch because it prevents me to used the
>>> rtdm_sem_down function ( which return -1) in my RTDM driver and I guess
>>> it
>>> must also deteriorate the real time performance of my application.
>>>
>>> Thanks for your help.
>>>
>>
>> If you are compiling with the Makefile in the attachment, your rt thread
>> is not in fact using xenomai.
>>
>>  Sorry, I mixed LDFLAGS and CFLAGS. Anyway, please try adding
> -Wl,--wrap,malloc to the LDFLAGS to see if you get the same issue.
>
> Also, if there is really an issue, I would expect this issue to be
> dependent on the architecture and I-pipe patch used, so, please give us
> this information.
>
> --
>                                             Gilles.
>

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Gilles Chanteperdrix]

On 12/18/2012 05:16 PM, alex alex wrote:
> My LDFLAGS is :
> -Wl,@/usr/lib/posix.wrappers -L/usr/lib -lpthread_rt -lxenomai -lpthread
> -lrt
>
> and I use :
> xenomai-2.6.1/ksrc/arch/x86/patches/adeos-ipipe-2.6.38.8-x86-2.11-01.patch

This patch is known to be bogus, try:
http://download.gna.org/adeos/patches/v2.6/x86/adeos-ipipe-2.6.38.8-x86-2.11-03.patch

>
> however I need to open my RTDM driver with blocking open during this
> phase,I usenativeskin
> semaphore for it. This page fault causes the blocking malfunction of the
> semaphores (rt_sem_p returns EPERM).

Using native skin semaphore is deprecated. If writing an RTDM driver, 
why not using an RTDM semaphore?

-- 
					    Gilles.

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[alex alex]

In fact, this is what I intend to use when the bug will be solved. I use
native because RTDM semaphore freezes my PC while with native,
semaphore is just
no longer blocking (so native helped me to find out the bug and it keeps me
from restart the PC when it bugs... )

Thanks for the link.

2012/12/18 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>

> On 12/18/2012 05:16 PM, alex alex wrote:
>
>> My LDFLAGS is :
>> -Wl,@/usr/lib/posix.wrappers -L/usr/lib -lpthread_rt -lxenomai -lpthread
>> -lrt
>>
>> and I use :
>> xenomai-2.6.1/ksrc/arch/x86/**patches/adeos-ipipe-2.6.38.8-**
>> x86-2.11-01.patch
>>
>
> This patch is known to be bogus, try:
> http://download.gna.org/adeos/**patches/v2.6/x86/adeos-ipipe-**
> 2.6.38.8-x86-2.11-03.patch<http://download.gna.org/adeos/patches/v2.6/x86/adeos-ipipe-2.6.38.8-x86-2.11-03.patch>
>
>
>
>> however I need to open my RTDM driver with blocking open during this
>> phase,I usenativeskin
>> semaphore for it. This page fault causes the blocking malfunction of the
>> semaphores (rt_sem_p returns EPERM).
>>
>
> Using native skin semaphore is deprecated. If writing an RTDM driver, why
> not using an RTDM semaphore?
>
> --
>                                             Gilles.
>

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[Gilles Chanteperdrix]

On 12/18/2012 08:27 PM, alex alex wrote:

> In fact, this is what I intend to use when the bug will be solved. I use
> native because RTDM semaphore freezes my PC while with native, semaphore
> is just no longer blocking (so native helped me to find out the bug and
> it keeps me from restart the PC when it bugs... )
> 
> Thanks for the link.


Please try the patch to see if it still causes the bug.

-- 
                                                                Gilles.

Re: [Xenomai] [Page fault - POSIX skin] Interraction between malloc en nanosleep

[alex alex]

> >> If you want to avoid  mode when using malloc, pre-allocate some memory
> for
> >> RT tasks,
> >> then use a custom allocator (not malloc), such as the TLSF allocator for
> >> instance.
> >>
> >
> >
> > Actually I carry out a frameworks portage so unfortunately I can not
> change
> > the malloc function.
>
> Yes you can, simply use the same --wrap technique as Xenomai.
>

I wrapped the malloc,calloc, realloc and free functions with those of the
TLSF allocator and I don't have the page fault any more then the rtdm
semaphore works properly since I stay in the primary mode.

Thank you for your advices.

Alex.