RFR (XXS): 8156660: JDK-8150393 does not set _scan_in_progress properly

[Thomas Schatzl]

Hi all,

  could I have reviews for this bug introduced in JDK-8150393 that
unconditionally sets G1CMRootRegions::_scan_in_progress to true when
preparing for root region scan?

The change causes some perf issue (since we are starting up threads
doing nothing), but with JDK-8155810 this causes a guarantee to fail (I
noticed this issue when testing JDK-8155810).

CR:
https://bugs.openjdk.java.net/browse/JDK-8156660
Webrev:
http://cr.openjdk.java.net/~tschatzl/8156660/webrev
Testing:
jprt (running), local testing

Thanks,
  Thomas