1 /*
2 * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "compiler/compiler_globals.hpp"
27 #include "interp_masm_x86.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "logging/log.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markWord.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/resolvedFieldEntry.hpp"
36 #include "oops/resolvedIndyEntry.hpp"
37 #include "oops/resolvedMethodEntry.hpp"
38 #include "prims/jvmtiExport.hpp"
39 #include "prims/jvmtiThreadState.hpp"
40 #include "runtime/basicLock.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/javaThread.hpp"
43 #include "runtime/runtimeUpcalls.hpp"
44 #include "runtime/safepointMechanism.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "utilities/powerOfTwo.hpp"
47
48 // Implementation of InterpreterMacroAssembler
49
50 void InterpreterMacroAssembler::jump_to_entry(address entry) {
51 assert(entry, "Entry must have been generated by now");
52 jump(RuntimeAddress(entry));
53 }
54
55 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
56 Label update, next, none;
57
58 #ifdef _LP64
59 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
60 #else
61 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
62 #endif
63
64 interp_verify_oop(obj, atos);
65
66 testptr(obj, obj);
67 jccb(Assembler::notZero, update);
68 testptr(mdo_addr, TypeEntries::null_seen);
69 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
70 // atomic update to prevent overwriting Klass* with 0
71 lock();
72 orptr(mdo_addr, TypeEntries::null_seen);
73 jmpb(next);
74
75 bind(update);
76 load_klass(obj, obj, rscratch1);
77 #ifdef _LP64
78 mov(rscratch1, obj);
79 #endif
80
81 xorptr(obj, mdo_addr);
82 testptr(obj, TypeEntries::type_klass_mask);
83 jccb(Assembler::zero, next); // klass seen before, nothing to
84 // do. The unknown bit may have been
85 // set already but no need to check.
86
87 testptr(obj, TypeEntries::type_unknown);
88 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
89
90 cmpptr(mdo_addr, 0);
91 jccb(Assembler::equal, none);
92 cmpptr(mdo_addr, TypeEntries::null_seen);
93 jccb(Assembler::equal, none);
94 #ifdef _LP64
95 // There is a chance that the checks above (re-reading profiling
96 // data from memory) fail if another thread has just set the
97 // profiling to this obj's klass
98 mov(obj, rscratch1);
99 xorptr(obj, mdo_addr);
100 testptr(obj, TypeEntries::type_klass_mask);
101 jccb(Assembler::zero, next);
102 #endif
103
104 // different than before. Cannot keep accurate profile.
105 orptr(mdo_addr, TypeEntries::type_unknown);
106 jmpb(next);
107
108 bind(none);
109 // first time here. Set profile type.
110 movptr(mdo_addr, obj);
111 #ifdef ASSERT
112 andptr(obj, TypeEntries::type_klass_mask);
113 verify_klass_ptr(obj);
114 #endif
115
116 bind(next);
117 }
118
119 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
120 if (!ProfileInterpreter) {
121 return;
122 }
123
124 if (MethodData::profile_arguments() || MethodData::profile_return()) {
125 Label profile_continue;
126
127 test_method_data_pointer(mdp, profile_continue);
128
129 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
130
131 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
132 jcc(Assembler::notEqual, profile_continue);
133
134 if (MethodData::profile_arguments()) {
135 Label done;
136 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
137 addptr(mdp, off_to_args);
138
139 for (int i = 0; i < TypeProfileArgsLimit; i++) {
140 if (i > 0 || MethodData::profile_return()) {
141 // If return value type is profiled we may have no argument to profile
142 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
143 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
144 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
145 jcc(Assembler::less, done);
146 }
147 movptr(tmp, Address(callee, Method::const_offset()));
148 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
149 // stack offset o (zero based) from the start of the argument
150 // list, for n arguments translates into offset n - o - 1 from
151 // the end of the argument list
152 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
153 subl(tmp, 1);
154 Address arg_addr = argument_address(tmp);
155 movptr(tmp, arg_addr);
156
157 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
158 profile_obj_type(tmp, mdo_arg_addr);
159
160 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
161 addptr(mdp, to_add);
162 off_to_args += to_add;
163 }
164
165 if (MethodData::profile_return()) {
166 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
167 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
168 }
169
170 bind(done);
171
172 if (MethodData::profile_return()) {
173 // We're right after the type profile for the last
174 // argument. tmp is the number of cells left in the
175 // CallTypeData/VirtualCallTypeData to reach its end. Non null
176 // if there's a return to profile.
177 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
178 shll(tmp, log2i_exact((int)DataLayout::cell_size));
179 addptr(mdp, tmp);
180 }
181 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
182 } else {
183 assert(MethodData::profile_return(), "either profile call args or call ret");
184 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
185 }
186
187 // mdp points right after the end of the
188 // CallTypeData/VirtualCallTypeData, right after the cells for the
189 // return value type if there's one
190
191 bind(profile_continue);
192 }
193 }
194
195 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
196 assert_different_registers(mdp, ret, tmp, _bcp_register);
197 if (ProfileInterpreter && MethodData::profile_return()) {
198 Label profile_continue;
199
200 test_method_data_pointer(mdp, profile_continue);
201
202 if (MethodData::profile_return_jsr292_only()) {
203 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
204
205 // If we don't profile all invoke bytecodes we must make sure
206 // it's a bytecode we indeed profile. We can't go back to the
207 // beginning of the ProfileData we intend to update to check its
208 // type because we're right after it and we don't known its
209 // length
210 Label do_profile;
211 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
212 jcc(Assembler::equal, do_profile);
213 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
214 jcc(Assembler::equal, do_profile);
215 get_method(tmp);
216 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
217 jcc(Assembler::notEqual, profile_continue);
218
219 bind(do_profile);
220 }
221
222 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
223 mov(tmp, ret);
224 profile_obj_type(tmp, mdo_ret_addr);
225
226 bind(profile_continue);
227 }
228 }
229
230 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
231 if (ProfileInterpreter && MethodData::profile_parameters()) {
232 Label profile_continue;
233
234 test_method_data_pointer(mdp, profile_continue);
235
236 // Load the offset of the area within the MDO used for
237 // parameters. If it's negative we're not profiling any parameters
238 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
239 testl(tmp1, tmp1);
240 jcc(Assembler::negative, profile_continue);
241
242 // Compute a pointer to the area for parameters from the offset
243 // and move the pointer to the slot for the last
244 // parameters. Collect profiling from last parameter down.
245 // mdo start + parameters offset + array length - 1
246 addptr(mdp, tmp1);
247 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
248 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
249
250 Label loop;
251 bind(loop);
252
253 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
254 int type_base = in_bytes(ParametersTypeData::type_offset(0));
255 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
256 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
257 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
258
259 // load offset on the stack from the slot for this parameter
260 movptr(tmp2, arg_off);
261 negptr(tmp2);
262 // read the parameter from the local area
263 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
264
265 // profile the parameter
266 profile_obj_type(tmp2, arg_type);
267
268 // go to next parameter
269 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
270 jcc(Assembler::positive, loop);
271
272 bind(profile_continue);
273 }
274 }
275
276 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
277 int number_of_arguments) {
278 // interpreter specific
279 //
280 // Note: No need to save/restore bcp & locals registers
281 // since these are callee saved registers and no blocking/
282 // GC can happen in leaf calls.
283 // Further Note: DO NOT save/restore bcp/locals. If a caller has
284 // already saved them so that it can use rsi/rdi as temporaries
285 // then a save/restore here will DESTROY the copy the caller
286 // saved! There used to be a save_bcp() that only happened in
287 // the ASSERT path (no restore_bcp). Which caused bizarre failures
288 // when jvm built with ASSERTs.
289 #ifdef ASSERT
290 {
291 Label L;
292 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
293 jcc(Assembler::equal, L);
294 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
295 " last_sp != null");
296 bind(L);
297 }
298 #endif
299 // super call
300 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
301 // interpreter specific
302 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
303 // but since they may not have been saved (and we don't want to
304 // save them here (see note above) the assert is invalid.
305 }
306
307 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
308 Register java_thread,
309 Register last_java_sp,
310 address entry_point,
311 int number_of_arguments,
312 bool check_exceptions) {
313 // interpreter specific
314 //
315 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
316 // really make a difference for these runtime calls, since they are
317 // slow anyway. Btw., bcp must be saved/restored since it may change
318 // due to GC.
319 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
320 save_bcp();
321 #ifdef ASSERT
322 {
323 Label L;
324 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
325 jcc(Assembler::equal, L);
326 stop("InterpreterMacroAssembler::call_VM_base:"
327 " last_sp isn't null");
328 bind(L);
329 }
330 #endif /* ASSERT */
331 // super call
332 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
333 entry_point, number_of_arguments,
334 check_exceptions);
335 // interpreter specific
336 restore_bcp();
337 restore_locals();
338 }
339
340 #ifdef _LP64
341 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
342 address entry_point,
343 Register arg_1) {
344 assert(arg_1 == c_rarg1, "");
345 Label resume_pc, not_preempted;
346
347 #ifdef ASSERT
348 {
349 Label L;
350 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
351 jcc(Assembler::equal, L);
352 stop("Should not have alternate return address set");
353 bind(L);
354 }
355 #endif /* ASSERT */
356
357 // Force freeze slow path.
358 push_cont_fastpath();
359
360 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
361 // Note: call_VM_helper requires last_Java_pc for anchor to be at the top of the stack.
362 lea(rscratch1, resume_pc);
363 push(rscratch1);
364 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/);
365 pop(rscratch1);
366
367 pop_cont_fastpath();
368
369 // Check if preempted.
370 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset()));
371 cmpptr(rscratch1, NULL_WORD);
372 jccb(Assembler::zero, not_preempted);
373 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD);
374 jmp(rscratch1);
375
376 // In case of preemption, this is where we will resume once we finally acquire the monitor.
377 bind(resume_pc);
378 restore_after_resume(false /* is_native */);
379
380 bind(not_preempted);
381 }
382
383 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
384 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
385 call(rscratch1);
386 if (is_native) {
387 // On resume we need to set up stack as expected.
388 push(dtos);
389 push(ltos);
390 }
391 }
392 #else
393 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
394 address entry_point,
395 Register arg_1) {
396 MacroAssembler::call_VM(oop_result, entry_point, arg_1);
397 }
398 #endif // _LP64
399
400 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
401 if (JvmtiExport::can_pop_frame()) {
402 Label L;
403 // Initiate popframe handling only if it is not already being
404 // processed. If the flag has the popframe_processing bit set, it
405 // means that this code is called *during* popframe handling - we
406 // don't want to reenter.
407 // This method is only called just after the call into the vm in
408 // call_VM_base, so the arg registers are available.
409 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
410 LP64_ONLY(c_rarg0);
411 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
412 testl(pop_cond, JavaThread::popframe_pending_bit);
413 jcc(Assembler::zero, L);
414 testl(pop_cond, JavaThread::popframe_processing_bit);
415 jcc(Assembler::notZero, L);
416 // Call Interpreter::remove_activation_preserving_args_entry() to get the
417 // address of the same-named entrypoint in the generated interpreter code.
418 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
419 jmp(rax);
420 bind(L);
421 NOT_LP64(get_thread(java_thread);)
422 }
423 }
424
425 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
426 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
427 NOT_LP64(get_thread(thread);)
428 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
429 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
430 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
431 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
432 #ifdef _LP64
433 switch (state) {
434 case atos: movptr(rax, oop_addr);
435 movptr(oop_addr, NULL_WORD);
436 interp_verify_oop(rax, state); break;
437 case ltos: movptr(rax, val_addr); break;
438 case btos: // fall through
439 case ztos: // fall through
440 case ctos: // fall through
441 case stos: // fall through
442 case itos: movl(rax, val_addr); break;
443 case ftos: load_float(val_addr); break;
444 case dtos: load_double(val_addr); break;
445 case vtos: /* nothing to do */ break;
446 default : ShouldNotReachHere();
447 }
448 // Clean up tos value in the thread object
449 movl(tos_addr, ilgl);
450 movl(val_addr, NULL_WORD);
451 #else
452 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
453 + in_ByteSize(wordSize));
454 switch (state) {
455 case atos: movptr(rax, oop_addr);
456 movptr(oop_addr, NULL_WORD);
457 interp_verify_oop(rax, state); break;
458 case ltos:
459 movl(rdx, val_addr1); // fall through
460 case btos: // fall through
461 case ztos: // fall through
462 case ctos: // fall through
463 case stos: // fall through
464 case itos: movl(rax, val_addr); break;
465 case ftos: load_float(val_addr); break;
466 case dtos: load_double(val_addr); break;
467 case vtos: /* nothing to do */ break;
468 default : ShouldNotReachHere();
469 }
470 #endif // _LP64
471 // Clean up tos value in the thread object
472 movl(tos_addr, ilgl);
473 movptr(val_addr, NULL_WORD);
474 NOT_LP64(movptr(val_addr1, NULL_WORD);)
475 }
476
477
478 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
479 if (JvmtiExport::can_force_early_return()) {
480 Label L;
481 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
482 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
483
484 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
485 testptr(tmp, tmp);
486 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
487
488 // Initiate earlyret handling only if it is not already being processed.
489 // If the flag has the earlyret_processing bit set, it means that this code
490 // is called *during* earlyret handling - we don't want to reenter.
491 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
492 cmpl(tmp, JvmtiThreadState::earlyret_pending);
493 jcc(Assembler::notEqual, L);
494
495 // Call Interpreter::remove_activation_early_entry() to get the address of the
496 // same-named entrypoint in the generated interpreter code.
497 NOT_LP64(get_thread(java_thread);)
498 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
499 #ifdef _LP64
500 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
501 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
502 #else
503 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
504 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
505 #endif // _LP64
506 jmp(rax);
507 bind(L);
508 NOT_LP64(get_thread(java_thread);)
509 }
510 }
511
512 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
513 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
514 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
515 bswapl(reg);
516 shrl(reg, 16);
517 }
518
519 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
520 int bcp_offset,
521 size_t index_size) {
522 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
523 if (index_size == sizeof(u2)) {
524 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
525 } else if (index_size == sizeof(u4)) {
526 movl(index, Address(_bcp_register, bcp_offset));
527 } else if (index_size == sizeof(u1)) {
528 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
529 } else {
530 ShouldNotReachHere();
531 }
532 }
533
534 // Load object from cpool->resolved_references(index)
535 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
536 Register index,
537 Register tmp) {
538 assert_different_registers(result, index);
539
540 get_constant_pool(result);
541 // load pointer for resolved_references[] objArray
542 movptr(result, Address(result, ConstantPool::cache_offset()));
543 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
544 resolve_oop_handle(result, tmp);
545 load_heap_oop(result, Address(result, index,
546 UseCompressedOops ? Address::times_4 : Address::times_ptr,
547 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
548 }
549
550 // load cpool->resolved_klass_at(index)
551 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
552 Register cpool,
553 Register index) {
554 assert_different_registers(cpool, index);
555
556 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
557 Register resolved_klasses = cpool;
558 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
559 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
560 }
561
562 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
563 // subtype of super_klass.
564 //
565 // Args:
566 // rax: superklass
567 // Rsub_klass: subklass
568 //
569 // Kills:
570 // rcx, rdi
571 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
572 Label& ok_is_subtype) {
573 assert(Rsub_klass != rax, "rax holds superklass");
574 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
575 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
576 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
577 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
578
579 // Profile the not-null value's klass.
580 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
581
582 // Do the check.
583 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
584 }
585
586
587 #ifndef _LP64
588 void InterpreterMacroAssembler::f2ieee() {
589 if (IEEEPrecision) {
590 fstp_s(Address(rsp, 0));
591 fld_s(Address(rsp, 0));
592 }
593 }
594
595
596 void InterpreterMacroAssembler::d2ieee() {
597 if (IEEEPrecision) {
598 fstp_d(Address(rsp, 0));
599 fld_d(Address(rsp, 0));
600 }
601 }
602 #endif // _LP64
603
604 // Java Expression Stack
605
606 void InterpreterMacroAssembler::pop_ptr(Register r) {
607 pop(r);
608 }
609
610 void InterpreterMacroAssembler::push_ptr(Register r) {
611 push(r);
612 }
613
614 void InterpreterMacroAssembler::push_i(Register r) {
615 push(r);
616 }
617
618 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
619 push(r);
620 }
621
622 void InterpreterMacroAssembler::push_f(XMMRegister r) {
623 subptr(rsp, wordSize);
624 movflt(Address(rsp, 0), r);
625 }
626
627 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
628 movflt(r, Address(rsp, 0));
629 addptr(rsp, wordSize);
630 }
631
632 void InterpreterMacroAssembler::push_d(XMMRegister r) {
633 subptr(rsp, 2 * wordSize);
634 movdbl(Address(rsp, 0), r);
635 }
636
637 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
638 movdbl(r, Address(rsp, 0));
639 addptr(rsp, 2 * Interpreter::stackElementSize);
640 }
641
642 #ifdef _LP64
643 void InterpreterMacroAssembler::pop_i(Register r) {
644 // XXX can't use pop currently, upper half non clean
645 movl(r, Address(rsp, 0));
646 addptr(rsp, wordSize);
647 }
648
649 void InterpreterMacroAssembler::pop_l(Register r) {
650 movq(r, Address(rsp, 0));
651 addptr(rsp, 2 * Interpreter::stackElementSize);
652 }
653
654 void InterpreterMacroAssembler::push_l(Register r) {
655 subptr(rsp, 2 * wordSize);
656 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
657 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
658 }
659
660 void InterpreterMacroAssembler::pop(TosState state) {
661 switch (state) {
662 case atos: pop_ptr(); break;
663 case btos:
664 case ztos:
665 case ctos:
666 case stos:
667 case itos: pop_i(); break;
668 case ltos: pop_l(); break;
669 case ftos: pop_f(xmm0); break;
670 case dtos: pop_d(xmm0); break;
671 case vtos: /* nothing to do */ break;
672 default: ShouldNotReachHere();
673 }
674 interp_verify_oop(rax, state);
675 }
676
677 void InterpreterMacroAssembler::push(TosState state) {
678 interp_verify_oop(rax, state);
679 switch (state) {
680 case atos: push_ptr(); break;
681 case btos:
682 case ztos:
683 case ctos:
684 case stos:
685 case itos: push_i(); break;
686 case ltos: push_l(); break;
687 case ftos: push_f(xmm0); break;
688 case dtos: push_d(xmm0); break;
689 case vtos: /* nothing to do */ break;
690 default : ShouldNotReachHere();
691 }
692 }
693 #else
694 void InterpreterMacroAssembler::pop_i(Register r) {
695 pop(r);
696 }
697
698 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
699 pop(lo);
700 pop(hi);
701 }
702
703 void InterpreterMacroAssembler::pop_f() {
704 fld_s(Address(rsp, 0));
705 addptr(rsp, 1 * wordSize);
706 }
707
708 void InterpreterMacroAssembler::pop_d() {
709 fld_d(Address(rsp, 0));
710 addptr(rsp, 2 * wordSize);
711 }
712
713
714 void InterpreterMacroAssembler::pop(TosState state) {
715 switch (state) {
716 case atos: pop_ptr(rax); break;
717 case btos: // fall through
718 case ztos: // fall through
719 case ctos: // fall through
720 case stos: // fall through
721 case itos: pop_i(rax); break;
722 case ltos: pop_l(rax, rdx); break;
723 case ftos:
724 if (UseSSE >= 1) {
725 pop_f(xmm0);
726 } else {
727 pop_f();
728 }
729 break;
730 case dtos:
731 if (UseSSE >= 2) {
732 pop_d(xmm0);
733 } else {
734 pop_d();
735 }
736 break;
737 case vtos: /* nothing to do */ break;
738 default : ShouldNotReachHere();
739 }
740 interp_verify_oop(rax, state);
741 }
742
743
744 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
745 push(hi);
746 push(lo);
747 }
748
749 void InterpreterMacroAssembler::push_f() {
750 // Do not schedule for no AGI! Never write beyond rsp!
751 subptr(rsp, 1 * wordSize);
752 fstp_s(Address(rsp, 0));
753 }
754
755 void InterpreterMacroAssembler::push_d() {
756 // Do not schedule for no AGI! Never write beyond rsp!
757 subptr(rsp, 2 * wordSize);
758 fstp_d(Address(rsp, 0));
759 }
760
761
762 void InterpreterMacroAssembler::push(TosState state) {
763 interp_verify_oop(rax, state);
764 switch (state) {
765 case atos: push_ptr(rax); break;
766 case btos: // fall through
767 case ztos: // fall through
768 case ctos: // fall through
769 case stos: // fall through
770 case itos: push_i(rax); break;
771 case ltos: push_l(rax, rdx); break;
772 case ftos:
773 if (UseSSE >= 1) {
774 push_f(xmm0);
775 } else {
776 push_f();
777 }
778 break;
779 case dtos:
780 if (UseSSE >= 2) {
781 push_d(xmm0);
782 } else {
783 push_d();
784 }
785 break;
786 case vtos: /* nothing to do */ break;
787 default : ShouldNotReachHere();
788 }
789 }
790 #endif // _LP64
791
792
793 // Helpers for swap and dup
794 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
795 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
796 }
797
798 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
799 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
800 }
801
802
803 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
804 // set sender sp
805 lea(_bcp_register, Address(rsp, wordSize));
806 // record last_sp
807 mov(rcx, _bcp_register);
808 subptr(rcx, rbp);
809 sarptr(rcx, LogBytesPerWord);
810 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
811 }
812
813
814 // Jump to from_interpreted entry of a call unless single stepping is possible
815 // in this thread in which case we must call the i2i entry
816 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
817 prepare_to_jump_from_interpreted();
818
819 if (JvmtiExport::can_post_interpreter_events()) {
820 Label run_compiled_code;
821 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
822 // compiled code in threads for which the event is enabled. Check here for
823 // interp_only_mode if these events CAN be enabled.
824 // interp_only is an int, on little endian it is sufficient to test the byte only
825 // Is a cmpl faster?
826 LP64_ONLY(temp = r15_thread;)
827 NOT_LP64(get_thread(temp);)
828 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
829 jccb(Assembler::zero, run_compiled_code);
830 jmp(Address(method, Method::interpreter_entry_offset()));
831 bind(run_compiled_code);
832 }
833
834 jmp(Address(method, Method::from_interpreted_offset()));
835 }
836
837 // The following two routines provide a hook so that an implementation
838 // can schedule the dispatch in two parts. x86 does not do this.
839 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
840 // Nothing x86 specific to be done here
841 }
842
843 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
844 dispatch_next(state, step);
845 }
846
847 void InterpreterMacroAssembler::dispatch_base(TosState state,
848 address* table,
849 bool verifyoop,
850 bool generate_poll) {
851 verify_FPU(1, state);
852 if (VerifyActivationFrameSize) {
853 Label L;
854 mov(rcx, rbp);
855 subptr(rcx, rsp);
856 int32_t min_frame_size =
857 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
858 wordSize;
859 cmpptr(rcx, min_frame_size);
860 jcc(Assembler::greaterEqual, L);
861 stop("broken stack frame");
862 bind(L);
863 }
864 if (verifyoop) {
865 interp_verify_oop(rax, state);
866 }
867
868 address* const safepoint_table = Interpreter::safept_table(state);
869 #ifdef _LP64
870 Label no_safepoint, dispatch;
871 if (table != safepoint_table && generate_poll) {
872 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
873 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
874
875 jccb(Assembler::zero, no_safepoint);
876 lea(rscratch1, ExternalAddress((address)safepoint_table));
877 jmpb(dispatch);
878 }
879
880 bind(no_safepoint);
881 lea(rscratch1, ExternalAddress((address)table));
882 bind(dispatch);
883 jmp(Address(rscratch1, rbx, Address::times_8));
884
885 #else
886 Address index(noreg, rbx, Address::times_ptr);
887 if (table != safepoint_table && generate_poll) {
888 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
889 Label no_safepoint;
890 const Register thread = rcx;
891 get_thread(thread);
892 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
893
894 jccb(Assembler::zero, no_safepoint);
895 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
896 jump(dispatch_addr, noreg);
897 bind(no_safepoint);
898 }
899
900 {
901 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
902 jump(dispatch_addr, noreg);
903 }
904 #endif // _LP64
905 }
906
907 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
908 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
909 }
910
911 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
912 dispatch_base(state, Interpreter::normal_table(state));
913 }
914
915 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
916 dispatch_base(state, Interpreter::normal_table(state), false);
917 }
918
919
920 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
921 // load next bytecode (load before advancing _bcp_register to prevent AGI)
922 load_unsigned_byte(rbx, Address(_bcp_register, step));
923 // advance _bcp_register
924 increment(_bcp_register, step);
925 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
926 }
927
928 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
929 // load current bytecode
930 load_unsigned_byte(rbx, Address(_bcp_register, 0));
931 dispatch_base(state, table);
932 }
933
934 void InterpreterMacroAssembler::narrow(Register result) {
935
936 // Get method->_constMethod->_result_type
937 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
938 movptr(rcx, Address(rcx, Method::const_offset()));
939 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
940
941 Label done, notBool, notByte, notChar;
942
943 // common case first
944 cmpl(rcx, T_INT);
945 jcc(Assembler::equal, done);
946
947 // mask integer result to narrower return type.
948 cmpl(rcx, T_BOOLEAN);
949 jcc(Assembler::notEqual, notBool);
950 andl(result, 0x1);
951 jmp(done);
952
953 bind(notBool);
954 cmpl(rcx, T_BYTE);
955 jcc(Assembler::notEqual, notByte);
956 LP64_ONLY(movsbl(result, result);)
957 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
958 NOT_LP64(sarl(result, 24);) // and sign-extend byte
959 jmp(done);
960
961 bind(notByte);
962 cmpl(rcx, T_CHAR);
963 jcc(Assembler::notEqual, notChar);
964 LP64_ONLY(movzwl(result, result);)
965 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
966 jmp(done);
967
968 bind(notChar);
969 // cmpl(rcx, T_SHORT); // all that's left
970 // jcc(Assembler::notEqual, done);
971 LP64_ONLY(movswl(result, result);)
972 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
973 NOT_LP64(sarl(result, 16);) // and sign-extend short
974
975 // Nothing to do for T_INT
976 bind(done);
977 }
978
979 // remove activation
980 //
981 // Apply stack watermark barrier.
982 // Unlock the receiver if this is a synchronized method.
983 // Unlock any Java monitors from synchronized blocks.
984 // Remove the activation from the stack.
985 //
986 // If there are locked Java monitors
987 // If throw_monitor_exception
988 // throws IllegalMonitorStateException
989 // Else if install_monitor_exception
990 // installs IllegalMonitorStateException
991 // Else
992 // no error processing
993 void InterpreterMacroAssembler::remove_activation(
994 TosState state,
995 Register ret_addr,
996 bool throw_monitor_exception,
997 bool install_monitor_exception,
998 bool notify_jvmdi) {
999 // Note: Registers rdx xmm0 may be in use for the
1000 // result check if synchronized method
1001 Label unlocked, unlock, no_unlock;
1002
1003 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1004 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
1005 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
1006 // monitor pointers need different register
1007 // because rdx may have the result in it
1008 NOT_LP64(get_thread(rthread);)
1009
1010 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
1011 // that would normally not be safe to use. Such bad returns into unsafe territory of
1012 // the stack, will call InterpreterRuntime::at_unwind.
1013 Label slow_path;
1014 Label fast_path;
1015 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
1016 jmp(fast_path);
1017 bind(slow_path);
1018 push(state);
1019 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
1020 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1021 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
1022 reset_last_Java_frame(rthread, true);
1023 pop(state);
1024 bind(fast_path);
1025
1026 // get the value of _do_not_unlock_if_synchronized into rdx
1027 const Address do_not_unlock_if_synchronized(rthread,
1028 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1029 movbool(rbx, do_not_unlock_if_synchronized);
1030 movbool(do_not_unlock_if_synchronized, false); // reset the flag
1031
1032 // get method access flags
1033 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1034 movl(rcx, Address(rcx, Method::access_flags_offset()));
1035 testl(rcx, JVM_ACC_SYNCHRONIZED);
1036 jcc(Assembler::zero, unlocked);
1037
1038 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1039 // is set.
1040 testbool(rbx);
1041 jcc(Assembler::notZero, no_unlock);
1042
1043 // unlock monitor
1044 push(state); // save result
1045
1046 // BasicObjectLock will be first in list, since this is a
1047 // synchronized method. However, need to check that the object has
1048 // not been unlocked by an explicit monitorexit bytecode.
1049 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
1050 wordSize - (int) sizeof(BasicObjectLock));
1051 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1052 // register for unlock_object to pass to VM directly
1053 lea(robj, monitor); // address of first monitor
1054
1055 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
1056 testptr(rax, rax);
1057 jcc(Assembler::notZero, unlock);
1058
1059 pop(state);
1060 if (throw_monitor_exception) {
1061 // Entry already unlocked, need to throw exception
1062 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1063 call_VM(noreg, CAST_FROM_FN_PTR(address,
1064 InterpreterRuntime::throw_illegal_monitor_state_exception));
1065 should_not_reach_here();
1066 } else {
1067 // Monitor already unlocked during a stack unroll. If requested,
1068 // install an illegal_monitor_state_exception. Continue with
1069 // stack unrolling.
1070 if (install_monitor_exception) {
1071 NOT_LP64(empty_FPU_stack();)
1072 call_VM(noreg, CAST_FROM_FN_PTR(address,
1073 InterpreterRuntime::new_illegal_monitor_state_exception));
1074 }
1075 jmp(unlocked);
1076 }
1077
1078 bind(unlock);
1079 unlock_object(robj);
1080 pop(state);
1081
1082 // Check that for block-structured locking (i.e., that all locked
1083 // objects has been unlocked)
1084 bind(unlocked);
1085
1086 // rax, rdx: Might contain return value
1087
1088 // Check that all monitors are unlocked
1089 {
1090 Label loop, exception, entry, restart;
1091 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1092 const Address monitor_block_top(
1093 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1094 const Address monitor_block_bot(
1095 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1096
1097 bind(restart);
1098 // We use c_rarg1 so that if we go slow path it will be the correct
1099 // register for unlock_object to pass to VM directly
1100 movptr(rmon, monitor_block_top); // derelativize pointer
1101 lea(rmon, Address(rbp, rmon, Address::times_ptr));
1102 // c_rarg1 points to current entry, starting with top-most entry
1103
1104 lea(rbx, monitor_block_bot); // points to word before bottom of
1105 // monitor block
1106 jmp(entry);
1107
1108 // Entry already locked, need to throw exception
1109 bind(exception);
1110
1111 if (throw_monitor_exception) {
1112 // Throw exception
1113 NOT_LP64(empty_FPU_stack();)
1114 MacroAssembler::call_VM(noreg,
1115 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1116 throw_illegal_monitor_state_exception));
1117 should_not_reach_here();
1118 } else {
1119 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1120 // Unlock does not block, so don't have to worry about the frame.
1121 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1122
1123 push(state);
1124 mov(robj, rmon); // nop if robj and rmon are the same
1125 unlock_object(robj);
1126 pop(state);
1127
1128 if (install_monitor_exception) {
1129 NOT_LP64(empty_FPU_stack();)
1130 call_VM(noreg, CAST_FROM_FN_PTR(address,
1131 InterpreterRuntime::
1132 new_illegal_monitor_state_exception));
1133 }
1134
1135 jmp(restart);
1136 }
1137
1138 bind(loop);
1139 // check if current entry is used
1140 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1141 jcc(Assembler::notEqual, exception);
1142
1143 addptr(rmon, entry_size); // otherwise advance to next entry
1144 bind(entry);
1145 cmpptr(rmon, rbx); // check if bottom reached
1146 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1147 }
1148
1149 bind(no_unlock);
1150
1151 // jvmti support
1152 if (notify_jvmdi) {
1153 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1154 } else {
1155 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1156 }
1157
1158 // remove activation
1159 // get sender sp
1160 movptr(rbx,
1161 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1162 if (StackReservedPages > 0) {
1163 // testing if reserved zone needs to be re-enabled
1164 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1165 Label no_reserved_zone_enabling;
1166
1167 NOT_LP64(get_thread(rthread);)
1168
1169 // check if already enabled - if so no re-enabling needed
1170 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1171 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1172 jcc(Assembler::equal, no_reserved_zone_enabling);
1173
1174 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1175 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1176
1177 call_VM_leaf(
1178 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1179 call_VM(noreg, CAST_FROM_FN_PTR(address,
1180 InterpreterRuntime::throw_delayed_StackOverflowError));
1181 should_not_reach_here();
1182
1183 bind(no_reserved_zone_enabling);
1184 }
1185 leave(); // remove frame anchor
1186 pop(ret_addr); // get return address
1187 mov(rsp, rbx); // set sp to sender sp
1188 pop_cont_fastpath();
1189 }
1190
1191 void InterpreterMacroAssembler::get_method_counters(Register method,
1192 Register mcs, Label& skip) {
1193 Label has_counters;
1194 movptr(mcs, Address(method, Method::method_counters_offset()));
1195 testptr(mcs, mcs);
1196 jcc(Assembler::notZero, has_counters);
1197 call_VM(noreg, CAST_FROM_FN_PTR(address,
1198 InterpreterRuntime::build_method_counters), method);
1199 movptr(mcs, Address(method,Method::method_counters_offset()));
1200 testptr(mcs, mcs);
1201 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1202 bind(has_counters);
1203 }
1204
1205
1206 // Lock object
1207 //
1208 // Args:
1209 // rdx, c_rarg1: BasicObjectLock to be used for locking
1210 //
1211 // Kills:
1212 // rax, rbx
1213 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1214 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1215 "The argument is only for looks. It must be c_rarg1");
1216
1217 if (LockingMode == LM_MONITOR) {
1218 call_VM_preemptable(noreg,
1219 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1220 lock_reg);
1221 } else {
1222 Label count_locking, done, slow_case;
1223
1224 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1225 const Register tmp_reg = rbx;
1226 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1227 const Register rklass_decode_tmp = rscratch1;
1228
1229 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1230 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1231 const int mark_offset = lock_offset +
1232 BasicLock::displaced_header_offset_in_bytes();
1233
1234 // Load object pointer into obj_reg
1235 movptr(obj_reg, Address(lock_reg, obj_offset));
1236
1237 if (DiagnoseSyncOnValueBasedClasses != 0) {
1238 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1239 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1240 jcc(Assembler::notZero, slow_case);
1241 }
1242
1243 if (LockingMode == LM_LIGHTWEIGHT) {
1244 #ifdef _LP64
1245 const Register thread = r15_thread;
1246 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1247 #else
1248 // Lacking registers and thread on x86_32. Always take slow path.
1249 jmp(slow_case);
1250 #endif
1251 } else if (LockingMode == LM_LEGACY) {
1252 // Load immediate 1 into swap_reg %rax
1253 movl(swap_reg, 1);
1254
1255 // Load (object->mark() | 1) into swap_reg %rax
1256 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1257
1258 // Save (object->mark() | 1) into BasicLock's displaced header
1259 movptr(Address(lock_reg, mark_offset), swap_reg);
1260
1261 assert(lock_offset == 0,
1262 "displaced header must be first word in BasicObjectLock");
1263
1264 lock();
1265 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1266 jcc(Assembler::zero, count_locking);
1267
1268 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1269
1270 // Fast check for recursive lock.
1271 //
1272 // Can apply the optimization only if this is a stack lock
1273 // allocated in this thread. For efficiency, we can focus on
1274 // recently allocated stack locks (instead of reading the stack
1275 // base and checking whether 'mark' points inside the current
1276 // thread stack):
1277 // 1) (mark & zero_bits) == 0, and
1278 // 2) rsp <= mark < mark + os::pagesize()
1279 //
1280 // Warning: rsp + os::pagesize can overflow the stack base. We must
1281 // neither apply the optimization for an inflated lock allocated
1282 // just above the thread stack (this is why condition 1 matters)
1283 // nor apply the optimization if the stack lock is inside the stack
1284 // of another thread. The latter is avoided even in case of overflow
1285 // because we have guard pages at the end of all stacks. Hence, if
1286 // we go over the stack base and hit the stack of another thread,
1287 // this should not be in a writeable area that could contain a
1288 // stack lock allocated by that thread. As a consequence, a stack
1289 // lock less than page size away from rsp is guaranteed to be
1290 // owned by the current thread.
1291 //
1292 // These 3 tests can be done by evaluating the following
1293 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1294 // assuming both stack pointer and pagesize have their
1295 // least significant bits clear.
1296 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1297 subptr(swap_reg, rsp);
1298 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1299
1300 // Save the test result, for recursive case, the result is zero
1301 movptr(Address(lock_reg, mark_offset), swap_reg);
1302 jcc(Assembler::notZero, slow_case);
1303
1304 bind(count_locking);
1305 inc_held_monitor_count();
1306 }
1307 jmp(done);
1308
1309 bind(slow_case);
1310
1311 // Call the runtime routine for slow case
1312 call_VM_preemptable(noreg,
1313 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1314 lock_reg);
1315 bind(done);
1316 }
1317 }
1318
1319
1320 // Unlocks an object. Used in monitorexit bytecode and
1321 // remove_activation. Throws an IllegalMonitorException if object is
1322 // not locked by current thread.
1323 //
1324 // Args:
1325 // rdx, c_rarg1: BasicObjectLock for lock
1326 //
1327 // Kills:
1328 // rax
1329 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1330 // rscratch1 (scratch reg)
1331 // rax, rbx, rcx, rdx
1332 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1333 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1334 "The argument is only for looks. It must be c_rarg1");
1335
1336 if (LockingMode == LM_MONITOR) {
1337 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1338 } else {
1339 Label count_locking, done, slow_case;
1340
1341 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1342 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1343 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1344
1345 save_bcp(); // Save in case of exception
1346
1347 if (LockingMode != LM_LIGHTWEIGHT) {
1348 // Convert from BasicObjectLock structure to object and BasicLock
1349 // structure Store the BasicLock address into %rax
1350 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1351 }
1352
1353 // Load oop into obj_reg(%c_rarg3)
1354 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1355
1356 // Free entry
1357 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1358
1359 if (LockingMode == LM_LIGHTWEIGHT) {
1360 #ifdef _LP64
1361 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1362 #else
1363 // Lacking registers and thread on x86_32. Always take slow path.
1364 jmp(slow_case);
1365 #endif
1366 } else if (LockingMode == LM_LEGACY) {
1367 // Load the old header from BasicLock structure
1368 movptr(header_reg, Address(swap_reg,
1369 BasicLock::displaced_header_offset_in_bytes()));
1370
1371 // Test for recursion
1372 testptr(header_reg, header_reg);
1373
1374 // zero for recursive case
1375 jcc(Assembler::zero, count_locking);
1376
1377 // Atomic swap back the old header
1378 lock();
1379 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1380
1381 // zero for simple unlock of a stack-lock case
1382 jcc(Assembler::notZero, slow_case);
1383
1384 bind(count_locking);
1385 dec_held_monitor_count();
1386 }
1387 jmp(done);
1388
1389 bind(slow_case);
1390 // Call the runtime routine for slow case.
1391 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1392 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1393
1394 bind(done);
1395
1396 restore_bcp();
1397 }
1398 }
1399
1400 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1401 Label& zero_continue) {
1402 assert(ProfileInterpreter, "must be profiling interpreter");
1403 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1404 testptr(mdp, mdp);
1405 jcc(Assembler::zero, zero_continue);
1406 }
1407
1408
1409 // Set the method data pointer for the current bcp.
1410 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1411 assert(ProfileInterpreter, "must be profiling interpreter");
1412 Label set_mdp;
1413 push(rax);
1414 push(rbx);
1415
1416 get_method(rbx);
1417 // Test MDO to avoid the call if it is null.
1418 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1419 testptr(rax, rax);
1420 jcc(Assembler::zero, set_mdp);
1421 // rbx: method
1422 // _bcp_register: bcp
1423 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1424 // rax: mdi
1425 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1426 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1427 addptr(rbx, in_bytes(MethodData::data_offset()));
1428 addptr(rax, rbx);
1429 bind(set_mdp);
1430 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1431 pop(rbx);
1432 pop(rax);
1433 }
1434
1435 void InterpreterMacroAssembler::verify_method_data_pointer() {
1436 assert(ProfileInterpreter, "must be profiling interpreter");
1437 #ifdef ASSERT
1438 Label verify_continue;
1439 push(rax);
1440 push(rbx);
1441 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1442 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1443 push(arg3_reg);
1444 push(arg2_reg);
1445 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1446 get_method(rbx);
1447
1448 // If the mdp is valid, it will point to a DataLayout header which is
1449 // consistent with the bcp. The converse is highly probable also.
1450 load_unsigned_short(arg2_reg,
1451 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1452 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1453 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1454 cmpptr(arg2_reg, _bcp_register);
1455 jcc(Assembler::equal, verify_continue);
1456 // rbx: method
1457 // _bcp_register: bcp
1458 // c_rarg3: mdp
1459 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1460 rbx, _bcp_register, arg3_reg);
1461 bind(verify_continue);
1462 pop(arg2_reg);
1463 pop(arg3_reg);
1464 pop(rbx);
1465 pop(rax);
1466 #endif // ASSERT
1467 }
1468
1469
1470 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1471 int constant,
1472 Register value) {
1473 assert(ProfileInterpreter, "must be profiling interpreter");
1474 Address data(mdp_in, constant);
1475 movptr(data, value);
1476 }
1477
1478
1479 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1480 int constant,
1481 bool decrement) {
1482 // Counter address
1483 Address data(mdp_in, constant);
1484
1485 increment_mdp_data_at(data, decrement);
1486 }
1487
1488 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1489 bool decrement) {
1490 assert(ProfileInterpreter, "must be profiling interpreter");
1491 // %%% this does 64bit counters at best it is wasting space
1492 // at worst it is a rare bug when counters overflow
1493
1494 if (decrement) {
1495 // Decrement the register. Set condition codes.
1496 addptr(data, -DataLayout::counter_increment);
1497 // If the decrement causes the counter to overflow, stay negative
1498 Label L;
1499 jcc(Assembler::negative, L);
1500 addptr(data, DataLayout::counter_increment);
1501 bind(L);
1502 } else {
1503 assert(DataLayout::counter_increment == 1,
1504 "flow-free idiom only works with 1");
1505 // Increment the register. Set carry flag.
1506 addptr(data, DataLayout::counter_increment);
1507 // If the increment causes the counter to overflow, pull back by 1.
1508 sbbptr(data, 0);
1509 }
1510 }
1511
1512
1513 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1514 Register reg,
1515 int constant,
1516 bool decrement) {
1517 Address data(mdp_in, reg, Address::times_1, constant);
1518
1519 increment_mdp_data_at(data, decrement);
1520 }
1521
1522 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1523 int flag_byte_constant) {
1524 assert(ProfileInterpreter, "must be profiling interpreter");
1525 int header_offset = in_bytes(DataLayout::flags_offset());
1526 int header_bits = flag_byte_constant;
1527 // Set the flag
1528 orb(Address(mdp_in, header_offset), header_bits);
1529 }
1530
1531
1532
1533 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1534 int offset,
1535 Register value,
1536 Register test_value_out,
1537 Label& not_equal_continue) {
1538 assert(ProfileInterpreter, "must be profiling interpreter");
1539 if (test_value_out == noreg) {
1540 cmpptr(value, Address(mdp_in, offset));
1541 } else {
1542 // Put the test value into a register, so caller can use it:
1543 movptr(test_value_out, Address(mdp_in, offset));
1544 cmpptr(test_value_out, value);
1545 }
1546 jcc(Assembler::notEqual, not_equal_continue);
1547 }
1548
1549
1550 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1551 int offset_of_disp) {
1552 assert(ProfileInterpreter, "must be profiling interpreter");
1553 Address disp_address(mdp_in, offset_of_disp);
1554 addptr(mdp_in, disp_address);
1555 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1556 }
1557
1558
1559 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1560 Register reg,
1561 int offset_of_disp) {
1562 assert(ProfileInterpreter, "must be profiling interpreter");
1563 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1564 addptr(mdp_in, disp_address);
1565 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1566 }
1567
1568
1569 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1570 int constant) {
1571 assert(ProfileInterpreter, "must be profiling interpreter");
1572 addptr(mdp_in, constant);
1573 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1574 }
1575
1576
1577 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1578 assert(ProfileInterpreter, "must be profiling interpreter");
1579 push(return_bci); // save/restore across call_VM
1580 call_VM(noreg,
1581 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1582 return_bci);
1583 pop(return_bci);
1584 }
1585
1586
1587 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1588 Register bumped_count) {
1589 if (ProfileInterpreter) {
1590 Label profile_continue;
1591
1592 // If no method data exists, go to profile_continue.
1593 // Otherwise, assign to mdp
1594 test_method_data_pointer(mdp, profile_continue);
1595
1596 // We are taking a branch. Increment the taken count.
1597 // We inline increment_mdp_data_at to return bumped_count in a register
1598 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1599 Address data(mdp, in_bytes(JumpData::taken_offset()));
1600 movptr(bumped_count, data);
1601 assert(DataLayout::counter_increment == 1,
1602 "flow-free idiom only works with 1");
1603 addptr(bumped_count, DataLayout::counter_increment);
1604 sbbptr(bumped_count, 0);
1605 movptr(data, bumped_count); // Store back out
1606
1607 // The method data pointer needs to be updated to reflect the new target.
1608 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1609 bind(profile_continue);
1610 }
1611 }
1612
1613
1614 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1615 if (ProfileInterpreter) {
1616 Label profile_continue;
1617
1618 // If no method data exists, go to profile_continue.
1619 test_method_data_pointer(mdp, profile_continue);
1620
1621 // We are taking a branch. Increment the not taken count.
1622 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1623
1624 // The method data pointer needs to be updated to correspond to
1625 // the next bytecode
1626 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1627 bind(profile_continue);
1628 }
1629 }
1630
1631 void InterpreterMacroAssembler::profile_call(Register mdp) {
1632 if (ProfileInterpreter) {
1633 Label profile_continue;
1634
1635 // If no method data exists, go to profile_continue.
1636 test_method_data_pointer(mdp, profile_continue);
1637
1638 // We are making a call. Increment the count.
1639 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1640
1641 // The method data pointer needs to be updated to reflect the new target.
1642 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1643 bind(profile_continue);
1644 }
1645 }
1646
1647
1648 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1649 if (ProfileInterpreter) {
1650 Label profile_continue;
1651
1652 // If no method data exists, go to profile_continue.
1653 test_method_data_pointer(mdp, profile_continue);
1654
1655 // We are making a call. Increment the count.
1656 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1657
1658 // The method data pointer needs to be updated to reflect the new target.
1659 update_mdp_by_constant(mdp,
1660 in_bytes(VirtualCallData::
1661 virtual_call_data_size()));
1662 bind(profile_continue);
1663 }
1664 }
1665
1666
1667 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1668 Register mdp,
1669 Register reg2,
1670 bool receiver_can_be_null) {
1671 if (ProfileInterpreter) {
1672 Label profile_continue;
1673
1674 // If no method data exists, go to profile_continue.
1675 test_method_data_pointer(mdp, profile_continue);
1676
1677 Label skip_receiver_profile;
1678 if (receiver_can_be_null) {
1679 Label not_null;
1680 testptr(receiver, receiver);
1681 jccb(Assembler::notZero, not_null);
1682 // We are making a call. Increment the count for null receiver.
1683 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1684 jmp(skip_receiver_profile);
1685 bind(not_null);
1686 }
1687
1688 // Record the receiver type.
1689 record_klass_in_profile(receiver, mdp, reg2, true);
1690 bind(skip_receiver_profile);
1691
1692 // The method data pointer needs to be updated to reflect the new target.
1693 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1694 bind(profile_continue);
1695 }
1696 }
1697
1698 // This routine creates a state machine for updating the multi-row
1699 // type profile at a virtual call site (or other type-sensitive bytecode).
1700 // The machine visits each row (of receiver/count) until the receiver type
1701 // is found, or until it runs out of rows. At the same time, it remembers
1702 // the location of the first empty row. (An empty row records null for its
1703 // receiver, and can be allocated for a newly-observed receiver type.)
1704 // Because there are two degrees of freedom in the state, a simple linear
1705 // search will not work; it must be a decision tree. Hence this helper
1706 // function is recursive, to generate the required tree structured code.
1707 // It's the interpreter, so we are trading off code space for speed.
1708 // See below for example code.
1709 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1710 Register receiver, Register mdp,
1711 Register reg2, int start_row,
1712 Label& done, bool is_virtual_call) {
1713 if (TypeProfileWidth == 0) {
1714 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1715 } else {
1716 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1717 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1718 }
1719 }
1720
1721 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1722 Label& done, int total_rows,
1723 OffsetFunction item_offset_fn,
1724 OffsetFunction item_count_offset_fn) {
1725 int last_row = total_rows - 1;
1726 assert(start_row <= last_row, "must be work left to do");
1727 // Test this row for both the item and for null.
1728 // Take any of three different outcomes:
1729 // 1. found item => increment count and goto done
1730 // 2. found null => keep looking for case 1, maybe allocate this cell
1731 // 3. found something else => keep looking for cases 1 and 2
1732 // Case 3 is handled by a recursive call.
1733 for (int row = start_row; row <= last_row; row++) {
1734 Label next_test;
1735 bool test_for_null_also = (row == start_row);
1736
1737 // See if the item is item[n].
1738 int item_offset = in_bytes(item_offset_fn(row));
1739 test_mdp_data_at(mdp, item_offset, item,
1740 (test_for_null_also ? reg2 : noreg),
1741 next_test);
1742 // (Reg2 now contains the item from the CallData.)
1743
1744 // The item is item[n]. Increment count[n].
1745 int count_offset = in_bytes(item_count_offset_fn(row));
1746 increment_mdp_data_at(mdp, count_offset);
1747 jmp(done);
1748 bind(next_test);
1749
1750 if (test_for_null_also) {
1751 // Failed the equality check on item[n]... Test for null.
1752 testptr(reg2, reg2);
1753 if (start_row == last_row) {
1754 // The only thing left to do is handle the null case.
1755 Label found_null;
1756 jccb(Assembler::zero, found_null);
1757 // Item did not match any saved item and there is no empty row for it.
1758 // Increment total counter to indicate polymorphic case.
1759 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1760 jmp(done);
1761 bind(found_null);
1762 break;
1763 }
1764 Label found_null;
1765 // Since null is rare, make it be the branch-taken case.
1766 jcc(Assembler::zero, found_null);
1767
1768 // Put all the "Case 3" tests here.
1769 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1770 item_offset_fn, item_count_offset_fn);
1771
1772 // Found a null. Keep searching for a matching item,
1773 // but remember that this is an empty (unused) slot.
1774 bind(found_null);
1775 }
1776 }
1777
1778 // In the fall-through case, we found no matching item, but we
1779 // observed the item[start_row] is null.
1780
1781 // Fill in the item field and increment the count.
1782 int item_offset = in_bytes(item_offset_fn(start_row));
1783 set_mdp_data_at(mdp, item_offset, item);
1784 int count_offset = in_bytes(item_count_offset_fn(start_row));
1785 movl(reg2, DataLayout::counter_increment);
1786 set_mdp_data_at(mdp, count_offset, reg2);
1787 if (start_row > 0) {
1788 jmp(done);
1789 }
1790 }
1791
1792 // Example state machine code for three profile rows:
1793 // // main copy of decision tree, rooted at row[1]
1794 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1795 // if (row[0].rec != nullptr) {
1796 // // inner copy of decision tree, rooted at row[1]
1797 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1798 // if (row[1].rec != nullptr) {
1799 // // degenerate decision tree, rooted at row[2]
1800 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1801 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1802 // row[2].init(rec); goto done;
1803 // } else {
1804 // // remember row[1] is empty
1805 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1806 // row[1].init(rec); goto done;
1807 // }
1808 // } else {
1809 // // remember row[0] is empty
1810 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1811 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1812 // row[0].init(rec); goto done;
1813 // }
1814 // done:
1815
1816 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1817 Register mdp, Register reg2,
1818 bool is_virtual_call) {
1819 assert(ProfileInterpreter, "must be profiling");
1820 Label done;
1821
1822 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1823
1824 bind (done);
1825 }
1826
1827 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1828 Register mdp) {
1829 if (ProfileInterpreter) {
1830 Label profile_continue;
1831 uint row;
1832
1833 // If no method data exists, go to profile_continue.
1834 test_method_data_pointer(mdp, profile_continue);
1835
1836 // Update the total ret count.
1837 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1838
1839 for (row = 0; row < RetData::row_limit(); row++) {
1840 Label next_test;
1841
1842 // See if return_bci is equal to bci[n]:
1843 test_mdp_data_at(mdp,
1844 in_bytes(RetData::bci_offset(row)),
1845 return_bci, noreg,
1846 next_test);
1847
1848 // return_bci is equal to bci[n]. Increment the count.
1849 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1850
1851 // The method data pointer needs to be updated to reflect the new target.
1852 update_mdp_by_offset(mdp,
1853 in_bytes(RetData::bci_displacement_offset(row)));
1854 jmp(profile_continue);
1855 bind(next_test);
1856 }
1857
1858 update_mdp_for_ret(return_bci);
1859
1860 bind(profile_continue);
1861 }
1862 }
1863
1864
1865 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1866 if (ProfileInterpreter) {
1867 Label profile_continue;
1868
1869 // If no method data exists, go to profile_continue.
1870 test_method_data_pointer(mdp, profile_continue);
1871
1872 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1873
1874 // The method data pointer needs to be updated.
1875 int mdp_delta = in_bytes(BitData::bit_data_size());
1876 if (TypeProfileCasts) {
1877 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1878 }
1879 update_mdp_by_constant(mdp, mdp_delta);
1880
1881 bind(profile_continue);
1882 }
1883 }
1884
1885
1886 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1887 if (ProfileInterpreter) {
1888 Label profile_continue;
1889
1890 // If no method data exists, go to profile_continue.
1891 test_method_data_pointer(mdp, profile_continue);
1892
1893 // The method data pointer needs to be updated.
1894 int mdp_delta = in_bytes(BitData::bit_data_size());
1895 if (TypeProfileCasts) {
1896 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1897
1898 // Record the object type.
1899 record_klass_in_profile(klass, mdp, reg2, false);
1900 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1901 NOT_LP64(restore_locals();) // Restore EDI
1902 }
1903 update_mdp_by_constant(mdp, mdp_delta);
1904
1905 bind(profile_continue);
1906 }
1907 }
1908
1909
1910 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1911 if (ProfileInterpreter) {
1912 Label profile_continue;
1913
1914 // If no method data exists, go to profile_continue.
1915 test_method_data_pointer(mdp, profile_continue);
1916
1917 // Update the default case count
1918 increment_mdp_data_at(mdp,
1919 in_bytes(MultiBranchData::default_count_offset()));
1920
1921 // The method data pointer needs to be updated.
1922 update_mdp_by_offset(mdp,
1923 in_bytes(MultiBranchData::
1924 default_displacement_offset()));
1925
1926 bind(profile_continue);
1927 }
1928 }
1929
1930
1931 void InterpreterMacroAssembler::profile_switch_case(Register index,
1932 Register mdp,
1933 Register reg2) {
1934 if (ProfileInterpreter) {
1935 Label profile_continue;
1936
1937 // If no method data exists, go to profile_continue.
1938 test_method_data_pointer(mdp, profile_continue);
1939
1940 // Build the base (index * per_case_size_in_bytes()) +
1941 // case_array_offset_in_bytes()
1942 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1943 imulptr(index, reg2); // XXX l ?
1944 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1945
1946 // Update the case count
1947 increment_mdp_data_at(mdp,
1948 index,
1949 in_bytes(MultiBranchData::relative_count_offset()));
1950
1951 // The method data pointer needs to be updated.
1952 update_mdp_by_offset(mdp,
1953 index,
1954 in_bytes(MultiBranchData::
1955 relative_displacement_offset()));
1956
1957 bind(profile_continue);
1958 }
1959 }
1960
1961
1962
1963 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1964 if (state == atos) {
1965 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1966 }
1967 }
1968
1969 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1970 #ifndef _LP64
1971 if ((state == ftos && UseSSE < 1) ||
1972 (state == dtos && UseSSE < 2)) {
1973 MacroAssembler::verify_FPU(stack_depth);
1974 }
1975 #endif
1976 }
1977
1978 // Jump if ((*counter_addr += increment) & mask) == 0
1979 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
1980 Register scratch, Label* where) {
1981 // This update is actually not atomic and can lose a number of updates
1982 // under heavy contention, but the alternative of using the (contended)
1983 // atomic update here penalizes profiling paths too much.
1984 movl(scratch, counter_addr);
1985 incrementl(scratch, InvocationCounter::count_increment);
1986 movl(counter_addr, scratch);
1987 andl(scratch, mask);
1988 if (where != nullptr) {
1989 jcc(Assembler::zero, *where);
1990 }
1991 }
1992
1993 void InterpreterMacroAssembler::generate_runtime_upcalls_on_method_entry()
1994 {
1995 address upcall = RuntimeUpcalls::on_method_entry_upcall_address();
1996 if (RuntimeUpcalls::does_upcall_need_method_parameter(upcall)) {
1997 get_method(c_rarg1);
1998 call_VM(noreg,upcall, c_rarg1);
1999 } else {
2000 call_VM(noreg,upcall);
2001 }
2002 }
2003
2004 void InterpreterMacroAssembler::notify_method_entry() {
2005 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2006 // track stack depth. If it is possible to enter interp_only_mode we add
2007 // the code to check if the event should be sent.
2008 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2009 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2010 if (JvmtiExport::can_post_interpreter_events()) {
2011 Label L;
2012 NOT_LP64(get_thread(rthread);)
2013 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2014 testl(rdx, rdx);
2015 jcc(Assembler::zero, L);
2016 call_VM(noreg, CAST_FROM_FN_PTR(address,
2017 InterpreterRuntime::post_method_entry));
2018 bind(L);
2019 }
2020
2021 if (DTraceMethodProbes) {
2022 NOT_LP64(get_thread(rthread);)
2023 get_method(rarg);
2024 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2025 rthread, rarg);
2026 }
2027
2028 // RedefineClasses() tracing support for obsolete method entry
2029 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2030 NOT_LP64(get_thread(rthread);)
2031 get_method(rarg);
2032 call_VM_leaf(
2033 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2034 rthread, rarg);
2035 }
2036 }
2037
2038
2039 void InterpreterMacroAssembler::notify_method_exit(
2040 TosState state, NotifyMethodExitMode mode) {
2041 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2042 // track stack depth. If it is possible to enter interp_only_mode we add
2043 // the code to check if the event should be sent.
2044 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2045 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2046 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2047 Label L;
2048 // Note: frame::interpreter_frame_result has a dependency on how the
2049 // method result is saved across the call to post_method_exit. If this
2050 // is changed then the interpreter_frame_result implementation will
2051 // need to be updated too.
2052
2053 // template interpreter will leave the result on the top of the stack.
2054 push(state);
2055 NOT_LP64(get_thread(rthread);)
2056 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2057 testl(rdx, rdx);
2058 jcc(Assembler::zero, L);
2059 call_VM(noreg,
2060 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2061 bind(L);
2062 pop(state);
2063 }
2064
2065 if (DTraceMethodProbes) {
2066 push(state);
2067 NOT_LP64(get_thread(rthread);)
2068 get_method(rarg);
2069 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2070 rthread, rarg);
2071 pop(state);
2072 }
2073 }
2074
2075 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2076 // Get index out of bytecode pointer
2077 get_cache_index_at_bcp(index, 1, sizeof(u4));
2078 // Get address of invokedynamic array
2079 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2080 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2081 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2082 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2083 } else {
2084 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2085 }
2086 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2087 }
2088
2089 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2090 // Get index out of bytecode pointer
2091 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2092 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2093
2094 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2095 // Take shortcut if the size is a power of 2
2096 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2097 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2098 } else {
2099 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2100 }
2101 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2102 }
2103
2104 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2105 // Get index out of bytecode pointer
2106 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2107 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2108
2109 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2110 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2111 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2112 }