[serializer] split up src/snapshot/serialize.*

R=rossberg@chromium.org, ulan@chromium.org, vogelheim@chromium.org

Review URL: https://codereview.chromium.org/1751863002

Cr-Commit-Position: refs/heads/master@{#34395}

BUILD.gn

@@ -704,6 +704,7 @@ source_set("v8_base") {
     "src/code-stubs-hydrogen.cc",
     "src/codegen.cc",
     "src/codegen.h",
+    "src/collector.h",
     "src/compilation-cache.cc",
     "src/compilation-cache.h",
     "src/compilation-dependencies.cc",
@@ -1282,13 +1283,23 @@ source_set("v8_base") {
     "src/signature.h",
     "src/simulator.h",
     "src/small-pointer-list.h",
+    "src/snapshot/code-serializer.cc",
+    "src/snapshot/code-serializer.h",
+    "src/snapshot/deserializer.cc",
+    "src/snapshot/deserializer.h",
     "src/snapshot/natives.h",
     "src/snapshot/natives-common.cc",
-    "src/snapshot/serialize.cc",
-    "src/snapshot/serialize.h",
+    "src/snapshot/partial-serializer.cc",
+    "src/snapshot/partial-serializer.h",
+    "src/snapshot/serializer.cc",
+    "src/snapshot/serializer.h",
+    "src/snapshot/serializer-common.cc",
+    "src/snapshot/serializer-common.h",
     "src/snapshot/snapshot-common.cc",
     "src/snapshot/snapshot-source-sink.cc",
     "src/snapshot/snapshot-source-sink.h",
+    "src/snapshot/startup-serializer.cc",
+    "src/snapshot/startup-serializer.h",
     "src/source-position.h",
     "src/splay-tree.h",
     "src/splay-tree-inl.h",

src/assembler.cc

@@ -60,7 +60,7 @@
 #include "src/register-configuration.h"
 #include "src/runtime/runtime.h"
 #include "src/simulator.h"  // For flushing instruction cache.
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/serializer-common.h"
 
 #if V8_TARGET_ARCH_IA32
 #include "src/ia32/assembler-ia32-inl.h"  // NOLINT

src/collector.h

+// Copyright 2016 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_COLLECTOR_H_
+#define V8_COLLECTOR_H_
+
+#include "src/checks.h"
+#include "src/list.h"
+#include "src/vector.h"
+
+namespace v8 {
+namespace internal {
+
+/*
+ * A class that collects values into a backing store.
+ * Specialized versions of the class can allow access to the backing store
+ * in different ways.
+ * There is no guarantee that the backing store is contiguous (and, as a
+ * consequence, no guarantees that consecutively added elements are adjacent
+ * in memory). The collector may move elements unless it has guaranteed not
+ * to.
+ */
+template <typename T, int growth_factor = 2, int max_growth = 1 * MB>
+class Collector {
+ public:
+  explicit Collector(int initial_capacity = kMinCapacity)
+      : index_(0), size_(0) {
+    current_chunk_ = Vector<T>::New(initial_capacity);
+  }
+
+  virtual ~Collector() {
+    // Free backing store (in reverse allocation order).
+    current_chunk_.Dispose();
+    for (int i = chunks_.length() - 1; i >= 0; i--) {
+      chunks_.at(i).Dispose();
+    }
+  }
+
+  // Add a single element.
+  inline void Add(T value) {
+    if (index_ >= current_chunk_.length()) {
+      Grow(1);
+    }
+    current_chunk_[index_] = value;
+    index_++;
+    size_++;
+  }
+
+  // Add a block of contiguous elements and return a Vector backed by the
+  // memory area.
+  // A basic Collector will keep this vector valid as long as the Collector
+  // is alive.
+  inline Vector<T> AddBlock(int size, T initial_value) {
+    DCHECK(size > 0);
+    if (size > current_chunk_.length() - index_) {
+      Grow(size);
+    }
+    T* position = current_chunk_.start() + index_;
+    index_ += size;
+    size_ += size;
+    for (int i = 0; i < size; i++) {
+      position[i] = initial_value;
+    }
+    return Vector<T>(position, size);
+  }
+
+  // Add a contiguous block of elements and return a vector backed
+  // by the added block.
+  // A basic Collector will keep this vector valid as long as the Collector
+  // is alive.
+  inline Vector<T> AddBlock(Vector<const T> source) {
+    if (source.length() > current_chunk_.length() - index_) {
+      Grow(source.length());
+    }
+    T* position = current_chunk_.start() + index_;
+    index_ += source.length();
+    size_ += source.length();
+    for (int i = 0; i < source.length(); i++) {
+      position[i] = source[i];
+    }
+    return Vector<T>(position, source.length());
+  }
+
+  // Write the contents of the collector into the provided vector.
+  void WriteTo(Vector<T> destination) {
+    DCHECK(size_ <= destination.length());
+    int position = 0;
+    for (int i = 0; i < chunks_.length(); i++) {
+      Vector<T> chunk = chunks_.at(i);
+      for (int j = 0; j < chunk.length(); j++) {
+        destination[position] = chunk[j];
+        position++;
+      }
+    }
+    for (int i = 0; i < index_; i++) {
+      destination[position] = current_chunk_[i];
+      position++;
+    }
+  }
+
+  // Allocate a single contiguous vector, copy all the collected
+  // elements to the vector, and return it.
+  // The caller is responsible for freeing the memory of the returned
+  // vector (e.g., using Vector::Dispose).
+  Vector<T> ToVector() {
+    Vector<T> new_store = Vector<T>::New(size_);
+    WriteTo(new_store);
+    return new_store;
+  }
+
+  // Resets the collector to be empty.
+  virtual void Reset() {
+    for (int i = chunks_.length() - 1; i >= 0; i--) {
+      chunks_.at(i).Dispose();
+    }
+    chunks_.Rewind(0);
+    index_ = 0;
+    size_ = 0;
+  }
+
+  // Total number of elements added to collector so far.
+  inline int size() { return size_; }
+
+ protected:
+  static const int kMinCapacity = 16;
+  List<Vector<T> > chunks_;
+  Vector<T> current_chunk_;  // Block of memory currently being written into.
+  int index_;                // Current index in current chunk.
+  int size_;                 // Total number of elements in collector.
+
+  // Creates a new current chunk, and stores the old chunk in the chunks_ list.
+  void Grow(int min_capacity) {
+    DCHECK(growth_factor > 1);
+    int new_capacity;
+    int current_length = current_chunk_.length();
+    if (current_length < kMinCapacity) {
+      // The collector started out as empty.
+      new_capacity = min_capacity * growth_factor;
+      if (new_capacity < kMinCapacity) new_capacity = kMinCapacity;
+    } else {
+      int growth = current_length * (growth_factor - 1);
+      if (growth > max_growth) {
+        growth = max_growth;
+      }
+      new_capacity = current_length + growth;
+      if (new_capacity < min_capacity) {
+        new_capacity = min_capacity + growth;
+      }
+    }
+    NewChunk(new_capacity);
+    DCHECK(index_ + min_capacity <= current_chunk_.length());
+  }
+
+  // Before replacing the current chunk, give a subclass the option to move
+  // some of the current data into the new chunk. The function may update
+  // the current index_ value to represent data no longer in the current chunk.
+  // Returns the initial index of the new chunk (after copied data).
+  virtual void NewChunk(int new_capacity) {
+    Vector<T> new_chunk = Vector<T>::New(new_capacity);
+    if (index_ > 0) {
+      chunks_.Add(current_chunk_.SubVector(0, index_));
+    } else {
+      current_chunk_.Dispose();
+    }
+    current_chunk_ = new_chunk;
+    index_ = 0;
+  }
+};
+
+/*
+ * A collector that allows sequences of values to be guaranteed to
+ * stay consecutive.
+ * If the backing store grows while a sequence is active, the current
+ * sequence might be moved, but after the sequence is ended, it will
+ * not move again.
+ * NOTICE: Blocks allocated using Collector::AddBlock(int) can move
+ * as well, if inside an active sequence where another element is added.
+ */
+template <typename T, int growth_factor = 2, int max_growth = 1 * MB>
+class SequenceCollector : public Collector<T, growth_factor, max_growth> {
+ public:
+  explicit SequenceCollector(int initial_capacity)
+      : Collector<T, growth_factor, max_growth>(initial_capacity),
+        sequence_start_(kNoSequence) {}
+
+  virtual ~SequenceCollector() {}
+
+  void StartSequence() {
+    DCHECK(sequence_start_ == kNoSequence);
+    sequence_start_ = this->index_;
+  }
+
+  Vector<T> EndSequence() {
+    DCHECK(sequence_start_ != kNoSequence);
+    int sequence_start = sequence_start_;
+    sequence_start_ = kNoSequence;
+    if (sequence_start == this->index_) return Vector<T>();
+    return this->current_chunk_.SubVector(sequence_start, this->index_);
+  }
+
+  // Drops the currently added sequence, and all collected elements in it.
+  void DropSequence() {
+    DCHECK(sequence_start_ != kNoSequence);
+    int sequence_length = this->index_ - sequence_start_;
+    this->index_ = sequence_start_;
+    this->size_ -= sequence_length;
+    sequence_start_ = kNoSequence;
+  }
+
+  virtual void Reset() {
+    sequence_start_ = kNoSequence;
+    this->Collector<T, growth_factor, max_growth>::Reset();
+  }
+
+ private:
+  static const int kNoSequence = -1;
+  int sequence_start_;
+
+  // Move the currently active sequence to the new chunk.
+  virtual void NewChunk(int new_capacity) {
+    if (sequence_start_ == kNoSequence) {
+      // Fall back on default behavior if no sequence has been started.
+      this->Collector<T, growth_factor, max_growth>::NewChunk(new_capacity);
+      return;
+    }
+    int sequence_length = this->index_ - sequence_start_;
+    Vector<T> new_chunk = Vector<T>::New(sequence_length + new_capacity);
+    DCHECK(sequence_length < new_chunk.length());
+    for (int i = 0; i < sequence_length; i++) {
+      new_chunk[i] = this->current_chunk_[sequence_start_ + i];
+    }
+    if (sequence_start_ > 0) {
+      this->chunks_.Add(this->current_chunk_.SubVector(0, sequence_start_));
+    } else {
+      this->current_chunk_.Dispose();
+    }
+    this->current_chunk_ = new_chunk;
+    this->index_ = sequence_length;
+    sequence_start_ = 0;
+  }
+};
+
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_COLLECTOR_H_

src/compiler.cc

@@ -31,7 +31,7 @@
 #include "src/parsing/scanner-character-streams.h"
 #include "src/profiler/cpu-profiler.h"
 #include "src/runtime-profiler.h"
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/code-serializer.h"
 #include "src/vm-state-inl.h"
 
 namespace v8 {

src/disassembler.cc

@@ -10,7 +10,7 @@
 #include "src/deoptimizer.h"
 #include "src/disasm.h"
 #include "src/macro-assembler.h"
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/serializer-common.h"
 #include "src/string-stream.h"
 
 namespace v8 {

src/heap/heap.cc

@@ -36,7 +36,7 @@
 #include "src/regexp/jsregexp.h"
 #include "src/runtime-profiler.h"
 #include "src/snapshot/natives.h"
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/serializer-common.h"
 #include "src/snapshot/snapshot.h"
 #include "src/tracing/trace-event.h"
 #include "src/type-feedback-vector.h"
@@ -1143,7 +1143,8 @@ bool Heap::ReserveSpace(Reservation* reservations) {
   static const int kThreshold = 20;
   while (gc_performed && counter++ < kThreshold) {
     gc_performed = false;
-    for (int space = NEW_SPACE; space < Serializer::kNumberOfSpaces; space++) {
+    for (int space = NEW_SPACE; space < SerializerDeserializer::kNumberOfSpaces;
+         space++) {
       Reservation* reservation = &reservations[space];
       DCHECK_LE(1, reservation->length());
       if (reservation->at(0).size == 0) continue;
@@ -1169,7 +1170,7 @@ bool Heap::ReserveSpace(Reservation* reservations) {
             Address free_space_address = free_space->address();
             CreateFillerObjectAt(free_space_address, size,
                                  ClearRecordedSlots::kNo);
-            DCHECK(space < Serializer::kNumberOfPreallocatedSpaces);
+            DCHECK(space < SerializerDeserializer::kNumberOfPreallocatedSpaces);
             chunk.start = free_space_address;
             chunk.end = free_space_address + size;
           } else {

src/isolate.cc

@@ -34,7 +34,8 @@
 #include "src/regexp/regexp-stack.h"
 #include "src/runtime-profiler.h"
 #include "src/simulator.h"
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/deserializer.h"
+#include "src/snapshot/serializer-common.h"
 #include "src/v8.h"
 #include "src/version.h"
 #include "src/vm-state-inl.h"
@@ -2299,9 +2300,8 @@ bool Isolate::Init(Deserializer* des) {
     // the snapshot.
     HandleScope scope(this);
     Deoptimizer::EnsureCodeForDeoptimizationEntry(
-        this,
-        Deoptimizer::LAZY,
-        kDeoptTableSerializeEntryCount - 1);
+        this, Deoptimizer::LAZY,
+        ExternalReferenceTable::kDeoptTableSerializeEntryCount - 1);
   }
 
   if (!serializer_enabled()) {

src/parsing/preparse-data.h

@@ -6,6 +6,7 @@
 #define V8_PARSING_PREPARSE_DATA_H_
 
 #include "src/allocation.h"
+#include "src/collector.h"
 #include "src/hashmap.h"
 #include "src/messages.h"
 #include "src/parsing/preparse-data-format.h"

src/parsing/scanner.h

@@ -10,13 +10,13 @@
 #include "src/allocation.h"
 #include "src/base/logging.h"
 #include "src/char-predicates.h"
+#include "src/collector.h"
 #include "src/globals.h"
 #include "src/hashmap.h"
 #include "src/list.h"
 #include "src/parsing/token.h"
 #include "src/unicode.h"
 #include "src/unicode-decoder.h"
-#include "src/utils.h"
 
 namespace v8 {
 namespace internal {

src/snapshot/code-serializer.h

+// Copyright 2016 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_SNAPSHOT_CODE_SERIALIZER_H_
+#define V8_SNAPSHOT_CODE_SERIALIZER_H_
+
+#include "src/parsing/preparse-data.h"
+#include "src/snapshot/serializer.h"
+
+namespace v8 {
+namespace internal {
+
+class CodeSerializer : public Serializer {
+ public:
+  static ScriptData* Serialize(Isolate* isolate,
+                               Handle<SharedFunctionInfo> info,
+                               Handle<String> source);
+
+  MUST_USE_RESULT static MaybeHandle<SharedFunctionInfo> Deserialize(
+      Isolate* isolate, ScriptData* cached_data, Handle<String> source);
+
+  static const int kSourceObjectIndex = 0;
+  STATIC_ASSERT(kSourceObjectReference == kSourceObjectIndex);
+
+  static const int kCodeStubsBaseIndex = 1;
+
+  String* source() const {
+    DCHECK(!AllowHeapAllocation::IsAllowed());
+    return source_;
+  }
+
+  const List<uint32_t>* stub_keys() const { return &stub_keys_; }
+
+ private:
+  CodeSerializer(Isolate* isolate, SnapshotByteSink* sink, String* source)
+      : Serializer(isolate, sink), source_(source) {
+    back_reference_map_.AddSourceString(source);
+  }
+
+  ~CodeSerializer() override { OutputStatistics("CodeSerializer"); }
+
+  void SerializeObject(HeapObject* o, HowToCode how_to_code,
+                       WhereToPoint where_to_point, int skip) override;
+
+  void SerializeBuiltin(int builtin_index, HowToCode how_to_code,
+                        WhereToPoint where_to_point);
+  void SerializeIC(Code* ic, HowToCode how_to_code,
+                   WhereToPoint where_to_point);
+  void SerializeCodeStub(uint32_t stub_key, HowToCode how_to_code,
+                         WhereToPoint where_to_point);
+  void SerializeGeneric(HeapObject* heap_object, HowToCode how_to_code,
+                        WhereToPoint where_to_point);
+  int AddCodeStubKey(uint32_t stub_key);
+
+  DisallowHeapAllocation no_gc_;
+  String* source_;
+  List<uint32_t> stub_keys_;
+  DISALLOW_COPY_AND_ASSIGN(CodeSerializer);
+};
+
+// Wrapper around ScriptData to provide code-serializer-specific functionality.
+class SerializedCodeData : public SerializedData {
+ public:
+  // Used when consuming.
+  static SerializedCodeData* FromCachedData(Isolate* isolate,
+                                            ScriptData* cached_data,
+                                            String* source);
+
+  // Used when producing.
+  SerializedCodeData(const List<byte>& payload, const CodeSerializer& cs);
+
+  // Return ScriptData object and relinquish ownership over it to the caller.
+  ScriptData* GetScriptData();
+
+  Vector<const Reservation> Reservations() const;
+  Vector<const byte> Payload() const;
+
+  Vector<const uint32_t> CodeStubKeys() const;
+
+ private:
+  explicit SerializedCodeData(ScriptData* data);
+
+  enum SanityCheckResult {
+    CHECK_SUCCESS = 0,
+    MAGIC_NUMBER_MISMATCH = 1,
+    VERSION_MISMATCH = 2,
+    SOURCE_MISMATCH = 3,
+    CPU_FEATURES_MISMATCH = 4,
+    FLAGS_MISMATCH = 5,
+    CHECKSUM_MISMATCH = 6
+  };
+
+  SanityCheckResult SanityCheck(Isolate* isolate, String* source) const;
+
+  uint32_t SourceHash(String* source) const;
+
+  // The data header consists of uint32_t-sized entries:
+  // [0] magic number and external reference count
+  // [1] version hash
+  // [2] source hash
+  // [3] cpu features
+  // [4] flag hash
+  // [5] number of code stub keys
+  // [6] number of reservation size entries
+  // [7] payload length
+  // [8] payload checksum part 1
+  // [9] payload checksum part 2
+  // ...  reservations
+  // ...  code stub keys
+  // ...  serialized payload
+  static const int kVersionHashOffset = kMagicNumberOffset + kInt32Size;
+  static const int kSourceHashOffset = kVersionHashOffset + kInt32Size;
+  static const int kCpuFeaturesOffset = kSourceHashOffset + kInt32Size;
+  static const int kFlagHashOffset = kCpuFeaturesOffset + kInt32Size;
+  static const int kNumReservationsOffset = kFlagHashOffset + kInt32Size;
+  static const int kNumCodeStubKeysOffset = kNumReservationsOffset + kInt32Size;
+  static const int kPayloadLengthOffset = kNumCodeStubKeysOffset + kInt32Size;
+  static const int kChecksum1Offset = kPayloadLengthOffset + kInt32Size;
+  static const int kChecksum2Offset = kChecksum1Offset + kInt32Size;
+  static const int kHeaderSize = kChecksum2Offset + kInt32Size;
+};
+
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_SNAPSHOT_CODE_SERIALIZER_H_

src/snapshot/deserializer.h

+// Copyright 2016 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_SNAPSHOT_DESERIALIZER_H_
+#define V8_SNAPSHOT_DESERIALIZER_H_
+
+#include "src/heap/heap.h"
+#include "src/objects.h"
+#include "src/snapshot/serializer-common.h"
+#include "src/snapshot/snapshot-source-sink.h"
+
+namespace v8 {
+namespace internal {
+
+class Heap;
+
+// 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.
+  template <class Data>
+  explicit Deserializer(Data* data)
+      : isolate_(NULL),
+        source_(data->Payload()),
+        magic_number_(data->GetMagicNumber()),
+        external_reference_table_(NULL),
+        deserialized_large_objects_(0),
+        deserializing_user_code_(false),
+        next_alignment_(kWordAligned) {
+    DecodeReservation(data->Reservations());
+  }
+
+  ~Deserializer() override;
+
+  // Deserialize the snapshot into an empty heap.
+  void Deserialize(Isolate* isolate);
+
+  // Deserialize a single object and the objects reachable from it.
+  MaybeHandle<Object> DeserializePartial(Isolate* isolate,
+                                         Handle<JSGlobalProxy> global_proxy);
+
+  // Deserialize a shared function info. Fail gracefully.
+  MaybeHandle<SharedFunctionInfo> DeserializeCode(Isolate* isolate);
+
+  // Pass a vector of externally-provided objects referenced by the snapshot.
+  // The ownership to its backing store is handed over as well.
+  void SetAttachedObjects(Vector<Handle<Object> > attached_objects) {
+    attached_objects_ = attached_objects;
+  }
+
+ private:
+  void VisitPointers(Object** start, Object** end) override;
+
+  void Synchronize(VisitorSynchronization::SyncTag tag) override;
+
+  void VisitRuntimeEntry(RelocInfo* rinfo) override { UNREACHABLE(); }
+
+  void Initialize(Isolate* isolate);
+
+  bool deserializing_user_code() { return deserializing_user_code_; }
+
+  void DecodeReservation(Vector<const SerializedData::Reservation> res);
+
+  bool ReserveSpace();
+
+  void UnalignedCopy(Object** dest, Object** src) {
+    memcpy(dest, src, sizeof(*src));
+  }
+
+  void SetAlignment(byte data) {
+    DCHECK_EQ(kWordAligned, next_alignment_);
+    int alignment = data - (kAlignmentPrefix - 1);
+    DCHECK_LE(kWordAligned, alignment);
+    DCHECK_LE(alignment, kSimd128Unaligned);
+    next_alignment_ = static_cast<AllocationAlignment>(alignment);
+  }
+
+  void DeserializeDeferredObjects();
+
+  void FlushICacheForNewIsolate();
+  void FlushICacheForNewCodeObjects();
+
+  void CommitPostProcessedObjects(Isolate* isolate);
+
+  // 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. Return false if the object content has been deferred.
+  bool 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* PostProcessNewObject(HeapObject* obj, int space);
+
+  // This returns the address of an object that has been described in the
+  // snapshot by chunk index and offset.
+  HeapObject* GetBackReferencedObject(int space);
+
+  Object** CopyInNativesSource(Vector<const char> source_vector,
+                               Object** current);
+
+  // Cached current isolate.
+  Isolate* isolate_;
+
+  // Objects from the attached object descriptions in the serialized user code.
+  Vector<Handle<Object> > attached_objects_;
+
+  SnapshotByteSource source_;
+  uint32_t magic_number_;
+
+  // 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];
+
+  ExternalReferenceTable* external_reference_table_;
+
+  List<HeapObject*> deserialized_large_objects_;
+  List<Code*> new_code_objects_;
+  List<Handle<String> > new_internalized_strings_;
+  List<Handle<Script> > new_scripts_;
+
+  bool deserializing_user_code_;
+
+  AllocationAlignment next_alignment_;
+
+  DISALLOW_COPY_AND_ASSIGN(Deserializer);
+};
+
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_SNAPSHOT_DESERIALIZER_H_

src/snapshot/mksnapshot.cc

@@ -12,8 +12,8 @@
 #include "src/flags.h"
 #include "src/list.h"
 #include "src/snapshot/natives.h"
-#include "src/snapshot/serialize.h"
-
+#include "src/snapshot/partial-serializer.h"
+#include "src/snapshot/startup-serializer.h"
 
 using namespace v8;
 

src/snapshot/partial-serializer.cc

+// Copyright 2016 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/snapshot/partial-serializer.h"
+
+#include "src/objects-inl.h"
+
+namespace v8 {
+namespace internal {
+
+PartialSerializer::PartialSerializer(Isolate* isolate,
+                                     Serializer* startup_snapshot_serializer,
+                                     SnapshotByteSink* sink)
+    : Serializer(isolate, sink),
+      startup_serializer_(startup_snapshot_serializer),
+      global_object_(NULL) {
+  InitializeCodeAddressMap();
+}
+
+PartialSerializer::~PartialSerializer() {
+  OutputStatistics("PartialSerializer");
+}
+
+void PartialSerializer::Serialize(Object** o) {
+  if ((*o)->IsContext()) {
+    Context* context = Context::cast(*o);
+    global_object_ = context->global_object();
+    back_reference_map()->AddGlobalProxy(context->global_proxy());
+    // The bootstrap snapshot has a code-stub context. When serializing the
+    // partial snapshot, it is chained into the weak context list on the isolate
+    // 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());
+    }
+  }
+  VisitPointer(o);
+  SerializeDeferredObjects();
+  Pad();
+}
+
+void PartialSerializer::SerializeObject(HeapObject* obj, HowToCode how_to_code,
+                                        WhereToPoint where_to_point, int skip) {
+  if (obj->IsMap()) {
+    // The code-caches link to context-specific code objects, which
+    // the startup and context serializes cannot currently handle.
+    DCHECK(Map::cast(obj)->code_cache() == obj->GetHeap()->empty_fixed_array());
+  }
+
+  // Replace typed arrays by undefined.
+  if (obj->IsJSTypedArray()) obj = isolate_->heap()->undefined_value();
+
+  int root_index = root_index_map_.Lookup(obj);
+  if (root_index != RootIndexMap::kInvalidRootIndex) {
+    PutRoot(root_index, obj, how_to_code, where_to_point, skip);
+    return;
+  }
+
+  if (ShouldBeInThePartialSnapshotCache(obj)) {
+    FlushSkip(skip);
+
+    int cache_index = PartialSnapshotCacheIndex(obj);
+    sink_->Put(kPartialSnapshotCache + how_to_code + where_to_point,
+               "PartialSnapshotCache");
+    sink_->PutInt(cache_index, "partial_snapshot_cache_index");
+    return;
+  }
+
+  // Pointers from the partial snapshot to the objects in the startup snapshot
+  // should go through the root array or through the partial snapshot cache.
+  // If this is not the case you may have to add something to the root array.
+  DCHECK(!startup_serializer_->back_reference_map()->Lookup(obj).is_valid());
+  // All the internalized strings that the partial snapshot needs should be
+  // either in the root table or in the partial snapshot cache.
+  DCHECK(!obj->IsInternalizedString());
+
+  if (SerializeKnownObject(obj, how_to_code, where_to_point, skip)) return;
+
+  FlushSkip(skip);
+
+  // Clear literal boilerplates.
+  if (obj->IsJSFunction()) {
+    FixedArray* literals = JSFunction::cast(obj)->literals();
+    for (int i = 0; i < literals->length(); i++) literals->set_undefined(i);
+  }
+
+  // Object has not yet been serialized.  Serialize it here.
+  ObjectSerializer serializer(this, obj, sink_, how_to_code, where_to_point);
+  serializer.Serialize();
+}
+
+int PartialSerializer::PartialSnapshotCacheIndex(HeapObject* heap_object) {
+  Isolate* isolate = this->isolate();
+  List<Object*>* cache = isolate->partial_snapshot_cache();
+  int new_index = cache->length();
+
+  int index = partial_cache_index_map_.LookupOrInsert(heap_object, new_index);
+  if (index == PartialCacheIndexMap::kInvalidIndex) {
+    // We didn't find the object in the cache.  So we add it to the cache and
+    // then visit the pointer so that it becomes part of the startup snapshot
+    // and we can refer to it from the partial snapshot.
+    cache->Add(heap_object);
+    startup_serializer_->VisitPointer(reinterpret_cast<Object**>(&heap_object));
+    // We don't recurse from the startup snapshot generator into the partial
+    // snapshot generator.
+    return new_index;
+  }
+  return index;
+}
+
+bool PartialSerializer::ShouldBeInThePartialSnapshotCache(HeapObject* o) {
+  // Scripts should be referred only through shared function infos.  We can't
+  // allow them to be part of the partial snapshot because they contain a
+  // unique ID, and deserializing several partial snapshots containing script
+  // would cause dupes.
+  DCHECK(!o->IsScript());
+  return o->IsName() || o->IsSharedFunctionInfo() || o->IsHeapNumber() ||
+         o->IsCode() || o->IsScopeInfo() || o->IsAccessorInfo() ||
+         o->map() ==
+             startup_serializer_->isolate()->heap()->fixed_cow_array_map();
+}
+
+}  // namespace internal
+}  // namespace v8

src/snapshot/partial-serializer.h

+// Copyright 2016 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_SNAPSHOT_PARTIAL_SERIALIZER_H_
+#define V8_SNAPSHOT_PARTIAL_SERIALIZER_H_
+
+#include "src/address-map.h"
+#include "src/snapshot/serializer.h"
+
+namespace v8 {
+namespace internal {
+
+class PartialSerializer : public Serializer {
+ public:
+  PartialSerializer(Isolate* isolate, Serializer* startup_snapshot_serializer,
+                    SnapshotByteSink* sink);
+
+  ~PartialSerializer() override;
+
+  // Serialize the objects reachable from a single object pointer.
+  void Serialize(Object** o);
+
+ private:
+  class PartialCacheIndexMap : public AddressMapBase {
+   public:
+    PartialCacheIndexMap() : map_(HashMap::PointersMatch) {}
+
+    static const int kInvalidIndex = -1;
+
+    // Lookup object in the map. Return its index if found, or create
+    // a new entry with new_index as value, and return kInvalidIndex.
+    int LookupOrInsert(HeapObject* obj, int new_index) {
+      HashMap::Entry* entry = LookupEntry(&map_, obj, false);
+      if (entry != NULL) return GetValue(entry);
+      SetValue(LookupEntry(&map_, obj, true), static_cast<uint32_t>(new_index));
+      return kInvalidIndex;
+    }
+
+   private:
+    HashMap map_;
+
+    DISALLOW_COPY_AND_ASSIGN(PartialCacheIndexMap);
+  };
+
+  void SerializeObject(HeapObject* o, HowToCode how_to_code,
+                       WhereToPoint where_to_point, int skip) override;
+
+  int PartialSnapshotCacheIndex(HeapObject* o);
+  bool ShouldBeInThePartialSnapshotCache(HeapObject* o);
+
+  Serializer* startup_serializer_;
+  Object* global_object_;
+  PartialCacheIndexMap partial_cache_index_map_;
+  DISALLOW_COPY_AND_ASSIGN(PartialSerializer);
+};
+
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_SNAPSHOT_PARTIAL_SERIALIZER_H_

src/snapshot/snapshot-common.cc

@@ -9,6 +9,9 @@
 #include "src/api.h"
 #include "src/base/platform/platform.h"
 #include "src/full-codegen/full-codegen.h"
+#include "src/snapshot/deserializer.h"
+#include "src/snapshot/snapshot-source-sink.h"
+#include "src/version.h"
 
 namespace v8 {
 namespace internal {
@@ -228,5 +231,52 @@ Vector<const byte> Snapshot::ExtractContextData(const v8::StartupData* data) {
   int context_length = data->raw_size - context_offset;
   return Vector<const byte>(context_data, context_length);
 }
+
+SnapshotData::SnapshotData(const Serializer& ser) {
+  DisallowHeapAllocation no_gc;
+  List<Reservation> reservations;
+  ser.EncodeReservations(&reservations);
+  const List<byte>& payload = ser.sink()->data();
+
+  // Calculate sizes.
+  int reservation_size = reservations.length() * kInt32Size;
+  int size = kHeaderSize + reservation_size + payload.length();
+
+  // Allocate backing store and create result data.
+  AllocateData(size);
+
+  // Set header values.
+  SetMagicNumber(ser.isolate());
+  SetHeaderValue(kCheckSumOffset, Version::Hash());
+  SetHeaderValue(kNumReservationsOffset, reservations.length());
+  SetHeaderValue(kPayloadLengthOffset, payload.length());
+
+  // Copy reservation chunk sizes.
+  CopyBytes(data_ + kHeaderSize, reinterpret_cast<byte*>(reservations.begin()),
+            reservation_size);
+
+  // Copy serialized data.
+  CopyBytes(data_ + kHeaderSize + reservation_size, payload.begin(),
+            static_cast<size_t>(payload.length()));
+}
+
+bool SnapshotData::IsSane() {
+  return GetHeaderValue(kCheckSumOffset) == Version::Hash();
+}
+
+Vector<const SerializedData::Reservation> SnapshotData::Reservations() const {
+  return Vector<const Reservation>(
+      reinterpret_cast<const Reservation*>(data_ + kHeaderSize),
+      GetHeaderValue(kNumReservationsOffset));
+}
+
+Vector<const byte> SnapshotData::Payload() const {
+  int reservations_size = GetHeaderValue(kNumReservationsOffset) * kInt32Size;
+  const byte* payload = data_ + kHeaderSize + reservations_size;
+  int length = GetHeaderValue(kPayloadLengthOffset);
+  DCHECK_EQ(data_ + size_, payload + length);
+  return Vector<const byte>(payload, length);
+}
+
 }  // namespace internal
 }  // namespace v8

src/snapshot/snapshot-external.cc

@@ -7,7 +7,6 @@
 #include "src/snapshot/snapshot.h"
 
 #include "src/base/platform/mutex.h"
-#include "src/snapshot/serialize.h"
 #include "src/snapshot/snapshot-source-sink.h"
 #include "src/v8.h"  // for V8::Initialize
 

src/snapshot/snapshot-source-sink.cc

@@ -7,7 +7,6 @@
 
 #include "src/base/logging.h"
 #include "src/handles-inl.h"
-#include "src/snapshot/serialize.h"  // for SerializerDeserializer::nop()
 
 
 namespace v8 {

src/snapshot/snapshot.h

@@ -5,7 +5,8 @@
 #ifndef V8_SNAPSHOT_SNAPSHOT_H_
 #define V8_SNAPSHOT_SNAPSHOT_H_
 
-#include "src/snapshot/serialize.h"
+#include "src/snapshot/partial-serializer.h"
+#include "src/snapshot/startup-serializer.h"
 
 namespace v8 {
 namespace internal {
@@ -88,6 +89,41 @@ class Snapshot : public AllStatic {
 void SetSnapshotFromFile(StartupData* snapshot_blob);
 #endif
 
+// Wrapper around reservation sizes and the serialization payload.
+class SnapshotData : public SerializedData {
+ public:
+  // Used when producing.
+  explicit SnapshotData(const Serializer& ser);
+
+  // Used when consuming.
+  explicit SnapshotData(const Vector<const byte> snapshot)
+      : SerializedData(const_cast<byte*>(snapshot.begin()), snapshot.length()) {
+    CHECK(IsSane());
+  }
+
+  Vector<const Reservation> Reservations() const;
+  Vector<const byte> Payload() const;
+
+  Vector<const byte> RawData() const {
+    return Vector<const byte>(data_, size_);
+  }
+
+ private:
+  bool IsSane();
+
+  // The data header consists of uint32_t-sized entries:
+  // [0] magic number and external reference count
+  // [1] version hash
+  // [2] number of reservation size entries
+  // [3] payload length
+  // ... reservations
+  // ... serialized payload
+  static const int kCheckSumOffset = kMagicNumberOffset + kInt32Size;
+  static const int kNumReservationsOffset = kCheckSumOffset + kInt32Size;
+  static const int kPayloadLengthOffset = kNumReservationsOffset + kInt32Size;
+  static const int kHeaderSize = kPayloadLengthOffset + kInt32Size;
+};
+
 }  // namespace internal
 }  // namespace v8
 

src/snapshot/startup-serializer.cc

+// Copyright 2016 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/snapshot/startup-serializer.h"
+
+#include "src/objects-inl.h"
+#include "src/v8threads.h"
+
+namespace v8 {
+namespace internal {
+
+StartupSerializer::StartupSerializer(Isolate* isolate, SnapshotByteSink* sink)
+    : Serializer(isolate, sink),
+      root_index_wave_front_(0),
+      serializing_builtins_(false) {
+  // Clear the cache of objects used by the partial snapshot.  After the
+  // strong roots have been serialized we can create a partial snapshot
+  // which will repopulate the cache with objects needed by that partial
+  // snapshot.
+  isolate->partial_snapshot_cache()->Clear();
+  InitializeCodeAddressMap();
+}
+
+StartupSerializer::~StartupSerializer() {
+  OutputStatistics("StartupSerializer");
+}
+
+void StartupSerializer::SerializeObject(HeapObject* obj, HowToCode how_to_code,
+                                        WhereToPoint where_to_point, int skip) {
+  DCHECK(!obj->IsJSFunction());
+
+  if (obj->IsCode()) {
+    Code* code = Code::cast(obj);
+    // If the function code is compiled (either as native code or bytecode),
+    // replace it with lazy-compile builtin. Only exception is when we are
+    // serializing the canonical interpreter-entry-trampoline builtin.
+    if (code->kind() == Code::FUNCTION ||
+        (!serializing_builtins_ && code->is_interpreter_entry_trampoline())) {
+      obj = isolate()->builtins()->builtin(Builtins::kCompileLazy);
+    }
+  } else if (obj->IsBytecodeArray()) {
+    obj = isolate()->heap()->undefined_value();
+  }
+
+  int root_index = root_index_map_.Lookup(obj);
+  bool is_immortal_immovable_root = false;
+  // We can only encode roots as such if it has already been serialized.
+  // That applies to root indices below the wave front.
+  if (root_index != RootIndexMap::kInvalidRootIndex) {
+    if (root_index < root_index_wave_front_) {
+      PutRoot(root_index, obj, how_to_code, where_to_point, skip);
+      return;
+    } else {
+      is_immortal_immovable_root = Heap::RootIsImmortalImmovable(root_index);
+    }
+  }
+
+  if (SerializeKnownObject(obj, how_to_code, where_to_point, skip)) return;
+
+  FlushSkip(skip);
+
+  // Object has not yet been serialized.  Serialize it here.
+  ObjectSerializer object_serializer(this, obj, sink_, how_to_code,
+                                     where_to_point);
+  object_serializer.Serialize();
+
+  if (is_immortal_immovable_root) {
+    // Make sure that the immortal immovable root has been included in the first
+    // chunk of its reserved space , so that it is deserialized onto the first
+    // page of its space and stays immortal immovable.
+    BackReference ref = back_reference_map_.Lookup(obj);
+    CHECK(ref.is_valid() && ref.chunk_index() == 0);
+  }
+}
+
+void StartupSerializer::SerializeWeakReferencesAndDeferred() {
+  // This phase comes right after the serialization (of the snapshot).
+  // After we have done the partial serialization the partial snapshot cache
+  // will contain some references needed to decode the partial snapshot.  We
+  // add one entry with 'undefined' which is the sentinel that the deserializer
+  // uses to know it is done deserializing the array.
+  Object* undefined = isolate()->heap()->undefined_value();
+  VisitPointer(&undefined);
+  isolate()->heap()->IterateWeakRoots(this, VISIT_ALL);
+  SerializeDeferredObjects();
+  Pad();
+}
+
+void StartupSerializer::Synchronize(VisitorSynchronization::SyncTag tag) {
+  // We expect the builtins tag after builtins have been serialized.
+  DCHECK(!serializing_builtins_ || tag == VisitorSynchronization::kBuiltins);
+  serializing_builtins_ = (tag == VisitorSynchronization::kHandleScope);
+  sink_->Put(kSynchronize, "Synchronize");
+}
+
+void StartupSerializer::SerializeStrongReferences() {
+  Isolate* isolate = this->isolate();
+  // No active threads.
+  CHECK_NULL(isolate->thread_manager()->FirstThreadStateInUse());
+  // No active or weak handles.
+  CHECK(isolate->handle_scope_implementer()->blocks()->is_empty());
+  CHECK_EQ(0, isolate->global_handles()->NumberOfWeakHandles());
+  CHECK_EQ(0, isolate->eternal_handles()->NumberOfHandles());
+  // We don't support serializing installed extensions.
+  CHECK(!isolate->has_installed_extensions());
+  isolate->heap()->IterateSmiRoots(this);
+  isolate->heap()->IterateStrongRoots(this, VISIT_ONLY_STRONG);
+}
+
+void StartupSerializer::VisitPointers(Object** start, Object** end) {
+  for (Object** current = start; current < end; current++) {
+    if (start == isolate()->heap()->roots_array_start()) {
+      root_index_wave_front_ =
+          Max(root_index_wave_front_, static_cast<intptr_t>(current - start));
+    }
+    if (ShouldBeSkipped(current)) {
+      sink_->Put(kSkip, "Skip");
+      sink_->PutInt(kPointerSize, "SkipOneWord");
+    } else if ((*current)->IsSmi()) {
+      sink_->Put(kOnePointerRawData, "Smi");
+      for (int i = 0; i < kPointerSize; i++) {
+        sink_->Put(reinterpret_cast<byte*>(current)[i], "Byte");
+      }
+    } else {
+      SerializeObject(HeapObject::cast(*current), kPlain, kStartOfObject, 0);
+    }
+  }
+}
+
+}  // namespace internal
+}  // namespace v8

src/v8.cc

@@ -19,7 +19,6 @@
 #include "src/profiler/sampler.h"
 #include "src/runtime-profiler.h"
 #include "src/snapshot/natives.h"
-#include "src/snapshot/serialize.h"
 #include "src/snapshot/snapshot.h"
 
 

test/cctest/test-heap-profiler.cc

@@ -32,6 +32,7 @@
 #include "src/v8.h"
 
 #include "include/v8-profiler.h"
+#include "src/collector.h"
 #include "src/debug/debug.h"
 #include "src/hashmap.h"
 #include "src/profiler/allocation-tracker.h"