RFR (S): 8155810: Bound the number of root region scan threads to the number of survivor regions

[Thomas Schatzl]

Hi all,

  can I have reviews for this change that originated from the
discussion about JDK-8150393 where it has been found
that G1ConcurrentMark::scan_root_regions() always starts all threads
for scanning the root regions independent of the number of root regions
available. This potentially leads to starting (and needing to
synchronize for) to too many threads on huge machines (and on smaller
ones if the conditions are met of course).

CR:
https://bugs.openjdk.java.net/browse/JDK-8155810
Webrev:
http://cr.openjdk.java.net/~tschatzl/8155810/webrev/
Testing:
jprt

Thanks,
  Thomas