Always update raw pointers when handling interrupts inside RegExp code.

src/x64/regexp-macro-assembler-x64.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"
 
 #if V8_TARGET_ARCH_X64
 
 #include "src/cpu-profiler.h"
 #include "src/log.h"
@@ skipping 1158 matching lines @@
 }
 
 
 // Helper function for reading a value out of a stack frame.
 template <typename T>
 static T& frame_entry(Address re_frame, int frame_offset) {
   return reinterpret_cast<T&>(Memory::int32_at(re_frame + frame_offset));
 }
 
 
+template <typename T>
+static T* frame_entry_address(Address re_frame, int frame_offset) {
+  return reinterpret_cast<T*>(re_frame + frame_offset);
+}
+
+
 int RegExpMacroAssemblerX64::CheckStackGuardState(Address* return_address,
                                                   Code* re_code,
                                                   Address re_frame) {
-  Isolate* isolate = frame_entry<Isolate*>(re_frame, kIsolate);
-  StackLimitCheck check(isolate);
-  if (check.JsHasOverflowed()) {
-    isolate->StackOverflow();
-    return EXCEPTION;
-  }
-
-  // If not real stack overflow the stack guard was used to interrupt
-  // execution for another purpose.
-
-  // If this is a direct call from JavaScript retry the RegExp forcing the call
-  // through the runtime system. Currently the direct call cannot handle a GC.
-  if (frame_entry<int>(re_frame, kDirectCall) == 1) {
-    return RETRY;
-  }
-
-  // Prepare for possible GC.
-  HandleScope handles(isolate);
-  Handle<Code> code_handle(re_code);
-
-  Handle<String> subject(frame_entry<String*>(re_frame, kInputString));
-
-  // Current string.
-  bool is_one_byte = subject->IsOneByteRepresentationUnderneath();
-
-  DCHECK(re_code->instruction_start() <= *return_address);
-  DCHECK(*return_address <=
-      re_code->instruction_start() + re_code->instruction_size());
-
-  Object* result = isolate->stack_guard()->HandleInterrupts();
-
-  if (*code_handle != re_code) {  // Return address no longer valid
-    intptr_t delta = code_handle->address() - re_code->address();
-    // Overwrite the return address on the stack.
-    *return_address += delta;
-  }
-
-  if (result->IsException()) {
-    return EXCEPTION;
-  }
-
-  Handle<String> subject_tmp = subject;
-  int slice_offset = 0;
-
-  // Extract the underlying string and the slice offset.
-  if (StringShape(*subject_tmp).IsCons()) {
-    subject_tmp = Handle<String>(ConsString::cast(*subject_tmp)->first());
-  } else if (StringShape(*subject_tmp).IsSliced()) {
-    SlicedString* slice = SlicedString::cast(*subject_tmp);
-    subject_tmp = Handle<String>(slice->parent());
-    slice_offset = slice->offset();
-  }
-
-  // String might have changed.
-  if (subject_tmp->IsOneByteRepresentation() != is_one_byte) {
-    // If we changed between an Latin1 and an UC16 string, the specialized
-    // code cannot be used, and we need to restart regexp matching from
-    // scratch (including, potentially, compiling a new version of the code).
-    return RETRY;
-  }
-
-  // Otherwise, the content of the string might have moved. It must still
-  // be a sequential or external string with the same content.
-  // Update the start and end pointers in the stack frame to the current
-  // location (whether it has actually moved or not).
-  DCHECK(StringShape(*subject_tmp).IsSequential() ||
-      StringShape(*subject_tmp).IsExternal());
-
-  // The original start address of the characters to match.
-  const byte* start_address = frame_entry<const byte*>(re_frame, kInputStart);
-
-  // Find the current start address of the same character at the current string
-  // position.
-  int start_index = frame_entry<int>(re_frame, kStartIndex);
-  const byte* new_address = StringCharacterPosition(*subject_tmp,
-                                                    start_index + slice_offset);
-
-  if (start_address != new_address) {
-    // If there is a difference, update the object pointer and start and end
-    // addresses in the RegExp stack frame to match the new value.
-    const byte* end_address = frame_entry<const byte* >(re_frame, kInputEnd);
-    int byte_length = static_cast<int>(end_address - start_address);
-    frame_entry<const String*>(re_frame, kInputString) = *subject;
-    frame_entry<const byte*>(re_frame, kInputStart) = new_address;
-    frame_entry<const byte*>(re_frame, kInputEnd) = new_address + byte_length;
-  } else if (frame_entry<const String*>(re_frame, kInputString) != *subject) {
-    // Subject string might have been a ConsString that underwent
-    // short-circuiting during GC. That will not change start_address but
-    // will change pointer inside the subject handle.
-    frame_entry<const String*>(re_frame, kInputString) = *subject;
-  }
-
-  return 0;
+  return NativeRegExpMacroAssembler::CheckStackGuardState(
+      frame_entry<Isolate*>(re_frame, kIsolate),
+      frame_entry<int>(re_frame, kStartIndex),
+      frame_entry<int>(re_frame, kDirectCall) == 1, return_address, re_code,
+      frame_entry_address<String*>(re_frame, kInputString),
+      frame_entry_address<const byte*>(re_frame, kInputStart),
+      frame_entry_address<const byte*>(re_frame, kInputEnd));
 }
 
 
 Operand RegExpMacroAssemblerX64::register_location(int register_index) {
   DCHECK(register_index < (1<<30));
   if (num_registers_ <= register_index) {
     num_registers_ = register_index + 1;
   }
   return Operand(rbp, kRegisterZero - register_index * kPointerSize);
 }
@@ skipping 147 matching lines @@
   }
 }
 
 #undef __
 
 #endif  // V8_INTERPRETED_REGEXP
 
 }}  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_X64