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/runtimeUpcalls.hpp"
48 #include "runtime/safepointMechanism.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "utilities/powerOfTwo.hpp"
51
52 void InterpreterMacroAssembler::narrow(Register result) {
53
54 // Get method->_constMethod->_result_type
55 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
56 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
57 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
58
59 Label done, notBool, notByte, notChar;
60
61 // common case first
62 cmpw(rscratch1, T_INT);
63 br(Assembler::EQ, done);
64
65 // mask integer result to narrower return type.
66 cmpw(rscratch1, T_BOOLEAN);
67 br(Assembler::NE, notBool);
68 andw(result, result, 0x1);
69 b(done);
70
71 bind(notBool);
72 cmpw(rscratch1, T_BYTE);
73 br(Assembler::NE, notByte);
74 sbfx(result, result, 0, 8);
75 b(done);
76
77 bind(notByte);
78 cmpw(rscratch1, T_CHAR);
79 br(Assembler::NE, notChar);
80 ubfx(result, result, 0, 16); // truncate upper 16 bits
81 b(done);
82
83 bind(notChar);
84 sbfx(result, result, 0, 16); // sign-extend short
85
86 // Nothing to do for T_INT
87 bind(done);
88 }
89
90 void InterpreterMacroAssembler::jump_to_entry(address entry) {
91 assert(entry, "Entry must have been generated by now");
92 b(entry);
93 }
94
95 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
96 if (JvmtiExport::can_pop_frame()) {
97 Label L;
98 // Initiate popframe handling only if it is not already being
99 // processed. If the flag has the popframe_processing bit set, it
100 // means that this code is called *during* popframe handling - we
101 // don't want to reenter.
102 // This method is only called just after the call into the vm in
103 // call_VM_base, so the arg registers are available.
104 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
105 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
106 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
107 // Call Interpreter::remove_activation_preserving_args_entry() to get the
108 // address of the same-named entrypoint in the generated interpreter code.
109 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
110 br(r0);
111 bind(L);
112 }
113 }
114
115
116 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
117 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
118 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
119 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
120 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
121 switch (state) {
122 case atos: ldr(r0, oop_addr);
123 str(zr, oop_addr);
124 interp_verify_oop(r0, state); break;
125 case ltos: ldr(r0, val_addr); break;
126 case btos: // fall through
127 case ztos: // fall through
128 case ctos: // fall through
129 case stos: // fall through
130 case itos: ldrw(r0, val_addr); break;
131 case ftos: ldrs(v0, val_addr); break;
132 case dtos: ldrd(v0, val_addr); break;
133 case vtos: /* nothing to do */ break;
134 default : ShouldNotReachHere();
135 }
136 // Clean up tos value in the thread object
137 movw(rscratch1, (int) ilgl);
138 strw(rscratch1, tos_addr);
139 strw(zr, val_addr);
140 }
141
142
143 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
144 if (JvmtiExport::can_force_early_return()) {
145 Label L;
146 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
147 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
148
149 // Initiate earlyret handling only if it is not already being processed.
150 // If the flag has the earlyret_processing bit set, it means that this code
151 // is called *during* earlyret handling - we don't want to reenter.
152 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
153 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
154 br(Assembler::NE, L);
155
156 // Call Interpreter::remove_activation_early_entry() to get the address of the
157 // same-named entrypoint in the generated interpreter code.
158 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
159 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
160 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
161 br(r0);
162 bind(L);
163 }
164 }
165
166 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
167 Register reg,
168 int bcp_offset) {
169 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
170 ldrh(reg, Address(rbcp, bcp_offset));
171 rev16(reg, reg);
172 }
173
174 void InterpreterMacroAssembler::get_dispatch() {
175 uint64_t offset;
176 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
177 // Use add() here after ARDP, rather than lea().
178 // lea() does not generate anything if its offset is zero.
179 // However, relocs expect to find either an ADD or a load/store
180 // insn after an ADRP. add() always generates an ADD insn, even
181 // for add(Rn, Rn, 0).
182 add(rdispatch, rdispatch, offset);
183 }
184
185 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
186 int bcp_offset,
187 size_t index_size) {
188 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
189 if (index_size == sizeof(u2)) {
190 load_unsigned_short(index, Address(rbcp, bcp_offset));
191 } else if (index_size == sizeof(u4)) {
192 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
193 ldrw(index, Address(rbcp, bcp_offset));
194 } else if (index_size == sizeof(u1)) {
195 load_unsigned_byte(index, Address(rbcp, bcp_offset));
196 } else {
197 ShouldNotReachHere();
198 }
199 }
200
201 void InterpreterMacroAssembler::get_method_counters(Register method,
202 Register mcs, Label& skip) {
203 Label has_counters;
204 ldr(mcs, Address(method, Method::method_counters_offset()));
205 cbnz(mcs, has_counters);
206 call_VM(noreg, CAST_FROM_FN_PTR(address,
207 InterpreterRuntime::build_method_counters), method);
208 ldr(mcs, Address(method, Method::method_counters_offset()));
209 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
210 bind(has_counters);
211 }
212
213 // Load object from cpool->resolved_references(index)
214 void InterpreterMacroAssembler::load_resolved_reference_at_index(
215 Register result, Register index, Register tmp) {
216 assert_different_registers(result, index);
217
218 get_constant_pool(result);
219 // load pointer for resolved_references[] objArray
220 ldr(result, Address(result, ConstantPool::cache_offset()));
221 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
222 resolve_oop_handle(result, tmp, rscratch2);
223 // Add in the index
224 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
225 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
226 }
227
228 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
229 Register cpool, Register index, Register klass, Register temp) {
230 add(temp, cpool, index, LSL, LogBytesPerWord);
231 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
232 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
233 add(klass, klass, temp, LSL, LogBytesPerWord);
234 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
235 }
236
237 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
238 // subtype of super_klass.
239 //
240 // Args:
241 // r0: superklass
242 // Rsub_klass: subklass
243 //
244 // Kills:
245 // r2, r5
246 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
247 Label& ok_is_subtype) {
248 assert(Rsub_klass != r0, "r0 holds superklass");
249 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
250 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
251
252 // Profile the not-null value's klass.
253 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
254
255 // Do the check.
256 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
257 }
258
259 // Java Expression Stack
260
261 void InterpreterMacroAssembler::pop_ptr(Register r) {
262 ldr(r, post(esp, wordSize));
263 }
264
265 void InterpreterMacroAssembler::pop_i(Register r) {
266 ldrw(r, post(esp, wordSize));
267 }
268
269 void InterpreterMacroAssembler::pop_l(Register r) {
270 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
271 }
272
273 void InterpreterMacroAssembler::push_ptr(Register r) {
274 str(r, pre(esp, -wordSize));
275 }
276
277 void InterpreterMacroAssembler::push_i(Register r) {
278 str(r, pre(esp, -wordSize));
279 }
280
281 void InterpreterMacroAssembler::push_l(Register r) {
282 str(zr, pre(esp, -wordSize));
283 str(r, pre(esp, - wordSize));
284 }
285
286 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
287 ldrs(r, post(esp, wordSize));
288 }
289
290 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
291 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
292 }
293
294 void InterpreterMacroAssembler::push_f(FloatRegister r) {
295 strs(r, pre(esp, -wordSize));
296 }
297
298 void InterpreterMacroAssembler::push_d(FloatRegister r) {
299 strd(r, pre(esp, 2* -wordSize));
300 }
301
302 void InterpreterMacroAssembler::pop(TosState state) {
303 switch (state) {
304 case atos: pop_ptr(); break;
305 case btos:
306 case ztos:
307 case ctos:
308 case stos:
309 case itos: pop_i(); break;
310 case ltos: pop_l(); break;
311 case ftos: pop_f(); break;
312 case dtos: pop_d(); break;
313 case vtos: /* nothing to do */ break;
314 default: ShouldNotReachHere();
315 }
316 interp_verify_oop(r0, state);
317 }
318
319 void InterpreterMacroAssembler::push(TosState state) {
320 interp_verify_oop(r0, state);
321 switch (state) {
322 case atos: push_ptr(); break;
323 case btos:
324 case ztos:
325 case ctos:
326 case stos:
327 case itos: push_i(); break;
328 case ltos: push_l(); break;
329 case ftos: push_f(); break;
330 case dtos: push_d(); break;
331 case vtos: /* nothing to do */ break;
332 default : ShouldNotReachHere();
333 }
334 }
335
336 // Helpers for swap and dup
337 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
338 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
339 }
340
341 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
342 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
343 }
344
345 void InterpreterMacroAssembler::load_float(Address src) {
346 ldrs(v0, src);
347 }
348
349 void InterpreterMacroAssembler::load_double(Address src) {
350 ldrd(v0, src);
351 }
352
353 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
354 // set sender sp
355 mov(r19_sender_sp, sp);
356 // record last_sp
357 sub(rscratch1, esp, rfp);
358 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
359 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
360 }
361
362 // Jump to from_interpreted entry of a call unless single stepping is possible
363 // in this thread in which case we must call the i2i entry
364 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
365 prepare_to_jump_from_interpreted();
366
367 if (JvmtiExport::can_post_interpreter_events()) {
368 Label run_compiled_code;
369 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
370 // compiled code in threads for which the event is enabled. Check here for
371 // interp_only_mode if these events CAN be enabled.
372 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
373 cbzw(rscratch1, run_compiled_code);
374 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
375 br(rscratch1);
376 bind(run_compiled_code);
377 }
378
379 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
380 br(rscratch1);
381 }
382
383 // The following two routines provide a hook so that an implementation
384 // can schedule the dispatch in two parts. amd64 does not do this.
385 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
386 }
387
388 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
389 dispatch_next(state, step);
390 }
391
392 void InterpreterMacroAssembler::dispatch_base(TosState state,
393 address* table,
394 bool verifyoop,
395 bool generate_poll) {
396 if (VerifyActivationFrameSize) {
397 Label L;
398 sub(rscratch2, rfp, esp);
399 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
400 subs(rscratch2, rscratch2, min_frame_size);
401 br(Assembler::GE, L);
402 stop("broken stack frame");
403 bind(L);
404 }
405 if (verifyoop) {
406 interp_verify_oop(r0, state);
407 }
408
409 Label safepoint;
410 address* const safepoint_table = Interpreter::safept_table(state);
411 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
412
413 if (needs_thread_local_poll) {
414 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
415 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
416 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
417 }
418
419 if (table == Interpreter::dispatch_table(state)) {
420 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
421 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
422 } else {
423 mov(rscratch2, (address)table);
424 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
425 }
426 br(rscratch2);
427
428 if (needs_thread_local_poll) {
429 bind(safepoint);
430 lea(rscratch2, ExternalAddress((address)safepoint_table));
431 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
432 br(rscratch2);
433 }
434 }
435
436 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
437 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
438 }
439
440 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
441 dispatch_base(state, Interpreter::normal_table(state));
442 }
443
444 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
445 dispatch_base(state, Interpreter::normal_table(state), false);
446 }
447
448
449 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
450 // load next bytecode
451 ldrb(rscratch1, Address(pre(rbcp, step)));
452 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
453 }
454
455 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
456 // load current bytecode
457 ldrb(rscratch1, Address(rbcp, 0));
458 dispatch_base(state, table);
459 }
460
461 // remove activation
462 //
463 // Apply stack watermark barrier.
464 // Unlock the receiver if this is a synchronized method.
465 // Unlock any Java monitors from synchronized blocks.
466 // Remove the activation from the stack.
467 //
468 // If there are locked Java monitors
469 // If throw_monitor_exception
470 // throws IllegalMonitorStateException
471 // Else if install_monitor_exception
472 // installs IllegalMonitorStateException
473 // Else
474 // no error processing
475 void InterpreterMacroAssembler::remove_activation(
476 TosState state,
477 bool throw_monitor_exception,
478 bool install_monitor_exception,
479 bool notify_jvmdi) {
480 // Note: Registers r3 xmm0 may be in use for the
481 // result check if synchronized method
482 Label unlocked, unlock, no_unlock;
483
484 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
485 // that would normally not be safe to use. Such bad returns into unsafe territory of
486 // the stack, will call InterpreterRuntime::at_unwind.
487 Label slow_path;
488 Label fast_path;
489 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
490 br(Assembler::AL, fast_path);
491 bind(slow_path);
492 push(state);
493 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
494 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
495 reset_last_Java_frame(true);
496 pop(state);
497 bind(fast_path);
498
499 // get the value of _do_not_unlock_if_synchronized into r3
500 const Address do_not_unlock_if_synchronized(rthread,
501 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
502 ldrb(r3, do_not_unlock_if_synchronized);
503 strb(zr, do_not_unlock_if_synchronized); // reset the flag
504
505 // get method access flags
506 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
507 ldrh(r2, Address(r1, Method::access_flags_offset()));
508 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
509
510 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
511 // is set.
512 cbnz(r3, no_unlock);
513
514 // unlock monitor
515 push(state); // save result
516
517 // BasicObjectLock will be first in list, since this is a
518 // synchronized method. However, need to check that the object has
519 // not been unlocked by an explicit monitorexit bytecode.
520 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
521 wordSize - (int) sizeof(BasicObjectLock));
522 // We use c_rarg1 so that if we go slow path it will be the correct
523 // register for unlock_object to pass to VM directly
524 lea(c_rarg1, monitor); // address of first monitor
525
526 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
527 cbnz(r0, unlock);
528
529 pop(state);
530 if (throw_monitor_exception) {
531 // Entry already unlocked, need to throw exception
532 call_VM(noreg, CAST_FROM_FN_PTR(address,
533 InterpreterRuntime::throw_illegal_monitor_state_exception));
534 should_not_reach_here();
535 } else {
536 // Monitor already unlocked during a stack unroll. If requested,
537 // install an illegal_monitor_state_exception. Continue with
538 // stack unrolling.
539 if (install_monitor_exception) {
540 call_VM(noreg, CAST_FROM_FN_PTR(address,
541 InterpreterRuntime::new_illegal_monitor_state_exception));
542 }
543 b(unlocked);
544 }
545
546 bind(unlock);
547 unlock_object(c_rarg1);
548 pop(state);
549
550 // Check that for block-structured locking (i.e., that all locked
551 // objects has been unlocked)
552 bind(unlocked);
553
554 // r0: Might contain return value
555
556 // Check that all monitors are unlocked
557 {
558 Label loop, exception, entry, restart;
559 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
560 const Address monitor_block_top(
561 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
562 const Address monitor_block_bot(
563 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
564
565 bind(restart);
566 // We use c_rarg1 so that if we go slow path it will be the correct
567 // register for unlock_object to pass to VM directly
568 ldr(c_rarg1, monitor_block_top); // derelativize pointer
569 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
570 // c_rarg1 points to current entry, starting with top-most entry
571
572 lea(r19, monitor_block_bot); // points to word before bottom of
573 // monitor block
574 b(entry);
575
576 // Entry already locked, need to throw exception
577 bind(exception);
578
579 if (throw_monitor_exception) {
580 // Throw exception
581 MacroAssembler::call_VM(noreg,
582 CAST_FROM_FN_PTR(address, InterpreterRuntime::
583 throw_illegal_monitor_state_exception));
584 should_not_reach_here();
585 } else {
586 // Stack unrolling. Unlock object and install illegal_monitor_exception.
587 // Unlock does not block, so don't have to worry about the frame.
588 // We don't have to preserve c_rarg1 since we are going to throw an exception.
589
590 push(state);
591 unlock_object(c_rarg1);
592 pop(state);
593
594 if (install_monitor_exception) {
595 call_VM(noreg, CAST_FROM_FN_PTR(address,
596 InterpreterRuntime::
597 new_illegal_monitor_state_exception));
598 }
599
600 b(restart);
601 }
602
603 bind(loop);
604 // check if current entry is used
605 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
606 cbnz(rscratch1, exception);
607
608 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
609 bind(entry);
610 cmp(c_rarg1, r19); // check if bottom reached
611 br(Assembler::NE, loop); // if not at bottom then check this entry
612 }
613
614 bind(no_unlock);
615
616 // jvmti support
617 if (notify_jvmdi) {
618 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
619 } else {
620 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
621 }
622
623 // remove activation
624 // get sender esp
625 ldr(rscratch2,
626 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
627 if (StackReservedPages > 0) {
628 // testing if reserved zone needs to be re-enabled
629 Label no_reserved_zone_enabling;
630
631 // check if already enabled - if so no re-enabling needed
632 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
633 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
634 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
635 br(Assembler::EQ, no_reserved_zone_enabling);
636
637 // look for an overflow into the stack reserved zone, i.e.
638 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
639 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
640 cmp(rscratch2, rscratch1);
641 br(Assembler::LS, no_reserved_zone_enabling);
642
643 call_VM_leaf(
644 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
645 call_VM(noreg, CAST_FROM_FN_PTR(address,
646 InterpreterRuntime::throw_delayed_StackOverflowError));
647 should_not_reach_here();
648
649 bind(no_reserved_zone_enabling);
650 }
651
652 // restore sender esp
653 mov(esp, rscratch2);
654 // remove frame anchor
655 leave();
656 // If we're returning to interpreted code we will shortly be
657 // adjusting SP to allow some space for ESP. If we're returning to
658 // compiled code the saved sender SP was saved in sender_sp, so this
659 // restores it.
660 andr(sp, esp, -16);
661 }
662
663 // Lock object
664 //
665 // Args:
666 // c_rarg1: BasicObjectLock to be used for locking
667 //
668 // Kills:
669 // r0
670 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
671 // rscratch1, rscratch2 (scratch regs)
672 void InterpreterMacroAssembler::lock_object(Register lock_reg)
673 {
674 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
675 if (LockingMode == LM_MONITOR) {
676 call_VM_preemptable(noreg,
677 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
678 lock_reg);
679 } else {
680 Label count, done;
681
682 const Register swap_reg = r0;
683 const Register tmp = c_rarg2;
684 const Register obj_reg = c_rarg3; // Will contain the oop
685 const Register tmp2 = c_rarg4;
686 const Register tmp3 = c_rarg5;
687
688 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
689 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
690 const int mark_offset = lock_offset +
691 BasicLock::displaced_header_offset_in_bytes();
692
693 Label slow_case;
694
695 // Load object pointer into obj_reg %c_rarg3
696 ldr(obj_reg, Address(lock_reg, obj_offset));
697
698 if (DiagnoseSyncOnValueBasedClasses != 0) {
699 load_klass(tmp, obj_reg);
700 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
701 tst(tmp, KlassFlags::_misc_is_value_based_class);
702 br(Assembler::NE, slow_case);
703 }
704
705 if (LockingMode == LM_LIGHTWEIGHT) {
706 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
707 b(done);
708 } else if (LockingMode == LM_LEGACY) {
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::verify_FPU(int stack_depth, TosState state) { ; }
1407
1408 void InterpreterMacroAssembler::generate_runtime_upcalls_on_method_entry()
1409 {
1410 address upcall = RuntimeUpcalls::on_method_entry_upcall_address();
1411 if (RuntimeUpcalls::does_upcall_need_method_parameter(upcall)) {
1412 get_method(c_rarg1);
1413 call_VM(noreg,upcall, c_rarg1);
1414 } else {
1415 call_VM(noreg,upcall);
1416 }
1417 }
1418
1419 void InterpreterMacroAssembler::notify_method_entry() {
1420 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1421 // track stack depth. If it is possible to enter interp_only_mode we add
1422 // the code to check if the event should be sent.
1423 if (JvmtiExport::can_post_interpreter_events()) {
1424 Label L;
1425 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1426 cbzw(r3, L);
1427 call_VM(noreg, CAST_FROM_FN_PTR(address,
1428 InterpreterRuntime::post_method_entry));
1429 bind(L);
1430 }
1431
1432 if (DTraceMethodProbes) {
1433 get_method(c_rarg1);
1434 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1435 rthread, c_rarg1);
1436 }
1437
1438 // RedefineClasses() tracing support for obsolete method entry
1439 if (log_is_enabled(Trace, redefine, class, obsolete) ||
1440 log_is_enabled(Trace, interpreter, bytecode)) {
1441 get_method(c_rarg1);
1442 call_VM_leaf(
1443 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1444 rthread, c_rarg1);
1445 }
1446
1447 }
1448
1449
1450 void InterpreterMacroAssembler::notify_method_exit(
1451 TosState state, NotifyMethodExitMode mode) {
1452 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1453 // track stack depth. If it is possible to enter interp_only_mode we add
1454 // the code to check if the event should be sent.
1455 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1456 Label L;
1457 // Note: frame::interpreter_frame_result has a dependency on how the
1458 // method result is saved across the call to post_method_exit. If this
1459 // is changed then the interpreter_frame_result implementation will
1460 // need to be updated too.
1461
1462 // template interpreter will leave the result on the top of the stack.
1463 push(state);
1464 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1465 cbz(r3, L);
1466 call_VM(noreg,
1467 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1468 bind(L);
1469 pop(state);
1470 }
1471
1472 if (DTraceMethodProbes) {
1473 push(state);
1474 get_method(c_rarg1);
1475 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1476 rthread, c_rarg1);
1477 pop(state);
1478 }
1479 }
1480
1481
1482 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1483 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1484 int increment, Address mask,
1485 Register scratch, Register scratch2,
1486 bool preloaded, Condition cond,
1487 Label* where) {
1488 if (!preloaded) {
1489 ldrw(scratch, counter_addr);
1490 }
1491 add(scratch, scratch, increment);
1492 strw(scratch, counter_addr);
1493 ldrw(scratch2, mask);
1494 ands(scratch, scratch, scratch2);
1495 br(cond, *where);
1496 }
1497
1498 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1499 int number_of_arguments) {
1500 // interpreter specific
1501 //
1502 // Note: No need to save/restore rbcp & rlocals pointer since these
1503 // are callee saved registers and no blocking/ GC can happen
1504 // in leaf calls.
1505 #ifdef ASSERT
1506 {
1507 Label L;
1508 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1509 cbz(rscratch1, L);
1510 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1511 " last_sp != nullptr");
1512 bind(L);
1513 }
1514 #endif /* ASSERT */
1515 // super call
1516 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1517 }
1518
1519 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1520 Register java_thread,
1521 Register last_java_sp,
1522 address entry_point,
1523 int number_of_arguments,
1524 bool check_exceptions) {
1525 // interpreter specific
1526 //
1527 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1528 // really make a difference for these runtime calls, since they are
1529 // slow anyway. Btw., bcp must be saved/restored since it may change
1530 // due to GC.
1531 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1532 save_bcp();
1533 #ifdef ASSERT
1534 {
1535 Label L;
1536 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1537 cbz(rscratch1, L);
1538 stop("InterpreterMacroAssembler::call_VM_base:"
1539 " last_sp != nullptr");
1540 bind(L);
1541 }
1542 #endif /* ASSERT */
1543 // super call
1544 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1545 entry_point, number_of_arguments,
1546 check_exceptions);
1547 // interpreter specific
1548 restore_bcp();
1549 restore_locals();
1550 }
1551
1552 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1553 address entry_point,
1554 Register arg_1) {
1555 assert(arg_1 == c_rarg1, "");
1556 Label resume_pc, not_preempted;
1557
1558 #ifdef ASSERT
1559 {
1560 Label L;
1561 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1562 cbz(rscratch1, L);
1563 stop("Should not have alternate return address set");
1564 bind(L);
1565 }
1566 #endif /* ASSERT */
1567
1568 // Force freeze slow path.
1569 push_cont_fastpath();
1570
1571 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1572 adr(rscratch1, resume_pc);
1573 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1574 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1575
1576 pop_cont_fastpath();
1577
1578 // Check if preempted.
1579 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1580 cbz(rscratch1, not_preempted);
1581 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1582 br(rscratch1);
1583
1584 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1585 bind(resume_pc);
1586 restore_after_resume(false /* is_native */);
1587
1588 bind(not_preempted);
1589 }
1590
1591 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1592 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1593 blr(rscratch1);
1594 if (is_native) {
1595 // On resume we need to set up stack as expected
1596 push(dtos);
1597 push(ltos);
1598 }
1599 }
1600
1601 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1602 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1603 Label update, next, none;
1604
1605 verify_oop(obj);
1606
1607 cbnz(obj, update);
1608 orptr(mdo_addr, TypeEntries::null_seen);
1609 b(next);
1610
1611 bind(update);
1612 load_klass(obj, obj);
1613
1614 ldr(rscratch1, mdo_addr);
1615 eor(obj, obj, rscratch1);
1616 tst(obj, TypeEntries::type_klass_mask);
1617 br(Assembler::EQ, next); // klass seen before, nothing to
1618 // do. The unknown bit may have been
1619 // set already but no need to check.
1620
1621 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1622 // already unknown. Nothing to do anymore.
1623
1624 cbz(rscratch1, none);
1625 cmp(rscratch1, (u1)TypeEntries::null_seen);
1626 br(Assembler::EQ, none);
1627 // There is a chance that the checks above
1628 // fail if another thread has just set the
1629 // profiling to this obj's klass
1630 eor(obj, obj, rscratch1); // get back original value before XOR
1631 ldr(rscratch1, mdo_addr);
1632 eor(obj, obj, rscratch1);
1633 tst(obj, TypeEntries::type_klass_mask);
1634 br(Assembler::EQ, next);
1635
1636 // different than before. Cannot keep accurate profile.
1637 orptr(mdo_addr, TypeEntries::type_unknown);
1638 b(next);
1639
1640 bind(none);
1641 // first time here. Set profile type.
1642 str(obj, mdo_addr);
1643 #ifdef ASSERT
1644 andr(obj, obj, TypeEntries::type_mask);
1645 verify_klass_ptr(obj);
1646 #endif
1647
1648 bind(next);
1649 }
1650
1651 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1652 if (!ProfileInterpreter) {
1653 return;
1654 }
1655
1656 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1657 Label profile_continue;
1658
1659 test_method_data_pointer(mdp, profile_continue);
1660
1661 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1662
1663 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1664 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1665 br(Assembler::NE, profile_continue);
1666
1667 if (MethodData::profile_arguments()) {
1668 Label done;
1669 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1670
1671 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1672 if (i > 0 || MethodData::profile_return()) {
1673 // If return value type is profiled we may have no argument to profile
1674 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1675 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1676 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1677 add(rscratch1, mdp, off_to_args);
1678 br(Assembler::LT, done);
1679 }
1680 ldr(tmp, Address(callee, Method::const_offset()));
1681 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1682 // stack offset o (zero based) from the start of the argument
1683 // list, for n arguments translates into offset n - o - 1 from
1684 // the end of the argument list
1685 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1686 sub(tmp, tmp, rscratch1);
1687 sub(tmp, tmp, 1);
1688 Address arg_addr = argument_address(tmp);
1689 ldr(tmp, arg_addr);
1690
1691 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1692 profile_obj_type(tmp, mdo_arg_addr);
1693
1694 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1695 off_to_args += to_add;
1696 }
1697
1698 if (MethodData::profile_return()) {
1699 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1700 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1701 }
1702
1703 add(rscratch1, mdp, off_to_args);
1704 bind(done);
1705 mov(mdp, rscratch1);
1706
1707 if (MethodData::profile_return()) {
1708 // We're right after the type profile for the last
1709 // argument. tmp is the number of cells left in the
1710 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1711 // if there's a return to profile.
1712 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1713 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1714 }
1715 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1716 } else {
1717 assert(MethodData::profile_return(), "either profile call args or call ret");
1718 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1719 }
1720
1721 // mdp points right after the end of the
1722 // CallTypeData/VirtualCallTypeData, right after the cells for the
1723 // return value type if there's one
1724
1725 bind(profile_continue);
1726 }
1727 }
1728
1729 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1730 assert_different_registers(mdp, ret, tmp, rbcp);
1731 if (ProfileInterpreter && MethodData::profile_return()) {
1732 Label profile_continue, done;
1733
1734 test_method_data_pointer(mdp, profile_continue);
1735
1736 if (MethodData::profile_return_jsr292_only()) {
1737 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1738
1739 // If we don't profile all invoke bytecodes we must make sure
1740 // it's a bytecode we indeed profile. We can't go back to the
1741 // beginning of the ProfileData we intend to update to check its
1742 // type because we're right after it and we don't known its
1743 // length
1744 Label do_profile;
1745 ldrb(rscratch1, Address(rbcp, 0));
1746 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1747 br(Assembler::EQ, do_profile);
1748 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1749 br(Assembler::EQ, do_profile);
1750 get_method(tmp);
1751 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1752 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1753 br(Assembler::NE, profile_continue);
1754
1755 bind(do_profile);
1756 }
1757
1758 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1759 mov(tmp, ret);
1760 profile_obj_type(tmp, mdo_ret_addr);
1761
1762 bind(profile_continue);
1763 }
1764 }
1765
1766 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1767 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1768 if (ProfileInterpreter && MethodData::profile_parameters()) {
1769 Label profile_continue, done;
1770
1771 test_method_data_pointer(mdp, profile_continue);
1772
1773 // Load the offset of the area within the MDO used for
1774 // parameters. If it's negative we're not profiling any parameters
1775 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1776 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1777
1778 // Compute a pointer to the area for parameters from the offset
1779 // and move the pointer to the slot for the last
1780 // parameters. Collect profiling from last parameter down.
1781 // mdo start + parameters offset + array length - 1
1782 add(mdp, mdp, tmp1);
1783 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1784 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1785
1786 Label loop;
1787 bind(loop);
1788
1789 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1790 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1791 int per_arg_scale = exact_log2(DataLayout::cell_size);
1792 add(rscratch1, mdp, off_base);
1793 add(rscratch2, mdp, type_base);
1794
1795 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1796 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1797
1798 // load offset on the stack from the slot for this parameter
1799 ldr(tmp2, arg_off);
1800 neg(tmp2, tmp2);
1801 // read the parameter from the local area
1802 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1803
1804 // profile the parameter
1805 profile_obj_type(tmp2, arg_type);
1806
1807 // go to next parameter
1808 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1809 br(Assembler::GE, loop);
1810
1811 bind(profile_continue);
1812 }
1813 }
1814
1815 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1816 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1817 get_cache_index_at_bcp(index, 1, sizeof(u4));
1818 // Get address of invokedynamic array
1819 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1820 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1821 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1822 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1823 lea(cache, Address(cache, index));
1824 }
1825
1826 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1827 // Get index out of bytecode pointer
1828 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1829 // Take shortcut if the size is a power of 2
1830 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1831 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1832 } else {
1833 mov(cache, sizeof(ResolvedFieldEntry));
1834 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1835 }
1836 // Get address of field entries array
1837 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1838 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1839 lea(cache, Address(cache, index));
1840 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1841 membar(MacroAssembler::LoadLoad);
1842 }
1843
1844 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1845 // Get index out of bytecode pointer
1846 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1847 mov(cache, sizeof(ResolvedMethodEntry));
1848 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1849
1850 // Get address of field entries array
1851 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1852 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1853 lea(cache, Address(cache, index));
1854 }