deps: backport rehash strings after deserialization

Original commit messages:
v8/v8@a2ab135
  [snapshot] Rehash strings after deserialization.

  See https://goo.gl/6aN8xA

  Bug: v8:6593
  Change-Id: Ic8b0b57195d01d41591397d5d45de3f0f3ebc3d9
  Reviewed-on: https://chromium-review.googlesource.com/574527
  Reviewed-by: Camillo Bruni <[email protected]>
  Reviewed-by: Jakob Gruber <[email protected]>
  Reviewed-by: Ulan Degenbaev <[email protected]>
  Commit-Queue: Yang Guo <[email protected]>
  Cr-Commit-Position: refs/heads/master@{#46732}

v8/v8@182caaf
  Do not track transitions for built-in objects.

  Objects created during bootstrapping do not need
  a transition tree except for elements kind transitions.

  Bug: v8:6596
  Change-Id: I237b8b2792f201336e1c9731c815095dd06bc182
  Reviewed-on: https://chromium-review.googlesource.com/571750
  Reviewed-by: Igor Sheludko <[email protected]>
  Commit-Queue: Yang Guo <[email protected]>
  Cr-Commit-Position: refs/heads/master@{#46693}

Fixes: #14171
PR-URL: #14385

Author: hashseed

configure

@@ -419,12 +419,12 @@ parser.add_option('--without-perfctr',
 # Dummy option for backwards compatibility
 parser.add_option('--with-snapshot',
     action='store_true',
-    dest='with_snapshot',
+    dest='unused_with_snapshot',
     help=optparse.SUPPRESS_HELP)
 
 parser.add_option('--without-snapshot',
     action='store_true',
-    dest='unused_without_snapshot',
+    dest='without_snapshot',
     help=optparse.SUPPRESS_HELP)
 
 parser.add_option('--without-ssl',
@@ -802,7 +802,7 @@ def configure_node(o):
   cross_compiling = (options.cross_compiling
                      if options.cross_compiling is not None
                      else target_arch != host_arch)
-  want_snapshots = 1 if options.with_snapshot else 0
+  want_snapshots = not options.without_snapshot
   o['variables']['want_separate_host_toolset'] = int(
       cross_compiling and want_snapshots)
 
@@ -946,7 +946,7 @@ def configure_v8(o):
   o['variables']['v8_no_strict_aliasing'] = 1  # Work around compiler bugs.
   o['variables']['v8_optimized_debug'] = 0  # Compile with -O0 in debug builds.
   o['variables']['v8_random_seed'] = 0  # Use a random seed for hash tables.
-  o['variables']['v8_use_snapshot'] = b(options.with_snapshot)
+  o['variables']['v8_use_snapshot'] = 'false' if options.without_snapshot else 'true'
   o['variables']['node_use_v8_platform'] = b(not options.without_v8_platform)
   o['variables']['node_use_bundled_v8'] = b(not options.without_bundled_v8)
   o['variables']['force_dynamic_crt'] = 1 if options.shared else 0

deps/v8/include/v8-version.h

@@ -11,7 +11,7 @@
 #define V8_MAJOR_VERSION 5
 #define V8_MINOR_VERSION 1
 #define V8_BUILD_NUMBER 281
-#define V8_PATCH_LEVEL 103
+#define V8_PATCH_LEVEL 104
 
 // Use 1 for candidates and 0 otherwise.
 // (Boolean macro values are not supported by all preprocessors.)

deps/v8/src/api.cc

@@ -391,6 +391,9 @@ StartupData SerializeIsolateAndContext(
 
   i::Isolate* internal_isolate = reinterpret_cast<i::Isolate*>(isolate);
 
+  // We might rehash strings and re-sort descriptors. Clear the lookup cache.
+  internal_isolate->descriptor_lookup_cache()->Clear();
+
   // If we don't do this then we end up with a stray root pointing at the
   // context even after we have disposed of the context.
   internal_isolate->heap()->CollectAllAvailableGarbage("mksnapshot");
@@ -428,6 +431,9 @@ StartupData SerializeIsolateAndContext(
   context_ser.Serialize(&raw_context);
   ser.SerializeWeakReferencesAndDeferred();
 
+  metadata.set_can_rehash(ser.can_be_rehashed() &&
+                          context_ser.can_be_rehashed());
+
   return i::Snapshot::CreateSnapshotBlob(ser, context_ser, metadata);
 }
 

deps/v8/src/bootstrapper.cc

@@ -655,6 +655,8 @@ Handle<JSFunction> Genesis::GetThrowTypeErrorIntrinsic(
     DCHECK(false);
   }
 
+  JSObject::MigrateSlowToFast(function, 0, "Bootstrapping");
+
   return function;
 }
 
@@ -1133,6 +1135,8 @@ void Genesis::InitializeGlobal(Handle<JSGlobalObject> global_object,
 
     sloppy_function_map_writable_prototype_->SetConstructor(*function_fun);
     strict_function_map_writable_prototype_->SetConstructor(*function_fun);
+
+    JSObject::MigrateSlowToFast(function_fun, 0, "Bootstrapping");
   }
 
   {  // --- A r r a y ---

deps/v8/src/flag-definitions.h

@@ -830,6 +830,8 @@ DEFINE_BOOL(abort_on_uncaught_exception, false,
 DEFINE_BOOL(randomize_hashes, true,
             "randomize hashes to avoid predictable hash collisions "
             "(with snapshots this option cannot override the baked-in seed)")
+DEFINE_BOOL(rehash_snapshot, true,
+            "rehash strings from the snapshot to override the baked-in seed")
 DEFINE_INT(hash_seed, 0,
            "Fixed seed to use to hash property keys (0 means random)"
            "(with snapshots this option cannot override the baked-in seed)")

deps/v8/src/heap/heap.cc

@@ -5348,14 +5348,7 @@ bool Heap::SetUp() {
 
   // Set up the seed that is used to randomize the string hash function.
   DCHECK(hash_seed() == 0);
-  if (FLAG_randomize_hashes) {
-    if (FLAG_hash_seed == 0) {
-      int rnd = isolate()->random_number_generator()->NextInt();
-      set_hash_seed(Smi::FromInt(rnd & Name::kHashBitMask));
-    } else {
-      set_hash_seed(Smi::FromInt(FLAG_hash_seed));
-    }
-  }
+  if (FLAG_randomize_hashes) InitializeHashSeed();
 
   for (int i = 0; i < static_cast<int>(v8::Isolate::kUseCounterFeatureCount);
        i++) {
@@ -5393,6 +5386,14 @@ bool Heap::SetUp() {
   return true;
 }
 
+void Heap::InitializeHashSeed() {
+  if (FLAG_hash_seed == 0) {
+    int rnd = isolate()->random_number_generator()->NextInt();
+    set_hash_seed(Smi::FromInt(rnd & Name::kHashBitMask));
+  } else {
+    set_hash_seed(Smi::FromInt(FLAG_hash_seed));
+  }
+}
 
 bool Heap::CreateHeapObjects() {
   // Create initial maps.

deps/v8/src/heap/heap.h

@@ -864,6 +864,9 @@ class Heap {
   // without actually creating any objects.
   bool SetUp();
 
+  // (Re-)Initialize hash seed from flag or RNG.
+  void InitializeHashSeed();
+
   // Bootstraps the object heap with the core set of objects required to run.
   // Returns whether it succeeded.
   bool CreateHeapObjects();

deps/v8/src/js/array.js

@@ -1831,6 +1831,8 @@ var unscopables = {
   keys: true,
 };
 
+%ToFastProperties(unscopables);
+
 %AddNamedProperty(GlobalArray.prototype, unscopablesSymbol, unscopables,
                   DONT_ENUM | READ_ONLY);
 

deps/v8/src/objects.cc

@@ -9469,6 +9469,12 @@ void Map::TraceAllTransitions(Map* map) {
 
 void Map::ConnectTransition(Handle<Map> parent, Handle<Map> child,
                             Handle<Name> name, SimpleTransitionFlag flag) {
+  // Do not track transitions during bootstrap except for element transitions.
+  Isolate* isolate = parent->GetIsolate();
+  if (isolate->bootstrapper()->IsActive() &&
+      !name.is_identical_to(isolate->factory()->elements_transition_symbol())) {
+    return;
+  }
   if (!parent->GetBackPointer()->IsUndefined()) {
     parent->set_owns_descriptors(false);
   } else {
@@ -17520,6 +17526,12 @@ template class Dictionary<UnseededNumberDictionary,
                           UnseededNumberDictionaryShape,
                           uint32_t>;
 
+template void
+HashTable<GlobalDictionary, GlobalDictionaryShape, Handle<Name> >::Rehash(Handle<Name> key);
+
+template void
+HashTable<NameDictionary, NameDictionaryShape, Handle<Name> >::Rehash(Handle<Name> key);
+
 template Handle<SeededNumberDictionary>
 Dictionary<SeededNumberDictionary, SeededNumberDictionaryShape, uint32_t>::
     New(Isolate*, int at_least_space_for, PretenureFlag pretenure);

deps/v8/src/snapshot/deserializer.cc

@@ -109,6 +109,8 @@ void Deserializer::Deserialize(Isolate* isolate) {
   LOG_CODE_EVENT(isolate_, LogCodeObjects());
   LOG_CODE_EVENT(isolate_, LogBytecodeHandlers());
   LOG_CODE_EVENT(isolate_, LogCompiledFunctions());
+
+  if (FLAG_rehash_snapshot && can_rehash_) Rehash();
 }
 
 MaybeHandle<Object> Deserializer::DeserializePartial(
@@ -138,6 +140,9 @@ MaybeHandle<Object> Deserializer::DeserializePartial(
   // changed and logging should be added to notify the profiler et al of the
   // new code, which also has to be flushed from instruction cache.
   CHECK_EQ(start_address, code_space->top());
+
+  if (FLAG_rehash_snapshot && can_rehash_) RehashContext(Context::cast(root));
+
   return Handle<Object>(root, isolate);
 }
 
@@ -164,6 +169,64 @@ MaybeHandle<SharedFunctionInfo> Deserializer::DeserializeCode(
   }
 }
 
+// We only really just need HashForObject here.
+class StringRehashKey : public HashTableKey {
+ public:
+  uint32_t HashForObject(Object* other) override {
+    return String::cast(other)->Hash();
+  }
+
+  static uint32_t StringHash(Object* obj) {
+    UNREACHABLE();
+    return String::cast(obj)->Hash();
+  }
+
+  bool IsMatch(Object* string) override {
+    UNREACHABLE();
+    return false;
+  }
+
+  uint32_t Hash() override {
+    UNREACHABLE();
+    return 0;
+  }
+
+  Handle<Object> AsHandle(Isolate* isolate) override {
+    UNREACHABLE();
+    return isolate->factory()->empty_string();
+  }
+};
+
+void Deserializer::Rehash() {
+  DCHECK(can_rehash_);
+  isolate_->heap()->InitializeHashSeed();
+  if (FLAG_profile_deserialization) {
+    PrintF("Re-initializing hash seed to %x\n",
+           isolate_->heap()->hash_seed()->value());
+  }
+  StringRehashKey string_rehash_key;
+  isolate_->heap()->string_table()->Rehash(&string_rehash_key);
+  isolate_->heap()->intrinsic_function_names()->Rehash(
+      isolate_->factory()->empty_string());
+  SortMapDescriptors();
+}
+
+void Deserializer::RehashContext(Context* context) {
+  DCHECK(can_rehash_);
+  for (const auto& array : transition_arrays_) array->Sort();
+  Handle<Name> dummy = isolate_->factory()->empty_string();
+  context->global_object()->global_dictionary()->Rehash(dummy);
+  SortMapDescriptors();
+}
+
+void Deserializer::SortMapDescriptors() {
+  for (const auto& map : maps_) {
+    if (map->instance_descriptors()->number_of_descriptors() > 1) {
+      map->instance_descriptors()->Sort();
+    }
+  }
+}
+
 Deserializer::~Deserializer() {
   // TODO(svenpanne) Re-enable this assertion when v8 initialization is fixed.
   // DCHECK(source_.AtEOF());
@@ -288,6 +351,18 @@ HeapObject* Deserializer::PostProcessNewObject(HeapObject* obj, int space) {
       new_code_objects_.Add(Code::cast(obj));
     }
   }
+  if (FLAG_rehash_snapshot && can_rehash_ && !deserializing_user_code()) {
+    if (obj->IsString()) {
+      // Uninitialize hash field as we are going to reinitialize the hash seed.
+      String* string = String::cast(obj);
+      string->set_hash_field(String::kEmptyHashField);
+    } else if (obj->IsTransitionArray() &&
+               TransitionArray::cast(obj)->number_of_entries() > 1) {
+      transition_arrays_.Add(TransitionArray::cast(obj));
+    } else if (obj->IsMap()) {
+      maps_.Add(Map::cast(obj));
+    }
+  }
   // Check alignment.
   DCHECK_EQ(0, Heap::GetFillToAlign(obj->address(), obj->RequiredAlignment()));
   return obj;

deps/v8/src/snapshot/deserializer.h

@@ -37,7 +37,8 @@ class Deserializer : public SerializerDeserializer {
         external_reference_table_(NULL),
         deserialized_large_objects_(0),
         deserializing_user_code_(false),
-        next_alignment_(kWordAligned) {
+        next_alignment_(kWordAligned),
+        can_rehash_(false) {
     DecodeReservation(data->Reservations());
   }
 
@@ -59,6 +60,8 @@ class Deserializer : public SerializerDeserializer {
     attached_objects_ = attached_objects;
   }
 
+  void SetRehashability(bool v) { can_rehash_ = v; }
+
  private:
   void VisitPointers(Object** start, Object** end) override;
 
@@ -113,6 +116,15 @@ class Deserializer : public SerializerDeserializer {
   Object** CopyInNativesSource(Vector<const char> source_vector,
                                Object** current);
 
+  // Rehash after deserializing an isolate.
+  void Rehash();
+
+  // Rehash after deserializing a context.
+  void RehashContext(Context* context);
+
+  // Sort descriptors of deserialized maps using new string hashes.
+  void SortMapDescriptors();
+
   // Cached current isolate.
   Isolate* isolate_;
 
@@ -136,11 +148,16 @@ class Deserializer : public SerializerDeserializer {
   List<Code*> new_code_objects_;
   List<Handle<String> > new_internalized_strings_;
   List<Handle<Script> > new_scripts_;
+  List<Map*> maps_;
+  List<TransitionArray*> transition_arrays_;
 
   bool deserializing_user_code_;
 
   AllocationAlignment next_alignment_;
 
+  // TODO(6593): generalize rehashing, and remove this flag.
+  bool can_rehash_;
+
   DISALLOW_COPY_AND_ASSIGN(Deserializer);
 };
 

deps/v8/src/snapshot/partial-serializer.cc

@@ -15,7 +15,9 @@ PartialSerializer::PartialSerializer(Isolate* isolate,
     : Serializer(isolate, sink),
       startup_serializer_(startup_snapshot_serializer),
       global_object_(NULL),
-      next_partial_cache_index_(0) {
+      next_partial_cache_index_(0),
+      rehashable_context_(nullptr),
+      can_be_rehashed_(true) {
   InitializeCodeAddressMap();
 }
 
@@ -24,7 +26,7 @@ PartialSerializer::~PartialSerializer() {
 }
 
 void PartialSerializer::Serialize(Object** o) {
-  if ((*o)->IsContext()) {
+  if ((*o)->IsNativeContext()) {
     Context* context = Context::cast(*o);
     global_object_ = context->global_object();
     back_reference_map()->AddGlobalProxy(context->global_proxy());
@@ -33,11 +35,14 @@ void PartialSerializer::Serialize(Object** o) {
     // and it's next context pointer may point to the code-stub context.  Clear
     // it before serializing, it will get re-added to the context list
     // explicitly when it's loaded.
-    if (context->IsNativeContext()) {
-      context->set(Context::NEXT_CONTEXT_LINK,
-                   isolate_->heap()->undefined_value());
-      DCHECK(!context->global_object()->IsUndefined());
-    }
+    context->set(Context::NEXT_CONTEXT_LINK,
+                 isolate_->heap()->undefined_value());
+    DCHECK(!context->global_object()->IsUndefined());
+    DCHECK_NULL(rehashable_context_);
+    rehashable_context_ = context;
+  } else {
+    // We only do rehashing for native contexts.
+    can_be_rehashed_ = false;
   }
   VisitPointer(o);
   SerializeDeferredObjects();
@@ -89,6 +94,8 @@ void PartialSerializer::SerializeObject(HeapObject* obj, HowToCode how_to_code,
     for (int i = 0; i < literals->length(); i++) literals->set_undefined(i);
   }
 
+  if (obj->IsHashTable()) CheckRehashability(obj);
+
   // Object has not yet been serialized.  Serialize it here.
   ObjectSerializer serializer(this, obj, sink_, how_to_code, where_to_point);
   serializer.Serialize();
@@ -119,5 +126,17 @@ bool PartialSerializer::ShouldBeInThePartialSnapshotCache(HeapObject* o) {
              startup_serializer_->isolate()->heap()->fixed_cow_array_map();
 }
 
+void PartialSerializer::CheckRehashability(HeapObject* table) {
+  DCHECK(table->IsHashTable());
+  if (!can_be_rehashed_) return;
+  // We can only correctly rehash if the global dictionary is the only hash
+  // table that we deserialize.
+  if (table == rehashable_context_->global_object()->global_dictionary()) {
+    return;
+  }
+  if (table == rehashable_context_->template_instantiations_cache()) return;
+  can_be_rehashed_ = false;
+}
+
 }  // namespace internal
 }  // namespace v8