It turns out that idr_for_each_entry is really slow compared to just
iterating over the slots. Based on measurements the difference is
estimated to be about a factor 64. That means using idr_for_each_entry
is only worth it with very few allocated events.