Functional:
  - dholmes CR:
    - Rename set_owner_from() to release_clear_owner_with_barrier(), drop
      the 'new_value' parameter and add a 'needs_fence' parameter; change
      logic from cmpxchg() to release_store() followed by fence() or a
      storeload() depending on 'needs_fence'.
    - Update the comments to explain why these two "Drop the lock" sites
      have different post release_store() barrier requirements.
  - eosterlund CR:
    - Fix FastHashCode() races with install_displaced_markword_in_object().
  - rehn CR:
    - Dropped volatile from fields and now using Atomic ops to keep the
      compiler from messing with them:
      - ObjectMonitor::_next_om and _ref_count
      - Thread::om_free_list, om_free_count, om_in_use_list and om_in_use_count
    - Dropped volatile from global variables:
      - g_block_list, g_free_list, g_om_in_use_list, g_wait_list,
        g_om_free_count, g_om_in_use_count, g_om_population, g_om_wait_count
    - Dropped volatile from function parameters that were passed
      previously volatile fields or volatile global vars:
      - deflate_monitor_list() and deflate_monitor_list_using_JT()
      - prepend_list_to_common(), prepend_to_common(),
        take_from_start_of_common()
    - Moved global variables into ListVars struct to put them on
      their own cache lines (also renamed):
      - g_free_list       -> LVars.free_list
      - g_om_in_use_list  -> LVars.in_use_list
      - g_wait_list       -> LVars.wait_list
      - g_om_free_count   -> LVars.free_count
      - g_om_in_use_count -> LVars.in_use_count
      - g_om_population   -> LVars.population
      - g_om_wait_count   -> LVars.wait_count
      - Rename some related functions:
        - prepend_list_to_g_om_in_use_list() -> prepend_list_to_global_in_use_list()
        - prepend_list_to_g_free_list() -> prepend_list_to_global_free_list()
        - prepend_list_to_g_wait_list() -> prepend_list_to_global_wait_list()
        - take_from_start_of_g_free_list() -> take_from_start_of_global_free_list()
    - Make the use of (spin) locking clear instead of calling it "marking"
      and clean up the use of Atomic ops:
      - Add "#define OM_LOCK_BIT 0x1" and switch literal 0x1 uses for clarity.
      - Rename is_next_marked() -> is_locked(); now uses Atomic::load()
        instead of Atomic::load_acquire().
      - Rename mark_next() -> try_om_lock(); now returns ObjectMonitor*
        instead of bool and drop the 'next' return parameter; now uses
        Atomic::load() to get the initial _next_om value.
      - Rename mark_next_loop() -> om_lock() and drop return of 'next'
        ObjectMonitor*.
      - set_next() now calls Atomic::store() instead of Atomic::release_store().
      - Add om_unlock() to unlock an ObjectMonitor*; uses Atomic::load()
        to get the initial _next_om value and Atomic::store() to update.
      - Rename mark_list_head() -> get_list_head_locked(); now returns
        ObjectMonitor* instead of bool and drop the 'next' return parameter;
        now uses Atomic::load() to get the list head values.
      - unmarked_next() now calls Atomic::load() instead of
        Atomic::load_acquire().
      - Rename mark_next_for_traversal() -> lock_next_for_traversal()
        and move it closer to the code that uses it.
    - List heads should be read with Atomic::load() and updated with
      Atomic::store() (when not updated with cmpxchg()).
    - Get rid of unnecessary OrderAccess::storestore() calls.
    - deflate_monitor_list_using_JT() and deflate_common_idle_monitors_using_JT()
      should call SafepointMechanism::should_block() instead of
      SafepointSynchronize::is_synchronizing().

Test update:
  - no changes

Collateral:
  - eosterlund CR:
    - Tighten save_om_ptr() mark parameter guarantee.

Cleanup:
  - dholmes CR:
    - Delete errant comments about MO_ACQ_REL and MO_SEQ_CST.
  - self review:
    - local variables that exist just for an assert() should
      be "#ifdef ASSERT...#endif" instead of DEBUG_ONLY(...).
    - correct description of AsyncDeflateIdleMonitors to
      "Deflate idle monitors using the ServiceThread."
  - rehn CR:
    - Add comments to ObjectMonitorHandle associated with
      Thread::current_pending_monitor() and Thread::current_waiting_monitor()
      calls to clarify why the ObjectMonitorHandle is there.
    - Clarify misc comments.
    - Revert increment of contentions to baseline's Atomic::inc() 
      instead of Atomic::add() since we no longer care about the
      return value; missed that change when ref_count was added.
    - Separate sections in synchronizer.cpp for "Spinlock functions"
      and "List Management functions".

Temporary:
  - dholmes CR:
    - Cleanup temporary debug code in ObjectMonitor::exit() for the
      "Non-balanced monitor enter/exit!" condition.