We are troubleshooting illegal-address issues in production software. As part of this effort, we are looking at whether our hardware may have "noise" on the NMI input. This input pin is disbled by pulling up with a resistor (it defaults to acting on the falling edge, which is how we leave it in software). But we thought that maybe electrical noise could be coupling from other traces onto the NMI pullup line and we want to know what might happen if it did.....
So, we wrote a software NMI handler, which toggles an unused port bit if the software sees an NMI. We are seeing that we we inject intentional pulses on the NMI input, we sometimes get more than one toggle per active (falling) edge of NMI. We were careful to have a clean edge, so we are trying to understand how we might possibly see more than one NMI event per NMI pin falling edge. In these events, most of the toggles occur a certain time after the falling edge but sometimes we see an additonal toggle some microseconds later.
Has anyone used NMI routines that act and return (as opposed to just going to reset the processor, which for example , might be a good thing to do if the NMI were being used to annnounce imminent power failure)?
Any ideas on how/why the NMI might give software more than 1 event per NMI pin signal edge?
Any more suggestions, thanks!