[Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

I downloaded the ipipe-2.6 from git.denx.de, head	is
"ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
kernel. Then I install the xenomai applications into our rootfs. I
just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
it gives out following line:

*
*
* Type ^C to stop this application.
*
*
0.00 0.05 0.03 2/24 1010

T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
      0 Max:-1000000
Segmentation fault

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Gilles Chanteperdrix]

Peter Pan wrote:
> I downloaded the ipipe-2.6 from git.denx.de, head	is
> "ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
> kernel. Then I install the xenomai applications into our rootfs. I
> just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
> it gives out following line:
> 
> *
> *
> * Type ^C to stop this application.
> *
> *
> 0.00 0.05 0.03 2/24 1010
> 
> T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
>       0 Max:-1000000
> Segmentation fault

Probably a stack overflow. Could you run the application inside gdb to
check that?

-- 
					    Gilles.

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

I searched through the archive, some one says that there is a git
commit to sole the problem. I downloaded the latest xenomai-2.5 git,
the problem is gone.

But I found another problem. When I use the "latency" test, all the
min latency time gives out a minus value. Is that normal?

2010/4/29 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>:
> Peter Pan wrote:
>> I downloaded the ipipe-2.6 from git.denx.de, head     is
>> "ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
>> kernel. Then I install the xenomai applications into our rootfs. I
>> just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
>> it gives out following line:
>>
>> *
>> *
>> * Type ^C to stop this application.
>> *
>> *
>> 0.00 0.05 0.03 2/24 1010
>>
>> T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
>>       0 Max:-1000000
>> Segmentation fault
>
> Probably a stack overflow. Could you run the application inside gdb to
> check that?
>
> --
>                                            Gilles.
>

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Philippe Gerum]

On Thu, 2010-04-29 at 23:20 +0800, Peter Pan wrote:
> I searched through the archive, some one says that there is a git
> commit to sole the problem. I downloaded the latest xenomai-2.5 git,
> the problem is gone.
> 
> But I found another problem. When I use the "latency" test, all the
> min latency time gives out a minus value. Is that normal?

Yes. Your board has better internal latency than the default calibration
expects (i.e. time to traverse the Xenomai stack from IRQ receipt to
activating the thread), so the timer shot is anticipated too much.
Try reducing the value stored in /proc/xenomai/latency, until your
system is calibrated. You can wire this value in your Kconfig
afterwards.

Which ppc hw are you running Xenomai on?

> 
> 2010/4/29 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>:
> > Peter Pan wrote:
> >> I downloaded the ipipe-2.6 from git.denx.de, head     is
> >> "ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
> >> kernel. Then I install the xenomai applications into our rootfs. I
> >> just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
> >> it gives out following line:
> >>
> >> *
> >> *
> >> * Type ^C to stop this application.
> >> *
> >> *
> >> 0.00 0.05 0.03 2/24 1010
> >>
> >> T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
> >>       0 Max:-1000000
> >> Segmentation fault
> >
> > Probably a stack overflow. Could you run the application inside gdb to
> > check that?
> >
> > --
> >                                            Gilles.
> >
> 
> _______________________________________________
> Xenomai-help mailing list
> Xenomai-help@domain.hid
> https://mail.gna.org/listinfo/xenomai-help


-- 
Philippe.

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

It's an EVM board of PPC460GT.

I think the processor is very powerful, that may make xenomai workable
on our system.

I'll try to reduce the latency to a proper value.

Thanks alot.

2010/4/29 Philippe Gerum <rpm@xenomai.org>:
> On Thu, 2010-04-29 at 23:20 +0800, Peter Pan wrote:
>> I searched through the archive, some one says that there is a git
>> commit to sole the problem. I downloaded the latest xenomai-2.5 git,
>> the problem is gone.
>>
>> But I found another problem. When I use the "latency" test, all the
>> min latency time gives out a minus value. Is that normal?
>
> Yes. Your board has better internal latency than the default calibration
> expects (i.e. time to traverse the Xenomai stack from IRQ receipt to
> activating the thread), so the timer shot is anticipated too much.
> Try reducing the value stored in /proc/xenomai/latency, until your
> system is calibrated. You can wire this value in your Kconfig
> afterwards.
>
> Which ppc hw are you running Xenomai on?
>
>>
>> 2010/4/29 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>:
>> > Peter Pan wrote:
>> >> I downloaded the ipipe-2.6 from git.denx.de, head     is
>> >> "ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
>> >> kernel. Then I install the xenomai applications into our rootfs. I
>> >> just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
>> >> it gives out following line:
>> >>
>> >> *
>> >> *
>> >> * Type ^C to stop this application.
>> >> *
>> >> *
>> >> 0.00 0.05 0.03 2/24 1010
>> >>
>> >> T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
>> >>       0 Max:-1000000
>> >> Segmentation fault
>> >
>> > Probably a stack overflow. Could you run the application inside gdb to
>> > check that?
>> >
>> > --
>> >                                            Gilles.
>> >
>>
>> _______________________________________________
>> Xenomai-help mailing list
>> Xenomai-help@domain.hid
>> https://mail.gna.org/listinfo/xenomai-help
>
>
> --
> Philippe.
>
>
>

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Philippe Gerum]

On Fri, 2010-04-30 at 08:16 +0800, Peter Pan wrote:
> It's an EVM board of PPC460GT.
> 
> I think the processor is very powerful, that may make xenomai workable
> on our system.

> I'll try to reduce the latency to a proper value.

When you are done, I'd be interested in knowing the proper calibration
value for this platform, so that I can pre-set it in
include/asm-powerpc/calibration.h for future users.

PS: echo 0 > /proc/xenomai/latency is the way to disable the
anticipation in setting timer shots, so that you can determine the
minimum latency of your platform, i.e. abs(min-lat) from the latency
test output, on an idle system.

TIA,

> 
> Thanks alot.
> 
> 2010/4/29 Philippe Gerum <rpm@xenomai.org>:
> > On Thu, 2010-04-29 at 23:20 +0800, Peter Pan wrote:
> >> I searched through the archive, some one says that there is a git
> >> commit to sole the problem. I downloaded the latest xenomai-2.5 git,
> >> the problem is gone.
> >>
> >> But I found another problem. When I use the "latency" test, all the
> >> min latency time gives out a minus value. Is that normal?
> >
> > Yes. Your board has better internal latency than the default calibration
> > expects (i.e. time to traverse the Xenomai stack from IRQ receipt to
> > activating the thread), so the timer shot is anticipated too much.
> > Try reducing the value stored in /proc/xenomai/latency, until your
> > system is calibrated. You can wire this value in your Kconfig
> > afterwards.
> >
> > Which ppc hw are you running Xenomai on?
> >
> >>
> >> 2010/4/29 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>:
> >> > Peter Pan wrote:
> >> >> I downloaded the ipipe-2.6 from git.denx.de, head     is
> >> >> "ipipe-2.6.32-powerpc". Then I downloaded xenomai-2.5.2 and patch the
> >> >> kernel. Then I install the xenomai applications into our rootfs. I
> >> >> just run cyclictest in /usr/xenomai/share/xenomai/testsuit/cyclic/run,
> >> >> it gives out following line:
> >> >>
> >> >> *
> >> >> *
> >> >> * Type ^C to stop this application.
> >> >> *
> >> >> *
> >> >> 0.00 0.05 0.03 2/24 1010
> >> >>
> >> >> T: 0 (    0) P:99 I: 1000000 C:       0 Min: 1000000 Act:       0 Avg:
> >> >>       0 Max:-1000000
> >> >> Segmentation fault
> >> >
> >> > Probably a stack overflow. Could you run the application inside gdb to
> >> > check that?
> >> >
> >> > --
> >> >                                            Gilles.
> >> >
> >>
> >> _______________________________________________
> >> Xenomai-help mailing list
> >> Xenomai-help@domain.hid
> >> https://mail.gna.org/listinfo/xenomai-help
> >
> >
> > --
> > Philippe.
> >
> >
> >


-- 
Philippe.

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

OK, I got that.

But which task mode should this latency test be run for?
user-mode task or kernel periodic task?

2010/4/30 Philippe Gerum <rpm@xenomai.org>:
> On Fri, 2010-04-30 at 17:17 +0800, Peter Pan wrote:
>> I set the /proc/xenomai/latency to 0, and get some latency results.
>> How could I set the latency to a proper value? I'm on vocation now. I
>> will post my latency result when I'm back to work.
>
> You should run the /usr/xenomai/bin/latency test when your calibrated
> latency is reset to 0, for a few seconds on your idle platform. In the
> left most column after you stopped the test, you should see the minimum
> latency value, which your reported to be negative. Use abs(min-lat) as
> the next /proc/xenomai/latency value, and redo the test, until you end
> up with a min-lat slightly higher than 0, from a few tenths of
> nanoseconds.
>
>>
>> 2010/4/30 Philippe Gerum <rpm@xenomai.org>:
>> > On Fri, 2010-04-30 at 08:16 +0800, Peter Pan wrote:
>> >> It's an EVM board of PPC460GT.
>> >>
>> >> I think the processor is very powerful, that may make xenomai workable
>> >> on our system.
>> >
>> >> I'll try to reduce the latency to a proper value.
>> >
>> > When you are done, I'd be interested in knowing the proper calibration
>> > value for this platform, so that I can pre-set it in
>> > include/asm-powerpc/calibration.h for future users.
>> >
>> > PS: echo 0 > /proc/xenomai/latency is the way to disable the
>> > anticipation in setting timer shots, so that you can determine the
>> > minimum latency of your platform, i.e. abs(min-lat) from the latency
>> > test output, on an idle system.
>> >
>> > TIA,
>> >
>
>
> --
> Philippe.
>
>
>

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Philippe Gerum]

On Fri, 2010-04-30 at 20:03 +0800, Peter Pan wrote:
> OK, I got that.
> 
> But which task mode should this latency test be run for?
> user-mode task or kernel periodic task?
> 

It depends on your particular use case, since the latency value is
global for all modes, but I guess you will likely run userland rt apps,
which is recommended, so latency [-t0] is what you look for. Userland
latency should be higher than any other modes.

> 2010/4/30 Philippe Gerum <rpm@xenomai.org>:
> > On Fri, 2010-04-30 at 17:17 +0800, Peter Pan wrote:
> >> I set the /proc/xenomai/latency to 0, and get some latency results.
> >> How could I set the latency to a proper value? I'm on vocation now. I
> >> will post my latency result when I'm back to work.
> >
> > You should run the /usr/xenomai/bin/latency test when your calibrated
> > latency is reset to 0, for a few seconds on your idle platform. In the
> > left most column after you stopped the test, you should see the minimum
> > latency value, which your reported to be negative. Use abs(min-lat) as
> > the next /proc/xenomai/latency value, and redo the test, until you end
> > up with a min-lat slightly higher than 0, from a few tenths of
> > nanoseconds.
> >
> >>
> >> 2010/4/30 Philippe Gerum <rpm@xenomai.org>:
> >> > On Fri, 2010-04-30 at 08:16 +0800, Peter Pan wrote:
> >> >> It's an EVM board of PPC460GT.
> >> >>
> >> >> I think the processor is very powerful, that may make xenomai workable
> >> >> on our system.
> >> >
> >> >> I'll try to reduce the latency to a proper value.
> >> >
> >> > When you are done, I'd be interested in knowing the proper calibration
> >> > value for this platform, so that I can pre-set it in
> >> > include/asm-powerpc/calibration.h for future users.
> >> >
> >> > PS: echo 0 > /proc/xenomai/latency is the way to disable the
> >> > anticipation in setting timer shots, so that you can determine the
> >> > minimum latency of your platform, i.e. abs(min-lat) from the latency
> >> > test output, on an idle system.
> >> >
> >> > TIA,
> >> >
> >
> >
> > --
> > Philippe.
> >
> >
> >


-- 
Philippe.

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

OK, got that.
I will test the latency both on AMCC PPC460GT and P2020E.

2010/4/30 Philippe Gerum <rpm@xenomai.org>:
> On Fri, 2010-04-30 at 20:03 +0800, Peter Pan wrote:
>> OK, I got that.
>>
>> But which task mode should this latency test be run for?
>> user-mode task or kernel periodic task?
>>
>
> It depends on your particular use case, since the latency value is
> global for all modes, but I guess you will likely run userland rt apps,
> which is recommended, so latency [-t0] is what you look for. Userland
> latency should be higher than any other modes.
>
>> 2010/4/30 Philippe Gerum <rpm@xenomai.org>:
>> > On Fri, 2010-04-30 at 17:17 +0800, Peter Pan wrote:
>> >> I set the /proc/xenomai/latency to 0, and get some latency results.
>> >> How could I set the latency to a proper value? I'm on vocation now. I
>> >> will post my latency result when I'm back to work.
>> >
>> > You should run the /usr/xenomai/bin/latency test when your calibrated
>> > latency is reset to 0, for a few seconds on your idle platform. In the
>> > left most column after you stopped the test, you should see the minimum
>> > latency value, which your reported to be negative. Use abs(min-lat) as
>> > the next /proc/xenomai/latency value, and redo the test, until you end
>> > up with a min-lat slightly higher than 0, from a few tenths of
>> > nanoseconds.
>> >
>> >>
>> >> 2010/4/30 Philippe Gerum <rpm@xenomai.org>:
>> >> > On Fri, 2010-04-30 at 08:16 +0800, Peter Pan wrote:
>> >> >> It's an EVM board of PPC460GT.
>> >> >>
>> >> >> I think the processor is very powerful, that may make xenomai workable
>> >> >> on our system.
>> >> >
>> >> >> I'll try to reduce the latency to a proper value.
>> >> >
>> >> > When you are done, I'd be interested in knowing the proper calibration
>> >> > value for this platform, so that I can pre-set it in
>> >> > include/asm-powerpc/calibration.h for future users.
>> >> >
>> >> > PS: echo 0 > /proc/xenomai/latency is the way to disable the
>> >> > anticipation in setting timer shots, so that you can determine the
>> >> > minimum latency of your platform, i.e. abs(min-lat) from the latency
>> >> > test output, on an idle system.
>> >> >
>> >> > TIA,
>> >> >
>> >
>> >
>> > --
>> > Philippe.
>> >
>> >
>> >
>
>
> --
> Philippe.
>
>
>

Re: [Xenomai-help] cyclictest gives out segmentation fault with xenomai 2.5.2 in powerpc arch

[Peter Pan]

The calibrated latency for PPC460GT:

latency : 2077
results :
periodic user-mode task:
RTS|      0.006|      0.095|      8.229|       0|     0|    00:00:10/00:00:10

The calibrated latency for P2020RDB:

latency : 2736
results :
periodic user-mode task:
RTS|      0.000|      0.064|      3.152|       0|     0|    00:00:20/00:00:20