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/runtimeUpcalls.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 (LockingMode == LM_LIGHTWEIGHT) {
698 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
699 b(done);
700 } else if (LockingMode == LM_LEGACY) {
701
702 if (DiagnoseSyncOnValueBasedClasses != 0) {
703 load_klass(tmp, obj_reg);
704 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
705 tst(tmp, KlassFlags::_misc_is_value_based_class);
706 br(Assembler::NE, slow_case);
707 }
708
709 // Load (object->mark() | 1) into swap_reg
710 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
711 orr(swap_reg, rscratch1, 1);
712
713 // Save (object->mark() | 1) into BasicLock's displaced header
714 str(swap_reg, Address(lock_reg, mark_offset));
715
716 assert(lock_offset == 0,
717 "displached header must be first word in BasicObjectLock");
718
719 Label fail;
720 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
721
722 // Fast check for recursive lock.
723 //
724 // Can apply the optimization only if this is a stack lock
725 // allocated in this thread. For efficiency, we can focus on
726 // recently allocated stack locks (instead of reading the stack
727 // base and checking whether 'mark' points inside the current
728 // thread stack):
729 // 1) (mark & 7) == 0, and
730 // 2) sp <= mark < mark + os::pagesize()
731 //
732 // Warning: sp + os::pagesize can overflow the stack base. We must
733 // neither apply the optimization for an inflated lock allocated
734 // just above the thread stack (this is why condition 1 matters)
735 // nor apply the optimization if the stack lock is inside the stack
736 // of another thread. The latter is avoided even in case of overflow
737 // because we have guard pages at the end of all stacks. Hence, if
738 // we go over the stack base and hit the stack of another thread,
739 // this should not be in a writeable area that could contain a
740 // stack lock allocated by that thread. As a consequence, a stack
741 // lock less than page size away from sp is guaranteed to be
742 // owned by the current thread.
743 //
744 // These 3 tests can be done by evaluating the following
745 // expression: ((mark - sp) & (7 - os::vm_page_size())),
746 // assuming both stack pointer and pagesize have their
747 // least significant 3 bits clear.
748 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
749 // NOTE2: aarch64 does not like to subtract sp from rn so take a
750 // copy
751 mov(rscratch1, sp);
752 sub(swap_reg, swap_reg, rscratch1);
753 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
754
755 // Save the test result, for recursive case, the result is zero
756 str(swap_reg, Address(lock_reg, mark_offset));
757 br(Assembler::NE, slow_case);
758
759 bind(count);
760 inc_held_monitor_count(rscratch1);
761 b(done);
762 }
763 bind(slow_case);
764
765 // Call the runtime routine for slow case
766 call_VM_preemptable(noreg,
767 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
768 lock_reg);
769
770 bind(done);
771 }
772 }
773
774
775 // Unlocks an object. Used in monitorexit bytecode and
776 // remove_activation. Throws an IllegalMonitorException if object is
777 // not locked by current thread.
778 //
779 // Args:
780 // c_rarg1: BasicObjectLock for lock
781 //
782 // Kills:
783 // r0
784 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
785 // rscratch1, rscratch2 (scratch regs)
786 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
787 {
788 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
789
790 if (LockingMode == LM_MONITOR) {
791 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
792 } else {
793 Label count, done;
794
795 const Register swap_reg = r0;
796 const Register header_reg = c_rarg2; // Will contain the old oopMark
797 const Register obj_reg = c_rarg3; // Will contain the oop
798 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
799
800 save_bcp(); // Save in case of exception
801
802 if (LockingMode != LM_LIGHTWEIGHT) {
803 // Convert from BasicObjectLock structure to object and BasicLock
804 // structure Store the BasicLock address into %r0
805 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
806 }
807
808 // Load oop into obj_reg(%c_rarg3)
809 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
810
811 // Free entry
812 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
813
814 Label slow_case;
815 if (LockingMode == LM_LIGHTWEIGHT) {
816 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
817 b(done);
818 } else if (LockingMode == LM_LEGACY) {
819 // Load the old header from BasicLock structure
820 ldr(header_reg, Address(swap_reg,
821 BasicLock::displaced_header_offset_in_bytes()));
822
823 // Test for recursion
824 cbz(header_reg, count);
825
826 // Atomic swap back the old header
827 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
828
829 bind(count);
830 dec_held_monitor_count(rscratch1);
831 b(done);
832 }
833
834 bind(slow_case);
835 // Call the runtime routine for slow case.
836 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
837 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
838 bind(done);
839 restore_bcp();
840 }
841 }
842
843 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
844 Label& zero_continue) {
845 assert(ProfileInterpreter, "must be profiling interpreter");
846 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
847 cbz(mdp, zero_continue);
848 }
849
850 // Set the method data pointer for the current bcp.
851 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
852 assert(ProfileInterpreter, "must be profiling interpreter");
853 Label set_mdp;
854 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
855
856 // Test MDO to avoid the call if it is null.
857 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
858 cbz(r0, set_mdp);
859 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
860 // r0: mdi
861 // mdo is guaranteed to be non-zero here, we checked for it before the call.
862 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
863 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
864 add(r0, r1, r0);
865 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
866 bind(set_mdp);
867 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
868 }
869
870 void InterpreterMacroAssembler::verify_method_data_pointer() {
871 assert(ProfileInterpreter, "must be profiling interpreter");
872 #ifdef ASSERT
873 Label verify_continue;
874 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
875 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
876 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
877 get_method(r1);
878
879 // If the mdp is valid, it will point to a DataLayout header which is
880 // consistent with the bcp. The converse is highly probable also.
881 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
882 ldr(rscratch1, Address(r1, Method::const_offset()));
883 add(r2, r2, rscratch1, Assembler::LSL);
884 lea(r2, Address(r2, ConstMethod::codes_offset()));
885 cmp(r2, rbcp);
886 br(Assembler::EQ, verify_continue);
887 // r1: method
888 // rbcp: bcp // rbcp == 22
889 // r3: mdp
890 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
891 r1, rbcp, r3);
892 bind(verify_continue);
893 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
894 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
895 #endif // ASSERT
896 }
897
898
899 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
900 int constant,
901 Register value) {
902 assert(ProfileInterpreter, "must be profiling interpreter");
903 Address data(mdp_in, constant);
904 str(value, data);
905 }
906
907
908 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
909 int constant,
910 bool decrement) {
911 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
912 }
913
914 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
915 Register reg,
916 int constant,
917 bool decrement) {
918 assert(ProfileInterpreter, "must be profiling interpreter");
919 // %%% this does 64bit counters at best it is wasting space
920 // at worst it is a rare bug when counters overflow
921
922 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
923
924 Address addr1(mdp_in, constant);
925 Address addr2(rscratch2, reg, Address::lsl(0));
926 Address &addr = addr1;
927 if (reg != noreg) {
928 lea(rscratch2, addr1);
929 addr = addr2;
930 }
931
932 if (decrement) {
933 // Decrement the register. Set condition codes.
934 // Intel does this
935 // addptr(data, (int32_t) -DataLayout::counter_increment);
936 // If the decrement causes the counter to overflow, stay negative
937 // Label L;
938 // jcc(Assembler::negative, L);
939 // addptr(data, (int32_t) DataLayout::counter_increment);
940 // so we do this
941 ldr(rscratch1, addr);
942 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
943 Label L;
944 br(Assembler::LO, L); // skip store if counter underflow
945 str(rscratch1, addr);
946 bind(L);
947 } else {
948 assert(DataLayout::counter_increment == 1,
949 "flow-free idiom only works with 1");
950 // Intel does this
951 // Increment the register. Set carry flag.
952 // addptr(data, DataLayout::counter_increment);
953 // If the increment causes the counter to overflow, pull back by 1.
954 // sbbptr(data, (int32_t)0);
955 // so we do this
956 ldr(rscratch1, addr);
957 adds(rscratch1, rscratch1, DataLayout::counter_increment);
958 Label L;
959 br(Assembler::CS, L); // skip store if counter overflow
960 str(rscratch1, addr);
961 bind(L);
962 }
963 }
964
965 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
966 int flag_byte_constant) {
967 assert(ProfileInterpreter, "must be profiling interpreter");
968 int flags_offset = in_bytes(DataLayout::flags_offset());
969 // Set the flag
970 ldrb(rscratch1, Address(mdp_in, flags_offset));
971 orr(rscratch1, rscratch1, flag_byte_constant);
972 strb(rscratch1, Address(mdp_in, flags_offset));
973 }
974
975
976 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
977 int offset,
978 Register value,
979 Register test_value_out,
980 Label& not_equal_continue) {
981 assert(ProfileInterpreter, "must be profiling interpreter");
982 if (test_value_out == noreg) {
983 ldr(rscratch1, Address(mdp_in, offset));
984 cmp(value, rscratch1);
985 } else {
986 // Put the test value into a register, so caller can use it:
987 ldr(test_value_out, Address(mdp_in, offset));
988 cmp(value, test_value_out);
989 }
990 br(Assembler::NE, not_equal_continue);
991 }
992
993
994 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
995 int offset_of_disp) {
996 assert(ProfileInterpreter, "must be profiling interpreter");
997 ldr(rscratch1, Address(mdp_in, offset_of_disp));
998 add(mdp_in, mdp_in, rscratch1, LSL);
999 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1000 }
1001
1002
1003 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1004 Register reg,
1005 int offset_of_disp) {
1006 assert(ProfileInterpreter, "must be profiling interpreter");
1007 lea(rscratch1, Address(mdp_in, offset_of_disp));
1008 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1009 add(mdp_in, mdp_in, rscratch1, LSL);
1010 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1011 }
1012
1013
1014 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1015 int constant) {
1016 assert(ProfileInterpreter, "must be profiling interpreter");
1017 add(mdp_in, mdp_in, (unsigned)constant);
1018 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1019 }
1020
1021
1022 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1023 assert(ProfileInterpreter, "must be profiling interpreter");
1024 // save/restore across call_VM
1025 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1026 call_VM(noreg,
1027 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1028 return_bci);
1029 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1030 }
1031
1032
1033 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1034 Register bumped_count) {
1035 if (ProfileInterpreter) {
1036 Label profile_continue;
1037
1038 // If no method data exists, go to profile_continue.
1039 // Otherwise, assign to mdp
1040 test_method_data_pointer(mdp, profile_continue);
1041
1042 // We are taking a branch. Increment the taken count.
1043 // We inline increment_mdp_data_at to return bumped_count in a register
1044 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1045 Address data(mdp, in_bytes(JumpData::taken_offset()));
1046 ldr(bumped_count, data);
1047 assert(DataLayout::counter_increment == 1,
1048 "flow-free idiom only works with 1");
1049 // Intel does this to catch overflow
1050 // addptr(bumped_count, DataLayout::counter_increment);
1051 // sbbptr(bumped_count, 0);
1052 // so we do this
1053 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1054 Label L;
1055 br(Assembler::CS, L); // skip store if counter overflow
1056 str(bumped_count, data);
1057 bind(L);
1058 // The method data pointer needs to be updated to reflect the new target.
1059 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1060 bind(profile_continue);
1061 }
1062 }
1063
1064
1065 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1066 if (ProfileInterpreter) {
1067 Label profile_continue;
1068
1069 // If no method data exists, go to profile_continue.
1070 test_method_data_pointer(mdp, profile_continue);
1071
1072 // We are taking a branch. Increment the not taken count.
1073 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1074
1075 // The method data pointer needs to be updated to correspond to
1076 // the next bytecode
1077 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1078 bind(profile_continue);
1079 }
1080 }
1081
1082
1083 void InterpreterMacroAssembler::profile_call(Register mdp) {
1084 if (ProfileInterpreter) {
1085 Label profile_continue;
1086
1087 // If no method data exists, go to profile_continue.
1088 test_method_data_pointer(mdp, profile_continue);
1089
1090 // We are making a call. Increment the count.
1091 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1092
1093 // The method data pointer needs to be updated to reflect the new target.
1094 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1095 bind(profile_continue);
1096 }
1097 }
1098
1099 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1100 if (ProfileInterpreter) {
1101 Label profile_continue;
1102
1103 // If no method data exists, go to profile_continue.
1104 test_method_data_pointer(mdp, profile_continue);
1105
1106 // We are making a call. Increment the count.
1107 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1108
1109 // The method data pointer needs to be updated to reflect the new target.
1110 update_mdp_by_constant(mdp,
1111 in_bytes(VirtualCallData::
1112 virtual_call_data_size()));
1113 bind(profile_continue);
1114 }
1115 }
1116
1117
1118 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1119 Register mdp,
1120 Register reg2,
1121 bool receiver_can_be_null) {
1122 if (ProfileInterpreter) {
1123 Label profile_continue;
1124
1125 // If no method data exists, go to profile_continue.
1126 test_method_data_pointer(mdp, profile_continue);
1127
1128 Label skip_receiver_profile;
1129 if (receiver_can_be_null) {
1130 Label not_null;
1131 // We are making a call. Increment the count for null receiver.
1132 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1133 b(skip_receiver_profile);
1134 bind(not_null);
1135 }
1136
1137 // Record the receiver type.
1138 record_klass_in_profile(receiver, mdp, reg2);
1139 bind(skip_receiver_profile);
1140
1141 // The method data pointer needs to be updated to reflect the new target.
1142 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1143 bind(profile_continue);
1144 }
1145 }
1146
1147 // This routine creates a state machine for updating the multi-row
1148 // type profile at a virtual call site (or other type-sensitive bytecode).
1149 // The machine visits each row (of receiver/count) until the receiver type
1150 // is found, or until it runs out of rows. At the same time, it remembers
1151 // the location of the first empty row. (An empty row records null for its
1152 // receiver, and can be allocated for a newly-observed receiver type.)
1153 // Because there are two degrees of freedom in the state, a simple linear
1154 // search will not work; it must be a decision tree. Hence this helper
1155 // function is recursive, to generate the required tree structured code.
1156 // It's the interpreter, so we are trading off code space for speed.
1157 // See below for example code.
1158 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1159 Register receiver, Register mdp,
1160 Register reg2, int start_row,
1161 Label& done) {
1162 if (TypeProfileWidth == 0) {
1163 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1164 } else {
1165 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1166 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1167 }
1168 }
1169
1170 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1171 Register reg2, int start_row, Label& done, int total_rows,
1172 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1173 int last_row = total_rows - 1;
1174 assert(start_row <= last_row, "must be work left to do");
1175 // Test this row for both the item and for null.
1176 // Take any of three different outcomes:
1177 // 1. found item => increment count and goto done
1178 // 2. found null => keep looking for case 1, maybe allocate this cell
1179 // 3. found something else => keep looking for cases 1 and 2
1180 // Case 3 is handled by a recursive call.
1181 for (int row = start_row; row <= last_row; row++) {
1182 Label next_test;
1183 bool test_for_null_also = (row == start_row);
1184
1185 // See if the item is item[n].
1186 int item_offset = in_bytes(item_offset_fn(row));
1187 test_mdp_data_at(mdp, item_offset, item,
1188 (test_for_null_also ? reg2 : noreg),
1189 next_test);
1190 // (Reg2 now contains the item from the CallData.)
1191
1192 // The item is item[n]. Increment count[n].
1193 int count_offset = in_bytes(item_count_offset_fn(row));
1194 increment_mdp_data_at(mdp, count_offset);
1195 b(done);
1196 bind(next_test);
1197
1198 if (test_for_null_also) {
1199 Label found_null;
1200 // Failed the equality check on item[n]... Test for null.
1201 if (start_row == last_row) {
1202 // The only thing left to do is handle the null case.
1203 cbz(reg2, found_null);
1204 // Item did not match any saved item and there is no empty row for it.
1205 // Increment total counter to indicate polymorphic case.
1206 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1207 b(done);
1208 bind(found_null);
1209 break;
1210 }
1211 // Since null is rare, make it be the branch-taken case.
1212 cbz(reg2, found_null);
1213
1214 // Put all the "Case 3" tests here.
1215 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1216 item_offset_fn, item_count_offset_fn);
1217
1218 // Found a null. Keep searching for a matching item,
1219 // but remember that this is an empty (unused) slot.
1220 bind(found_null);
1221 }
1222 }
1223
1224 // In the fall-through case, we found no matching item, but we
1225 // observed the item[start_row] is null.
1226
1227 // Fill in the item field and increment the count.
1228 int item_offset = in_bytes(item_offset_fn(start_row));
1229 set_mdp_data_at(mdp, item_offset, item);
1230 int count_offset = in_bytes(item_count_offset_fn(start_row));
1231 mov(reg2, DataLayout::counter_increment);
1232 set_mdp_data_at(mdp, count_offset, reg2);
1233 if (start_row > 0) {
1234 b(done);
1235 }
1236 }
1237
1238 // Example state machine code for three profile rows:
1239 // // main copy of decision tree, rooted at row[1]
1240 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1241 // if (row[0].rec != nullptr) {
1242 // // inner copy of decision tree, rooted at row[1]
1243 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1244 // if (row[1].rec != nullptr) {
1245 // // degenerate decision tree, rooted at row[2]
1246 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1247 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1248 // row[2].init(rec); goto done;
1249 // } else {
1250 // // remember row[1] is empty
1251 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1252 // row[1].init(rec); goto done;
1253 // }
1254 // } else {
1255 // // remember row[0] is empty
1256 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1257 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1258 // row[0].init(rec); goto done;
1259 // }
1260 // done:
1261
1262 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1263 Register mdp, Register reg2) {
1264 assert(ProfileInterpreter, "must be profiling");
1265 Label done;
1266
1267 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1268
1269 bind (done);
1270 }
1271
1272 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1273 Register mdp) {
1274 if (ProfileInterpreter) {
1275 Label profile_continue;
1276 uint row;
1277
1278 // If no method data exists, go to profile_continue.
1279 test_method_data_pointer(mdp, profile_continue);
1280
1281 // Update the total ret count.
1282 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1283
1284 for (row = 0; row < RetData::row_limit(); row++) {
1285 Label next_test;
1286
1287 // See if return_bci is equal to bci[n]:
1288 test_mdp_data_at(mdp,
1289 in_bytes(RetData::bci_offset(row)),
1290 return_bci, noreg,
1291 next_test);
1292
1293 // return_bci is equal to bci[n]. Increment the count.
1294 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1295
1296 // The method data pointer needs to be updated to reflect the new target.
1297 update_mdp_by_offset(mdp,
1298 in_bytes(RetData::bci_displacement_offset(row)));
1299 b(profile_continue);
1300 bind(next_test);
1301 }
1302
1303 update_mdp_for_ret(return_bci);
1304
1305 bind(profile_continue);
1306 }
1307 }
1308
1309 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1310 if (ProfileInterpreter) {
1311 Label profile_continue;
1312
1313 // If no method data exists, go to profile_continue.
1314 test_method_data_pointer(mdp, profile_continue);
1315
1316 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1317
1318 // The method data pointer needs to be updated.
1319 int mdp_delta = in_bytes(BitData::bit_data_size());
1320 if (TypeProfileCasts) {
1321 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1322 }
1323 update_mdp_by_constant(mdp, mdp_delta);
1324
1325 bind(profile_continue);
1326 }
1327 }
1328
1329 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1330 if (ProfileInterpreter) {
1331 Label profile_continue;
1332
1333 // If no method data exists, go to profile_continue.
1334 test_method_data_pointer(mdp, profile_continue);
1335
1336 // The method data pointer needs to be updated.
1337 int mdp_delta = in_bytes(BitData::bit_data_size());
1338 if (TypeProfileCasts) {
1339 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1340
1341 // Record the object type.
1342 record_klass_in_profile(klass, mdp, reg2);
1343 }
1344 update_mdp_by_constant(mdp, mdp_delta);
1345
1346 bind(profile_continue);
1347 }
1348 }
1349
1350 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1351 if (ProfileInterpreter) {
1352 Label profile_continue;
1353
1354 // If no method data exists, go to profile_continue.
1355 test_method_data_pointer(mdp, profile_continue);
1356
1357 // Update the default case count
1358 increment_mdp_data_at(mdp,
1359 in_bytes(MultiBranchData::default_count_offset()));
1360
1361 // The method data pointer needs to be updated.
1362 update_mdp_by_offset(mdp,
1363 in_bytes(MultiBranchData::
1364 default_displacement_offset()));
1365
1366 bind(profile_continue);
1367 }
1368 }
1369
1370 void InterpreterMacroAssembler::profile_switch_case(Register index,
1371 Register mdp,
1372 Register reg2) {
1373 if (ProfileInterpreter) {
1374 Label profile_continue;
1375
1376 // If no method data exists, go to profile_continue.
1377 test_method_data_pointer(mdp, profile_continue);
1378
1379 // Build the base (index * per_case_size_in_bytes()) +
1380 // case_array_offset_in_bytes()
1381 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1382 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1383 Assembler::maddw(index, index, reg2, rscratch1);
1384
1385 // Update the case count
1386 increment_mdp_data_at(mdp,
1387 index,
1388 in_bytes(MultiBranchData::relative_count_offset()));
1389
1390 // The method data pointer needs to be updated.
1391 update_mdp_by_offset(mdp,
1392 index,
1393 in_bytes(MultiBranchData::
1394 relative_displacement_offset()));
1395
1396 bind(profile_continue);
1397 }
1398 }
1399
1400 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1401 if (state == atos) {
1402 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1403 }
1404 }
1405
1406 void InterpreterMacroAssembler::generate_runtime_upcalls_on_method_entry()
1407 {
1408 address upcall = RuntimeUpcalls::on_method_entry_upcall_address();
1409 if (RuntimeUpcalls::does_upcall_need_method_parameter(upcall)) {
1410 get_method(c_rarg1);
1411 call_VM(noreg,upcall, c_rarg1);
1412 } else {
1413 call_VM(noreg,upcall);
1414 }
1415 }
1416
1417 void InterpreterMacroAssembler::notify_method_entry() {
1418 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1419 // track stack depth. If it is possible to enter interp_only_mode we add
1420 // the code to check if the event should be sent.
1421 if (JvmtiExport::can_post_interpreter_events()) {
1422 Label L;
1423 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1424 cbzw(r3, L);
1425 call_VM(noreg, CAST_FROM_FN_PTR(address,
1426 InterpreterRuntime::post_method_entry));
1427 bind(L);
1428 }
1429
1430 if (DTraceMethodProbes) {
1431 get_method(c_rarg1);
1432 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1433 rthread, c_rarg1);
1434 }
1435
1436 // RedefineClasses() tracing support for obsolete method entry
1437 if (log_is_enabled(Trace, redefine, class, obsolete) ||
1438 log_is_enabled(Trace, interpreter, bytecode)) {
1439 get_method(c_rarg1);
1440 call_VM_leaf(
1441 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1442 rthread, c_rarg1);
1443 }
1444
1445 }
1446
1447
1448 void InterpreterMacroAssembler::notify_method_exit(
1449 TosState state, NotifyMethodExitMode mode) {
1450 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1451 // track stack depth. If it is possible to enter interp_only_mode we add
1452 // the code to check if the event should be sent.
1453 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1454 Label L;
1455 // Note: frame::interpreter_frame_result has a dependency on how the
1456 // method result is saved across the call to post_method_exit. If this
1457 // is changed then the interpreter_frame_result implementation will
1458 // need to be updated too.
1459
1460 // template interpreter will leave the result on the top of the stack.
1461 push(state);
1462 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1463 cbz(r3, L);
1464 call_VM(noreg,
1465 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1466 bind(L);
1467 pop(state);
1468 }
1469
1470 if (DTraceMethodProbes) {
1471 push(state);
1472 get_method(c_rarg1);
1473 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1474 rthread, c_rarg1);
1475 pop(state);
1476 }
1477 }
1478
1479
1480 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1481 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1482 int increment, Address mask,
1483 Register scratch, Register scratch2,
1484 bool preloaded, Condition cond,
1485 Label* where) {
1486 if (!preloaded) {
1487 ldrw(scratch, counter_addr);
1488 }
1489 add(scratch, scratch, increment);
1490 strw(scratch, counter_addr);
1491 ldrw(scratch2, mask);
1492 ands(scratch, scratch, scratch2);
1493 br(cond, *where);
1494 }
1495
1496 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1497 int number_of_arguments) {
1498 // interpreter specific
1499 //
1500 // Note: No need to save/restore rbcp & rlocals pointer since these
1501 // are callee saved registers and no blocking/ GC can happen
1502 // in leaf calls.
1503 #ifdef ASSERT
1504 {
1505 Label L;
1506 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1507 cbz(rscratch1, L);
1508 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1509 " last_sp != nullptr");
1510 bind(L);
1511 }
1512 #endif /* ASSERT */
1513 // super call
1514 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1515 }
1516
1517 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1518 Register java_thread,
1519 Register last_java_sp,
1520 address entry_point,
1521 int number_of_arguments,
1522 bool check_exceptions) {
1523 // interpreter specific
1524 //
1525 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1526 // really make a difference for these runtime calls, since they are
1527 // slow anyway. Btw., bcp must be saved/restored since it may change
1528 // due to GC.
1529 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1530 save_bcp();
1531 #ifdef ASSERT
1532 {
1533 Label L;
1534 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1535 cbz(rscratch1, L);
1536 stop("InterpreterMacroAssembler::call_VM_base:"
1537 " last_sp != nullptr");
1538 bind(L);
1539 }
1540 #endif /* ASSERT */
1541 // super call
1542 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1543 entry_point, number_of_arguments,
1544 check_exceptions);
1545 // interpreter specific
1546 restore_bcp();
1547 restore_locals();
1548 }
1549
1550 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1551 address entry_point,
1552 Register arg_1) {
1553 assert(arg_1 == c_rarg1, "");
1554 Label resume_pc, not_preempted;
1555
1556 #ifdef ASSERT
1557 {
1558 Label L;
1559 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1560 cbz(rscratch1, L);
1561 stop("Should not have alternate return address set");
1562 bind(L);
1563 }
1564 #endif /* ASSERT */
1565
1566 // Force freeze slow path.
1567 push_cont_fastpath();
1568
1569 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1570 adr(rscratch1, resume_pc);
1571 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1572 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1573
1574 pop_cont_fastpath();
1575
1576 // Check if preempted.
1577 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1578 cbz(rscratch1, not_preempted);
1579 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1580 br(rscratch1);
1581
1582 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1583 bind(resume_pc);
1584 restore_after_resume(false /* is_native */);
1585
1586 bind(not_preempted);
1587 }
1588
1589 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1590 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1591 blr(rscratch1);
1592 if (is_native) {
1593 // On resume we need to set up stack as expected
1594 push(dtos);
1595 push(ltos);
1596 }
1597 }
1598
1599 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1600 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1601 Label update, next, none;
1602
1603 verify_oop(obj);
1604
1605 cbnz(obj, update);
1606 orptr(mdo_addr, TypeEntries::null_seen);
1607 b(next);
1608
1609 bind(update);
1610 load_klass(obj, obj);
1611
1612 ldr(rscratch1, mdo_addr);
1613 eor(obj, obj, rscratch1);
1614 tst(obj, TypeEntries::type_klass_mask);
1615 br(Assembler::EQ, next); // klass seen before, nothing to
1616 // do. The unknown bit may have been
1617 // set already but no need to check.
1618
1619 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1620 // already unknown. Nothing to do anymore.
1621
1622 cbz(rscratch1, none);
1623 cmp(rscratch1, (u1)TypeEntries::null_seen);
1624 br(Assembler::EQ, none);
1625 // There is a chance that the checks above
1626 // fail if another thread has just set the
1627 // profiling to this obj's klass
1628 eor(obj, obj, rscratch1); // get back original value before XOR
1629 ldr(rscratch1, mdo_addr);
1630 eor(obj, obj, rscratch1);
1631 tst(obj, TypeEntries::type_klass_mask);
1632 br(Assembler::EQ, next);
1633
1634 // different than before. Cannot keep accurate profile.
1635 orptr(mdo_addr, TypeEntries::type_unknown);
1636 b(next);
1637
1638 bind(none);
1639 // first time here. Set profile type.
1640 str(obj, mdo_addr);
1641 #ifdef ASSERT
1642 andr(obj, obj, TypeEntries::type_mask);
1643 verify_klass_ptr(obj);
1644 #endif
1645
1646 bind(next);
1647 }
1648
1649 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1650 if (!ProfileInterpreter) {
1651 return;
1652 }
1653
1654 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1655 Label profile_continue;
1656
1657 test_method_data_pointer(mdp, profile_continue);
1658
1659 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1660
1661 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1662 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1663 br(Assembler::NE, profile_continue);
1664
1665 if (MethodData::profile_arguments()) {
1666 Label done;
1667 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1668
1669 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1670 if (i > 0 || MethodData::profile_return()) {
1671 // If return value type is profiled we may have no argument to profile
1672 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1673 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1674 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1675 add(rscratch1, mdp, off_to_args);
1676 br(Assembler::LT, done);
1677 }
1678 ldr(tmp, Address(callee, Method::const_offset()));
1679 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1680 // stack offset o (zero based) from the start of the argument
1681 // list, for n arguments translates into offset n - o - 1 from
1682 // the end of the argument list
1683 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1684 sub(tmp, tmp, rscratch1);
1685 sub(tmp, tmp, 1);
1686 Address arg_addr = argument_address(tmp);
1687 ldr(tmp, arg_addr);
1688
1689 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1690 profile_obj_type(tmp, mdo_arg_addr);
1691
1692 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1693 off_to_args += to_add;
1694 }
1695
1696 if (MethodData::profile_return()) {
1697 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1698 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1699 }
1700
1701 add(rscratch1, mdp, off_to_args);
1702 bind(done);
1703 mov(mdp, rscratch1);
1704
1705 if (MethodData::profile_return()) {
1706 // We're right after the type profile for the last
1707 // argument. tmp is the number of cells left in the
1708 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1709 // if there's a return to profile.
1710 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1711 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1712 }
1713 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1714 } else {
1715 assert(MethodData::profile_return(), "either profile call args or call ret");
1716 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1717 }
1718
1719 // mdp points right after the end of the
1720 // CallTypeData/VirtualCallTypeData, right after the cells for the
1721 // return value type if there's one
1722
1723 bind(profile_continue);
1724 }
1725 }
1726
1727 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1728 assert_different_registers(mdp, ret, tmp, rbcp);
1729 if (ProfileInterpreter && MethodData::profile_return()) {
1730 Label profile_continue, done;
1731
1732 test_method_data_pointer(mdp, profile_continue);
1733
1734 if (MethodData::profile_return_jsr292_only()) {
1735 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1736
1737 // If we don't profile all invoke bytecodes we must make sure
1738 // it's a bytecode we indeed profile. We can't go back to the
1739 // beginning of the ProfileData we intend to update to check its
1740 // type because we're right after it and we don't known its
1741 // length
1742 Label do_profile;
1743 ldrb(rscratch1, Address(rbcp, 0));
1744 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1745 br(Assembler::EQ, do_profile);
1746 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1747 br(Assembler::EQ, do_profile);
1748 get_method(tmp);
1749 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1750 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1751 br(Assembler::NE, profile_continue);
1752
1753 bind(do_profile);
1754 }
1755
1756 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1757 mov(tmp, ret);
1758 profile_obj_type(tmp, mdo_ret_addr);
1759
1760 bind(profile_continue);
1761 }
1762 }
1763
1764 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1765 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1766 if (ProfileInterpreter && MethodData::profile_parameters()) {
1767 Label profile_continue, done;
1768
1769 test_method_data_pointer(mdp, profile_continue);
1770
1771 // Load the offset of the area within the MDO used for
1772 // parameters. If it's negative we're not profiling any parameters
1773 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1774 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1775
1776 // Compute a pointer to the area for parameters from the offset
1777 // and move the pointer to the slot for the last
1778 // parameters. Collect profiling from last parameter down.
1779 // mdo start + parameters offset + array length - 1
1780 add(mdp, mdp, tmp1);
1781 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1782 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1783
1784 Label loop;
1785 bind(loop);
1786
1787 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1788 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1789 int per_arg_scale = exact_log2(DataLayout::cell_size);
1790 add(rscratch1, mdp, off_base);
1791 add(rscratch2, mdp, type_base);
1792
1793 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1794 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1795
1796 // load offset on the stack from the slot for this parameter
1797 ldr(tmp2, arg_off);
1798 neg(tmp2, tmp2);
1799 // read the parameter from the local area
1800 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1801
1802 // profile the parameter
1803 profile_obj_type(tmp2, arg_type);
1804
1805 // go to next parameter
1806 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1807 br(Assembler::GE, loop);
1808
1809 bind(profile_continue);
1810 }
1811 }
1812
1813 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1814 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1815 get_cache_index_at_bcp(index, 1, sizeof(u4));
1816 // Get address of invokedynamic array
1817 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1818 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1819 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1820 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1821 lea(cache, Address(cache, index));
1822 }
1823
1824 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1825 // Get index out of bytecode pointer
1826 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1827 // Take shortcut if the size is a power of 2
1828 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1829 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1830 } else {
1831 mov(cache, sizeof(ResolvedFieldEntry));
1832 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1833 }
1834 // Get address of field entries array
1835 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1836 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1837 lea(cache, Address(cache, index));
1838 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1839 membar(MacroAssembler::LoadLoad);
1840 }
1841
1842 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1843 // Get index out of bytecode pointer
1844 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1845 mov(cache, sizeof(ResolvedMethodEntry));
1846 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1847
1848 // Get address of field entries array
1849 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1850 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1851 lea(cache, Address(cache, index));
1852 }