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