On Friday 25 February 2011 14:18:18 Peter Soetens wrote:
> On Friday 25 February 2011 13:48:45 Ruben Smits wrote:
>
> > On Friday 25 February 2011 13:35:37 Ruben Smits wrote:
> >
> > > On Friday 25 February 2011 13:32:06 Ruben Smits wrote:
> > >
> > > > On Friday 25 February 2011 13:18:18 Ruben Smits wrote:
> > > >
> > > > > On Friday 25 February 2011 13:16:20 Ruben Smits wrote:
> > > > >
> > > > > > On Friday 25 February 2011 13:10:43 Peter Soetens
> > > > > > wrote:
> > > > > >
> > > > > > > On Friday 25 February 2011 12:45:54 Ruben Smits
> > > > > > > wrote:
> > > > > > >
> > > > > > > > On Friday 25 February 2011 12:29:19 Peter
> > > > > > > > Soetens wrote:
> > > > > > > >
> > > > > > > > > On Friday 25 February 2011 12:19:55 Ruben
> > > > > > > > > Smits wrote:
> > > > > > > > >
> > > > > > > > > > Hi,
> > > > > > > > > >
> > > > > > > > > > I'm running RTT/OCL master (from
> > > > > > > > > > yesterday) on a
> > > > > > > > > > 2.6.35.7-xenomai-2.5.5.2
> > > > > > > > > > system and wanted to test if my
> > > > > > > > > > component does
> > > > > > > > > > not
> > > > > > > > > > cause
> > > > > > > > > > context
> > > > > > > > > > switches
> > > > > > > > > > between hard and soft realtime by using
> > > > > > > > > > the
> > > > > > > > > > rtos_enable_rt_warning()
> > > > > > > > > > function in my startHook. The component
> > > > > > > > > > is
> > > > > > > > > > running
> > > > > > > > > > fine
> > > > > > > > > > without
> > > > > > > > > > this
> > > > > > > > > > function.
> > > > > > > > > >
> > > > > > > > > > When I start my component, the process
> > > > > > > > > > is
> > > > > > > > > > directly
> > > > > > > > > > killed
> > > > > > > > > > with
> > > > > > > > > > the
> > > > > > > > > > following message:
> > > > > > > > > >
> > > > > > > > > > Deployer [S]> LWR.start
> > > > > > > > > > CPU time limit exceeded
> > > > > > > > > >
> > > > > > > > > > I checked the what this message means
> > > > > > > > > > and
> > > > > > > > > > <http://www.xenomai.org/documentation/xe
> > > > > > > > > > nomai-2.
> > > > > > > > > > 0/pd
> > > > > > > > > > f/Na
> > > > > > > > > > tive
> > > > > > > > > > -API
> > > > > > > > > > -Tour.p d f> tells me:
> > > > > > > > > > <quote>
> > > > > > > > > > Detecting unwanted switches to the
> > > > > > > > > > secondary
> > > > > > > > > > execution
> > > > > > > > > > mode
> > > > > > > > > > Sometimes, switching to the secondary
> > > > > > > > > > execution
> > > > > > > > > > mode
> > > > > > > > > > for
> > > > > > > > > > a
> > > > > > > > > > real­
> > > > > > > > > > time task so that it can
> > > > > > > > > > re­
> > > > > > > > > > enter the Linux kernel context may be
> > > > > > > > > > an
> > > > > > > > > > unwanted
> > > > > > > > > > behaviour,
> > > > > > > > > > likely due to invoking a Linux system
> > > > > > > > > > call
> > > > > > > > > > involuntarily. In
> > > > > > > > > > order to trap this situation, the native
> > > > > > > > > > API
> > > > > > > > > > allows
> > > > > > > > > > real­
> > > > > > > > > > time
> > > > > > > > > > tasks in user­
> > > > > > > > > > space to require the SIGXCPU signal
> > > > > > > > > > (i.e. CPU
> > > > > > > > > > time
> > > > > > > > > > limit
> > > > > > > > > > exceeded) to
> > > > > > > > > > be sent to them each time they switch
> > > > > > > > > > from the
> > > > > > > > > > primary
> > > > > > > > > > to
> > > > > > > > > > the
> > > > > > > > > > secondary execution mode. To use this
> > > > > > > > > > feature,
> > > > > > > > > > the
> > > > > > > > > > application
> > > > > > > > > > needs to install a handler for the
> > > > > > > > > > SIGXCPU
> > > > > > > > > > signal
> > > > > > > > > > (e.g.
> > > > > > > > > > using
> > > > > > > > > > sigaction(2)), then set the T_WARNSW
> > > > > > > > > > mode bit
> > > > > > > > > > for
> > > > > > > > > > the
> > > > > > > > > > task
> > > > > > > > > > to
> > > > > > > > > > monitor, using
> > > > > > > > > > the
> > > > > > > > > > rt_task_set_mode() system call. If no
> > > > > > > > > > SIGXCPU
> > > > > > > > > > handler
> > > > > > > > > > has
> > > > > > > > > > been
> > > > > > > > > > set, the whole application is terminated
> > > > > > > > > > upon
> > > > > > > > > > receipt of
> > > > > > > > > > this
> > > > > > > > > > signal. <quote>
> > > > > > > > > >
> > > > > > > > > > So apparently no SIGXCPU handler has
> > > > > > > > > > been set
> > > > > > > > > > and
> > > > > > > > > > this a
> > > > > > > > > > bug? Or
> > > > > > > > > > I do
> > > > > > > > > > something completely wrong and this is a
> > > > > > > > > > call
> > > > > > > > > > for
> > > > > > > > > > help
> > > > > > > > > > ;)
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > We definately have a SIGXCPU handler setup.
> > > > > > > > > The fact
> > > > > > > > > that
> > > > > > > > > you
> > > > > > > > > get
> > > > > > > > > the
> > > > > > > > > sigxcpu is at least a good sign. You should
> > > > > > > > > also see
> > > > > > > > > a
> > > > > > > > > log
> > > > > > > > > [Info]
> > > > > > > > > message with : "Installing SIGXCPU handler."
> > > > > > > > > right
> > > > > > > > > after
> > > > > > > > > "Xenomai
> > > > > > > > > Periodic Timer runs in preemptive 'one-shot'
> > > > > > > > > mode."
> > > > > > > >
> > > > > > > >
> > > > > > > > Ok, I see this stuff.
> > > > > > > >
> > > > > > > >
> > > > > > > > > Maybe the readline magic is blocking some
> > > > > > > > > signals,
> > > > > > > > > you
> > > > > > > > > could
> > > > > > > > > try
> > > > > > > > > to
> > > > > > > > > compile with editline (install
> > > > > > > > > libeditline-dev) and
> > >
> > >
> > > This should have been install libedit-dev.
> > >
> > > It's working now, I'll send a patch for the CMakeLists.txt fix
> >
> >
> > I had to set NO_GPL to ONN, to make sure that libreadline does not get
> > linked to the taskbrowser.
> >
> > But it is still not working, I've made sure that the deployer is not
> > linked
with readline:
> >
> > ldd
> > /home/rsmits/ros/kul-ros-pkg/stacks/orocos_toolchain_ros/ocl/bin/deploye
> > r-
xenomai
> >
> > linux-vdso.so.1 => (0x00007ffffebff000)
> > liblog4cpp.so => /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/log4cpp/install/lib/liblog4cpp.so
> > (0x00007f50e1ada000)
> >
> > liborocos-ocl-log4cpp-xenomai.so.2.3.0 =>
> > /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/ocl/lib/liborocos-ocl-log4cpp-xenomai.so
> > .2.
3.0 (0x00007f50e186d000)
> >
> > liborocos-rtt-xenomai.so.2.3 =>
> > /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/rtt/install/lib/liborocos-rtt-xenomai.so
> > .2.
3 (0x00007f50e14c7000)
> >
> > liborocos-ocl-taskbrowser-xenomai.so.2.3.0 =>
> > /home/rsmits/ros/kul-
> >
> > ros-pkg/stacks/orocos_toolchain_ros/ocl/lib/liborocos-ocl-taskbrowser-
> > xenomai.so.2.3.0 (0x00007f50e128d000)
> >
> > liborocos-ocl-deployment-xenomai.so.2.3.0 =>
> >
> > /home/rsmits/ros/kul-ros-
> > pkg/stacks/orocos_toolchain_ros/ocl/lib/liborocos-ocl-deployment-
> > xenomai.so.2.3.0 (0x00007f50e0eb3000)
> >
> > libboost_program_options.so.1.42.0 =>
> >
> > /usr/lib/libboost_program_options.so.1.42.0 (0x00007f50e0c30000)
> >
> > librtt-typekit-xenomai.so.2.3.0 =>
> > /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/rtt/install/lib/orocos/types/librtt-type
> > kit
- xenomai.so.2.3.0 (0x00007f50e03e0000)
> >
> > libboost_filesystem.so.1.42.0 =>
> >
> > /usr/lib/libboost_filesystem.so.1.42.0 (0x00007f50e01cb000)
> >
> > libboost_system.so.1.42.0 =>
> > /usr/lib/libboost_system.so.1.42.0
> >
> > (0x00007f50dffc6000)
> >
> > libboost_serialization.so.1.42.0 =>
> >
> > /usr/lib/libboost_serialization.so.1.42.0 (0x00007f50dfd4f000)
> >
> > libxenomai.so.0 => /usr/lib/libxenomai.so.0
> > (0x00007f50dfb4a000)
> > libnative.so.3 => /usr/lib/libnative.so.3
> > (0x00007f50df93e000)
> > libpthread.so.0 => /lib/libpthread.so.0
> > (0x00007f50df721000)
> > libdl.so.2 => /lib/libdl.so.2 (0x00007f50df51d000)
> > libstdc++.so.6 => /usr/lib/libstdc++.so.6
> > (0x00007f50df216000)
> > libm.so.6 => /lib/libm.so.6 (0x00007f50def93000)
> > libgcc_s.so.1 => /lib/libgcc_s.so.1
> > (0x00007f50ded7d000)
> > libc.so.6 => /lib/libc.so.6 (0x00007f50de9f9000)
> > libedit.so.2 => /usr/lib/libedit.so.2
> > (0x00007f50de7d1000)
> > libncurses.so.5 => /lib/libncurses.so.5
> > (0x00007f50de58d000)
> > librtt-scripting-xenomai.so.2.3.0 =>
> > /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/rtt/install/lib/orocos/plugins/librtt-
> > scripting-xenomai.so.2.3.0 (0x00007f50ddf45000)
> >
> > librtt-marshalling-xenomai.so.2.3.0 =>
> > /home/rsmits/ros/kul-ros-
> >
> > pkg/stacks/orocos_toolchain_ros/rtt/install/lib/orocos/plugins/librtt-
> > marshalling-xenomai.so.2.3.0 (0x00007f50ddc8f000)
> >
> > libroslib.so =>
> > /opt/ros/unstable/ros/core/roslib/lib/libroslib.so
> >
> > (0x00007f50dda82000)
> >
> > librt.so.1 => /lib/librt.so.1 (0x00007f50dd87a000)
> > /lib64/ld-linux-x86-64.so.2 (0x00007f50e1d61000)
> > libbsd.so.0 => /lib/libbsd.so.0 (0x00007f50dd66f000)
> > librospack.so =>
> >
> > /opt/ros/unstable/ros/tools/rospack/lib/librospack.so
> > (0x00007f50dd441000)
> >
> > librosstack.so =>
> >
> > /opt/ros/unstable/ros/tools/rospack/lib/librosstack.so
> > (0x00007f50dd22e000)
> >
> > libboost_thread.so.1.42.0 =>
> > /usr/lib/libboost_thread.so.1.42.0
> >
> > (0x00007f50dd019000)
> >
> > How can I check if other linked in libraries are not bugging me? (I'm
> > suspecting the ros related libs)
>
>
> I don't know. We could install the signal handler explicitly or try to force
>
>to reinstall it by calling
> rtos_task_create_main ( MainThread::Instance()->getTask() );
> from the main thread. But i don't know if the double shadow will work in
> Xenomai.
>
> Maybe first try the patch in attachment which uses the recommended sigaction
>
>instead of signal...

The patch did not work. But I've backtraced the signal using gdb, it shows
that the context switch is caused by the TaskBrowser, I also see that the
deployer's activity is a RT activity by default, I've tried to set its
activity to ORO_SCHED_OTHER, but that does not seem to work (It still shows up
as rt task in /proc/xenomai/sched, but maybe that's normal).

Shouldn't the deployer by default have a non-RT activitity attached? Why
doesn't the setActivity with ORO_SCHED_OTHER work?

-- Ruben

> Peter