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