[wasm] Split off debug info from wasm object

All debugging-related information is now stored inside a dedicated
object, which is only allocated if debugging support is needed.
This is also where later a reference to the interpreter will be stored
for executing to-be-debugged functions and providing stack inspection.

R=titzer@chromium.org, ahaas@chromium.org
BUG=chromium:613110

Review-Url: https://codereview.chromium.org/2050953003
Cr-Commit-Position: refs/heads/master@{#37055}

BUILD.gn

@@ -1478,6 +1478,8 @@ v8_source_set("v8_base") {
     "src/wasm/module-decoder.h",
     "src/wasm/switch-logic.cc",
     "src/wasm/switch-logic.h",
+    "src/wasm/wasm-debug.cc",
+    "src/wasm/wasm-debug.h",
     "src/wasm/wasm-external-refs.cc",
     "src/wasm/wasm-external-refs.h",
     "src/wasm/wasm-function-name-table.cc",

src/runtime/runtime-internal.cc

@@ -619,9 +619,9 @@ RUNTIME_FUNCTION(Runtime_OrdinaryHasInstance) {
 RUNTIME_FUNCTION(Runtime_IsWasmObject) {
   HandleScope scope(isolate);
   DCHECK_EQ(1, args.length());
-  CONVERT_ARG_HANDLE_CHECKED(Object, object, 0);
-  bool is_wasm_object = object->IsJSObject() &&
-                        wasm::IsWasmObject(Handle<JSObject>::cast(object));
+  CONVERT_ARG_CHECKED(Object, object, 0);
+  bool is_wasm_object =
+      object->IsJSObject() && wasm::IsWasmObject(JSObject::cast(object));
   return *isolate->factory()->ToBoolean(is_wasm_object);
 }
 

src/v8.gyp

@@ -1162,6 +1162,8 @@
         'wasm/module-decoder.h',
         'wasm/switch-logic.h',
         'wasm/switch-logic.cc',
+        'wasm/wasm-debug.cc',
+        'wasm/wasm-debug.h',
         'wasm/wasm-external-refs.cc',
         'wasm/wasm-external-refs.h',
         'wasm/wasm-function-name-table.cc',

src/wasm/wasm-debug.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/wasm/wasm-debug.h"
+
+#include "src/assert-scope.h"
+#include "src/debug/debug.h"
+#include "src/factory.h"
+#include "src/isolate.h"
+#include "src/wasm/module-decoder.h"
+#include "src/wasm/wasm-module.h"
+
+using namespace v8::internal;
+using namespace v8::internal::wasm;
+
+namespace {
+
+enum {
+  kWasmDebugInfoWasmObj,
+  kWasmDebugInfoWasmBytesHash,
+  kWasmDebugInfoFunctionByteOffsets,
+  kWasmDebugInfoNumEntries
+};
+
+ByteArray *GetOrCreateFunctionOffsetTable(WasmDebugInfo *debug_info) {
+  Object *offset_table = debug_info->get(kWasmDebugInfoFunctionByteOffsets);
+  Isolate *isolate = debug_info->GetIsolate();
+  if (!offset_table->IsUndefined(isolate)) return ByteArray::cast(offset_table);
+
+  FunctionOffsetsResult function_offsets;
+  {
+    DisallowHeapAllocation no_gc;
+    SeqOneByteString *wasm_bytes =
+        wasm::GetWasmBytes(debug_info->wasm_object());
+    const byte *bytes_start = wasm_bytes->GetChars();
+    const byte *bytes_end = bytes_start + wasm_bytes->length();
+    function_offsets = wasm::DecodeWasmFunctionOffsets(bytes_start, bytes_end);
+  }
+  DCHECK(function_offsets.ok());
+  size_t array_size = 2 * kIntSize * function_offsets.val.size();
+  CHECK_LE(array_size, static_cast<size_t>(kMaxInt));
+  ByteArray *arr =
+      *isolate->factory()->NewByteArray(static_cast<int>(array_size));
+  int idx = 0;
+  for (std::pair<int, int> p : function_offsets.val) {
+    arr->set_int(idx++, p.first);
+    arr->set_int(idx++, p.second);
+  }
+  DCHECK_EQ(arr->length(), idx * kIntSize);
+  debug_info->set(kWasmDebugInfoFunctionByteOffsets, arr);
+
+  return arr;
+}
+
+std::pair<int, int> GetFunctionOffsetAndLength(WasmDebugInfo *debug_info,
+                                               int func_index) {
+  ByteArray *arr = GetOrCreateFunctionOffsetTable(debug_info);
+  DCHECK(func_index >= 0 && func_index < arr->length() / kIntSize / 2);
+
+  int offset = arr->get_int(2 * func_index);
+  int length = arr->get_int(2 * func_index + 1);
+  // Assert that it's distinguishable from the "illegal function index" return.
+  DCHECK(offset > 0 && length > 0);
+  return {offset, length};
+}
+
+Vector<const uint8_t> GetFunctionBytes(WasmDebugInfo *debug_info,
+                                       int func_index) {
+  DCHECK(!AllowHeapAllocation::IsAllowed());
+  SeqOneByteString *module_bytes =
+      wasm::GetWasmBytes(debug_info->wasm_object());
+  std::pair<int, int> offset_and_length =
+      GetFunctionOffsetAndLength(debug_info, func_index);
+  return Vector<const uint8_t>(
+      module_bytes->GetChars() + offset_and_length.first,
+      offset_and_length.second);
+}
+
+}  // namespace
+
+Handle<WasmDebugInfo> WasmDebugInfo::New(Handle<JSObject> wasm) {
+  Isolate *isolate = wasm->GetIsolate();
+  Factory *factory = isolate->factory();
+  Handle<FixedArray> arr =
+      factory->NewFixedArray(kWasmDebugInfoNumEntries, TENURED);
+  arr->set(kWasmDebugInfoWasmObj, *wasm);
+  int hash = 0;
+  Handle<SeqOneByteString> wasm_bytes(GetWasmBytes(*wasm), isolate);
+  {
+    DisallowHeapAllocation no_gc;
+    hash = StringHasher::HashSequentialString(
+        wasm_bytes->GetChars(), wasm_bytes->length(), kZeroHashSeed);
+  }
+  Handle<Object> hash_obj = factory->NewNumberFromInt(hash, TENURED);
+  arr->set(kWasmDebugInfoWasmBytesHash, *hash_obj);
+
+  return Handle<WasmDebugInfo>::cast(arr);
+}
+
+bool WasmDebugInfo::IsDebugInfo(Object *object) {
+  if (!object->IsFixedArray()) return false;
+  FixedArray *arr = FixedArray::cast(object);
+  Isolate *isolate = arr->GetIsolate();
+  return arr->length() == kWasmDebugInfoNumEntries &&
+         IsWasmObject(arr->get(kWasmDebugInfoWasmObj)) &&
+         arr->get(kWasmDebugInfoWasmBytesHash)->IsNumber() &&
+         (arr->get(kWasmDebugInfoFunctionByteOffsets)->IsUndefined(isolate) ||
+          arr->get(kWasmDebugInfoFunctionByteOffsets)->IsByteArray());
+}
+
+WasmDebugInfo *WasmDebugInfo::cast(Object *object) {
+  DCHECK(IsDebugInfo(object));
+  return reinterpret_cast<WasmDebugInfo *>(object);
+}
+
+JSObject *WasmDebugInfo::wasm_object() {
+  return JSObject::cast(get(kWasmDebugInfoWasmObj));
+}
+
+bool WasmDebugInfo::SetBreakPoint(int byte_offset) {
+  // TODO(clemensh): Implement this.
+  return false;
+}
+
+Handle<String> WasmDebugInfo::DisassembleFunction(int func_index) {
+  std::ostringstream disassembly_os;
+
+  {
+    DisallowHeapAllocation no_gc;
+    Vector<const uint8_t> bytes_vec = GetFunctionBytes(this, func_index);
+
+    base::AccountingAllocator allocator;
+    bool ok = PrintAst(
+        &allocator, FunctionBodyForTesting(bytes_vec.start(), bytes_vec.end()),
+        disassembly_os, nullptr);
+    DCHECK(ok);
+    USE(ok);
+  }
+
+  // Unfortunately, we have to copy the string here.
+  std::string code_str = disassembly_os.str();
+  CHECK_LE(code_str.length(), static_cast<size_t>(kMaxInt));
+  return GetIsolate()
+      ->factory()
+      ->NewStringFromAscii(Vector<const char>(
+          code_str.data(), static_cast<int>(code_str.length())))
+      .ToHandleChecked();
+}
+
+Handle<FixedArray> WasmDebugInfo::GetFunctionOffsetTable(int func_index) {
+  class NullBuf : public std::streambuf {};
+  NullBuf null_buf;
+  std::ostream null_stream(&null_buf);
+
+  std::vector<std::tuple<uint32_t, int, int>> offset_table_vec;
+
+  {
+    DisallowHeapAllocation no_gc;
+    Vector<const uint8_t> bytes_vec = GetFunctionBytes(this, func_index);
+
+    v8::base::AccountingAllocator allocator;
+    bool ok = PrintAst(
+        &allocator, FunctionBodyForTesting(bytes_vec.start(), bytes_vec.end()),
+        null_stream, &offset_table_vec);
+    DCHECK(ok);
+    USE(ok);
+  }
+
+  size_t arr_size = 3 * offset_table_vec.size();
+  CHECK_LE(arr_size, static_cast<size_t>(kMaxInt));
+  Handle<FixedArray> offset_table = GetIsolate()->factory()->NewFixedArray(
+      static_cast<int>(arr_size), TENURED);
+
+  int idx = 0;
+  for (std::tuple<uint32_t, int, int> elem : offset_table_vec) {
+    offset_table->set(idx++, Smi::FromInt(std::get<0>(elem)));
+    offset_table->set(idx++, Smi::FromInt(std::get<1>(elem)));
+    offset_table->set(idx++, Smi::FromInt(std::get<2>(elem)));
+  }
+  DCHECK_EQ(idx, offset_table->length());
+
+  return offset_table;
+}

src/wasm/wasm-debug.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_WASM_DEBUG_H_
+#define V8_WASM_DEBUG_H_
+
+#include "src/handles.h"
+#include "src/objects.h"
+
+namespace v8 {
+namespace internal {
+namespace wasm {
+
+class WasmDebugInfo : public FixedArray {
+ public:
+  static Handle<WasmDebugInfo> New(Handle<JSObject> wasm);
+
+  static bool IsDebugInfo(Object* object);
+  static WasmDebugInfo* cast(Object* object);
+
+  JSObject* wasm_object();
+
+  bool SetBreakPoint(int byte_offset);
+
+  // Disassemble the specified function from this module.
+  Handle<String> DisassembleFunction(int func_index);
+
+  // Get the offset table for the specified function.
+  // Returns an array with three entries per instruction: byte offset, line and
+  // column.
+  Handle<FixedArray> GetFunctionOffsetTable(int func_index);
+};
+
+}  // namespace wasm
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_WASM_DEBUG_H_

src/wasm/wasm-module.cc

@@ -12,6 +12,7 @@
 
 #include "src/wasm/ast-decoder.h"
 #include "src/wasm/module-decoder.h"
+#include "src/wasm/wasm-debug.h"
 #include "src/wasm/wasm-function-name-table.h"
 #include "src/wasm/wasm-module.h"
 #include "src/wasm/wasm-result.h"
@@ -205,12 +206,15 @@ class WasmLinker {
 
 namespace {
 // Internal constants for the layout of the module object.
-const int kWasmModuleInternalFieldCount = 5;
 const int kWasmModuleFunctionTable = 0;
 const int kWasmModuleCodeTable = 1;
 const int kWasmMemArrayBuffer = 2;
 const int kWasmGlobalsArrayBuffer = 3;
+// TODO(clemensh): Remove function name array, extract names from module bytes.
 const int kWasmFunctionNamesArray = 4;
+const int kWasmModuleBytesString = 5;
+const int kWasmDebugInfo = 6;
+const int kWasmModuleInternalFieldCount = 7;
 
 void LoadDataSegments(const WasmModule* module, byte* mem_addr,
                       size_t mem_size) {
@@ -701,6 +705,16 @@ MaybeHandle<JSObject> WasmModule::Instantiate(
   Handle<FixedArray> code_table =
       factory->NewFixedArray(static_cast<int>(functions.size()), TENURED);
   instance.js_object->SetInternalField(kWasmModuleCodeTable, *code_table);
+  size_t module_bytes_len =
+      instance.module->module_end - instance.module->module_start;
+  DCHECK_LE(module_bytes_len, static_cast<size_t>(kMaxInt));
+  Vector<const uint8_t> module_bytes_vec(instance.module->module_start,
+                                         static_cast<int>(module_bytes_len));
+  Handle<String> module_bytes_string =
+      factory->NewStringFromOneByte(module_bytes_vec, TENURED)
+          .ToHandleChecked();
+  instance.js_object->SetInternalField(kWasmModuleBytesString,
+                                       *module_bytes_string);
 
   //-------------------------------------------------------------------------
   // Allocate and initialize the linear memory.
@@ -1010,12 +1024,36 @@ Handle<String> GetWasmFunctionName(Isolate* isolate, Handle<Object> wasm,
   return isolate->factory()->NewStringFromStaticChars("<WASM UNNAMED>");
 }
 
-bool IsWasmObject(Handle<JSObject> object) {
-  // TODO(clemensh): Check wasm byte header once we store a copy of the bytes.
-  return object->GetInternalFieldCount() == kWasmModuleInternalFieldCount &&
-         object->GetInternalField(kWasmModuleCodeTable)->IsFixedArray() &&
-         object->GetInternalField(kWasmMemArrayBuffer)->IsJSArrayBuffer() &&
-         object->GetInternalField(kWasmFunctionNamesArray)->IsByteArray();
+bool IsWasmObject(Object* object) {
+  if (!object->IsJSObject()) return false;
+  JSObject* obj = JSObject::cast(object);
+  if (obj->GetInternalFieldCount() != kWasmModuleInternalFieldCount ||
+      !obj->GetInternalField(kWasmModuleCodeTable)->IsFixedArray() ||
+      !obj->GetInternalField(kWasmMemArrayBuffer)->IsJSArrayBuffer() ||
+      !obj->GetInternalField(kWasmFunctionNamesArray)->IsByteArray() ||
+      !obj->GetInternalField(kWasmModuleBytesString)->IsSeqOneByteString()) {
+    return false;
+  }
+  DisallowHeapAllocation no_gc;
+  SeqOneByteString* bytes =
+      SeqOneByteString::cast(obj->GetInternalField(kWasmModuleBytesString));
+  if (bytes->length() < 4) return false;
+  if (memcmp(bytes->GetChars(), "\0asm", 4)) return false;
+
+  // All checks passed.
+  return true;
+}
+
+SeqOneByteString* GetWasmBytes(JSObject* wasm) {
+  return SeqOneByteString::cast(wasm->GetInternalField(kWasmModuleBytesString));
+}
+
+WasmDebugInfo* GetDebugInfo(JSObject* wasm) {
+  Object* info = wasm->GetInternalField(kWasmDebugInfo);
+  if (!info->IsUndefined(wasm->GetIsolate())) return WasmDebugInfo::cast(info);
+  Handle<WasmDebugInfo> new_info = WasmDebugInfo::New(handle(wasm));
+  wasm->SetInternalField(kWasmDebugInfo, *new_info);
+  return *new_info;
 }
 
 }  // namespace wasm

src/wasm/wasm-module.h

@@ -73,6 +73,8 @@ const uint8_t kWasmFunctionTypeForm = 0x40;
 #define WASM_SECTION_FUNCTION_BODIES_SIZE ((size_t)5)
 #define WASM_SECTION_NAMES_SIZE ((size_t)5)
 
+class WasmDebugInfo;
+
 struct WasmSection {
   enum class Code : uint32_t {
 #define F(enumerator, order, string) enumerator,
@@ -331,12 +333,19 @@ Handle<String> GetWasmFunctionName(Isolate* isolate, Handle<Object> wasm,
 Handle<Object> GetWasmFunctionNameOrNull(Isolate* isolate, Handle<Object> wasm,
                                          uint32_t func_index);
 
+// Return the binary source bytes of a wasm module.
+SeqOneByteString* GetWasmBytes(JSObject* wasm);
+
+// Get the debug info associated with the given wasm object.
+// If no debug info exists yet, it is created automatically.
+WasmDebugInfo* GetDebugInfo(JSObject* wasm);
+
 // Check whether the given object is a wasm object.
 // This checks the number and type of internal fields, so it's not 100 percent
 // secure. If it turns out that we need more complete checks, we could add a
 // special marker as internal field, which will definitely never occur anywhere
 // else.
-bool IsWasmObject(Handle<JSObject> object);
+bool IsWasmObject(Object* object);
 
 }  // namespace wasm
 }  // namespace internal