Personas aren't about a particular system so much as they are about a need that users have and the way(s) they currently address it. Your questions depend on and need to fit into the problem-space.
For example, suppose you are in the research department of an auto manufacturer trying to design the next great hit. You might ask questions about the last time each person bought a car, such as:
- What factors were most important to you? (Follow-up questions should emerge from this answer.)
- What kind of research did you do before going shopping?
You might ask questions about the car itself, like:
- What are all the ways you use your car? (Commuting, hauling four kids around, camping trips, etc.)
- What do you especially like about this car?
- What bugs you about this car? (These last two questions should prompt more discussion.)
From these kinds of discussions you might learn that Bob is a commuter who cares about fuel efficiency not so much because of price but because refueling frequently is a hassle, so maybe he would accept a larger gas tank instead of better MPG. You might also learn that Bob goes camping for two weeks every summer, so while he wants a small car to fit into those tight city parking spots, he needs to occasionally cram a lot of stuff into it, and you might think about flexible options with fold-down seats and the like. That's just one user; you'll talk to a bunch, learn each of their stories, assess what they have in common (affinity groups), and from that data you'll start to build your personas.