hi precision timing

[Gregory.D.Menke.1 at gsfc.nasa.gov]

 > > 
 > > We're using a mcp750, and would like to get at some of the timers
 > > available in the chipset.  After wading thru the openpic code, I found
 > > a few functions that look nice for that purpose.
 > > 
 > > - does RTEMS use any of the Raven timers?  If so, are any free for
 > >   use?
 > 
 > The BSP seems to be using the decrementer register for the clock
 > tick.  I don't know what else would be using any other timers.

Well I'll give the 2nd timer a shot...  ;)

 > > - Our application needs an incrementing counter for timing at a
 > >   resolution of 50 to 100us, and we would prefer to poll it rather
 > >   than have an interrupt.  Is there an RTEMS kernel feature that would
 > >   expedite setting up this kind of thing?  A fast isr that increments
 > >   an int somewhere would be fine- I'm mostly wondering if this
 > >   approach is reasonable.
 > 
 > You could certainly do this but I would think the overhead is worse
 > than simply reading the chip when you want to know something.  How
 > often are you going to read it?  

The counter should run for an extended period (many minutes).  With
the ~8mhz clock, the Raven counters will roll over in a couple
minutes.  So, I was thinking a fast interrupt routine incrementing an
int counter via a Raven timer set to roll over at ~100us would scale
things well.  Our application isn't adversely affected by additional
kernel overhead, as long as its consistent.


 >  
 > > - Does RTEMS prefer PCI work be done via some internal API, or is
 > >   direct access down on the bare metal OK?
 > 
 > The RTEMS PCI code is minimal at the moment.  As with all things,
 > it is better to stick to the API even if you have to augment it.
 > That way the same code will work on another board/CPU easier.

Cool.  Thanks, it may be coming in handy shortly.  We're getting some
driver code for a CPCI serial bus tranceiver...

Thanks Joel,

Gregm