JDK-8280940 fixed up os.release_multi_mappings_vm to make its handling of multiple neighboring reservations less racy, but I forgot to fix up the reserve_multiple() function, which follows the same pattern and is still racy. It reserves an area, releases it, then re-reserves stripes in that area.

We see problems on Linux PPC, which makes sense because we have here 64k pages, and therefore the stripe size is large, and we have a higher chance of someone grabbing that address space concurrently between the release and the stripe-wise re-reserve.