Am 14.11.2012 um 08:19 schrieb John Morris:

> On 11/12/2012 11:27 PM, John Morris wrote:
>> I'm going back to try to recreate your setup.  Found your kernel config
>> file.  If you have one around, please send a LinuxCNC build log to help
>> guess what the differences might be.
> 
> While sitting around waiting for the umpteenth time for the Xenomai 
> kernel to build, this time with Michael's config just to help narrow 
> things down, I compiled up LinuxCNC xenomai-user.
> 
> Latency-test works.  And on the same kernel that, combined with 
> xenomai-kernel, hangs!  64-bit uniprocessor running several glxgears 
> (why's that the test we always use?), loading up Firefox with a buggy 
> load of saved tabs, compiling that same xenomai kernel and spinning 
> shapes around in FreeCAD, latency-test reported 7ms/6ms jitter.

That result both Kent and me are puzzled about.. conventional wisdom would 
imply 'kernel threads might have lower latency'. 

The really hard question for the future (LinuxCNC3) is: does it make sense to 
support kernel threads builds at all? if the above results turn out to be 
reproducible, the answer to this might be 'sorry, to gain exactly _what_?' 
which opens a wide range of options to simplify things in LinuxCNC3

> Also, Michael told me today that xenomai-user will also compile for 
> Raspberry Pi, one of which my friend dropped in my lap a few weeks ago 
> and which I had no idea what to do with until now.

note operative word 'compile':

I didnt get the GUI's to run because of some Python OpenGL support missing, 
this could be me or support for OpenGL on the Pi; likely me. runtests is ok.

what does work nicely is HAL+RTAPI; also I cooked up a primitive Pi GPIO driver 
(src/hal/drivers/hal_gpio) which wiggles pins just fine. Now if only motion had 
a sane message interface we already would have a working 'RT outboard' by now.

With xenomai as RT base, the range of architectures is in fact wide open:

http://www.xenomai.org/index.php/Embedded_Device_Support

the architecture dependent code in src/rtapi/rtapi_bitops.h can nowadays be 
handled by GCC builtin atomic operations which obviates the need for assembly, 
this is enabled by --enable-gcc-atomic-ops and forced by 
--with-platform=raspberry . That should in fact compile on any washer & dryer 
supported by gcc.

I need to write up the status, and the pre-merge laundry list. 

The viability of PCI I/O in userthreads builds is still pending verification. 
The upside of doing this: debugging RT code with gdb like any other user 
program simplifies things drastically.

- Michael


> 
> Woo hoo!  What a wonderful turn of events after a crummy week.
> 
>       John
> 
> ------------------------------------------------------------------------------
> Monitor your physical, virtual and cloud infrastructure from a single
> web console. Get in-depth insight into apps, servers, databases, vmware,
> SAP, cloud infrastructure, etc. Download 30-day Free Trial.
> Pricing starts from $795 for 25 servers or applications!
> http://p.sf.net/sfu/zoho_dev2dev_nov
> _______________________________________________
> Emc-developers mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/emc-developers


------------------------------------------------------------------------------
Monitor your physical, virtual and cloud infrastructure from a single
web console. Get in-depth insight into apps, servers, databases, vmware,
SAP, cloud infrastructure, etc. Download 30-day Free Trial.
Pricing starts from $795 for 25 servers or applications!
http://p.sf.net/sfu/zoho_dev2dev_nov
_______________________________________________
Emc-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/emc-developers

Reply via email to