We use planning poker when we do backlog planning/grooming, and also when we do
task breakdown for a sprint.

Some of our projects are "fixed" scope (we know things will change, but to
provide estimates for time/price the requirements backlog is estimated, and a
high-level sprint/release plan is developed. In this case we use planning poker
to estimate the entire backlog. (We recognized that this approach generates some
waste, since we know the backlog will change, but this is expected with "fixed"
scope projects. The backlog actually provides a strong mechanism for managing
change, which is essential with this type of project.)

If the project isn't "fixed", we still estimate the backlog, but only the
highest priority items required to consume our capacity (as predicted by
velocity) for a sprint. That is, in these cases we don't estimate the entire
backlog.

During sprint planning we break items into tasks, and use planning poker to
estimate the tasks. We use the same measurement (story points) to estimate both
the backlog and tasks, so we can sum up our task estimates to help validate that
we weren't way off track when we estimated the requirement when we did the
estimate while it was on the backlog. If we were the sprint planning meeting is
the best time to identify and expose this problem. Transparency is critical to
project success.

The entire team participates in both the backlog and sprint planning meetings.
In some cases the full team has not been identified for a project (responding to
RFPs, for example), the identified subset of the proposed team will play
planning poker.

As noted by others in this thread, when planning poker takes a long time it
usually means requirements are unclear, so the discussion is very important. We
also often find that initial planning takes quite a bit of time, but that the
amount of time required normalizes as the team does (moves from
forming->storming->norming).

I always try to remind people that Scrum is simple, but it's not easy. Have some
patience and diligence in applying planning poker, and I think you'll see the
time required normalize and the benefits of the process really manifest itself.