From: Ford Sleeman <wcsleeman@vc...> - 2006-05-23 02:21:42
From reading the Wiki it looks like some people have been able to
generate PWM signals from software using a Gumstix board. To those who
have tried that, what kind of timing accuracy can you get?
Virginia Commonwealth University
From: Pascal A. Brisset <pascal.brisset-ml@wa...> - 2006-05-23 14:15:34
Ford Sleeman wrote:
> From reading the Wiki it looks like some people have been able to
> generate PWM signals from software using a Gumstix board. To those who
> have tried that, what kind of timing accuracy can you get?
I posted these results about interrupt latency on 2005-10-30.
Since then I have improved the handler a little. Average jitter
is pretty good but worst-case is still around 4 microseconds.
| I get 2 uS 95 % of the time and 4 uS worst case on a heavily loaded
| 200 MHz gumstix. When idle, the system does 500 nS 95 % of the time.
| This requires using a FIQ (fast interrupt) with a handler locked into
| the instruction cache. With a regular Linux IRQ, worst case latency
| is above 100 uS.
| These measurements are done entirely in software by comparing
| the value of OSCR when entering the handler with the value that
| was written to OSMR to schedule the interrupt. This might be
| inaccurate or suboptimal because the peripheral bus is involved
| (both when signaling the interrupt and when reading OSCR).
| Performance might be much better with the memory bus.