Break deserializer reservations into chunks that fit onto a page.

src/serialize.h

 // 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.
 
 #ifndef V8_SERIALIZE_H_
 #define V8_SERIALIZE_H_
 
 #include "src/compiler.h"
 #include "src/hashmap.h"
 #include "src/heap-profiler.h"
@@ skipping 134 matching lines @@
 class SerializerDeserializer: public ObjectVisitor {
  public:
   static void Iterate(Isolate* isolate, ObjectVisitor* visitor);
 
   static int nop() { return kNop; }
 
   // No reservation for large object space necessary.
   static const int kNumberOfPreallocatedSpaces = LO_SPACE;
   static const int kNumberOfSpaces = INVALID_SPACE;
 
+  // To encode object for back-references.
+  class OffsetBits : public BitField<uint32_t, 0, kPageSizeBits> {};
+  class ChunkIndexBits
+      : public BitField<uint32_t, kPageSizeBits, 32 - kPageSizeBits> {};
+
  protected:
   // Where the pointed-to object can be found:
   enum Where {
     kNewObject = 0,  // Object is next in snapshot.
     // 1-7                             One per space.
     kRootArray = 0x9,             // Object is found in root array.
     kPartialSnapshotCache = 0xa,  // Object is in the cache.
     kExternalReference = 0xb,     // Pointer to an external reference.
     kSkip = 0xc,                  // Skip n bytes.
     kBuiltin = 0xd,               // Builtin code object.
@@ skipping 76 matching lines @@
 class Deserializer: public SerializerDeserializer {
  public:
   // Create a deserializer from a snapshot byte source.
   explicit Deserializer(SnapshotByteSource* source);
 
   virtual ~Deserializer();
 
   // Deserialize the snapshot into an empty heap.
   void Deserialize(Isolate* isolate);
 
+  enum OnOOM { FATAL_ON_OOM, NULL_ON_OOM };
+
   // Deserialize a single object and the objects reachable from it.
-  void DeserializePartial(Isolate* isolate, Object** root);
+  // We may want to abort and gracefully even if deserialization fails.

[mvstanton, 2014/10/15 12:30:56]

Remove the and to get "...abort gracefully..."

+  void DeserializePartial(Isolate* isolate, Object** root,
+                          OnOOM on_oom = FATAL_ON_OOM);
 
-  void set_reservation(int space_number, int reservation) {
-    DCHECK(space_number >= 0);
-    DCHECK(space_number < kNumberOfSpaces);
-    reservations_[space_number] = reservation;
+  void AddReservation(int space, uint32_t chunk) {
+    DCHECK(space >= 0);
+    DCHECK(space < kNumberOfSpaces);
+    DCHECK(space == LO_SPACE || chunk < Page::kMaxRegularHeapObjectSize);
+    reservations_[space].Add({chunk, NULL, NULL});
   }
 
   void FlushICacheForNewCodeObjects();
 
   // Serialized user code reference certain objects that are provided in a list
   // By calling this method, we assume that we are deserializing user code.
   void SetAttachedObjects(Vector<Handle<Object> >* attached_objects) {
     attached_objects_ = attached_objects;
   }
 
   bool deserializing_user_code() { return attached_objects_ != NULL; }
 
  private:
   virtual void VisitPointers(Object** start, Object** end);
 
   virtual void VisitRuntimeEntry(RelocInfo* rinfo) {
     UNREACHABLE();
   }
 
+  bool ReserveSpace();
+
   // Allocation sites are present in the snapshot, and must be linked into
   // a list at deserialization time.
   void RelinkAllocationSite(AllocationSite* site);
 
   // Fills in some heap data in an area from start to end (non-inclusive).  The
   // space id is used for the write barrier.  The object_address is the address
   // of the object we are writing into, or NULL if we are not writing into an
   // object, i.e. if we are writing a series of tagged values that are not on
   // the heap.
-  void ReadChunk(
-      Object** start, Object** end, int space, Address object_address);
+  void ReadData(Object** start, Object** end, int space,
+                Address object_address);
   void ReadObject(int space_number, Object** write_back);
   Address Allocate(int space_index, int size);
 
   // Special handling for serialized code like hooking up internalized strings.
   HeapObject* ProcessNewObjectFromSerializedCode(HeapObject* obj);
   Object* ProcessBackRefInSerializedCode(Object* obj);
 
   // This returns the address of an object that has been described in the
-  // snapshot as being offset bytes back in a particular space.
-  HeapObject* GetAddressFromEnd(int space) {
-    int offset = source_->GetInt();
-    if (space == LO_SPACE) return deserialized_large_objects_[offset];
-    DCHECK(space < kNumberOfPreallocatedSpaces);
-    offset <<= kObjectAlignmentBits;
-    return HeapObject::FromAddress(high_water_[space] - offset);
+  // snapshot by chunk index and offset.
+  HeapObject* GetBackReferencedObject(int space) {
+    if (space == LO_SPACE) {
+      uint32_t index = source_->GetInt();
+      return deserialized_large_objects_[index];
+    } else {
+      uint32_t allocation = source_->GetInt() << kObjectAlignmentBits;
+      DCHECK(space < kNumberOfPreallocatedSpaces);
+      uint32_t chunk_index = ChunkIndexBits::decode(allocation);
+      uint32_t offset = OffsetBits::decode(allocation);
+      DCHECK_LE(chunk_index, current_chunk_[space]);
+      return HeapObject::FromAddress(reservations_[space][chunk_index].start +
+                                     offset);
+    }
   }
 
   // Cached current isolate.
   Isolate* isolate_;
 
   // Objects from the attached object descriptions in the serialized user code.
   Vector<Handle<Object> >* attached_objects_;
 
   SnapshotByteSource* source_;
-  // This is the address of the next object that will be allocated in each
-  // space.  It is used to calculate the addresses of back-references.
+  // The address of the next object that will be allocated in each space.
+  // Each space has a number of chunks reserved by the GC, with each chunk
+  // fitting into a page. Deserialized objects are allocated into the
+  // current chunk of the target space by bumping up high water mark.
+  Heap::Reservation reservations_[kNumberOfSpaces];
+  uint32_t current_chunk_[kNumberOfPreallocatedSpaces];
   Address high_water_[kNumberOfPreallocatedSpaces];
 
-  int reservations_[kNumberOfSpaces];
-  static const intptr_t kUninitializedReservation = -1;
-
   ExternalReferenceDecoder* external_reference_decoder_;
 
   List<HeapObject*> deserialized_large_objects_;
 
   DISALLOW_COPY_AND_ASSIGN(Deserializer);
 };
 
 
 // Mapping objects to their location after deserialization.
 // This is used during building, but not at runtime by V8.
@@ skipping 44 matching lines @@
 
 
 class CodeAddressMap;
 
 // There can be only one serializer per V8 process.
 class Serializer : public SerializerDeserializer {
  public:
   Serializer(Isolate* isolate, SnapshotByteSink* sink);
   ~Serializer();
   void VisitPointers(Object** start, Object** end);
-  // You can call this after serialization to find out how much space was used
-  // in each space.
-  int CurrentAllocationAddress(int space) const {
-    DCHECK(space < kNumberOfSpaces);
-    return fullness_[space];
+
+  void FinalizeAllocation();
+
+  Vector<const uint32_t> FinalAllocationChunks(int space) const {
+    DCHECK_EQ(1, completed_chunks_[LO_SPACE].length());  // Already finalized.
+    DCHECK_EQ(0, pending_chunk_[space]);                 // No pending chunks.
+    return completed_chunks_[space].ToConstVector();
   }
 
   Isolate* isolate() const { return isolate_; }
 
   SerializationAddressMapper* address_mapper() { return &address_mapper_; }
   void PutRoot(int index,
                HeapObject* object,
                HowToCode how,
                WhereToPoint where,
                int skip);
@@ skipping 65 matching lines @@
                                HowToCode how_to_code,
                                WhereToPoint where_to_point,
                                int skip) = 0;
   void SerializeReferenceToPreviousObject(HeapObject* heap_object,
                                           HowToCode how_to_code,
                                           WhereToPoint where_to_point,
                                           int skip);
   void InitializeAllocators();
   // This will return the space for an object.
   static int SpaceOfObject(HeapObject* object);
-  int AllocateLargeObject(int size);
-  int Allocate(int space, int size);
+  uint32_t AllocateLargeObject(int size);
+  uint32_t Allocate(int space, int size);
   int EncodeExternalReference(Address addr) {
     return external_reference_encoder_->Encode(addr);
   }
 
   int SpaceAreaSize(int space);
 
   // Some roots should not be serialized, because their actual value depends on
   // absolute addresses and they are reset after deserialization, anyway.
   bool ShouldBeSkipped(Object** current);
 
   Isolate* isolate_;
-  // Keep track of the fullness of each space in order to generate
-  // relative addresses for back references.
-  int fullness_[kNumberOfSpaces];
+
+  // Objects from the same space are put into chunks for bulk-allocation
+  // when deserializing. We have to make sure that each chunk fits into a
+  // page. So we track the chunk size in pending_chunk_ of a space, but
+  // when it exceeds a page, we complete the current chunk and start a new one.
+  uint32_t pending_chunk_[kNumberOfSpaces];
+  List<uint32_t> completed_chunks_[kNumberOfSpaces];
+
   SnapshotByteSink* sink_;
   ExternalReferenceEncoder* external_reference_encoder_;
 
   SerializationAddressMapper address_mapper_;
   intptr_t root_index_wave_front_;
   void Pad();
 
   friend class ObjectSerializer;
   friend class Deserializer;
 
   // We may not need the code address map for logging for every instance
   // of the serializer.  Initialize it on demand.
   void InitializeCodeAddressMap();
 
  private:
   CodeAddressMap* code_address_map_;
   // We map serialized large objects to indexes for back-referencing.
-  int seen_large_objects_index_;
+  uint32_t seen_large_objects_index_;
   DISALLOW_COPY_AND_ASSIGN(Serializer);
 };
 
 
 class PartialSerializer : public Serializer {
  public:
   PartialSerializer(Isolate* isolate,
                     Serializer* startup_snapshot_serializer,
                     SnapshotByteSink* sink)
     : Serializer(isolate, sink),
@@ skipping 61 matching lines @@
   DISALLOW_COPY_AND_ASSIGN(StartupSerializer);
 };
 
 
 class CodeSerializer : public Serializer {
  public:
   static ScriptData* Serialize(Isolate* isolate,
                                Handle<SharedFunctionInfo> info,
                                Handle<String> source);
 
-  static Handle<SharedFunctionInfo> Deserialize(Isolate* isolate,
-                                                ScriptData* data,
-                                                Handle<String> source);
+  MUST_USE_RESULT static MaybeHandle<SharedFunctionInfo> Deserialize(
+      Isolate* isolate, ScriptData* data, Handle<String> source);
 
   static const int kSourceObjectIndex = 0;
   static const int kCodeStubsBaseIndex = 1;
 
   String* source() {
     DCHECK(!AllowHeapAllocation::IsAllowed());
     return source_;
   }
 
   List<uint32_t>* stub_keys() { return &stub_keys_; }
@@ skipping 46 matching lines @@
 
   // Return ScriptData object and relinquish ownership over it to the caller.
   ScriptData* GetScriptData() {
     ScriptData* result = script_data_;
     script_data_ = NULL;
     DCHECK(owns_script_data_);
     owns_script_data_ = false;
     return result;
   }
 
+  class Reservation {
+   public:
+    uint32_t chunk_size() const { return ChunkSizeBits::decode(reservation); }
+    bool is_last_chunk() const { return IsLastChunkBits::decode(reservation); }
+
+   private:
+    uint32_t reservation;
+
+    DISALLOW_COPY_AND_ASSIGN(Reservation);
+  };
+
+  Vector<const Reservation> Reservations() const {
+    return Vector<const Reservation>(reinterpret_cast<const Reservation*>(
+                                         script_data_->data() + kHeaderSize),
+                                     GetHeaderValue(kReservationsOffset));
+  }
+
   Vector<const uint32_t> CodeStubKeys() const {
-    return Vector<const uint32_t>(
-        reinterpret_cast<const uint32_t*>(script_data_->data() + kHeaderSize),
-        GetHeaderValue(kNumCodeStubKeysOffset));
+    int reservations_size = GetHeaderValue(kReservationsOffset) * kInt32Size;
+    const byte* start = script_data_->data() + kHeaderSize + reservations_size;
+    return Vector<const uint32_t>(reinterpret_cast<const uint32_t*>(start),
+                                  GetHeaderValue(kNumCodeStubKeysOffset));
   }
 
   const byte* Payload() const {
+    int reservations_size = GetHeaderValue(kReservationsOffset) * kInt32Size;
     int code_stubs_size = GetHeaderValue(kNumCodeStubKeysOffset) * kInt32Size;
-    return script_data_->data() + kHeaderSize + code_stubs_size;
+    return script_data_->data() + kHeaderSize + reservations_size +
+           code_stubs_size;
   }
 
   int PayloadLength() const {
     int payload_length = GetHeaderValue(kPayloadLengthOffset);
     DCHECK_EQ(script_data_->data() + script_data_->length(),
               Payload() + payload_length);
     return payload_length;
   }
 
-  int GetReservation(int space) const {
-    return GetHeaderValue(kReservationsOffset + space);
-  }
-
  private:
   void SetHeaderValue(int offset, int value) {
     reinterpret_cast<int*>(const_cast<byte*>(script_data_->data()))[offset] =
         value;
   }
 
   int GetHeaderValue(int offset) const {
     return reinterpret_cast<const int*>(script_data_->data())[offset];
   }
 
   bool IsSane(String* source);
 
   int CheckSum(String* source);
 
   // The data header consists of int-sized entries:
   // [0] version hash
   // [1] number of code stub keys
   // [2] payload length
   // [3..9] reservation sizes for spaces from NEW_SPACE to PROPERTY_CELL_SPACE.
   static const int kCheckSumOffset = 0;
-  static const int kNumCodeStubKeysOffset = 1;
-  static const int kPayloadLengthOffset = 2;
-  static const int kReservationsOffset = 3;
+  static const int kReservationsOffset = 1;
+  static const int kNumCodeStubKeysOffset = 2;
+  static const int kPayloadLengthOffset = 3;
+  static const int kHeaderSize = (kPayloadLengthOffset + 1) * kIntSize;
 
-  static const int kHeaderEntries =
-      kReservationsOffset + SerializerDeserializer::kNumberOfSpaces;
-  static const int kHeaderSize = kHeaderEntries * kIntSize;
+  class ChunkSizeBits : public BitField<uint32_t, 0, 31> {};
+  class IsLastChunkBits : public BitField<bool, 31, 1> {};
 
   // Following the header, we store, in sequential order
   // - code stub keys
   // - serialization payload
 
   ScriptData* script_data_;
   bool owns_script_data_;
 };
 } }  // namespace v8::internal
 
 #endif  // V8_SERIALIZE_H_