gumstix

Pascal,

let me know if you'd like write-access to the Gumstix svn  
repository.  We'd be happy to host your software through our site.   
This is a great writeup -- I look forward to following your progress!

C

On Jun 2, 2005, at 8:43 AM, pascal.brisset-ml@... wrote:

> Hi,
>
> I am releasing a timer-based PWM signal generator for the PXA255.
> In a nutshell, this should allow you to control any (non-PCM)
> RC model with only three wires between a Gumstix and the legacy
> FM RC receiver: GND, +5V, and multiplexed PWM.
>
> For source code, usage examples and related software (including
> remote control with a PC joystick over Bluetooth), see:
> "Cheap Robotic Microhelicopter HOWTO"
> http://perso.wanadoo.fr/pascal.brisset/chromicro/doc/ 
> chromicro.html#software
>
> Jitter is acceptable now that I have switched from using an IRQ
> to a PXA255 FIQ.  It might be possible to bring it down to less
> than 1 microsecond worst case with additional tricks.
> Does anyone here have experience with locking the instruction cache
> of the PXA255 ?  Especially with ensuring that the locking code is
> in a non-cacheable region, as required by the documentation ?

I didn't say anything when you originally posted your implementation of
a timer-based PWM generator for the Gumstix.  At the time, I had almost
finished a multiplexed PWM generator (actually, positive-modulation 2-8
channel PPM) based on the PXA's built-in PWM generators, but it wasn't
working yet.  Now it is.  I've tested it using a Hitec Laser 4 model
aircraft RC transmitter and a homemade signal inverter.  Yes, that means
the gumstik is on the ground and not in the air.

I am using a GPIO to feedback the PWM signal to the gumstix and get
rising-edge interrupts from it.  On the interrupt, I update the period
and duty cycle registers for the affected PWM line to the values for the
next pulse.  Because the hardware controls the timing for both the
rising and falling edges of the square waves, there is no jitter at all.
I am using an IRQ on the GPIO rather than an FIQ, since the ISR has a
realtime deadline of at least 700 microseconds (usually longer).  The
PWM source timer is disabled after the last channel is output, and a
20ms periodic OS timer restarts the pulse train.

My driver is rough around the /dev interface.  Pascal, which major
device number are you using?  Are you dynamically allocating them?  I
found that the busybox mknod utility does not support the >8-bit device
numbers found in 2.6 kernels, so I just picked one that wasn't occupied
by an active driver in my particular configuration :/

The only depmod I could find in the buildroot was a perl script.  I have
set up my code to be built as a third-party kernel driver rather than a
patch to the main kernel sources.  How can I install this driver such
that modprobe will work, rather than manually running insmod?

Thanks,
-Jonathan

http://www4.ncsu.edu/~jdbrandm/gumstix-pwm.tar.gz

Jonathan Brandmeyer writes:
 > I am using a GPIO to feedback the PWM signal to the gumstix and get
 > rising-edge interrupts from it.

Nice trick.  Indeed this should give you two jitter-free signals.

 > My driver is rough around the /dev interface.  Pascal, which major
 > device number are you using?

Mine is configurable with a module parameter (dev_major).
Default is 60, as permitted by linux/Documentation/devices.txt.

 > How can I install this driver such
 > that modprobe will work, rather than manually running insmod?

Just add "/tmp/multi-pwm.ko:" (or whatever the path is) at the
end of /lib/modules/linux-X.Y.Z/modules.dep.
You may want to update /lib/modules/linux-X.Y.Z/modules.alias too.

Cheers,

-- Pascal

On Jun 9, 2005, at 4:08 AM, pascal.brisset-ml@... wrote:

> Jonathan Brandmeyer writes:
>
>> I am using a GPIO to feedback the PWM signal to the gumstix and get
>> rising-edge interrupts from it.
>>
>
> Nice trick.  Indeed this should give you two jitter-free signals.
>
>
>> My driver is rough around the /dev interface.  Pascal, which major
>> device number are you using?
>>
>
> Mine is configurable with a module parameter (dev_major).
> Default is 60, as permitted by linux/Documentation/devices.txt.
>
>
>> How can I install this driver such
>> that modprobe will work, rather than manually running insmod?
>>
>
> Just add "/tmp/multi-pwm.ko:" (or whatever the path is) at the
> end of /lib/modules/linux-X.Y.Z/modules.dep.
> You may want to update /lib/modules/linux-X.Y.Z/modules.alias too.

Guys, if either of you would like me to merge your drivers into the  
kernel tree in the buildroot (and even send them upstream to the  
mainline kernel), pass me the code -- over the next week I'm  
travelling, and will be spending a lot of time heads-down on  
buildroot improvements and updates.

C

PS If anyone's in the Toledo, OH area and wants to say hi, let me  
know -- I'll be there from monday through saturday next week.