diff --git a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/KonanConfig.kt b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/KonanConfig.kt
index d4449b7ae1f..bf0f408e862 100644
--- a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/KonanConfig.kt
+++ b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/KonanConfig.kt
@@ -186,7 +186,7 @@ class KonanConfig(val project: Project, val configuration: CompilerConfiguration
         } ?: arg
     }
 
-    private val defaultGcMarkSingleThreaded get() = target.family == Family.MINGW
+    private val defaultGcMarkSingleThreaded get() = target.family == Family.MINGW && gc == GC.PARALLEL_MARK_CONCURRENT_SWEEP
 
     val gcMarkSingleThreaded: Boolean by lazy {
         configuration.get(BinaryOptions.gcMarkSingleThreaded) ?: defaultGcMarkSingleThreaded
@@ -203,6 +203,12 @@ class KonanConfig(val project: Project, val configuration: CompilerConfiguration
                         "Mutators cooperation is not supported during single threaded mark")
             }
             false
+        } else if (gc == GC.CONCURRENT_MARK_AND_SWEEP) {
+            if (mutatorsCooperate == true) {
+                configuration.report(CompilerMessageSeverity.STRONG_WARNING,
+                        "Mutators cooperation is not yet supported in CMS GC")
+            }
+            false
         } else {
             mutatorsCooperate ?: true
         }
diff --git a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/RuntimeLogging.kt b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/RuntimeLogging.kt
index e07486815f7..9f7b8d70de4 100644
--- a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/RuntimeLogging.kt
+++ b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/RuntimeLogging.kt
@@ -30,6 +30,7 @@ enum class LoggingTag(val ord: Int) {
     Alloc(6),
     Balancing(7),
     Barriers(8),
+    GCMark(9),
     ;
 
     companion object {
diff --git a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/llvm/ContextUtils.kt b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/llvm/ContextUtils.kt
index 841d2be1241..ef44f283336 100644
--- a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/llvm/ContextUtils.kt
+++ b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/llvm/ContextUtils.kt
@@ -481,7 +481,6 @@ internal class CodegenLlvmHelpers(private val generationState: NativeGenerationS
 
     val Kotlin_processObjectInMark by lazyRtFunction
     val Kotlin_processArrayInMark by lazyRtFunction
-    val Kotlin_processFieldInMark by lazyRtFunction
     val Kotlin_processEmptyObjectInMark by lazyRtFunction
 
     val UpdateVolatileHeapRef by lazyRtFunction
diff --git a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/optimizations/EscapeAnalysis.kt b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/optimizations/EscapeAnalysis.kt
index 7223f12e34d..0a2ecb218c2 100644
--- a/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/optimizations/EscapeAnalysis.kt
+++ b/kotlin-native/backend.native/compiler/ir/backend.native/src/org/jetbrains/kotlin/backend/konan/optimizations/EscapeAnalysis.kt
@@ -1828,6 +1828,10 @@ internal object EscapeAnalysis {
                     IntraproceduralAnalysis(context, moduleDFG, externalModulesDFG, callGraph).analyze()
             InterproceduralAnalysis(context, generationState, callGraph, intraproceduralAnalysisResult, externalModulesDFG, lifetimes,
                     propagateExiledToHeapObjects = context.config.memoryModel != MemoryModel.EXPERIMENTAL
+                            // The GC must be careful not to scan exiled objects, that have already became dead,
+                            // as they may reference other already destroyed stack-allocated objects.
+                            // TODO somehow tag these object, so that GC could handle them properly.
+                            || context.config.gc == GC.CONCURRENT_MARK_AND_SWEEP
             ).analyze()
         } catch (t: Throwable) {
             val extraUserInfo =
diff --git a/kotlin-native/runtime/src/compiler_interface/cpp/CompilerCInterface.cpp b/kotlin-native/runtime/src/compiler_interface/cpp/CompilerCInterface.cpp
index 706d1257705..25e7e853ab3 100644
--- a/kotlin-native/runtime/src/compiler_interface/cpp/CompilerCInterface.cpp
+++ b/kotlin-native/runtime/src/compiler_interface/cpp/CompilerCInterface.cpp
@@ -81,7 +81,6 @@ touchFunction(Kotlin_mm_safePointWhileLoopBody)
 
 touchFunction(Kotlin_processObjectInMark)
 touchFunction(Kotlin_processArrayInMark)
-touchFunction(Kotlin_processFieldInMark)
 touchFunction(Kotlin_processEmptyObjectInMark)
 
 touchFunction(Kotlin_arrayGetElementAddress)
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/Barriers.cpp b/kotlin-native/runtime/src/gc/cms/cpp/Barriers.cpp
index dbd14300d64..461c48a94e1 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/Barriers.cpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/Barriers.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2010-2023 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
+ * Copyright 2010-2024 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
  * that can be found in the LICENSE file.
  */
 
@@ -33,12 +33,14 @@ void gc::barriers::BarriersThreadData::onThreadRegistration() noexcept {
 ALWAYS_INLINE void gc::barriers::BarriersThreadData::onSafePoint() noexcept {}
 
 void gc::barriers::BarriersThreadData::startMarkingNewObjects(gc::GCHandle gcHandle) noexcept {
-    RuntimeAssert(markBarriersEnabled.load(std::memory_order_relaxed), "New allocations marking may only be requested by mark barriers");
+    RuntimeAssert(markBarriersEnabled.load(std::memory_order_relaxed),
+                  "New allocations marking may only be requested by mark barriers");
     markHandle_ = gcHandle.mark();
 }
 
 void gc::barriers::BarriersThreadData::stopMarkingNewObjects() noexcept {
-    RuntimeAssert(!markBarriersEnabled.load(std::memory_order_relaxed), "New allocations marking could only been requested by mark barriers");
+    RuntimeAssert(!markBarriersEnabled.load(std::memory_order_relaxed),
+                  "New allocations marking could only been requested by mark barriers");
     markHandle_ = std::nullopt;
 }
 
@@ -48,7 +50,8 @@ bool gc::barriers::BarriersThreadData::shouldMarkNewObjects() const noexcept {
 
 ALWAYS_INLINE void gc::barriers::BarriersThreadData::onAllocation(ObjHeader* allocated) {
     bool shouldMark = shouldMarkNewObjects();
-    RuntimeAssert(shouldMark == markBarriersEnabled.load(std::memory_order_relaxed), "New allocations marking must happen with and only with mark barriers");
+    RuntimeAssert(shouldMark == markBarriersEnabled.load(std::memory_order_relaxed),
+                  "New allocations marking must happen with and only with mark barriers");
     BarriersLogDebug(shouldMark, "Allocation %p", allocated);
     if (shouldMark) {
         auto& objectData = alloc::objectDataForObject(allocated);
@@ -87,7 +90,7 @@ NO_INLINE void beforeHeapRefUpdateSlowPath(mm::DirectRefAccessor ref, ObjHeader*
         // TODO perhaps it would be better to path the thread data from outside
         auto& threadData = *mm::ThreadRegistry::Instance().CurrentThreadData();
         auto& markQueue = *threadData.gc().impl().gc().mark().markQueue();
-        gc::mark::ParallelMark::MarkTraits::tryEnqueue(markQueue, prev);
+        gc::mark::ConcurrentMark::MarkTraits::tryEnqueue(markQueue, prev);
         // No need to add the marked object in statistics here.
         // Objects will be counted on dequeue.
     }
@@ -102,3 +105,30 @@ ALWAYS_INLINE void gc::barriers::beforeHeapRefUpdate(mm::DirectRefAccessor ref,
         BarriersLogDebug(false, "Write *%p <- %p (%p overwritten)", ref.location(), value, ref.load());
     }
 }
+
+namespace {
+
+// TODO decide whether it's really beneficial to NO_INLINE the slow path
+NO_INLINE void weakRefReadInMarkSlowPath(ObjHeader* weakReferee) noexcept {
+    auto& threadData = *mm::ThreadRegistry::Instance().CurrentThreadData();
+    auto& markQueue = *threadData.gc().impl().gc().mark().markQueue();
+    gc::mark::ConcurrentMark::MarkTraits::tryEnqueue(markQueue, weakReferee);
+}
+
+} // namespace
+
+
+ALWAYS_INLINE OBJ_GETTER(gc::barriers::weakRefReadBarrier, std::atomic<ObjHeader*>& weakReferee) noexcept {
+    // TODO be careful with atomics when conucrrent weak sweep is supported
+    auto weak = weakReferee.load(std::memory_order_relaxed);
+    if (!weak) return nullptr;
+    bool mark = markBarriersEnabled.load(std::memory_order_acquire);
+    if (__builtin_expect(mark, false)) {
+        BarriersLogDebug(true, "[mark] Weak read %p", weak);
+        weakRefReadInMarkSlowPath(weak);
+    } else {
+        // TODO reintroduce after-mark barriers that check mark bit like in PMCS, when concurrent weak sweep is supported
+        BarriersLogDebug(false, "Weak read %p", weak);
+    }
+    return weak;
+}
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/Barriers.hpp b/kotlin-native/runtime/src/gc/cms/cpp/Barriers.hpp
index ffb3504580d..238f50c105b 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/Barriers.hpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/Barriers.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2010-2023 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
+ * Copyright 2010-2024 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
  * that can be found in the LICENSE file.
  */
 
@@ -12,13 +12,14 @@
 #include "GCStatistics.hpp"
 #include "ReferenceOps.hpp"
 
+/** See. `ConcurrentMark` */
 namespace kotlin::gc::barriers {
 
 class BarriersThreadData : private Pinned {
 public:
     void onThreadRegistration() noexcept;
     void onSafePoint() noexcept;
-    
+
     void startMarkingNewObjects(GCHandle gcHandle) noexcept;
     void stopMarkingNewObjects() noexcept;
     bool shouldMarkNewObjects() const noexcept;
@@ -34,6 +35,6 @@ void disableMarkBarriers() noexcept;
 
 void beforeHeapRefUpdate(mm::DirectRefAccessor ref, ObjHeader* value) noexcept;
 
-// TODO re-introduce weak barriers again
+OBJ_GETTER(weakRefReadBarrier, std::atomic<ObjHeader*>& weakReferee) noexcept;
 
 } // namespace kotlin::gc
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/BarriersTest.cpp b/kotlin-native/runtime/src/gc/cms/cpp/BarriersTest.cpp
index 6f4c31d213e..0d6e21d3125 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/BarriersTest.cpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/BarriersTest.cpp
@@ -7,9 +7,8 @@
 #include "gtest/gtest.h"
 
 #include "Barriers.hpp"
-#include "ParallelMark.hpp"
+#include "ConcurrentMark.hpp"
 #include "GCImpl.hpp"
-#include "ManuallyScoped.hpp"
 #include "ObjectTestSupport.hpp"
 #include "ObjectOps.hpp"
 #include "ReferenceOps.hpp"
@@ -57,7 +56,7 @@ public:
     }
 
 private:
-    gc::mark::ParallelMark::ParallelProcessor parProc_;
+    gc::mark::ConcurrentMark::ParallelProcessor parProc_;
 };
 
 } // namespace
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.cpp b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.cpp
new file mode 100644
index 00000000000..e4e31db2e85
--- /dev/null
+++ b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.cpp
@@ -0,0 +1,112 @@
+/*
+ * Copyright 2010-2024 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
+ * that can be found in the LICENSE file.
+ */
+
+#include "ConcurrentMark.hpp"
+
+#include "MarkAndSweepUtils.hpp"
+#include "GCStatistics.hpp"
+#include "Utils.hpp"
+#include "GCImpl.hpp"
+
+using namespace kotlin;
+
+void gc::mark::ConcurrentMark::beginMarkingEpoch(gc::GCHandle gcHandle) {
+    gcHandle_ = gcHandle;
+
+    lockedMutatorsList_ = mm::ThreadRegistry::Instance().LockForIter();
+
+    parallelProcessor_.construct();
+}
+
+void gc::mark::ConcurrentMark::endMarkingEpoch() {
+    parallelProcessor_.destroy();
+    resetMutatorFlags();
+    lockedMutatorsList_ = std::nullopt;
+}
+
+void gc::mark::ConcurrentMark::runMainInSTW() {
+    ParallelProcessor::Worker mainWorker(*parallelProcessor_);
+    GCLogDebug(gcHandle().getEpoch(), "Creating main (#0) mark worker");
+
+    // create mutator mark queues
+    for (auto& thread: *lockedMutatorsList_) {
+        thread.gc().impl().gc().mark().markQueue().construct(*parallelProcessor_);
+    }
+
+    completeMutatorsRootSet(mainWorker);
+
+    // global root set must be collected after all the mutator's global data have been published
+    collectRootSetGlobals <MarkTraits>(gcHandle(), mainWorker);
+
+    barriers::enableMarkBarriers(gcHandle().getEpoch());
+
+    resumeTheWorld(gcHandle());
+
+    // build mark closure
+    parallelMark(mainWorker);
+
+    // TODO resume the world much later when the mark closure is completed
+    stopTheWorld(gcHandle(), "GC stop the world #2: complete mark closure");
+
+    barriers::disableMarkBarriers();
+
+    bool refsRemainInMutatorQueues = false;
+    do {
+        for (auto& mutator: *lockedMutatorsList_) {
+            const bool markQueueNowEmpty = mutator.gc().impl().gc().mark().markQueue()->forceFlush();
+            if (!markQueueNowEmpty) {
+                refsRemainInMutatorQueues = true;
+            }
+        }
+
+        parallelProcessor_->resetForNewWork();
+        // complete mark closure form newly found objects
+        parallelMark(mainWorker);
+    } while (refsRemainInMutatorQueues);
+
+    for (auto& thread: *lockedMutatorsList_) {
+        auto& markQueue = thread.gc().impl().gc().mark().markQueue();
+        RuntimeAssert(markQueue->retainsNoWork(), ""); // TODO move into queue's destuctor?
+        markQueue.destroy();
+    }
+}
+
+void gc::mark::ConcurrentMark::runOnMutator(mm::ThreadData&) {
+    // no-op
+}
+
+
+gc::GCHandle& gc::mark::ConcurrentMark::gcHandle() {
+    RuntimeAssert(gcHandle_.isValid(), "GCHandle must be initialized");
+    return gcHandle_;
+}
+
+
+void gc::mark::ConcurrentMark::completeMutatorsRootSet(MarkTraits::MarkQueue& markQueue) {
+    // workers compete for mutators to collect their root set
+    for (auto& thread: *lockedMutatorsList_) {
+        tryCollectRootSet(thread, markQueue);
+    }
+}
+
+void gc::mark::ConcurrentMark::tryCollectRootSet(mm::ThreadData& thread, MarkTraits::MarkQueue& markQueue) {
+    auto& gcData = thread.gc().impl().gc();
+    if (!gcData.tryLockRootSet()) return;
+
+    GCLogDebug(gcHandle().getEpoch(), "Root set collection on thread %d for thread %d", konan::currentThreadId(), thread.threadId());
+    gcData.publish();
+    collectRootSetForThread <MarkTraits>(gcHandle(), markQueue, thread);
+}
+
+void gc::mark::ConcurrentMark::parallelMark(ParallelProcessor::Worker& worker) {
+    GCLogDebug(gcHandle().getEpoch(), "Mark loop has begun");
+    Mark <MarkTraits>(gcHandle(), worker);
+}
+
+void gc::mark::ConcurrentMark::resetMutatorFlags() {
+    for (auto& mut: *lockedMutatorsList_) {
+        mut.gc().impl().gc().clearMarkFlags();
+    }
+}
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.hpp b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.hpp
new file mode 100644
index 00000000000..61a43e99e2e
--- /dev/null
+++ b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMark.hpp
@@ -0,0 +1,125 @@
+/*
+ * Copyright 2010-2023 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
+ * that can be found in the LICENSE file.
+ */
+
+#pragma once
+
+#include <condition_variable>
+#include <mutex>
+
+#include "GCStatistics.hpp"
+#include "ManuallyScoped.hpp"
+#include "ObjectData.hpp"
+#include "ParallelProcessor.hpp"
+#include "SafePoint.hpp"
+#include "ThreadRegistry.hpp"
+#include "Utils.hpp"
+
+namespace kotlin::gc::mark {
+
+/**
+ * Implementation of a Mark GC phase, that runs concurrent with mutator threads, using a "Snapshot at the Beginning" approach.
+ * Steps:
+ * 1. Pause all mutators, collect their root sets.
+ * 2. Mutators resume, maintaining a weak tri-collor invariant with the help of:
+ *    - Deletion write barrier (Dijkstra et al):
+ *      Remembers overwritten values in a thread-local mark queue.
+ *      Barrier+write combination doesn't need to be atomic,
+ *      as only references from mark phase beginning matter for the SatB approach.
+ *    - Read barrier for weak refs:
+ *      Remembers each object read via a weak reference in a thread-local mark queue.
+ *      Prevents the possibility of inserting a strong reference to a ewakly-reachable object behind the mark front.
+ * 3. Concurrently, the marker thread builds a mark closure. Unmarked objects hidden in a mutator mark queues may still exist.
+ * 4. Pause mutators once more, drain local mark queues, and complete the mark closure, this time non-concurrently.
+ *    // TODO build closure fully concurrent
+ * 5. Process and clean weak references. // TODO process weak refs concurrently
+ * 6. Prepare mared heap for sweeping and resume mutation. // TODO prepare heap without a pause
+ */
+class ConcurrentMark : private Pinned {
+    using MarkStackImpl = intrusive_forward_list<GC::ObjectData>;
+
+public:
+    // work balancing parameters were chosen pretty arbitrary
+    using ParallelProcessor = ParallelProcessor<MarkStackImpl, 512, 4096>;
+    using MutatorQueue = ParallelProcessor::WorkSource;
+
+    class MarkTraits {
+    public:
+        using MarkQueue = ParallelProcessor::Worker;
+        using AnyQueue = ParallelProcessor::WorkSource;
+
+        static constexpr auto kAllowHeapToStackRefs = false;
+
+        static void clear(AnyQueue& queue) noexcept {
+            RuntimeAssert(queue.localEmpty(), "Mark queue must be empty");
+        }
+
+        static ALWAYS_INLINE ObjHeader* tryDequeue(MarkQueue& queue) noexcept {
+            auto* obj = queue.tryPop();
+            if (obj) {
+                auto object = alloc::objectForObjectData(*obj);
+                RuntimeLogDebug({logging::Tag::kGCMark}, "Dequeued %p", object);
+                return object;
+            }
+            return nullptr;
+        }
+
+        static ALWAYS_INLINE bool tryEnqueue(AnyQueue& queue, ObjHeader* object) noexcept {
+            auto& objectData = alloc::objectDataForObject(object);
+            bool pushed = queue.tryPush(objectData);
+            if (pushed) {
+                RuntimeLogDebug({logging::Tag::kGCMark}, "Enqueued %p", object);
+            }
+            return pushed;
+        }
+
+        static ALWAYS_INLINE bool tryMark(ObjHeader* object) noexcept {
+            auto& objectData = alloc::objectDataForObject(object);
+            bool pushed = objectData.tryMark();
+            if (pushed) {
+                RuntimeLogDebug({logging::Tag::kGCMark}, "Marked %p", object);
+            }
+            return pushed;
+        }
+
+        static ALWAYS_INLINE void processInMark(MarkQueue& markQueue, ObjHeader* object) noexcept {
+            auto process = object->type_info()->processObjectInMark;
+            RuntimeAssert(process != nullptr, "Got null processObjectInMark for object %p", object);
+            process(static_cast<void*>(&markQueue), object);
+        }
+    };
+
+    class ThreadData : private Pinned {
+    public:
+        auto& markQueue() noexcept { return markQueue_; }
+    private:
+        ManuallyScoped<MutatorQueue> markQueue_{};
+    };
+
+    void beginMarkingEpoch(GCHandle gcHandle);
+    void endMarkingEpoch();
+
+    /** To be run by a single "main" GC thread during STW. */
+    void runMainInSTW();
+
+    /**
+     * To be run by mutator threads that would like to participate in mark.
+     * Will wait for STW detection by a "main" routine.
+     */
+    void runOnMutator(mm::ThreadData& mutatorThread);
+
+private:
+    GCHandle& gcHandle();
+
+    void completeMutatorsRootSet(MarkTraits::MarkQueue& markQueue);
+    void tryCollectRootSet(mm::ThreadData& thread, ParallelProcessor::Worker& markQueue);
+    void parallelMark(ParallelProcessor::Worker& worker);
+
+    void resetMutatorFlags();
+
+    GCHandle gcHandle_ = GCHandle::invalid();
+    std::optional<mm::ThreadRegistry::Iterable> lockedMutatorsList_;
+    ManuallyScoped<ParallelProcessor> parallelProcessor_{};
+};
+} // namespace kotlin::gc::mark
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.cpp b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.cpp
index c6ec04689ea..bf4dcc9b02f 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.cpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.cpp
@@ -55,8 +55,6 @@ ScopedThread createGCThread(const char* name, Body&& body) {
 } // namespace
 
 void gc::ConcurrentMarkAndSweep::ThreadData::OnSuspendForGC() noexcept {
-    CallsCheckerIgnoreGuard guard;
-
     gc_.markDispatcher_.runOnMutator(commonThreadData());
 }
 
@@ -91,12 +89,10 @@ gc::ConcurrentMarkAndSweep::ConcurrentMarkAndSweep(
         GCHandle::getByEpoch(epoch).finalizersDone();
         state_.finalized(epoch);
     }),
-    markDispatcher_(mutatorsCooperate),
     mainThread_(createGCThread("Main GC thread", [this] { mainGCThreadBody(); })) {
-    for (std::size_t i = 0; i < auxGCThreads; ++i) {
-        auxThreads_.emplace_back(createGCThread("Auxiliary GC thread", [this] { auxiliaryGCThreadBody(); }));
-    }
-    RuntimeLogInfo({kTagGC}, "Parallel Mark & Concurrent Sweep GC initialized");
+    RuntimeAssert(!mutatorsCooperate, "Cooperative mutators aren't supported yet");
+    RuntimeAssert(auxGCThreads == 0, "Auxiliary GC threads aren't supported yet");
+    RuntimeLogInfo({kTagGC}, "Concurrent Mark & Sweep GC initialized");
 }
 
 gc::ConcurrentMarkAndSweep::~ConcurrentMarkAndSweep() {
@@ -119,7 +115,7 @@ bool gc::ConcurrentMarkAndSweep::FinalizersThreadIsRunning() noexcept {
 }
 
 void gc::ConcurrentMarkAndSweep::mainGCThreadBody() {
-    RuntimeLogWarning({kTagGC}, "Initializing Concurrent Mark and Sweep GC. Concurrent mark is not implemented yet.");
+    RuntimeLogWarning({kTagGC}, "Initializing Concurrent Mark and Sweep GC.");
     while (true) {
         auto epoch = state_.waitScheduled();
         if (epoch.has_value()) {
@@ -128,14 +124,6 @@ void gc::ConcurrentMarkAndSweep::mainGCThreadBody() {
             break;
         }
     }
-    markDispatcher_.requestShutdown();
-}
-
-void gc::ConcurrentMarkAndSweep::auxiliaryGCThreadBody() {
-    RuntimeAssert(!compiler::gcMarkSingleThreaded(), "Should not reach here during single threaded mark");
-    while (!markDispatcher_.shutdownRequested()) {
-        markDispatcher_.runAuxiliary();
-    }
 }
 
 void gc::ConcurrentMarkAndSweep::PerformFullGC(int64_t epoch) noexcept {
@@ -145,7 +133,7 @@ void gc::ConcurrentMarkAndSweep::PerformFullGC(int64_t epoch) noexcept {
     markDispatcher_.beginMarkingEpoch(gcHandle);
     GCLogDebug(epoch, "Main GC requested marking in mutators");
 
-    stopTheWorld(gcHandle);
+    stopTheWorld(gcHandle, "GC stop the world #1: collect root set");
 
     auto& scheduler = gcScheduler_;
     scheduler.onGCStart();
@@ -207,15 +195,3 @@ void gc::ConcurrentMarkAndSweep::PerformFullGC(int64_t epoch) noexcept {
     // TODO: Consider having an always on sleeping finalizer thread.
     finalizerProcessor_.ScheduleTasks(std::move(finalizerQueue), epoch);
 }
-
-void gc::ConcurrentMarkAndSweep::reconfigure(std::size_t maxParallelism, bool mutatorsCooperate, std::size_t auxGCThreads) noexcept {
-    if (compiler::gcMarkSingleThreaded()) {
-        RuntimeCheck(auxGCThreads == 0, "Auxiliary GC threads must not be created with gcMarkSingleThread");
-        return;
-    }
-    std::unique_lock mainGCLock(gcMutex);
-    markDispatcher_.reset(maxParallelism, mutatorsCooperate, [this] { auxThreads_.clear(); });
-    for (std::size_t i = 0; i < auxGCThreads; ++i) {
-        auxThreads_.emplace_back(createGCThread("Auxiliary GC thread", [this] { auxiliaryGCThreadBody(); }));
-    }
-}
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.hpp b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.hpp
index fdb463491f4..43422f405c8 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.hpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweep.hpp
@@ -19,7 +19,7 @@
 #include "IntrusiveList.hpp"
 #include "MarkAndSweepUtils.hpp"
 #include "ObjectData.hpp"
-#include "ParallelMark.hpp"
+#include "ConcurrentMark.hpp"
 #include "ScopedThread.hpp"
 #include "ThreadData.hpp"
 #include "Types.h"
@@ -57,7 +57,7 @@ public:
         ConcurrentMarkAndSweep& gc_;
         mm::ThreadData& threadData_;
         barriers::BarriersThreadData barriers_;
-        mark::ParallelMark::ThreadData mark_;
+        mark::ConcurrentMark::ThreadData mark_;
 
         std::atomic<bool> rootSetLocked_ = false;
         std::atomic<bool> published_ = false;
@@ -71,13 +71,10 @@ public:
     void StopFinalizerThreadIfRunning() noexcept;
     bool FinalizersThreadIsRunning() noexcept;
 
-    void reconfigure(std::size_t maxParallelism, bool mutatorsCooperate, size_t auxGCThreads) noexcept;
-
     GCStateHolder& state() noexcept { return state_; }
 
 private:
     void mainGCThreadBody();
-    void auxiliaryGCThreadBody();
     void PerformFullGC(int64_t epoch) noexcept;
 
     alloc::Allocator& allocator_;
@@ -86,7 +83,7 @@ private:
     GCStateHolder state_;
     FinalizerProcessor<alloc::FinalizerQueue, alloc::FinalizerQueueTraits> finalizerProcessor_;
 
-    mark::ParallelMark markDispatcher_;
+    mark::ConcurrentMark markDispatcher_;
     ScopedThread mainThread_;
     std::vector<ScopedThread> auxThreads_;
 };
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweepTest.cpp b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweepTest.cpp
index f4bc1b3cf89..64ed1cadcb9 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweepTest.cpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/ConcurrentMarkAndSweepTest.cpp
@@ -11,7 +11,6 @@
 
 #include "GCImpl.hpp"
 #include "GlobalData.hpp"
-#include "SafePoint.hpp"
 #include "StableRef.hpp"
 #include "TestSupport.hpp"
 #include "TracingGCTest.hpp"
@@ -34,65 +33,5 @@ public:
 
 } // namespace
 
-TYPED_TEST_P(TracingGCTest, CMSMultipleMutatorsWeak) {
-    std::vector<Mutator> mutators(kDefaultThreadCount);
-    ObjHeader* globalRoot = nullptr;
-    test_support::RegularWeakReferenceImpl* weak = nullptr;
-
-    mutators[0]
-            .Execute([&weak, &globalRoot](mm::ThreadData& threadData, Mutator& mutator) {
-                auto& global = mutator.AddGlobalRoot();
-
-                auto& object = AllocateObject(threadData);
-                auto& objectWeak = ([&threadData, &object]() -> test_support::RegularWeakReferenceImpl& {
-                    ObjHolder holder;
-                    return test_support::InstallWeakReference(threadData, object.header(), holder.slot());
-                })();
-                global->field1 = objectWeak.header();
-                weak = &objectWeak;
-                globalRoot = global.header();
-            })
-            .wait();
-
-    // Make sure all mutators are initialized.
-    for (int i = 1; i < kDefaultThreadCount; ++i) {
-        mutators[i].Execute([](mm::ThreadData& threadData, Mutator& mutator) {}).wait();
-    }
-
-    std::vector<std::future<void>> gcFutures;
-    auto epoch = mm::GlobalData::Instance().gc().Schedule();
-    std::atomic<bool> gcDone = false;
-
-    // Spin until thread suspension is requested.
-    while (!mm::IsThreadSuspensionRequested()) {}
-
-    for (auto& mutator : mutators) {
-        gcFutures.emplace_back(mutator.Execute([&](mm::ThreadData& threadData, Mutator& mutator) noexcept {
-            bool dead = false;
-            while (!gcDone.load(std::memory_order_relaxed)) {
-                mm::safePoint(threadData);
-                auto weakReferee = weak->get();
-                if (dead) {
-                    EXPECT_THAT(weakReferee, nullptr);
-                } else if (weakReferee == nullptr) {
-                    dead = true;
-                }
-            }
-            EXPECT_THAT(weak->get(), nullptr);
-        }));
-    }
-
-    mm::GlobalData::Instance().gc().WaitFinalizers(epoch);
-    gcDone.store(true, std::memory_order_relaxed);
-
-    for (auto& future : gcFutures) {
-        future.wait();
-    }
-
-    for (auto& mutator : mutators) {
-        EXPECT_THAT(mutator.Alive(), testing::UnorderedElementsAre(globalRoot, weak->header()));
-    }
-}
-
-REGISTER_TYPED_TEST_SUITE_WITH_LISTS(TracingGCTest, TRACING_GC_TEST_LIST, CMSMultipleMutatorsWeak);
+REGISTER_TYPED_TEST_SUITE_WITH_LISTS(TracingGCTest, TRACING_GC_TEST_LIST);
 INSTANTIATE_TYPED_TEST_SUITE_P(CMS, TracingGCTest, ConcurrentMarkAndSweepTest);
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/GCImpl.cpp b/kotlin-native/runtime/src/gc/cms/cpp/GCImpl.cpp
index 42c3c428ad7..21991220dc9 100644
--- a/kotlin-native/runtime/src/gc/cms/cpp/GCImpl.cpp
+++ b/kotlin-native/runtime/src/gc/cms/cpp/GCImpl.cpp
@@ -59,17 +59,12 @@ bool gc::GC::FinalizersThreadIsRunning() noexcept {
 
 // static
 ALWAYS_INLINE void gc::GC::processObjectInMark(void* state, ObjHeader* object) noexcept {
-    gc::internal::processObjectInMark<gc::mark::ParallelMark::MarkTraits>(state, object);
+    gc::internal::processObjectInMark<gc::mark::ConcurrentMark::MarkTraits>(state, object);
 }
 
 // static
 ALWAYS_INLINE void gc::GC::processArrayInMark(void* state, ArrayHeader* array) noexcept {
-    gc::internal::processArrayInMark<gc::mark::ParallelMark::MarkTraits>(state, array);
-}
-
-// static
-ALWAYS_INLINE void gc::GC::processFieldInMark(void* state, ObjHeader* field) noexcept {
-    gc::internal::processFieldInMark<gc::mark::ParallelMark::MarkTraits>(state, field);
+    gc::internal::processArrayInMark<gc::mark::ConcurrentMark::MarkTraits>(state, array);
 }
 
 int64_t gc::GC::Schedule() noexcept {
@@ -89,7 +84,7 @@ ALWAYS_INLINE void gc::beforeHeapRefUpdate(mm::DirectRefAccessor ref, ObjHeader*
 }
 
 ALWAYS_INLINE OBJ_GETTER(gc::weakRefReadBarrier, std::atomic<ObjHeader*>& weakReferee) noexcept {
-    RETURN_OBJ(weakReferee.load(std::memory_order_relaxed));
+    RETURN_RESULT_OF(gc::barriers::weakRefReadBarrier, weakReferee);
 }
 
 bool gc::isMarked(ObjHeader* object) noexcept {
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.cpp b/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.cpp
deleted file mode 100644
index 6820d972418..00000000000
--- a/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.cpp
+++ /dev/null
@@ -1,235 +0,0 @@
-/*
- * Copyright 2010-2023 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
- * that can be found in the LICENSE file.
- */
-
-#include "ParallelMark.hpp"
-
-#include "MarkAndSweepUtils.hpp"
-#include "GCStatistics.hpp"
-#include "Utils.hpp"
-
-// required to access gc thread data
-#include "GCImpl.hpp"
-
-using namespace kotlin;
-
-namespace {
-
-template<typename Cond>
-void spinWait(Cond&& until) {
-    while (!until()) {
-        std::this_thread::yield();
-    }
-}
-
-} // namespace
-
-bool gc::mark::MarkPacer::is(gc::mark::MarkPacer::Phase phase) const {
-    std::unique_lock lock(mutex_);
-    return phase_ == phase;
-}
-
-void gc::mark::MarkPacer::begin(gc::mark::MarkPacer::Phase phase) {
-    {
-        std::unique_lock lock(mutex_);
-        phase_ = phase;
-    }
-    cond_.notify_all();
-}
-
-void gc::mark::MarkPacer::wait(gc::mark::MarkPacer::Phase phase) {
-    std::unique_lock lock(mutex_);
-    cond_.wait(lock, [=]() { return phase_ >= phase; });
-}
-
-void gc::mark::MarkPacer::beginEpoch(uint64_t epoch) {
-    epoch_ = epoch;
-    begin(Phase::kReady);
-    GCLogDebug(epoch_.load(), "Mark is ready to recruit workers in a new epoch.");
-}
-
-void gc::mark::MarkPacer::waitNewEpochReadyOrShutdown() const {
-    std::unique_lock lock(mutex_);
-    cond_.wait(lock, [this]() { return phase_ >= Phase::kReady; });
-}
-
-void gc::mark::MarkPacer::waitEpochFinished(uint64_t currentEpoch) const {
-    std::unique_lock lock(mutex_);
-    cond_.wait(lock, [this, currentEpoch]() {
-        return phase_ == Phase::kIdle || phase_ == Phase::kShutdown || epoch_.load(std::memory_order_relaxed) > currentEpoch;
-    });
-}
-
-bool gc::mark::MarkPacer::acceptingNewWorkers() const {
-    std::unique_lock lock(mutex_);
-    return Phase::kReady <= phase_ && phase_ <= Phase::kParallelMark;
-}
-
-
-gc::mark::ParallelMark::ParallelMark(bool mutatorsCooperate) {
-    std::size_t maxParallelism = std::thread::hardware_concurrency();
-    if (maxParallelism == 0) {
-        maxParallelism = std::numeric_limits<std::size_t>::max();
-    }
-    setParallelismLevel(maxParallelism, mutatorsCooperate);
-}
-
-void gc::mark::ParallelMark::beginMarkingEpoch(gc::GCHandle gcHandle) {
-    gcHandle_ = gcHandle;
-
-    lockedMutatorsList_ = mm::ThreadRegistry::Instance().LockForIter();
-
-    parallelProcessor_.construct();
-
-    if (!compiler::gcMarkSingleThreaded()) {
-        std::unique_lock guard(workerCreationMutex_);
-        pacer_.beginEpoch(gcHandle.getEpoch());
-        // main worker is always accounted, so others would not be able to exhaust all the parallelism before main is instantiated
-        activeWorkersCount_ = 1;
-    }
-}
-
-void gc::mark::ParallelMark::endMarkingEpoch() {
-    if (!compiler::gcMarkSingleThreaded()) {
-        // We must now wait for every worker to finish the Mark procedure:
-        // wake up from possible waiting, publish statistics, etc.
-        // Only then it's safe to destroy the parallelProcessor and proceed to other GC tasks such as sweep.
-        spinWait([=]() { return activeWorkersCount_.load(std::memory_order_relaxed) == 0; });
-
-        std::unique_lock guard(workerCreationMutex_);
-        RuntimeAssert(activeWorkersCount_ == 0, "All the workers must already finish");
-        pacer_.begin(MarkPacer::Phase::kIdle);
-    }
-    parallelProcessor_.destroy();
-    resetMutatorFlags();
-    lockedMutatorsList_ = std::nullopt;
-}
-
-void gc::mark::ParallelMark::runMainInSTW() {
-    if (compiler::gcMarkSingleThreaded()) {
-        ParallelProcessor::Worker worker(*parallelProcessor_);
-        gc::collectRootSet<MarkTraits>(gcHandle(), worker, [] (mm::ThreadData&) { return true; });
-        gc::Mark<MarkTraits>(gcHandle(), worker);
-    } else {
-        RuntimeAssert(activeWorkersCount_ > 0, "Main worker must always be accounted");
-        ParallelProcessor::Worker mainWorker(*parallelProcessor_);
-        GCLogDebug(gcHandle().getEpoch(), "Creating main (#0) mark worker");
-
-        pacer_.begin(MarkPacer::Phase::kRootSet);
-        completeMutatorsRootSet(mainWorker);
-        spinWait([this] {
-            return allMutators([](mm::ThreadData& mut) { return mut.gc().impl().gc().published(); });
-        });
-        // global root set must be collected after all the mutator's global data have been published
-        collectRootSetGlobals<MarkTraits>(gcHandle(), mainWorker);
-
-        pacer_.begin(MarkPacer::Phase::kParallelMark);
-        parallelMark(mainWorker);
-    }
-}
-
-void gc::mark::ParallelMark::runOnMutator(mm::ThreadData& mutatorThread) {
-    if (compiler::gcMarkSingleThreaded() || !mutatorsCooperate_) return;
-
-    auto parallelWorker = createWorker();
-    if (parallelWorker) {
-        GCLogDebug(gcHandle().getEpoch(), "Mutator thread cooperates in marking");
-
-        tryCollectRootSet(mutatorThread, *parallelWorker);
-
-        completeRootSetAndMark(*parallelWorker);
-    }
-}
-
-void gc::mark::ParallelMark::runAuxiliary() {
-    RuntimeAssert(!compiler::gcMarkSingleThreaded(), "Should not reach here during single threaded mark");
-
-    pacer_.waitNewEpochReadyOrShutdown();
-    if (pacer_.is(MarkPacer::Phase::kShutdown)) return;
-
-    auto curEpoch = gcHandle().getEpoch();
-    auto parallelWorker = createWorker();
-    if (parallelWorker) {
-        completeRootSetAndMark(*parallelWorker);
-    }
-
-    pacer_.waitEpochFinished(curEpoch);
-}
-
-void gc::mark::ParallelMark::requestShutdown() {
-    pacer_.begin(MarkPacer::Phase::kShutdown);
-}
-
-bool gc::mark::ParallelMark::shutdownRequested() const {
-    return pacer_.is(MarkPacer::Phase::kShutdown);
-}
-
-
-gc::GCHandle& gc::mark::ParallelMark::gcHandle() {
-    RuntimeAssert(gcHandle_.isValid(), "GCHandle must be initialized");
-    return gcHandle_;
-}
-
-void gc::mark::ParallelMark::setParallelismLevel(size_t maxParallelism, bool mutatorsCooperate) {
-    RuntimeCheck(maxParallelism > 0, "Parallelism level can't be 0");
-    maxParallelism_ = maxParallelism;
-    mutatorsCooperate_ = mutatorsCooperate;
-    RuntimeLogInfo({kTagGC},
-                   "Set up parallel mark with maxParallelism = %zu and %s" "cooperative mutators",
-                   maxParallelism_, (mutatorsCooperate_ ? "" : "non-"));
-}
-
-void gc::mark::ParallelMark::completeRootSetAndMark(ParallelProcessor::Worker& parallelWorker) {
-    pacer_.wait(MarkPacer::Phase::kRootSet);
-    completeMutatorsRootSet(parallelWorker);
-    pacer_.wait(MarkPacer::Phase::kParallelMark);
-    parallelMark(parallelWorker);
-}
-
-void gc::mark::ParallelMark::completeMutatorsRootSet(MarkTraits::MarkQueue& markQueue) {
-    // workers compete for mutators to collect their root set
-    for (auto& thread: *lockedMutatorsList_) {
-        tryCollectRootSet(thread, markQueue);
-    }
-}
-
-void gc::mark::ParallelMark::tryCollectRootSet(mm::ThreadData& thread, MarkTraits::MarkQueue& markQueue) {
-    auto& gcData = thread.gc().impl().gc();
-    if (!gcData.tryLockRootSet()) return;
-
-    GCLogDebug(gcHandle().getEpoch(), "Root set collection on thread %d for thread %d",
-               konan::currentThreadId(), thread.threadId());
-    gcData.publish();
-    collectRootSetForThread<MarkTraits>(gcHandle(), markQueue, thread);
-}
-
-void gc::mark::ParallelMark::parallelMark(ParallelProcessor::Worker& worker) {
-    GCLogDebug(gcHandle().getEpoch(), "Mark loop has begun");
-    Mark<MarkTraits>(gcHandle(), worker);
-
-    std::unique_lock guard(workerCreationMutex_);
-    activeWorkersCount_.fetch_sub(1, std::memory_order_relaxed);
-}
-
-std::optional<gc::mark::ParallelMark::ParallelProcessor::Worker> gc::mark::ParallelMark::createWorker() {
-    std::unique_lock guard(workerCreationMutex_);
-    if (!pacer_.acceptingNewWorkers() ||
-        activeWorkersCount_.load(std::memory_order_relaxed) >= maxParallelism_ ||
-        activeWorkersCount_.load(std::memory_order_relaxed) == 0) return std::nullopt;
-
-    auto num = activeWorkersCount_.fetch_add(1, std::memory_order_relaxed);
-    GCLogDebug(gcHandle().getEpoch(), "Creating mark worker #%zu", num);
-    return std::make_optional<ParallelProcessor::Worker>(*parallelProcessor_);
-}
-
-void gc::mark::ParallelMark::resetMutatorFlags() {
-    for (auto& mut: *lockedMutatorsList_) {
-        auto& gcData = mut.gc().impl().gc();
-        if (!compiler::gcMarkSingleThreaded()) {
-            // single threaded mark do not use this flag
-            RuntimeAssert(gcData.published(), "Must have been published during mark");
-        }
-        gcData.clearMarkFlags();
-    }
-}
diff --git a/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.hpp b/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.hpp
deleted file mode 100644
index 957679fcbe7..00000000000
--- a/kotlin-native/runtime/src/gc/cms/cpp/ParallelMark.hpp
+++ /dev/null
@@ -1,193 +0,0 @@
-/*
- * Copyright 2010-2023 JetBrains s.r.o. Use of this source code is governed by the Apache 2.0 license
- * that can be found in the LICENSE file.
- */
-
-#pragma once
-
-#include <condition_variable>
-#include <mutex>
-
-#include "GCStatistics.hpp"
-#include "ManuallyScoped.hpp"
-#include "ObjectData.hpp"
-#include "ParallelProcessor.hpp"
-#include "ThreadRegistry.hpp"
-#include "Utils.hpp"
-
-namespace kotlin::gc::mark {
-
-class MarkPacer : private Pinned {
-public:
-    enum class Phase {
-        /** Mark is not in progress. */
-        kIdle,
-        /**
-         * MarkDispatcher is ready to recruit new workers.
-         *
-         * In case of cooperative mark mutator threads are welcome to mark their own root sets.
-         * Each thread is free to start as soon as it reaches a safe point.
-         * No need to wait for others.
-         */
-        kReady,
-        /**
-         * All mutator threads must be in a safe state at this point:
-         * 1) Suspended on a safe point;
-         * 2) In the native code;
-         * 3) Registered as cooperative markers during previous phase.
-         *
-         * Now all the GC workers are summoned to participate in a root set collection.
-         */
-        kRootSet,
-        /**
-         * Root set is collected. No more workers can be instantiated, time to begin parallel mark.
-         * Parallel mark can't stop before all the created workers begin the marking.
-         */
-        kParallelMark,
-        /** A shutdown was requested. There is nothing more to wait for. */
-        kShutdown,
-    };
-
-    bool is(Phase phase) const;
-    void begin(Phase phase);
-    void wait(Phase phase);
-
-    void beginEpoch(uint64_t epoch);
-    void waitNewEpochReadyOrShutdown() const;
-    void waitEpochFinished(uint64_t epoch) const;
-
-    bool acceptingNewWorkers() const;
-
-private:
-    std::atomic<uint64_t> epoch_ = 0;
-    Phase phase_ = Phase::kIdle;
-    mutable std::mutex mutex_;
-    mutable std::condition_variable cond_;
-};
-
-/**
- * Parallel mark dispatcher.
- * Mark can be performed on one or more threads.
- * Each threads wanting to participate have to execute an appropriate run- routine when ready to mark.
- * There must be exactly one executor of a `runMainInSTW()`.
- *
- * Mark workers are able to balance work between each other through sharing/stealing.
- */
-class ParallelMark : private Pinned {
-    using MarkStackImpl = intrusive_forward_list<GC::ObjectData>;
-    // work balancing parameters were chosen pretty arbitrary
-
-public:
-    using ParallelProcessor = ParallelProcessor<MarkStackImpl, 512, 4096>;
-    using MutatorQueue = ParallelProcessor::WorkSource;
-
-    class MarkTraits {
-    public:
-        using MarkQueue = ParallelProcessor::Worker;
-        using AnyQueue = ParallelProcessor::WorkSource;
-
-        static void clear(AnyQueue& queue) noexcept {
-            RuntimeAssert(queue.localEmpty(), "Mark queue must be empty");
-        }
-
-        static ALWAYS_INLINE ObjHeader* tryDequeue(MarkQueue& queue) noexcept {
-            auto* obj = compiler::gcMarkSingleThreaded() ? queue.tryPopLocal() : queue.tryPop();
-            if (obj) {
-                return alloc::objectForObjectData(*obj);
-            }
-            return nullptr;
-        }
-
-        static ALWAYS_INLINE bool tryEnqueue(AnyQueue& queue, ObjHeader* object) noexcept {
-            auto& objectData = alloc::objectDataForObject(object);
-            return compiler::gcMarkSingleThreaded() ? queue.tryPushLocal(objectData) : queue.tryPush(objectData);
-        }
-
-        static ALWAYS_INLINE bool tryMark(ObjHeader* object) noexcept {
-            auto& objectData = alloc::objectDataForObject(object);
-            return objectData.tryMark();
-        }
-
-        static ALWAYS_INLINE void processInMark(MarkQueue& markQueue, ObjHeader* object) noexcept {
-            auto process = object->type_info()->processObjectInMark;
-            RuntimeAssert(process != nullptr, "Got null processObjectInMark for object %p", object);
-            process(static_cast<void*>(&markQueue), object);
-        }
-    };
-
-    class ThreadData : private Pinned {
-    public:
-        auto& markQueue() noexcept { return markQueue_; }
-    private:
-        ManuallyScoped<MutatorQueue> markQueue_{};
-    };
-
-    explicit ParallelMark(bool mutatorsCooperate);
-
-    void beginMarkingEpoch(gc::GCHandle gcHandle);
-    void endMarkingEpoch();
-
-    /** To be run by a single "main" GC thread during STW. */
-    void runMainInSTW();
-
-    /**
-     * To be run by mutator threads that would like to participate in mark.
-     * Will wait for STW detection by a "main" routine.
-     */
-    void runOnMutator(mm::ThreadData& mutatorThread);
-
-    /**
-     * To be run by auxiliary GC threads.
-     * Will wait for STW detection by a "main" routine.
-     */
-    void runAuxiliary();
-
-    void requestShutdown();
-    bool shutdownRequested() const;
-
-    template<typename Pred>
-    void reset(std::size_t maxParallelism, bool mutatorsCooperate, Pred waitForWorkersToFinish) {
-        pacer_.begin(MarkPacer::Phase::kShutdown);
-        waitForWorkersToFinish();
-        pacer_.begin(MarkPacer::Phase::kIdle);
-        setParallelismLevel(maxParallelism, mutatorsCooperate);
-    }
-
-private:
-    GCHandle& gcHandle();
-
-    void setParallelismLevel(size_t maxParallelism, bool mutatorsCooperate);
-
-    template <typename Pred>
-    bool allMutators(Pred predicate) noexcept {
-        for (auto& thread : *lockedMutatorsList_) {
-            if (!predicate(thread)) {
-                return false;
-            }
-        }
-        return true;
-    }
-
-    void completeRootSetAndMark(ParallelProcessor::Worker& parallelWorker);
-    void completeMutatorsRootSet(MarkTraits::MarkQueue& markQueue);
-    void tryCollectRootSet(mm::ThreadData& thread, ParallelProcessor::Worker& markQueue);
-    void parallelMark(ParallelProcessor::Worker& worker);
-
-    std::optional<ParallelProcessor::Worker> createWorker();
-
-    void resetMutatorFlags();
-
-    std::size_t maxParallelism_ = 1;
-    bool mutatorsCooperate_ = false;
-
-    GCHandle gcHandle_ = GCHandle::invalid();
-    MarkPacer pacer_;
-    std::optional<mm::ThreadRegistry::Iterable> lockedMutatorsList_;
-    ManuallyScoped<ParallelProcessor> parallelProcessor_{};
-
-    std::mutex workerCreationMutex_;
-    std::atomic<std::size_t> activeWorkersCount_ = 0;
-};
-
-} // namespace kotlin::gc::mark
-
diff --git a/kotlin-native/runtime/src/gc/common/cpp/GC.hpp b/kotlin-native/runtime/src/gc/common/cpp/GC.hpp
index 4ef88a3ce2b..cbedd069f74 100644
--- a/kotlin-native/runtime/src/gc/common/cpp/GC.hpp
+++ b/kotlin-native/runtime/src/gc/common/cpp/GC.hpp
@@ -72,7 +72,6 @@ public:
 
     static void processObjectInMark(void* state, ObjHeader* object) noexcept;
     static void processArrayInMark(void* state, ArrayHeader* array) noexcept;
-    static void processFieldInMark(void* state, ObjHeader* field) noexcept;
 
     // TODO: These should exist only in the scheduler.
     int64_t Schedule() noexcept;
diff --git a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.cpp b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.cpp
index 14a88907558..f5c9d943440 100644
--- a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.cpp
+++ b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.cpp
@@ -5,8 +5,8 @@
 
 #include "MarkAndSweepUtils.hpp"
 
-void kotlin::gc::stopTheWorld(kotlin::gc::GCHandle gcHandle) noexcept {
-    bool didSuspend = mm::RequestThreadsSuspension();
+void kotlin::gc::stopTheWorld(GCHandle gcHandle, const char* reason) noexcept {
+    bool didSuspend = mm::RequestThreadsSuspension(reason);
     RuntimeAssert(didSuspend, "Only GC thread can request suspension");
     gcHandle.suspensionRequested();
 
diff --git a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.hpp b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.hpp
index 4538a841f55..d1b06577d63 100644
--- a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.hpp
+++ b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtils.hpp
@@ -26,27 +26,32 @@ namespace gc {
 namespace internal {
 
 template <typename Traits>
-void processFieldInMark(void* state, ObjHeader* field) noexcept {
+void processFieldInMark(void* state, ObjHeader* object, ObjHeader* field) noexcept {
     auto& markQueue = *static_cast<typename Traits::MarkQueue*>(state);
     if (field->heap()) {
         Traits::tryEnqueue(markQueue, field);
     }
+    if constexpr (!Traits::kAllowHeapToStackRefs) {
+        if (object->heap()) {
+            RuntimeAssert(!field->local(), "Heap object %p references stack object %p[typeInfo=%p]", object, field, field->type_info());
+        }
+    }
 }
 
 template <typename Traits>
 void processObjectInMark(void* state, ObjHeader* object) noexcept {
-    traverseClassObjectFields(object, [state] (auto fieldAccessor) noexcept {
+    traverseClassObjectFields(object, [=] (auto fieldAccessor) noexcept {
         if (ObjHeader* field = fieldAccessor.direct()) {
-            processFieldInMark<Traits>(state, field);
+            processFieldInMark<Traits>(state, object, field);
         }
     });
 }
 
 template <typename Traits>
 void processArrayInMark(void* state, ArrayHeader* array) noexcept {
-    traverseArrayOfObjectsElements(array, [state] (auto elemAccessor) noexcept {
+    traverseArrayOfObjectsElements(array, [=] (auto elemAccessor) noexcept {
         if (ObjHeader* elem = elemAccessor.direct()) {
-            processFieldInMark<Traits>(state, elem);
+            processFieldInMark<Traits>(state, array->obj(), elem);
         }
     });
 }
@@ -180,11 +185,11 @@ struct DefaultProcessWeaksTraits {
     static bool IsMarked(ObjHeader* obj) noexcept { return gc::isMarked(obj); }
 };
 
-void stopTheWorld(GCHandle gcHandle) noexcept;
+void stopTheWorld(GCHandle gcHandle, const char* reason) noexcept;
 void resumeTheWorld(GCHandle gcHandle) noexcept;
 
 [[nodiscard]] inline auto stopTheWorldInScope(GCHandle gcHandle) noexcept {
-    return ScopeGuard([=]() { stopTheWorld(gcHandle); }, [=]() { resumeTheWorld(gcHandle); });
+    return ScopeGuard([=]() { stopTheWorld(gcHandle, "GC stop the world"); }, [=]() { resumeTheWorld(gcHandle); });
 }
 
 } // namespace gc
diff --git a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtilsMarkTest.cpp b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtilsMarkTest.cpp
index d922c418d5b..329c7f351d9 100644
--- a/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtilsMarkTest.cpp
+++ b/kotlin-native/runtime/src/gc/common/cpp/MarkAndSweepUtilsMarkTest.cpp
@@ -75,6 +75,8 @@ class ScopedMarkTraits : private Pinned {
 public:
     using MarkQueue = std::vector<ObjHeader*>;
 
+    static constexpr auto kAllowHeapToStackRefs = true;
+
     ScopedMarkTraits() {
         RuntimeAssert(instance_ == nullptr, "Only one ScopedMarkTraits is allowed");
         instance_ = this;
diff --git a/kotlin-native/runtime/src/gc/common/cpp/TracingGCTest.hpp b/kotlin-native/runtime/src/gc/common/cpp/TracingGCTest.hpp
index 435c8402c3b..94ddab71e3a 100644
--- a/kotlin-native/runtime/src/gc/common/cpp/TracingGCTest.hpp
+++ b/kotlin-native/runtime/src/gc/common/cpp/TracingGCTest.hpp
@@ -1073,6 +1073,180 @@ TYPED_TEST_P(TracingGCTest, FreeObjectWithFreeWeakReversedOrder) {
     f1.wait();
 }
 
+TYPED_TEST_P(TracingGCTest, MutateBetweenSafePoints) {
+    constexpr auto kQuant = 5;
+    constexpr auto kGcNumber = 5;
+    constexpr auto kMaxItersPerGC = 10;
+
+    std::atomic<bool> stopMutation = false;
+    std::atomic<bool> startMutation = false;
+    std::atomic<std::size_t> initializedMutators = 0;
+    std::atomic<int> gcEpoch = 0;
+
+    Mutator scheduler;
+    std::future<void> schedulerFuture = scheduler.Execute([&](mm::ThreadData& threadData, Mutator& mutator) {
+        while (initializedMutators < kDefaultThreadCount) { /* wait */ }
+        startMutation = true;
+        for (int i = 0; i < kGcNumber; ++i) {
+            gcEpoch = i;
+            mm::GlobalData::Instance().gcScheduler().scheduleAndWaitFinalized();
+        }
+        stopMutation = true;
+    });
+
+    std::vector<Mutator> mutators(kDefaultThreadCount);
+    std::vector<std::future<void>> mutatorFutures;
+
+    for (int i = 0; i < kDefaultThreadCount; ++i) {
+        mutatorFutures.emplace_back(mutators[i].Execute([&](mm::ThreadData& threadData, Mutator& mutator) {
+            StackObjectHolder head{threadData};
+            ObjHeader* tail = head.header();
+
+            auto append = [&](int count) {
+                for (int i = 0; i < count; ++i) {
+                    auto& next = AllocateObject(threadData);
+                    auto& tailObj = test_support::Object<Payload>::FromObjHeader(tail);
+                    tailObj->field1 = next.header();
+                    tail = next.header();
+                }
+            };
+
+            auto cut = [&](ObjHeader* first, int count) {
+                ObjHeader* last = first;
+                for (int i = 0; i < count; ++i) {
+                    auto& lastObj = test_support::Object<Payload>::FromObjHeader(last);
+                    last = lastObj->field1.accessor().load();
+                }
+                auto& firstObj = test_support::Object<Payload>::FromObjHeader(first);
+                auto& lastObj = test_support::Object<Payload>::FromObjHeader(last);
+                firstObj->field1.accessor() = lastObj->field1.accessor().load();
+            };
+
+            // Initialize
+            append(kQuant * 2);
+            ++initializedMutators;
+            while (!startMutation.load()) { /* wait */ };
+
+            int iter = 0;
+            while (!stopMutation.load()) {
+                // do not let mutator outpace gc to much
+                while (iter > (gcEpoch.load() * kMaxItersPerGC) && !stopMutation.load()) {
+                    mm::safePoint(threadData);
+                }
+
+                auto oldTail = tail;
+                append(kQuant);
+                // [head]->(a)->...->(b)->...->(oldTail)->(c)->...->(d)
+                mm::safePoint(threadData);
+
+                append(kQuant);
+                // [head]->(a)->...->(b)->...->(oldTail)->(c)->...->(d)->...->(tail)
+                mm::safePoint(threadData);
+
+                cut(head.header(), kQuant);
+                // (a)->...-+
+                //          v
+                // [head]->(b)->...->(oldTail)->(c)->...->(d)->...->(tail)
+                mm::safePoint(threadData);
+
+                cut(oldTail, kQuant);
+                // (a)->...-+           (c)->...-+
+                //          v                    v
+                // [head]->(b)->...->(oldTail)->(d)->...->(tail)
+                mm::safePoint(threadData);
+                ++iter;
+            }
+
+            std::size_t length = 0;
+            auto* node = &test_support::Object<Payload>::FromObjHeader(head.header());
+            while ((*node)->field1.accessor().load() != nullptr) {
+                node = &test_support::Object<Payload>::FromObjHeader((*node)->field1.accessor().load());
+                ++length;
+            }
+
+            EXPECT_THAT(length, kQuant * 2);
+        }));
+    }
+
+    schedulerFuture.wait();
+    for (auto& future : mutatorFutures) {
+        future.wait();
+    }
+}
+
+TYPED_TEST_P(TracingGCTest, WeakResuractionInMark) {
+    constexpr auto kObjectsPerThread = 100;
+    std::vector<Mutator> mutators(kDefaultThreadCount);
+    std::vector<test_support::RegularWeakReferenceImpl*> weaks(kDefaultThreadCount);
+    std::vector<test_support::Object<Payload>*> roots(kDefaultThreadCount);
+
+    // initialize threads
+    for (int i = 0; i < kDefaultThreadCount; ++i) {
+        mutators[i].Execute([&, i](mm::ThreadData& threadData, Mutator& mutator) {
+            // make sure GC will have somthing to do whil we play with weak references
+            for (int j = 0; j < kObjectsPerThread; ++j) {
+                auto& object = AllocateObject(threadData);
+                object->field1 = roots[i]->header();
+                roots[i] = &object;
+            }
+            mutator.AddGlobalRoot(roots[i]->header());
+
+            auto& weakReferee = AllocateObject(threadData);
+            auto& weakRef = [&threadData, &weakReferee]() -> test_support::RegularWeakReferenceImpl& {
+                ObjHolder holder;
+                return test_support::InstallWeakReference(threadData, weakReferee.header(), holder.slot());
+            }();
+            EXPECT_NE(weakRef.get(), nullptr);
+            weaks[i] = &weakRef;
+            mutator.AddGlobalRoot(weakRef.header());
+        }).wait();
+    }
+
+    auto epoch = mm::GlobalData::Instance().gc().Schedule();
+    std::atomic gcDone = false;
+
+    // Spin until thread suspension is requested.
+    while (!mm::IsThreadSuspensionRequested()) {}
+
+    std::vector<std::future<void>> mutatorFutures;
+    for (int i = 0; i < kDefaultThreadCount; ++i) {
+        mutatorFutures.emplace_back(mutators[i].Execute([&, i](mm::ThreadData& threadData, Mutator& mutator) noexcept {
+            safePoint(threadData);
+
+            auto weakReferee = weaks[i]->get();
+            (*roots[i])->field2 = weakReferee;
+            bool resurected = weakReferee != nullptr;
+
+            while (!gcDone.load(std::memory_order_relaxed)) {
+                safePoint(threadData);
+            }
+            if (resurected) {
+                EXPECT_NE(weaks[i]->get(), nullptr);
+            } else {
+                EXPECT_EQ(weaks[i]->get(), nullptr);
+            }
+        }));
+    }
+
+    mm::GlobalData::Instance().gc().WaitFinalizers(epoch);
+    gcDone = true;
+
+    for (auto& future : mutatorFutures) {
+        future.wait();
+    }
+
+    for (int i = 0; i < kDefaultThreadCount; ++i) {
+        mutators[i].Execute([&, i](mm::ThreadData&, Mutator&) noexcept {
+            if (auto weakReferee = weaks[i]->get()) {
+                auto& extraObj = *mm::ExtraObjectData::Get(weakReferee);
+                extraObj.ClearRegularWeakReferenceImpl();
+                extraObj.Uninstall();
+                alloc::destroyExtraObjectData(extraObj);
+            }
+        }).wait();
+    }
+}
+
 #define TRACING_GC_TEST_LIST \
     RootSet, \
     InterconnectedRootSet, \
@@ -1092,7 +1266,9 @@ TYPED_TEST_P(TracingGCTest, FreeObjectWithFreeWeakReversedOrder) {
     MultipleMutatorsAddToRootSetAfterCollectionRequested, \
     CrossThreadReference, \
     NewThreadsWhileRequestingCollection, \
-    FreeObjectWithFreeWeakReversedOrder
+    FreeObjectWithFreeWeakReversedOrder, \
+    MutateBetweenSafePoints, \
+    WeakResuractionInMark
 
 template <typename FixtureImpl>
 class STWMarkGCTest : public TracingGCTest<FixtureImpl> {};
diff --git a/kotlin-native/runtime/src/gc/noop/cpp/GCImpl.cpp b/kotlin-native/runtime/src/gc/noop/cpp/GCImpl.cpp
index bcedacaa980..7e4b72cfbd0 100644
--- a/kotlin-native/runtime/src/gc/noop/cpp/GCImpl.cpp
+++ b/kotlin-native/runtime/src/gc/noop/cpp/GCImpl.cpp
@@ -48,9 +48,6 @@ ALWAYS_INLINE void gc::GC::processObjectInMark(void* state, ObjHeader* object) n
 // static
 ALWAYS_INLINE void gc::GC::processArrayInMark(void* state, ArrayHeader* array) noexcept {}
 
-// static
-ALWAYS_INLINE void gc::GC::processFieldInMark(void* state, ObjHeader* field) noexcept {}
-
 int64_t gc::GC::Schedule() noexcept {
     return 0;
 }
diff --git a/kotlin-native/runtime/src/gc/pmcs/cpp/GCImpl.cpp b/kotlin-native/runtime/src/gc/pmcs/cpp/GCImpl.cpp
index 06b37d3a11f..5a5025fb391 100644
--- a/kotlin-native/runtime/src/gc/pmcs/cpp/GCImpl.cpp
+++ b/kotlin-native/runtime/src/gc/pmcs/cpp/GCImpl.cpp
@@ -67,11 +67,6 @@ ALWAYS_INLINE void gc::GC::processArrayInMark(void* state, ArrayHeader* array) n
     gc::internal::processArrayInMark<gc::mark::ParallelMark::MarkTraits>(state, array);
 }
 
-// static
-ALWAYS_INLINE void gc::GC::processFieldInMark(void* state, ObjHeader* field) noexcept {
-    gc::internal::processFieldInMark<gc::mark::ParallelMark::MarkTraits>(state, field);
-}
-
 int64_t gc::GC::Schedule() noexcept {
     return impl_->gc().state().schedule();
 }
diff --git a/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMark.hpp b/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMark.hpp
index 6d41d096411..7b6173e87f4 100644
--- a/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMark.hpp
+++ b/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMark.hpp
@@ -82,6 +82,8 @@ public:
     public:
         using MarkQueue = ParallelProcessor::Worker;
 
+        static constexpr auto kAllowHeapToStackRefs = true;
+
         static void clear(MarkQueue& queue) noexcept {
             RuntimeAssert(queue.localEmpty(), "Mark queue must be empty");
         }
diff --git a/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMarkConcurrentSweep.cpp b/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMarkConcurrentSweep.cpp
index cda9871e130..f24c07aa9c0 100644
--- a/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMarkConcurrentSweep.cpp
+++ b/kotlin-native/runtime/src/gc/pmcs/cpp/ParallelMarkConcurrentSweep.cpp
@@ -148,7 +148,7 @@ void gc::ParallelMarkConcurrentSweep::PerformFullGC(int64_t epoch) noexcept {
     markDispatcher_.beginMarkingEpoch(gcHandle);
     GCLogDebug(epoch, "Main GC requested marking in mutators");
 
-    stopTheWorld(gcHandle);
+    stopTheWorld(gcHandle, "GC stop the world #1: mark");
 
     auto& scheduler = gcScheduler_;
     scheduler.onGCStart();
@@ -168,7 +168,7 @@ void gc::ParallelMarkConcurrentSweep::PerformFullGC(int64_t epoch) noexcept {
     gc::processWeaks<DefaultProcessWeaksTraits>(gcHandle, mm::SpecialRefRegistry::instance());
 
     if (compiler::concurrentWeakSweep()) {
-        stopTheWorld(gcHandle);
+        stopTheWorld(gcHandle, "GC stop the world #2: prepare heap for sweep");
         gc::DisableWeakRefBarriers();
     }
 
diff --git a/kotlin-native/runtime/src/gc/stms/cpp/GCImpl.cpp b/kotlin-native/runtime/src/gc/stms/cpp/GCImpl.cpp
index 35fc1fe38f7..e3e103ab449 100644
--- a/kotlin-native/runtime/src/gc/stms/cpp/GCImpl.cpp
+++ b/kotlin-native/runtime/src/gc/stms/cpp/GCImpl.cpp
@@ -60,11 +60,6 @@ ALWAYS_INLINE void gc::GC::processArrayInMark(void* state, ArrayHeader* array) n
     gc::internal::processArrayInMark<gc::internal::MarkTraits>(state, array);
 }
 
-// static
-ALWAYS_INLINE void gc::GC::processFieldInMark(void* state, ObjHeader* field) noexcept {
-    gc::internal::processFieldInMark<gc::internal::MarkTraits>(state, field);
-}
-
 int64_t gc::GC::Schedule() noexcept {
     return impl_->gc().state().schedule();
 }
diff --git a/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.cpp b/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.cpp
index 03f722d02b1..91c2a6ed5a4 100644
--- a/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.cpp
+++ b/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.cpp
@@ -57,7 +57,7 @@ bool gc::SameThreadMarkAndSweep::FinalizersThreadIsRunning() noexcept {
 void gc::SameThreadMarkAndSweep::PerformFullGC(int64_t epoch) noexcept {
     auto gcHandle = GCHandle::create(epoch);
 
-    stopTheWorld(gcHandle);
+    stopTheWorld(gcHandle, "GC stop the world");
 
     auto& scheduler = gcScheduler_;
     scheduler.onGCStart();
diff --git a/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.hpp b/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.hpp
index cd2f86eb7d6..602f426a269 100644
--- a/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.hpp
+++ b/kotlin-native/runtime/src/gc/stms/cpp/SameThreadMarkAndSweep.hpp
@@ -68,6 +68,8 @@ namespace internal {
 struct MarkTraits {
     using MarkQueue = gc::SameThreadMarkAndSweep::MarkQueue;
 
+    static constexpr auto kAllowHeapToStackRefs = true;
+
     static void clear(MarkQueue& queue) noexcept { queue.clear(); }
 
     static ObjHeader* tryDequeue(MarkQueue& queue) noexcept {
diff --git a/kotlin-native/runtime/src/main/cpp/Logging.hpp b/kotlin-native/runtime/src/main/cpp/Logging.hpp
index 82a7edb237e..d7427b8f974 100644
--- a/kotlin-native/runtime/src/main/cpp/Logging.hpp
+++ b/kotlin-native/runtime/src/main/cpp/Logging.hpp
@@ -38,8 +38,9 @@ enum class Tag : int32_t {
     kAlloc = 6,
     kBalancing = 7,
     kBarriers = 8,
+    kGCMark = 9,
 
-    kEnumSize = 9
+    kEnumSize = 10
 };
 
 namespace internal {
@@ -65,6 +66,7 @@ inline const char* name(Tag tag) {
         case Tag::kAlloc: return "alloc";
         case Tag::kBalancing: return "balancing";
         case Tag::kBarriers: return "barriers";
+        case Tag::kGCMark: return "gcMark";
 
         case Tag::kEnumSize: break;
     }
diff --git a/kotlin-native/runtime/src/main/cpp/Memory.h b/kotlin-native/runtime/src/main/cpp/Memory.h
index 32d5c8e45bd..9ff035f680e 100644
--- a/kotlin-native/runtime/src/main/cpp/Memory.h
+++ b/kotlin-native/runtime/src/main/cpp/Memory.h
@@ -77,7 +77,9 @@ struct ObjHeader {
    * Hardware guaranties on many supported platforms doesn't allow this to happen.
    */
   const TypeInfo* type_info() const {
-      return atomicGetRelaxed(&clearPointerBits(typeInfoOrMetaRelaxed(), OBJECT_TAG_MASK)->typeInfo_);
+      const TypeInfo* typeInfo = atomicGetRelaxed(&clearPointerBits(typeInfoOrMetaRelaxed(), OBJECT_TAG_MASK)->typeInfo_);
+      RuntimeAssert(typeInfo != nullptr, "TypeInfo ptr in object %p in null", this);
+      return typeInfo;
   }
 
   bool has_meta_object() const {
@@ -599,7 +601,6 @@ void compactObjectPoolInCurrentThread() noexcept;
 
 RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processObjectInMark(void* state, ObjHeader* object);
 RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processArrayInMark(void* state, ObjHeader* object);
-RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processFieldInMark(void* state, ObjHeader* field);
 RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processEmptyObjectInMark(void* state, ObjHeader* object);
 
 #endif // RUNTIME_MEMORY_H
diff --git a/kotlin-native/runtime/src/main/cpp/ParallelProcessor.hpp b/kotlin-native/runtime/src/main/cpp/ParallelProcessor.hpp
index 74cec8eadf4..e8bab4a6536 100644
--- a/kotlin-native/runtime/src/main/cpp/ParallelProcessor.hpp
+++ b/kotlin-native/runtime/src/main/cpp/ParallelProcessor.hpp
@@ -95,7 +95,7 @@ public:
         explicit WorkSource(ParallelProcessor& dispatcher) : dispatcher_(dispatcher) {}
 
         ALWAYS_INLINE bool retainsNoWork() const noexcept {
-            return batch_.empty() && this->localEmpty();
+            return batch_.empty() && overflowList().empty();
         }
 
         ALWAYS_INLINE bool tryPush(typename ListImpl::reference value) noexcept {
@@ -119,7 +119,7 @@ public:
                 if (!batch_.empty()) {
                     bool released = dispatcher_.releaseBatch(std::move(batch_));
                     if (released) {
-                        RuntimeLogDebug({ kTagBalancing }, "Work butch flushed");
+                        RuntimeLogDebug({ kTagBalancing }, "Work batch flushed");
                         batch_ = Batch{};
                     } else {
                         RuntimeLogDebug({ kTagBalancing }, "Failed to force flush work queue");
@@ -130,7 +130,7 @@ public:
                 if (overflowList().empty()) {
                     return true;
                 } else {
-                    RuntimeLogDebug({ kTagBalancing }, "Refiling batch from overflow list");
+                    RuntimeLogDebug({ kTagBalancing }, "Refilling batch from overflow list");
                     batch_.fillFrom(overflowList());
                 }
             }
@@ -141,6 +141,10 @@ public:
             return this->localQueue_;
         }
 
+        const ListImpl& overflowList() const noexcept {
+            return this->localQueue_;
+        }
+
         ParallelProcessor& dispatcher_;
         Batch batch_;
     };
@@ -222,6 +226,13 @@ public:
         return registeredWorkers_.load(std::memory_order_relaxed);
     }
 
+    /** Prepare for a new round of work processing. Must be called only after a previous round is fully finished. */
+    void resetForNewWork() noexcept {
+        RuntimeAssert(allDone_, "A work processing iteration must be finished");
+        RuntimeAssert(waitingWorkers_ == 0, "There must be no workers sleeping inside processing loop");
+        allDone_ = false;
+    }
+
 private:
     bool releaseBatch(Batch&& batch) {
         RuntimeAssert(!batch.empty(), "A batch to release into shared pool must be non-empty");
diff --git a/kotlin-native/runtime/src/mm/cpp/Memory.cpp b/kotlin-native/runtime/src/mm/cpp/Memory.cpp
index d08b33d89c3..cc1b4196f40 100644
--- a/kotlin-native/runtime/src/mm/cpp/Memory.cpp
+++ b/kotlin-native/runtime/src/mm/cpp/Memory.cpp
@@ -575,10 +575,6 @@ RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processArrayInMark(void* st
     gc::GC::processArrayInMark(state, object->array());
 }
 
-RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processFieldInMark(void* state, ObjHeader* field) {
-    gc::GC::processFieldInMark(state, field);
-}
-
 RUNTIME_NOTHROW ALWAYS_INLINE extern "C" void Kotlin_processEmptyObjectInMark(void* state, ObjHeader* object) {
     // Empty object. Nothing to do.
     // TODO: Try to generate it in the code generator.
diff --git a/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.cpp b/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.cpp
index 3e8d9bd2d9b..1e9da672b6e 100644
--- a/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.cpp
+++ b/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.cpp
@@ -23,7 +23,7 @@ namespace {
 
 } // namespace
 
-std::atomic<bool> kotlin::mm::internal::gSuspensionRequested = false;
+std::atomic<mm::internal::SuspensionReason> mm::internal::gSuspensionReauestReason = nullptr;
 
 ALWAYS_INLINE mm::ThreadSuspensionData::MutatorPauseHandle::MutatorPauseHandle(const char* reason, mm::ThreadData& threadData) noexcept
     : reason_(reason), threadData_(threadData), pauseStartTimeMicros_(konan::getTimeMicros())
@@ -65,7 +65,7 @@ kotlin::ThreadState kotlin::mm::ThreadSuspensionData::setState(kotlin::ThreadSta
 
 NO_EXTERNAL_CALLS_CHECK void kotlin::mm::ThreadSuspensionData::suspendIfRequested() noexcept {
     if (IsThreadSuspensionRequested()) {
-        auto pauseHandle = pauseMutationInScope("stop the world");
+        auto pauseHandle = pauseMutationInScope(internal::gSuspensionReauestReason.load(std::memory_order_relaxed));
 
         threadData_.gc().OnSuspendForGC();
         std::unique_lock lock(gSuspensionMutex);
@@ -80,18 +80,18 @@ ALWAYS_INLINE mm::ThreadSuspensionData::MutatorPauseHandle mm::ThreadSuspensionD
     return MutatorPauseHandle(reason, threadData_);
 }
 
-bool kotlin::mm::RequestThreadsSuspension() noexcept {
+bool kotlin::mm::RequestThreadsSuspension(internal::SuspensionReason reason) noexcept {
     CallsCheckerIgnoreGuard guard;
 
     RuntimeAssert(gSafePointActivator == std::nullopt, "Current thread already suspended threads.");
     {
         std::unique_lock lock(gSuspensionMutex);
         // Someone else has already suspended threads.
-        if (internal::gSuspensionRequested.load(std::memory_order_relaxed)) {
+        if (internal::gSuspensionReauestReason.load(std::memory_order_relaxed) != nullptr) {
             return false;
         }
         gSafePointActivator = mm::SafePointActivator();
-        internal::gSuspensionRequested.store(true);
+        internal::gSuspensionReauestReason.store(reason);
     }
 
     return true;
@@ -116,7 +116,7 @@ void kotlin::mm::ResumeThreads() noexcept {
     // https://en.cppreference.com/w/cpp/thread/condition_variable
     {
         std::unique_lock lock(gSuspensionMutex);
-        internal::gSuspensionRequested.store(false);
+        internal::gSuspensionReauestReason.store(nullptr);
     }
     gSuspensionCondVar.notify_all();
 }
diff --git a/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.hpp b/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.hpp
index d94ef8431eb..f6383b4bb1a 100644
--- a/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.hpp
+++ b/kotlin-native/runtime/src/mm/cpp/ThreadSuspension.hpp
@@ -9,6 +9,7 @@
 #include <atomic>
 
 #include "Memory.h"
+#include "Utils.hpp"
 
 namespace kotlin {
 namespace mm {
@@ -17,14 +18,15 @@ class ThreadData;
 
 namespace internal {
 
-extern std::atomic<bool> gSuspensionRequested;
+using SuspensionReason = const char*;
+extern std::atomic<SuspensionReason> gSuspensionReauestReason;
 
 } // namespace internal
 
 inline bool IsThreadSuspensionRequested() noexcept {
     // Must use seq_cst ordering for synchronization with GC
     // in native->runnable transition.
-    return internal::gSuspensionRequested.load();
+    return internal::gSuspensionReauestReason.load();
 }
 
 class ThreadSuspensionData : private Pinned {
@@ -71,7 +73,7 @@ private:
     mm::ThreadData& threadData_;
 };
 
-bool RequestThreadsSuspension() noexcept;
+bool RequestThreadsSuspension(const char* reason) noexcept;
 void WaitForThreadsSuspension() noexcept;
 
 /**
@@ -80,8 +82,8 @@ void WaitForThreadsSuspension() noexcept;
  * of this call will be suspended on exit from the Native state.
  * Returns false if some other thread has suspended the threads.
  */
-inline bool SuspendThreads() noexcept {
-    if (!RequestThreadsSuspension()) {
+inline bool SuspendThreads(internal::SuspensionReason reason) noexcept {
+    if (!RequestThreadsSuspension(reason)) {
         return false;
     }
     WaitForThreadsSuspension();
diff --git a/kotlin-native/runtime/src/mm/cpp/ThreadSuspensionTest.cpp b/kotlin-native/runtime/src/mm/cpp/ThreadSuspensionTest.cpp
index 3bdd90ae458..051f7a3f353 100644
--- a/kotlin-native/runtime/src/mm/cpp/ThreadSuspensionTest.cpp
+++ b/kotlin-native/runtime/src/mm/cpp/ThreadSuspensionTest.cpp
@@ -128,7 +128,7 @@ TEST_F(ThreadSuspensionTest, SimpleStartStop) {
         reportProgress(i, kIterations);
         canStart = true;
 
-        mm::SuspendThreads();
+        mm::SuspendThreads("test");
         auto suspended = collectSuspended();
         EXPECT_THAT(suspended, testing::Each(true));
         EXPECT_EQ(mm::IsThreadSuspensionRequested(), true);
@@ -171,7 +171,7 @@ TEST_F(ThreadSuspensionTest, SwitchStateToNative) {
         reportProgress(i, kIterations);
         canStart = true;
 
-        mm::SuspendThreads();
+        mm::SuspendThreads("test");
         EXPECT_EQ(mm::IsThreadSuspensionRequested(), true);
 
         mm::ResumeThreads();
@@ -196,7 +196,7 @@ TEST_F(ThreadSuspensionTest, ConcurrentSuspend) {
             ready[i] = true;
             waitUntilThreadsAreReady();
 
-            bool success = mm::SuspendThreads();
+            bool success = mm::SuspendThreads("test");
             if (success) {
                 successCount++;
                 auto allThreadData = collectThreadData();
@@ -248,7 +248,7 @@ TEST_F(ThreadSuspensionTest, FileInitializationWithSuspend) {
     waitUntilThreadsAreReady();
 
     auto gcThread = std::async(std::launch::async, [] {
-        mm::RequestThreadsSuspension();
+        mm::RequestThreadsSuspension("test");
         mm::WaitForThreadsSuspension();
         mm::ResumeThreads();
     });