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