/*

 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.

 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.

 *

 * This code is free software; you can redistribute it and/or modify it

 * under the terms of the GNU General Public License version 2 only, as

 * published by the Free Software Foundation.

 *

 * This code is distributed in the hope that it will be useful, but WITHOUT

 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or

 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License

 * version 2 for more details (a copy is included in the LICENSE file that

 * accompanied this code).

 *

 * You should have received a copy of the GNU General Public License version

 * 2 along with this work; if not, write to the Free Software Foundation,

 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.

 *

 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA

 * or visit www.oracle.com if you need additional information or have any

 * questions.

 *

 */

#include "incls/_precompiled.incl"

#include "incls/_methodHandles_x86.cpp.incl"

#define __ _masm->

#ifdef PRODUCT

#define BLOCK_COMMENT(str) /* nothing */

#else

#define BLOCK_COMMENT(str) __ block_comment(str)

#endif

#define BIND(label) bind(label); BLOCK_COMMENT(#label ":")

address MethodHandleEntry::start_compiled_entry(MacroAssembler* _masm,

                                                address interpreted_entry) {

  // Just before the actual machine code entry point, allocate space

  // for a MethodHandleEntry::Data record, so that we can manage everything

  // from one base pointer.

  __ align(wordSize);

  address target = __ pc() + sizeof(Data);

  while (__ pc() < target) {

    __ nop();

    __ align(wordSize);

  }

  MethodHandleEntry* me = (MethodHandleEntry*) __ pc();

  me->set_end_address(__ pc());         // set a temporary end_address

  me->set_from_interpreted_entry(interpreted_entry);

  me->set_type_checking_entry(NULL);

  return (address) me;

}

MethodHandleEntry* MethodHandleEntry::finish_compiled_entry(MacroAssembler* _masm,

                                                address start_addr) {

  MethodHandleEntry* me = (MethodHandleEntry*) start_addr;

  assert(me->end_address() == start_addr, "valid ME");

  // Fill in the real end_address:

  __ align(wordSize);

  me->set_end_address(__ pc());

  return me;

}

#ifdef ASSERT

static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,

                           const char* error_message) {

  // Verify that argslot lies within (rsp, rbp].

  Label L_ok, L_bad;

  BLOCK_COMMENT("{ verify_argslot");

  __ cmpptr(argslot_reg, rbp);

  __ jccb(Assembler::above, L_bad);

  __ cmpptr(rsp, argslot_reg);

  __ jccb(Assembler::below, L_ok);

  __ bind(L_bad);

  __ stop(error_message);

  __ bind(L_ok);

  BLOCK_COMMENT("} verify_argslot");

}

#endif

// Code generation

address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {

  // rbx: methodOop

  // rcx: receiver method handle (must load from sp[MethodTypeForm.vmslots])

  // rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)

  // rdx, rdi: garbage temp, blown away

  Register rbx_method = rbx;

  Register rcx_recv   = rcx;

  Register rax_mtype  = rax;

  Register rdx_temp   = rdx;

  Register rdi_temp   = rdi;

  // emit WrongMethodType path first, to enable jccb back-branch from main path

  Label wrong_method_type;

  __ bind(wrong_method_type);

  Label invoke_generic_slow_path;

  assert(methodOopDesc::intrinsic_id_size_in_bytes() == sizeof(u1), "");;

  __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeExact);

  __ jcc(Assembler::notEqual, invoke_generic_slow_path);

  __ push(rax_mtype);       // required mtype

  __ push(rcx_recv);        // bad mh (1st stacked argument)

  __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));

  // here's where control starts out:

  __ align(CodeEntryAlignment);

  address entry_point = __ pc();

  // fetch the MethodType from the method handle into rax (the 'check' register)

  {

    Register tem = rbx_method;

    for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {

      __ movptr(rax_mtype, Address(tem, *pchase));

      tem = rax_mtype;          // in case there is another indirection

    }

  }

  // given the MethodType, find out where the MH argument is buried

  __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp)));

  Register rdx_vmslots = rdx_temp;

  __ movl(rdx_vmslots, Address(rdx_temp, __ delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp)));

  __ movptr(rcx_recv, __ argument_address(rdx_vmslots));

  trace_method_handle(_masm, "invokeExact");

  __ check_method_handle_type(rax_mtype, rcx_recv, rdi_temp, wrong_method_type);

  __ jump_to_method_handle_entry(rcx_recv, rdi_temp);

  // for invokeGeneric (only), apply argument and result conversions on the fly

  __ bind(invoke_generic_slow_path);

#ifdef ASSERT

  { Label L;

    __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeGeneric);

    __ jcc(Assembler::equal, L);

    __ stop("bad methodOop::intrinsic_id");

    __ bind(L);

  }

#endif //ASSERT

  Register rbx_temp = rbx_method;  // don't need it now

  // make room on the stack for another pointer:

  Register rcx_argslot = rcx_recv;

  __ lea(rcx_argslot, __ argument_address(rdx_vmslots, 1));

  insert_arg_slots(_masm, 2 * stack_move_unit(), _INSERT_REF_MASK,

                   rcx_argslot, rbx_temp, rdx_temp);

  // load up an adapter from the calling type (Java weaves this)

  __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_dyn_MethodType::form_offset_in_bytes, rdi_temp)));

  Register rdx_adapter = rdx_temp;

  // __ load_heap_oop(rdx_adapter, Address(rdx_temp, java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes()));

  // deal with old JDK versions:

  __ lea(rdi_temp, Address(rdx_temp, __ delayed_value(java_dyn_MethodTypeForm::genericInvoker_offset_in_bytes, rdi_temp)));

  __ cmpptr(rdi_temp, rdx_temp);

  Label sorry_no_invoke_generic;

  __ jcc(Assembler::below, sorry_no_invoke_generic);

  __ load_heap_oop(rdx_adapter, Address(rdi_temp, 0));

  __ testptr(rdx_adapter, rdx_adapter);

  __ jcc(Assembler::zero, sorry_no_invoke_generic);

  __ movptr(Address(rcx_argslot, 1 * Interpreter::stackElementSize), rdx_adapter);

  // As a trusted first argument, pass the type being called, so the adapter knows

  // the actual types of the arguments and return values.

  // (Generic invokers are shared among form-families of method-type.)

  __ movptr(Address(rcx_argslot, 0 * Interpreter::stackElementSize), rax_mtype);

  // FIXME: assert that rdx_adapter is of the right method-type.

  __ mov(rcx, rdx_adapter);

  trace_method_handle(_masm, "invokeGeneric");

  __ jump_to_method_handle_entry(rcx, rdi_temp);

  __ bind(sorry_no_invoke_generic); // no invokeGeneric implementation available!

  __ movptr(rcx_recv, Address(rcx_argslot, -1 * Interpreter::stackElementSize));  // recover original MH

  __ push(rax_mtype);       // required mtype

  __ push(rcx_recv);        // bad mh (1st stacked argument)

  __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));

  return entry_point;

}

// Helper to insert argument slots into the stack.

// arg_slots must be a multiple of stack_move_unit() and <= 0

void MethodHandles::insert_arg_slots(MacroAssembler* _masm,

                                     RegisterOrConstant arg_slots,

                                     int arg_mask,

                                     Register rax_argslot,

                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {

  assert(temp3_reg == noreg, "temp3 not required");

  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,

                             (!arg_slots.is_register() ? rsp : arg_slots.as_register()));

#ifdef ASSERT

  verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");

  if (arg_slots.is_register()) {

    Label L_ok, L_bad;

    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);

    __ jccb(Assembler::greater, L_bad);

    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);

    __ jccb(Assembler::zero, L_ok);

    __ bind(L_bad);

    __ stop("assert arg_slots <= 0 and clear low bits");

    __ bind(L_ok);

  } else {

    assert(arg_slots.as_constant() <= 0, "");

    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");

  }

#endif //ASSERT

#ifdef _LP64

  if (arg_slots.is_register()) {

    // clean high bits of stack motion register (was loaded as an int)

    __ movslq(arg_slots.as_register(), arg_slots.as_register());

  }

#endif

  // Make space on the stack for the inserted argument(s).

  // Then pull down everything shallower than rax_argslot.

  // The stacked return address gets pulled down with everything else.

  // That is, copy [rsp, argslot) downward by -size words.  In pseudo-code:

  //   rsp -= size;

  //   for (rdx = rsp + size; rdx < argslot; rdx++)

  //     rdx[-size] = rdx[0]

  //   argslot -= size;

  BLOCK_COMMENT("insert_arg_slots {");

  __ mov(rdx_temp, rsp);                        // source pointer for copy

  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));

  {

    Label loop;

    __ BIND(loop);

    // pull one word down each time through the loop

    __ movptr(rbx_temp, Address(rdx_temp, 0));

    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);

    __ addptr(rdx_temp, wordSize);

    __ cmpptr(rdx_temp, rax_argslot);

    __ jccb(Assembler::less, loop);

  }

  // Now move the argslot down, to point to the opened-up space.

  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));

  BLOCK_COMMENT("} insert_arg_slots");

}

// Helper to remove argument slots from the stack.

// arg_slots must be a multiple of stack_move_unit() and >= 0

void MethodHandles::remove_arg_slots(MacroAssembler* _masm,

                                    RegisterOrConstant arg_slots,

                                    Register rax_argslot,

                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {

  assert(temp3_reg == noreg, "temp3 not required");

  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,

                             (!arg_slots.is_register() ? rsp : arg_slots.as_register()));

#ifdef ASSERT

  // Verify that [argslot..argslot+size) lies within (rsp, rbp).

  __ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr));

  verify_argslot(_masm, rbx_temp, "deleted argument(s) must fall within current frame");

  if (arg_slots.is_register()) {

    Label L_ok, L_bad;

    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);

    __ jccb(Assembler::less, L_bad);

    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);

    __ jccb(Assembler::zero, L_ok);

    __ bind(L_bad);

    __ stop("assert arg_slots >= 0 and clear low bits");

    __ bind(L_ok);

  } else {

    assert(arg_slots.as_constant() >= 0, "");

    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");

  }

#endif //ASSERT

#ifdef _LP64

  if (false) {                  // not needed, since register is positive

    // clean high bits of stack motion register (was loaded as an int)

    if (arg_slots.is_register())

      __ movslq(arg_slots.as_register(), arg_slots.as_register());

  }

#endif

  BLOCK_COMMENT("remove_arg_slots {");

  // Pull up everything shallower than rax_argslot.

  // Then remove the excess space on the stack.

  // The stacked return address gets pulled up with everything else.

  // That is, copy [rsp, argslot) upward by size words.  In pseudo-code:

  //   for (rdx = argslot-1; rdx >= rsp; --rdx)

  //     rdx[size] = rdx[0]

  //   argslot += size;

  //   rsp += size;

  __ lea(rdx_temp, Address(rax_argslot, -wordSize)); // source pointer for copy

  {

    Label loop;

    __ BIND(loop);

    // pull one word up each time through the loop

    __ movptr(rbx_temp, Address(rdx_temp, 0));

    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);

    __ addptr(rdx_temp, -wordSize);

    __ cmpptr(rdx_temp, rsp);

    __ jccb(Assembler::greaterEqual, loop);

  }

  // Now move the argslot up, to point to the just-copied block.

  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));

  // And adjust the argslot address to point at the deletion point.

  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));

  BLOCK_COMMENT("} remove_arg_slots");

}

#ifndef PRODUCT

extern "C" void print_method_handle(oop mh);

void trace_method_handle_stub(const char* adaptername,

                              oop mh,

                              intptr_t* saved_regs,

                              intptr_t* entry_sp,

                              intptr_t* saved_sp,

                              intptr_t* saved_bp) {

  // called as a leaf from native code: do not block the JVM!

  intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];

  intptr_t* base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset];

  printf("MH %s mh="INTPTR_FORMAT" sp=("INTPTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="INTPTR_FORMAT"\n",

         adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp);

  if (last_sp != saved_sp && last_sp != NULL)

    printf("*** last_sp="INTPTR_FORMAT"\n", (intptr_t)last_sp);

  if (Verbose) {

    printf(" reg dump: ");

    int saved_regs_count = (entry_sp-1) - saved_regs;

    // 32 bit: rdi rsi rbp rsp; rbx rdx rcx (*) rax

    int i;

    for (i = 0; i <= saved_regs_count; i++) {

      if (i > 0 && i % 4 == 0 && i != saved_regs_count)

        printf("\n   + dump: ");

      printf(" %d: "INTPTR_FORMAT, i, saved_regs[i]);

    }

    printf("\n");

    int stack_dump_count = 16;

    if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))

      stack_dump_count = (int)(saved_bp + 2 - saved_sp);

    if (stack_dump_count > 64)  stack_dump_count = 48;

    for (i = 0; i < stack_dump_count; i += 4) {

      printf(" dump at SP[%d] "INTPTR_FORMAT": "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT"\n",

             i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);

    }

    print_method_handle(mh);

  }

}

void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {

  if (!TraceMethodHandles)  return;

  BLOCK_COMMENT("trace_method_handle {");

  __ push(rax);

  __ lea(rax, Address(rsp, wordSize*6)); // entry_sp

  __ pusha();

  // arguments:

  __ push(rbp);               // interpreter frame pointer

  __ push(rsi);               // saved_sp

  __ push(rax);               // entry_sp

  __ push(rcx);               // mh

  __ push(rcx);

  __ movptr(Address(rsp, 0), (intptr_t) adaptername);

  __ call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), 5);

  __ popa();

  __ pop(rax);

  BLOCK_COMMENT("} trace_method_handle");

}

#endif //PRODUCT

// which conversion op types are implemented here?

int MethodHandles::adapter_conversion_ops_supported_mask() {

  return ((1<<sun_dyn_AdapterMethodHandle::OP_RETYPE_ONLY)

         |(1<<sun_dyn_AdapterMethodHandle::OP_RETYPE_RAW)

         |(1<<sun_dyn_AdapterMethodHandle::OP_CHECK_CAST)

         |(1<<sun_dyn_AdapterMethodHandle::OP_PRIM_TO_PRIM)

         |(1<<sun_dyn_AdapterMethodHandle::OP_REF_TO_PRIM)

         |(1<<sun_dyn_AdapterMethodHandle::OP_SWAP_ARGS)

         |(1<<sun_dyn_AdapterMethodHandle::OP_ROT_ARGS)

         |(1<<sun_dyn_AdapterMethodHandle::OP_DUP_ARGS)

         |(1<<sun_dyn_AdapterMethodHandle::OP_DROP_ARGS)

         //|(1<<sun_dyn_AdapterMethodHandle::OP_SPREAD_ARGS) //BUG!

         );

  // FIXME: MethodHandlesTest gets a crash if we enable OP_SPREAD_ARGS.

}

// Generate an "entry" field for a method handle.

// This determines how the method handle will respond to calls.

void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {

  // Here is the register state during an interpreted call,

  // as set up by generate_method_handle_interpreter_entry():

  // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)

  // - rcx: receiver method handle

  // - rax: method handle type (only used by the check_mtype entry point)

  // - rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)

  // - rdx: garbage temp, can blow away

  Register rcx_recv    = rcx;

  Register rax_argslot = rax;

  Register rbx_temp    = rbx;

  Register rdx_temp    = rdx;

  // This guy is set up by prepare_to_jump_from_interpreted (from interpreted calls)

  // and gen_c2i_adapter (from compiled calls):

  Register saved_last_sp = LP64_ONLY(r13) NOT_LP64(rsi);

  guarantee(java_dyn_MethodHandle::vmentry_offset_in_bytes() != 0, "must have offsets");

  // some handy addresses

  Address rbx_method_fie(     rbx,      methodOopDesc::from_interpreted_offset() );

  Address rcx_mh_vmtarget(    rcx_recv, java_dyn_MethodHandle::vmtarget_offset_in_bytes() );

  Address rcx_dmh_vmindex(    rcx_recv, sun_dyn_DirectMethodHandle::vmindex_offset_in_bytes() );

  Address rcx_bmh_vmargslot(  rcx_recv, sun_dyn_BoundMethodHandle::vmargslot_offset_in_bytes() );

  Address rcx_bmh_argument(   rcx_recv, sun_dyn_BoundMethodHandle::argument_offset_in_bytes() );

  Address rcx_amh_vmargslot(  rcx_recv, sun_dyn_AdapterMethodHandle::vmargslot_offset_in_bytes() );

  Address rcx_amh_argument(   rcx_recv, sun_dyn_AdapterMethodHandle::argument_offset_in_bytes() );

  Address rcx_amh_conversion( rcx_recv, sun_dyn_AdapterMethodHandle::conversion_offset_in_bytes() );

  Address vmarg;                // __ argument_address(vmargslot)

  const int java_mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();

  if (have_entry(ek)) {

    __ nop();                   // empty stubs make SG sick

    return;

  }

  address interp_entry = __ pc();

  trace_method_handle(_masm, entry_name(ek));

  BLOCK_COMMENT(entry_name(ek));

  switch ((int) ek) {

  case _raise_exception:

    {

      // Not a real MH entry, but rather shared code for raising an exception.

      // Extra local arguments are pushed on stack, as required type at TOS+8,

      // failing object (or NULL) at TOS+4, failing bytecode type at TOS.

      // Beyond those local arguments are the PC, of course.

      Register rdx_code = rdx_temp;

      Register rcx_fail = rcx_recv;

      Register rax_want = rax_argslot;

      Register rdi_pc   = rdi;

      __ pop(rdx_code);  // TOS+0

      __ pop(rcx_fail);  // TOS+4

      __ pop(rax_want);  // TOS+8

      __ pop(rdi_pc);    // caller PC

      __ mov(rsp, rsi);   // cut the stack back to where the caller started

      // Repush the arguments as if coming from the interpreter.

      __ push(rdx_code);

      __ push(rcx_fail);

      __ push(rax_want);

      Register rbx_method = rbx_temp;

      Label no_method;

      // FIXME: fill in _raise_exception_method with a suitable sun.dyn method

      __ movptr(rbx_method, ExternalAddress((address) &_raise_exception_method));

      __ testptr(rbx_method, rbx_method);

      __ jccb(Assembler::zero, no_method);

      int jobject_oop_offset = 0;

      __ movptr(rbx_method, Address(rbx_method, jobject_oop_offset));  // dereference the jobject

      __ testptr(rbx_method, rbx_method);

      __ jccb(Assembler::zero, no_method);

      __ verify_oop(rbx_method);

      __ push(rdi_pc);          // and restore caller PC

      __ jmp(rbx_method_fie);

      // If we get here, the Java runtime did not do its job of creating the exception.

      // Do something that is at least causes a valid throw from the interpreter.

      __ bind(no_method);

      __ pop(rax_want);

      __ pop(rcx_fail);

      __ push(rax_want);

      __ push(rcx_fail);

      __ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));

    }

    break;

  case _invokestatic_mh:

  case _invokespecial_mh:

    {

      Register rbx_method = rbx_temp;

      __ load_heap_oop(rbx_method, rcx_mh_vmtarget); // target is a methodOop

      __ verify_oop(rbx_method);

      // same as TemplateTable::invokestatic or invokespecial,

      // minus the CP setup and profiling:

      if (ek == _invokespecial_mh) {

        // Must load & check the first argument before entering the target method.

        __ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);

        __ movptr(rcx_recv, __ argument_address(rax_argslot, -1));

        __ null_check(rcx_recv);

        __ verify_oop(rcx_recv);

      }

      __ jmp(rbx_method_fie);

    }

    break;

  case _invokevirtual_mh:

    {

      // same as TemplateTable::invokevirtual,

      // minus the CP setup and profiling:

      // pick out the vtable index and receiver offset from the MH,