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