JDK-8303215 added a workaround to deal with too-eager THP page coalescation for thread stacks that could lead to a much higher thread stack use than planned.

The workaround helps a lot already, but it is still incomplete in light of JDK-8310233.

We do this workaround even if unnecessary (when THPs are disabled on the system). And we may use the wrong page size when multiple *static* hugepage sizes are configured on the system, which may make the fix ineffectual for java threads (e.g. THP assumes lp size is 1G, causing the condition in os::create_thread to fail).

I am working on a solution for JDK-8310233. I plan to separate cleanly, at the os::Linux namespace, *static* hugepages from THPs such that we can query the THP page size and mode independently from the large-page-abstraction in the os namespace. Once done, the JDK-8303215 workaround can be altered to use THP state and THP page size instead.