1 /*
2 * Copyright (c) 1997, 2025, 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 "compiler/compiler_globals.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/constMethodFlags.hpp"
32 #include "oops/markWord.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/inlineKlass.hpp"
36 #include "oops/resolvedFieldEntry.hpp"
37 #include "oops/resolvedIndyEntry.hpp"
38 #include "oops/resolvedMethodEntry.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/basicLock.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/javaThread.hpp"
44 #include "runtime/safepointMechanism.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "utilities/powerOfTwo.hpp"
47
48 // Implementation of InterpreterMacroAssembler
49
50 void InterpreterMacroAssembler::jump_to_entry(address entry) {
51 assert(entry, "Entry must have been generated by now");
52 jump(RuntimeAddress(entry));
53 }
54
55 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
56 Label update, next, none;
57
58 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
59
60 interp_verify_oop(obj, atos);
61
62 testptr(obj, obj);
63 jccb(Assembler::notZero, update);
64 testptr(mdo_addr, TypeEntries::null_seen);
65 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
66 // atomic update to prevent overwriting Klass* with 0
67 lock();
68 orptr(mdo_addr, TypeEntries::null_seen);
69 jmpb(next);
70
71 bind(update);
72 load_klass(obj, obj, rscratch1);
73 mov(rscratch1, obj);
74
75 xorptr(obj, mdo_addr);
76 testptr(obj, TypeEntries::type_klass_mask);
77 jccb(Assembler::zero, next); // klass seen before, nothing to
78 // do. The unknown bit may have been
79 // set already but no need to check.
80
81 testptr(obj, TypeEntries::type_unknown);
82 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
83
84 cmpptr(mdo_addr, 0);
85 jccb(Assembler::equal, none);
86 cmpptr(mdo_addr, TypeEntries::null_seen);
87 jccb(Assembler::equal, none);
88
89 // There is a chance that the checks above (re-reading profiling
90 // data from memory) fail if another thread has just set the
91 // profiling to this obj's klass
92 mov(obj, rscratch1);
93 xorptr(obj, mdo_addr);
94 testptr(obj, TypeEntries::type_klass_mask);
95 jccb(Assembler::zero, next);
96
97 // different than before. Cannot keep accurate profile.
98 orptr(mdo_addr, TypeEntries::type_unknown);
99 jmpb(next);
100
101 bind(none);
102 // first time here. Set profile type.
103 movptr(mdo_addr, obj);
104 #ifdef ASSERT
105 andptr(obj, TypeEntries::type_klass_mask);
106 verify_klass_ptr(obj);
107 #endif
108
109 bind(next);
110 }
111
112 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
113 if (!ProfileInterpreter) {
114 return;
115 }
116
117 if (MethodData::profile_arguments() || MethodData::profile_return()) {
118 Label profile_continue;
119
120 test_method_data_pointer(mdp, profile_continue);
121
122 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
123
124 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
125 jcc(Assembler::notEqual, profile_continue);
126
127 if (MethodData::profile_arguments()) {
128 Label done;
129 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
130 addptr(mdp, off_to_args);
131
132 for (int i = 0; i < TypeProfileArgsLimit; i++) {
133 if (i > 0 || MethodData::profile_return()) {
134 // If return value type is profiled we may have no argument to profile
135 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
136 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
137 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
138 jcc(Assembler::less, done);
139 }
140 movptr(tmp, Address(callee, Method::const_offset()));
141 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
142 // stack offset o (zero based) from the start of the argument
143 // list, for n arguments translates into offset n - o - 1 from
144 // the end of the argument list
145 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
146 subl(tmp, 1);
147 Address arg_addr = argument_address(tmp);
148 movptr(tmp, arg_addr);
149
150 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
151 profile_obj_type(tmp, mdo_arg_addr);
152
153 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
154 addptr(mdp, to_add);
155 off_to_args += to_add;
156 }
157
158 if (MethodData::profile_return()) {
159 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
160 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
161 }
162
163 bind(done);
164
165 if (MethodData::profile_return()) {
166 // We're right after the type profile for the last
167 // argument. tmp is the number of cells left in the
168 // CallTypeData/VirtualCallTypeData to reach its end. Non null
169 // if there's a return to profile.
170 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
171 shll(tmp, log2i_exact((int)DataLayout::cell_size));
172 addptr(mdp, tmp);
173 }
174 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
175 } else {
176 assert(MethodData::profile_return(), "either profile call args or call ret");
177 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
178 }
179
180 // mdp points right after the end of the
181 // CallTypeData/VirtualCallTypeData, right after the cells for the
182 // return value type if there's one
183
184 bind(profile_continue);
185 }
186 }
187
188 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
189 assert_different_registers(mdp, ret, tmp, _bcp_register);
190 if (ProfileInterpreter && MethodData::profile_return()) {
191 Label profile_continue;
192
193 test_method_data_pointer(mdp, profile_continue);
194
195 if (MethodData::profile_return_jsr292_only()) {
196 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
197
198 // If we don't profile all invoke bytecodes we must make sure
199 // it's a bytecode we indeed profile. We can't go back to the
200 // beginning of the ProfileData we intend to update to check its
201 // type because we're right after it and we don't known its
202 // length
203 Label do_profile;
204 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
205 jcc(Assembler::equal, do_profile);
206 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
207 jcc(Assembler::equal, do_profile);
208 get_method(tmp);
209 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
210 jcc(Assembler::notEqual, profile_continue);
211
212 bind(do_profile);
213 }
214
215 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
216 mov(tmp, ret);
217 profile_obj_type(tmp, mdo_ret_addr);
218
219 bind(profile_continue);
220 }
221 }
222
223 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
224 if (ProfileInterpreter && MethodData::profile_parameters()) {
225 Label profile_continue;
226
227 test_method_data_pointer(mdp, profile_continue);
228
229 // Load the offset of the area within the MDO used for
230 // parameters. If it's negative we're not profiling any parameters
231 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
232 testl(tmp1, tmp1);
233 jcc(Assembler::negative, profile_continue);
234
235 // Compute a pointer to the area for parameters from the offset
236 // and move the pointer to the slot for the last
237 // parameters. Collect profiling from last parameter down.
238 // mdo start + parameters offset + array length - 1
239 addptr(mdp, tmp1);
240 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
241 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
242
243 Label loop;
244 bind(loop);
245
246 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
247 int type_base = in_bytes(ParametersTypeData::type_offset(0));
248 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
249 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
250 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
251
252 // load offset on the stack from the slot for this parameter
253 movptr(tmp2, arg_off);
254 negptr(tmp2);
255 // read the parameter from the local area
256 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
257
258 // profile the parameter
259 profile_obj_type(tmp2, arg_type);
260
261 // go to next parameter
262 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
263 jcc(Assembler::positive, loop);
264
265 bind(profile_continue);
266 }
267 }
268
269 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
270 int number_of_arguments) {
271 // interpreter specific
272 //
273 // Note: No need to save/restore bcp & locals registers
274 // since these are callee saved registers and no blocking/
275 // GC can happen in leaf calls.
276 // Further Note: DO NOT save/restore bcp/locals. If a caller has
277 // already saved them so that it can use rsi/rdi as temporaries
278 // then a save/restore here will DESTROY the copy the caller
279 // saved! There used to be a save_bcp() that only happened in
280 // the ASSERT path (no restore_bcp). Which caused bizarre failures
281 // when jvm built with ASSERTs.
282 #ifdef ASSERT
283 {
284 Label L;
285 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
286 jcc(Assembler::equal, L);
287 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
288 " last_sp != null");
289 bind(L);
290 }
291 #endif
292 // super call
293 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
294 // interpreter specific
295 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
296 // but since they may not have been saved (and we don't want to
297 // save them here (see note above) the assert is invalid.
298 }
299
300 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
301 Register java_thread,
302 Register last_java_sp,
303 address entry_point,
304 int number_of_arguments,
305 bool check_exceptions) {
306 // interpreter specific
307 //
308 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
309 // really make a difference for these runtime calls, since they are
310 // slow anyway. Btw., bcp must be saved/restored since it may change
311 // due to GC.
312 save_bcp();
313 #ifdef ASSERT
314 {
315 Label L;
316 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
317 jcc(Assembler::equal, L);
318 stop("InterpreterMacroAssembler::call_VM_base:"
319 " last_sp isn't null");
320 bind(L);
321 }
322 #endif /* ASSERT */
323 // super call
324 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
325 entry_point, number_of_arguments,
326 check_exceptions);
327 // interpreter specific
328 restore_bcp();
329 restore_locals();
330 }
331
332 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
333 address entry_point,
334 Register arg_1) {
335 assert(arg_1 == c_rarg1, "");
336 Label resume_pc, not_preempted;
337
338 #ifdef ASSERT
339 {
340 Label L;
341 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
342 jcc(Assembler::equal, L);
343 stop("Should not have alternate return address set");
344 bind(L);
345 }
346 #endif /* ASSERT */
347
348 // Force freeze slow path.
349 push_cont_fastpath();
350
351 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
352 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
353 lea(rscratch1, resume_pc);
354 push(rscratch1);
355 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
356 pop(rscratch1);
357
358 pop_cont_fastpath();
359
360 // Check if preempted.
361 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
362 cmpptr(rscratch1, NULL_WORD);
363 jccb(Assembler::zero, not_preempted);
364 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
365 jmp(rscratch1);
366
367 // In case of preemption, this is where we will resume once we finally acquire the monitor.
368 bind(resume_pc);
369 restore_after_resume(false /* is_native */);
370
371 bind(not_preempted);
372 }
373
374 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
375 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
376 call(rscratch1);
377 if (is_native) {
378 // On resume we need to set up stack as expected.
379 push(dtos);
380 push(ltos);
381 }
382 }
383
384 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
385 if (JvmtiExport::can_pop_frame()) {
386 Label L;
387 // Initiate popframe handling only if it is not already being
388 // processed. If the flag has the popframe_processing bit set, it
389 // means that this code is called *during* popframe handling - we
390 // don't want to reenter.
391 // This method is only called just after the call into the vm in
392 // call_VM_base, so the arg registers are available.
393 Register pop_cond = c_rarg0;
394 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
395 testl(pop_cond, JavaThread::popframe_pending_bit);
396 jcc(Assembler::zero, L);
397 testl(pop_cond, JavaThread::popframe_processing_bit);
398 jcc(Assembler::notZero, L);
399 // Call Interpreter::remove_activation_preserving_args_entry() to get the
400 // address of the same-named entrypoint in the generated interpreter code.
401 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
402 jmp(rax);
403 bind(L);
404 }
405 }
406
407 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
408 movptr(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
409 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
410 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
411 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
412
413 switch (state) {
414 case atos: movptr(rax, oop_addr);
415 movptr(oop_addr, NULL_WORD);
416 interp_verify_oop(rax, state); break;
417 case ltos: movptr(rax, val_addr); break;
418 case btos: // fall through
419 case ztos: // fall through
420 case ctos: // fall through
421 case stos: // fall through
422 case itos: movl(rax, val_addr); break;
423 case ftos: load_float(val_addr); break;
424 case dtos: load_double(val_addr); break;
425 case vtos: /* nothing to do */ break;
426 default : ShouldNotReachHere();
427 }
428
429 // Clean up tos value in the thread object
430 movl(tos_addr, ilgl);
431 movptr(val_addr, NULL_WORD);
432 }
433
434
435 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
436 if (JvmtiExport::can_force_early_return()) {
437 Label L;
438 Register tmp = c_rarg0;
439 Register rthread = r15_thread;
440
441 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
442 testptr(tmp, tmp);
443 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
444
445 // Initiate earlyret handling only if it is not already being processed.
446 // If the flag has the earlyret_processing bit set, it means that this code
447 // is called *during* earlyret handling - we don't want to reenter.
448 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
449 cmpl(tmp, JvmtiThreadState::earlyret_pending);
450 jcc(Assembler::notEqual, L);
451
452 // Call Interpreter::remove_activation_early_entry() to get the address of the
453 // same-named entrypoint in the generated interpreter code.
454 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
455 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
456 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
457 jmp(rax);
458 bind(L);
459 }
460 }
461
462 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
463 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
464 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
465 bswapl(reg);
466 shrl(reg, 16);
467 }
468
469 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
470 int bcp_offset,
471 size_t index_size) {
472 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
473 if (index_size == sizeof(u2)) {
474 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
475 } else if (index_size == sizeof(u4)) {
476 movl(index, Address(_bcp_register, bcp_offset));
477 } else if (index_size == sizeof(u1)) {
478 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
479 } else {
480 ShouldNotReachHere();
481 }
482 }
483
484 // Load object from cpool->resolved_references(index)
485 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
486 Register index,
487 Register tmp) {
488 assert_different_registers(result, index);
489
490 get_constant_pool(result);
491 // load pointer for resolved_references[] objArray
492 movptr(result, Address(result, ConstantPool::cache_offset()));
493 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
494 resolve_oop_handle(result, tmp);
495 load_heap_oop(result, Address(result, index,
496 UseCompressedOops ? Address::times_4 : Address::times_ptr,
497 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
498 }
499
500 // load cpool->resolved_klass_at(index)
501 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
502 Register cpool,
503 Register index) {
504 assert_different_registers(cpool, index);
505
506 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
507 Register resolved_klasses = cpool;
508 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
509 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
510 }
511
512 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
513 // subtype of super_klass.
514 //
515 // Args:
516 // rax: superklass
517 // Rsub_klass: subklass
518 //
519 // Kills:
520 // rcx, rdi
521 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
522 Label& ok_is_subtype,
523 bool profile) {
524 assert(Rsub_klass != rax, "rax holds superklass");
525 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
526 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
527 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
528 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
529
530 // Profile the not-null value's klass.
531 if (profile) {
532 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
533 }
534
535 // Do the check.
536 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
537 }
538
539
540 // Java Expression Stack
541
542 void InterpreterMacroAssembler::pop_ptr(Register r) {
543 pop(r);
544 }
545
546 void InterpreterMacroAssembler::push_ptr(Register r) {
547 push(r);
548 }
549
550 void InterpreterMacroAssembler::push_i(Register r) {
551 push(r);
552 }
553
554 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
555 push(r);
556 }
557
558 void InterpreterMacroAssembler::push_f(XMMRegister r) {
559 subptr(rsp, wordSize);
560 movflt(Address(rsp, 0), r);
561 }
562
563 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
564 movflt(r, Address(rsp, 0));
565 addptr(rsp, wordSize);
566 }
567
568 void InterpreterMacroAssembler::push_d(XMMRegister r) {
569 subptr(rsp, 2 * wordSize);
570 movdbl(Address(rsp, 0), r);
571 }
572
573 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
574 movdbl(r, Address(rsp, 0));
575 addptr(rsp, 2 * Interpreter::stackElementSize);
576 }
577
578 void InterpreterMacroAssembler::pop_i(Register r) {
579 // XXX can't use pop currently, upper half non clean
580 movl(r, Address(rsp, 0));
581 addptr(rsp, wordSize);
582 }
583
584 void InterpreterMacroAssembler::pop_l(Register r) {
585 movq(r, Address(rsp, 0));
586 addptr(rsp, 2 * Interpreter::stackElementSize);
587 }
588
589 void InterpreterMacroAssembler::push_l(Register r) {
590 subptr(rsp, 2 * wordSize);
591 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
592 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
593 }
594
595 void InterpreterMacroAssembler::pop(TosState state) {
596 switch (state) {
597 case atos: pop_ptr(); break;
598 case btos:
599 case ztos:
600 case ctos:
601 case stos:
602 case itos: pop_i(); break;
603 case ltos: pop_l(); break;
604 case ftos: pop_f(xmm0); break;
605 case dtos: pop_d(xmm0); break;
606 case vtos: /* nothing to do */ break;
607 default: ShouldNotReachHere();
608 }
609 interp_verify_oop(rax, state);
610 }
611
612 void InterpreterMacroAssembler::push(TosState state) {
613 interp_verify_oop(rax, state);
614 switch (state) {
615 case atos: push_ptr(); break;
616 case btos:
617 case ztos:
618 case ctos:
619 case stos:
620 case itos: push_i(); break;
621 case ltos: push_l(); break;
622 case ftos: push_f(xmm0); break;
623 case dtos: push_d(xmm0); break;
624 case vtos: /* nothing to do */ break;
625 default : ShouldNotReachHere();
626 }
627 }
628
629 // Helpers for swap and dup
630 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
631 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
632 }
633
634 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
635 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
636 }
637
638
639 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
640 // set sender sp
641 lea(_bcp_register, Address(rsp, wordSize));
642 // record last_sp
643 mov(rcx, _bcp_register);
644 subptr(rcx, rbp);
645 sarptr(rcx, LogBytesPerWord);
646 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
647 }
648
649
650 // Jump to from_interpreted entry of a call unless single stepping is possible
651 // in this thread in which case we must call the i2i entry
652 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
653 prepare_to_jump_from_interpreted();
654
655 if (JvmtiExport::can_post_interpreter_events()) {
656 Label run_compiled_code;
657 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
658 // compiled code in threads for which the event is enabled. Check here for
659 // interp_only_mode if these events CAN be enabled.
660 // interp_only is an int, on little endian it is sufficient to test the byte only
661 // Is a cmpl faster?
662 cmpb(Address(r15_thread, JavaThread::interp_only_mode_offset()), 0);
663 jccb(Assembler::zero, run_compiled_code);
664 jmp(Address(method, Method::interpreter_entry_offset()));
665 bind(run_compiled_code);
666 }
667
668 jmp(Address(method, Method::from_interpreted_offset()));
669 }
670
671 // The following two routines provide a hook so that an implementation
672 // can schedule the dispatch in two parts. x86 does not do this.
673 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
674 // Nothing x86 specific to be done here
675 }
676
677 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
678 dispatch_next(state, step);
679 }
680
681 void InterpreterMacroAssembler::dispatch_base(TosState state,
682 address* table,
683 bool verifyoop,
684 bool generate_poll) {
685 if (VerifyActivationFrameSize) {
686 Label L;
687 mov(rcx, rbp);
688 subptr(rcx, rsp);
689 int32_t min_frame_size =
690 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
691 wordSize;
692 cmpptr(rcx, min_frame_size);
693 jcc(Assembler::greaterEqual, L);
694 stop("broken stack frame");
695 bind(L);
696 }
697 if (verifyoop) {
698 interp_verify_oop(rax, state);
699 }
700
701 address* const safepoint_table = Interpreter::safept_table(state);
702 Label no_safepoint, dispatch;
703 if (table != safepoint_table && generate_poll) {
704 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
705 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
706
707 jccb(Assembler::zero, no_safepoint);
708 lea(rscratch1, ExternalAddress((address)safepoint_table));
709 jmpb(dispatch);
710 }
711
712 bind(no_safepoint);
713 lea(rscratch1, ExternalAddress((address)table));
714 bind(dispatch);
715 jmp(Address(rscratch1, rbx, Address::times_8));
716 }
717
718 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
719 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
720 }
721
722 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
723 dispatch_base(state, Interpreter::normal_table(state));
724 }
725
726 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
727 dispatch_base(state, Interpreter::normal_table(state), false);
728 }
729
730
731 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
732 // load next bytecode (load before advancing _bcp_register to prevent AGI)
733 load_unsigned_byte(rbx, Address(_bcp_register, step));
734 // advance _bcp_register
735 increment(_bcp_register, step);
736 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
737 }
738
739 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
740 // load current bytecode
741 load_unsigned_byte(rbx, Address(_bcp_register, 0));
742 dispatch_base(state, table);
743 }
744
745 void InterpreterMacroAssembler::narrow(Register result) {
746
747 // Get method->_constMethod->_result_type
748 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
749 movptr(rcx, Address(rcx, Method::const_offset()));
750 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
751
752 Label done, notBool, notByte, notChar;
753
754 // common case first
755 cmpl(rcx, T_INT);
756 jcc(Assembler::equal, done);
757
758 // mask integer result to narrower return type.
759 cmpl(rcx, T_BOOLEAN);
760 jcc(Assembler::notEqual, notBool);
761 andl(result, 0x1);
762 jmp(done);
763
764 bind(notBool);
765 cmpl(rcx, T_BYTE);
766 jcc(Assembler::notEqual, notByte);
767 movsbl(result, result);
768 jmp(done);
769
770 bind(notByte);
771 cmpl(rcx, T_CHAR);
772 jcc(Assembler::notEqual, notChar);
773 movzwl(result, result);
774 jmp(done);
775
776 bind(notChar);
777 // cmpl(rcx, T_SHORT); // all that's left
778 // jcc(Assembler::notEqual, done);
779 movswl(result, result);
780
781 // Nothing to do for T_INT
782 bind(done);
783 }
784
785 // remove activation
786 //
787 // Apply stack watermark barrier.
788 // Unlock the receiver if this is a synchronized method.
789 // Unlock any Java monitors from synchronized blocks.
790 // Remove the activation from the stack.
791 //
792 // If there are locked Java monitors
793 // If throw_monitor_exception
794 // throws IllegalMonitorStateException
795 // Else if install_monitor_exception
796 // installs IllegalMonitorStateException
797 // Else
798 // no error processing
799 void InterpreterMacroAssembler::remove_activation(
800 TosState state,
801 Register ret_addr,
802 bool throw_monitor_exception,
803 bool install_monitor_exception,
804 bool notify_jvmdi) {
805 // Note: Registers rdx xmm0 may be in use for the
806 // result check if synchronized method
807 Label unlocked, unlock, no_unlock;
808
809 const Register rthread = r15_thread;
810 const Register robj = c_rarg1;
811 const Register rmon = c_rarg1;
812
813 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
814 // that would normally not be safe to use. Such bad returns into unsafe territory of
815 // the stack, will call InterpreterRuntime::at_unwind.
816 Label slow_path;
817 Label fast_path;
818 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
819 jmp(fast_path);
820 bind(slow_path);
821 push(state);
822 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
823 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
824 reset_last_Java_frame(rthread, true);
825 pop(state);
826 bind(fast_path);
827
828 // get the value of _do_not_unlock_if_synchronized into rdx
829 const Address do_not_unlock_if_synchronized(rthread,
830 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
831 movbool(rbx, do_not_unlock_if_synchronized);
832 movbool(do_not_unlock_if_synchronized, false); // reset the flag
833
834 // get method access flags
835 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
836 load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
837 testl(rcx, JVM_ACC_SYNCHRONIZED);
838 jcc(Assembler::zero, unlocked);
839
840 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
841 // is set.
842 testbool(rbx);
843 jcc(Assembler::notZero, no_unlock);
844
845 // unlock monitor
846 push(state); // save result
847
848 // BasicObjectLock will be first in list, since this is a
849 // synchronized method. However, need to check that the object has
850 // not been unlocked by an explicit monitorexit bytecode.
851 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
852 wordSize - (int) sizeof(BasicObjectLock));
853 // We use c_rarg1/rdx so that if we go slow path it will be the correct
854 // register for unlock_object to pass to VM directly
855 lea(robj, monitor); // address of first monitor
856
857 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
858 testptr(rax, rax);
859 jcc(Assembler::notZero, unlock);
860
861 pop(state);
862 if (throw_monitor_exception) {
863 // Entry already unlocked, need to throw exception
864 call_VM(noreg, CAST_FROM_FN_PTR(address,
865 InterpreterRuntime::throw_illegal_monitor_state_exception));
866 should_not_reach_here();
867 } else {
868 // Monitor already unlocked during a stack unroll. If requested,
869 // install an illegal_monitor_state_exception. Continue with
870 // stack unrolling.
871 if (install_monitor_exception) {
872 call_VM(noreg, CAST_FROM_FN_PTR(address,
873 InterpreterRuntime::new_illegal_monitor_state_exception));
874 }
875 jmp(unlocked);
876 }
877
878 bind(unlock);
879 unlock_object(robj);
880 pop(state);
881
882 // Check that for block-structured locking (i.e., that all locked
883 // objects has been unlocked)
884 bind(unlocked);
885
886 // rax, rdx: Might contain return value
887
888 // Check that all monitors are unlocked
889 {
890 Label loop, exception, entry, restart;
891 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
892 const Address monitor_block_top(
893 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
894 const Address monitor_block_bot(
895 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
896
897 bind(restart);
898 // We use c_rarg1 so that if we go slow path it will be the correct
899 // register for unlock_object to pass to VM directly
900 movptr(rmon, monitor_block_top); // derelativize pointer
901 lea(rmon, Address(rbp, rmon, Address::times_ptr));
902 // c_rarg1 points to current entry, starting with top-most entry
903
904 lea(rbx, monitor_block_bot); // points to word before bottom of
905 // monitor block
906 jmp(entry);
907
908 // Entry already locked, need to throw exception
909 bind(exception);
910
911 if (throw_monitor_exception) {
912 // Throw exception
913 MacroAssembler::call_VM(noreg,
914 CAST_FROM_FN_PTR(address, InterpreterRuntime::
915 throw_illegal_monitor_state_exception));
916 should_not_reach_here();
917 } else {
918 // Stack unrolling. Unlock object and install illegal_monitor_exception.
919 // Unlock does not block, so don't have to worry about the frame.
920 // We don't have to preserve c_rarg1 since we are going to throw an exception.
921
922 push(state);
923 mov(robj, rmon); // nop if robj and rmon are the same
924 unlock_object(robj);
925 pop(state);
926
927 if (install_monitor_exception) {
928 call_VM(noreg, CAST_FROM_FN_PTR(address,
929 InterpreterRuntime::
930 new_illegal_monitor_state_exception));
931 }
932
933 jmp(restart);
934 }
935
936 bind(loop);
937 // check if current entry is used
938 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
939 jcc(Assembler::notEqual, exception);
940
941 addptr(rmon, entry_size); // otherwise advance to next entry
942 bind(entry);
943 cmpptr(rmon, rbx); // check if bottom reached
944 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
945 }
946
947 bind(no_unlock);
948
949 // jvmti support
950 if (notify_jvmdi) {
951 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
952 } else {
953 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
954 }
955
956 if (StackReservedPages > 0) {
957 movptr(rbx,
958 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
959 // testing if reserved zone needs to be re-enabled
960 Register rthread = r15_thread;
961 Label no_reserved_zone_enabling;
962
963 // check if already enabled - if so no re-enabling needed
964 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
965 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
966 jcc(Assembler::equal, no_reserved_zone_enabling);
967
968 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
969 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
970
971 call_VM_leaf(
972 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
973 call_VM(noreg, CAST_FROM_FN_PTR(address,
974 InterpreterRuntime::throw_delayed_StackOverflowError));
975 should_not_reach_here();
976
977 bind(no_reserved_zone_enabling);
978 }
979
980 // remove activation
981 // get sender sp
982 movptr(rbx,
983 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
984
985 if (state == atos && InlineTypeReturnedAsFields) {
986 // Check if we are returning an non-null inline type and load its fields into registers
987 Label skip;
988 test_oop_is_not_inline_type(rax, rscratch1, skip);
989
990 #ifndef _LP64
991 super_call_VM_leaf(StubRoutines::load_inline_type_fields_in_regs());
992 #else
993 // Load fields from a buffered value with an inline class specific handler
994 load_klass(rdi, rax, rscratch1);
995 movptr(rdi, Address(rdi, InstanceKlass::adr_inlineklass_fixed_block_offset()));
996 movptr(rdi, Address(rdi, InlineKlass::unpack_handler_offset()));
997 // Unpack handler can be null if inline type is not scalarizable in returns
998 testptr(rdi, rdi);
999 jcc(Assembler::zero, skip);
1000 call(rdi);
1001 #endif
1002 #ifdef ASSERT
1003 // TODO 8284443 Enable
1004 if (StressCallingConvention && false) {
1005 Label skip_stress;
1006 movptr(rscratch1, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1007 movl(rscratch1, Address(rscratch1, Method::flags_offset()));
1008 testl(rcx, MethodFlags::has_scalarized_return_flag());
1009 jcc(Assembler::zero, skip_stress);
1010 load_klass(rax, rax, rscratch1);
1011 orptr(rax, 1);
1012 bind(skip_stress);
1013 }
1014 #endif
1015 // call above kills the value in rbx. Reload it.
1016 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1017 bind(skip);
1018 }
1019 leave(); // remove frame anchor
1020 pop(ret_addr); // get return address
1021 mov(rsp, rbx); // set sp to sender sp
1022 pop_cont_fastpath();
1023 }
1024
1025 void InterpreterMacroAssembler::get_method_counters(Register method,
1026 Register mcs, Label& skip) {
1027 Label has_counters;
1028 movptr(mcs, Address(method, Method::method_counters_offset()));
1029 testptr(mcs, mcs);
1030 jcc(Assembler::notZero, has_counters);
1031 call_VM(noreg, CAST_FROM_FN_PTR(address,
1032 InterpreterRuntime::build_method_counters), method);
1033 movptr(mcs, Address(method,Method::method_counters_offset()));
1034 testptr(mcs, mcs);
1035 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1036 bind(has_counters);
1037 }
1038
1039 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
1040 Register t1, Register t2,
1041 bool clear_fields, Label& alloc_failed) {
1042 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
1043 if (DTraceMethodProbes) {
1044 // Trigger dtrace event for fastpath
1045 push(atos);
1046 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
1047 pop(atos);
1048 }
1049 }
1050
1051 void InterpreterMacroAssembler::read_flat_field(Register entry, Register tmp1, Register tmp2, Register obj) {
1052 Label alloc_failed, done;
1053 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1054 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1055 assert_different_registers(obj, entry, tmp1, tmp2, dst_temp, r8, r9);
1056
1057 // FIXME: code below could be re-written to better use InlineLayoutInfo data structure
1058 // see aarch64 version
1059
1060 // Grap the inline field klass
1061 const Register field_klass = tmp1;
1062 load_unsigned_short(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
1063 movptr(tmp1, Address(entry, ResolvedFieldEntry::field_holder_offset()));
1064 get_inline_type_field_klass(tmp1, tmp2, field_klass);
1065
1066 // allocate buffer
1067 push(obj); // push object being read from // FIXME spilling on stack could probably be avoided by using tmp2
1068 allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1069
1070 // Have an oop instance buffer, copy into it
1071 load_unsigned_short(r9, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
1072 movptr(r8, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
1073 inline_layout_info(r8, r9, r8); // holder, index, info => InlineLayoutInfo into r8
1074
1075 payload_addr(obj, dst_temp, field_klass);
1076 pop(alloc_temp); // restore object being read from
1077 load_sized_value(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
1078 lea(tmp2, Address(alloc_temp, tmp2));
1079 // call_VM_leaf, clobbers a few regs, save restore new obj
1080 push(obj);
1081 // access_value_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, field_klass);
1082 flat_field_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, r8);
1083 pop(obj);
1084 jmp(done);
1085
1086 bind(alloc_failed);
1087 pop(obj);
1088 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
1089 obj, entry);
1090 get_vm_result(obj, r15_thread);
1091 bind(done);
1092 }
1093
1094 // Lock object
1095 //
1096 // Args:
1097 // rdx, c_rarg1: BasicObjectLock to be used for locking
1098 //
1099 // Kills:
1100 // rax, rbx
1101 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1102 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1103
1104 if (LockingMode == LM_MONITOR) {
1105 call_VM_preemptable(noreg,
1106 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1107 lock_reg);
1108 } else {
1109 Label count_locking, done, slow_case;
1110
1111 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1112 const Register tmp_reg = rbx;
1113 const Register obj_reg = c_rarg3; // Will contain the oop
1114 const Register rklass_decode_tmp = rscratch1;
1115
1116 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1117 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1118 const int mark_offset = lock_offset +
1119 BasicLock::displaced_header_offset_in_bytes();
1120
1121 // Load object pointer into obj_reg
1122 movptr(obj_reg, Address(lock_reg, obj_offset));
1123
1124 if (DiagnoseSyncOnValueBasedClasses != 0) {
1125 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1126 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1127 jcc(Assembler::notZero, slow_case);
1128 }
1129
1130 if (LockingMode == LM_LIGHTWEIGHT) {
1131 const Register thread = r15_thread;
1132 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1133 } else if (LockingMode == LM_LEGACY) {
1134 // Load immediate 1 into swap_reg %rax
1135 movl(swap_reg, 1);
1136
1137 // Load (object->mark() | 1) into swap_reg %rax
1138 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1139 if (EnableValhalla) {
1140 // Mask inline_type bit such that we go to the slow path if object is an inline type
1141 andptr(swap_reg, ~((int) markWord::inline_type_bit_in_place));
1142 }
1143
1144 // Save (object->mark() | 1) into BasicLock's displaced header
1145 movptr(Address(lock_reg, mark_offset), swap_reg);
1146
1147 assert(lock_offset == 0,
1148 "displaced header must be first word in BasicObjectLock");
1149
1150 lock();
1151 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1152 jcc(Assembler::zero, count_locking);
1153
1154 const int zero_bits = 7;
1155
1156 // Fast check for recursive lock.
1157 //
1158 // Can apply the optimization only if this is a stack lock
1159 // allocated in this thread. For efficiency, we can focus on
1160 // recently allocated stack locks (instead of reading the stack
1161 // base and checking whether 'mark' points inside the current
1162 // thread stack):
1163 // 1) (mark & zero_bits) == 0, and
1164 // 2) rsp <= mark < mark + os::pagesize()
1165 //
1166 // Warning: rsp + os::pagesize can overflow the stack base. We must
1167 // neither apply the optimization for an inflated lock allocated
1168 // just above the thread stack (this is why condition 1 matters)
1169 // nor apply the optimization if the stack lock is inside the stack
1170 // of another thread. The latter is avoided even in case of overflow
1171 // because we have guard pages at the end of all stacks. Hence, if
1172 // we go over the stack base and hit the stack of another thread,
1173 // this should not be in a writeable area that could contain a
1174 // stack lock allocated by that thread. As a consequence, a stack
1175 // lock less than page size away from rsp is guaranteed to be
1176 // owned by the current thread.
1177 //
1178 // These 3 tests can be done by evaluating the following
1179 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1180 // assuming both stack pointer and pagesize have their
1181 // least significant bits clear.
1182 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1183 subptr(swap_reg, rsp);
1184 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1185
1186 // Save the test result, for recursive case, the result is zero
1187 movptr(Address(lock_reg, mark_offset), swap_reg);
1188 jcc(Assembler::notZero, slow_case);
1189
1190 bind(count_locking);
1191 inc_held_monitor_count();
1192 }
1193 jmp(done);
1194
1195 bind(slow_case);
1196
1197 // Call the runtime routine for slow case
1198 call_VM_preemptable(noreg,
1199 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1200 lock_reg);
1201 bind(done);
1202 }
1203 }
1204
1205
1206 // Unlocks an object. Used in monitorexit bytecode and
1207 // remove_activation. Throws an IllegalMonitorException if object is
1208 // not locked by current thread.
1209 //
1210 // Args:
1211 // rdx, c_rarg1: BasicObjectLock for lock
1212 //
1213 // Kills:
1214 // rax
1215 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1216 // rscratch1 (scratch reg)
1217 // rax, rbx, rcx, rdx
1218 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1219 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
1220
1221 if (LockingMode == LM_MONITOR) {
1222 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1223 } else {
1224 Label count_locking, done, slow_case;
1225
1226 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1227 const Register header_reg = c_rarg2; // Will contain the old oopMark
1228 const Register obj_reg = c_rarg3; // Will contain the oop
1229
1230 save_bcp(); // Save in case of exception
1231
1232 if (LockingMode != LM_LIGHTWEIGHT) {
1233 // Convert from BasicObjectLock structure to object and BasicLock
1234 // structure Store the BasicLock address into %rax
1235 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1236 }
1237
1238 // Load oop into obj_reg(%c_rarg3)
1239 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1240
1241 // Free entry
1242 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1243
1244 if (LockingMode == LM_LIGHTWEIGHT) {
1245 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1246 } else if (LockingMode == LM_LEGACY) {
1247 // Load the old header from BasicLock structure
1248 movptr(header_reg, Address(swap_reg,
1249 BasicLock::displaced_header_offset_in_bytes()));
1250
1251 // Test for recursion
1252 testptr(header_reg, header_reg);
1253
1254 // zero for recursive case
1255 jcc(Assembler::zero, count_locking);
1256
1257 // Atomic swap back the old header
1258 lock();
1259 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1260
1261 // zero for simple unlock of a stack-lock case
1262 jcc(Assembler::notZero, slow_case);
1263
1264 bind(count_locking);
1265 dec_held_monitor_count();
1266 }
1267 jmp(done);
1268
1269 bind(slow_case);
1270 // Call the runtime routine for slow case.
1271 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1272 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1273
1274 bind(done);
1275
1276 restore_bcp();
1277 }
1278 }
1279
1280 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1281 Label& zero_continue) {
1282 assert(ProfileInterpreter, "must be profiling interpreter");
1283 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1284 testptr(mdp, mdp);
1285 jcc(Assembler::zero, zero_continue);
1286 }
1287
1288
1289 // Set the method data pointer for the current bcp.
1290 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1291 assert(ProfileInterpreter, "must be profiling interpreter");
1292 Label set_mdp;
1293 push(rax);
1294 push(rbx);
1295
1296 get_method(rbx);
1297 // Test MDO to avoid the call if it is null.
1298 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1299 testptr(rax, rax);
1300 jcc(Assembler::zero, set_mdp);
1301 // rbx: method
1302 // _bcp_register: bcp
1303 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1304 // rax: mdi
1305 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1306 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1307 addptr(rbx, in_bytes(MethodData::data_offset()));
1308 addptr(rax, rbx);
1309 bind(set_mdp);
1310 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1311 pop(rbx);
1312 pop(rax);
1313 }
1314
1315 void InterpreterMacroAssembler::verify_method_data_pointer() {
1316 assert(ProfileInterpreter, "must be profiling interpreter");
1317 #ifdef ASSERT
1318 Label verify_continue;
1319 push(rax);
1320 push(rbx);
1321 Register arg3_reg = c_rarg3;
1322 Register arg2_reg = c_rarg2;
1323 push(arg3_reg);
1324 push(arg2_reg);
1325 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1326 get_method(rbx);
1327
1328 // If the mdp is valid, it will point to a DataLayout header which is
1329 // consistent with the bcp. The converse is highly probable also.
1330 load_unsigned_short(arg2_reg,
1331 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1332 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1333 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1334 cmpptr(arg2_reg, _bcp_register);
1335 jcc(Assembler::equal, verify_continue);
1336 // rbx: method
1337 // _bcp_register: bcp
1338 // c_rarg3: mdp
1339 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1340 rbx, _bcp_register, arg3_reg);
1341 bind(verify_continue);
1342 pop(arg2_reg);
1343 pop(arg3_reg);
1344 pop(rbx);
1345 pop(rax);
1346 #endif // ASSERT
1347 }
1348
1349
1350 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1351 int constant,
1352 Register value) {
1353 assert(ProfileInterpreter, "must be profiling interpreter");
1354 Address data(mdp_in, constant);
1355 movptr(data, value);
1356 }
1357
1358
1359 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1360 int constant,
1361 bool decrement) {
1362 // Counter address
1363 Address data(mdp_in, constant);
1364
1365 increment_mdp_data_at(data, decrement);
1366 }
1367
1368 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1369 bool decrement) {
1370 assert(ProfileInterpreter, "must be profiling interpreter");
1371 // %%% this does 64bit counters at best it is wasting space
1372 // at worst it is a rare bug when counters overflow
1373
1374 if (decrement) {
1375 // Decrement the register. Set condition codes.
1376 addptr(data, -DataLayout::counter_increment);
1377 // If the decrement causes the counter to overflow, stay negative
1378 Label L;
1379 jcc(Assembler::negative, L);
1380 addptr(data, DataLayout::counter_increment);
1381 bind(L);
1382 } else {
1383 assert(DataLayout::counter_increment == 1,
1384 "flow-free idiom only works with 1");
1385 // Increment the register. Set carry flag.
1386 addptr(data, DataLayout::counter_increment);
1387 // If the increment causes the counter to overflow, pull back by 1.
1388 sbbptr(data, 0);
1389 }
1390 }
1391
1392
1393 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1394 Register reg,
1395 int constant,
1396 bool decrement) {
1397 Address data(mdp_in, reg, Address::times_1, constant);
1398
1399 increment_mdp_data_at(data, decrement);
1400 }
1401
1402 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1403 int flag_byte_constant) {
1404 assert(ProfileInterpreter, "must be profiling interpreter");
1405 int header_offset = in_bytes(DataLayout::flags_offset());
1406 int header_bits = flag_byte_constant;
1407 // Set the flag
1408 orb(Address(mdp_in, header_offset), header_bits);
1409 }
1410
1411
1412
1413 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1414 int offset,
1415 Register value,
1416 Register test_value_out,
1417 Label& not_equal_continue) {
1418 assert(ProfileInterpreter, "must be profiling interpreter");
1419 if (test_value_out == noreg) {
1420 cmpptr(value, Address(mdp_in, offset));
1421 } else {
1422 // Put the test value into a register, so caller can use it:
1423 movptr(test_value_out, Address(mdp_in, offset));
1424 cmpptr(test_value_out, value);
1425 }
1426 jcc(Assembler::notEqual, not_equal_continue);
1427 }
1428
1429
1430 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1431 int offset_of_disp) {
1432 assert(ProfileInterpreter, "must be profiling interpreter");
1433 Address disp_address(mdp_in, offset_of_disp);
1434 addptr(mdp_in, disp_address);
1435 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1436 }
1437
1438
1439 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1440 Register reg,
1441 int offset_of_disp) {
1442 assert(ProfileInterpreter, "must be profiling interpreter");
1443 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1444 addptr(mdp_in, disp_address);
1445 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1446 }
1447
1448
1449 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1450 int constant) {
1451 assert(ProfileInterpreter, "must be profiling interpreter");
1452 addptr(mdp_in, constant);
1453 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1454 }
1455
1456
1457 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1458 assert(ProfileInterpreter, "must be profiling interpreter");
1459 push(return_bci); // save/restore across call_VM
1460 call_VM(noreg,
1461 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1462 return_bci);
1463 pop(return_bci);
1464 }
1465
1466
1467 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1468 Register bumped_count) {
1469 if (ProfileInterpreter) {
1470 Label profile_continue;
1471
1472 // If no method data exists, go to profile_continue.
1473 // Otherwise, assign to mdp
1474 test_method_data_pointer(mdp, profile_continue);
1475
1476 // We are taking a branch. Increment the taken count.
1477 // We inline increment_mdp_data_at to return bumped_count in a register
1478 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1479 Address data(mdp, in_bytes(JumpData::taken_offset()));
1480 movptr(bumped_count, data);
1481 assert(DataLayout::counter_increment == 1,
1482 "flow-free idiom only works with 1");
1483 addptr(bumped_count, DataLayout::counter_increment);
1484 sbbptr(bumped_count, 0);
1485 movptr(data, bumped_count); // Store back out
1486
1487 // The method data pointer needs to be updated to reflect the new target.
1488 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1489 bind(profile_continue);
1490 }
1491 }
1492
1493
1494 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1495 if (ProfileInterpreter) {
1496 Label profile_continue;
1497
1498 // If no method data exists, go to profile_continue.
1499 test_method_data_pointer(mdp, profile_continue);
1500
1501 // We are taking a branch. Increment the not taken count.
1502 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1503
1504 // The method data pointer needs to be updated to correspond to
1505 // the next bytecode
1506 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()): in_bytes(BranchData::branch_data_size()));
1507 bind(profile_continue);
1508 }
1509 }
1510
1511 void InterpreterMacroAssembler::profile_call(Register mdp) {
1512 if (ProfileInterpreter) {
1513 Label profile_continue;
1514
1515 // If no method data exists, go to profile_continue.
1516 test_method_data_pointer(mdp, profile_continue);
1517
1518 // We are making a call. Increment the count.
1519 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1520
1521 // The method data pointer needs to be updated to reflect the new target.
1522 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1523 bind(profile_continue);
1524 }
1525 }
1526
1527
1528 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1529 if (ProfileInterpreter) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(mdp, profile_continue);
1534
1535 // We are making a call. Increment the count.
1536 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1537
1538 // The method data pointer needs to be updated to reflect the new target.
1539 update_mdp_by_constant(mdp,
1540 in_bytes(VirtualCallData::
1541 virtual_call_data_size()));
1542 bind(profile_continue);
1543 }
1544 }
1545
1546
1547 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1548 Register mdp,
1549 Register reg2,
1550 bool receiver_can_be_null) {
1551 if (ProfileInterpreter) {
1552 Label profile_continue;
1553
1554 // If no method data exists, go to profile_continue.
1555 test_method_data_pointer(mdp, profile_continue);
1556
1557 Label skip_receiver_profile;
1558 if (receiver_can_be_null) {
1559 Label not_null;
1560 testptr(receiver, receiver);
1561 jccb(Assembler::notZero, not_null);
1562 // We are making a call. Increment the count for null receiver.
1563 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1564 jmp(skip_receiver_profile);
1565 bind(not_null);
1566 }
1567
1568 // Record the receiver type.
1569 record_klass_in_profile(receiver, mdp, reg2);
1570 bind(skip_receiver_profile);
1571
1572 // The method data pointer needs to be updated to reflect the new target.
1573 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1574 bind(profile_continue);
1575 }
1576 }
1577
1578 // This routine creates a state machine for updating the multi-row
1579 // type profile at a virtual call site (or other type-sensitive bytecode).
1580 // The machine visits each row (of receiver/count) until the receiver type
1581 // is found, or until it runs out of rows. At the same time, it remembers
1582 // the location of the first empty row. (An empty row records null for its
1583 // receiver, and can be allocated for a newly-observed receiver type.)
1584 // Because there are two degrees of freedom in the state, a simple linear
1585 // search will not work; it must be a decision tree. Hence this helper
1586 // function is recursive, to generate the required tree structured code.
1587 // It's the interpreter, so we are trading off code space for speed.
1588 // See below for example code.
1589 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register mdp,
1590 Register reg2, int start_row,
1591 Label& done) {
1592 if (TypeProfileWidth == 0) {
1593 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1594 } else {
1595 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1596 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1597 }
1598 }
1599
1600 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1601 Label& done, int total_rows,
1602 OffsetFunction item_offset_fn,
1603 OffsetFunction item_count_offset_fn) {
1604 int last_row = total_rows - 1;
1605 assert(start_row <= last_row, "must be work left to do");
1606 // Test this row for both the item and for null.
1607 // Take any of three different outcomes:
1608 // 1. found item => increment count and goto done
1609 // 2. found null => keep looking for case 1, maybe allocate this cell
1610 // 3. found something else => keep looking for cases 1 and 2
1611 // Case 3 is handled by a recursive call.
1612 for (int row = start_row; row <= last_row; row++) {
1613 Label next_test;
1614 bool test_for_null_also = (row == start_row);
1615
1616 // See if the item is item[n].
1617 int item_offset = in_bytes(item_offset_fn(row));
1618 test_mdp_data_at(mdp, item_offset, item,
1619 (test_for_null_also ? reg2 : noreg),
1620 next_test);
1621 // (Reg2 now contains the item from the CallData.)
1622
1623 // The item is item[n]. Increment count[n].
1624 int count_offset = in_bytes(item_count_offset_fn(row));
1625 increment_mdp_data_at(mdp, count_offset);
1626 jmp(done);
1627 bind(next_test);
1628
1629 if (test_for_null_also) {
1630 // Failed the equality check on item[n]... Test for null.
1631 testptr(reg2, reg2);
1632 if (start_row == last_row) {
1633 // The only thing left to do is handle the null case.
1634 Label found_null;
1635 jccb(Assembler::zero, found_null);
1636 // Item did not match any saved item and there is no empty row for it.
1637 // Increment total counter to indicate polymorphic case.
1638 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1639 jmp(done);
1640 bind(found_null);
1641 break;
1642 }
1643 Label found_null;
1644 // Since null is rare, make it be the branch-taken case.
1645 jcc(Assembler::zero, found_null);
1646
1647 // Put all the "Case 3" tests here.
1648 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1649 item_offset_fn, item_count_offset_fn);
1650
1651 // Found a null. Keep searching for a matching item,
1652 // but remember that this is an empty (unused) slot.
1653 bind(found_null);
1654 }
1655 }
1656
1657 // In the fall-through case, we found no matching item, but we
1658 // observed the item[start_row] is null.
1659
1660 // Fill in the item field and increment the count.
1661 int item_offset = in_bytes(item_offset_fn(start_row));
1662 set_mdp_data_at(mdp, item_offset, item);
1663 int count_offset = in_bytes(item_count_offset_fn(start_row));
1664 movl(reg2, DataLayout::counter_increment);
1665 set_mdp_data_at(mdp, count_offset, reg2);
1666 if (start_row > 0) {
1667 jmp(done);
1668 }
1669 }
1670
1671 // Example state machine code for three profile rows:
1672 // // main copy of decision tree, rooted at row[1]
1673 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1674 // if (row[0].rec != nullptr) {
1675 // // inner copy of decision tree, rooted at row[1]
1676 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1677 // if (row[1].rec != nullptr) {
1678 // // degenerate decision tree, rooted at row[2]
1679 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1680 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1681 // row[2].init(rec); goto done;
1682 // } else {
1683 // // remember row[1] is empty
1684 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1685 // row[1].init(rec); goto done;
1686 // }
1687 // } else {
1688 // // remember row[0] is empty
1689 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1690 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1691 // row[0].init(rec); goto done;
1692 // }
1693 // done:
1694
1695 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, Register mdp, Register reg2) {
1696 assert(ProfileInterpreter, "must be profiling");
1697 Label done;
1698
1699 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1700
1701 bind (done);
1702 }
1703
1704 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1705 Register mdp) {
1706 if (ProfileInterpreter) {
1707 Label profile_continue;
1708 uint row;
1709
1710 // If no method data exists, go to profile_continue.
1711 test_method_data_pointer(mdp, profile_continue);
1712
1713 // Update the total ret count.
1714 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1715
1716 for (row = 0; row < RetData::row_limit(); row++) {
1717 Label next_test;
1718
1719 // See if return_bci is equal to bci[n]:
1720 test_mdp_data_at(mdp,
1721 in_bytes(RetData::bci_offset(row)),
1722 return_bci, noreg,
1723 next_test);
1724
1725 // return_bci is equal to bci[n]. Increment the count.
1726 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1727
1728 // The method data pointer needs to be updated to reflect the new target.
1729 update_mdp_by_offset(mdp,
1730 in_bytes(RetData::bci_displacement_offset(row)));
1731 jmp(profile_continue);
1732 bind(next_test);
1733 }
1734
1735 update_mdp_for_ret(return_bci);
1736
1737 bind(profile_continue);
1738 }
1739 }
1740
1741
1742 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1743 if (ProfileInterpreter) {
1744 Label profile_continue;
1745
1746 // If no method data exists, go to profile_continue.
1747 test_method_data_pointer(mdp, profile_continue);
1748
1749 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1750
1751 // The method data pointer needs to be updated.
1752 int mdp_delta = in_bytes(BitData::bit_data_size());
1753 if (TypeProfileCasts) {
1754 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1755 }
1756 update_mdp_by_constant(mdp, mdp_delta);
1757
1758 bind(profile_continue);
1759 }
1760 }
1761
1762
1763 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1764 if (ProfileInterpreter) {
1765 Label profile_continue;
1766
1767 // If no method data exists, go to profile_continue.
1768 test_method_data_pointer(mdp, profile_continue);
1769
1770 // The method data pointer needs to be updated.
1771 int mdp_delta = in_bytes(BitData::bit_data_size());
1772 if (TypeProfileCasts) {
1773 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1774
1775 // Record the object type.
1776 record_klass_in_profile(klass, mdp, reg2);
1777 }
1778 update_mdp_by_constant(mdp, mdp_delta);
1779
1780 bind(profile_continue);
1781 }
1782 }
1783
1784
1785 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1786 if (ProfileInterpreter) {
1787 Label profile_continue;
1788
1789 // If no method data exists, go to profile_continue.
1790 test_method_data_pointer(mdp, profile_continue);
1791
1792 // Update the default case count
1793 increment_mdp_data_at(mdp,
1794 in_bytes(MultiBranchData::default_count_offset()));
1795
1796 // The method data pointer needs to be updated.
1797 update_mdp_by_offset(mdp,
1798 in_bytes(MultiBranchData::
1799 default_displacement_offset()));
1800
1801 bind(profile_continue);
1802 }
1803 }
1804
1805
1806 void InterpreterMacroAssembler::profile_switch_case(Register index,
1807 Register mdp,
1808 Register reg2) {
1809 if (ProfileInterpreter) {
1810 Label profile_continue;
1811
1812 // If no method data exists, go to profile_continue.
1813 test_method_data_pointer(mdp, profile_continue);
1814
1815 // Build the base (index * per_case_size_in_bytes()) +
1816 // case_array_offset_in_bytes()
1817 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1818 imulptr(index, reg2); // XXX l ?
1819 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1820
1821 // Update the case count
1822 increment_mdp_data_at(mdp,
1823 index,
1824 in_bytes(MultiBranchData::relative_count_offset()));
1825
1826 // The method data pointer needs to be updated.
1827 update_mdp_by_offset(mdp,
1828 index,
1829 in_bytes(MultiBranchData::
1830 relative_displacement_offset()));
1831
1832 bind(profile_continue);
1833 }
1834 }
1835
1836 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1837 Register array,
1838 Register tmp) {
1839 if (ProfileInterpreter) {
1840 Label profile_continue;
1841
1842 // If no method data exists, go to profile_continue.
1843 test_method_data_pointer(mdp, profile_continue);
1844
1845 mov(tmp, array);
1846 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1847
1848 Label not_flat;
1849 test_non_flat_array_oop(array, tmp, not_flat);
1850
1851 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1852
1853 bind(not_flat);
1854
1855 Label not_null_free;
1856 test_non_null_free_array_oop(array, tmp, not_null_free);
1857
1858 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1859
1860 bind(not_null_free);
1861
1862 bind(profile_continue);
1863 }
1864 }
1865
1866 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1867 Register array,
1868 Register tmp);
1869 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1870 Register array,
1871 Register tmp);
1872
1873
1874 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1875 if (ProfileInterpreter) {
1876 Label profile_continue;
1877
1878 // If no method data exists, go to profile_continue.
1879 test_method_data_pointer(mdp, profile_continue);
1880
1881 Label done, update;
1882 testptr(element, element);
1883 jccb(Assembler::notZero, update);
1884 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1885 jmp(done);
1886
1887 bind(update);
1888 load_klass(tmp, element, rscratch1);
1889
1890 // Record the object type.
1891 record_klass_in_profile(tmp, mdp, tmp2);
1892
1893 bind(done);
1894
1895 // The method data pointer needs to be updated.
1896 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1897
1898 bind(profile_continue);
1899 }
1900 }
1901
1902 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1903 Register element,
1904 Register tmp) {
1905 if (ProfileInterpreter) {
1906 Label profile_continue;
1907
1908 // If no method data exists, go to profile_continue.
1909 test_method_data_pointer(mdp, profile_continue);
1910
1911 mov(tmp, element);
1912 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1913
1914 // The method data pointer needs to be updated.
1915 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1916
1917 bind(profile_continue);
1918 }
1919 }
1920
1921 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1922 Register left,
1923 Register right,
1924 Register tmp) {
1925 if (ProfileInterpreter) {
1926 Label profile_continue;
1927
1928 // If no method data exists, go to profile_continue.
1929 test_method_data_pointer(mdp, profile_continue);
1930
1931 mov(tmp, left);
1932 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1933
1934 Label left_not_inline_type;
1935 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1936 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1937 bind(left_not_inline_type);
1938
1939 mov(tmp, right);
1940 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1941
1942 Label right_not_inline_type;
1943 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1944 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1945 bind(right_not_inline_type);
1946
1947 bind(profile_continue);
1948 }
1949 }
1950
1951
1952 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1953 if (state == atos) {
1954 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1955 }
1956 }
1957
1958
1959 // Jump if ((*counter_addr += increment) & mask) == 0
1960 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1961 Register scratch, Label* where) {
1962 // This update is actually not atomic and can lose a number of updates
1963 // under heavy contention, but the alternative of using the (contended)
1964 // atomic update here penalizes profiling paths too much.
1965 movl(scratch, counter_addr);
1966 incrementl(scratch, InvocationCounter::count_increment);
1967 movl(counter_addr, scratch);
1968 andl(scratch, mask);
1969 if (where != nullptr) {
1970 jcc(Assembler::zero, *where);
1971 }
1972 }
1973
1974 void InterpreterMacroAssembler::notify_method_entry() {
1975 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1976 // track stack depth. If it is possible to enter interp_only_mode we add
1977 // the code to check if the event should be sent.
1978 Register rthread = r15_thread;
1979 Register rarg = c_rarg1;
1980 if (JvmtiExport::can_post_interpreter_events()) {
1981 Label L;
1982 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1983 testl(rdx, rdx);
1984 jcc(Assembler::zero, L);
1985 call_VM(noreg, CAST_FROM_FN_PTR(address,
1986 InterpreterRuntime::post_method_entry));
1987 bind(L);
1988 }
1989
1990 if (DTraceMethodProbes) {
1991 get_method(rarg);
1992 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1993 rthread, rarg);
1994 }
1995
1996 // RedefineClasses() tracing support for obsolete method entry
1997 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1998 get_method(rarg);
1999 call_VM_leaf(
2000 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2001 rthread, rarg);
2002 }
2003 }
2004
2005
2006 void InterpreterMacroAssembler::notify_method_exit(
2007 TosState state, NotifyMethodExitMode mode) {
2008 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2009 // track stack depth. If it is possible to enter interp_only_mode we add
2010 // the code to check if the event should be sent.
2011 Register rthread = r15_thread;
2012 Register rarg = c_rarg1;
2013 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2014 Label L;
2015 // Note: frame::interpreter_frame_result has a dependency on how the
2016 // method result is saved across the call to post_method_exit. If this
2017 // is changed then the interpreter_frame_result implementation will
2018 // need to be updated too.
2019
2020 // template interpreter will leave the result on the top of the stack.
2021 push(state);
2022 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2023 testl(rdx, rdx);
2024 jcc(Assembler::zero, L);
2025 call_VM(noreg,
2026 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2027 bind(L);
2028 pop(state);
2029 }
2030
2031 if (DTraceMethodProbes) {
2032 push(state);
2033 get_method(rarg);
2034 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2035 rthread, rarg);
2036 pop(state);
2037 }
2038 }
2039
2040 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2041 // Get index out of bytecode pointer
2042 get_cache_index_at_bcp(index, 1, sizeof(u4));
2043 // Get address of invokedynamic array
2044 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2045 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2046 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2047 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2048 } else {
2049 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2050 }
2051 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2052 }
2053
2054 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2055 // Get index out of bytecode pointer
2056 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2057 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2058
2059 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2060 // Take shortcut if the size is a power of 2
2061 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2062 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2063 } else {
2064 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2065 }
2066 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2067 }
2068
2069 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2070 // Get index out of bytecode pointer
2071 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2072 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2073
2074 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2075 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2076 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2077 }