[Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Wolfgang Denk]

Hello all,

DENX Software Engineering is proud to announce Xenomai/SOLO, the new
RTOS emulator that runs as a normal user-space application on a
standard Linux kernel. Of course, when timing requirements have to be
met, a real-time capable Linux kernel with the PREEMPT_RT support
enabled should be used.

Philippe Gerum, the author of the code, writes:

    Some weeks ago, I started laying the groundwork for porting the
    Xenomai emulators natively over the PREEMPT_RT kernel. Unlike the
    co-kernel based Xenomai version, SOLO does not require any kernel
    support from additional modules or patches. It is fully based on
    the standard POSIX library, and runs as a regular process
    controlled by a single image Linux kernel.

    As a first step, a VxWorks emulator has just been rebuilt over
    this new framework.

    This framework is not another window-dressing of the existing
    Xenomai 2.x core, but really something 100% redesigned and
    rewritten for running on a native real-time Linux implementation,
    while keeping the opportunity to adapt it back to a co-kernel
    system. At some point, I'd like to make both the native and
    co-kernel Xenomai incarnations converge on the basis of this new
    core interface for Xenomai 3, so that we would be able to share
    the APIs between them. In short, SOLO is a necessary intermediate
    step on the road to Xenomai 3.

This work has been co-sponsored by the OSADL (http://www.osadl.org)
and DENX Software Engineering (http://www.denx.de).

For details, please see the README at
http://www.denx.de/cgi-bin/gitweb.cgi?p=xenomai-solo.git;a=blob;f=README

You can pull the whole code from
git://www.denx.de/git/xenomai-solo.git  resp.
http://www.denx.de/git/xenomai-solo.git  or browse it at
http://www.denx.de/cgi-bin/gitweb.cgi?p=xenomai-solo.git;a=summary


Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@domain.hid
Quote from the Boss... "I didn't say it was your fault.  I said I was
going to blame it on you."

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Roland Stigge]

Hi,

Wolfgang Denk wrote:
> DENX Software Engineering is proud to announce Xenomai/SOLO, the new
> RTOS emulator that runs as a normal user-space application on a
> standard Linux kernel. Of course, when timing requirements have to be
> met, a real-time capable Linux kernel with the PREEMPT_RT support
> enabled should be used.

Thanks very much for this package!

Is it worth providing a separate xenomai-solo package in Debian right
now (considering it being propagated to Debian 5.0 and supported there
in this form until ca. 2010), or should we wait until Xenomai/SOLO is
integrated into Xenomai mainline?

bye,
  Roland

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Wolfgang Denk]

Dear Roland,

in message <47E62D72.4060306@domain.hid> you wrote:
> 
> Is it worth providing a separate xenomai-solo package in Debian right
> now (considering it being propagated to Debian 5.0 and supported there
> in this form until ca. 2010), or should we wait until Xenomai/SOLO is
> integrated into Xenomai mainline?

Integration into Xenomai mainline means waiting for Xenomai 3, which
seems still a pretty long way to go.

On the other hand, you need a PREEMPT_RT enhanced  Linux  kernel  for
Xenomai/SOLO  to provide real-time behaviour which is probably needed
in most cases when you try and emulate a RTOS.  This  is  probably  a
bigger hurdle?

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@domain.hid
It is clear that the individual who persecutes a  man,  his  brother,
because he is not of the same opinion, is a monster.       - Voltaire

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Roland Stigge]

Hi,

Wolfgang Denk wrote:
>> Is it worth providing a separate xenomai-solo package in Debian right
>> now (considering it being propagated to Debian 5.0 and supported there
>> in this form until ca. 2010), or should we wait until Xenomai/SOLO is
>> integrated into Xenomai mainline?
> 
> Integration into Xenomai mainline means waiting for Xenomai 3, which
> seems still a pretty long way to go.
> 
> On the other hand, you need a PREEMPT_RT enhanced  Linux  kernel  for
> Xenomai/SOLO  to provide real-time behaviour which is probably needed
> in most cases when you try and emulate a RTOS.  This  is  probably  a
> bigger hurdle?

I think an RT-patched kernel is quite common, at least if I consider the
realtime conscious Debian user community. For other realtime
applications such as JACK (though in a completely different application
domain), the respective kernels are also often presumed.

So if we are talking about Xenomai 3 in the order of years, Xenomai/SOLO
as a separate package seems worth considering if it matures reasonably.

The other "desktop suitable" approach is the simulator. Both
Xenomai/SOLO and the simulator are good candidates for inclusion in the
OS distribution since (apart from certain kernel requirements) they can
be provided as universal binary packages in userspace.

The problem with xenosim is its dependency upon gcc-2.95.6 sources.
Since Debian packages must provide the complete sources for building the
binary packages or alternatively/additionally have build dependencies
satisfied by other (binary) Debian packages in the distribution, this is
not an easy task (without gcc 2.95.x being in Debian anymore). One way
to go could be utilizing a newer GCC (4.x) instead of 2.95.x, but that
doesn't work with current xenosim. Is it worth it to put porting effort
into this spot?

Thanks!

bye,
  Roland

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Steven A. Falco]

I'll of course have to make my own tests, but I am curious - do folks
expect that Xenomai/SOLO will be able to equal the interrupt performance
of Xenomai/IPIPE?  I guess my intuition says that the IPIPE approach
would guarantee better interrupt response, but maybe my intuition is
completely wrong.  I'll try to post some results in a few weeks...

    Steve Falco

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Wolfgang Denk]

In message <47EA4F65.7050600@domain.hid> you wrote:
> I'll of course have to make my own tests, but I am curious - do folks
> expect that Xenomai/SOLO will be able to equal the interrupt performance
> of Xenomai/IPIPE?  I guess my intuition says that the IPIPE approach
> would guarantee better interrupt response, but maybe my intuition is
> completely wrong.  I'll try to post some results in a few weeks...

Well, interrupt performance is just one thing.  The  whole  real-time
behaviour  depends on the underlying OS. And frankly, what we've seen
so far means that PREEMPT_RT can deliver "probabilistic" real-time at
best. Take a test case that has been running fine  and  just  put  it
into  a  new environment (like attach it to a different network), and
it will behave differently. Just plug in a new USB device that hasn't
been tested before, and nobody can tell what will happen.

If it comes to  hard,  reliable  real-time  behaviour,  we  recommend
Xenomai/ipipe to all our customers. However, there are some who think
it  is  important  to have an "original, unpatched" kernel.org source
tree. These obviously run for PREEMPT_RT, and SOLO.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd@domain.hid
"You know, after a woman's raised a family and so on,  she  wants  to
start living her own life."   "Whose life she's _been_ living, then?"
                                  - Terry Pratchett, _Witches Abroad_

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Roland Stigge]

Wolfgang Denk wrote:
> If it comes to  hard,  reliable  real-time  behaviour,  we  recommend
> Xenomai/ipipe to all our customers. However, there are some who think
> it  is  important  to have an "original, unpatched" kernel.org source
> tree. These obviously run for PREEMPT_RT, and SOLO.

Please remember that PREEMPT_RT currently doesn't mean "unpatched", so
you could theoretically patch every PREEMPT_RT enhanced kernel with
ipipe instead... ;->

(Feelings of customers are a completely different thing, though...)

Roland

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Philippe Gerum]

Steven A. Falco wrote:
> I'll of course have to make my own tests, but I am curious - do folks
> expect that Xenomai/SOLO will be able to equal the interrupt performance
> of Xenomai/IPIPE?  I guess my intuition says that the IPIPE approach
> would guarantee better interrupt response, but maybe my intuition is
> completely wrong.  I'll try to post some results in a few weeks...
>

SOLO is not about pretending that native preemption should now be the one and
only solution to real-time requirements. The point is that there are different
levels of real-time requirements, different real-time applications, different
real-time environments. Since I just don't believe in the one-fits-it-all
marketroïd message, SOLO is there to bring the Xenomai emulators to the native
preemption world, because a significant portion of the application base to be
migrated to Linux will be just fine in that environment.

FWIW, I like the co-kernel approach for the principle of least surprise it
brings with respect to latency; if the latency tests run fine, it is extremely
likely that deadlines will always be met within the application, provided the
latter behaves properly, even if you upgrade your target kernel. But on the
other hand, I like the native preemption as well, for its ability to keep things
reasonably simple when it comes to port large legacy applications, which used to
rely on zillions of libraries; in that case, the necessary co-kernel/Linux
programming model split is just a massive pain in the neck (i.e. primary &
secondary runtime modes, restriction on using the glibc in real-time mode and so
on).

But obviously, the co-kernel mode based on the I-pipe is here to stay, and the
purpose of Xenomai 3 is to allow the emulators to be usable on top of both
real-time cores (i.e. PREEMPT_RT, or I-pipe + nucleus), using a simple
recompilation. SOLO is an intermediate step, before both approaches are
reconciled with a common emulator code base in Xenomai 3. In other words: two
real-time cores, one set of emulators. Well, that's the plan.

I will send a roadmap to Xenomai 3 asap (take this literally, because of hectic
schedule here), to explain where we are heading to.

>     Steve Falco
> 
> 
> _______________________________________________
> Xenomai-core mailing list
> Xenomai-core@domain.hid
> https://mail.gna.org/listinfo/xenomai-core
> 


-- 
Philippe.

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Steven A. Falco]

Philippe Gerum wrote:
> But obviously, the co-kernel mode based on the I-pipe is here to stay, and the
> purpose of Xenomai 3 is to allow the emulators to be usable on top of both
> real-time cores (i.e. PREEMPT_RT, or I-pipe + nucleus), using a simple
> recompilation. SOLO is an intermediate step, before both approaches are
> reconciled with a common emulator code base in Xenomai 3. In other words: two
> real-time cores, one set of emulators. Well, that's the plan.
>
> I will send a roadmap to Xenomai 3 asap (take this literally, because of hectic
> schedule here), to explain where we are heading to.
>
>   
Thanks for the clarification - I was not sure of the intent for SOLO,
but now it makes sense.  I'm glad that IPIPE is not going away, as my
application will need hard real time.

    Steve

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Philippe Gerum]

Roland Stigge wrote:
> 
> So if we are talking about Xenomai 3 in the order of years, Xenomai/SOLO
> as a separate package seems worth considering if it matures reasonably.
>

It is not in the order of years, but likely weeks, to get the four traditional
RTOS emulators currently available from 2.x running on top of the SOLO core
(Btw, I've just committed the early version of the pSOS emulator to the GIT tree
which still needs heavy testing though).

Before packaging this, I would recommend to wait for significant test suites to
be available for all of them; I guess this should be the case by the end of May,
if the current pace can be sustained.

> The other "desktop suitable" approach is the simulator. Both
> Xenomai/SOLO and the simulator are good candidates for inclusion in the
> OS distribution since (apart from certain kernel requirements) they can
> be provided as universal binary packages in userspace.
> 
> The problem with xenosim is its dependency upon gcc-2.95.6 sources.
> Since Debian packages must provide the complete sources for building the
> binary packages or alternatively/additionally have build dependencies
> satisfied by other (binary) Debian packages in the distribution, this is
> not an easy task (without gcc 2.95.x being in Debian anymore). One way
> to go could be utilizing a newer GCC (4.x) instead of 2.95.x, but that
> doesn't work with current xenosim. Is it worth it to put porting effort
> into this spot?
> 

The simulator is really a nice tool, but its code instrumenter was written ten
years ago as a patch against egcs-1.0, then subsequently upgraded a number of
times, up to 2.95.3. Unfortunately, upgrading this patch to gcc 3.x or even 4.x
would be a very significant work, because I hooked the instrumentation engine
right into the C/C++ parser stage, and this area of code has been very deeply
refactored over the years (actually, even replaced IIRC).

To my knowledge, only two people "officially" use the simulator, specifically
for debugging the Xenomai core, i.e. Gilles and myself, so there is not much
traction on this side. We just grab some outdated box on which we can rebuild
xenosim, and we live with it this way.

I feel it's sort of a shame, since IMHO, event-driven simulation is just the
ultimate way for efficient debugging and modelling real-time behaviour, but
well, it would really be a project on its own to revamp it, and lack of time
bites everyone.

-- 
Philippe.

Re: [Xenomai-core] Xenomai/SOLO - RTOS emulation for standard Linux

[Philippe Gerum]

Philippe Gerum wrote:
> Roland Stigge wrote:
>> So if we are talking about Xenomai 3 in the order of years, Xenomai/SOLO
>> as a separate package seems worth considering if it matures reasonably.
>>
> 
> It is not in the order of years, but likely weeks, to get the four traditional
> RTOS emulators currently available from 2.x running on top of the SOLO core

Just to prevent any confusion:

- Xenomai/SOLO as a Debian package would make sense in a few weeks
- Xenomai 3 is a matter of months, given that a significant part of the job will
 be directly available from the SOLO effort (i.e. have the RTOS emulators run in
user-space natively).

-- 
Philippe.