The method G1CollectedHeap::determine_start_concurrent_mark_gc() basically calls G1Policy::decide_on_conc_mark_initiation(), with the difference mostly because the former also checks whether we are currently terminating the VM (and decide that we should not start a concurrent mark then).

This is apparently to avoid publishing _cm_thread in a getter (for ConcurrentMarkThread::should_terminate); the suggested fix is to merge both methods and provide the result of ConcurrentMarkThread::should_terminate() in G1CollectedHeap.

I think this is an okay change because
  * ConcurrentMarkThread::should_terminate is just another factor for determining whether to start marking
  * ConcurrentMarkThread::should_terminate is actually asked for in another place too
  * it helps with avoiding publishing G1CollectedHeap::_cm_thread for the move of young collector code to separate files