Race between ciEnv::register_method and nmethod::make_not_entrant_or_zombie

Tests nsk/jvmti/RedefineClasses/StressRedefine and nsk/jvmti/RedefineClasses/StressRedefineWithoutBytecodeCorruption do the following: class methods being invoked aggressively while class itself being redefined with jvmti with multiple threads. These tests hang sometimes. pstack shows that hotspot is working and doing some compiler work that never ends.

Steps to reproduce: go to sc11136470.us.oracle.com, cd to /export/local/repr1 and run "repr.sh" script in background. Wait until test run takes more than couple of minutes. It took 9 attempts to hang last time.

I've launched script (it's dump_loop.sh in the same dir) that launches pstack every 5 seconds. Results can be found in "10029" dir. Several core files can be found there as well. Also I'll attach several pstack snapshots here.

RULE nsk/jvmti/RedefineClasses/StressRedefineWithoutBytecodeCorruption Timeout none
RULE nsk/jvmti/RedefineClasses/StressRedefine Timeout none

Build that I used:
java version "1.8.0-ea-fastdebug"
Java(TM) SE Runtime Environment (build 1.8.0-ea-fastdebug-b102)
Java HotSpot(TM) Server VM (build 25.0-b44-fastdebug, mixed mode)