Support large objects in the serializer/deserializer.

src/serialize.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/v8.h"
 
 #include "src/accessors.h"
 #include "src/api.h"
 #include "src/base/platform/platform.h"
 #include "src/bootstrapper.h"
@@ skipping 578 matching lines @@
 
   NameMap address_to_name_map_;
   Isolate* isolate_;
 };
 
 
 Deserializer::Deserializer(SnapshotByteSource* source)
     : isolate_(NULL),
       attached_objects_(NULL),
       source_(source),
-      external_reference_decoder_(NULL) {
-  for (int i = 0; i < LAST_SPACE + 1; i++) {
+      external_reference_decoder_(NULL),
+      deserialized_large_objects_(0) {
+  for (int i = 0; i < kNumberOfSpaces; i++) {
     reservations_[i] = kUninitializedReservation;
   }
 }
 
 
 void Deserializer::FlushICacheForNewCodeObjects() {
   PageIterator it(isolate_->heap()->code_space());
   while (it.has_next()) {
     Page* p = it.next();
     CpuFeatures::FlushICache(p->area_start(), p->area_end() - p->area_start());
   }
 }
 
 
 void Deserializer::Deserialize(Isolate* isolate) {
   isolate_ = isolate;
   DCHECK(isolate_ != NULL);
-  isolate_->heap()->ReserveSpace(reservations_, &high_water_[0]);
+  isolate_->heap()->ReserveSpace(reservations_, high_water_);
   // No active threads.
   DCHECK_EQ(NULL, isolate_->thread_manager()->FirstThreadStateInUse());
   // No active handles.
   DCHECK(isolate_->handle_scope_implementer()->blocks()->is_empty());
   DCHECK_EQ(NULL, external_reference_decoder_);
   external_reference_decoder_ = new ExternalReferenceDecoder(isolate);
   isolate_->heap()->IterateSmiRoots(this);
   isolate_->heap()->IterateStrongRoots(this, VISIT_ONLY_STRONG);
   isolate_->heap()->RepairFreeListsAfterBoot();
   isolate_->heap()->IterateWeakRoots(this, VISIT_ALL);
@@ skipping 26 matching lines @@
   LOG_CODE_EVENT(isolate_, LogCodeObjects());
   LOG_CODE_EVENT(isolate_, LogCompiledFunctions());
 }
 
 
 void Deserializer::DeserializePartial(Isolate* isolate, Object** root) {
   isolate_ = isolate;
   for (int i = NEW_SPACE; i < kNumberOfSpaces; i++) {
     DCHECK(reservations_[i] != kUninitializedReservation);
   }
-  isolate_->heap()->ReserveSpace(reservations_, &high_water_[0]);
+  Heap* heap = isolate->heap();
+  heap->ReserveSpace(reservations_, high_water_);
   if (external_reference_decoder_ == NULL) {
     external_reference_decoder_ = new ExternalReferenceDecoder(isolate);
   }
 
   DisallowHeapAllocation no_gc;
 
   // Keep track of the code space start and end pointers in case new
   // code objects were unserialized
   OldSpace* code_space = isolate_->heap()->code_space();
   Address start_address = code_space->top();
@@ skipping 115 matching lines @@
   // TODO(mvstanton): consider treating the heap()->allocation_sites_list()
   // as a (weak) root. If this root is relocated correctly,
   // RelinkAllocationSite() isn't necessary.
   if (obj->IsAllocationSite()) RelinkAllocationSite(AllocationSite::cast(obj));
 
   // Fix up strings from serialized user code.
   if (deserializing_user_code()) obj = ProcessNewObjectFromSerializedCode(obj);
 
   *write_back = obj;
 #ifdef DEBUG
-  bool is_codespace = (space_number == CODE_SPACE);
-  DCHECK(obj->IsCode() == is_codespace);
+  if (obj->IsCode()) {
+    DCHECK(space_number == CODE_SPACE || space_number == LO_SPACE);
+  } else {
+    DCHECK(space_number != CODE_SPACE);
+  }
 #endif
 }
 
+
+// We know the space requirements before deserialization and can
+// pre-allocate that reserved space. During deserialization, all we need
+// to do is to bump up the pointer for each space in the reserved
+// space. This is also used for fixing back references.
+// Since multiple large objects cannot be folded into one large object
+// space allocation, we have to do an actual allocation when deserializing
+// each large object. Instead of tracking offset for back references, we
+// reference large objects by index.
+Address Deserializer::Allocate(int space_index, int size) {
+  if (space_index == LO_SPACE) {
+    AlwaysAllocateScope scope(isolate_);
+    LargeObjectSpace* lo_space = isolate_->heap()->lo_space();
+    Executability exec = static_cast<Executability>(source_->GetInt());
+    AllocationResult result = lo_space->AllocateRaw(size, exec);
+    HeapObject* obj = HeapObject::cast(result.ToObjectChecked());
+    deserialized_large_objects_.Add(obj);
+    return obj->address();
+  } else {
+    DCHECK(space_index < kNumberOfPreallocatedSpaces);
+    Address address = high_water_[space_index];
+    high_water_[space_index] = address + size;
+    return address;
+  }
+}
+
 void Deserializer::ReadChunk(Object** current,
                              Object** limit,
                              int source_space,
                              Address current_object_address) {
   Isolate* const isolate = isolate_;
   // Write barrier support costs around 1% in startup time.  In fact there
   // are no new space objects in current boot snapshots, so it's not needed,
   // but that may change.
   bool write_barrier_needed = (current_object_address != NULL &&
                                source_space != NEW_SPACE &&
@@ skipping 102 matching lines @@
     }                                                                          \
     if (!current_was_incremented) {                                            \
       current++;                                                               \
     }                                                                          \
     break;                                                                     \
   }
 
 // This generates a case and a body for the new space (which has to do extra
 // write barrier handling) and handles the other spaces with 8 fall-through
 // cases and one body.
-#define ALL_SPACES(where, how, within)                                         \
-  CASE_STATEMENT(where, how, within, NEW_SPACE)                                \
-  CASE_BODY(where, how, within, NEW_SPACE)                                     \
-  CASE_STATEMENT(where, how, within, OLD_DATA_SPACE)                           \
-  CASE_STATEMENT(where, how, within, OLD_POINTER_SPACE)                        \
-  CASE_STATEMENT(where, how, within, CODE_SPACE)                               \
-  CASE_STATEMENT(where, how, within, CELL_SPACE)                               \
-  CASE_STATEMENT(where, how, within, PROPERTY_CELL_SPACE)                      \
-  CASE_STATEMENT(where, how, within, MAP_SPACE)                                \
+#define ALL_SPACES(where, how, within)                    \
+  CASE_STATEMENT(where, how, within, NEW_SPACE)           \
+  CASE_BODY(where, how, within, NEW_SPACE)                \
+  CASE_STATEMENT(where, how, within, OLD_DATA_SPACE)      \
+  CASE_STATEMENT(where, how, within, OLD_POINTER_SPACE)   \
+  CASE_STATEMENT(where, how, within, CODE_SPACE)          \
+  CASE_STATEMENT(where, how, within, MAP_SPACE)           \
+  CASE_STATEMENT(where, how, within, CELL_SPACE)          \
+  CASE_STATEMENT(where, how, within, PROPERTY_CELL_SPACE) \
+  CASE_STATEMENT(where, how, within, LO_SPACE)            \
   CASE_BODY(where, how, within, kAnyOldSpace)
 
 #define FOUR_CASES(byte_code)             \
   case byte_code:                         \
   case byte_code + 1:                     \
   case byte_code + 2:                     \
   case byte_code + 3:
 
 #define SIXTEEN_CASES(byte_code)          \
   FOUR_CASES(byte_code)                   \
@@ skipping 230 matching lines @@
   }
   DCHECK_EQ(limit, current);
 }
 
 
 Serializer::Serializer(Isolate* isolate, SnapshotByteSink* sink)
     : isolate_(isolate),
       sink_(sink),
       external_reference_encoder_(new ExternalReferenceEncoder(isolate)),
       root_index_wave_front_(0),
-      code_address_map_(NULL) {
+      code_address_map_(NULL),
+      seen_large_objects_index_(0) {
   // The serializer is meant to be used only to generate initial heap images
   // from a context in which there is only one isolate.
-  for (int i = 0; i <= LAST_SPACE; i++) {
-    fullness_[i] = 0;
-  }
+  for (int i = 0; i < kNumberOfSpaces; i++) fullness_[i] = 0;
 }
 
 
 Serializer::~Serializer() {
   delete external_reference_encoder_;
   if (code_address_map_ != NULL) delete code_address_map_;
 }
 
 
 void StartupSerializer::SerializeStrongReferences() {
@@ skipping 114 matching lines @@
 
 // Encode the location of an already deserialized object in order to write its
 // location into a later object.  We can encode the location as an offset from
 // the start of the deserialized objects or as an offset backwards from the
 // current allocation pointer.
 void Serializer::SerializeReferenceToPreviousObject(HeapObject* heap_object,
                                                     HowToCode how_to_code,
                                                     WhereToPoint where_to_point,
                                                     int skip) {
   int space = SpaceOfObject(heap_object);
-  int address = address_mapper_.MappedTo(heap_object);
-  int offset = CurrentAllocationAddress(space) - address;
-  // Shift out the bits that are always 0.
-  offset >>= kObjectAlignmentBits;
+
   if (skip == 0) {
     sink_->Put(kBackref + how_to_code + where_to_point + space, "BackRefSer");
   } else {
     sink_->Put(kBackrefWithSkip + how_to_code + where_to_point + space,
                "BackRefSerWithSkip");
     sink_->PutInt(skip, "BackRefSkipDistance");
   }
-  sink_->PutInt(offset, "offset");
+
+  if (space == LO_SPACE) {
+    int index = address_mapper_.MappedTo(heap_object);
+    sink_->PutInt(index, "large object index");
+  } else {
+    int address = address_mapper_.MappedTo(heap_object);
+    int offset = CurrentAllocationAddress(space) - address;
+    // Shift out the bits that are always 0.
+    offset >>= kObjectAlignmentBits;
+    sink_->PutInt(offset, "offset");
+  }
 }
 
 
 void StartupSerializer::SerializeObject(
     Object* o,
     HowToCode how_to_code,
     WhereToPoint where_to_point,
     int skip) {
   CHECK(o->IsHeapObject());
   HeapObject* heap_object = HeapObject::cast(o);
@@ skipping 138 matching lines @@
   if (serializer_->code_address_map_) {
     const char* code_name =
         serializer_->code_address_map_->Lookup(object_->address());
     LOG(serializer_->isolate_,
         CodeNameEvent(object_->address(), sink_->Position(), code_name));
     LOG(serializer_->isolate_,
         SnapshotPositionEvent(object_->address(), sink_->Position()));
   }
 
   // Mark this object as already serialized.
-  int offset = serializer_->Allocate(space, size);
-  serializer_->address_mapper()->AddMapping(object_, offset);
+  if (space == LO_SPACE) {
+    if (object_->IsCode()) {
+      sink_->PutInt(EXECUTABLE, "executable large object");
+    } else {
+      sink_->PutInt(NOT_EXECUTABLE, "not executable large object");
+    }
+    int index = serializer_->AllocateLargeObject(size);
+    serializer_->address_mapper()->AddMapping(object_, index);
+  } else {
+    int offset = serializer_->Allocate(space, size);
+    serializer_->address_mapper()->AddMapping(object_, offset);
+  }
 
   // Serialize the map (first word of the object).
   serializer_->SerializeObject(object_->map(), kPlain, kStartOfObject, 0);
 
   // Serialize the rest of the object.
   CHECK_EQ(0, bytes_processed_so_far_);
   bytes_processed_so_far_ = kPointerSize;
   object_->IterateBody(object_->map()->instance_type(), size, this);
   OutputRawData(object_->address() + size);
 }
@@ skipping 231 matching lines @@
     if (object->GetHeap()->InSpace(object, s)) {
       DCHECK(i < kNumberOfSpaces);
       return i;
     }
   }
   UNREACHABLE();
   return 0;
 }
 
 
+int Serializer::AllocateLargeObject(int size) {
+  fullness_[LO_SPACE] += size;
+  return seen_large_objects_index_++;
+}
+
+
 int Serializer::Allocate(int space, int size) {
-  CHECK(space >= 0 && space < kNumberOfSpaces);
+  CHECK(space >= 0 && space < kNumberOfPreallocatedSpaces);
   int allocation_address = fullness_[space];
   fullness_[space] = allocation_address + size;
   return allocation_address;
 }
 
 
 int Serializer::SpaceAreaSize(int space) {
   if (space == CODE_SPACE) {
     return isolate_->memory_allocator()->CodePageAreaSize();
   } else {
@@ skipping 71 matching lines @@
   DCHECK(!heap_object->IsHashTable());
 
   if (address_mapper_.IsMapped(heap_object)) {
     SerializeReferenceToPreviousObject(heap_object, how_to_code, where_to_point,
                                        skip);
     return;
   }
 
   if (heap_object->IsCode()) {
     Code* code_object = Code::cast(heap_object);
+    DCHECK(!code_object->is_optimized_code());
     if (code_object->kind() == Code::BUILTIN) {
       SerializeBuiltin(code_object, how_to_code, where_to_point, skip);
       return;
-    }
-    if (code_object->IsCodeStubOrIC()) {
+    } else if (code_object->IsCodeStubOrIC()) {
       SerializeCodeStub(code_object, how_to_code, where_to_point, skip);
       return;
     }
     code_object->ClearInlineCaches();
   }
 
   if (heap_object == source_) {
     SerializeSourceObject(how_to_code, where_to_point, skip);
     return;
   }
@@ skipping 126 matching lines @@
 
   Object* root;
 
   {
     HandleScope scope(isolate);
 
     SerializedCodeData scd(data, *source);
     SnapshotByteSource payload(scd.Payload(), scd.PayloadLength());
     Deserializer deserializer(&payload);
     STATIC_ASSERT(NEW_SPACE == 0);
-    for (int i = NEW_SPACE; i <= PROPERTY_CELL_SPACE; i++) {
+    for (int i = NEW_SPACE; i < kNumberOfSpaces; i++) {
       deserializer.set_reservation(i, scd.GetReservation(i));
     }
 
     // Prepare and register list of attached objects.
     Vector<const uint32_t> code_stub_keys = scd.CodeStubKeys();
     Vector<Handle<Object> > attached_objects = Vector<Handle<Object> >::New(
         code_stub_keys.length() + kCodeStubsBaseIndex);
     attached_objects[kSourceObjectIndex] = source;
     for (int i = 0; i < code_stub_keys.length(); i++) {
       attached_objects[i + kCodeStubsBaseIndex] =
@@ skipping 29 matching lines @@
   byte* data = NewArray<byte>(data_length);
   DCHECK(IsAligned(reinterpret_cast<intptr_t>(data), kPointerAlignment));
   script_data_ = new ScriptData(data, data_length);
   script_data_->AcquireDataOwnership();
 
   // Set header values.
   SetHeaderValue(kCheckSumOffset, CheckSum(cs->source()));
   SetHeaderValue(kNumCodeStubKeysOffset, num_stub_keys);
   SetHeaderValue(kPayloadLengthOffset, payload->length());
   STATIC_ASSERT(NEW_SPACE == 0);
-  for (int i = NEW_SPACE; i <= PROPERTY_CELL_SPACE; i++) {
+  for (int i = 0; i < SerializerDeserializer::kNumberOfSpaces; i++) {
     SetHeaderValue(kReservationsOffset + i, cs->CurrentAllocationAddress(i));
   }
 
   // Copy code stub keys.
   CopyBytes(data + kHeaderSize, reinterpret_cast<byte*>(stub_keys->begin()),
             stub_keys_size);
 
   // Copy serialized data.
   CopyBytes(data + kHeaderSize + stub_keys_size, payload->begin(),
             static_cast<size_t>(payload->length()));
 }
 
 
 bool SerializedCodeData::IsSane(String* source) {
   return GetHeaderValue(kCheckSumOffset) == CheckSum(source) &&
          PayloadLength() >= SharedFunctionInfo::kSize;
 }
 
 
 int SerializedCodeData::CheckSum(String* string) {
   int checksum = Version::Hash();
 #ifdef DEBUG
   uint32_t seed = static_cast<uint32_t>(checksum);
   checksum = static_cast<int>(IteratingStringHasher::Hash(string, seed));
 #endif  // DEBUG
   return checksum;
 }
 } }  // namespace v8::internal