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/constMethodFlags.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/inlineKlass.hpp"
40 #include "oops/resolvedFieldEntry.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/jvmtiThreadState.hpp"
45 #include "runtime/basicLock.hpp"
46 #include "runtime/frame.inline.hpp"
47 #include "runtime/javaThread.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 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
214 Register t1, Register t2,
215 bool clear_fields, Label& alloc_failed) {
216 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
217 if (DTraceAllocProbes) {
218 // Trigger dtrace event for fastpath
219 push(atos);
220 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
221 pop(atos);
222 }
223 }
224
225 void InterpreterMacroAssembler::read_flat_field(Register entry, Register obj) {
226 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field), obj, entry);
227 membar(Assembler::StoreStore);
228 }
229
230 void InterpreterMacroAssembler::write_flat_field(Register entry, Register field_offset,
231 Register tmp1, Register tmp2,
232 Register obj) {
233 assert_different_registers(entry, field_offset, tmp1, tmp2, obj);
234 Label slow_path, done;
235
236 load_unsigned_byte(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::flags_offset())));
237 test_field_is_not_null_free_inline_type(tmp1, noreg /* temp */, slow_path);
238
239 null_check(r0); // FIXME JDK-8341120
240
241 add(obj, obj, field_offset);
242
243 load_klass(tmp1, r0);
244 payload_address(r0, r0, tmp1);
245
246 Register layout_info = field_offset;
247 load_unsigned_short(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
248 ldr(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
249 inline_layout_info(tmp2, tmp1, layout_info);
250
251 flat_field_copy(IN_HEAP, r0, obj, layout_info);
252 b(done);
253
254 bind(slow_path);
255 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flat_field), obj, r0, entry);
256 bind(done);
257 }
258
259 // Load object from cpool->resolved_references(index)
260 void InterpreterMacroAssembler::load_resolved_reference_at_index(
261 Register result, Register index, Register tmp) {
262 assert_different_registers(result, index);
263
264 get_constant_pool(result);
265 // load pointer for resolved_references[] objArray
266 ldr(result, Address(result, ConstantPool::cache_offset()));
267 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
268 resolve_oop_handle(result, tmp, rscratch2);
269 // Add in the index
270 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
271 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
272 }
273
274 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
275 Register cpool, Register index, Register klass, Register temp) {
276 add(temp, cpool, index, LSL, LogBytesPerWord);
277 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
278 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
279 add(klass, klass, temp, LSL, LogBytesPerWord);
280 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
281 }
282
283 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
284 // subtype of super_klass.
285 //
286 // Args:
287 // r0: superklass
288 // Rsub_klass: subklass
289 //
290 // Kills:
291 // r2, r5
292 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
293 Label& ok_is_subtype,
294 bool profile) {
295 assert(Rsub_klass != r0, "r0 holds superklass");
296 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
297 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
298
299 // Profile the not-null value's klass.
300 if (profile) {
301 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
302 }
303
304 // Do the check.
305 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
306 }
307
308 // Java Expression Stack
309
310 void InterpreterMacroAssembler::pop_ptr(Register r) {
311 ldr(r, post(esp, wordSize));
312 }
313
314 void InterpreterMacroAssembler::pop_i(Register r) {
315 ldrw(r, post(esp, wordSize));
316 }
317
318 void InterpreterMacroAssembler::pop_l(Register r) {
319 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
320 }
321
322 void InterpreterMacroAssembler::push_ptr(Register r) {
323 str(r, pre(esp, -wordSize));
324 }
325
326 void InterpreterMacroAssembler::push_i(Register r) {
327 str(r, pre(esp, -wordSize));
328 }
329
330 void InterpreterMacroAssembler::push_l(Register r) {
331 str(zr, pre(esp, -wordSize));
332 str(r, pre(esp, - wordSize));
333 }
334
335 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
336 ldrs(r, post(esp, wordSize));
337 }
338
339 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
340 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
341 }
342
343 void InterpreterMacroAssembler::push_f(FloatRegister r) {
344 strs(r, pre(esp, -wordSize));
345 }
346
347 void InterpreterMacroAssembler::push_d(FloatRegister r) {
348 strd(r, pre(esp, 2* -wordSize));
349 }
350
351 void InterpreterMacroAssembler::pop(TosState state) {
352 switch (state) {
353 case atos: pop_ptr(); break;
354 case btos:
355 case ztos:
356 case ctos:
357 case stos:
358 case itos: pop_i(); break;
359 case ltos: pop_l(); break;
360 case ftos: pop_f(); break;
361 case dtos: pop_d(); break;
362 case vtos: /* nothing to do */ break;
363 default: ShouldNotReachHere();
364 }
365 interp_verify_oop(r0, state);
366 }
367
368 void InterpreterMacroAssembler::push(TosState state) {
369 interp_verify_oop(r0, state);
370 switch (state) {
371 case atos: push_ptr(); break;
372 case btos:
373 case ztos:
374 case ctos:
375 case stos:
376 case itos: push_i(); break;
377 case ltos: push_l(); break;
378 case ftos: push_f(); break;
379 case dtos: push_d(); break;
380 case vtos: /* nothing to do */ break;
381 default : ShouldNotReachHere();
382 }
383 }
384
385 // Helpers for swap and dup
386 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
387 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
388 }
389
390 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
391 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
392 }
393
394 void InterpreterMacroAssembler::load_float(Address src) {
395 ldrs(v0, src);
396 }
397
398 void InterpreterMacroAssembler::load_double(Address src) {
399 ldrd(v0, src);
400 }
401
402 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
403 // set sender sp
404 mov(r19_sender_sp, sp);
405 // record last_sp
406 sub(rscratch1, esp, rfp);
407 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
408 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
409 }
410
411 // Jump to from_interpreted entry of a call unless single stepping is possible
412 // in this thread in which case we must call the i2i entry
413 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
414 prepare_to_jump_from_interpreted();
415
416 if (JvmtiExport::can_post_interpreter_events()) {
417 Label run_compiled_code;
418 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
419 // compiled code in threads for which the event is enabled. Check here for
420 // interp_only_mode if these events CAN be enabled.
421 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
422 cbzw(rscratch1, run_compiled_code);
423 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
424 br(rscratch1);
425 bind(run_compiled_code);
426 }
427
428 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
429 br(rscratch1);
430 }
431
432 // The following two routines provide a hook so that an implementation
433 // can schedule the dispatch in two parts. amd64 does not do this.
434 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
435 }
436
437 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
438 dispatch_next(state, step);
439 }
440
441 void InterpreterMacroAssembler::dispatch_base(TosState state,
442 address* table,
443 bool verifyoop,
444 bool generate_poll) {
445 if (VerifyActivationFrameSize) {
446 Label L;
447 sub(rscratch2, rfp, esp);
448 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
449 subs(rscratch2, rscratch2, min_frame_size);
450 br(Assembler::GE, L);
451 stop("broken stack frame");
452 bind(L);
453 }
454 if (verifyoop) {
455 interp_verify_oop(r0, state);
456 }
457
458 Label safepoint;
459 address* const safepoint_table = Interpreter::safept_table(state);
460 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
461
462 if (needs_thread_local_poll) {
463 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
464 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
465 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
466 }
467
468 if (table == Interpreter::dispatch_table(state)) {
469 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
470 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
471 } else {
472 mov(rscratch2, (address)table);
473 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
474 }
475 br(rscratch2);
476
477 if (needs_thread_local_poll) {
478 bind(safepoint);
479 lea(rscratch2, ExternalAddress((address)safepoint_table));
480 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
481 br(rscratch2);
482 }
483 }
484
485 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
486 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
487 }
488
489 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
490 dispatch_base(state, Interpreter::normal_table(state));
491 }
492
493 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
494 dispatch_base(state, Interpreter::normal_table(state), false);
495 }
496
497
498 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
499 // load next bytecode
500 ldrb(rscratch1, Address(pre(rbcp, step)));
501 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
502 }
503
504 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
505 // load current bytecode
506 ldrb(rscratch1, Address(rbcp, 0));
507 dispatch_base(state, table);
508 }
509
510 // remove activation
511 //
512 // Unlock the receiver if this is a synchronized method.
513 // Unlock any Java monitors from synchronized blocks.
514 // Apply stack watermark barrier.
515 // Notify JVMTI.
516 // Remove the activation from the stack.
517 //
518 // If there are locked Java monitors
519 // If throw_monitor_exception
520 // throws IllegalMonitorStateException
521 // Else if install_monitor_exception
522 // installs IllegalMonitorStateException
523 // Else
524 // no error processing
525 void InterpreterMacroAssembler::remove_activation(TosState state,
526 bool throw_monitor_exception,
527 bool install_monitor_exception,
528 bool notify_jvmdi) {
529 // Note: Registers r3 xmm0 may be in use for the
530 // result check if synchronized method
531 Label unlocked, unlock, no_unlock;
532
533 #ifdef ASSERT
534 Label not_preempted;
535 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
536 cbz(rscratch1, not_preempted);
537 stop("remove_activation: should not have alternate return address set");
538 bind(not_preempted);
539 #endif /* ASSERT */
540
541 // get the value of _do_not_unlock_if_synchronized into r3
542 const Address do_not_unlock_if_synchronized(rthread,
543 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
544 ldrb(r3, do_not_unlock_if_synchronized);
545 strb(zr, do_not_unlock_if_synchronized); // reset the flag
546
547 // get method access flags
548 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
549 ldrh(r2, Address(r1, Method::access_flags_offset()));
550 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
551
552 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
553 // is set.
554 cbnz(r3, no_unlock);
555
556 // unlock monitor
557 push(state); // save result
558
559 // BasicObjectLock will be first in list, since this is a
560 // synchronized method. However, need to check that the object has
561 // not been unlocked by an explicit monitorexit bytecode.
562 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
563 wordSize - (int) sizeof(BasicObjectLock));
564 // We use c_rarg1 so that if we go slow path it will be the correct
565 // register for unlock_object to pass to VM directly
566 lea(c_rarg1, monitor); // address of first monitor
567
568 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
569 cbnz(r0, unlock);
570
571 pop(state);
572 if (throw_monitor_exception) {
573 // Entry already unlocked, need to throw exception
574 call_VM(noreg, CAST_FROM_FN_PTR(address,
575 InterpreterRuntime::throw_illegal_monitor_state_exception));
576 should_not_reach_here();
577 } else {
578 // Monitor already unlocked during a stack unroll. If requested,
579 // install an illegal_monitor_state_exception. Continue with
580 // stack unrolling.
581 if (install_monitor_exception) {
582 call_VM(noreg, CAST_FROM_FN_PTR(address,
583 InterpreterRuntime::new_illegal_monitor_state_exception));
584 }
585 b(unlocked);
586 }
587
588 bind(unlock);
589 unlock_object(c_rarg1);
590 pop(state);
591
592 // Check that for block-structured locking (i.e., that all locked
593 // objects has been unlocked)
594 bind(unlocked);
595
596 // r0: Might contain return value
597
598 // Check that all monitors are unlocked
599 {
600 Label loop, exception, entry, restart;
601 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
602 const Address monitor_block_top(
603 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
604 const Address monitor_block_bot(
605 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
606
607 bind(restart);
608 // We use c_rarg1 so that if we go slow path it will be the correct
609 // register for unlock_object to pass to VM directly
610 ldr(c_rarg1, monitor_block_top); // derelativize pointer
611 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
612 // c_rarg1 points to current entry, starting with top-most entry
613
614 lea(r19, monitor_block_bot); // points to word before bottom of
615 // monitor block
616 b(entry);
617
618 // Entry already locked, need to throw exception
619 bind(exception);
620
621 if (throw_monitor_exception) {
622 // Throw exception
623 MacroAssembler::call_VM(noreg,
624 CAST_FROM_FN_PTR(address, InterpreterRuntime::
625 throw_illegal_monitor_state_exception));
626 should_not_reach_here();
627 } else {
628 // Stack unrolling. Unlock object and install illegal_monitor_exception.
629 // Unlock does not block, so don't have to worry about the frame.
630 // We don't have to preserve c_rarg1 since we are going to throw an exception.
631
632 push(state);
633 unlock_object(c_rarg1);
634 pop(state);
635
636 if (install_monitor_exception) {
637 call_VM(noreg, CAST_FROM_FN_PTR(address,
638 InterpreterRuntime::
639 new_illegal_monitor_state_exception));
640 }
641
642 b(restart);
643 }
644
645 bind(loop);
646 // check if current entry is used
647 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
648 cbnz(rscratch1, exception);
649
650 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
651 bind(entry);
652 cmp(c_rarg1, r19); // check if bottom reached
653 br(Assembler::NE, loop); // if not at bottom then check this entry
654 }
655
656 bind(no_unlock);
657
658 JFR_ONLY(enter_jfr_critical_section();)
659
660 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
661 // that would normally not be safe to use. Such bad returns into unsafe territory of
662 // the stack, will call InterpreterRuntime::at_unwind.
663 Label slow_path;
664 Label fast_path;
665 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
666 br(Assembler::AL, fast_path);
667 bind(slow_path);
668 push(state);
669 set_last_Java_frame(esp, rfp, pc(), rscratch1);
670 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
671 reset_last_Java_frame(true);
672 pop(state);
673 bind(fast_path);
674
675 // JVMTI support. Make sure the safepoint poll test is issued prior.
676 if (notify_jvmdi) {
677 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
678 } else {
679 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
680 }
681
682 // remove activation
683 // get sender esp
684 ldr(rscratch2,
685 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
686
687 if (StackReservedPages > 0) {
688 // testing if reserved zone needs to be re-enabled
689 Label no_reserved_zone_enabling;
690
691 // check if already enabled - if so no re-enabling needed
692 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
693 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
694 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
695 br(Assembler::EQ, no_reserved_zone_enabling);
696
697 // look for an overflow into the stack reserved zone, i.e.
698 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
699 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
700 cmp(rscratch2, rscratch1);
701 br(Assembler::LS, no_reserved_zone_enabling);
702
703 JFR_ONLY(leave_jfr_critical_section();)
704
705 call_VM_leaf(
706 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
707 call_VM(noreg, CAST_FROM_FN_PTR(address,
708 InterpreterRuntime::throw_delayed_StackOverflowError));
709 should_not_reach_here();
710
711 bind(no_reserved_zone_enabling);
712 }
713
714 if (state == atos && InlineTypeReturnedAsFields) {
715 Label skip;
716 Label not_null;
717 cbnz(r0, not_null);
718 // Returned value is null, zero all return registers because they may belong to oop fields
719 mov(j_rarg1, zr);
720 mov(j_rarg2, zr);
721 mov(j_rarg3, zr);
722 mov(j_rarg4, zr);
723 mov(j_rarg5, zr);
724 mov(j_rarg6, zr);
725 mov(j_rarg7, zr);
726 b(skip);
727 bind(not_null);
728
729 // Check if we are returning an non-null inline type and load its fields into registers
730 test_oop_is_not_inline_type(r0, rscratch2, skip, /* can_be_null= */ false);
731
732 // Load fields from a buffered value with an inline class specific handler
733 load_klass(rscratch1 /*dst*/, r0 /*src*/);
734 ldr(rscratch1, Address(rscratch1, InlineKlass::adr_members_offset()));
735 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
736 // Unpack handler can be null if inline type is not scalarizable in returns
737 cbz(rscratch1, skip);
738
739 blr(rscratch1);
740 #ifdef ASSERT
741 // TODO 8284443 Enable
742 if (StressCallingConvention && false) {
743 Label skip_stress;
744 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
745 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
746 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
747 br(Assembler::EQ, skip_stress);
748 load_klass(r0, r0);
749 orr(r0, r0, 1);
750 bind(skip_stress);
751 }
752 #endif
753 bind(skip);
754 // Check above kills sender esp in rscratch2. Reload it.
755 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
756 }
757
758 // remove frame anchor
759 leave();
760
761 JFR_ONLY(leave_jfr_critical_section();)
762
763 // restore sender esp
764 mov(esp, rscratch2);
765
766 // If we're returning to interpreted code we will shortly be
767 // adjusting SP to allow some space for ESP. If we're returning to
768 // compiled code the saved sender SP was saved in sender_sp, so this
769 // restores it.
770 andr(sp, esp, -16);
771 }
772
773 #if INCLUDE_JFR
774 void InterpreterMacroAssembler::enter_jfr_critical_section() {
775 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
776 mov(rscratch1, true);
777 strb(rscratch1, sampling_critical_section);
778 }
779
780 void InterpreterMacroAssembler::leave_jfr_critical_section() {
781 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
782 strb(zr, sampling_critical_section);
783 }
784 #endif // INCLUDE_JFR
785
786 // Lock object
787 //
788 // Args:
789 // c_rarg1: BasicObjectLock to be used for locking
790 //
791 // Kills:
792 // r0
793 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
794 // rscratch1, rscratch2 (scratch regs)
795 void InterpreterMacroAssembler::lock_object(Register lock_reg)
796 {
797 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
798
799 const Register tmp = c_rarg2;
800 const Register obj_reg = c_rarg3; // Will contain the oop
801 const Register tmp2 = c_rarg4;
802 const Register tmp3 = c_rarg5;
803
804 // Load object pointer into obj_reg %c_rarg3
805 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
806
807 Label slow_case, done;
808 fast_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
809 b(done);
810
811 bind(slow_case);
812
813 // Call the runtime routine for slow case
814 call_VM_preemptable(noreg,
815 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
816 lock_reg);
817
818 bind(done);
819 }
820
821
822 // Unlocks an object. Used in monitorexit bytecode and
823 // remove_activation. Throws an IllegalMonitorException if object is
824 // not locked by current thread.
825 //
826 // Args:
827 // c_rarg1: BasicObjectLock for lock
828 //
829 // Kills:
830 // r0
831 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
832 // rscratch1, rscratch2 (scratch regs)
833 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
834 {
835 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
836
837 const Register swap_reg = r0;
838 const Register header_reg = c_rarg2; // Will contain the old oopMark
839 const Register obj_reg = c_rarg3; // Will contain the oop
840 const Register tmp_reg = c_rarg4; // Temporary used by fast_unlock
841
842 save_bcp(); // Save in case of exception
843
844 // Load oop into obj_reg(%c_rarg3)
845 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
846
847 // Free entry
848 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
849
850 Label slow_case, done;
851 fast_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
852 b(done);
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 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
863 Label& zero_continue) {
864 assert(ProfileInterpreter, "must be profiling interpreter");
865 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
866 cbz(mdp, zero_continue);
867 }
868
869 // Set the method data pointer for the current bcp.
870 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
871 assert(ProfileInterpreter, "must be profiling interpreter");
872 Label set_mdp;
873 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
874
875 // Test MDO to avoid the call if it is null.
876 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
877 cbz(r0, set_mdp);
878 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
879 // r0: mdi
880 // mdo is guaranteed to be non-zero here, we checked for it before the call.
881 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
882 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
883 add(r0, r1, r0);
884 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
885 bind(set_mdp);
886 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
887 }
888
889 void InterpreterMacroAssembler::verify_method_data_pointer() {
890 assert(ProfileInterpreter, "must be profiling interpreter");
891 #ifdef ASSERT
892 Label verify_continue;
893 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
894 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
895 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
896 get_method(r1);
897
898 // If the mdp is valid, it will point to a DataLayout header which is
899 // consistent with the bcp. The converse is highly probable also.
900 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
901 ldr(rscratch1, Address(r1, Method::const_offset()));
902 add(r2, r2, rscratch1, Assembler::LSL);
903 lea(r2, Address(r2, ConstMethod::codes_offset()));
904 cmp(r2, rbcp);
905 br(Assembler::EQ, verify_continue);
906 // r1: method
907 // rbcp: bcp // rbcp == 22
908 // r3: mdp
909 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
910 r1, rbcp, r3);
911 bind(verify_continue);
912 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
913 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
914 #endif // ASSERT
915 }
916
917
918 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
919 int constant,
920 Register value) {
921 assert(ProfileInterpreter, "must be profiling interpreter");
922 Address data(mdp_in, constant);
923 str(value, data);
924 }
925
926
927 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
928 int constant) {
929 increment_mdp_data_at(mdp_in, noreg, constant);
930 }
931
932 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
933 Register index,
934 int constant) {
935 assert(ProfileInterpreter, "must be profiling interpreter");
936
937 assert_different_registers(rscratch2, rscratch1, mdp_in, index);
938
939 Address addr1(mdp_in, constant);
940 Address addr2(rscratch2, index, Address::lsl(0));
941 Address &addr = addr1;
942 if (index != noreg) {
943 lea(rscratch2, addr1);
944 addr = addr2;
945 }
946
947 increment(addr, DataLayout::counter_increment);
948 }
949
950 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
951 int flag_byte_constant) {
952 assert(ProfileInterpreter, "must be profiling interpreter");
953 int flags_offset = in_bytes(DataLayout::flags_offset());
954 // Set the flag
955 ldrb(rscratch1, Address(mdp_in, flags_offset));
956 orr(rscratch1, rscratch1, flag_byte_constant);
957 strb(rscratch1, Address(mdp_in, flags_offset));
958 }
959
960
961 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
962 int offset,
963 Register value,
964 Register test_value_out,
965 Label& not_equal_continue) {
966 assert(ProfileInterpreter, "must be profiling interpreter");
967 if (test_value_out == noreg) {
968 ldr(rscratch1, Address(mdp_in, offset));
969 cmp(value, rscratch1);
970 } else {
971 // Put the test value into a register, so caller can use it:
972 ldr(test_value_out, Address(mdp_in, offset));
973 cmp(value, test_value_out);
974 }
975 br(Assembler::NE, not_equal_continue);
976 }
977
978
979 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
980 int offset_of_disp) {
981 assert(ProfileInterpreter, "must be profiling interpreter");
982 ldr(rscratch1, Address(mdp_in, offset_of_disp));
983 add(mdp_in, mdp_in, rscratch1, LSL);
984 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
985 }
986
987
988 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
989 Register reg,
990 int offset_of_disp) {
991 assert(ProfileInterpreter, "must be profiling interpreter");
992 lea(rscratch1, Address(mdp_in, offset_of_disp));
993 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
994 add(mdp_in, mdp_in, rscratch1, LSL);
995 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
996 }
997
998
999 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1000 int constant) {
1001 assert(ProfileInterpreter, "must be profiling interpreter");
1002 add(mdp_in, mdp_in, (unsigned)constant);
1003 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1004 }
1005
1006
1007 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1008 assert(ProfileInterpreter, "must be profiling interpreter");
1009 // save/restore across call_VM
1010 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1011 call_VM(noreg,
1012 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1013 return_bci);
1014 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1015 }
1016
1017
1018 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
1019 if (ProfileInterpreter) {
1020 Label profile_continue;
1021
1022 // If no method data exists, go to profile_continue.
1023 test_method_data_pointer(mdp, profile_continue);
1024
1025 // We are taking a branch. Increment the taken count.
1026 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1027
1028 // The method data pointer needs to be updated to reflect the new target.
1029 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1030 bind(profile_continue);
1031 }
1032 }
1033
1034
1035 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1036 if (ProfileInterpreter) {
1037 Label profile_continue;
1038
1039 // If no method data exists, go to profile_continue.
1040 test_method_data_pointer(mdp, profile_continue);
1041
1042 // We are not taking a branch. Increment the not taken count.
1043 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1044
1045 // The method data pointer needs to be updated to correspond to
1046 // the next bytecode
1047 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1048 bind(profile_continue);
1049 }
1050 }
1051
1052
1053 void InterpreterMacroAssembler::profile_call(Register mdp) {
1054 if (ProfileInterpreter) {
1055 Label profile_continue;
1056
1057 // If no method data exists, go to profile_continue.
1058 test_method_data_pointer(mdp, profile_continue);
1059
1060 // We are making a call. Increment the count.
1061 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1062
1063 // The method data pointer needs to be updated to reflect the new target.
1064 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1065 bind(profile_continue);
1066 }
1067 }
1068
1069 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1070 if (ProfileInterpreter) {
1071 Label profile_continue;
1072
1073 // If no method data exists, go to profile_continue.
1074 test_method_data_pointer(mdp, profile_continue);
1075
1076 // We are making a call. Increment the count.
1077 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1078
1079 // The method data pointer needs to be updated to reflect the new target.
1080 update_mdp_by_constant(mdp,
1081 in_bytes(VirtualCallData::
1082 virtual_call_data_size()));
1083 bind(profile_continue);
1084 }
1085 }
1086
1087
1088 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1089 Register mdp,
1090 Register reg2,
1091 bool receiver_can_be_null) {
1092 if (ProfileInterpreter) {
1093 Label profile_continue;
1094
1095 // If no method data exists, go to profile_continue.
1096 test_method_data_pointer(mdp, profile_continue);
1097
1098 Label skip_receiver_profile;
1099 if (receiver_can_be_null) {
1100 Label not_null;
1101 // We are making a call. Increment the count for null receiver.
1102 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1103 b(skip_receiver_profile);
1104 bind(not_null);
1105 }
1106
1107 // Record the receiver type.
1108 record_klass_in_profile(receiver, mdp, reg2);
1109 bind(skip_receiver_profile);
1110
1111 // The method data pointer needs to be updated to reflect the new target.
1112 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1113 bind(profile_continue);
1114 }
1115 }
1116
1117 // This routine creates a state machine for updating the multi-row
1118 // type profile at a virtual call site (or other type-sensitive bytecode).
1119 // The machine visits each row (of receiver/count) until the receiver type
1120 // is found, or until it runs out of rows. At the same time, it remembers
1121 // the location of the first empty row. (An empty row records null for its
1122 // receiver, and can be allocated for a newly-observed receiver type.)
1123 // Because there are two degrees of freedom in the state, a simple linear
1124 // search will not work; it must be a decision tree. Hence this helper
1125 // function is recursive, to generate the required tree structured code.
1126 // It's the interpreter, so we are trading off code space for speed.
1127 // See below for example code.
1128 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1129 Register receiver, Register mdp,
1130 Register reg2, int start_row,
1131 Label& done) {
1132 if (TypeProfileWidth == 0) {
1133 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1134 } else {
1135 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1136 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1137 }
1138 }
1139
1140 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1141 Register reg2, int start_row, Label& done, int total_rows,
1142 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1143 int last_row = total_rows - 1;
1144 assert(start_row <= last_row, "must be work left to do");
1145 // Test this row for both the item and for null.
1146 // Take any of three different outcomes:
1147 // 1. found item => increment count and goto done
1148 // 2. found null => keep looking for case 1, maybe allocate this cell
1149 // 3. found something else => keep looking for cases 1 and 2
1150 // Case 3 is handled by a recursive call.
1151 for (int row = start_row; row <= last_row; row++) {
1152 Label next_test;
1153 bool test_for_null_also = (row == start_row);
1154
1155 // See if the item is item[n].
1156 int item_offset = in_bytes(item_offset_fn(row));
1157 test_mdp_data_at(mdp, item_offset, item,
1158 (test_for_null_also ? reg2 : noreg),
1159 next_test);
1160 // (Reg2 now contains the item from the CallData.)
1161
1162 // The item is item[n]. Increment count[n].
1163 int count_offset = in_bytes(item_count_offset_fn(row));
1164 increment_mdp_data_at(mdp, count_offset);
1165 b(done);
1166 bind(next_test);
1167
1168 if (test_for_null_also) {
1169 Label found_null;
1170 // Failed the equality check on item[n]... Test for null.
1171 if (start_row == last_row) {
1172 // The only thing left to do is handle the null case.
1173 cbz(reg2, found_null);
1174 // Item did not match any saved item and there is no empty row for it.
1175 // Increment total counter to indicate polymorphic case.
1176 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1177 b(done);
1178 bind(found_null);
1179 break;
1180 }
1181 // Since null is rare, make it be the branch-taken case.
1182 cbz(reg2, found_null);
1183
1184 // Put all the "Case 3" tests here.
1185 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1186 item_offset_fn, item_count_offset_fn);
1187
1188 // Found a null. Keep searching for a matching item,
1189 // but remember that this is an empty (unused) slot.
1190 bind(found_null);
1191 }
1192 }
1193
1194 // In the fall-through case, we found no matching item, but we
1195 // observed the item[start_row] is null.
1196
1197 // Fill in the item field and increment the count.
1198 int item_offset = in_bytes(item_offset_fn(start_row));
1199 set_mdp_data_at(mdp, item_offset, item);
1200 int count_offset = in_bytes(item_count_offset_fn(start_row));
1201 mov(reg2, DataLayout::counter_increment);
1202 set_mdp_data_at(mdp, count_offset, reg2);
1203 if (start_row > 0) {
1204 b(done);
1205 }
1206 }
1207
1208 // Example state machine code for three profile rows:
1209 // // main copy of decision tree, rooted at row[1]
1210 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1211 // if (row[0].rec != nullptr) {
1212 // // inner copy of decision tree, rooted at row[1]
1213 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1214 // if (row[1].rec != nullptr) {
1215 // // degenerate decision tree, rooted at row[2]
1216 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1217 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1218 // row[2].init(rec); goto done;
1219 // } else {
1220 // // remember row[1] is empty
1221 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1222 // row[1].init(rec); goto done;
1223 // }
1224 // } else {
1225 // // remember row[0] is empty
1226 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1227 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1228 // row[0].init(rec); goto done;
1229 // }
1230 // done:
1231
1232 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1233 Register mdp, Register reg2) {
1234 assert(ProfileInterpreter, "must be profiling");
1235 Label done;
1236
1237 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1238
1239 bind (done);
1240 }
1241
1242 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1243 Register mdp) {
1244 if (ProfileInterpreter) {
1245 Label profile_continue;
1246 uint row;
1247
1248 // If no method data exists, go to profile_continue.
1249 test_method_data_pointer(mdp, profile_continue);
1250
1251 // Update the total ret count.
1252 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1253
1254 for (row = 0; row < RetData::row_limit(); row++) {
1255 Label next_test;
1256
1257 // See if return_bci is equal to bci[n]:
1258 test_mdp_data_at(mdp,
1259 in_bytes(RetData::bci_offset(row)),
1260 return_bci, noreg,
1261 next_test);
1262
1263 // return_bci is equal to bci[n]. Increment the count.
1264 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1265
1266 // The method data pointer needs to be updated to reflect the new target.
1267 update_mdp_by_offset(mdp,
1268 in_bytes(RetData::bci_displacement_offset(row)));
1269 b(profile_continue);
1270 bind(next_test);
1271 }
1272
1273 update_mdp_for_ret(return_bci);
1274
1275 bind(profile_continue);
1276 }
1277 }
1278
1279 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1280 if (ProfileInterpreter) {
1281 Label profile_continue;
1282
1283 // If no method data exists, go to profile_continue.
1284 test_method_data_pointer(mdp, profile_continue);
1285
1286 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1287
1288 // The method data pointer needs to be updated.
1289 int mdp_delta = in_bytes(BitData::bit_data_size());
1290 if (TypeProfileCasts) {
1291 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1292 }
1293 update_mdp_by_constant(mdp, mdp_delta);
1294
1295 bind(profile_continue);
1296 }
1297 }
1298
1299 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1300 if (ProfileInterpreter) {
1301 Label profile_continue;
1302
1303 // If no method data exists, go to profile_continue.
1304 test_method_data_pointer(mdp, profile_continue);
1305
1306 // The method data pointer needs to be updated.
1307 int mdp_delta = in_bytes(BitData::bit_data_size());
1308 if (TypeProfileCasts) {
1309 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1310
1311 // Record the object type.
1312 record_klass_in_profile(klass, mdp, reg2);
1313 }
1314 update_mdp_by_constant(mdp, mdp_delta);
1315
1316 bind(profile_continue);
1317 }
1318 }
1319
1320 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1321 if (ProfileInterpreter) {
1322 Label profile_continue;
1323
1324 // If no method data exists, go to profile_continue.
1325 test_method_data_pointer(mdp, profile_continue);
1326
1327 // Update the default case count
1328 increment_mdp_data_at(mdp,
1329 in_bytes(MultiBranchData::default_count_offset()));
1330
1331 // The method data pointer needs to be updated.
1332 update_mdp_by_offset(mdp,
1333 in_bytes(MultiBranchData::
1334 default_displacement_offset()));
1335
1336 bind(profile_continue);
1337 }
1338 }
1339
1340 void InterpreterMacroAssembler::profile_switch_case(Register index,
1341 Register mdp,
1342 Register reg2) {
1343 if (ProfileInterpreter) {
1344 Label profile_continue;
1345
1346 // If no method data exists, go to profile_continue.
1347 test_method_data_pointer(mdp, profile_continue);
1348
1349 // Build the base (index * per_case_size_in_bytes()) +
1350 // case_array_offset_in_bytes()
1351 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1352 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1353 Assembler::maddw(index, index, reg2, rscratch1);
1354
1355 // Update the case count
1356 increment_mdp_data_at(mdp,
1357 index,
1358 in_bytes(MultiBranchData::relative_count_offset()));
1359
1360 // The method data pointer needs to be updated.
1361 update_mdp_by_offset(mdp,
1362 index,
1363 in_bytes(MultiBranchData::
1364 relative_displacement_offset()));
1365
1366 bind(profile_continue);
1367 }
1368 }
1369
1370 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1371 Register array,
1372 Register tmp) {
1373 if (ProfileInterpreter) {
1374 Label profile_continue;
1375
1376 // If no method data exists, go to profile_continue.
1377 test_method_data_pointer(mdp, profile_continue);
1378
1379 mov(tmp, array);
1380 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1381
1382 Label not_flat;
1383 test_non_flat_array_oop(array, tmp, not_flat);
1384
1385 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1386
1387 bind(not_flat);
1388
1389 Label not_null_free;
1390 test_non_null_free_array_oop(array, tmp, not_null_free);
1391
1392 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1393
1394 bind(not_null_free);
1395
1396 bind(profile_continue);
1397 }
1398 }
1399
1400 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1401 Register array,
1402 Register tmp);
1403 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1404 Register array,
1405 Register tmp);
1406
1407 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1408 if (ProfileInterpreter) {
1409 Label profile_continue;
1410
1411 // If no method data exists, go to profile_continue.
1412 test_method_data_pointer(mdp, profile_continue);
1413
1414 Label done, update;
1415 cbnz(element, update);
1416 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1417 b(done);
1418
1419 bind(update);
1420 load_klass(tmp, element);
1421
1422 // Record the object type.
1423 record_klass_in_profile(tmp, mdp, tmp2);
1424
1425 bind(done);
1426
1427 // The method data pointer needs to be updated.
1428 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1429
1430 bind(profile_continue);
1431 }
1432 }
1433
1434
1435 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1436 Register element,
1437 Register tmp) {
1438 if (ProfileInterpreter) {
1439 Label profile_continue;
1440
1441 // If no method data exists, go to profile_continue.
1442 test_method_data_pointer(mdp, profile_continue);
1443
1444 mov(tmp, element);
1445 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1446
1447 // The method data pointer needs to be updated.
1448 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1449
1450 bind(profile_continue);
1451 }
1452 }
1453
1454 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1455 Register left,
1456 Register right,
1457 Register tmp) {
1458 if (ProfileInterpreter) {
1459 Label profile_continue;
1460
1461 // If no method data exists, go to profile_continue.
1462 test_method_data_pointer(mdp, profile_continue);
1463
1464 mov(tmp, left);
1465 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1466
1467 Label left_not_inline_type;
1468 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1469 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1470 bind(left_not_inline_type);
1471
1472 mov(tmp, right);
1473 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1474
1475 Label right_not_inline_type;
1476 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1477 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1478 bind(right_not_inline_type);
1479
1480 bind(profile_continue);
1481 }
1482 }
1483
1484 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1485 if (state == atos) {
1486 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1487 }
1488 }
1489
1490 void InterpreterMacroAssembler::notify_method_entry() {
1491 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1492 // track stack depth. If it is possible to enter interp_only_mode we add
1493 // the code to check if the event should be sent.
1494 if (JvmtiExport::can_post_interpreter_events()) {
1495 Label L;
1496 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1497 cbzw(r3, L);
1498 call_VM(noreg, CAST_FROM_FN_PTR(address,
1499 InterpreterRuntime::post_method_entry));
1500 bind(L);
1501 }
1502
1503 if (DTraceMethodProbes) {
1504 get_method(c_rarg1);
1505 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1506 rthread, c_rarg1);
1507 }
1508
1509 // RedefineClasses() tracing support for obsolete method entry
1510 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1511 get_method(c_rarg1);
1512 call_VM_leaf(
1513 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1514 rthread, c_rarg1);
1515 }
1516
1517 }
1518
1519
1520 void InterpreterMacroAssembler::notify_method_exit(
1521 TosState state, NotifyMethodExitMode mode) {
1522 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1523 // track stack depth. If it is possible to enter interp_only_mode we add
1524 // the code to check if the event should be sent.
1525 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1526 Label L;
1527 // Note: frame::interpreter_frame_result has a dependency on how the
1528 // method result is saved across the call to post_method_exit. If this
1529 // is changed then the interpreter_frame_result implementation will
1530 // need to be updated too.
1531
1532 // template interpreter will leave the result on the top of the stack.
1533 push(state);
1534 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1535 cbz(r3, L);
1536 call_VM(noreg,
1537 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1538 bind(L);
1539 pop(state);
1540 }
1541
1542 if (DTraceMethodProbes) {
1543 push(state);
1544 get_method(c_rarg1);
1545 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1546 rthread, c_rarg1);
1547 pop(state);
1548 }
1549 }
1550
1551
1552 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1553 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1554 int increment, Address mask,
1555 Register scratch, Register scratch2,
1556 bool preloaded, Condition cond,
1557 Label* where) {
1558 if (!preloaded) {
1559 ldrw(scratch, counter_addr);
1560 }
1561 add(scratch, scratch, increment);
1562 strw(scratch, counter_addr);
1563 ldrw(scratch2, mask);
1564 ands(scratch, scratch, scratch2);
1565 br(cond, *where);
1566 }
1567
1568 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1569 int number_of_arguments) {
1570 // interpreter specific
1571 //
1572 // Note: No need to save/restore rbcp & rlocals pointer since these
1573 // are callee saved registers and no blocking/ GC can happen
1574 // in leaf calls.
1575 #ifdef ASSERT
1576 {
1577 Label L;
1578 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1579 cbz(rscratch1, L);
1580 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1581 " last_sp != nullptr");
1582 bind(L);
1583 }
1584 #endif /* ASSERT */
1585 // super call
1586 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1587 }
1588
1589 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1590 Register java_thread,
1591 Register last_java_sp,
1592 Label* return_pc,
1593 address entry_point,
1594 int number_of_arguments,
1595 bool check_exceptions) {
1596 // interpreter specific
1597 //
1598 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1599 // really make a difference for these runtime calls, since they are
1600 // slow anyway. Btw., bcp must be saved/restored since it may change
1601 // due to GC.
1602 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1603 save_bcp();
1604 #ifdef ASSERT
1605 {
1606 Label L;
1607 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1608 cbz(rscratch1, L);
1609 stop("InterpreterMacroAssembler::call_VM_base:"
1610 " last_sp != nullptr");
1611 bind(L);
1612 }
1613 #endif /* ASSERT */
1614 // super call
1615 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1616 return_pc, entry_point,
1617 number_of_arguments, check_exceptions);
1618 // interpreter specific
1619 restore_bcp();
1620 restore_locals();
1621 }
1622
1623 void InterpreterMacroAssembler::call_VM_preemptable_helper(Register oop_result,
1624 address entry_point,
1625 int number_of_arguments,
1626 bool check_exceptions) {
1627 assert(InterpreterRuntime::is_preemptable_call(entry_point), "VM call not preemptable, should use call_VM()");
1628 Label resume_pc, not_preempted;
1629
1630 #ifdef ASSERT
1631 {
1632 Label L1, L2;
1633 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1634 cbz(rscratch1, L1);
1635 stop("call_VM_preemptable_helper: Should not have alternate return address set");
1636 bind(L1);
1637 // We check this counter in patch_return_pc_with_preempt_stub() during freeze.
1638 incrementw(Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1639 ldrw(rscratch1, Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1640 cmpw(rscratch1, 0);
1641 br(Assembler::GT, L2);
1642 stop("call_VM_preemptable_helper: should be > 0");
1643 bind(L2);
1644 }
1645 #endif /* ASSERT */
1646
1647 // Force freeze slow path.
1648 push_cont_fastpath();
1649
1650 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1651 // Note: call_VM_base will use resume_pc label to set last_Java_pc.
1652 call_VM_base(noreg, noreg, noreg, &resume_pc, entry_point, number_of_arguments, false /*check_exceptions*/);
1653
1654 pop_cont_fastpath();
1655
1656 #ifdef ASSERT
1657 {
1658 Label L;
1659 decrementw(Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1660 ldrw(rscratch1, Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1661 cmpw(rscratch1, 0);
1662 br(Assembler::GE, L);
1663 stop("call_VM_preemptable_helper: should be >= 0");
1664 bind(L);
1665 }
1666 #endif /* ASSERT */
1667
1668 // Check if preempted.
1669 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1670 cbz(rscratch1, not_preempted);
1671 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1672 br(rscratch1);
1673
1674 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1675 bind(resume_pc);
1676 restore_after_resume(false /* is_native */);
1677
1678 bind(not_preempted);
1679 if (check_exceptions) {
1680 // check for pending exceptions
1681 ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1682 Label ok;
1683 cbz(rscratch1, ok);
1684 lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
1685 br(rscratch1);
1686 bind(ok);
1687 }
1688
1689 // get oop result if there is one and reset the value in the thread
1690 if (oop_result->is_valid()) {
1691 get_vm_result_oop(oop_result, rthread);
1692 }
1693 }
1694
1695 static void pass_arg1(MacroAssembler* masm, Register arg) {
1696 if (c_rarg1 != arg ) {
1697 masm->mov(c_rarg1, arg);
1698 }
1699 }
1700
1701 static void pass_arg2(MacroAssembler* masm, Register arg) {
1702 if (c_rarg2 != arg ) {
1703 masm->mov(c_rarg2, arg);
1704 }
1705 }
1706
1707 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1708 address entry_point,
1709 Register arg_1,
1710 bool check_exceptions) {
1711 pass_arg1(this, arg_1);
1712 call_VM_preemptable_helper(oop_result, entry_point, 1, check_exceptions);
1713 }
1714
1715 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1716 address entry_point,
1717 Register arg_1,
1718 Register arg_2,
1719 bool check_exceptions) {
1720 LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
1721 pass_arg2(this, arg_2);
1722 pass_arg1(this, arg_1);
1723 call_VM_preemptable_helper(oop_result, entry_point, 2, check_exceptions);
1724 }
1725
1726 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1727 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1728 blr(rscratch1);
1729 if (is_native) {
1730 // On resume we need to set up stack as expected
1731 push(dtos);
1732 push(ltos);
1733 }
1734 }
1735
1736 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1737 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1738 Label update, next, none;
1739
1740 verify_oop(obj);
1741
1742 cbnz(obj, update);
1743 orptr(mdo_addr, TypeEntries::null_seen);
1744 b(next);
1745
1746 bind(update);
1747 load_klass(obj, obj);
1748
1749 ldr(rscratch1, mdo_addr);
1750 eor(obj, obj, rscratch1);
1751 tst(obj, TypeEntries::type_klass_mask);
1752 br(Assembler::EQ, next); // klass seen before, nothing to
1753 // do. The unknown bit may have been
1754 // set already but no need to check.
1755
1756 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1757 // already unknown. Nothing to do anymore.
1758
1759 cbz(rscratch1, none);
1760 cmp(rscratch1, (u1)TypeEntries::null_seen);
1761 br(Assembler::EQ, none);
1762 // There is a chance that the checks above
1763 // fail if another thread has just set the
1764 // profiling to this obj's klass
1765 eor(obj, obj, rscratch1); // get back original value before XOR
1766 ldr(rscratch1, mdo_addr);
1767 eor(obj, obj, rscratch1);
1768 tst(obj, TypeEntries::type_klass_mask);
1769 br(Assembler::EQ, next);
1770
1771 // different than before. Cannot keep accurate profile.
1772 orptr(mdo_addr, TypeEntries::type_unknown);
1773 b(next);
1774
1775 bind(none);
1776 // first time here. Set profile type.
1777 str(obj, mdo_addr);
1778 #ifdef ASSERT
1779 andr(obj, obj, TypeEntries::type_mask);
1780 verify_klass_ptr(obj);
1781 #endif
1782
1783 bind(next);
1784 }
1785
1786 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1787 if (!ProfileInterpreter) {
1788 return;
1789 }
1790
1791 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1792 Label profile_continue;
1793
1794 test_method_data_pointer(mdp, profile_continue);
1795
1796 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1797
1798 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1799 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1800 br(Assembler::NE, profile_continue);
1801
1802 if (MethodData::profile_arguments()) {
1803 Label done;
1804 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1805
1806 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1807 if (i > 0 || MethodData::profile_return()) {
1808 // If return value type is profiled we may have no argument to profile
1809 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1810 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1811 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1812 add(rscratch1, mdp, off_to_args);
1813 br(Assembler::LT, done);
1814 }
1815 ldr(tmp, Address(callee, Method::const_offset()));
1816 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1817 // stack offset o (zero based) from the start of the argument
1818 // list, for n arguments translates into offset n - o - 1 from
1819 // the end of the argument list
1820 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1821 sub(tmp, tmp, rscratch1);
1822 sub(tmp, tmp, 1);
1823 Address arg_addr = argument_address(tmp);
1824 ldr(tmp, arg_addr);
1825
1826 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1827 profile_obj_type(tmp, mdo_arg_addr);
1828
1829 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1830 off_to_args += to_add;
1831 }
1832
1833 if (MethodData::profile_return()) {
1834 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1835 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1836 }
1837
1838 add(rscratch1, mdp, off_to_args);
1839 bind(done);
1840 mov(mdp, rscratch1);
1841
1842 if (MethodData::profile_return()) {
1843 // We're right after the type profile for the last
1844 // argument. tmp is the number of cells left in the
1845 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1846 // if there's a return to profile.
1847 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1848 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1849 }
1850 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1851 } else {
1852 assert(MethodData::profile_return(), "either profile call args or call ret");
1853 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1854 }
1855
1856 // mdp points right after the end of the
1857 // CallTypeData/VirtualCallTypeData, right after the cells for the
1858 // return value type if there's one
1859
1860 bind(profile_continue);
1861 }
1862 }
1863
1864 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1865 assert_different_registers(mdp, ret, tmp, rbcp);
1866 if (ProfileInterpreter && MethodData::profile_return()) {
1867 Label profile_continue, done;
1868
1869 test_method_data_pointer(mdp, profile_continue);
1870
1871 if (MethodData::profile_return_jsr292_only()) {
1872 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1873
1874 // If we don't profile all invoke bytecodes we must make sure
1875 // it's a bytecode we indeed profile. We can't go back to the
1876 // beginning of the ProfileData we intend to update to check its
1877 // type because we're right after it and we don't known its
1878 // length
1879 Label do_profile;
1880 ldrb(rscratch1, Address(rbcp, 0));
1881 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1882 br(Assembler::EQ, do_profile);
1883 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1884 br(Assembler::EQ, do_profile);
1885 get_method(tmp);
1886 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1887 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1888 br(Assembler::NE, profile_continue);
1889
1890 bind(do_profile);
1891 }
1892
1893 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1894 mov(tmp, ret);
1895 profile_obj_type(tmp, mdo_ret_addr);
1896
1897 bind(profile_continue);
1898 }
1899 }
1900
1901 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1902 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1903 if (ProfileInterpreter && MethodData::profile_parameters()) {
1904 Label profile_continue, done;
1905
1906 test_method_data_pointer(mdp, profile_continue);
1907
1908 // Load the offset of the area within the MDO used for
1909 // parameters. If it's negative we're not profiling any parameters
1910 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1911 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1912
1913 // Compute a pointer to the area for parameters from the offset
1914 // and move the pointer to the slot for the last
1915 // parameters. Collect profiling from last parameter down.
1916 // mdo start + parameters offset + array length - 1
1917 add(mdp, mdp, tmp1);
1918 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1919 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1920
1921 Label loop;
1922 bind(loop);
1923
1924 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1925 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1926 int per_arg_scale = exact_log2(DataLayout::cell_size);
1927 add(rscratch1, mdp, off_base);
1928 add(rscratch2, mdp, type_base);
1929
1930 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1931 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1932
1933 // load offset on the stack from the slot for this parameter
1934 ldr(tmp2, arg_off);
1935 neg(tmp2, tmp2);
1936 // read the parameter from the local area
1937 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1938
1939 // profile the parameter
1940 profile_obj_type(tmp2, arg_type);
1941
1942 // go to next parameter
1943 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1944 br(Assembler::GE, loop);
1945
1946 bind(profile_continue);
1947 }
1948 }
1949
1950 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1951 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1952 get_cache_index_at_bcp(index, 1, sizeof(u4));
1953 // Get address of invokedynamic array
1954 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1955 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1956 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1957 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1958 lea(cache, Address(cache, index));
1959 }
1960
1961 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1962 // Get index out of bytecode pointer
1963 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1964 // Take shortcut if the size is a power of 2
1965 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1966 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1967 } else {
1968 mov(cache, sizeof(ResolvedFieldEntry));
1969 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1970 }
1971 // Get address of field entries array
1972 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1973 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1974 lea(cache, Address(cache, index));
1975 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1976 membar(MacroAssembler::LoadLoad);
1977 }
1978
1979 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1980 // Get index out of bytecode pointer
1981 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1982 mov(cache, sizeof(ResolvedMethodEntry));
1983 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1984
1985 // Get address of field entries array
1986 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1987 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1988 lea(cache, Address(cache, index));
1989 }
1990
1991 #ifdef ASSERT
1992 void InterpreterMacroAssembler::verify_field_offset(Register reg) {
1993 // Verify the field offset is not in the header, implicitly checks for 0
1994 Label L;
1995 subs(zr, reg, oopDesc::base_offset_in_bytes());
1996 br(Assembler::GE, L);
1997 stop("bad field offset");
1998 bind(L);
1999 }
2000 #endif