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/constMethodFlags.hpp"
33 #include "oops/markWord.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/inlineKlass.hpp"
37 #include "oops/resolvedFieldEntry.hpp"
38 #include "oops/resolvedIndyEntry.hpp"
39 #include "oops/resolvedMethodEntry.hpp"
40 #include "prims/jvmtiExport.hpp"
41 #include "prims/jvmtiThreadState.hpp"
42 #include "runtime/basicLock.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/javaThread.hpp"
45 #include "runtime/safepointMechanism.hpp"
46 #include "runtime/sharedRuntime.hpp"
47 #include "utilities/powerOfTwo.hpp"
48
49 // Implementation of InterpreterMacroAssembler
50
51 void InterpreterMacroAssembler::jump_to_entry(address entry) {
52 assert(entry, "Entry must have been generated by now");
53 jump(RuntimeAddress(entry));
54 }
55
56 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
57 Label update, next, none;
58
59 #ifdef _LP64
60 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
61 #else
62 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index());
63 #endif
64
65 interp_verify_oop(obj, atos);
66
67 testptr(obj, obj);
68 jccb(Assembler::notZero, update);
69 testptr(mdo_addr, TypeEntries::null_seen);
70 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore.
71 // atomic update to prevent overwriting Klass* with 0
72 lock();
73 orptr(mdo_addr, TypeEntries::null_seen);
74 jmpb(next);
75
76 bind(update);
77 load_klass(obj, obj, rscratch1);
78 #ifdef _LP64
79 mov(rscratch1, obj);
80 #endif
81
82 xorptr(obj, mdo_addr);
83 testptr(obj, TypeEntries::type_klass_mask);
84 jccb(Assembler::zero, next); // klass seen before, nothing to
85 // do. The unknown bit may have been
86 // set already but no need to check.
87
88 testptr(obj, TypeEntries::type_unknown);
89 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
90
91 cmpptr(mdo_addr, 0);
92 jccb(Assembler::equal, none);
93 cmpptr(mdo_addr, TypeEntries::null_seen);
94 jccb(Assembler::equal, none);
95 #ifdef _LP64
96 // There is a chance that the checks above (re-reading profiling
97 // data from memory) fail if another thread has just set the
98 // profiling to this obj's klass
99 mov(obj, rscratch1);
100 xorptr(obj, mdo_addr);
101 testptr(obj, TypeEntries::type_klass_mask);
102 jccb(Assembler::zero, next);
103 #endif
104
105 // different than before. Cannot keep accurate profile.
106 orptr(mdo_addr, TypeEntries::type_unknown);
107 jmpb(next);
108
109 bind(none);
110 // first time here. Set profile type.
111 movptr(mdo_addr, obj);
112 #ifdef ASSERT
113 andptr(obj, TypeEntries::type_klass_mask);
114 verify_klass_ptr(obj);
115 #endif
116
117 bind(next);
118 }
119
120 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
121 if (!ProfileInterpreter) {
122 return;
123 }
124
125 if (MethodData::profile_arguments() || MethodData::profile_return()) {
126 Label profile_continue;
127
128 test_method_data_pointer(mdp, profile_continue);
129
130 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
131
132 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
133 jcc(Assembler::notEqual, profile_continue);
134
135 if (MethodData::profile_arguments()) {
136 Label done;
137 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
138 addptr(mdp, off_to_args);
139
140 for (int i = 0; i < TypeProfileArgsLimit; i++) {
141 if (i > 0 || MethodData::profile_return()) {
142 // If return value type is profiled we may have no argument to profile
143 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
144 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
145 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
146 jcc(Assembler::less, done);
147 }
148 movptr(tmp, Address(callee, Method::const_offset()));
149 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
150 // stack offset o (zero based) from the start of the argument
151 // list, for n arguments translates into offset n - o - 1 from
152 // the end of the argument list
153 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
154 subl(tmp, 1);
155 Address arg_addr = argument_address(tmp);
156 movptr(tmp, arg_addr);
157
158 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
159 profile_obj_type(tmp, mdo_arg_addr);
160
161 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
162 addptr(mdp, to_add);
163 off_to_args += to_add;
164 }
165
166 if (MethodData::profile_return()) {
167 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
168 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
169 }
170
171 bind(done);
172
173 if (MethodData::profile_return()) {
174 // We're right after the type profile for the last
175 // argument. tmp is the number of cells left in the
176 // CallTypeData/VirtualCallTypeData to reach its end. Non null
177 // if there's a return to profile.
178 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
179 shll(tmp, log2i_exact((int)DataLayout::cell_size));
180 addptr(mdp, tmp);
181 }
182 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp);
183 } else {
184 assert(MethodData::profile_return(), "either profile call args or call ret");
185 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
186 }
187
188 // mdp points right after the end of the
189 // CallTypeData/VirtualCallTypeData, right after the cells for the
190 // return value type if there's one
191
192 bind(profile_continue);
193 }
194 }
195
196 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
197 assert_different_registers(mdp, ret, tmp, _bcp_register);
198 if (ProfileInterpreter && MethodData::profile_return()) {
199 Label profile_continue;
200
201 test_method_data_pointer(mdp, profile_continue);
202
203 if (MethodData::profile_return_jsr292_only()) {
204 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
205
206 // If we don't profile all invoke bytecodes we must make sure
207 // it's a bytecode we indeed profile. We can't go back to the
208 // beginning of the ProfileData we intend to update to check its
209 // type because we're right after it and we don't known its
210 // length
211 Label do_profile;
212 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
213 jcc(Assembler::equal, do_profile);
214 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
215 jcc(Assembler::equal, do_profile);
216 get_method(tmp);
217 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm));
218 jcc(Assembler::notEqual, profile_continue);
219
220 bind(do_profile);
221 }
222
223 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
224 mov(tmp, ret);
225 profile_obj_type(tmp, mdo_ret_addr);
226
227 bind(profile_continue);
228 }
229 }
230
231 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
232 if (ProfileInterpreter && MethodData::profile_parameters()) {
233 Label profile_continue;
234
235 test_method_data_pointer(mdp, profile_continue);
236
237 // Load the offset of the area within the MDO used for
238 // parameters. If it's negative we're not profiling any parameters
239 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
240 testl(tmp1, tmp1);
241 jcc(Assembler::negative, profile_continue);
242
243 // Compute a pointer to the area for parameters from the offset
244 // and move the pointer to the slot for the last
245 // parameters. Collect profiling from last parameter down.
246 // mdo start + parameters offset + array length - 1
247 addptr(mdp, tmp1);
248 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
249 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
250
251 Label loop;
252 bind(loop);
253
254 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
255 int type_base = in_bytes(ParametersTypeData::type_offset(0));
256 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
257 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
258 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
259
260 // load offset on the stack from the slot for this parameter
261 movptr(tmp2, arg_off);
262 negptr(tmp2);
263 // read the parameter from the local area
264 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
265
266 // profile the parameter
267 profile_obj_type(tmp2, arg_type);
268
269 // go to next parameter
270 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
271 jcc(Assembler::positive, loop);
272
273 bind(profile_continue);
274 }
275 }
276
277 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
278 int number_of_arguments) {
279 // interpreter specific
280 //
281 // Note: No need to save/restore bcp & locals registers
282 // since these are callee saved registers and no blocking/
283 // GC can happen in leaf calls.
284 // Further Note: DO NOT save/restore bcp/locals. If a caller has
285 // already saved them so that it can use rsi/rdi as temporaries
286 // then a save/restore here will DESTROY the copy the caller
287 // saved! There used to be a save_bcp() that only happened in
288 // the ASSERT path (no restore_bcp). Which caused bizarre failures
289 // when jvm built with ASSERTs.
290 #ifdef ASSERT
291 {
292 Label L;
293 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
294 jcc(Assembler::equal, L);
295 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
296 " last_sp != null");
297 bind(L);
298 }
299 #endif
300 // super call
301 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
302 // interpreter specific
303 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals
304 // but since they may not have been saved (and we don't want to
305 // save them here (see note above) the assert is invalid.
306 }
307
308 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
309 Register java_thread,
310 Register last_java_sp,
311 address entry_point,
312 int number_of_arguments,
313 bool check_exceptions) {
314 // interpreter specific
315 //
316 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
317 // really make a difference for these runtime calls, since they are
318 // slow anyway. Btw., bcp must be saved/restored since it may change
319 // due to GC.
320 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
321 save_bcp();
322 #ifdef ASSERT
323 {
324 Label L;
325 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
326 jcc(Assembler::equal, L);
327 stop("InterpreterMacroAssembler::call_VM_base:"
328 " last_sp isn't null");
329 bind(L);
330 }
331 #endif /* ASSERT */
332 // super call
333 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
334 entry_point, number_of_arguments,
335 check_exceptions);
336 // interpreter specific
337 restore_bcp();
338 restore_locals();
339 }
340
341 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
342 if (JvmtiExport::can_pop_frame()) {
343 Label L;
344 // Initiate popframe handling only if it is not already being
345 // processed. If the flag has the popframe_processing bit set, it
346 // means that this code is called *during* popframe handling - we
347 // don't want to reenter.
348 // This method is only called just after the call into the vm in
349 // call_VM_base, so the arg registers are available.
350 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
351 LP64_ONLY(c_rarg0);
352 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
353 testl(pop_cond, JavaThread::popframe_pending_bit);
354 jcc(Assembler::zero, L);
355 testl(pop_cond, JavaThread::popframe_processing_bit);
356 jcc(Assembler::notZero, L);
357 // Call Interpreter::remove_activation_preserving_args_entry() to get the
358 // address of the same-named entrypoint in the generated interpreter code.
359 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
360 jmp(rax);
361 bind(L);
362 NOT_LP64(get_thread(java_thread);)
363 }
364 }
365
366 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
367 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
368 NOT_LP64(get_thread(thread);)
369 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
370 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
371 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
372 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
373 #ifdef _LP64
374 switch (state) {
375 case atos: movptr(rax, oop_addr);
376 movptr(oop_addr, NULL_WORD);
377 interp_verify_oop(rax, state); break;
378 case ltos: movptr(rax, val_addr); break;
379 case btos: // fall through
380 case ztos: // fall through
381 case ctos: // fall through
382 case stos: // fall through
383 case itos: movl(rax, val_addr); break;
384 case ftos: load_float(val_addr); break;
385 case dtos: load_double(val_addr); break;
386 case vtos: /* nothing to do */ break;
387 default : ShouldNotReachHere();
388 }
389 // Clean up tos value in the thread object
390 movl(tos_addr, ilgl);
391 movl(val_addr, NULL_WORD);
392 #else
393 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
394 + in_ByteSize(wordSize));
395 switch (state) {
396 case atos: movptr(rax, oop_addr);
397 movptr(oop_addr, NULL_WORD);
398 interp_verify_oop(rax, state); break;
399 case ltos:
400 movl(rdx, val_addr1); // fall through
401 case btos: // fall through
402 case ztos: // fall through
403 case ctos: // fall through
404 case stos: // fall through
405 case itos: movl(rax, val_addr); break;
406 case ftos: load_float(val_addr); break;
407 case dtos: load_double(val_addr); break;
408 case vtos: /* nothing to do */ break;
409 default : ShouldNotReachHere();
410 }
411 #endif // _LP64
412 // Clean up tos value in the thread object
413 movl(tos_addr, ilgl);
414 movptr(val_addr, NULL_WORD);
415 NOT_LP64(movptr(val_addr1, NULL_WORD);)
416 }
417
418
419 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
420 if (JvmtiExport::can_force_early_return()) {
421 Label L;
422 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
423 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
424
425 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
426 testptr(tmp, tmp);
427 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit;
428
429 // Initiate earlyret handling only if it is not already being processed.
430 // If the flag has the earlyret_processing bit set, it means that this code
431 // is called *during* earlyret handling - we don't want to reenter.
432 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
433 cmpl(tmp, JvmtiThreadState::earlyret_pending);
434 jcc(Assembler::notEqual, L);
435
436 // Call Interpreter::remove_activation_early_entry() to get the address of the
437 // same-named entrypoint in the generated interpreter code.
438 NOT_LP64(get_thread(java_thread);)
439 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
440 #ifdef _LP64
441 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
442 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
443 #else
444 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
445 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
446 #endif // _LP64
447 jmp(rax);
448 bind(L);
449 NOT_LP64(get_thread(java_thread);)
450 }
451 }
452
453 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
454 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
455 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
456 bswapl(reg);
457 shrl(reg, 16);
458 }
459
460 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
461 int bcp_offset,
462 size_t index_size) {
463 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
464 if (index_size == sizeof(u2)) {
465 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
466 } else if (index_size == sizeof(u4)) {
467 movl(index, Address(_bcp_register, bcp_offset));
468 // Check if the secondary index definition is still ~x, otherwise
469 // we have to change the following assembler code to calculate the
470 // plain index.
471 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
472 notl(index); // convert to plain index
473 } else if (index_size == sizeof(u1)) {
474 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
475 } else {
476 ShouldNotReachHere();
477 }
478 }
479
480 // Load object from cpool->resolved_references(index)
481 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
482 Register index,
483 Register tmp) {
484 assert_different_registers(result, index);
485
486 get_constant_pool(result);
487 // load pointer for resolved_references[] objArray
488 movptr(result, Address(result, ConstantPool::cache_offset()));
489 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
490 resolve_oop_handle(result, tmp);
491 load_heap_oop(result, Address(result, index,
492 UseCompressedOops ? Address::times_4 : Address::times_ptr,
493 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
494 }
495
496 // load cpool->resolved_klass_at(index)
497 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
498 Register cpool,
499 Register index) {
500 assert_different_registers(cpool, index);
501
502 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
503 Register resolved_klasses = cpool;
504 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
505 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
506 }
507
508 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
509 // subtype of super_klass.
510 //
511 // Args:
512 // rax: superklass
513 // Rsub_klass: subklass
514 //
515 // Kills:
516 // rcx, rdi
517 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
518 Label& ok_is_subtype,
519 bool profile) {
520 assert(Rsub_klass != rax, "rax holds superklass");
521 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
522 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
523 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
524 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
525
526 // Profile the not-null value's klass.
527 if (profile) {
528 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
529 }
530
531 // Do the check.
532 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
533 }
534
535
536 #ifndef _LP64
537 void InterpreterMacroAssembler::f2ieee() {
538 if (IEEEPrecision) {
539 fstp_s(Address(rsp, 0));
540 fld_s(Address(rsp, 0));
541 }
542 }
543
544
545 void InterpreterMacroAssembler::d2ieee() {
546 if (IEEEPrecision) {
547 fstp_d(Address(rsp, 0));
548 fld_d(Address(rsp, 0));
549 }
550 }
551 #endif // _LP64
552
553 // Java Expression Stack
554
555 void InterpreterMacroAssembler::pop_ptr(Register r) {
556 pop(r);
557 }
558
559 void InterpreterMacroAssembler::push_ptr(Register r) {
560 push(r);
561 }
562
563 void InterpreterMacroAssembler::push_i(Register r) {
564 push(r);
565 }
566
567 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
568 push(r);
569 }
570
571 void InterpreterMacroAssembler::push_f(XMMRegister r) {
572 subptr(rsp, wordSize);
573 movflt(Address(rsp, 0), r);
574 }
575
576 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
577 movflt(r, Address(rsp, 0));
578 addptr(rsp, wordSize);
579 }
580
581 void InterpreterMacroAssembler::push_d(XMMRegister r) {
582 subptr(rsp, 2 * wordSize);
583 movdbl(Address(rsp, 0), r);
584 }
585
586 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
587 movdbl(r, Address(rsp, 0));
588 addptr(rsp, 2 * Interpreter::stackElementSize);
589 }
590
591 #ifdef _LP64
592 void InterpreterMacroAssembler::pop_i(Register r) {
593 // XXX can't use pop currently, upper half non clean
594 movl(r, Address(rsp, 0));
595 addptr(rsp, wordSize);
596 }
597
598 void InterpreterMacroAssembler::pop_l(Register r) {
599 movq(r, Address(rsp, 0));
600 addptr(rsp, 2 * Interpreter::stackElementSize);
601 }
602
603 void InterpreterMacroAssembler::push_l(Register r) {
604 subptr(rsp, 2 * wordSize);
605 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
606 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
607 }
608
609 void InterpreterMacroAssembler::pop(TosState state) {
610 switch (state) {
611 case atos: pop_ptr(); break;
612 case btos:
613 case ztos:
614 case ctos:
615 case stos:
616 case itos: pop_i(); break;
617 case ltos: pop_l(); break;
618 case ftos: pop_f(xmm0); break;
619 case dtos: pop_d(xmm0); break;
620 case vtos: /* nothing to do */ break;
621 default: ShouldNotReachHere();
622 }
623 interp_verify_oop(rax, state);
624 }
625
626 void InterpreterMacroAssembler::push(TosState state) {
627 interp_verify_oop(rax, state);
628 switch (state) {
629 case atos: push_ptr(); break;
630 case btos:
631 case ztos:
632 case ctos:
633 case stos:
634 case itos: push_i(); break;
635 case ltos: push_l(); break;
636 case ftos: push_f(xmm0); break;
637 case dtos: push_d(xmm0); break;
638 case vtos: /* nothing to do */ break;
639 default : ShouldNotReachHere();
640 }
641 }
642 #else
643 void InterpreterMacroAssembler::pop_i(Register r) {
644 pop(r);
645 }
646
647 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
648 pop(lo);
649 pop(hi);
650 }
651
652 void InterpreterMacroAssembler::pop_f() {
653 fld_s(Address(rsp, 0));
654 addptr(rsp, 1 * wordSize);
655 }
656
657 void InterpreterMacroAssembler::pop_d() {
658 fld_d(Address(rsp, 0));
659 addptr(rsp, 2 * wordSize);
660 }
661
662
663 void InterpreterMacroAssembler::pop(TosState state) {
664 switch (state) {
665 case atos: pop_ptr(rax); break;
666 case btos: // fall through
667 case ztos: // fall through
668 case ctos: // fall through
669 case stos: // fall through
670 case itos: pop_i(rax); break;
671 case ltos: pop_l(rax, rdx); break;
672 case ftos:
673 if (UseSSE >= 1) {
674 pop_f(xmm0);
675 } else {
676 pop_f();
677 }
678 break;
679 case dtos:
680 if (UseSSE >= 2) {
681 pop_d(xmm0);
682 } else {
683 pop_d();
684 }
685 break;
686 case vtos: /* nothing to do */ break;
687 default : ShouldNotReachHere();
688 }
689 interp_verify_oop(rax, state);
690 }
691
692
693 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
694 push(hi);
695 push(lo);
696 }
697
698 void InterpreterMacroAssembler::push_f() {
699 // Do not schedule for no AGI! Never write beyond rsp!
700 subptr(rsp, 1 * wordSize);
701 fstp_s(Address(rsp, 0));
702 }
703
704 void InterpreterMacroAssembler::push_d() {
705 // Do not schedule for no AGI! Never write beyond rsp!
706 subptr(rsp, 2 * wordSize);
707 fstp_d(Address(rsp, 0));
708 }
709
710
711 void InterpreterMacroAssembler::push(TosState state) {
712 interp_verify_oop(rax, state);
713 switch (state) {
714 case atos: push_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: push_i(rax); break;
720 case ltos: push_l(rax, rdx); break;
721 case ftos:
722 if (UseSSE >= 1) {
723 push_f(xmm0);
724 } else {
725 push_f();
726 }
727 break;
728 case dtos:
729 if (UseSSE >= 2) {
730 push_d(xmm0);
731 } else {
732 push_d();
733 }
734 break;
735 case vtos: /* nothing to do */ break;
736 default : ShouldNotReachHere();
737 }
738 }
739 #endif // _LP64
740
741
742 // Helpers for swap and dup
743 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
744 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
745 }
746
747 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
748 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
749 }
750
751
752 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
753 // set sender sp
754 lea(_bcp_register, Address(rsp, wordSize));
755 // record last_sp
756 mov(rcx, _bcp_register);
757 subptr(rcx, rbp);
758 sarptr(rcx, LogBytesPerWord);
759 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
760 }
761
762
763 // Jump to from_interpreted entry of a call unless single stepping is possible
764 // in this thread in which case we must call the i2i entry
765 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
766 prepare_to_jump_from_interpreted();
767
768 if (JvmtiExport::can_post_interpreter_events()) {
769 Label run_compiled_code;
770 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
771 // compiled code in threads for which the event is enabled. Check here for
772 // interp_only_mode if these events CAN be enabled.
773 // interp_only is an int, on little endian it is sufficient to test the byte only
774 // Is a cmpl faster?
775 LP64_ONLY(temp = r15_thread;)
776 NOT_LP64(get_thread(temp);)
777 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
778 jccb(Assembler::zero, run_compiled_code);
779 jmp(Address(method, Method::interpreter_entry_offset()));
780 bind(run_compiled_code);
781 }
782
783 jmp(Address(method, Method::from_interpreted_offset()));
784 }
785
786 // The following two routines provide a hook so that an implementation
787 // can schedule the dispatch in two parts. x86 does not do this.
788 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
789 // Nothing x86 specific to be done here
790 }
791
792 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
793 dispatch_next(state, step);
794 }
795
796 void InterpreterMacroAssembler::dispatch_base(TosState state,
797 address* table,
798 bool verifyoop,
799 bool generate_poll) {
800 verify_FPU(1, state);
801 if (VerifyActivationFrameSize) {
802 Label L;
803 mov(rcx, rbp);
804 subptr(rcx, rsp);
805 int32_t min_frame_size =
806 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
807 wordSize;
808 cmpptr(rcx, min_frame_size);
809 jcc(Assembler::greaterEqual, L);
810 stop("broken stack frame");
811 bind(L);
812 }
813 if (verifyoop) {
814 interp_verify_oop(rax, state);
815 }
816
817 address* const safepoint_table = Interpreter::safept_table(state);
818 #ifdef _LP64
819 Label no_safepoint, dispatch;
820 if (table != safepoint_table && generate_poll) {
821 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
822 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
823
824 jccb(Assembler::zero, no_safepoint);
825 lea(rscratch1, ExternalAddress((address)safepoint_table));
826 jmpb(dispatch);
827 }
828
829 bind(no_safepoint);
830 lea(rscratch1, ExternalAddress((address)table));
831 bind(dispatch);
832 jmp(Address(rscratch1, rbx, Address::times_8));
833
834 #else
835 Address index(noreg, rbx, Address::times_ptr);
836 if (table != safepoint_table && generate_poll) {
837 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
838 Label no_safepoint;
839 const Register thread = rcx;
840 get_thread(thread);
841 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
842
843 jccb(Assembler::zero, no_safepoint);
844 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
845 jump(dispatch_addr, noreg);
846 bind(no_safepoint);
847 }
848
849 {
850 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
851 jump(dispatch_addr, noreg);
852 }
853 #endif // _LP64
854 }
855
856 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
857 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
858 }
859
860 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
861 dispatch_base(state, Interpreter::normal_table(state));
862 }
863
864 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
865 dispatch_base(state, Interpreter::normal_table(state), false);
866 }
867
868
869 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
870 // load next bytecode (load before advancing _bcp_register to prevent AGI)
871 load_unsigned_byte(rbx, Address(_bcp_register, step));
872 // advance _bcp_register
873 increment(_bcp_register, step);
874 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
875 }
876
877 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
878 // load current bytecode
879 load_unsigned_byte(rbx, Address(_bcp_register, 0));
880 dispatch_base(state, table);
881 }
882
883 void InterpreterMacroAssembler::narrow(Register result) {
884
885 // Get method->_constMethod->_result_type
886 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
887 movptr(rcx, Address(rcx, Method::const_offset()));
888 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
889
890 Label done, notBool, notByte, notChar;
891
892 // common case first
893 cmpl(rcx, T_INT);
894 jcc(Assembler::equal, done);
895
896 // mask integer result to narrower return type.
897 cmpl(rcx, T_BOOLEAN);
898 jcc(Assembler::notEqual, notBool);
899 andl(result, 0x1);
900 jmp(done);
901
902 bind(notBool);
903 cmpl(rcx, T_BYTE);
904 jcc(Assembler::notEqual, notByte);
905 LP64_ONLY(movsbl(result, result);)
906 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
907 NOT_LP64(sarl(result, 24);) // and sign-extend byte
908 jmp(done);
909
910 bind(notByte);
911 cmpl(rcx, T_CHAR);
912 jcc(Assembler::notEqual, notChar);
913 LP64_ONLY(movzwl(result, result);)
914 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
915 jmp(done);
916
917 bind(notChar);
918 // cmpl(rcx, T_SHORT); // all that's left
919 // jcc(Assembler::notEqual, done);
920 LP64_ONLY(movswl(result, result);)
921 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
922 NOT_LP64(sarl(result, 16);) // and sign-extend short
923
924 // Nothing to do for T_INT
925 bind(done);
926 }
927
928 // remove activation
929 //
930 // Apply stack watermark barrier.
931 // Unlock the receiver if this is a synchronized method.
932 // Unlock any Java monitors from synchronized blocks.
933 // Remove the activation from the stack.
934 //
935 // If there are locked Java monitors
936 // If throw_monitor_exception
937 // throws IllegalMonitorStateException
938 // Else if install_monitor_exception
939 // installs IllegalMonitorStateException
940 // Else
941 // no error processing
942 void InterpreterMacroAssembler::remove_activation(
943 TosState state,
944 Register ret_addr,
945 bool throw_monitor_exception,
946 bool install_monitor_exception,
947 bool notify_jvmdi) {
948 // Note: Registers rdx xmm0 may be in use for the
949 // result check if synchronized method
950 Label unlocked, unlock, no_unlock;
951
952 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
953 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
954 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
955 // monitor pointers need different register
956 // because rdx may have the result in it
957 NOT_LP64(get_thread(rthread);)
958
959 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
960 // that would normally not be safe to use. Such bad returns into unsafe territory of
961 // the stack, will call InterpreterRuntime::at_unwind.
962 Label slow_path;
963 Label fast_path;
964 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
965 jmp(fast_path);
966 bind(slow_path);
967 push(state);
968 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
969 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
970 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
971 reset_last_Java_frame(rthread, true);
972 pop(state);
973 bind(fast_path);
974
975 // get the value of _do_not_unlock_if_synchronized into rdx
976 const Address do_not_unlock_if_synchronized(rthread,
977 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
978 movbool(rbx, do_not_unlock_if_synchronized);
979 movbool(do_not_unlock_if_synchronized, false); // reset the flag
980
981 // get method access flags
982 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
983 movl(rcx, Address(rcx, Method::access_flags_offset()));
984 testl(rcx, JVM_ACC_SYNCHRONIZED);
985 jcc(Assembler::zero, unlocked);
986
987 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
988 // is set.
989 testbool(rbx);
990 jcc(Assembler::notZero, no_unlock);
991
992 // unlock monitor
993 push(state); // save result
994
995 // BasicObjectLock will be first in list, since this is a
996 // synchronized method. However, need to check that the object has
997 // not been unlocked by an explicit monitorexit bytecode.
998 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
999 wordSize - (int) sizeof(BasicObjectLock));
1000 // We use c_rarg1/rdx so that if we go slow path it will be the correct
1001 // register for unlock_object to pass to VM directly
1002 lea(robj, monitor); // address of first monitor
1003
1004 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
1005 testptr(rax, rax);
1006 jcc(Assembler::notZero, unlock);
1007
1008 pop(state);
1009 if (throw_monitor_exception) {
1010 // Entry already unlocked, need to throw exception
1011 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1012 call_VM(noreg, CAST_FROM_FN_PTR(address,
1013 InterpreterRuntime::throw_illegal_monitor_state_exception));
1014 should_not_reach_here();
1015 } else {
1016 // Monitor already unlocked during a stack unroll. If requested,
1017 // install an illegal_monitor_state_exception. Continue with
1018 // stack unrolling.
1019 if (install_monitor_exception) {
1020 NOT_LP64(empty_FPU_stack();)
1021 call_VM(noreg, CAST_FROM_FN_PTR(address,
1022 InterpreterRuntime::new_illegal_monitor_state_exception));
1023 }
1024 jmp(unlocked);
1025 }
1026
1027 bind(unlock);
1028 unlock_object(robj);
1029 pop(state);
1030
1031 // Check that for block-structured locking (i.e., that all locked
1032 // objects has been unlocked)
1033 bind(unlocked);
1034
1035 // rax, rdx: Might contain return value
1036
1037 // Check that all monitors are unlocked
1038 {
1039 Label loop, exception, entry, restart;
1040 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1041 const Address monitor_block_top(
1042 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1043 const Address monitor_block_bot(
1044 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1045
1046 bind(restart);
1047 // We use c_rarg1 so that if we go slow path it will be the correct
1048 // register for unlock_object to pass to VM directly
1049 movptr(rmon, monitor_block_top); // derelativize pointer
1050 lea(rmon, Address(rbp, rmon, Address::times_ptr));
1051 // c_rarg1 points to current entry, starting with top-most entry
1052
1053 lea(rbx, monitor_block_bot); // points to word before bottom of
1054 // monitor block
1055 jmp(entry);
1056
1057 // Entry already locked, need to throw exception
1058 bind(exception);
1059
1060 if (throw_monitor_exception) {
1061 // Throw exception
1062 NOT_LP64(empty_FPU_stack();)
1063 MacroAssembler::call_VM(noreg,
1064 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1065 throw_illegal_monitor_state_exception));
1066 should_not_reach_here();
1067 } else {
1068 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1069 // Unlock does not block, so don't have to worry about the frame.
1070 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1071
1072 push(state);
1073 mov(robj, rmon); // nop if robj and rmon are the same
1074 unlock_object(robj);
1075 pop(state);
1076
1077 if (install_monitor_exception) {
1078 NOT_LP64(empty_FPU_stack();)
1079 call_VM(noreg, CAST_FROM_FN_PTR(address,
1080 InterpreterRuntime::
1081 new_illegal_monitor_state_exception));
1082 }
1083
1084 jmp(restart);
1085 }
1086
1087 bind(loop);
1088 // check if current entry is used
1089 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1090 jcc(Assembler::notEqual, exception);
1091
1092 addptr(rmon, entry_size); // otherwise advance to next entry
1093 bind(entry);
1094 cmpptr(rmon, rbx); // check if bottom reached
1095 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1096 }
1097
1098 bind(no_unlock);
1099
1100 // jvmti support
1101 if (notify_jvmdi) {
1102 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1103 } else {
1104 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1105 }
1106
1107 if (StackReservedPages > 0) {
1108 movptr(rbx,
1109 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1110 // testing if reserved zone needs to be re-enabled
1111 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1112 Label no_reserved_zone_enabling;
1113
1114 NOT_LP64(get_thread(rthread);)
1115
1116 // check if already enabled - if so no re-enabling needed
1117 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1118 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1119 jcc(Assembler::equal, no_reserved_zone_enabling);
1120
1121 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1122 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1123
1124 call_VM_leaf(
1125 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1126 call_VM(noreg, CAST_FROM_FN_PTR(address,
1127 InterpreterRuntime::throw_delayed_StackOverflowError));
1128 should_not_reach_here();
1129
1130 bind(no_reserved_zone_enabling);
1131 }
1132
1133 // remove activation
1134 // get sender sp
1135 movptr(rbx,
1136 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1137
1138 if (state == atos && InlineTypeReturnedAsFields) {
1139 // Check if we are returning an non-null inline type and load its fields into registers
1140 Label skip;
1141 test_oop_is_not_inline_type(rax, rscratch1, skip);
1142
1143 #ifndef _LP64
1144 super_call_VM_leaf(StubRoutines::load_inline_type_fields_in_regs());
1145 #else
1146 // Load fields from a buffered value with an inline class specific handler
1147 load_klass(rdi, rax, rscratch1);
1148 movptr(rdi, Address(rdi, InstanceKlass::adr_inlineklass_fixed_block_offset()));
1149 movptr(rdi, Address(rdi, InlineKlass::unpack_handler_offset()));
1150 // Unpack handler can be null if inline type is not scalarizable in returns
1151 testptr(rdi, rdi);
1152 jcc(Assembler::zero, skip);
1153 call(rdi);
1154 #endif
1155 #ifdef ASSERT
1156 // TODO 8284443 Enable
1157 if (StressCallingConvention && false) {
1158 Label skip_stress;
1159 movptr(rscratch1, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1160 movl(rscratch1, Address(rscratch1, Method::flags_offset()));
1161 testl(rcx, MethodFlags::has_scalarized_return_flag());
1162 jcc(Assembler::zero, skip_stress);
1163 load_klass(rax, rax, rscratch1);
1164 orptr(rax, 1);
1165 bind(skip_stress);
1166 }
1167 #endif
1168 // call above kills the value in rbx. Reload it.
1169 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1170 bind(skip);
1171 }
1172 leave(); // remove frame anchor
1173 pop(ret_addr); // get return address
1174 mov(rsp, rbx); // set sp to sender sp
1175 pop_cont_fastpath();
1176 }
1177
1178 void InterpreterMacroAssembler::get_method_counters(Register method,
1179 Register mcs, Label& skip) {
1180 Label has_counters;
1181 movptr(mcs, Address(method, Method::method_counters_offset()));
1182 testptr(mcs, mcs);
1183 jcc(Assembler::notZero, has_counters);
1184 call_VM(noreg, CAST_FROM_FN_PTR(address,
1185 InterpreterRuntime::build_method_counters), method);
1186 movptr(mcs, Address(method,Method::method_counters_offset()));
1187 testptr(mcs, mcs);
1188 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1189 bind(has_counters);
1190 }
1191
1192 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
1193 Register t1, Register t2,
1194 bool clear_fields, Label& alloc_failed) {
1195 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
1196 {
1197 SkipIfEqual skip_if(this, &DTraceAllocProbes, 0, rscratch1);
1198 // Trigger dtrace event for fastpath
1199 push(atos);
1200 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
1201 pop(atos);
1202 }
1203 }
1204
1205
1206 void InterpreterMacroAssembler::read_flat_field(Register holder_klass,
1207 Register field_index, Register field_offset,
1208 Register obj) {
1209 Label alloc_failed, empty_value, done;
1210 const Register src = field_offset;
1211 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1212 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1213 assert_different_registers(obj, holder_klass, field_index, field_offset, dst_temp);
1214
1215 // Grap the inline field klass
1216 push(holder_klass);
1217 const Register field_klass = holder_klass;
1218 get_inline_type_field_klass(holder_klass, field_index, field_klass);
1219
1220 //check for empty value klass
1221 test_klass_is_empty_inline_type(field_klass, dst_temp, empty_value);
1222
1223 // allocate buffer
1224 push(obj); // save holder
1225 allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1226
1227 // Have an oop instance buffer, copy into it
1228 data_for_oop(obj, dst_temp, field_klass);
1229 pop(alloc_temp); // restore holder
1230 lea(src, Address(alloc_temp, field_offset));
1231 // call_VM_leaf, clobbers a few regs, save restore new obj
1232 push(obj);
1233 access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, field_klass);
1234 pop(obj);
1235 pop(holder_klass);
1236 jmp(done);
1237
1238 bind(empty_value);
1239 get_empty_inline_type_oop(field_klass, dst_temp, obj);
1240 pop(holder_klass);
1241 jmp(done);
1242
1243 bind(alloc_failed);
1244 pop(obj);
1245 pop(holder_klass);
1246 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
1247 obj, field_index, holder_klass);
1248
1249 bind(done);
1250 }
1251
1252 void InterpreterMacroAssembler::read_flat_element(Register array, Register index,
1253 Register t1, Register t2,
1254 Register obj) {
1255 assert_different_registers(array, index, t1, t2);
1256 Label alloc_failed, empty_value, done;
1257 const Register array_klass = t2;
1258 const Register elem_klass = t1;
1259 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1260 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1261
1262 // load in array->klass()->element_klass()
1263 Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1264 load_klass(array_klass, array, tmp_load_klass);
1265 movptr(elem_klass, Address(array_klass, ArrayKlass::element_klass_offset()));
1266
1267 //check for empty value klass
1268 test_klass_is_empty_inline_type(elem_klass, dst_temp, empty_value);
1269
1270 // calc source into "array_klass" and free up some regs
1271 const Register src = array_klass;
1272 push(index); // preserve index reg in case alloc_failed
1273 data_for_value_array_index(array, array_klass, index, src);
1274
1275 allocate_instance(elem_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1276 // Have an oop instance buffer, copy into it
1277 store_ptr(0, obj); // preserve obj (overwrite index, no longer needed)
1278 data_for_oop(obj, dst_temp, elem_klass);
1279 access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, elem_klass);
1280 pop(obj);
1281 jmp(done);
1282
1283 bind(empty_value);
1284 get_empty_inline_type_oop(elem_klass, dst_temp, obj);
1285 jmp(done);
1286
1287 bind(alloc_failed);
1288 pop(index);
1289 if (array == c_rarg2) {
1290 mov(elem_klass, array);
1291 array = elem_klass;
1292 }
1293 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), array, index);
1294
1295 bind(done);
1296 }
1297
1298
1299 // Lock object
1300 //
1301 // Args:
1302 // rdx, c_rarg1: BasicObjectLock to be used for locking
1303 //
1304 // Kills:
1305 // rax, rbx
1306 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1307 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1308 "The argument is only for looks. It must be c_rarg1");
1309
1310 if (LockingMode == LM_MONITOR) {
1311 call_VM(noreg,
1312 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1313 lock_reg);
1314 } else {
1315 Label count_locking, done, slow_case;
1316
1317 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1318 const Register tmp_reg = rbx;
1319 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1320 const Register rklass_decode_tmp = rscratch1;
1321
1322 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1323 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1324 const int mark_offset = lock_offset +
1325 BasicLock::displaced_header_offset_in_bytes();
1326
1327 // Load object pointer into obj_reg
1328 movptr(obj_reg, Address(lock_reg, obj_offset));
1329
1330 if (DiagnoseSyncOnValueBasedClasses != 0) {
1331 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1332 movl(tmp_reg, Address(tmp_reg, Klass::access_flags_offset()));
1333 testl(tmp_reg, JVM_ACC_IS_VALUE_BASED_CLASS);
1334 jcc(Assembler::notZero, slow_case);
1335 }
1336
1337 if (LockingMode == LM_LIGHTWEIGHT) {
1338 #ifdef _LP64
1339 const Register thread = r15_thread;
1340 #else
1341 const Register thread = lock_reg;
1342 get_thread(thread);
1343 #endif
1344 lightweight_lock(obj_reg, swap_reg, thread, tmp_reg, slow_case);
1345 } else if (LockingMode == LM_LEGACY) {
1346 // Load immediate 1 into swap_reg %rax
1347 movl(swap_reg, 1);
1348
1349 // Load (object->mark() | 1) into swap_reg %rax
1350 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1351 if (EnableValhalla) {
1352 // Mask inline_type bit such that we go to the slow path if object is an inline type
1353 andptr(swap_reg, ~((int) markWord::inline_type_bit_in_place));
1354 }
1355
1356 // Save (object->mark() | 1) into BasicLock's displaced header
1357 movptr(Address(lock_reg, mark_offset), swap_reg);
1358
1359 assert(lock_offset == 0,
1360 "displaced header must be first word in BasicObjectLock");
1361
1362 lock();
1363 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1364 jcc(Assembler::zero, count_locking);
1365
1366 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1367
1368 // Fast check for recursive lock.
1369 //
1370 // Can apply the optimization only if this is a stack lock
1371 // allocated in this thread. For efficiency, we can focus on
1372 // recently allocated stack locks (instead of reading the stack
1373 // base and checking whether 'mark' points inside the current
1374 // thread stack):
1375 // 1) (mark & zero_bits) == 0, and
1376 // 2) rsp <= mark < mark + os::pagesize()
1377 //
1378 // Warning: rsp + os::pagesize can overflow the stack base. We must
1379 // neither apply the optimization for an inflated lock allocated
1380 // just above the thread stack (this is why condition 1 matters)
1381 // nor apply the optimization if the stack lock is inside the stack
1382 // of another thread. The latter is avoided even in case of overflow
1383 // because we have guard pages at the end of all stacks. Hence, if
1384 // we go over the stack base and hit the stack of another thread,
1385 // this should not be in a writeable area that could contain a
1386 // stack lock allocated by that thread. As a consequence, a stack
1387 // lock less than page size away from rsp is guaranteed to be
1388 // owned by the current thread.
1389 //
1390 // These 3 tests can be done by evaluating the following
1391 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1392 // assuming both stack pointer and pagesize have their
1393 // least significant bits clear.
1394 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1395 subptr(swap_reg, rsp);
1396 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1397
1398 // Save the test result, for recursive case, the result is zero
1399 movptr(Address(lock_reg, mark_offset), swap_reg);
1400 jcc(Assembler::notZero, slow_case);
1401
1402 bind(count_locking);
1403 }
1404 inc_held_monitor_count();
1405 jmp(done);
1406
1407 bind(slow_case);
1408
1409 // Call the runtime routine for slow case
1410 if (LockingMode == LM_LIGHTWEIGHT) {
1411 call_VM(noreg,
1412 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
1413 obj_reg);
1414 } else {
1415 call_VM(noreg,
1416 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1417 lock_reg);
1418 }
1419 bind(done);
1420 }
1421 }
1422
1423
1424 // Unlocks an object. Used in monitorexit bytecode and
1425 // remove_activation. Throws an IllegalMonitorException if object is
1426 // not locked by current thread.
1427 //
1428 // Args:
1429 // rdx, c_rarg1: BasicObjectLock for lock
1430 //
1431 // Kills:
1432 // rax
1433 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1434 // rscratch1 (scratch reg)
1435 // rax, rbx, rcx, rdx
1436 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1437 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1438 "The argument is only for looks. It must be c_rarg1");
1439
1440 if (LockingMode == LM_MONITOR) {
1441 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1442 } else {
1443 Label count_locking, done, slow_case;
1444
1445 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1446 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1447 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1448
1449 save_bcp(); // Save in case of exception
1450
1451 if (LockingMode != LM_LIGHTWEIGHT) {
1452 // Convert from BasicObjectLock structure to object and BasicLock
1453 // structure Store the BasicLock address into %rax
1454 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1455 }
1456
1457 // Load oop into obj_reg(%c_rarg3)
1458 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1459
1460 // Free entry
1461 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1462
1463 if (LockingMode == LM_LIGHTWEIGHT) {
1464 #ifdef _LP64
1465 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1466 #else
1467 // This relies on the implementation of lightweight_unlock being able to handle
1468 // that the reg_rax and thread Register parameters may alias each other.
1469 get_thread(swap_reg);
1470 lightweight_unlock(obj_reg, swap_reg, swap_reg, header_reg, slow_case);
1471 #endif
1472 } else if (LockingMode == LM_LEGACY) {
1473 // Load the old header from BasicLock structure
1474 movptr(header_reg, Address(swap_reg,
1475 BasicLock::displaced_header_offset_in_bytes()));
1476
1477 // Test for recursion
1478 testptr(header_reg, header_reg);
1479
1480 // zero for recursive case
1481 jcc(Assembler::zero, count_locking);
1482
1483 // Atomic swap back the old header
1484 lock();
1485 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1486
1487 // zero for simple unlock of a stack-lock case
1488 jcc(Assembler::notZero, slow_case);
1489
1490 bind(count_locking);
1491 }
1492 dec_held_monitor_count();
1493 jmp(done);
1494
1495 bind(slow_case);
1496 // Call the runtime routine for slow case.
1497 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1498 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1499
1500 bind(done);
1501
1502 restore_bcp();
1503 }
1504 }
1505
1506 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1507 Label& zero_continue) {
1508 assert(ProfileInterpreter, "must be profiling interpreter");
1509 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1510 testptr(mdp, mdp);
1511 jcc(Assembler::zero, zero_continue);
1512 }
1513
1514
1515 // Set the method data pointer for the current bcp.
1516 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1517 assert(ProfileInterpreter, "must be profiling interpreter");
1518 Label set_mdp;
1519 push(rax);
1520 push(rbx);
1521
1522 get_method(rbx);
1523 // Test MDO to avoid the call if it is null.
1524 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1525 testptr(rax, rax);
1526 jcc(Assembler::zero, set_mdp);
1527 // rbx: method
1528 // _bcp_register: bcp
1529 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1530 // rax: mdi
1531 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1532 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1533 addptr(rbx, in_bytes(MethodData::data_offset()));
1534 addptr(rax, rbx);
1535 bind(set_mdp);
1536 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1537 pop(rbx);
1538 pop(rax);
1539 }
1540
1541 void InterpreterMacroAssembler::verify_method_data_pointer() {
1542 assert(ProfileInterpreter, "must be profiling interpreter");
1543 #ifdef ASSERT
1544 Label verify_continue;
1545 push(rax);
1546 push(rbx);
1547 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1548 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1549 push(arg3_reg);
1550 push(arg2_reg);
1551 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1552 get_method(rbx);
1553
1554 // If the mdp is valid, it will point to a DataLayout header which is
1555 // consistent with the bcp. The converse is highly probable also.
1556 load_unsigned_short(arg2_reg,
1557 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1558 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1559 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1560 cmpptr(arg2_reg, _bcp_register);
1561 jcc(Assembler::equal, verify_continue);
1562 // rbx: method
1563 // _bcp_register: bcp
1564 // c_rarg3: mdp
1565 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1566 rbx, _bcp_register, arg3_reg);
1567 bind(verify_continue);
1568 pop(arg2_reg);
1569 pop(arg3_reg);
1570 pop(rbx);
1571 pop(rax);
1572 #endif // ASSERT
1573 }
1574
1575
1576 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1577 int constant,
1578 Register value) {
1579 assert(ProfileInterpreter, "must be profiling interpreter");
1580 Address data(mdp_in, constant);
1581 movptr(data, value);
1582 }
1583
1584
1585 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1586 int constant,
1587 bool decrement) {
1588 // Counter address
1589 Address data(mdp_in, constant);
1590
1591 increment_mdp_data_at(data, decrement);
1592 }
1593
1594 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1595 bool decrement) {
1596 assert(ProfileInterpreter, "must be profiling interpreter");
1597 // %%% this does 64bit counters at best it is wasting space
1598 // at worst it is a rare bug when counters overflow
1599
1600 if (decrement) {
1601 // Decrement the register. Set condition codes.
1602 addptr(data, -DataLayout::counter_increment);
1603 // If the decrement causes the counter to overflow, stay negative
1604 Label L;
1605 jcc(Assembler::negative, L);
1606 addptr(data, DataLayout::counter_increment);
1607 bind(L);
1608 } else {
1609 assert(DataLayout::counter_increment == 1,
1610 "flow-free idiom only works with 1");
1611 // Increment the register. Set carry flag.
1612 addptr(data, DataLayout::counter_increment);
1613 // If the increment causes the counter to overflow, pull back by 1.
1614 sbbptr(data, 0);
1615 }
1616 }
1617
1618
1619 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1620 Register reg,
1621 int constant,
1622 bool decrement) {
1623 Address data(mdp_in, reg, Address::times_1, constant);
1624
1625 increment_mdp_data_at(data, decrement);
1626 }
1627
1628 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1629 int flag_byte_constant) {
1630 assert(ProfileInterpreter, "must be profiling interpreter");
1631 int header_offset = in_bytes(DataLayout::flags_offset());
1632 int header_bits = flag_byte_constant;
1633 // Set the flag
1634 orb(Address(mdp_in, header_offset), header_bits);
1635 }
1636
1637
1638
1639 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1640 int offset,
1641 Register value,
1642 Register test_value_out,
1643 Label& not_equal_continue) {
1644 assert(ProfileInterpreter, "must be profiling interpreter");
1645 if (test_value_out == noreg) {
1646 cmpptr(value, Address(mdp_in, offset));
1647 } else {
1648 // Put the test value into a register, so caller can use it:
1649 movptr(test_value_out, Address(mdp_in, offset));
1650 cmpptr(test_value_out, value);
1651 }
1652 jcc(Assembler::notEqual, not_equal_continue);
1653 }
1654
1655
1656 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1657 int offset_of_disp) {
1658 assert(ProfileInterpreter, "must be profiling interpreter");
1659 Address disp_address(mdp_in, offset_of_disp);
1660 addptr(mdp_in, disp_address);
1661 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1662 }
1663
1664
1665 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1666 Register reg,
1667 int offset_of_disp) {
1668 assert(ProfileInterpreter, "must be profiling interpreter");
1669 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1670 addptr(mdp_in, disp_address);
1671 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1672 }
1673
1674
1675 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1676 int constant) {
1677 assert(ProfileInterpreter, "must be profiling interpreter");
1678 addptr(mdp_in, constant);
1679 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1680 }
1681
1682
1683 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1684 assert(ProfileInterpreter, "must be profiling interpreter");
1685 push(return_bci); // save/restore across call_VM
1686 call_VM(noreg,
1687 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1688 return_bci);
1689 pop(return_bci);
1690 }
1691
1692
1693 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1694 Register bumped_count) {
1695 if (ProfileInterpreter) {
1696 Label profile_continue;
1697
1698 // If no method data exists, go to profile_continue.
1699 // Otherwise, assign to mdp
1700 test_method_data_pointer(mdp, profile_continue);
1701
1702 // We are taking a branch. Increment the taken count.
1703 // We inline increment_mdp_data_at to return bumped_count in a register
1704 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1705 Address data(mdp, in_bytes(JumpData::taken_offset()));
1706 movptr(bumped_count, data);
1707 assert(DataLayout::counter_increment == 1,
1708 "flow-free idiom only works with 1");
1709 addptr(bumped_count, DataLayout::counter_increment);
1710 sbbptr(bumped_count, 0);
1711 movptr(data, bumped_count); // Store back out
1712
1713 // The method data pointer needs to be updated to reflect the new target.
1714 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1715 bind(profile_continue);
1716 }
1717 }
1718
1719
1720 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1721 if (ProfileInterpreter) {
1722 Label profile_continue;
1723
1724 // If no method data exists, go to profile_continue.
1725 test_method_data_pointer(mdp, profile_continue);
1726
1727 // We are taking a branch. Increment the not taken count.
1728 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1729
1730 // The method data pointer needs to be updated to correspond to
1731 // the next bytecode
1732 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()): in_bytes(BranchData::branch_data_size()));
1733 bind(profile_continue);
1734 }
1735 }
1736
1737 void InterpreterMacroAssembler::profile_call(Register mdp) {
1738 if (ProfileInterpreter) {
1739 Label profile_continue;
1740
1741 // If no method data exists, go to profile_continue.
1742 test_method_data_pointer(mdp, profile_continue);
1743
1744 // We are making a call. Increment the count.
1745 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1746
1747 // The method data pointer needs to be updated to reflect the new target.
1748 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1749 bind(profile_continue);
1750 }
1751 }
1752
1753
1754 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1755 if (ProfileInterpreter) {
1756 Label profile_continue;
1757
1758 // If no method data exists, go to profile_continue.
1759 test_method_data_pointer(mdp, profile_continue);
1760
1761 // We are making a call. Increment the count.
1762 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1763
1764 // The method data pointer needs to be updated to reflect the new target.
1765 update_mdp_by_constant(mdp,
1766 in_bytes(VirtualCallData::
1767 virtual_call_data_size()));
1768 bind(profile_continue);
1769 }
1770 }
1771
1772
1773 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1774 Register mdp,
1775 Register reg2,
1776 bool receiver_can_be_null) {
1777 if (ProfileInterpreter) {
1778 Label profile_continue;
1779
1780 // If no method data exists, go to profile_continue.
1781 test_method_data_pointer(mdp, profile_continue);
1782
1783 Label skip_receiver_profile;
1784 if (receiver_can_be_null) {
1785 Label not_null;
1786 testptr(receiver, receiver);
1787 jccb(Assembler::notZero, not_null);
1788 // We are making a call. Increment the count for null receiver.
1789 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1790 jmp(skip_receiver_profile);
1791 bind(not_null);
1792 }
1793
1794 // Record the receiver type.
1795 record_klass_in_profile(receiver, mdp, reg2);
1796 bind(skip_receiver_profile);
1797
1798 // The method data pointer needs to be updated to reflect the new target.
1799 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1800 bind(profile_continue);
1801 }
1802 }
1803
1804 // This routine creates a state machine for updating the multi-row
1805 // type profile at a virtual call site (or other type-sensitive bytecode).
1806 // The machine visits each row (of receiver/count) until the receiver type
1807 // is found, or until it runs out of rows. At the same time, it remembers
1808 // the location of the first empty row. (An empty row records null for its
1809 // receiver, and can be allocated for a newly-observed receiver type.)
1810 // Because there are two degrees of freedom in the state, a simple linear
1811 // search will not work; it must be a decision tree. Hence this helper
1812 // function is recursive, to generate the required tree structured code.
1813 // It's the interpreter, so we are trading off code space for speed.
1814 // See below for example code.
1815 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register mdp,
1816 Register reg2, int start_row,
1817 Label& done) {
1818 if (TypeProfileWidth == 0) {
1819 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1820 } else {
1821 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1822 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1823 }
1824 }
1825
1826 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1827 Label& done, int total_rows,
1828 OffsetFunction item_offset_fn,
1829 OffsetFunction item_count_offset_fn) {
1830 int last_row = total_rows - 1;
1831 assert(start_row <= last_row, "must be work left to do");
1832 // Test this row for both the item and for null.
1833 // Take any of three different outcomes:
1834 // 1. found item => increment count and goto done
1835 // 2. found null => keep looking for case 1, maybe allocate this cell
1836 // 3. found something else => keep looking for cases 1 and 2
1837 // Case 3 is handled by a recursive call.
1838 for (int row = start_row; row <= last_row; row++) {
1839 Label next_test;
1840 bool test_for_null_also = (row == start_row);
1841
1842 // See if the item is item[n].
1843 int item_offset = in_bytes(item_offset_fn(row));
1844 test_mdp_data_at(mdp, item_offset, item,
1845 (test_for_null_also ? reg2 : noreg),
1846 next_test);
1847 // (Reg2 now contains the item from the CallData.)
1848
1849 // The item is item[n]. Increment count[n].
1850 int count_offset = in_bytes(item_count_offset_fn(row));
1851 increment_mdp_data_at(mdp, count_offset);
1852 jmp(done);
1853 bind(next_test);
1854
1855 if (test_for_null_also) {
1856 // Failed the equality check on item[n]... Test for null.
1857 testptr(reg2, reg2);
1858 if (start_row == last_row) {
1859 // The only thing left to do is handle the null case.
1860 Label found_null;
1861 jccb(Assembler::zero, found_null);
1862 // Item did not match any saved item and there is no empty row for it.
1863 // Increment total counter to indicate polymorphic case.
1864 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1865 jmp(done);
1866 bind(found_null);
1867 break;
1868 }
1869 Label found_null;
1870 // Since null is rare, make it be the branch-taken case.
1871 jcc(Assembler::zero, found_null);
1872
1873 // Put all the "Case 3" tests here.
1874 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1875 item_offset_fn, item_count_offset_fn);
1876
1877 // Found a null. Keep searching for a matching item,
1878 // but remember that this is an empty (unused) slot.
1879 bind(found_null);
1880 }
1881 }
1882
1883 // In the fall-through case, we found no matching item, but we
1884 // observed the item[start_row] is null.
1885
1886 // Fill in the item field and increment the count.
1887 int item_offset = in_bytes(item_offset_fn(start_row));
1888 set_mdp_data_at(mdp, item_offset, item);
1889 int count_offset = in_bytes(item_count_offset_fn(start_row));
1890 movl(reg2, DataLayout::counter_increment);
1891 set_mdp_data_at(mdp, count_offset, reg2);
1892 if (start_row > 0) {
1893 jmp(done);
1894 }
1895 }
1896
1897 // Example state machine code for three profile rows:
1898 // // main copy of decision tree, rooted at row[1]
1899 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1900 // if (row[0].rec != nullptr) {
1901 // // inner copy of decision tree, rooted at row[1]
1902 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1903 // if (row[1].rec != nullptr) {
1904 // // degenerate decision tree, rooted at row[2]
1905 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1906 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1907 // row[2].init(rec); goto done;
1908 // } else {
1909 // // remember row[1] is empty
1910 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1911 // row[1].init(rec); goto done;
1912 // }
1913 // } else {
1914 // // remember row[0] is empty
1915 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1916 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1917 // row[0].init(rec); goto done;
1918 // }
1919 // done:
1920
1921 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, Register mdp, Register reg2) {
1922 assert(ProfileInterpreter, "must be profiling");
1923 Label done;
1924
1925 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1926
1927 bind (done);
1928 }
1929
1930 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1931 Register mdp) {
1932 if (ProfileInterpreter) {
1933 Label profile_continue;
1934 uint row;
1935
1936 // If no method data exists, go to profile_continue.
1937 test_method_data_pointer(mdp, profile_continue);
1938
1939 // Update the total ret count.
1940 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1941
1942 for (row = 0; row < RetData::row_limit(); row++) {
1943 Label next_test;
1944
1945 // See if return_bci is equal to bci[n]:
1946 test_mdp_data_at(mdp,
1947 in_bytes(RetData::bci_offset(row)),
1948 return_bci, noreg,
1949 next_test);
1950
1951 // return_bci is equal to bci[n]. Increment the count.
1952 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1953
1954 // The method data pointer needs to be updated to reflect the new target.
1955 update_mdp_by_offset(mdp,
1956 in_bytes(RetData::bci_displacement_offset(row)));
1957 jmp(profile_continue);
1958 bind(next_test);
1959 }
1960
1961 update_mdp_for_ret(return_bci);
1962
1963 bind(profile_continue);
1964 }
1965 }
1966
1967
1968 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1969 if (ProfileInterpreter) {
1970 Label profile_continue;
1971
1972 // If no method data exists, go to profile_continue.
1973 test_method_data_pointer(mdp, profile_continue);
1974
1975 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1976
1977 // The method data pointer needs to be updated.
1978 int mdp_delta = in_bytes(BitData::bit_data_size());
1979 if (TypeProfileCasts) {
1980 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1981 }
1982 update_mdp_by_constant(mdp, mdp_delta);
1983
1984 bind(profile_continue);
1985 }
1986 }
1987
1988
1989 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1990 if (ProfileInterpreter) {
1991 Label profile_continue;
1992
1993 // If no method data exists, go to profile_continue.
1994 test_method_data_pointer(mdp, profile_continue);
1995
1996 // The method data pointer needs to be updated.
1997 int mdp_delta = in_bytes(BitData::bit_data_size());
1998 if (TypeProfileCasts) {
1999 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
2000
2001 // Record the object type.
2002 record_klass_in_profile(klass, mdp, reg2);
2003 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
2004 NOT_LP64(restore_locals();) // Restore EDI
2005 }
2006 update_mdp_by_constant(mdp, mdp_delta);
2007
2008 bind(profile_continue);
2009 }
2010 }
2011
2012
2013 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
2014 if (ProfileInterpreter) {
2015 Label profile_continue;
2016
2017 // If no method data exists, go to profile_continue.
2018 test_method_data_pointer(mdp, profile_continue);
2019
2020 // Update the default case count
2021 increment_mdp_data_at(mdp,
2022 in_bytes(MultiBranchData::default_count_offset()));
2023
2024 // The method data pointer needs to be updated.
2025 update_mdp_by_offset(mdp,
2026 in_bytes(MultiBranchData::
2027 default_displacement_offset()));
2028
2029 bind(profile_continue);
2030 }
2031 }
2032
2033
2034 void InterpreterMacroAssembler::profile_switch_case(Register index,
2035 Register mdp,
2036 Register reg2) {
2037 if (ProfileInterpreter) {
2038 Label profile_continue;
2039
2040 // If no method data exists, go to profile_continue.
2041 test_method_data_pointer(mdp, profile_continue);
2042
2043 // Build the base (index * per_case_size_in_bytes()) +
2044 // case_array_offset_in_bytes()
2045 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
2046 imulptr(index, reg2); // XXX l ?
2047 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
2048
2049 // Update the case count
2050 increment_mdp_data_at(mdp,
2051 index,
2052 in_bytes(MultiBranchData::relative_count_offset()));
2053
2054 // The method data pointer needs to be updated.
2055 update_mdp_by_offset(mdp,
2056 index,
2057 in_bytes(MultiBranchData::
2058 relative_displacement_offset()));
2059
2060 bind(profile_continue);
2061 }
2062 }
2063
2064 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
2065 Register array,
2066 Register tmp) {
2067 if (ProfileInterpreter) {
2068 Label profile_continue;
2069
2070 // If no method data exists, go to profile_continue.
2071 test_method_data_pointer(mdp, profile_continue);
2072
2073 mov(tmp, array);
2074 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
2075
2076 Label not_flat;
2077 test_non_flat_array_oop(array, tmp, not_flat);
2078
2079 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
2080
2081 bind(not_flat);
2082
2083 Label not_null_free;
2084 test_non_null_free_array_oop(array, tmp, not_null_free);
2085
2086 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
2087
2088 bind(not_null_free);
2089
2090 bind(profile_continue);
2091 }
2092 }
2093
2094 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
2095 Register array,
2096 Register tmp);
2097 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
2098 Register array,
2099 Register tmp);
2100
2101
2102 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
2103 if (ProfileInterpreter) {
2104 Label profile_continue;
2105
2106 // If no method data exists, go to profile_continue.
2107 test_method_data_pointer(mdp, profile_continue);
2108
2109 Label done, update;
2110 testptr(element, element);
2111 jccb(Assembler::notZero, update);
2112 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
2113 jmp(done);
2114
2115 bind(update);
2116 load_klass(tmp, element, rscratch1);
2117
2118 // Record the object type.
2119 record_klass_in_profile(tmp, mdp, tmp2);
2120
2121 bind(done);
2122
2123 // The method data pointer needs to be updated.
2124 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
2125
2126 bind(profile_continue);
2127 }
2128 }
2129
2130 void InterpreterMacroAssembler::profile_element_type(Register mdp,
2131 Register element,
2132 Register tmp) {
2133 if (ProfileInterpreter) {
2134 Label profile_continue;
2135
2136 // If no method data exists, go to profile_continue.
2137 test_method_data_pointer(mdp, profile_continue);
2138
2139 mov(tmp, element);
2140 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
2141
2142 // The method data pointer needs to be updated.
2143 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
2144
2145 bind(profile_continue);
2146 }
2147 }
2148
2149 void InterpreterMacroAssembler::profile_acmp(Register mdp,
2150 Register left,
2151 Register right,
2152 Register tmp) {
2153 if (ProfileInterpreter) {
2154 Label profile_continue;
2155
2156 // If no method data exists, go to profile_continue.
2157 test_method_data_pointer(mdp, profile_continue);
2158
2159 mov(tmp, left);
2160 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
2161
2162 Label left_not_inline_type;
2163 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
2164 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
2165 bind(left_not_inline_type);
2166
2167 mov(tmp, right);
2168 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
2169
2170 Label right_not_inline_type;
2171 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
2172 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
2173 bind(right_not_inline_type);
2174
2175 bind(profile_continue);
2176 }
2177 }
2178
2179
2180 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
2181 if (state == atos) {
2182 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
2183 }
2184 }
2185
2186 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2187 #ifndef _LP64
2188 if ((state == ftos && UseSSE < 1) ||
2189 (state == dtos && UseSSE < 2)) {
2190 MacroAssembler::verify_FPU(stack_depth);
2191 }
2192 #endif
2193 }
2194
2195 // Jump if ((*counter_addr += increment) & mask) == 0
2196 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
2197 Register scratch, Label* where) {
2198 // This update is actually not atomic and can lose a number of updates
2199 // under heavy contention, but the alternative of using the (contended)
2200 // atomic update here penalizes profiling paths too much.
2201 movl(scratch, counter_addr);
2202 incrementl(scratch, InvocationCounter::count_increment);
2203 movl(counter_addr, scratch);
2204 andl(scratch, mask);
2205 if (where != nullptr) {
2206 jcc(Assembler::zero, *where);
2207 }
2208 }
2209
2210 void InterpreterMacroAssembler::notify_method_entry() {
2211 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2212 // track stack depth. If it is possible to enter interp_only_mode we add
2213 // the code to check if the event should be sent.
2214 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2215 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2216 if (JvmtiExport::can_post_interpreter_events()) {
2217 Label L;
2218 NOT_LP64(get_thread(rthread);)
2219 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2220 testl(rdx, rdx);
2221 jcc(Assembler::zero, L);
2222 call_VM(noreg, CAST_FROM_FN_PTR(address,
2223 InterpreterRuntime::post_method_entry));
2224 bind(L);
2225 }
2226
2227 {
2228 SkipIfEqual skip(this, &DTraceMethodProbes, false, rscratch1);
2229 NOT_LP64(get_thread(rthread);)
2230 get_method(rarg);
2231 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2232 rthread, rarg);
2233 }
2234
2235 // RedefineClasses() tracing support for obsolete method entry
2236 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2237 NOT_LP64(get_thread(rthread);)
2238 get_method(rarg);
2239 call_VM_leaf(
2240 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2241 rthread, rarg);
2242 }
2243 }
2244
2245
2246 void InterpreterMacroAssembler::notify_method_exit(
2247 TosState state, NotifyMethodExitMode mode) {
2248 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2249 // track stack depth. If it is possible to enter interp_only_mode we add
2250 // the code to check if the event should be sent.
2251 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2252 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2253 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2254 Label L;
2255 // Note: frame::interpreter_frame_result has a dependency on how the
2256 // method result is saved across the call to post_method_exit. If this
2257 // is changed then the interpreter_frame_result implementation will
2258 // need to be updated too.
2259
2260 // template interpreter will leave the result on the top of the stack.
2261 push(state);
2262 NOT_LP64(get_thread(rthread);)
2263 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2264 testl(rdx, rdx);
2265 jcc(Assembler::zero, L);
2266 call_VM(noreg,
2267 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2268 bind(L);
2269 pop(state);
2270 }
2271
2272 {
2273 SkipIfEqual skip(this, &DTraceMethodProbes, false, rscratch1);
2274 push(state);
2275 NOT_LP64(get_thread(rthread);)
2276 get_method(rarg);
2277 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2278 rthread, rarg);
2279 pop(state);
2280 }
2281 }
2282
2283 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2284 // Get index out of bytecode pointer
2285 get_cache_index_at_bcp(index, 1, sizeof(u4));
2286 // Get address of invokedynamic array
2287 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2288 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2289 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2290 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2291 } else {
2292 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2293 }
2294 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2295 }
2296
2297 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2298 // Get index out of bytecode pointer
2299 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2300 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2301
2302 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2303 // Take shortcut if the size is a power of 2
2304 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2305 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2306 } else {
2307 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2308 }
2309 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2310 }
2311
2312 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2313 // Get index out of bytecode pointer
2314 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2315 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2316
2317 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2318 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2319 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2320 }