1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_x86.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "logging/log.hpp"
  30 #include "oops/arrayOop.hpp"
  31 #include "oops/markOop.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/method.hpp"
  34 #include "oops/valueKlass.hpp"
  35 #include "prims/jvmtiExport.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/frame.inline.hpp"
  40 #include "runtime/safepointMechanism.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/thread.inline.hpp"
  43 
  44 // Implementation of InterpreterMacroAssembler
  45 
  46 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  47   assert(entry, "Entry must have been generated by now");
  48   jump(RuntimeAddress(entry));
  49 }
  50 
  51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
  52   Label update, next, none;
  53 
  54   verify_oop(obj);
  55 
  56   testptr(obj, obj);
  57   jccb(Assembler::notZero, update);
  58   orptr(mdo_addr, TypeEntries::null_seen);
  59   jmpb(next);
  60 
  61   bind(update);
  62   load_klass(obj, obj);
  63 
  64   xorptr(obj, mdo_addr);
  65   testptr(obj, TypeEntries::type_klass_mask);
  66   jccb(Assembler::zero, next); // klass seen before, nothing to
  67                                // do. The unknown bit may have been
  68                                // set already but no need to check.
  69 
  70   testptr(obj, TypeEntries::type_unknown);
  71   jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
  72 
  73   cmpptr(mdo_addr, 0);
  74   jccb(Assembler::equal, none);
  75   cmpptr(mdo_addr, TypeEntries::null_seen);
  76   jccb(Assembler::equal, none);
  77   // There is a chance that the checks above (re-reading profiling
  78   // data from memory) fail if another thread has just set the
  79   // profiling to this obj's klass
  80   xorptr(obj, mdo_addr);
  81   testptr(obj, TypeEntries::type_klass_mask);
  82   jccb(Assembler::zero, next);
  83 
  84   // different than before. Cannot keep accurate profile.
  85   orptr(mdo_addr, TypeEntries::type_unknown);
  86   jmpb(next);
  87 
  88   bind(none);
  89   // first time here. Set profile type.
  90   movptr(mdo_addr, obj);
  91 
  92   bind(next);
  93 }
  94 
  95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
  96   if (!ProfileInterpreter) {
  97     return;
  98   }
  99 
 100   if (MethodData::profile_arguments() || MethodData::profile_return()) {
 101     Label profile_continue;
 102 
 103     test_method_data_pointer(mdp, profile_continue);
 104 
 105     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
 106 
 107     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
 108     jcc(Assembler::notEqual, profile_continue);
 109 
 110     if (MethodData::profile_arguments()) {
 111       Label done;
 112       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
 113       addptr(mdp, off_to_args);
 114 
 115       for (int i = 0; i < TypeProfileArgsLimit; i++) {
 116         if (i > 0 || MethodData::profile_return()) {
 117           // If return value type is profiled we may have no argument to profile
 118           movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 119           subl(tmp, i*TypeStackSlotEntries::per_arg_count());
 120           cmpl(tmp, TypeStackSlotEntries::per_arg_count());
 121           jcc(Assembler::less, done);
 122         }
 123         movptr(tmp, Address(callee, Method::const_offset()));
 124         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
 125         // stack offset o (zero based) from the start of the argument
 126         // list, for n arguments translates into offset n - o - 1 from
 127         // the end of the argument list
 128         subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
 129         subl(tmp, 1);
 130         Address arg_addr = argument_address(tmp);
 131         movptr(tmp, arg_addr);
 132 
 133         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
 134         profile_obj_type(tmp, mdo_arg_addr);
 135 
 136         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
 137         addptr(mdp, to_add);
 138         off_to_args += to_add;
 139       }
 140 
 141       if (MethodData::profile_return()) {
 142         movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 143         subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
 144       }
 145 
 146       bind(done);
 147 
 148       if (MethodData::profile_return()) {
 149         // We're right after the type profile for the last
 150         // argument. tmp is the number of cells left in the
 151         // CallTypeData/VirtualCallTypeData to reach its end. Non null
 152         // if there's a return to profile.
 153         assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
 154         shll(tmp, exact_log2(DataLayout::cell_size));
 155         addptr(mdp, tmp);
 156       }
 157       movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
 158     } else {
 159       assert(MethodData::profile_return(), "either profile call args or call ret");
 160       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
 161     }
 162 
 163     // mdp points right after the end of the
 164     // CallTypeData/VirtualCallTypeData, right after the cells for the
 165     // return value type if there's one
 166 
 167     bind(profile_continue);
 168   }
 169 }
 170 
 171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
 172   assert_different_registers(mdp, ret, tmp, _bcp_register);
 173   if (ProfileInterpreter && MethodData::profile_return()) {
 174     Label profile_continue;
 175 
 176     test_method_data_pointer(mdp, profile_continue);
 177 
 178     if (MethodData::profile_return_jsr292_only()) {
 179       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
 180 
 181       // If we don't profile all invoke bytecodes we must make sure
 182       // it's a bytecode we indeed profile. We can't go back to the
 183       // begining of the ProfileData we intend to update to check its
 184       // type because we're right after it and we don't known its
 185       // length
 186       Label do_profile;
 187       cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
 188       jcc(Assembler::equal, do_profile);
 189       cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
 190       jcc(Assembler::equal, do_profile);
 191       get_method(tmp);
 192       cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
 193       jcc(Assembler::notEqual, profile_continue);
 194 
 195       bind(do_profile);
 196     }
 197 
 198     Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
 199     mov(tmp, ret);
 200     profile_obj_type(tmp, mdo_ret_addr);
 201 
 202     bind(profile_continue);
 203   }
 204 }
 205 
 206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
 207   if (ProfileInterpreter && MethodData::profile_parameters()) {
 208     Label profile_continue;
 209 
 210     test_method_data_pointer(mdp, profile_continue);
 211 
 212     // Load the offset of the area within the MDO used for
 213     // parameters. If it's negative we're not profiling any parameters
 214     movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
 215     testl(tmp1, tmp1);
 216     jcc(Assembler::negative, profile_continue);
 217 
 218     // Compute a pointer to the area for parameters from the offset
 219     // and move the pointer to the slot for the last
 220     // parameters. Collect profiling from last parameter down.
 221     // mdo start + parameters offset + array length - 1
 222     addptr(mdp, tmp1);
 223     movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
 224     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 225 
 226     Label loop;
 227     bind(loop);
 228 
 229     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
 230     int type_base = in_bytes(ParametersTypeData::type_offset(0));
 231     Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
 232     Address arg_off(mdp, tmp1, per_arg_scale, off_base);
 233     Address arg_type(mdp, tmp1, per_arg_scale, type_base);
 234 
 235     // load offset on the stack from the slot for this parameter
 236     movptr(tmp2, arg_off);
 237     negptr(tmp2);
 238     // read the parameter from the local area
 239     movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
 240 
 241     // profile the parameter
 242     profile_obj_type(tmp2, arg_type);
 243 
 244     // go to next parameter
 245     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 246     jcc(Assembler::positive, loop);
 247 
 248     bind(profile_continue);
 249   }
 250 }
 251 
 252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
 253                                                   int number_of_arguments) {
 254   // interpreter specific
 255   //
 256   // Note: No need to save/restore bcp & locals registers
 257   //       since these are callee saved registers and no blocking/
 258   //       GC can happen in leaf calls.
 259   // Further Note: DO NOT save/restore bcp/locals. If a caller has
 260   // already saved them so that it can use rsi/rdi as temporaries
 261   // then a save/restore here will DESTROY the copy the caller
 262   // saved! There used to be a save_bcp() that only happened in
 263   // the ASSERT path (no restore_bcp). Which caused bizarre failures
 264   // when jvm built with ASSERTs.
 265 #ifdef ASSERT
 266   {
 267     Label L;
 268     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 269     jcc(Assembler::equal, L);
 270     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 271          " last_sp != NULL");
 272     bind(L);
 273   }
 274 #endif
 275   // super call
 276   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 277   // interpreter specific
 278   // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
 279   // but since they may not have been saved (and we don't want to
 280   // save them here (see note above) the assert is invalid.
 281 }
 282 
 283 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
 284                                              Register java_thread,
 285                                              Register last_java_sp,
 286                                              address  entry_point,
 287                                              int      number_of_arguments,
 288                                              bool     check_exceptions) {
 289   // interpreter specific
 290   //
 291   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 292   //       really make a difference for these runtime calls, since they are
 293   //       slow anyway. Btw., bcp must be saved/restored since it may change
 294   //       due to GC.
 295   NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
 296   save_bcp();
 297 #ifdef ASSERT
 298   {
 299     Label L;
 300     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 301     jcc(Assembler::equal, L);
 302     stop("InterpreterMacroAssembler::call_VM_base:"
 303          " last_sp != NULL");
 304     bind(L);
 305   }
 306 #endif /* ASSERT */
 307   // super call
 308   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
 309                                entry_point, number_of_arguments,
 310                                check_exceptions);
 311   // interpreter specific
 312   restore_bcp();
 313   restore_locals();
 314 }
 315 
 316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
 317   if (JvmtiExport::can_pop_frame()) {
 318     Label L;
 319     // Initiate popframe handling only if it is not already being
 320     // processed.  If the flag has the popframe_processing bit set, it
 321     // means that this code is called *during* popframe handling - we
 322     // don't want to reenter.
 323     // This method is only called just after the call into the vm in
 324     // call_VM_base, so the arg registers are available.
 325     Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
 326                         LP64_ONLY(c_rarg0);
 327     movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
 328     testl(pop_cond, JavaThread::popframe_pending_bit);
 329     jcc(Assembler::zero, L);
 330     testl(pop_cond, JavaThread::popframe_processing_bit);
 331     jcc(Assembler::notZero, L);
 332     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 333     // address of the same-named entrypoint in the generated interpreter code.
 334     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 335     jmp(rax);
 336     bind(L);
 337     NOT_LP64(get_thread(java_thread);)
 338   }
 339 }
 340 
 341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 342   Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 343   NOT_LP64(get_thread(thread);)
 344   movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
 345   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 346   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 347   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 348 #ifdef _LP64
 349   switch (state) {
 350     case atos: movptr(rax, oop_addr);
 351                movptr(oop_addr, (int32_t)NULL_WORD);
 352                verify_oop(rax, state);              break;
 353     case ltos: movptr(rax, val_addr);                 break;
 354     case btos:                                   // fall through
 355     case ztos:                                   // fall through
 356     case ctos:                                   // fall through
 357     case stos:                                   // fall through
 358     case itos: movl(rax, val_addr);                 break;
 359     case ftos: load_float(val_addr);                break;
 360     case dtos: load_double(val_addr);               break;
 361     case vtos: /* nothing to do */                  break;
 362     default  : ShouldNotReachHere();
 363   }
 364   // Clean up tos value in the thread object
 365   movl(tos_addr,  (int) ilgl);
 366   movl(val_addr,  (int32_t) NULL_WORD);
 367 #else
 368   const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
 369                              + in_ByteSize(wordSize));
 370   switch (state) {
 371     case atos: movptr(rax, oop_addr);
 372                movptr(oop_addr, NULL_WORD);
 373                verify_oop(rax, state);                break;
 374     case ltos:
 375                movl(rdx, val_addr1);               // fall through
 376     case btos:                                     // fall through
 377     case ztos:                                     // fall through
 378     case ctos:                                     // fall through
 379     case stos:                                     // fall through
 380     case itos: movl(rax, val_addr);                   break;
 381     case ftos: load_float(val_addr);                  break;
 382     case dtos: load_double(val_addr);                 break;
 383     case vtos: /* nothing to do */                    break;
 384     default  : ShouldNotReachHere();
 385   }
 386 #endif // _LP64
 387   // Clean up tos value in the thread object
 388   movl(tos_addr,  (int32_t) ilgl);
 389   movptr(val_addr,  NULL_WORD);
 390   NOT_LP64(movptr(val_addr1, NULL_WORD);)
 391 }
 392 
 393 
 394 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 395   if (JvmtiExport::can_force_early_return()) {
 396     Label L;
 397     Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
 398     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
 399 
 400     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 401     testptr(tmp, tmp);
 402     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
 403 
 404     // Initiate earlyret handling only if it is not already being processed.
 405     // If the flag has the earlyret_processing bit set, it means that this code
 406     // is called *during* earlyret handling - we don't want to reenter.
 407     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
 408     cmpl(tmp, JvmtiThreadState::earlyret_pending);
 409     jcc(Assembler::notEqual, L);
 410 
 411     // Call Interpreter::remove_activation_early_entry() to get the address of the
 412     // same-named entrypoint in the generated interpreter code.
 413     NOT_LP64(get_thread(java_thread);)
 414     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 415 #ifdef _LP64
 416     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 417     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
 418 #else
 419     pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 420     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
 421 #endif // _LP64
 422     jmp(rax);
 423     bind(L);
 424     NOT_LP64(get_thread(java_thread);)
 425   }
 426 }
 427 
 428 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 429   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 430   load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
 431   bswapl(reg);
 432   shrl(reg, 16);
 433 }
 434 
 435 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 436                                                        int bcp_offset,
 437                                                        size_t index_size) {
 438   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 439   if (index_size == sizeof(u2)) {
 440     load_unsigned_short(index, Address(_bcp_register, bcp_offset));
 441   } else if (index_size == sizeof(u4)) {
 442     movl(index, Address(_bcp_register, bcp_offset));
 443     // Check if the secondary index definition is still ~x, otherwise
 444     // we have to change the following assembler code to calculate the
 445     // plain index.
 446     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 447     notl(index);  // convert to plain index
 448   } else if (index_size == sizeof(u1)) {
 449     load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
 450   } else {
 451     ShouldNotReachHere();
 452   }
 453 }
 454 
 455 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 456                                                            Register index,
 457                                                            int bcp_offset,
 458                                                            size_t index_size) {
 459   assert_different_registers(cache, index);
 460   get_cache_index_at_bcp(index, bcp_offset, index_size);
 461   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 462   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 463   // convert from field index to ConstantPoolCacheEntry index
 464   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
 465   shll(index, 2);
 466 }
 467 
 468 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 469                                                                         Register index,
 470                                                                         Register bytecode,
 471                                                                         int byte_no,
 472                                                                         int bcp_offset,
 473                                                                         size_t index_size) {
 474   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
 475   // We use a 32-bit load here since the layout of 64-bit words on
 476   // little-endian machines allow us that.
 477   movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 478   const int shift_count = (1 + byte_no) * BitsPerByte;
 479   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 480          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 481          "correct shift count");
 482   shrl(bytecode, shift_count);
 483   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 484   andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
 485 }
 486 
 487 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 488                                                                Register tmp,
 489                                                                int bcp_offset,
 490                                                                size_t index_size) {
 491   assert(cache != tmp, "must use different register");
 492   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 493   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 494   // convert from field index to ConstantPoolCacheEntry index
 495   // and from word offset to byte offset
 496   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
 497   shll(tmp, 2 + LogBytesPerWord);
 498   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 499   // skip past the header
 500   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
 501   addptr(cache, tmp);  // construct pointer to cache entry
 502 }
 503 
 504 // Load object from cpool->resolved_references(index)
 505 void InterpreterMacroAssembler::load_resolved_reference_at_index(
 506                                            Register result, Register index, Register tmp) {
 507   assert_different_registers(result, index);
 508 
 509   get_constant_pool(result);
 510   // load pointer for resolved_references[] objArray
 511   movptr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
 512   movptr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
 513   resolve_oop_handle(result, tmp);
 514   load_heap_oop(result, Address(result, index,
 515                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 516                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
 517 }
 518 
 519 // load cpool->resolved_klass_at(index)
 520 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register cpool,
 521                                            Register index, Register klass) {
 522   movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
 523   Register resolved_klasses = cpool;
 524   movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
 525   movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
 526 }
 527 
 528 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 529 // subtype of super_klass.
 530 //
 531 // Args:
 532 //      rax: superklass
 533 //      Rsub_klass: subklass
 534 //
 535 // Kills:
 536 //      rcx, rdi
 537 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 538                                                   Label& ok_is_subtype) {
 539   assert(Rsub_klass != rax, "rax holds superklass");
 540   LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
 541   LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
 542   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 543   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 544 
 545   // Profile the not-null value's klass.
 546   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 547 
 548   // Do the check.
 549   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 550 
 551   // Profile the failure of the check.
 552   profile_typecheck_failed(rcx); // blows rcx
 553 }
 554 
 555 
 556 #ifndef _LP64
 557 void InterpreterMacroAssembler::f2ieee() {
 558   if (IEEEPrecision) {
 559     fstp_s(Address(rsp, 0));
 560     fld_s(Address(rsp, 0));
 561   }
 562 }
 563 
 564 
 565 void InterpreterMacroAssembler::d2ieee() {
 566   if (IEEEPrecision) {
 567     fstp_d(Address(rsp, 0));
 568     fld_d(Address(rsp, 0));
 569   }
 570 }
 571 #endif // _LP64
 572 
 573 // Java Expression Stack
 574 
 575 void InterpreterMacroAssembler::pop_ptr(Register r) {
 576   pop(r);
 577 }
 578 
 579 void InterpreterMacroAssembler::push_ptr(Register r) {
 580   push(r);
 581 }
 582 
 583 void InterpreterMacroAssembler::push_i(Register r) {
 584   push(r);
 585 }
 586 
 587 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 588   subptr(rsp, wordSize);
 589   movflt(Address(rsp, 0), r);
 590 }
 591 
 592 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 593   movflt(r, Address(rsp, 0));
 594   addptr(rsp, wordSize);
 595 }
 596 
 597 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 598   subptr(rsp, 2 * wordSize);
 599   movdbl(Address(rsp, 0), r);
 600 }
 601 
 602 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 603   movdbl(r, Address(rsp, 0));
 604   addptr(rsp, 2 * Interpreter::stackElementSize);
 605 }
 606 
 607 #ifdef _LP64
 608 void InterpreterMacroAssembler::pop_i(Register r) {
 609   // XXX can't use pop currently, upper half non clean
 610   movl(r, Address(rsp, 0));
 611   addptr(rsp, wordSize);
 612 }
 613 
 614 void InterpreterMacroAssembler::pop_l(Register r) {
 615   movq(r, Address(rsp, 0));
 616   addptr(rsp, 2 * Interpreter::stackElementSize);
 617 }
 618 
 619 void InterpreterMacroAssembler::push_l(Register r) {
 620   subptr(rsp, 2 * wordSize);
 621   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
 622   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
 623 }
 624 
 625 void InterpreterMacroAssembler::pop(TosState state) {
 626   switch (state) {
 627   case atos: pop_ptr();                 break;
 628   case btos:
 629   case ztos:
 630   case ctos:
 631   case stos:
 632   case itos: pop_i();                   break;
 633   case ltos: pop_l();                   break;
 634   case ftos: pop_f(xmm0);               break;
 635   case dtos: pop_d(xmm0);               break;
 636   case vtos: /* nothing to do */        break;
 637   default:   ShouldNotReachHere();
 638   }
 639   verify_oop(rax, state);
 640 }
 641 
 642 void InterpreterMacroAssembler::push(TosState state) {
 643   verify_oop(rax, state);
 644   switch (state) {
 645   case atos: push_ptr();                break;
 646   case btos:
 647   case ztos:
 648   case ctos:
 649   case stos:
 650   case itos: push_i();                  break;
 651   case ltos: push_l();                  break;
 652   case ftos: push_f(xmm0);              break;
 653   case dtos: push_d(xmm0);              break;
 654   case vtos: /* nothing to do */        break;
 655   default  : ShouldNotReachHere();
 656   }
 657 }
 658 #else
 659 void InterpreterMacroAssembler::pop_i(Register r) {
 660   pop(r);
 661 }
 662 
 663 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
 664   pop(lo);
 665   pop(hi);
 666 }
 667 
 668 void InterpreterMacroAssembler::pop_f() {
 669   fld_s(Address(rsp, 0));
 670   addptr(rsp, 1 * wordSize);
 671 }
 672 
 673 void InterpreterMacroAssembler::pop_d() {
 674   fld_d(Address(rsp, 0));
 675   addptr(rsp, 2 * wordSize);
 676 }
 677 
 678 
 679 void InterpreterMacroAssembler::pop(TosState state) {
 680   switch (state) {
 681     case atos: pop_ptr(rax);                                 break;
 682     case btos:                                               // fall through
 683     case ztos:                                               // fall through
 684     case ctos:                                               // fall through
 685     case stos:                                               // fall through
 686     case itos: pop_i(rax);                                   break;
 687     case ltos: pop_l(rax, rdx);                              break;
 688     case ftos:
 689       if (UseSSE >= 1) {
 690         pop_f(xmm0);
 691       } else {
 692         pop_f();
 693       }
 694       break;
 695     case dtos:
 696       if (UseSSE >= 2) {
 697         pop_d(xmm0);
 698       } else {
 699         pop_d();
 700       }
 701       break;
 702     case vtos: /* nothing to do */                           break;
 703     default  : ShouldNotReachHere();
 704   }
 705   verify_oop(rax, state);
 706 }
 707 
 708 
 709 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
 710   push(hi);
 711   push(lo);
 712 }
 713 
 714 void InterpreterMacroAssembler::push_f() {
 715   // Do not schedule for no AGI! Never write beyond rsp!
 716   subptr(rsp, 1 * wordSize);
 717   fstp_s(Address(rsp, 0));
 718 }
 719 
 720 void InterpreterMacroAssembler::push_d() {
 721   // Do not schedule for no AGI! Never write beyond rsp!
 722   subptr(rsp, 2 * wordSize);
 723   fstp_d(Address(rsp, 0));
 724 }
 725 
 726 
 727 void InterpreterMacroAssembler::push(TosState state) {
 728   verify_oop(rax, state);
 729   switch (state) {
 730     case atos: push_ptr(rax); break;
 731     case btos:                                               // fall through
 732     case ztos:                                               // fall through
 733     case ctos:                                               // fall through
 734     case stos:                                               // fall through
 735     case itos: push_i(rax);                                    break;
 736     case ltos: push_l(rax, rdx);                               break;
 737     case ftos:
 738       if (UseSSE >= 1) {
 739         push_f(xmm0);
 740       } else {
 741         push_f();
 742       }
 743       break;
 744     case dtos:
 745       if (UseSSE >= 2) {
 746         push_d(xmm0);
 747       } else {
 748         push_d();
 749       }
 750       break;
 751     case vtos: /* nothing to do */                             break;
 752     default  : ShouldNotReachHere();
 753   }
 754 }
 755 #endif // _LP64
 756 
 757 
 758 // Helpers for swap and dup
 759 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 760   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 761 }
 762 
 763 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 764   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 765 }
 766 
 767 
 768 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 769   // set sender sp
 770   lea(_bcp_register, Address(rsp, wordSize));
 771   // record last_sp
 772   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
 773 }
 774 
 775 
 776 // Jump to from_interpreted entry of a call unless single stepping is possible
 777 // in this thread in which case we must call the i2i entry
 778 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 779   prepare_to_jump_from_interpreted();
 780 
 781   if (JvmtiExport::can_post_interpreter_events()) {
 782     Label run_compiled_code;
 783     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 784     // compiled code in threads for which the event is enabled.  Check here for
 785     // interp_only_mode if these events CAN be enabled.
 786     // interp_only is an int, on little endian it is sufficient to test the byte only
 787     // Is a cmpl faster?
 788     LP64_ONLY(temp = r15_thread;)
 789     NOT_LP64(get_thread(temp);)
 790     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
 791     jccb(Assembler::zero, run_compiled_code);
 792     jmp(Address(method, Method::interpreter_entry_offset()));
 793     bind(run_compiled_code);
 794   }
 795 
 796   jmp(Address(method, Method::from_interpreted_offset()));
 797 }
 798 
 799 // The following two routines provide a hook so that an implementation
 800 // can schedule the dispatch in two parts.  x86 does not do this.
 801 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 802   // Nothing x86 specific to be done here
 803 }
 804 
 805 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 806   dispatch_next(state, step);
 807 }
 808 
 809 void InterpreterMacroAssembler::dispatch_base(TosState state,
 810                                               address* table,
 811                                               bool verifyoop,
 812                                               bool generate_poll) {
 813   verify_FPU(1, state);
 814   if (VerifyActivationFrameSize) {
 815     Label L;
 816     mov(rcx, rbp);
 817     subptr(rcx, rsp);
 818     int32_t min_frame_size =
 819       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 820       wordSize;
 821     cmpptr(rcx, (int32_t)min_frame_size);
 822     jcc(Assembler::greaterEqual, L);
 823     stop("broken stack frame");
 824     bind(L);
 825   }
 826   if (verifyoop) {
 827     verify_oop(rax, state);
 828   }
 829 
 830   address* const safepoint_table = Interpreter::safept_table(state);
 831 #ifdef _LP64
 832   Label no_safepoint, dispatch;
 833   if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
 834     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 835     testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
 836 
 837     jccb(Assembler::zero, no_safepoint);
 838     lea(rscratch1, ExternalAddress((address)safepoint_table));
 839     jmpb(dispatch);
 840   }
 841 
 842   bind(no_safepoint);
 843   lea(rscratch1, ExternalAddress((address)table));
 844   bind(dispatch);
 845   jmp(Address(rscratch1, rbx, Address::times_8));
 846 
 847 #else
 848   Address index(noreg, rbx, Address::times_ptr);
 849   if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
 850     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 851     Label no_safepoint;
 852     const Register thread = rcx;
 853     get_thread(thread);
 854     testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
 855 
 856     jccb(Assembler::zero, no_safepoint);
 857     ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
 858     jump(dispatch_addr);
 859     bind(no_safepoint);
 860   }
 861 
 862   {
 863     ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
 864     jump(dispatch_addr);
 865   }
 866 #endif // _LP64
 867 }
 868 
 869 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 870   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 871 }
 872 
 873 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 874   dispatch_base(state, Interpreter::normal_table(state));
 875 }
 876 
 877 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 878   dispatch_base(state, Interpreter::normal_table(state), false);
 879 }
 880 
 881 
 882 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 883   // load next bytecode (load before advancing _bcp_register to prevent AGI)
 884   load_unsigned_byte(rbx, Address(_bcp_register, step));
 885   // advance _bcp_register
 886   increment(_bcp_register, step);
 887   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 888 }
 889 
 890 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 891   // load current bytecode
 892   load_unsigned_byte(rbx, Address(_bcp_register, 0));
 893   dispatch_base(state, table);
 894 }
 895 
 896 void InterpreterMacroAssembler::narrow(Register result) {
 897 
 898   // Get method->_constMethod->_result_type
 899   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 900   movptr(rcx, Address(rcx, Method::const_offset()));
 901   load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
 902 
 903   Label done, notBool, notByte, notChar;
 904 
 905   // common case first
 906   cmpl(rcx, T_INT);
 907   jcc(Assembler::equal, done);
 908 
 909   // mask integer result to narrower return type.
 910   cmpl(rcx, T_BOOLEAN);
 911   jcc(Assembler::notEqual, notBool);
 912   andl(result, 0x1);
 913   jmp(done);
 914 
 915   bind(notBool);
 916   cmpl(rcx, T_BYTE);
 917   jcc(Assembler::notEqual, notByte);
 918   LP64_ONLY(movsbl(result, result);)
 919   NOT_LP64(shll(result, 24);)      // truncate upper 24 bits
 920   NOT_LP64(sarl(result, 24);)      // and sign-extend byte
 921   jmp(done);
 922 
 923   bind(notByte);
 924   cmpl(rcx, T_CHAR);
 925   jcc(Assembler::notEqual, notChar);
 926   LP64_ONLY(movzwl(result, result);)
 927   NOT_LP64(andl(result, 0xFFFF);)  // truncate upper 16 bits
 928   jmp(done);
 929 
 930   bind(notChar);
 931   // cmpl(rcx, T_SHORT);  // all that's left
 932   // jcc(Assembler::notEqual, done);
 933   LP64_ONLY(movswl(result, result);)
 934   NOT_LP64(shll(result, 16);)      // truncate upper 16 bits
 935   NOT_LP64(sarl(result, 16);)      // and sign-extend short
 936 
 937   // Nothing to do for T_INT
 938   bind(done);
 939 }
 940 
 941 // remove activation
 942 //
 943 // Unlock the receiver if this is a synchronized method.
 944 // Unlock any Java monitors from syncronized blocks.
 945 // Remove the activation from the stack.
 946 //
 947 // If there are locked Java monitors
 948 //    If throw_monitor_exception
 949 //       throws IllegalMonitorStateException
 950 //    Else if install_monitor_exception
 951 //       installs IllegalMonitorStateException
 952 //    Else
 953 //       no error processing
 954 void InterpreterMacroAssembler::remove_activation(
 955         TosState state,
 956         Register ret_addr,
 957         bool throw_monitor_exception,
 958         bool install_monitor_exception,
 959         bool notify_jvmdi) {
 960   // Note: Registers rdx xmm0 may be in use for the
 961   // result check if synchronized method
 962   Label unlocked, unlock, no_unlock;
 963 
 964   const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 965   const Register robj    = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
 966   const Register rmon    = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
 967                               // monitor pointers need different register
 968                               // because rdx may have the result in it
 969   NOT_LP64(get_thread(rcx);)
 970 
 971   // get the value of _do_not_unlock_if_synchronized into rdx
 972   const Address do_not_unlock_if_synchronized(rthread,
 973     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 974   movbool(rbx, do_not_unlock_if_synchronized);
 975   movbool(do_not_unlock_if_synchronized, false); // reset the flag
 976 
 977   // get method access flags
 978   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 979   movl(rcx, Address(rcx, Method::access_flags_offset()));
 980   testl(rcx, JVM_ACC_SYNCHRONIZED);
 981   jcc(Assembler::zero, unlocked);
 982 
 983   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
 984   // is set.
 985   testbool(rbx);
 986   jcc(Assembler::notZero, no_unlock);
 987 
 988   // unlock monitor
 989   push(state); // save result
 990 
 991   // BasicObjectLock will be first in list, since this is a
 992   // synchronized method. However, need to check that the object has
 993   // not been unlocked by an explicit monitorexit bytecode.
 994   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
 995                         wordSize - (int) sizeof(BasicObjectLock));
 996   // We use c_rarg1/rdx so that if we go slow path it will be the correct
 997   // register for unlock_object to pass to VM directly
 998   lea(robj, monitor); // address of first monitor
 999 
1000   movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1001   testptr(rax, rax);
1002   jcc(Assembler::notZero, unlock);
1003 
1004   pop(state);
1005   if (throw_monitor_exception) {
1006     // Entry already unlocked, need to throw exception
1007     NOT_LP64(empty_FPU_stack();)  // remove possible return value from FPU-stack, otherwise stack could overflow
1008     call_VM(noreg, CAST_FROM_FN_PTR(address,
1009                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1010     should_not_reach_here();
1011   } else {
1012     // Monitor already unlocked during a stack unroll. If requested,
1013     // install an illegal_monitor_state_exception.  Continue with
1014     // stack unrolling.
1015     if (install_monitor_exception) {
1016       NOT_LP64(empty_FPU_stack();)
1017       call_VM(noreg, CAST_FROM_FN_PTR(address,
1018                      InterpreterRuntime::new_illegal_monitor_state_exception));
1019     }
1020     jmp(unlocked);
1021   }
1022 
1023   bind(unlock);
1024   unlock_object(robj);
1025   pop(state);
1026 
1027   // Check that for block-structured locking (i.e., that all locked
1028   // objects has been unlocked)
1029   bind(unlocked);
1030 
1031   // rax, rdx: Might contain return value
1032 
1033   // Check that all monitors are unlocked
1034   {
1035     Label loop, exception, entry, restart;
1036     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1037     const Address monitor_block_top(
1038         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1039     const Address monitor_block_bot(
1040         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1041 
1042     bind(restart);
1043     // We use c_rarg1 so that if we go slow path it will be the correct
1044     // register for unlock_object to pass to VM directly
1045     movptr(rmon, monitor_block_top); // points to current entry, starting
1046                                   // with top-most entry
1047     lea(rbx, monitor_block_bot);  // points to word before bottom of
1048                                   // monitor block
1049     jmp(entry);
1050 
1051     // Entry already locked, need to throw exception
1052     bind(exception);
1053 
1054     if (throw_monitor_exception) {
1055       // Throw exception
1056       NOT_LP64(empty_FPU_stack();)
1057       MacroAssembler::call_VM(noreg,
1058                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
1059                                    throw_illegal_monitor_state_exception));
1060       should_not_reach_here();
1061     } else {
1062       // Stack unrolling. Unlock object and install illegal_monitor_exception.
1063       // Unlock does not block, so don't have to worry about the frame.
1064       // We don't have to preserve c_rarg1 since we are going to throw an exception.
1065 
1066       push(state);
1067       mov(robj, rmon);   // nop if robj and rmon are the same
1068       unlock_object(robj);
1069       pop(state);
1070 
1071       if (install_monitor_exception) {
1072         NOT_LP64(empty_FPU_stack();)
1073         call_VM(noreg, CAST_FROM_FN_PTR(address,
1074                                         InterpreterRuntime::
1075                                         new_illegal_monitor_state_exception));
1076       }
1077 
1078       jmp(restart);
1079     }
1080 
1081     bind(loop);
1082     // check if current entry is used
1083     cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1084     jcc(Assembler::notEqual, exception);
1085 
1086     addptr(rmon, entry_size); // otherwise advance to next entry
1087     bind(entry);
1088     cmpptr(rmon, rbx); // check if bottom reached
1089     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1090   }
1091 
1092   bind(no_unlock);
1093 
1094   // jvmti support
1095   if (notify_jvmdi) {
1096     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
1097   } else {
1098     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1099   }
1100 
1101   if (StackReservedPages > 0) {
1102     movptr(rbx,
1103                Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1104     // testing if reserved zone needs to be re-enabled
1105     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1106     Label no_reserved_zone_enabling;
1107 
1108     NOT_LP64(get_thread(rthread);)
1109 
1110     cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1111     jcc(Assembler::equal, no_reserved_zone_enabling);
1112 
1113     cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1114     jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1115 
1116     call_VM_leaf(
1117       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1118     call_VM(noreg, CAST_FROM_FN_PTR(address,
1119                    InterpreterRuntime::throw_delayed_StackOverflowError));
1120     should_not_reach_here();
1121 
1122     bind(no_reserved_zone_enabling);
1123   }
1124 
1125   // remove activation
1126   // get sender sp
1127   movptr(rbx,
1128          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1129 
1130   if (state == atos && ValueTypeReturnedAsFields) {
1131     Label skip;
1132     // Test if the return type is a value type
1133     movptr(rdi, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1134     movptr(rdi, Address(rdi, Method::const_offset()));
1135     load_unsigned_byte(rdi, Address(rdi, ConstMethod::result_type_offset()));
1136     cmpl(rdi, T_VALUETYPE);
1137     jcc(Assembler::notEqual, skip);
1138 
1139     // We are returning a value type, load its fields into registers
1140 #ifndef _LP64
1141     super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs());
1142 #else
1143     // Load fields from a buffered value with a value class specific handler
1144     load_klass(rdi, rax);
1145     movptr(rdi, Address(rdi, InstanceKlass::adr_valueklass_fixed_block_offset()));
1146     movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset()));
1147 
1148     testptr(rdi, rdi);
1149     jcc(Assembler::equal, skip);
1150 
1151     call(rdi);
1152 #endif
1153     // call above kills the value in rbx. Reload it.
1154     movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1155     bind(skip);
1156   }
1157   leave();                           // remove frame anchor
1158   pop(ret_addr);                     // get return address
1159   mov(rsp, rbx);                     // set sp to sender sp
1160 }
1161 
1162 void InterpreterMacroAssembler::get_method_counters(Register method,
1163                                                     Register mcs, Label& skip) {
1164   Label has_counters;
1165   movptr(mcs, Address(method, Method::method_counters_offset()));
1166   testptr(mcs, mcs);
1167   jcc(Assembler::notZero, has_counters);
1168   call_VM(noreg, CAST_FROM_FN_PTR(address,
1169           InterpreterRuntime::build_method_counters), method);
1170   movptr(mcs, Address(method,Method::method_counters_offset()));
1171   testptr(mcs, mcs);
1172   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1173   bind(has_counters);
1174 }
1175 
1176 
1177 // Lock object
1178 //
1179 // Args:
1180 //      rdx, c_rarg1: BasicObjectLock to be used for locking
1181 //
1182 // Kills:
1183 //      rax, rbx
1184 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1185   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1186          "The argument is only for looks. It must be c_rarg1");
1187 
1188   if (UseHeavyMonitors) {
1189     call_VM(noreg,
1190             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1191             lock_reg);
1192   } else {
1193     Label done;
1194 
1195     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1196     const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1197                                   // problematic case where tmp_reg = no_reg.
1198     const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1199 
1200     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1201     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1202     const int mark_offset = lock_offset +
1203                             BasicLock::displaced_header_offset_in_bytes();
1204 
1205     Label slow_case;
1206 
1207     // Load object pointer into obj_reg
1208     movptr(obj_reg, Address(lock_reg, obj_offset));
1209 
1210     if (UseBiasedLocking) {
1211       biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1212     }
1213 
1214     // Load immediate 1 into swap_reg %rax
1215     movl(swap_reg, (int32_t)1);
1216 
1217     // Load (object->mark() | 1) into swap_reg %rax
1218     orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1219     if (EnableValhalla && !UseBiasedLocking) {
1220       // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking
1221       andptr(swap_reg, ~markOopDesc::biased_lock_bit_in_place);
1222     }
1223 
1224     // Save (object->mark() | 1) into BasicLock's displaced header
1225     movptr(Address(lock_reg, mark_offset), swap_reg);
1226 
1227     assert(lock_offset == 0,
1228            "displaced header must be first word in BasicObjectLock");
1229 
1230     lock();
1231     cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1232     if (PrintBiasedLockingStatistics) {
1233       cond_inc32(Assembler::zero,
1234                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1235     }
1236     jcc(Assembler::zero, done);
1237 
1238     const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1239 
1240     // Test if the oopMark is an obvious stack pointer, i.e.,
1241     //  1) (mark & zero_bits) == 0, and
1242     //  2) rsp <= mark < mark + os::pagesize()
1243     //
1244     // These 3 tests can be done by evaluating the following
1245     // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1246     // assuming both stack pointer and pagesize have their
1247     // least significant bits clear.
1248     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1249     subptr(swap_reg, rsp);
1250     andptr(swap_reg, zero_bits - os::vm_page_size());
1251 
1252     // Save the test result, for recursive case, the result is zero
1253     movptr(Address(lock_reg, mark_offset), swap_reg);
1254 
1255     if (PrintBiasedLockingStatistics) {
1256       cond_inc32(Assembler::zero,
1257                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1258     }
1259     jcc(Assembler::zero, done);
1260 
1261     bind(slow_case);
1262 
1263     // Call the runtime routine for slow case
1264     call_VM(noreg,
1265             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1266             lock_reg);
1267 
1268     bind(done);
1269   }
1270 }
1271 
1272 
1273 // Unlocks an object. Used in monitorexit bytecode and
1274 // remove_activation.  Throws an IllegalMonitorException if object is
1275 // not locked by current thread.
1276 //
1277 // Args:
1278 //      rdx, c_rarg1: BasicObjectLock for lock
1279 //
1280 // Kills:
1281 //      rax
1282 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1283 //      rscratch1 (scratch reg)
1284 // rax, rbx, rcx, rdx
1285 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1286   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1287          "The argument is only for looks. It must be c_rarg1");
1288 
1289   if (UseHeavyMonitors) {
1290     call_VM(noreg,
1291             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1292             lock_reg);
1293   } else {
1294     Label done;
1295 
1296     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1297     const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx);  // Will contain the old oopMark
1298     const Register obj_reg    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);  // Will contain the oop
1299 
1300     save_bcp(); // Save in case of exception
1301 
1302     // Convert from BasicObjectLock structure to object and BasicLock
1303     // structure Store the BasicLock address into %rax
1304     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1305 
1306     // Load oop into obj_reg(%c_rarg3)
1307     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1308 
1309     // Free entry
1310     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1311 
1312     if (UseBiasedLocking) {
1313       biased_locking_exit(obj_reg, header_reg, done);
1314     }
1315 
1316     // Load the old header from BasicLock structure
1317     movptr(header_reg, Address(swap_reg,
1318                                BasicLock::displaced_header_offset_in_bytes()));
1319 
1320     // Test for recursion
1321     testptr(header_reg, header_reg);
1322 
1323     // zero for recursive case
1324     jcc(Assembler::zero, done);
1325 
1326     // Atomic swap back the old header
1327     lock();
1328     cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1329 
1330     // zero for simple unlock of a stack-lock case
1331     jcc(Assembler::zero, done);
1332 
1333     // Call the runtime routine for slow case.
1334     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1335          obj_reg); // restore obj
1336     call_VM(noreg,
1337             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1338             lock_reg);
1339 
1340     bind(done);
1341 
1342     restore_bcp();
1343   }
1344 }
1345 
1346 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1347                                                          Label& zero_continue) {
1348   assert(ProfileInterpreter, "must be profiling interpreter");
1349   movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1350   testptr(mdp, mdp);
1351   jcc(Assembler::zero, zero_continue);
1352 }
1353 
1354 
1355 // Set the method data pointer for the current bcp.
1356 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1357   assert(ProfileInterpreter, "must be profiling interpreter");
1358   Label set_mdp;
1359   push(rax);
1360   push(rbx);
1361 
1362   get_method(rbx);
1363   // Test MDO to avoid the call if it is NULL.
1364   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1365   testptr(rax, rax);
1366   jcc(Assembler::zero, set_mdp);
1367   // rbx: method
1368   // _bcp_register: bcp
1369   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1370   // rax: mdi
1371   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1372   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1373   addptr(rbx, in_bytes(MethodData::data_offset()));
1374   addptr(rax, rbx);
1375   bind(set_mdp);
1376   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1377   pop(rbx);
1378   pop(rax);
1379 }
1380 
1381 void InterpreterMacroAssembler::verify_method_data_pointer() {
1382   assert(ProfileInterpreter, "must be profiling interpreter");
1383 #ifdef ASSERT
1384   Label verify_continue;
1385   push(rax);
1386   push(rbx);
1387   Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1388   Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1389   push(arg3_reg);
1390   push(arg2_reg);
1391   test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1392   get_method(rbx);
1393 
1394   // If the mdp is valid, it will point to a DataLayout header which is
1395   // consistent with the bcp.  The converse is highly probable also.
1396   load_unsigned_short(arg2_reg,
1397                       Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1398   addptr(arg2_reg, Address(rbx, Method::const_offset()));
1399   lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1400   cmpptr(arg2_reg, _bcp_register);
1401   jcc(Assembler::equal, verify_continue);
1402   // rbx: method
1403   // _bcp_register: bcp
1404   // c_rarg3: mdp
1405   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1406                rbx, _bcp_register, arg3_reg);
1407   bind(verify_continue);
1408   pop(arg2_reg);
1409   pop(arg3_reg);
1410   pop(rbx);
1411   pop(rax);
1412 #endif // ASSERT
1413 }
1414 
1415 
1416 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1417                                                 int constant,
1418                                                 Register value) {
1419   assert(ProfileInterpreter, "must be profiling interpreter");
1420   Address data(mdp_in, constant);
1421   movptr(data, value);
1422 }
1423 
1424 
1425 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1426                                                       int constant,
1427                                                       bool decrement) {
1428   // Counter address
1429   Address data(mdp_in, constant);
1430 
1431   increment_mdp_data_at(data, decrement);
1432 }
1433 
1434 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1435                                                       bool decrement) {
1436   assert(ProfileInterpreter, "must be profiling interpreter");
1437   // %%% this does 64bit counters at best it is wasting space
1438   // at worst it is a rare bug when counters overflow
1439 
1440   if (decrement) {
1441     // Decrement the register.  Set condition codes.
1442     addptr(data, (int32_t) -DataLayout::counter_increment);
1443     // If the decrement causes the counter to overflow, stay negative
1444     Label L;
1445     jcc(Assembler::negative, L);
1446     addptr(data, (int32_t) DataLayout::counter_increment);
1447     bind(L);
1448   } else {
1449     assert(DataLayout::counter_increment == 1,
1450            "flow-free idiom only works with 1");
1451     // Increment the register.  Set carry flag.
1452     addptr(data, DataLayout::counter_increment);
1453     // If the increment causes the counter to overflow, pull back by 1.
1454     sbbptr(data, (int32_t)0);
1455   }
1456 }
1457 
1458 
1459 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1460                                                       Register reg,
1461                                                       int constant,
1462                                                       bool decrement) {
1463   Address data(mdp_in, reg, Address::times_1, constant);
1464 
1465   increment_mdp_data_at(data, decrement);
1466 }
1467 
1468 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1469                                                 int flag_byte_constant) {
1470   assert(ProfileInterpreter, "must be profiling interpreter");
1471   int header_offset = in_bytes(DataLayout::flags_offset());
1472   int header_bits = flag_byte_constant;
1473   // Set the flag
1474   orb(Address(mdp_in, header_offset), header_bits);
1475 }
1476 
1477 
1478 
1479 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1480                                                  int offset,
1481                                                  Register value,
1482                                                  Register test_value_out,
1483                                                  Label& not_equal_continue) {
1484   assert(ProfileInterpreter, "must be profiling interpreter");
1485   if (test_value_out == noreg) {
1486     cmpptr(value, Address(mdp_in, offset));
1487   } else {
1488     // Put the test value into a register, so caller can use it:
1489     movptr(test_value_out, Address(mdp_in, offset));
1490     cmpptr(test_value_out, value);
1491   }
1492   jcc(Assembler::notEqual, not_equal_continue);
1493 }
1494 
1495 
1496 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1497                                                      int offset_of_disp) {
1498   assert(ProfileInterpreter, "must be profiling interpreter");
1499   Address disp_address(mdp_in, offset_of_disp);
1500   addptr(mdp_in, disp_address);
1501   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1502 }
1503 
1504 
1505 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1506                                                      Register reg,
1507                                                      int offset_of_disp) {
1508   assert(ProfileInterpreter, "must be profiling interpreter");
1509   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1510   addptr(mdp_in, disp_address);
1511   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1512 }
1513 
1514 
1515 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1516                                                        int constant) {
1517   assert(ProfileInterpreter, "must be profiling interpreter");
1518   addptr(mdp_in, constant);
1519   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1520 }
1521 
1522 
1523 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1524   assert(ProfileInterpreter, "must be profiling interpreter");
1525   push(return_bci); // save/restore across call_VM
1526   call_VM(noreg,
1527           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1528           return_bci);
1529   pop(return_bci);
1530 }
1531 
1532 
1533 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1534                                                      Register bumped_count) {
1535   if (ProfileInterpreter) {
1536     Label profile_continue;
1537 
1538     // If no method data exists, go to profile_continue.
1539     // Otherwise, assign to mdp
1540     test_method_data_pointer(mdp, profile_continue);
1541 
1542     // We are taking a branch.  Increment the taken count.
1543     // We inline increment_mdp_data_at to return bumped_count in a register
1544     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1545     Address data(mdp, in_bytes(JumpData::taken_offset()));
1546     movptr(bumped_count, data);
1547     assert(DataLayout::counter_increment == 1,
1548             "flow-free idiom only works with 1");
1549     addptr(bumped_count, DataLayout::counter_increment);
1550     sbbptr(bumped_count, 0);
1551     movptr(data, bumped_count); // Store back out
1552 
1553     // The method data pointer needs to be updated to reflect the new target.
1554     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1555     bind(profile_continue);
1556   }
1557 }
1558 
1559 
1560 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1561   if (ProfileInterpreter) {
1562     Label profile_continue;
1563 
1564     // If no method data exists, go to profile_continue.
1565     test_method_data_pointer(mdp, profile_continue);
1566 
1567     // We are taking a branch.  Increment the not taken count.
1568     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1569 
1570     // The method data pointer needs to be updated to correspond to
1571     // the next bytecode
1572     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1573     bind(profile_continue);
1574   }
1575 }
1576 
1577 void InterpreterMacroAssembler::profile_call(Register mdp) {
1578   if (ProfileInterpreter) {
1579     Label profile_continue;
1580 
1581     // If no method data exists, go to profile_continue.
1582     test_method_data_pointer(mdp, profile_continue);
1583 
1584     // We are making a call.  Increment the count.
1585     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1586 
1587     // The method data pointer needs to be updated to reflect the new target.
1588     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1589     bind(profile_continue);
1590   }
1591 }
1592 
1593 
1594 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1595   if (ProfileInterpreter) {
1596     Label profile_continue;
1597 
1598     // If no method data exists, go to profile_continue.
1599     test_method_data_pointer(mdp, profile_continue);
1600 
1601     // We are making a call.  Increment the count.
1602     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1603 
1604     // The method data pointer needs to be updated to reflect the new target.
1605     update_mdp_by_constant(mdp,
1606                            in_bytes(VirtualCallData::
1607                                     virtual_call_data_size()));
1608     bind(profile_continue);
1609   }
1610 }
1611 
1612 
1613 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1614                                                      Register mdp,
1615                                                      Register reg2,
1616                                                      bool receiver_can_be_null) {
1617   if (ProfileInterpreter) {
1618     Label profile_continue;
1619 
1620     // If no method data exists, go to profile_continue.
1621     test_method_data_pointer(mdp, profile_continue);
1622 
1623     Label skip_receiver_profile;
1624     if (receiver_can_be_null) {
1625       Label not_null;
1626       testptr(receiver, receiver);
1627       jccb(Assembler::notZero, not_null);
1628       // We are making a call.  Increment the count for null receiver.
1629       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1630       jmp(skip_receiver_profile);
1631       bind(not_null);
1632     }
1633 
1634     // Record the receiver type.
1635     record_klass_in_profile(receiver, mdp, reg2, true);
1636     bind(skip_receiver_profile);
1637 
1638     // The method data pointer needs to be updated to reflect the new target.
1639 #if INCLUDE_JVMCI
1640     if (MethodProfileWidth == 0) {
1641       update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1642     }
1643 #else // INCLUDE_JVMCI
1644     update_mdp_by_constant(mdp,
1645                            in_bytes(VirtualCallData::
1646                                     virtual_call_data_size()));
1647 #endif // INCLUDE_JVMCI
1648     bind(profile_continue);
1649   }
1650 }
1651 
1652 #if INCLUDE_JVMCI
1653 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1654   assert_different_registers(method, mdp, reg2);
1655   if (ProfileInterpreter && MethodProfileWidth > 0) {
1656     Label profile_continue;
1657 
1658     // If no method data exists, go to profile_continue.
1659     test_method_data_pointer(mdp, profile_continue);
1660 
1661     Label done;
1662     record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1663       &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1664     bind(done);
1665 
1666     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1667     bind(profile_continue);
1668   }
1669 }
1670 #endif // INCLUDE_JVMCI
1671 
1672 // This routine creates a state machine for updating the multi-row
1673 // type profile at a virtual call site (or other type-sensitive bytecode).
1674 // The machine visits each row (of receiver/count) until the receiver type
1675 // is found, or until it runs out of rows.  At the same time, it remembers
1676 // the location of the first empty row.  (An empty row records null for its
1677 // receiver, and can be allocated for a newly-observed receiver type.)
1678 // Because there are two degrees of freedom in the state, a simple linear
1679 // search will not work; it must be a decision tree.  Hence this helper
1680 // function is recursive, to generate the required tree structured code.
1681 // It's the interpreter, so we are trading off code space for speed.
1682 // See below for example code.
1683 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1684                                         Register receiver, Register mdp,
1685                                         Register reg2, int start_row,
1686                                         Label& done, bool is_virtual_call) {
1687   if (TypeProfileWidth == 0) {
1688     if (is_virtual_call) {
1689       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1690     }
1691 #if INCLUDE_JVMCI
1692     else if (EnableJVMCI) {
1693       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1694     }
1695 #endif // INCLUDE_JVMCI
1696   } else {
1697     int non_profiled_offset = -1;
1698     if (is_virtual_call) {
1699       non_profiled_offset = in_bytes(CounterData::count_offset());
1700     }
1701 #if INCLUDE_JVMCI
1702     else if (EnableJVMCI) {
1703       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1704     }
1705 #endif // INCLUDE_JVMCI
1706 
1707     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1708         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1709   }
1710 }
1711 
1712 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1713                                         Register reg2, int start_row, Label& done, int total_rows,
1714                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1715                                         int non_profiled_offset) {
1716   int last_row = total_rows - 1;
1717   assert(start_row <= last_row, "must be work left to do");
1718   // Test this row for both the item and for null.
1719   // Take any of three different outcomes:
1720   //   1. found item => increment count and goto done
1721   //   2. found null => keep looking for case 1, maybe allocate this cell
1722   //   3. found something else => keep looking for cases 1 and 2
1723   // Case 3 is handled by a recursive call.
1724   for (int row = start_row; row <= last_row; row++) {
1725     Label next_test;
1726     bool test_for_null_also = (row == start_row);
1727 
1728     // See if the item is item[n].
1729     int item_offset = in_bytes(item_offset_fn(row));
1730     test_mdp_data_at(mdp, item_offset, item,
1731                      (test_for_null_also ? reg2 : noreg),
1732                      next_test);
1733     // (Reg2 now contains the item from the CallData.)
1734 
1735     // The item is item[n].  Increment count[n].
1736     int count_offset = in_bytes(item_count_offset_fn(row));
1737     increment_mdp_data_at(mdp, count_offset);
1738     jmp(done);
1739     bind(next_test);
1740 
1741     if (test_for_null_also) {
1742       // Failed the equality check on item[n]...  Test for null.
1743       testptr(reg2, reg2);
1744       if (start_row == last_row) {
1745         // The only thing left to do is handle the null case.
1746         if (non_profiled_offset >= 0) {
1747           Label found_null;
1748           jccb(Assembler::zero, found_null);
1749           // Item did not match any saved item and there is no empty row for it.
1750           // Increment total counter to indicate polymorphic case.
1751           increment_mdp_data_at(mdp, non_profiled_offset);
1752           jmp(done);
1753           bind(found_null);
1754         } else {
1755           jcc(Assembler::notZero, done);
1756         }
1757         break;
1758       }
1759       Label found_null;
1760       // Since null is rare, make it be the branch-taken case.
1761       jcc(Assembler::zero, found_null);
1762 
1763       // Put all the "Case 3" tests here.
1764       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1765         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1766 
1767       // Found a null.  Keep searching for a matching item,
1768       // but remember that this is an empty (unused) slot.
1769       bind(found_null);
1770     }
1771   }
1772 
1773   // In the fall-through case, we found no matching item, but we
1774   // observed the item[start_row] is NULL.
1775 
1776   // Fill in the item field and increment the count.
1777   int item_offset = in_bytes(item_offset_fn(start_row));
1778   set_mdp_data_at(mdp, item_offset, item);
1779   int count_offset = in_bytes(item_count_offset_fn(start_row));
1780   movl(reg2, DataLayout::counter_increment);
1781   set_mdp_data_at(mdp, count_offset, reg2);
1782   if (start_row > 0) {
1783     jmp(done);
1784   }
1785 }
1786 
1787 // Example state machine code for three profile rows:
1788 //   // main copy of decision tree, rooted at row[1]
1789 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1790 //   if (row[0].rec != NULL) {
1791 //     // inner copy of decision tree, rooted at row[1]
1792 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1793 //     if (row[1].rec != NULL) {
1794 //       // degenerate decision tree, rooted at row[2]
1795 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1796 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1797 //       row[2].init(rec); goto done;
1798 //     } else {
1799 //       // remember row[1] is empty
1800 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1801 //       row[1].init(rec); goto done;
1802 //     }
1803 //   } else {
1804 //     // remember row[0] is empty
1805 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1806 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1807 //     row[0].init(rec); goto done;
1808 //   }
1809 //   done:
1810 
1811 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1812                                                         Register mdp, Register reg2,
1813                                                         bool is_virtual_call) {
1814   assert(ProfileInterpreter, "must be profiling");
1815   Label done;
1816 
1817   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1818 
1819   bind (done);
1820 }
1821 
1822 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1823                                             Register mdp) {
1824   if (ProfileInterpreter) {
1825     Label profile_continue;
1826     uint row;
1827 
1828     // If no method data exists, go to profile_continue.
1829     test_method_data_pointer(mdp, profile_continue);
1830 
1831     // Update the total ret count.
1832     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1833 
1834     for (row = 0; row < RetData::row_limit(); row++) {
1835       Label next_test;
1836 
1837       // See if return_bci is equal to bci[n]:
1838       test_mdp_data_at(mdp,
1839                        in_bytes(RetData::bci_offset(row)),
1840                        return_bci, noreg,
1841                        next_test);
1842 
1843       // return_bci is equal to bci[n].  Increment the count.
1844       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1845 
1846       // The method data pointer needs to be updated to reflect the new target.
1847       update_mdp_by_offset(mdp,
1848                            in_bytes(RetData::bci_displacement_offset(row)));
1849       jmp(profile_continue);
1850       bind(next_test);
1851     }
1852 
1853     update_mdp_for_ret(return_bci);
1854 
1855     bind(profile_continue);
1856   }
1857 }
1858 
1859 
1860 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1861   if (ProfileInterpreter) {
1862     Label profile_continue;
1863 
1864     // If no method data exists, go to profile_continue.
1865     test_method_data_pointer(mdp, profile_continue);
1866 
1867     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1868 
1869     // The method data pointer needs to be updated.
1870     int mdp_delta = in_bytes(BitData::bit_data_size());
1871     if (TypeProfileCasts) {
1872       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1873     }
1874     update_mdp_by_constant(mdp, mdp_delta);
1875 
1876     bind(profile_continue);
1877   }
1878 }
1879 
1880 
1881 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1882   if (ProfileInterpreter && TypeProfileCasts) {
1883     Label profile_continue;
1884 
1885     // If no method data exists, go to profile_continue.
1886     test_method_data_pointer(mdp, profile_continue);
1887 
1888     int count_offset = in_bytes(CounterData::count_offset());
1889     // Back up the address, since we have already bumped the mdp.
1890     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1891 
1892     // *Decrement* the counter.  We expect to see zero or small negatives.
1893     increment_mdp_data_at(mdp, count_offset, true);
1894 
1895     bind (profile_continue);
1896   }
1897 }
1898 
1899 
1900 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1901   if (ProfileInterpreter) {
1902     Label profile_continue;
1903 
1904     // If no method data exists, go to profile_continue.
1905     test_method_data_pointer(mdp, profile_continue);
1906 
1907     // The method data pointer needs to be updated.
1908     int mdp_delta = in_bytes(BitData::bit_data_size());
1909     if (TypeProfileCasts) {
1910       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1911 
1912       // Record the object type.
1913       record_klass_in_profile(klass, mdp, reg2, false);
1914       NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1915       NOT_LP64(restore_locals();)         // Restore EDI
1916     }
1917     update_mdp_by_constant(mdp, mdp_delta);
1918 
1919     bind(profile_continue);
1920   }
1921 }
1922 
1923 
1924 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1925   if (ProfileInterpreter) {
1926     Label profile_continue;
1927 
1928     // If no method data exists, go to profile_continue.
1929     test_method_data_pointer(mdp, profile_continue);
1930 
1931     // Update the default case count
1932     increment_mdp_data_at(mdp,
1933                           in_bytes(MultiBranchData::default_count_offset()));
1934 
1935     // The method data pointer needs to be updated.
1936     update_mdp_by_offset(mdp,
1937                          in_bytes(MultiBranchData::
1938                                   default_displacement_offset()));
1939 
1940     bind(profile_continue);
1941   }
1942 }
1943 
1944 
1945 void InterpreterMacroAssembler::profile_switch_case(Register index,
1946                                                     Register mdp,
1947                                                     Register reg2) {
1948   if (ProfileInterpreter) {
1949     Label profile_continue;
1950 
1951     // If no method data exists, go to profile_continue.
1952     test_method_data_pointer(mdp, profile_continue);
1953 
1954     // Build the base (index * per_case_size_in_bytes()) +
1955     // case_array_offset_in_bytes()
1956     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1957     imulptr(index, reg2); // XXX l ?
1958     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1959 
1960     // Update the case count
1961     increment_mdp_data_at(mdp,
1962                           index,
1963                           in_bytes(MultiBranchData::relative_count_offset()));
1964 
1965     // The method data pointer needs to be updated.
1966     update_mdp_by_offset(mdp,
1967                          index,
1968                          in_bytes(MultiBranchData::
1969                                   relative_displacement_offset()));
1970 
1971     bind(profile_continue);
1972   }
1973 }
1974 
1975 
1976 
1977 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1978   if (state == atos) {
1979     MacroAssembler::verify_oop(reg);
1980   }
1981 }
1982 
1983 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1984 #ifndef _LP64
1985   if ((state == ftos && UseSSE < 1) ||
1986       (state == dtos && UseSSE < 2)) {
1987     MacroAssembler::verify_FPU(stack_depth);
1988   }
1989 #endif
1990 }
1991 
1992 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1993 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1994                                                         int increment, Address mask,
1995                                                         Register scratch, bool preloaded,
1996                                                         Condition cond, Label* where) {
1997   if (!preloaded) {
1998     movl(scratch, counter_addr);
1999   }
2000   incrementl(scratch, increment);
2001   movl(counter_addr, scratch);
2002   andl(scratch, mask);
2003   if (where != NULL) {
2004     jcc(cond, *where);
2005   }
2006 }
2007 
2008 void InterpreterMacroAssembler::notify_method_entry() {
2009   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2010   // track stack depth.  If it is possible to enter interp_only_mode we add
2011   // the code to check if the event should be sent.
2012   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2013   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2014   if (JvmtiExport::can_post_interpreter_events()) {
2015     Label L;
2016     NOT_LP64(get_thread(rthread);)
2017     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2018     testl(rdx, rdx);
2019     jcc(Assembler::zero, L);
2020     call_VM(noreg, CAST_FROM_FN_PTR(address,
2021                                     InterpreterRuntime::post_method_entry));
2022     bind(L);
2023   }
2024 
2025   {
2026     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2027     NOT_LP64(get_thread(rthread);)
2028     get_method(rarg);
2029     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2030                  rthread, rarg);
2031   }
2032 
2033   // RedefineClasses() tracing support for obsolete method entry
2034   if (log_is_enabled(Trace, redefine, class, obsolete)) {
2035     NOT_LP64(get_thread(rthread);)
2036     get_method(rarg);
2037     call_VM_leaf(
2038       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2039       rthread, rarg);
2040   }
2041 }
2042 
2043 
2044 void InterpreterMacroAssembler::notify_method_exit(
2045     TosState state, NotifyMethodExitMode mode) {
2046   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2047   // track stack depth.  If it is possible to enter interp_only_mode we add
2048   // the code to check if the event should be sent.
2049   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2050   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2051   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2052     Label L;
2053     // Note: frame::interpreter_frame_result has a dependency on how the
2054     // method result is saved across the call to post_method_exit. If this
2055     // is changed then the interpreter_frame_result implementation will
2056     // need to be updated too.
2057 
2058     // template interpreter will leave the result on the top of the stack.
2059     push(state);
2060     NOT_LP64(get_thread(rthread);)
2061     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2062     testl(rdx, rdx);
2063     jcc(Assembler::zero, L);
2064     call_VM(noreg,
2065             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2066     bind(L);
2067     pop(state);
2068   }
2069 
2070   {
2071     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2072     push(state);
2073     NOT_LP64(get_thread(rthread);)
2074     get_method(rarg);
2075     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2076                  rthread, rarg);
2077     pop(state);
2078   }
2079 }