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/markWord.hpp"
36 #include "oops/method.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/resolvedFieldEntry.hpp"
39 #include "oops/resolvedIndyEntry.hpp"
40 #include "oops/resolvedMethodEntry.hpp"
41 #include "prims/jvmtiExport.hpp"
42 #include "prims/jvmtiThreadState.hpp"
43 #include "runtime/basicLock.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/javaThread.hpp"
46 #include "runtime/safepointMechanism.hpp"
47 #include "runtime/sharedRuntime.hpp"
48 #include "utilities/powerOfTwo.hpp"
49
50 void InterpreterMacroAssembler::narrow(Register result) {
51
52 // Get method->_constMethod->_result_type
53 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
54 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
55 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
56
57 Label done, notBool, notByte, notChar;
58
59 // common case first
60 cmpw(rscratch1, T_INT);
61 br(Assembler::EQ, done);
62
63 // mask integer result to narrower return type.
64 cmpw(rscratch1, T_BOOLEAN);
65 br(Assembler::NE, notBool);
66 andw(result, result, 0x1);
67 b(done);
68
69 bind(notBool);
70 cmpw(rscratch1, T_BYTE);
71 br(Assembler::NE, notByte);
72 sbfx(result, result, 0, 8);
73 b(done);
74
75 bind(notByte);
76 cmpw(rscratch1, T_CHAR);
77 br(Assembler::NE, notChar);
78 ubfx(result, result, 0, 16); // truncate upper 16 bits
79 b(done);
80
81 bind(notChar);
82 sbfx(result, result, 0, 16); // sign-extend short
83
84 // Nothing to do for T_INT
85 bind(done);
86 }
87
88 void InterpreterMacroAssembler::jump_to_entry(address entry) {
89 assert(entry, "Entry must have been generated by now");
90 b(entry);
91 }
92
93 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
94 if (JvmtiExport::can_pop_frame()) {
95 Label L;
96 // Initiate popframe handling only if it is not already being
97 // processed. If the flag has the popframe_processing bit set, it
98 // means that this code is called *during* popframe handling - we
99 // don't want to reenter.
100 // This method is only called just after the call into the vm in
101 // call_VM_base, so the arg registers are available.
102 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
103 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
104 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
105 // Call Interpreter::remove_activation_preserving_args_entry() to get the
106 // address of the same-named entrypoint in the generated interpreter code.
107 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
108 br(r0);
109 bind(L);
110 }
111 }
112
113
114 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
115 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
116 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
117 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
118 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
119 switch (state) {
120 case atos: ldr(r0, oop_addr);
121 str(zr, oop_addr);
122 interp_verify_oop(r0, state); break;
123 case ltos: ldr(r0, val_addr); break;
124 case btos: // fall through
125 case ztos: // fall through
126 case ctos: // fall through
127 case stos: // fall through
128 case itos: ldrw(r0, val_addr); break;
129 case ftos: ldrs(v0, val_addr); break;
130 case dtos: ldrd(v0, val_addr); break;
131 case vtos: /* nothing to do */ break;
132 default : ShouldNotReachHere();
133 }
134 // Clean up tos value in the thread object
135 movw(rscratch1, (int) ilgl);
136 strw(rscratch1, tos_addr);
137 strw(zr, val_addr);
138 }
139
140
141 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
142 if (JvmtiExport::can_force_early_return()) {
143 Label L;
144 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
145 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
146
147 // Initiate earlyret handling only if it is not already being processed.
148 // If the flag has the earlyret_processing bit set, it means that this code
149 // is called *during* earlyret handling - we don't want to reenter.
150 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
151 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
152 br(Assembler::NE, L);
153
154 // Call Interpreter::remove_activation_early_entry() to get the address of the
155 // same-named entrypoint in the generated interpreter code.
156 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
157 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
158 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
159 br(r0);
160 bind(L);
161 }
162 }
163
164 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
165 Register reg,
166 int bcp_offset) {
167 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
168 ldrh(reg, Address(rbcp, bcp_offset));
169 rev16(reg, reg);
170 }
171
172 void InterpreterMacroAssembler::get_dispatch() {
173 uint64_t offset;
174 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
175 // Use add() here after ARDP, rather than lea().
176 // lea() does not generate anything if its offset is zero.
177 // However, relocs expect to find either an ADD or a load/store
178 // insn after an ADRP. add() always generates an ADD insn, even
179 // for add(Rn, Rn, 0).
180 add(rdispatch, rdispatch, offset);
181 }
182
183 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
184 int bcp_offset,
185 size_t index_size) {
186 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
187 if (index_size == sizeof(u2)) {
188 load_unsigned_short(index, Address(rbcp, bcp_offset));
189 } else if (index_size == sizeof(u4)) {
190 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
191 ldrw(index, Address(rbcp, bcp_offset));
192 } else if (index_size == sizeof(u1)) {
193 load_unsigned_byte(index, Address(rbcp, bcp_offset));
194 } else {
195 ShouldNotReachHere();
196 }
197 }
198
199 void InterpreterMacroAssembler::get_method_counters(Register method,
200 Register mcs, Label& skip) {
201 Label has_counters;
202 ldr(mcs, Address(method, Method::method_counters_offset()));
203 cbnz(mcs, has_counters);
204 call_VM(noreg, CAST_FROM_FN_PTR(address,
205 InterpreterRuntime::build_method_counters), method);
206 ldr(mcs, Address(method, Method::method_counters_offset()));
207 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
208 bind(has_counters);
209 }
210
211 // Load object from cpool->resolved_references(index)
212 void InterpreterMacroAssembler::load_resolved_reference_at_index(
213 Register result, Register index, Register tmp) {
214 assert_different_registers(result, index);
215
216 get_constant_pool(result);
217 // load pointer for resolved_references[] objArray
218 ldr(result, Address(result, ConstantPool::cache_offset()));
219 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
220 resolve_oop_handle(result, tmp, rscratch2);
221 // Add in the index
222 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
223 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
224 }
225
226 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
227 Register cpool, Register index, Register klass, Register temp) {
228 add(temp, cpool, index, LSL, LogBytesPerWord);
229 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
230 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
231 add(klass, klass, temp, LSL, LogBytesPerWord);
232 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
233 }
234
235 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
236 // subtype of super_klass.
237 //
238 // Args:
239 // r0: superklass
240 // Rsub_klass: subklass
241 //
242 // Kills:
243 // r2, r5
244 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
245 Label& ok_is_subtype) {
246 assert(Rsub_klass != r0, "r0 holds superklass");
247 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
248 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
249
250 // Profile the not-null value's klass.
251 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
252
253 // Do the check.
254 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
255 }
256
257 // Java Expression Stack
258
259 void InterpreterMacroAssembler::pop_ptr(Register r) {
260 ldr(r, post(esp, wordSize));
261 }
262
263 void InterpreterMacroAssembler::pop_i(Register r) {
264 ldrw(r, post(esp, wordSize));
265 }
266
267 void InterpreterMacroAssembler::pop_l(Register r) {
268 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
269 }
270
271 void InterpreterMacroAssembler::push_ptr(Register r) {
272 str(r, pre(esp, -wordSize));
273 }
274
275 void InterpreterMacroAssembler::push_i(Register r) {
276 str(r, pre(esp, -wordSize));
277 }
278
279 void InterpreterMacroAssembler::push_l(Register r) {
280 str(zr, pre(esp, -wordSize));
281 str(r, pre(esp, - wordSize));
282 }
283
284 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
285 ldrs(r, post(esp, wordSize));
286 }
287
288 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
289 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
290 }
291
292 void InterpreterMacroAssembler::push_f(FloatRegister r) {
293 strs(r, pre(esp, -wordSize));
294 }
295
296 void InterpreterMacroAssembler::push_d(FloatRegister r) {
297 strd(r, pre(esp, 2* -wordSize));
298 }
299
300 void InterpreterMacroAssembler::pop(TosState state) {
301 switch (state) {
302 case atos: pop_ptr(); break;
303 case btos:
304 case ztos:
305 case ctos:
306 case stos:
307 case itos: pop_i(); break;
308 case ltos: pop_l(); break;
309 case ftos: pop_f(); break;
310 case dtos: pop_d(); break;
311 case vtos: /* nothing to do */ break;
312 default: ShouldNotReachHere();
313 }
314 interp_verify_oop(r0, state);
315 }
316
317 void InterpreterMacroAssembler::push(TosState state) {
318 interp_verify_oop(r0, state);
319 switch (state) {
320 case atos: push_ptr(); break;
321 case btos:
322 case ztos:
323 case ctos:
324 case stos:
325 case itos: push_i(); break;
326 case ltos: push_l(); break;
327 case ftos: push_f(); break;
328 case dtos: push_d(); break;
329 case vtos: /* nothing to do */ break;
330 default : ShouldNotReachHere();
331 }
332 }
333
334 // Helpers for swap and dup
335 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
336 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
337 }
338
339 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
340 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
341 }
342
343 void InterpreterMacroAssembler::load_float(Address src) {
344 ldrs(v0, src);
345 }
346
347 void InterpreterMacroAssembler::load_double(Address src) {
348 ldrd(v0, src);
349 }
350
351 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
352 // set sender sp
353 mov(r19_sender_sp, sp);
354 // record last_sp
355 sub(rscratch1, esp, rfp);
356 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
357 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
358 }
359
360 // Jump to from_interpreted entry of a call unless single stepping is possible
361 // in this thread in which case we must call the i2i entry
362 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
363 prepare_to_jump_from_interpreted();
364
365 if (JvmtiExport::can_post_interpreter_events()) {
366 Label run_compiled_code;
367 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
368 // compiled code in threads for which the event is enabled. Check here for
369 // interp_only_mode if these events CAN be enabled.
370 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
371 cbzw(rscratch1, run_compiled_code);
372 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
373 br(rscratch1);
374 bind(run_compiled_code);
375 }
376
377 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
378 br(rscratch1);
379 }
380
381 // The following two routines provide a hook so that an implementation
382 // can schedule the dispatch in two parts. amd64 does not do this.
383 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
384 }
385
386 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
387 dispatch_next(state, step);
388 }
389
390 void InterpreterMacroAssembler::dispatch_base(TosState state,
391 address* table,
392 bool verifyoop,
393 bool generate_poll) {
394 if (VerifyActivationFrameSize) {
395 Label L;
396 sub(rscratch2, rfp, esp);
397 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
398 subs(rscratch2, rscratch2, min_frame_size);
399 br(Assembler::GE, L);
400 stop("broken stack frame");
401 bind(L);
402 }
403 if (verifyoop) {
404 interp_verify_oop(r0, state);
405 }
406
407 Label safepoint;
408 address* const safepoint_table = Interpreter::safept_table(state);
409 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
410
411 if (needs_thread_local_poll) {
412 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
413 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
414 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
415 }
416
417 if (table == Interpreter::dispatch_table(state)) {
418 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
419 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
420 } else {
421 mov(rscratch2, (address)table);
422 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
423 }
424 br(rscratch2);
425
426 if (needs_thread_local_poll) {
427 bind(safepoint);
428 lea(rscratch2, ExternalAddress((address)safepoint_table));
429 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
430 br(rscratch2);
431 }
432 }
433
434 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
435 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
436 }
437
438 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
439 dispatch_base(state, Interpreter::normal_table(state));
440 }
441
442 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
443 dispatch_base(state, Interpreter::normal_table(state), false);
444 }
445
446
447 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
448 // load next bytecode
449 ldrb(rscratch1, Address(pre(rbcp, step)));
450 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
451 }
452
453 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
454 // load current bytecode
455 ldrb(rscratch1, Address(rbcp, 0));
456 dispatch_base(state, table);
457 }
458
459 // remove activation
460 //
461 // Apply stack watermark barrier.
462 // Unlock the receiver if this is a synchronized method.
463 // Unlock any Java monitors from synchronized blocks.
464 // Remove the activation from the stack.
465 //
466 // If there are locked Java monitors
467 // If throw_monitor_exception
468 // throws IllegalMonitorStateException
469 // Else if install_monitor_exception
470 // installs IllegalMonitorStateException
471 // Else
472 // no error processing
473 void InterpreterMacroAssembler::remove_activation(
474 TosState state,
475 bool throw_monitor_exception,
476 bool install_monitor_exception,
477 bool notify_jvmdi) {
478 // Note: Registers r3 xmm0 may be in use for the
479 // result check if synchronized method
480 Label unlocked, unlock, no_unlock;
481
482 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
483 // that would normally not be safe to use. Such bad returns into unsafe territory of
484 // the stack, will call InterpreterRuntime::at_unwind.
485 Label slow_path;
486 Label fast_path;
487 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
488 br(Assembler::AL, fast_path);
489 bind(slow_path);
490 push(state);
491 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
492 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
493 reset_last_Java_frame(true);
494 pop(state);
495 bind(fast_path);
496
497 // get the value of _do_not_unlock_if_synchronized into r3
498 const Address do_not_unlock_if_synchronized(rthread,
499 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
500 ldrb(r3, do_not_unlock_if_synchronized);
501 strb(zr, do_not_unlock_if_synchronized); // reset the flag
502
503 // get method access flags
504 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
505 ldrh(r2, Address(r1, Method::access_flags_offset()));
506 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
507
508 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
509 // is set.
510 cbnz(r3, no_unlock);
511
512 // unlock monitor
513 push(state); // save result
514
515 // BasicObjectLock will be first in list, since this is a
516 // synchronized method. However, need to check that the object has
517 // not been unlocked by an explicit monitorexit bytecode.
518 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
519 wordSize - (int) sizeof(BasicObjectLock));
520 // We use c_rarg1 so that if we go slow path it will be the correct
521 // register for unlock_object to pass to VM directly
522 lea(c_rarg1, monitor); // address of first monitor
523
524 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
525 cbnz(r0, unlock);
526
527 pop(state);
528 if (throw_monitor_exception) {
529 // Entry already unlocked, need to throw exception
530 call_VM(noreg, CAST_FROM_FN_PTR(address,
531 InterpreterRuntime::throw_illegal_monitor_state_exception));
532 should_not_reach_here();
533 } else {
534 // Monitor already unlocked during a stack unroll. If requested,
535 // install an illegal_monitor_state_exception. Continue with
536 // stack unrolling.
537 if (install_monitor_exception) {
538 call_VM(noreg, CAST_FROM_FN_PTR(address,
539 InterpreterRuntime::new_illegal_monitor_state_exception));
540 }
541 b(unlocked);
542 }
543
544 bind(unlock);
545 unlock_object(c_rarg1);
546 pop(state);
547
548 // Check that for block-structured locking (i.e., that all locked
549 // objects has been unlocked)
550 bind(unlocked);
551
552 // r0: Might contain return value
553
554 // Check that all monitors are unlocked
555 {
556 Label loop, exception, entry, restart;
557 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
558 const Address monitor_block_top(
559 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
560 const Address monitor_block_bot(
561 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
562
563 bind(restart);
564 // We use c_rarg1 so that if we go slow path it will be the correct
565 // register for unlock_object to pass to VM directly
566 ldr(c_rarg1, monitor_block_top); // derelativize pointer
567 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
568 // c_rarg1 points to current entry, starting with top-most entry
569
570 lea(r19, monitor_block_bot); // points to word before bottom of
571 // monitor block
572 b(entry);
573
574 // Entry already locked, need to throw exception
575 bind(exception);
576
577 if (throw_monitor_exception) {
578 // Throw exception
579 MacroAssembler::call_VM(noreg,
580 CAST_FROM_FN_PTR(address, InterpreterRuntime::
581 throw_illegal_monitor_state_exception));
582 should_not_reach_here();
583 } else {
584 // Stack unrolling. Unlock object and install illegal_monitor_exception.
585 // Unlock does not block, so don't have to worry about the frame.
586 // We don't have to preserve c_rarg1 since we are going to throw an exception.
587
588 push(state);
589 unlock_object(c_rarg1);
590 pop(state);
591
592 if (install_monitor_exception) {
593 call_VM(noreg, CAST_FROM_FN_PTR(address,
594 InterpreterRuntime::
595 new_illegal_monitor_state_exception));
596 }
597
598 b(restart);
599 }
600
601 bind(loop);
602 // check if current entry is used
603 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
604 cbnz(rscratch1, exception);
605
606 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
607 bind(entry);
608 cmp(c_rarg1, r19); // check if bottom reached
609 br(Assembler::NE, loop); // if not at bottom then check this entry
610 }
611
612 bind(no_unlock);
613
614 // jvmti support
615 if (notify_jvmdi) {
616 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
617 } else {
618 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
619 }
620
621 // remove activation
622 // get sender esp
623 ldr(rscratch2,
624 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
625 if (StackReservedPages > 0) {
626 // testing if reserved zone needs to be re-enabled
627 Label no_reserved_zone_enabling;
628
629 // check if already enabled - if so no re-enabling needed
630 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
631 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
632 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
633 br(Assembler::EQ, no_reserved_zone_enabling);
634
635 // look for an overflow into the stack reserved zone, i.e.
636 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
637 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
638 cmp(rscratch2, rscratch1);
639 br(Assembler::LS, no_reserved_zone_enabling);
640
641 call_VM_leaf(
642 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
643 call_VM(noreg, CAST_FROM_FN_PTR(address,
644 InterpreterRuntime::throw_delayed_StackOverflowError));
645 should_not_reach_here();
646
647 bind(no_reserved_zone_enabling);
648 }
649
650 // restore sender esp
651 mov(esp, rscratch2);
652 // remove frame anchor
653 leave();
654 // If we're returning to interpreted code we will shortly be
655 // adjusting SP to allow some space for ESP. If we're returning to
656 // compiled code the saved sender SP was saved in sender_sp, so this
657 // restores it.
658 andr(sp, esp, -16);
659 }
660
661 // Lock object
662 //
663 // Args:
664 // c_rarg1: BasicObjectLock to be used for locking
665 //
666 // Kills:
667 // r0
668 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
669 // rscratch1, rscratch2 (scratch regs)
670 void InterpreterMacroAssembler::lock_object(Register lock_reg)
671 {
672 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
673 if (LockingMode == LM_MONITOR) {
674 call_VM_preemptable(noreg,
675 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
676 lock_reg);
677 } else {
678 Label count, done;
679
680 const Register swap_reg = r0;
681 const Register tmp = c_rarg2;
682 const Register obj_reg = c_rarg3; // Will contain the oop
683 const Register tmp2 = c_rarg4;
684 const Register tmp3 = c_rarg5;
685
686 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
687 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
688 const int mark_offset = lock_offset +
689 BasicLock::displaced_header_offset_in_bytes();
690
691 Label slow_case;
692
693 // Load object pointer into obj_reg %c_rarg3
694 ldr(obj_reg, Address(lock_reg, obj_offset));
695
696 if (LockingMode == LM_LIGHTWEIGHT) {
697 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
698 b(done);
699 } else if (LockingMode == LM_LEGACY) {
700
701 if (DiagnoseSyncOnValueBasedClasses != 0) {
702 load_klass(tmp, obj_reg);
703 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
704 tst(tmp, KlassFlags::_misc_is_value_based_class);
705 br(Assembler::NE, slow_case);
706 }
707
708 // Load (object->mark() | 1) into swap_reg
709 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
710 orr(swap_reg, rscratch1, 1);
711
712 // Save (object->mark() | 1) into BasicLock's displaced header
713 str(swap_reg, Address(lock_reg, mark_offset));
714
715 assert(lock_offset == 0,
716 "displached header must be first word in BasicObjectLock");
717
718 Label fail;
719 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
720
721 // Fast check for recursive lock.
722 //
723 // Can apply the optimization only if this is a stack lock
724 // allocated in this thread. For efficiency, we can focus on
725 // recently allocated stack locks (instead of reading the stack
726 // base and checking whether 'mark' points inside the current
727 // thread stack):
728 // 1) (mark & 7) == 0, and
729 // 2) sp <= mark < mark + os::pagesize()
730 //
731 // Warning: sp + os::pagesize can overflow the stack base. We must
732 // neither apply the optimization for an inflated lock allocated
733 // just above the thread stack (this is why condition 1 matters)
734 // nor apply the optimization if the stack lock is inside the stack
735 // of another thread. The latter is avoided even in case of overflow
736 // because we have guard pages at the end of all stacks. Hence, if
737 // we go over the stack base and hit the stack of another thread,
738 // this should not be in a writeable area that could contain a
739 // stack lock allocated by that thread. As a consequence, a stack
740 // lock less than page size away from sp is guaranteed to be
741 // owned by the current thread.
742 //
743 // These 3 tests can be done by evaluating the following
744 // expression: ((mark - sp) & (7 - os::vm_page_size())),
745 // assuming both stack pointer and pagesize have their
746 // least significant 3 bits clear.
747 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
748 // NOTE2: aarch64 does not like to subtract sp from rn so take a
749 // copy
750 mov(rscratch1, sp);
751 sub(swap_reg, swap_reg, rscratch1);
752 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
753
754 // Save the test result, for recursive case, the result is zero
755 str(swap_reg, Address(lock_reg, mark_offset));
756 br(Assembler::NE, slow_case);
757
758 bind(count);
759 inc_held_monitor_count(rscratch1);
760 b(done);
761 }
762 bind(slow_case);
763
764 // Call the runtime routine for slow case
765 call_VM_preemptable(noreg,
766 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
767 lock_reg);
768
769 bind(done);
770 }
771 }
772
773
774 // Unlocks an object. Used in monitorexit bytecode and
775 // remove_activation. Throws an IllegalMonitorException if object is
776 // not locked by current thread.
777 //
778 // Args:
779 // c_rarg1: BasicObjectLock for lock
780 //
781 // Kills:
782 // r0
783 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
784 // rscratch1, rscratch2 (scratch regs)
785 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
786 {
787 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
788
789 if (LockingMode == LM_MONITOR) {
790 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
791 } else {
792 Label count, done;
793
794 const Register swap_reg = r0;
795 const Register header_reg = c_rarg2; // Will contain the old oopMark
796 const Register obj_reg = c_rarg3; // Will contain the oop
797 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
798
799 save_bcp(); // Save in case of exception
800
801 if (LockingMode != LM_LIGHTWEIGHT) {
802 // Convert from BasicObjectLock structure to object and BasicLock
803 // structure Store the BasicLock address into %r0
804 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
805 }
806
807 // Load oop into obj_reg(%c_rarg3)
808 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
809
810 // Free entry
811 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
812
813 Label slow_case;
814 if (LockingMode == LM_LIGHTWEIGHT) {
815 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
816 b(done);
817 } else if (LockingMode == LM_LEGACY) {
818 // Load the old header from BasicLock structure
819 ldr(header_reg, Address(swap_reg,
820 BasicLock::displaced_header_offset_in_bytes()));
821
822 // Test for recursion
823 cbz(header_reg, count);
824
825 // Atomic swap back the old header
826 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
827
828 bind(count);
829 dec_held_monitor_count(rscratch1);
830 b(done);
831 }
832
833 bind(slow_case);
834 // Call the runtime routine for slow case.
835 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
836 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
837 bind(done);
838 restore_bcp();
839 }
840 }
841
842 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
843 Label& zero_continue) {
844 assert(ProfileInterpreter, "must be profiling interpreter");
845 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
846 cbz(mdp, zero_continue);
847 }
848
849 // Set the method data pointer for the current bcp.
850 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
851 assert(ProfileInterpreter, "must be profiling interpreter");
852 Label set_mdp;
853 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
854
855 // Test MDO to avoid the call if it is null.
856 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
857 cbz(r0, set_mdp);
858 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
859 // r0: mdi
860 // mdo is guaranteed to be non-zero here, we checked for it before the call.
861 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
862 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
863 add(r0, r1, r0);
864 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
865 bind(set_mdp);
866 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
867 }
868
869 void InterpreterMacroAssembler::verify_method_data_pointer() {
870 assert(ProfileInterpreter, "must be profiling interpreter");
871 #ifdef ASSERT
872 Label verify_continue;
873 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
874 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
875 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
876 get_method(r1);
877
878 // If the mdp is valid, it will point to a DataLayout header which is
879 // consistent with the bcp. The converse is highly probable also.
880 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
881 ldr(rscratch1, Address(r1, Method::const_offset()));
882 add(r2, r2, rscratch1, Assembler::LSL);
883 lea(r2, Address(r2, ConstMethod::codes_offset()));
884 cmp(r2, rbcp);
885 br(Assembler::EQ, verify_continue);
886 // r1: method
887 // rbcp: bcp // rbcp == 22
888 // r3: mdp
889 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
890 r1, rbcp, r3);
891 bind(verify_continue);
892 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
893 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
894 #endif // ASSERT
895 }
896
897
898 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
899 int constant,
900 Register value) {
901 assert(ProfileInterpreter, "must be profiling interpreter");
902 Address data(mdp_in, constant);
903 str(value, data);
904 }
905
906
907 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
908 int constant,
909 bool decrement) {
910 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
911 }
912
913 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
914 Register reg,
915 int constant,
916 bool decrement) {
917 assert(ProfileInterpreter, "must be profiling interpreter");
918 // %%% this does 64bit counters at best it is wasting space
919 // at worst it is a rare bug when counters overflow
920
921 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
922
923 Address addr1(mdp_in, constant);
924 Address addr2(rscratch2, reg, Address::lsl(0));
925 Address &addr = addr1;
926 if (reg != noreg) {
927 lea(rscratch2, addr1);
928 addr = addr2;
929 }
930
931 if (decrement) {
932 // Decrement the register. Set condition codes.
933 // Intel does this
934 // addptr(data, (int32_t) -DataLayout::counter_increment);
935 // If the decrement causes the counter to overflow, stay negative
936 // Label L;
937 // jcc(Assembler::negative, L);
938 // addptr(data, (int32_t) DataLayout::counter_increment);
939 // so we do this
940 ldr(rscratch1, addr);
941 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
942 Label L;
943 br(Assembler::LO, L); // skip store if counter underflow
944 str(rscratch1, addr);
945 bind(L);
946 } else {
947 assert(DataLayout::counter_increment == 1,
948 "flow-free idiom only works with 1");
949 // Intel does this
950 // Increment the register. Set carry flag.
951 // addptr(data, DataLayout::counter_increment);
952 // If the increment causes the counter to overflow, pull back by 1.
953 // sbbptr(data, (int32_t)0);
954 // so we do this
955 ldr(rscratch1, addr);
956 adds(rscratch1, rscratch1, DataLayout::counter_increment);
957 Label L;
958 br(Assembler::CS, L); // skip store if counter overflow
959 str(rscratch1, addr);
960 bind(L);
961 }
962 }
963
964 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
965 int flag_byte_constant) {
966 assert(ProfileInterpreter, "must be profiling interpreter");
967 int flags_offset = in_bytes(DataLayout::flags_offset());
968 // Set the flag
969 ldrb(rscratch1, Address(mdp_in, flags_offset));
970 orr(rscratch1, rscratch1, flag_byte_constant);
971 strb(rscratch1, Address(mdp_in, flags_offset));
972 }
973
974
975 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
976 int offset,
977 Register value,
978 Register test_value_out,
979 Label& not_equal_continue) {
980 assert(ProfileInterpreter, "must be profiling interpreter");
981 if (test_value_out == noreg) {
982 ldr(rscratch1, Address(mdp_in, offset));
983 cmp(value, rscratch1);
984 } else {
985 // Put the test value into a register, so caller can use it:
986 ldr(test_value_out, Address(mdp_in, offset));
987 cmp(value, test_value_out);
988 }
989 br(Assembler::NE, not_equal_continue);
990 }
991
992
993 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
994 int offset_of_disp) {
995 assert(ProfileInterpreter, "must be profiling interpreter");
996 ldr(rscratch1, Address(mdp_in, offset_of_disp));
997 add(mdp_in, mdp_in, rscratch1, LSL);
998 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
999 }
1000
1001
1002 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1003 Register reg,
1004 int offset_of_disp) {
1005 assert(ProfileInterpreter, "must be profiling interpreter");
1006 lea(rscratch1, Address(mdp_in, offset_of_disp));
1007 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1008 add(mdp_in, mdp_in, rscratch1, LSL);
1009 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1010 }
1011
1012
1013 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1014 int constant) {
1015 assert(ProfileInterpreter, "must be profiling interpreter");
1016 add(mdp_in, mdp_in, (unsigned)constant);
1017 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1018 }
1019
1020
1021 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1022 assert(ProfileInterpreter, "must be profiling interpreter");
1023 // save/restore across call_VM
1024 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1025 call_VM(noreg,
1026 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1027 return_bci);
1028 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1029 }
1030
1031
1032 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1033 Register bumped_count) {
1034 if (ProfileInterpreter) {
1035 Label profile_continue;
1036
1037 // If no method data exists, go to profile_continue.
1038 // Otherwise, assign to mdp
1039 test_method_data_pointer(mdp, profile_continue);
1040
1041 // We are taking a branch. Increment the taken count.
1042 // We inline increment_mdp_data_at to return bumped_count in a register
1043 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1044 Address data(mdp, in_bytes(JumpData::taken_offset()));
1045 ldr(bumped_count, data);
1046 assert(DataLayout::counter_increment == 1,
1047 "flow-free idiom only works with 1");
1048 // Intel does this to catch overflow
1049 // addptr(bumped_count, DataLayout::counter_increment);
1050 // sbbptr(bumped_count, 0);
1051 // so we do this
1052 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1053 Label L;
1054 br(Assembler::CS, L); // skip store if counter overflow
1055 str(bumped_count, data);
1056 bind(L);
1057 // The method data pointer needs to be updated to reflect the new target.
1058 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1059 bind(profile_continue);
1060 }
1061 }
1062
1063
1064 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1065 if (ProfileInterpreter) {
1066 Label profile_continue;
1067
1068 // If no method data exists, go to profile_continue.
1069 test_method_data_pointer(mdp, profile_continue);
1070
1071 // We are taking a branch. Increment the not taken count.
1072 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1073
1074 // The method data pointer needs to be updated to correspond to
1075 // the next bytecode
1076 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1077 bind(profile_continue);
1078 }
1079 }
1080
1081
1082 void InterpreterMacroAssembler::profile_call(Register mdp) {
1083 if (ProfileInterpreter) {
1084 Label profile_continue;
1085
1086 // If no method data exists, go to profile_continue.
1087 test_method_data_pointer(mdp, profile_continue);
1088
1089 // We are making a call. Increment the count.
1090 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1091
1092 // The method data pointer needs to be updated to reflect the new target.
1093 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1094 bind(profile_continue);
1095 }
1096 }
1097
1098 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1099 if (ProfileInterpreter) {
1100 Label profile_continue;
1101
1102 // If no method data exists, go to profile_continue.
1103 test_method_data_pointer(mdp, profile_continue);
1104
1105 // We are making a call. Increment the count.
1106 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1107
1108 // The method data pointer needs to be updated to reflect the new target.
1109 update_mdp_by_constant(mdp,
1110 in_bytes(VirtualCallData::
1111 virtual_call_data_size()));
1112 bind(profile_continue);
1113 }
1114 }
1115
1116
1117 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1118 Register mdp,
1119 Register reg2,
1120 bool receiver_can_be_null) {
1121 if (ProfileInterpreter) {
1122 Label profile_continue;
1123
1124 // If no method data exists, go to profile_continue.
1125 test_method_data_pointer(mdp, profile_continue);
1126
1127 Label skip_receiver_profile;
1128 if (receiver_can_be_null) {
1129 Label not_null;
1130 // We are making a call. Increment the count for null receiver.
1131 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1132 b(skip_receiver_profile);
1133 bind(not_null);
1134 }
1135
1136 // Record the receiver type.
1137 record_klass_in_profile(receiver, mdp, reg2);
1138 bind(skip_receiver_profile);
1139
1140 // The method data pointer needs to be updated to reflect the new target.
1141 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1142 bind(profile_continue);
1143 }
1144 }
1145
1146 // This routine creates a state machine for updating the multi-row
1147 // type profile at a virtual call site (or other type-sensitive bytecode).
1148 // The machine visits each row (of receiver/count) until the receiver type
1149 // is found, or until it runs out of rows. At the same time, it remembers
1150 // the location of the first empty row. (An empty row records null for its
1151 // receiver, and can be allocated for a newly-observed receiver type.)
1152 // Because there are two degrees of freedom in the state, a simple linear
1153 // search will not work; it must be a decision tree. Hence this helper
1154 // function is recursive, to generate the required tree structured code.
1155 // It's the interpreter, so we are trading off code space for speed.
1156 // See below for example code.
1157 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1158 Register receiver, Register mdp,
1159 Register reg2, int start_row,
1160 Label& done) {
1161 if (TypeProfileWidth == 0) {
1162 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1163 } else {
1164 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1165 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1166 }
1167 }
1168
1169 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1170 Register reg2, int start_row, Label& done, int total_rows,
1171 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1172 int last_row = total_rows - 1;
1173 assert(start_row <= last_row, "must be work left to do");
1174 // Test this row for both the item and for null.
1175 // Take any of three different outcomes:
1176 // 1. found item => increment count and goto done
1177 // 2. found null => keep looking for case 1, maybe allocate this cell
1178 // 3. found something else => keep looking for cases 1 and 2
1179 // Case 3 is handled by a recursive call.
1180 for (int row = start_row; row <= last_row; row++) {
1181 Label next_test;
1182 bool test_for_null_also = (row == start_row);
1183
1184 // See if the item is item[n].
1185 int item_offset = in_bytes(item_offset_fn(row));
1186 test_mdp_data_at(mdp, item_offset, item,
1187 (test_for_null_also ? reg2 : noreg),
1188 next_test);
1189 // (Reg2 now contains the item from the CallData.)
1190
1191 // The item is item[n]. Increment count[n].
1192 int count_offset = in_bytes(item_count_offset_fn(row));
1193 increment_mdp_data_at(mdp, count_offset);
1194 b(done);
1195 bind(next_test);
1196
1197 if (test_for_null_also) {
1198 Label found_null;
1199 // Failed the equality check on item[n]... Test for null.
1200 if (start_row == last_row) {
1201 // The only thing left to do is handle the null case.
1202 cbz(reg2, found_null);
1203 // Item did not match any saved item and there is no empty row for it.
1204 // Increment total counter to indicate polymorphic case.
1205 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1206 b(done);
1207 bind(found_null);
1208 break;
1209 }
1210 // Since null is rare, make it be the branch-taken case.
1211 cbz(reg2, found_null);
1212
1213 // Put all the "Case 3" tests here.
1214 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1215 item_offset_fn, item_count_offset_fn);
1216
1217 // Found a null. Keep searching for a matching item,
1218 // but remember that this is an empty (unused) slot.
1219 bind(found_null);
1220 }
1221 }
1222
1223 // In the fall-through case, we found no matching item, but we
1224 // observed the item[start_row] is null.
1225
1226 // Fill in the item field and increment the count.
1227 int item_offset = in_bytes(item_offset_fn(start_row));
1228 set_mdp_data_at(mdp, item_offset, item);
1229 int count_offset = in_bytes(item_count_offset_fn(start_row));
1230 mov(reg2, DataLayout::counter_increment);
1231 set_mdp_data_at(mdp, count_offset, reg2);
1232 if (start_row > 0) {
1233 b(done);
1234 }
1235 }
1236
1237 // Example state machine code for three profile rows:
1238 // // main copy of decision tree, rooted at row[1]
1239 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1240 // if (row[0].rec != nullptr) {
1241 // // inner copy of decision tree, rooted at row[1]
1242 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1243 // if (row[1].rec != nullptr) {
1244 // // degenerate decision tree, rooted at row[2]
1245 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1246 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1247 // row[2].init(rec); goto done;
1248 // } else {
1249 // // remember row[1] is empty
1250 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1251 // row[1].init(rec); goto done;
1252 // }
1253 // } else {
1254 // // remember row[0] is empty
1255 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1256 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1257 // row[0].init(rec); goto done;
1258 // }
1259 // done:
1260
1261 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1262 Register mdp, Register reg2) {
1263 assert(ProfileInterpreter, "must be profiling");
1264 Label done;
1265
1266 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1267
1268 bind (done);
1269 }
1270
1271 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1272 Register mdp) {
1273 if (ProfileInterpreter) {
1274 Label profile_continue;
1275 uint row;
1276
1277 // If no method data exists, go to profile_continue.
1278 test_method_data_pointer(mdp, profile_continue);
1279
1280 // Update the total ret count.
1281 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1282
1283 for (row = 0; row < RetData::row_limit(); row++) {
1284 Label next_test;
1285
1286 // See if return_bci is equal to bci[n]:
1287 test_mdp_data_at(mdp,
1288 in_bytes(RetData::bci_offset(row)),
1289 return_bci, noreg,
1290 next_test);
1291
1292 // return_bci is equal to bci[n]. Increment the count.
1293 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1294
1295 // The method data pointer needs to be updated to reflect the new target.
1296 update_mdp_by_offset(mdp,
1297 in_bytes(RetData::bci_displacement_offset(row)));
1298 b(profile_continue);
1299 bind(next_test);
1300 }
1301
1302 update_mdp_for_ret(return_bci);
1303
1304 bind(profile_continue);
1305 }
1306 }
1307
1308 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1309 if (ProfileInterpreter) {
1310 Label profile_continue;
1311
1312 // If no method data exists, go to profile_continue.
1313 test_method_data_pointer(mdp, profile_continue);
1314
1315 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1316
1317 // The method data pointer needs to be updated.
1318 int mdp_delta = in_bytes(BitData::bit_data_size());
1319 if (TypeProfileCasts) {
1320 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1321 }
1322 update_mdp_by_constant(mdp, mdp_delta);
1323
1324 bind(profile_continue);
1325 }
1326 }
1327
1328 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1329 if (ProfileInterpreter) {
1330 Label profile_continue;
1331
1332 // If no method data exists, go to profile_continue.
1333 test_method_data_pointer(mdp, profile_continue);
1334
1335 // The method data pointer needs to be updated.
1336 int mdp_delta = in_bytes(BitData::bit_data_size());
1337 if (TypeProfileCasts) {
1338 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1339
1340 // Record the object type.
1341 record_klass_in_profile(klass, mdp, reg2);
1342 }
1343 update_mdp_by_constant(mdp, mdp_delta);
1344
1345 bind(profile_continue);
1346 }
1347 }
1348
1349 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1350 if (ProfileInterpreter) {
1351 Label profile_continue;
1352
1353 // If no method data exists, go to profile_continue.
1354 test_method_data_pointer(mdp, profile_continue);
1355
1356 // Update the default case count
1357 increment_mdp_data_at(mdp,
1358 in_bytes(MultiBranchData::default_count_offset()));
1359
1360 // The method data pointer needs to be updated.
1361 update_mdp_by_offset(mdp,
1362 in_bytes(MultiBranchData::
1363 default_displacement_offset()));
1364
1365 bind(profile_continue);
1366 }
1367 }
1368
1369 void InterpreterMacroAssembler::profile_switch_case(Register index,
1370 Register mdp,
1371 Register reg2) {
1372 if (ProfileInterpreter) {
1373 Label profile_continue;
1374
1375 // If no method data exists, go to profile_continue.
1376 test_method_data_pointer(mdp, profile_continue);
1377
1378 // Build the base (index * per_case_size_in_bytes()) +
1379 // case_array_offset_in_bytes()
1380 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1381 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1382 Assembler::maddw(index, index, reg2, rscratch1);
1383
1384 // Update the case count
1385 increment_mdp_data_at(mdp,
1386 index,
1387 in_bytes(MultiBranchData::relative_count_offset()));
1388
1389 // The method data pointer needs to be updated.
1390 update_mdp_by_offset(mdp,
1391 index,
1392 in_bytes(MultiBranchData::
1393 relative_displacement_offset()));
1394
1395 bind(profile_continue);
1396 }
1397 }
1398
1399 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1400 if (state == atos) {
1401 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1402 }
1403 }
1404
1405 void InterpreterMacroAssembler::notify_method_entry() {
1406 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1407 // track stack depth. If it is possible to enter interp_only_mode we add
1408 // the code to check if the event should be sent.
1409 if (JvmtiExport::can_post_interpreter_events()) {
1410 Label L;
1411 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1412 cbzw(r3, L);
1413 call_VM(noreg, CAST_FROM_FN_PTR(address,
1414 InterpreterRuntime::post_method_entry));
1415 bind(L);
1416 }
1417
1418 if (DTraceMethodProbes) {
1419 get_method(c_rarg1);
1420 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1421 rthread, c_rarg1);
1422 }
1423
1424 // RedefineClasses() tracing support for obsolete method entry
1425 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1426 get_method(c_rarg1);
1427 call_VM_leaf(
1428 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1429 rthread, c_rarg1);
1430 }
1431
1432 }
1433
1434
1435 void InterpreterMacroAssembler::notify_method_exit(
1436 TosState state, NotifyMethodExitMode mode) {
1437 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1438 // track stack depth. If it is possible to enter interp_only_mode we add
1439 // the code to check if the event should be sent.
1440 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1441 Label L;
1442 // Note: frame::interpreter_frame_result has a dependency on how the
1443 // method result is saved across the call to post_method_exit. If this
1444 // is changed then the interpreter_frame_result implementation will
1445 // need to be updated too.
1446
1447 // template interpreter will leave the result on the top of the stack.
1448 push(state);
1449 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1450 cbz(r3, L);
1451 call_VM(noreg,
1452 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1453 bind(L);
1454 pop(state);
1455 }
1456
1457 if (DTraceMethodProbes) {
1458 push(state);
1459 get_method(c_rarg1);
1460 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1461 rthread, c_rarg1);
1462 pop(state);
1463 }
1464 }
1465
1466
1467 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1468 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1469 int increment, Address mask,
1470 Register scratch, Register scratch2,
1471 bool preloaded, Condition cond,
1472 Label* where) {
1473 if (!preloaded) {
1474 ldrw(scratch, counter_addr);
1475 }
1476 add(scratch, scratch, increment);
1477 strw(scratch, counter_addr);
1478 ldrw(scratch2, mask);
1479 ands(scratch, scratch, scratch2);
1480 br(cond, *where);
1481 }
1482
1483 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1484 int number_of_arguments) {
1485 // interpreter specific
1486 //
1487 // Note: No need to save/restore rbcp & rlocals pointer since these
1488 // are callee saved registers and no blocking/ GC can happen
1489 // in leaf calls.
1490 #ifdef ASSERT
1491 {
1492 Label L;
1493 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1494 cbz(rscratch1, L);
1495 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1496 " last_sp != nullptr");
1497 bind(L);
1498 }
1499 #endif /* ASSERT */
1500 // super call
1501 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1502 }
1503
1504 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1505 Register java_thread,
1506 Register last_java_sp,
1507 address entry_point,
1508 int number_of_arguments,
1509 bool check_exceptions) {
1510 // interpreter specific
1511 //
1512 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1513 // really make a difference for these runtime calls, since they are
1514 // slow anyway. Btw., bcp must be saved/restored since it may change
1515 // due to GC.
1516 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1517 save_bcp();
1518 #ifdef ASSERT
1519 {
1520 Label L;
1521 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1522 cbz(rscratch1, L);
1523 stop("InterpreterMacroAssembler::call_VM_base:"
1524 " last_sp != nullptr");
1525 bind(L);
1526 }
1527 #endif /* ASSERT */
1528 // super call
1529 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1530 entry_point, number_of_arguments,
1531 check_exceptions);
1532 // interpreter specific
1533 restore_bcp();
1534 restore_locals();
1535 }
1536
1537 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1538 address entry_point,
1539 Register arg_1) {
1540 assert(arg_1 == c_rarg1, "");
1541 Label resume_pc, not_preempted;
1542
1543 #ifdef ASSERT
1544 {
1545 Label L;
1546 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1547 cbz(rscratch1, L);
1548 stop("Should not have alternate return address set");
1549 bind(L);
1550 }
1551 #endif /* ASSERT */
1552
1553 // Force freeze slow path.
1554 push_cont_fastpath();
1555
1556 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1557 adr(rscratch1, resume_pc);
1558 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1559 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1560
1561 pop_cont_fastpath();
1562
1563 // Check if preempted.
1564 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1565 cbz(rscratch1, not_preempted);
1566 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1567 br(rscratch1);
1568
1569 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1570 bind(resume_pc);
1571 restore_after_resume(false /* is_native */);
1572
1573 bind(not_preempted);
1574 }
1575
1576 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1577 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1578 blr(rscratch1);
1579 if (is_native) {
1580 // On resume we need to set up stack as expected
1581 push(dtos);
1582 push(ltos);
1583 }
1584 }
1585
1586 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1587 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1588 Label update, next, none;
1589
1590 verify_oop(obj);
1591
1592 cbnz(obj, update);
1593 orptr(mdo_addr, TypeEntries::null_seen);
1594 b(next);
1595
1596 bind(update);
1597 load_klass(obj, obj);
1598
1599 ldr(rscratch1, mdo_addr);
1600 eor(obj, obj, rscratch1);
1601 tst(obj, TypeEntries::type_klass_mask);
1602 br(Assembler::EQ, next); // klass seen before, nothing to
1603 // do. The unknown bit may have been
1604 // set already but no need to check.
1605
1606 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1607 // already unknown. Nothing to do anymore.
1608
1609 cbz(rscratch1, none);
1610 cmp(rscratch1, (u1)TypeEntries::null_seen);
1611 br(Assembler::EQ, none);
1612 // There is a chance that the checks above
1613 // fail if another thread has just set the
1614 // profiling to this obj's klass
1615 eor(obj, obj, rscratch1); // get back original value before XOR
1616 ldr(rscratch1, mdo_addr);
1617 eor(obj, obj, rscratch1);
1618 tst(obj, TypeEntries::type_klass_mask);
1619 br(Assembler::EQ, next);
1620
1621 // different than before. Cannot keep accurate profile.
1622 orptr(mdo_addr, TypeEntries::type_unknown);
1623 b(next);
1624
1625 bind(none);
1626 // first time here. Set profile type.
1627 str(obj, mdo_addr);
1628 #ifdef ASSERT
1629 andr(obj, obj, TypeEntries::type_mask);
1630 verify_klass_ptr(obj);
1631 #endif
1632
1633 bind(next);
1634 }
1635
1636 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1637 if (!ProfileInterpreter) {
1638 return;
1639 }
1640
1641 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1642 Label profile_continue;
1643
1644 test_method_data_pointer(mdp, profile_continue);
1645
1646 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1647
1648 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1649 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1650 br(Assembler::NE, profile_continue);
1651
1652 if (MethodData::profile_arguments()) {
1653 Label done;
1654 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1655
1656 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1657 if (i > 0 || MethodData::profile_return()) {
1658 // If return value type is profiled we may have no argument to profile
1659 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1660 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1661 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1662 add(rscratch1, mdp, off_to_args);
1663 br(Assembler::LT, done);
1664 }
1665 ldr(tmp, Address(callee, Method::const_offset()));
1666 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1667 // stack offset o (zero based) from the start of the argument
1668 // list, for n arguments translates into offset n - o - 1 from
1669 // the end of the argument list
1670 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1671 sub(tmp, tmp, rscratch1);
1672 sub(tmp, tmp, 1);
1673 Address arg_addr = argument_address(tmp);
1674 ldr(tmp, arg_addr);
1675
1676 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1677 profile_obj_type(tmp, mdo_arg_addr);
1678
1679 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1680 off_to_args += to_add;
1681 }
1682
1683 if (MethodData::profile_return()) {
1684 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1685 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1686 }
1687
1688 add(rscratch1, mdp, off_to_args);
1689 bind(done);
1690 mov(mdp, rscratch1);
1691
1692 if (MethodData::profile_return()) {
1693 // We're right after the type profile for the last
1694 // argument. tmp is the number of cells left in the
1695 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1696 // if there's a return to profile.
1697 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1698 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1699 }
1700 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1701 } else {
1702 assert(MethodData::profile_return(), "either profile call args or call ret");
1703 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1704 }
1705
1706 // mdp points right after the end of the
1707 // CallTypeData/VirtualCallTypeData, right after the cells for the
1708 // return value type if there's one
1709
1710 bind(profile_continue);
1711 }
1712 }
1713
1714 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1715 assert_different_registers(mdp, ret, tmp, rbcp);
1716 if (ProfileInterpreter && MethodData::profile_return()) {
1717 Label profile_continue, done;
1718
1719 test_method_data_pointer(mdp, profile_continue);
1720
1721 if (MethodData::profile_return_jsr292_only()) {
1722 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1723
1724 // If we don't profile all invoke bytecodes we must make sure
1725 // it's a bytecode we indeed profile. We can't go back to the
1726 // beginning of the ProfileData we intend to update to check its
1727 // type because we're right after it and we don't known its
1728 // length
1729 Label do_profile;
1730 ldrb(rscratch1, Address(rbcp, 0));
1731 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1732 br(Assembler::EQ, do_profile);
1733 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1734 br(Assembler::EQ, do_profile);
1735 get_method(tmp);
1736 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1737 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1738 br(Assembler::NE, profile_continue);
1739
1740 bind(do_profile);
1741 }
1742
1743 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1744 mov(tmp, ret);
1745 profile_obj_type(tmp, mdo_ret_addr);
1746
1747 bind(profile_continue);
1748 }
1749 }
1750
1751 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1752 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1753 if (ProfileInterpreter && MethodData::profile_parameters()) {
1754 Label profile_continue, done;
1755
1756 test_method_data_pointer(mdp, profile_continue);
1757
1758 // Load the offset of the area within the MDO used for
1759 // parameters. If it's negative we're not profiling any parameters
1760 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1761 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1762
1763 // Compute a pointer to the area for parameters from the offset
1764 // and move the pointer to the slot for the last
1765 // parameters. Collect profiling from last parameter down.
1766 // mdo start + parameters offset + array length - 1
1767 add(mdp, mdp, tmp1);
1768 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1769 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1770
1771 Label loop;
1772 bind(loop);
1773
1774 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1775 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1776 int per_arg_scale = exact_log2(DataLayout::cell_size);
1777 add(rscratch1, mdp, off_base);
1778 add(rscratch2, mdp, type_base);
1779
1780 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1781 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1782
1783 // load offset on the stack from the slot for this parameter
1784 ldr(tmp2, arg_off);
1785 neg(tmp2, tmp2);
1786 // read the parameter from the local area
1787 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1788
1789 // profile the parameter
1790 profile_obj_type(tmp2, arg_type);
1791
1792 // go to next parameter
1793 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1794 br(Assembler::GE, loop);
1795
1796 bind(profile_continue);
1797 }
1798 }
1799
1800 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1801 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1802 get_cache_index_at_bcp(index, 1, sizeof(u4));
1803 // Get address of invokedynamic array
1804 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1805 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1806 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1807 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1808 lea(cache, Address(cache, index));
1809 }
1810
1811 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1812 // Get index out of bytecode pointer
1813 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1814 // Take shortcut if the size is a power of 2
1815 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1816 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1817 } else {
1818 mov(cache, sizeof(ResolvedFieldEntry));
1819 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1820 }
1821 // Get address of field entries array
1822 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1823 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1824 lea(cache, Address(cache, index));
1825 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1826 membar(MacroAssembler::LoadLoad);
1827 }
1828
1829 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1830 // Get index out of bytecode pointer
1831 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1832 mov(cache, sizeof(ResolvedMethodEntry));
1833 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1834
1835 // Get address of field entries array
1836 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1837 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1838 lea(cache, Address(cache, index));
1839 }