Make code_data_container and kind_specific_flags accessors atomic.

The concurrent marker needs to access the kind_specific_flags to decide
whether an embedded object reference is weak or not.

This patch turns the Code::code_data_container() into an acquire/release
atomic accessor and makes CodeDataContainer::kind_specific_flags a
relaxed atomic accessor.

Bug: v8:8459
Change-Id: I5251fed4e7b3315f8e229dfcfe2c23f611f4b333
Reviewed-on: https://chromium-review.googlesource.com/c/1337746Reviewed-by: Hannes Payer <hpayer@chromium.org>
Reviewed-by: Michael Lippautz <mlippautz@chromium.org>
Commit-Queue: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#57638}

src/objects/code-inl.h

@@ -187,10 +187,15 @@ INT_ACCESSORS(Code, raw_instruction_size, kInstructionSizeOffset)
 INT_ACCESSORS(Code, handler_table_offset, kHandlerTableOffsetOffset)
 #define CODE_ACCESSORS(name, type, offset) \
   ACCESSORS_CHECKED2(Code, name, type, offset, true, !Heap::InNewSpace(value))
+#define SYNCHRONIZED_CODE_ACCESSORS(name, type, offset)           \
+  SYNCHRONIZED_ACCESSORS_CHECKED2(Code, name, type, offset, true, \
+                                  !Heap::InNewSpace(value))
 CODE_ACCESSORS(relocation_info, ByteArray, kRelocationInfoOffset)
 CODE_ACCESSORS(deoptimization_data, FixedArray, kDeoptimizationDataOffset)
 CODE_ACCESSORS(source_position_table, Object, kSourcePositionTableOffset)
-CODE_ACCESSORS(code_data_container, CodeDataContainer, kCodeDataContainerOffset)
+// Concurrent marker needs to access kind specific flags in code data container.
+SYNCHRONIZED_CODE_ACCESSORS(code_data_container, CodeDataContainer,
+                            kCodeDataContainerOffset)
 #undef CODE_ACCESSORS
 
 void Code::WipeOutHeader() {
@@ -404,53 +409,53 @@ inline bool Code::is_turbofanned() const {
 
 inline bool Code::can_have_weak_objects() const {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return CanHaveWeakObjectsField::decode(flags);
 }
 
 inline void Code::set_can_have_weak_objects(bool value) {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = CanHaveWeakObjectsField::update(previous, value);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = CanHaveWeakObjectsField::update(previous, value);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
 inline bool Code::is_construct_stub() const {
   DCHECK(kind() == BUILTIN);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return IsConstructStubField::decode(flags);
 }
 
 inline void Code::set_is_construct_stub(bool value) {
   DCHECK(kind() == BUILTIN);
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = IsConstructStubField::update(previous, value);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = IsConstructStubField::update(previous, value);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
 inline bool Code::is_promise_rejection() const {
   DCHECK(kind() == BUILTIN);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return IsPromiseRejectionField::decode(flags);
 }
 
 inline void Code::set_is_promise_rejection(bool value) {
   DCHECK(kind() == BUILTIN);
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = IsPromiseRejectionField::update(previous, value);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = IsPromiseRejectionField::update(previous, value);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
 inline bool Code::is_exception_caught() const {
   DCHECK(kind() == BUILTIN);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return IsExceptionCaughtField::decode(flags);
 }
 
 inline void Code::set_is_exception_caught(bool value) {
   DCHECK(kind() == BUILTIN);
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = IsExceptionCaughtField::update(previous, value);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = IsExceptionCaughtField::update(previous, value);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
@@ -500,43 +505,43 @@ void Code::set_safepoint_table_offset(int offset) {
 
 bool Code::marked_for_deoptimization() const {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return MarkedForDeoptimizationField::decode(flags);
 }
 
 void Code::set_marked_for_deoptimization(bool flag) {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
   DCHECK_IMPLIES(flag, AllowDeoptimization::IsAllowed(GetIsolate()));
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = MarkedForDeoptimizationField::update(previous, flag);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = MarkedForDeoptimizationField::update(previous, flag);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
 bool Code::embedded_objects_cleared() const {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return EmbeddedObjectsClearedField::decode(flags);
 }
 
 void Code::set_embedded_objects_cleared(bool flag) {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
   DCHECK_IMPLIES(flag, marked_for_deoptimization());
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = EmbeddedObjectsClearedField::update(previous, flag);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = EmbeddedObjectsClearedField::update(previous, flag);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
 bool Code::deopt_already_counted() const {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
-  int flags = code_data_container()->kind_specific_flags();
+  int32_t flags = code_data_container()->kind_specific_flags();
   return DeoptAlreadyCountedField::decode(flags);
 }
 
 void Code::set_deopt_already_counted(bool flag) {
   DCHECK(kind() == OPTIMIZED_FUNCTION);
   DCHECK_IMPLIES(flag, AllowDeoptimization::IsAllowed(GetIsolate()));
-  int previous = code_data_container()->kind_specific_flags();
-  int updated = DeoptAlreadyCountedField::update(previous, flag);
+  int32_t previous = code_data_container()->kind_specific_flags();
+  int32_t updated = DeoptAlreadyCountedField::update(previous, flag);
   code_data_container()->set_kind_specific_flags(updated);
 }
 
@@ -606,7 +611,10 @@ bool Code::IsWeakObjectInOptimizedCode(HeapObject* object) {
          InstanceTypeChecker::IsContext(instance_type);
 }
 
-INT_ACCESSORS(CodeDataContainer, kind_specific_flags, kKindSpecificFlagsOffset)
+// This field has to have relaxed atomic accessors because it is accessed in the
+// concurrent marker.
+RELAXED_INT32_ACCESSORS(CodeDataContainer, kind_specific_flags,
+                        kKindSpecificFlagsOffset)
 ACCESSORS(CodeDataContainer, next_code_link, Object, kNextCodeLinkOffset)
 
 void CodeDataContainer::clear_padding() {

src/objects/object-macros.h

@@ -118,6 +118,14 @@
     WRITE_INT32_FIELD(this, offset, value);                               \
   }
 
+#define RELAXED_INT32_ACCESSORS(holder, name, offset) \
+  int32_t holder::name() const {                      \
+    return RELAXED_READ_INT32_FIELD(this, offset);    \
+  }                                                   \
+  void holder::set_##name(int32_t value) {            \
+    RELAXED_WRITE_INT32_FIELD(this, offset, value);   \
+  }
+
 #define UINT16_ACCESSORS(holder, name, offset)                              \
   uint16_t holder::name() const { return READ_UINT16_FIELD(this, offset); } \
   void holder::set_##name(int value) {                                      \
@@ -425,9 +433,18 @@
 #define READ_INT32_FIELD(p, offset) \
   (*reinterpret_cast<const int32_t*>(FIELD_ADDR(p, offset)))
 
+#define RELAXED_READ_INT32_FIELD(p, offset) \
+  static_cast<int32_t>(base::Relaxed_Load(  \
+      reinterpret_cast<const base::Atomic32*>(FIELD_ADDR(p, offset))))
+
 #define WRITE_INT32_FIELD(p, offset, value) \
   (*reinterpret_cast<int32_t*>(FIELD_ADDR(p, offset)) = value)
 
+#define RELAXED_WRITE_INT32_FIELD(p, offset, value)             \
+  base::Relaxed_Store(                                          \
+      reinterpret_cast<base::Atomic32*>(FIELD_ADDR(p, offset)), \
+      static_cast<base::Atomic32>(value));
+
 #define READ_FLOAT_FIELD(p, offset) \
   (*reinterpret_cast<const float*>(FIELD_ADDR(p, offset)))