Update constant pool correctly when resuming a generator

src/arm/full-codegen-arm.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 2176 matching lines @@
   __ ldr(r3, FieldMemOperand(r3, FixedArray::kLengthOffset));
   __ SmiUntag(r3);
 
   // If we are sending a value and there is no operand stack, we can jump back
   // in directly.
   if (resume_mode == JSGeneratorObject::NEXT) {
     Label slow_resume;
     __ cmp(r3, Operand(0));
     __ b(ne, &slow_resume);
     __ ldr(r3, FieldMemOperand(r4, JSFunction::kCodeEntryOffset));
+
+    ConstantPoolUnavailableScope constant_pool_unavailable(masm_);

[ulan, 2014/03/11 13:07:39]

Shouldn't this scope be
- at the beginning of the "if (resume_mode == JSGeneratorObject::NEXT)" branch?
- or inside the "if (FLAG_enable_ool_constant_pool)" branch?

[rmcilroy, 2014/03/11 16:14:53]

The Constant Pool is unavailable (for this code object) from the ldr(pp, ...) 
until the Jump(r3) instruction (since between these instructions it contains the
pp for the target function).  

I can't put it inside the "if (FLAG_enable_ool_constant_pool) because the scope
would go out of scope before the Jump(r3).  I could put it at the beginning of
the if (resume_mode == JSGeneratorObject::NEXT) scope if you prefer, but I put
it here to make it clear that this was related to the ldr(pp, ..) instruction. 
If you still prefer moving it to the beginning of the scope though I'm happy to
move it. WDYT?

[ulan, 2014/03/11 16:19:57]

I see, thanks for explaining! What about putting instructions from ldr(pp, ...) 
until the Jump(r3) into a new block { }? That would make clear when exactly the
scope is active.

[rmcilroy, 2014/03/11 16:42:23]

Good call! Done.

+    if (FLAG_enable_ool_constant_pool) {
+      // Load the new code object's constant pool pointer.
+      __ ldr(pp, MemOperand(r3, Code::kConstantPoolOffset - Code::kHeaderSize));
+    }
+
     __ ldr(r2, FieldMemOperand(r1, JSGeneratorObject::kContinuationOffset));
     __ SmiUntag(r2);
     __ add(r3, r3, r2);
     __ mov(r2, Operand(Smi::FromInt(JSGeneratorObject::kGeneratorExecuting)));
     __ str(r2, FieldMemOperand(r1, JSGeneratorObject::kContinuationOffset));
     __ Jump(r3);
+
     __ bind(&slow_resume);
   }
 
   // Otherwise, we push holes for the operand stack and call the runtime to fix
   // up the stack and the handlers.
   Label push_operand_holes, call_resume;
   __ bind(&push_operand_holes);
   __ sub(r3, r3, Operand(1), SetCC);
   __ b(mi, &call_resume);
   __ push(r2);
@@ skipping 2704 matching lines @@
   ASSERT(Memory::uint32_at(interrupt_address_pointer) ==
          reinterpret_cast<uint32_t>(
              isolate->builtins()->OsrAfterStackCheck()->entry()));
   return OSR_AFTER_STACK_CHECK;
 }
 
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_ARM