diff --git a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.inline.hpp
index f3ce9418d35de..58522871338b7 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.inline.hpp
@@ -67,8 +67,7 @@ inline oop ShenandoahBarrierSet::load_reference_barrier_mutator(oop obj, T* load
 
   oop fwd = resolve_forwarded_not_null_mutator(obj);
   if (obj == fwd) {
-    assert(_heap->is_evacuation_in_progress(),
-           "evac should be in progress");
+    assert(_heap->is_evacuation_in_progress(), "evac should be in progress");
     Thread* const t = Thread::current();
     ShenandoahEvacOOMScope scope(t);
     fwd = _heap->evacuate_object(obj, t);
@@ -86,8 +85,8 @@ inline oop ShenandoahBarrierSet::load_reference_barrier(oop obj) {
   if (!ShenandoahLoadRefBarrier) {
     return obj;
   }
-  if (_heap->has_forwarded_objects() &&
-      _heap->in_collection_set(obj)) { // Subsumes null-check
+  if (_heap->has_forwarded_objects() && _heap->in_collection_set(obj)) {
+    // Subsumes null-check
     assert(obj != nullptr, "cset check must have subsumed null-check");
     oop fwd = resolve_forwarded_not_null(obj);
     if (obj == fwd && _heap->is_evacuation_in_progress()) {
@@ -381,7 +380,7 @@ void ShenandoahBarrierSet::arraycopy_work(T* src, size_t count) {
   // this barrier will be called with ENQUEUE=true and HAS_FWD=false, even though the young generation
   // may have forwarded objects. In this case, the `arraycopy_work` is first called with HAS_FWD=true and
   // ENQUEUE=false.
-  assert(HAS_FWD == _heap->has_forwarded_objects() || (_heap->gc_state() & ShenandoahHeap::OLD_MARKING) != 0,
+  assert(HAS_FWD == _heap->has_forwarded_objects() || _heap->is_concurrent_old_mark_in_progress(),
          "Forwarded object status is sane");
   // This function cannot be called to handle marking and evacuation at the same time (they operate on
   // different sides of the copy).
@@ -418,7 +417,7 @@ void ShenandoahBarrierSet::arraycopy_barrier(T* src, T* dst, size_t count) {
     return;
   }
 
-  int gc_state = _heap->gc_state();
+  char gc_state = ShenandoahThreadLocalData::gc_state(Thread::current());
   if ((gc_state & ShenandoahHeap::EVACUATION) != 0) {
     arraycopy_evacuation(src, count);
   } else if ((gc_state & ShenandoahHeap::UPDATEREFS) != 0) {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSetClone.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSetClone.inline.hpp
index 3b7bf9864deb0..8b83cc6b32cf9 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSetClone.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSetClone.inline.hpp
@@ -95,8 +95,7 @@ void ShenandoahBarrierSet::clone_barrier(oop obj) {
   assert(ShenandoahCloneBarrier, "only get here with clone barriers enabled");
   shenandoah_assert_correct(nullptr, obj);
 
-  int gc_state = _heap->gc_state();
-  if ((gc_state & ShenandoahHeap::EVACUATION) != 0) {
+  if (_heap->is_evacuation_in_progress()) {
     clone_evacuation(obj);
   } else {
     clone_update(obj);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
index b1c2e72ef82b6..bd703cdb96f00 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahConcurrentGC.cpp
@@ -192,8 +192,14 @@ bool ShenandoahConcurrentGC::collect(GCCause::Cause cause) {
       return false;
     }
 
+    // Evacuation is complete, retire gc labs
+    heap->concurrent_prepare_for_update_refs();
+
     // Perform update-refs phase.
-    vmop_entry_init_updaterefs();
+    if (ShenandoahVerify || ShenandoahPacing) {
+      vmop_entry_init_updaterefs();
+    }
+
     entry_updaterefs();
     if (check_cancellation_and_abort(ShenandoahDegenPoint::_degenerated_updaterefs)) {
       return false;
@@ -748,10 +754,6 @@ void ShenandoahConcurrentGC::op_final_mark() {
           heap->verifier()->verify_after_concmark();
         }
       }
-
-      if (VerifyAfterGC) {
-        Universe::verify();
-      }
     }
   }
 }
@@ -920,8 +922,8 @@ class ShenandoahConcurrentWeakRootsEvacUpdateTask : public WorkerTask {
     }
 
     // If we are going to perform concurrent class unloading later on, we need to
-    // cleanup the weak oops in CLD and determinate nmethod's unloading state, so that we
-    // can cleanup immediate garbage sooner.
+    // clean up the weak oops in CLD and determine nmethod's unloading state, so that we
+    // can clean up immediate garbage sooner.
     if (ShenandoahHeap::heap()->unload_classes()) {
       // Applies ShenandoahIsCLDAlive closure to CLDs, native barrier will either null the
       // CLD's holder or evacuate it.
@@ -947,21 +949,10 @@ void ShenandoahConcurrentGC::op_weak_roots() {
   ShenandoahHeap* const heap = ShenandoahHeap::heap();
   assert(heap->is_concurrent_weak_root_in_progress(), "Only during this phase");
   // Concurrent weak root processing
-  {
-    ShenandoahTimingsTracker t(ShenandoahPhaseTimings::conc_weak_roots_work);
-    ShenandoahGCWorkerPhase worker_phase(ShenandoahPhaseTimings::conc_weak_roots_work);
-    ShenandoahConcurrentWeakRootsEvacUpdateTask task(ShenandoahPhaseTimings::conc_weak_roots_work);
-    heap->workers()->run_task(&task);
-  }
-
-  // Perform handshake to flush out dead oops
-  {
-    ShenandoahTimingsTracker t(ShenandoahPhaseTimings::conc_weak_roots_rendezvous);
-    heap->rendezvous_threads("Shenandoah Concurrent Weak Roots");
-  }
-  // We can only toggle concurrent_weak_root_in_progress flag
-  // at a safepoint, so that mutators see a consistent
-  // value. The flag will be cleared at the next safepoint.
+  ShenandoahTimingsTracker t(ShenandoahPhaseTimings::conc_weak_roots_work);
+  ShenandoahGCWorkerPhase worker_phase(ShenandoahPhaseTimings::conc_weak_roots_work);
+  ShenandoahConcurrentWeakRootsEvacUpdateTask task(ShenandoahPhaseTimings::conc_weak_roots_work);
+  heap->workers()->run_task(&task);
 }
 
 void ShenandoahConcurrentGC::op_class_unloading() {
@@ -1058,10 +1049,6 @@ void ShenandoahConcurrentGC::op_evacuate() {
 
 void ShenandoahConcurrentGC::op_init_updaterefs() {
   ShenandoahHeap* const heap = ShenandoahHeap::heap();
-  heap->set_evacuation_in_progress(false);
-  heap->set_concurrent_weak_root_in_progress(false);
-  heap->prepare_update_heap_references(true /*concurrent*/);
-  heap->set_update_refs_in_progress(true);
   if (ShenandoahVerify) {
     heap->verifier()->verify_before_updaterefs();
   }
@@ -1178,6 +1165,10 @@ void ShenandoahConcurrentGC::op_final_roots() {
       ShenandoahGenerationalHeap::heap()->update_region_ages(_generation->complete_marking_context());
     }
   }
+
+  if (VerifyAfterGC) {
+    Universe::verify();
+  }
 }
 
 void ShenandoahConcurrentGC::op_cleanup_complete() {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
index 89a04d23cec92..ec0bb5afbc6fb 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahFreeSet.cpp
@@ -998,6 +998,8 @@ HeapWord* ShenandoahFreeSet::allocate_aligned_plab(size_t size, ShenandoahAllocR
 HeapWord* ShenandoahFreeSet::try_allocate_in(ShenandoahHeapRegion* r, ShenandoahAllocRequest& req, bool& in_new_region) {
   assert (has_alloc_capacity(r), "Performance: should avoid full regions on this path: " SIZE_FORMAT, r->index());
   if (_heap->is_concurrent_weak_root_in_progress() && r->is_trash()) {
+    // We cannot use this region for allocation when weak roots are in progress because the collector may need
+    // to reference unmarked oops during concurrent classunloading.
     return nullptr;
   }
   HeapWord* result = nullptr;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
index d75a9234a63d2..018540b33b70b 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
@@ -102,6 +102,7 @@
 #include "runtime/orderAccess.hpp"
 #include "runtime/safepointMechanism.hpp"
 #include "runtime/stackWatermarkSet.hpp"
+#include "runtime/threads.hpp"
 #include "runtime/vmThread.hpp"
 #include "utilities/globalDefinitions.hpp"
 #include "utilities/events.hpp"
@@ -644,7 +645,6 @@ class ShenandoahInitWorkerGCLABClosure : public ThreadClosure {
 public:
   void do_thread(Thread* thread) {
     assert(thread != nullptr, "Sanity");
-    assert(thread->is_Worker_thread(), "Only worker thread expected");
     ShenandoahThreadLocalData::initialize_gclab(thread);
   }
 };
@@ -663,6 +663,9 @@ void ShenandoahHeap::post_initialize() {
   // gclab can not be initialized early during VM startup, as it can not determinate its max_size.
   // Now, we will let WorkerThreads to initialize gclab when new worker is created.
   _workers->set_initialize_gclab();
+
+  // Note that the safepoint workers may require gclabs if the threads are used to create a heap dump
+  // during a concurrent evacuation phase.
   if (_safepoint_workers != nullptr) {
     _safepoint_workers->threads_do(&init_gclabs);
     _safepoint_workers->set_initialize_gclab();
@@ -1190,11 +1193,86 @@ class ShenandoahEvacuationTask : public WorkerTask {
   }
 };
 
+class ShenandoahRetireGCLABClosure : public ThreadClosure {
+private:
+  bool const _resize;
+public:
+  explicit ShenandoahRetireGCLABClosure(bool resize) : _resize(resize) {}
+  void do_thread(Thread* thread) override {
+    PLAB* gclab = ShenandoahThreadLocalData::gclab(thread);
+    assert(gclab != nullptr, "GCLAB should be initialized for %s", thread->name());
+    gclab->retire();
+    if (_resize && ShenandoahThreadLocalData::gclab_size(thread) > 0) {
+      ShenandoahThreadLocalData::set_gclab_size(thread, 0);
+    }
+
+    if (ShenandoahHeap::heap()->mode()->is_generational()) {
+      PLAB* plab = ShenandoahThreadLocalData::plab(thread);
+      assert(plab != nullptr, "PLAB should be initialized for %s", thread->name());
+
+      // There are two reasons to retire all plabs between old-gen evacuation passes.
+      //  1. We need to make the plab memory parsable by remembered-set scanning.
+      //  2. We need to establish a trustworthy UpdateWaterMark value within each old-gen heap region
+      ShenandoahGenerationalHeap::heap()->retire_plab(plab, thread);
+      if (_resize && ShenandoahThreadLocalData::plab_size(thread) > 0) {
+        ShenandoahThreadLocalData::set_plab_size(thread, 0);
+      }
+    }
+  }
+};
+
+class ShenandoahGCStatePropagator : public ThreadClosure {
+public:
+  explicit ShenandoahGCStatePropagator(char gc_state) : _gc_state(gc_state) {}
+
+  void do_thread(Thread* thread) override {
+    ShenandoahThreadLocalData::set_gc_state(thread, _gc_state);
+  }
+private:
+  char _gc_state;
+};
+
+class ShenandoahPrepareForUpdateRefs : public HandshakeClosure {
+public:
+  explicit ShenandoahPrepareForUpdateRefs(char gc_state) :
+    HandshakeClosure("Shenandoah Prepare for Update Refs"),
+    _retire(ResizeTLAB), _propagator(gc_state) {}
+
+  void do_thread(Thread* thread) override {
+    _propagator.do_thread(thread);
+    if (ShenandoahThreadLocalData::gclab(thread) != nullptr) {
+      _retire.do_thread(thread);
+    }
+  }
+private:
+  ShenandoahRetireGCLABClosure _retire;
+  ShenandoahGCStatePropagator _propagator;
+};
+
 void ShenandoahHeap::evacuate_collection_set(bool concurrent) {
   ShenandoahEvacuationTask task(this, _collection_set, concurrent);
   workers()->run_task(&task);
 }
 
+void ShenandoahHeap::concurrent_prepare_for_update_refs() {
+  // It's possible that evacuation succeeded, but we could still be cancelled when we get here.
+  // A cancellation at this point means the degenerated cycle must resume from update-refs.
+  set_gc_state_concurrent(EVACUATION, false);
+  set_gc_state_concurrent(WEAK_ROOTS, false);
+  set_gc_state_concurrent(UPDATEREFS, true);
+
+  // This will propagate the gc state and retire gclabs and plabs for threads that require it.
+  ShenandoahPrepareForUpdateRefs prepare_for_update_refs(_gc_state.raw_value());
+
+  // The handshake won't touch worker threads (or control thread, or VM thread), so do those separately.
+  Threads::non_java_threads_do(&prepare_for_update_refs);
+
+  // Now retire gclabs and plabs and propagate gc_state for mutator threads
+  Handshake::execute(&prepare_for_update_refs);
+
+  _update_refs_iterator.reset();
+}
+
 oop ShenandoahHeap::evacuate_object(oop p, Thread* thread) {
   assert(thread == Thread::current(), "Expected thread parameter to be current thread.");
   if (ShenandoahThreadLocalData::is_oom_during_evac(thread)) {
@@ -1349,34 +1427,6 @@ class ShenandoahCheckCleanGCLABClosure : public ThreadClosure {
   }
 };
 
-class ShenandoahRetireGCLABClosure : public ThreadClosure {
-private:
-  bool const _resize;
-public:
-  ShenandoahRetireGCLABClosure(bool resize) : _resize(resize) {}
-  void do_thread(Thread* thread) {
-    PLAB* gclab = ShenandoahThreadLocalData::gclab(thread);
-    assert(gclab != nullptr, "GCLAB should be initialized for %s", thread->name());
-    gclab->retire();
-    if (_resize && ShenandoahThreadLocalData::gclab_size(thread) > 0) {
-      ShenandoahThreadLocalData::set_gclab_size(thread, 0);
-    }
-
-    if (ShenandoahHeap::heap()->mode()->is_generational()) {
-      PLAB* plab = ShenandoahThreadLocalData::plab(thread);
-      assert(plab != nullptr, "PLAB should be initialized for %s", thread->name());
-
-      // There are two reasons to retire all plabs between old-gen evacuation passes.
-      //  1. We need to make the plab memory parsable by remembered-set scanning.
-      //  2. We need to establish a trustworthy UpdateWaterMark value within each old-gen heap region
-      ShenandoahGenerationalHeap::heap()->retire_plab(plab, thread);
-      if (_resize && ShenandoahThreadLocalData::plab_size(thread) > 0) {
-        ShenandoahThreadLocalData::set_plab_size(thread, 0);
-      }
-    }
-  }
-};
-
 void ShenandoahHeap::labs_make_parsable() {
   assert(UseTLAB, "Only call with UseTLAB");
 
@@ -1389,6 +1439,10 @@ void ShenandoahHeap::labs_make_parsable() {
   }
 
   workers()->threads_do(&cl);
+
+  if (safepoint_workers() != nullptr) {
+    safepoint_workers()->threads_do(&cl);
+  }
 }
 
 void ShenandoahHeap::tlabs_retire(bool resize) {
@@ -1424,6 +1478,7 @@ void ShenandoahHeap::gclabs_retire(bool resize) {
   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *t = jtiwh.next(); ) {
     cl.do_thread(t);
   }
+
   workers()->threads_do(&cl);
 
   if (safepoint_workers() != nullptr) {
@@ -1933,23 +1988,25 @@ void ShenandoahHeap::prepare_update_heap_references(bool concurrent) {
   _update_refs_iterator.reset();
 }
 
-void ShenandoahHeap::propagate_gc_state_to_java_threads() {
+void ShenandoahHeap::propagate_gc_state_to_all_threads() {
   assert(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "Must be at Shenandoah safepoint");
   if (_gc_state_changed) {
+    ShenandoahGCStatePropagator propagator(_gc_state.raw_value());
+    Threads::threads_do(&propagator);
     _gc_state_changed = false;
-    char state = gc_state();
-    for (JavaThreadIteratorWithHandle jtiwh; JavaThread *t = jtiwh.next(); ) {
-      ShenandoahThreadLocalData::set_gc_state(t, state);
-    }
   }
 }
 
-void ShenandoahHeap::set_gc_state(uint mask, bool value) {
+void ShenandoahHeap::set_gc_state_at_safepoint(uint mask, bool value) {
   assert(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "Must be at Shenandoah safepoint");
   _gc_state.set_cond(mask, value);
   _gc_state_changed = true;
 }
 
+void ShenandoahHeap::set_gc_state_concurrent(uint mask, bool value) {
+  _gc_state.set_cond(mask, value);
+}
+
 void ShenandoahHeap::set_concurrent_young_mark_in_progress(bool in_progress) {
   uint mask;
   assert(!has_forwarded_objects(), "Young marking is not concurrent with evacuation");
@@ -1961,7 +2018,7 @@ void ShenandoahHeap::set_concurrent_young_mark_in_progress(bool in_progress) {
   } else {
     mask = MARKING | YOUNG_MARKING;
   }
-  set_gc_state(mask, in_progress);
+  set_gc_state_at_safepoint(mask, in_progress);
   manage_satb_barrier(in_progress);
 }
 
@@ -1977,9 +2034,9 @@ void ShenandoahHeap::set_concurrent_old_mark_in_progress(bool in_progress) {
   if (!in_progress && is_concurrent_young_mark_in_progress()) {
     // If young-marking is in progress when we turn off OLD_MARKING, leave MARKING (and YOUNG_MARKING) on
     assert(_gc_state.is_set(MARKING), "concurrent_young_marking_in_progress implies MARKING");
-    set_gc_state(OLD_MARKING, in_progress);
+    set_gc_state_at_safepoint(OLD_MARKING, in_progress);
   } else {
-    set_gc_state(MARKING | OLD_MARKING, in_progress);
+    set_gc_state_at_safepoint(MARKING | OLD_MARKING, in_progress);
   }
   manage_satb_barrier(in_progress);
 }
@@ -2006,7 +2063,7 @@ void ShenandoahHeap::manage_satb_barrier(bool active) {
 
 void ShenandoahHeap::set_evacuation_in_progress(bool in_progress) {
   assert(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "Only call this at safepoint");
-  set_gc_state(EVACUATION, in_progress);
+  set_gc_state_at_safepoint(EVACUATION, in_progress);
 }
 
 void ShenandoahHeap::set_concurrent_strong_root_in_progress(bool in_progress) {
@@ -2018,7 +2075,7 @@ void ShenandoahHeap::set_concurrent_strong_root_in_progress(bool in_progress) {
 }
 
 void ShenandoahHeap::set_concurrent_weak_root_in_progress(bool cond) {
-  set_gc_state(WEAK_ROOTS, cond);
+  set_gc_state_at_safepoint(WEAK_ROOTS, cond);
 }
 
 GCTracer* ShenandoahHeap::tracer() {
@@ -2165,7 +2222,7 @@ void ShenandoahHeap::parallel_cleaning(bool full_gc) {
 }
 
 void ShenandoahHeap::set_has_forwarded_objects(bool cond) {
-  set_gc_state(HAS_FORWARDED, cond);
+  set_gc_state_at_safepoint(HAS_FORWARDED, cond);
 }
 
 void ShenandoahHeap::set_unload_classes(bool uc) {
@@ -2205,7 +2262,7 @@ void ShenandoahHeap::set_full_gc_move_in_progress(bool in_progress) {
 }
 
 void ShenandoahHeap::set_update_refs_in_progress(bool in_progress) {
-  set_gc_state(UPDATEREFS, in_progress);
+  set_gc_state_at_safepoint(UPDATEREFS, in_progress);
 }
 
 void ShenandoahHeap::register_nmethod(nmethod* nm) {
@@ -2610,6 +2667,14 @@ char ShenandoahHeap::gc_state() const {
   return _gc_state.raw_value();
 }
 
+bool ShenandoahHeap::is_gc_state(GCState state) const {
+  // If the global gc state has been changed, but hasn't yet been propagated to all threads, then
+  // the global gc state is the correct value. Once the gc state has been synchronized with all threads,
+  // _gc_state_changed will be toggled to false and we need to use the thread local state.
+  return _gc_state_changed ? _gc_state.is_set(state) : ShenandoahThreadLocalData::is_gc_state(state);
+}
+
+
 ShenandoahLiveData* ShenandoahHeap::get_liveness_cache(uint worker_id) {
 #ifdef ASSERT
   assert(_liveness_cache != nullptr, "sanity");
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
index 5beced0544cf6..508f993b3003f 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
@@ -363,18 +363,35 @@ class ShenandoahHeap : public CollectedHeap {
 
   size_t _gc_no_progress_count;
 
-  // This updates the singlular, global gc state. This must happen on a safepoint.
-  void set_gc_state(uint mask, bool value);
+  // This updates the singular, global gc state. This call must happen on a safepoint.
+  void set_gc_state_at_safepoint(uint mask, bool value);
+
+  // This also updates the global gc state, but does not need to be called on a safepoint.
+  // Critically, this method will _not_ flag that the global gc state has changed and threads
+  // will continue to use their thread local copy. This is expected to be used in conjunction
+  // with a handshake operation to propagate the new gc state.
+  void set_gc_state_concurrent(uint mask, bool value);
 
 public:
+  // This returns the raw value of the singular, global gc state.
   char gc_state() const;
 
-  // This copies the global gc state into a thread local variable for java threads.
-  // It is primarily intended to support quick access at barriers.
-  void propagate_gc_state_to_java_threads();
+  // Compares the given state against either the global gc state, or the thread local state.
+  // The global gc state may change on a safepoint and is the correct value to use until
+  // the global gc state has been propagated to all threads (after which, this method will
+  // compare against the thread local state). The thread local gc state may also be changed
+  // by a handshake operation, in which case, this function continues using the updated thread
+  // local value.
+  bool is_gc_state(GCState state) const;
+
+  // This copies the global gc state into a thread local variable for all threads.
+  // The thread local gc state is primarily intended to support quick access at barriers.
+  // All threads are updated because in some cases the control thread or the vm thread may
+  // need to execute the load reference barrier.
+  void propagate_gc_state_to_all_threads();
 
   // This is public to support assertions that the state hasn't been changed off of
-  // a safepoint and that any changes were propagated to java threads after the safepoint.
+  // a safepoint and that any changes were propagated to threads after the safepoint.
   bool has_gc_state_changed() const { return _gc_state_changed; }
 
   // Returns true if allocations have occurred in new regions or if regions have been
@@ -394,9 +411,7 @@ class ShenandoahHeap : public CollectedHeap {
   void set_concurrent_strong_root_in_progress(bool cond);
   void set_concurrent_weak_root_in_progress(bool cond);
 
-  inline bool is_stable() const;
   inline bool is_idle() const;
-
   inline bool is_concurrent_mark_in_progress() const;
   inline bool is_concurrent_young_mark_in_progress() const;
   inline bool is_concurrent_old_mark_in_progress() const;
@@ -464,6 +479,10 @@ class ShenandoahHeap : public CollectedHeap {
   void do_class_unloading();
   // Reference updating
   void prepare_update_heap_references(bool concurrent);
+
+  // Retires LABs used for evacuation
+  void concurrent_prepare_for_update_refs();
+
   virtual void update_heap_references(bool concurrent);
   // Final update region states
   void update_heap_region_states(bool concurrent);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
index 6a38266489e03..f879188431b5e 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
@@ -64,10 +64,6 @@ inline ShenandoahHeapRegion* ShenandoahRegionIterator::next() {
   return _heap->get_region(new_index - 1);
 }
 
-inline bool ShenandoahHeap::has_forwarded_objects() const {
-  return _gc_state.is_set(HAS_FORWARDED);
-}
-
 inline WorkerThreads* ShenandoahHeap::workers() const {
   return _workers;
 }
@@ -450,28 +446,36 @@ inline bool ShenandoahHeap::in_collection_set_loc(void* p) const {
   return collection_set()->is_in_loc(p);
 }
 
-inline bool ShenandoahHeap::is_stable() const {
-  return _gc_state.is_clear();
+inline bool ShenandoahHeap::is_idle() const {
+  return _gc_state_changed ? _gc_state.is_clear() : ShenandoahThreadLocalData::gc_state(Thread::current()) == 0;
 }
 
-inline bool ShenandoahHeap::is_idle() const {
-  return _gc_state.is_unset(MARKING | EVACUATION | UPDATEREFS);
+inline bool ShenandoahHeap::has_forwarded_objects() const {
+  return is_gc_state(HAS_FORWARDED);
 }
 
 inline bool ShenandoahHeap::is_concurrent_mark_in_progress() const {
-  return _gc_state.is_set(MARKING);
+  return is_gc_state(MARKING);
 }
 
 inline bool ShenandoahHeap::is_concurrent_young_mark_in_progress() const {
-  return _gc_state.is_set(YOUNG_MARKING);
+  return is_gc_state(YOUNG_MARKING);
 }
 
 inline bool ShenandoahHeap::is_concurrent_old_mark_in_progress() const {
-  return _gc_state.is_set(OLD_MARKING);
+  return is_gc_state(OLD_MARKING);
 }
 
 inline bool ShenandoahHeap::is_evacuation_in_progress() const {
-  return _gc_state.is_set(EVACUATION);
+  return is_gc_state(EVACUATION);
+}
+
+inline bool ShenandoahHeap::is_update_refs_in_progress() const {
+  return is_gc_state(UPDATEREFS);
+}
+
+inline bool ShenandoahHeap::is_concurrent_weak_root_in_progress() const {
+  return is_gc_state(WEAK_ROOTS);
 }
 
 inline bool ShenandoahHeap::is_degenerated_gc_in_progress() const {
@@ -486,10 +490,6 @@ inline bool ShenandoahHeap::is_full_gc_move_in_progress() const {
   return _full_gc_move_in_progress.is_set();
 }
 
-inline bool ShenandoahHeap::is_update_refs_in_progress() const {
-  return _gc_state.is_set(UPDATEREFS);
-}
-
 inline bool ShenandoahHeap::is_stw_gc_in_progress() const {
   return is_full_gc_in_progress() || is_degenerated_gc_in_progress();
 }
@@ -498,10 +498,6 @@ inline bool ShenandoahHeap::is_concurrent_strong_root_in_progress() const {
   return _concurrent_strong_root_in_progress.is_set();
 }
 
-inline bool ShenandoahHeap::is_concurrent_weak_root_in_progress() const {
-  return _gc_state.is_set(WEAK_ROOTS);
-}
-
 template<class T>
 inline void ShenandoahHeap::marked_object_iterate(ShenandoahHeapRegion* region, T* cl) {
   marked_object_iterate(region, cl, region->top());
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegionCounters.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegionCounters.cpp
index 73250cecd6fe5..f04aa7ce8ed1f 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegionCounters.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegionCounters.cpp
@@ -147,10 +147,10 @@ static int encode_phase(ShenandoahHeap* heap) {
   if (heap->is_update_refs_in_progress() || heap->is_full_gc_move_in_progress()) {
     return 3;
   }
-  if (heap->is_concurrent_mark_in_progress() || heap->is_full_gc_in_progress()) {
+  if (heap->is_concurrent_mark_in_progress() || heap->is_concurrent_weak_root_in_progress() || heap->is_full_gc_in_progress()) {
     return 1;
   }
-  assert(heap->is_idle(), "What is it doing?");
+  assert(heap->is_idle(), "Unexpected gc_state: %d", heap->gc_state());
   return 0;
 }
 
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahThreadLocalData.hpp b/src/hotspot/share/gc/shenandoah/shenandoahThreadLocalData.hpp
index c226b1ad2545e..9e1777dd82c4f 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahThreadLocalData.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahThreadLocalData.hpp
@@ -111,12 +111,18 @@ class ShenandoahThreadLocalData {
   }
 
   static char gc_state(Thread* thread) {
-    assert(thread->is_Java_thread(), "GC state is only synchronized to java threads");
     return data(thread)->_gc_state;
   }
 
+  static bool is_gc_state(Thread* thread, ShenandoahHeap::GCState state) {
+    return (gc_state(thread) & state) != 0;
+  }
+
+  static bool is_gc_state(ShenandoahHeap::GCState state) {
+    return is_gc_state(Thread::current(), state);
+  }
+
   static void initialize_gclab(Thread* thread) {
-    assert (thread->is_Java_thread() || thread->is_Worker_thread(), "Only Java and GC worker threads are allowed to get GCLABs");
     assert(data(thread)->_gclab == nullptr, "Only initialize once");
     data(thread)->_gclab = new PLAB(PLAB::min_size());
     data(thread)->_gclab_size = 0;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVMOperations.cpp b/src/hotspot/share/gc/shenandoah/shenandoahVMOperations.cpp
index da32601eed7f0..67813be36eeb3 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVMOperations.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVMOperations.cpp
@@ -82,47 +82,47 @@ void VM_ShenandoahInitMark::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Init Mark", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_init_mark();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahFinalMarkStartEvac::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Final Mark", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_final_mark();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahFullGC::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Full GC", SvcGCMarker::FULL);
   set_active_generation();
   _full_gc->entry_full(_gc_cause);
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahDegeneratedGC::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Degenerated GC", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_degenerated();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahInitUpdateRefs::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Init Update Refs", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_init_updaterefs();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahFinalUpdateRefs::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Final Update Refs", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_final_updaterefs();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
 
 void VM_ShenandoahFinalRoots::doit() {
   ShenandoahGCPauseMark mark(_gc_id, "Final Roots", SvcGCMarker::CONCURRENT);
   set_active_generation();
   _gc->entry_final_roots();
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 }
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
index bde8638140b52..a85bbba7a1a6c 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
@@ -796,7 +796,7 @@ void ShenandoahVerifier::verify_at_safepoint(const char* label,
   guarantee(ShenandoahSafepoint::is_at_shenandoah_safepoint(), "only when nothing else happens");
   guarantee(ShenandoahVerify, "only when enabled, and bitmap is initialized in ShenandoahHeap::initialize");
 
-  ShenandoahHeap::heap()->propagate_gc_state_to_java_threads();
+  ShenandoahHeap::heap()->propagate_gc_state_to_all_threads();
 
   // Avoid side-effect of changing workers' active thread count, but bypass concurrent/parallel protocol check
   ShenandoahPushWorkerScope verify_worker_scope(_heap->workers(), _heap->max_workers(), false /*bypass check*/);