1 /*
   2  * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "interp_masm_x86.hpp"
  27 #include "interpreter/interpreter.hpp"
  28 #include "interpreter/interpreterRuntime.hpp"
  29 #include "logging/log.hpp"
  30 #include "oops/arrayOop.hpp"
  31 #include "oops/markWord.hpp"
  32 #include "oops/methodData.hpp"
  33 #include "oops/method.hpp"
  34 #include "oops/valueKlass.hpp"
  35 #include "prims/jvmtiExport.hpp"
  36 #include "prims/jvmtiThreadState.hpp"
  37 #include "runtime/basicLock.hpp"
  38 #include "runtime/biasedLocking.hpp"
  39 #include "runtime/frame.inline.hpp"
  40 #include "runtime/safepointMechanism.hpp"
  41 #include "runtime/sharedRuntime.hpp"
  42 #include "runtime/thread.inline.hpp"
  43 
  44 // Implementation of InterpreterMacroAssembler
  45 
  46 void InterpreterMacroAssembler::jump_to_entry(address entry) {
  47   assert(entry, "Entry must have been generated by now");
  48   jump(RuntimeAddress(entry));
  49 }
  50 
  51 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
  52   Label update, next, none;
  53 
  54   verify_oop(obj);
  55 
  56   testptr(obj, obj);
  57   jccb(Assembler::notZero, update);
  58   orptr(mdo_addr, TypeEntries::null_seen);
  59   jmpb(next);
  60 
  61   bind(update);
  62   load_klass(obj, obj);
  63 
  64   xorptr(obj, mdo_addr);
  65   testptr(obj, TypeEntries::type_klass_mask);
  66   jccb(Assembler::zero, next); // klass seen before, nothing to
  67                                // do. The unknown bit may have been
  68                                // set already but no need to check.
  69 
  70   testptr(obj, TypeEntries::type_unknown);
  71   jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
  72 
  73   cmpptr(mdo_addr, 0);
  74   jccb(Assembler::equal, none);
  75   cmpptr(mdo_addr, TypeEntries::null_seen);
  76   jccb(Assembler::equal, none);
  77   // There is a chance that the checks above (re-reading profiling
  78   // data from memory) fail if another thread has just set the
  79   // profiling to this obj's klass
  80   xorptr(obj, mdo_addr);
  81   testptr(obj, TypeEntries::type_klass_mask);
  82   jccb(Assembler::zero, next);
  83 
  84   // different than before. Cannot keep accurate profile.
  85   orptr(mdo_addr, TypeEntries::type_unknown);
  86   jmpb(next);
  87 
  88   bind(none);
  89   // first time here. Set profile type.
  90   movptr(mdo_addr, obj);
  91 
  92   bind(next);
  93 }
  94 
  95 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
  96   if (!ProfileInterpreter) {
  97     return;
  98   }
  99 
 100   if (MethodData::profile_arguments() || MethodData::profile_return()) {
 101     Label profile_continue;
 102 
 103     test_method_data_pointer(mdp, profile_continue);
 104 
 105     int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
 106 
 107     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
 108     jcc(Assembler::notEqual, profile_continue);
 109 
 110     if (MethodData::profile_arguments()) {
 111       Label done;
 112       int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
 113       addptr(mdp, off_to_args);
 114 
 115       for (int i = 0; i < TypeProfileArgsLimit; i++) {
 116         if (i > 0 || MethodData::profile_return()) {
 117           // If return value type is profiled we may have no argument to profile
 118           movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 119           subl(tmp, i*TypeStackSlotEntries::per_arg_count());
 120           cmpl(tmp, TypeStackSlotEntries::per_arg_count());
 121           jcc(Assembler::less, done);
 122         }
 123         movptr(tmp, Address(callee, Method::const_offset()));
 124         load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
 125         // stack offset o (zero based) from the start of the argument
 126         // list, for n arguments translates into offset n - o - 1 from
 127         // the end of the argument list
 128         subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
 129         subl(tmp, 1);
 130         Address arg_addr = argument_address(tmp);
 131         movptr(tmp, arg_addr);
 132 
 133         Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
 134         profile_obj_type(tmp, mdo_arg_addr);
 135 
 136         int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
 137         addptr(mdp, to_add);
 138         off_to_args += to_add;
 139       }
 140 
 141       if (MethodData::profile_return()) {
 142         movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
 143         subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
 144       }
 145 
 146       bind(done);
 147 
 148       if (MethodData::profile_return()) {
 149         // We're right after the type profile for the last
 150         // argument. tmp is the number of cells left in the
 151         // CallTypeData/VirtualCallTypeData to reach its end. Non null
 152         // if there's a return to profile.
 153         assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
 154         shll(tmp, exact_log2(DataLayout::cell_size));
 155         addptr(mdp, tmp);
 156       }
 157       movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
 158     } else {
 159       assert(MethodData::profile_return(), "either profile call args or call ret");
 160       update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
 161     }
 162 
 163     // mdp points right after the end of the
 164     // CallTypeData/VirtualCallTypeData, right after the cells for the
 165     // return value type if there's one
 166 
 167     bind(profile_continue);
 168   }
 169 }
 170 
 171 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
 172   assert_different_registers(mdp, ret, tmp, _bcp_register);
 173   if (ProfileInterpreter && MethodData::profile_return()) {
 174     Label profile_continue;
 175 
 176     test_method_data_pointer(mdp, profile_continue);
 177 
 178     if (MethodData::profile_return_jsr292_only()) {
 179       assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
 180 
 181       // If we don't profile all invoke bytecodes we must make sure
 182       // it's a bytecode we indeed profile. We can't go back to the
 183       // begining of the ProfileData we intend to update to check its
 184       // type because we're right after it and we don't known its
 185       // length
 186       Label do_profile;
 187       cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
 188       jcc(Assembler::equal, do_profile);
 189       cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
 190       jcc(Assembler::equal, do_profile);
 191       get_method(tmp);
 192       cmpw(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
 193       jcc(Assembler::notEqual, profile_continue);
 194 
 195       bind(do_profile);
 196     }
 197 
 198     Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
 199     mov(tmp, ret);
 200     profile_obj_type(tmp, mdo_ret_addr);
 201 
 202     bind(profile_continue);
 203   }
 204 }
 205 
 206 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
 207   if (ProfileInterpreter && MethodData::profile_parameters()) {
 208     Label profile_continue;
 209 
 210     test_method_data_pointer(mdp, profile_continue);
 211 
 212     // Load the offset of the area within the MDO used for
 213     // parameters. If it's negative we're not profiling any parameters
 214     movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
 215     testl(tmp1, tmp1);
 216     jcc(Assembler::negative, profile_continue);
 217 
 218     // Compute a pointer to the area for parameters from the offset
 219     // and move the pointer to the slot for the last
 220     // parameters. Collect profiling from last parameter down.
 221     // mdo start + parameters offset + array length - 1
 222     addptr(mdp, tmp1);
 223     movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
 224     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 225 
 226     Label loop;
 227     bind(loop);
 228 
 229     int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
 230     int type_base = in_bytes(ParametersTypeData::type_offset(0));
 231     Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
 232     Address arg_off(mdp, tmp1, per_arg_scale, off_base);
 233     Address arg_type(mdp, tmp1, per_arg_scale, type_base);
 234 
 235     // load offset on the stack from the slot for this parameter
 236     movptr(tmp2, arg_off);
 237     negptr(tmp2);
 238     // read the parameter from the local area
 239     movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
 240 
 241     // profile the parameter
 242     profile_obj_type(tmp2, arg_type);
 243 
 244     // go to next parameter
 245     decrement(tmp1, TypeStackSlotEntries::per_arg_count());
 246     jcc(Assembler::positive, loop);
 247 
 248     bind(profile_continue);
 249   }
 250 }
 251 
 252 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
 253                                                   int number_of_arguments) {
 254   // interpreter specific
 255   //
 256   // Note: No need to save/restore bcp & locals registers
 257   //       since these are callee saved registers and no blocking/
 258   //       GC can happen in leaf calls.
 259   // Further Note: DO NOT save/restore bcp/locals. If a caller has
 260   // already saved them so that it can use rsi/rdi as temporaries
 261   // then a save/restore here will DESTROY the copy the caller
 262   // saved! There used to be a save_bcp() that only happened in
 263   // the ASSERT path (no restore_bcp). Which caused bizarre failures
 264   // when jvm built with ASSERTs.
 265 #ifdef ASSERT
 266   {
 267     Label L;
 268     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 269     jcc(Assembler::equal, L);
 270     stop("InterpreterMacroAssembler::call_VM_leaf_base:"
 271          " last_sp != NULL");
 272     bind(L);
 273   }
 274 #endif
 275   // super call
 276   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
 277   // interpreter specific
 278   // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
 279   // but since they may not have been saved (and we don't want to
 280   // save them here (see note above) the assert is invalid.
 281 }
 282 
 283 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
 284                                              Register java_thread,
 285                                              Register last_java_sp,
 286                                              address  entry_point,
 287                                              int      number_of_arguments,
 288                                              bool     check_exceptions) {
 289   // interpreter specific
 290   //
 291   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
 292   //       really make a difference for these runtime calls, since they are
 293   //       slow anyway. Btw., bcp must be saved/restored since it may change
 294   //       due to GC.
 295   NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
 296   save_bcp();
 297 #ifdef ASSERT
 298   {
 299     Label L;
 300     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
 301     jcc(Assembler::equal, L);
 302     stop("InterpreterMacroAssembler::call_VM_base:"
 303          " last_sp != NULL");
 304     bind(L);
 305   }
 306 #endif /* ASSERT */
 307   // super call
 308   MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
 309                                entry_point, number_of_arguments,
 310                                check_exceptions);
 311   // interpreter specific
 312   restore_bcp();
 313   restore_locals();
 314 }
 315 
 316 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
 317   if (JvmtiExport::can_pop_frame()) {
 318     Label L;
 319     // Initiate popframe handling only if it is not already being
 320     // processed.  If the flag has the popframe_processing bit set, it
 321     // means that this code is called *during* popframe handling - we
 322     // don't want to reenter.
 323     // This method is only called just after the call into the vm in
 324     // call_VM_base, so the arg registers are available.
 325     Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
 326                         LP64_ONLY(c_rarg0);
 327     movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
 328     testl(pop_cond, JavaThread::popframe_pending_bit);
 329     jcc(Assembler::zero, L);
 330     testl(pop_cond, JavaThread::popframe_processing_bit);
 331     jcc(Assembler::notZero, L);
 332     // Call Interpreter::remove_activation_preserving_args_entry() to get the
 333     // address of the same-named entrypoint in the generated interpreter code.
 334     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
 335     jmp(rax);
 336     bind(L);
 337     NOT_LP64(get_thread(java_thread);)
 338   }
 339 }
 340 
 341 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
 342   Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 343   NOT_LP64(get_thread(thread);)
 344   movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
 345   const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
 346   const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
 347   const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
 348 #ifdef _LP64
 349   switch (state) {
 350     case atos: movptr(rax, oop_addr);
 351                movptr(oop_addr, (int32_t)NULL_WORD);
 352                verify_oop(rax, state);              break;
 353     case ltos: movptr(rax, val_addr);                 break;
 354     case btos:                                   // fall through
 355     case ztos:                                   // fall through
 356     case ctos:                                   // fall through
 357     case stos:                                   // fall through
 358     case itos: movl(rax, val_addr);                 break;
 359     case ftos: load_float(val_addr);                break;
 360     case dtos: load_double(val_addr);               break;
 361     case vtos: /* nothing to do */                  break;
 362     default  : ShouldNotReachHere();
 363   }
 364   // Clean up tos value in the thread object
 365   movl(tos_addr,  (int) ilgl);
 366   movl(val_addr,  (int32_t) NULL_WORD);
 367 #else
 368   const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
 369                              + in_ByteSize(wordSize));
 370   switch (state) {
 371     case atos: movptr(rax, oop_addr);
 372                movptr(oop_addr, NULL_WORD);
 373                verify_oop(rax, state);                break;
 374     case ltos:
 375                movl(rdx, val_addr1);               // fall through
 376     case btos:                                     // fall through
 377     case ztos:                                     // fall through
 378     case ctos:                                     // fall through
 379     case stos:                                     // fall through
 380     case itos: movl(rax, val_addr);                   break;
 381     case ftos: load_float(val_addr);                  break;
 382     case dtos: load_double(val_addr);                 break;
 383     case vtos: /* nothing to do */                    break;
 384     default  : ShouldNotReachHere();
 385   }
 386 #endif // _LP64
 387   // Clean up tos value in the thread object
 388   movl(tos_addr,  (int32_t) ilgl);
 389   movptr(val_addr,  NULL_WORD);
 390   NOT_LP64(movptr(val_addr1, NULL_WORD);)
 391 }
 392 
 393 
 394 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
 395   if (JvmtiExport::can_force_early_return()) {
 396     Label L;
 397     Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
 398     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
 399 
 400     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 401     testptr(tmp, tmp);
 402     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
 403 
 404     // Initiate earlyret handling only if it is not already being processed.
 405     // If the flag has the earlyret_processing bit set, it means that this code
 406     // is called *during* earlyret handling - we don't want to reenter.
 407     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
 408     cmpl(tmp, JvmtiThreadState::earlyret_pending);
 409     jcc(Assembler::notEqual, L);
 410 
 411     // Call Interpreter::remove_activation_early_entry() to get the address of the
 412     // same-named entrypoint in the generated interpreter code.
 413     NOT_LP64(get_thread(java_thread);)
 414     movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
 415 #ifdef _LP64
 416     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 417     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
 418 #else
 419     pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
 420     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
 421 #endif // _LP64
 422     jmp(rax);
 423     bind(L);
 424     NOT_LP64(get_thread(java_thread);)
 425   }
 426 }
 427 
 428 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
 429   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
 430   load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
 431   bswapl(reg);
 432   shrl(reg, 16);
 433 }
 434 
 435 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
 436                                                        int bcp_offset,
 437                                                        size_t index_size) {
 438   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
 439   if (index_size == sizeof(u2)) {
 440     load_unsigned_short(index, Address(_bcp_register, bcp_offset));
 441   } else if (index_size == sizeof(u4)) {
 442     movl(index, Address(_bcp_register, bcp_offset));
 443     // Check if the secondary index definition is still ~x, otherwise
 444     // we have to change the following assembler code to calculate the
 445     // plain index.
 446     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
 447     notl(index);  // convert to plain index
 448   } else if (index_size == sizeof(u1)) {
 449     load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
 450   } else {
 451     ShouldNotReachHere();
 452   }
 453 }
 454 
 455 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
 456                                                            Register index,
 457                                                            int bcp_offset,
 458                                                            size_t index_size) {
 459   assert_different_registers(cache, index);
 460   get_cache_index_at_bcp(index, bcp_offset, index_size);
 461   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 462   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 463   // convert from field index to ConstantPoolCacheEntry index
 464   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
 465   shll(index, 2);
 466 }
 467 
 468 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
 469                                                                         Register index,
 470                                                                         Register bytecode,
 471                                                                         int byte_no,
 472                                                                         int bcp_offset,
 473                                                                         size_t index_size) {
 474   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
 475   // We use a 32-bit load here since the layout of 64-bit words on
 476   // little-endian machines allow us that.
 477   movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
 478   const int shift_count = (1 + byte_no) * BitsPerByte;
 479   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
 480          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
 481          "correct shift count");
 482   shrl(bytecode, shift_count);
 483   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
 484   andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
 485 }
 486 
 487 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
 488                                                                Register tmp,
 489                                                                int bcp_offset,
 490                                                                size_t index_size) {
 491   assert_different_registers(cache, tmp);
 492 
 493   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
 494   assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
 495   // convert from field index to ConstantPoolCacheEntry index
 496   // and from word offset to byte offset
 497   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
 498   shll(tmp, 2 + LogBytesPerWord);
 499   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
 500   // skip past the header
 501   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
 502   addptr(cache, tmp);  // construct pointer to cache entry
 503 }
 504 
 505 // Load object from cpool->resolved_references(index)
 506 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
 507                                                                  Register index,
 508                                                                  Register tmp) {
 509   assert_different_registers(result, index);
 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, tmp);
 516   load_heap_oop(result, Address(result, index,
 517                                 UseCompressedOops ? Address::times_4 : Address::times_ptr,
 518                                 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
 519 }
 520 
 521 // load cpool->resolved_klass_at(index)
 522 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
 523                                                              Register cpool,
 524                                                              Register index) {
 525   assert_different_registers(cpool, index);
 526 
 527   movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
 528   Register resolved_klasses = cpool;
 529   movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes()));
 530   movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
 531 }
 532 
 533 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
 534                                                               Register method,
 535                                                               Register cache,
 536                                                               Register index) {
 537   assert_different_registers(cache, index);
 538 
 539   const int method_offset = in_bytes(
 540     ConstantPoolCache::base_offset() +
 541       ((byte_no == TemplateTable::f2_byte)
 542        ? ConstantPoolCacheEntry::f2_offset()
 543        : ConstantPoolCacheEntry::f1_offset()));
 544 
 545   movptr(method, Address(cache, index, Address::times_ptr, method_offset)); // get f1 Method*
 546 }
 547 
 548 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
 549 // subtype of super_klass.
 550 //
 551 // Args:
 552 //      rax: superklass
 553 //      Rsub_klass: subklass
 554 //
 555 // Kills:
 556 //      rcx, rdi
 557 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
 558                                                   Label& ok_is_subtype,
 559                                                   bool profile) {
 560   assert(Rsub_klass != rax, "rax holds superklass");
 561   LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
 562   LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
 563   assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
 564   assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
 565 
 566   // Profile the not-null value's klass.
 567   if (profile) {
 568     profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
 569   }
 570 
 571   // Do the check.
 572   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
 573 
 574   // Profile the failure of the check.
 575   if (profile) {
 576     profile_typecheck_failed(rcx); // blows rcx
 577   }
 578 }
 579 
 580 
 581 #ifndef _LP64
 582 void InterpreterMacroAssembler::f2ieee() {
 583   if (IEEEPrecision) {
 584     fstp_s(Address(rsp, 0));
 585     fld_s(Address(rsp, 0));
 586   }
 587 }
 588 
 589 
 590 void InterpreterMacroAssembler::d2ieee() {
 591   if (IEEEPrecision) {
 592     fstp_d(Address(rsp, 0));
 593     fld_d(Address(rsp, 0));
 594   }
 595 }
 596 #endif // _LP64
 597 
 598 // Java Expression Stack
 599 
 600 void InterpreterMacroAssembler::pop_ptr(Register r) {
 601   pop(r);
 602 }
 603 
 604 void InterpreterMacroAssembler::push_ptr(Register r) {
 605   push(r);
 606 }
 607 
 608 void InterpreterMacroAssembler::push_i(Register r) {
 609   push(r);
 610 }
 611 
 612 void InterpreterMacroAssembler::push_f(XMMRegister r) {
 613   subptr(rsp, wordSize);
 614   movflt(Address(rsp, 0), r);
 615 }
 616 
 617 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
 618   movflt(r, Address(rsp, 0));
 619   addptr(rsp, wordSize);
 620 }
 621 
 622 void InterpreterMacroAssembler::push_d(XMMRegister r) {
 623   subptr(rsp, 2 * wordSize);
 624   movdbl(Address(rsp, 0), r);
 625 }
 626 
 627 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
 628   movdbl(r, Address(rsp, 0));
 629   addptr(rsp, 2 * Interpreter::stackElementSize);
 630 }
 631 
 632 #ifdef _LP64
 633 void InterpreterMacroAssembler::pop_i(Register r) {
 634   // XXX can't use pop currently, upper half non clean
 635   movl(r, Address(rsp, 0));
 636   addptr(rsp, wordSize);
 637 }
 638 
 639 void InterpreterMacroAssembler::pop_l(Register r) {
 640   movq(r, Address(rsp, 0));
 641   addptr(rsp, 2 * Interpreter::stackElementSize);
 642 }
 643 
 644 void InterpreterMacroAssembler::push_l(Register r) {
 645   subptr(rsp, 2 * wordSize);
 646   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
 647   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
 648 }
 649 
 650 void InterpreterMacroAssembler::pop(TosState state) {
 651   switch (state) {
 652   case atos: pop_ptr();                 break;
 653   case btos:
 654   case ztos:
 655   case ctos:
 656   case stos:
 657   case itos: pop_i();                   break;
 658   case ltos: pop_l();                   break;
 659   case ftos: pop_f(xmm0);               break;
 660   case dtos: pop_d(xmm0);               break;
 661   case vtos: /* nothing to do */        break;
 662   default:   ShouldNotReachHere();
 663   }
 664   verify_oop(rax, state);
 665 }
 666 
 667 void InterpreterMacroAssembler::push(TosState state) {
 668   verify_oop(rax, state);
 669   switch (state) {
 670   case atos: push_ptr();                break;
 671   case btos:
 672   case ztos:
 673   case ctos:
 674   case stos:
 675   case itos: push_i();                  break;
 676   case ltos: push_l();                  break;
 677   case ftos: push_f(xmm0);              break;
 678   case dtos: push_d(xmm0);              break;
 679   case vtos: /* nothing to do */        break;
 680   default  : ShouldNotReachHere();
 681   }
 682 }
 683 #else
 684 void InterpreterMacroAssembler::pop_i(Register r) {
 685   pop(r);
 686 }
 687 
 688 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
 689   pop(lo);
 690   pop(hi);
 691 }
 692 
 693 void InterpreterMacroAssembler::pop_f() {
 694   fld_s(Address(rsp, 0));
 695   addptr(rsp, 1 * wordSize);
 696 }
 697 
 698 void InterpreterMacroAssembler::pop_d() {
 699   fld_d(Address(rsp, 0));
 700   addptr(rsp, 2 * wordSize);
 701 }
 702 
 703 
 704 void InterpreterMacroAssembler::pop(TosState state) {
 705   switch (state) {
 706     case atos: pop_ptr(rax);                                 break;
 707     case btos:                                               // fall through
 708     case ztos:                                               // fall through
 709     case ctos:                                               // fall through
 710     case stos:                                               // fall through
 711     case itos: pop_i(rax);                                   break;
 712     case ltos: pop_l(rax, rdx);                              break;
 713     case ftos:
 714       if (UseSSE >= 1) {
 715         pop_f(xmm0);
 716       } else {
 717         pop_f();
 718       }
 719       break;
 720     case dtos:
 721       if (UseSSE >= 2) {
 722         pop_d(xmm0);
 723       } else {
 724         pop_d();
 725       }
 726       break;
 727     case vtos: /* nothing to do */                           break;
 728     default  : ShouldNotReachHere();
 729   }
 730   verify_oop(rax, state);
 731 }
 732 
 733 
 734 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
 735   push(hi);
 736   push(lo);
 737 }
 738 
 739 void InterpreterMacroAssembler::push_f() {
 740   // Do not schedule for no AGI! Never write beyond rsp!
 741   subptr(rsp, 1 * wordSize);
 742   fstp_s(Address(rsp, 0));
 743 }
 744 
 745 void InterpreterMacroAssembler::push_d() {
 746   // Do not schedule for no AGI! Never write beyond rsp!
 747   subptr(rsp, 2 * wordSize);
 748   fstp_d(Address(rsp, 0));
 749 }
 750 
 751 
 752 void InterpreterMacroAssembler::push(TosState state) {
 753   verify_oop(rax, state);
 754   switch (state) {
 755     case atos: push_ptr(rax); break;
 756     case btos:                                               // fall through
 757     case ztos:                                               // fall through
 758     case ctos:                                               // fall through
 759     case stos:                                               // fall through
 760     case itos: push_i(rax);                                    break;
 761     case ltos: push_l(rax, rdx);                               break;
 762     case ftos:
 763       if (UseSSE >= 1) {
 764         push_f(xmm0);
 765       } else {
 766         push_f();
 767       }
 768       break;
 769     case dtos:
 770       if (UseSSE >= 2) {
 771         push_d(xmm0);
 772       } else {
 773         push_d();
 774       }
 775       break;
 776     case vtos: /* nothing to do */                             break;
 777     default  : ShouldNotReachHere();
 778   }
 779 }
 780 #endif // _LP64
 781 
 782 
 783 // Helpers for swap and dup
 784 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
 785   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
 786 }
 787 
 788 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
 789   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
 790 }
 791 
 792 
 793 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
 794   // set sender sp
 795   lea(_bcp_register, Address(rsp, wordSize));
 796   // record last_sp
 797   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
 798 }
 799 
 800 
 801 // Jump to from_interpreted entry of a call unless single stepping is possible
 802 // in this thread in which case we must call the i2i entry
 803 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
 804   prepare_to_jump_from_interpreted();
 805 
 806   if (JvmtiExport::can_post_interpreter_events()) {
 807     Label run_compiled_code;
 808     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
 809     // compiled code in threads for which the event is enabled.  Check here for
 810     // interp_only_mode if these events CAN be enabled.
 811     // interp_only is an int, on little endian it is sufficient to test the byte only
 812     // Is a cmpl faster?
 813     LP64_ONLY(temp = r15_thread;)
 814     NOT_LP64(get_thread(temp);)
 815     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
 816     jccb(Assembler::zero, run_compiled_code);
 817     jmp(Address(method, Method::interpreter_entry_offset()));
 818     bind(run_compiled_code);
 819   }
 820 
 821   jmp(Address(method, Method::from_interpreted_offset()));
 822 }
 823 
 824 // The following two routines provide a hook so that an implementation
 825 // can schedule the dispatch in two parts.  x86 does not do this.
 826 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
 827   // Nothing x86 specific to be done here
 828 }
 829 
 830 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
 831   dispatch_next(state, step);
 832 }
 833 
 834 void InterpreterMacroAssembler::dispatch_base(TosState state,
 835                                               address* table,
 836                                               bool verifyoop,
 837                                               bool generate_poll) {
 838   verify_FPU(1, state);
 839   if (VerifyActivationFrameSize) {
 840     Label L;
 841     mov(rcx, rbp);
 842     subptr(rcx, rsp);
 843     int32_t min_frame_size =
 844       (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
 845       wordSize;
 846     cmpptr(rcx, (int32_t)min_frame_size);
 847     jcc(Assembler::greaterEqual, L);
 848     stop("broken stack frame");
 849     bind(L);
 850   }
 851   if (verifyoop) {
 852     verify_oop(rax, state);
 853   }
 854 
 855   address* const safepoint_table = Interpreter::safept_table(state);
 856 #ifdef _LP64
 857   Label no_safepoint, dispatch;
 858   if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
 859     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 860     testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
 861 
 862     jccb(Assembler::zero, no_safepoint);
 863     lea(rscratch1, ExternalAddress((address)safepoint_table));
 864     jmpb(dispatch);
 865   }
 866 
 867   bind(no_safepoint);
 868   lea(rscratch1, ExternalAddress((address)table));
 869   bind(dispatch);
 870   jmp(Address(rscratch1, rbx, Address::times_8));
 871 
 872 #else
 873   Address index(noreg, rbx, Address::times_ptr);
 874   if (SafepointMechanism::uses_thread_local_poll() && table != safepoint_table && generate_poll) {
 875     NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
 876     Label no_safepoint;
 877     const Register thread = rcx;
 878     get_thread(thread);
 879     testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
 880 
 881     jccb(Assembler::zero, no_safepoint);
 882     ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
 883     jump(dispatch_addr);
 884     bind(no_safepoint);
 885   }
 886 
 887   {
 888     ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
 889     jump(dispatch_addr);
 890   }
 891 #endif // _LP64
 892 }
 893 
 894 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
 895   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 896 }
 897 
 898 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
 899   dispatch_base(state, Interpreter::normal_table(state));
 900 }
 901 
 902 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
 903   dispatch_base(state, Interpreter::normal_table(state), false);
 904 }
 905 
 906 
 907 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
 908   // load next bytecode (load before advancing _bcp_register to prevent AGI)
 909   load_unsigned_byte(rbx, Address(_bcp_register, step));
 910   // advance _bcp_register
 911   increment(_bcp_register, step);
 912   dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
 913 }
 914 
 915 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
 916   // load current bytecode
 917   load_unsigned_byte(rbx, Address(_bcp_register, 0));
 918   dispatch_base(state, table);
 919 }
 920 
 921 void InterpreterMacroAssembler::narrow(Register result) {
 922 
 923   // Get method->_constMethod->_result_type
 924   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
 925   movptr(rcx, Address(rcx, Method::const_offset()));
 926   load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
 927 
 928   Label done, notBool, notByte, notChar;
 929 
 930   // common case first
 931   cmpl(rcx, T_INT);
 932   jcc(Assembler::equal, done);
 933 
 934   // mask integer result to narrower return type.
 935   cmpl(rcx, T_BOOLEAN);
 936   jcc(Assembler::notEqual, notBool);
 937   andl(result, 0x1);
 938   jmp(done);
 939 
 940   bind(notBool);
 941   cmpl(rcx, T_BYTE);
 942   jcc(Assembler::notEqual, notByte);
 943   LP64_ONLY(movsbl(result, result);)
 944   NOT_LP64(shll(result, 24);)      // truncate upper 24 bits
 945   NOT_LP64(sarl(result, 24);)      // and sign-extend byte
 946   jmp(done);
 947 
 948   bind(notByte);
 949   cmpl(rcx, T_CHAR);
 950   jcc(Assembler::notEqual, notChar);
 951   LP64_ONLY(movzwl(result, result);)
 952   NOT_LP64(andl(result, 0xFFFF);)  // truncate upper 16 bits
 953   jmp(done);
 954 
 955   bind(notChar);
 956   // cmpl(rcx, T_SHORT);  // all that's left
 957   // jcc(Assembler::notEqual, done);
 958   LP64_ONLY(movswl(result, result);)
 959   NOT_LP64(shll(result, 16);)      // truncate upper 16 bits
 960   NOT_LP64(sarl(result, 16);)      // and sign-extend short
 961 
 962   // Nothing to do for T_INT
 963   bind(done);
 964 }
 965 
 966 // remove activation
 967 //
 968 // Unlock the receiver if this is a synchronized method.
 969 // Unlock any Java monitors from syncronized blocks.
 970 // Remove the activation from the stack.
 971 //
 972 // If there are locked Java monitors
 973 //    If throw_monitor_exception
 974 //       throws IllegalMonitorStateException
 975 //    Else if install_monitor_exception
 976 //       installs IllegalMonitorStateException
 977 //    Else
 978 //       no error processing
 979 void InterpreterMacroAssembler::remove_activation(
 980         TosState state,
 981         Register ret_addr,
 982         bool throw_monitor_exception,
 983         bool install_monitor_exception,
 984         bool notify_jvmdi) {
 985   // Note: Registers rdx xmm0 may be in use for the
 986   // result check if synchronized method
 987   Label unlocked, unlock, no_unlock;
 988 
 989   const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
 990   const Register robj    = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
 991   const Register rmon    = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
 992                               // monitor pointers need different register
 993                               // because rdx may have the result in it
 994   NOT_LP64(get_thread(rcx);)
 995 
 996   // get the value of _do_not_unlock_if_synchronized into rdx
 997   const Address do_not_unlock_if_synchronized(rthread,
 998     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
 999   movbool(rbx, do_not_unlock_if_synchronized);
1000   movbool(do_not_unlock_if_synchronized, false); // reset the flag
1001 
1002   // get method access flags
1003   movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1004   movl(rcx, Address(rcx, Method::access_flags_offset()));
1005   testl(rcx, JVM_ACC_SYNCHRONIZED);
1006   jcc(Assembler::zero, unlocked);
1007 
1008   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1009   // is set.
1010   testbool(rbx);
1011   jcc(Assembler::notZero, no_unlock);
1012 
1013   // unlock monitor
1014   push(state); // save result
1015 
1016   // BasicObjectLock will be first in list, since this is a
1017   // synchronized method. However, need to check that the object has
1018   // not been unlocked by an explicit monitorexit bytecode.
1019   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1020                         wordSize - (int) sizeof(BasicObjectLock));
1021   // We use c_rarg1/rdx so that if we go slow path it will be the correct
1022   // register for unlock_object to pass to VM directly
1023   lea(robj, monitor); // address of first monitor
1024 
1025   movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
1026   testptr(rax, rax);
1027   jcc(Assembler::notZero, unlock);
1028 
1029   pop(state);
1030   if (throw_monitor_exception) {
1031     // Entry already unlocked, need to throw exception
1032     NOT_LP64(empty_FPU_stack();)  // remove possible return value from FPU-stack, otherwise stack could overflow
1033     call_VM(noreg, CAST_FROM_FN_PTR(address,
1034                    InterpreterRuntime::throw_illegal_monitor_state_exception));
1035     should_not_reach_here();
1036   } else {
1037     // Monitor already unlocked during a stack unroll. If requested,
1038     // install an illegal_monitor_state_exception.  Continue with
1039     // stack unrolling.
1040     if (install_monitor_exception) {
1041       NOT_LP64(empty_FPU_stack();)
1042       call_VM(noreg, CAST_FROM_FN_PTR(address,
1043                      InterpreterRuntime::new_illegal_monitor_state_exception));
1044     }
1045     jmp(unlocked);
1046   }
1047 
1048   bind(unlock);
1049   unlock_object(robj);
1050   pop(state);
1051 
1052   // Check that for block-structured locking (i.e., that all locked
1053   // objects has been unlocked)
1054   bind(unlocked);
1055 
1056   // rax, rdx: Might contain return value
1057 
1058   // Check that all monitors are unlocked
1059   {
1060     Label loop, exception, entry, restart;
1061     const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
1062     const Address monitor_block_top(
1063         rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1064     const Address monitor_block_bot(
1065         rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1066 
1067     bind(restart);
1068     // We use c_rarg1 so that if we go slow path it will be the correct
1069     // register for unlock_object to pass to VM directly
1070     movptr(rmon, monitor_block_top); // points to current entry, starting
1071                                   // with top-most entry
1072     lea(rbx, monitor_block_bot);  // points to word before bottom of
1073                                   // monitor block
1074     jmp(entry);
1075 
1076     // Entry already locked, need to throw exception
1077     bind(exception);
1078 
1079     if (throw_monitor_exception) {
1080       // Throw exception
1081       NOT_LP64(empty_FPU_stack();)
1082       MacroAssembler::call_VM(noreg,
1083                               CAST_FROM_FN_PTR(address, InterpreterRuntime::
1084                                    throw_illegal_monitor_state_exception));
1085       should_not_reach_here();
1086     } else {
1087       // Stack unrolling. Unlock object and install illegal_monitor_exception.
1088       // Unlock does not block, so don't have to worry about the frame.
1089       // We don't have to preserve c_rarg1 since we are going to throw an exception.
1090 
1091       push(state);
1092       mov(robj, rmon);   // nop if robj and rmon are the same
1093       unlock_object(robj);
1094       pop(state);
1095 
1096       if (install_monitor_exception) {
1097         NOT_LP64(empty_FPU_stack();)
1098         call_VM(noreg, CAST_FROM_FN_PTR(address,
1099                                         InterpreterRuntime::
1100                                         new_illegal_monitor_state_exception));
1101       }
1102 
1103       jmp(restart);
1104     }
1105 
1106     bind(loop);
1107     // check if current entry is used
1108     cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
1109     jcc(Assembler::notEqual, exception);
1110 
1111     addptr(rmon, entry_size); // otherwise advance to next entry
1112     bind(entry);
1113     cmpptr(rmon, rbx); // check if bottom reached
1114     jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1115   }
1116 
1117   bind(no_unlock);
1118 
1119   // jvmti support
1120   if (notify_jvmdi) {
1121     notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
1122   } else {
1123     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1124   }
1125 
1126   if (StackReservedPages > 0) {
1127     movptr(rbx,
1128                Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1129     // testing if reserved zone needs to be re-enabled
1130     Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1131     Label no_reserved_zone_enabling;
1132 
1133     NOT_LP64(get_thread(rthread);)
1134 
1135     cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_enabled);
1136     jcc(Assembler::equal, no_reserved_zone_enabling);
1137 
1138     cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1139     jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1140 
1141     call_VM_leaf(
1142       CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1143     call_VM(noreg, CAST_FROM_FN_PTR(address,
1144                    InterpreterRuntime::throw_delayed_StackOverflowError));
1145     should_not_reach_here();
1146 
1147     bind(no_reserved_zone_enabling);
1148   }
1149 
1150   // remove activation
1151   // get sender sp
1152   movptr(rbx,
1153          Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1154 
1155   if (state == atos && ValueTypeReturnedAsFields) {
1156     Label skip;
1157     // Test if the return type is a value type
1158     movptr(rdi, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1159     movptr(rdi, Address(rdi, Method::const_offset()));
1160     load_unsigned_byte(rdi, Address(rdi, ConstMethod::result_type_offset()));
1161     cmpl(rdi, T_VALUETYPE);
1162     jcc(Assembler::notEqual, skip);
1163 
1164     // We are returning a value type, load its fields into registers
1165 #ifndef _LP64
1166     super_call_VM_leaf(StubRoutines::load_value_type_fields_in_regs());
1167 #else
1168     // Load fields from a buffered value with a value class specific handler
1169     load_klass(rdi, rax);
1170     movptr(rdi, Address(rdi, InstanceKlass::adr_valueklass_fixed_block_offset()));
1171     movptr(rdi, Address(rdi, ValueKlass::unpack_handler_offset()));
1172 
1173     testptr(rdi, rdi);
1174     jcc(Assembler::equal, skip);
1175 
1176     call(rdi);
1177 #endif
1178     // call above kills the value in rbx. Reload it.
1179     movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1180     bind(skip);
1181   }
1182   leave();                           // remove frame anchor
1183   pop(ret_addr);                     // get return address
1184   mov(rsp, rbx);                     // set sp to sender sp
1185 }
1186 
1187 void InterpreterMacroAssembler::get_method_counters(Register method,
1188                                                     Register mcs, Label& skip) {
1189   Label has_counters;
1190   movptr(mcs, Address(method, Method::method_counters_offset()));
1191   testptr(mcs, mcs);
1192   jcc(Assembler::notZero, has_counters);
1193   call_VM(noreg, CAST_FROM_FN_PTR(address,
1194           InterpreterRuntime::build_method_counters), method);
1195   movptr(mcs, Address(method,Method::method_counters_offset()));
1196   testptr(mcs, mcs);
1197   jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1198   bind(has_counters);
1199 }
1200 
1201 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
1202                                                   Register t1, Register t2,
1203                                                   bool clear_fields, Label& alloc_failed) {
1204   MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
1205   {
1206     SkipIfEqual skip_if(this, &DTraceAllocProbes, 0);
1207     // Trigger dtrace event for fastpath
1208     push(atos);
1209     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), new_obj);
1210     pop(atos);
1211   }
1212 }
1213 
1214 
1215 void InterpreterMacroAssembler::read_flattened_field(Register holder_klass,
1216                                                      Register field_index, Register field_offset,
1217                                                      Register obj) {
1218   Label alloc_failed, empty_value, done;
1219   const Register src = field_offset;
1220   const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1221   const Register dst_temp   = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1222   assert_different_registers(obj, holder_klass, field_index, field_offset, dst_temp);
1223 
1224   // Grap the inline field klass
1225   push(holder_klass);
1226   const Register field_klass = holder_klass;
1227   get_value_field_klass(holder_klass, field_index, field_klass);
1228 
1229   //check for empty value klass
1230   test_klass_is_empty_value(field_klass, dst_temp, empty_value);
1231 
1232   // allocate buffer
1233   push(obj); // save holder
1234   allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1235 
1236   // Have an oop instance buffer, copy into it
1237   data_for_oop(obj, dst_temp, field_klass);
1238   pop(alloc_temp);             // restore holder
1239   lea(src, Address(alloc_temp, field_offset));
1240   // call_VM_leaf, clobbers a few regs, save restore new obj
1241   push(obj);
1242   access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, field_klass);
1243   pop(obj);
1244   pop(holder_klass);
1245   jmp(done);
1246 
1247   bind(empty_value);
1248   get_empty_value_oop(field_klass, dst_temp, obj);
1249   pop(holder_klass);
1250   jmp(done);
1251 
1252   bind(alloc_failed);
1253   pop(obj);
1254   pop(holder_klass);
1255   call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
1256           obj, field_index, holder_klass);
1257 
1258   bind(done);
1259 }
1260 
1261 void InterpreterMacroAssembler::read_flattened_element(Register array, Register index,
1262                                                        Register t1, Register t2,
1263                                                        Register obj) {
1264   assert_different_registers(array, index, t1, t2);
1265   Label alloc_failed, empty_value, done;
1266   const Register array_klass = t2;
1267   const Register elem_klass = t1;
1268   const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1269   const Register dst_temp   = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1270 
1271   // load in array->klass()->element_klass()
1272   load_klass(array_klass, array);
1273   movptr(elem_klass, Address(array_klass, ArrayKlass::element_klass_offset()));
1274 
1275   //check for empty value klass
1276   test_klass_is_empty_value(elem_klass, dst_temp, empty_value);
1277 
1278   // calc source into "array_klass" and free up some regs
1279   const Register src = array_klass;
1280   push(index); // preserve index reg in case alloc_failed
1281   data_for_value_array_index(array, array_klass, index, src);
1282 
1283   allocate_instance(elem_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1284   // Have an oop instance buffer, copy into it
1285   movptr(Address(rsp, 0), obj); // preserve obj (overwrite index, no longer needed)
1286   data_for_oop(obj, dst_temp, elem_klass);
1287   access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, elem_klass);
1288   pop(obj);
1289   jmp(done);
1290 
1291   bind(empty_value);
1292   get_empty_value_oop(elem_klass, dst_temp, obj);
1293   jmp(done);
1294 
1295   bind(alloc_failed);
1296   pop(index);
1297   if (array == c_rarg2) {
1298     mov(elem_klass, array);
1299     array = elem_klass;
1300   }
1301   call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), array, index);
1302 
1303   bind(done);
1304 }
1305 
1306 
1307 // Lock object
1308 //
1309 // Args:
1310 //      rdx, c_rarg1: BasicObjectLock to be used for locking
1311 //
1312 // Kills:
1313 //      rax, rbx
1314 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1315   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1316          "The argument is only for looks. It must be c_rarg1");
1317 
1318   if (UseHeavyMonitors) {
1319     call_VM(noreg,
1320             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1321             lock_reg);
1322   } else {
1323     Label done;
1324 
1325     const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1326     const Register tmp_reg = rbx; // Will be passed to biased_locking_enter to avoid a
1327                                   // problematic case where tmp_reg = no_reg.
1328     const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1329 
1330     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
1331     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
1332     const int mark_offset = lock_offset +
1333                             BasicLock::displaced_header_offset_in_bytes();
1334 
1335     Label slow_case;
1336 
1337     // Load object pointer into obj_reg
1338     movptr(obj_reg, Address(lock_reg, obj_offset));
1339 
1340     if (UseBiasedLocking) {
1341       biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp_reg, false, done, &slow_case);
1342     }
1343 
1344     // Load immediate 1 into swap_reg %rax
1345     movl(swap_reg, (int32_t)1);
1346 
1347     // Load (object->mark() | 1) into swap_reg %rax
1348     orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1349     if (EnableValhalla && !UseBiasedLocking) {
1350       // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking
1351       andptr(swap_reg, ~((int) markWord::biased_lock_bit_in_place));
1352     }
1353 
1354     // Save (object->mark() | 1) into BasicLock's displaced header
1355     movptr(Address(lock_reg, mark_offset), swap_reg);
1356 
1357     assert(lock_offset == 0,
1358            "displaced header must be first word in BasicObjectLock");
1359 
1360     lock();
1361     cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1362     if (PrintBiasedLockingStatistics) {
1363       cond_inc32(Assembler::zero,
1364                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1365     }
1366     jcc(Assembler::zero, done);
1367 
1368     const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1369 
1370     // Test if the oopMark is an obvious stack pointer, i.e.,
1371     //  1) (mark & zero_bits) == 0, and
1372     //  2) rsp <= mark < mark + os::pagesize()
1373     //
1374     // These 3 tests can be done by evaluating the following
1375     // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1376     // assuming both stack pointer and pagesize have their
1377     // least significant bits clear.
1378     // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
1379     subptr(swap_reg, rsp);
1380     andptr(swap_reg, zero_bits - os::vm_page_size());
1381 
1382     // Save the test result, for recursive case, the result is zero
1383     movptr(Address(lock_reg, mark_offset), swap_reg);
1384 
1385     if (PrintBiasedLockingStatistics) {
1386       cond_inc32(Assembler::zero,
1387                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
1388     }
1389     jcc(Assembler::zero, done);
1390 
1391     bind(slow_case);
1392 
1393     // Call the runtime routine for slow case
1394     call_VM(noreg,
1395             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1396             lock_reg);
1397 
1398     bind(done);
1399   }
1400 }
1401 
1402 
1403 // Unlocks an object. Used in monitorexit bytecode and
1404 // remove_activation.  Throws an IllegalMonitorException if object is
1405 // not locked by current thread.
1406 //
1407 // Args:
1408 //      rdx, c_rarg1: BasicObjectLock for lock
1409 //
1410 // Kills:
1411 //      rax
1412 //      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1413 //      rscratch1 (scratch reg)
1414 // rax, rbx, rcx, rdx
1415 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1416   assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1417          "The argument is only for looks. It must be c_rarg1");
1418 
1419   if (UseHeavyMonitors) {
1420     call_VM(noreg,
1421             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1422             lock_reg);
1423   } else {
1424     Label done;
1425 
1426     const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1427     const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx);  // Will contain the old oopMark
1428     const Register obj_reg    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);  // Will contain the oop
1429 
1430     save_bcp(); // Save in case of exception
1431 
1432     // Convert from BasicObjectLock structure to object and BasicLock
1433     // structure Store the BasicLock address into %rax
1434     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
1435 
1436     // Load oop into obj_reg(%c_rarg3)
1437     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
1438 
1439     // Free entry
1440     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
1441 
1442     if (UseBiasedLocking) {
1443       biased_locking_exit(obj_reg, header_reg, done);
1444     }
1445 
1446     // Load the old header from BasicLock structure
1447     movptr(header_reg, Address(swap_reg,
1448                                BasicLock::displaced_header_offset_in_bytes()));
1449 
1450     // Test for recursion
1451     testptr(header_reg, header_reg);
1452 
1453     // zero for recursive case
1454     jcc(Assembler::zero, done);
1455 
1456     // Atomic swap back the old header
1457     lock();
1458     cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1459 
1460     // zero for simple unlock of a stack-lock case
1461     jcc(Assembler::zero, done);
1462 
1463     // Call the runtime routine for slow case.
1464     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
1465          obj_reg); // restore obj
1466     call_VM(noreg,
1467             CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1468             lock_reg);
1469 
1470     bind(done);
1471 
1472     restore_bcp();
1473   }
1474 }
1475 
1476 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1477                                                          Label& zero_continue) {
1478   assert(ProfileInterpreter, "must be profiling interpreter");
1479   movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1480   testptr(mdp, mdp);
1481   jcc(Assembler::zero, zero_continue);
1482 }
1483 
1484 
1485 // Set the method data pointer for the current bcp.
1486 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1487   assert(ProfileInterpreter, "must be profiling interpreter");
1488   Label set_mdp;
1489   push(rax);
1490   push(rbx);
1491 
1492   get_method(rbx);
1493   // Test MDO to avoid the call if it is NULL.
1494   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1495   testptr(rax, rax);
1496   jcc(Assembler::zero, set_mdp);
1497   // rbx: method
1498   // _bcp_register: bcp
1499   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1500   // rax: mdi
1501   // mdo is guaranteed to be non-zero here, we checked for it before the call.
1502   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1503   addptr(rbx, in_bytes(MethodData::data_offset()));
1504   addptr(rax, rbx);
1505   bind(set_mdp);
1506   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1507   pop(rbx);
1508   pop(rax);
1509 }
1510 
1511 void InterpreterMacroAssembler::verify_method_data_pointer() {
1512   assert(ProfileInterpreter, "must be profiling interpreter");
1513 #ifdef ASSERT
1514   Label verify_continue;
1515   push(rax);
1516   push(rbx);
1517   Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1518   Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1519   push(arg3_reg);
1520   push(arg2_reg);
1521   test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1522   get_method(rbx);
1523 
1524   // If the mdp is valid, it will point to a DataLayout header which is
1525   // consistent with the bcp.  The converse is highly probable also.
1526   load_unsigned_short(arg2_reg,
1527                       Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1528   addptr(arg2_reg, Address(rbx, Method::const_offset()));
1529   lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1530   cmpptr(arg2_reg, _bcp_register);
1531   jcc(Assembler::equal, verify_continue);
1532   // rbx: method
1533   // _bcp_register: bcp
1534   // c_rarg3: mdp
1535   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1536                rbx, _bcp_register, arg3_reg);
1537   bind(verify_continue);
1538   pop(arg2_reg);
1539   pop(arg3_reg);
1540   pop(rbx);
1541   pop(rax);
1542 #endif // ASSERT
1543 }
1544 
1545 
1546 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1547                                                 int constant,
1548                                                 Register value) {
1549   assert(ProfileInterpreter, "must be profiling interpreter");
1550   Address data(mdp_in, constant);
1551   movptr(data, value);
1552 }
1553 
1554 
1555 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1556                                                       int constant,
1557                                                       bool decrement) {
1558   // Counter address
1559   Address data(mdp_in, constant);
1560 
1561   increment_mdp_data_at(data, decrement);
1562 }
1563 
1564 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1565                                                       bool decrement) {
1566   assert(ProfileInterpreter, "must be profiling interpreter");
1567   // %%% this does 64bit counters at best it is wasting space
1568   // at worst it is a rare bug when counters overflow
1569 
1570   if (decrement) {
1571     // Decrement the register.  Set condition codes.
1572     addptr(data, (int32_t) -DataLayout::counter_increment);
1573     // If the decrement causes the counter to overflow, stay negative
1574     Label L;
1575     jcc(Assembler::negative, L);
1576     addptr(data, (int32_t) DataLayout::counter_increment);
1577     bind(L);
1578   } else {
1579     assert(DataLayout::counter_increment == 1,
1580            "flow-free idiom only works with 1");
1581     // Increment the register.  Set carry flag.
1582     addptr(data, DataLayout::counter_increment);
1583     // If the increment causes the counter to overflow, pull back by 1.
1584     sbbptr(data, (int32_t)0);
1585   }
1586 }
1587 
1588 
1589 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1590                                                       Register reg,
1591                                                       int constant,
1592                                                       bool decrement) {
1593   Address data(mdp_in, reg, Address::times_1, constant);
1594 
1595   increment_mdp_data_at(data, decrement);
1596 }
1597 
1598 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1599                                                 int flag_byte_constant) {
1600   assert(ProfileInterpreter, "must be profiling interpreter");
1601   int header_offset = in_bytes(DataLayout::flags_offset());
1602   int header_bits = flag_byte_constant;
1603   // Set the flag
1604   orb(Address(mdp_in, header_offset), header_bits);
1605 }
1606 
1607 
1608 
1609 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1610                                                  int offset,
1611                                                  Register value,
1612                                                  Register test_value_out,
1613                                                  Label& not_equal_continue) {
1614   assert(ProfileInterpreter, "must be profiling interpreter");
1615   if (test_value_out == noreg) {
1616     cmpptr(value, Address(mdp_in, offset));
1617   } else {
1618     // Put the test value into a register, so caller can use it:
1619     movptr(test_value_out, Address(mdp_in, offset));
1620     cmpptr(test_value_out, value);
1621   }
1622   jcc(Assembler::notEqual, not_equal_continue);
1623 }
1624 
1625 
1626 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1627                                                      int offset_of_disp) {
1628   assert(ProfileInterpreter, "must be profiling interpreter");
1629   Address disp_address(mdp_in, offset_of_disp);
1630   addptr(mdp_in, disp_address);
1631   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1632 }
1633 
1634 
1635 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1636                                                      Register reg,
1637                                                      int offset_of_disp) {
1638   assert(ProfileInterpreter, "must be profiling interpreter");
1639   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1640   addptr(mdp_in, disp_address);
1641   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1642 }
1643 
1644 
1645 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1646                                                        int constant) {
1647   assert(ProfileInterpreter, "must be profiling interpreter");
1648   addptr(mdp_in, constant);
1649   movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1650 }
1651 
1652 
1653 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1654   assert(ProfileInterpreter, "must be profiling interpreter");
1655   push(return_bci); // save/restore across call_VM
1656   call_VM(noreg,
1657           CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1658           return_bci);
1659   pop(return_bci);
1660 }
1661 
1662 
1663 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1664                                                      Register bumped_count) {
1665   if (ProfileInterpreter) {
1666     Label profile_continue;
1667 
1668     // If no method data exists, go to profile_continue.
1669     // Otherwise, assign to mdp
1670     test_method_data_pointer(mdp, profile_continue);
1671 
1672     // We are taking a branch.  Increment the taken count.
1673     // We inline increment_mdp_data_at to return bumped_count in a register
1674     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1675     Address data(mdp, in_bytes(JumpData::taken_offset()));
1676     movptr(bumped_count, data);
1677     assert(DataLayout::counter_increment == 1,
1678             "flow-free idiom only works with 1");
1679     addptr(bumped_count, DataLayout::counter_increment);
1680     sbbptr(bumped_count, 0);
1681     movptr(data, bumped_count); // Store back out
1682 
1683     // The method data pointer needs to be updated to reflect the new target.
1684     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1685     bind(profile_continue);
1686   }
1687 }
1688 
1689 
1690 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1691   if (ProfileInterpreter) {
1692     Label profile_continue;
1693 
1694     // If no method data exists, go to profile_continue.
1695     test_method_data_pointer(mdp, profile_continue);
1696 
1697     // We are taking a branch.  Increment the not taken count.
1698     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1699 
1700     // The method data pointer needs to be updated to correspond to
1701     // the next bytecode
1702     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1703     bind(profile_continue);
1704   }
1705 }
1706 
1707 void InterpreterMacroAssembler::profile_call(Register mdp) {
1708   if (ProfileInterpreter) {
1709     Label profile_continue;
1710 
1711     // If no method data exists, go to profile_continue.
1712     test_method_data_pointer(mdp, profile_continue);
1713 
1714     // We are making a call.  Increment the count.
1715     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1716 
1717     // The method data pointer needs to be updated to reflect the new target.
1718     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1719     bind(profile_continue);
1720   }
1721 }
1722 
1723 
1724 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1725   if (ProfileInterpreter) {
1726     Label profile_continue;
1727 
1728     // If no method data exists, go to profile_continue.
1729     test_method_data_pointer(mdp, profile_continue);
1730 
1731     // We are making a call.  Increment the count.
1732     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1733 
1734     // The method data pointer needs to be updated to reflect the new target.
1735     update_mdp_by_constant(mdp,
1736                            in_bytes(VirtualCallData::
1737                                     virtual_call_data_size()));
1738     bind(profile_continue);
1739   }
1740 }
1741 
1742 
1743 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1744                                                      Register mdp,
1745                                                      Register reg2,
1746                                                      bool receiver_can_be_null) {
1747   if (ProfileInterpreter) {
1748     Label profile_continue;
1749 
1750     // If no method data exists, go to profile_continue.
1751     test_method_data_pointer(mdp, profile_continue);
1752 
1753     Label skip_receiver_profile;
1754     if (receiver_can_be_null) {
1755       Label not_null;
1756       testptr(receiver, receiver);
1757       jccb(Assembler::notZero, not_null);
1758       // We are making a call.  Increment the count for null receiver.
1759       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1760       jmp(skip_receiver_profile);
1761       bind(not_null);
1762     }
1763 
1764     // Record the receiver type.
1765     record_klass_in_profile(receiver, mdp, reg2, true);
1766     bind(skip_receiver_profile);
1767 
1768     // The method data pointer needs to be updated to reflect the new target.
1769 #if INCLUDE_JVMCI
1770     if (MethodProfileWidth == 0) {
1771       update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1772     }
1773 #else // INCLUDE_JVMCI
1774     update_mdp_by_constant(mdp,
1775                            in_bytes(VirtualCallData::
1776                                     virtual_call_data_size()));
1777 #endif // INCLUDE_JVMCI
1778     bind(profile_continue);
1779   }
1780 }
1781 
1782 #if INCLUDE_JVMCI
1783 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1784   assert_different_registers(method, mdp, reg2);
1785   if (ProfileInterpreter && MethodProfileWidth > 0) {
1786     Label profile_continue;
1787 
1788     // If no method data exists, go to profile_continue.
1789     test_method_data_pointer(mdp, profile_continue);
1790 
1791     Label done;
1792     record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1793       &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1794     bind(done);
1795 
1796     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1797     bind(profile_continue);
1798   }
1799 }
1800 #endif // INCLUDE_JVMCI
1801 
1802 // This routine creates a state machine for updating the multi-row
1803 // type profile at a virtual call site (or other type-sensitive bytecode).
1804 // The machine visits each row (of receiver/count) until the receiver type
1805 // is found, or until it runs out of rows.  At the same time, it remembers
1806 // the location of the first empty row.  (An empty row records null for its
1807 // receiver, and can be allocated for a newly-observed receiver type.)
1808 // Because there are two degrees of freedom in the state, a simple linear
1809 // search will not work; it must be a decision tree.  Hence this helper
1810 // function is recursive, to generate the required tree structured code.
1811 // It's the interpreter, so we are trading off code space for speed.
1812 // See below for example code.
1813 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1814                                         Register receiver, Register mdp,
1815                                         Register reg2, int start_row,
1816                                         Label& done, bool is_virtual_call) {
1817   if (TypeProfileWidth == 0) {
1818     if (is_virtual_call) {
1819       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1820     }
1821 #if INCLUDE_JVMCI
1822     else if (EnableJVMCI) {
1823       increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1824     }
1825 #endif // INCLUDE_JVMCI
1826   } else {
1827     int non_profiled_offset = -1;
1828     if (is_virtual_call) {
1829       non_profiled_offset = in_bytes(CounterData::count_offset());
1830     }
1831 #if INCLUDE_JVMCI
1832     else if (EnableJVMCI) {
1833       non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1834     }
1835 #endif // INCLUDE_JVMCI
1836 
1837     record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1838         &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1839   }
1840 }
1841 
1842 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1843                                         Register reg2, int start_row, Label& done, int total_rows,
1844                                         OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1845                                         int non_profiled_offset) {
1846   int last_row = total_rows - 1;
1847   assert(start_row <= last_row, "must be work left to do");
1848   // Test this row for both the item and for null.
1849   // Take any of three different outcomes:
1850   //   1. found item => increment count and goto done
1851   //   2. found null => keep looking for case 1, maybe allocate this cell
1852   //   3. found something else => keep looking for cases 1 and 2
1853   // Case 3 is handled by a recursive call.
1854   for (int row = start_row; row <= last_row; row++) {
1855     Label next_test;
1856     bool test_for_null_also = (row == start_row);
1857 
1858     // See if the item is item[n].
1859     int item_offset = in_bytes(item_offset_fn(row));
1860     test_mdp_data_at(mdp, item_offset, item,
1861                      (test_for_null_also ? reg2 : noreg),
1862                      next_test);
1863     // (Reg2 now contains the item from the CallData.)
1864 
1865     // The item is item[n].  Increment count[n].
1866     int count_offset = in_bytes(item_count_offset_fn(row));
1867     increment_mdp_data_at(mdp, count_offset);
1868     jmp(done);
1869     bind(next_test);
1870 
1871     if (test_for_null_also) {
1872       // Failed the equality check on item[n]...  Test for null.
1873       testptr(reg2, reg2);
1874       if (start_row == last_row) {
1875         // The only thing left to do is handle the null case.
1876         if (non_profiled_offset >= 0) {
1877           Label found_null;
1878           jccb(Assembler::zero, found_null);
1879           // Item did not match any saved item and there is no empty row for it.
1880           // Increment total counter to indicate polymorphic case.
1881           increment_mdp_data_at(mdp, non_profiled_offset);
1882           jmp(done);
1883           bind(found_null);
1884         } else {
1885           jcc(Assembler::notZero, done);
1886         }
1887         break;
1888       }
1889       Label found_null;
1890       // Since null is rare, make it be the branch-taken case.
1891       jcc(Assembler::zero, found_null);
1892 
1893       // Put all the "Case 3" tests here.
1894       record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1895         item_offset_fn, item_count_offset_fn, non_profiled_offset);
1896 
1897       // Found a null.  Keep searching for a matching item,
1898       // but remember that this is an empty (unused) slot.
1899       bind(found_null);
1900     }
1901   }
1902 
1903   // In the fall-through case, we found no matching item, but we
1904   // observed the item[start_row] is NULL.
1905 
1906   // Fill in the item field and increment the count.
1907   int item_offset = in_bytes(item_offset_fn(start_row));
1908   set_mdp_data_at(mdp, item_offset, item);
1909   int count_offset = in_bytes(item_count_offset_fn(start_row));
1910   movl(reg2, DataLayout::counter_increment);
1911   set_mdp_data_at(mdp, count_offset, reg2);
1912   if (start_row > 0) {
1913     jmp(done);
1914   }
1915 }
1916 
1917 // Example state machine code for three profile rows:
1918 //   // main copy of decision tree, rooted at row[1]
1919 //   if (row[0].rec == rec) { row[0].incr(); goto done; }
1920 //   if (row[0].rec != NULL) {
1921 //     // inner copy of decision tree, rooted at row[1]
1922 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1923 //     if (row[1].rec != NULL) {
1924 //       // degenerate decision tree, rooted at row[2]
1925 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1926 //       if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1927 //       row[2].init(rec); goto done;
1928 //     } else {
1929 //       // remember row[1] is empty
1930 //       if (row[2].rec == rec) { row[2].incr(); goto done; }
1931 //       row[1].init(rec); goto done;
1932 //     }
1933 //   } else {
1934 //     // remember row[0] is empty
1935 //     if (row[1].rec == rec) { row[1].incr(); goto done; }
1936 //     if (row[2].rec == rec) { row[2].incr(); goto done; }
1937 //     row[0].init(rec); goto done;
1938 //   }
1939 //   done:
1940 
1941 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1942                                                         Register mdp, Register reg2,
1943                                                         bool is_virtual_call) {
1944   assert(ProfileInterpreter, "must be profiling");
1945   Label done;
1946 
1947   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1948 
1949   bind (done);
1950 }
1951 
1952 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1953                                             Register mdp) {
1954   if (ProfileInterpreter) {
1955     Label profile_continue;
1956     uint row;
1957 
1958     // If no method data exists, go to profile_continue.
1959     test_method_data_pointer(mdp, profile_continue);
1960 
1961     // Update the total ret count.
1962     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1963 
1964     for (row = 0; row < RetData::row_limit(); row++) {
1965       Label next_test;
1966 
1967       // See if return_bci is equal to bci[n]:
1968       test_mdp_data_at(mdp,
1969                        in_bytes(RetData::bci_offset(row)),
1970                        return_bci, noreg,
1971                        next_test);
1972 
1973       // return_bci is equal to bci[n].  Increment the count.
1974       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1975 
1976       // The method data pointer needs to be updated to reflect the new target.
1977       update_mdp_by_offset(mdp,
1978                            in_bytes(RetData::bci_displacement_offset(row)));
1979       jmp(profile_continue);
1980       bind(next_test);
1981     }
1982 
1983     update_mdp_for_ret(return_bci);
1984 
1985     bind(profile_continue);
1986   }
1987 }
1988 
1989 
1990 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1991   if (ProfileInterpreter) {
1992     Label profile_continue;
1993 
1994     // If no method data exists, go to profile_continue.
1995     test_method_data_pointer(mdp, profile_continue);
1996 
1997     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1998 
1999     // The method data pointer needs to be updated.
2000     int mdp_delta = in_bytes(BitData::bit_data_size());
2001     if (TypeProfileCasts) {
2002       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
2003     }
2004     update_mdp_by_constant(mdp, mdp_delta);
2005 
2006     bind(profile_continue);
2007   }
2008 }
2009 
2010 
2011 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
2012   if (ProfileInterpreter && TypeProfileCasts) {
2013     Label profile_continue;
2014 
2015     // If no method data exists, go to profile_continue.
2016     test_method_data_pointer(mdp, profile_continue);
2017 
2018     int count_offset = in_bytes(CounterData::count_offset());
2019     // Back up the address, since we have already bumped the mdp.
2020     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
2021 
2022     // *Decrement* the counter.  We expect to see zero or small negatives.
2023     increment_mdp_data_at(mdp, count_offset, true);
2024 
2025     bind (profile_continue);
2026   }
2027 }
2028 
2029 
2030 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
2031   if (ProfileInterpreter) {
2032     Label profile_continue;
2033 
2034     // If no method data exists, go to profile_continue.
2035     test_method_data_pointer(mdp, profile_continue);
2036 
2037     // The method data pointer needs to be updated.
2038     int mdp_delta = in_bytes(BitData::bit_data_size());
2039     if (TypeProfileCasts) {
2040       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
2041 
2042       // Record the object type.
2043       record_klass_in_profile(klass, mdp, reg2, false);
2044       NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
2045       NOT_LP64(restore_locals();)         // Restore EDI
2046     }
2047     update_mdp_by_constant(mdp, mdp_delta);
2048 
2049     bind(profile_continue);
2050   }
2051 }
2052 
2053 
2054 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
2055   if (ProfileInterpreter) {
2056     Label profile_continue;
2057 
2058     // If no method data exists, go to profile_continue.
2059     test_method_data_pointer(mdp, profile_continue);
2060 
2061     // Update the default case count
2062     increment_mdp_data_at(mdp,
2063                           in_bytes(MultiBranchData::default_count_offset()));
2064 
2065     // The method data pointer needs to be updated.
2066     update_mdp_by_offset(mdp,
2067                          in_bytes(MultiBranchData::
2068                                   default_displacement_offset()));
2069 
2070     bind(profile_continue);
2071   }
2072 }
2073 
2074 
2075 void InterpreterMacroAssembler::profile_switch_case(Register index,
2076                                                     Register mdp,
2077                                                     Register reg2) {
2078   if (ProfileInterpreter) {
2079     Label profile_continue;
2080 
2081     // If no method data exists, go to profile_continue.
2082     test_method_data_pointer(mdp, profile_continue);
2083 
2084     // Build the base (index * per_case_size_in_bytes()) +
2085     // case_array_offset_in_bytes()
2086     movl(reg2, in_bytes(MultiBranchData::per_case_size()));
2087     imulptr(index, reg2); // XXX l ?
2088     addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
2089 
2090     // Update the case count
2091     increment_mdp_data_at(mdp,
2092                           index,
2093                           in_bytes(MultiBranchData::relative_count_offset()));
2094 
2095     // The method data pointer needs to be updated.
2096     update_mdp_by_offset(mdp,
2097                          index,
2098                          in_bytes(MultiBranchData::
2099                                   relative_displacement_offset()));
2100 
2101     bind(profile_continue);
2102   }
2103 }
2104 
2105 void InterpreterMacroAssembler::profile_array(Register mdp,
2106                                               Register array,
2107                                               Register tmp) {
2108   if (ProfileInterpreter) {
2109     Label profile_continue;
2110 
2111     // If no method data exists, go to profile_continue.
2112     test_method_data_pointer(mdp, profile_continue);
2113 
2114     mov(tmp, array);
2115     profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadStoreData::array_offset())));
2116 
2117     Label not_flat;
2118     test_non_flattened_array_oop(array, tmp, not_flat);
2119 
2120     set_mdp_flag_at(mdp, ArrayLoadStoreData::flat_array_byte_constant());
2121 
2122     bind(not_flat);
2123 
2124     Label not_null_free;
2125     test_non_null_free_array_oop(array, tmp, not_null_free);
2126 
2127     set_mdp_flag_at(mdp, ArrayLoadStoreData::null_free_array_byte_constant());
2128 
2129     bind(not_null_free);
2130 
2131     bind(profile_continue);
2132   }
2133 }
2134 
2135 void InterpreterMacroAssembler::profile_element(Register mdp,
2136                                                 Register element,
2137                                                 Register tmp) {
2138   if (ProfileInterpreter) {
2139     Label profile_continue;
2140 
2141     // If no method data exists, go to profile_continue.
2142     test_method_data_pointer(mdp, profile_continue);
2143 
2144     mov(tmp, element);
2145     profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadStoreData::element_offset())));
2146 
2147     // The method data pointer needs to be updated.
2148     update_mdp_by_constant(mdp, in_bytes(ArrayLoadStoreData::array_load_store_data_size()));
2149 
2150     bind(profile_continue);
2151   }
2152 }
2153 
2154 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
2155   if (state == atos) {
2156     MacroAssembler::verify_oop(reg);
2157   }
2158 }
2159 
2160 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2161 #ifndef _LP64
2162   if ((state == ftos && UseSSE < 1) ||
2163       (state == dtos && UseSSE < 2)) {
2164     MacroAssembler::verify_FPU(stack_depth);
2165   }
2166 #endif
2167 }
2168 
2169 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2170 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2171                                                         int increment, Address mask,
2172                                                         Register scratch, bool preloaded,
2173                                                         Condition cond, Label* where) {
2174   if (!preloaded) {
2175     movl(scratch, counter_addr);
2176   }
2177   incrementl(scratch, increment);
2178   movl(counter_addr, scratch);
2179   andl(scratch, mask);
2180   if (where != NULL) {
2181     jcc(cond, *where);
2182   }
2183 }
2184 
2185 void InterpreterMacroAssembler::notify_method_entry() {
2186   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2187   // track stack depth.  If it is possible to enter interp_only_mode we add
2188   // the code to check if the event should be sent.
2189   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2190   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2191   if (JvmtiExport::can_post_interpreter_events()) {
2192     Label L;
2193     NOT_LP64(get_thread(rthread);)
2194     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2195     testl(rdx, rdx);
2196     jcc(Assembler::zero, L);
2197     call_VM(noreg, CAST_FROM_FN_PTR(address,
2198                                     InterpreterRuntime::post_method_entry));
2199     bind(L);
2200   }
2201 
2202   {
2203     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2204     NOT_LP64(get_thread(rthread);)
2205     get_method(rarg);
2206     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2207                  rthread, rarg);
2208   }
2209 
2210   // RedefineClasses() tracing support for obsolete method entry
2211   if (log_is_enabled(Trace, redefine, class, obsolete)) {
2212     NOT_LP64(get_thread(rthread);)
2213     get_method(rarg);
2214     call_VM_leaf(
2215       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2216       rthread, rarg);
2217   }
2218 }
2219 
2220 
2221 void InterpreterMacroAssembler::notify_method_exit(
2222     TosState state, NotifyMethodExitMode mode) {
2223   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2224   // track stack depth.  If it is possible to enter interp_only_mode we add
2225   // the code to check if the event should be sent.
2226   Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2227   Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2228   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2229     Label L;
2230     // Note: frame::interpreter_frame_result has a dependency on how the
2231     // method result is saved across the call to post_method_exit. If this
2232     // is changed then the interpreter_frame_result implementation will
2233     // need to be updated too.
2234 
2235     // template interpreter will leave the result on the top of the stack.
2236     push(state);
2237     NOT_LP64(get_thread(rthread);)
2238     movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2239     testl(rdx, rdx);
2240     jcc(Assembler::zero, L);
2241     call_VM(noreg,
2242             CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2243     bind(L);
2244     pop(state);
2245   }
2246 
2247   {
2248     SkipIfEqual skip(this, &DTraceMethodProbes, false);
2249     push(state);
2250     NOT_LP64(get_thread(rthread);)
2251     get_method(rarg);
2252     call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2253                  rthread, rarg);
2254     pop(state);
2255   }
2256 }