1 /*
2 * Copyright (c) 2003, 2024, 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 "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_aarch64.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/constMethodFlags.hpp"
37 #include "oops/markWord.hpp"
38 #include "oops/method.hpp"
39 #include "oops/methodData.hpp"
40 #include "oops/inlineKlass.hpp"
41 #include "oops/resolvedFieldEntry.hpp"
42 #include "oops/resolvedIndyEntry.hpp"
43 #include "oops/resolvedMethodEntry.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/jvmtiThreadState.hpp"
46 #include "runtime/basicLock.hpp"
47 #include "runtime/frame.inline.hpp"
48 #include "runtime/javaThread.hpp"
49 #include "runtime/safepointMechanism.hpp"
50 #include "runtime/sharedRuntime.hpp"
51 #include "utilities/powerOfTwo.hpp"
52
53 void InterpreterMacroAssembler::narrow(Register result) {
54
55 // Get method->_constMethod->_result_type
56 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
57 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
58 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
59
60 Label done, notBool, notByte, notChar;
61
62 // common case first
63 cmpw(rscratch1, T_INT);
64 br(Assembler::EQ, done);
65
66 // mask integer result to narrower return type.
67 cmpw(rscratch1, T_BOOLEAN);
68 br(Assembler::NE, notBool);
69 andw(result, result, 0x1);
70 b(done);
71
72 bind(notBool);
73 cmpw(rscratch1, T_BYTE);
74 br(Assembler::NE, notByte);
75 sbfx(result, result, 0, 8);
76 b(done);
77
78 bind(notByte);
79 cmpw(rscratch1, T_CHAR);
80 br(Assembler::NE, notChar);
81 ubfx(result, result, 0, 16); // truncate upper 16 bits
82 b(done);
83
84 bind(notChar);
85 sbfx(result, result, 0, 16); // sign-extend short
86
87 // Nothing to do for T_INT
88 bind(done);
89 }
90
91 void InterpreterMacroAssembler::jump_to_entry(address entry) {
92 assert(entry, "Entry must have been generated by now");
93 b(entry);
94 }
95
96 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
97 if (JvmtiExport::can_pop_frame()) {
98 Label L;
99 // Initiate popframe handling only if it is not already being
100 // processed. If the flag has the popframe_processing bit set, it
101 // means that this code is called *during* popframe handling - we
102 // don't want to reenter.
103 // This method is only called just after the call into the vm in
104 // call_VM_base, so the arg registers are available.
105 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
106 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
107 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
108 // Call Interpreter::remove_activation_preserving_args_entry() to get the
109 // address of the same-named entrypoint in the generated interpreter code.
110 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
111 br(r0);
112 bind(L);
113 }
114 }
115
116
117 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
118 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
119 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
120 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
121 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
122 switch (state) {
123 case atos: ldr(r0, oop_addr);
124 str(zr, oop_addr);
125 interp_verify_oop(r0, state); break;
126 case ltos: ldr(r0, val_addr); break;
127 case btos: // fall through
128 case ztos: // fall through
129 case ctos: // fall through
130 case stos: // fall through
131 case itos: ldrw(r0, val_addr); break;
132 case ftos: ldrs(v0, val_addr); break;
133 case dtos: ldrd(v0, val_addr); break;
134 case vtos: /* nothing to do */ break;
135 default : ShouldNotReachHere();
136 }
137 // Clean up tos value in the thread object
138 movw(rscratch1, (int) ilgl);
139 strw(rscratch1, tos_addr);
140 strw(zr, val_addr);
141 }
142
143
144 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
145 if (JvmtiExport::can_force_early_return()) {
146 Label L;
147 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
148 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
149
150 // Initiate earlyret handling only if it is not already being processed.
151 // If the flag has the earlyret_processing bit set, it means that this code
152 // is called *during* earlyret handling - we don't want to reenter.
153 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
154 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
155 br(Assembler::NE, L);
156
157 // Call Interpreter::remove_activation_early_entry() to get the address of the
158 // same-named entrypoint in the generated interpreter code.
159 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
160 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
161 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
162 br(r0);
163 bind(L);
164 }
165 }
166
167 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
168 Register reg,
169 int bcp_offset) {
170 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
171 ldrh(reg, Address(rbcp, bcp_offset));
172 rev16(reg, reg);
173 }
174
175 void InterpreterMacroAssembler::get_dispatch() {
176 uint64_t offset;
177 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
178 // Use add() here after ARDP, rather than lea().
179 // lea() does not generate anything if its offset is zero.
180 // However, relocs expect to find either an ADD or a load/store
181 // insn after an ADRP. add() always generates an ADD insn, even
182 // for add(Rn, Rn, 0).
183 add(rdispatch, rdispatch, offset);
184 }
185
186 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
187 int bcp_offset,
188 size_t index_size) {
189 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
190 if (index_size == sizeof(u2)) {
191 load_unsigned_short(index, Address(rbcp, bcp_offset));
192 } else if (index_size == sizeof(u4)) {
193 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
194 ldrw(index, Address(rbcp, bcp_offset));
195 } else if (index_size == sizeof(u1)) {
196 load_unsigned_byte(index, Address(rbcp, bcp_offset));
197 } else {
198 ShouldNotReachHere();
199 }
200 }
201
202 void InterpreterMacroAssembler::get_method_counters(Register method,
203 Register mcs, Label& skip) {
204 Label has_counters;
205 ldr(mcs, Address(method, Method::method_counters_offset()));
206 cbnz(mcs, has_counters);
207 call_VM(noreg, CAST_FROM_FN_PTR(address,
208 InterpreterRuntime::build_method_counters), method);
209 ldr(mcs, Address(method, Method::method_counters_offset()));
210 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
211 bind(has_counters);
212 }
213
214 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
215 Register t1, Register t2,
216 bool clear_fields, Label& alloc_failed) {
217 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
218 if (DTraceMethodProbes) {
219 // Trigger dtrace event for fastpath
220 push(atos);
221 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
222 pop(atos);
223 }
224 }
225
226 void InterpreterMacroAssembler::read_flat_field(Register entry,
227 Register field_index, Register field_offset,
228 Register temp, Register obj) {
229 Label alloc_failed, empty_value, done;
230 const Register src = field_offset;
231 const Register alloc_temp = r10;
232 const Register dst_temp = field_index;
233 const Register layout_info = temp;
234 assert_different_registers(obj, entry, field_index, field_offset, temp, alloc_temp);
235
236 // Grab the inline field klass
237 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
238 inline_layout_info(rscratch1, field_index, layout_info);
239
240 const Register field_klass = dst_temp;
241 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
242
243 // check for empty value klass
244 test_klass_is_empty_inline_type(field_klass, rscratch1, empty_value);
245
246 // allocate buffer
247 push(obj); // save holder
248 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, alloc_failed);
249
250 // Have an oop instance buffer, copy into it
251 data_for_oop(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(empty_value);
261 get_empty_inline_type_oop(field_klass, alloc_temp, obj);
262 b(done);
263
264 bind(alloc_failed);
265 pop(obj);
266 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
267 obj, entry);
268
269 bind(done);
270 membar(Assembler::StoreStore);
271 }
272
273 // Load object from cpool->resolved_references(index)
274 void InterpreterMacroAssembler::load_resolved_reference_at_index(
275 Register result, Register index, Register tmp) {
276 assert_different_registers(result, index);
277
278 get_constant_pool(result);
279 // load pointer for resolved_references[] objArray
280 ldr(result, Address(result, ConstantPool::cache_offset()));
281 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
282 resolve_oop_handle(result, tmp, rscratch2);
283 // Add in the index
284 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
285 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
286 }
287
288 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
289 Register cpool, Register index, Register klass, Register temp) {
290 add(temp, cpool, index, LSL, LogBytesPerWord);
291 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
292 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
293 add(klass, klass, temp, LSL, LogBytesPerWord);
294 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
295 }
296
297 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
298 // subtype of super_klass.
299 //
300 // Args:
301 // r0: superklass
302 // Rsub_klass: subklass
303 //
304 // Kills:
305 // r2, r5
306 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
307 Label& ok_is_subtype,
308 bool profile) {
309 assert(Rsub_klass != r0, "r0 holds superklass");
310 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
311 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
312
313 // Profile the not-null value's klass.
314 if (profile) {
315 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
316 }
317
318 // Do the check.
319 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
320 }
321
322 // Java Expression Stack
323
324 void InterpreterMacroAssembler::pop_ptr(Register r) {
325 ldr(r, post(esp, wordSize));
326 }
327
328 void InterpreterMacroAssembler::pop_i(Register r) {
329 ldrw(r, post(esp, wordSize));
330 }
331
332 void InterpreterMacroAssembler::pop_l(Register r) {
333 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
334 }
335
336 void InterpreterMacroAssembler::push_ptr(Register r) {
337 str(r, pre(esp, -wordSize));
338 }
339
340 void InterpreterMacroAssembler::push_i(Register r) {
341 str(r, pre(esp, -wordSize));
342 }
343
344 void InterpreterMacroAssembler::push_l(Register r) {
345 str(zr, pre(esp, -wordSize));
346 str(r, pre(esp, - wordSize));
347 }
348
349 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
350 ldrs(r, post(esp, wordSize));
351 }
352
353 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
354 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
355 }
356
357 void InterpreterMacroAssembler::push_f(FloatRegister r) {
358 strs(r, pre(esp, -wordSize));
359 }
360
361 void InterpreterMacroAssembler::push_d(FloatRegister r) {
362 strd(r, pre(esp, 2* -wordSize));
363 }
364
365 void InterpreterMacroAssembler::pop(TosState state) {
366 switch (state) {
367 case atos: pop_ptr(); break;
368 case btos:
369 case ztos:
370 case ctos:
371 case stos:
372 case itos: pop_i(); break;
373 case ltos: pop_l(); break;
374 case ftos: pop_f(); break;
375 case dtos: pop_d(); break;
376 case vtos: /* nothing to do */ break;
377 default: ShouldNotReachHere();
378 }
379 interp_verify_oop(r0, state);
380 }
381
382 void InterpreterMacroAssembler::push(TosState state) {
383 interp_verify_oop(r0, state);
384 switch (state) {
385 case atos: push_ptr(); break;
386 case btos:
387 case ztos:
388 case ctos:
389 case stos:
390 case itos: push_i(); break;
391 case ltos: push_l(); break;
392 case ftos: push_f(); break;
393 case dtos: push_d(); break;
394 case vtos: /* nothing to do */ break;
395 default : ShouldNotReachHere();
396 }
397 }
398
399 // Helpers for swap and dup
400 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
401 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
402 }
403
404 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
405 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
406 }
407
408 void InterpreterMacroAssembler::load_float(Address src) {
409 ldrs(v0, src);
410 }
411
412 void InterpreterMacroAssembler::load_double(Address src) {
413 ldrd(v0, src);
414 }
415
416 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
417 // set sender sp
418 mov(r19_sender_sp, sp);
419 // record last_sp
420 sub(rscratch1, esp, rfp);
421 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
422 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
423 }
424
425 // Jump to from_interpreted entry of a call unless single stepping is possible
426 // in this thread in which case we must call the i2i entry
427 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
428 prepare_to_jump_from_interpreted();
429
430 if (JvmtiExport::can_post_interpreter_events()) {
431 Label run_compiled_code;
432 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
433 // compiled code in threads for which the event is enabled. Check here for
434 // interp_only_mode if these events CAN be enabled.
435 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
436 cbzw(rscratch1, run_compiled_code);
437 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
438 br(rscratch1);
439 bind(run_compiled_code);
440 }
441
442 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
443 br(rscratch1);
444 }
445
446 // The following two routines provide a hook so that an implementation
447 // can schedule the dispatch in two parts. amd64 does not do this.
448 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
449 }
450
451 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
452 dispatch_next(state, step);
453 }
454
455 void InterpreterMacroAssembler::dispatch_base(TosState state,
456 address* table,
457 bool verifyoop,
458 bool generate_poll) {
459 if (VerifyActivationFrameSize) {
460 Unimplemented();
461 }
462 if (verifyoop) {
463 interp_verify_oop(r0, state);
464 }
465
466 Label safepoint;
467 address* const safepoint_table = Interpreter::safept_table(state);
468 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
469
470 if (needs_thread_local_poll) {
471 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
472 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
473 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
474 }
475
476 if (table == Interpreter::dispatch_table(state)) {
477 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
478 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
479 } else {
480 mov(rscratch2, (address)table);
481 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
482 }
483 br(rscratch2);
484
485 if (needs_thread_local_poll) {
486 bind(safepoint);
487 lea(rscratch2, ExternalAddress((address)safepoint_table));
488 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
489 br(rscratch2);
490 }
491 }
492
493 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
494 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
495 }
496
497 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
498 dispatch_base(state, Interpreter::normal_table(state));
499 }
500
501 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
502 dispatch_base(state, Interpreter::normal_table(state), false);
503 }
504
505
506 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
507 // load next bytecode
508 ldrb(rscratch1, Address(pre(rbcp, step)));
509 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
510 }
511
512 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
513 // load current bytecode
514 ldrb(rscratch1, Address(rbcp, 0));
515 dispatch_base(state, table);
516 }
517
518 // remove activation
519 //
520 // Apply stack watermark barrier.
521 // Unlock the receiver if this is a synchronized method.
522 // Unlock any Java monitors from synchronized blocks.
523 // Remove the activation from the stack.
524 //
525 // If there are locked Java monitors
526 // If throw_monitor_exception
527 // throws IllegalMonitorStateException
528 // Else if install_monitor_exception
529 // installs IllegalMonitorStateException
530 // Else
531 // no error processing
532 void InterpreterMacroAssembler::remove_activation(
533 TosState state,
534 bool throw_monitor_exception,
535 bool install_monitor_exception,
536 bool notify_jvmdi) {
537 // Note: Registers r3 xmm0 may be in use for the
538 // result check if synchronized method
539 Label unlocked, unlock, no_unlock;
540
541 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
542 // that would normally not be safe to use. Such bad returns into unsafe territory of
543 // the stack, will call InterpreterRuntime::at_unwind.
544 Label slow_path;
545 Label fast_path;
546 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
547 br(Assembler::AL, fast_path);
548 bind(slow_path);
549 push(state);
550 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
551 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
552 reset_last_Java_frame(true);
553 pop(state);
554 bind(fast_path);
555
556 // get the value of _do_not_unlock_if_synchronized into r3
557 const Address do_not_unlock_if_synchronized(rthread,
558 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
559 ldrb(r3, do_not_unlock_if_synchronized);
560 strb(zr, do_not_unlock_if_synchronized); // reset the flag
561
562 // get method access flags
563 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
564 ldr(r2, Address(r1, Method::access_flags_offset()));
565 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
566
567 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
568 // is set.
569 cbnz(r3, no_unlock);
570
571 // unlock monitor
572 push(state); // save result
573
574 // BasicObjectLock will be first in list, since this is a
575 // synchronized method. However, need to check that the object has
576 // not been unlocked by an explicit monitorexit bytecode.
577 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
578 wordSize - (int) sizeof(BasicObjectLock));
579 // We use c_rarg1 so that if we go slow path it will be the correct
580 // register for unlock_object to pass to VM directly
581 lea(c_rarg1, monitor); // address of first monitor
582
583 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
584 cbnz(r0, unlock);
585
586 pop(state);
587 if (throw_monitor_exception) {
588 // Entry already unlocked, need to throw exception
589 call_VM(noreg, CAST_FROM_FN_PTR(address,
590 InterpreterRuntime::throw_illegal_monitor_state_exception));
591 should_not_reach_here();
592 } else {
593 // Monitor already unlocked during a stack unroll. If requested,
594 // install an illegal_monitor_state_exception. Continue with
595 // stack unrolling.
596 if (install_monitor_exception) {
597 call_VM(noreg, CAST_FROM_FN_PTR(address,
598 InterpreterRuntime::new_illegal_monitor_state_exception));
599 }
600 b(unlocked);
601 }
602
603 bind(unlock);
604 unlock_object(c_rarg1);
605 pop(state);
606
607 // Check that for block-structured locking (i.e., that all locked
608 // objects has been unlocked)
609 bind(unlocked);
610
611 // r0: Might contain return value
612
613 // Check that all monitors are unlocked
614 {
615 Label loop, exception, entry, restart;
616 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
617 const Address monitor_block_top(
618 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
619 const Address monitor_block_bot(
620 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
621
622 bind(restart);
623 // We use c_rarg1 so that if we go slow path it will be the correct
624 // register for unlock_object to pass to VM directly
625 ldr(c_rarg1, monitor_block_top); // derelativize pointer
626 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
627 // c_rarg1 points to current entry, starting with top-most entry
628
629 lea(r19, monitor_block_bot); // points to word before bottom of
630 // monitor block
631 b(entry);
632
633 // Entry already locked, need to throw exception
634 bind(exception);
635
636 if (throw_monitor_exception) {
637 // Throw exception
638 MacroAssembler::call_VM(noreg,
639 CAST_FROM_FN_PTR(address, InterpreterRuntime::
640 throw_illegal_monitor_state_exception));
641 should_not_reach_here();
642 } else {
643 // Stack unrolling. Unlock object and install illegal_monitor_exception.
644 // Unlock does not block, so don't have to worry about the frame.
645 // We don't have to preserve c_rarg1 since we are going to throw an exception.
646
647 push(state);
648 unlock_object(c_rarg1);
649 pop(state);
650
651 if (install_monitor_exception) {
652 call_VM(noreg, CAST_FROM_FN_PTR(address,
653 InterpreterRuntime::
654 new_illegal_monitor_state_exception));
655 }
656
657 b(restart);
658 }
659
660 bind(loop);
661 // check if current entry is used
662 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
663 cbnz(rscratch1, exception);
664
665 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
666 bind(entry);
667 cmp(c_rarg1, r19); // check if bottom reached
668 br(Assembler::NE, loop); // if not at bottom then check this entry
669 }
670
671 bind(no_unlock);
672
673 // jvmti support
674 if (notify_jvmdi) {
675 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
676 } else {
677 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
678 }
679
680 // remove activation
681 // get sender esp
682 ldr(rscratch2,
683 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
684
685 if (StackReservedPages > 0) {
686 // testing if reserved zone needs to be re-enabled
687 Label no_reserved_zone_enabling;
688
689 // check if already enabled - if so no re-enabling needed
690 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
691 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
692 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
693 br(Assembler::EQ, no_reserved_zone_enabling);
694
695 // look for an overflow into the stack reserved zone, i.e.
696 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
697 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
698 cmp(rscratch2, rscratch1);
699 br(Assembler::LS, no_reserved_zone_enabling);
700
701 call_VM_leaf(
702 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
703 call_VM(noreg, CAST_FROM_FN_PTR(address,
704 InterpreterRuntime::throw_delayed_StackOverflowError));
705 should_not_reach_here();
706
707 bind(no_reserved_zone_enabling);
708 }
709
710 if (state == atos && InlineTypeReturnedAsFields) {
711 // Check if we are returning an non-null inline type and load its fields into registers
712 Label skip;
713 test_oop_is_not_inline_type(r0, rscratch2, skip);
714
715 // Load fields from a buffered value with an inline class specific handler
716 load_klass(rscratch1 /*dst*/, r0 /*src*/);
717 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
718 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
719 // Unpack handler can be null if inline type is not scalarizable in returns
720 cbz(rscratch1, skip);
721
722 blr(rscratch1);
723 #ifdef ASSERT
724 // TODO 8284443 Enable
725 if (StressCallingConvention && false) {
726 Label skip_stress;
727 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
728 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
729 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
730 br(Assembler::EQ, skip_stress);
731 load_klass(r0, r0);
732 orr(r0, r0, 1);
733 bind(skip_stress);
734 }
735 #endif
736 bind(skip);
737 // Check above kills sender esp in rscratch2. Reload it.
738 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
739 }
740
741 // restore sender esp
742 mov(esp, rscratch2);
743 // remove frame anchor
744 leave();
745 // If we're returning to interpreted code we will shortly be
746 // adjusting SP to allow some space for ESP. If we're returning to
747 // compiled code the saved sender SP was saved in sender_sp, so this
748 // restores it.
749 andr(sp, esp, -16);
750 }
751
752 // Lock object
753 //
754 // Args:
755 // c_rarg1: BasicObjectLock to be used for locking
756 //
757 // Kills:
758 // r0
759 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
760 // rscratch1, rscratch2 (scratch regs)
761 void InterpreterMacroAssembler::lock_object(Register lock_reg)
762 {
763 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
764 if (LockingMode == LM_MONITOR) {
765 call_VM(noreg,
766 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
767 lock_reg);
768 } else {
769 Label count, done;
770
771 const Register swap_reg = r0;
772 const Register tmp = c_rarg2;
773 const Register obj_reg = c_rarg3; // Will contain the oop
774 const Register tmp2 = c_rarg4;
775 const Register tmp3 = c_rarg5;
776
777 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
778 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
779 const int mark_offset = lock_offset +
780 BasicLock::displaced_header_offset_in_bytes();
781
782 Label slow_case;
783
784 // Load object pointer into obj_reg %c_rarg3
785 ldr(obj_reg, Address(lock_reg, obj_offset));
786
787 if (DiagnoseSyncOnValueBasedClasses != 0) {
788 load_klass(tmp, obj_reg);
789 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
790 tst(tmp, KlassFlags::_misc_is_value_based_class);
791 br(Assembler::NE, slow_case);
792 }
793
794 if (LockingMode == LM_LIGHTWEIGHT) {
795 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
796 b(count);
797 } else if (LockingMode == LM_LEGACY) {
798 // Load (object->mark() | 1) into swap_reg
799 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
800 orr(swap_reg, rscratch1, 1);
801 if (EnableValhalla) {
802 // Mask inline_type bit such that we go to the slow path if object is an inline type
803 andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
804 }
805
806 // Save (object->mark() | 1) into BasicLock's displaced header
807 str(swap_reg, Address(lock_reg, mark_offset));
808
809 assert(lock_offset == 0,
810 "displached header must be first word in BasicObjectLock");
811
812 Label fail;
813 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
814
815 // Fast check for recursive lock.
816 //
817 // Can apply the optimization only if this is a stack lock
818 // allocated in this thread. For efficiency, we can focus on
819 // recently allocated stack locks (instead of reading the stack
820 // base and checking whether 'mark' points inside the current
821 // thread stack):
822 // 1) (mark & 7) == 0, and
823 // 2) sp <= mark < mark + os::pagesize()
824 //
825 // Warning: sp + os::pagesize can overflow the stack base. We must
826 // neither apply the optimization for an inflated lock allocated
827 // just above the thread stack (this is why condition 1 matters)
828 // nor apply the optimization if the stack lock is inside the stack
829 // of another thread. The latter is avoided even in case of overflow
830 // because we have guard pages at the end of all stacks. Hence, if
831 // we go over the stack base and hit the stack of another thread,
832 // this should not be in a writeable area that could contain a
833 // stack lock allocated by that thread. As a consequence, a stack
834 // lock less than page size away from sp is guaranteed to be
835 // owned by the current thread.
836 //
837 // These 3 tests can be done by evaluating the following
838 // expression: ((mark - sp) & (7 - os::vm_page_size())),
839 // assuming both stack pointer and pagesize have their
840 // least significant 3 bits clear.
841 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
842 // NOTE2: aarch64 does not like to subtract sp from rn so take a
843 // copy
844 mov(rscratch1, sp);
845 sub(swap_reg, swap_reg, rscratch1);
846 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
847
848 // Save the test result, for recursive case, the result is zero
849 str(swap_reg, Address(lock_reg, mark_offset));
850 br(Assembler::EQ, count);
851 }
852 bind(slow_case);
853
854 // Call the runtime routine for slow case
855 call_VM(noreg,
856 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
857 lock_reg);
858 b(done);
859
860 bind(count);
861 increment(Address(rthread, JavaThread::held_monitor_count_offset()));
862
863 bind(done);
864 }
865 }
866
867
868 // Unlocks an object. Used in monitorexit bytecode and
869 // remove_activation. Throws an IllegalMonitorException if object is
870 // not locked by current thread.
871 //
872 // Args:
873 // c_rarg1: BasicObjectLock for lock
874 //
875 // Kills:
876 // r0
877 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
878 // rscratch1, rscratch2 (scratch regs)
879 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
880 {
881 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
882
883 if (LockingMode == LM_MONITOR) {
884 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
885 } else {
886 Label count, done;
887
888 const Register swap_reg = r0;
889 const Register header_reg = c_rarg2; // Will contain the old oopMark
890 const Register obj_reg = c_rarg3; // Will contain the oop
891 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
892
893 save_bcp(); // Save in case of exception
894
895 if (LockingMode != LM_LIGHTWEIGHT) {
896 // Convert from BasicObjectLock structure to object and BasicLock
897 // structure Store the BasicLock address into %r0
898 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
899 }
900
901 // Load oop into obj_reg(%c_rarg3)
902 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
903
904 // Free entry
905 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
906
907 if (LockingMode == LM_LIGHTWEIGHT) {
908 Label slow_case;
909 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
910 b(count);
911 bind(slow_case);
912 } else if (LockingMode == LM_LEGACY) {
913 // Load the old header from BasicLock structure
914 ldr(header_reg, Address(swap_reg,
915 BasicLock::displaced_header_offset_in_bytes()));
916
917 // Test for recursion
918 cbz(header_reg, count);
919
920 // Atomic swap back the old header
921 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
922 }
923 // Call the runtime routine for slow case.
924 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
925 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
926 b(done);
927
928 bind(count);
929 decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
930
931 bind(done);
932 restore_bcp();
933 }
934 }
935
936 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
937 Label& zero_continue) {
938 assert(ProfileInterpreter, "must be profiling interpreter");
939 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
940 cbz(mdp, zero_continue);
941 }
942
943 // Set the method data pointer for the current bcp.
944 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
945 assert(ProfileInterpreter, "must be profiling interpreter");
946 Label set_mdp;
947 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
948
949 // Test MDO to avoid the call if it is null.
950 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
951 cbz(r0, set_mdp);
952 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
953 // r0: mdi
954 // mdo is guaranteed to be non-zero here, we checked for it before the call.
955 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
956 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
957 add(r0, r1, r0);
958 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
959 bind(set_mdp);
960 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
961 }
962
963 void InterpreterMacroAssembler::verify_method_data_pointer() {
964 assert(ProfileInterpreter, "must be profiling interpreter");
965 #ifdef ASSERT
966 Label verify_continue;
967 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
968 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
969 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
970 get_method(r1);
971
972 // If the mdp is valid, it will point to a DataLayout header which is
973 // consistent with the bcp. The converse is highly probable also.
974 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
975 ldr(rscratch1, Address(r1, Method::const_offset()));
976 add(r2, r2, rscratch1, Assembler::LSL);
977 lea(r2, Address(r2, ConstMethod::codes_offset()));
978 cmp(r2, rbcp);
979 br(Assembler::EQ, verify_continue);
980 // r1: method
981 // rbcp: bcp // rbcp == 22
982 // r3: mdp
983 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
984 r1, rbcp, r3);
985 bind(verify_continue);
986 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
987 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
988 #endif // ASSERT
989 }
990
991
992 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
993 int constant,
994 Register value) {
995 assert(ProfileInterpreter, "must be profiling interpreter");
996 Address data(mdp_in, constant);
997 str(value, data);
998 }
999
1000
1001 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1002 int constant,
1003 bool decrement) {
1004 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1005 }
1006
1007 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1008 Register reg,
1009 int constant,
1010 bool decrement) {
1011 assert(ProfileInterpreter, "must be profiling interpreter");
1012 // %%% this does 64bit counters at best it is wasting space
1013 // at worst it is a rare bug when counters overflow
1014
1015 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1016
1017 Address addr1(mdp_in, constant);
1018 Address addr2(rscratch2, reg, Address::lsl(0));
1019 Address &addr = addr1;
1020 if (reg != noreg) {
1021 lea(rscratch2, addr1);
1022 addr = addr2;
1023 }
1024
1025 if (decrement) {
1026 // Decrement the register. Set condition codes.
1027 // Intel does this
1028 // addptr(data, (int32_t) -DataLayout::counter_increment);
1029 // If the decrement causes the counter to overflow, stay negative
1030 // Label L;
1031 // jcc(Assembler::negative, L);
1032 // addptr(data, (int32_t) DataLayout::counter_increment);
1033 // so we do this
1034 ldr(rscratch1, addr);
1035 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1036 Label L;
1037 br(Assembler::LO, L); // skip store if counter underflow
1038 str(rscratch1, addr);
1039 bind(L);
1040 } else {
1041 assert(DataLayout::counter_increment == 1,
1042 "flow-free idiom only works with 1");
1043 // Intel does this
1044 // Increment the register. Set carry flag.
1045 // addptr(data, DataLayout::counter_increment);
1046 // If the increment causes the counter to overflow, pull back by 1.
1047 // sbbptr(data, (int32_t)0);
1048 // so we do this
1049 ldr(rscratch1, addr);
1050 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1051 Label L;
1052 br(Assembler::CS, L); // skip store if counter overflow
1053 str(rscratch1, addr);
1054 bind(L);
1055 }
1056 }
1057
1058 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1059 int flag_byte_constant) {
1060 assert(ProfileInterpreter, "must be profiling interpreter");
1061 int flags_offset = in_bytes(DataLayout::flags_offset());
1062 // Set the flag
1063 ldrb(rscratch1, Address(mdp_in, flags_offset));
1064 orr(rscratch1, rscratch1, flag_byte_constant);
1065 strb(rscratch1, Address(mdp_in, flags_offset));
1066 }
1067
1068
1069 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1070 int offset,
1071 Register value,
1072 Register test_value_out,
1073 Label& not_equal_continue) {
1074 assert(ProfileInterpreter, "must be profiling interpreter");
1075 if (test_value_out == noreg) {
1076 ldr(rscratch1, Address(mdp_in, offset));
1077 cmp(value, rscratch1);
1078 } else {
1079 // Put the test value into a register, so caller can use it:
1080 ldr(test_value_out, Address(mdp_in, offset));
1081 cmp(value, test_value_out);
1082 }
1083 br(Assembler::NE, not_equal_continue);
1084 }
1085
1086
1087 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1088 int offset_of_disp) {
1089 assert(ProfileInterpreter, "must be profiling interpreter");
1090 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1091 add(mdp_in, mdp_in, rscratch1, LSL);
1092 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1093 }
1094
1095
1096 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1097 Register reg,
1098 int offset_of_disp) {
1099 assert(ProfileInterpreter, "must be profiling interpreter");
1100 lea(rscratch1, Address(mdp_in, offset_of_disp));
1101 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1102 add(mdp_in, mdp_in, rscratch1, LSL);
1103 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1104 }
1105
1106
1107 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1108 int constant) {
1109 assert(ProfileInterpreter, "must be profiling interpreter");
1110 add(mdp_in, mdp_in, (unsigned)constant);
1111 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1112 }
1113
1114
1115 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1116 assert(ProfileInterpreter, "must be profiling interpreter");
1117 // save/restore across call_VM
1118 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1119 call_VM(noreg,
1120 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1121 return_bci);
1122 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1123 }
1124
1125
1126 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1127 Register bumped_count) {
1128 if (ProfileInterpreter) {
1129 Label profile_continue;
1130
1131 // If no method data exists, go to profile_continue.
1132 // Otherwise, assign to mdp
1133 test_method_data_pointer(mdp, profile_continue);
1134
1135 // We are taking a branch. Increment the taken count.
1136 // We inline increment_mdp_data_at to return bumped_count in a register
1137 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1138 Address data(mdp, in_bytes(JumpData::taken_offset()));
1139 ldr(bumped_count, data);
1140 assert(DataLayout::counter_increment == 1,
1141 "flow-free idiom only works with 1");
1142 // Intel does this to catch overflow
1143 // addptr(bumped_count, DataLayout::counter_increment);
1144 // sbbptr(bumped_count, 0);
1145 // so we do this
1146 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1147 Label L;
1148 br(Assembler::CS, L); // skip store if counter overflow
1149 str(bumped_count, data);
1150 bind(L);
1151 // The method data pointer needs to be updated to reflect the new target.
1152 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1153 bind(profile_continue);
1154 }
1155 }
1156
1157
1158 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1159 if (ProfileInterpreter) {
1160 Label profile_continue;
1161
1162 // If no method data exists, go to profile_continue.
1163 test_method_data_pointer(mdp, profile_continue);
1164
1165 // We are taking a branch. Increment the not taken count.
1166 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1167
1168 // The method data pointer needs to be updated to correspond to
1169 // the next bytecode
1170 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1171 bind(profile_continue);
1172 }
1173 }
1174
1175
1176 void InterpreterMacroAssembler::profile_call(Register mdp) {
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 making a call. Increment the count.
1184 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1185
1186 // The method data pointer needs to be updated to reflect the new target.
1187 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1188 bind(profile_continue);
1189 }
1190 }
1191
1192 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1193 if (ProfileInterpreter) {
1194 Label profile_continue;
1195
1196 // If no method data exists, go to profile_continue.
1197 test_method_data_pointer(mdp, profile_continue);
1198
1199 // We are making a call. Increment the count.
1200 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1201
1202 // The method data pointer needs to be updated to reflect the new target.
1203 update_mdp_by_constant(mdp,
1204 in_bytes(VirtualCallData::
1205 virtual_call_data_size()));
1206 bind(profile_continue);
1207 }
1208 }
1209
1210
1211 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1212 Register mdp,
1213 Register reg2,
1214 bool receiver_can_be_null) {
1215 if (ProfileInterpreter) {
1216 Label profile_continue;
1217
1218 // If no method data exists, go to profile_continue.
1219 test_method_data_pointer(mdp, profile_continue);
1220
1221 Label skip_receiver_profile;
1222 if (receiver_can_be_null) {
1223 Label not_null;
1224 // We are making a call. Increment the count for null receiver.
1225 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1226 b(skip_receiver_profile);
1227 bind(not_null);
1228 }
1229
1230 // Record the receiver type.
1231 record_klass_in_profile(receiver, mdp, reg2);
1232 bind(skip_receiver_profile);
1233
1234 // The method data pointer needs to be updated to reflect the new target.
1235 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1236 bind(profile_continue);
1237 }
1238 }
1239
1240 // This routine creates a state machine for updating the multi-row
1241 // type profile at a virtual call site (or other type-sensitive bytecode).
1242 // The machine visits each row (of receiver/count) until the receiver type
1243 // is found, or until it runs out of rows. At the same time, it remembers
1244 // the location of the first empty row. (An empty row records null for its
1245 // receiver, and can be allocated for a newly-observed receiver type.)
1246 // Because there are two degrees of freedom in the state, a simple linear
1247 // search will not work; it must be a decision tree. Hence this helper
1248 // function is recursive, to generate the required tree structured code.
1249 // It's the interpreter, so we are trading off code space for speed.
1250 // See below for example code.
1251 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1252 Register receiver, Register mdp,
1253 Register reg2, int start_row,
1254 Label& done) {
1255 if (TypeProfileWidth == 0) {
1256 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1257 } else {
1258 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1259 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1260 }
1261 }
1262
1263 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1264 Register reg2, int start_row, Label& done, int total_rows,
1265 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1266 int last_row = total_rows - 1;
1267 assert(start_row <= last_row, "must be work left to do");
1268 // Test this row for both the item and for null.
1269 // Take any of three different outcomes:
1270 // 1. found item => increment count and goto done
1271 // 2. found null => keep looking for case 1, maybe allocate this cell
1272 // 3. found something else => keep looking for cases 1 and 2
1273 // Case 3 is handled by a recursive call.
1274 for (int row = start_row; row <= last_row; row++) {
1275 Label next_test;
1276 bool test_for_null_also = (row == start_row);
1277
1278 // See if the item is item[n].
1279 int item_offset = in_bytes(item_offset_fn(row));
1280 test_mdp_data_at(mdp, item_offset, item,
1281 (test_for_null_also ? reg2 : noreg),
1282 next_test);
1283 // (Reg2 now contains the item from the CallData.)
1284
1285 // The item is item[n]. Increment count[n].
1286 int count_offset = in_bytes(item_count_offset_fn(row));
1287 increment_mdp_data_at(mdp, count_offset);
1288 b(done);
1289 bind(next_test);
1290
1291 if (test_for_null_also) {
1292 Label found_null;
1293 // Failed the equality check on item[n]... Test for null.
1294 if (start_row == last_row) {
1295 // The only thing left to do is handle the null case.
1296 cbz(reg2, found_null);
1297 // Item did not match any saved item and there is no empty row for it.
1298 // Increment total counter to indicate polymorphic case.
1299 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1300 b(done);
1301 bind(found_null);
1302 break;
1303 }
1304 // Since null is rare, make it be the branch-taken case.
1305 cbz(reg2, found_null);
1306
1307 // Put all the "Case 3" tests here.
1308 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1309 item_offset_fn, item_count_offset_fn);
1310
1311 // Found a null. Keep searching for a matching item,
1312 // but remember that this is an empty (unused) slot.
1313 bind(found_null);
1314 }
1315 }
1316
1317 // In the fall-through case, we found no matching item, but we
1318 // observed the item[start_row] is null.
1319
1320 // Fill in the item field and increment the count.
1321 int item_offset = in_bytes(item_offset_fn(start_row));
1322 set_mdp_data_at(mdp, item_offset, item);
1323 int count_offset = in_bytes(item_count_offset_fn(start_row));
1324 mov(reg2, DataLayout::counter_increment);
1325 set_mdp_data_at(mdp, count_offset, reg2);
1326 if (start_row > 0) {
1327 b(done);
1328 }
1329 }
1330
1331 // Example state machine code for three profile rows:
1332 // // main copy of decision tree, rooted at row[1]
1333 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1334 // if (row[0].rec != nullptr) {
1335 // // inner copy of decision tree, rooted at row[1]
1336 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1337 // if (row[1].rec != nullptr) {
1338 // // degenerate decision tree, rooted at row[2]
1339 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1340 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1341 // row[2].init(rec); goto done;
1342 // } else {
1343 // // remember row[1] is empty
1344 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1345 // row[1].init(rec); goto done;
1346 // }
1347 // } else {
1348 // // remember row[0] is empty
1349 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1350 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1351 // row[0].init(rec); goto done;
1352 // }
1353 // done:
1354
1355 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1356 Register mdp, Register reg2) {
1357 assert(ProfileInterpreter, "must be profiling");
1358 Label done;
1359
1360 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1361
1362 bind (done);
1363 }
1364
1365 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1366 Register mdp) {
1367 if (ProfileInterpreter) {
1368 Label profile_continue;
1369 uint row;
1370
1371 // If no method data exists, go to profile_continue.
1372 test_method_data_pointer(mdp, profile_continue);
1373
1374 // Update the total ret count.
1375 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1376
1377 for (row = 0; row < RetData::row_limit(); row++) {
1378 Label next_test;
1379
1380 // See if return_bci is equal to bci[n]:
1381 test_mdp_data_at(mdp,
1382 in_bytes(RetData::bci_offset(row)),
1383 return_bci, noreg,
1384 next_test);
1385
1386 // return_bci is equal to bci[n]. Increment the count.
1387 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1388
1389 // The method data pointer needs to be updated to reflect the new target.
1390 update_mdp_by_offset(mdp,
1391 in_bytes(RetData::bci_displacement_offset(row)));
1392 b(profile_continue);
1393 bind(next_test);
1394 }
1395
1396 update_mdp_for_ret(return_bci);
1397
1398 bind(profile_continue);
1399 }
1400 }
1401
1402 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1403 if (ProfileInterpreter) {
1404 Label profile_continue;
1405
1406 // If no method data exists, go to profile_continue.
1407 test_method_data_pointer(mdp, profile_continue);
1408
1409 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1410
1411 // The method data pointer needs to be updated.
1412 int mdp_delta = in_bytes(BitData::bit_data_size());
1413 if (TypeProfileCasts) {
1414 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1415 }
1416 update_mdp_by_constant(mdp, mdp_delta);
1417
1418 bind(profile_continue);
1419 }
1420 }
1421
1422 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1423 if (ProfileInterpreter) {
1424 Label profile_continue;
1425
1426 // If no method data exists, go to profile_continue.
1427 test_method_data_pointer(mdp, profile_continue);
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 // Record the object type.
1435 record_klass_in_profile(klass, mdp, reg2);
1436 }
1437 update_mdp_by_constant(mdp, mdp_delta);
1438
1439 bind(profile_continue);
1440 }
1441 }
1442
1443 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1444 if (ProfileInterpreter) {
1445 Label profile_continue;
1446
1447 // If no method data exists, go to profile_continue.
1448 test_method_data_pointer(mdp, profile_continue);
1449
1450 // Update the default case count
1451 increment_mdp_data_at(mdp,
1452 in_bytes(MultiBranchData::default_count_offset()));
1453
1454 // The method data pointer needs to be updated.
1455 update_mdp_by_offset(mdp,
1456 in_bytes(MultiBranchData::
1457 default_displacement_offset()));
1458
1459 bind(profile_continue);
1460 }
1461 }
1462
1463 void InterpreterMacroAssembler::profile_switch_case(Register index,
1464 Register mdp,
1465 Register reg2) {
1466 if (ProfileInterpreter) {
1467 Label profile_continue;
1468
1469 // If no method data exists, go to profile_continue.
1470 test_method_data_pointer(mdp, profile_continue);
1471
1472 // Build the base (index * per_case_size_in_bytes()) +
1473 // case_array_offset_in_bytes()
1474 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1475 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1476 Assembler::maddw(index, index, reg2, rscratch1);
1477
1478 // Update the case count
1479 increment_mdp_data_at(mdp,
1480 index,
1481 in_bytes(MultiBranchData::relative_count_offset()));
1482
1483 // The method data pointer needs to be updated.
1484 update_mdp_by_offset(mdp,
1485 index,
1486 in_bytes(MultiBranchData::
1487 relative_displacement_offset()));
1488
1489 bind(profile_continue);
1490 }
1491 }
1492
1493 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1494 Register array,
1495 Register tmp) {
1496 if (ProfileInterpreter) {
1497 Label profile_continue;
1498
1499 // If no method data exists, go to profile_continue.
1500 test_method_data_pointer(mdp, profile_continue);
1501
1502 mov(tmp, array);
1503 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1504
1505 Label not_flat;
1506 test_non_flat_array_oop(array, tmp, not_flat);
1507
1508 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1509
1510 bind(not_flat);
1511
1512 Label not_null_free;
1513 test_non_null_free_array_oop(array, tmp, not_null_free);
1514
1515 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1516
1517 bind(not_null_free);
1518
1519 bind(profile_continue);
1520 }
1521 }
1522
1523 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1524 Register array,
1525 Register tmp);
1526 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1527 Register array,
1528 Register tmp);
1529
1530 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1531 if (ProfileInterpreter) {
1532 Label profile_continue;
1533
1534 // If no method data exists, go to profile_continue.
1535 test_method_data_pointer(mdp, profile_continue);
1536
1537 Label done, update;
1538 cbnz(element, update);
1539 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1540 b(done);
1541
1542 bind(update);
1543 load_klass(tmp, element);
1544
1545 // Record the object type.
1546 record_klass_in_profile(tmp, mdp, tmp2);
1547
1548 bind(done);
1549
1550 // The method data pointer needs to be updated.
1551 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1552
1553 bind(profile_continue);
1554 }
1555 }
1556
1557
1558 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1559 Register element,
1560 Register tmp) {
1561 if (ProfileInterpreter) {
1562 Label profile_continue;
1563
1564 // If no method data exists, go to profile_continue.
1565 test_method_data_pointer(mdp, profile_continue);
1566
1567 mov(tmp, element);
1568 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1569
1570 // The method data pointer needs to be updated.
1571 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1572
1573 bind(profile_continue);
1574 }
1575 }
1576
1577 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1578 Register left,
1579 Register right,
1580 Register tmp) {
1581 if (ProfileInterpreter) {
1582 Label profile_continue;
1583
1584 // If no method data exists, go to profile_continue.
1585 test_method_data_pointer(mdp, profile_continue);
1586
1587 mov(tmp, left);
1588 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1589
1590 Label left_not_inline_type;
1591 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1592 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1593 bind(left_not_inline_type);
1594
1595 mov(tmp, right);
1596 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1597
1598 Label right_not_inline_type;
1599 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1600 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1601 bind(right_not_inline_type);
1602
1603 bind(profile_continue);
1604 }
1605 }
1606
1607 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1608 if (state == atos) {
1609 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1610 }
1611 }
1612
1613 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1614
1615
1616 void InterpreterMacroAssembler::notify_method_entry() {
1617 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1618 // track stack depth. If it is possible to enter interp_only_mode we add
1619 // the code to check if the event should be sent.
1620 if (JvmtiExport::can_post_interpreter_events()) {
1621 Label L;
1622 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1623 cbzw(r3, L);
1624 call_VM(noreg, CAST_FROM_FN_PTR(address,
1625 InterpreterRuntime::post_method_entry));
1626 bind(L);
1627 }
1628
1629 if (DTraceMethodProbes) {
1630 get_method(c_rarg1);
1631 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1632 rthread, c_rarg1);
1633 }
1634
1635 // RedefineClasses() tracing support for obsolete method entry
1636 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1637 get_method(c_rarg1);
1638 call_VM_leaf(
1639 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1640 rthread, c_rarg1);
1641 }
1642
1643 }
1644
1645
1646 void InterpreterMacroAssembler::notify_method_exit(
1647 TosState state, NotifyMethodExitMode mode) {
1648 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1649 // track stack depth. If it is possible to enter interp_only_mode we add
1650 // the code to check if the event should be sent.
1651 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1652 Label L;
1653 // Note: frame::interpreter_frame_result has a dependency on how the
1654 // method result is saved across the call to post_method_exit. If this
1655 // is changed then the interpreter_frame_result implementation will
1656 // need to be updated too.
1657
1658 // template interpreter will leave the result on the top of the stack.
1659 push(state);
1660 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1661 cbz(r3, L);
1662 call_VM(noreg,
1663 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1664 bind(L);
1665 pop(state);
1666 }
1667
1668 if (DTraceMethodProbes) {
1669 push(state);
1670 get_method(c_rarg1);
1671 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1672 rthread, c_rarg1);
1673 pop(state);
1674 }
1675 }
1676
1677
1678 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1679 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1680 int increment, Address mask,
1681 Register scratch, Register scratch2,
1682 bool preloaded, Condition cond,
1683 Label* where) {
1684 if (!preloaded) {
1685 ldrw(scratch, counter_addr);
1686 }
1687 add(scratch, scratch, increment);
1688 strw(scratch, counter_addr);
1689 ldrw(scratch2, mask);
1690 ands(scratch, scratch, scratch2);
1691 br(cond, *where);
1692 }
1693
1694 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1695 int number_of_arguments) {
1696 // interpreter specific
1697 //
1698 // Note: No need to save/restore rbcp & rlocals pointer since these
1699 // are callee saved registers and no blocking/ GC can happen
1700 // in leaf calls.
1701 #ifdef ASSERT
1702 {
1703 Label L;
1704 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1705 cbz(rscratch1, L);
1706 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1707 " last_sp != nullptr");
1708 bind(L);
1709 }
1710 #endif /* ASSERT */
1711 // super call
1712 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1713 }
1714
1715 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1716 Register java_thread,
1717 Register last_java_sp,
1718 address entry_point,
1719 int number_of_arguments,
1720 bool check_exceptions) {
1721 // interpreter specific
1722 //
1723 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1724 // really make a difference for these runtime calls, since they are
1725 // slow anyway. Btw., bcp must be saved/restored since it may change
1726 // due to GC.
1727 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1728 save_bcp();
1729 #ifdef ASSERT
1730 {
1731 Label L;
1732 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1733 cbz(rscratch1, L);
1734 stop("InterpreterMacroAssembler::call_VM_base:"
1735 " last_sp != nullptr");
1736 bind(L);
1737 }
1738 #endif /* ASSERT */
1739 // super call
1740 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1741 entry_point, number_of_arguments,
1742 check_exceptions);
1743 // interpreter specific
1744 restore_bcp();
1745 restore_locals();
1746 }
1747
1748 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1749 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1750 Label update, next, none;
1751
1752 verify_oop(obj);
1753
1754 cbnz(obj, update);
1755 orptr(mdo_addr, TypeEntries::null_seen);
1756 b(next);
1757
1758 bind(update);
1759 load_klass(obj, obj);
1760
1761 ldr(rscratch1, mdo_addr);
1762 eor(obj, obj, rscratch1);
1763 tst(obj, TypeEntries::type_klass_mask);
1764 br(Assembler::EQ, next); // klass seen before, nothing to
1765 // do. The unknown bit may have been
1766 // set already but no need to check.
1767
1768 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1769 // already unknown. Nothing to do anymore.
1770
1771 cbz(rscratch1, none);
1772 cmp(rscratch1, (u1)TypeEntries::null_seen);
1773 br(Assembler::EQ, none);
1774 // There is a chance that the checks above
1775 // fail if another thread has just set the
1776 // profiling to this obj's klass
1777 eor(obj, obj, rscratch1); // get back original value before XOR
1778 ldr(rscratch1, mdo_addr);
1779 eor(obj, obj, rscratch1);
1780 tst(obj, TypeEntries::type_klass_mask);
1781 br(Assembler::EQ, next);
1782
1783 // different than before. Cannot keep accurate profile.
1784 orptr(mdo_addr, TypeEntries::type_unknown);
1785 b(next);
1786
1787 bind(none);
1788 // first time here. Set profile type.
1789 str(obj, mdo_addr);
1790 #ifdef ASSERT
1791 andr(obj, obj, TypeEntries::type_mask);
1792 verify_klass_ptr(obj);
1793 #endif
1794
1795 bind(next);
1796 }
1797
1798 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1799 if (!ProfileInterpreter) {
1800 return;
1801 }
1802
1803 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1804 Label profile_continue;
1805
1806 test_method_data_pointer(mdp, profile_continue);
1807
1808 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1809
1810 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1811 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1812 br(Assembler::NE, profile_continue);
1813
1814 if (MethodData::profile_arguments()) {
1815 Label done;
1816 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1817
1818 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1819 if (i > 0 || MethodData::profile_return()) {
1820 // If return value type is profiled we may have no argument to profile
1821 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1822 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1823 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1824 add(rscratch1, mdp, off_to_args);
1825 br(Assembler::LT, done);
1826 }
1827 ldr(tmp, Address(callee, Method::const_offset()));
1828 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1829 // stack offset o (zero based) from the start of the argument
1830 // list, for n arguments translates into offset n - o - 1 from
1831 // the end of the argument list
1832 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1833 sub(tmp, tmp, rscratch1);
1834 sub(tmp, tmp, 1);
1835 Address arg_addr = argument_address(tmp);
1836 ldr(tmp, arg_addr);
1837
1838 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1839 profile_obj_type(tmp, mdo_arg_addr);
1840
1841 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1842 off_to_args += to_add;
1843 }
1844
1845 if (MethodData::profile_return()) {
1846 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1847 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1848 }
1849
1850 add(rscratch1, mdp, off_to_args);
1851 bind(done);
1852 mov(mdp, rscratch1);
1853
1854 if (MethodData::profile_return()) {
1855 // We're right after the type profile for the last
1856 // argument. tmp is the number of cells left in the
1857 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1858 // if there's a return to profile.
1859 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1860 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1861 }
1862 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1863 } else {
1864 assert(MethodData::profile_return(), "either profile call args or call ret");
1865 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1866 }
1867
1868 // mdp points right after the end of the
1869 // CallTypeData/VirtualCallTypeData, right after the cells for the
1870 // return value type if there's one
1871
1872 bind(profile_continue);
1873 }
1874 }
1875
1876 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1877 assert_different_registers(mdp, ret, tmp, rbcp);
1878 if (ProfileInterpreter && MethodData::profile_return()) {
1879 Label profile_continue, done;
1880
1881 test_method_data_pointer(mdp, profile_continue);
1882
1883 if (MethodData::profile_return_jsr292_only()) {
1884 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1885
1886 // If we don't profile all invoke bytecodes we must make sure
1887 // it's a bytecode we indeed profile. We can't go back to the
1888 // beginning of the ProfileData we intend to update to check its
1889 // type because we're right after it and we don't known its
1890 // length
1891 Label do_profile;
1892 ldrb(rscratch1, Address(rbcp, 0));
1893 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1894 br(Assembler::EQ, do_profile);
1895 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1896 br(Assembler::EQ, do_profile);
1897 get_method(tmp);
1898 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1899 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1900 br(Assembler::NE, profile_continue);
1901
1902 bind(do_profile);
1903 }
1904
1905 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1906 mov(tmp, ret);
1907 profile_obj_type(tmp, mdo_ret_addr);
1908
1909 bind(profile_continue);
1910 }
1911 }
1912
1913 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1914 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1915 if (ProfileInterpreter && MethodData::profile_parameters()) {
1916 Label profile_continue, done;
1917
1918 test_method_data_pointer(mdp, profile_continue);
1919
1920 // Load the offset of the area within the MDO used for
1921 // parameters. If it's negative we're not profiling any parameters
1922 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1923 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1924
1925 // Compute a pointer to the area for parameters from the offset
1926 // and move the pointer to the slot for the last
1927 // parameters. Collect profiling from last parameter down.
1928 // mdo start + parameters offset + array length - 1
1929 add(mdp, mdp, tmp1);
1930 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1931 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1932
1933 Label loop;
1934 bind(loop);
1935
1936 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1937 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1938 int per_arg_scale = exact_log2(DataLayout::cell_size);
1939 add(rscratch1, mdp, off_base);
1940 add(rscratch2, mdp, type_base);
1941
1942 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1943 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1944
1945 // load offset on the stack from the slot for this parameter
1946 ldr(tmp2, arg_off);
1947 neg(tmp2, tmp2);
1948 // read the parameter from the local area
1949 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1950
1951 // profile the parameter
1952 profile_obj_type(tmp2, arg_type);
1953
1954 // go to next parameter
1955 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1956 br(Assembler::GE, loop);
1957
1958 bind(profile_continue);
1959 }
1960 }
1961
1962 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1963 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1964 get_cache_index_at_bcp(index, 1, sizeof(u4));
1965 // Get address of invokedynamic array
1966 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1967 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1968 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1969 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1970 lea(cache, Address(cache, index));
1971 }
1972
1973 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1974 // Get index out of bytecode pointer
1975 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1976 // Take shortcut if the size is a power of 2
1977 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1978 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1979 } else {
1980 mov(cache, sizeof(ResolvedFieldEntry));
1981 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1982 }
1983 // Get address of field entries array
1984 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1985 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1986 lea(cache, Address(cache, index));
1987 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1988 membar(MacroAssembler::LoadLoad);
1989 }
1990
1991 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1992 // Get index out of bytecode pointer
1993 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1994 mov(cache, sizeof(ResolvedMethodEntry));
1995 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1996
1997 // Get address of field entries array
1998 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1999 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2000 lea(cache, Address(cache, index));
2001 }