While working on JDK-8256641 we found that the code could be brushed up a bit:

- improve handling of VM_GC_Operations that do not participate in the operation skipping protocol like VM_GC_HeapInspection: either let them inherit from VM_GC_Sync_Operation (and add the check that GC must have been initialized when calling this, see VM_GC_Operation::doit_prologue()), or try to change the order of constructor parameters so that not every one of those needs to add an "this is ignored" comment

- maybe GCCause can be made an enum class

- some of those operation skipping VM_GC_Operations seem to duplicate the _full member

- some of those operation skipping  VM_GC_Operations pass a GCCause in the VM_GC_Operation but then do not use the _gc_cause member

- use initializer lists

- documentation/indentation style (single space before visibility modifier, column aligned member names, ...) could be brushed up

- not all VM_GC_Operations are in that file (but maybe ignore this)

- those not operation-skipping VM_Operations kind of have the same helper method to do the gc e.g.

bool VM_GC_HeapInspection::collect() {
  if (GCLocker::is_active()) {
    return false;
  }
  Universe::heap()->collect_as_vm_thread(GCCause::_heap_inspection);
  return true;
}

[...]

Universe::heap()->ensure_parsability(...);
if (!collect()) {
  log_warning(gc)(...
}

Maybe there could be a base class for this case unifying this.

Probably other issues. Create new issues as appropriate.