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 if (LockingMode == LM_MONITOR) {
831 call_VM_preemptable(noreg,
832 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
833 lock_reg);
834 } else {
835 Label count, done;
836
837 const Register swap_reg = r0;
838 const Register tmp = c_rarg2;
839 const Register obj_reg = c_rarg3; // Will contain the oop
840 const Register tmp2 = c_rarg4;
841 const Register tmp3 = c_rarg5;
842
843 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
844 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
845 const int mark_offset = lock_offset +
846 BasicLock::displaced_header_offset_in_bytes();
847
848 Label slow_case;
849
850 // Load object pointer into obj_reg %c_rarg3
851 ldr(obj_reg, Address(lock_reg, obj_offset));
852
853 if (LockingMode == LM_LIGHTWEIGHT) {
854 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
855 b(done);
856 } else if (LockingMode == LM_LEGACY) {
857
858 if (DiagnoseSyncOnValueBasedClasses != 0) {
859 load_klass(tmp, obj_reg);
860 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
861 tst(tmp, KlassFlags::_misc_is_value_based_class);
862 br(Assembler::NE, slow_case);
863 }
864
865 // Load (object->mark() | 1) into swap_reg
866 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
867 orr(swap_reg, rscratch1, 1);
868 if (EnableValhalla) {
869 // Mask inline_type bit such that we go to the slow path if object is an inline type
870 andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
871 }
872
873 // Save (object->mark() | 1) into BasicLock's displaced header
874 str(swap_reg, Address(lock_reg, mark_offset));
875
876 assert(lock_offset == 0,
877 "displached header must be first word in BasicObjectLock");
878
879 Label fail;
880 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
881
882 // Fast check for recursive lock.
883 //
884 // Can apply the optimization only if this is a stack lock
885 // allocated in this thread. For efficiency, we can focus on
886 // recently allocated stack locks (instead of reading the stack
887 // base and checking whether 'mark' points inside the current
888 // thread stack):
889 // 1) (mark & 7) == 0, and
890 // 2) sp <= mark < mark + os::pagesize()
891 //
892 // Warning: sp + os::pagesize can overflow the stack base. We must
893 // neither apply the optimization for an inflated lock allocated
894 // just above the thread stack (this is why condition 1 matters)
895 // nor apply the optimization if the stack lock is inside the stack
896 // of another thread. The latter is avoided even in case of overflow
897 // because we have guard pages at the end of all stacks. Hence, if
898 // we go over the stack base and hit the stack of another thread,
899 // this should not be in a writeable area that could contain a
900 // stack lock allocated by that thread. As a consequence, a stack
901 // lock less than page size away from sp is guaranteed to be
902 // owned by the current thread.
903 //
904 // These 3 tests can be done by evaluating the following
905 // expression: ((mark - sp) & (7 - os::vm_page_size())),
906 // assuming both stack pointer and pagesize have their
907 // least significant 3 bits clear.
908 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
909 // NOTE2: aarch64 does not like to subtract sp from rn so take a
910 // copy
911 mov(rscratch1, sp);
912 sub(swap_reg, swap_reg, rscratch1);
913 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
914
915 // Save the test result, for recursive case, the result is zero
916 str(swap_reg, Address(lock_reg, mark_offset));
917 br(Assembler::NE, slow_case);
918
919 bind(count);
920 inc_held_monitor_count(rscratch1);
921 b(done);
922 }
923 bind(slow_case);
924
925 // Call the runtime routine for slow case
926 call_VM_preemptable(noreg,
927 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
928 lock_reg);
929
930 bind(done);
931 }
932 }
933
934
935 // Unlocks an object. Used in monitorexit bytecode and
936 // remove_activation. Throws an IllegalMonitorException if object is
937 // not locked by current thread.
938 //
939 // Args:
940 // c_rarg1: BasicObjectLock for lock
941 //
942 // Kills:
943 // r0
944 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
945 // rscratch1, rscratch2 (scratch regs)
946 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
947 {
948 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
949
950 if (LockingMode == LM_MONITOR) {
951 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
952 } else {
953 Label count, done;
954
955 const Register swap_reg = r0;
956 const Register header_reg = c_rarg2; // Will contain the old oopMark
957 const Register obj_reg = c_rarg3; // Will contain the oop
958 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
959
960 save_bcp(); // Save in case of exception
961
962 if (LockingMode != LM_LIGHTWEIGHT) {
963 // Convert from BasicObjectLock structure to object and BasicLock
964 // structure Store the BasicLock address into %r0
965 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
966 }
967
968 // Load oop into obj_reg(%c_rarg3)
969 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
970
971 // Free entry
972 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
973
974 Label slow_case;
975 if (LockingMode == LM_LIGHTWEIGHT) {
976 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
977 b(done);
978 } else if (LockingMode == LM_LEGACY) {
979 // Load the old header from BasicLock structure
980 ldr(header_reg, Address(swap_reg,
981 BasicLock::displaced_header_offset_in_bytes()));
982
983 // Test for recursion
984 cbz(header_reg, count);
985
986 // Atomic swap back the old header
987 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
988
989 bind(count);
990 dec_held_monitor_count(rscratch1);
991 b(done);
992 }
993
994 bind(slow_case);
995 // Call the runtime routine for slow case.
996 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
997 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
998 bind(done);
999 restore_bcp();
1000 }
1001 }
1002
1003 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1004 Label& zero_continue) {
1005 assert(ProfileInterpreter, "must be profiling interpreter");
1006 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1007 cbz(mdp, zero_continue);
1008 }
1009
1010 // Set the method data pointer for the current bcp.
1011 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1012 assert(ProfileInterpreter, "must be profiling interpreter");
1013 Label set_mdp;
1014 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
1015
1016 // Test MDO to avoid the call if it is null.
1017 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
1018 cbz(r0, set_mdp);
1019 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
1020 // r0: mdi
1021 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1022 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1023 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
1024 add(r0, r1, r0);
1025 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1026 bind(set_mdp);
1027 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
1028 }
1029
1030 void InterpreterMacroAssembler::verify_method_data_pointer() {
1031 assert(ProfileInterpreter, "must be profiling interpreter");
1032 #ifdef ASSERT
1033 Label verify_continue;
1034 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
1035 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
1036 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
1037 get_method(r1);
1038
1039 // If the mdp is valid, it will point to a DataLayout header which is
1040 // consistent with the bcp. The converse is highly probable also.
1041 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
1042 ldr(rscratch1, Address(r1, Method::const_offset()));
1043 add(r2, r2, rscratch1, Assembler::LSL);
1044 lea(r2, Address(r2, ConstMethod::codes_offset()));
1045 cmp(r2, rbcp);
1046 br(Assembler::EQ, verify_continue);
1047 // r1: method
1048 // rbcp: bcp // rbcp == 22
1049 // r3: mdp
1050 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1051 r1, rbcp, r3);
1052 bind(verify_continue);
1053 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
1054 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
1055 #endif // ASSERT
1056 }
1057
1058
1059 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1060 int constant,
1061 Register value) {
1062 assert(ProfileInterpreter, "must be profiling interpreter");
1063 Address data(mdp_in, constant);
1064 str(value, data);
1065 }
1066
1067
1068 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1069 int constant) {
1070 increment_mdp_data_at(mdp_in, noreg, constant);
1071 }
1072
1073 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1074 Register index,
1075 int constant) {
1076 assert(ProfileInterpreter, "must be profiling interpreter");
1077
1078 assert_different_registers(rscratch2, rscratch1, mdp_in, index);
1079
1080 Address addr1(mdp_in, constant);
1081 Address addr2(rscratch2, index, Address::lsl(0));
1082 Address &addr = addr1;
1083 if (index != noreg) {
1084 lea(rscratch2, addr1);
1085 addr = addr2;
1086 }
1087
1088 increment(addr, DataLayout::counter_increment);
1089 }
1090
1091 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1092 int flag_byte_constant) {
1093 assert(ProfileInterpreter, "must be profiling interpreter");
1094 int flags_offset = in_bytes(DataLayout::flags_offset());
1095 // Set the flag
1096 ldrb(rscratch1, Address(mdp_in, flags_offset));
1097 orr(rscratch1, rscratch1, flag_byte_constant);
1098 strb(rscratch1, Address(mdp_in, flags_offset));
1099 }
1100
1101
1102 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1103 int offset,
1104 Register value,
1105 Register test_value_out,
1106 Label& not_equal_continue) {
1107 assert(ProfileInterpreter, "must be profiling interpreter");
1108 if (test_value_out == noreg) {
1109 ldr(rscratch1, Address(mdp_in, offset));
1110 cmp(value, rscratch1);
1111 } else {
1112 // Put the test value into a register, so caller can use it:
1113 ldr(test_value_out, Address(mdp_in, offset));
1114 cmp(value, test_value_out);
1115 }
1116 br(Assembler::NE, not_equal_continue);
1117 }
1118
1119
1120 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1121 int offset_of_disp) {
1122 assert(ProfileInterpreter, "must be profiling interpreter");
1123 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1124 add(mdp_in, mdp_in, rscratch1, LSL);
1125 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1126 }
1127
1128
1129 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1130 Register reg,
1131 int offset_of_disp) {
1132 assert(ProfileInterpreter, "must be profiling interpreter");
1133 lea(rscratch1, Address(mdp_in, offset_of_disp));
1134 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1135 add(mdp_in, mdp_in, rscratch1, LSL);
1136 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1137 }
1138
1139
1140 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1141 int constant) {
1142 assert(ProfileInterpreter, "must be profiling interpreter");
1143 add(mdp_in, mdp_in, (unsigned)constant);
1144 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1145 }
1146
1147
1148 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1149 assert(ProfileInterpreter, "must be profiling interpreter");
1150 // save/restore across call_VM
1151 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1152 call_VM(noreg,
1153 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1154 return_bci);
1155 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1156 }
1157
1158
1159 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
1160 if (ProfileInterpreter) {
1161 Label profile_continue;
1162
1163 // If no method data exists, go to profile_continue.
1164 test_method_data_pointer(mdp, profile_continue);
1165
1166 // We are taking a branch. Increment the taken count.
1167 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1168
1169 // The method data pointer needs to be updated to reflect the new target.
1170 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1171 bind(profile_continue);
1172 }
1173 }
1174
1175
1176 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1177 if (ProfileInterpreter) {
1178 Label profile_continue;
1179
1180 // If no method data exists, go to profile_continue.
1181 test_method_data_pointer(mdp, profile_continue);
1182
1183 // We are not taking a branch. Increment the not taken count.
1184 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1185
1186 // The method data pointer needs to be updated to correspond to
1187 // the next bytecode
1188 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1189 bind(profile_continue);
1190 }
1191 }
1192
1193
1194 void InterpreterMacroAssembler::profile_call(Register mdp) {
1195 if (ProfileInterpreter) {
1196 Label profile_continue;
1197
1198 // If no method data exists, go to profile_continue.
1199 test_method_data_pointer(mdp, profile_continue);
1200
1201 // We are making a call. Increment the count.
1202 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1203
1204 // The method data pointer needs to be updated to reflect the new target.
1205 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1206 bind(profile_continue);
1207 }
1208 }
1209
1210 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1211 if (ProfileInterpreter) {
1212 Label profile_continue;
1213
1214 // If no method data exists, go to profile_continue.
1215 test_method_data_pointer(mdp, profile_continue);
1216
1217 // We are making a call. Increment the count.
1218 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1219
1220 // The method data pointer needs to be updated to reflect the new target.
1221 update_mdp_by_constant(mdp,
1222 in_bytes(VirtualCallData::
1223 virtual_call_data_size()));
1224 bind(profile_continue);
1225 }
1226 }
1227
1228
1229 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1230 Register mdp,
1231 Register reg2,
1232 bool receiver_can_be_null) {
1233 if (ProfileInterpreter) {
1234 Label profile_continue;
1235
1236 // If no method data exists, go to profile_continue.
1237 test_method_data_pointer(mdp, profile_continue);
1238
1239 Label skip_receiver_profile;
1240 if (receiver_can_be_null) {
1241 Label not_null;
1242 // We are making a call. Increment the count for null receiver.
1243 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1244 b(skip_receiver_profile);
1245 bind(not_null);
1246 }
1247
1248 // Record the receiver type.
1249 record_klass_in_profile(receiver, mdp, reg2);
1250 bind(skip_receiver_profile);
1251
1252 // The method data pointer needs to be updated to reflect the new target.
1253 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1254 bind(profile_continue);
1255 }
1256 }
1257
1258 // This routine creates a state machine for updating the multi-row
1259 // type profile at a virtual call site (or other type-sensitive bytecode).
1260 // The machine visits each row (of receiver/count) until the receiver type
1261 // is found, or until it runs out of rows. At the same time, it remembers
1262 // the location of the first empty row. (An empty row records null for its
1263 // receiver, and can be allocated for a newly-observed receiver type.)
1264 // Because there are two degrees of freedom in the state, a simple linear
1265 // search will not work; it must be a decision tree. Hence this helper
1266 // function is recursive, to generate the required tree structured code.
1267 // It's the interpreter, so we are trading off code space for speed.
1268 // See below for example code.
1269 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1270 Register receiver, Register mdp,
1271 Register reg2, int start_row,
1272 Label& done) {
1273 if (TypeProfileWidth == 0) {
1274 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1275 } else {
1276 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1277 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1278 }
1279 }
1280
1281 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1282 Register reg2, int start_row, Label& done, int total_rows,
1283 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1284 int last_row = total_rows - 1;
1285 assert(start_row <= last_row, "must be work left to do");
1286 // Test this row for both the item and for null.
1287 // Take any of three different outcomes:
1288 // 1. found item => increment count and goto done
1289 // 2. found null => keep looking for case 1, maybe allocate this cell
1290 // 3. found something else => keep looking for cases 1 and 2
1291 // Case 3 is handled by a recursive call.
1292 for (int row = start_row; row <= last_row; row++) {
1293 Label next_test;
1294 bool test_for_null_also = (row == start_row);
1295
1296 // See if the item is item[n].
1297 int item_offset = in_bytes(item_offset_fn(row));
1298 test_mdp_data_at(mdp, item_offset, item,
1299 (test_for_null_also ? reg2 : noreg),
1300 next_test);
1301 // (Reg2 now contains the item from the CallData.)
1302
1303 // The item is item[n]. Increment count[n].
1304 int count_offset = in_bytes(item_count_offset_fn(row));
1305 increment_mdp_data_at(mdp, count_offset);
1306 b(done);
1307 bind(next_test);
1308
1309 if (test_for_null_also) {
1310 Label found_null;
1311 // Failed the equality check on item[n]... Test for null.
1312 if (start_row == last_row) {
1313 // The only thing left to do is handle the null case.
1314 cbz(reg2, found_null);
1315 // Item did not match any saved item and there is no empty row for it.
1316 // Increment total counter to indicate polymorphic case.
1317 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1318 b(done);
1319 bind(found_null);
1320 break;
1321 }
1322 // Since null is rare, make it be the branch-taken case.
1323 cbz(reg2, found_null);
1324
1325 // Put all the "Case 3" tests here.
1326 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1327 item_offset_fn, item_count_offset_fn);
1328
1329 // Found a null. Keep searching for a matching item,
1330 // but remember that this is an empty (unused) slot.
1331 bind(found_null);
1332 }
1333 }
1334
1335 // In the fall-through case, we found no matching item, but we
1336 // observed the item[start_row] is null.
1337
1338 // Fill in the item field and increment the count.
1339 int item_offset = in_bytes(item_offset_fn(start_row));
1340 set_mdp_data_at(mdp, item_offset, item);
1341 int count_offset = in_bytes(item_count_offset_fn(start_row));
1342 mov(reg2, DataLayout::counter_increment);
1343 set_mdp_data_at(mdp, count_offset, reg2);
1344 if (start_row > 0) {
1345 b(done);
1346 }
1347 }
1348
1349 // Example state machine code for three profile rows:
1350 // // main copy of decision tree, rooted at row[1]
1351 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1352 // if (row[0].rec != nullptr) {
1353 // // inner copy of decision tree, rooted at row[1]
1354 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1355 // if (row[1].rec != nullptr) {
1356 // // degenerate decision tree, rooted at row[2]
1357 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1358 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1359 // row[2].init(rec); goto done;
1360 // } else {
1361 // // remember row[1] is empty
1362 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1363 // row[1].init(rec); goto done;
1364 // }
1365 // } else {
1366 // // remember row[0] is empty
1367 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1368 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1369 // row[0].init(rec); goto done;
1370 // }
1371 // done:
1372
1373 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1374 Register mdp, Register reg2) {
1375 assert(ProfileInterpreter, "must be profiling");
1376 Label done;
1377
1378 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1379
1380 bind (done);
1381 }
1382
1383 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1384 Register mdp) {
1385 if (ProfileInterpreter) {
1386 Label profile_continue;
1387 uint row;
1388
1389 // If no method data exists, go to profile_continue.
1390 test_method_data_pointer(mdp, profile_continue);
1391
1392 // Update the total ret count.
1393 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1394
1395 for (row = 0; row < RetData::row_limit(); row++) {
1396 Label next_test;
1397
1398 // See if return_bci is equal to bci[n]:
1399 test_mdp_data_at(mdp,
1400 in_bytes(RetData::bci_offset(row)),
1401 return_bci, noreg,
1402 next_test);
1403
1404 // return_bci is equal to bci[n]. Increment the count.
1405 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1406
1407 // The method data pointer needs to be updated to reflect the new target.
1408 update_mdp_by_offset(mdp,
1409 in_bytes(RetData::bci_displacement_offset(row)));
1410 b(profile_continue);
1411 bind(next_test);
1412 }
1413
1414 update_mdp_for_ret(return_bci);
1415
1416 bind(profile_continue);
1417 }
1418 }
1419
1420 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1421 if (ProfileInterpreter) {
1422 Label profile_continue;
1423
1424 // If no method data exists, go to profile_continue.
1425 test_method_data_pointer(mdp, profile_continue);
1426
1427 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1428
1429 // The method data pointer needs to be updated.
1430 int mdp_delta = in_bytes(BitData::bit_data_size());
1431 if (TypeProfileCasts) {
1432 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1433 }
1434 update_mdp_by_constant(mdp, mdp_delta);
1435
1436 bind(profile_continue);
1437 }
1438 }
1439
1440 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1441 if (ProfileInterpreter) {
1442 Label profile_continue;
1443
1444 // If no method data exists, go to profile_continue.
1445 test_method_data_pointer(mdp, profile_continue);
1446
1447 // The method data pointer needs to be updated.
1448 int mdp_delta = in_bytes(BitData::bit_data_size());
1449 if (TypeProfileCasts) {
1450 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1451
1452 // Record the object type.
1453 record_klass_in_profile(klass, mdp, reg2);
1454 }
1455 update_mdp_by_constant(mdp, mdp_delta);
1456
1457 bind(profile_continue);
1458 }
1459 }
1460
1461 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1462 if (ProfileInterpreter) {
1463 Label profile_continue;
1464
1465 // If no method data exists, go to profile_continue.
1466 test_method_data_pointer(mdp, profile_continue);
1467
1468 // Update the default case count
1469 increment_mdp_data_at(mdp,
1470 in_bytes(MultiBranchData::default_count_offset()));
1471
1472 // The method data pointer needs to be updated.
1473 update_mdp_by_offset(mdp,
1474 in_bytes(MultiBranchData::
1475 default_displacement_offset()));
1476
1477 bind(profile_continue);
1478 }
1479 }
1480
1481 void InterpreterMacroAssembler::profile_switch_case(Register index,
1482 Register mdp,
1483 Register reg2) {
1484 if (ProfileInterpreter) {
1485 Label profile_continue;
1486
1487 // If no method data exists, go to profile_continue.
1488 test_method_data_pointer(mdp, profile_continue);
1489
1490 // Build the base (index * per_case_size_in_bytes()) +
1491 // case_array_offset_in_bytes()
1492 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1493 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1494 Assembler::maddw(index, index, reg2, rscratch1);
1495
1496 // Update the case count
1497 increment_mdp_data_at(mdp,
1498 index,
1499 in_bytes(MultiBranchData::relative_count_offset()));
1500
1501 // The method data pointer needs to be updated.
1502 update_mdp_by_offset(mdp,
1503 index,
1504 in_bytes(MultiBranchData::
1505 relative_displacement_offset()));
1506
1507 bind(profile_continue);
1508 }
1509 }
1510
1511 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1512 Register array,
1513 Register tmp) {
1514 if (ProfileInterpreter) {
1515 Label profile_continue;
1516
1517 // If no method data exists, go to profile_continue.
1518 test_method_data_pointer(mdp, profile_continue);
1519
1520 mov(tmp, array);
1521 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1522
1523 Label not_flat;
1524 test_non_flat_array_oop(array, tmp, not_flat);
1525
1526 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1527
1528 bind(not_flat);
1529
1530 Label not_null_free;
1531 test_non_null_free_array_oop(array, tmp, not_null_free);
1532
1533 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1534
1535 bind(not_null_free);
1536
1537 bind(profile_continue);
1538 }
1539 }
1540
1541 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1542 Register array,
1543 Register tmp);
1544 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1545 Register array,
1546 Register tmp);
1547
1548 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1549 if (ProfileInterpreter) {
1550 Label profile_continue;
1551
1552 // If no method data exists, go to profile_continue.
1553 test_method_data_pointer(mdp, profile_continue);
1554
1555 Label done, update;
1556 cbnz(element, update);
1557 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1558 b(done);
1559
1560 bind(update);
1561 load_klass(tmp, element);
1562
1563 // Record the object type.
1564 record_klass_in_profile(tmp, mdp, tmp2);
1565
1566 bind(done);
1567
1568 // The method data pointer needs to be updated.
1569 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1570
1571 bind(profile_continue);
1572 }
1573 }
1574
1575
1576 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1577 Register element,
1578 Register tmp) {
1579 if (ProfileInterpreter) {
1580 Label profile_continue;
1581
1582 // If no method data exists, go to profile_continue.
1583 test_method_data_pointer(mdp, profile_continue);
1584
1585 mov(tmp, element);
1586 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1587
1588 // The method data pointer needs to be updated.
1589 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1590
1591 bind(profile_continue);
1592 }
1593 }
1594
1595 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1596 Register left,
1597 Register right,
1598 Register tmp) {
1599 if (ProfileInterpreter) {
1600 Label profile_continue;
1601
1602 // If no method data exists, go to profile_continue.
1603 test_method_data_pointer(mdp, profile_continue);
1604
1605 mov(tmp, left);
1606 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1607
1608 Label left_not_inline_type;
1609 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1610 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1611 bind(left_not_inline_type);
1612
1613 mov(tmp, right);
1614 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1615
1616 Label right_not_inline_type;
1617 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1618 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1619 bind(right_not_inline_type);
1620
1621 bind(profile_continue);
1622 }
1623 }
1624
1625 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1626 if (state == atos) {
1627 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1628 }
1629 }
1630
1631 void InterpreterMacroAssembler::notify_method_entry() {
1632 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1633 // track stack depth. If it is possible to enter interp_only_mode we add
1634 // the code to check if the event should be sent.
1635 if (JvmtiExport::can_post_interpreter_events()) {
1636 Label L;
1637 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1638 cbzw(r3, L);
1639 call_VM(noreg, CAST_FROM_FN_PTR(address,
1640 InterpreterRuntime::post_method_entry));
1641 bind(L);
1642 }
1643
1644 if (DTraceMethodProbes) {
1645 get_method(c_rarg1);
1646 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1647 rthread, c_rarg1);
1648 }
1649
1650 // RedefineClasses() tracing support for obsolete method entry
1651 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1652 get_method(c_rarg1);
1653 call_VM_leaf(
1654 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1655 rthread, c_rarg1);
1656 }
1657
1658 }
1659
1660
1661 void InterpreterMacroAssembler::notify_method_exit(
1662 TosState state, NotifyMethodExitMode mode) {
1663 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1664 // track stack depth. If it is possible to enter interp_only_mode we add
1665 // the code to check if the event should be sent.
1666 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1667 Label L;
1668 // Note: frame::interpreter_frame_result has a dependency on how the
1669 // method result is saved across the call to post_method_exit. If this
1670 // is changed then the interpreter_frame_result implementation will
1671 // need to be updated too.
1672
1673 // template interpreter will leave the result on the top of the stack.
1674 push(state);
1675 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1676 cbz(r3, L);
1677 call_VM(noreg,
1678 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1679 bind(L);
1680 pop(state);
1681 }
1682
1683 if (DTraceMethodProbes) {
1684 push(state);
1685 get_method(c_rarg1);
1686 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1687 rthread, c_rarg1);
1688 pop(state);
1689 }
1690 }
1691
1692
1693 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1694 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1695 int increment, Address mask,
1696 Register scratch, Register scratch2,
1697 bool preloaded, Condition cond,
1698 Label* where) {
1699 if (!preloaded) {
1700 ldrw(scratch, counter_addr);
1701 }
1702 add(scratch, scratch, increment);
1703 strw(scratch, counter_addr);
1704 ldrw(scratch2, mask);
1705 ands(scratch, scratch, scratch2);
1706 br(cond, *where);
1707 }
1708
1709 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1710 int number_of_arguments) {
1711 // interpreter specific
1712 //
1713 // Note: No need to save/restore rbcp & rlocals pointer since these
1714 // are callee saved registers and no blocking/ GC can happen
1715 // in leaf calls.
1716 #ifdef ASSERT
1717 {
1718 Label L;
1719 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1720 cbz(rscratch1, L);
1721 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1722 " last_sp != nullptr");
1723 bind(L);
1724 }
1725 #endif /* ASSERT */
1726 // super call
1727 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1728 }
1729
1730 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1731 Register java_thread,
1732 Register last_java_sp,
1733 address entry_point,
1734 int number_of_arguments,
1735 bool check_exceptions) {
1736 // interpreter specific
1737 //
1738 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1739 // really make a difference for these runtime calls, since they are
1740 // slow anyway. Btw., bcp must be saved/restored since it may change
1741 // due to GC.
1742 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1743 save_bcp();
1744 #ifdef ASSERT
1745 {
1746 Label L;
1747 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1748 cbz(rscratch1, L);
1749 stop("InterpreterMacroAssembler::call_VM_base:"
1750 " last_sp != nullptr");
1751 bind(L);
1752 }
1753 #endif /* ASSERT */
1754 // super call
1755 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1756 entry_point, number_of_arguments,
1757 check_exceptions);
1758 // interpreter specific
1759 restore_bcp();
1760 restore_locals();
1761 }
1762
1763 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1764 address entry_point,
1765 Register arg_1) {
1766 assert(arg_1 == c_rarg1, "");
1767 Label resume_pc, not_preempted;
1768
1769 #ifdef ASSERT
1770 {
1771 Label L;
1772 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1773 cbz(rscratch1, L);
1774 stop("Should not have alternate return address set");
1775 bind(L);
1776 }
1777 #endif /* ASSERT */
1778
1779 // Force freeze slow path.
1780 push_cont_fastpath();
1781
1782 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1783 adr(rscratch1, resume_pc);
1784 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1785 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1786
1787 pop_cont_fastpath();
1788
1789 // Check if preempted.
1790 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1791 cbz(rscratch1, not_preempted);
1792 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1793 br(rscratch1);
1794
1795 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1796 bind(resume_pc);
1797 restore_after_resume(false /* is_native */);
1798
1799 bind(not_preempted);
1800 }
1801
1802 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1803 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1804 blr(rscratch1);
1805 if (is_native) {
1806 // On resume we need to set up stack as expected
1807 push(dtos);
1808 push(ltos);
1809 }
1810 }
1811
1812 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1813 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1814 Label update, next, none;
1815
1816 verify_oop(obj);
1817
1818 cbnz(obj, update);
1819 orptr(mdo_addr, TypeEntries::null_seen);
1820 b(next);
1821
1822 bind(update);
1823 load_klass(obj, obj);
1824
1825 ldr(rscratch1, mdo_addr);
1826 eor(obj, obj, rscratch1);
1827 tst(obj, TypeEntries::type_klass_mask);
1828 br(Assembler::EQ, next); // klass seen before, nothing to
1829 // do. The unknown bit may have been
1830 // set already but no need to check.
1831
1832 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1833 // already unknown. Nothing to do anymore.
1834
1835 cbz(rscratch1, none);
1836 cmp(rscratch1, (u1)TypeEntries::null_seen);
1837 br(Assembler::EQ, none);
1838 // There is a chance that the checks above
1839 // fail if another thread has just set the
1840 // profiling to this obj's klass
1841 eor(obj, obj, rscratch1); // get back original value before XOR
1842 ldr(rscratch1, mdo_addr);
1843 eor(obj, obj, rscratch1);
1844 tst(obj, TypeEntries::type_klass_mask);
1845 br(Assembler::EQ, next);
1846
1847 // different than before. Cannot keep accurate profile.
1848 orptr(mdo_addr, TypeEntries::type_unknown);
1849 b(next);
1850
1851 bind(none);
1852 // first time here. Set profile type.
1853 str(obj, mdo_addr);
1854 #ifdef ASSERT
1855 andr(obj, obj, TypeEntries::type_mask);
1856 verify_klass_ptr(obj);
1857 #endif
1858
1859 bind(next);
1860 }
1861
1862 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1863 if (!ProfileInterpreter) {
1864 return;
1865 }
1866
1867 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1868 Label profile_continue;
1869
1870 test_method_data_pointer(mdp, profile_continue);
1871
1872 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1873
1874 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1875 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1876 br(Assembler::NE, profile_continue);
1877
1878 if (MethodData::profile_arguments()) {
1879 Label done;
1880 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1881
1882 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1883 if (i > 0 || MethodData::profile_return()) {
1884 // If return value type is profiled we may have no argument to profile
1885 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1886 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1887 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1888 add(rscratch1, mdp, off_to_args);
1889 br(Assembler::LT, done);
1890 }
1891 ldr(tmp, Address(callee, Method::const_offset()));
1892 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1893 // stack offset o (zero based) from the start of the argument
1894 // list, for n arguments translates into offset n - o - 1 from
1895 // the end of the argument list
1896 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1897 sub(tmp, tmp, rscratch1);
1898 sub(tmp, tmp, 1);
1899 Address arg_addr = argument_address(tmp);
1900 ldr(tmp, arg_addr);
1901
1902 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1903 profile_obj_type(tmp, mdo_arg_addr);
1904
1905 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1906 off_to_args += to_add;
1907 }
1908
1909 if (MethodData::profile_return()) {
1910 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1911 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1912 }
1913
1914 add(rscratch1, mdp, off_to_args);
1915 bind(done);
1916 mov(mdp, rscratch1);
1917
1918 if (MethodData::profile_return()) {
1919 // We're right after the type profile for the last
1920 // argument. tmp is the number of cells left in the
1921 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1922 // if there's a return to profile.
1923 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1924 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1925 }
1926 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1927 } else {
1928 assert(MethodData::profile_return(), "either profile call args or call ret");
1929 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1930 }
1931
1932 // mdp points right after the end of the
1933 // CallTypeData/VirtualCallTypeData, right after the cells for the
1934 // return value type if there's one
1935
1936 bind(profile_continue);
1937 }
1938 }
1939
1940 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1941 assert_different_registers(mdp, ret, tmp, rbcp);
1942 if (ProfileInterpreter && MethodData::profile_return()) {
1943 Label profile_continue, done;
1944
1945 test_method_data_pointer(mdp, profile_continue);
1946
1947 if (MethodData::profile_return_jsr292_only()) {
1948 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1949
1950 // If we don't profile all invoke bytecodes we must make sure
1951 // it's a bytecode we indeed profile. We can't go back to the
1952 // beginning of the ProfileData we intend to update to check its
1953 // type because we're right after it and we don't known its
1954 // length
1955 Label do_profile;
1956 ldrb(rscratch1, Address(rbcp, 0));
1957 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1958 br(Assembler::EQ, do_profile);
1959 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1960 br(Assembler::EQ, do_profile);
1961 get_method(tmp);
1962 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1963 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1964 br(Assembler::NE, profile_continue);
1965
1966 bind(do_profile);
1967 }
1968
1969 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1970 mov(tmp, ret);
1971 profile_obj_type(tmp, mdo_ret_addr);
1972
1973 bind(profile_continue);
1974 }
1975 }
1976
1977 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1978 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1979 if (ProfileInterpreter && MethodData::profile_parameters()) {
1980 Label profile_continue, done;
1981
1982 test_method_data_pointer(mdp, profile_continue);
1983
1984 // Load the offset of the area within the MDO used for
1985 // parameters. If it's negative we're not profiling any parameters
1986 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1987 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1988
1989 // Compute a pointer to the area for parameters from the offset
1990 // and move the pointer to the slot for the last
1991 // parameters. Collect profiling from last parameter down.
1992 // mdo start + parameters offset + array length - 1
1993 add(mdp, mdp, tmp1);
1994 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1995 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1996
1997 Label loop;
1998 bind(loop);
1999
2000 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2001 int type_base = in_bytes(ParametersTypeData::type_offset(0));
2002 int per_arg_scale = exact_log2(DataLayout::cell_size);
2003 add(rscratch1, mdp, off_base);
2004 add(rscratch2, mdp, type_base);
2005
2006 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
2007 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
2008
2009 // load offset on the stack from the slot for this parameter
2010 ldr(tmp2, arg_off);
2011 neg(tmp2, tmp2);
2012 // read the parameter from the local area
2013 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
2014
2015 // profile the parameter
2016 profile_obj_type(tmp2, arg_type);
2017
2018 // go to next parameter
2019 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
2020 br(Assembler::GE, loop);
2021
2022 bind(profile_continue);
2023 }
2024 }
2025
2026 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2027 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
2028 get_cache_index_at_bcp(index, 1, sizeof(u4));
2029 // Get address of invokedynamic array
2030 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2031 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2032 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
2033 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
2034 lea(cache, Address(cache, index));
2035 }
2036
2037 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2038 // Get index out of bytecode pointer
2039 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2040 // Take shortcut if the size is a power of 2
2041 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2042 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2043 } else {
2044 mov(cache, sizeof(ResolvedFieldEntry));
2045 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2046 }
2047 // Get address of field entries array
2048 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
2049 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
2050 lea(cache, Address(cache, index));
2051 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
2052 membar(MacroAssembler::LoadLoad);
2053 }
2054
2055 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2056 // Get index out of bytecode pointer
2057 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2058 mov(cache, sizeof(ResolvedMethodEntry));
2059 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2060
2061 // Get address of field entries array
2062 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
2063 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2064 lea(cache, Address(cache, index));
2065 }