1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/macroAssembler.inline.hpp"
27 #include "compiler/compiler_globals.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interp_masm_aarch64.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "logging/log.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/constMethodFlags.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/inlineKlass.hpp"
40 #include "oops/resolvedFieldEntry.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/jvmtiThreadState.hpp"
45 #include "runtime/basicLock.hpp"
46 #include "runtime/frame.inline.hpp"
47 #include "runtime/javaThread.hpp"
48 #include "runtime/safepointMechanism.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "utilities/powerOfTwo.hpp"
51
52 void InterpreterMacroAssembler::narrow(Register result) {
53
54 // Get method->_constMethod->_result_type
55 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
56 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
57 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
58
59 Label done, notBool, notByte, notChar;
60
61 // common case first
62 cmpw(rscratch1, T_INT);
63 br(Assembler::EQ, done);
64
65 // mask integer result to narrower return type.
66 cmpw(rscratch1, T_BOOLEAN);
67 br(Assembler::NE, notBool);
68 andw(result, result, 0x1);
69 b(done);
70
71 bind(notBool);
72 cmpw(rscratch1, T_BYTE);
73 br(Assembler::NE, notByte);
74 sbfx(result, result, 0, 8);
75 b(done);
76
77 bind(notByte);
78 cmpw(rscratch1, T_CHAR);
79 br(Assembler::NE, notChar);
80 ubfx(result, result, 0, 16); // truncate upper 16 bits
81 b(done);
82
83 bind(notChar);
84 sbfx(result, result, 0, 16); // sign-extend short
85
86 // Nothing to do for T_INT
87 bind(done);
88 }
89
90 void InterpreterMacroAssembler::jump_to_entry(address entry) {
91 assert(entry, "Entry must have been generated by now");
92 b(entry);
93 }
94
95 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
96 if (JvmtiExport::can_pop_frame()) {
97 Label L;
98 // Initiate popframe handling only if it is not already being
99 // processed. If the flag has the popframe_processing bit set, it
100 // means that this code is called *during* popframe handling - we
101 // don't want to reenter.
102 // This method is only called just after the call into the vm in
103 // call_VM_base, so the arg registers are available.
104 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
105 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
106 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
107 // Call Interpreter::remove_activation_preserving_args_entry() to get the
108 // address of the same-named entrypoint in the generated interpreter code.
109 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
110 br(r0);
111 bind(L);
112 }
113 }
114
115
116 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
117 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
118 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
119 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
120 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
121 switch (state) {
122 case atos: ldr(r0, oop_addr);
123 str(zr, oop_addr);
124 interp_verify_oop(r0, state); break;
125 case ltos: ldr(r0, val_addr); break;
126 case btos: // fall through
127 case ztos: // fall through
128 case ctos: // fall through
129 case stos: // fall through
130 case itos: ldrw(r0, val_addr); break;
131 case ftos: ldrs(v0, val_addr); break;
132 case dtos: ldrd(v0, val_addr); break;
133 case vtos: /* nothing to do */ break;
134 default : ShouldNotReachHere();
135 }
136 // Clean up tos value in the thread object
137 movw(rscratch1, (int) ilgl);
138 strw(rscratch1, tos_addr);
139 strw(zr, val_addr);
140 }
141
142
143 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
144 if (JvmtiExport::can_force_early_return()) {
145 Label L;
146 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
147 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
148
149 // Initiate earlyret handling only if it is not already being processed.
150 // If the flag has the earlyret_processing bit set, it means that this code
151 // is called *during* earlyret handling - we don't want to reenter.
152 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
153 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
154 br(Assembler::NE, L);
155
156 // Call Interpreter::remove_activation_early_entry() to get the address of the
157 // same-named entrypoint in the generated interpreter code.
158 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
159 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
160 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
161 br(r0);
162 bind(L);
163 }
164 }
165
166 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
167 Register reg,
168 int bcp_offset) {
169 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
170 ldrh(reg, Address(rbcp, bcp_offset));
171 rev16(reg, reg);
172 }
173
174 void InterpreterMacroAssembler::get_dispatch() {
175 uint64_t offset;
176 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
177 // Use add() here after ARDP, rather than lea().
178 // lea() does not generate anything if its offset is zero.
179 // However, relocs expect to find either an ADD or a load/store
180 // insn after an ADRP. add() always generates an ADD insn, even
181 // for add(Rn, Rn, 0).
182 add(rdispatch, rdispatch, offset);
183 }
184
185 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
186 int bcp_offset,
187 size_t index_size) {
188 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
189 if (index_size == sizeof(u2)) {
190 load_unsigned_short(index, Address(rbcp, bcp_offset));
191 } else if (index_size == sizeof(u4)) {
192 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
193 ldrw(index, Address(rbcp, bcp_offset));
194 } else if (index_size == sizeof(u1)) {
195 load_unsigned_byte(index, Address(rbcp, bcp_offset));
196 } else {
197 ShouldNotReachHere();
198 }
199 }
200
201 void InterpreterMacroAssembler::get_method_counters(Register method,
202 Register mcs, Label& skip) {
203 Label has_counters;
204 ldr(mcs, Address(method, Method::method_counters_offset()));
205 cbnz(mcs, has_counters);
206 call_VM(noreg, CAST_FROM_FN_PTR(address,
207 InterpreterRuntime::build_method_counters), method);
208 ldr(mcs, Address(method, Method::method_counters_offset()));
209 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
210 bind(has_counters);
211 }
212
213 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
214 Register t1, Register t2,
215 bool clear_fields, Label& alloc_failed) {
216 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
217 if (DTraceAllocProbes) {
218 // Trigger dtrace event for fastpath
219 push(atos);
220 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
221 pop(atos);
222 }
223 }
224
225 void InterpreterMacroAssembler::read_flat_field(Register entry,
226 Register field_index, Register field_offset,
227 Register temp, Register obj) {
228 Label failed_alloc, slow_path, done;
229 const Register src = field_offset;
230 const Register alloc_temp = r10;
231 const Register dst_temp = field_index;
232 const Register layout_info = temp;
233 assert_different_registers(obj, entry, field_index, field_offset, temp, alloc_temp, rscratch1);
234
235 load_unsigned_byte(temp, Address(entry, in_bytes(ResolvedFieldEntry::flags_offset())));
236 // If the field is nullable, jump to slow path
237 tbz(temp, ResolvedFieldEntry::is_null_free_inline_type_shift, slow_path);
238
239 // Grab the inline field klass
240 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
241 inline_layout_info(rscratch1, field_index, layout_info);
242
243 const Register field_klass = dst_temp;
244 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
245
246 // allocate buffer
247 push(obj); // save holder
248 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, failed_alloc);
249
250 // Have an oop instance buffer, copy into it
251 payload_address(obj, dst_temp, field_klass); // danger, uses rscratch1
252 pop(alloc_temp); // restore holder
253 lea(src, Address(alloc_temp, field_offset));
254 // call_VM_leaf, clobbers a few regs, save restore new obj
255 push(obj);
256 flat_field_copy(IS_DEST_UNINITIALIZED, src, dst_temp, layout_info);
257 pop(obj);
258 b(done);
259
260 bind(failed_alloc);
261 pop(obj);
262 bind(slow_path);
263 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
264 obj, entry);
265
266 bind(done);
267 membar(Assembler::StoreStore);
268 }
269
270 void InterpreterMacroAssembler::write_flat_field(Register entry, Register field_offset,
271 Register tmp1, Register tmp2,
272 Register obj) {
273 assert_different_registers(entry, field_offset, tmp1, tmp2, obj);
274 Label slow_path, done;
275
276 load_unsigned_byte(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::flags_offset())));
277 test_field_is_not_null_free_inline_type(tmp1, noreg /* temp */, slow_path);
278
279 null_check(r0); // FIXME JDK-8341120
280
281 add(obj, obj, field_offset);
282
283 load_klass(tmp1, r0);
284 payload_address(r0, r0, tmp1);
285
286 Register layout_info = field_offset;
287 load_unsigned_short(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
288 ldr(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
289 inline_layout_info(tmp2, tmp1, layout_info);
290
291 flat_field_copy(IN_HEAP, r0, obj, layout_info);
292 b(done);
293
294 bind(slow_path);
295 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flat_field), obj, r0, entry);
296 bind(done);
297 }
298
299 // Load object from cpool->resolved_references(index)
300 void InterpreterMacroAssembler::load_resolved_reference_at_index(
301 Register result, Register index, Register tmp) {
302 assert_different_registers(result, index);
303
304 get_constant_pool(result);
305 // load pointer for resolved_references[] objArray
306 ldr(result, Address(result, ConstantPool::cache_offset()));
307 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
308 resolve_oop_handle(result, tmp, rscratch2);
309 // Add in the index
310 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
311 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
312 }
313
314 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
315 Register cpool, Register index, Register klass, Register temp) {
316 add(temp, cpool, index, LSL, LogBytesPerWord);
317 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
318 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
319 add(klass, klass, temp, LSL, LogBytesPerWord);
320 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
321 }
322
323 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
324 // subtype of super_klass.
325 //
326 // Args:
327 // r0: superklass
328 // Rsub_klass: subklass
329 //
330 // Kills:
331 // r2, r5
332 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
333 Label& ok_is_subtype,
334 bool profile) {
335 assert(Rsub_klass != r0, "r0 holds superklass");
336 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
337 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
338
339 // Profile the not-null value's klass.
340 if (profile) {
341 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
342 }
343
344 // Do the check.
345 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
346 }
347
348 // Java Expression Stack
349
350 void InterpreterMacroAssembler::pop_ptr(Register r) {
351 ldr(r, post(esp, wordSize));
352 }
353
354 void InterpreterMacroAssembler::pop_i(Register r) {
355 ldrw(r, post(esp, wordSize));
356 }
357
358 void InterpreterMacroAssembler::pop_l(Register r) {
359 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
360 }
361
362 void InterpreterMacroAssembler::push_ptr(Register r) {
363 str(r, pre(esp, -wordSize));
364 }
365
366 void InterpreterMacroAssembler::push_i(Register r) {
367 str(r, pre(esp, -wordSize));
368 }
369
370 void InterpreterMacroAssembler::push_l(Register r) {
371 str(zr, pre(esp, -wordSize));
372 str(r, pre(esp, - wordSize));
373 }
374
375 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
376 ldrs(r, post(esp, wordSize));
377 }
378
379 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
380 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
381 }
382
383 void InterpreterMacroAssembler::push_f(FloatRegister r) {
384 strs(r, pre(esp, -wordSize));
385 }
386
387 void InterpreterMacroAssembler::push_d(FloatRegister r) {
388 strd(r, pre(esp, 2* -wordSize));
389 }
390
391 void InterpreterMacroAssembler::pop(TosState state) {
392 switch (state) {
393 case atos: pop_ptr(); break;
394 case btos:
395 case ztos:
396 case ctos:
397 case stos:
398 case itos: pop_i(); break;
399 case ltos: pop_l(); break;
400 case ftos: pop_f(); break;
401 case dtos: pop_d(); break;
402 case vtos: /* nothing to do */ break;
403 default: ShouldNotReachHere();
404 }
405 interp_verify_oop(r0, state);
406 }
407
408 void InterpreterMacroAssembler::push(TosState state) {
409 interp_verify_oop(r0, state);
410 switch (state) {
411 case atos: push_ptr(); break;
412 case btos:
413 case ztos:
414 case ctos:
415 case stos:
416 case itos: push_i(); break;
417 case ltos: push_l(); break;
418 case ftos: push_f(); break;
419 case dtos: push_d(); break;
420 case vtos: /* nothing to do */ break;
421 default : ShouldNotReachHere();
422 }
423 }
424
425 // Helpers for swap and dup
426 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
427 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
428 }
429
430 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
431 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
432 }
433
434 void InterpreterMacroAssembler::load_float(Address src) {
435 ldrs(v0, src);
436 }
437
438 void InterpreterMacroAssembler::load_double(Address src) {
439 ldrd(v0, src);
440 }
441
442 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
443 // set sender sp
444 mov(r19_sender_sp, sp);
445 // record last_sp
446 sub(rscratch1, esp, rfp);
447 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
448 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
449 }
450
451 // Jump to from_interpreted entry of a call unless single stepping is possible
452 // in this thread in which case we must call the i2i entry
453 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
454 prepare_to_jump_from_interpreted();
455
456 if (JvmtiExport::can_post_interpreter_events()) {
457 Label run_compiled_code;
458 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
459 // compiled code in threads for which the event is enabled. Check here for
460 // interp_only_mode if these events CAN be enabled.
461 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
462 cbzw(rscratch1, run_compiled_code);
463 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
464 br(rscratch1);
465 bind(run_compiled_code);
466 }
467
468 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
469 br(rscratch1);
470 }
471
472 // The following two routines provide a hook so that an implementation
473 // can schedule the dispatch in two parts. amd64 does not do this.
474 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
475 }
476
477 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
478 dispatch_next(state, step);
479 }
480
481 void InterpreterMacroAssembler::dispatch_base(TosState state,
482 address* table,
483 bool verifyoop,
484 bool generate_poll) {
485 if (VerifyActivationFrameSize) {
486 Label L;
487 sub(rscratch2, rfp, esp);
488 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
489 subs(rscratch2, rscratch2, min_frame_size);
490 br(Assembler::GE, L);
491 stop("broken stack frame");
492 bind(L);
493 }
494 if (verifyoop) {
495 interp_verify_oop(r0, state);
496 }
497
498 Label safepoint;
499 address* const safepoint_table = Interpreter::safept_table(state);
500 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
501
502 if (needs_thread_local_poll) {
503 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
504 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
505 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
506 }
507
508 if (table == Interpreter::dispatch_table(state)) {
509 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
510 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
511 } else {
512 mov(rscratch2, (address)table);
513 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
514 }
515 br(rscratch2);
516
517 if (needs_thread_local_poll) {
518 bind(safepoint);
519 lea(rscratch2, ExternalAddress((address)safepoint_table));
520 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
521 br(rscratch2);
522 }
523 }
524
525 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
526 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
527 }
528
529 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
530 dispatch_base(state, Interpreter::normal_table(state));
531 }
532
533 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
534 dispatch_base(state, Interpreter::normal_table(state), false);
535 }
536
537
538 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
539 // load next bytecode
540 ldrb(rscratch1, Address(pre(rbcp, step)));
541 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
542 }
543
544 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
545 // load current bytecode
546 ldrb(rscratch1, Address(rbcp, 0));
547 dispatch_base(state, table);
548 }
549
550 // remove activation
551 //
552 // Unlock the receiver if this is a synchronized method.
553 // Unlock any Java monitors from synchronized blocks.
554 // Apply stack watermark barrier.
555 // Notify JVMTI.
556 // Remove the activation from the stack.
557 //
558 // If there are locked Java monitors
559 // If throw_monitor_exception
560 // throws IllegalMonitorStateException
561 // Else if install_monitor_exception
562 // installs IllegalMonitorStateException
563 // Else
564 // no error processing
565 void InterpreterMacroAssembler::remove_activation(TosState state,
566 bool throw_monitor_exception,
567 bool install_monitor_exception,
568 bool notify_jvmdi) {
569 // Note: Registers r3 xmm0 may be in use for the
570 // result check if synchronized method
571 Label unlocked, unlock, no_unlock;
572
573 // get the value of _do_not_unlock_if_synchronized into r3
574 const Address do_not_unlock_if_synchronized(rthread,
575 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
576 ldrb(r3, do_not_unlock_if_synchronized);
577 strb(zr, do_not_unlock_if_synchronized); // reset the flag
578
579 // get method access flags
580 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
581 ldrh(r2, Address(r1, Method::access_flags_offset()));
582 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
583
584 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
585 // is set.
586 cbnz(r3, no_unlock);
587
588 // unlock monitor
589 push(state); // save result
590
591 // BasicObjectLock will be first in list, since this is a
592 // synchronized method. However, need to check that the object has
593 // not been unlocked by an explicit monitorexit bytecode.
594 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
595 wordSize - (int) sizeof(BasicObjectLock));
596 // We use c_rarg1 so that if we go slow path it will be the correct
597 // register for unlock_object to pass to VM directly
598 lea(c_rarg1, monitor); // address of first monitor
599
600 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
601 cbnz(r0, unlock);
602
603 pop(state);
604 if (throw_monitor_exception) {
605 // Entry already unlocked, need to throw exception
606 call_VM(noreg, CAST_FROM_FN_PTR(address,
607 InterpreterRuntime::throw_illegal_monitor_state_exception));
608 should_not_reach_here();
609 } else {
610 // Monitor already unlocked during a stack unroll. If requested,
611 // install an illegal_monitor_state_exception. Continue with
612 // stack unrolling.
613 if (install_monitor_exception) {
614 call_VM(noreg, CAST_FROM_FN_PTR(address,
615 InterpreterRuntime::new_illegal_monitor_state_exception));
616 }
617 b(unlocked);
618 }
619
620 bind(unlock);
621 unlock_object(c_rarg1);
622 pop(state);
623
624 // Check that for block-structured locking (i.e., that all locked
625 // objects has been unlocked)
626 bind(unlocked);
627
628 // r0: Might contain return value
629
630 // Check that all monitors are unlocked
631 {
632 Label loop, exception, entry, restart;
633 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
634 const Address monitor_block_top(
635 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
636 const Address monitor_block_bot(
637 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
638
639 bind(restart);
640 // We use c_rarg1 so that if we go slow path it will be the correct
641 // register for unlock_object to pass to VM directly
642 ldr(c_rarg1, monitor_block_top); // derelativize pointer
643 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
644 // c_rarg1 points to current entry, starting with top-most entry
645
646 lea(r19, monitor_block_bot); // points to word before bottom of
647 // monitor block
648 b(entry);
649
650 // Entry already locked, need to throw exception
651 bind(exception);
652
653 if (throw_monitor_exception) {
654 // Throw exception
655 MacroAssembler::call_VM(noreg,
656 CAST_FROM_FN_PTR(address, InterpreterRuntime::
657 throw_illegal_monitor_state_exception));
658 should_not_reach_here();
659 } else {
660 // Stack unrolling. Unlock object and install illegal_monitor_exception.
661 // Unlock does not block, so don't have to worry about the frame.
662 // We don't have to preserve c_rarg1 since we are going to throw an exception.
663
664 push(state);
665 unlock_object(c_rarg1);
666 pop(state);
667
668 if (install_monitor_exception) {
669 call_VM(noreg, CAST_FROM_FN_PTR(address,
670 InterpreterRuntime::
671 new_illegal_monitor_state_exception));
672 }
673
674 b(restart);
675 }
676
677 bind(loop);
678 // check if current entry is used
679 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
680 cbnz(rscratch1, exception);
681
682 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
683 bind(entry);
684 cmp(c_rarg1, r19); // check if bottom reached
685 br(Assembler::NE, loop); // if not at bottom then check this entry
686 }
687
688 bind(no_unlock);
689
690 JFR_ONLY(enter_jfr_critical_section();)
691
692 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
693 // that would normally not be safe to use. Such bad returns into unsafe territory of
694 // the stack, will call InterpreterRuntime::at_unwind.
695 Label slow_path;
696 Label fast_path;
697 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
698 br(Assembler::AL, fast_path);
699 bind(slow_path);
700 push(state);
701 set_last_Java_frame(esp, rfp, pc(), rscratch1);
702 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
703 reset_last_Java_frame(true);
704 pop(state);
705 bind(fast_path);
706
707 // JVMTI support. Make sure the safepoint poll test is issued prior.
708 if (notify_jvmdi) {
709 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
710 } else {
711 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
712 }
713
714 // remove activation
715 // get sender esp
716 ldr(rscratch2,
717 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
718
719 if (StackReservedPages > 0) {
720 // testing if reserved zone needs to be re-enabled
721 Label no_reserved_zone_enabling;
722
723 // check if already enabled - if so no re-enabling needed
724 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
725 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
726 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
727 br(Assembler::EQ, no_reserved_zone_enabling);
728
729 // look for an overflow into the stack reserved zone, i.e.
730 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
731 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
732 cmp(rscratch2, rscratch1);
733 br(Assembler::LS, no_reserved_zone_enabling);
734
735 JFR_ONLY(leave_jfr_critical_section();)
736
737 call_VM_leaf(
738 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
739 call_VM(noreg, CAST_FROM_FN_PTR(address,
740 InterpreterRuntime::throw_delayed_StackOverflowError));
741 should_not_reach_here();
742
743 bind(no_reserved_zone_enabling);
744 }
745
746 if (state == atos && InlineTypeReturnedAsFields) {
747 Label skip;
748 Label not_null;
749 cbnz(r0, not_null);
750 // Returned value is null, zero all return registers because they may belong to oop fields
751 mov(j_rarg1, zr);
752 mov(j_rarg2, zr);
753 mov(j_rarg3, zr);
754 mov(j_rarg4, zr);
755 mov(j_rarg5, zr);
756 mov(j_rarg6, zr);
757 mov(j_rarg7, zr);
758 b(skip);
759 bind(not_null);
760
761 // Check if we are returning an non-null inline type and load its fields into registers
762 test_oop_is_not_inline_type(r0, rscratch2, skip, /* can_be_null= */ false);
763
764 // Load fields from a buffered value with an inline class specific handler
765 load_klass(rscratch1 /*dst*/, r0 /*src*/);
766 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
767 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
768 // Unpack handler can be null if inline type is not scalarizable in returns
769 cbz(rscratch1, skip);
770
771 blr(rscratch1);
772 #ifdef ASSERT
773 // TODO 8284443 Enable
774 if (StressCallingConvention && false) {
775 Label skip_stress;
776 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
777 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
778 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
779 br(Assembler::EQ, skip_stress);
780 load_klass(r0, r0);
781 orr(r0, r0, 1);
782 bind(skip_stress);
783 }
784 #endif
785 bind(skip);
786 // Check above kills sender esp in rscratch2. Reload it.
787 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
788 }
789
790 // remove frame anchor
791 leave();
792
793 JFR_ONLY(leave_jfr_critical_section();)
794
795 // restore sender esp
796 mov(esp, rscratch2);
797
798 // If we're returning to interpreted code we will shortly be
799 // adjusting SP to allow some space for ESP. If we're returning to
800 // compiled code the saved sender SP was saved in sender_sp, so this
801 // restores it.
802 andr(sp, esp, -16);
803 }
804
805 #if INCLUDE_JFR
806 void InterpreterMacroAssembler::enter_jfr_critical_section() {
807 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
808 mov(rscratch1, true);
809 strb(rscratch1, sampling_critical_section);
810 }
811
812 void InterpreterMacroAssembler::leave_jfr_critical_section() {
813 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
814 strb(zr, sampling_critical_section);
815 }
816 #endif // INCLUDE_JFR
817
818 // Lock object
819 //
820 // Args:
821 // c_rarg1: BasicObjectLock to be used for locking
822 //
823 // Kills:
824 // r0
825 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
826 // rscratch1, rscratch2 (scratch regs)
827 void InterpreterMacroAssembler::lock_object(Register lock_reg)
828 {
829 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
830
831 const Register tmp = c_rarg2;
832 const Register obj_reg = c_rarg3; // Will contain the oop
833 const Register tmp2 = c_rarg4;
834 const Register tmp3 = c_rarg5;
835
836 // Load object pointer into obj_reg %c_rarg3
837 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
838
839 Label slow_case, done;
840 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
841 b(done);
842
843 bind(slow_case);
844
845 // Call the runtime routine for slow case
846 call_VM_preemptable(noreg,
847 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
848 lock_reg);
849
850 bind(done);
851 }
852
853
854 // Unlocks an object. Used in monitorexit bytecode and
855 // remove_activation. Throws an IllegalMonitorException if object is
856 // not locked by current thread.
857 //
858 // Args:
859 // c_rarg1: BasicObjectLock for lock
860 //
861 // Kills:
862 // r0
863 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
864 // rscratch1, rscratch2 (scratch regs)
865 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
866 {
867 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
868
869 const Register swap_reg = r0;
870 const Register header_reg = c_rarg2; // Will contain the old oopMark
871 const Register obj_reg = c_rarg3; // Will contain the oop
872 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
873
874 save_bcp(); // Save in case of exception
875
876 // Load oop into obj_reg(%c_rarg3)
877 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
878
879 // Free entry
880 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
881
882 Label slow_case, done;
883 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
884 b(done);
885
886 bind(slow_case);
887 // Call the runtime routine for slow case.
888 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
889 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
890 bind(done);
891 restore_bcp();
892 }
893
894 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
895 Label& zero_continue) {
896 assert(ProfileInterpreter, "must be profiling interpreter");
897 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
898 cbz(mdp, zero_continue);
899 }
900
901 // Set the method data pointer for the current bcp.
902 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
903 assert(ProfileInterpreter, "must be profiling interpreter");
904 Label set_mdp;
905 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
906
907 // Test MDO to avoid the call if it is null.
908 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
909 cbz(r0, set_mdp);
910 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
911 // r0: mdi
912 // mdo is guaranteed to be non-zero here, we checked for it before the call.
913 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
914 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
915 add(r0, r1, r0);
916 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
917 bind(set_mdp);
918 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
919 }
920
921 void InterpreterMacroAssembler::verify_method_data_pointer() {
922 assert(ProfileInterpreter, "must be profiling interpreter");
923 #ifdef ASSERT
924 Label verify_continue;
925 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
926 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
927 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
928 get_method(r1);
929
930 // If the mdp is valid, it will point to a DataLayout header which is
931 // consistent with the bcp. The converse is highly probable also.
932 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
933 ldr(rscratch1, Address(r1, Method::const_offset()));
934 add(r2, r2, rscratch1, Assembler::LSL);
935 lea(r2, Address(r2, ConstMethod::codes_offset()));
936 cmp(r2, rbcp);
937 br(Assembler::EQ, verify_continue);
938 // r1: method
939 // rbcp: bcp // rbcp == 22
940 // r3: mdp
941 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
942 r1, rbcp, r3);
943 bind(verify_continue);
944 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
945 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
946 #endif // ASSERT
947 }
948
949
950 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
951 int constant,
952 Register value) {
953 assert(ProfileInterpreter, "must be profiling interpreter");
954 Address data(mdp_in, constant);
955 str(value, data);
956 }
957
958
959 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
960 int constant) {
961 increment_mdp_data_at(mdp_in, noreg, constant);
962 }
963
964 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
965 Register index,
966 int constant) {
967 assert(ProfileInterpreter, "must be profiling interpreter");
968
969 assert_different_registers(rscratch2, rscratch1, mdp_in, index);
970
971 Address addr1(mdp_in, constant);
972 Address addr2(rscratch2, index, Address::lsl(0));
973 Address &addr = addr1;
974 if (index != noreg) {
975 lea(rscratch2, addr1);
976 addr = addr2;
977 }
978
979 increment(addr, DataLayout::counter_increment);
980 }
981
982 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
983 int flag_byte_constant) {
984 assert(ProfileInterpreter, "must be profiling interpreter");
985 int flags_offset = in_bytes(DataLayout::flags_offset());
986 // Set the flag
987 ldrb(rscratch1, Address(mdp_in, flags_offset));
988 orr(rscratch1, rscratch1, flag_byte_constant);
989 strb(rscratch1, Address(mdp_in, flags_offset));
990 }
991
992
993 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
994 int offset,
995 Register value,
996 Register test_value_out,
997 Label& not_equal_continue) {
998 assert(ProfileInterpreter, "must be profiling interpreter");
999 if (test_value_out == noreg) {
1000 ldr(rscratch1, Address(mdp_in, offset));
1001 cmp(value, rscratch1);
1002 } else {
1003 // Put the test value into a register, so caller can use it:
1004 ldr(test_value_out, Address(mdp_in, offset));
1005 cmp(value, test_value_out);
1006 }
1007 br(Assembler::NE, not_equal_continue);
1008 }
1009
1010
1011 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1012 int offset_of_disp) {
1013 assert(ProfileInterpreter, "must be profiling interpreter");
1014 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1015 add(mdp_in, mdp_in, rscratch1, LSL);
1016 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1017 }
1018
1019
1020 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1021 Register reg,
1022 int offset_of_disp) {
1023 assert(ProfileInterpreter, "must be profiling interpreter");
1024 lea(rscratch1, Address(mdp_in, offset_of_disp));
1025 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1026 add(mdp_in, mdp_in, rscratch1, LSL);
1027 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1028 }
1029
1030
1031 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1032 int constant) {
1033 assert(ProfileInterpreter, "must be profiling interpreter");
1034 add(mdp_in, mdp_in, (unsigned)constant);
1035 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1036 }
1037
1038
1039 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1040 assert(ProfileInterpreter, "must be profiling interpreter");
1041 // save/restore across call_VM
1042 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1043 call_VM(noreg,
1044 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1045 return_bci);
1046 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1047 }
1048
1049
1050 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
1051 if (ProfileInterpreter) {
1052 Label profile_continue;
1053
1054 // If no method data exists, go to profile_continue.
1055 test_method_data_pointer(mdp, profile_continue);
1056
1057 // We are taking a branch. Increment the taken count.
1058 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1059
1060 // The method data pointer needs to be updated to reflect the new target.
1061 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1062 bind(profile_continue);
1063 }
1064 }
1065
1066
1067 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1068 if (ProfileInterpreter) {
1069 Label profile_continue;
1070
1071 // If no method data exists, go to profile_continue.
1072 test_method_data_pointer(mdp, profile_continue);
1073
1074 // We are not taking a branch. Increment the not taken count.
1075 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1076
1077 // The method data pointer needs to be updated to correspond to
1078 // the next bytecode
1079 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1080 bind(profile_continue);
1081 }
1082 }
1083
1084
1085 void InterpreterMacroAssembler::profile_call(Register mdp) {
1086 if (ProfileInterpreter) {
1087 Label profile_continue;
1088
1089 // If no method data exists, go to profile_continue.
1090 test_method_data_pointer(mdp, profile_continue);
1091
1092 // We are making a call. Increment the count.
1093 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1094
1095 // The method data pointer needs to be updated to reflect the new target.
1096 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1097 bind(profile_continue);
1098 }
1099 }
1100
1101 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1102 if (ProfileInterpreter) {
1103 Label profile_continue;
1104
1105 // If no method data exists, go to profile_continue.
1106 test_method_data_pointer(mdp, profile_continue);
1107
1108 // We are making a call. Increment the count.
1109 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1110
1111 // The method data pointer needs to be updated to reflect the new target.
1112 update_mdp_by_constant(mdp,
1113 in_bytes(VirtualCallData::
1114 virtual_call_data_size()));
1115 bind(profile_continue);
1116 }
1117 }
1118
1119
1120 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1121 Register mdp,
1122 Register reg2,
1123 bool receiver_can_be_null) {
1124 if (ProfileInterpreter) {
1125 Label profile_continue;
1126
1127 // If no method data exists, go to profile_continue.
1128 test_method_data_pointer(mdp, profile_continue);
1129
1130 Label skip_receiver_profile;
1131 if (receiver_can_be_null) {
1132 Label not_null;
1133 // We are making a call. Increment the count for null receiver.
1134 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1135 b(skip_receiver_profile);
1136 bind(not_null);
1137 }
1138
1139 // Record the receiver type.
1140 record_klass_in_profile(receiver, mdp, reg2);
1141 bind(skip_receiver_profile);
1142
1143 // The method data pointer needs to be updated to reflect the new target.
1144 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1145 bind(profile_continue);
1146 }
1147 }
1148
1149 // This routine creates a state machine for updating the multi-row
1150 // type profile at a virtual call site (or other type-sensitive bytecode).
1151 // The machine visits each row (of receiver/count) until the receiver type
1152 // is found, or until it runs out of rows. At the same time, it remembers
1153 // the location of the first empty row. (An empty row records null for its
1154 // receiver, and can be allocated for a newly-observed receiver type.)
1155 // Because there are two degrees of freedom in the state, a simple linear
1156 // search will not work; it must be a decision tree. Hence this helper
1157 // function is recursive, to generate the required tree structured code.
1158 // It's the interpreter, so we are trading off code space for speed.
1159 // See below for example code.
1160 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1161 Register receiver, Register mdp,
1162 Register reg2, int start_row,
1163 Label& done) {
1164 if (TypeProfileWidth == 0) {
1165 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1166 } else {
1167 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1168 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1169 }
1170 }
1171
1172 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1173 Register reg2, int start_row, Label& done, int total_rows,
1174 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1175 int last_row = total_rows - 1;
1176 assert(start_row <= last_row, "must be work left to do");
1177 // Test this row for both the item and for null.
1178 // Take any of three different outcomes:
1179 // 1. found item => increment count and goto done
1180 // 2. found null => keep looking for case 1, maybe allocate this cell
1181 // 3. found something else => keep looking for cases 1 and 2
1182 // Case 3 is handled by a recursive call.
1183 for (int row = start_row; row <= last_row; row++) {
1184 Label next_test;
1185 bool test_for_null_also = (row == start_row);
1186
1187 // See if the item is item[n].
1188 int item_offset = in_bytes(item_offset_fn(row));
1189 test_mdp_data_at(mdp, item_offset, item,
1190 (test_for_null_also ? reg2 : noreg),
1191 next_test);
1192 // (Reg2 now contains the item from the CallData.)
1193
1194 // The item is item[n]. Increment count[n].
1195 int count_offset = in_bytes(item_count_offset_fn(row));
1196 increment_mdp_data_at(mdp, count_offset);
1197 b(done);
1198 bind(next_test);
1199
1200 if (test_for_null_also) {
1201 Label found_null;
1202 // Failed the equality check on item[n]... Test for null.
1203 if (start_row == last_row) {
1204 // The only thing left to do is handle the null case.
1205 cbz(reg2, found_null);
1206 // Item did not match any saved item and there is no empty row for it.
1207 // Increment total counter to indicate polymorphic case.
1208 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1209 b(done);
1210 bind(found_null);
1211 break;
1212 }
1213 // Since null is rare, make it be the branch-taken case.
1214 cbz(reg2, found_null);
1215
1216 // Put all the "Case 3" tests here.
1217 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1218 item_offset_fn, item_count_offset_fn);
1219
1220 // Found a null. Keep searching for a matching item,
1221 // but remember that this is an empty (unused) slot.
1222 bind(found_null);
1223 }
1224 }
1225
1226 // In the fall-through case, we found no matching item, but we
1227 // observed the item[start_row] is null.
1228
1229 // Fill in the item field and increment the count.
1230 int item_offset = in_bytes(item_offset_fn(start_row));
1231 set_mdp_data_at(mdp, item_offset, item);
1232 int count_offset = in_bytes(item_count_offset_fn(start_row));
1233 mov(reg2, DataLayout::counter_increment);
1234 set_mdp_data_at(mdp, count_offset, reg2);
1235 if (start_row > 0) {
1236 b(done);
1237 }
1238 }
1239
1240 // Example state machine code for three profile rows:
1241 // // main copy of decision tree, rooted at row[1]
1242 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1243 // if (row[0].rec != nullptr) {
1244 // // inner copy of decision tree, rooted at row[1]
1245 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1246 // if (row[1].rec != nullptr) {
1247 // // degenerate decision tree, rooted at row[2]
1248 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1249 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1250 // row[2].init(rec); goto done;
1251 // } else {
1252 // // remember row[1] is empty
1253 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1254 // row[1].init(rec); goto done;
1255 // }
1256 // } else {
1257 // // remember row[0] is empty
1258 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1259 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1260 // row[0].init(rec); goto done;
1261 // }
1262 // done:
1263
1264 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1265 Register mdp, Register reg2) {
1266 assert(ProfileInterpreter, "must be profiling");
1267 Label done;
1268
1269 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1270
1271 bind (done);
1272 }
1273
1274 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1275 Register mdp) {
1276 if (ProfileInterpreter) {
1277 Label profile_continue;
1278 uint row;
1279
1280 // If no method data exists, go to profile_continue.
1281 test_method_data_pointer(mdp, profile_continue);
1282
1283 // Update the total ret count.
1284 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1285
1286 for (row = 0; row < RetData::row_limit(); row++) {
1287 Label next_test;
1288
1289 // See if return_bci is equal to bci[n]:
1290 test_mdp_data_at(mdp,
1291 in_bytes(RetData::bci_offset(row)),
1292 return_bci, noreg,
1293 next_test);
1294
1295 // return_bci is equal to bci[n]. Increment the count.
1296 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1297
1298 // The method data pointer needs to be updated to reflect the new target.
1299 update_mdp_by_offset(mdp,
1300 in_bytes(RetData::bci_displacement_offset(row)));
1301 b(profile_continue);
1302 bind(next_test);
1303 }
1304
1305 update_mdp_for_ret(return_bci);
1306
1307 bind(profile_continue);
1308 }
1309 }
1310
1311 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1312 if (ProfileInterpreter) {
1313 Label profile_continue;
1314
1315 // If no method data exists, go to profile_continue.
1316 test_method_data_pointer(mdp, profile_continue);
1317
1318 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1319
1320 // The method data pointer needs to be updated.
1321 int mdp_delta = in_bytes(BitData::bit_data_size());
1322 if (TypeProfileCasts) {
1323 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1324 }
1325 update_mdp_by_constant(mdp, mdp_delta);
1326
1327 bind(profile_continue);
1328 }
1329 }
1330
1331 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1332 if (ProfileInterpreter) {
1333 Label profile_continue;
1334
1335 // If no method data exists, go to profile_continue.
1336 test_method_data_pointer(mdp, profile_continue);
1337
1338 // The method data pointer needs to be updated.
1339 int mdp_delta = in_bytes(BitData::bit_data_size());
1340 if (TypeProfileCasts) {
1341 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1342
1343 // Record the object type.
1344 record_klass_in_profile(klass, mdp, reg2);
1345 }
1346 update_mdp_by_constant(mdp, mdp_delta);
1347
1348 bind(profile_continue);
1349 }
1350 }
1351
1352 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1353 if (ProfileInterpreter) {
1354 Label profile_continue;
1355
1356 // If no method data exists, go to profile_continue.
1357 test_method_data_pointer(mdp, profile_continue);
1358
1359 // Update the default case count
1360 increment_mdp_data_at(mdp,
1361 in_bytes(MultiBranchData::default_count_offset()));
1362
1363 // The method data pointer needs to be updated.
1364 update_mdp_by_offset(mdp,
1365 in_bytes(MultiBranchData::
1366 default_displacement_offset()));
1367
1368 bind(profile_continue);
1369 }
1370 }
1371
1372 void InterpreterMacroAssembler::profile_switch_case(Register index,
1373 Register mdp,
1374 Register reg2) {
1375 if (ProfileInterpreter) {
1376 Label profile_continue;
1377
1378 // If no method data exists, go to profile_continue.
1379 test_method_data_pointer(mdp, profile_continue);
1380
1381 // Build the base (index * per_case_size_in_bytes()) +
1382 // case_array_offset_in_bytes()
1383 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1384 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1385 Assembler::maddw(index, index, reg2, rscratch1);
1386
1387 // Update the case count
1388 increment_mdp_data_at(mdp,
1389 index,
1390 in_bytes(MultiBranchData::relative_count_offset()));
1391
1392 // The method data pointer needs to be updated.
1393 update_mdp_by_offset(mdp,
1394 index,
1395 in_bytes(MultiBranchData::
1396 relative_displacement_offset()));
1397
1398 bind(profile_continue);
1399 }
1400 }
1401
1402 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1403 Register array,
1404 Register tmp) {
1405 if (ProfileInterpreter) {
1406 Label profile_continue;
1407
1408 // If no method data exists, go to profile_continue.
1409 test_method_data_pointer(mdp, profile_continue);
1410
1411 mov(tmp, array);
1412 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1413
1414 Label not_flat;
1415 test_non_flat_array_oop(array, tmp, not_flat);
1416
1417 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1418
1419 bind(not_flat);
1420
1421 Label not_null_free;
1422 test_non_null_free_array_oop(array, tmp, not_null_free);
1423
1424 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1425
1426 bind(not_null_free);
1427
1428 bind(profile_continue);
1429 }
1430 }
1431
1432 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1433 Register array,
1434 Register tmp);
1435 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1436 Register array,
1437 Register tmp);
1438
1439 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1440 if (ProfileInterpreter) {
1441 Label profile_continue;
1442
1443 // If no method data exists, go to profile_continue.
1444 test_method_data_pointer(mdp, profile_continue);
1445
1446 Label done, update;
1447 cbnz(element, update);
1448 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1449 b(done);
1450
1451 bind(update);
1452 load_klass(tmp, element);
1453
1454 // Record the object type.
1455 record_klass_in_profile(tmp, mdp, tmp2);
1456
1457 bind(done);
1458
1459 // The method data pointer needs to be updated.
1460 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1461
1462 bind(profile_continue);
1463 }
1464 }
1465
1466
1467 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1468 Register element,
1469 Register tmp) {
1470 if (ProfileInterpreter) {
1471 Label profile_continue;
1472
1473 // If no method data exists, go to profile_continue.
1474 test_method_data_pointer(mdp, profile_continue);
1475
1476 mov(tmp, element);
1477 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1478
1479 // The method data pointer needs to be updated.
1480 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1481
1482 bind(profile_continue);
1483 }
1484 }
1485
1486 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1487 Register left,
1488 Register right,
1489 Register tmp) {
1490 if (ProfileInterpreter) {
1491 Label profile_continue;
1492
1493 // If no method data exists, go to profile_continue.
1494 test_method_data_pointer(mdp, profile_continue);
1495
1496 mov(tmp, left);
1497 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1498
1499 Label left_not_inline_type;
1500 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1501 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1502 bind(left_not_inline_type);
1503
1504 mov(tmp, right);
1505 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1506
1507 Label right_not_inline_type;
1508 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1509 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1510 bind(right_not_inline_type);
1511
1512 bind(profile_continue);
1513 }
1514 }
1515
1516 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1517 if (state == atos) {
1518 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1519 }
1520 }
1521
1522 void InterpreterMacroAssembler::notify_method_entry() {
1523 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1524 // track stack depth. If it is possible to enter interp_only_mode we add
1525 // the code to check if the event should be sent.
1526 if (JvmtiExport::can_post_interpreter_events()) {
1527 Label L;
1528 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1529 cbzw(r3, L);
1530 call_VM(noreg, CAST_FROM_FN_PTR(address,
1531 InterpreterRuntime::post_method_entry));
1532 bind(L);
1533 }
1534
1535 if (DTraceMethodProbes) {
1536 get_method(c_rarg1);
1537 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1538 rthread, c_rarg1);
1539 }
1540
1541 // RedefineClasses() tracing support for obsolete method entry
1542 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1543 get_method(c_rarg1);
1544 call_VM_leaf(
1545 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1546 rthread, c_rarg1);
1547 }
1548
1549 }
1550
1551
1552 void InterpreterMacroAssembler::notify_method_exit(
1553 TosState state, NotifyMethodExitMode mode) {
1554 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1555 // track stack depth. If it is possible to enter interp_only_mode we add
1556 // the code to check if the event should be sent.
1557 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1558 Label L;
1559 // Note: frame::interpreter_frame_result has a dependency on how the
1560 // method result is saved across the call to post_method_exit. If this
1561 // is changed then the interpreter_frame_result implementation will
1562 // need to be updated too.
1563
1564 // template interpreter will leave the result on the top of the stack.
1565 push(state);
1566 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1567 cbz(r3, L);
1568 call_VM(noreg,
1569 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1570 bind(L);
1571 pop(state);
1572 }
1573
1574 if (DTraceMethodProbes) {
1575 push(state);
1576 get_method(c_rarg1);
1577 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1578 rthread, c_rarg1);
1579 pop(state);
1580 }
1581 }
1582
1583
1584 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1585 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1586 int increment, Address mask,
1587 Register scratch, Register scratch2,
1588 bool preloaded, Condition cond,
1589 Label* where) {
1590 if (!preloaded) {
1591 ldrw(scratch, counter_addr);
1592 }
1593 add(scratch, scratch, increment);
1594 strw(scratch, counter_addr);
1595 ldrw(scratch2, mask);
1596 ands(scratch, scratch, scratch2);
1597 br(cond, *where);
1598 }
1599
1600 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1601 int number_of_arguments) {
1602 // interpreter specific
1603 //
1604 // Note: No need to save/restore rbcp & rlocals pointer since these
1605 // are callee saved registers and no blocking/ GC can happen
1606 // in leaf calls.
1607 #ifdef ASSERT
1608 {
1609 Label L;
1610 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1611 cbz(rscratch1, L);
1612 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1613 " last_sp != nullptr");
1614 bind(L);
1615 }
1616 #endif /* ASSERT */
1617 // super call
1618 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1619 }
1620
1621 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1622 Register java_thread,
1623 Register last_java_sp,
1624 address entry_point,
1625 int number_of_arguments,
1626 bool check_exceptions) {
1627 // interpreter specific
1628 //
1629 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1630 // really make a difference for these runtime calls, since they are
1631 // slow anyway. Btw., bcp must be saved/restored since it may change
1632 // due to GC.
1633 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1634 save_bcp();
1635 #ifdef ASSERT
1636 {
1637 Label L;
1638 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1639 cbz(rscratch1, L);
1640 stop("InterpreterMacroAssembler::call_VM_base:"
1641 " last_sp != nullptr");
1642 bind(L);
1643 }
1644 #endif /* ASSERT */
1645 // super call
1646 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1647 entry_point, number_of_arguments,
1648 check_exceptions);
1649 // interpreter specific
1650 restore_bcp();
1651 restore_locals();
1652 }
1653
1654 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1655 address entry_point,
1656 Register arg_1) {
1657 assert(arg_1 == c_rarg1, "");
1658 Label resume_pc, not_preempted;
1659
1660 #ifdef ASSERT
1661 {
1662 Label L;
1663 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1664 cbz(rscratch1, L);
1665 stop("Should not have alternate return address set");
1666 bind(L);
1667 }
1668 #endif /* ASSERT */
1669
1670 // Force freeze slow path.
1671 push_cont_fastpath();
1672
1673 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1674 adr(rscratch1, resume_pc);
1675 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1676 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1677
1678 pop_cont_fastpath();
1679
1680 // Check if preempted.
1681 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1682 cbz(rscratch1, not_preempted);
1683 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1684 br(rscratch1);
1685
1686 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1687 bind(resume_pc);
1688 restore_after_resume(false /* is_native */);
1689
1690 bind(not_preempted);
1691 }
1692
1693 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1694 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1695 blr(rscratch1);
1696 if (is_native) {
1697 // On resume we need to set up stack as expected
1698 push(dtos);
1699 push(ltos);
1700 }
1701 }
1702
1703 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1704 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1705 Label update, next, none;
1706
1707 verify_oop(obj);
1708
1709 cbnz(obj, update);
1710 orptr(mdo_addr, TypeEntries::null_seen);
1711 b(next);
1712
1713 bind(update);
1714 load_klass(obj, obj);
1715
1716 ldr(rscratch1, mdo_addr);
1717 eor(obj, obj, rscratch1);
1718 tst(obj, TypeEntries::type_klass_mask);
1719 br(Assembler::EQ, next); // klass seen before, nothing to
1720 // do. The unknown bit may have been
1721 // set already but no need to check.
1722
1723 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1724 // already unknown. Nothing to do anymore.
1725
1726 cbz(rscratch1, none);
1727 cmp(rscratch1, (u1)TypeEntries::null_seen);
1728 br(Assembler::EQ, none);
1729 // There is a chance that the checks above
1730 // fail if another thread has just set the
1731 // profiling to this obj's klass
1732 eor(obj, obj, rscratch1); // get back original value before XOR
1733 ldr(rscratch1, mdo_addr);
1734 eor(obj, obj, rscratch1);
1735 tst(obj, TypeEntries::type_klass_mask);
1736 br(Assembler::EQ, next);
1737
1738 // different than before. Cannot keep accurate profile.
1739 orptr(mdo_addr, TypeEntries::type_unknown);
1740 b(next);
1741
1742 bind(none);
1743 // first time here. Set profile type.
1744 str(obj, mdo_addr);
1745 #ifdef ASSERT
1746 andr(obj, obj, TypeEntries::type_mask);
1747 verify_klass_ptr(obj);
1748 #endif
1749
1750 bind(next);
1751 }
1752
1753 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1754 if (!ProfileInterpreter) {
1755 return;
1756 }
1757
1758 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1759 Label profile_continue;
1760
1761 test_method_data_pointer(mdp, profile_continue);
1762
1763 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1764
1765 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1766 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1767 br(Assembler::NE, profile_continue);
1768
1769 if (MethodData::profile_arguments()) {
1770 Label done;
1771 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1772
1773 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1774 if (i > 0 || MethodData::profile_return()) {
1775 // If return value type is profiled we may have no argument to profile
1776 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1777 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1778 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1779 add(rscratch1, mdp, off_to_args);
1780 br(Assembler::LT, done);
1781 }
1782 ldr(tmp, Address(callee, Method::const_offset()));
1783 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1784 // stack offset o (zero based) from the start of the argument
1785 // list, for n arguments translates into offset n - o - 1 from
1786 // the end of the argument list
1787 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1788 sub(tmp, tmp, rscratch1);
1789 sub(tmp, tmp, 1);
1790 Address arg_addr = argument_address(tmp);
1791 ldr(tmp, arg_addr);
1792
1793 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1794 profile_obj_type(tmp, mdo_arg_addr);
1795
1796 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1797 off_to_args += to_add;
1798 }
1799
1800 if (MethodData::profile_return()) {
1801 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1802 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1803 }
1804
1805 add(rscratch1, mdp, off_to_args);
1806 bind(done);
1807 mov(mdp, rscratch1);
1808
1809 if (MethodData::profile_return()) {
1810 // We're right after the type profile for the last
1811 // argument. tmp is the number of cells left in the
1812 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1813 // if there's a return to profile.
1814 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1815 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1816 }
1817 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1818 } else {
1819 assert(MethodData::profile_return(), "either profile call args or call ret");
1820 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1821 }
1822
1823 // mdp points right after the end of the
1824 // CallTypeData/VirtualCallTypeData, right after the cells for the
1825 // return value type if there's one
1826
1827 bind(profile_continue);
1828 }
1829 }
1830
1831 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1832 assert_different_registers(mdp, ret, tmp, rbcp);
1833 if (ProfileInterpreter && MethodData::profile_return()) {
1834 Label profile_continue, done;
1835
1836 test_method_data_pointer(mdp, profile_continue);
1837
1838 if (MethodData::profile_return_jsr292_only()) {
1839 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1840
1841 // If we don't profile all invoke bytecodes we must make sure
1842 // it's a bytecode we indeed profile. We can't go back to the
1843 // beginning of the ProfileData we intend to update to check its
1844 // type because we're right after it and we don't known its
1845 // length
1846 Label do_profile;
1847 ldrb(rscratch1, Address(rbcp, 0));
1848 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1849 br(Assembler::EQ, do_profile);
1850 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1851 br(Assembler::EQ, do_profile);
1852 get_method(tmp);
1853 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1854 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1855 br(Assembler::NE, profile_continue);
1856
1857 bind(do_profile);
1858 }
1859
1860 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1861 mov(tmp, ret);
1862 profile_obj_type(tmp, mdo_ret_addr);
1863
1864 bind(profile_continue);
1865 }
1866 }
1867
1868 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1869 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1870 if (ProfileInterpreter && MethodData::profile_parameters()) {
1871 Label profile_continue, done;
1872
1873 test_method_data_pointer(mdp, profile_continue);
1874
1875 // Load the offset of the area within the MDO used for
1876 // parameters. If it's negative we're not profiling any parameters
1877 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1878 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1879
1880 // Compute a pointer to the area for parameters from the offset
1881 // and move the pointer to the slot for the last
1882 // parameters. Collect profiling from last parameter down.
1883 // mdo start + parameters offset + array length - 1
1884 add(mdp, mdp, tmp1);
1885 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1886 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1887
1888 Label loop;
1889 bind(loop);
1890
1891 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1892 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1893 int per_arg_scale = exact_log2(DataLayout::cell_size);
1894 add(rscratch1, mdp, off_base);
1895 add(rscratch2, mdp, type_base);
1896
1897 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1898 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1899
1900 // load offset on the stack from the slot for this parameter
1901 ldr(tmp2, arg_off);
1902 neg(tmp2, tmp2);
1903 // read the parameter from the local area
1904 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1905
1906 // profile the parameter
1907 profile_obj_type(tmp2, arg_type);
1908
1909 // go to next parameter
1910 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1911 br(Assembler::GE, loop);
1912
1913 bind(profile_continue);
1914 }
1915 }
1916
1917 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1918 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1919 get_cache_index_at_bcp(index, 1, sizeof(u4));
1920 // Get address of invokedynamic array
1921 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1922 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1923 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1924 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1925 lea(cache, Address(cache, index));
1926 }
1927
1928 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1929 // Get index out of bytecode pointer
1930 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1931 // Take shortcut if the size is a power of 2
1932 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1933 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1934 } else {
1935 mov(cache, sizeof(ResolvedFieldEntry));
1936 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1937 }
1938 // Get address of field entries array
1939 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1940 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1941 lea(cache, Address(cache, index));
1942 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1943 membar(MacroAssembler::LoadLoad);
1944 }
1945
1946 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1947 // Get index out of bytecode pointer
1948 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1949 mov(cache, sizeof(ResolvedMethodEntry));
1950 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1951
1952 // Get address of field entries array
1953 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1954 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1955 lea(cache, Address(cache, index));
1956 }