Encode reservation meta data in the snapshot blob.

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 372 matching lines @@
 
   // A bitmask for getting the space out of an instruction.
   static const int kSpaceMask = 7;
   STATIC_ASSERT(kNumberOfSpaces <= kSpaceMask + 1);
 
   // Sentinel after a new object to indicate that double alignment is needed.
   static const int kDoubleAlignmentSentinel = 0;
 };
 
 
+class SerializedData {
+ public:
+  class Reservation {
+   public:
+    explicit Reservation(uint32_t size)
+        : reservation_(ChunkSizeBits::encode(size)) {}
+
+    uint32_t chunk_size() const { return ChunkSizeBits::decode(reservation_); }
+    bool is_last() const { return IsLastChunkBits::decode(reservation_); }
+
+    void mark_as_last() { reservation_ |= IsLastChunkBits::encode(true); }
+
+   private:
+    uint32_t reservation_;
+  };
+
+  SerializedData(byte* data, int size)
+      : data_(data), size_(size), owns_data_(false) {}
+  SerializedData() : data_(NULL), size_(0), owns_data_(true) {}

[vogelheim, 2014/12/03 18:49:25]

Here, owns_data_ == (data_ == NULL), which doesn't seem to make sense.

If I got this right, several subclass constructors will put 'owned' data_ there
by allocating memory. In that case it may make more sense to also let those
subclasses delete that memory in their destructors.

Your choice, though...

 

[Yang, 2014/12/04 08:51:43]

Acknowledged.

+
+  virtual ~SerializedData() {
+    if (owns_data_) DeleteArray<byte>(data_);
+  }
+
+  virtual Vector<const Reservation> Reservations() const = 0;
+  virtual Vector<const byte> Payload() const = 0;
+
+  class ChunkSizeBits : public BitField<uint32_t, 0, 31> {};
+  class IsLastChunkBits : public BitField<bool, 31, 1> {};
+
+ protected:
+  void SetHeaderValue(int offset, int value) {
+    reinterpret_cast<int*>(data_)[offset] = value;
+  }
+
+  int GetHeaderValue(int offset) const {
+    return reinterpret_cast<const int*>(data_)[offset];
+  }
+
+  byte* data_;
+  int size_;
+  bool owns_data_;
+};
+
+
 // A Deserializer reads a snapshot and reconstructs the Object graph it defines.
 class Deserializer: public SerializerDeserializer {
  public:
   // Create a deserializer from a snapshot byte source.
-  explicit Deserializer(SnapshotByteSource* source);
+  explicit Deserializer(SerializedData* serialized_data);
 
   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.
   // We may want to abort gracefully even if deserialization fails.
   void DeserializePartial(Isolate* isolate, Object** root,
                           OnOOM on_oom = FATAL_ON_OOM);
 
-  void AddReservation(int space, uint32_t chunk) {
-    DCHECK(space >= 0);
-    DCHECK(space < kNumberOfSpaces);
-    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; }
 
@@ skipping 21 matching lines @@
   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 by chunk index and offset.
   HeapObject* GetBackReferencedObject(int space) {
     if (space == LO_SPACE) {
-      uint32_t index = source_->GetInt();
+      uint32_t index = source_.GetInt();
       return deserialized_large_objects_[index];
     } else {
-      BackReference back_reference(source_->GetInt());
+      BackReference back_reference(source_.GetInt());
       DCHECK(space < kNumberOfPreallocatedSpaces);
       uint32_t chunk_index = back_reference.chunk_index();
       DCHECK_LE(chunk_index, current_chunk_[space]);
       uint32_t chunk_offset = back_reference.chunk_offset();
       return HeapObject::FromAddress(reservations_[space][chunk_index].start +
                                      chunk_offset);
     }
   }
 
   // Cached current isolate.
   Isolate* isolate_;
 
   // Objects from the attached object descriptions in the serialized user code.
   Vector<Handle<Object> >* attached_objects_;
 
-  SnapshotByteSource* source_;
+  SnapshotByteSource source_;
   // 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];
 
   ExternalReferenceDecoder* external_reference_decoder_;
 
   List<HeapObject*> deserialized_large_objects_;
 
   DISALLOW_COPY_AND_ASSIGN(Deserializer);
 };
 
 
 class CodeAddressMap;
 
 // There can be only one serializer per V8 process.
 class Serializer : public SerializerDeserializer {
  public:
   Serializer(Isolate* isolate, SnapshotByteSink* sink);
   ~Serializer();
   virtual void VisitPointers(Object** start, Object** end) OVERRIDE;
 
-  void FinalizeAllocation();
-
-  Vector<const uint32_t> FinalAllocationChunks(int space) const {
-    if (space == LO_SPACE) {
-      return Vector<const uint32_t>(&large_objects_total_size_, 1);
-    } else {
-      DCHECK_EQ(0, pending_chunk_[space]);  // No pending chunks.
-      return completed_chunks_[space].ToConstVector();
-    }
-  }
+  void EncodeReservations(List<SerializedData::Reservation>* out) const;
 
   Isolate* isolate() const { return isolate_; }
 
   BackReferenceMap* back_reference_map() { return &back_reference_map_; }
   RootIndexMap* root_index_map() { return &root_index_map_; }
 
  protected:
   class ObjectSerializer : public ObjectVisitor {
    public:
     ObjectSerializer(Serializer* serializer,
@@ skipping 193 matching lines @@
       Isolate* isolate, ScriptData* cached_data, Handle<String> source);
 
   static const int kSourceObjectIndex = 0;
   static const int kCodeStubsBaseIndex = 1;
 
   String* source() const {
     DCHECK(!AllowHeapAllocation::IsAllowed());
     return source_;
   }
 
-  List<uint32_t>* stub_keys() { return &stub_keys_; }
+  const List<uint32_t>* stub_keys() const { return &stub_keys_; }
   int num_internalized_strings() const { return num_internalized_strings_; }
 
  private:
   CodeSerializer(Isolate* isolate, SnapshotByteSink* sink, String* source,
                  Code* main_code)
       : Serializer(isolate, sink),
         source_(source),
         main_code_(main_code),
         num_internalized_strings_(0) {
     back_reference_map_.AddSourceString(source);
@@ skipping 16 matching lines @@
 
   DisallowHeapAllocation no_gc_;
   String* source_;
   Code* main_code_;
   int num_internalized_strings_;
   List<uint32_t> stub_keys_;
   DISALLOW_COPY_AND_ASSIGN(CodeSerializer);
 };
 
 
+// Wrapper around reservation sizes and the serialization payload.
+class SnapshotData : public SerializedData {
+ public:
+  // Used when producing.
+  SnapshotData(const SnapshotByteSink& sink, const Serializer& ser);
+
+  // Used when consuming.
+  explicit SnapshotData(const byte* data, int size)
+      : SerializedData(const_cast<byte*>(data), size) {
+    CHECK(IsSane());
+  }
+
+  virtual Vector<const Reservation> Reservations() const OVERRIDE;
+  virtual Vector<const byte> Payload() const OVERRIDE;
+
+  Vector<const byte> RawData() const {
+    return Vector<const byte>(data_, size_);
+  }
+
+ private:
+  bool IsSane();
+  // The data header consists of int-sized entries:
+  // [0] version hash
+  // [1] number of reservation size entries
+  // [2] payload length
+  static const int kCheckSumOffset = 0;
+  static const int kReservationsOffset = 1;
+  static const int kPayloadLengthOffset = 2;
+  static const int kHeaderSize = (kPayloadLengthOffset + 1) * kIntSize;
+};
+
+
 // Wrapper around ScriptData to provide code-serializer-specific functionality.
-class SerializedCodeData {
+class SerializedCodeData : public SerializedData {
  public:
-  // Used by when consuming.
+  // Used when consuming.
   static SerializedCodeData* FromCachedData(ScriptData* cached_data,
                                             String* source) {
     DisallowHeapAllocation no_gc;
     SerializedCodeData* scd = new SerializedCodeData(cached_data);
     if (scd->IsSane(source)) return scd;
     cached_data->Reject();
     delete scd;
     return NULL;
   }
 
   // Used when producing.
-  SerializedCodeData(const List<byte>& payload, CodeSerializer* cs);
-
-  ~SerializedCodeData() {
-    if (owns_script_data_) DeleteArray<byte>(data_);
-  }
+  SerializedCodeData(const List<byte>& payload, const CodeSerializer& cs);
 
   // Return ScriptData object and relinquish ownership over it to the caller.
-  ScriptData* GetScriptData() {
-    DCHECK(owns_script_data_);
-    ScriptData* result = new ScriptData(data_, size_);
-    result->AcquireDataOwnership();
-    return result;
-  }
+  ScriptData* GetScriptData();
 
-  class Reservation {
-   public:
-    uint32_t chunk_size() const { return ChunkSizeBits::decode(reservation); }
-    bool is_last_chunk() const { return IsLastChunkBits::decode(reservation); }
+  virtual Vector<const Reservation> Reservations() const OVERRIDE;
+  virtual Vector<const byte> Payload() const OVERRIDE;
 
-   private:
-    uint32_t reservation;
-
-    DISALLOW_COPY_AND_ASSIGN(Reservation);
-  };
-
-  int NumInternalizedStrings() const {
-    return GetHeaderValue(kNumInternalizedStringsOffset);
-  }
-
-  Vector<const Reservation> Reservations() const {
-    return Vector<const Reservation>(
-        reinterpret_cast<const Reservation*>(data_ + kHeaderSize),
-        GetHeaderValue(kReservationsOffset));
-  }
-
-  Vector<const uint32_t> CodeStubKeys() const {
-    int reservations_size = GetHeaderValue(kReservationsOffset) * kInt32Size;
-    const byte* start = 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 data_ + kHeaderSize + reservations_size + code_stubs_size;
-  }
-
-  int PayloadLength() const {
-    int payload_length = GetHeaderValue(kPayloadLengthOffset);
-    DCHECK_EQ(data_ + size_, Payload() + payload_length);
-    return payload_length;
-  }
+  int NumInternalizedStrings() const;
+  Vector<const uint32_t> CodeStubKeys() const;
 
  private:
   explicit SerializedCodeData(ScriptData* data)
-      : data_(const_cast<byte*>(data->data())),
-        size_(data->length()),
-        owns_script_data_(false) {}
-
-  void SetHeaderValue(int offset, int value) {
-    reinterpret_cast<int*>(data_)[offset] = value;
-  }
-
-  int GetHeaderValue(int offset) const {
-    return reinterpret_cast<const int*>(data_)[offset];
-  }
+      : SerializedData(const_cast<byte*>(data->data()), data->length()) {}
 
   bool IsSane(String* source);
 
   int CheckSum(String* source);
 
   // The data header consists of int-sized entries:
   // [0] version hash
   // [1] number of internalized strings
   // [2] number of code stub keys
   // [3] number of reservation size entries
-  // [3] payload length
+  // [4] payload length
   static const int kCheckSumOffset = 0;
   static const int kNumInternalizedStringsOffset = 1;
   static const int kReservationsOffset = 2;
   static const int kNumCodeStubKeysOffset = 3;
   static const int kPayloadLengthOffset = 4;
   static const int kHeaderSize = (kPayloadLengthOffset + 1) * kIntSize;
-
-  class ChunkSizeBits : public BitField<uint32_t, 0, 31> {};
-  class IsLastChunkBits : public BitField<bool, 31, 1> {};
-
-  byte* data_;
-  int size_;
-  bool owns_script_data_;
 };
 } }  // namespace v8::internal
 
 #endif  // V8_SERIALIZE_H_