[sh/jdk8u] RFR [backport] 8232747: Shenandoah: Concurrent GC should deactivate SATB before processing weak roots

[Zhengyu Gu]

This is *not* straight forward backport, as sh/jdk8u has yet adopted 
parallel_cleaning() from upstream, which is too risky to backport, IMHO.

The key point of this patch is that, we should deactivate SATB before 
processing weak roots.

Original bug: https://bugs.openjdk.java.net/browse/JDK-8232747
Original patch: https://hg.openjdk.java.net/jdk/jdk/rev/ef2b75750838

sh/jdk8u webrev: 
http://cr.openjdk.java.net/~zgu/shenandoah/JDK-8232747-8u/webrev.00/

Test:
   hotspot_gc_shenandoah Linux x86_64 and x86_32

Thanks,

-Zhengyu