1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_aarch64.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/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 Unimplemented();
398 }
399 if (verifyoop) {
400 interp_verify_oop(r0, state);
401 }
402
403 Label safepoint;
404 address* const safepoint_table = Interpreter::safept_table(state);
405 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
406
407 if (needs_thread_local_poll) {
408 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
409 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
410 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
411 }
412
413 if (table == Interpreter::dispatch_table(state)) {
414 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
415 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
416 } else {
417 mov(rscratch2, (address)table);
418 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
419 }
420 br(rscratch2);
421
422 if (needs_thread_local_poll) {
423 bind(safepoint);
424 lea(rscratch2, ExternalAddress((address)safepoint_table));
425 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
426 br(rscratch2);
427 }
428 }
429
430 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
431 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
432 }
433
434 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
435 dispatch_base(state, Interpreter::normal_table(state));
436 }
437
438 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
439 dispatch_base(state, Interpreter::normal_table(state), false);
440 }
441
442
443 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
444 // load next bytecode
445 ldrb(rscratch1, Address(pre(rbcp, step)));
446 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
447 }
448
449 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
450 // load current bytecode
451 ldrb(rscratch1, Address(rbcp, 0));
452 dispatch_base(state, table);
453 }
454
455 // remove activation
456 //
457 // Apply stack watermark barrier.
458 // Unlock the receiver if this is a synchronized method.
459 // Unlock any Java monitors from synchronized blocks.
460 // Remove the activation from the stack.
461 //
462 // If there are locked Java monitors
463 // If throw_monitor_exception
464 // throws IllegalMonitorStateException
465 // Else if install_monitor_exception
466 // installs IllegalMonitorStateException
467 // Else
468 // no error processing
469 void InterpreterMacroAssembler::remove_activation(
470 TosState state,
471 bool throw_monitor_exception,
472 bool install_monitor_exception,
473 bool notify_jvmdi) {
474 // Note: Registers r3 xmm0 may be in use for the
475 // result check if synchronized method
476 Label unlocked, unlock, no_unlock;
477
478 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
479 // that would normally not be safe to use. Such bad returns into unsafe territory of
480 // the stack, will call InterpreterRuntime::at_unwind.
481 Label slow_path;
482 Label fast_path;
483 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
484 br(Assembler::AL, fast_path);
485 bind(slow_path);
486 push(state);
487 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
488 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
489 reset_last_Java_frame(true);
490 pop(state);
491 bind(fast_path);
492
493 // get the value of _do_not_unlock_if_synchronized into r3
494 const Address do_not_unlock_if_synchronized(rthread,
495 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
496 ldrb(r3, do_not_unlock_if_synchronized);
497 strb(zr, do_not_unlock_if_synchronized); // reset the flag
498
499 // get method access flags
500 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
501 ldr(r2, Address(r1, Method::access_flags_offset()));
502 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
503
504 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
505 // is set.
506 cbnz(r3, no_unlock);
507
508 // unlock monitor
509 push(state); // save result
510
511 // BasicObjectLock will be first in list, since this is a
512 // synchronized method. However, need to check that the object has
513 // not been unlocked by an explicit monitorexit bytecode.
514 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
515 wordSize - (int) sizeof(BasicObjectLock));
516 // We use c_rarg1 so that if we go slow path it will be the correct
517 // register for unlock_object to pass to VM directly
518 lea(c_rarg1, monitor); // address of first monitor
519
520 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
521 cbnz(r0, unlock);
522
523 pop(state);
524 if (throw_monitor_exception) {
525 // Entry already unlocked, need to throw exception
526 call_VM(noreg, CAST_FROM_FN_PTR(address,
527 InterpreterRuntime::throw_illegal_monitor_state_exception));
528 should_not_reach_here();
529 } else {
530 // Monitor already unlocked during a stack unroll. If requested,
531 // install an illegal_monitor_state_exception. Continue with
532 // stack unrolling.
533 if (install_monitor_exception) {
534 call_VM(noreg, CAST_FROM_FN_PTR(address,
535 InterpreterRuntime::new_illegal_monitor_state_exception));
536 }
537 b(unlocked);
538 }
539
540 bind(unlock);
541 unlock_object(c_rarg1);
542 pop(state);
543
544 // Check that for block-structured locking (i.e., that all locked
545 // objects has been unlocked)
546 bind(unlocked);
547
548 // r0: Might contain return value
549
550 // Check that all monitors are unlocked
551 {
552 Label loop, exception, entry, restart;
553 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
554 const Address monitor_block_top(
555 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
556 const Address monitor_block_bot(
557 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
558
559 bind(restart);
560 // We use c_rarg1 so that if we go slow path it will be the correct
561 // register for unlock_object to pass to VM directly
562 ldr(c_rarg1, monitor_block_top); // derelativize pointer
563 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
564 // c_rarg1 points to current entry, starting with top-most entry
565
566 lea(r19, monitor_block_bot); // points to word before bottom of
567 // monitor block
568 b(entry);
569
570 // Entry already locked, need to throw exception
571 bind(exception);
572
573 if (throw_monitor_exception) {
574 // Throw exception
575 MacroAssembler::call_VM(noreg,
576 CAST_FROM_FN_PTR(address, InterpreterRuntime::
577 throw_illegal_monitor_state_exception));
578 should_not_reach_here();
579 } else {
580 // Stack unrolling. Unlock object and install illegal_monitor_exception.
581 // Unlock does not block, so don't have to worry about the frame.
582 // We don't have to preserve c_rarg1 since we are going to throw an exception.
583
584 push(state);
585 unlock_object(c_rarg1);
586 pop(state);
587
588 if (install_monitor_exception) {
589 call_VM(noreg, CAST_FROM_FN_PTR(address,
590 InterpreterRuntime::
591 new_illegal_monitor_state_exception));
592 }
593
594 b(restart);
595 }
596
597 bind(loop);
598 // check if current entry is used
599 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
600 cbnz(rscratch1, exception);
601
602 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
603 bind(entry);
604 cmp(c_rarg1, r19); // check if bottom reached
605 br(Assembler::NE, loop); // if not at bottom then check this entry
606 }
607
608 bind(no_unlock);
609
610 // jvmti support
611 if (notify_jvmdi) {
612 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
613 } else {
614 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
615 }
616
617 // remove activation
618 // get sender esp
619 ldr(rscratch2,
620 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
621 if (StackReservedPages > 0) {
622 // testing if reserved zone needs to be re-enabled
623 Label no_reserved_zone_enabling;
624
625 // check if already enabled - if so no re-enabling needed
626 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
627 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
628 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
629 br(Assembler::EQ, no_reserved_zone_enabling);
630
631 // look for an overflow into the stack reserved zone, i.e.
632 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
633 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
634 cmp(rscratch2, rscratch1);
635 br(Assembler::LS, no_reserved_zone_enabling);
636
637 call_VM_leaf(
638 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
639 call_VM(noreg, CAST_FROM_FN_PTR(address,
640 InterpreterRuntime::throw_delayed_StackOverflowError));
641 should_not_reach_here();
642
643 bind(no_reserved_zone_enabling);
644 }
645
646 // restore sender esp
647 mov(esp, rscratch2);
648 // remove frame anchor
649 leave();
650 // If we're returning to interpreted code we will shortly be
651 // adjusting SP to allow some space for ESP. If we're returning to
652 // compiled code the saved sender SP was saved in sender_sp, so this
653 // restores it.
654 andr(sp, esp, -16);
655 }
656
657 // Lock object
658 //
659 // Args:
660 // c_rarg1: BasicObjectLock to be used for locking
661 //
662 // Kills:
663 // r0
664 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
665 // rscratch1, rscratch2 (scratch regs)
666 void InterpreterMacroAssembler::lock_object(Register lock_reg)
667 {
668 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
669 if (LockingMode == LM_MONITOR) {
670 call_VM_preemptable(noreg,
671 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
672 lock_reg);
673 } else {
674 Label count, done;
675
676 const Register swap_reg = r0;
677 const Register tmp = c_rarg2;
678 const Register obj_reg = c_rarg3; // Will contain the oop
679 const Register tmp2 = c_rarg4;
680 const Register tmp3 = c_rarg5;
681
682 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
683 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
684 const int mark_offset = lock_offset +
685 BasicLock::displaced_header_offset_in_bytes();
686
687 Label slow_case;
688
689 // Load object pointer into obj_reg %c_rarg3
690 ldr(obj_reg, Address(lock_reg, obj_offset));
691
692 if (DiagnoseSyncOnValueBasedClasses != 0) {
693 load_klass(tmp, obj_reg);
694 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
695 tst(tmp, KlassFlags::_misc_is_value_based_class);
696 br(Assembler::NE, slow_case);
697 }
698
699 if (LockingMode == LM_LIGHTWEIGHT) {
700 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
701 b(done);
702 } else if (LockingMode == LM_LEGACY) {
703 // Load (object->mark() | 1) into swap_reg
704 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
705 orr(swap_reg, rscratch1, 1);
706
707 // Save (object->mark() | 1) into BasicLock's displaced header
708 str(swap_reg, Address(lock_reg, mark_offset));
709
710 assert(lock_offset == 0,
711 "displached header must be first word in BasicObjectLock");
712
713 Label fail;
714 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
715
716 // Fast check for recursive lock.
717 //
718 // Can apply the optimization only if this is a stack lock
719 // allocated in this thread. For efficiency, we can focus on
720 // recently allocated stack locks (instead of reading the stack
721 // base and checking whether 'mark' points inside the current
722 // thread stack):
723 // 1) (mark & 7) == 0, and
724 // 2) sp <= mark < mark + os::pagesize()
725 //
726 // Warning: sp + os::pagesize can overflow the stack base. We must
727 // neither apply the optimization for an inflated lock allocated
728 // just above the thread stack (this is why condition 1 matters)
729 // nor apply the optimization if the stack lock is inside the stack
730 // of another thread. The latter is avoided even in case of overflow
731 // because we have guard pages at the end of all stacks. Hence, if
732 // we go over the stack base and hit the stack of another thread,
733 // this should not be in a writeable area that could contain a
734 // stack lock allocated by that thread. As a consequence, a stack
735 // lock less than page size away from sp is guaranteed to be
736 // owned by the current thread.
737 //
738 // These 3 tests can be done by evaluating the following
739 // expression: ((mark - sp) & (7 - os::vm_page_size())),
740 // assuming both stack pointer and pagesize have their
741 // least significant 3 bits clear.
742 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
743 // NOTE2: aarch64 does not like to subtract sp from rn so take a
744 // copy
745 mov(rscratch1, sp);
746 sub(swap_reg, swap_reg, rscratch1);
747 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
748
749 // Save the test result, for recursive case, the result is zero
750 str(swap_reg, Address(lock_reg, mark_offset));
751 br(Assembler::NE, slow_case);
752
753 bind(count);
754 inc_held_monitor_count(rscratch1);
755 b(done);
756 }
757 bind(slow_case);
758
759 // Call the runtime routine for slow case
760 call_VM_preemptable(noreg,
761 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
762 lock_reg);
763
764 bind(done);
765 }
766 }
767
768
769 // Unlocks an object. Used in monitorexit bytecode and
770 // remove_activation. Throws an IllegalMonitorException if object is
771 // not locked by current thread.
772 //
773 // Args:
774 // c_rarg1: BasicObjectLock for lock
775 //
776 // Kills:
777 // r0
778 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
779 // rscratch1, rscratch2 (scratch regs)
780 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
781 {
782 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
783
784 if (LockingMode == LM_MONITOR) {
785 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
786 } else {
787 Label count, done;
788
789 const Register swap_reg = r0;
790 const Register header_reg = c_rarg2; // Will contain the old oopMark
791 const Register obj_reg = c_rarg3; // Will contain the oop
792 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
793
794 save_bcp(); // Save in case of exception
795
796 if (LockingMode != LM_LIGHTWEIGHT) {
797 // Convert from BasicObjectLock structure to object and BasicLock
798 // structure Store the BasicLock address into %r0
799 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
800 }
801
802 // Load oop into obj_reg(%c_rarg3)
803 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
804
805 // Free entry
806 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
807
808 Label slow_case;
809 if (LockingMode == LM_LIGHTWEIGHT) {
810 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
811 b(done);
812 } else if (LockingMode == LM_LEGACY) {
813 // Load the old header from BasicLock structure
814 ldr(header_reg, Address(swap_reg,
815 BasicLock::displaced_header_offset_in_bytes()));
816
817 // Test for recursion
818 cbz(header_reg, count);
819
820 // Atomic swap back the old header
821 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
822
823 bind(count);
824 dec_held_monitor_count(rscratch1);
825 b(done);
826 }
827
828 bind(slow_case);
829 // Call the runtime routine for slow case.
830 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
831 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
832 bind(done);
833 restore_bcp();
834 }
835 }
836
837 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
838 Label& zero_continue) {
839 assert(ProfileInterpreter, "must be profiling interpreter");
840 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
841 cbz(mdp, zero_continue);
842 }
843
844 // Set the method data pointer for the current bcp.
845 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
846 assert(ProfileInterpreter, "must be profiling interpreter");
847 Label set_mdp;
848 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
849
850 // Test MDO to avoid the call if it is null.
851 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
852 cbz(r0, set_mdp);
853 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
854 // r0: mdi
855 // mdo is guaranteed to be non-zero here, we checked for it before the call.
856 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
857 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
858 add(r0, r1, r0);
859 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
860 bind(set_mdp);
861 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
862 }
863
864 void InterpreterMacroAssembler::verify_method_data_pointer() {
865 assert(ProfileInterpreter, "must be profiling interpreter");
866 #ifdef ASSERT
867 Label verify_continue;
868 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
869 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
870 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
871 get_method(r1);
872
873 // If the mdp is valid, it will point to a DataLayout header which is
874 // consistent with the bcp. The converse is highly probable also.
875 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
876 ldr(rscratch1, Address(r1, Method::const_offset()));
877 add(r2, r2, rscratch1, Assembler::LSL);
878 lea(r2, Address(r2, ConstMethod::codes_offset()));
879 cmp(r2, rbcp);
880 br(Assembler::EQ, verify_continue);
881 // r1: method
882 // rbcp: bcp // rbcp == 22
883 // r3: mdp
884 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
885 r1, rbcp, r3);
886 bind(verify_continue);
887 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
888 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
889 #endif // ASSERT
890 }
891
892
893 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
894 int constant,
895 Register value) {
896 assert(ProfileInterpreter, "must be profiling interpreter");
897 Address data(mdp_in, constant);
898 str(value, data);
899 }
900
901
902 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
903 int constant,
904 bool decrement) {
905 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
906 }
907
908 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
909 Register reg,
910 int constant,
911 bool decrement) {
912 assert(ProfileInterpreter, "must be profiling interpreter");
913 // %%% this does 64bit counters at best it is wasting space
914 // at worst it is a rare bug when counters overflow
915
916 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
917
918 Address addr1(mdp_in, constant);
919 Address addr2(rscratch2, reg, Address::lsl(0));
920 Address &addr = addr1;
921 if (reg != noreg) {
922 lea(rscratch2, addr1);
923 addr = addr2;
924 }
925
926 if (decrement) {
927 // Decrement the register. Set condition codes.
928 // Intel does this
929 // addptr(data, (int32_t) -DataLayout::counter_increment);
930 // If the decrement causes the counter to overflow, stay negative
931 // Label L;
932 // jcc(Assembler::negative, L);
933 // addptr(data, (int32_t) DataLayout::counter_increment);
934 // so we do this
935 ldr(rscratch1, addr);
936 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
937 Label L;
938 br(Assembler::LO, L); // skip store if counter underflow
939 str(rscratch1, addr);
940 bind(L);
941 } else {
942 assert(DataLayout::counter_increment == 1,
943 "flow-free idiom only works with 1");
944 // Intel does this
945 // Increment the register. Set carry flag.
946 // addptr(data, DataLayout::counter_increment);
947 // If the increment causes the counter to overflow, pull back by 1.
948 // sbbptr(data, (int32_t)0);
949 // so we do this
950 ldr(rscratch1, addr);
951 adds(rscratch1, rscratch1, DataLayout::counter_increment);
952 Label L;
953 br(Assembler::CS, L); // skip store if counter overflow
954 str(rscratch1, addr);
955 bind(L);
956 }
957 }
958
959 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
960 int flag_byte_constant) {
961 assert(ProfileInterpreter, "must be profiling interpreter");
962 int flags_offset = in_bytes(DataLayout::flags_offset());
963 // Set the flag
964 ldrb(rscratch1, Address(mdp_in, flags_offset));
965 orr(rscratch1, rscratch1, flag_byte_constant);
966 strb(rscratch1, Address(mdp_in, flags_offset));
967 }
968
969
970 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
971 int offset,
972 Register value,
973 Register test_value_out,
974 Label& not_equal_continue) {
975 assert(ProfileInterpreter, "must be profiling interpreter");
976 if (test_value_out == noreg) {
977 ldr(rscratch1, Address(mdp_in, offset));
978 cmp(value, rscratch1);
979 } else {
980 // Put the test value into a register, so caller can use it:
981 ldr(test_value_out, Address(mdp_in, offset));
982 cmp(value, test_value_out);
983 }
984 br(Assembler::NE, not_equal_continue);
985 }
986
987
988 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
989 int offset_of_disp) {
990 assert(ProfileInterpreter, "must be profiling interpreter");
991 ldr(rscratch1, Address(mdp_in, offset_of_disp));
992 add(mdp_in, mdp_in, rscratch1, LSL);
993 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
994 }
995
996
997 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
998 Register reg,
999 int offset_of_disp) {
1000 assert(ProfileInterpreter, "must be profiling interpreter");
1001 lea(rscratch1, Address(mdp_in, offset_of_disp));
1002 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1003 add(mdp_in, mdp_in, rscratch1, LSL);
1004 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1005 }
1006
1007
1008 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1009 int constant) {
1010 assert(ProfileInterpreter, "must be profiling interpreter");
1011 add(mdp_in, mdp_in, (unsigned)constant);
1012 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1013 }
1014
1015
1016 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1017 assert(ProfileInterpreter, "must be profiling interpreter");
1018 // save/restore across call_VM
1019 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1020 call_VM(noreg,
1021 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1022 return_bci);
1023 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1024 }
1025
1026
1027 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1028 Register bumped_count) {
1029 if (ProfileInterpreter) {
1030 Label profile_continue;
1031
1032 // If no method data exists, go to profile_continue.
1033 // Otherwise, assign to mdp
1034 test_method_data_pointer(mdp, profile_continue);
1035
1036 // We are taking a branch. Increment the taken count.
1037 // We inline increment_mdp_data_at to return bumped_count in a register
1038 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1039 Address data(mdp, in_bytes(JumpData::taken_offset()));
1040 ldr(bumped_count, data);
1041 assert(DataLayout::counter_increment == 1,
1042 "flow-free idiom only works with 1");
1043 // Intel does this to catch overflow
1044 // addptr(bumped_count, DataLayout::counter_increment);
1045 // sbbptr(bumped_count, 0);
1046 // so we do this
1047 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1048 Label L;
1049 br(Assembler::CS, L); // skip store if counter overflow
1050 str(bumped_count, data);
1051 bind(L);
1052 // The method data pointer needs to be updated to reflect the new target.
1053 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1054 bind(profile_continue);
1055 }
1056 }
1057
1058
1059 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1060 if (ProfileInterpreter) {
1061 Label profile_continue;
1062
1063 // If no method data exists, go to profile_continue.
1064 test_method_data_pointer(mdp, profile_continue);
1065
1066 // We are taking a branch. Increment the not taken count.
1067 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1068
1069 // The method data pointer needs to be updated to correspond to
1070 // the next bytecode
1071 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1072 bind(profile_continue);
1073 }
1074 }
1075
1076
1077 void InterpreterMacroAssembler::profile_call(Register mdp) {
1078 if (ProfileInterpreter) {
1079 Label profile_continue;
1080
1081 // If no method data exists, go to profile_continue.
1082 test_method_data_pointer(mdp, profile_continue);
1083
1084 // We are making a call. Increment the count.
1085 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1086
1087 // The method data pointer needs to be updated to reflect the new target.
1088 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1089 bind(profile_continue);
1090 }
1091 }
1092
1093 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1094 if (ProfileInterpreter) {
1095 Label profile_continue;
1096
1097 // If no method data exists, go to profile_continue.
1098 test_method_data_pointer(mdp, profile_continue);
1099
1100 // We are making a call. Increment the count.
1101 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1102
1103 // The method data pointer needs to be updated to reflect the new target.
1104 update_mdp_by_constant(mdp,
1105 in_bytes(VirtualCallData::
1106 virtual_call_data_size()));
1107 bind(profile_continue);
1108 }
1109 }
1110
1111
1112 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1113 Register mdp,
1114 Register reg2,
1115 bool receiver_can_be_null) {
1116 if (ProfileInterpreter) {
1117 Label profile_continue;
1118
1119 // If no method data exists, go to profile_continue.
1120 test_method_data_pointer(mdp, profile_continue);
1121
1122 Label skip_receiver_profile;
1123 if (receiver_can_be_null) {
1124 Label not_null;
1125 // We are making a call. Increment the count for null receiver.
1126 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1127 b(skip_receiver_profile);
1128 bind(not_null);
1129 }
1130
1131 // Record the receiver type.
1132 record_klass_in_profile(receiver, mdp, reg2);
1133 bind(skip_receiver_profile);
1134
1135 // The method data pointer needs to be updated to reflect the new target.
1136 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1137 bind(profile_continue);
1138 }
1139 }
1140
1141 // This routine creates a state machine for updating the multi-row
1142 // type profile at a virtual call site (or other type-sensitive bytecode).
1143 // The machine visits each row (of receiver/count) until the receiver type
1144 // is found, or until it runs out of rows. At the same time, it remembers
1145 // the location of the first empty row. (An empty row records null for its
1146 // receiver, and can be allocated for a newly-observed receiver type.)
1147 // Because there are two degrees of freedom in the state, a simple linear
1148 // search will not work; it must be a decision tree. Hence this helper
1149 // function is recursive, to generate the required tree structured code.
1150 // It's the interpreter, so we are trading off code space for speed.
1151 // See below for example code.
1152 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1153 Register receiver, Register mdp,
1154 Register reg2, int start_row,
1155 Label& done) {
1156 if (TypeProfileWidth == 0) {
1157 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1158 } else {
1159 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1160 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1161 }
1162 }
1163
1164 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1165 Register reg2, int start_row, Label& done, int total_rows,
1166 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1167 int last_row = total_rows - 1;
1168 assert(start_row <= last_row, "must be work left to do");
1169 // Test this row for both the item and for null.
1170 // Take any of three different outcomes:
1171 // 1. found item => increment count and goto done
1172 // 2. found null => keep looking for case 1, maybe allocate this cell
1173 // 3. found something else => keep looking for cases 1 and 2
1174 // Case 3 is handled by a recursive call.
1175 for (int row = start_row; row <= last_row; row++) {
1176 Label next_test;
1177 bool test_for_null_also = (row == start_row);
1178
1179 // See if the item is item[n].
1180 int item_offset = in_bytes(item_offset_fn(row));
1181 test_mdp_data_at(mdp, item_offset, item,
1182 (test_for_null_also ? reg2 : noreg),
1183 next_test);
1184 // (Reg2 now contains the item from the CallData.)
1185
1186 // The item is item[n]. Increment count[n].
1187 int count_offset = in_bytes(item_count_offset_fn(row));
1188 increment_mdp_data_at(mdp, count_offset);
1189 b(done);
1190 bind(next_test);
1191
1192 if (test_for_null_also) {
1193 Label found_null;
1194 // Failed the equality check on item[n]... Test for null.
1195 if (start_row == last_row) {
1196 // The only thing left to do is handle the null case.
1197 cbz(reg2, found_null);
1198 // Item did not match any saved item and there is no empty row for it.
1199 // Increment total counter to indicate polymorphic case.
1200 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1201 b(done);
1202 bind(found_null);
1203 break;
1204 }
1205 // Since null is rare, make it be the branch-taken case.
1206 cbz(reg2, found_null);
1207
1208 // Put all the "Case 3" tests here.
1209 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1210 item_offset_fn, item_count_offset_fn);
1211
1212 // Found a null. Keep searching for a matching item,
1213 // but remember that this is an empty (unused) slot.
1214 bind(found_null);
1215 }
1216 }
1217
1218 // In the fall-through case, we found no matching item, but we
1219 // observed the item[start_row] is null.
1220
1221 // Fill in the item field and increment the count.
1222 int item_offset = in_bytes(item_offset_fn(start_row));
1223 set_mdp_data_at(mdp, item_offset, item);
1224 int count_offset = in_bytes(item_count_offset_fn(start_row));
1225 mov(reg2, DataLayout::counter_increment);
1226 set_mdp_data_at(mdp, count_offset, reg2);
1227 if (start_row > 0) {
1228 b(done);
1229 }
1230 }
1231
1232 // Example state machine code for three profile rows:
1233 // // main copy of decision tree, rooted at row[1]
1234 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1235 // if (row[0].rec != nullptr) {
1236 // // inner copy of decision tree, rooted at row[1]
1237 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1238 // if (row[1].rec != nullptr) {
1239 // // degenerate decision tree, rooted at row[2]
1240 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1241 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1242 // row[2].init(rec); goto done;
1243 // } else {
1244 // // remember row[1] is empty
1245 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1246 // row[1].init(rec); goto done;
1247 // }
1248 // } else {
1249 // // remember row[0] is empty
1250 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1251 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1252 // row[0].init(rec); goto done;
1253 // }
1254 // done:
1255
1256 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1257 Register mdp, Register reg2) {
1258 assert(ProfileInterpreter, "must be profiling");
1259 Label done;
1260
1261 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1262
1263 bind (done);
1264 }
1265
1266 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1267 Register mdp) {
1268 if (ProfileInterpreter) {
1269 Label profile_continue;
1270 uint row;
1271
1272 // If no method data exists, go to profile_continue.
1273 test_method_data_pointer(mdp, profile_continue);
1274
1275 // Update the total ret count.
1276 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1277
1278 for (row = 0; row < RetData::row_limit(); row++) {
1279 Label next_test;
1280
1281 // See if return_bci is equal to bci[n]:
1282 test_mdp_data_at(mdp,
1283 in_bytes(RetData::bci_offset(row)),
1284 return_bci, noreg,
1285 next_test);
1286
1287 // return_bci is equal to bci[n]. Increment the count.
1288 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1289
1290 // The method data pointer needs to be updated to reflect the new target.
1291 update_mdp_by_offset(mdp,
1292 in_bytes(RetData::bci_displacement_offset(row)));
1293 b(profile_continue);
1294 bind(next_test);
1295 }
1296
1297 update_mdp_for_ret(return_bci);
1298
1299 bind(profile_continue);
1300 }
1301 }
1302
1303 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1304 if (ProfileInterpreter) {
1305 Label profile_continue;
1306
1307 // If no method data exists, go to profile_continue.
1308 test_method_data_pointer(mdp, profile_continue);
1309
1310 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1311
1312 // The method data pointer needs to be updated.
1313 int mdp_delta = in_bytes(BitData::bit_data_size());
1314 if (TypeProfileCasts) {
1315 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1316 }
1317 update_mdp_by_constant(mdp, mdp_delta);
1318
1319 bind(profile_continue);
1320 }
1321 }
1322
1323 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1324 if (ProfileInterpreter) {
1325 Label profile_continue;
1326
1327 // If no method data exists, go to profile_continue.
1328 test_method_data_pointer(mdp, profile_continue);
1329
1330 // The method data pointer needs to be updated.
1331 int mdp_delta = in_bytes(BitData::bit_data_size());
1332 if (TypeProfileCasts) {
1333 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1334
1335 // Record the object type.
1336 record_klass_in_profile(klass, mdp, reg2);
1337 }
1338 update_mdp_by_constant(mdp, mdp_delta);
1339
1340 bind(profile_continue);
1341 }
1342 }
1343
1344 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1345 if (ProfileInterpreter) {
1346 Label profile_continue;
1347
1348 // If no method data exists, go to profile_continue.
1349 test_method_data_pointer(mdp, profile_continue);
1350
1351 // Update the default case count
1352 increment_mdp_data_at(mdp,
1353 in_bytes(MultiBranchData::default_count_offset()));
1354
1355 // The method data pointer needs to be updated.
1356 update_mdp_by_offset(mdp,
1357 in_bytes(MultiBranchData::
1358 default_displacement_offset()));
1359
1360 bind(profile_continue);
1361 }
1362 }
1363
1364 void InterpreterMacroAssembler::profile_switch_case(Register index,
1365 Register mdp,
1366 Register reg2) {
1367 if (ProfileInterpreter) {
1368 Label profile_continue;
1369
1370 // If no method data exists, go to profile_continue.
1371 test_method_data_pointer(mdp, profile_continue);
1372
1373 // Build the base (index * per_case_size_in_bytes()) +
1374 // case_array_offset_in_bytes()
1375 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1376 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1377 Assembler::maddw(index, index, reg2, rscratch1);
1378
1379 // Update the case count
1380 increment_mdp_data_at(mdp,
1381 index,
1382 in_bytes(MultiBranchData::relative_count_offset()));
1383
1384 // The method data pointer needs to be updated.
1385 update_mdp_by_offset(mdp,
1386 index,
1387 in_bytes(MultiBranchData::
1388 relative_displacement_offset()));
1389
1390 bind(profile_continue);
1391 }
1392 }
1393
1394 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1395 if (state == atos) {
1396 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1397 }
1398 }
1399
1400 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1401
1402 void InterpreterMacroAssembler::generate_runtime_upcalls_on_method_entry()
1403 {
1404 address upcall = RuntimeUpcalls::on_method_entry_upcall_address();
1405 if (RuntimeUpcalls::does_upcall_need_method_parameter(upcall)) {
1406 get_method(c_rarg1);
1407 call_VM(noreg,upcall, c_rarg1);
1408 } else {
1409 call_VM(noreg,upcall);
1410 }
1411 }
1412
1413 void InterpreterMacroAssembler::notify_method_entry() {
1414 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1415 // track stack depth. If it is possible to enter interp_only_mode we add
1416 // the code to check if the event should be sent.
1417 if (JvmtiExport::can_post_interpreter_events()) {
1418 Label L;
1419 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1420 cbzw(r3, L);
1421 call_VM(noreg, CAST_FROM_FN_PTR(address,
1422 InterpreterRuntime::post_method_entry));
1423 bind(L);
1424 }
1425
1426 if (DTraceMethodProbes) {
1427 get_method(c_rarg1);
1428 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1429 rthread, c_rarg1);
1430 }
1431
1432 // RedefineClasses() tracing support for obsolete method entry
1433 if (log_is_enabled(Trace, redefine, class, obsolete) ||
1434 log_is_enabled(Trace, interpreter, bytecode)) {
1435 get_method(c_rarg1);
1436 call_VM_leaf(
1437 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1438 rthread, c_rarg1);
1439 }
1440
1441 }
1442
1443
1444 void InterpreterMacroAssembler::notify_method_exit(
1445 TosState state, NotifyMethodExitMode mode) {
1446 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1447 // track stack depth. If it is possible to enter interp_only_mode we add
1448 // the code to check if the event should be sent.
1449 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1450 Label L;
1451 // Note: frame::interpreter_frame_result has a dependency on how the
1452 // method result is saved across the call to post_method_exit. If this
1453 // is changed then the interpreter_frame_result implementation will
1454 // need to be updated too.
1455
1456 // template interpreter will leave the result on the top of the stack.
1457 push(state);
1458 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1459 cbz(r3, L);
1460 call_VM(noreg,
1461 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1462 bind(L);
1463 pop(state);
1464 }
1465
1466 if (DTraceMethodProbes) {
1467 push(state);
1468 get_method(c_rarg1);
1469 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1470 rthread, c_rarg1);
1471 pop(state);
1472 }
1473 }
1474
1475
1476 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1477 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1478 int increment, Address mask,
1479 Register scratch, Register scratch2,
1480 bool preloaded, Condition cond,
1481 Label* where) {
1482 if (!preloaded) {
1483 ldrw(scratch, counter_addr);
1484 }
1485 add(scratch, scratch, increment);
1486 strw(scratch, counter_addr);
1487 ldrw(scratch2, mask);
1488 ands(scratch, scratch, scratch2);
1489 br(cond, *where);
1490 }
1491
1492 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1493 int number_of_arguments) {
1494 // interpreter specific
1495 //
1496 // Note: No need to save/restore rbcp & rlocals pointer since these
1497 // are callee saved registers and no blocking/ GC can happen
1498 // in leaf calls.
1499 #ifdef ASSERT
1500 {
1501 Label L;
1502 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1503 cbz(rscratch1, L);
1504 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1505 " last_sp != nullptr");
1506 bind(L);
1507 }
1508 #endif /* ASSERT */
1509 // super call
1510 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1511 }
1512
1513 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1514 Register java_thread,
1515 Register last_java_sp,
1516 address entry_point,
1517 int number_of_arguments,
1518 bool check_exceptions) {
1519 // interpreter specific
1520 //
1521 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1522 // really make a difference for these runtime calls, since they are
1523 // slow anyway. Btw., bcp must be saved/restored since it may change
1524 // due to GC.
1525 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1526 save_bcp();
1527 #ifdef ASSERT
1528 {
1529 Label L;
1530 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1531 cbz(rscratch1, L);
1532 stop("InterpreterMacroAssembler::call_VM_base:"
1533 " last_sp != nullptr");
1534 bind(L);
1535 }
1536 #endif /* ASSERT */
1537 // super call
1538 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1539 entry_point, number_of_arguments,
1540 check_exceptions);
1541 // interpreter specific
1542 restore_bcp();
1543 restore_locals();
1544 }
1545
1546 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1547 address entry_point,
1548 Register arg_1) {
1549 assert(arg_1 == c_rarg1, "");
1550 Label resume_pc, not_preempted;
1551
1552 #ifdef ASSERT
1553 {
1554 Label L;
1555 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1556 cbz(rscratch1, L);
1557 stop("Should not have alternate return address set");
1558 bind(L);
1559 }
1560 #endif /* ASSERT */
1561
1562 // Force freeze slow path.
1563 push_cont_fastpath();
1564
1565 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1566 adr(rscratch1, resume_pc);
1567 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1568 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1569
1570 pop_cont_fastpath();
1571
1572 // Check if preempted.
1573 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1574 cbz(rscratch1, not_preempted);
1575 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1576 br(rscratch1);
1577
1578 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1579 bind(resume_pc);
1580 restore_after_resume(false /* is_native */);
1581
1582 bind(not_preempted);
1583 }
1584
1585 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1586 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1587 blr(rscratch1);
1588 if (is_native) {
1589 // On resume we need to set up stack as expected
1590 push(dtos);
1591 push(ltos);
1592 }
1593 }
1594
1595 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1596 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1597 Label update, next, none;
1598
1599 verify_oop(obj);
1600
1601 cbnz(obj, update);
1602 orptr(mdo_addr, TypeEntries::null_seen);
1603 b(next);
1604
1605 bind(update);
1606 load_klass(obj, obj);
1607
1608 ldr(rscratch1, mdo_addr);
1609 eor(obj, obj, rscratch1);
1610 tst(obj, TypeEntries::type_klass_mask);
1611 br(Assembler::EQ, next); // klass seen before, nothing to
1612 // do. The unknown bit may have been
1613 // set already but no need to check.
1614
1615 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1616 // already unknown. Nothing to do anymore.
1617
1618 cbz(rscratch1, none);
1619 cmp(rscratch1, (u1)TypeEntries::null_seen);
1620 br(Assembler::EQ, none);
1621 // There is a chance that the checks above
1622 // fail if another thread has just set the
1623 // profiling to this obj's klass
1624 eor(obj, obj, rscratch1); // get back original value before XOR
1625 ldr(rscratch1, mdo_addr);
1626 eor(obj, obj, rscratch1);
1627 tst(obj, TypeEntries::type_klass_mask);
1628 br(Assembler::EQ, next);
1629
1630 // different than before. Cannot keep accurate profile.
1631 orptr(mdo_addr, TypeEntries::type_unknown);
1632 b(next);
1633
1634 bind(none);
1635 // first time here. Set profile type.
1636 str(obj, mdo_addr);
1637 #ifdef ASSERT
1638 andr(obj, obj, TypeEntries::type_mask);
1639 verify_klass_ptr(obj);
1640 #endif
1641
1642 bind(next);
1643 }
1644
1645 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1646 if (!ProfileInterpreter) {
1647 return;
1648 }
1649
1650 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1651 Label profile_continue;
1652
1653 test_method_data_pointer(mdp, profile_continue);
1654
1655 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1656
1657 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1658 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1659 br(Assembler::NE, profile_continue);
1660
1661 if (MethodData::profile_arguments()) {
1662 Label done;
1663 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1664
1665 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1666 if (i > 0 || MethodData::profile_return()) {
1667 // If return value type is profiled we may have no argument to profile
1668 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1669 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1670 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1671 add(rscratch1, mdp, off_to_args);
1672 br(Assembler::LT, done);
1673 }
1674 ldr(tmp, Address(callee, Method::const_offset()));
1675 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1676 // stack offset o (zero based) from the start of the argument
1677 // list, for n arguments translates into offset n - o - 1 from
1678 // the end of the argument list
1679 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1680 sub(tmp, tmp, rscratch1);
1681 sub(tmp, tmp, 1);
1682 Address arg_addr = argument_address(tmp);
1683 ldr(tmp, arg_addr);
1684
1685 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1686 profile_obj_type(tmp, mdo_arg_addr);
1687
1688 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1689 off_to_args += to_add;
1690 }
1691
1692 if (MethodData::profile_return()) {
1693 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1694 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1695 }
1696
1697 add(rscratch1, mdp, off_to_args);
1698 bind(done);
1699 mov(mdp, rscratch1);
1700
1701 if (MethodData::profile_return()) {
1702 // We're right after the type profile for the last
1703 // argument. tmp is the number of cells left in the
1704 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1705 // if there's a return to profile.
1706 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1707 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1708 }
1709 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1710 } else {
1711 assert(MethodData::profile_return(), "either profile call args or call ret");
1712 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1713 }
1714
1715 // mdp points right after the end of the
1716 // CallTypeData/VirtualCallTypeData, right after the cells for the
1717 // return value type if there's one
1718
1719 bind(profile_continue);
1720 }
1721 }
1722
1723 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1724 assert_different_registers(mdp, ret, tmp, rbcp);
1725 if (ProfileInterpreter && MethodData::profile_return()) {
1726 Label profile_continue, done;
1727
1728 test_method_data_pointer(mdp, profile_continue);
1729
1730 if (MethodData::profile_return_jsr292_only()) {
1731 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1732
1733 // If we don't profile all invoke bytecodes we must make sure
1734 // it's a bytecode we indeed profile. We can't go back to the
1735 // beginning of the ProfileData we intend to update to check its
1736 // type because we're right after it and we don't known its
1737 // length
1738 Label do_profile;
1739 ldrb(rscratch1, Address(rbcp, 0));
1740 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1741 br(Assembler::EQ, do_profile);
1742 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1743 br(Assembler::EQ, do_profile);
1744 get_method(tmp);
1745 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1746 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1747 br(Assembler::NE, profile_continue);
1748
1749 bind(do_profile);
1750 }
1751
1752 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1753 mov(tmp, ret);
1754 profile_obj_type(tmp, mdo_ret_addr);
1755
1756 bind(profile_continue);
1757 }
1758 }
1759
1760 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1761 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1762 if (ProfileInterpreter && MethodData::profile_parameters()) {
1763 Label profile_continue, done;
1764
1765 test_method_data_pointer(mdp, profile_continue);
1766
1767 // Load the offset of the area within the MDO used for
1768 // parameters. If it's negative we're not profiling any parameters
1769 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1770 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1771
1772 // Compute a pointer to the area for parameters from the offset
1773 // and move the pointer to the slot for the last
1774 // parameters. Collect profiling from last parameter down.
1775 // mdo start + parameters offset + array length - 1
1776 add(mdp, mdp, tmp1);
1777 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1778 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1779
1780 Label loop;
1781 bind(loop);
1782
1783 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1784 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1785 int per_arg_scale = exact_log2(DataLayout::cell_size);
1786 add(rscratch1, mdp, off_base);
1787 add(rscratch2, mdp, type_base);
1788
1789 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1790 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1791
1792 // load offset on the stack from the slot for this parameter
1793 ldr(tmp2, arg_off);
1794 neg(tmp2, tmp2);
1795 // read the parameter from the local area
1796 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1797
1798 // profile the parameter
1799 profile_obj_type(tmp2, arg_type);
1800
1801 // go to next parameter
1802 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1803 br(Assembler::GE, loop);
1804
1805 bind(profile_continue);
1806 }
1807 }
1808
1809 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1810 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1811 get_cache_index_at_bcp(index, 1, sizeof(u4));
1812 // Get address of invokedynamic array
1813 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1814 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1815 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1816 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1817 lea(cache, Address(cache, index));
1818 }
1819
1820 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1821 // Get index out of bytecode pointer
1822 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1823 // Take shortcut if the size is a power of 2
1824 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1825 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1826 } else {
1827 mov(cache, sizeof(ResolvedFieldEntry));
1828 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1829 }
1830 // Get address of field entries array
1831 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1832 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1833 lea(cache, Address(cache, index));
1834 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1835 membar(MacroAssembler::LoadLoad);
1836 }
1837
1838 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1839 // Get index out of bytecode pointer
1840 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1841 mov(cache, sizeof(ResolvedMethodEntry));
1842 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1843
1844 // Get address of field entries array
1845 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1846 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1847 lea(cache, Address(cache, index));
1848 }