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 } else if (index_size == sizeof(u1)) {
469 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
470 } else {
471 ShouldNotReachHere();
472 }
473 }
474
475 // Load object from cpool->resolved_references(index)
476 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result,
477 Register index,
478 Register tmp) {
479 assert_different_registers(result, index);
480
481 get_constant_pool(result);
482 // load pointer for resolved_references[] objArray
483 movptr(result, Address(result, ConstantPool::cache_offset()));
484 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
485 resolve_oop_handle(result, tmp);
486 load_heap_oop(result, Address(result, index,
487 UseCompressedOops ? Address::times_4 : Address::times_ptr,
488 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp);
489 }
490
491 // load cpool->resolved_klass_at(index)
492 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass,
493 Register cpool,
494 Register index) {
495 assert_different_registers(cpool, index);
496
497 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool)));
498 Register resolved_klasses = cpool;
499 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset()));
500 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes()));
501 }
502
503 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
504 // subtype of super_klass.
505 //
506 // Args:
507 // rax: superklass
508 // Rsub_klass: subklass
509 //
510 // Kills:
511 // rcx, rdi
512 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
513 Label& ok_is_subtype,
514 bool profile) {
515 assert(Rsub_klass != rax, "rax holds superklass");
516 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
517 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
518 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
519 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
520
521 // Profile the not-null value's klass.
522 if (profile) {
523 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
524 }
525
526 // Do the check.
527 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
528 }
529
530
531 #ifndef _LP64
532 void InterpreterMacroAssembler::f2ieee() {
533 if (IEEEPrecision) {
534 fstp_s(Address(rsp, 0));
535 fld_s(Address(rsp, 0));
536 }
537 }
538
539
540 void InterpreterMacroAssembler::d2ieee() {
541 if (IEEEPrecision) {
542 fstp_d(Address(rsp, 0));
543 fld_d(Address(rsp, 0));
544 }
545 }
546 #endif // _LP64
547
548 // Java Expression Stack
549
550 void InterpreterMacroAssembler::pop_ptr(Register r) {
551 pop(r);
552 }
553
554 void InterpreterMacroAssembler::push_ptr(Register r) {
555 push(r);
556 }
557
558 void InterpreterMacroAssembler::push_i(Register r) {
559 push(r);
560 }
561
562 void InterpreterMacroAssembler::push_i_or_ptr(Register r) {
563 push(r);
564 }
565
566 void InterpreterMacroAssembler::push_f(XMMRegister r) {
567 subptr(rsp, wordSize);
568 movflt(Address(rsp, 0), r);
569 }
570
571 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
572 movflt(r, Address(rsp, 0));
573 addptr(rsp, wordSize);
574 }
575
576 void InterpreterMacroAssembler::push_d(XMMRegister r) {
577 subptr(rsp, 2 * wordSize);
578 movdbl(Address(rsp, 0), r);
579 }
580
581 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
582 movdbl(r, Address(rsp, 0));
583 addptr(rsp, 2 * Interpreter::stackElementSize);
584 }
585
586 #ifdef _LP64
587 void InterpreterMacroAssembler::pop_i(Register r) {
588 // XXX can't use pop currently, upper half non clean
589 movl(r, Address(rsp, 0));
590 addptr(rsp, wordSize);
591 }
592
593 void InterpreterMacroAssembler::pop_l(Register r) {
594 movq(r, Address(rsp, 0));
595 addptr(rsp, 2 * Interpreter::stackElementSize);
596 }
597
598 void InterpreterMacroAssembler::push_l(Register r) {
599 subptr(rsp, 2 * wordSize);
600 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r );
601 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD );
602 }
603
604 void InterpreterMacroAssembler::pop(TosState state) {
605 switch (state) {
606 case atos: pop_ptr(); break;
607 case btos:
608 case ztos:
609 case ctos:
610 case stos:
611 case itos: pop_i(); break;
612 case ltos: pop_l(); break;
613 case ftos: pop_f(xmm0); break;
614 case dtos: pop_d(xmm0); break;
615 case vtos: /* nothing to do */ break;
616 default: ShouldNotReachHere();
617 }
618 interp_verify_oop(rax, state);
619 }
620
621 void InterpreterMacroAssembler::push(TosState state) {
622 interp_verify_oop(rax, state);
623 switch (state) {
624 case atos: push_ptr(); break;
625 case btos:
626 case ztos:
627 case ctos:
628 case stos:
629 case itos: push_i(); break;
630 case ltos: push_l(); break;
631 case ftos: push_f(xmm0); break;
632 case dtos: push_d(xmm0); break;
633 case vtos: /* nothing to do */ break;
634 default : ShouldNotReachHere();
635 }
636 }
637 #else
638 void InterpreterMacroAssembler::pop_i(Register r) {
639 pop(r);
640 }
641
642 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
643 pop(lo);
644 pop(hi);
645 }
646
647 void InterpreterMacroAssembler::pop_f() {
648 fld_s(Address(rsp, 0));
649 addptr(rsp, 1 * wordSize);
650 }
651
652 void InterpreterMacroAssembler::pop_d() {
653 fld_d(Address(rsp, 0));
654 addptr(rsp, 2 * wordSize);
655 }
656
657
658 void InterpreterMacroAssembler::pop(TosState state) {
659 switch (state) {
660 case atos: pop_ptr(rax); break;
661 case btos: // fall through
662 case ztos: // fall through
663 case ctos: // fall through
664 case stos: // fall through
665 case itos: pop_i(rax); break;
666 case ltos: pop_l(rax, rdx); break;
667 case ftos:
668 if (UseSSE >= 1) {
669 pop_f(xmm0);
670 } else {
671 pop_f();
672 }
673 break;
674 case dtos:
675 if (UseSSE >= 2) {
676 pop_d(xmm0);
677 } else {
678 pop_d();
679 }
680 break;
681 case vtos: /* nothing to do */ break;
682 default : ShouldNotReachHere();
683 }
684 interp_verify_oop(rax, state);
685 }
686
687
688 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
689 push(hi);
690 push(lo);
691 }
692
693 void InterpreterMacroAssembler::push_f() {
694 // Do not schedule for no AGI! Never write beyond rsp!
695 subptr(rsp, 1 * wordSize);
696 fstp_s(Address(rsp, 0));
697 }
698
699 void InterpreterMacroAssembler::push_d() {
700 // Do not schedule for no AGI! Never write beyond rsp!
701 subptr(rsp, 2 * wordSize);
702 fstp_d(Address(rsp, 0));
703 }
704
705
706 void InterpreterMacroAssembler::push(TosState state) {
707 interp_verify_oop(rax, state);
708 switch (state) {
709 case atos: push_ptr(rax); break;
710 case btos: // fall through
711 case ztos: // fall through
712 case ctos: // fall through
713 case stos: // fall through
714 case itos: push_i(rax); break;
715 case ltos: push_l(rax, rdx); break;
716 case ftos:
717 if (UseSSE >= 1) {
718 push_f(xmm0);
719 } else {
720 push_f();
721 }
722 break;
723 case dtos:
724 if (UseSSE >= 2) {
725 push_d(xmm0);
726 } else {
727 push_d();
728 }
729 break;
730 case vtos: /* nothing to do */ break;
731 default : ShouldNotReachHere();
732 }
733 }
734 #endif // _LP64
735
736
737 // Helpers for swap and dup
738 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
739 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
740 }
741
742 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
743 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
744 }
745
746
747 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
748 // set sender sp
749 lea(_bcp_register, Address(rsp, wordSize));
750 // record last_sp
751 mov(rcx, _bcp_register);
752 subptr(rcx, rbp);
753 sarptr(rcx, LogBytesPerWord);
754 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx);
755 }
756
757
758 // Jump to from_interpreted entry of a call unless single stepping is possible
759 // in this thread in which case we must call the i2i entry
760 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
761 prepare_to_jump_from_interpreted();
762
763 if (JvmtiExport::can_post_interpreter_events()) {
764 Label run_compiled_code;
765 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
766 // compiled code in threads for which the event is enabled. Check here for
767 // interp_only_mode if these events CAN be enabled.
768 // interp_only is an int, on little endian it is sufficient to test the byte only
769 // Is a cmpl faster?
770 LP64_ONLY(temp = r15_thread;)
771 NOT_LP64(get_thread(temp);)
772 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
773 jccb(Assembler::zero, run_compiled_code);
774 jmp(Address(method, Method::interpreter_entry_offset()));
775 bind(run_compiled_code);
776 }
777
778 jmp(Address(method, Method::from_interpreted_offset()));
779 }
780
781 // The following two routines provide a hook so that an implementation
782 // can schedule the dispatch in two parts. x86 does not do this.
783 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
784 // Nothing x86 specific to be done here
785 }
786
787 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
788 dispatch_next(state, step);
789 }
790
791 void InterpreterMacroAssembler::dispatch_base(TosState state,
792 address* table,
793 bool verifyoop,
794 bool generate_poll) {
795 verify_FPU(1, state);
796 if (VerifyActivationFrameSize) {
797 Label L;
798 mov(rcx, rbp);
799 subptr(rcx, rsp);
800 int32_t min_frame_size =
801 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
802 wordSize;
803 cmpptr(rcx, min_frame_size);
804 jcc(Assembler::greaterEqual, L);
805 stop("broken stack frame");
806 bind(L);
807 }
808 if (verifyoop) {
809 interp_verify_oop(rax, state);
810 }
811
812 address* const safepoint_table = Interpreter::safept_table(state);
813 #ifdef _LP64
814 Label no_safepoint, dispatch;
815 if (table != safepoint_table && generate_poll) {
816 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
817 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
818
819 jccb(Assembler::zero, no_safepoint);
820 lea(rscratch1, ExternalAddress((address)safepoint_table));
821 jmpb(dispatch);
822 }
823
824 bind(no_safepoint);
825 lea(rscratch1, ExternalAddress((address)table));
826 bind(dispatch);
827 jmp(Address(rscratch1, rbx, Address::times_8));
828
829 #else
830 Address index(noreg, rbx, Address::times_ptr);
831 if (table != safepoint_table && generate_poll) {
832 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
833 Label no_safepoint;
834 const Register thread = rcx;
835 get_thread(thread);
836 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit());
837
838 jccb(Assembler::zero, no_safepoint);
839 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index);
840 jump(dispatch_addr, noreg);
841 bind(no_safepoint);
842 }
843
844 {
845 ArrayAddress dispatch_addr(ExternalAddress((address)table), index);
846 jump(dispatch_addr, noreg);
847 }
848 #endif // _LP64
849 }
850
851 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
852 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
853 }
854
855 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
856 dispatch_base(state, Interpreter::normal_table(state));
857 }
858
859 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
860 dispatch_base(state, Interpreter::normal_table(state), false);
861 }
862
863
864 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
865 // load next bytecode (load before advancing _bcp_register to prevent AGI)
866 load_unsigned_byte(rbx, Address(_bcp_register, step));
867 // advance _bcp_register
868 increment(_bcp_register, step);
869 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
870 }
871
872 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
873 // load current bytecode
874 load_unsigned_byte(rbx, Address(_bcp_register, 0));
875 dispatch_base(state, table);
876 }
877
878 void InterpreterMacroAssembler::narrow(Register result) {
879
880 // Get method->_constMethod->_result_type
881 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
882 movptr(rcx, Address(rcx, Method::const_offset()));
883 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
884
885 Label done, notBool, notByte, notChar;
886
887 // common case first
888 cmpl(rcx, T_INT);
889 jcc(Assembler::equal, done);
890
891 // mask integer result to narrower return type.
892 cmpl(rcx, T_BOOLEAN);
893 jcc(Assembler::notEqual, notBool);
894 andl(result, 0x1);
895 jmp(done);
896
897 bind(notBool);
898 cmpl(rcx, T_BYTE);
899 jcc(Assembler::notEqual, notByte);
900 LP64_ONLY(movsbl(result, result);)
901 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
902 NOT_LP64(sarl(result, 24);) // and sign-extend byte
903 jmp(done);
904
905 bind(notByte);
906 cmpl(rcx, T_CHAR);
907 jcc(Assembler::notEqual, notChar);
908 LP64_ONLY(movzwl(result, result);)
909 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
910 jmp(done);
911
912 bind(notChar);
913 // cmpl(rcx, T_SHORT); // all that's left
914 // jcc(Assembler::notEqual, done);
915 LP64_ONLY(movswl(result, result);)
916 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
917 NOT_LP64(sarl(result, 16);) // and sign-extend short
918
919 // Nothing to do for T_INT
920 bind(done);
921 }
922
923 // remove activation
924 //
925 // Apply stack watermark barrier.
926 // Unlock the receiver if this is a synchronized method.
927 // Unlock any Java monitors from synchronized blocks.
928 // Remove the activation from the stack.
929 //
930 // If there are locked Java monitors
931 // If throw_monitor_exception
932 // throws IllegalMonitorStateException
933 // Else if install_monitor_exception
934 // installs IllegalMonitorStateException
935 // Else
936 // no error processing
937 void InterpreterMacroAssembler::remove_activation(
938 TosState state,
939 Register ret_addr,
940 bool throw_monitor_exception,
941 bool install_monitor_exception,
942 bool notify_jvmdi) {
943 // Note: Registers rdx xmm0 may be in use for the
944 // result check if synchronized method
945 Label unlocked, unlock, no_unlock;
946
947 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
948 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
949 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
950 // monitor pointers need different register
951 // because rdx may have the result in it
952 NOT_LP64(get_thread(rthread);)
953
954 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
955 // that would normally not be safe to use. Such bad returns into unsafe territory of
956 // the stack, will call InterpreterRuntime::at_unwind.
957 Label slow_path;
958 Label fast_path;
959 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */);
960 jmp(fast_path);
961 bind(slow_path);
962 push(state);
963 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1);
964 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
965 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore
966 reset_last_Java_frame(rthread, true);
967 pop(state);
968 bind(fast_path);
969
970 // get the value of _do_not_unlock_if_synchronized into rdx
971 const Address do_not_unlock_if_synchronized(rthread,
972 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
973 movbool(rbx, do_not_unlock_if_synchronized);
974 movbool(do_not_unlock_if_synchronized, false); // reset the flag
975
976 // get method access flags
977 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
978 movl(rcx, Address(rcx, Method::access_flags_offset()));
979 testl(rcx, JVM_ACC_SYNCHRONIZED);
980 jcc(Assembler::zero, unlocked);
981
982 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
983 // is set.
984 testbool(rbx);
985 jcc(Assembler::notZero, no_unlock);
986
987 // unlock monitor
988 push(state); // save result
989
990 // BasicObjectLock will be first in list, since this is a
991 // synchronized method. However, need to check that the object has
992 // not been unlocked by an explicit monitorexit bytecode.
993 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
994 wordSize - (int) sizeof(BasicObjectLock));
995 // We use c_rarg1/rdx so that if we go slow path it will be the correct
996 // register for unlock_object to pass to VM directly
997 lea(robj, monitor); // address of first monitor
998
999 movptr(rax, Address(robj, BasicObjectLock::obj_offset()));
1000 testptr(rax, rax);
1001 jcc(Assembler::notZero, unlock);
1002
1003 pop(state);
1004 if (throw_monitor_exception) {
1005 // Entry already unlocked, need to throw exception
1006 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
1007 call_VM(noreg, CAST_FROM_FN_PTR(address,
1008 InterpreterRuntime::throw_illegal_monitor_state_exception));
1009 should_not_reach_here();
1010 } else {
1011 // Monitor already unlocked during a stack unroll. If requested,
1012 // install an illegal_monitor_state_exception. Continue with
1013 // stack unrolling.
1014 if (install_monitor_exception) {
1015 NOT_LP64(empty_FPU_stack();)
1016 call_VM(noreg, CAST_FROM_FN_PTR(address,
1017 InterpreterRuntime::new_illegal_monitor_state_exception));
1018 }
1019 jmp(unlocked);
1020 }
1021
1022 bind(unlock);
1023 unlock_object(robj);
1024 pop(state);
1025
1026 // Check that for block-structured locking (i.e., that all locked
1027 // objects has been unlocked)
1028 bind(unlocked);
1029
1030 // rax, rdx: Might contain return value
1031
1032 // Check that all monitors are unlocked
1033 {
1034 Label loop, exception, entry, restart;
1035 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
1036 const Address monitor_block_top(
1037 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1038 const Address monitor_block_bot(
1039 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
1040
1041 bind(restart);
1042 // We use c_rarg1 so that if we go slow path it will be the correct
1043 // register for unlock_object to pass to VM directly
1044 movptr(rmon, monitor_block_top); // derelativize pointer
1045 lea(rmon, Address(rbp, rmon, Address::times_ptr));
1046 // c_rarg1 points to current entry, starting with top-most entry
1047
1048 lea(rbx, monitor_block_bot); // points to word before bottom of
1049 // monitor block
1050 jmp(entry);
1051
1052 // Entry already locked, need to throw exception
1053 bind(exception);
1054
1055 if (throw_monitor_exception) {
1056 // Throw exception
1057 NOT_LP64(empty_FPU_stack();)
1058 MacroAssembler::call_VM(noreg,
1059 CAST_FROM_FN_PTR(address, InterpreterRuntime::
1060 throw_illegal_monitor_state_exception));
1061 should_not_reach_here();
1062 } else {
1063 // Stack unrolling. Unlock object and install illegal_monitor_exception.
1064 // Unlock does not block, so don't have to worry about the frame.
1065 // We don't have to preserve c_rarg1 since we are going to throw an exception.
1066
1067 push(state);
1068 mov(robj, rmon); // nop if robj and rmon are the same
1069 unlock_object(robj);
1070 pop(state);
1071
1072 if (install_monitor_exception) {
1073 NOT_LP64(empty_FPU_stack();)
1074 call_VM(noreg, CAST_FROM_FN_PTR(address,
1075 InterpreterRuntime::
1076 new_illegal_monitor_state_exception));
1077 }
1078
1079 jmp(restart);
1080 }
1081
1082 bind(loop);
1083 // check if current entry is used
1084 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD);
1085 jcc(Assembler::notEqual, exception);
1086
1087 addptr(rmon, entry_size); // otherwise advance to next entry
1088 bind(entry);
1089 cmpptr(rmon, rbx); // check if bottom reached
1090 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
1091 }
1092
1093 bind(no_unlock);
1094
1095 // jvmti support
1096 if (notify_jvmdi) {
1097 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
1098 } else {
1099 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
1100 }
1101
1102 if (StackReservedPages > 0) {
1103 movptr(rbx,
1104 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1105 // testing if reserved zone needs to be re-enabled
1106 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1107 Label no_reserved_zone_enabling;
1108
1109 NOT_LP64(get_thread(rthread);)
1110
1111 // check if already enabled - if so no re-enabling needed
1112 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
1113 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled);
1114 jcc(Assembler::equal, no_reserved_zone_enabling);
1115
1116 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset()));
1117 jcc(Assembler::lessEqual, no_reserved_zone_enabling);
1118
1119 call_VM_leaf(
1120 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
1121 call_VM(noreg, CAST_FROM_FN_PTR(address,
1122 InterpreterRuntime::throw_delayed_StackOverflowError));
1123 should_not_reach_here();
1124
1125 bind(no_reserved_zone_enabling);
1126 }
1127
1128 // remove activation
1129 // get sender sp
1130 movptr(rbx,
1131 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1132
1133 if (state == atos && InlineTypeReturnedAsFields) {
1134 // Check if we are returning an non-null inline type and load its fields into registers
1135 Label skip;
1136 test_oop_is_not_inline_type(rax, rscratch1, skip);
1137
1138 #ifndef _LP64
1139 super_call_VM_leaf(StubRoutines::load_inline_type_fields_in_regs());
1140 #else
1141 // Load fields from a buffered value with an inline class specific handler
1142 load_klass(rdi, rax, rscratch1);
1143 movptr(rdi, Address(rdi, InstanceKlass::adr_inlineklass_fixed_block_offset()));
1144 movptr(rdi, Address(rdi, InlineKlass::unpack_handler_offset()));
1145 // Unpack handler can be null if inline type is not scalarizable in returns
1146 testptr(rdi, rdi);
1147 jcc(Assembler::zero, skip);
1148 call(rdi);
1149 #endif
1150 #ifdef ASSERT
1151 // TODO 8284443 Enable
1152 if (StressCallingConvention && false) {
1153 Label skip_stress;
1154 movptr(rscratch1, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
1155 movl(rscratch1, Address(rscratch1, Method::flags_offset()));
1156 testl(rcx, MethodFlags::has_scalarized_return_flag());
1157 jcc(Assembler::zero, skip_stress);
1158 load_klass(rax, rax, rscratch1);
1159 orptr(rax, 1);
1160 bind(skip_stress);
1161 }
1162 #endif
1163 // call above kills the value in rbx. Reload it.
1164 movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
1165 bind(skip);
1166 }
1167 leave(); // remove frame anchor
1168 pop(ret_addr); // get return address
1169 mov(rsp, rbx); // set sp to sender sp
1170 pop_cont_fastpath();
1171 }
1172
1173 void InterpreterMacroAssembler::get_method_counters(Register method,
1174 Register mcs, Label& skip) {
1175 Label has_counters;
1176 movptr(mcs, Address(method, Method::method_counters_offset()));
1177 testptr(mcs, mcs);
1178 jcc(Assembler::notZero, has_counters);
1179 call_VM(noreg, CAST_FROM_FN_PTR(address,
1180 InterpreterRuntime::build_method_counters), method);
1181 movptr(mcs, Address(method,Method::method_counters_offset()));
1182 testptr(mcs, mcs);
1183 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
1184 bind(has_counters);
1185 }
1186
1187 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
1188 Register t1, Register t2,
1189 bool clear_fields, Label& alloc_failed) {
1190 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
1191 if (DTraceMethodProbes) {
1192 // Trigger dtrace event for fastpath
1193 push(atos);
1194 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
1195 pop(atos);
1196 }
1197 }
1198
1199 void InterpreterMacroAssembler::read_flat_field(Register entry, Register tmp1, Register tmp2, Register obj) {
1200 Label alloc_failed, empty_value, done;
1201 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1202 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1203 assert_different_registers(obj, entry, tmp1, tmp2, dst_temp, r8, r9);
1204
1205 // FIXME: code below could be re-written to better use InlineLayoutInfo data structure
1206 // see aarch64 version
1207
1208 // Grap the inline field klass
1209 const Register field_klass = tmp1;
1210 load_unsigned_short(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
1211 movptr(tmp1, Address(entry, ResolvedFieldEntry::field_holder_offset()));
1212 get_inline_type_field_klass(tmp1, tmp2, field_klass);
1213
1214 //check for empty value klass
1215 test_klass_is_empty_inline_type(field_klass, dst_temp, empty_value);
1216
1217 // allocate buffer
1218 push(obj); // push object being read from // FIXME spilling on stack could probably be avoided by using tmp2
1219 allocate_instance(field_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1220
1221 // Have an oop instance buffer, copy into it
1222 load_unsigned_short(r9, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
1223 movptr(r8, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
1224 inline_layout_info(r8, r9, r8); // holder, index, info => InlineLayoutInfo into r8
1225
1226 data_for_oop(obj, dst_temp, field_klass);
1227 pop(alloc_temp); // restore object being read from
1228 load_sized_value(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_offset_offset())), sizeof(int), true /*is_signed*/);
1229 lea(tmp2, Address(alloc_temp, tmp2));
1230 // call_VM_leaf, clobbers a few regs, save restore new obj
1231 push(obj);
1232 // access_value_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, field_klass);
1233 flat_field_copy(IS_DEST_UNINITIALIZED, tmp2, dst_temp, r8);
1234 pop(obj);
1235 jmp(done);
1236
1237 bind(empty_value);
1238 get_empty_inline_type_oop(field_klass, dst_temp, obj);
1239 jmp(done);
1240
1241 bind(alloc_failed);
1242 pop(obj);
1243 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
1244 obj, entry);
1245 get_vm_result(obj, r15_thread);
1246 bind(done);
1247 }
1248
1249 void InterpreterMacroAssembler::read_flat_element(Register array, Register index,
1250 Register t1, Register t2,
1251 Register obj) {
1252 assert_different_registers(array, index, t1, t2);
1253 Label alloc_failed, empty_value, done;
1254 const Register array_klass = t2;
1255 const Register elem_klass = t1;
1256 const Register alloc_temp = LP64_ONLY(rscratch1) NOT_LP64(rsi);
1257 const Register dst_temp = LP64_ONLY(rscratch2) NOT_LP64(rdi);
1258
1259 // load in array->klass()->element_klass()
1260 Register tmp_load_klass = LP64_ONLY(rscratch1) NOT_LP64(noreg);
1261 load_klass(array_klass, array, tmp_load_klass);
1262 movptr(elem_klass, Address(array_klass, ArrayKlass::element_klass_offset()));
1263
1264 //check for empty value klass
1265 test_klass_is_empty_inline_type(elem_klass, dst_temp, empty_value);
1266
1267 // calc source into "array_klass" and free up some regs
1268 const Register src = array_klass;
1269 push(index); // preserve index reg in case alloc_failed
1270 data_for_value_array_index(array, array_klass, index, src);
1271
1272 allocate_instance(elem_klass, obj, alloc_temp, dst_temp, false, alloc_failed);
1273 // Have an oop instance buffer, copy into it
1274 store_ptr(0, obj); // preserve obj (overwrite index, no longer needed)
1275 data_for_oop(obj, dst_temp, elem_klass);
1276 access_value_copy(IS_DEST_UNINITIALIZED, src, dst_temp, elem_klass);
1277 pop(obj);
1278 jmp(done);
1279
1280 bind(empty_value);
1281 get_empty_inline_type_oop(elem_klass, dst_temp, obj);
1282 jmp(done);
1283
1284 bind(alloc_failed);
1285 pop(index);
1286 if (array == c_rarg2) {
1287 mov(elem_klass, array);
1288 array = elem_klass;
1289 }
1290 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), array, index);
1291
1292 bind(done);
1293 }
1294
1295
1296 // Lock object
1297 //
1298 // Args:
1299 // rdx, c_rarg1: BasicObjectLock to be used for locking
1300 //
1301 // Kills:
1302 // rax, rbx
1303 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
1304 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1305 "The argument is only for looks. It must be c_rarg1");
1306
1307 if (LockingMode == LM_MONITOR) {
1308 call_VM(noreg,
1309 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1310 lock_reg);
1311 } else {
1312 Label count_locking, done, slow_case;
1313
1314 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1315 const Register tmp_reg = rbx;
1316 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1317 const Register rklass_decode_tmp = rscratch1;
1318
1319 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
1320 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
1321 const int mark_offset = lock_offset +
1322 BasicLock::displaced_header_offset_in_bytes();
1323
1324 // Load object pointer into obj_reg
1325 movptr(obj_reg, Address(lock_reg, obj_offset));
1326
1327 if (DiagnoseSyncOnValueBasedClasses != 0) {
1328 load_klass(tmp_reg, obj_reg, rklass_decode_tmp);
1329 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class);
1330 jcc(Assembler::notZero, slow_case);
1331 }
1332
1333 if (LockingMode == LM_LIGHTWEIGHT) {
1334 #ifdef _LP64
1335 const Register thread = r15_thread;
1336 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case);
1337 #else
1338 // Lacking registers and thread on x86_32. Always take slow path.
1339 jmp(slow_case);
1340 #endif
1341 } else if (LockingMode == LM_LEGACY) {
1342 // Load immediate 1 into swap_reg %rax
1343 movl(swap_reg, 1);
1344
1345 // Load (object->mark() | 1) into swap_reg %rax
1346 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1347 if (EnableValhalla) {
1348 // Mask inline_type bit such that we go to the slow path if object is an inline type
1349 andptr(swap_reg, ~((int) markWord::inline_type_bit_in_place));
1350 }
1351
1352 // Save (object->mark() | 1) into BasicLock's displaced header
1353 movptr(Address(lock_reg, mark_offset), swap_reg);
1354
1355 assert(lock_offset == 0,
1356 "displaced header must be first word in BasicObjectLock");
1357
1358 lock();
1359 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1360 jcc(Assembler::zero, count_locking);
1361
1362 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
1363
1364 // Fast check for recursive lock.
1365 //
1366 // Can apply the optimization only if this is a stack lock
1367 // allocated in this thread. For efficiency, we can focus on
1368 // recently allocated stack locks (instead of reading the stack
1369 // base and checking whether 'mark' points inside the current
1370 // thread stack):
1371 // 1) (mark & zero_bits) == 0, and
1372 // 2) rsp <= mark < mark + os::pagesize()
1373 //
1374 // Warning: rsp + os::pagesize can overflow the stack base. We must
1375 // neither apply the optimization for an inflated lock allocated
1376 // just above the thread stack (this is why condition 1 matters)
1377 // nor apply the optimization if the stack lock is inside the stack
1378 // of another thread. The latter is avoided even in case of overflow
1379 // because we have guard pages at the end of all stacks. Hence, if
1380 // we go over the stack base and hit the stack of another thread,
1381 // this should not be in a writeable area that could contain a
1382 // stack lock allocated by that thread. As a consequence, a stack
1383 // lock less than page size away from rsp is guaranteed to be
1384 // owned by the current thread.
1385 //
1386 // These 3 tests can be done by evaluating the following
1387 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
1388 // assuming both stack pointer and pagesize have their
1389 // least significant bits clear.
1390 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg
1391 subptr(swap_reg, rsp);
1392 andptr(swap_reg, zero_bits - (int)os::vm_page_size());
1393
1394 // Save the test result, for recursive case, the result is zero
1395 movptr(Address(lock_reg, mark_offset), swap_reg);
1396 jcc(Assembler::notZero, slow_case);
1397
1398 bind(count_locking);
1399 }
1400 inc_held_monitor_count();
1401 jmp(done);
1402
1403 bind(slow_case);
1404
1405 // Call the runtime routine for slow case
1406 call_VM(noreg,
1407 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
1408 lock_reg);
1409 bind(done);
1410 }
1411 }
1412
1413
1414 // Unlocks an object. Used in monitorexit bytecode and
1415 // remove_activation. Throws an IllegalMonitorException if object is
1416 // not locked by current thread.
1417 //
1418 // Args:
1419 // rdx, c_rarg1: BasicObjectLock for lock
1420 //
1421 // Kills:
1422 // rax
1423 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
1424 // rscratch1 (scratch reg)
1425 // rax, rbx, rcx, rdx
1426 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1427 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
1428 "The argument is only for looks. It must be c_rarg1");
1429
1430 if (LockingMode == LM_MONITOR) {
1431 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1432 } else {
1433 Label count_locking, done, slow_case;
1434
1435 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
1436 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
1437 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
1438
1439 save_bcp(); // Save in case of exception
1440
1441 if (LockingMode != LM_LIGHTWEIGHT) {
1442 // Convert from BasicObjectLock structure to object and BasicLock
1443 // structure Store the BasicLock address into %rax
1444 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
1445 }
1446
1447 // Load oop into obj_reg(%c_rarg3)
1448 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
1449
1450 // Free entry
1451 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD);
1452
1453 if (LockingMode == LM_LIGHTWEIGHT) {
1454 #ifdef _LP64
1455 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case);
1456 #else
1457 // Lacking registers and thread on x86_32. Always take slow path.
1458 jmp(slow_case);
1459 #endif
1460 } else if (LockingMode == LM_LEGACY) {
1461 // Load the old header from BasicLock structure
1462 movptr(header_reg, Address(swap_reg,
1463 BasicLock::displaced_header_offset_in_bytes()));
1464
1465 // Test for recursion
1466 testptr(header_reg, header_reg);
1467
1468 // zero for recursive case
1469 jcc(Assembler::zero, count_locking);
1470
1471 // Atomic swap back the old header
1472 lock();
1473 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
1474
1475 // zero for simple unlock of a stack-lock case
1476 jcc(Assembler::notZero, slow_case);
1477
1478 bind(count_locking);
1479 }
1480 dec_held_monitor_count();
1481 jmp(done);
1482
1483 bind(slow_case);
1484 // Call the runtime routine for slow case.
1485 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj
1486 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1487
1488 bind(done);
1489
1490 restore_bcp();
1491 }
1492 }
1493
1494 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1495 Label& zero_continue) {
1496 assert(ProfileInterpreter, "must be profiling interpreter");
1497 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
1498 testptr(mdp, mdp);
1499 jcc(Assembler::zero, zero_continue);
1500 }
1501
1502
1503 // Set the method data pointer for the current bcp.
1504 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1505 assert(ProfileInterpreter, "must be profiling interpreter");
1506 Label set_mdp;
1507 push(rax);
1508 push(rbx);
1509
1510 get_method(rbx);
1511 // Test MDO to avoid the call if it is null.
1512 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1513 testptr(rax, rax);
1514 jcc(Assembler::zero, set_mdp);
1515 // rbx: method
1516 // _bcp_register: bcp
1517 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1518 // rax: mdi
1519 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1520 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1521 addptr(rbx, in_bytes(MethodData::data_offset()));
1522 addptr(rax, rbx);
1523 bind(set_mdp);
1524 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1525 pop(rbx);
1526 pop(rax);
1527 }
1528
1529 void InterpreterMacroAssembler::verify_method_data_pointer() {
1530 assert(ProfileInterpreter, "must be profiling interpreter");
1531 #ifdef ASSERT
1532 Label verify_continue;
1533 push(rax);
1534 push(rbx);
1535 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1536 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1537 push(arg3_reg);
1538 push(arg2_reg);
1539 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1540 get_method(rbx);
1541
1542 // If the mdp is valid, it will point to a DataLayout header which is
1543 // consistent with the bcp. The converse is highly probable also.
1544 load_unsigned_short(arg2_reg,
1545 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1546 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1547 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1548 cmpptr(arg2_reg, _bcp_register);
1549 jcc(Assembler::equal, verify_continue);
1550 // rbx: method
1551 // _bcp_register: bcp
1552 // c_rarg3: mdp
1553 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1554 rbx, _bcp_register, arg3_reg);
1555 bind(verify_continue);
1556 pop(arg2_reg);
1557 pop(arg3_reg);
1558 pop(rbx);
1559 pop(rax);
1560 #endif // ASSERT
1561 }
1562
1563
1564 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1565 int constant,
1566 Register value) {
1567 assert(ProfileInterpreter, "must be profiling interpreter");
1568 Address data(mdp_in, constant);
1569 movptr(data, value);
1570 }
1571
1572
1573 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1574 int constant,
1575 bool decrement) {
1576 // Counter address
1577 Address data(mdp_in, constant);
1578
1579 increment_mdp_data_at(data, decrement);
1580 }
1581
1582 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1583 bool decrement) {
1584 assert(ProfileInterpreter, "must be profiling interpreter");
1585 // %%% this does 64bit counters at best it is wasting space
1586 // at worst it is a rare bug when counters overflow
1587
1588 if (decrement) {
1589 // Decrement the register. Set condition codes.
1590 addptr(data, -DataLayout::counter_increment);
1591 // If the decrement causes the counter to overflow, stay negative
1592 Label L;
1593 jcc(Assembler::negative, L);
1594 addptr(data, DataLayout::counter_increment);
1595 bind(L);
1596 } else {
1597 assert(DataLayout::counter_increment == 1,
1598 "flow-free idiom only works with 1");
1599 // Increment the register. Set carry flag.
1600 addptr(data, DataLayout::counter_increment);
1601 // If the increment causes the counter to overflow, pull back by 1.
1602 sbbptr(data, 0);
1603 }
1604 }
1605
1606
1607 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1608 Register reg,
1609 int constant,
1610 bool decrement) {
1611 Address data(mdp_in, reg, Address::times_1, constant);
1612
1613 increment_mdp_data_at(data, decrement);
1614 }
1615
1616 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1617 int flag_byte_constant) {
1618 assert(ProfileInterpreter, "must be profiling interpreter");
1619 int header_offset = in_bytes(DataLayout::flags_offset());
1620 int header_bits = flag_byte_constant;
1621 // Set the flag
1622 orb(Address(mdp_in, header_offset), header_bits);
1623 }
1624
1625
1626
1627 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1628 int offset,
1629 Register value,
1630 Register test_value_out,
1631 Label& not_equal_continue) {
1632 assert(ProfileInterpreter, "must be profiling interpreter");
1633 if (test_value_out == noreg) {
1634 cmpptr(value, Address(mdp_in, offset));
1635 } else {
1636 // Put the test value into a register, so caller can use it:
1637 movptr(test_value_out, Address(mdp_in, offset));
1638 cmpptr(test_value_out, value);
1639 }
1640 jcc(Assembler::notEqual, not_equal_continue);
1641 }
1642
1643
1644 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1645 int offset_of_disp) {
1646 assert(ProfileInterpreter, "must be profiling interpreter");
1647 Address disp_address(mdp_in, offset_of_disp);
1648 addptr(mdp_in, disp_address);
1649 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1650 }
1651
1652
1653 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1654 Register reg,
1655 int offset_of_disp) {
1656 assert(ProfileInterpreter, "must be profiling interpreter");
1657 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1658 addptr(mdp_in, disp_address);
1659 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1660 }
1661
1662
1663 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1664 int constant) {
1665 assert(ProfileInterpreter, "must be profiling interpreter");
1666 addptr(mdp_in, constant);
1667 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1668 }
1669
1670
1671 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1672 assert(ProfileInterpreter, "must be profiling interpreter");
1673 push(return_bci); // save/restore across call_VM
1674 call_VM(noreg,
1675 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1676 return_bci);
1677 pop(return_bci);
1678 }
1679
1680
1681 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1682 Register bumped_count) {
1683 if (ProfileInterpreter) {
1684 Label profile_continue;
1685
1686 // If no method data exists, go to profile_continue.
1687 // Otherwise, assign to mdp
1688 test_method_data_pointer(mdp, profile_continue);
1689
1690 // We are taking a branch. Increment the taken count.
1691 // We inline increment_mdp_data_at to return bumped_count in a register
1692 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1693 Address data(mdp, in_bytes(JumpData::taken_offset()));
1694 movptr(bumped_count, data);
1695 assert(DataLayout::counter_increment == 1,
1696 "flow-free idiom only works with 1");
1697 addptr(bumped_count, DataLayout::counter_increment);
1698 sbbptr(bumped_count, 0);
1699 movptr(data, bumped_count); // Store back out
1700
1701 // The method data pointer needs to be updated to reflect the new target.
1702 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1703 bind(profile_continue);
1704 }
1705 }
1706
1707
1708 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1709 if (ProfileInterpreter) {
1710 Label profile_continue;
1711
1712 // If no method data exists, go to profile_continue.
1713 test_method_data_pointer(mdp, profile_continue);
1714
1715 // We are taking a branch. Increment the not taken count.
1716 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1717
1718 // The method data pointer needs to be updated to correspond to
1719 // the next bytecode
1720 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()): in_bytes(BranchData::branch_data_size()));
1721 bind(profile_continue);
1722 }
1723 }
1724
1725 void InterpreterMacroAssembler::profile_call(Register mdp) {
1726 if (ProfileInterpreter) {
1727 Label profile_continue;
1728
1729 // If no method data exists, go to profile_continue.
1730 test_method_data_pointer(mdp, profile_continue);
1731
1732 // We are making a call. Increment the count.
1733 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1734
1735 // The method data pointer needs to be updated to reflect the new target.
1736 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1737 bind(profile_continue);
1738 }
1739 }
1740
1741
1742 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1743 if (ProfileInterpreter) {
1744 Label profile_continue;
1745
1746 // If no method data exists, go to profile_continue.
1747 test_method_data_pointer(mdp, profile_continue);
1748
1749 // We are making a call. Increment the count.
1750 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1751
1752 // The method data pointer needs to be updated to reflect the new target.
1753 update_mdp_by_constant(mdp,
1754 in_bytes(VirtualCallData::
1755 virtual_call_data_size()));
1756 bind(profile_continue);
1757 }
1758 }
1759
1760
1761 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1762 Register mdp,
1763 Register reg2,
1764 bool receiver_can_be_null) {
1765 if (ProfileInterpreter) {
1766 Label profile_continue;
1767
1768 // If no method data exists, go to profile_continue.
1769 test_method_data_pointer(mdp, profile_continue);
1770
1771 Label skip_receiver_profile;
1772 if (receiver_can_be_null) {
1773 Label not_null;
1774 testptr(receiver, receiver);
1775 jccb(Assembler::notZero, not_null);
1776 // We are making a call. Increment the count for null receiver.
1777 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1778 jmp(skip_receiver_profile);
1779 bind(not_null);
1780 }
1781
1782 // Record the receiver type.
1783 record_klass_in_profile(receiver, mdp, reg2);
1784 bind(skip_receiver_profile);
1785
1786 // The method data pointer needs to be updated to reflect the new target.
1787 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1788 bind(profile_continue);
1789 }
1790 }
1791
1792 // This routine creates a state machine for updating the multi-row
1793 // type profile at a virtual call site (or other type-sensitive bytecode).
1794 // The machine visits each row (of receiver/count) until the receiver type
1795 // is found, or until it runs out of rows. At the same time, it remembers
1796 // the location of the first empty row. (An empty row records null for its
1797 // receiver, and can be allocated for a newly-observed receiver type.)
1798 // Because there are two degrees of freedom in the state, a simple linear
1799 // search will not work; it must be a decision tree. Hence this helper
1800 // function is recursive, to generate the required tree structured code.
1801 // It's the interpreter, so we are trading off code space for speed.
1802 // See below for example code.
1803 void InterpreterMacroAssembler::record_klass_in_profile_helper(Register receiver, Register mdp,
1804 Register reg2, int start_row,
1805 Label& done) {
1806 if (TypeProfileWidth == 0) {
1807 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1808 } else {
1809 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1810 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1811 }
1812 }
1813
1814 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row,
1815 Label& done, int total_rows,
1816 OffsetFunction item_offset_fn,
1817 OffsetFunction item_count_offset_fn) {
1818 int last_row = total_rows - 1;
1819 assert(start_row <= last_row, "must be work left to do");
1820 // Test this row for both the item and for null.
1821 // Take any of three different outcomes:
1822 // 1. found item => increment count and goto done
1823 // 2. found null => keep looking for case 1, maybe allocate this cell
1824 // 3. found something else => keep looking for cases 1 and 2
1825 // Case 3 is handled by a recursive call.
1826 for (int row = start_row; row <= last_row; row++) {
1827 Label next_test;
1828 bool test_for_null_also = (row == start_row);
1829
1830 // See if the item is item[n].
1831 int item_offset = in_bytes(item_offset_fn(row));
1832 test_mdp_data_at(mdp, item_offset, item,
1833 (test_for_null_also ? reg2 : noreg),
1834 next_test);
1835 // (Reg2 now contains the item from the CallData.)
1836
1837 // The item is item[n]. Increment count[n].
1838 int count_offset = in_bytes(item_count_offset_fn(row));
1839 increment_mdp_data_at(mdp, count_offset);
1840 jmp(done);
1841 bind(next_test);
1842
1843 if (test_for_null_also) {
1844 // Failed the equality check on item[n]... Test for null.
1845 testptr(reg2, reg2);
1846 if (start_row == last_row) {
1847 // The only thing left to do is handle the null case.
1848 Label found_null;
1849 jccb(Assembler::zero, found_null);
1850 // Item did not match any saved item and there is no empty row for it.
1851 // Increment total counter to indicate polymorphic case.
1852 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1853 jmp(done);
1854 bind(found_null);
1855 break;
1856 }
1857 Label found_null;
1858 // Since null is rare, make it be the branch-taken case.
1859 jcc(Assembler::zero, found_null);
1860
1861 // Put all the "Case 3" tests here.
1862 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1863 item_offset_fn, item_count_offset_fn);
1864
1865 // Found a null. Keep searching for a matching item,
1866 // but remember that this is an empty (unused) slot.
1867 bind(found_null);
1868 }
1869 }
1870
1871 // In the fall-through case, we found no matching item, but we
1872 // observed the item[start_row] is null.
1873
1874 // Fill in the item field and increment the count.
1875 int item_offset = in_bytes(item_offset_fn(start_row));
1876 set_mdp_data_at(mdp, item_offset, item);
1877 int count_offset = in_bytes(item_count_offset_fn(start_row));
1878 movl(reg2, DataLayout::counter_increment);
1879 set_mdp_data_at(mdp, count_offset, reg2);
1880 if (start_row > 0) {
1881 jmp(done);
1882 }
1883 }
1884
1885 // Example state machine code for three profile rows:
1886 // // main copy of decision tree, rooted at row[1]
1887 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1888 // if (row[0].rec != nullptr) {
1889 // // inner copy of decision tree, rooted at row[1]
1890 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1891 // if (row[1].rec != nullptr) {
1892 // // degenerate decision tree, rooted at row[2]
1893 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1894 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1895 // row[2].init(rec); goto done;
1896 // } else {
1897 // // remember row[1] is empty
1898 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1899 // row[1].init(rec); goto done;
1900 // }
1901 // } else {
1902 // // remember row[0] is empty
1903 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1904 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1905 // row[0].init(rec); goto done;
1906 // }
1907 // done:
1908
1909 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, Register mdp, Register reg2) {
1910 assert(ProfileInterpreter, "must be profiling");
1911 Label done;
1912
1913 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1914
1915 bind (done);
1916 }
1917
1918 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1919 Register mdp) {
1920 if (ProfileInterpreter) {
1921 Label profile_continue;
1922 uint row;
1923
1924 // If no method data exists, go to profile_continue.
1925 test_method_data_pointer(mdp, profile_continue);
1926
1927 // Update the total ret count.
1928 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1929
1930 for (row = 0; row < RetData::row_limit(); row++) {
1931 Label next_test;
1932
1933 // See if return_bci is equal to bci[n]:
1934 test_mdp_data_at(mdp,
1935 in_bytes(RetData::bci_offset(row)),
1936 return_bci, noreg,
1937 next_test);
1938
1939 // return_bci is equal to bci[n]. Increment the count.
1940 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1941
1942 // The method data pointer needs to be updated to reflect the new target.
1943 update_mdp_by_offset(mdp,
1944 in_bytes(RetData::bci_displacement_offset(row)));
1945 jmp(profile_continue);
1946 bind(next_test);
1947 }
1948
1949 update_mdp_for_ret(return_bci);
1950
1951 bind(profile_continue);
1952 }
1953 }
1954
1955
1956 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1957 if (ProfileInterpreter) {
1958 Label profile_continue;
1959
1960 // If no method data exists, go to profile_continue.
1961 test_method_data_pointer(mdp, profile_continue);
1962
1963 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1964
1965 // The method data pointer needs to be updated.
1966 int mdp_delta = in_bytes(BitData::bit_data_size());
1967 if (TypeProfileCasts) {
1968 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1969 }
1970 update_mdp_by_constant(mdp, mdp_delta);
1971
1972 bind(profile_continue);
1973 }
1974 }
1975
1976
1977 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1978 if (ProfileInterpreter) {
1979 Label profile_continue;
1980
1981 // If no method data exists, go to profile_continue.
1982 test_method_data_pointer(mdp, profile_continue);
1983
1984 // The method data pointer needs to be updated.
1985 int mdp_delta = in_bytes(BitData::bit_data_size());
1986 if (TypeProfileCasts) {
1987 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1988
1989 // Record the object type.
1990 record_klass_in_profile(klass, mdp, reg2);
1991 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1992 NOT_LP64(restore_locals();) // Restore EDI
1993 }
1994 update_mdp_by_constant(mdp, mdp_delta);
1995
1996 bind(profile_continue);
1997 }
1998 }
1999
2000
2001 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
2002 if (ProfileInterpreter) {
2003 Label profile_continue;
2004
2005 // If no method data exists, go to profile_continue.
2006 test_method_data_pointer(mdp, profile_continue);
2007
2008 // Update the default case count
2009 increment_mdp_data_at(mdp,
2010 in_bytes(MultiBranchData::default_count_offset()));
2011
2012 // The method data pointer needs to be updated.
2013 update_mdp_by_offset(mdp,
2014 in_bytes(MultiBranchData::
2015 default_displacement_offset()));
2016
2017 bind(profile_continue);
2018 }
2019 }
2020
2021
2022 void InterpreterMacroAssembler::profile_switch_case(Register index,
2023 Register mdp,
2024 Register reg2) {
2025 if (ProfileInterpreter) {
2026 Label profile_continue;
2027
2028 // If no method data exists, go to profile_continue.
2029 test_method_data_pointer(mdp, profile_continue);
2030
2031 // Build the base (index * per_case_size_in_bytes()) +
2032 // case_array_offset_in_bytes()
2033 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
2034 imulptr(index, reg2); // XXX l ?
2035 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
2036
2037 // Update the case count
2038 increment_mdp_data_at(mdp,
2039 index,
2040 in_bytes(MultiBranchData::relative_count_offset()));
2041
2042 // The method data pointer needs to be updated.
2043 update_mdp_by_offset(mdp,
2044 index,
2045 in_bytes(MultiBranchData::
2046 relative_displacement_offset()));
2047
2048 bind(profile_continue);
2049 }
2050 }
2051
2052 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
2053 Register array,
2054 Register tmp) {
2055 if (ProfileInterpreter) {
2056 Label profile_continue;
2057
2058 // If no method data exists, go to profile_continue.
2059 test_method_data_pointer(mdp, profile_continue);
2060
2061 mov(tmp, array);
2062 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
2063
2064 Label not_flat;
2065 test_non_flat_array_oop(array, tmp, not_flat);
2066
2067 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
2068
2069 bind(not_flat);
2070
2071 Label not_null_free;
2072 test_non_null_free_array_oop(array, tmp, not_null_free);
2073
2074 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
2075
2076 bind(not_null_free);
2077
2078 bind(profile_continue);
2079 }
2080 }
2081
2082 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
2083 Register array,
2084 Register tmp);
2085 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
2086 Register array,
2087 Register tmp);
2088
2089
2090 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
2091 if (ProfileInterpreter) {
2092 Label profile_continue;
2093
2094 // If no method data exists, go to profile_continue.
2095 test_method_data_pointer(mdp, profile_continue);
2096
2097 Label done, update;
2098 testptr(element, element);
2099 jccb(Assembler::notZero, update);
2100 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
2101 jmp(done);
2102
2103 bind(update);
2104 load_klass(tmp, element, rscratch1);
2105
2106 // Record the object type.
2107 record_klass_in_profile(tmp, mdp, tmp2);
2108
2109 bind(done);
2110
2111 // The method data pointer needs to be updated.
2112 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
2113
2114 bind(profile_continue);
2115 }
2116 }
2117
2118 void InterpreterMacroAssembler::profile_element_type(Register mdp,
2119 Register element,
2120 Register tmp) {
2121 if (ProfileInterpreter) {
2122 Label profile_continue;
2123
2124 // If no method data exists, go to profile_continue.
2125 test_method_data_pointer(mdp, profile_continue);
2126
2127 mov(tmp, element);
2128 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
2129
2130 // The method data pointer needs to be updated.
2131 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
2132
2133 bind(profile_continue);
2134 }
2135 }
2136
2137 void InterpreterMacroAssembler::profile_acmp(Register mdp,
2138 Register left,
2139 Register right,
2140 Register tmp) {
2141 if (ProfileInterpreter) {
2142 Label profile_continue;
2143
2144 // If no method data exists, go to profile_continue.
2145 test_method_data_pointer(mdp, profile_continue);
2146
2147 mov(tmp, left);
2148 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
2149
2150 Label left_not_inline_type;
2151 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
2152 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
2153 bind(left_not_inline_type);
2154
2155 mov(tmp, right);
2156 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
2157
2158 Label right_not_inline_type;
2159 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
2160 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
2161 bind(right_not_inline_type);
2162
2163 bind(profile_continue);
2164 }
2165 }
2166
2167
2168 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
2169 if (state == atos) {
2170 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
2171 }
2172 }
2173
2174 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2175 #ifndef _LP64
2176 if ((state == ftos && UseSSE < 1) ||
2177 (state == dtos && UseSSE < 2)) {
2178 MacroAssembler::verify_FPU(stack_depth);
2179 }
2180 #endif
2181 }
2182
2183 // Jump if ((*counter_addr += increment) & mask) == 0
2184 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask,
2185 Register scratch, Label* where) {
2186 // This update is actually not atomic and can lose a number of updates
2187 // under heavy contention, but the alternative of using the (contended)
2188 // atomic update here penalizes profiling paths too much.
2189 movl(scratch, counter_addr);
2190 incrementl(scratch, InvocationCounter::count_increment);
2191 movl(counter_addr, scratch);
2192 andl(scratch, mask);
2193 if (where != nullptr) {
2194 jcc(Assembler::zero, *where);
2195 }
2196 }
2197
2198 void InterpreterMacroAssembler::notify_method_entry() {
2199 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2200 // track stack depth. If it is possible to enter interp_only_mode we add
2201 // the code to check if the event should be sent.
2202 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2203 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2204 if (JvmtiExport::can_post_interpreter_events()) {
2205 Label L;
2206 NOT_LP64(get_thread(rthread);)
2207 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2208 testl(rdx, rdx);
2209 jcc(Assembler::zero, L);
2210 call_VM(noreg, CAST_FROM_FN_PTR(address,
2211 InterpreterRuntime::post_method_entry));
2212 bind(L);
2213 }
2214
2215 if (DTraceMethodProbes) {
2216 NOT_LP64(get_thread(rthread);)
2217 get_method(rarg);
2218 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2219 rthread, rarg);
2220 }
2221
2222 // RedefineClasses() tracing support for obsolete method entry
2223 if (log_is_enabled(Trace, redefine, class, obsolete)) {
2224 NOT_LP64(get_thread(rthread);)
2225 get_method(rarg);
2226 call_VM_leaf(
2227 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2228 rthread, rarg);
2229 }
2230 }
2231
2232
2233 void InterpreterMacroAssembler::notify_method_exit(
2234 TosState state, NotifyMethodExitMode mode) {
2235 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
2236 // track stack depth. If it is possible to enter interp_only_mode we add
2237 // the code to check if the event should be sent.
2238 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
2239 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2240 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2241 Label L;
2242 // Note: frame::interpreter_frame_result has a dependency on how the
2243 // method result is saved across the call to post_method_exit. If this
2244 // is changed then the interpreter_frame_result implementation will
2245 // need to be updated too.
2246
2247 // template interpreter will leave the result on the top of the stack.
2248 push(state);
2249 NOT_LP64(get_thread(rthread);)
2250 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
2251 testl(rdx, rdx);
2252 jcc(Assembler::zero, L);
2253 call_VM(noreg,
2254 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2255 bind(L);
2256 pop(state);
2257 }
2258
2259 if (DTraceMethodProbes) {
2260 push(state);
2261 NOT_LP64(get_thread(rthread);)
2262 get_method(rarg);
2263 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2264 rthread, rarg);
2265 pop(state);
2266 }
2267 }
2268
2269 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2270 // Get index out of bytecode pointer
2271 get_cache_index_at_bcp(index, 1, sizeof(u4));
2272 // Get address of invokedynamic array
2273 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2274 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2275 if (is_power_of_2(sizeof(ResolvedIndyEntry))) {
2276 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2
2277 } else {
2278 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2279 }
2280 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes()));
2281 }
2282
2283 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2284 // Get index out of bytecode pointer
2285 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2286 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2287
2288 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset()));
2289 // Take shortcut if the size is a power of 2
2290 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2291 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2292 } else {
2293 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2294 }
2295 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes()));
2296 }
2297
2298 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2299 // Get index out of bytecode pointer
2300 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
2301 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2302
2303 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset()));
2304 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2305 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes()));
2306 }