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/safepointMechanism.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "utilities/powerOfTwo.hpp"
45
46 // Implementation of InterpreterMacroAssembler
47
48 void InterpreterMacroAssembler::jump_to_entry(address entry) {
49 assert(entry, "Entry must have been generated by now");
50 jump(RuntimeAddress(entry));
51 }
52
53 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
54 Label update, next, none;
55
56 #ifdef _LP64
57 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
58 #else
59 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
60 #endif
61
62 interp_verify_oop(obj, atos);
63
64 testptr(obj, obj);
65 jccb(Assembler::notZero, update);
66 testptr(mdo_addr, TypeEntries::null_seen);
67 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
68 // atomic update to prevent overwriting Klass* with 0
69 lock();
70 orptr(mdo_addr, TypeEntries::null_seen);
71 jmpb(next);
72
73 bind(update);
74 load_klass(obj, obj, rscratch1);
75 #ifdef _LP64
76 mov(rscratch1, obj);
77 #endif
78
79 xorptr(obj, mdo_addr);
80 testptr(obj, TypeEntries::type_klass_mask);
81 jccb(Assembler::zero, next); // klass seen before, nothing to
82 // do. The unknown bit may have been
83 // set already but no need to check.
84
85 testptr(obj, TypeEntries::type_unknown);
86 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
87
88 cmpptr(mdo_addr, 0);
89 jccb(Assembler::equal, none);
90 cmpptr(mdo_addr, TypeEntries::null_seen);
91 jccb(Assembler::equal, none);
92 #ifdef _LP64
93 // There is a chance that the checks above (re-reading profiling
94 // data from memory) fail if another thread has just set the
95 // profiling to this obj's klass
96 mov(obj, rscratch1);
97 xorptr(obj, mdo_addr);
98 testptr(obj, TypeEntries::type_klass_mask);
99 jccb(Assembler::zero, next);
100 #endif
101
102 // different than before. Cannot keep accurate profile.
103 orptr(mdo_addr, TypeEntries::type_unknown);
104 jmpb(next);
105
106 bind(none);
107 // first time here. Set profile type.
108 movptr(mdo_addr, obj);
109 #ifdef ASSERT
110 andptr(obj, TypeEntries::type_klass_mask);
111 verify_klass_ptr(obj);
112 #endif
113
114 bind(next);
115 }
116
117 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
118 if (!ProfileInterpreter) {
119 return;
120 }
121
122 if (MethodData::profile_arguments() || MethodData::profile_return()) {
123 Label profile_continue;
124
125 test_method_data_pointer(mdp, profile_continue);
126
127 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
128
129 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
130 jcc(Assembler::notEqual, profile_continue);
131
132 if (MethodData::profile_arguments()) {
133 Label done;
134 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
135 addptr(mdp, off_to_args);
136
137 for (int i = 0; i < TypeProfileArgsLimit; i++) {
138 if (i > 0 || MethodData::profile_return()) {
139 // If return value type is profiled we may have no argument to profile
140 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
141 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
142 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
143 jcc(Assembler::less, done);
144 }
145 movptr(tmp, Address(callee, Method::const_offset()));
146 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
147 // stack offset o (zero based) from the start of the argument
148 // list, for n arguments translates into offset n - o - 1 from
149 // the end of the argument list
150 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
151 subl(tmp, 1);
152 Address arg_addr = argument_address(tmp);
153 movptr(tmp, arg_addr);
154
155 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
156 profile_obj_type(tmp, mdo_arg_addr);
157
158 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
159 addptr(mdp, to_add);
160 off_to_args += to_add;
161 }
162
163 if (MethodData::profile_return()) {
164 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
165 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
166 }
167
168 bind(done);
169
170 if (MethodData::profile_return()) {
171 // We're right after the type profile for the last
172 // argument. tmp is the number of cells left in the
173 // CallTypeData/VirtualCallTypeData to reach its end. Non null
174 // if there's a return to profile.
175 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
176 shll(tmp, log2i_exact((int)DataLayout::cell_size));
177 addptr(mdp, tmp);
178 }
179 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
180 } else {
181 assert(MethodData::profile_return(), "either profile call args or call ret");
182 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
183 }
184
185 // mdp points right after the end of the
186 // CallTypeData/VirtualCallTypeData, right after the cells for the
187 // return value type if there's one
188
189 bind(profile_continue);
190 }
191 }
192
193 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
194 assert_different_registers(mdp, ret, tmp, _bcp_register);
195 if (ProfileInterpreter && MethodData::profile_return()) {
196 Label profile_continue;
197
198 test_method_data_pointer(mdp, profile_continue);
199
200 if (MethodData::profile_return_jsr292_only()) {
201 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
202
203 // If we don't profile all invoke bytecodes we must make sure
204 // it's a bytecode we indeed profile. We can't go back to the
205 // beginning of the ProfileData we intend to update to check its
206 // type because we're right after it and we don't known its
207 // length
208 Label do_profile;
209 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
210 jcc(Assembler::equal, do_profile);
211 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
212 jcc(Assembler::equal, do_profile);
213 get_method(tmp);
214 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
215 jcc(Assembler::notEqual, profile_continue);
216
217 bind(do_profile);
218 }
219
220 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
221 mov(tmp, ret);
222 profile_obj_type(tmp, mdo_ret_addr);
223
224 bind(profile_continue);
225 }
226 }
227
228 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
229 if (ProfileInterpreter && MethodData::profile_parameters()) {
230 Label profile_continue;
231
232 test_method_data_pointer(mdp, profile_continue);
233
234 // Load the offset of the area within the MDO used for
235 // parameters. If it's negative we're not profiling any parameters
236 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
237 testl(tmp1, tmp1);
238 jcc(Assembler::negative, profile_continue);
239
240 // Compute a pointer to the area for parameters from the offset
241 // and move the pointer to the slot for the last
242 // parameters. Collect profiling from last parameter down.
243 // mdo start + parameters offset + array length - 1
244 addptr(mdp, tmp1);
245 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
246 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
247
248 Label loop;
249 bind(loop);
250
251 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
252 int type_base = in_bytes(ParametersTypeData::type_offset(0));
253 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
254 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
255 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
256
257 // load offset on the stack from the slot for this parameter
258 movptr(tmp2, arg_off);
259 negptr(tmp2);
260 // read the parameter from the local area
261 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
262
263 // profile the parameter
264 profile_obj_type(tmp2, arg_type);
265
266 // go to next parameter
267 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
268 jcc(Assembler::positive, loop);
269
270 bind(profile_continue);
271 }
272 }
273
274 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
275 int number_of_arguments) {
276 // interpreter specific
277 //
278 // Note: No need to save/restore bcp & locals registers
279 // since these are callee saved registers and no blocking/
280 // GC can happen in leaf calls.
281 // Further Note: DO NOT save/restore bcp/locals. If a caller has
282 // already saved them so that it can use rsi/rdi as temporaries
283 // then a save/restore here will DESTROY the copy the caller
284 // saved! There used to be a save_bcp() that only happened in
285 // the ASSERT path (no restore_bcp). Which caused bizarre failures
286 // when jvm built with ASSERTs.
287 #ifdef ASSERT
288 {
289 Label L;
290 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
291 jcc(Assembler::equal, L);
292 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
293 " last_sp != null");
294 bind(L);
295 }
296 #endif
297 // super call
298 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
299 // interpreter specific
300 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
301 // but since they may not have been saved (and we don't want to
302 // save them here (see note above) the assert is invalid.
303 }
304
305 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
306 Register java_thread,
307 Register last_java_sp,
308 address entry_point,
309 int number_of_arguments,
310 bool check_exceptions) {
311 // interpreter specific
312 //
313 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
314 // really make a difference for these runtime calls, since they are
315 // slow anyway. Btw., bcp must be saved/restored since it may change
316 // due to GC.
317 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
318 save_bcp();
319 #ifdef ASSERT
320 {
321 Label L;
322 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
323 jcc(Assembler::equal, L);
324 stop("InterpreterMacroAssembler::call_VM_base:"
325 " last_sp isn't null");
326 bind(L);
327 }
328 #endif /* ASSERT */
329 // super call
330 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
331 entry_point, number_of_arguments,
332 check_exceptions);
333 // interpreter specific
334 restore_bcp();
335 restore_locals();
336 }
337
338 #ifdef _LP64
339 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
340 address entry_point,
341 Register arg_1) {
342 assert(arg_1 == c_rarg1, "");
343 Label resume_pc, not_preempted;
344
345 #ifdef ASSERT
346 {
347 Label L;
348 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
349 jcc(Assembler::equal, L);
350 stop("Should not have alternate return address set");
351 bind(L);
352 }
353 #endif /* ASSERT */
354
355 // Force freeze slow path.
356 push_cont_fastpath();
357
358 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
359 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
360 lea(rscratch1, resume_pc);
361 push(rscratch1);
362 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
363 pop(rscratch1);
364
365 pop_cont_fastpath();
366
367 // Check if preempted.
368 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
369 cmpptr(rscratch1, NULL_WORD);
370 jccb(Assembler::zero, not_preempted);
371 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
372 jmp(rscratch1);
373
374 // In case of preemption, this is where we will resume once we finally acquire the monitor.
375 bind(resume_pc);
376 restore_after_resume(false /* is_native */);
377
378 bind(not_preempted);
379 }
380
381 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
382 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
383 call(rscratch1);
384 if (is_native) {
385 // On resume we need to set up stack as expected.
386 push(dtos);
387 push(ltos);
388 }
389 }
390 #else
391 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
392 address entry_point,
393 Register arg_1) {
394 MacroAssembler::call_VM(oop_result, entry_point, arg_1);
395 }
396 #endif // _LP64
397
398 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
399 if (JvmtiExport::can_pop_frame()) {
400 Label L;
401 // Initiate popframe handling only if it is not already being
402 // processed. If the flag has the popframe_processing bit set, it
403 // means that this code is called *during* popframe handling - we
404 // don't want to reenter.
405 // This method is only called just after the call into the vm in
406 // call_VM_base, so the arg registers are available.
407 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
408 LP64_ONLY(c_rarg0);
409 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
410 testl(pop_cond, JavaThread::popframe_pending_bit);
411 jcc(Assembler::zero, L);
412 testl(pop_cond, JavaThread::popframe_processing_bit);
413 jcc(Assembler::notZero, L);
414 // Call Interpreter::remove_activation_preserving_args_entry() to get the
415 // address of the same-named entrypoint in the generated interpreter code.
416 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
417 jmp(rax);
418 bind(L);
419 NOT_LP64(get_thread(java_thread);)
420 }
421 }
422
423 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
424 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
425 NOT_LP64(get_thread(thread);)
426 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
427 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
428 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
429 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
430 #ifdef _LP64
431 switch (state) {
432 case atos: movptr(rax, oop_addr);
433 movptr(oop_addr, NULL_WORD);
434 interp_verify_oop(rax, state); break;
435 case ltos: movptr(rax, val_addr); break;
436 case btos: // fall through
437 case ztos: // fall through
438 case ctos: // fall through
439 case stos: // fall through
440 case itos: movl(rax, val_addr); break;
441 case ftos: load_float(val_addr); break;
442 case dtos: load_double(val_addr); break;
443 case vtos: /* nothing to do */ break;
444 default : ShouldNotReachHere();
445 }
446 // Clean up tos value in the thread object
447 movl(tos_addr, ilgl);
448 movl(val_addr, NULL_WORD);
449 #else
450 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
451 + in_ByteSize(wordSize));
452 switch (state) {
453 case atos: movptr(rax, oop_addr);
454 movptr(oop_addr, NULL_WORD);
455 interp_verify_oop(rax, state); break;
456 case ltos:
457 movl(rdx, val_addr1); // fall through
458 case btos: // fall through
459 case ztos: // fall through
460 case ctos: // fall through
461 case stos: // fall through
462 case itos: movl(rax, val_addr); break;
463 case ftos: load_float(val_addr); break;
464 case dtos: load_double(val_addr); break;
465 case vtos: /* nothing to do */ break;
466 default : ShouldNotReachHere();
467 }
468 #endif // _LP64
469 // Clean up tos value in the thread object
470 movl(tos_addr, ilgl);
471 movptr(val_addr, NULL_WORD);
472 NOT_LP64(movptr(val_addr1, NULL_WORD);)
473 }
474
475
476 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
477 if (JvmtiExport::can_force_early_return()) {
478 Label L;
479 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
480 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
481
482 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
483 testptr(tmp, tmp);
484 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
485
486 // Initiate earlyret handling only if it is not already being processed.
487 // If the flag has the earlyret_processing bit set, it means that this code
488 // is called *during* earlyret handling - we don't want to reenter.
489 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
490 cmpl(tmp, JvmtiThreadState::earlyret_pending);
491 jcc(Assembler::notEqual, L);
492
493 // Call Interpreter::remove_activation_early_entry() to get the address of the
494 // same-named entrypoint in the generated interpreter code.
495 NOT_LP64(get_thread(java_thread);)
496 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
497 #ifdef _LP64
498 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
499 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
500 #else
501 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
502 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
503 #endif // _LP64
504 jmp(rax);
505 bind(L);
506 NOT_LP64(get_thread(java_thread);)
507 }
508 }
509
510 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
511 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
512 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
513 bswapl(reg);
514 shrl(reg, 16);
515 }
516
517 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
518 int bcp_offset,
519 size_t index_size) {
520 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
521 if (index_size == sizeof(u2)) {
522 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
523 } else if (index_size == sizeof(u4)) {
524 movl(index, Address(_bcp_register, bcp_offset));
525 } else if (index_size == sizeof(u1)) {
526 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
527 } else {
528 ShouldNotReachHere();
529 }
530 }
531
532 // Load object from cpool->resolved_references(index)
533 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
534 Register index,
535 Register tmp) {
536 assert_different_registers(result, index);
537
538 get_constant_pool(result);
539 // load pointer for resolved_references[] objArray
540 movptr(result, Address(result, ConstantPool::cache_offset()));
541 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
542 resolve_oop_handle(result, tmp);
543 load_heap_oop(result, Address(result, index,
544 UseCompressedOops ? Address::times_4 : Address::times_ptr,
545 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
546 }
547
548 // load cpool->resolved_klass_at(index)
549 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
550 Register cpool,
551 Register index) {
552 assert_different_registers(cpool, index);
553
554 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
555 Register resolved_klasses = cpool;
556 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
557 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
558 }
559
560 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
561 // subtype of super_klass.
562 //
563 // Args:
564 // rax: superklass
565 // Rsub_klass: subklass
566 //
567 // Kills:
568 // rcx, rdi
569 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
570 Label& ok_is_subtype) {
571 assert(Rsub_klass != rax, "rax holds superklass");
572 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
573 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
574 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
575 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
576
577 // Profile the not-null value's klass.
578 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
579
580 // Do the check.
581 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
582 }
583
584
585 #ifndef _LP64
586 void InterpreterMacroAssembler::f2ieee() {
587 if (IEEEPrecision) {
588 fstp_s(Address(rsp, 0));
589 fld_s(Address(rsp, 0));
590 }
591 }
592
593
594 void InterpreterMacroAssembler::d2ieee() {
595 if (IEEEPrecision) {
596 fstp_d(Address(rsp, 0));
597 fld_d(Address(rsp, 0));
598 }
599 }
600 #endif // _LP64
601
602 // Java Expression Stack
603
604 void InterpreterMacroAssembler::pop_ptr(Register r) {
605 pop(r);
606 }
607
608 void InterpreterMacroAssembler::push_ptr(Register r) {
609 push(r);
610 }
611
612 void InterpreterMacroAssembler::push_i(Register r) {
613 push(r);
614 }
615
616 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
617 push(r);
618 }
619
620 void InterpreterMacroAssembler::push_f(XMMRegister r) {
621 subptr(rsp, wordSize);
622 movflt(Address(rsp, 0), r);
623 }
624
625 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
626 movflt(r, Address(rsp, 0));
627 addptr(rsp, wordSize);
628 }
629
630 void InterpreterMacroAssembler::push_d(XMMRegister r) {
631 subptr(rsp, 2 * wordSize);
632 movdbl(Address(rsp, 0), r);
633 }
634
635 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
636 movdbl(r, Address(rsp, 0));
637 addptr(rsp, 2 * Interpreter::stackElementSize);
638 }
639
640 #ifdef _LP64
641 void InterpreterMacroAssembler::pop_i(Register r) {
642 // XXX can't use pop currently, upper half non clean
643 movl(r, Address(rsp, 0));
644 addptr(rsp, wordSize);
645 }
646
647 void InterpreterMacroAssembler::pop_l(Register r) {
648 movq(r, Address(rsp, 0));
649 addptr(rsp, 2 * Interpreter::stackElementSize);
650 }
651
652 void InterpreterMacroAssembler::push_l(Register r) {
653 subptr(rsp, 2 * wordSize);
654 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
655 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
656 }
657
658 void InterpreterMacroAssembler::pop(TosState state) {
659 switch (state) {
660 case atos: pop_ptr(); break;
661 case btos:
662 case ztos:
663 case ctos:
664 case stos:
665 case itos: pop_i(); break;
666 case ltos: pop_l(); break;
667 case ftos: pop_f(xmm0); break;
668 case dtos: pop_d(xmm0); break;
669 case vtos: /* nothing to do */ break;
670 default: ShouldNotReachHere();
671 }
672 interp_verify_oop(rax, state);
673 }
674
675 void InterpreterMacroAssembler::push(TosState state) {
676 interp_verify_oop(rax, state);
677 switch (state) {
678 case atos: push_ptr(); break;
679 case btos:
680 case ztos:
681 case ctos:
682 case stos:
683 case itos: push_i(); break;
684 case ltos: push_l(); break;
685 case ftos: push_f(xmm0); break;
686 case dtos: push_d(xmm0); break;
687 case vtos: /* nothing to do */ break;
688 default : ShouldNotReachHere();
689 }
690 }
691 #else
692 void InterpreterMacroAssembler::pop_i(Register r) {
693 pop(r);
694 }
695
696 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
697 pop(lo);
698 pop(hi);
699 }
700
701 void InterpreterMacroAssembler::pop_f() {
702 fld_s(Address(rsp, 0));
703 addptr(rsp, 1 * wordSize);
704 }
705
706 void InterpreterMacroAssembler::pop_d() {
707 fld_d(Address(rsp, 0));
708 addptr(rsp, 2 * wordSize);
709 }
710
711
712 void InterpreterMacroAssembler::pop(TosState state) {
713 switch (state) {
714 case atos: pop_ptr(rax); break;
715 case btos: // fall through
716 case ztos: // fall through
717 case ctos: // fall through
718 case stos: // fall through
719 case itos: pop_i(rax); break;
720 case ltos: pop_l(rax, rdx); break;
721 case ftos:
722 if (UseSSE >= 1) {
723 pop_f(xmm0);
724 } else {
725 pop_f();
726 }
727 break;
728 case dtos:
729 if (UseSSE >= 2) {
730 pop_d(xmm0);
731 } else {
732 pop_d();
733 }
734 break;
735 case vtos: /* nothing to do */ break;
736 default : ShouldNotReachHere();
737 }
738 interp_verify_oop(rax, state);
739 }
740
741
742 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
743 push(hi);
744 push(lo);
745 }
746
747 void InterpreterMacroAssembler::push_f() {
748 // Do not schedule for no AGI! Never write beyond rsp!
749 subptr(rsp, 1 * wordSize);
750 fstp_s(Address(rsp, 0));
751 }
752
753 void InterpreterMacroAssembler::push_d() {
754 // Do not schedule for no AGI! Never write beyond rsp!
755 subptr(rsp, 2 * wordSize);
756 fstp_d(Address(rsp, 0));
757 }
758
759
760 void InterpreterMacroAssembler::push(TosState state) {
761 interp_verify_oop(rax, state);
762 switch (state) {
763 case atos: push_ptr(rax); break;
764 case btos: // fall through
765 case ztos: // fall through
766 case ctos: // fall through
767 case stos: // fall through
768 case itos: push_i(rax); break;
769 case ltos: push_l(rax, rdx); break;
770 case ftos:
771 if (UseSSE >= 1) {
772 push_f(xmm0);
773 } else {
774 push_f();
775 }
776 break;
777 case dtos:
778 if (UseSSE >= 2) {
779 push_d(xmm0);
780 } else {
781 push_d();
782 }
783 break;
784 case vtos: /* nothing to do */ break;
785 default : ShouldNotReachHere();
786 }
787 }
788 #endif // _LP64
789
790
791 // Helpers for swap and dup
792 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
793 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
794 }
795
796 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
797 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
798 }
799
800
801 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
802 // set sender sp
803 lea(_bcp_register, Address(rsp, wordSize));
804 // record last_sp
805 mov(rcx, _bcp_register);
806 subptr(rcx, rbp);
807 sarptr(rcx, LogBytesPerWord);
808 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
809 }
810
811
812 // Jump to from_interpreted entry of a call unless single stepping is possible
813 // in this thread in which case we must call the i2i entry
814 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
815 prepare_to_jump_from_interpreted();
816
817 if (JvmtiExport::can_post_interpreter_events()) {
818 Label run_compiled_code;
819 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
820 // compiled code in threads for which the event is enabled. Check here for
821 // interp_only_mode if these events CAN be enabled.
822 // interp_only is an int, on little endian it is sufficient to test the byte only
823 // Is a cmpl faster?
824 LP64_ONLY(temp = r15_thread;)
825 NOT_LP64(get_thread(temp);)
826 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
827 jccb(Assembler::zero, run_compiled_code);
828 jmp(Address(method, Method::interpreter_entry_offset()));
829 bind(run_compiled_code);
830 }
831
832 jmp(Address(method, Method::from_interpreted_offset()));
833 }
834
835 // The following two routines provide a hook so that an implementation
836 // can schedule the dispatch in two parts. x86 does not do this.
837 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
838 // Nothing x86 specific to be done here
839 }
840
841 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
842 dispatch_next(state, step);
843 }
844
845 void InterpreterMacroAssembler::dispatch_base(TosState state,
846 address* table,
847 bool verifyoop,
848 bool generate_poll) {
849 verify_FPU(1, state);
850 if (VerifyActivationFrameSize) {
851 Label L;
852 mov(rcx, rbp);
853 subptr(rcx, rsp);
854 int32_t min_frame_size =
855 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
856 wordSize;
857 cmpptr(rcx, min_frame_size);
858 jcc(Assembler::greaterEqual, L);
859 stop("broken stack frame");
860 bind(L);
861 }
862 if (verifyoop) {
863 interp_verify_oop(rax, state);
864 }
865
866 address* const safepoint_table = Interpreter::safept_table(state);
867 #ifdef _LP64
868 Label no_safepoint, dispatch;
869 if (table != safepoint_table && generate_poll) {
870 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
871 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
872
873 jccb(Assembler::zero, no_safepoint);
874 lea(rscratch1, ExternalAddress((address)safepoint_table));
875 jmpb(dispatch);
876 }
877
878 bind(no_safepoint);
879 lea(rscratch1, ExternalAddress((address)table));
880 bind(dispatch);
881 jmp(Address(rscratch1, rbx, Address::times_8));
882
883 #else
884 Address index(noreg, rbx, Address::times_ptr);
885 if (table != safepoint_table && generate_poll) {
886 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
887 Label no_safepoint;
888 const Register thread = rcx;
889 get_thread(thread);
890 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
891
892 jccb(Assembler::zero, no_safepoint);
893 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
894 jump(dispatch_addr, noreg);
895 bind(no_safepoint);
896 }
897
898 {
899 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
900 jump(dispatch_addr, noreg);
901 }
902 #endif // _LP64
903 }
904
905 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
906 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
907 }
908
909 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
910 dispatch_base(state, Interpreter::normal_table(state));
911 }
912
913 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
914 dispatch_base(state, Interpreter::normal_table(state), false);
915 }
916
917
918 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
919 // load next bytecode (load before advancing _bcp_register to prevent AGI)
920 load_unsigned_byte(rbx, Address(_bcp_register, step));
921 // advance _bcp_register
922 increment(_bcp_register, step);
923 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
924 }
925
926 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
927 // load current bytecode
928 load_unsigned_byte(rbx, Address(_bcp_register, 0));
929 dispatch_base(state, table);
930 }
931
932 void InterpreterMacroAssembler::narrow(Register result) {
933
934 // Get method->_constMethod->_result_type
935 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
936 movptr(rcx, Address(rcx, Method::const_offset()));
937 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
938
939 Label done, notBool, notByte, notChar;
940
941 // common case first
942 cmpl(rcx, T_INT);
943 jcc(Assembler::equal, done);
944
945 // mask integer result to narrower return type.
946 cmpl(rcx, T_BOOLEAN);
947 jcc(Assembler::notEqual, notBool);
948 andl(result, 0x1);
949 jmp(done);
950
951 bind(notBool);
952 cmpl(rcx, T_BYTE);
953 jcc(Assembler::notEqual, notByte);
954 LP64_ONLY(movsbl(result, result);)
955 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
956 NOT_LP64(sarl(result, 24);) // and sign-extend byte
957 jmp(done);
958
959 bind(notByte);
960 cmpl(rcx, T_CHAR);
961 jcc(Assembler::notEqual, notChar);
962 LP64_ONLY(movzwl(result, result);)
963 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
964 jmp(done);
965
966 bind(notChar);
967 // cmpl(rcx, T_SHORT); // all that's left
968 // jcc(Assembler::notEqual, done);
969 LP64_ONLY(movswl(result, result);)
970 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
971 NOT_LP64(sarl(result, 16);) // and sign-extend short
972
973 // Nothing to do for T_INT
974 bind(done);
975 }
976
977 // remove activation
978 //
979 // Apply stack watermark barrier.
980 // Unlock the receiver if this is a synchronized method.
981 // Unlock any Java monitors from synchronized blocks.
982 // Remove the activation from the stack.
983 //
984 // If there are locked Java monitors
985 // If throw_monitor_exception
986 // throws IllegalMonitorStateException
987 // Else if install_monitor_exception
988 // installs IllegalMonitorStateException
989 // Else
990 // no error processing
991 void InterpreterMacroAssembler::remove_activation(
992 TosState state,
993 Register ret_addr,
994 bool throw_monitor_exception,
995 bool install_monitor_exception,
996 bool notify_jvmdi) {
997 // Note: Registers rdx xmm0 may be in use for the
998 // result check if synchronized method
999 Label unlocked, unlock, no_unlock;
1000
1001 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1002 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
1003 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
1004 // monitor pointers need different register
1005 // because rdx may have the result in it
1006 NOT_LP64(get_thread(rthread);)
1007
1008 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
1009 // that would normally not be safe to use. Such bad returns into unsafe territory of
1010 // the stack, will call InterpreterRuntime::at_unwind.
1011 Label slow_path;
1012 Label fast_path;
1013 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
1014 jmp(fast_path);
1015 bind(slow_path);
1016 push(state);
1017 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
1018 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1019 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
1020 reset_last_Java_frame(rthread, true);
1021 pop(state);
1022 bind(fast_path);
1023
1024 // get the value of _do_not_unlock_if_synchronized into rdx
1025 const Address do_not_unlock_if_synchronized(rthread,
1026 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1027 movbool(rbx, do_not_unlock_if_synchronized);
1028 movbool(do_not_unlock_if_synchronized, false); // reset the flag
1029
1030 // get method access flags
1031 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1032 load_unsigned_short(rcx, Address(rcx, Method::access_flags_offset()));
1033 testl(rcx, JVM_ACC_SYNCHRONIZED);
1034 jcc(Assembler::zero, unlocked);
1035
1036 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1037 // is set.
1038 testbool(rbx);
1039 jcc(Assembler::notZero, no_unlock);
1040
1041 // unlock monitor
1042 push(state); // save result
1043
1044 // BasicObjectLock will be first in list, since this is a
1045 // synchronized method. However, need to check that the object has
1046 // not been unlocked by an explicit monitorexit bytecode.
1047 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1048 wordSize - (int) sizeof(BasicObjectLock));
1049 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1050 // register for unlock_object to pass to VM directly
1051 lea(robj, monitor); // address of first monitor
1052
1053 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
1054 testptr(rax, rax);
1055 jcc(Assembler::notZero, unlock);
1056
1057 pop(state);
1058 if (throw_monitor_exception) {
1059 // Entry already unlocked, need to throw exception
1060 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1061 call_VM(noreg, CAST_FROM_FN_PTR(address,
1062 InterpreterRuntime::throw_illegal_monitor_state_exception));
1063 should_not_reach_here();
1064 } else {
1065 // Monitor already unlocked during a stack unroll. If requested,
1066 // install an illegal_monitor_state_exception. Continue with
1067 // stack unrolling.
1068 if (install_monitor_exception) {
1069 NOT_LP64(empty_FPU_stack();)
1070 call_VM(noreg, CAST_FROM_FN_PTR(address,
1071 InterpreterRuntime::new_illegal_monitor_state_exception));
1072 }
1073 jmp(unlocked);
1074 }
1075
1076 bind(unlock);
1077 unlock_object(robj);
1078 pop(state);
1079
1080 // Check that for block-structured locking (i.e., that all locked
1081 // objects has been unlocked)
1082 bind(unlocked);
1083
1084 // rax, rdx: Might contain return value
1085
1086 // Check that all monitors are unlocked
1087 {
1088 Label loop, exception, entry, restart;
1089 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1090 const Address monitor_block_top(
1091 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1092 const Address monitor_block_bot(
1093 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1094
1095 bind(restart);
1096 // We use c_rarg1 so that if we go slow path it will be the correct
1097 // register for unlock_object to pass to VM directly
1098 movptr(rmon, monitor_block_top); // derelativize pointer
1099 lea(rmon, Address(rbp, rmon, Address::times_ptr));
1100 // c_rarg1 points to current entry, starting with top-most entry
1101
1102 lea(rbx, monitor_block_bot); // points to word before bottom of
1103 // monitor block
1104 jmp(entry);
1105
1106 // Entry already locked, need to throw exception
1107 bind(exception);
1108
1109 if (throw_monitor_exception) {
1110 // Throw exception
1111 NOT_LP64(empty_FPU_stack();)
1112 MacroAssembler::call_VM(noreg,
1113 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1114 throw_illegal_monitor_state_exception));
1115 should_not_reach_here();
1116 } else {
1117 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1118 // Unlock does not block, so don't have to worry about the frame.
1119 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1120
1121 push(state);
1122 mov(robj, rmon); // nop if robj and rmon are the same
1123 unlock_object(robj);
1124 pop(state);
1125
1126 if (install_monitor_exception) {
1127 NOT_LP64(empty_FPU_stack();)
1128 call_VM(noreg, CAST_FROM_FN_PTR(address,
1129 InterpreterRuntime::
1130 new_illegal_monitor_state_exception));
1131 }
1132
1133 jmp(restart);
1134 }
1135
1136 bind(loop);
1137 // check if current entry is used
1138 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1139 jcc(Assembler::notEqual, exception);
1140
1141 addptr(rmon, entry_size); // otherwise advance to next entry
1142 bind(entry);
1143 cmpptr(rmon, rbx); // check if bottom reached
1144 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1145 }
1146
1147 bind(no_unlock);
1148
1149 // jvmti support
1150 if (notify_jvmdi) {
1151 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1152 } else {
1153 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1154 }
1155
1156 // remove activation
1157 // get sender sp
1158 movptr(rbx,
1159 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1160 if (StackReservedPages > 0) {
1161 // testing if reserved zone needs to be re-enabled
1162 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1163 Label no_reserved_zone_enabling;
1164
1165 NOT_LP64(get_thread(rthread);)
1166
1167 // check if already enabled - if so no re-enabling needed
1168 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1169 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1170 jcc(Assembler::equal, no_reserved_zone_enabling);
1171
1172 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1173 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1174
1175 call_VM_leaf(
1176 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1177 call_VM(noreg, CAST_FROM_FN_PTR(address,
1178 InterpreterRuntime::throw_delayed_StackOverflowError));
1179 should_not_reach_here();
1180
1181 bind(no_reserved_zone_enabling);
1182 }
1183 leave(); // remove frame anchor
1184 pop(ret_addr); // get return address
1185 mov(rsp, rbx); // set sp to sender sp
1186 pop_cont_fastpath();
1187 }
1188
1189 void InterpreterMacroAssembler::get_method_counters(Register method,
1190 Register mcs, Label& skip) {
1191 Label has_counters;
1192 movptr(mcs, Address(method, Method::method_counters_offset()));
1193 testptr(mcs, mcs);
1194 jcc(Assembler::notZero, has_counters);
1195 call_VM(noreg, CAST_FROM_FN_PTR(address,
1196 InterpreterRuntime::build_method_counters), method);
1197 movptr(mcs, Address(method,Method::method_counters_offset()));
1198 testptr(mcs, mcs);
1199 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1200 bind(has_counters);
1201 }
1202
1203
1204 // Lock object
1205 //
1206 // Args:
1207 // rdx, c_rarg1: BasicObjectLock to be used for locking
1208 //
1209 // Kills:
1210 // rax, rbx
1211 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1212 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1213 "The argument is only for looks. It must be c_rarg1");
1214
1215 if (LockingMode == LM_MONITOR) {
1216 call_VM_preemptable(noreg,
1217 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1218 lock_reg);
1219 } else {
1220 Label count_locking, done, slow_case;
1221
1222 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1223 const Register tmp_reg = rbx;
1224 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1225 const Register rklass_decode_tmp = rscratch1;
1226
1227 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1228 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1229 const int mark_offset = lock_offset +
1230 BasicLock::displaced_header_offset_in_bytes();
1231
1232 // Load object pointer into obj_reg
1233 movptr(obj_reg, Address(lock_reg, obj_offset));
1234
1235 if (DiagnoseSyncOnValueBasedClasses != 0) {
1236 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1237 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1238 jcc(Assembler::notZero, slow_case);
1239 }
1240
1241 if (LockingMode == LM_LIGHTWEIGHT) {
1242 #ifdef _LP64
1243 const Register thread = r15_thread;
1244 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1245 #else
1246 // Lacking registers and thread on x86_32. Always take slow path.
1247 jmp(slow_case);
1248 #endif
1249 } else if (LockingMode == LM_LEGACY) {
1250 // Load immediate 1 into swap_reg %rax
1251 movl(swap_reg, 1);
1252
1253 // Load (object->mark() | 1) into swap_reg %rax
1254 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1255
1256 // Save (object->mark() | 1) into BasicLock's displaced header
1257 movptr(Address(lock_reg, mark_offset), swap_reg);
1258
1259 assert(lock_offset == 0,
1260 "displaced header must be first word in BasicObjectLock");
1261
1262 lock();
1263 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1264 jcc(Assembler::zero, count_locking);
1265
1266 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1267
1268 // Fast check for recursive lock.
1269 //
1270 // Can apply the optimization only if this is a stack lock
1271 // allocated in this thread. For efficiency, we can focus on
1272 // recently allocated stack locks (instead of reading the stack
1273 // base and checking whether 'mark' points inside the current
1274 // thread stack):
1275 // 1) (mark & zero_bits) == 0, and
1276 // 2) rsp <= mark < mark + os::pagesize()
1277 //
1278 // Warning: rsp + os::pagesize can overflow the stack base. We must
1279 // neither apply the optimization for an inflated lock allocated
1280 // just above the thread stack (this is why condition 1 matters)
1281 // nor apply the optimization if the stack lock is inside the stack
1282 // of another thread. The latter is avoided even in case of overflow
1283 // because we have guard pages at the end of all stacks. Hence, if
1284 // we go over the stack base and hit the stack of another thread,
1285 // this should not be in a writeable area that could contain a
1286 // stack lock allocated by that thread. As a consequence, a stack
1287 // lock less than page size away from rsp is guaranteed to be
1288 // owned by the current thread.
1289 //
1290 // These 3 tests can be done by evaluating the following
1291 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1292 // assuming both stack pointer and pagesize have their
1293 // least significant bits clear.
1294 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1295 subptr(swap_reg, rsp);
1296 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1297
1298 // Save the test result, for recursive case, the result is zero
1299 movptr(Address(lock_reg, mark_offset), swap_reg);
1300 jcc(Assembler::notZero, slow_case);
1301
1302 bind(count_locking);
1303 inc_held_monitor_count();
1304 }
1305 jmp(done);
1306
1307 bind(slow_case);
1308
1309 // Call the runtime routine for slow case
1310 call_VM_preemptable(noreg,
1311 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1312 lock_reg);
1313 bind(done);
1314 }
1315 }
1316
1317
1318 // Unlocks an object. Used in monitorexit bytecode and
1319 // remove_activation. Throws an IllegalMonitorException if object is
1320 // not locked by current thread.
1321 //
1322 // Args:
1323 // rdx, c_rarg1: BasicObjectLock for lock
1324 //
1325 // Kills:
1326 // rax
1327 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1328 // rscratch1 (scratch reg)
1329 // rax, rbx, rcx, rdx
1330 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1331 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1332 "The argument is only for looks. It must be c_rarg1");
1333
1334 if (LockingMode == LM_MONITOR) {
1335 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1336 } else {
1337 Label count_locking, done, slow_case;
1338
1339 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1340 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1341 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1342
1343 save_bcp(); // Save in case of exception
1344
1345 if (LockingMode != LM_LIGHTWEIGHT) {
1346 // Convert from BasicObjectLock structure to object and BasicLock
1347 // structure Store the BasicLock address into %rax
1348 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1349 }
1350
1351 // Load oop into obj_reg(%c_rarg3)
1352 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1353
1354 // Free entry
1355 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1356
1357 if (LockingMode == LM_LIGHTWEIGHT) {
1358 #ifdef _LP64
1359 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1360 #else
1361 // Lacking registers and thread on x86_32. Always take slow path.
1362 jmp(slow_case);
1363 #endif
1364 } else if (LockingMode == LM_LEGACY) {
1365 // Load the old header from BasicLock structure
1366 movptr(header_reg, Address(swap_reg,
1367 BasicLock::displaced_header_offset_in_bytes()));
1368
1369 // Test for recursion
1370 testptr(header_reg, header_reg);
1371
1372 // zero for recursive case
1373 jcc(Assembler::zero, count_locking);
1374
1375 // Atomic swap back the old header
1376 lock();
1377 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1378
1379 // zero for simple unlock of a stack-lock case
1380 jcc(Assembler::notZero, slow_case);
1381
1382 bind(count_locking);
1383 dec_held_monitor_count();
1384 }
1385 jmp(done);
1386
1387 bind(slow_case);
1388 // Call the runtime routine for slow case.
1389 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1390 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1391
1392 bind(done);
1393
1394 restore_bcp();
1395 }
1396 }
1397
1398 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1399 Label& zero_continue) {
1400 assert(ProfileInterpreter, "must be profiling interpreter");
1401 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1402 testptr(mdp, mdp);
1403 jcc(Assembler::zero, zero_continue);
1404 }
1405
1406
1407 // Set the method data pointer for the current bcp.
1408 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1409 assert(ProfileInterpreter, "must be profiling interpreter");
1410 Label set_mdp;
1411 push(rax);
1412 push(rbx);
1413
1414 get_method(rbx);
1415 // Test MDO to avoid the call if it is null.
1416 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1417 testptr(rax, rax);
1418 jcc(Assembler::zero, set_mdp);
1419 // rbx: method
1420 // _bcp_register: bcp
1421 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1422 // rax: mdi
1423 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1424 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1425 addptr(rbx, in_bytes(MethodData::data_offset()));
1426 addptr(rax, rbx);
1427 bind(set_mdp);
1428 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1429 pop(rbx);
1430 pop(rax);
1431 }
1432
1433 void InterpreterMacroAssembler::verify_method_data_pointer() {
1434 assert(ProfileInterpreter, "must be profiling interpreter");
1435 #ifdef ASSERT
1436 Label verify_continue;
1437 push(rax);
1438 push(rbx);
1439 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1440 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1441 push(arg3_reg);
1442 push(arg2_reg);
1443 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1444 get_method(rbx);
1445
1446 // If the mdp is valid, it will point to a DataLayout header which is
1447 // consistent with the bcp. The converse is highly probable also.
1448 load_unsigned_short(arg2_reg,
1449 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1450 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1451 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1452 cmpptr(arg2_reg, _bcp_register);
1453 jcc(Assembler::equal, verify_continue);
1454 // rbx: method
1455 // _bcp_register: bcp
1456 // c_rarg3: mdp
1457 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1458 rbx, _bcp_register, arg3_reg);
1459 bind(verify_continue);
1460 pop(arg2_reg);
1461 pop(arg3_reg);
1462 pop(rbx);
1463 pop(rax);
1464 #endif // ASSERT
1465 }
1466
1467
1468 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1469 int constant,
1470 Register value) {
1471 assert(ProfileInterpreter, "must be profiling interpreter");
1472 Address data(mdp_in, constant);
1473 movptr(data, value);
1474 }
1475
1476
1477 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1478 int constant,
1479 bool decrement) {
1480 // Counter address
1481 Address data(mdp_in, constant);
1482
1483 increment_mdp_data_at(data, decrement);
1484 }
1485
1486 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1487 bool decrement) {
1488 assert(ProfileInterpreter, "must be profiling interpreter");
1489 // %%% this does 64bit counters at best it is wasting space
1490 // at worst it is a rare bug when counters overflow
1491
1492 if (decrement) {
1493 // Decrement the register. Set condition codes.
1494 addptr(data, -DataLayout::counter_increment);
1495 // If the decrement causes the counter to overflow, stay negative
1496 Label L;
1497 jcc(Assembler::negative, L);
1498 addptr(data, DataLayout::counter_increment);
1499 bind(L);
1500 } else {
1501 assert(DataLayout::counter_increment == 1,
1502 "flow-free idiom only works with 1");
1503 // Increment the register. Set carry flag.
1504 addptr(data, DataLayout::counter_increment);
1505 // If the increment causes the counter to overflow, pull back by 1.
1506 sbbptr(data, 0);
1507 }
1508 }
1509
1510
1511 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1512 Register reg,
1513 int constant,
1514 bool decrement) {
1515 Address data(mdp_in, reg, Address::times_1, constant);
1516
1517 increment_mdp_data_at(data, decrement);
1518 }
1519
1520 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1521 int flag_byte_constant) {
1522 assert(ProfileInterpreter, "must be profiling interpreter");
1523 int header_offset = in_bytes(DataLayout::flags_offset());
1524 int header_bits = flag_byte_constant;
1525 // Set the flag
1526 orb(Address(mdp_in, header_offset), header_bits);
1527 }
1528
1529
1530
1531 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1532 int offset,
1533 Register value,
1534 Register test_value_out,
1535 Label& not_equal_continue) {
1536 assert(ProfileInterpreter, "must be profiling interpreter");
1537 if (test_value_out == noreg) {
1538 cmpptr(value, Address(mdp_in, offset));
1539 } else {
1540 // Put the test value into a register, so caller can use it:
1541 movptr(test_value_out, Address(mdp_in, offset));
1542 cmpptr(test_value_out, value);
1543 }
1544 jcc(Assembler::notEqual, not_equal_continue);
1545 }
1546
1547
1548 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1549 int offset_of_disp) {
1550 assert(ProfileInterpreter, "must be profiling interpreter");
1551 Address disp_address(mdp_in, offset_of_disp);
1552 addptr(mdp_in, disp_address);
1553 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1554 }
1555
1556
1557 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1558 Register reg,
1559 int offset_of_disp) {
1560 assert(ProfileInterpreter, "must be profiling interpreter");
1561 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1562 addptr(mdp_in, disp_address);
1563 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1564 }
1565
1566
1567 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1568 int constant) {
1569 assert(ProfileInterpreter, "must be profiling interpreter");
1570 addptr(mdp_in, constant);
1571 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1572 }
1573
1574
1575 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1576 assert(ProfileInterpreter, "must be profiling interpreter");
1577 push(return_bci); // save/restore across call_VM
1578 call_VM(noreg,
1579 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1580 return_bci);
1581 pop(return_bci);
1582 }
1583
1584
1585 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1586 Register bumped_count) {
1587 if (ProfileInterpreter) {
1588 Label profile_continue;
1589
1590 // If no method data exists, go to profile_continue.
1591 // Otherwise, assign to mdp
1592 test_method_data_pointer(mdp, profile_continue);
1593
1594 // We are taking a branch. Increment the taken count.
1595 // We inline increment_mdp_data_at to return bumped_count in a register
1596 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1597 Address data(mdp, in_bytes(JumpData::taken_offset()));
1598 movptr(bumped_count, data);
1599 assert(DataLayout::counter_increment == 1,
1600 "flow-free idiom only works with 1");
1601 addptr(bumped_count, DataLayout::counter_increment);
1602 sbbptr(bumped_count, 0);
1603 movptr(data, bumped_count); // Store back out
1604
1605 // The method data pointer needs to be updated to reflect the new target.
1606 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1607 bind(profile_continue);
1608 }
1609 }
1610
1611
1612 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1613 if (ProfileInterpreter) {
1614 Label profile_continue;
1615
1616 // If no method data exists, go to profile_continue.
1617 test_method_data_pointer(mdp, profile_continue);
1618
1619 // We are taking a branch. Increment the not taken count.
1620 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1621
1622 // The method data pointer needs to be updated to correspond to
1623 // the next bytecode
1624 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1625 bind(profile_continue);
1626 }
1627 }
1628
1629 void InterpreterMacroAssembler::profile_call(Register mdp) {
1630 if (ProfileInterpreter) {
1631 Label profile_continue;
1632
1633 // If no method data exists, go to profile_continue.
1634 test_method_data_pointer(mdp, profile_continue);
1635
1636 // We are making a call. Increment the count.
1637 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1638
1639 // The method data pointer needs to be updated to reflect the new target.
1640 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1641 bind(profile_continue);
1642 }
1643 }
1644
1645
1646 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1647 if (ProfileInterpreter) {
1648 Label profile_continue;
1649
1650 // If no method data exists, go to profile_continue.
1651 test_method_data_pointer(mdp, profile_continue);
1652
1653 // We are making a call. Increment the count.
1654 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1655
1656 // The method data pointer needs to be updated to reflect the new target.
1657 update_mdp_by_constant(mdp,
1658 in_bytes(VirtualCallData::
1659 virtual_call_data_size()));
1660 bind(profile_continue);
1661 }
1662 }
1663
1664
1665 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1666 Register mdp,
1667 Register reg2,
1668 bool receiver_can_be_null) {
1669 if (ProfileInterpreter) {
1670 Label profile_continue;
1671
1672 // If no method data exists, go to profile_continue.
1673 test_method_data_pointer(mdp, profile_continue);
1674
1675 Label skip_receiver_profile;
1676 if (receiver_can_be_null) {
1677 Label not_null;
1678 testptr(receiver, receiver);
1679 jccb(Assembler::notZero, not_null);
1680 // We are making a call. Increment the count for null receiver.
1681 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1682 jmp(skip_receiver_profile);
1683 bind(not_null);
1684 }
1685
1686 // Record the receiver type.
1687 record_klass_in_profile(receiver, mdp, reg2, true);
1688 bind(skip_receiver_profile);
1689
1690 // The method data pointer needs to be updated to reflect the new target.
1691 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1692 bind(profile_continue);
1693 }
1694 }
1695
1696 // This routine creates a state machine for updating the multi-row
1697 // type profile at a virtual call site (or other type-sensitive bytecode).
1698 // The machine visits each row (of receiver/count) until the receiver type
1699 // is found, or until it runs out of rows. At the same time, it remembers
1700 // the location of the first empty row. (An empty row records null for its
1701 // receiver, and can be allocated for a newly-observed receiver type.)
1702 // Because there are two degrees of freedom in the state, a simple linear
1703 // search will not work; it must be a decision tree. Hence this helper
1704 // function is recursive, to generate the required tree structured code.
1705 // It's the interpreter, so we are trading off code space for speed.
1706 // See below for example code.
1707 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1708 Register receiver, Register mdp,
1709 Register reg2, int start_row,
1710 Label& done, bool is_virtual_call) {
1711 if (TypeProfileWidth == 0) {
1712 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1713 } else {
1714 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1715 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1716 }
1717 }
1718
1719 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1720 Label& done, int total_rows,
1721 OffsetFunction item_offset_fn,
1722 OffsetFunction item_count_offset_fn) {
1723 int last_row = total_rows - 1;
1724 assert(start_row <= last_row, "must be work left to do");
1725 // Test this row for both the item and for null.
1726 // Take any of three different outcomes:
1727 // 1. found item => increment count and goto done
1728 // 2. found null => keep looking for case 1, maybe allocate this cell
1729 // 3. found something else => keep looking for cases 1 and 2
1730 // Case 3 is handled by a recursive call.
1731 for (int row = start_row; row <= last_row; row++) {
1732 Label next_test;
1733 bool test_for_null_also = (row == start_row);
1734
1735 // See if the item is item[n].
1736 int item_offset = in_bytes(item_offset_fn(row));
1737 test_mdp_data_at(mdp, item_offset, item,
1738 (test_for_null_also ? reg2 : noreg),
1739 next_test);
1740 // (Reg2 now contains the item from the CallData.)
1741
1742 // The item is item[n]. Increment count[n].
1743 int count_offset = in_bytes(item_count_offset_fn(row));
1744 increment_mdp_data_at(mdp, count_offset);
1745 jmp(done);
1746 bind(next_test);
1747
1748 if (test_for_null_also) {
1749 // Failed the equality check on item[n]... Test for null.
1750 testptr(reg2, reg2);
1751 if (start_row == last_row) {
1752 // The only thing left to do is handle the null case.
1753 Label found_null;
1754 jccb(Assembler::zero, found_null);
1755 // Item did not match any saved item and there is no empty row for it.
1756 // Increment total counter to indicate polymorphic case.
1757 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1758 jmp(done);
1759 bind(found_null);
1760 break;
1761 }
1762 Label found_null;
1763 // Since null is rare, make it be the branch-taken case.
1764 jcc(Assembler::zero, found_null);
1765
1766 // Put all the "Case 3" tests here.
1767 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1768 item_offset_fn, item_count_offset_fn);
1769
1770 // Found a null. Keep searching for a matching item,
1771 // but remember that this is an empty (unused) slot.
1772 bind(found_null);
1773 }
1774 }
1775
1776 // In the fall-through case, we found no matching item, but we
1777 // observed the item[start_row] is null.
1778
1779 // Fill in the item field and increment the count.
1780 int item_offset = in_bytes(item_offset_fn(start_row));
1781 set_mdp_data_at(mdp, item_offset, item);
1782 int count_offset = in_bytes(item_count_offset_fn(start_row));
1783 movl(reg2, DataLayout::counter_increment);
1784 set_mdp_data_at(mdp, count_offset, reg2);
1785 if (start_row > 0) {
1786 jmp(done);
1787 }
1788 }
1789
1790 // Example state machine code for three profile rows:
1791 // // main copy of decision tree, rooted at row[1]
1792 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1793 // if (row[0].rec != nullptr) {
1794 // // inner copy of decision tree, rooted at row[1]
1795 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1796 // if (row[1].rec != nullptr) {
1797 // // degenerate decision tree, rooted at row[2]
1798 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1799 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1800 // row[2].init(rec); goto done;
1801 // } else {
1802 // // remember row[1] is empty
1803 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1804 // row[1].init(rec); goto done;
1805 // }
1806 // } else {
1807 // // remember row[0] is empty
1808 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1809 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1810 // row[0].init(rec); goto done;
1811 // }
1812 // done:
1813
1814 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1815 Register mdp, Register reg2,
1816 bool is_virtual_call) {
1817 assert(ProfileInterpreter, "must be profiling");
1818 Label done;
1819
1820 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1821
1822 bind (done);
1823 }
1824
1825 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1826 Register mdp) {
1827 if (ProfileInterpreter) {
1828 Label profile_continue;
1829 uint row;
1830
1831 // If no method data exists, go to profile_continue.
1832 test_method_data_pointer(mdp, profile_continue);
1833
1834 // Update the total ret count.
1835 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1836
1837 for (row = 0; row < RetData::row_limit(); row++) {
1838 Label next_test;
1839
1840 // See if return_bci is equal to bci[n]:
1841 test_mdp_data_at(mdp,
1842 in_bytes(RetData::bci_offset(row)),
1843 return_bci, noreg,
1844 next_test);
1845
1846 // return_bci is equal to bci[n]. Increment the count.
1847 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1848
1849 // The method data pointer needs to be updated to reflect the new target.
1850 update_mdp_by_offset(mdp,
1851 in_bytes(RetData::bci_displacement_offset(row)));
1852 jmp(profile_continue);
1853 bind(next_test);
1854 }
1855
1856 update_mdp_for_ret(return_bci);
1857
1858 bind(profile_continue);
1859 }
1860 }
1861
1862
1863 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1864 if (ProfileInterpreter) {
1865 Label profile_continue;
1866
1867 // If no method data exists, go to profile_continue.
1868 test_method_data_pointer(mdp, profile_continue);
1869
1870 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1871
1872 // The method data pointer needs to be updated.
1873 int mdp_delta = in_bytes(BitData::bit_data_size());
1874 if (TypeProfileCasts) {
1875 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1876 }
1877 update_mdp_by_constant(mdp, mdp_delta);
1878
1879 bind(profile_continue);
1880 }
1881 }
1882
1883
1884 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1885 if (ProfileInterpreter) {
1886 Label profile_continue;
1887
1888 // If no method data exists, go to profile_continue.
1889 test_method_data_pointer(mdp, profile_continue);
1890
1891 // The method data pointer needs to be updated.
1892 int mdp_delta = in_bytes(BitData::bit_data_size());
1893 if (TypeProfileCasts) {
1894 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1895
1896 // Record the object type.
1897 record_klass_in_profile(klass, mdp, reg2, false);
1898 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1899 NOT_LP64(restore_locals();) // Restore EDI
1900 }
1901 update_mdp_by_constant(mdp, mdp_delta);
1902
1903 bind(profile_continue);
1904 }
1905 }
1906
1907
1908 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1909 if (ProfileInterpreter) {
1910 Label profile_continue;
1911
1912 // If no method data exists, go to profile_continue.
1913 test_method_data_pointer(mdp, profile_continue);
1914
1915 // Update the default case count
1916 increment_mdp_data_at(mdp,
1917 in_bytes(MultiBranchData::default_count_offset()));
1918
1919 // The method data pointer needs to be updated.
1920 update_mdp_by_offset(mdp,
1921 in_bytes(MultiBranchData::
1922 default_displacement_offset()));
1923
1924 bind(profile_continue);
1925 }
1926 }
1927
1928
1929 void InterpreterMacroAssembler::profile_switch_case(Register index,
1930 Register mdp,
1931 Register reg2) {
1932 if (ProfileInterpreter) {
1933 Label profile_continue;
1934
1935 // If no method data exists, go to profile_continue.
1936 test_method_data_pointer(mdp, profile_continue);
1937
1938 // Build the base (index * per_case_size_in_bytes()) +
1939 // case_array_offset_in_bytes()
1940 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1941 imulptr(index, reg2); // XXX l ?
1942 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1943
1944 // Update the case count
1945 increment_mdp_data_at(mdp,
1946 index,
1947 in_bytes(MultiBranchData::relative_count_offset()));
1948
1949 // The method data pointer needs to be updated.
1950 update_mdp_by_offset(mdp,
1951 index,
1952 in_bytes(MultiBranchData::
1953 relative_displacement_offset()));
1954
1955 bind(profile_continue);
1956 }
1957 }
1958
1959
1960
1961 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1962 if (state == atos) {
1963 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1964 }
1965 }
1966
1967 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1968 #ifndef _LP64
1969 if ((state == ftos && UseSSE < 1) ||
1970 (state == dtos && UseSSE < 2)) {
1971 MacroAssembler::verify_FPU(stack_depth);
1972 }
1973 #endif
1974 }
1975
1976 // Jump if ((*counter_addr += increment) & mask) == 0
1977 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1978 Register scratch, Label* where) {
1979 // This update is actually not atomic and can lose a number of updates
1980 // under heavy contention, but the alternative of using the (contended)
1981 // atomic update here penalizes profiling paths too much.
1982 movl(scratch, counter_addr);
1983 incrementl(scratch, InvocationCounter::count_increment);
1984 movl(counter_addr, scratch);
1985 andl(scratch, mask);
1986 if (where != nullptr) {
1987 jcc(Assembler::zero, *where);
1988 }
1989 }
1990
1991 void InterpreterMacroAssembler::notify_method_entry() {
1992 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1993 // track stack depth. If it is possible to enter interp_only_mode we add
1994 // the code to check if the event should be sent.
1995 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1996 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1997 if (JvmtiExport::can_post_interpreter_events()) {
1998 Label L;
1999 NOT_LP64(get_thread(rthread);)
2000 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2001 testl(rdx, rdx);
2002 jcc(Assembler::zero, L);
2003 call_VM(noreg, CAST_FROM_FN_PTR(address,
2004 InterpreterRuntime::post_method_entry));
2005 bind(L);
2006 }
2007
2008 if (DTraceMethodProbes) {
2009 NOT_LP64(get_thread(rthread);)
2010 get_method(rarg);
2011 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2012 rthread, rarg);
2013 }
2014
2015 // RedefineClasses() tracing support for obsolete method entry
2016 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2017 NOT_LP64(get_thread(rthread);)
2018 get_method(rarg);
2019 call_VM_leaf(
2020 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2021 rthread, rarg);
2022 }
2023 }
2024
2025
2026 void InterpreterMacroAssembler::notify_method_exit(
2027 TosState state, NotifyMethodExitMode mode) {
2028 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2029 // track stack depth. If it is possible to enter interp_only_mode we add
2030 // the code to check if the event should be sent.
2031 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2032 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2033 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2034 Label L;
2035 // Note: frame::interpreter_frame_result has a dependency on how the
2036 // method result is saved across the call to post_method_exit. If this
2037 // is changed then the interpreter_frame_result implementation will
2038 // need to be updated too.
2039
2040 // template interpreter will leave the result on the top of the stack.
2041 push(state);
2042 NOT_LP64(get_thread(rthread);)
2043 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2044 testl(rdx, rdx);
2045 jcc(Assembler::zero, L);
2046 call_VM(noreg,
2047 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2048 bind(L);
2049 pop(state);
2050 }
2051
2052 if (DTraceMethodProbes) {
2053 push(state);
2054 NOT_LP64(get_thread(rthread);)
2055 get_method(rarg);
2056 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2057 rthread, rarg);
2058 pop(state);
2059 }
2060 }
2061
2062 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2063 // Get index out of bytecode pointer
2064 get_cache_index_at_bcp(index, 1, sizeof(u4));
2065 // Get address of invokedynamic array
2066 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2067 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2068 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2069 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2070 } else {
2071 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2072 }
2073 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2074 }
2075
2076 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2077 // Get index out of bytecode pointer
2078 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2079 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2080
2081 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2082 // Take shortcut if the size is a power of 2
2083 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2084 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2085 } else {
2086 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2087 }
2088 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2089 }
2090
2091 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2092 // Get index out of bytecode pointer
2093 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2094 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2095
2096 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2097 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2098 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2099 }