src/jni_internal_arm.cc

/*
 * Copyright (C) 2011 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <stdint.h>

#include <algorithm>

#include "asm_support.h"
#include "assembler.h"
#include "compiled_method.h"
#include "compiler.h"
#include "object.h"

namespace art {
namespace arm {

// Creates a function which invokes a managed method with an array of
// arguments.
//
// At the time of call, the environment looks something like this:
//
// R0 = method pointer
// R1 = receiver pointer or NULL for static methods
// R2 = (managed) thread pointer
// R3 = argument array or NULL for no argument methods
// [SP] = JValue* result or NULL for void returns
//
// As the JNI call has already transitioned the thread into the
// "running" state the remaining responsibilities of this routine are
// to save the native register value and restore the managed thread
// register and transfer arguments from the array into register and on
// the stack, if needed.  On return, the thread register must be
// shuffled and the return value must be store into the result JValue.
CompiledInvokeStub* CreateInvokeStub(bool is_static, const char* shorty, uint32_t shorty_len) {
  UniquePtr<ArmAssembler> assembler(down_cast<ArmAssembler*>(Assembler::Create(kArm)));
#define __ assembler->
  size_t num_arg_array_bytes = NumArgArrayBytes(shorty, shorty_len);
  // Size of frame = spill of R4,R9/LR + Method* + possible receiver + arg array size
  // Note, space is left in the frame to flush arguments in registers back to out locations.
  size_t unpadded_frame_size = (4 * kPointerSize) +
                               (is_static ? 0 : kPointerSize) +
                               num_arg_array_bytes;
  size_t frame_size = RoundUp(unpadded_frame_size, kStackAlignment);

  // Spill R4,R9 and LR
  RegList save = (1 << R9) | (1 << R4);
  __ PushList(save | (1 << LR));

  // Move the managed thread pointer into R9.
  __ mov(R9, ShifterOperand(R2));

  // Reset R4 to suspend check interval
  __ LoadImmediate(R4, SUSPEND_CHECK_INTERVAL);

  // Move frame down for arguments less 3 pushed values above
  __ AddConstant(SP, -frame_size + (3 * kPointerSize));

  // Can either get 3 or 2 arguments into registers
  size_t reg_bytes = (is_static ? 3 : 2) * kPointerSize;
  // Bytes passed by stack
  size_t stack_bytes;
  if (num_arg_array_bytes > reg_bytes) {
    stack_bytes = num_arg_array_bytes - reg_bytes;
  } else {
    stack_bytes = 0;
    reg_bytes = num_arg_array_bytes;
  }

  // Method* at bottom of frame is null thereby terminating managed stack crawls
  __ LoadImmediate(IP, 0, AL);
  __ StoreToOffset(kStoreWord, IP, SP, 0);

  // Copy values onto the stack.
  size_t src_offset = 0;
  size_t dst_offset = (is_static ? 1 : 2) * kPointerSize;
  for (size_t i = 1; i < shorty_len; ++i) {
    switch (shorty[i]) {
      case 'D':
      case 'J':
        // Move both pointers 64 bits.
        __ LoadFromOffset(kLoadWord, IP, R3, src_offset);
        src_offset += kPointerSize;
        __ StoreToOffset(kStoreWord, IP, SP, dst_offset);
        dst_offset += kPointerSize;

        __ LoadFromOffset(kLoadWord, IP, R3, src_offset);
        src_offset += kPointerSize;
        __ StoreToOffset(kStoreWord, IP, SP, dst_offset);
        dst_offset += kPointerSize;
        break;
      default:
        // Move the source pointer sizeof(JValue) and the destination pointer 32 bits.
        __ LoadFromOffset(kLoadWord, IP, R3, src_offset);
        src_offset += sizeof(JValue);
        __ StoreToOffset(kStoreWord, IP, SP, dst_offset);
        dst_offset += kPointerSize;
        break;
    }
  }

  // Move all the register arguments into place.
  dst_offset = (is_static ? 1 : 2) * kPointerSize;
  if (is_static) {
    if (reg_bytes > 0 && num_arg_array_bytes > 0) {
      __ LoadFromOffset(kLoadWord, R1, SP, dst_offset + 0);
      if (reg_bytes > 4 && num_arg_array_bytes > 4) {
        __ LoadFromOffset(kLoadWord, R2, SP, dst_offset + 4);
        if (reg_bytes > 8 && num_arg_array_bytes > 8) {
          __ LoadFromOffset(kLoadWord, R3, SP, dst_offset + 8);
        }
      }
    }
  } else {
    if (reg_bytes > 0 && num_arg_array_bytes > 0) {
      __ LoadFromOffset(kLoadWord, R2, SP, dst_offset + 0);
      if (reg_bytes > 4 && num_arg_array_bytes > 4) {
        __ LoadFromOffset(kLoadWord, R3, SP, dst_offset + 4);
      }
    }
  }

  // Load the code pointer we are about to call.
  __ LoadFromOffset(kLoadWord, IP, R0, Method::GetCodeOffset().Int32Value());

  // Do the call.
  __ blx(IP);

  // If the method returns a value, store it to the result pointer.
  if (shorty[0] != 'V') {
    // Load the result JValue pointer of the stub caller's out args.
    __ LoadFromOffset(kLoadWord, IP, SP, frame_size);
    StoreOperandType type = (shorty[0] == 'J' || shorty[0] == 'D') ? kStoreWordPair : kStoreWord;
    __ StoreToOffset(type, R0, IP, 0);
  }

  // Remove the frame less the spilled R4, R9 and LR
  __ AddConstant(SP, frame_size - (3 * kPointerSize));

  // Pop R4, R9 and the LR into PC
  __ PopList(save | (1 << PC));
  // TODO: store native_entry in the stub table
  std::vector<uint8_t> code(assembler->CodeSize());
  MemoryRegion region(&code[0], code.size());
  assembler->FinalizeInstructions(region);
  return new CompiledInvokeStub(code);
#undef __
}

}  // namespace arm
}  // namespace art

extern "C" art::CompiledInvokeStub* ArtCreateInvokeStub(art::Compiler& compiler, bool is_static, const char* shorty, uint32_t shorty_len) {
  return art::arm::CreateInvokeStub(is_static, shorty, shorty_len);
}