[Xenomai] about switchtest ...

[Xenomai] about switchtest ...

[raespi]

Hi ... I'm using the 2.6.0 Xenomai version in userspace.  I managed to 
forwardport the *adeos-ipipe-3.0.36-arm-1.18-11* patch provided to the 
3.1.10 kernel version since it's the one I'm currently using.  I managed 
to modify it in a way that it compiles and runs on my SMDK2440 ARM 
target.  I attached the patch for your consideration.  If you consider 
it valuable I can send you another version at a later time with the 
other marchs included ( I removed them from the patched and left the 
modifications important to me ) ...

I followed the guide and everything works fine, I see the boot messages 
and the *xeno latency* test runs without any problem.  The problem 
arises when I try the *xeno-test *program, it starts with the first test 
and then crashes with an error:

switchtest: Unable to open switchtest device.
(modprobe xeno_switchtest ?)

I issued a *cat /proc/xenomai/rtdm/named_devices* and got:

Hash    Name                Driver        /proc
24    rttest-timerbench0                 xeno_timerbench    
rttest-timerbench0
55    rttest-switchtest0                 xeno_switchtest    
rttest-switchtest0

Meaning the rttest-switchtest0 device is there right?? I looked for it 
here after googling this problem.  In some mailing lists it's said that 
this problem was fixed in the 2.5.1 head version, but as you can see 
it's still popping up ...

Any ideas ??

Thanks in advance ...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: adeos-ipipe-3.1.10-arm-001.patch
Type: text/x-diff
Size: 460278 bytes
Desc: not available
URL: <http://www.xenomai.org/pipermail/xenomai/attachments/20130204/7310252d/attachment.patch>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/04/2013 07:28 PM, raespi wrote:

> Hi ... I'm using the 2.6.0 Xenomai version in userspace.  I managed to 
> forwardport the *adeos-ipipe-3.0.36-arm-1.18-11* patch provided to the 
> 3.1.10 kernel version since it's the one I'm currently using.  I managed 
> to modify it in a way that it compiles and runs on my SMDK2440 ARM 
> target.  I attached the patch for your consideration.  If you consider 
> it valuable I can send you another version at a later time with the 
> other marchs included ( I removed them from the patched and left the 
> modifications important to me ) ...
> 
> I followed the guide and everything works fine, I see the boot messages 
> and the *xeno latency* test runs without any problem.  The problem 
> arises when I try the *xeno-test *program, it starts with the first test 
> and then crashes with an error:
> 
> switchtest: Unable to open switchtest device.
> (modprobe xeno_switchtest ?)
> 
> I issued a *cat /proc/xenomai/rtdm/named_devices* and got:
> 
> Hash    Name                Driver        /proc
> 24    rttest-timerbench0                 xeno_timerbench    
> rttest-timerbench0
> 55    rttest-switchtest0                 xeno_switchtest    
> rttest-switchtest0
> 
> Meaning the rttest-switchtest0 device is there right?? I looked for it 
> here after googling this problem.  In some mailing lists it's said that 
> this problem was fixed in the 2.5.1 head version, but as you can see 
> it's still popping up ...
> 
> Any ideas ??


In order to avoid looking for a bug which has already been fixed, please
try Xenomai 2.6.2.1.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

Ok, I downloaded the 2.6.2.1 stable version now.  Everything compiles 
without problems ... Still I get the same error when running *xeno-latency*:

switchtest: Unable to open switchtest device.
(modprobe xeno_switchtest ?)

Any ideas .. ?


El 02/04/2013 01:31 PM, Gilles Chanteperdrix escribió:
> On 02/04/2013 07:28 PM, raespi wrote:
>
>> Hi ... I'm using the 2.6.0 Xenomai version in userspace.  I managed to
>> forwardport the *adeos-ipipe-3.0.36-arm-1.18-11* patch provided to the
>> 3.1.10 kernel version since it's the one I'm currently using.  I managed
>> to modify it in a way that it compiles and runs on my SMDK2440 ARM
>> target.  I attached the patch for your consideration.  If you consider
>> it valuable I can send you another version at a later time with the
>> other marchs included ( I removed them from the patched and left the
>> modifications important to me ) ...
>>
>> I followed the guide and everything works fine, I see the boot messages
>> and the *xeno latency* test runs without any problem.  The problem
>> arises when I try the *xeno-test *program, it starts with the first test
>> and then crashes with an error:
>>
>> switchtest: Unable to open switchtest device.
>> (modprobe xeno_switchtest ?)
>>
>> I issued a *cat /proc/xenomai/rtdm/named_devices* and got:
>>
>> Hash    Name                Driver        /proc
>> 24    rttest-timerbench0                 xeno_timerbench
>> rttest-timerbench0
>> 55    rttest-switchtest0                 xeno_switchtest
>> rttest-switchtest0
>>
>> Meaning the rttest-switchtest0 device is there right?? I looked for it
>> here after googling this problem.  In some mailing lists it's said that
>> this problem was fixed in the 2.5.1 head version, but as you can see
>> it's still popping up ...
>>
>> Any ideas ??
>
> In order to avoid looking for a bug which has already been fixed, please
> try Xenomai 2.6.2.1.
>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/04/2013 08:22 PM, raespi wrote:

>  Ok, I downloaded the 2.6.2.1 stable version now.  Everything compiles
> without problems ... Still I get the same error when running *xeno-latency*:
> 
> switchtest: Unable to open switchtest device.
> (modprobe xeno_switchtest ?)
> 
> Any ideas .. ?


No, we do not have enough details. What toolchain do you use? What
arguments do you pass to configure? Did you recompile the kernel too?

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[Jan Kiszka]

On 2013-02-04 20:55, Gilles Chanteperdrix wrote:
> On 02/04/2013 08:22 PM, raespi wrote:
> 
>>  Ok, I downloaded the 2.6.2.1 stable version now.  Everything compiles
>> without problems ... Still I get the same error when running *xeno-latency*:
>>
>> switchtest: Unable to open switchtest device.
>> (modprobe xeno_switchtest ?)
>>
>> Any ideas .. ?
> 
> 
> No, we do not have enough details. What toolchain do you use? What
> arguments do you pass to configure? Did you recompile the kernel too?

I remember stumbling over this issue recently as well, on x86. Hmm, I
think it was not a missing device but failing
RTTST_RTIOC_SWTEST_SET_TASKS_COUNT due to too few resources of the
nucleus - but that was on 16 or even 32 CPU box.

Anyway, my point is that the failure might be in that IOCTL. Instrument
the code to find out.

Jan

-- 
Siemens AG, Corporate Technology, CT RTC ITP SDP-DE
Corporate Competence Center Embedded Linux

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/04/2013 10:02 PM, raespi wrote:

>  Right, I'm using these tools and options:
> 
> *- toolchain:*
> 4.6.3 (Sourcery CodeBench Lite 2012.03-57)
> 
> *- configure options ( Running under buildroot, no special recipe, it
> comes with one for the 2.6.0 version ):*
> $ ./configure --target=arm-linux --host=arm-linux
> --build=i686-pc-linux-gnu --prefix=/usr --exec-prefix=/usr
> --sysconfdir=/etc --program-prefix= --disable-gtk-doc --enable-ipv6
> --enable-static --enable-shared --includedir=/usr/include/xenomai/
> 
> *- Prepin' the kernel:*
> $(XENOMAI_DIR)/scripts/prepare-kernel.sh --linux=$(LINUX_DIR) --arch=arm
> --adeos=adeos-ipipe-3.1.10-arm-001.patch
> 
> I think I saw the problem.   Buildroot doesn't pass the " -march=armv4t
> -msoft-float"  CFLAGS and LDFLAGS options to the xenomai configure line
> for the s3c24xx.  At least the *xeno-test* is running now ...


Good news. It is strange that not passing these options does not cause
other run-time errors though (like invalid instructions).

> 
> Thanks for your time, are you interested in the complete patch for the
> 3.1.10 version ?? ...


Yes, that is interesting, but 3.1 is a bit old, I am not sure you have
all the fixes, any reason for not using 3.2?

Please keep the mailing list in CC.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/04/2013 09:30 PM, Jan Kiszka wrote:

> On 2013-02-04 20:55, Gilles Chanteperdrix wrote:
>> On 02/04/2013 08:22 PM, raespi wrote:
>>
>>>  Ok, I downloaded the 2.6.2.1 stable version now.  Everything compiles
>>> without problems ... Still I get the same error when running *xeno-latency*:
>>>
>>> switchtest: Unable to open switchtest device.
>>> (modprobe xeno_switchtest ?)
>>>
>>> Any ideas .. ?
>>
>>
>> No, we do not have enough details. What toolchain do you use? What
>> arguments do you pass to configure? Did you recompile the kernel too?
> 
> I remember stumbling over this issue recently as well, on x86. Hmm, I
> think it was not a missing device but failing
> RTTST_RTIOC_SWTEST_SET_TASKS_COUNT due to too few resources of the
> nucleus - but that was on 16 or even 32 CPU box.


It is true that any error in the ioctl will trigger this message,
whereas we should treat anything else than -ENOSYS as an error.


-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

Well we have a very big team working with linux on the smdk2440, from 
what we've seen, the 3.4 version breaks the audio in a very weird way.   
We're also merging other changes like the ones from the AF_DBUS project, 
but they're using other versions and it's a bit complicated.  Still all 
our changes will be submitted back to you ... I want to test Xenomai a 
bit more and see how it improves my system's response times, when I have 
more info I will send the full changes ...

Thanks again ..

El 02/04/2013 04:07 PM, Gilles Chanteperdrix escribió:
> On 02/04/2013 10:02 PM, raespi wrote:
>
>>   Right, I'm using these tools and options:
>>
>> *- toolchain:*
>> 4.6.3 (Sourcery CodeBench Lite 2012.03-57)
>>
>> *- configure options ( Running under buildroot, no special recipe, it
>> comes with one for the 2.6.0 version ):*
>> $ ./configure --target=arm-linux --host=arm-linux
>> --build=i686-pc-linux-gnu --prefix=/usr --exec-prefix=/usr
>> --sysconfdir=/etc --program-prefix= --disable-gtk-doc --enable-ipv6
>> --enable-static --enable-shared --includedir=/usr/include/xenomai/
>>
>> *- Prepin' the kernel:*
>> $(XENOMAI_DIR)/scripts/prepare-kernel.sh --linux=$(LINUX_DIR) --arch=arm
>> --adeos=adeos-ipipe-3.1.10-arm-001.patch
>>
>> I think I saw the problem.   Buildroot doesn't pass the " -march=armv4t
>> -msoft-float"  CFLAGS and LDFLAGS options to the xenomai configure line
>> for the s3c24xx.  At least the *xeno-test* is running now ...
>
> Good news. It is strange that not passing these options does not cause
> other run-time errors though (like invalid instructions).
>
>> Thanks for your time, are you interested in the complete patch for the
>> 3.1.10 version ?? ...
>
> Yes, that is interesting, but 3.1 is a bit old, I am not sure you have
> all the fixes, any reason for not using 3.2?
>
> Please keep the mailing list in CC.
>

Re: [Xenomai] about switchtest ...

[raespi]

Still on this topic ... do the CFLAGS and LDFLAGS parameters are checked 
only from the ./configure command line or are they also checked by 
Xenomai as an environment variable ??  I'm currently in discussion with 
the buildroot maintainer about this topic to fix this problem ...

El 02/04/2013 04:28 PM, raespi escribió:
> Well we have a very big team working with linux on the smdk2440, from 
> what we've seen, the 3.4 version breaks the audio in a very weird 
> way.   We're also merging other changes like the ones from the AF_DBUS 
> project, but they're using other versions and it's a bit complicated.  
> Still all our changes will be submitted back to you ... I want to test 
> Xenomai a bit more and see how it improves my system's response times, 
> when I have more info I will send the full changes ...
>
> Thanks again ..
>
> El 02/04/2013 04:07 PM, Gilles Chanteperdrix escribió:
>> On 02/04/2013 10:02 PM, raespi wrote:
>>
>>>   Right, I'm using these tools and options:
>>>
>>> *- toolchain:*
>>> 4.6.3 (Sourcery CodeBench Lite 2012.03-57)
>>>
>>> *- configure options ( Running under buildroot, no special recipe, it
>>> comes with one for the 2.6.0 version ):*
>>> $ ./configure --target=arm-linux --host=arm-linux
>>> --build=i686-pc-linux-gnu --prefix=/usr --exec-prefix=/usr
>>> --sysconfdir=/etc --program-prefix= --disable-gtk-doc --enable-ipv6
>>> --enable-static --enable-shared --includedir=/usr/include/xenomai/
>>>
>>> *- Prepin' the kernel:*
>>> $(XENOMAI_DIR)/scripts/prepare-kernel.sh --linux=$(LINUX_DIR) 
>>> --arch=arm
>>> --adeos=adeos-ipipe-3.1.10-arm-001.patch
>>>
>>> I think I saw the problem.   Buildroot doesn't pass the " -march=armv4t
>>> -msoft-float"  CFLAGS and LDFLAGS options to the xenomai configure line
>>> for the s3c24xx.  At least the *xeno-test* is running now ...
>>
>> Good news. It is strange that not passing these options does not cause
>> other run-time errors though (like invalid instructions).
>>
>>> Thanks for your time, are you interested in the complete patch for the
>>> 3.1.10 version ?? ...
>>
>> Yes, that is interesting, but 3.1 is a bit old, I am not sure you have
>> all the fixes, any reason for not using 3.2?
>>
>> Please keep the mailing list in CC.
>>
>
>
>
> _______________________________________________
> Xenomai mailing list
> Xenomai@xenomai.org
> http://www.xenomai.org/mailman/listinfo/xenomai

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 02:34 PM, raespi wrote:

> Still on this topic ... do the CFLAGS and LDFLAGS parameters are checked 
> only from the ./configure command line or are they also checked by 
> Xenomai as an environment variable ??  I'm currently in discussion with 
> the buildroot maintainer about this topic to fix this problem ...


Xenomai does not do anything special, it uses the autotools default
which is:
- if you pass CFLAGS and LDFLAGS on the right hand of the configure
command line, the generated makefiles are hardcoded with these flags
values, and you need not do anything special later on to get these flags
used for all the compilations. This has been the recommended way of
passing CFLAGS and LDFLAGS when building autotools-based projects for
many years.

- if you want to pass them as environment variables (so, on the left
hand of the configure command line), the makefiles are not generated
with these flags, so, you have to pass them in the environment of the
"make" command.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 02:34 PM, raespi wrote:

> Still on this topic ... do the CFLAGS and LDFLAGS parameters are checked 
> only from the ./configure command line or are they also checked by 
> Xenomai as an environment variable ??  I'm currently in discussion with 
> the buildroot maintainer about this topic to fix this problem ...


I am not sure you should bother the buildroot maintainer until you have
understood what the real problem is. Because if you do not compile for
the right machine, you should get illegal instructions, not the error
you report.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

The problem was fixed ... The buildroot system didn't pass the specific 
CFLAGS and LDFLAGS for my machine.  In your Getting Started guide this 
is something necessary.  That's what fixed the problem and allowed me to 
run the tests.  What I'm checking with him is how those flags are passed 
to Xenomai.  Perhaps he didn't notice it before since no one ran the 
test suite for this machine ?? IDK ...

El 02/05/2013 09:52 AM, Gilles Chanteperdrix escribió:
> On 02/05/2013 02:34 PM, raespi wrote:
>
>> Still on this topic ... do the CFLAGS and LDFLAGS parameters are checked
>> only from the ./configure command line or are they also checked by
>> Xenomai as an environment variable ??  I'm currently in discussion with
>> the buildroot maintainer about this topic to fix this problem ...
>
> I am not sure you should bother the buildroot maintainer until you have
> understood what the real problem is. Because if you do not compile for
> the right machine, you should get illegal instructions, not the error
> you report.
>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 03:56 PM, raespi wrote:

> The problem was fixed ... The buildroot system didn't pass the specific 
> CFLAGS and LDFLAGS for my machine.  In your Getting Started guide this 

> is something necessary.  That's what fixed the problem and allowed me to 
> run the tests.  What I'm checking with him is how those flags are passed 
> to Xenomai.  Perhaps he didn't notice it before since no one ran the 
> test suite for this machine ?? IDK ...


No, read my answer again. If buildroot did not pass correct CFLAGS or
LDFLAGS, you would get an "illegal instruction" trap. So, you have not
really understood what your problem is, and you have not understood why
passing specific CFLAGS and LDFLAGS in addition to the ones passed by
buildroot fixes it.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

Yes I saw your answer and understood what you were talking about.  
Again, the "-march=armv4t -msoft-float" options are passed to every 
package Buildroot builds since it's an option I enable in the Buildroot 
config file.  The problem may lie in how it passes them to Xenomai 
specifically.  For a fact, I can assure you they weren't passed to the 
right of the ./configure script as you suggested since I double checked 
the config.log for the Xenomai build.   Why the "Illegal Instruction" 
message wasn't showing up I don't know but it's not something that 
popped up to me before for not passing the flags, at least not in the 
smdk2440 target.

Facts:
1- Buildroot doesn't build correctly the Xenomai package ( test suite 
fails to launch )
2- When passing the flags in Buildroot manually ( for now ) to the 
Xenomai package the test suite runs normally
3- The solution is either fixing the parameter passing scheme on Xenomai 
or how Buildroot passes the parameters ( I go with this last one ).

El 02/05/2013 09:59 AM, Gilles Chanteperdrix escribió:
> On 02/05/2013 03:56 PM, raespi wrote:
>
>> The problem was fixed ... The buildroot system didn't pass the specific
>> CFLAGS and LDFLAGS for my machine.  In your Getting Started guide this
>> is something necessary.  That's what fixed the problem and allowed me to
>> run the tests.  What I'm checking with him is how those flags are passed
>> to Xenomai.  Perhaps he didn't notice it before since no one ran the
>> test suite for this machine ?? IDK ...
>
> No, read my answer again. If buildroot did not pass correct CFLAGS or
> LDFLAGS, you would get an "illegal instruction" trap. So, you have not
> really understood what your problem is, and you have not understood why
> passing specific CFLAGS and LDFLAGS in addition to the ones passed by
> buildroot fixes it.
>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 04:18 PM, raespi wrote:

> Facts:
> 1- Buildroot doesn't build correctly the Xenomai package ( test suite 
> fails to launch )
> 2- When passing the flags in Buildroot manually ( for now ) to the 
> Xenomai package the test suite runs normally


I Agree with these two, but what I ask you is why?

> 3- The solution is either fixing the parameter passing scheme on Xenomai 
> or how Buildroot passes the parameters ( I go with this last one ).


Disagree with this one, which is a shortcut, not a real conclusion.
There may be a bug, what you have found is a workaround, not a solution.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

I know, which is why I'm currently between two mailing lists trying to 
solve it.  I have already registered the bug on the Buildroot site.  For 
followups this is the url:
https://bugs.busybox.net/show_bug.cgi?id=5900

El 02/05/2013 10:25 AM, Gilles Chanteperdrix escribió:
> On 02/05/2013 04:18 PM, raespi wrote:
>
>> Facts:
>> 1- Buildroot doesn't build correctly the Xenomai package ( test suite
>> fails to launch )
>> 2- When passing the flags in Buildroot manually ( for now ) to the
>> Xenomai package the test suite runs normally
>
> I Agree with these two, but what I ask you is why?
>
>> 3- The solution is either fixing the parameter passing scheme on Xenomai
>> or how Buildroot passes the parameters ( I go with this last one ).
>
> Disagree with this one, which is a shortcut, not a real conclusion.
> There may be a bug, what you have found is a workaround, not a solution.
>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 04:31 PM, raespi wrote:

>  I know, which is why I'm currently between two mailing lists trying to
> solve it.  I have already registered the bug on the Buildroot site.  For
> followups this is the url:
> https://bugs.busybox.net/show_bug.cgi?id=5900


And again: this probably has nothing to do with buildroot.

-- 
                                                                Gilles.

Re: [Xenomai] about switchtest ...

[raespi]

So you're saying that I'm correctly passing the CFLAGS and LDFLAGS for 
all the other packages on my system ( kernel, busybox, dbus, my apps, 
etc ... ) but not to the Xenomai build ?? I fail to understand your 
point ...

On Buildroot I have:
1- BR2_GCC_TARGET_ARCH="armv4t"
2- BR2_SOFT_FLOAT=y

Both translated in to good old GCC parameters are "-march=arm4vt 
-msoft-float". So ???

El 02/05/2013 10:32 AM, Gilles Chanteperdrix escribió:
> On 02/05/2013 04:31 PM, raespi wrote:
>
>>   I know, which is why I'm currently between two mailing lists trying to
>> solve it.  I have already registered the bug on the Buildroot site.  For
>> followups this is the url:
>> https://bugs.busybox.net/show_bug.cgi?id=5900
>
> And again: this probably has nothing to do with buildroot.
>

Re: [Xenomai] about switchtest ...

[Gilles Chanteperdrix]

On 02/05/2013 04:38 PM, raespi wrote:

> So you're saying that I'm correctly passing the CFLAGS and LDFLAGS for 
> all the other packages on my system ( kernel, busybox, dbus, my apps, 
> etc ... ) but not to the Xenomai build ?? I fail to understand your 
> point ...
> 
> On Buildroot I have:
> 1- BR2_GCC_TARGET_ARCH="armv4t"
> 2- BR2_SOFT_FLOAT=y
> 
> Both translated in to good old GCC parameters are "-march=arm4vt 
> -msoft-float". So ???


So, there is probably a problem in Xenomai build system, which has
nothing to do with buildroot, but is related to the (deprecated) way
buildroot passes the CFLAGS and LDFLAGS to the build system.

-- 
                                                                Gilles.