Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Karl Reichert]

Jan Kiszka wrote:
> Karl Reichert wrote:
> > Hello,
> > 
> > I have two stations running RTnet, one as a master and one as a slave. I
> want 
> > to measure the time, which a message, sent from master to slave, takes.
> But, 
> > I don't want the time of the transmission only, but the complete time of
> all 
> > software layers this data is processed through.
> > 
> > _______________	_______________
> > |		A		  |	|		D		  |
> > |______________|	|______________|
> > |		B		  |	|		C		  |
> > |______________|	|______________|
> > |   RTnet (Master)	  |	|    RTnet (Slave)	  |
> > |______________|	|______________|
> > 		|				  |
> > 		|______________|
> > 
> > So, I create the data, that should be send, in Layer A (Master) and send
> it 
> > over RTnet to Layer D (Slave).
> > 
> > I want to take the time when the data leaves A and when it arrives D
> with 
> > rtdm_clock_read(). If I have the offset between master and slave, I can 
> > calculate the time it took to pass A, B, both RTnet stacks, C and D.
> > 
> > As RTnet keeps track of the offset in it's stack, I want to use this
> value. Is 
> > it possible via the API? I didn't found anything. Or do I have to
> manipulate 
> > the stack to pass the offset to the higher layers (C and D)?
> 
> If RTmac/TDMA is in use, you can obtain the clock offset via
> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where "<x>"
> corresponds to "rteth<x>").
> 
> Jan
> 

When I try to compile my application (xenomai native skin, user task), I get an error, saying rtdm_driver.h (which contains the prototype for rtdm_clock_read()) is for kernel mode tasks only.

How can I get a timestamp in a xenomai native user task? Is there sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?

Thanks
Karl
-- 
von Karl Reichert

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Jan Kiszka]

Karl Reichert wrote:
> Jan Kiszka wrote:
>> Karl Reichert wrote:
>>> Hello,
>>>
>>> I have two stations running RTnet, one as a master and one as a slave. I
>> want 
>>> to measure the time, which a message, sent from master to slave, takes.
>> But, 
>>> I don't want the time of the transmission only, but the complete time of
>> all 
>>> software layers this data is processed through.
>>>
>>> _______________	_______________
>>> |		A		  |	|		D		  |
>>> |______________|	|______________|
>>> |		B		  |	|		C		  |
>>> |______________|	|______________|
>>> |   RTnet (Master)	  |	|    RTnet (Slave)	  |
>>> |______________|	|______________|
>>> 		|				  |
>>> 		|______________|
>>>
>>> So, I create the data, that should be send, in Layer A (Master) and send
>> it 
>>> over RTnet to Layer D (Slave).
>>>
>>> I want to take the time when the data leaves A and when it arrives D
>> with 
>>> rtdm_clock_read(). If I have the offset between master and slave, I can 
>>> calculate the time it took to pass A, B, both RTnet stacks, C and D.
>>>
>>> As RTnet keeps track of the offset in it's stack, I want to use this
>> value. Is 
>>> it possible via the API? I didn't found anything. Or do I have to
>> manipulate 
>>> the stack to pass the offset to the higher layers (C and D)?
>> If RTmac/TDMA is in use, you can obtain the clock offset via
>> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where "<x>"
>> corresponds to "rteth<x>").
>>
>> Jan
>>
> 
> When I try to compile my application (xenomai native skin, user task), I get an error, saying rtdm_driver.h (which contains the prototype for rtdm_clock_read()) is for kernel mode tasks only.

So you are in userland, obviously...

> 
> How can I get a timestamp in a xenomai native user task? Is there sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?

Nope, there is rt_timer_read & Co. for native user-mode applications.

I was about to remark that everything can be found in the API docs - but
it can't! Any volunteer to add the missing functions from native/timer.h
to the docs? Maybe there is just something broken for doxygen. TIA!

Jan

-- 
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Gilles Chanteperdrix]

On Wed, May 14, 2008 at 5:04 PM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>
> Karl Reichert wrote:
>  > Jan Kiszka wrote:
>  >> Karl Reichert wrote:
>  >>> Hello,
>  >>>
>  >>> I have two stations running RTnet, one as a master and one as a slave. I
>  >> want
>  >>> to measure the time, which a message, sent from master to slave, takes.
>  >> But,
>  >>> I don't want the time of the transmission only, but the complete time of
>  >> all
>  >>> software layers this data is processed through.
>  >>>
>  >>> _______________     _______________
>  >>> |           A                 |     |               D                 |
>  >>> |______________|    |______________|
>  >>> |           B                 |     |               C                 |
>  >>> |______________|    |______________|
>  >>> |   RTnet (Master)    |     |    RTnet (Slave)        |
>  >>> |______________|    |______________|
>  >>>             |                                 |
>  >>>             |______________|
>  >>>
>  >>> So, I create the data, that should be send, in Layer A (Master) and send
>  >> it
>  >>> over RTnet to Layer D (Slave).
>  >>>
>  >>> I want to take the time when the data leaves A and when it arrives D
>  >> with
>  >>> rtdm_clock_read(). If I have the offset between master and slave, I can
>  >>> calculate the time it took to pass A, B, both RTnet stacks, C and D.
>  >>>
>  >>> As RTnet keeps track of the offset in it's stack, I want to use this
>  >> value. Is
>  >>> it possible via the API? I didn't found anything. Or do I have to
>  >> manipulate
>  >>> the stack to pass the offset to the higher layers (C and D)?
>  >> If RTmac/TDMA is in use, you can obtain the clock offset via
>  >> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where "<x>"
>  >> corresponds to "rteth<x>").
>  >>
>  >> Jan
>  >>
>  >
>  > When I try to compile my application (xenomai native skin, user task), I get an error, saying rtdm_driver.h (which contains the prototype for rtdm_clock_read()) is for kernel mode tasks only.
>
>  So you are in userland, obviously...
>
>
>  >
>  > How can I get a timestamp in a xenomai native user task? Is there sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?
>
>  Nope, there is rt_timer_read & Co. for native user-mode applications.
>
>  I was about to remark that everything can be found in the API docs - but
>  it can't! Any volunteer to add the missing functions from native/timer.h
>  to the docs? Maybe there is just something broken for doxygen. TIA!

The problem is that __KERNEL__ is in the list of PREDEFINED macros,
and the comments in native/timer.h are in the !__KERNEL__ section.

-- 
 Gilles

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Gilles Chanteperdrix]

On Wed, May 14, 2008 at 5:17 PM, Gilles Chanteperdrix
<gilles.chanteperdrix@xenomai.org> wrote:
>
> On Wed, May 14, 2008 at 5:04 PM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
>  >
>  > Karl Reichert wrote:
>  >  > Jan Kiszka wrote:
>  >  >> Karl Reichert wrote:
>  >  >>> Hello,
>  >  >>>
>  >  >>> I have two stations running RTnet, one as a master and one as a slave. I
>  >  >> want
>  >  >>> to measure the time, which a message, sent from master to slave, takes.
>  >  >> But,
>  >  >>> I don't want the time of the transmission only, but the complete time of
>  >  >> all
>  >  >>> software layers this data is processed through.
>  >  >>>
>  >  >>> _______________     _______________
>  >  >>> |           A                 |     |               D                 |
>  >  >>> |______________|    |______________|
>  >  >>> |           B                 |     |               C                 |
>  >  >>> |______________|    |______________|
>  >  >>> |   RTnet (Master)    |     |    RTnet (Slave)        |
>  >  >>> |______________|    |______________|
>  >  >>>             |                                 |
>  >  >>>             |______________|
>  >  >>>
>  >  >>> So, I create the data, that should be send, in Layer A (Master) and send
>  >  >> it
>  >  >>> over RTnet to Layer D (Slave).
>  >  >>>
>  >  >>> I want to take the time when the data leaves A and when it arrives D
>  >  >> with
>  >  >>> rtdm_clock_read(). If I have the offset between master and slave, I can
>  >  >>> calculate the time it took to pass A, B, both RTnet stacks, C and D.
>  >  >>>
>  >  >>> As RTnet keeps track of the offset in it's stack, I want to use this
>  >  >> value. Is
>  >  >>> it possible via the API? I didn't found anything. Or do I have to
>  >  >> manipulate
>  >  >>> the stack to pass the offset to the higher layers (C and D)?
>  >  >> If RTmac/TDMA is in use, you can obtain the clock offset via
>  >  >> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where "<x>"
>  >  >> corresponds to "rteth<x>").
>  >  >>
>  >  >> Jan
>  >  >>
>  >  >
>  >  > When I try to compile my application (xenomai native skin, user task), I get an error, saying rtdm_driver.h (which contains the prototype for rtdm_clock_read()) is for kernel mode tasks only.
>  >
>  >  So you are in userland, obviously...
>  >
>  >
>  >  >
>  >  > How can I get a timestamp in a xenomai native user task? Is there sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?
>  >
>  >  Nope, there is rt_timer_read & Co. for native user-mode applications.
>  >
>  >  I was about to remark that everything can be found in the API docs - but
>  >  it can't! Any volunteer to add the missing functions from native/timer.h
>  >  to the docs? Maybe there is just something broken for doxygen. TIA!
>
>  The problem is that __KERNEL__ is in the list of PREDEFINED macros,
>  and the comments in native/timer.h are in the !__KERNEL__ section.

Other headers do #if (defined(__KERNEL__) || defined(__XENO_SIM__)) &&
!defined(DOXYGEN_CPP)
Maybe we should use the same construction in native/timer.h ?

-- 
 Gilles

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Karl Reichert]

> Gilles Chanteperdrix wrote:
> > Jan Kiszka wrote:
> >  > Karl Reichert wrote:
> >  >  > Jan Kiszka wrote:
> >  >  >> Karl Reichert wrote:
> >  >  >>> Hello,
> >  >  >>>
> >  >  >>> I have two stations running RTnet, one as a master and one as a
> slave. I
> >  >  >> want
> >  >  >>> to measure the time, which a message, sent from master to slave,
> takes.
> >  >  >> But,
> >  >  >>> I don't want the time of the transmission only, but the complete
> time of
> >  >  >> all
> >  >  >>> software layers this data is processed through.
> >  >  >>>
> >  >  >>> _______________     _______________
> >  >  >>> |           A                 |     |               D           
>      |
> >  >  >>> |______________|    |______________|
> >  >  >>> |           B                 |     |               C           
>      |
> >  >  >>> |______________|    |______________|
> >  >  >>> |   RTnet (Master)    |     |    RTnet (Slave)        |
> >  >  >>> |______________|    |______________|
> >  >  >>>             |                                 |
> >  >  >>>             |______________|
> >  >  >>>
> >  >  >>> So, I create the data, that should be send, in Layer A (Master)
> and send
> >  >  >> it
> >  >  >>> over RTnet to Layer D (Slave).
> >  >  >>>
> >  >  >>> I want to take the time when the data leaves A and when it
> arrives D
> >  >  >> with
> >  >  >>> rtdm_clock_read(). If I have the offset between master and
> slave, I can
> >  >  >>> calculate the time it took to pass A, B, both RTnet stacks, C
> and D.
> >  >  >>>
> >  >  >>> As RTnet keeps track of the offset in it's stack, I want to use
> this
> >  >  >> value. Is
> >  >  >>> it possible via the API? I didn't found anything. Or do I have
> to
> >  >  >> manipulate
> >  >  >>> the stack to pass the offset to the higher layers (C and D)?
> >  >  >> If RTmac/TDMA is in use, you can obtain the clock offset via
> >  >  >> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where
> "<x>"
> >  >  >> corresponds to "rteth<x>").
> >  >  >>
> >  >  >> Jan
> >  >  >>
> >  >  >
> >  >  > When I try to compile my application (xenomai native skin, user
> task), I get an error, saying rtdm_driver.h (which contains the prototype for
> rtdm_clock_read()) is for kernel mode tasks only.
> >  >
> >  >  So you are in userland, obviously...
> >  >
> >  >
> >  >  >
> >  >  > How can I get a timestamp in a xenomai native user task? Is there
> sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?
> >  >
> >  >  Nope, there is rt_timer_read & Co. for native user-mode
> applications.
> >  >
> >  >  I was about to remark that everything can be found in the API docs -
> but
> >  >  it can't! Any volunteer to add the missing functions from
> native/timer.h
> >  >  to the docs? Maybe there is just something broken for doxygen. TIA!
> >
> >  The problem is that __KERNEL__ is in the list of PREDEFINED macros,
> >  and the comments in native/timer.h are in the !__KERNEL__ section.
> 
> Other headers do #if (defined(__KERNEL__) || defined(__XENO_SIM__)) &&
> !defined(DOXYGEN_CPP)
> Maybe we should use the same construction in native/timer.h ?
> 
> -- 
>  Gilles

Hmm ... what is the difference between rt_timer_read, rt_timer_inquire and rt_timer_tsc? I think rt_timer_inquire is giving the same results (and a little more, the period) like the other two ones, but in one call?

As I understand, TSC is a reliable value on single processor machines with CONFIG_CPU_FREQ, CONFIG_ACPI_PROCESSOR and CONFIG_APM disabled?! So I can use rt_timer_tsc?! Or should I better use rt_timer_read? (Please see above for my UseCase)

Thanks in advance.
Karl
-- 
von Karl Reichert

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
 > On Wed, May 14, 2008 at 5:17 PM, Gilles Chanteperdrix
 > <gilles.chanteperdrix@xenomai.org> wrote:
 > >
 > > On Wed, May 14, 2008 at 5:04 PM, Jan Kiszka <jan.kiszka@domain.hid> wrote:
 > >  >
 > >  > Karl Reichert wrote:
 > >  >  > Jan Kiszka wrote:
 > >  >  >> Karl Reichert wrote:
 > >  >  >>> Hello,
 > >  >  >>>
 > >  >  >>> I have two stations running RTnet, one as a master and one as a slave. I
 > >  >  >> want
 > >  >  >>> to measure the time, which a message, sent from master to slave, takes.
 > >  >  >> But,
 > >  >  >>> I don't want the time of the transmission only, but the complete time of
 > >  >  >> all
 > >  >  >>> software layers this data is processed through.
 > >  >  >>>
 > >  >  >>> _______________     _______________
 > >  >  >>> |           A                 |     |               D                 |
 > >  >  >>> |______________|    |______________|
 > >  >  >>> |           B                 |     |               C                 |
 > >  >  >>> |______________|    |______________|
 > >  >  >>> |   RTnet (Master)    |     |    RTnet (Slave)        |
 > >  >  >>> |______________|    |______________|
 > >  >  >>>             |                                 |
 > >  >  >>>             |______________|
 > >  >  >>>
 > >  >  >>> So, I create the data, that should be send, in Layer A (Master) and send
 > >  >  >> it
 > >  >  >>> over RTnet to Layer D (Slave).
 > >  >  >>>
 > >  >  >>> I want to take the time when the data leaves A and when it arrives D
 > >  >  >> with
 > >  >  >>> rtdm_clock_read(). If I have the offset between master and slave, I can
 > >  >  >>> calculate the time it took to pass A, B, both RTnet stacks, C and D.
 > >  >  >>>
 > >  >  >>> As RTnet keeps track of the offset in it's stack, I want to use this
 > >  >  >> value. Is
 > >  >  >>> it possible via the API? I didn't found anything. Or do I have to
 > >  >  >> manipulate
 > >  >  >>> the stack to pass the offset to the higher layers (C and D)?
 > >  >  >> If RTmac/TDMA is in use, you can obtain the clock offset via
 > >  >  >> RTMAC_RTIOC_TIMEOFFSET from the RTDM device "TDMA<x>" (where "<x>"
 > >  >  >> corresponds to "rteth<x>").
 > >  >  >>
 > >  >  >> Jan
 > >  >  >>
 > >  >  >
 > >  >  > When I try to compile my application (xenomai native skin, user task), I get an error, saying rtdm_driver.h (which contains the prototype for rtdm_clock_read()) is for kernel mode tasks only.
 > >  >
 > >  >  So you are in userland, obviously...
 > >  >
 > >  >
 > >  >  >
 > >  >  > How can I get a timestamp in a xenomai native user task? Is there sth equivalent to rtdm_clock_read()? Or do I have to write a kernel task?
 > >  >
 > >  >  Nope, there is rt_timer_read & Co. for native user-mode applications.
 > >  >
 > >  >  I was about to remark that everything can be found in the API docs - but
 > >  >  it can't! Any volunteer to add the missing functions from native/timer.h
 > >  >  to the docs? Maybe there is just something broken for doxygen. TIA!
 > >
 > >  The problem is that __KERNEL__ is in the list of PREDEFINED macros,
 > >  and the comments in native/timer.h are in the !__KERNEL__ section.
 > 
 > Other headers do #if (defined(__KERNEL__) || defined(__XENO_SIM__)) &&
 > !defined(DOXYGEN_CPP)
 > Maybe we should use the same construction in native/timer.h ?

Fixed in trunk and v2.4.x. Next documentation generation should include
native/timer.h inlines documentation.

-- 


					    Gilles.

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Gilles Chanteperdrix]

Karl Reichert wrote:
 > Hmm ... what is the difference between rt_timer_read, rt_timer_inquire and rt_timer_tsc? I think rt_timer_inquire is giving the same results (and a little more, the period) like the other two ones, but in one call?
 > 
 > As I understand, TSC is a reliable value on single processor machines with CONFIG_CPU_FREQ, CONFIG_ACPI_PROCESSOR and CONFIG_APM disabled?! So I can use rt_timer_tsc?! Or should I better use rt_timer_read? (Please see above for my UseCase)

rt_timer_tsc has less overhead than rt_timer_read (it does not issue a
syscall, and it does not do any conversion).

On the other hand, rt_timer_read gives you an absolute value whose
reference is the well known 1st january 1970.

With rt_timer_inquire, you get rt_timer_tsc and rt_timer_read at the
same time, and you know the relation between the two values.

-- 


					    Gilles.

Re: [Xenomai-help] [RTnet-users] Ideas how to measure delays with RTnet

[Stéphane ANCELOT]

[-- Attachment #1: Type: text/html, Size: 3185 bytes --]