diff --git a/src/hotspot/share/c1/c1_GraphBuilder.cpp b/src/hotspot/share/c1/c1_GraphBuilder.cpp
index f910ecadc164a..db7722ff09004 100644
--- a/src/hotspot/share/c1/c1_GraphBuilder.cpp
+++ b/src/hotspot/share/c1/c1_GraphBuilder.cpp
@@ -3560,6 +3560,14 @@ const char* GraphBuilder::check_can_parse(ciMethod* callee) const {
 const char* GraphBuilder::should_not_inline(ciMethod* callee) const {
   if ( compilation()->directive()->should_not_inline(callee)) return "disallowed by CompileCommand";
   if ( callee->dont_inline())          return "don't inline by annotation";
+
+  // Don't inline a method that changes Thread.currentThread() except
+  // into another method that is annotated @ChangesCurrentThread.
+  if (callee->changes_current_thread()
+      && !compilation()->method()->changes_current_thread()) {
+    return "method changes current thread";
+  }
+
   return nullptr;
 }
 
diff --git a/src/hotspot/share/c1/c1_LIRGenerator.cpp b/src/hotspot/share/c1/c1_LIRGenerator.cpp
index f6807abcd7acd..ea29940daf66f 100644
--- a/src/hotspot/share/c1/c1_LIRGenerator.cpp
+++ b/src/hotspot/share/c1/c1_LIRGenerator.cpp
@@ -1379,7 +1379,14 @@ void LIRGenerator::do_JavaThreadField(Intrinsic* x, ByteSize offset) {
   LIR_Opr temp = new_register(T_ADDRESS);
   LIR_Opr reg = rlock_result(x);
   __ move(new LIR_Address(getThreadPointer(), in_bytes(offset), T_ADDRESS), temp);
-  access_load(IN_NATIVE, T_OBJECT,
+  // OopHandle stores uncompressed oops in native memory.
+  // Use IN_NATIVE to ensure a raw 64-bit load without compressed oop handling.
+  // Use MO_ACQUIRE so that subsequent loads (e.g. Thread.cont used by
+  // Continuation.yield) cannot float above this load on weakly-ordered
+  // architectures such as AArch64.  Without acquire semantics the hardware
+  // may reorder a later field load before the OopHandle dereference,
+  // observing a stale value after a virtual thread migrates between carriers.
+  access_load(IN_NATIVE | MO_ACQUIRE, T_OBJECT,
               LIR_OprFact::address(new LIR_Address(temp, T_OBJECT)), reg);
 }
 
diff --git a/src/hotspot/share/gc/shared/c1/barrierSetC1.cpp b/src/hotspot/share/gc/shared/c1/barrierSetC1.cpp
index a31078f7e6790..b4ad8f9a29f55 100644
--- a/src/hotspot/share/gc/shared/c1/barrierSetC1.cpp
+++ b/src/hotspot/share/gc/shared/c1/barrierSetC1.cpp
@@ -179,6 +179,7 @@ void BarrierSetC1::load_at_resolved(LIRAccess& access, LIR_Opr result) {
   LIRGenerator *gen = access.gen();
   DecoratorSet decorators = access.decorators();
   bool is_volatile = (decorators & MO_SEQ_CST) != 0;
+  bool is_acquire  = (decorators & MO_ACQUIRE) != 0;
   bool needs_atomic = AlwaysAtomicAccesses && !access_is_atomic(result->type());
   bool needs_patching = (decorators & C1_NEEDS_PATCHING) != 0;
   bool mask_boolean = (decorators & C1_MASK_BOOLEAN) != 0;
@@ -197,7 +198,7 @@ void BarrierSetC1::load_at_resolved(LIRAccess& access, LIR_Opr result) {
     __ load(access.resolved_addr()->as_address_ptr(), result, access.access_emit_info(), patch_code);
   }
 
-  if (is_volatile) {
+  if (is_volatile || is_acquire) {
     __ membar_acquire();
   }
 
diff --git a/src/hotspot/share/opto/library_call.cpp b/src/hotspot/share/opto/library_call.cpp
index 3627d06a87a2d..966a8fb7e5e0f 100644
--- a/src/hotspot/share/opto/library_call.cpp
+++ b/src/hotspot/share/opto/library_call.cpp
@@ -1000,7 +1000,14 @@ Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_o
       : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
   thread_obj_handle = _gvn.transform(thread_obj_handle);
 
-  DecoratorSet decorators = IN_NATIVE;
+  // OopHandle stores uncompressed oops in native memory.
+  // Use IN_NATIVE to ensure proper handling without compressed oop decoding.
+  // Use MO_ACQUIRE so that subsequent loads (e.g. Thread.cont used by
+  // Continuation.yield) cannot float above this load on weakly-ordered
+  // architectures such as AArch64.  Without acquire semantics the hardware
+  // may reorder a later field load before the OopHandle dereference,
+  // observing a stale value after a virtual thread migrates between carriers.
+  DecoratorSet decorators = IN_NATIVE | MO_ACQUIRE;
   if (is_immutable) {
     decorators |= C2_IMMUTABLE_MEMORY;
   }
diff --git a/src/java.base/share/classes/jdk/internal/vm/Continuation.java b/src/java.base/share/classes/jdk/internal/vm/Continuation.java
index a7eb3ea6a9fa5..147cdaaae6fd5 100644
--- a/src/java.base/share/classes/jdk/internal/vm/Continuation.java
+++ b/src/java.base/share/classes/jdk/internal/vm/Continuation.java
@@ -342,6 +342,7 @@ private boolean isEmpty() {
      * @throws IllegalStateException if not currently in the given {@code scope},
      */
     @Hidden
+    @DontInline
     @JvmtiHideEvents
     public static boolean yield(ContinuationScope scope) {
         Continuation cont = JLA.getContinuation(currentCarrierThread());