In my experience with similar things, the idea is that your code gets called (via signal or callback) when the program dips into the emergency pool ($^M). It should then immediately rummage around and try to free stuff, then try to reallocate the emergency pool. If it fails, then you're probably doomed.

Unfortunately, I don't see a relevant callback for $^M. If so, then there's no way to tell when $^M has been used, and I don't see how it can be useful.