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 "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/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/resolvedFieldEntry.hpp"
40 #include "oops/resolvedIndyEntry.hpp"
41 #include "oops/resolvedMethodEntry.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "prims/jvmtiThreadState.hpp"
44 #include "runtime/basicLock.hpp"
45 #include "runtime/frame.inline.hpp"
46 #include "runtime/javaThread.hpp"
47 #include "runtime/safepointMechanism.hpp"
48 #include "runtime/sharedRuntime.hpp"
49 #include "utilities/powerOfTwo.hpp"
50
51 void InterpreterMacroAssembler::narrow(Register result) {
52
53 // Get method->_constMethod->_result_type
54 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
55 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
56 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
57
58 Label done, notBool, notByte, notChar;
59
60 // common case first
61 cmpw(rscratch1, T_INT);
62 br(Assembler::EQ, done);
63
64 // mask integer result to narrower return type.
65 cmpw(rscratch1, T_BOOLEAN);
66 br(Assembler::NE, notBool);
67 andw(result, result, 0x1);
68 b(done);
69
70 bind(notBool);
71 cmpw(rscratch1, T_BYTE);
72 br(Assembler::NE, notByte);
73 sbfx(result, result, 0, 8);
74 b(done);
75
76 bind(notByte);
77 cmpw(rscratch1, T_CHAR);
78 br(Assembler::NE, notChar);
79 ubfx(result, result, 0, 16); // truncate upper 16 bits
80 b(done);
81
82 bind(notChar);
83 sbfx(result, result, 0, 16); // sign-extend short
84
85 // Nothing to do for T_INT
86 bind(done);
87 }
88
89 void InterpreterMacroAssembler::jump_to_entry(address entry) {
90 assert(entry, "Entry must have been generated by now");
91 b(entry);
92 }
93
94 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
95 if (JvmtiExport::can_pop_frame()) {
96 Label L;
97 // Initiate popframe handling only if it is not already being
98 // processed. If the flag has the popframe_processing bit set, it
99 // means that this code is called *during* popframe handling - we
100 // don't want to reenter.
101 // This method is only called just after the call into the vm in
102 // call_VM_base, so the arg registers are available.
103 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
104 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
105 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
106 // Call Interpreter::remove_activation_preserving_args_entry() to get the
107 // address of the same-named entrypoint in the generated interpreter code.
108 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
109 br(r0);
110 bind(L);
111 }
112 }
113
114
115 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
116 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
117 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
118 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
119 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
120 switch (state) {
121 case atos: ldr(r0, oop_addr);
122 str(zr, oop_addr);
123 interp_verify_oop(r0, state); break;
124 case ltos: ldr(r0, val_addr); break;
125 case btos: // fall through
126 case ztos: // fall through
127 case ctos: // fall through
128 case stos: // fall through
129 case itos: ldrw(r0, val_addr); break;
130 case ftos: ldrs(v0, val_addr); break;
131 case dtos: ldrd(v0, val_addr); break;
132 case vtos: /* nothing to do */ break;
133 default : ShouldNotReachHere();
134 }
135 // Clean up tos value in the thread object
136 movw(rscratch1, (int) ilgl);
137 strw(rscratch1, tos_addr);
138 strw(zr, val_addr);
139 }
140
141
142 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
143 if (JvmtiExport::can_force_early_return()) {
144 Label L;
145 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
146 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
147
148 // Initiate earlyret handling only if it is not already being processed.
149 // If the flag has the earlyret_processing bit set, it means that this code
150 // is called *during* earlyret handling - we don't want to reenter.
151 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
152 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
153 br(Assembler::NE, L);
154
155 // Call Interpreter::remove_activation_early_entry() to get the address of the
156 // same-named entrypoint in the generated interpreter code.
157 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
158 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
159 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
160 br(r0);
161 bind(L);
162 }
163 }
164
165 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
166 Register reg,
167 int bcp_offset) {
168 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
169 ldrh(reg, Address(rbcp, bcp_offset));
170 rev16(reg, reg);
171 }
172
173 void InterpreterMacroAssembler::get_dispatch() {
174 uint64_t offset;
175 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
176 // Use add() here after ARDP, rather than lea().
177 // lea() does not generate anything if its offset is zero.
178 // However, relocs expect to find either an ADD or a load/store
179 // insn after an ADRP. add() always generates an ADD insn, even
180 // for add(Rn, Rn, 0).
181 add(rdispatch, rdispatch, offset);
182 }
183
184 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
185 int bcp_offset,
186 size_t index_size) {
187 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
188 if (index_size == sizeof(u2)) {
189 load_unsigned_short(index, Address(rbcp, bcp_offset));
190 } else if (index_size == sizeof(u4)) {
191 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
192 ldrw(index, Address(rbcp, bcp_offset));
193 } else if (index_size == sizeof(u1)) {
194 load_unsigned_byte(index, Address(rbcp, bcp_offset));
195 } else {
196 ShouldNotReachHere();
197 }
198 }
199
200 void InterpreterMacroAssembler::get_method_counters(Register method,
201 Register mcs, Label& skip) {
202 Label has_counters;
203 ldr(mcs, Address(method, Method::method_counters_offset()));
204 cbnz(mcs, has_counters);
205 call_VM(noreg, CAST_FROM_FN_PTR(address,
206 InterpreterRuntime::build_method_counters), method);
207 ldr(mcs, Address(method, Method::method_counters_offset()));
208 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
209 bind(has_counters);
210 }
211
212 // Load object from cpool->resolved_references(index)
213 void InterpreterMacroAssembler::load_resolved_reference_at_index(
214 Register result, Register index, Register tmp) {
215 assert_different_registers(result, index);
216
217 get_constant_pool(result);
218 // load pointer for resolved_references[] objArray
219 ldr(result, Address(result, ConstantPool::cache_offset()));
220 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
221 resolve_oop_handle(result, tmp, rscratch2);
222 // Add in the index
223 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
224 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
225 }
226
227 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
228 Register cpool, Register index, Register klass, Register temp) {
229 add(temp, cpool, index, LSL, LogBytesPerWord);
230 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
231 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
232 add(klass, klass, temp, LSL, LogBytesPerWord);
233 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
234 }
235
236 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
237 // subtype of super_klass.
238 //
239 // Args:
240 // r0: superklass
241 // Rsub_klass: subklass
242 //
243 // Kills:
244 // r2, r5
245 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
246 Label& ok_is_subtype) {
247 assert(Rsub_klass != r0, "r0 holds superklass");
248 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
249 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
250
251 // Profile the not-null value's klass.
252 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
253
254 // Do the check.
255 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
256 }
257
258 // Java Expression Stack
259
260 void InterpreterMacroAssembler::pop_ptr(Register r) {
261 ldr(r, post(esp, wordSize));
262 }
263
264 void InterpreterMacroAssembler::pop_i(Register r) {
265 ldrw(r, post(esp, wordSize));
266 }
267
268 void InterpreterMacroAssembler::pop_l(Register r) {
269 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
270 }
271
272 void InterpreterMacroAssembler::push_ptr(Register r) {
273 str(r, pre(esp, -wordSize));
274 }
275
276 void InterpreterMacroAssembler::push_i(Register r) {
277 str(r, pre(esp, -wordSize));
278 }
279
280 void InterpreterMacroAssembler::push_l(Register r) {
281 str(zr, pre(esp, -wordSize));
282 str(r, pre(esp, - wordSize));
283 }
284
285 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
286 ldrs(r, post(esp, wordSize));
287 }
288
289 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
290 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
291 }
292
293 void InterpreterMacroAssembler::push_f(FloatRegister r) {
294 strs(r, pre(esp, -wordSize));
295 }
296
297 void InterpreterMacroAssembler::push_d(FloatRegister r) {
298 strd(r, pre(esp, 2* -wordSize));
299 }
300
301 void InterpreterMacroAssembler::pop(TosState state) {
302 switch (state) {
303 case atos: pop_ptr(); break;
304 case btos:
305 case ztos:
306 case ctos:
307 case stos:
308 case itos: pop_i(); break;
309 case ltos: pop_l(); break;
310 case ftos: pop_f(); break;
311 case dtos: pop_d(); break;
312 case vtos: /* nothing to do */ break;
313 default: ShouldNotReachHere();
314 }
315 interp_verify_oop(r0, state);
316 }
317
318 void InterpreterMacroAssembler::push(TosState state) {
319 interp_verify_oop(r0, state);
320 switch (state) {
321 case atos: push_ptr(); break;
322 case btos:
323 case ztos:
324 case ctos:
325 case stos:
326 case itos: push_i(); break;
327 case ltos: push_l(); break;
328 case ftos: push_f(); break;
329 case dtos: push_d(); break;
330 case vtos: /* nothing to do */ break;
331 default : ShouldNotReachHere();
332 }
333 }
334
335 // Helpers for swap and dup
336 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
337 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
338 }
339
340 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
341 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
342 }
343
344 void InterpreterMacroAssembler::load_float(Address src) {
345 ldrs(v0, src);
346 }
347
348 void InterpreterMacroAssembler::load_double(Address src) {
349 ldrd(v0, src);
350 }
351
352 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
353 // set sender sp
354 mov(r19_sender_sp, sp);
355 // record last_sp
356 sub(rscratch1, esp, rfp);
357 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
358 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
359 }
360
361 // Jump to from_interpreted entry of a call unless single stepping is possible
362 // in this thread in which case we must call the i2i entry
363 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
364 prepare_to_jump_from_interpreted();
365
366 if (JvmtiExport::can_post_interpreter_events()) {
367 Label run_compiled_code;
368 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
369 // compiled code in threads for which the event is enabled. Check here for
370 // interp_only_mode if these events CAN be enabled.
371 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
372 cbzw(rscratch1, run_compiled_code);
373 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
374 br(rscratch1);
375 bind(run_compiled_code);
376 }
377
378 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
379 br(rscratch1);
380 }
381
382 // The following two routines provide a hook so that an implementation
383 // can schedule the dispatch in two parts. amd64 does not do this.
384 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
385 }
386
387 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
388 dispatch_next(state, step);
389 }
390
391 void InterpreterMacroAssembler::dispatch_base(TosState state,
392 address* table,
393 bool verifyoop,
394 bool generate_poll) {
395 if (VerifyActivationFrameSize) {
396 Label L;
397 sub(rscratch2, rfp, esp);
398 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
399 subs(rscratch2, rscratch2, min_frame_size);
400 br(Assembler::GE, L);
401 stop("broken stack frame");
402 bind(L);
403 }
404 if (verifyoop) {
405 interp_verify_oop(r0, state);
406 }
407
408 Label safepoint;
409 address* const safepoint_table = Interpreter::safept_table(state);
410 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
411
412 if (needs_thread_local_poll) {
413 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
414 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
415 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
416 }
417
418 if (table == Interpreter::dispatch_table(state)) {
419 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
420 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
421 } else {
422 mov(rscratch2, (address)table);
423 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
424 }
425 br(rscratch2);
426
427 if (needs_thread_local_poll) {
428 bind(safepoint);
429 lea(rscratch2, ExternalAddress((address)safepoint_table));
430 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
431 br(rscratch2);
432 }
433 }
434
435 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
436 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
437 }
438
439 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
440 dispatch_base(state, Interpreter::normal_table(state));
441 }
442
443 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
444 dispatch_base(state, Interpreter::normal_table(state), false);
445 }
446
447
448 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
449 // load next bytecode
450 ldrb(rscratch1, Address(pre(rbcp, step)));
451 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
452 }
453
454 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
455 // load current bytecode
456 ldrb(rscratch1, Address(rbcp, 0));
457 dispatch_base(state, table);
458 }
459
460 // remove activation
461 //
462 // Apply stack watermark barrier.
463 // Unlock the receiver if this is a synchronized method.
464 // Unlock any Java monitors from synchronized blocks.
465 // Remove the activation from the stack.
466 //
467 // If there are locked Java monitors
468 // If throw_monitor_exception
469 // throws IllegalMonitorStateException
470 // Else if install_monitor_exception
471 // installs IllegalMonitorStateException
472 // Else
473 // no error processing
474 void InterpreterMacroAssembler::remove_activation(
475 TosState state,
476 bool throw_monitor_exception,
477 bool install_monitor_exception,
478 bool notify_jvmdi) {
479 // Note: Registers r3 xmm0 may be in use for the
480 // result check if synchronized method
481 Label unlocked, unlock, no_unlock;
482
483 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
484 // that would normally not be safe to use. Such bad returns into unsafe territory of
485 // the stack, will call InterpreterRuntime::at_unwind.
486 Label slow_path;
487 Label fast_path;
488 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
489 br(Assembler::AL, fast_path);
490 bind(slow_path);
491 push(state);
492 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
493 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
494 reset_last_Java_frame(true);
495 pop(state);
496 bind(fast_path);
497
498 // get the value of _do_not_unlock_if_synchronized into r3
499 const Address do_not_unlock_if_synchronized(rthread,
500 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
501 ldrb(r3, do_not_unlock_if_synchronized);
502 strb(zr, do_not_unlock_if_synchronized); // reset the flag
503
504 // get method access flags
505 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
506 ldrh(r2, Address(r1, Method::access_flags_offset()));
507 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
508
509 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
510 // is set.
511 cbnz(r3, no_unlock);
512
513 // unlock monitor
514 push(state); // save result
515
516 // BasicObjectLock will be first in list, since this is a
517 // synchronized method. However, need to check that the object has
518 // not been unlocked by an explicit monitorexit bytecode.
519 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
520 wordSize - (int) sizeof(BasicObjectLock));
521 // We use c_rarg1 so that if we go slow path it will be the correct
522 // register for unlock_object to pass to VM directly
523 lea(c_rarg1, monitor); // address of first monitor
524
525 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
526 cbnz(r0, unlock);
527
528 pop(state);
529 if (throw_monitor_exception) {
530 // Entry already unlocked, need to throw exception
531 call_VM(noreg, CAST_FROM_FN_PTR(address,
532 InterpreterRuntime::throw_illegal_monitor_state_exception));
533 should_not_reach_here();
534 } else {
535 // Monitor already unlocked during a stack unroll. If requested,
536 // install an illegal_monitor_state_exception. Continue with
537 // stack unrolling.
538 if (install_monitor_exception) {
539 call_VM(noreg, CAST_FROM_FN_PTR(address,
540 InterpreterRuntime::new_illegal_monitor_state_exception));
541 }
542 b(unlocked);
543 }
544
545 bind(unlock);
546 unlock_object(c_rarg1);
547 pop(state);
548
549 // Check that for block-structured locking (i.e., that all locked
550 // objects has been unlocked)
551 bind(unlocked);
552
553 // r0: Might contain return value
554
555 // Check that all monitors are unlocked
556 {
557 Label loop, exception, entry, restart;
558 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
559 const Address monitor_block_top(
560 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
561 const Address monitor_block_bot(
562 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
563
564 bind(restart);
565 // We use c_rarg1 so that if we go slow path it will be the correct
566 // register for unlock_object to pass to VM directly
567 ldr(c_rarg1, monitor_block_top); // derelativize pointer
568 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
569 // c_rarg1 points to current entry, starting with top-most entry
570
571 lea(r19, monitor_block_bot); // points to word before bottom of
572 // monitor block
573 b(entry);
574
575 // Entry already locked, need to throw exception
576 bind(exception);
577
578 if (throw_monitor_exception) {
579 // Throw exception
580 MacroAssembler::call_VM(noreg,
581 CAST_FROM_FN_PTR(address, InterpreterRuntime::
582 throw_illegal_monitor_state_exception));
583 should_not_reach_here();
584 } else {
585 // Stack unrolling. Unlock object and install illegal_monitor_exception.
586 // Unlock does not block, so don't have to worry about the frame.
587 // We don't have to preserve c_rarg1 since we are going to throw an exception.
588
589 push(state);
590 unlock_object(c_rarg1);
591 pop(state);
592
593 if (install_monitor_exception) {
594 call_VM(noreg, CAST_FROM_FN_PTR(address,
595 InterpreterRuntime::
596 new_illegal_monitor_state_exception));
597 }
598
599 b(restart);
600 }
601
602 bind(loop);
603 // check if current entry is used
604 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
605 cbnz(rscratch1, exception);
606
607 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
608 bind(entry);
609 cmp(c_rarg1, r19); // check if bottom reached
610 br(Assembler::NE, loop); // if not at bottom then check this entry
611 }
612
613 bind(no_unlock);
614
615 // jvmti support
616 if (notify_jvmdi) {
617 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
618 } else {
619 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
620 }
621
622 // remove activation
623 // get sender esp
624 ldr(rscratch2,
625 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
626 if (StackReservedPages > 0) {
627 // testing if reserved zone needs to be re-enabled
628 Label no_reserved_zone_enabling;
629
630 // check if already enabled - if so no re-enabling needed
631 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
632 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
633 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
634 br(Assembler::EQ, no_reserved_zone_enabling);
635
636 // look for an overflow into the stack reserved zone, i.e.
637 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
638 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
639 cmp(rscratch2, rscratch1);
640 br(Assembler::LS, no_reserved_zone_enabling);
641
642 call_VM_leaf(
643 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
644 call_VM(noreg, CAST_FROM_FN_PTR(address,
645 InterpreterRuntime::throw_delayed_StackOverflowError));
646 should_not_reach_here();
647
648 bind(no_reserved_zone_enabling);
649 }
650
651 // restore sender esp
652 mov(esp, rscratch2);
653 // remove frame anchor
654 leave();
655 // If we're returning to interpreted code we will shortly be
656 // adjusting SP to allow some space for ESP. If we're returning to
657 // compiled code the saved sender SP was saved in sender_sp, so this
658 // restores it.
659 andr(sp, esp, -16);
660 }
661
662 // Lock object
663 //
664 // Args:
665 // c_rarg1: BasicObjectLock to be used for locking
666 //
667 // Kills:
668 // r0
669 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
670 // rscratch1, rscratch2 (scratch regs)
671 void InterpreterMacroAssembler::lock_object(Register lock_reg)
672 {
673 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
674 if (LockingMode == LM_MONITOR) {
675 call_VM_preemptable(noreg,
676 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
677 lock_reg);
678 } else {
679 Label count, done;
680
681 const Register swap_reg = r0;
682 const Register tmp = c_rarg2;
683 const Register obj_reg = c_rarg3; // Will contain the oop
684 const Register tmp2 = c_rarg4;
685 const Register tmp3 = c_rarg5;
686
687 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
688 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
689 const int mark_offset = lock_offset +
690 BasicLock::displaced_header_offset_in_bytes();
691
692 Label slow_case;
693
694 // Load object pointer into obj_reg %c_rarg3
695 ldr(obj_reg, Address(lock_reg, obj_offset));
696
697 if (DiagnoseSyncOnValueBasedClasses != 0) {
698 load_klass(tmp, obj_reg);
699 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
700 tst(tmp, KlassFlags::_misc_is_value_based_class);
701 br(Assembler::NE, slow_case);
702 }
703
704 if (LockingMode == LM_LIGHTWEIGHT) {
705 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
706 b(done);
707 } else if (LockingMode == LM_LEGACY) {
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::verify_FPU(int stack_depth, TosState state) { ; }
1406
1407
1408 void InterpreterMacroAssembler::notify_method_entry() {
1409 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1410 // track stack depth. If it is possible to enter interp_only_mode we add
1411 // the code to check if the event should be sent.
1412 if (JvmtiExport::can_post_interpreter_events()) {
1413 Label L;
1414 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1415 cbzw(r3, L);
1416 call_VM(noreg, CAST_FROM_FN_PTR(address,
1417 InterpreterRuntime::post_method_entry));
1418 bind(L);
1419 }
1420
1421 if (DTraceMethodProbes) {
1422 get_method(c_rarg1);
1423 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1424 rthread, c_rarg1);
1425 }
1426
1427 // RedefineClasses() tracing support for obsolete method entry
1428 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1429 get_method(c_rarg1);
1430 call_VM_leaf(
1431 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1432 rthread, c_rarg1);
1433 }
1434
1435 }
1436
1437
1438 void InterpreterMacroAssembler::notify_method_exit(
1439 TosState state, NotifyMethodExitMode mode) {
1440 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1441 // track stack depth. If it is possible to enter interp_only_mode we add
1442 // the code to check if the event should be sent.
1443 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1444 Label L;
1445 // Note: frame::interpreter_frame_result has a dependency on how the
1446 // method result is saved across the call to post_method_exit. If this
1447 // is changed then the interpreter_frame_result implementation will
1448 // need to be updated too.
1449
1450 // template interpreter will leave the result on the top of the stack.
1451 push(state);
1452 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1453 cbz(r3, L);
1454 call_VM(noreg,
1455 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1456 bind(L);
1457 pop(state);
1458 }
1459
1460 if (DTraceMethodProbes) {
1461 push(state);
1462 get_method(c_rarg1);
1463 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1464 rthread, c_rarg1);
1465 pop(state);
1466 }
1467 }
1468
1469
1470 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1471 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1472 int increment, Address mask,
1473 Register scratch, Register scratch2,
1474 bool preloaded, Condition cond,
1475 Label* where) {
1476 if (!preloaded) {
1477 ldrw(scratch, counter_addr);
1478 }
1479 add(scratch, scratch, increment);
1480 strw(scratch, counter_addr);
1481 ldrw(scratch2, mask);
1482 ands(scratch, scratch, scratch2);
1483 br(cond, *where);
1484 }
1485
1486 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1487 int number_of_arguments) {
1488 // interpreter specific
1489 //
1490 // Note: No need to save/restore rbcp & rlocals pointer since these
1491 // are callee saved registers and no blocking/ GC can happen
1492 // in leaf calls.
1493 #ifdef ASSERT
1494 {
1495 Label L;
1496 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1497 cbz(rscratch1, L);
1498 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1499 " last_sp != nullptr");
1500 bind(L);
1501 }
1502 #endif /* ASSERT */
1503 // super call
1504 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1505 }
1506
1507 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1508 Register java_thread,
1509 Register last_java_sp,
1510 address entry_point,
1511 int number_of_arguments,
1512 bool check_exceptions) {
1513 // interpreter specific
1514 //
1515 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1516 // really make a difference for these runtime calls, since they are
1517 // slow anyway. Btw., bcp must be saved/restored since it may change
1518 // due to GC.
1519 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1520 save_bcp();
1521 #ifdef ASSERT
1522 {
1523 Label L;
1524 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1525 cbz(rscratch1, L);
1526 stop("InterpreterMacroAssembler::call_VM_base:"
1527 " last_sp != nullptr");
1528 bind(L);
1529 }
1530 #endif /* ASSERT */
1531 // super call
1532 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1533 entry_point, number_of_arguments,
1534 check_exceptions);
1535 // interpreter specific
1536 restore_bcp();
1537 restore_locals();
1538 }
1539
1540 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1541 address entry_point,
1542 Register arg_1) {
1543 assert(arg_1 == c_rarg1, "");
1544 Label resume_pc, not_preempted;
1545
1546 #ifdef ASSERT
1547 {
1548 Label L;
1549 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1550 cbz(rscratch1, L);
1551 stop("Should not have alternate return address set");
1552 bind(L);
1553 }
1554 #endif /* ASSERT */
1555
1556 // Force freeze slow path.
1557 push_cont_fastpath();
1558
1559 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1560 adr(rscratch1, resume_pc);
1561 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1562 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1563
1564 pop_cont_fastpath();
1565
1566 // Check if preempted.
1567 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1568 cbz(rscratch1, not_preempted);
1569 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1570 br(rscratch1);
1571
1572 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1573 bind(resume_pc);
1574 restore_after_resume(false /* is_native */);
1575
1576 bind(not_preempted);
1577 }
1578
1579 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1580 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1581 blr(rscratch1);
1582 if (is_native) {
1583 // On resume we need to set up stack as expected
1584 push(dtos);
1585 push(ltos);
1586 }
1587 }
1588
1589 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1590 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1591 Label update, next, none;
1592
1593 verify_oop(obj);
1594
1595 cbnz(obj, update);
1596 orptr(mdo_addr, TypeEntries::null_seen);
1597 b(next);
1598
1599 bind(update);
1600 load_klass(obj, obj);
1601
1602 ldr(rscratch1, mdo_addr);
1603 eor(obj, obj, rscratch1);
1604 tst(obj, TypeEntries::type_klass_mask);
1605 br(Assembler::EQ, next); // klass seen before, nothing to
1606 // do. The unknown bit may have been
1607 // set already but no need to check.
1608
1609 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1610 // already unknown. Nothing to do anymore.
1611
1612 cbz(rscratch1, none);
1613 cmp(rscratch1, (u1)TypeEntries::null_seen);
1614 br(Assembler::EQ, none);
1615 // There is a chance that the checks above
1616 // fail if another thread has just set the
1617 // profiling to this obj's klass
1618 eor(obj, obj, rscratch1); // get back original value before XOR
1619 ldr(rscratch1, mdo_addr);
1620 eor(obj, obj, rscratch1);
1621 tst(obj, TypeEntries::type_klass_mask);
1622 br(Assembler::EQ, next);
1623
1624 // different than before. Cannot keep accurate profile.
1625 orptr(mdo_addr, TypeEntries::type_unknown);
1626 b(next);
1627
1628 bind(none);
1629 // first time here. Set profile type.
1630 str(obj, mdo_addr);
1631 #ifdef ASSERT
1632 andr(obj, obj, TypeEntries::type_mask);
1633 verify_klass_ptr(obj);
1634 #endif
1635
1636 bind(next);
1637 }
1638
1639 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1640 if (!ProfileInterpreter) {
1641 return;
1642 }
1643
1644 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1645 Label profile_continue;
1646
1647 test_method_data_pointer(mdp, profile_continue);
1648
1649 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1650
1651 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1652 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1653 br(Assembler::NE, profile_continue);
1654
1655 if (MethodData::profile_arguments()) {
1656 Label done;
1657 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1658
1659 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1660 if (i > 0 || MethodData::profile_return()) {
1661 // If return value type is profiled we may have no argument to profile
1662 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1663 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1664 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1665 add(rscratch1, mdp, off_to_args);
1666 br(Assembler::LT, done);
1667 }
1668 ldr(tmp, Address(callee, Method::const_offset()));
1669 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1670 // stack offset o (zero based) from the start of the argument
1671 // list, for n arguments translates into offset n - o - 1 from
1672 // the end of the argument list
1673 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1674 sub(tmp, tmp, rscratch1);
1675 sub(tmp, tmp, 1);
1676 Address arg_addr = argument_address(tmp);
1677 ldr(tmp, arg_addr);
1678
1679 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1680 profile_obj_type(tmp, mdo_arg_addr);
1681
1682 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1683 off_to_args += to_add;
1684 }
1685
1686 if (MethodData::profile_return()) {
1687 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1688 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1689 }
1690
1691 add(rscratch1, mdp, off_to_args);
1692 bind(done);
1693 mov(mdp, rscratch1);
1694
1695 if (MethodData::profile_return()) {
1696 // We're right after the type profile for the last
1697 // argument. tmp is the number of cells left in the
1698 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1699 // if there's a return to profile.
1700 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1701 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1702 }
1703 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1704 } else {
1705 assert(MethodData::profile_return(), "either profile call args or call ret");
1706 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1707 }
1708
1709 // mdp points right after the end of the
1710 // CallTypeData/VirtualCallTypeData, right after the cells for the
1711 // return value type if there's one
1712
1713 bind(profile_continue);
1714 }
1715 }
1716
1717 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1718 assert_different_registers(mdp, ret, tmp, rbcp);
1719 if (ProfileInterpreter && MethodData::profile_return()) {
1720 Label profile_continue, done;
1721
1722 test_method_data_pointer(mdp, profile_continue);
1723
1724 if (MethodData::profile_return_jsr292_only()) {
1725 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1726
1727 // If we don't profile all invoke bytecodes we must make sure
1728 // it's a bytecode we indeed profile. We can't go back to the
1729 // beginning of the ProfileData we intend to update to check its
1730 // type because we're right after it and we don't known its
1731 // length
1732 Label do_profile;
1733 ldrb(rscratch1, Address(rbcp, 0));
1734 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1735 br(Assembler::EQ, do_profile);
1736 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1737 br(Assembler::EQ, do_profile);
1738 get_method(tmp);
1739 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1740 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1741 br(Assembler::NE, profile_continue);
1742
1743 bind(do_profile);
1744 }
1745
1746 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1747 mov(tmp, ret);
1748 profile_obj_type(tmp, mdo_ret_addr);
1749
1750 bind(profile_continue);
1751 }
1752 }
1753
1754 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1755 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1756 if (ProfileInterpreter && MethodData::profile_parameters()) {
1757 Label profile_continue, done;
1758
1759 test_method_data_pointer(mdp, profile_continue);
1760
1761 // Load the offset of the area within the MDO used for
1762 // parameters. If it's negative we're not profiling any parameters
1763 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1764 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1765
1766 // Compute a pointer to the area for parameters from the offset
1767 // and move the pointer to the slot for the last
1768 // parameters. Collect profiling from last parameter down.
1769 // mdo start + parameters offset + array length - 1
1770 add(mdp, mdp, tmp1);
1771 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1772 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1773
1774 Label loop;
1775 bind(loop);
1776
1777 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1778 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1779 int per_arg_scale = exact_log2(DataLayout::cell_size);
1780 add(rscratch1, mdp, off_base);
1781 add(rscratch2, mdp, type_base);
1782
1783 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1784 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1785
1786 // load offset on the stack from the slot for this parameter
1787 ldr(tmp2, arg_off);
1788 neg(tmp2, tmp2);
1789 // read the parameter from the local area
1790 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1791
1792 // profile the parameter
1793 profile_obj_type(tmp2, arg_type);
1794
1795 // go to next parameter
1796 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1797 br(Assembler::GE, loop);
1798
1799 bind(profile_continue);
1800 }
1801 }
1802
1803 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1804 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1805 get_cache_index_at_bcp(index, 1, sizeof(u4));
1806 // Get address of invokedynamic array
1807 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1808 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1809 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1810 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1811 lea(cache, Address(cache, index));
1812 }
1813
1814 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1815 // Get index out of bytecode pointer
1816 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1817 // Take shortcut if the size is a power of 2
1818 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1819 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1820 } else {
1821 mov(cache, sizeof(ResolvedFieldEntry));
1822 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1823 }
1824 // Get address of field entries array
1825 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1826 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1827 lea(cache, Address(cache, index));
1828 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1829 membar(MacroAssembler::LoadLoad);
1830 }
1831
1832 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1833 // Get index out of bytecode pointer
1834 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1835 mov(cache, sizeof(ResolvedMethodEntry));
1836 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1837
1838 // Get address of field entries array
1839 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1840 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1841 lea(cache, Address(cache, index));
1842 }