1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/macroAssembler.inline.hpp"
27 #include "compiler/compiler_globals.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interp_masm_aarch64.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "logging/log.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/markWord.hpp"
36 #include "oops/method.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/resolvedFieldEntry.hpp"
39 #include "oops/resolvedIndyEntry.hpp"
40 #include "oops/resolvedMethodEntry.hpp"
41 #include "prims/jvmtiExport.hpp"
42 #include "prims/jvmtiThreadState.hpp"
43 #include "runtime/basicLock.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/javaThread.hpp"
46 #include "runtime/safepointMechanism.hpp"
47 #include "runtime/sharedRuntime.hpp"
48 #include "utilities/powerOfTwo.hpp"
49
50 void InterpreterMacroAssembler::narrow(Register result) {
51
52 // Get method->_constMethod->_result_type
53 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
54 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
55 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
56
57 Label done, notBool, notByte, notChar;
58
59 // common case first
60 cmpw(rscratch1, T_INT);
61 br(Assembler::EQ, done);
62
63 // mask integer result to narrower return type.
64 cmpw(rscratch1, T_BOOLEAN);
65 br(Assembler::NE, notBool);
66 andw(result, result, 0x1);
67 b(done);
68
69 bind(notBool);
70 cmpw(rscratch1, T_BYTE);
71 br(Assembler::NE, notByte);
72 sbfx(result, result, 0, 8);
73 b(done);
74
75 bind(notByte);
76 cmpw(rscratch1, T_CHAR);
77 br(Assembler::NE, notChar);
78 ubfx(result, result, 0, 16); // truncate upper 16 bits
79 b(done);
80
81 bind(notChar);
82 sbfx(result, result, 0, 16); // sign-extend short
83
84 // Nothing to do for T_INT
85 bind(done);
86 }
87
88 void InterpreterMacroAssembler::jump_to_entry(address entry) {
89 assert(entry, "Entry must have been generated by now");
90 b(entry);
91 }
92
93 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
94 if (JvmtiExport::can_pop_frame()) {
95 Label L;
96 // Initiate popframe handling only if it is not already being
97 // processed. If the flag has the popframe_processing bit set, it
98 // means that this code is called *during* popframe handling - we
99 // don't want to reenter.
100 // This method is only called just after the call into the vm in
101 // call_VM_base, so the arg registers are available.
102 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
103 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
104 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
105 // Call Interpreter::remove_activation_preserving_args_entry() to get the
106 // address of the same-named entrypoint in the generated interpreter code.
107 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
108 br(r0);
109 bind(L);
110 }
111 }
112
113
114 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
115 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
116 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
117 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
118 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
119 switch (state) {
120 case atos: ldr(r0, oop_addr);
121 str(zr, oop_addr);
122 interp_verify_oop(r0, state); break;
123 case ltos: ldr(r0, val_addr); break;
124 case btos: // fall through
125 case ztos: // fall through
126 case ctos: // fall through
127 case stos: // fall through
128 case itos: ldrw(r0, val_addr); break;
129 case ftos: ldrs(v0, val_addr); break;
130 case dtos: ldrd(v0, val_addr); break;
131 case vtos: /* nothing to do */ break;
132 default : ShouldNotReachHere();
133 }
134 // Clean up tos value in the thread object
135 movw(rscratch1, (int) ilgl);
136 strw(rscratch1, tos_addr);
137 strw(zr, val_addr);
138 }
139
140
141 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
142 if (JvmtiExport::can_force_early_return()) {
143 Label L;
144 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
145 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
146
147 // Initiate earlyret handling only if it is not already being processed.
148 // If the flag has the earlyret_processing bit set, it means that this code
149 // is called *during* earlyret handling - we don't want to reenter.
150 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
151 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
152 br(Assembler::NE, L);
153
154 // Call Interpreter::remove_activation_early_entry() to get the address of the
155 // same-named entrypoint in the generated interpreter code.
156 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
157 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
158 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
159 br(r0);
160 bind(L);
161 }
162 }
163
164 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
165 Register reg,
166 int bcp_offset) {
167 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
168 ldrh(reg, Address(rbcp, bcp_offset));
169 rev16(reg, reg);
170 }
171
172 void InterpreterMacroAssembler::get_dispatch() {
173 uint64_t offset;
174 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
175 // Use add() here after ARDP, rather than lea().
176 // lea() does not generate anything if its offset is zero.
177 // However, relocs expect to find either an ADD or a load/store
178 // insn after an ADRP. add() always generates an ADD insn, even
179 // for add(Rn, Rn, 0).
180 add(rdispatch, rdispatch, offset);
181 }
182
183 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
184 int bcp_offset,
185 size_t index_size) {
186 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
187 if (index_size == sizeof(u2)) {
188 load_unsigned_short(index, Address(rbcp, bcp_offset));
189 } else if (index_size == sizeof(u4)) {
190 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
191 ldrw(index, Address(rbcp, bcp_offset));
192 } else if (index_size == sizeof(u1)) {
193 load_unsigned_byte(index, Address(rbcp, bcp_offset));
194 } else {
195 ShouldNotReachHere();
196 }
197 }
198
199 void InterpreterMacroAssembler::get_method_counters(Register method,
200 Register mcs, Label& skip) {
201 Label has_counters;
202 ldr(mcs, Address(method, Method::method_counters_offset()));
203 cbnz(mcs, has_counters);
204 call_VM(noreg, CAST_FROM_FN_PTR(address,
205 InterpreterRuntime::build_method_counters), method);
206 ldr(mcs, Address(method, Method::method_counters_offset()));
207 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
208 bind(has_counters);
209 }
210
211 // Load object from cpool->resolved_references(index)
212 void InterpreterMacroAssembler::load_resolved_reference_at_index(
213 Register result, Register index, Register tmp) {
214 assert_different_registers(result, index);
215
216 get_constant_pool(result);
217 // load pointer for resolved_references[] objArray
218 ldr(result, Address(result, ConstantPool::cache_offset()));
219 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
220 resolve_oop_handle(result, tmp, rscratch2);
221 // Add in the index
222 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
223 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
224 }
225
226 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
227 Register cpool, Register index, Register klass, Register temp) {
228 add(temp, cpool, index, LSL, LogBytesPerWord);
229 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
230 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
231 add(klass, klass, temp, LSL, LogBytesPerWord);
232 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
233 }
234
235 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
236 // subtype of super_klass.
237 //
238 // Args:
239 // r0: superklass
240 // Rsub_klass: subklass
241 //
242 // Kills:
243 // r2, r5
244 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
245 Label& ok_is_subtype) {
246 assert(Rsub_klass != r0, "r0 holds superklass");
247 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
248 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
249
250 // Profile the not-null value's klass.
251 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
252
253 // Do the check.
254 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
255 }
256
257 // Java Expression Stack
258
259 void InterpreterMacroAssembler::pop_ptr(Register r) {
260 ldr(r, post(esp, wordSize));
261 }
262
263 void InterpreterMacroAssembler::pop_i(Register r) {
264 ldrw(r, post(esp, wordSize));
265 }
266
267 void InterpreterMacroAssembler::pop_l(Register r) {
268 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
269 }
270
271 void InterpreterMacroAssembler::push_ptr(Register r) {
272 str(r, pre(esp, -wordSize));
273 }
274
275 void InterpreterMacroAssembler::push_i(Register r) {
276 str(r, pre(esp, -wordSize));
277 }
278
279 void InterpreterMacroAssembler::push_l(Register r) {
280 str(zr, pre(esp, -wordSize));
281 str(r, pre(esp, - wordSize));
282 }
283
284 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
285 ldrs(r, post(esp, wordSize));
286 }
287
288 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
289 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
290 }
291
292 void InterpreterMacroAssembler::push_f(FloatRegister r) {
293 strs(r, pre(esp, -wordSize));
294 }
295
296 void InterpreterMacroAssembler::push_d(FloatRegister r) {
297 strd(r, pre(esp, 2* -wordSize));
298 }
299
300 void InterpreterMacroAssembler::pop(TosState state) {
301 switch (state) {
302 case atos: pop_ptr(); break;
303 case btos:
304 case ztos:
305 case ctos:
306 case stos:
307 case itos: pop_i(); break;
308 case ltos: pop_l(); break;
309 case ftos: pop_f(); break;
310 case dtos: pop_d(); break;
311 case vtos: /* nothing to do */ break;
312 default: ShouldNotReachHere();
313 }
314 interp_verify_oop(r0, state);
315 }
316
317 void InterpreterMacroAssembler::push(TosState state) {
318 interp_verify_oop(r0, state);
319 switch (state) {
320 case atos: push_ptr(); break;
321 case btos:
322 case ztos:
323 case ctos:
324 case stos:
325 case itos: push_i(); break;
326 case ltos: push_l(); break;
327 case ftos: push_f(); break;
328 case dtos: push_d(); break;
329 case vtos: /* nothing to do */ break;
330 default : ShouldNotReachHere();
331 }
332 }
333
334 // Helpers for swap and dup
335 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
336 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
337 }
338
339 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
340 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
341 }
342
343 void InterpreterMacroAssembler::load_float(Address src) {
344 ldrs(v0, src);
345 }
346
347 void InterpreterMacroAssembler::load_double(Address src) {
348 ldrd(v0, src);
349 }
350
351 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
352 // set sender sp
353 mov(r19_sender_sp, sp);
354 // record last_sp
355 sub(rscratch1, esp, rfp);
356 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
357 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
358 }
359
360 // Jump to from_interpreted entry of a call unless single stepping is possible
361 // in this thread in which case we must call the i2i entry
362 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
363 prepare_to_jump_from_interpreted();
364
365 if (JvmtiExport::can_post_interpreter_events()) {
366 Label run_compiled_code;
367 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
368 // compiled code in threads for which the event is enabled. Check here for
369 // interp_only_mode if these events CAN be enabled.
370 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
371 cbzw(rscratch1, run_compiled_code);
372 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
373 br(rscratch1);
374 bind(run_compiled_code);
375 }
376
377 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
378 br(rscratch1);
379 }
380
381 // The following two routines provide a hook so that an implementation
382 // can schedule the dispatch in two parts. amd64 does not do this.
383 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
384 }
385
386 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
387 dispatch_next(state, step);
388 }
389
390 void InterpreterMacroAssembler::dispatch_base(TosState state,
391 address* table,
392 bool verifyoop,
393 bool generate_poll) {
394 if (VerifyActivationFrameSize) {
395 Label L;
396 sub(rscratch2, rfp, esp);
397 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
398 subs(rscratch2, rscratch2, min_frame_size);
399 br(Assembler::GE, L);
400 stop("broken stack frame");
401 bind(L);
402 }
403 if (verifyoop) {
404 interp_verify_oop(r0, state);
405 }
406
407 Label safepoint;
408 address* const safepoint_table = Interpreter::safept_table(state);
409 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
410
411 if (needs_thread_local_poll) {
412 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
413 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
414 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
415 }
416
417 if (table == Interpreter::dispatch_table(state)) {
418 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
419 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
420 } else {
421 mov(rscratch2, (address)table);
422 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
423 }
424 br(rscratch2);
425
426 if (needs_thread_local_poll) {
427 bind(safepoint);
428 lea(rscratch2, ExternalAddress((address)safepoint_table));
429 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
430 br(rscratch2);
431 }
432 }
433
434 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
435 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
436 }
437
438 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
439 dispatch_base(state, Interpreter::normal_table(state));
440 }
441
442 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
443 dispatch_base(state, Interpreter::normal_table(state), false);
444 }
445
446
447 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
448 // load next bytecode
449 ldrb(rscratch1, Address(pre(rbcp, step)));
450 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
451 }
452
453 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
454 // load current bytecode
455 ldrb(rscratch1, Address(rbcp, 0));
456 dispatch_base(state, table);
457 }
458
459 // remove activation
460 //
461 // Unlock the receiver if this is a synchronized method.
462 // Unlock any Java monitors from synchronized blocks.
463 // Apply stack watermark barrier.
464 // Notify JVMTI.
465 // Remove the activation from the stack.
466 //
467 // If there are locked Java monitors
468 // If throw_monitor_exception
469 // throws IllegalMonitorStateException
470 // Else if install_monitor_exception
471 // installs IllegalMonitorStateException
472 // Else
473 // no error processing
474 void InterpreterMacroAssembler::remove_activation(TosState state,
475 bool throw_monitor_exception,
476 bool install_monitor_exception,
477 bool notify_jvmdi) {
478 // Note: Registers r3 xmm0 may be in use for the
479 // result check if synchronized method
480 Label unlocked, unlock, no_unlock;
481
482 // get the value of _do_not_unlock_if_synchronized into r3
483 const Address do_not_unlock_if_synchronized(rthread,
484 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
485 ldrb(r3, do_not_unlock_if_synchronized);
486 strb(zr, do_not_unlock_if_synchronized); // reset the flag
487
488 // get method access flags
489 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
490 ldrh(r2, Address(r1, Method::access_flags_offset()));
491 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
492
493 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
494 // is set.
495 cbnz(r3, no_unlock);
496
497 // unlock monitor
498 push(state); // save result
499
500 // BasicObjectLock will be first in list, since this is a
501 // synchronized method. However, need to check that the object has
502 // not been unlocked by an explicit monitorexit bytecode.
503 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
504 wordSize - (int) sizeof(BasicObjectLock));
505 // We use c_rarg1 so that if we go slow path it will be the correct
506 // register for unlock_object to pass to VM directly
507 lea(c_rarg1, monitor); // address of first monitor
508
509 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
510 cbnz(r0, unlock);
511
512 pop(state);
513 if (throw_monitor_exception) {
514 // Entry already unlocked, need to throw exception
515 call_VM(noreg, CAST_FROM_FN_PTR(address,
516 InterpreterRuntime::throw_illegal_monitor_state_exception));
517 should_not_reach_here();
518 } else {
519 // Monitor already unlocked during a stack unroll. If requested,
520 // install an illegal_monitor_state_exception. Continue with
521 // stack unrolling.
522 if (install_monitor_exception) {
523 call_VM(noreg, CAST_FROM_FN_PTR(address,
524 InterpreterRuntime::new_illegal_monitor_state_exception));
525 }
526 b(unlocked);
527 }
528
529 bind(unlock);
530 unlock_object(c_rarg1);
531 pop(state);
532
533 // Check that for block-structured locking (i.e., that all locked
534 // objects has been unlocked)
535 bind(unlocked);
536
537 // r0: Might contain return value
538
539 // Check that all monitors are unlocked
540 {
541 Label loop, exception, entry, restart;
542 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
543 const Address monitor_block_top(
544 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
545 const Address monitor_block_bot(
546 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
547
548 bind(restart);
549 // We use c_rarg1 so that if we go slow path it will be the correct
550 // register for unlock_object to pass to VM directly
551 ldr(c_rarg1, monitor_block_top); // derelativize pointer
552 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
553 // c_rarg1 points to current entry, starting with top-most entry
554
555 lea(r19, monitor_block_bot); // points to word before bottom of
556 // monitor block
557 b(entry);
558
559 // Entry already locked, need to throw exception
560 bind(exception);
561
562 if (throw_monitor_exception) {
563 // Throw exception
564 MacroAssembler::call_VM(noreg,
565 CAST_FROM_FN_PTR(address, InterpreterRuntime::
566 throw_illegal_monitor_state_exception));
567 should_not_reach_here();
568 } else {
569 // Stack unrolling. Unlock object and install illegal_monitor_exception.
570 // Unlock does not block, so don't have to worry about the frame.
571 // We don't have to preserve c_rarg1 since we are going to throw an exception.
572
573 push(state);
574 unlock_object(c_rarg1);
575 pop(state);
576
577 if (install_monitor_exception) {
578 call_VM(noreg, CAST_FROM_FN_PTR(address,
579 InterpreterRuntime::
580 new_illegal_monitor_state_exception));
581 }
582
583 b(restart);
584 }
585
586 bind(loop);
587 // check if current entry is used
588 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
589 cbnz(rscratch1, exception);
590
591 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
592 bind(entry);
593 cmp(c_rarg1, r19); // check if bottom reached
594 br(Assembler::NE, loop); // if not at bottom then check this entry
595 }
596
597 bind(no_unlock);
598
599 JFR_ONLY(enter_jfr_critical_section();)
600
601 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
602 // that would normally not be safe to use. Such bad returns into unsafe territory of
603 // the stack, will call InterpreterRuntime::at_unwind.
604 Label slow_path;
605 Label fast_path;
606 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
607 br(Assembler::AL, fast_path);
608 bind(slow_path);
609 push(state);
610 set_last_Java_frame(esp, rfp, pc(), rscratch1);
611 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
612 reset_last_Java_frame(true);
613 pop(state);
614 bind(fast_path);
615
616 // JVMTI support. Make sure the safepoint poll test is issued prior.
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 JFR_ONLY(leave_jfr_critical_section();)
644
645 call_VM_leaf(
646 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
647 call_VM(noreg, CAST_FROM_FN_PTR(address,
648 InterpreterRuntime::throw_delayed_StackOverflowError));
649 should_not_reach_here();
650
651 bind(no_reserved_zone_enabling);
652 }
653
654 // remove frame anchor
655 leave();
656
657 JFR_ONLY(leave_jfr_critical_section();)
658
659 // restore sender esp
660 mov(esp, rscratch2);
661
662 // If we're returning to interpreted code we will shortly be
663 // adjusting SP to allow some space for ESP. If we're returning to
664 // compiled code the saved sender SP was saved in sender_sp, so this
665 // restores it.
666 andr(sp, esp, -16);
667 }
668
669 #if INCLUDE_JFR
670 void InterpreterMacroAssembler::enter_jfr_critical_section() {
671 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
672 mov(rscratch1, true);
673 strb(rscratch1, sampling_critical_section);
674 }
675
676 void InterpreterMacroAssembler::leave_jfr_critical_section() {
677 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
678 strb(zr, sampling_critical_section);
679 }
680 #endif // INCLUDE_JFR
681
682 // Lock object
683 //
684 // Args:
685 // c_rarg1: BasicObjectLock to be used for locking
686 //
687 // Kills:
688 // r0
689 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
690 // rscratch1, rscratch2 (scratch regs)
691 void InterpreterMacroAssembler::lock_object(Register lock_reg)
692 {
693 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
694 if (LockingMode == LM_MONITOR) {
695 call_VM_preemptable(noreg,
696 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
697 lock_reg);
698 } else {
699 Label count, done;
700
701 const Register swap_reg = r0;
702 const Register tmp = c_rarg2;
703 const Register obj_reg = c_rarg3; // Will contain the oop
704 const Register tmp2 = c_rarg4;
705 const Register tmp3 = c_rarg5;
706
707 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
708 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
709 const int mark_offset = lock_offset +
710 BasicLock::displaced_header_offset_in_bytes();
711
712 Label slow_case;
713
714 // Load object pointer into obj_reg %c_rarg3
715 ldr(obj_reg, Address(lock_reg, obj_offset));
716
717 if (LockingMode == LM_LIGHTWEIGHT) {
718 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
719 b(done);
720 } else if (LockingMode == LM_LEGACY) {
721
722 if (DiagnoseSyncOnValueBasedClasses != 0) {
723 load_klass(tmp, obj_reg);
724 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
725 tst(tmp, KlassFlags::_misc_is_value_based_class);
726 br(Assembler::NE, slow_case);
727 }
728
729 // Load (object->mark() | 1) into swap_reg
730 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
731 orr(swap_reg, rscratch1, 1);
732
733 // Save (object->mark() | 1) into BasicLock's displaced header
734 str(swap_reg, Address(lock_reg, mark_offset));
735
736 assert(lock_offset == 0,
737 "displached header must be first word in BasicObjectLock");
738
739 Label fail;
740 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
741
742 // Fast check for recursive lock.
743 //
744 // Can apply the optimization only if this is a stack lock
745 // allocated in this thread. For efficiency, we can focus on
746 // recently allocated stack locks (instead of reading the stack
747 // base and checking whether 'mark' points inside the current
748 // thread stack):
749 // 1) (mark & 7) == 0, and
750 // 2) sp <= mark < mark + os::pagesize()
751 //
752 // Warning: sp + os::pagesize can overflow the stack base. We must
753 // neither apply the optimization for an inflated lock allocated
754 // just above the thread stack (this is why condition 1 matters)
755 // nor apply the optimization if the stack lock is inside the stack
756 // of another thread. The latter is avoided even in case of overflow
757 // because we have guard pages at the end of all stacks. Hence, if
758 // we go over the stack base and hit the stack of another thread,
759 // this should not be in a writeable area that could contain a
760 // stack lock allocated by that thread. As a consequence, a stack
761 // lock less than page size away from sp is guaranteed to be
762 // owned by the current thread.
763 //
764 // These 3 tests can be done by evaluating the following
765 // expression: ((mark - sp) & (7 - os::vm_page_size())),
766 // assuming both stack pointer and pagesize have their
767 // least significant 3 bits clear.
768 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
769 // NOTE2: aarch64 does not like to subtract sp from rn so take a
770 // copy
771 mov(rscratch1, sp);
772 sub(swap_reg, swap_reg, rscratch1);
773 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
774
775 // Save the test result, for recursive case, the result is zero
776 str(swap_reg, Address(lock_reg, mark_offset));
777 br(Assembler::NE, slow_case);
778
779 bind(count);
780 inc_held_monitor_count(rscratch1);
781 b(done);
782 }
783 bind(slow_case);
784
785 // Call the runtime routine for slow case
786 call_VM_preemptable(noreg,
787 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
788 lock_reg);
789
790 bind(done);
791 }
792 }
793
794
795 // Unlocks an object. Used in monitorexit bytecode and
796 // remove_activation. Throws an IllegalMonitorException if object is
797 // not locked by current thread.
798 //
799 // Args:
800 // c_rarg1: BasicObjectLock for lock
801 //
802 // Kills:
803 // r0
804 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
805 // rscratch1, rscratch2 (scratch regs)
806 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
807 {
808 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
809
810 if (LockingMode == LM_MONITOR) {
811 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
812 } else {
813 Label count, done;
814
815 const Register swap_reg = r0;
816 const Register header_reg = c_rarg2; // Will contain the old oopMark
817 const Register obj_reg = c_rarg3; // Will contain the oop
818 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
819
820 save_bcp(); // Save in case of exception
821
822 if (LockingMode != LM_LIGHTWEIGHT) {
823 // Convert from BasicObjectLock structure to object and BasicLock
824 // structure Store the BasicLock address into %r0
825 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
826 }
827
828 // Load oop into obj_reg(%c_rarg3)
829 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
830
831 // Free entry
832 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
833
834 Label slow_case;
835 if (LockingMode == LM_LIGHTWEIGHT) {
836 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
837 b(done);
838 } else if (LockingMode == LM_LEGACY) {
839 // Load the old header from BasicLock structure
840 ldr(header_reg, Address(swap_reg,
841 BasicLock::displaced_header_offset_in_bytes()));
842
843 // Test for recursion
844 cbz(header_reg, count);
845
846 // Atomic swap back the old header
847 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
848
849 bind(count);
850 dec_held_monitor_count(rscratch1);
851 b(done);
852 }
853
854 bind(slow_case);
855 // Call the runtime routine for slow case.
856 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
857 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
858 bind(done);
859 restore_bcp();
860 }
861 }
862
863 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
864 Label& zero_continue) {
865 assert(ProfileInterpreter, "must be profiling interpreter");
866 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
867 cbz(mdp, zero_continue);
868 }
869
870 // Set the method data pointer for the current bcp.
871 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
872 assert(ProfileInterpreter, "must be profiling interpreter");
873 Label set_mdp;
874 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
875
876 // Test MDO to avoid the call if it is null.
877 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
878 cbz(r0, set_mdp);
879 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
880 // r0: mdi
881 // mdo is guaranteed to be non-zero here, we checked for it before the call.
882 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
883 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
884 add(r0, r1, r0);
885 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
886 bind(set_mdp);
887 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
888 }
889
890 void InterpreterMacroAssembler::verify_method_data_pointer() {
891 assert(ProfileInterpreter, "must be profiling interpreter");
892 #ifdef ASSERT
893 Label verify_continue;
894 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
895 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
896 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
897 get_method(r1);
898
899 // If the mdp is valid, it will point to a DataLayout header which is
900 // consistent with the bcp. The converse is highly probable also.
901 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
902 ldr(rscratch1, Address(r1, Method::const_offset()));
903 add(r2, r2, rscratch1, Assembler::LSL);
904 lea(r2, Address(r2, ConstMethod::codes_offset()));
905 cmp(r2, rbcp);
906 br(Assembler::EQ, verify_continue);
907 // r1: method
908 // rbcp: bcp // rbcp == 22
909 // r3: mdp
910 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
911 r1, rbcp, r3);
912 bind(verify_continue);
913 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
914 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
915 #endif // ASSERT
916 }
917
918
919 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
920 int constant,
921 Register value) {
922 assert(ProfileInterpreter, "must be profiling interpreter");
923 Address data(mdp_in, constant);
924 str(value, data);
925 }
926
927
928 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
929 int constant,
930 bool decrement) {
931 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
932 }
933
934 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
935 Register reg,
936 int constant,
937 bool decrement) {
938 assert(ProfileInterpreter, "must be profiling interpreter");
939 // %%% this does 64bit counters at best it is wasting space
940 // at worst it is a rare bug when counters overflow
941
942 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
943
944 Address addr1(mdp_in, constant);
945 Address addr2(rscratch2, reg, Address::lsl(0));
946 Address &addr = addr1;
947 if (reg != noreg) {
948 lea(rscratch2, addr1);
949 addr = addr2;
950 }
951
952 if (decrement) {
953 // Decrement the register. Set condition codes.
954 // Intel does this
955 // addptr(data, (int32_t) -DataLayout::counter_increment);
956 // If the decrement causes the counter to overflow, stay negative
957 // Label L;
958 // jcc(Assembler::negative, L);
959 // addptr(data, (int32_t) DataLayout::counter_increment);
960 // so we do this
961 ldr(rscratch1, addr);
962 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
963 Label L;
964 br(Assembler::LO, L); // skip store if counter underflow
965 str(rscratch1, addr);
966 bind(L);
967 } else {
968 assert(DataLayout::counter_increment == 1,
969 "flow-free idiom only works with 1");
970 // Intel does this
971 // Increment the register. Set carry flag.
972 // addptr(data, DataLayout::counter_increment);
973 // If the increment causes the counter to overflow, pull back by 1.
974 // sbbptr(data, (int32_t)0);
975 // so we do this
976 ldr(rscratch1, addr);
977 adds(rscratch1, rscratch1, DataLayout::counter_increment);
978 Label L;
979 br(Assembler::CS, L); // skip store if counter overflow
980 str(rscratch1, addr);
981 bind(L);
982 }
983 }
984
985 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
986 int flag_byte_constant) {
987 assert(ProfileInterpreter, "must be profiling interpreter");
988 int flags_offset = in_bytes(DataLayout::flags_offset());
989 // Set the flag
990 ldrb(rscratch1, Address(mdp_in, flags_offset));
991 orr(rscratch1, rscratch1, flag_byte_constant);
992 strb(rscratch1, Address(mdp_in, flags_offset));
993 }
994
995
996 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
997 int offset,
998 Register value,
999 Register test_value_out,
1000 Label& not_equal_continue) {
1001 assert(ProfileInterpreter, "must be profiling interpreter");
1002 if (test_value_out == noreg) {
1003 ldr(rscratch1, Address(mdp_in, offset));
1004 cmp(value, rscratch1);
1005 } else {
1006 // Put the test value into a register, so caller can use it:
1007 ldr(test_value_out, Address(mdp_in, offset));
1008 cmp(value, test_value_out);
1009 }
1010 br(Assembler::NE, not_equal_continue);
1011 }
1012
1013
1014 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1015 int offset_of_disp) {
1016 assert(ProfileInterpreter, "must be profiling interpreter");
1017 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1018 add(mdp_in, mdp_in, rscratch1, LSL);
1019 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1020 }
1021
1022
1023 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1024 Register reg,
1025 int offset_of_disp) {
1026 assert(ProfileInterpreter, "must be profiling interpreter");
1027 lea(rscratch1, Address(mdp_in, offset_of_disp));
1028 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1029 add(mdp_in, mdp_in, rscratch1, LSL);
1030 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1031 }
1032
1033
1034 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1035 int constant) {
1036 assert(ProfileInterpreter, "must be profiling interpreter");
1037 add(mdp_in, mdp_in, (unsigned)constant);
1038 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1039 }
1040
1041
1042 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1043 assert(ProfileInterpreter, "must be profiling interpreter");
1044 // save/restore across call_VM
1045 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1046 call_VM(noreg,
1047 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1048 return_bci);
1049 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1050 }
1051
1052
1053 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1054 Register bumped_count) {
1055 if (ProfileInterpreter) {
1056 Label profile_continue;
1057
1058 // If no method data exists, go to profile_continue.
1059 // Otherwise, assign to mdp
1060 test_method_data_pointer(mdp, profile_continue);
1061
1062 // We are taking a branch. Increment the taken count.
1063 // We inline increment_mdp_data_at to return bumped_count in a register
1064 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1065 Address data(mdp, in_bytes(JumpData::taken_offset()));
1066 ldr(bumped_count, data);
1067 assert(DataLayout::counter_increment == 1,
1068 "flow-free idiom only works with 1");
1069 // Intel does this to catch overflow
1070 // addptr(bumped_count, DataLayout::counter_increment);
1071 // sbbptr(bumped_count, 0);
1072 // so we do this
1073 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1074 Label L;
1075 br(Assembler::CS, L); // skip store if counter overflow
1076 str(bumped_count, data);
1077 bind(L);
1078 // The method data pointer needs to be updated to reflect the new target.
1079 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1080 bind(profile_continue);
1081 }
1082 }
1083
1084
1085 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1086 if (ProfileInterpreter) {
1087 Label profile_continue;
1088
1089 // If no method data exists, go to profile_continue.
1090 test_method_data_pointer(mdp, profile_continue);
1091
1092 // We are taking a branch. Increment the not taken count.
1093 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1094
1095 // The method data pointer needs to be updated to correspond to
1096 // the next bytecode
1097 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1098 bind(profile_continue);
1099 }
1100 }
1101
1102
1103 void InterpreterMacroAssembler::profile_call(Register mdp) {
1104 if (ProfileInterpreter) {
1105 Label profile_continue;
1106
1107 // If no method data exists, go to profile_continue.
1108 test_method_data_pointer(mdp, profile_continue);
1109
1110 // We are making a call. Increment the count.
1111 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1112
1113 // The method data pointer needs to be updated to reflect the new target.
1114 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1115 bind(profile_continue);
1116 }
1117 }
1118
1119 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1120 if (ProfileInterpreter) {
1121 Label profile_continue;
1122
1123 // If no method data exists, go to profile_continue.
1124 test_method_data_pointer(mdp, profile_continue);
1125
1126 // We are making a call. Increment the count.
1127 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1128
1129 // The method data pointer needs to be updated to reflect the new target.
1130 update_mdp_by_constant(mdp,
1131 in_bytes(VirtualCallData::
1132 virtual_call_data_size()));
1133 bind(profile_continue);
1134 }
1135 }
1136
1137
1138 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1139 Register mdp,
1140 Register reg2,
1141 bool receiver_can_be_null) {
1142 if (ProfileInterpreter) {
1143 Label profile_continue;
1144
1145 // If no method data exists, go to profile_continue.
1146 test_method_data_pointer(mdp, profile_continue);
1147
1148 Label skip_receiver_profile;
1149 if (receiver_can_be_null) {
1150 Label not_null;
1151 // We are making a call. Increment the count for null receiver.
1152 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1153 b(skip_receiver_profile);
1154 bind(not_null);
1155 }
1156
1157 // Record the receiver type.
1158 record_klass_in_profile(receiver, mdp, reg2);
1159 bind(skip_receiver_profile);
1160
1161 // The method data pointer needs to be updated to reflect the new target.
1162 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1163 bind(profile_continue);
1164 }
1165 }
1166
1167 // This routine creates a state machine for updating the multi-row
1168 // type profile at a virtual call site (or other type-sensitive bytecode).
1169 // The machine visits each row (of receiver/count) until the receiver type
1170 // is found, or until it runs out of rows. At the same time, it remembers
1171 // the location of the first empty row. (An empty row records null for its
1172 // receiver, and can be allocated for a newly-observed receiver type.)
1173 // Because there are two degrees of freedom in the state, a simple linear
1174 // search will not work; it must be a decision tree. Hence this helper
1175 // function is recursive, to generate the required tree structured code.
1176 // It's the interpreter, so we are trading off code space for speed.
1177 // See below for example code.
1178 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1179 Register receiver, Register mdp,
1180 Register reg2, int start_row,
1181 Label& done) {
1182 if (TypeProfileWidth == 0) {
1183 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1184 } else {
1185 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1186 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1187 }
1188 }
1189
1190 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1191 Register reg2, int start_row, Label& done, int total_rows,
1192 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1193 int last_row = total_rows - 1;
1194 assert(start_row <= last_row, "must be work left to do");
1195 // Test this row for both the item and for null.
1196 // Take any of three different outcomes:
1197 // 1. found item => increment count and goto done
1198 // 2. found null => keep looking for case 1, maybe allocate this cell
1199 // 3. found something else => keep looking for cases 1 and 2
1200 // Case 3 is handled by a recursive call.
1201 for (int row = start_row; row <= last_row; row++) {
1202 Label next_test;
1203 bool test_for_null_also = (row == start_row);
1204
1205 // See if the item is item[n].
1206 int item_offset = in_bytes(item_offset_fn(row));
1207 test_mdp_data_at(mdp, item_offset, item,
1208 (test_for_null_also ? reg2 : noreg),
1209 next_test);
1210 // (Reg2 now contains the item from the CallData.)
1211
1212 // The item is item[n]. Increment count[n].
1213 int count_offset = in_bytes(item_count_offset_fn(row));
1214 increment_mdp_data_at(mdp, count_offset);
1215 b(done);
1216 bind(next_test);
1217
1218 if (test_for_null_also) {
1219 Label found_null;
1220 // Failed the equality check on item[n]... Test for null.
1221 if (start_row == last_row) {
1222 // The only thing left to do is handle the null case.
1223 cbz(reg2, found_null);
1224 // Item did not match any saved item and there is no empty row for it.
1225 // Increment total counter to indicate polymorphic case.
1226 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1227 b(done);
1228 bind(found_null);
1229 break;
1230 }
1231 // Since null is rare, make it be the branch-taken case.
1232 cbz(reg2, found_null);
1233
1234 // Put all the "Case 3" tests here.
1235 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1236 item_offset_fn, item_count_offset_fn);
1237
1238 // Found a null. Keep searching for a matching item,
1239 // but remember that this is an empty (unused) slot.
1240 bind(found_null);
1241 }
1242 }
1243
1244 // In the fall-through case, we found no matching item, but we
1245 // observed the item[start_row] is null.
1246
1247 // Fill in the item field and increment the count.
1248 int item_offset = in_bytes(item_offset_fn(start_row));
1249 set_mdp_data_at(mdp, item_offset, item);
1250 int count_offset = in_bytes(item_count_offset_fn(start_row));
1251 mov(reg2, DataLayout::counter_increment);
1252 set_mdp_data_at(mdp, count_offset, reg2);
1253 if (start_row > 0) {
1254 b(done);
1255 }
1256 }
1257
1258 // Example state machine code for three profile rows:
1259 // // main copy of decision tree, rooted at row[1]
1260 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1261 // if (row[0].rec != nullptr) {
1262 // // inner copy of decision tree, rooted at row[1]
1263 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1264 // if (row[1].rec != nullptr) {
1265 // // degenerate decision tree, rooted at row[2]
1266 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1267 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1268 // row[2].init(rec); goto done;
1269 // } else {
1270 // // remember row[1] is empty
1271 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1272 // row[1].init(rec); goto done;
1273 // }
1274 // } else {
1275 // // remember row[0] is empty
1276 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1277 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1278 // row[0].init(rec); goto done;
1279 // }
1280 // done:
1281
1282 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1283 Register mdp, Register reg2) {
1284 assert(ProfileInterpreter, "must be profiling");
1285 Label done;
1286
1287 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1288
1289 bind (done);
1290 }
1291
1292 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1293 Register mdp) {
1294 if (ProfileInterpreter) {
1295 Label profile_continue;
1296 uint row;
1297
1298 // If no method data exists, go to profile_continue.
1299 test_method_data_pointer(mdp, profile_continue);
1300
1301 // Update the total ret count.
1302 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1303
1304 for (row = 0; row < RetData::row_limit(); row++) {
1305 Label next_test;
1306
1307 // See if return_bci is equal to bci[n]:
1308 test_mdp_data_at(mdp,
1309 in_bytes(RetData::bci_offset(row)),
1310 return_bci, noreg,
1311 next_test);
1312
1313 // return_bci is equal to bci[n]. Increment the count.
1314 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1315
1316 // The method data pointer needs to be updated to reflect the new target.
1317 update_mdp_by_offset(mdp,
1318 in_bytes(RetData::bci_displacement_offset(row)));
1319 b(profile_continue);
1320 bind(next_test);
1321 }
1322
1323 update_mdp_for_ret(return_bci);
1324
1325 bind(profile_continue);
1326 }
1327 }
1328
1329 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
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 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1337
1338 // The method data pointer needs to be updated.
1339 int mdp_delta = in_bytes(BitData::bit_data_size());
1340 if (TypeProfileCasts) {
1341 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1342 }
1343 update_mdp_by_constant(mdp, mdp_delta);
1344
1345 bind(profile_continue);
1346 }
1347 }
1348
1349 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1350 if (ProfileInterpreter) {
1351 Label profile_continue;
1352
1353 // If no method data exists, go to profile_continue.
1354 test_method_data_pointer(mdp, profile_continue);
1355
1356 // The method data pointer needs to be updated.
1357 int mdp_delta = in_bytes(BitData::bit_data_size());
1358 if (TypeProfileCasts) {
1359 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1360
1361 // Record the object type.
1362 record_klass_in_profile(klass, mdp, reg2);
1363 }
1364 update_mdp_by_constant(mdp, mdp_delta);
1365
1366 bind(profile_continue);
1367 }
1368 }
1369
1370 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1371 if (ProfileInterpreter) {
1372 Label profile_continue;
1373
1374 // If no method data exists, go to profile_continue.
1375 test_method_data_pointer(mdp, profile_continue);
1376
1377 // Update the default case count
1378 increment_mdp_data_at(mdp,
1379 in_bytes(MultiBranchData::default_count_offset()));
1380
1381 // The method data pointer needs to be updated.
1382 update_mdp_by_offset(mdp,
1383 in_bytes(MultiBranchData::
1384 default_displacement_offset()));
1385
1386 bind(profile_continue);
1387 }
1388 }
1389
1390 void InterpreterMacroAssembler::profile_switch_case(Register index,
1391 Register mdp,
1392 Register reg2) {
1393 if (ProfileInterpreter) {
1394 Label profile_continue;
1395
1396 // If no method data exists, go to profile_continue.
1397 test_method_data_pointer(mdp, profile_continue);
1398
1399 // Build the base (index * per_case_size_in_bytes()) +
1400 // case_array_offset_in_bytes()
1401 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1402 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1403 Assembler::maddw(index, index, reg2, rscratch1);
1404
1405 // Update the case count
1406 increment_mdp_data_at(mdp,
1407 index,
1408 in_bytes(MultiBranchData::relative_count_offset()));
1409
1410 // The method data pointer needs to be updated.
1411 update_mdp_by_offset(mdp,
1412 index,
1413 in_bytes(MultiBranchData::
1414 relative_displacement_offset()));
1415
1416 bind(profile_continue);
1417 }
1418 }
1419
1420 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1421 if (state == atos) {
1422 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1423 }
1424 }
1425
1426 void InterpreterMacroAssembler::notify_method_entry() {
1427 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1428 // track stack depth. If it is possible to enter interp_only_mode we add
1429 // the code to check if the event should be sent.
1430 if (JvmtiExport::can_post_interpreter_events()) {
1431 Label L;
1432 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1433 cbzw(r3, L);
1434 call_VM(noreg, CAST_FROM_FN_PTR(address,
1435 InterpreterRuntime::post_method_entry));
1436 bind(L);
1437 }
1438
1439 if (DTraceMethodProbes) {
1440 get_method(c_rarg1);
1441 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1442 rthread, c_rarg1);
1443 }
1444
1445 // RedefineClasses() tracing support for obsolete method entry
1446 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1447 get_method(c_rarg1);
1448 call_VM_leaf(
1449 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1450 rthread, c_rarg1);
1451 }
1452
1453 }
1454
1455
1456 void InterpreterMacroAssembler::notify_method_exit(
1457 TosState state, NotifyMethodExitMode mode) {
1458 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1459 // track stack depth. If it is possible to enter interp_only_mode we add
1460 // the code to check if the event should be sent.
1461 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1462 Label L;
1463 // Note: frame::interpreter_frame_result has a dependency on how the
1464 // method result is saved across the call to post_method_exit. If this
1465 // is changed then the interpreter_frame_result implementation will
1466 // need to be updated too.
1467
1468 // template interpreter will leave the result on the top of the stack.
1469 push(state);
1470 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1471 cbz(r3, L);
1472 call_VM(noreg,
1473 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1474 bind(L);
1475 pop(state);
1476 }
1477
1478 if (DTraceMethodProbes) {
1479 push(state);
1480 get_method(c_rarg1);
1481 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1482 rthread, c_rarg1);
1483 pop(state);
1484 }
1485 }
1486
1487
1488 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1489 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1490 int increment, Address mask,
1491 Register scratch, Register scratch2,
1492 bool preloaded, Condition cond,
1493 Label* where) {
1494 if (!preloaded) {
1495 ldrw(scratch, counter_addr);
1496 }
1497 add(scratch, scratch, increment);
1498 strw(scratch, counter_addr);
1499 ldrw(scratch2, mask);
1500 ands(scratch, scratch, scratch2);
1501 br(cond, *where);
1502 }
1503
1504 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1505 int number_of_arguments) {
1506 // interpreter specific
1507 //
1508 // Note: No need to save/restore rbcp & rlocals pointer since these
1509 // are callee saved registers and no blocking/ GC can happen
1510 // in leaf calls.
1511 #ifdef ASSERT
1512 {
1513 Label L;
1514 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1515 cbz(rscratch1, L);
1516 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1517 " last_sp != nullptr");
1518 bind(L);
1519 }
1520 #endif /* ASSERT */
1521 // super call
1522 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1523 }
1524
1525 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1526 Register java_thread,
1527 Register last_java_sp,
1528 address entry_point,
1529 int number_of_arguments,
1530 bool check_exceptions) {
1531 // interpreter specific
1532 //
1533 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1534 // really make a difference for these runtime calls, since they are
1535 // slow anyway. Btw., bcp must be saved/restored since it may change
1536 // due to GC.
1537 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1538 save_bcp();
1539 #ifdef ASSERT
1540 {
1541 Label L;
1542 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1543 cbz(rscratch1, L);
1544 stop("InterpreterMacroAssembler::call_VM_base:"
1545 " last_sp != nullptr");
1546 bind(L);
1547 }
1548 #endif /* ASSERT */
1549 // super call
1550 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1551 entry_point, number_of_arguments,
1552 check_exceptions);
1553 // interpreter specific
1554 restore_bcp();
1555 restore_locals();
1556 }
1557
1558 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1559 address entry_point,
1560 Register arg_1) {
1561 assert(arg_1 == c_rarg1, "");
1562 Label resume_pc, not_preempted;
1563
1564 #ifdef ASSERT
1565 {
1566 Label L;
1567 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1568 cbz(rscratch1, L);
1569 stop("Should not have alternate return address set");
1570 bind(L);
1571 }
1572 #endif /* ASSERT */
1573
1574 // Force freeze slow path.
1575 push_cont_fastpath();
1576
1577 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1578 adr(rscratch1, resume_pc);
1579 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1580 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1581
1582 pop_cont_fastpath();
1583
1584 // Check if preempted.
1585 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1586 cbz(rscratch1, not_preempted);
1587 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1588 br(rscratch1);
1589
1590 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1591 bind(resume_pc);
1592 restore_after_resume(false /* is_native */);
1593
1594 bind(not_preempted);
1595 }
1596
1597 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1598 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1599 blr(rscratch1);
1600 if (is_native) {
1601 // On resume we need to set up stack as expected
1602 push(dtos);
1603 push(ltos);
1604 }
1605 }
1606
1607 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1608 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1609 Label update, next, none;
1610
1611 verify_oop(obj);
1612
1613 cbnz(obj, update);
1614 orptr(mdo_addr, TypeEntries::null_seen);
1615 b(next);
1616
1617 bind(update);
1618 load_klass(obj, obj);
1619
1620 ldr(rscratch1, mdo_addr);
1621 eor(obj, obj, rscratch1);
1622 tst(obj, TypeEntries::type_klass_mask);
1623 br(Assembler::EQ, next); // klass seen before, nothing to
1624 // do. The unknown bit may have been
1625 // set already but no need to check.
1626
1627 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1628 // already unknown. Nothing to do anymore.
1629
1630 cbz(rscratch1, none);
1631 cmp(rscratch1, (u1)TypeEntries::null_seen);
1632 br(Assembler::EQ, none);
1633 // There is a chance that the checks above
1634 // fail if another thread has just set the
1635 // profiling to this obj's klass
1636 eor(obj, obj, rscratch1); // get back original value before XOR
1637 ldr(rscratch1, mdo_addr);
1638 eor(obj, obj, rscratch1);
1639 tst(obj, TypeEntries::type_klass_mask);
1640 br(Assembler::EQ, next);
1641
1642 // different than before. Cannot keep accurate profile.
1643 orptr(mdo_addr, TypeEntries::type_unknown);
1644 b(next);
1645
1646 bind(none);
1647 // first time here. Set profile type.
1648 str(obj, mdo_addr);
1649 #ifdef ASSERT
1650 andr(obj, obj, TypeEntries::type_mask);
1651 verify_klass_ptr(obj);
1652 #endif
1653
1654 bind(next);
1655 }
1656
1657 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1658 if (!ProfileInterpreter) {
1659 return;
1660 }
1661
1662 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1663 Label profile_continue;
1664
1665 test_method_data_pointer(mdp, profile_continue);
1666
1667 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1668
1669 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1670 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1671 br(Assembler::NE, profile_continue);
1672
1673 if (MethodData::profile_arguments()) {
1674 Label done;
1675 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1676
1677 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1678 if (i > 0 || MethodData::profile_return()) {
1679 // If return value type is profiled we may have no argument to profile
1680 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1681 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1682 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1683 add(rscratch1, mdp, off_to_args);
1684 br(Assembler::LT, done);
1685 }
1686 ldr(tmp, Address(callee, Method::const_offset()));
1687 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1688 // stack offset o (zero based) from the start of the argument
1689 // list, for n arguments translates into offset n - o - 1 from
1690 // the end of the argument list
1691 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1692 sub(tmp, tmp, rscratch1);
1693 sub(tmp, tmp, 1);
1694 Address arg_addr = argument_address(tmp);
1695 ldr(tmp, arg_addr);
1696
1697 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1698 profile_obj_type(tmp, mdo_arg_addr);
1699
1700 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1701 off_to_args += to_add;
1702 }
1703
1704 if (MethodData::profile_return()) {
1705 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1706 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1707 }
1708
1709 add(rscratch1, mdp, off_to_args);
1710 bind(done);
1711 mov(mdp, rscratch1);
1712
1713 if (MethodData::profile_return()) {
1714 // We're right after the type profile for the last
1715 // argument. tmp is the number of cells left in the
1716 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1717 // if there's a return to profile.
1718 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1719 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1720 }
1721 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1722 } else {
1723 assert(MethodData::profile_return(), "either profile call args or call ret");
1724 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1725 }
1726
1727 // mdp points right after the end of the
1728 // CallTypeData/VirtualCallTypeData, right after the cells for the
1729 // return value type if there's one
1730
1731 bind(profile_continue);
1732 }
1733 }
1734
1735 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1736 assert_different_registers(mdp, ret, tmp, rbcp);
1737 if (ProfileInterpreter && MethodData::profile_return()) {
1738 Label profile_continue, done;
1739
1740 test_method_data_pointer(mdp, profile_continue);
1741
1742 if (MethodData::profile_return_jsr292_only()) {
1743 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1744
1745 // If we don't profile all invoke bytecodes we must make sure
1746 // it's a bytecode we indeed profile. We can't go back to the
1747 // beginning of the ProfileData we intend to update to check its
1748 // type because we're right after it and we don't known its
1749 // length
1750 Label do_profile;
1751 ldrb(rscratch1, Address(rbcp, 0));
1752 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1753 br(Assembler::EQ, do_profile);
1754 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1755 br(Assembler::EQ, do_profile);
1756 get_method(tmp);
1757 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1758 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1759 br(Assembler::NE, profile_continue);
1760
1761 bind(do_profile);
1762 }
1763
1764 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1765 mov(tmp, ret);
1766 profile_obj_type(tmp, mdo_ret_addr);
1767
1768 bind(profile_continue);
1769 }
1770 }
1771
1772 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1773 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1774 if (ProfileInterpreter && MethodData::profile_parameters()) {
1775 Label profile_continue, done;
1776
1777 test_method_data_pointer(mdp, profile_continue);
1778
1779 // Load the offset of the area within the MDO used for
1780 // parameters. If it's negative we're not profiling any parameters
1781 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1782 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1783
1784 // Compute a pointer to the area for parameters from the offset
1785 // and move the pointer to the slot for the last
1786 // parameters. Collect profiling from last parameter down.
1787 // mdo start + parameters offset + array length - 1
1788 add(mdp, mdp, tmp1);
1789 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1790 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1791
1792 Label loop;
1793 bind(loop);
1794
1795 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1796 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1797 int per_arg_scale = exact_log2(DataLayout::cell_size);
1798 add(rscratch1, mdp, off_base);
1799 add(rscratch2, mdp, type_base);
1800
1801 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1802 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1803
1804 // load offset on the stack from the slot for this parameter
1805 ldr(tmp2, arg_off);
1806 neg(tmp2, tmp2);
1807 // read the parameter from the local area
1808 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1809
1810 // profile the parameter
1811 profile_obj_type(tmp2, arg_type);
1812
1813 // go to next parameter
1814 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1815 br(Assembler::GE, loop);
1816
1817 bind(profile_continue);
1818 }
1819 }
1820
1821 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1822 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1823 get_cache_index_at_bcp(index, 1, sizeof(u4));
1824 // Get address of invokedynamic array
1825 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1826 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1827 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1828 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1829 lea(cache, Address(cache, index));
1830 }
1831
1832 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1833 // Get index out of bytecode pointer
1834 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1835 // Take shortcut if the size is a power of 2
1836 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1837 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1838 } else {
1839 mov(cache, sizeof(ResolvedFieldEntry));
1840 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1841 }
1842 // Get address of field entries array
1843 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1844 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1845 lea(cache, Address(cache, index));
1846 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1847 membar(MacroAssembler::LoadLoad);
1848 }
1849
1850 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1851 // Get index out of bytecode pointer
1852 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1853 mov(cache, sizeof(ResolvedMethodEntry));
1854 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1855
1856 // Get address of field entries array
1857 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1858 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1859 lea(cache, Address(cache, index));
1860 }