RFR: Humongous regions should support explicit pinning

[Roman Kennke]

Am 29.11.2017 um 20:12 schrieb Aleksey Shipilev:
> http://cr.openjdk.java.net/~shade/shenandoah/humongous-pinning/webrev.01/
> 
> Another stepping stone for humongous object moves: we need to handle pinning explicitly. That is,
> when humongous regions moves are allowed, humongous-pinned should still be unmovable. We can handle
> the humongous_start-s for simplicity, leaving continuations implicit, as pinning/unpinning
> continuations is the logic error anyway. Experimental option reverts the humongous behavior back to
> "legacy" behavior. It is now "false", but subsequent changes would flip that option to "true".
> 
> As the added benefit, this also mechanically checks that no pinned humongous region can be reclaimed
> without first being unpinned -- something that is implicitly handled by retaining the reference to
> that humongous region object via JNI handle, but still fragile in my mind.
> 
> Testing: hotspot_gc_shenandoah (fastdebug)
> 
> Thanks,
> -Aleksey
> 

Good.
If we keep adding states like this, we need a 3D-ASCII-Flowchart ;-)

Roman