# HG changeset patch
# User smonteith
# Date 1587034905 -3600
#      Thu Apr 16 12:01:45 2020 +0100
# Node ID edd3e9bdae09ec2d5d84c57f138df1d58ceabbee
# Parent  2cd3a8f8be61768b79119298f309bcf2bad9c939
[mq]: aarch64-jdk-nmethod-barriers-3.patch

diff --git a/src/hotspot/cpu/aarch64/aarch64.ad b/src/hotspot/cpu/aarch64/aarch64.ad
--- a/src/hotspot/cpu/aarch64/aarch64.ad
+++ b/src/hotspot/cpu/aarch64/aarch64.ad
@@ -1635,6 +1635,20 @@
     st->print("mov  rscratch1, #%d\n\t", framesize - 2 * wordSize);
     st->print("sub  sp, sp, rscratch1");
   }
+  if (C->stub_function() == NULL && BarrierSet::barrier_set()->barrier_set_nmethod() != NULL) {
+    st->print("\n\t");
+    st->print("ldr  rscratch1, [guard]\n\t");
+    st->print("dmb ishld\n\t");
+    st->print("ldr  rscratch2, [rthread, #thread_disarmed_offset]\n\t");
+    st->print("cmp  rscratch1, rscratch2\n\t");
+    st->print("b.eq skip");
+    st->print("\n\t");
+    st->print("blr #nmethod_entry_barrier_stub\n\t");
+    st->print("b skip\n\t");
+    st->print("guard: int\n\t");
+    st->print("\n\t");
+    st->print("skip:\n\t");
+  }
 }
 #endif
 
@@ -1667,6 +1681,11 @@
 
   __ build_frame(framesize);
 
+  if (C->stub_function() == NULL) {
+    BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+    bs->nmethod_entry_barrier(&_masm);
+  }
+
   if (VerifyStackAtCalls) {
     Unimplemented();
   }
diff --git a/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/c1_MacroAssembler_aarch64.cpp
@@ -28,6 +28,7 @@
 #include "c1/c1_Runtime1.hpp"
 #include "classfile/systemDictionary.hpp"
 #include "gc/shared/collectedHeap.hpp"
+#include "gc/shared/barrierSetAssembler.hpp"
 #include "interpreter/interpreter.hpp"
 #include "oops/arrayOop.hpp"
 #include "oops/markWord.hpp"
@@ -336,6 +337,10 @@
   // Note that we do this before doing an enter().
   generate_stack_overflow_check(bang_size_in_bytes);
   MacroAssembler::build_frame(framesize + 2 * wordSize);
+
+  // Insert nmethod entry barrier into frame.
+  BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+  bs->nmethod_entry_barrier(this);
 }
 
 void C1_MacroAssembler::remove_frame(int framesize) {
diff --git a/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -23,12 +23,17 @@
  */
 
 #include "precompiled.hpp"
+#include "gc/shared/barrierSet.hpp"
 #include "gc/shared/barrierSetAssembler.hpp"
+#include "gc/shared/barrierSetNMethod.hpp"
 #include "gc/shared/collectedHeap.hpp"
+#include "interpreter/interp_masm.hpp"
 #include "memory/universe.hpp"
 #include "runtime/jniHandles.hpp"
+#include "runtime/sharedRuntime.hpp"
 #include "runtime/thread.hpp"
 
+
 #define __ masm->
 
 void BarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
@@ -229,3 +234,67 @@
   }
   __ str(t1, Address(rthread, in_bytes(JavaThread::allocated_bytes_offset())));
 }
+
+void BarrierSetAssembler::nmethod_entry_barrier(MacroAssembler* masm) {
+  BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
+
+  if (bs_nm == NULL) {
+    return;
+  }
+
+  Label skip, guard;
+  Address thread_disarmed_addr(rthread, in_bytes(bs_nm->thread_disarmed_offset()));
+
+  __ ldrw(rscratch1, guard);
+
+  // Subsequent loads of oops must occur after load of guard value.
+  // BarrierSetNMethod::disarm sets guard with release semantics.
+  __ membar(__ LoadLoad);
+  __ ldrw(rscratch2, thread_disarmed_addr);
+  __ cmpw(rscratch1, rscratch2);
+  __ br(Assembler::EQ, skip);
+
+  __ mov(rscratch1, StubRoutines::aarch64::method_entry_barrier());
+  __ blr(rscratch1);
+  __ b(skip);
+
+  __ bind(guard);
+
+  __ emit_int32(0);   // nmethod guard value. Skipped over in common case.
+
+  __ bind(skip);
+}
+
+void BarrierSetAssembler::c2i_entry_barrier(MacroAssembler* masm) {
+  BarrierSetNMethod* bs = BarrierSet::barrier_set()->barrier_set_nmethod();
+  if (bs == NULL) {
+    return;
+  }
+
+  Label bad_call;
+  __ cbz(rmethod, bad_call);
+
+  // Pointer chase to the method holder to find out if the method is concurrently unloading.
+  Label method_live;
+  __ load_method_holder_cld(rscratch1, rmethod);
+
+  // Is it a strong CLD?
+  __ ldr(rscratch2, Address(rscratch1, ClassLoaderData::keep_alive_offset()));
+  __ cbnz(rscratch2, method_live);
+
+  // Is it a weak but alive CLD?
+  __ stp(r10, r11, Address(__ pre(sp, -2 * wordSize)));
+  __ ldr(r10, Address(rscratch1, ClassLoaderData::holder_offset()));
+
+  // Uses rscratch1 & rscratch2, so we must pass new temporaries.
+  __ resolve_weak_handle(r10, r11);
+  __ mov(rscratch1, r10);
+  __ ldp(r10, r11, Address(__ post(sp, 2 * wordSize)));
+  __ cbnz(rscratch1, method_live);
+
+  __ bind(bad_call);
+
+  __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+  __ bind(method_live);
+}
+
diff --git a/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -26,6 +26,8 @@
 #define CPU_AARCH64_GC_SHARED_BARRIERSETASSEMBLER_AARCH64_HPP
 
 #include "asm/macroAssembler.hpp"
+#include "gc/shared/barrierSet.hpp"
+#include "gc/shared/barrierSetNMethod.hpp"
 #include "memory/allocation.hpp"
 #include "oops/access.hpp"
 
@@ -72,6 +74,10 @@
     Label&   slow_case                 // continuation point if fast allocation fails
   );
   virtual void barrier_stubs_init() {}
+
+  virtual void nmethod_entry_barrier(MacroAssembler* masm);
+  virtual void c2i_entry_barrier(MacroAssembler* masm);
+
 };
 
 #endif // CPU_AARCH64_GC_SHARED_BARRIERSETASSEMBLER_AARCH64_HPP
diff --git a/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -23,18 +23,144 @@
  */
 
 #include "precompiled.hpp"
+#include "code/codeCache.hpp"
+#include "code/nativeInst.hpp"
 #include "gc/shared/barrierSetNMethod.hpp"
+#include "logging/log.hpp"
+#include "memory/resourceArea.hpp"
+#include "runtime/sharedRuntime.hpp"
+#include "runtime/thread.hpp"
+#include "utilities/align.hpp"
 #include "utilities/debug.hpp"
+#include "icache_aarch64.hpp"
+
+class NativeNMethodBarrier: public NativeInstruction {
+  address instruction_address() const { return addr_at(0); }
+
+  int *guard_addr() {
+    return reinterpret_cast<int*>(instruction_address() + 10 * 4);
+  }
+
+public:
+  int get_value() {
+    return Atomic::load_acquire(guard_addr());
+  }
+
+  void set_value(int value) {
+    Atomic::release_store(guard_addr(), value);
+  }
+
+  void verify() const;
+};
+
+// Store the instruction bitmask, bits and name for checking the barrier.
+struct CheckInsn {
+  uint32_t mask;
+  uint32_t bits;
+  const char *name;
+};
+
+static const struct CheckInsn barrierInsn[] = {
+  { 0xff000000, 0x18000000, "ldr (literal)" },
+  { 0xfffff0ff, 0xd50330bf, "dmb" },
+  { 0xffc00000, 0xb9400000, "ldr"},
+  { 0x7f20001f, 0x6b00001f, "cmp"},
+  { 0xff00001f, 0x54000000, "b.eq"},
+  { 0xff800000, 0xd2800000, "mov"},
+  { 0xff800000, 0xf2800000, "movk"},
+  { 0xff800000, 0xf2800000, "movk"},
+  { 0xfffffc1f, 0xd63f0000, "blr"},
+  { 0xfc000000, 0x14000000, "b"}
+};
 
+// The encodings must match the instructions emitted by
+// BarrierSetAssembler::nmethod_entry_barrier. The matching ignores the specific
+// register numbers and immediate values in the encoding.
+void NativeNMethodBarrier::verify() const {
+  intptr_t addr = (intptr_t) instruction_address();
+  for(unsigned int i = 0; i < sizeof(barrierInsn)/sizeof(struct CheckInsn); i++ ) {
+    uint32_t inst = *((uint32_t*) addr);
+    if ((inst & barrierInsn[i].mask) != barrierInsn[i].bits) {
+      tty->print_cr("Addr: " INTPTR_FORMAT " Code: 0x%x", addr, inst);
+      fatal("not an %s instruction.", barrierInsn[i].name);
+    }
+    addr +=4;
+  }
+}
+
+
+/* We're called from an nmethod when we need to deoptimize it. We do
+   this by throwing away the nmethod's frame and jumping to the
+   ic_miss stub. This looks like there has been an IC miss at the
+   entry of the nmethod, so we resolve the call, which will fall back
+   to the interpreter if the nmethod has been unloaded. */
 void BarrierSetNMethod::deoptimize(nmethod* nm, address* return_address_ptr) {
-  ShouldNotReachHere();
+
+  typedef struct {
+    intptr_t *sp; intptr_t *fp; address lr; address pc;
+  } frame_pointers_t;
+
+  frame_pointers_t *new_frame = (frame_pointers_t *)(return_address_ptr - 5);
+
+  JavaThread *thread = (JavaThread*)Thread::current();
+  RegisterMap reg_map(thread, false);
+  frame frame = thread->last_frame();
+
+  assert(frame.is_compiled_frame() || frame.is_native_frame(), "must be");
+  assert(frame.cb() == nm, "must be");
+  frame = frame.sender(&reg_map);
+
+  LogTarget(Trace, nmethod, barrier) out;
+  if (out.is_enabled()) {
+    Thread* thread = Thread::current();
+    assert(thread->is_Java_thread(), "must be JavaThread");
+    JavaThread* jth = (JavaThread*) thread;
+    ResourceMark mark;
+    log_trace(nmethod, barrier)("deoptimize(nmethod: %s(%p), return_addr: %p, osr: %d, thread: %p(%s), making rsp: %p) -> %p",
+                                nm->method()->name_and_sig_as_C_string(),
+                                nm, *(address *) return_address_ptr, nm->is_osr_method(), jth,
+                                jth->get_thread_name(), frame.sp(), nm->verified_entry_point());
+  }
+
+  new_frame->sp = frame.sp();
+  new_frame->fp = frame.fp();
+  new_frame->lr = frame.pc();
+  new_frame->pc = SharedRuntime::get_handle_wrong_method_stub();
+}
+
+// This is the offset of the entry barrier from where the frame is completed.
+// If any code changes between the end of the verified entry where the entry
+// barrier resides, and the completion of the frame, then
+// NativeNMethodCmpBarrier::verify() will immediately complain when it does
+// not find the expected native instruction at this offset, which needs updating.
+// Note that this offset is invariant of PreserveFramePointer.
+
+static const int entry_barrier_offset = -4 * 11;
+
+static NativeNMethodBarrier* native_nmethod_barrier(nmethod* nm) {
+  address barrier_address = nm->code_begin() + nm->frame_complete_offset() + entry_barrier_offset;
+  NativeNMethodBarrier* barrier = reinterpret_cast<NativeNMethodBarrier*>(barrier_address);
+  debug_only(barrier->verify());
+  return barrier;
 }
 
 void BarrierSetNMethod::disarm(nmethod* nm) {
-  ShouldNotReachHere();
+  if (!supports_entry_barrier(nm)) {
+    return;
+  }
+
+  // Disarms the nmethod guard emitted by BarrierSetAssembler::nmethod_entry_barrier.
+  // Symmetric "LDR; DMB ISHLD" is in the nmethod barrier.
+  NativeNMethodBarrier* barrier = native_nmethod_barrier(nm);
+
+  barrier->set_value(disarmed_value());
 }
 
 bool BarrierSetNMethod::is_armed(nmethod* nm) {
-  ShouldNotReachHere();
-  return false;
+  if (!supports_entry_barrier(nm)) {
+    return false;
+  }
+
+  NativeNMethodBarrier* barrier = native_nmethod_barrier(nm);
+  return barrier->get_value() != disarmed_value();
 }
diff --git a/src/hotspot/cpu/aarch64/gc/z/zArguments_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/z/zArguments_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/gc/z/zArguments_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/gc/z/zArguments_aarch64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -23,12 +23,7 @@
 
 #include "precompiled.hpp"
 #include "gc/z/zArguments.hpp"
-#include "runtime/globals.hpp"
-#include "runtime/globals_extension.hpp"
-#include "utilities/debug.hpp"
 
 void ZArguments::initialize_platform() {
-  // Disable class unloading - we don't support concurrent class unloading yet.
-  FLAG_SET_DEFAULT(ClassUnloading, false);
-  FLAG_SET_DEFAULT(ClassUnloadingWithConcurrentMark, false);
+    // Does nothing
 }
diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
@@ -3689,6 +3689,11 @@
   bs->obj_equals(this, obj1, obj2);
 }
 
+void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
+  load_method_holder(rresult, rmethod);
+  ldr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
+}
+
 void MacroAssembler::load_method_holder(Register holder, Register method) {
   ldr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
   ldr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
@@ -3710,6 +3715,22 @@
   access_load_at(T_OBJECT, IN_NATIVE, result, Address(result, 0), tmp, noreg);
 }
 
+// ((WeakHandle)result).resolve();
+void MacroAssembler::resolve_weak_handle(Register rresult, Register rtmp) {
+  assert_different_registers(rresult, rtmp);
+  Label resolved;
+
+  // A null weak handle resolves to null.
+  cbz(rresult, resolved);
+
+  // Only 64 bit platforms support GCs that require a tmp register
+  // Only IN_HEAP loads require a thread_tmp register
+  // WeakHandle::resolve is an indirection like jweak.
+  access_load_at(T_OBJECT, IN_NATIVE | ON_PHANTOM_OOP_REF,
+                 rresult, Address(rresult), rtmp, /*tmp_thread*/noreg);
+  bind(resolved);
+}
+
 void MacroAssembler::load_mirror(Register dst, Register method, Register tmp) {
   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
   ldr(dst, Address(rmethod, Method::const_offset()));
@@ -4104,9 +4125,9 @@
 }
 
 // Move an oop into a register.  immediate is true if we want
-// immediate instrcutions, i.e. we are not going to patch this
-// instruction while the code is being executed by another thread.  In
-// that case we can use move immediates rather than the constant pool.
+// immediate instructions and nmethod entry barriers are not enabled.
+// i.e. we are not going to patch this instruction while the code is being
+// executed by another thread.
 void MacroAssembler::movoop(Register dst, jobject obj, bool immediate) {
   int oop_index;
   if (obj == NULL) {
@@ -4121,7 +4142,11 @@
     oop_index = oop_recorder()->find_index(obj);
   }
   RelocationHolder rspec = oop_Relocation::spec(oop_index);
-  if (! immediate) {
+
+  // nmethod entry barrier necessitate using the constant pool. They have to be
+  // ordered with respected to oop accesses.
+  // Using immediate literals would necessitate ISBs.
+  if (BarrierSet::barrier_set()->barrier_set_nmethod() != NULL || !immediate) {
     address dummy = address(uintptr_t(pc()) & -wordSize); // A nearby aligned address
     ldr_constant(dst, Address(dummy, rspec));
   } else
diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, 2019, Red Hat Inc. All rights reserved.
+ * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -176,6 +176,8 @@
 
   using Assembler::ldr;
   using Assembler::str;
+  using Assembler::ldrw;
+  using Assembler::strw;
 
   void ldr(Register Rx, const Address &adr);
   void ldrw(Register Rw, const Address &adr);
@@ -809,6 +811,7 @@
   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
   void c2bool(Register x);
 
+  void load_method_holder_cld(Register rresult, Register rmethod);
   void load_method_holder(Register holder, Register method);
 
   // oop manipulations
@@ -816,6 +819,7 @@
   void store_klass(Register dst, Register src);
   void cmp_klass(Register oop, Register trial_klass, Register tmp);
 
+  void resolve_weak_handle(Register result, Register tmp);
   void resolve_oop_handle(Register result, Register tmp = r5);
   void load_mirror(Register dst, Register method, Register tmp = r5);
 
diff --git a/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp b/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp
@@ -732,6 +732,9 @@
     c2i_no_clinit_check_entry = __ pc();
   }
 
+  BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+  bs->c2i_entry_barrier(masm);
+
   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
 
   __ flush();
@@ -1504,6 +1507,9 @@
   // -2 because return address is already present and so is saved rfp
   __ sub(sp, sp, stack_size - 2*wordSize);
 
+  BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+  bs->nmethod_entry_barrier(masm);
+
   // Frame is now completed as far as size and linkage.
   int frame_complete = ((intptr_t)__ pc()) - start;
 
diff --git a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
@@ -1,6 +1,6 @@
 /*
- * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, 2019, Red Hat Inc. All rights reserved.
+ * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -4155,6 +4155,50 @@
     return entry;
   }
 
+    address generate_method_entry_barrier() {
+    __ align(CodeEntryAlignment);
+    StubCodeMark mark(this, "StubRoutines", "nmethod_entry_barrier");
+
+    Label deoptimize_label;
+
+    address start = __ pc();
+
+    __ set_last_Java_frame(sp, rfp, lr, rscratch1);
+
+    __ enter();
+    __ add(rscratch2, sp, wordSize);  // rscratch2 points to the saved lr
+
+    __ sub(sp, sp, 4 * wordSize);  // four words for the returned {sp, fp, lr, pc}
+
+    __ push_call_clobbered_registers();
+
+    __ mov(c_rarg0, rscratch2);
+    __ call_VM_leaf
+         (CAST_FROM_FN_PTR
+          (address, BarrierSetNMethod::nmethod_stub_entry_barrier), 1);
+
+    __ reset_last_Java_frame(true);
+
+    __ mov(rscratch1, r0);
+
+    __ pop_call_clobbered_registers();
+
+    __ cbnz(rscratch1, deoptimize_label);
+
+    __ leave();
+    __ ret(lr);
+
+    __ BIND(deoptimize_label);
+
+    __ ldp(/* new sp */ rscratch1, rfp, Address(sp, 0 * wordSize));
+    __ ldp(lr, /* new pc*/ rscratch2, Address(sp, 2 * wordSize));
+
+    __ mov(sp, rscratch1);
+    __ br(rscratch2);
+
+    return start;
+  }
+
   // r0  = result
   // r1  = str1
   // r2  = cnt1
@@ -5745,6 +5789,10 @@
     // byte_array_inflate stub for large arrays.
     StubRoutines::aarch64::_large_byte_array_inflate = generate_large_byte_array_inflate();
 
+    BarrierSetNMethod* bs_nm = BarrierSet::barrier_set()->barrier_set_nmethod();
+    if (bs_nm != NULL) {
+      StubRoutines::aarch64::_method_entry_barrier = generate_method_entry_barrier();
+    }
 #ifdef COMPILER2
     if (UseMultiplyToLenIntrinsic) {
       StubRoutines::_multiplyToLen = generate_multiplyToLen();
diff --git a/src/hotspot/cpu/aarch64/stubRoutines_aarch64.cpp b/src/hotspot/cpu/aarch64/stubRoutines_aarch64.cpp
--- a/src/hotspot/cpu/aarch64/stubRoutines_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/stubRoutines_aarch64.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -56,6 +56,7 @@
 address StubRoutines::aarch64::_string_indexof_linear_uu = NULL;
 address StubRoutines::aarch64::_string_indexof_linear_ul = NULL;
 address StubRoutines::aarch64::_large_byte_array_inflate = NULL;
+address StubRoutines::aarch64::_method_entry_barrier = NULL;
 bool StubRoutines::aarch64::_completed = false;
 
 /**
diff --git a/src/hotspot/cpu/aarch64/stubRoutines_aarch64.hpp b/src/hotspot/cpu/aarch64/stubRoutines_aarch64.hpp
--- a/src/hotspot/cpu/aarch64/stubRoutines_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/stubRoutines_aarch64.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2020, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -69,6 +69,9 @@
   static address _string_indexof_linear_uu;
   static address _string_indexof_linear_ul;
   static address _large_byte_array_inflate;
+
+  static address _method_entry_barrier;
+
   static bool _completed;
 
  public:
@@ -171,6 +174,10 @@
       return _large_byte_array_inflate;
   }
 
+  static address method_entry_barrier() {
+    return _method_entry_barrier;
+  }
+
   static bool complete() {
     return _completed;
   }
diff --git a/src/hotspot/share/gc/shared/barrierSetNMethod.cpp b/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
--- a/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
+++ b/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2020, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -62,6 +62,16 @@
   assert(!nm->is_osr_method(), "Should not reach here");
   // Called upon first entry after being armed
   bool may_enter = bs_nm->nmethod_entry_barrier(nm);
+
+  // Diagnostic option to force deoptimization 1 in 3 times. It is otherwise
+  // a very rare event.
+  if (DeoptimizeNMethodBarriersALot) {
+    static volatile uint32_t counter=0;
+    if (Atomic::add(&counter, 1u) % 3 == 0) {
+      may_enter = false;
+    }
+  }
+
   if (!may_enter) {
     log_trace(nmethod, barrier)("Deoptimizing nmethod: " PTR_FORMAT, p2i(nm));
     bs_nm->deoptimize(nm, return_address_ptr);
diff --git a/src/hotspot/share/runtime/globals.hpp b/src/hotspot/share/runtime/globals.hpp
--- a/src/hotspot/share/runtime/globals.hpp
+++ b/src/hotspot/share/runtime/globals.hpp
@@ -2489,7 +2489,8 @@
   product(bool, UseEmptySlotsInSupers, true,                                \
                 "Allow allocating fields in empty slots of super-classes")  \
                                                                             \
-
+  diagnostic(bool, DeoptimizeNMethodBarriersALot, false,                    \
+                "Make nmethod barriers deoptimise a lot.")                  \
 
 // Interface macros
 #define DECLARE_PRODUCT_FLAG(type, name, value, doc)      extern "C" type name;
diff --git a/src/hotspot/share/runtime/sharedRuntime.cpp b/src/hotspot/share/runtime/sharedRuntime.cpp
--- a/src/hotspot/share/runtime/sharedRuntime.cpp
+++ b/src/hotspot/share/runtime/sharedRuntime.cpp
@@ -2873,6 +2873,12 @@
       CodeBuffer buffer(buf);
       double locs_buf[20];
       buffer.insts()->initialize_shared_locs((relocInfo*)locs_buf, sizeof(locs_buf) / sizeof(relocInfo));
+#if defined(AARCH64)
+      // On AArch64 with ZGC and nmethod entry barriers, we need all oops to be
+      // in the constant pool to ensure ordering between the barrier and oops
+      // accesses. For native_wrappers we need a constant.
+      buffer.initialize_consts_size(8);
+#endif
       MacroAssembler _masm(&buffer);
 
       // Fill in the signature array, for the calling-convention call.
diff --git a/test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithZ.java b/test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithZ.java
--- a/test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithZ.java
+++ b/test/hotspot/jtreg/gc/stress/gcbasher/TestGCBasherWithZ.java
@@ -35,6 +35,18 @@
  * @summary Stress ZGC
  * @run main/othervm/timeout=200 -Xlog:gc*=info -Xmx384m -server -XX:+UseZGC gc.stress.gcbasher.TestGCBasherWithZ 120000
  */
+
+/*
+ * @test TestGCBasherDeoptWithZ
+ * @key gc stress
+ * @library /
+ * @requires vm.gc.Z
+ * @requires vm.flavor == "server" & !vm.emulatedClient & !vm.graal.enabled & vm.opt.ClassUnloading != false
+ * @summary Stress ZGC with nmethod barrier forced deoptimization enabled.
+ * @run main/othervm/timeout=200 -Xlog:gc*=info,nmethod+barrier=trace -Xmx384m -server -XX:+UnlockExperimentalVMOptions -XX:+UseZGC
+ *  -XX:+UnlockDiagnosticVMOptions -XX:+DeoptimizeNMethodBarriersALot -XX:-Inline
+ *   gc.stress.gcbasher.TestGCBasherWithZ 120000
+ */
 public class TestGCBasherWithZ {
     public static void main(String[] args) throws IOException {
         TestGCBasher.main(args);