With most ARM processors you can also use the DWT to set hardware watchpoints on data accesses. The advantage of that approach is that it runs at full speed and you don't have to instrument your code so you can set guards around your stack and the debugger/trace unit catches bad accesses.
Interesting. I believe this is the same as the hardware watchpoint I mentioned. I'm not usually working with ARMs, so I do not know for sure.
How would you deal with multiple tasks running? Do you have multiple watchpoints you can set with an ARM, or would the watch space need to be changed when switching tasks.
It's at the lowest level so it's unaware of multiple tasks. If there are multiple stacks you'd have to switch. At that point an MMU would be more suited to the job but of course that's a whole different environment.