jdk

1
/*
2
 * Copyright (c) 1997, 2024, 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 "compiler/compiler_globals.hpp"
27
#include "interp_masm_x86.hpp"
28
#include "interpreter/interpreter.hpp"
29
#include "interpreter/interpreterRuntime.hpp"
30
#include "logging/log.hpp"
31
#include "oops/arrayOop.hpp"
32
#include "oops/markWord.hpp"
33
#include "oops/methodData.hpp"
34
#include "oops/method.hpp"
35
#include "oops/resolvedFieldEntry.hpp"
36
#include "oops/resolvedIndyEntry.hpp"
37
#include "oops/resolvedMethodEntry.hpp"
38
#include "prims/jvmtiExport.hpp"
39
#include "prims/jvmtiThreadState.hpp"
40
#include "runtime/basicLock.hpp"
41
#include "runtime/frame.inline.hpp"
42
#include "runtime/javaThread.hpp"
43
#include "runtime/safepointMechanism.hpp"
44
#include "runtime/sharedRuntime.hpp"
45
#include "utilities/powerOfTwo.hpp"
46

47
// Implementation of InterpreterMacroAssembler
48

49
void InterpreterMacroAssembler::jump_to_entry(address entry) {
50
  assert(entry, "Entry must have been generated by now");
51
  jump(RuntimeAddress(entry));
52
}
53

54
void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
55
  Label update, next, none;
56

57
#ifdef _LP64
58
  assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
59
#else
60
  assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
61
#endif
62

63
  interp_verify_oop(obj, atos);
64

65
  testptr(obj, obj);
66
  jccb(Assembler::notZero, update);
67
  testptr(mdo_addr, TypeEntries::null_seen);
68
  jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
69
  // atomic update to prevent overwriting Klass* with 0
70
  lock();
71
  orptr(mdo_addr, TypeEntries::null_seen);
72
  jmpb(next);
73

74
  bind(update);
75
  load_klass(obj, obj, rscratch1);
76
#ifdef _LP64
77
  mov(rscratch1, obj);
78
#endif
79

80
  xorptr(obj, mdo_addr);
81
  testptr(obj, TypeEntries::type_klass_mask);
82
  jccb(Assembler::zero, next); // klass seen before, nothing to
83
                               // do. The unknown bit may have been
84
                               // set already but no need to check.
85

86
  testptr(obj, TypeEntries::type_unknown);
87
  jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
88

89
  cmpptr(mdo_addr, 0);
90
  jccb(Assembler::equal, none);
91
  cmpptr(mdo_addr, TypeEntries::null_seen);
92
  jccb(Assembler::equal, none);
93
#ifdef _LP64
94
  // There is a chance that the checks above (re-reading profiling
95
  // data from memory) fail if another thread has just set the
96
  // profiling to this obj's klass
97
  mov(obj, rscratch1);
98
  xorptr(obj, mdo_addr);
99
  testptr(obj, TypeEntries::type_klass_mask);
100
  jccb(Assembler::zero, next);
101
#endif
102

103
  // different than before. Cannot keep accurate profile.
104
  orptr(mdo_addr, TypeEntries::type_unknown);
105
  jmpb(next);
106

107
  bind(none);
108
  // first time here. Set profile type.
109
  movptr(mdo_addr, obj);
110
#ifdef ASSERT
111
  andptr(obj, TypeEntries::type_klass_mask);
112
  verify_klass_ptr(obj);
113
#endif
114

115
  bind(next);
116
}
117

118
void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
119
  if (!ProfileInterpreter) {
120
    return;
121
  }
122

123
  if (MethodData::profile_arguments() || MethodData::profile_return()) {
124
    Label profile_continue;
125

126
    test_method_data_pointer(mdp, profile_continue);
127

128
    int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
129

130
    cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
131
    jcc(Assembler::notEqual, profile_continue);
132

133
    if (MethodData::profile_arguments()) {
134
      Label done;
135
      int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
136
      addptr(mdp, off_to_args);
137

138
      for (int i = 0; i < TypeProfileArgsLimit; i++) {
139
        if (i > 0 || MethodData::profile_return()) {
140
          // If return value type is profiled we may have no argument to profile
141
          movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
142
          subl(tmp, i*TypeStackSlotEntries::per_arg_count());
143
          cmpl(tmp, TypeStackSlotEntries::per_arg_count());
144
          jcc(Assembler::less, done);
145
        }
146
        movptr(tmp, Address(callee, Method::const_offset()));
147
        load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
148
        // stack offset o (zero based) from the start of the argument
149
        // list, for n arguments translates into offset n - o - 1 from
150
        // the end of the argument list
151
        subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
152
        subl(tmp, 1);
153
        Address arg_addr = argument_address(tmp);
154
        movptr(tmp, arg_addr);
155

156
        Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
157
        profile_obj_type(tmp, mdo_arg_addr);
158

159
        int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
160
        addptr(mdp, to_add);
161
        off_to_args += to_add;
162
      }
163

164
      if (MethodData::profile_return()) {
165
        movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
166
        subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
167
      }
168

169
      bind(done);
170

171
      if (MethodData::profile_return()) {
172
        // We're right after the type profile for the last
173
        // argument. tmp is the number of cells left in the
174
        // CallTypeData/VirtualCallTypeData to reach its end. Non null
175
        // if there's a return to profile.
176
        assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
177
        shll(tmp, log2i_exact((int)DataLayout::cell_size));
178
        addptr(mdp, tmp);
179
      }
180
      movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
181
    } else {
182
      assert(MethodData::profile_return(), "either profile call args or call ret");
183
      update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
184
    }
185

186
    // mdp points right after the end of the
187
    // CallTypeData/VirtualCallTypeData, right after the cells for the
188
    // return value type if there's one
189

190
    bind(profile_continue);
191
  }
192
}
193

194
void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
195
  assert_different_registers(mdp, ret, tmp, _bcp_register);
196
  if (ProfileInterpreter && MethodData::profile_return()) {
197
    Label profile_continue;
198

199
    test_method_data_pointer(mdp, profile_continue);
200

201
    if (MethodData::profile_return_jsr292_only()) {
202
      assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
203

204
      // If we don't profile all invoke bytecodes we must make sure
205
      // it's a bytecode we indeed profile. We can't go back to the
206
      // beginning of the ProfileData we intend to update to check its
207
      // type because we're right after it and we don't known its
208
      // length
209
      Label do_profile;
210
      cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
211
      jcc(Assembler::equal, do_profile);
212
      cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
213
      jcc(Assembler::equal, do_profile);
214
      get_method(tmp);
215
      cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
216
      jcc(Assembler::notEqual, profile_continue);
217

218
      bind(do_profile);
219
    }
220

221
    Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
222
    mov(tmp, ret);
223
    profile_obj_type(tmp, mdo_ret_addr);
224

225
    bind(profile_continue);
226
  }
227
}
228

229
void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
230
  if (ProfileInterpreter && MethodData::profile_parameters()) {
231
    Label profile_continue;
232

233
    test_method_data_pointer(mdp, profile_continue);
234

235
    // Load the offset of the area within the MDO used for
236
    // parameters. If it's negative we're not profiling any parameters
237
    movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
238
    testl(tmp1, tmp1);
239
    jcc(Assembler::negative, profile_continue);
240

241
    // Compute a pointer to the area for parameters from the offset
242
    // and move the pointer to the slot for the last
243
    // parameters. Collect profiling from last parameter down.
244
    // mdo start + parameters offset + array length - 1
245
    addptr(mdp, tmp1);
246
    movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
247
    decrement(tmp1, TypeStackSlotEntries::per_arg_count());
248

249
    Label loop;
250
    bind(loop);
251

252
    int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
253
    int type_base = in_bytes(ParametersTypeData::type_offset(0));
254
    Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
255
    Address arg_off(mdp, tmp1, per_arg_scale, off_base);
256
    Address arg_type(mdp, tmp1, per_arg_scale, type_base);
257

258
    // load offset on the stack from the slot for this parameter
259
    movptr(tmp2, arg_off);
260
    negptr(tmp2);
261
    // read the parameter from the local area
262
    movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
263

264
    // profile the parameter
265
    profile_obj_type(tmp2, arg_type);
266

267
    // go to next parameter
268
    decrement(tmp1, TypeStackSlotEntries::per_arg_count());
269
    jcc(Assembler::positive, loop);
270

271
    bind(profile_continue);
272
  }
273
}
274

275
void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
276
                                                  int number_of_arguments) {
277
  // interpreter specific
278
  //
279
  // Note: No need to save/restore bcp & locals registers
280
  //       since these are callee saved registers and no blocking/
281
  //       GC can happen in leaf calls.
282
  // Further Note: DO NOT save/restore bcp/locals. If a caller has
283
  // already saved them so that it can use rsi/rdi as temporaries
284
  // then a save/restore here will DESTROY the copy the caller
285
  // saved! There used to be a save_bcp() that only happened in
286
  // the ASSERT path (no restore_bcp). Which caused bizarre failures
287
  // when jvm built with ASSERTs.
288
#ifdef ASSERT
289
  {
290
    Label L;
291
    cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
292
    jcc(Assembler::equal, L);
293
    stop("InterpreterMacroAssembler::call_VM_leaf_base:"
294
         " last_sp != null");
295
    bind(L);
296
  }
297
#endif
298
  // super call
299
  MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
300
  // interpreter specific
301
  // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
302
  // but since they may not have been saved (and we don't want to
303
  // save them here (see note above) the assert is invalid.
304
}
305

306
void InterpreterMacroAssembler::call_VM_base(Register oop_result,
307
                                             Register java_thread,
308
                                             Register last_java_sp,
309
                                             address  entry_point,
310
                                             int      number_of_arguments,
311
                                             bool     check_exceptions) {
312
  // interpreter specific
313
  //
314
  // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
315
  //       really make a difference for these runtime calls, since they are
316
  //       slow anyway. Btw., bcp must be saved/restored since it may change
317
  //       due to GC.
318
  NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
319
  save_bcp();
320
#ifdef ASSERT
321
  {
322
    Label L;
323
    cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
324
    jcc(Assembler::equal, L);
325
    stop("InterpreterMacroAssembler::call_VM_base:"
326
         " last_sp isn't null");
327
    bind(L);
328
  }
329
#endif /* ASSERT */
330
  // super call
331
  MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
332
                               entry_point, number_of_arguments,
333
                               check_exceptions);
334
  // interpreter specific
335
  restore_bcp();
336
  restore_locals();
337
}
338

339
void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
340
  if (JvmtiExport::can_pop_frame()) {
341
    Label L;
342
    // Initiate popframe handling only if it is not already being
343
    // processed.  If the flag has the popframe_processing bit set, it
344
    // means that this code is called *during* popframe handling - we
345
    // don't want to reenter.
346
    // This method is only called just after the call into the vm in
347
    // call_VM_base, so the arg registers are available.
348
    Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
349
                        LP64_ONLY(c_rarg0);
350
    movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
351
    testl(pop_cond, JavaThread::popframe_pending_bit);
352
    jcc(Assembler::zero, L);
353
    testl(pop_cond, JavaThread::popframe_processing_bit);
354
    jcc(Assembler::notZero, L);
355
    // Call Interpreter::remove_activation_preserving_args_entry() to get the
356
    // address of the same-named entrypoint in the generated interpreter code.
357
    call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
358
    jmp(rax);
359
    bind(L);
360
    NOT_LP64(get_thread(java_thread);)
361
  }
362
}
363

364
void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
365
  Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
366
  NOT_LP64(get_thread(thread);)
367
  movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
368
  const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
369
  const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
370
  const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
371
#ifdef _LP64
372
  switch (state) {
373
    case atos: movptr(rax, oop_addr);
374
               movptr(oop_addr, NULL_WORD);
375
               interp_verify_oop(rax, state);         break;
376
    case ltos: movptr(rax, val_addr);                 break;
377
    case btos:                                   // fall through
378
    case ztos:                                   // fall through
379
    case ctos:                                   // fall through
380
    case stos:                                   // fall through
381
    case itos: movl(rax, val_addr);                 break;
382
    case ftos: load_float(val_addr);                break;
383
    case dtos: load_double(val_addr);               break;
384
    case vtos: /* nothing to do */                  break;
385
    default  : ShouldNotReachHere();
386
  }
387
  // Clean up tos value in the thread object
388
  movl(tos_addr, ilgl);
389
  movl(val_addr, NULL_WORD);
390
#else
391
  const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
392
                             + in_ByteSize(wordSize));
393
  switch (state) {
394
    case atos: movptr(rax, oop_addr);
395
               movptr(oop_addr, NULL_WORD);
396
               interp_verify_oop(rax, state);         break;
397
    case ltos:
398
               movl(rdx, val_addr1);               // fall through
399
    case btos:                                     // fall through
400
    case ztos:                                     // fall through
401
    case ctos:                                     // fall through
402
    case stos:                                     // fall through
403
    case itos: movl(rax, val_addr);                   break;
404
    case ftos: load_float(val_addr);                  break;
405
    case dtos: load_double(val_addr);                 break;
406
    case vtos: /* nothing to do */                    break;
407
    default  : ShouldNotReachHere();
408
  }
409
#endif // _LP64
410
  // Clean up tos value in the thread object
411
  movl(tos_addr, ilgl);
412
  movptr(val_addr, NULL_WORD);
413
  NOT_LP64(movptr(val_addr1, NULL_WORD);)
414
}
415

416

417
void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
418
  if (JvmtiExport::can_force_early_return()) {
419
    Label L;
420
    Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
421
    Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
422

423
    movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
424
    testptr(tmp, tmp);
425
    jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
426

427
    // Initiate earlyret handling only if it is not already being processed.
428
    // If the flag has the earlyret_processing bit set, it means that this code
429
    // is called *during* earlyret handling - we don't want to reenter.
430
    movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
431
    cmpl(tmp, JvmtiThreadState::earlyret_pending);
432
    jcc(Assembler::notEqual, L);
433

434
    // Call Interpreter::remove_activation_early_entry() to get the address of the
435
    // same-named entrypoint in the generated interpreter code.
436
    NOT_LP64(get_thread(java_thread);)
437
    movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
438
#ifdef _LP64
439
    movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
440
    call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
441
#else
442
    pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
443
    call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
444
#endif // _LP64
445
    jmp(rax);
446
    bind(L);
447
    NOT_LP64(get_thread(java_thread);)
448
  }
449
}
450

451
void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
452
  assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
453
  load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
454
  bswapl(reg);
455
  shrl(reg, 16);
456
}
457

458
void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
459
                                                       int bcp_offset,
460
                                                       size_t index_size) {
461
  assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
462
  if (index_size == sizeof(u2)) {
463
    load_unsigned_short(index, Address(_bcp_register, bcp_offset));
464
  } else if (index_size == sizeof(u4)) {
465
    movl(index, Address(_bcp_register, bcp_offset));
466
  } else if (index_size == sizeof(u1)) {
467
    load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
468
  } else {
469
    ShouldNotReachHere();
470
  }
471
}
472

473
// Load object from cpool->resolved_references(index)
474
void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
475
                                                                 Register index,
476
                                                                 Register tmp) {
477
  assert_different_registers(result, index);
478

479
  get_constant_pool(result);
480
  // load pointer for resolved_references[] objArray
481
  movptr(result, Address(result, ConstantPool::cache_offset()));
482
  movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
483
  resolve_oop_handle(result, tmp);
484
  load_heap_oop(result, Address(result, index,
485
                                UseCompressedOops ? Address::times_4 : Address::times_ptr,
486
                                arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
487
}
488

489
// load cpool->resolved_klass_at(index)
490
void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
491
                                                             Register cpool,
492
                                                             Register index) {
493
  assert_different_registers(cpool, index);
494

495
  movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
496
  Register resolved_klasses = cpool;
497
  movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
498
  movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
499
}
500

501
// Generate a subtype check: branch to ok_is_subtype if sub_klass is a
502
// subtype of super_klass.
503
//
504
// Args:
505
//      rax: superklass
506
//      Rsub_klass: subklass
507
//
508
// Kills:
509
//      rcx, rdi
510
void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
511
                                                  Label& ok_is_subtype) {
512
  assert(Rsub_klass != rax, "rax holds superklass");
513
  LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
514
  LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
515
  assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
516
  assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
517

518
  // Profile the not-null value's klass.
519
  profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
520

521
  // Do the check.
522
  check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
523
}
524

525

526
#ifndef _LP64
527
void InterpreterMacroAssembler::f2ieee() {
528
  if (IEEEPrecision) {
529
    fstp_s(Address(rsp, 0));
530
    fld_s(Address(rsp, 0));
531
  }
532
}
533

534

535
void InterpreterMacroAssembler::d2ieee() {
536
  if (IEEEPrecision) {
537
    fstp_d(Address(rsp, 0));
538
    fld_d(Address(rsp, 0));
539
  }
540
}
541
#endif // _LP64
542

543
// Java Expression Stack
544

545
void InterpreterMacroAssembler::pop_ptr(Register r) {
546
  pop(r);
547
}
548

549
void InterpreterMacroAssembler::push_ptr(Register r) {
550
  push(r);
551
}
552

553
void InterpreterMacroAssembler::push_i(Register r) {
554
  push(r);
555
}
556

557
void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
558
  push(r);
559
}
560

561
void InterpreterMacroAssembler::push_f(XMMRegister r) {
562
  subptr(rsp, wordSize);
563
  movflt(Address(rsp, 0), r);
564
}
565

566
void InterpreterMacroAssembler::pop_f(XMMRegister r) {
567
  movflt(r, Address(rsp, 0));
568
  addptr(rsp, wordSize);
569
}
570

571
void InterpreterMacroAssembler::push_d(XMMRegister r) {
572
  subptr(rsp, 2 * wordSize);
573
  movdbl(Address(rsp, 0), r);
574
}
575

576
void InterpreterMacroAssembler::pop_d(XMMRegister r) {
577
  movdbl(r, Address(rsp, 0));
578
  addptr(rsp, 2 * Interpreter::stackElementSize);
579
}
580

581
#ifdef _LP64
582
void InterpreterMacroAssembler::pop_i(Register r) {
583
  // XXX can't use pop currently, upper half non clean
584
  movl(r, Address(rsp, 0));
585
  addptr(rsp, wordSize);
586
}
587

588
void InterpreterMacroAssembler::pop_l(Register r) {
589
  movq(r, Address(rsp, 0));
590
  addptr(rsp, 2 * Interpreter::stackElementSize);
591
}
592

593
void InterpreterMacroAssembler::push_l(Register r) {
594
  subptr(rsp, 2 * wordSize);
595
  movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r         );
596
  movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
597
}
598

599
void InterpreterMacroAssembler::pop(TosState state) {
600
  switch (state) {
601
  case atos: pop_ptr();                 break;
602
  case btos:
603
  case ztos:
604
  case ctos:
605
  case stos:
606
  case itos: pop_i();                   break;
607
  case ltos: pop_l();                   break;
608
  case ftos: pop_f(xmm0);               break;
609
  case dtos: pop_d(xmm0);               break;
610
  case vtos: /* nothing to do */        break;
611
  default:   ShouldNotReachHere();
612
  }
613
  interp_verify_oop(rax, state);
614
}
615

616
void InterpreterMacroAssembler::push(TosState state) {
617
  interp_verify_oop(rax, state);
618
  switch (state) {
619
  case atos: push_ptr();                break;
620
  case btos:
621
  case ztos:
622
  case ctos:
623
  case stos:
624
  case itos: push_i();                  break;
625
  case ltos: push_l();                  break;
626
  case ftos: push_f(xmm0);              break;
627
  case dtos: push_d(xmm0);              break;
628
  case vtos: /* nothing to do */        break;
629
  default  : ShouldNotReachHere();
630
  }
631
}
632
#else
633
void InterpreterMacroAssembler::pop_i(Register r) {
634
  pop(r);
635
}
636

637
void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
638
  pop(lo);
639
  pop(hi);
640
}
641

642
void InterpreterMacroAssembler::pop_f() {
643
  fld_s(Address(rsp, 0));
644
  addptr(rsp, 1 * wordSize);
645
}
646

647
void InterpreterMacroAssembler::pop_d() {
648
  fld_d(Address(rsp, 0));
649
  addptr(rsp, 2 * wordSize);
650
}
651

652

653
void InterpreterMacroAssembler::pop(TosState state) {
654
  switch (state) {
655
    case atos: pop_ptr(rax);                                 break;
656
    case btos:                                               // fall through
657
    case ztos:                                               // fall through
658
    case ctos:                                               // fall through
659
    case stos:                                               // fall through
660
    case itos: pop_i(rax);                                   break;
661
    case ltos: pop_l(rax, rdx);                              break;
662
    case ftos:
663
      if (UseSSE >= 1) {
664
        pop_f(xmm0);
665
      } else {
666
        pop_f();
667
      }
668
      break;
669
    case dtos:
670
      if (UseSSE >= 2) {
671
        pop_d(xmm0);
672
      } else {
673
        pop_d();
674
      }
675
      break;
676
    case vtos: /* nothing to do */                           break;
677
    default  : ShouldNotReachHere();
678
  }
679
  interp_verify_oop(rax, state);
680
}
681

682

683
void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
684
  push(hi);
685
  push(lo);
686
}
687

688
void InterpreterMacroAssembler::push_f() {
689
  // Do not schedule for no AGI! Never write beyond rsp!
690
  subptr(rsp, 1 * wordSize);
691
  fstp_s(Address(rsp, 0));
692
}
693

694
void InterpreterMacroAssembler::push_d() {
695
  // Do not schedule for no AGI! Never write beyond rsp!
696
  subptr(rsp, 2 * wordSize);
697
  fstp_d(Address(rsp, 0));
698
}
699

700

701
void InterpreterMacroAssembler::push(TosState state) {
702
  interp_verify_oop(rax, state);
703
  switch (state) {
704
    case atos: push_ptr(rax); break;
705
    case btos:                                               // fall through
706
    case ztos:                                               // fall through
707
    case ctos:                                               // fall through
708
    case stos:                                               // fall through
709
    case itos: push_i(rax);                                    break;
710
    case ltos: push_l(rax, rdx);                               break;
711
    case ftos:
712
      if (UseSSE >= 1) {
713
        push_f(xmm0);
714
      } else {
715
        push_f();
716
      }
717
      break;
718
    case dtos:
719
      if (UseSSE >= 2) {
720
        push_d(xmm0);
721
      } else {
722
        push_d();
723
      }
724
      break;
725
    case vtos: /* nothing to do */                             break;
726
    default  : ShouldNotReachHere();
727
  }
728
}
729
#endif // _LP64
730

731

732
// Helpers for swap and dup
733
void InterpreterMacroAssembler::load_ptr(int n, Register val) {
734
  movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
735
}
736

737
void InterpreterMacroAssembler::store_ptr(int n, Register val) {
738
  movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
739
}
740

741

742
void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
743
  // set sender sp
744
  lea(_bcp_register, Address(rsp, wordSize));
745
  // record last_sp
746
  mov(rcx, _bcp_register);
747
  subptr(rcx, rbp);
748
  sarptr(rcx, LogBytesPerWord);
749
  movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
750
}
751

752

753
// Jump to from_interpreted entry of a call unless single stepping is possible
754
// in this thread in which case we must call the i2i entry
755
void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
756
  prepare_to_jump_from_interpreted();
757

758
  if (JvmtiExport::can_post_interpreter_events()) {
759
    Label run_compiled_code;
760
    // JVMTI events, such as single-stepping, are implemented partly by avoiding running
761
    // compiled code in threads for which the event is enabled.  Check here for
762
    // interp_only_mode if these events CAN be enabled.
763
    // interp_only is an int, on little endian it is sufficient to test the byte only
764
    // Is a cmpl faster?
765
    LP64_ONLY(temp = r15_thread;)
766
    NOT_LP64(get_thread(temp);)
767
    cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
768
    jccb(Assembler::zero, run_compiled_code);
769
    jmp(Address(method, Method::interpreter_entry_offset()));
770
    bind(run_compiled_code);
771
  }
772

773
  jmp(Address(method, Method::from_interpreted_offset()));
774
}
775

776
// The following two routines provide a hook so that an implementation
777
// can schedule the dispatch in two parts.  x86 does not do this.
778
void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
779
  // Nothing x86 specific to be done here
780
}
781

782
void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
783
  dispatch_next(state, step);
784
}
785

786
void InterpreterMacroAssembler::dispatch_base(TosState state,
787
                                              address* table,
788
                                              bool verifyoop,
789
                                              bool generate_poll) {
790
  verify_FPU(1, state);
791
  if (VerifyActivationFrameSize) {
792
    Label L;
793
    mov(rcx, rbp);
794
    subptr(rcx, rsp);
795
    int32_t min_frame_size =
796
      (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
797
      wordSize;
798
    cmpptr(rcx, min_frame_size);
799
    jcc(Assembler::greaterEqual, L);
800
    stop("broken stack frame");
801
    bind(L);
802
  }
803
  if (verifyoop) {
804
    interp_verify_oop(rax, state);
805
  }
806

807
  address* const safepoint_table = Interpreter::safept_table(state);
808
#ifdef _LP64
809
  Label no_safepoint, dispatch;
810
  if (table != safepoint_table && generate_poll) {
811
    NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
812
    testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
813

814
    jccb(Assembler::zero, no_safepoint);
815
    lea(rscratch1, ExternalAddress((address)safepoint_table));
816
    jmpb(dispatch);
817
  }
818

819
  bind(no_safepoint);
820
  lea(rscratch1, ExternalAddress((address)table));
821
  bind(dispatch);
822
  jmp(Address(rscratch1, rbx, Address::times_8));
823

824
#else
825
  Address index(noreg, rbx, Address::times_ptr);
826
  if (table != safepoint_table && generate_poll) {
827
    NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
828
    Label no_safepoint;
829
    const Register thread = rcx;
830
    get_thread(thread);
831
    testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
832

833
    jccb(Assembler::zero, no_safepoint);
834
    ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
835
    jump(dispatch_addr, noreg);
836
    bind(no_safepoint);
837
  }
838

839
  {
840
    ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
841
    jump(dispatch_addr, noreg);
842
  }
843
#endif // _LP64
844
}
845

846
void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
847
  dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
848
}
849

850
void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
851
  dispatch_base(state, Interpreter::normal_table(state));
852
}
853

854
void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
855
  dispatch_base(state, Interpreter::normal_table(state), false);
856
}
857

858

859
void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
860
  // load next bytecode (load before advancing _bcp_register to prevent AGI)
861
  load_unsigned_byte(rbx, Address(_bcp_register, step));
862
  // advance _bcp_register
863
  increment(_bcp_register, step);
864
  dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
865
}
866

867
void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
868
  // load current bytecode
869
  load_unsigned_byte(rbx, Address(_bcp_register, 0));
870
  dispatch_base(state, table);
871
}
872

873
void InterpreterMacroAssembler::narrow(Register result) {
874

875
  // Get method->_constMethod->_result_type
876
  movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
877
  movptr(rcx, Address(rcx, Method::const_offset()));
878
  load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
879

880
  Label done, notBool, notByte, notChar;
881

882
  // common case first
883
  cmpl(rcx, T_INT);
884
  jcc(Assembler::equal, done);
885

886
  // mask integer result to narrower return type.
887
  cmpl(rcx, T_BOOLEAN);
888
  jcc(Assembler::notEqual, notBool);
889
  andl(result, 0x1);
890
  jmp(done);
891

892
  bind(notBool);
893
  cmpl(rcx, T_BYTE);
894
  jcc(Assembler::notEqual, notByte);
895
  LP64_ONLY(movsbl(result, result);)
896
  NOT_LP64(shll(result, 24);)      // truncate upper 24 bits
897
  NOT_LP64(sarl(result, 24);)      // and sign-extend byte
898
  jmp(done);
899

900
  bind(notByte);
901
  cmpl(rcx, T_CHAR);
902
  jcc(Assembler::notEqual, notChar);
903
  LP64_ONLY(movzwl(result, result);)
904
  NOT_LP64(andl(result, 0xFFFF);)  // truncate upper 16 bits
905
  jmp(done);
906

907
  bind(notChar);
908
  // cmpl(rcx, T_SHORT);  // all that's left
909
  // jcc(Assembler::notEqual, done);
910
  LP64_ONLY(movswl(result, result);)
911
  NOT_LP64(shll(result, 16);)      // truncate upper 16 bits
912
  NOT_LP64(sarl(result, 16);)      // and sign-extend short
913

914
  // Nothing to do for T_INT
915
  bind(done);
916
}
917

918
// remove activation
919
//
920
// Apply stack watermark barrier.
921
// Unlock the receiver if this is a synchronized method.
922
// Unlock any Java monitors from synchronized blocks.
923
// Remove the activation from the stack.
924
//
925
// If there are locked Java monitors
926
//    If throw_monitor_exception
927
//       throws IllegalMonitorStateException
928
//    Else if install_monitor_exception
929
//       installs IllegalMonitorStateException
930
//    Else
931
//       no error processing
932
void InterpreterMacroAssembler::remove_activation(
933
        TosState state,
934
        Register ret_addr,
935
        bool throw_monitor_exception,
936
        bool install_monitor_exception,
937
        bool notify_jvmdi) {
938
  // Note: Registers rdx xmm0 may be in use for the
939
  // result check if synchronized method
940
  Label unlocked, unlock, no_unlock;
941

942
  const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
943
  const Register robj    = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
944
  const Register rmon    = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
945
                              // monitor pointers need different register
946
                              // because rdx may have the result in it
947
  NOT_LP64(get_thread(rthread);)
948

949
  // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
950
  // that would normally not be safe to use. Such bad returns into unsafe territory of
951
  // the stack, will call InterpreterRuntime::at_unwind.
952
  Label slow_path;
953
  Label fast_path;
954
  safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
955
  jmp(fast_path);
956
  bind(slow_path);
957
  push(state);
958
  set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
959
  super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
960
  NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
961
  reset_last_Java_frame(rthread, true);
962
  pop(state);
963
  bind(fast_path);
964

965
  // get the value of _do_not_unlock_if_synchronized into rdx
966
  const Address do_not_unlock_if_synchronized(rthread,
967
    in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
968
  movbool(rbx, do_not_unlock_if_synchronized);
969
  movbool(do_not_unlock_if_synchronized, false); // reset the flag
970

971
 // get method access flags
972
  movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
973
  movl(rcx, Address(rcx, Method::access_flags_offset()));
974
  testl(rcx, JVM_ACC_SYNCHRONIZED);
975
  jcc(Assembler::zero, unlocked);
976

977
  // Don't unlock anything if the _do_not_unlock_if_synchronized flag
978
  // is set.
979
  testbool(rbx);
980
  jcc(Assembler::notZero, no_unlock);
981

982
  // unlock monitor
983
  push(state); // save result
984

985
  // BasicObjectLock will be first in list, since this is a
986
  // synchronized method. However, need to check that the object has
987
  // not been unlocked by an explicit monitorexit bytecode.
988
  const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
989
                        wordSize - (int) sizeof(BasicObjectLock));
990
  // We use c_rarg1/rdx so that if we go slow path it will be the correct
991
  // register for unlock_object to pass to VM directly
992
  lea(robj, monitor); // address of first monitor
993

994
  movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
995
  testptr(rax, rax);
996
  jcc(Assembler::notZero, unlock);
997

998
  pop(state);
999
  if (throw_monitor_exception) {
1000
    // Entry already unlocked, need to throw exception
1001
    NOT_LP64(empty_FPU_stack();)  // remove possible return value from FPU-stack, otherwise stack could overflow
1002
    call_VM(noreg, CAST_FROM_FN_PTR(address,
1003
                   InterpreterRuntime::throw_illegal_monitor_state_exception));
1004
    should_not_reach_here();
1005
  } else {
1006
    // Monitor already unlocked during a stack unroll. If requested,
1007
    // install an illegal_monitor_state_exception.  Continue with
1008
    // stack unrolling.
1009
    if (install_monitor_exception) {
1010
      NOT_LP64(empty_FPU_stack();)
1011
      call_VM(noreg, CAST_FROM_FN_PTR(address,
1012
                     InterpreterRuntime::new_illegal_monitor_state_exception));
1013
    }
1014
    jmp(unlocked);
1015
  }
1016

1017
  bind(unlock);
1018
  unlock_object(robj);
1019
  pop(state);
1020

1021
  // Check that for block-structured locking (i.e., that all locked
1022
  // objects has been unlocked)
1023
  bind(unlocked);
1024

1025
  // rax, rdx: Might contain return value
1026

1027
  // Check that all monitors are unlocked
1028
  {
1029
    Label loop, exception, entry, restart;
1030
    const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1031
    const Address monitor_block_top(
1032
        rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1033
    const Address monitor_block_bot(
1034
        rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1035

1036
    bind(restart);
1037
    // We use c_rarg1 so that if we go slow path it will be the correct
1038
    // register for unlock_object to pass to VM directly
1039
    movptr(rmon, monitor_block_top); // derelativize pointer
1040
    lea(rmon, Address(rbp, rmon, Address::times_ptr));
1041
    // c_rarg1 points to current entry, starting with top-most entry
1042

1043
    lea(rbx, monitor_block_bot);  // points to word before bottom of
1044
                                  // monitor block
1045
    jmp(entry);
1046

1047
    // Entry already locked, need to throw exception
1048
    bind(exception);
1049

1050
    if (throw_monitor_exception) {
1051
      // Throw exception
1052
      NOT_LP64(empty_FPU_stack();)
1053
      MacroAssembler::call_VM(noreg,
1054
                              CAST_FROM_FN_PTR(address, InterpreterRuntime::
1055
                                   throw_illegal_monitor_state_exception));
1056
      should_not_reach_here();
1057
    } else {
1058
      // Stack unrolling. Unlock object and install illegal_monitor_exception.
1059
      // Unlock does not block, so don't have to worry about the frame.
1060
      // We don't have to preserve c_rarg1 since we are going to throw an exception.
1061

1062
      push(state);
1063
      mov(robj, rmon);   // nop if robj and rmon are the same
1064
      unlock_object(robj);
1065
      pop(state);
1066

1067
      if (install_monitor_exception) {
1068
        NOT_LP64(empty_FPU_stack();)
1069
        call_VM(noreg, CAST_FROM_FN_PTR(address,
1070
                                        InterpreterRuntime::
1071
                                        new_illegal_monitor_state_exception));
1072
      }
1073

1074
      jmp(restart);
1075
    }
1076

1077
    bind(loop);
1078
    // check if current entry is used
1079
    cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1080
    jcc(Assembler::notEqual, exception);
1081

1082
    addptr(rmon, entry_size); // otherwise advance to next entry
1083
    bind(entry);
1084
    cmpptr(rmon, rbx); // check if bottom reached
1085
    jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1086
  }
1087

1088
  bind(no_unlock);
1089

1090
  // jvmti support
1091
  if (notify_jvmdi) {
1092
    notify_method_exit(state, NotifyJVMTI);    // preserve TOSCA
1093
  } else {
1094
    notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1095
  }
1096

1097
  // remove activation
1098
  // get sender sp
1099
  movptr(rbx,
1100
         Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1101
  if (StackReservedPages > 0) {
1102
    // testing if reserved zone needs to be re-enabled
1103
    Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1104
    Label no_reserved_zone_enabling;
1105

1106
    NOT_LP64(get_thread(rthread);)
1107

1108
    // check if already enabled - if so no re-enabling needed
1109
    assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1110
    cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1111
    jcc(Assembler::equal, no_reserved_zone_enabling);
1112

1113
    cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1114
    jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1115

1116
    call_VM_leaf(
1117
      CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1118
    call_VM(noreg, CAST_FROM_FN_PTR(address,
1119
                   InterpreterRuntime::throw_delayed_StackOverflowError));
1120
    should_not_reach_here();
1121

1122
    bind(no_reserved_zone_enabling);
1123
  }
1124
  leave();                           // remove frame anchor
1125
  pop(ret_addr);                     // get return address
1126
  mov(rsp, rbx);                     // set sp to sender sp
1127
  pop_cont_fastpath();
1128
}
1129

1130
void InterpreterMacroAssembler::get_method_counters(Register method,
1131
                                                    Register mcs, Label& skip) {
1132
  Label has_counters;
1133
  movptr(mcs, Address(method, Method::method_counters_offset()));
1134
  testptr(mcs, mcs);
1135
  jcc(Assembler::notZero, has_counters);
1136
  call_VM(noreg, CAST_FROM_FN_PTR(address,
1137
          InterpreterRuntime::build_method_counters), method);
1138
  movptr(mcs, Address(method,Method::method_counters_offset()));
1139
  testptr(mcs, mcs);
1140
  jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1141
  bind(has_counters);
1142
}
1143

1144

1145
// Lock object
1146
//
1147
// Args:
1148
//      rdx, c_rarg1: BasicObjectLock to be used for locking
1149
//
1150
// Kills:
1151
//      rax, rbx
1152
void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1153
  assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1154
         "The argument is only for looks. It must be c_rarg1");
1155

1156
  if (LockingMode == LM_MONITOR) {
1157
    call_VM(noreg,
1158
            CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1159
            lock_reg);
1160
  } else {
1161
    Label count_locking, done, slow_case;
1162

1163
    const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1164
    const Register tmp_reg = rbx;
1165
    const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1166
    const Register rklass_decode_tmp = rscratch1;
1167

1168
    const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1169
    const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1170
    const int mark_offset = lock_offset +
1171
                            BasicLock::displaced_header_offset_in_bytes();
1172

1173
    // Load object pointer into obj_reg
1174
    movptr(obj_reg, Address(lock_reg, obj_offset));
1175

1176
    if (DiagnoseSyncOnValueBasedClasses != 0) {
1177
      load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1178
      movl(tmp_reg, Address(tmp_reg, Klass::access_flags_offset()));
1179
      testl(tmp_reg, JVM_ACC_IS_VALUE_BASED_CLASS);
1180
      jcc(Assembler::notZero, slow_case);
1181
    }
1182

1183
    if (LockingMode == LM_LIGHTWEIGHT) {
1184
#ifdef _LP64
1185
      const Register thread = r15_thread;
1186
#else
1187
      const Register thread = lock_reg;
1188
      get_thread(thread);
1189
#endif
1190
      lightweight_lock(obj_reg, swap_reg, thread, tmp_reg, slow_case);
1191
    } else if (LockingMode == LM_LEGACY) {
1192
      // Load immediate 1 into swap_reg %rax
1193
      movl(swap_reg, 1);
1194

1195
      // Load (object->mark() | 1) into swap_reg %rax
1196
      orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1197

1198
      // Save (object->mark() | 1) into BasicLock's displaced header
1199
      movptr(Address(lock_reg, mark_offset), swap_reg);
1200

1201
      assert(lock_offset == 0,
1202
             "displaced header must be first word in BasicObjectLock");
1203

1204
      lock();
1205
      cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1206
      jcc(Assembler::zero, count_locking);
1207

1208
      const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1209

1210
      // Fast check for recursive lock.
1211
      //
1212
      // Can apply the optimization only if this is a stack lock
1213
      // allocated in this thread. For efficiency, we can focus on
1214
      // recently allocated stack locks (instead of reading the stack
1215
      // base and checking whether 'mark' points inside the current
1216
      // thread stack):
1217
      //  1) (mark & zero_bits) == 0, and
1218
      //  2) rsp <= mark < mark + os::pagesize()
1219
      //
1220
      // Warning: rsp + os::pagesize can overflow the stack base. We must
1221
      // neither apply the optimization for an inflated lock allocated
1222
      // just above the thread stack (this is why condition 1 matters)
1223
      // nor apply the optimization if the stack lock is inside the stack
1224
      // of another thread. The latter is avoided even in case of overflow
1225
      // because we have guard pages at the end of all stacks. Hence, if
1226
      // we go over the stack base and hit the stack of another thread,
1227
      // this should not be in a writeable area that could contain a
1228
      // stack lock allocated by that thread. As a consequence, a stack
1229
      // lock less than page size away from rsp is guaranteed to be
1230
      // owned by the current thread.
1231
      //
1232
      // These 3 tests can be done by evaluating the following
1233
      // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1234
      // assuming both stack pointer and pagesize have their
1235
      // least significant bits clear.
1236
      // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1237
      subptr(swap_reg, rsp);
1238
      andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1239

1240
      // Save the test result, for recursive case, the result is zero
1241
      movptr(Address(lock_reg, mark_offset), swap_reg);
1242
      jcc(Assembler::notZero, slow_case);
1243

1244
      bind(count_locking);
1245
    }
1246
    inc_held_monitor_count();
1247
    jmp(done);
1248

1249
    bind(slow_case);
1250

1251
    // Call the runtime routine for slow case
1252
    if (LockingMode == LM_LIGHTWEIGHT) {
1253
      call_VM(noreg,
1254
              CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
1255
              obj_reg);
1256
    } else {
1257
      call_VM(noreg,
1258
              CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1259
              lock_reg);
1260
    }
1261
    bind(done);
1262
  }
1263
}
1264

1265

1266
// Unlocks an object. Used in monitorexit bytecode and
1267
// remove_activation.  Throws an IllegalMonitorException if object is
1268
// not locked by current thread.
1269
//
1270
// Args:
1271
//      rdx, c_rarg1: BasicObjectLock for lock
1272
//
1273
// Kills:
1274
//      rax
1275
//      c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1276
//      rscratch1 (scratch reg)
1277
// rax, rbx, rcx, rdx
1278
void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1279
  assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1280
         "The argument is only for looks. It must be c_rarg1");
1281

1282
  if (LockingMode == LM_MONITOR) {
1283
    call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1284
  } else {
1285
    Label count_locking, done, slow_case;
1286

1287
    const Register swap_reg   = rax;  // Must use rax for cmpxchg instruction
1288
    const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx);  // Will contain the old oopMark
1289
    const Register obj_reg    = LP64_ONLY(c_rarg3) NOT_LP64(rcx);  // Will contain the oop
1290

1291
    save_bcp(); // Save in case of exception
1292

1293
    if (LockingMode != LM_LIGHTWEIGHT) {
1294
      // Convert from BasicObjectLock structure to object and BasicLock
1295
      // structure Store the BasicLock address into %rax
1296
      lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1297
    }
1298

1299
    // Load oop into obj_reg(%c_rarg3)
1300
    movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1301

1302
    // Free entry
1303
    movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1304

1305
    if (LockingMode == LM_LIGHTWEIGHT) {
1306
#ifdef _LP64
1307
      lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1308
#else
1309
      // This relies on the implementation of lightweight_unlock being able to handle
1310
      // that the reg_rax and thread Register parameters may alias each other.
1311
      get_thread(swap_reg);
1312
      lightweight_unlock(obj_reg, swap_reg, swap_reg, header_reg, slow_case);
1313
#endif
1314
    } else if (LockingMode == LM_LEGACY) {
1315
      // Load the old header from BasicLock structure
1316
      movptr(header_reg, Address(swap_reg,
1317
                                 BasicLock::displaced_header_offset_in_bytes()));
1318

1319
      // Test for recursion
1320
      testptr(header_reg, header_reg);
1321

1322
      // zero for recursive case
1323
      jcc(Assembler::zero, count_locking);
1324

1325
      // Atomic swap back the old header
1326
      lock();
1327
      cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1328

1329
      // zero for simple unlock of a stack-lock case
1330
      jcc(Assembler::notZero, slow_case);
1331

1332
      bind(count_locking);
1333
    }
1334
    dec_held_monitor_count();
1335
    jmp(done);
1336

1337
    bind(slow_case);
1338
    // Call the runtime routine for slow case.
1339
    movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1340
    call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1341

1342
    bind(done);
1343

1344
    restore_bcp();
1345
  }
1346
}
1347

1348
void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1349
                                                         Label& zero_continue) {
1350
  assert(ProfileInterpreter, "must be profiling interpreter");
1351
  movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1352
  testptr(mdp, mdp);
1353
  jcc(Assembler::zero, zero_continue);
1354
}
1355

1356

1357
// Set the method data pointer for the current bcp.
1358
void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1359
  assert(ProfileInterpreter, "must be profiling interpreter");
1360
  Label set_mdp;
1361
  push(rax);
1362
  push(rbx);
1363

1364
  get_method(rbx);
1365
  // Test MDO to avoid the call if it is null.
1366
  movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1367
  testptr(rax, rax);
1368
  jcc(Assembler::zero, set_mdp);
1369
  // rbx: method
1370
  // _bcp_register: bcp
1371
  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1372
  // rax: mdi
1373
  // mdo is guaranteed to be non-zero here, we checked for it before the call.
1374
  movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1375
  addptr(rbx, in_bytes(MethodData::data_offset()));
1376
  addptr(rax, rbx);
1377
  bind(set_mdp);
1378
  movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1379
  pop(rbx);
1380
  pop(rax);
1381
}
1382

1383
void InterpreterMacroAssembler::verify_method_data_pointer() {
1384
  assert(ProfileInterpreter, "must be profiling interpreter");
1385
#ifdef ASSERT
1386
  Label verify_continue;
1387
  push(rax);
1388
  push(rbx);
1389
  Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1390
  Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1391
  push(arg3_reg);
1392
  push(arg2_reg);
1393
  test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1394
  get_method(rbx);
1395

1396
  // If the mdp is valid, it will point to a DataLayout header which is
1397
  // consistent with the bcp.  The converse is highly probable also.
1398
  load_unsigned_short(arg2_reg,
1399
                      Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1400
  addptr(arg2_reg, Address(rbx, Method::const_offset()));
1401
  lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1402
  cmpptr(arg2_reg, _bcp_register);
1403
  jcc(Assembler::equal, verify_continue);
1404
  // rbx: method
1405
  // _bcp_register: bcp
1406
  // c_rarg3: mdp
1407
  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1408
               rbx, _bcp_register, arg3_reg);
1409
  bind(verify_continue);
1410
  pop(arg2_reg);
1411
  pop(arg3_reg);
1412
  pop(rbx);
1413
  pop(rax);
1414
#endif // ASSERT
1415
}
1416

1417

1418
void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1419
                                                int constant,
1420
                                                Register value) {
1421
  assert(ProfileInterpreter, "must be profiling interpreter");
1422
  Address data(mdp_in, constant);
1423
  movptr(data, value);
1424
}
1425

1426

1427
void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1428
                                                      int constant,
1429
                                                      bool decrement) {
1430
  // Counter address
1431
  Address data(mdp_in, constant);
1432

1433
  increment_mdp_data_at(data, decrement);
1434
}
1435

1436
void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1437
                                                      bool decrement) {
1438
  assert(ProfileInterpreter, "must be profiling interpreter");
1439
  // %%% this does 64bit counters at best it is wasting space
1440
  // at worst it is a rare bug when counters overflow
1441

1442
  if (decrement) {
1443
    // Decrement the register.  Set condition codes.
1444
    addptr(data, -DataLayout::counter_increment);
1445
    // If the decrement causes the counter to overflow, stay negative
1446
    Label L;
1447
    jcc(Assembler::negative, L);
1448
    addptr(data, DataLayout::counter_increment);
1449
    bind(L);
1450
  } else {
1451
    assert(DataLayout::counter_increment == 1,
1452
           "flow-free idiom only works with 1");
1453
    // Increment the register.  Set carry flag.
1454
    addptr(data, DataLayout::counter_increment);
1455
    // If the increment causes the counter to overflow, pull back by 1.
1456
    sbbptr(data, 0);
1457
  }
1458
}
1459

1460

1461
void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1462
                                                      Register reg,
1463
                                                      int constant,
1464
                                                      bool decrement) {
1465
  Address data(mdp_in, reg, Address::times_1, constant);
1466

1467
  increment_mdp_data_at(data, decrement);
1468
}
1469

1470
void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1471
                                                int flag_byte_constant) {
1472
  assert(ProfileInterpreter, "must be profiling interpreter");
1473
  int header_offset = in_bytes(DataLayout::flags_offset());
1474
  int header_bits = flag_byte_constant;
1475
  // Set the flag
1476
  orb(Address(mdp_in, header_offset), header_bits);
1477
}
1478

1479

1480

1481
void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1482
                                                 int offset,
1483
                                                 Register value,
1484
                                                 Register test_value_out,
1485
                                                 Label& not_equal_continue) {
1486
  assert(ProfileInterpreter, "must be profiling interpreter");
1487
  if (test_value_out == noreg) {
1488
    cmpptr(value, Address(mdp_in, offset));
1489
  } else {
1490
    // Put the test value into a register, so caller can use it:
1491
    movptr(test_value_out, Address(mdp_in, offset));
1492
    cmpptr(test_value_out, value);
1493
  }
1494
  jcc(Assembler::notEqual, not_equal_continue);
1495
}
1496

1497

1498
void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1499
                                                     int offset_of_disp) {
1500
  assert(ProfileInterpreter, "must be profiling interpreter");
1501
  Address disp_address(mdp_in, offset_of_disp);
1502
  addptr(mdp_in, disp_address);
1503
  movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1504
}
1505

1506

1507
void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1508
                                                     Register reg,
1509
                                                     int offset_of_disp) {
1510
  assert(ProfileInterpreter, "must be profiling interpreter");
1511
  Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1512
  addptr(mdp_in, disp_address);
1513
  movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1514
}
1515

1516

1517
void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1518
                                                       int constant) {
1519
  assert(ProfileInterpreter, "must be profiling interpreter");
1520
  addptr(mdp_in, constant);
1521
  movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1522
}
1523

1524

1525
void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1526
  assert(ProfileInterpreter, "must be profiling interpreter");
1527
  push(return_bci); // save/restore across call_VM
1528
  call_VM(noreg,
1529
          CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1530
          return_bci);
1531
  pop(return_bci);
1532
}
1533

1534

1535
void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1536
                                                     Register bumped_count) {
1537
  if (ProfileInterpreter) {
1538
    Label profile_continue;
1539

1540
    // If no method data exists, go to profile_continue.
1541
    // Otherwise, assign to mdp
1542
    test_method_data_pointer(mdp, profile_continue);
1543

1544
    // We are taking a branch.  Increment the taken count.
1545
    // We inline increment_mdp_data_at to return bumped_count in a register
1546
    //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1547
    Address data(mdp, in_bytes(JumpData::taken_offset()));
1548
    movptr(bumped_count, data);
1549
    assert(DataLayout::counter_increment == 1,
1550
            "flow-free idiom only works with 1");
1551
    addptr(bumped_count, DataLayout::counter_increment);
1552
    sbbptr(bumped_count, 0);
1553
    movptr(data, bumped_count); // Store back out
1554

1555
    // The method data pointer needs to be updated to reflect the new target.
1556
    update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1557
    bind(profile_continue);
1558
  }
1559
}
1560

1561

1562
void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1563
  if (ProfileInterpreter) {
1564
    Label profile_continue;
1565

1566
    // If no method data exists, go to profile_continue.
1567
    test_method_data_pointer(mdp, profile_continue);
1568

1569
    // We are taking a branch.  Increment the not taken count.
1570
    increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1571

1572
    // The method data pointer needs to be updated to correspond to
1573
    // the next bytecode
1574
    update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1575
    bind(profile_continue);
1576
  }
1577
}
1578

1579
void InterpreterMacroAssembler::profile_call(Register mdp) {
1580
  if (ProfileInterpreter) {
1581
    Label profile_continue;
1582

1583
    // If no method data exists, go to profile_continue.
1584
    test_method_data_pointer(mdp, profile_continue);
1585

1586
    // We are making a call.  Increment the count.
1587
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1588

1589
    // The method data pointer needs to be updated to reflect the new target.
1590
    update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1591
    bind(profile_continue);
1592
  }
1593
}
1594

1595

1596
void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1597
  if (ProfileInterpreter) {
1598
    Label profile_continue;
1599

1600
    // If no method data exists, go to profile_continue.
1601
    test_method_data_pointer(mdp, profile_continue);
1602

1603
    // We are making a call.  Increment the count.
1604
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1605

1606
    // The method data pointer needs to be updated to reflect the new target.
1607
    update_mdp_by_constant(mdp,
1608
                           in_bytes(VirtualCallData::
1609
                                    virtual_call_data_size()));
1610
    bind(profile_continue);
1611
  }
1612
}
1613

1614

1615
void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1616
                                                     Register mdp,
1617
                                                     Register reg2,
1618
                                                     bool receiver_can_be_null) {
1619
  if (ProfileInterpreter) {
1620
    Label profile_continue;
1621

1622
    // If no method data exists, go to profile_continue.
1623
    test_method_data_pointer(mdp, profile_continue);
1624

1625
    Label skip_receiver_profile;
1626
    if (receiver_can_be_null) {
1627
      Label not_null;
1628
      testptr(receiver, receiver);
1629
      jccb(Assembler::notZero, not_null);
1630
      // We are making a call.  Increment the count for null receiver.
1631
      increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1632
      jmp(skip_receiver_profile);
1633
      bind(not_null);
1634
    }
1635

1636
    // Record the receiver type.
1637
    record_klass_in_profile(receiver, mdp, reg2, true);
1638
    bind(skip_receiver_profile);
1639

1640
    // The method data pointer needs to be updated to reflect the new target.
1641
    update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1642
    bind(profile_continue);
1643
  }
1644
}
1645

1646
// This routine creates a state machine for updating the multi-row
1647
// type profile at a virtual call site (or other type-sensitive bytecode).
1648
// The machine visits each row (of receiver/count) until the receiver type
1649
// is found, or until it runs out of rows.  At the same time, it remembers
1650
// the location of the first empty row.  (An empty row records null for its
1651
// receiver, and can be allocated for a newly-observed receiver type.)
1652
// Because there are two degrees of freedom in the state, a simple linear
1653
// search will not work; it must be a decision tree.  Hence this helper
1654
// function is recursive, to generate the required tree structured code.
1655
// It's the interpreter, so we are trading off code space for speed.
1656
// See below for example code.
1657
void InterpreterMacroAssembler::record_klass_in_profile_helper(
1658
                                        Register receiver, Register mdp,
1659
                                        Register reg2, int start_row,
1660
                                        Label& done, bool is_virtual_call) {
1661
  if (TypeProfileWidth == 0) {
1662
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1663
  } else {
1664
    record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1665
                                  &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1666
  }
1667
}
1668

1669
void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1670
                                                              Label& done, int total_rows,
1671
                                                              OffsetFunction item_offset_fn,
1672
                                                              OffsetFunction item_count_offset_fn) {
1673
  int last_row = total_rows - 1;
1674
  assert(start_row <= last_row, "must be work left to do");
1675
  // Test this row for both the item and for null.
1676
  // Take any of three different outcomes:
1677
  //   1. found item => increment count and goto done
1678
  //   2. found null => keep looking for case 1, maybe allocate this cell
1679
  //   3. found something else => keep looking for cases 1 and 2
1680
  // Case 3 is handled by a recursive call.
1681
  for (int row = start_row; row <= last_row; row++) {
1682
    Label next_test;
1683
    bool test_for_null_also = (row == start_row);
1684

1685
    // See if the item is item[n].
1686
    int item_offset = in_bytes(item_offset_fn(row));
1687
    test_mdp_data_at(mdp, item_offset, item,
1688
                     (test_for_null_also ? reg2 : noreg),
1689
                     next_test);
1690
    // (Reg2 now contains the item from the CallData.)
1691

1692
    // The item is item[n].  Increment count[n].
1693
    int count_offset = in_bytes(item_count_offset_fn(row));
1694
    increment_mdp_data_at(mdp, count_offset);
1695
    jmp(done);
1696
    bind(next_test);
1697

1698
    if (test_for_null_also) {
1699
      // Failed the equality check on item[n]...  Test for null.
1700
      testptr(reg2, reg2);
1701
      if (start_row == last_row) {
1702
        // The only thing left to do is handle the null case.
1703
        Label found_null;
1704
        jccb(Assembler::zero, found_null);
1705
        // Item did not match any saved item and there is no empty row for it.
1706
        // Increment total counter to indicate polymorphic case.
1707
        increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1708
        jmp(done);
1709
        bind(found_null);
1710
        break;
1711
      }
1712
      Label found_null;
1713
      // Since null is rare, make it be the branch-taken case.
1714
      jcc(Assembler::zero, found_null);
1715

1716
      // Put all the "Case 3" tests here.
1717
      record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1718
                                    item_offset_fn, item_count_offset_fn);
1719

1720
      // Found a null.  Keep searching for a matching item,
1721
      // but remember that this is an empty (unused) slot.
1722
      bind(found_null);
1723
    }
1724
  }
1725

1726
  // In the fall-through case, we found no matching item, but we
1727
  // observed the item[start_row] is null.
1728

1729
  // Fill in the item field and increment the count.
1730
  int item_offset = in_bytes(item_offset_fn(start_row));
1731
  set_mdp_data_at(mdp, item_offset, item);
1732
  int count_offset = in_bytes(item_count_offset_fn(start_row));
1733
  movl(reg2, DataLayout::counter_increment);
1734
  set_mdp_data_at(mdp, count_offset, reg2);
1735
  if (start_row > 0) {
1736
    jmp(done);
1737
  }
1738
}
1739

1740
// Example state machine code for three profile rows:
1741
//   // main copy of decision tree, rooted at row[1]
1742
//   if (row[0].rec == rec) { row[0].incr(); goto done; }
1743
//   if (row[0].rec != nullptr) {
1744
//     // inner copy of decision tree, rooted at row[1]
1745
//     if (row[1].rec == rec) { row[1].incr(); goto done; }
1746
//     if (row[1].rec != nullptr) {
1747
//       // degenerate decision tree, rooted at row[2]
1748
//       if (row[2].rec == rec) { row[2].incr(); goto done; }
1749
//       if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1750
//       row[2].init(rec); goto done;
1751
//     } else {
1752
//       // remember row[1] is empty
1753
//       if (row[2].rec == rec) { row[2].incr(); goto done; }
1754
//       row[1].init(rec); goto done;
1755
//     }
1756
//   } else {
1757
//     // remember row[0] is empty
1758
//     if (row[1].rec == rec) { row[1].incr(); goto done; }
1759
//     if (row[2].rec == rec) { row[2].incr(); goto done; }
1760
//     row[0].init(rec); goto done;
1761
//   }
1762
//   done:
1763

1764
void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1765
                                                        Register mdp, Register reg2,
1766
                                                        bool is_virtual_call) {
1767
  assert(ProfileInterpreter, "must be profiling");
1768
  Label done;
1769

1770
  record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1771

1772
  bind (done);
1773
}
1774

1775
void InterpreterMacroAssembler::profile_ret(Register return_bci,
1776
                                            Register mdp) {
1777
  if (ProfileInterpreter) {
1778
    Label profile_continue;
1779
    uint row;
1780

1781
    // If no method data exists, go to profile_continue.
1782
    test_method_data_pointer(mdp, profile_continue);
1783

1784
    // Update the total ret count.
1785
    increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1786

1787
    for (row = 0; row < RetData::row_limit(); row++) {
1788
      Label next_test;
1789

1790
      // See if return_bci is equal to bci[n]:
1791
      test_mdp_data_at(mdp,
1792
                       in_bytes(RetData::bci_offset(row)),
1793
                       return_bci, noreg,
1794
                       next_test);
1795

1796
      // return_bci is equal to bci[n].  Increment the count.
1797
      increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1798

1799
      // The method data pointer needs to be updated to reflect the new target.
1800
      update_mdp_by_offset(mdp,
1801
                           in_bytes(RetData::bci_displacement_offset(row)));
1802
      jmp(profile_continue);
1803
      bind(next_test);
1804
    }
1805

1806
    update_mdp_for_ret(return_bci);
1807

1808
    bind(profile_continue);
1809
  }
1810
}
1811

1812

1813
void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1814
  if (ProfileInterpreter) {
1815
    Label profile_continue;
1816

1817
    // If no method data exists, go to profile_continue.
1818
    test_method_data_pointer(mdp, profile_continue);
1819

1820
    set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1821

1822
    // The method data pointer needs to be updated.
1823
    int mdp_delta = in_bytes(BitData::bit_data_size());
1824
    if (TypeProfileCasts) {
1825
      mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1826
    }
1827
    update_mdp_by_constant(mdp, mdp_delta);
1828

1829
    bind(profile_continue);
1830
  }
1831
}
1832

1833

1834
void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1835
  if (ProfileInterpreter) {
1836
    Label profile_continue;
1837

1838
    // If no method data exists, go to profile_continue.
1839
    test_method_data_pointer(mdp, profile_continue);
1840

1841
    // The method data pointer needs to be updated.
1842
    int mdp_delta = in_bytes(BitData::bit_data_size());
1843
    if (TypeProfileCasts) {
1844
      mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1845

1846
      // Record the object type.
1847
      record_klass_in_profile(klass, mdp, reg2, false);
1848
      NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1849
      NOT_LP64(restore_locals();)         // Restore EDI
1850
    }
1851
    update_mdp_by_constant(mdp, mdp_delta);
1852

1853
    bind(profile_continue);
1854
  }
1855
}
1856

1857

1858
void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1859
  if (ProfileInterpreter) {
1860
    Label profile_continue;
1861

1862
    // If no method data exists, go to profile_continue.
1863
    test_method_data_pointer(mdp, profile_continue);
1864

1865
    // Update the default case count
1866
    increment_mdp_data_at(mdp,
1867
                          in_bytes(MultiBranchData::default_count_offset()));
1868

1869
    // The method data pointer needs to be updated.
1870
    update_mdp_by_offset(mdp,
1871
                         in_bytes(MultiBranchData::
1872
                                  default_displacement_offset()));
1873

1874
    bind(profile_continue);
1875
  }
1876
}
1877

1878

1879
void InterpreterMacroAssembler::profile_switch_case(Register index,
1880
                                                    Register mdp,
1881
                                                    Register reg2) {
1882
  if (ProfileInterpreter) {
1883
    Label profile_continue;
1884

1885
    // If no method data exists, go to profile_continue.
1886
    test_method_data_pointer(mdp, profile_continue);
1887

1888
    // Build the base (index * per_case_size_in_bytes()) +
1889
    // case_array_offset_in_bytes()
1890
    movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1891
    imulptr(index, reg2); // XXX l ?
1892
    addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1893

1894
    // Update the case count
1895
    increment_mdp_data_at(mdp,
1896
                          index,
1897
                          in_bytes(MultiBranchData::relative_count_offset()));
1898

1899
    // The method data pointer needs to be updated.
1900
    update_mdp_by_offset(mdp,
1901
                         index,
1902
                         in_bytes(MultiBranchData::
1903
                                  relative_displacement_offset()));
1904

1905
    bind(profile_continue);
1906
  }
1907
}
1908

1909

1910

1911
void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1912
  if (state == atos) {
1913
    MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1914
  }
1915
}
1916

1917
void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1918
#ifndef _LP64
1919
  if ((state == ftos && UseSSE < 1) ||
1920
      (state == dtos && UseSSE < 2)) {
1921
    MacroAssembler::verify_FPU(stack_depth);
1922
  }
1923
#endif
1924
}
1925

1926
// Jump if ((*counter_addr += increment) & mask) == 0
1927
void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1928
                                                        Register scratch, Label* where) {
1929
  // This update is actually not atomic and can lose a number of updates
1930
  // under heavy contention, but the alternative of using the (contended)
1931
  // atomic update here penalizes profiling paths too much.
1932
  movl(scratch, counter_addr);
1933
  incrementl(scratch, InvocationCounter::count_increment);
1934
  movl(counter_addr, scratch);
1935
  andl(scratch, mask);
1936
  if (where != nullptr) {
1937
    jcc(Assembler::zero, *where);
1938
  }
1939
}
1940

1941
void InterpreterMacroAssembler::notify_method_entry() {
1942
  // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1943
  // track stack depth.  If it is possible to enter interp_only_mode we add
1944
  // the code to check if the event should be sent.
1945
  Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1946
  Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1947
  if (JvmtiExport::can_post_interpreter_events()) {
1948
    Label L;
1949
    NOT_LP64(get_thread(rthread);)
1950
    movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1951
    testl(rdx, rdx);
1952
    jcc(Assembler::zero, L);
1953
    call_VM(noreg, CAST_FROM_FN_PTR(address,
1954
                                    InterpreterRuntime::post_method_entry));
1955
    bind(L);
1956
  }
1957

1958
  if (DTraceMethodProbes) {
1959
    NOT_LP64(get_thread(rthread);)
1960
    get_method(rarg);
1961
    call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1962
                 rthread, rarg);
1963
  }
1964

1965
  // RedefineClasses() tracing support for obsolete method entry
1966
  if (log_is_enabled(Trace, redefine, class, obsolete)) {
1967
    NOT_LP64(get_thread(rthread);)
1968
    get_method(rarg);
1969
    call_VM_leaf(
1970
      CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1971
      rthread, rarg);
1972
  }
1973
}
1974

1975

1976
void InterpreterMacroAssembler::notify_method_exit(
1977
    TosState state, NotifyMethodExitMode mode) {
1978
  // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1979
  // track stack depth.  If it is possible to enter interp_only_mode we add
1980
  // the code to check if the event should be sent.
1981
  Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1982
  Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1983
  if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1984
    Label L;
1985
    // Note: frame::interpreter_frame_result has a dependency on how the
1986
    // method result is saved across the call to post_method_exit. If this
1987
    // is changed then the interpreter_frame_result implementation will
1988
    // need to be updated too.
1989

1990
    // template interpreter will leave the result on the top of the stack.
1991
    push(state);
1992
    NOT_LP64(get_thread(rthread);)
1993
    movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1994
    testl(rdx, rdx);
1995
    jcc(Assembler::zero, L);
1996
    call_VM(noreg,
1997
            CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1998
    bind(L);
1999
    pop(state);
2000
  }
2001

2002
  if (DTraceMethodProbes) {
2003
    push(state);
2004
    NOT_LP64(get_thread(rthread);)
2005
    get_method(rarg);
2006
    call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2007
                 rthread, rarg);
2008
    pop(state);
2009
  }
2010
}
2011

2012
void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2013
  // Get index out of bytecode pointer
2014
  get_cache_index_at_bcp(index, 1, sizeof(u4));
2015
  // Get address of invokedynamic array
2016
  movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2017
  movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2018
  if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2019
    shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2020
  } else {
2021
    imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2022
  }
2023
  lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2024
}
2025

2026
void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2027
  // Get index out of bytecode pointer
2028
  movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2029
  get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2030

2031
  movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2032
  // Take shortcut if the size is a power of 2
2033
  if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2034
    shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2035
  } else {
2036
    imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2037
  }
2038
  lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2039
}
2040

2041
void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2042
  // Get index out of bytecode pointer
2043
  movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2044
  get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2045

2046
  movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2047
  imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2048
  lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2049
}
2050