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 (DTraceMethodProbes) {
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,
226 Register field_index, Register field_offset,
227 Register temp, Register obj) {
228 Label alloc_failed, done;
229 const Register src = field_offset;
230 const Register alloc_temp = r10;
231 const Register dst_temp = field_index;
232 const Register layout_info = temp;
233 assert_different_registers(obj, entry, field_index, field_offset, temp, alloc_temp);
234
235 // Grab the inline field klass
236 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
237 inline_layout_info(rscratch1, field_index, layout_info);
238
239 const Register field_klass = dst_temp;
240 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
241
242 // allocate buffer
243 push(obj); // save holder
244 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, alloc_failed);
245
246 // Have an oop instance buffer, copy into it
247 payload_address(obj, dst_temp, field_klass); // danger, uses rscratch1
248 pop(alloc_temp); // restore holder
249 lea(src, Address(alloc_temp, field_offset));
250 // call_VM_leaf, clobbers a few regs, save restore new obj
251 push(obj);
252 flat_field_copy(IS_DEST_UNINITIALIZED, src, dst_temp, layout_info);
253 pop(obj);
254 b(done);
255
256 bind(alloc_failed);
257 pop(obj);
258 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
259 obj, entry);
260
261 bind(done);
262 membar(Assembler::StoreStore);
263 }
264
265 // Load object from cpool->resolved_references(index)
266 void InterpreterMacroAssembler::load_resolved_reference_at_index(
267 Register result, Register index, Register tmp) {
268 assert_different_registers(result, index);
269
270 get_constant_pool(result);
271 // load pointer for resolved_references[] objArray
272 ldr(result, Address(result, ConstantPool::cache_offset()));
273 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
274 resolve_oop_handle(result, tmp, rscratch2);
275 // Add in the index
276 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
277 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
278 }
279
280 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
281 Register cpool, Register index, Register klass, Register temp) {
282 add(temp, cpool, index, LSL, LogBytesPerWord);
283 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
284 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
285 add(klass, klass, temp, LSL, LogBytesPerWord);
286 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
287 }
288
289 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
290 // subtype of super_klass.
291 //
292 // Args:
293 // r0: superklass
294 // Rsub_klass: subklass
295 //
296 // Kills:
297 // r2, r5
298 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
299 Label& ok_is_subtype,
300 bool profile) {
301 assert(Rsub_klass != r0, "r0 holds superklass");
302 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
303 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
304
305 // Profile the not-null value's klass.
306 if (profile) {
307 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
308 }
309
310 // Do the check.
311 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
312 }
313
314 // Java Expression Stack
315
316 void InterpreterMacroAssembler::pop_ptr(Register r) {
317 ldr(r, post(esp, wordSize));
318 }
319
320 void InterpreterMacroAssembler::pop_i(Register r) {
321 ldrw(r, post(esp, wordSize));
322 }
323
324 void InterpreterMacroAssembler::pop_l(Register r) {
325 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
326 }
327
328 void InterpreterMacroAssembler::push_ptr(Register r) {
329 str(r, pre(esp, -wordSize));
330 }
331
332 void InterpreterMacroAssembler::push_i(Register r) {
333 str(r, pre(esp, -wordSize));
334 }
335
336 void InterpreterMacroAssembler::push_l(Register r) {
337 str(zr, pre(esp, -wordSize));
338 str(r, pre(esp, - wordSize));
339 }
340
341 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
342 ldrs(r, post(esp, wordSize));
343 }
344
345 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
346 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
347 }
348
349 void InterpreterMacroAssembler::push_f(FloatRegister r) {
350 strs(r, pre(esp, -wordSize));
351 }
352
353 void InterpreterMacroAssembler::push_d(FloatRegister r) {
354 strd(r, pre(esp, 2* -wordSize));
355 }
356
357 void InterpreterMacroAssembler::pop(TosState state) {
358 switch (state) {
359 case atos: pop_ptr(); break;
360 case btos:
361 case ztos:
362 case ctos:
363 case stos:
364 case itos: pop_i(); break;
365 case ltos: pop_l(); break;
366 case ftos: pop_f(); break;
367 case dtos: pop_d(); break;
368 case vtos: /* nothing to do */ break;
369 default: ShouldNotReachHere();
370 }
371 interp_verify_oop(r0, state);
372 }
373
374 void InterpreterMacroAssembler::push(TosState state) {
375 interp_verify_oop(r0, state);
376 switch (state) {
377 case atos: push_ptr(); break;
378 case btos:
379 case ztos:
380 case ctos:
381 case stos:
382 case itos: push_i(); break;
383 case ltos: push_l(); break;
384 case ftos: push_f(); break;
385 case dtos: push_d(); break;
386 case vtos: /* nothing to do */ break;
387 default : ShouldNotReachHere();
388 }
389 }
390
391 // Helpers for swap and dup
392 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
393 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
394 }
395
396 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
397 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
398 }
399
400 void InterpreterMacroAssembler::load_float(Address src) {
401 ldrs(v0, src);
402 }
403
404 void InterpreterMacroAssembler::load_double(Address src) {
405 ldrd(v0, src);
406 }
407
408 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
409 // set sender sp
410 mov(r19_sender_sp, sp);
411 // record last_sp
412 sub(rscratch1, esp, rfp);
413 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
414 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
415 }
416
417 // Jump to from_interpreted entry of a call unless single stepping is possible
418 // in this thread in which case we must call the i2i entry
419 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
420 prepare_to_jump_from_interpreted();
421
422 if (JvmtiExport::can_post_interpreter_events()) {
423 Label run_compiled_code;
424 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
425 // compiled code in threads for which the event is enabled. Check here for
426 // interp_only_mode if these events CAN be enabled.
427 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
428 cbzw(rscratch1, run_compiled_code);
429 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
430 br(rscratch1);
431 bind(run_compiled_code);
432 }
433
434 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
435 br(rscratch1);
436 }
437
438 // The following two routines provide a hook so that an implementation
439 // can schedule the dispatch in two parts. amd64 does not do this.
440 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
441 }
442
443 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
444 dispatch_next(state, step);
445 }
446
447 void InterpreterMacroAssembler::dispatch_base(TosState state,
448 address* table,
449 bool verifyoop,
450 bool generate_poll) {
451 if (VerifyActivationFrameSize) {
452 Label L;
453 sub(rscratch2, rfp, esp);
454 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
455 subs(rscratch2, rscratch2, min_frame_size);
456 br(Assembler::GE, L);
457 stop("broken stack frame");
458 bind(L);
459 }
460 if (verifyoop) {
461 interp_verify_oop(r0, state);
462 }
463
464 Label safepoint;
465 address* const safepoint_table = Interpreter::safept_table(state);
466 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
467
468 if (needs_thread_local_poll) {
469 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
470 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
471 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
472 }
473
474 if (table == Interpreter::dispatch_table(state)) {
475 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
476 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
477 } else {
478 mov(rscratch2, (address)table);
479 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
480 }
481 br(rscratch2);
482
483 if (needs_thread_local_poll) {
484 bind(safepoint);
485 lea(rscratch2, ExternalAddress((address)safepoint_table));
486 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
487 br(rscratch2);
488 }
489 }
490
491 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
492 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
493 }
494
495 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
496 dispatch_base(state, Interpreter::normal_table(state));
497 }
498
499 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
500 dispatch_base(state, Interpreter::normal_table(state), false);
501 }
502
503
504 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
505 // load next bytecode
506 ldrb(rscratch1, Address(pre(rbcp, step)));
507 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
508 }
509
510 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
511 // load current bytecode
512 ldrb(rscratch1, Address(rbcp, 0));
513 dispatch_base(state, table);
514 }
515
516 // remove activation
517 //
518 // Apply stack watermark barrier.
519 // Unlock the receiver if this is a synchronized method.
520 // Unlock any Java monitors from synchronized blocks.
521 // Remove the activation from the stack.
522 //
523 // If there are locked Java monitors
524 // If throw_monitor_exception
525 // throws IllegalMonitorStateException
526 // Else if install_monitor_exception
527 // installs IllegalMonitorStateException
528 // Else
529 // no error processing
530 void InterpreterMacroAssembler::remove_activation(
531 TosState state,
532 bool throw_monitor_exception,
533 bool install_monitor_exception,
534 bool notify_jvmdi) {
535 // Note: Registers r3 xmm0 may be in use for the
536 // result check if synchronized method
537 Label unlocked, unlock, no_unlock;
538
539 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
540 // that would normally not be safe to use. Such bad returns into unsafe territory of
541 // the stack, will call InterpreterRuntime::at_unwind.
542 Label slow_path;
543 Label fast_path;
544 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
545 br(Assembler::AL, fast_path);
546 bind(slow_path);
547 push(state);
548 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
549 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
550 reset_last_Java_frame(true);
551 pop(state);
552 bind(fast_path);
553
554 // get the value of _do_not_unlock_if_synchronized into r3
555 const Address do_not_unlock_if_synchronized(rthread,
556 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
557 ldrb(r3, do_not_unlock_if_synchronized);
558 strb(zr, do_not_unlock_if_synchronized); // reset the flag
559
560 // get method access flags
561 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
562 ldrh(r2, Address(r1, Method::access_flags_offset()));
563 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
564
565 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
566 // is set.
567 cbnz(r3, no_unlock);
568
569 // unlock monitor
570 push(state); // save result
571
572 // BasicObjectLock will be first in list, since this is a
573 // synchronized method. However, need to check that the object has
574 // not been unlocked by an explicit monitorexit bytecode.
575 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
576 wordSize - (int) sizeof(BasicObjectLock));
577 // We use c_rarg1 so that if we go slow path it will be the correct
578 // register for unlock_object to pass to VM directly
579 lea(c_rarg1, monitor); // address of first monitor
580
581 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
582 cbnz(r0, unlock);
583
584 pop(state);
585 if (throw_monitor_exception) {
586 // Entry already unlocked, need to throw exception
587 call_VM(noreg, CAST_FROM_FN_PTR(address,
588 InterpreterRuntime::throw_illegal_monitor_state_exception));
589 should_not_reach_here();
590 } else {
591 // Monitor already unlocked during a stack unroll. If requested,
592 // install an illegal_monitor_state_exception. Continue with
593 // stack unrolling.
594 if (install_monitor_exception) {
595 call_VM(noreg, CAST_FROM_FN_PTR(address,
596 InterpreterRuntime::new_illegal_monitor_state_exception));
597 }
598 b(unlocked);
599 }
600
601 bind(unlock);
602 unlock_object(c_rarg1);
603 pop(state);
604
605 // Check that for block-structured locking (i.e., that all locked
606 // objects has been unlocked)
607 bind(unlocked);
608
609 // r0: Might contain return value
610
611 // Check that all monitors are unlocked
612 {
613 Label loop, exception, entry, restart;
614 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
615 const Address monitor_block_top(
616 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
617 const Address monitor_block_bot(
618 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
619
620 bind(restart);
621 // We use c_rarg1 so that if we go slow path it will be the correct
622 // register for unlock_object to pass to VM directly
623 ldr(c_rarg1, monitor_block_top); // derelativize pointer
624 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
625 // c_rarg1 points to current entry, starting with top-most entry
626
627 lea(r19, monitor_block_bot); // points to word before bottom of
628 // monitor block
629 b(entry);
630
631 // Entry already locked, need to throw exception
632 bind(exception);
633
634 if (throw_monitor_exception) {
635 // Throw exception
636 MacroAssembler::call_VM(noreg,
637 CAST_FROM_FN_PTR(address, InterpreterRuntime::
638 throw_illegal_monitor_state_exception));
639 should_not_reach_here();
640 } else {
641 // Stack unrolling. Unlock object and install illegal_monitor_exception.
642 // Unlock does not block, so don't have to worry about the frame.
643 // We don't have to preserve c_rarg1 since we are going to throw an exception.
644
645 push(state);
646 unlock_object(c_rarg1);
647 pop(state);
648
649 if (install_monitor_exception) {
650 call_VM(noreg, CAST_FROM_FN_PTR(address,
651 InterpreterRuntime::
652 new_illegal_monitor_state_exception));
653 }
654
655 b(restart);
656 }
657
658 bind(loop);
659 // check if current entry is used
660 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
661 cbnz(rscratch1, exception);
662
663 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
664 bind(entry);
665 cmp(c_rarg1, r19); // check if bottom reached
666 br(Assembler::NE, loop); // if not at bottom then check this entry
667 }
668
669 bind(no_unlock);
670
671 // jvmti support
672 if (notify_jvmdi) {
673 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
674 } else {
675 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
676 }
677
678 // remove activation
679 // get sender esp
680 ldr(rscratch2,
681 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
682
683 if (StackReservedPages > 0) {
684 // testing if reserved zone needs to be re-enabled
685 Label no_reserved_zone_enabling;
686
687 // check if already enabled - if so no re-enabling needed
688 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
689 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
690 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
691 br(Assembler::EQ, no_reserved_zone_enabling);
692
693 // look for an overflow into the stack reserved zone, i.e.
694 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
695 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
696 cmp(rscratch2, rscratch1);
697 br(Assembler::LS, no_reserved_zone_enabling);
698
699 call_VM_leaf(
700 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
701 call_VM(noreg, CAST_FROM_FN_PTR(address,
702 InterpreterRuntime::throw_delayed_StackOverflowError));
703 should_not_reach_here();
704
705 bind(no_reserved_zone_enabling);
706 }
707
708 if (state == atos && InlineTypeReturnedAsFields) {
709 // Check if we are returning an non-null inline type and load its fields into registers
710 Label skip;
711 test_oop_is_not_inline_type(r0, rscratch2, skip);
712
713 // Load fields from a buffered value with an inline class specific handler
714 load_klass(rscratch1 /*dst*/, r0 /*src*/);
715 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
716 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
717 // Unpack handler can be null if inline type is not scalarizable in returns
718 cbz(rscratch1, skip);
719
720 blr(rscratch1);
721 #ifdef ASSERT
722 // TODO 8284443 Enable
723 if (StressCallingConvention && false) {
724 Label skip_stress;
725 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
726 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
727 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
728 br(Assembler::EQ, skip_stress);
729 load_klass(r0, r0);
730 orr(r0, r0, 1);
731 bind(skip_stress);
732 }
733 #endif
734 bind(skip);
735 // Check above kills sender esp in rscratch2. Reload it.
736 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
737 }
738
739 // restore sender esp
740 mov(esp, rscratch2);
741 // remove frame anchor
742 leave();
743 // If we're returning to interpreted code we will shortly be
744 // adjusting SP to allow some space for ESP. If we're returning to
745 // compiled code the saved sender SP was saved in sender_sp, so this
746 // restores it.
747 andr(sp, esp, -16);
748 }
749
750 // Lock object
751 //
752 // Args:
753 // c_rarg1: BasicObjectLock to be used for locking
754 //
755 // Kills:
756 // r0
757 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
758 // rscratch1, rscratch2 (scratch regs)
759 void InterpreterMacroAssembler::lock_object(Register lock_reg)
760 {
761 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
762 if (LockingMode == LM_MONITOR) {
763 call_VM_preemptable(noreg,
764 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
765 lock_reg);
766 } else {
767 Label count, done;
768
769 const Register swap_reg = r0;
770 const Register tmp = c_rarg2;
771 const Register obj_reg = c_rarg3; // Will contain the oop
772 const Register tmp2 = c_rarg4;
773 const Register tmp3 = c_rarg5;
774
775 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
776 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
777 const int mark_offset = lock_offset +
778 BasicLock::displaced_header_offset_in_bytes();
779
780 Label slow_case;
781
782 // Load object pointer into obj_reg %c_rarg3
783 ldr(obj_reg, Address(lock_reg, obj_offset));
784
785 if (DiagnoseSyncOnValueBasedClasses != 0) {
786 load_klass(tmp, obj_reg);
787 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
788 tst(tmp, KlassFlags::_misc_is_value_based_class);
789 br(Assembler::NE, slow_case);
790 }
791
792 if (LockingMode == LM_LIGHTWEIGHT) {
793 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
794 b(done);
795 } else if (LockingMode == LM_LEGACY) {
796 // Load (object->mark() | 1) into swap_reg
797 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
798 orr(swap_reg, rscratch1, 1);
799 if (EnableValhalla) {
800 // Mask inline_type bit such that we go to the slow path if object is an inline type
801 andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
802 }
803
804 // Save (object->mark() | 1) into BasicLock's displaced header
805 str(swap_reg, Address(lock_reg, mark_offset));
806
807 assert(lock_offset == 0,
808 "displached header must be first word in BasicObjectLock");
809
810 Label fail;
811 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
812
813 // Fast check for recursive lock.
814 //
815 // Can apply the optimization only if this is a stack lock
816 // allocated in this thread. For efficiency, we can focus on
817 // recently allocated stack locks (instead of reading the stack
818 // base and checking whether 'mark' points inside the current
819 // thread stack):
820 // 1) (mark & 7) == 0, and
821 // 2) sp <= mark < mark + os::pagesize()
822 //
823 // Warning: sp + os::pagesize can overflow the stack base. We must
824 // neither apply the optimization for an inflated lock allocated
825 // just above the thread stack (this is why condition 1 matters)
826 // nor apply the optimization if the stack lock is inside the stack
827 // of another thread. The latter is avoided even in case of overflow
828 // because we have guard pages at the end of all stacks. Hence, if
829 // we go over the stack base and hit the stack of another thread,
830 // this should not be in a writeable area that could contain a
831 // stack lock allocated by that thread. As a consequence, a stack
832 // lock less than page size away from sp is guaranteed to be
833 // owned by the current thread.
834 //
835 // These 3 tests can be done by evaluating the following
836 // expression: ((mark - sp) & (7 - os::vm_page_size())),
837 // assuming both stack pointer and pagesize have their
838 // least significant 3 bits clear.
839 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
840 // NOTE2: aarch64 does not like to subtract sp from rn so take a
841 // copy
842 mov(rscratch1, sp);
843 sub(swap_reg, swap_reg, rscratch1);
844 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
845
846 // Save the test result, for recursive case, the result is zero
847 str(swap_reg, Address(lock_reg, mark_offset));
848 br(Assembler::NE, slow_case);
849
850 bind(count);
851 inc_held_monitor_count(rscratch1);
852 b(done);
853 }
854 bind(slow_case);
855
856 // Call the runtime routine for slow case
857 call_VM_preemptable(noreg,
858 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
859 lock_reg);
860
861 bind(done);
862 }
863 }
864
865
866 // Unlocks an object. Used in monitorexit bytecode and
867 // remove_activation. Throws an IllegalMonitorException if object is
868 // not locked by current thread.
869 //
870 // Args:
871 // c_rarg1: BasicObjectLock for lock
872 //
873 // Kills:
874 // r0
875 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
876 // rscratch1, rscratch2 (scratch regs)
877 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
878 {
879 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
880
881 if (LockingMode == LM_MONITOR) {
882 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
883 } else {
884 Label count, done;
885
886 const Register swap_reg = r0;
887 const Register header_reg = c_rarg2; // Will contain the old oopMark
888 const Register obj_reg = c_rarg3; // Will contain the oop
889 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
890
891 save_bcp(); // Save in case of exception
892
893 if (LockingMode != LM_LIGHTWEIGHT) {
894 // Convert from BasicObjectLock structure to object and BasicLock
895 // structure Store the BasicLock address into %r0
896 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
897 }
898
899 // Load oop into obj_reg(%c_rarg3)
900 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
901
902 // Free entry
903 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
904
905 Label slow_case;
906 if (LockingMode == LM_LIGHTWEIGHT) {
907 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
908 b(done);
909 } else if (LockingMode == LM_LEGACY) {
910 // Load the old header from BasicLock structure
911 ldr(header_reg, Address(swap_reg,
912 BasicLock::displaced_header_offset_in_bytes()));
913
914 // Test for recursion
915 cbz(header_reg, count);
916
917 // Atomic swap back the old header
918 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
919
920 bind(count);
921 dec_held_monitor_count(rscratch1);
922 b(done);
923 }
924
925 bind(slow_case);
926 // Call the runtime routine for slow case.
927 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
928 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
929 bind(done);
930 restore_bcp();
931 }
932 }
933
934 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
935 Label& zero_continue) {
936 assert(ProfileInterpreter, "must be profiling interpreter");
937 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
938 cbz(mdp, zero_continue);
939 }
940
941 // Set the method data pointer for the current bcp.
942 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
943 assert(ProfileInterpreter, "must be profiling interpreter");
944 Label set_mdp;
945 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
946
947 // Test MDO to avoid the call if it is null.
948 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
949 cbz(r0, set_mdp);
950 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
951 // r0: mdi
952 // mdo is guaranteed to be non-zero here, we checked for it before the call.
953 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
954 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
955 add(r0, r1, r0);
956 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
957 bind(set_mdp);
958 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
959 }
960
961 void InterpreterMacroAssembler::verify_method_data_pointer() {
962 assert(ProfileInterpreter, "must be profiling interpreter");
963 #ifdef ASSERT
964 Label verify_continue;
965 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
966 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
967 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
968 get_method(r1);
969
970 // If the mdp is valid, it will point to a DataLayout header which is
971 // consistent with the bcp. The converse is highly probable also.
972 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
973 ldr(rscratch1, Address(r1, Method::const_offset()));
974 add(r2, r2, rscratch1, Assembler::LSL);
975 lea(r2, Address(r2, ConstMethod::codes_offset()));
976 cmp(r2, rbcp);
977 br(Assembler::EQ, verify_continue);
978 // r1: method
979 // rbcp: bcp // rbcp == 22
980 // r3: mdp
981 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
982 r1, rbcp, r3);
983 bind(verify_continue);
984 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
985 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
986 #endif // ASSERT
987 }
988
989
990 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
991 int constant,
992 Register value) {
993 assert(ProfileInterpreter, "must be profiling interpreter");
994 Address data(mdp_in, constant);
995 str(value, data);
996 }
997
998
999 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1000 int constant,
1001 bool decrement) {
1002 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1003 }
1004
1005 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1006 Register reg,
1007 int constant,
1008 bool decrement) {
1009 assert(ProfileInterpreter, "must be profiling interpreter");
1010 // %%% this does 64bit counters at best it is wasting space
1011 // at worst it is a rare bug when counters overflow
1012
1013 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1014
1015 Address addr1(mdp_in, constant);
1016 Address addr2(rscratch2, reg, Address::lsl(0));
1017 Address &addr = addr1;
1018 if (reg != noreg) {
1019 lea(rscratch2, addr1);
1020 addr = addr2;
1021 }
1022
1023 if (decrement) {
1024 // Decrement the register. Set condition codes.
1025 // Intel does this
1026 // addptr(data, (int32_t) -DataLayout::counter_increment);
1027 // If the decrement causes the counter to overflow, stay negative
1028 // Label L;
1029 // jcc(Assembler::negative, L);
1030 // addptr(data, (int32_t) DataLayout::counter_increment);
1031 // so we do this
1032 ldr(rscratch1, addr);
1033 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1034 Label L;
1035 br(Assembler::LO, L); // skip store if counter underflow
1036 str(rscratch1, addr);
1037 bind(L);
1038 } else {
1039 assert(DataLayout::counter_increment == 1,
1040 "flow-free idiom only works with 1");
1041 // Intel does this
1042 // Increment the register. Set carry flag.
1043 // addptr(data, DataLayout::counter_increment);
1044 // If the increment causes the counter to overflow, pull back by 1.
1045 // sbbptr(data, (int32_t)0);
1046 // so we do this
1047 ldr(rscratch1, addr);
1048 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1049 Label L;
1050 br(Assembler::CS, L); // skip store if counter overflow
1051 str(rscratch1, addr);
1052 bind(L);
1053 }
1054 }
1055
1056 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1057 int flag_byte_constant) {
1058 assert(ProfileInterpreter, "must be profiling interpreter");
1059 int flags_offset = in_bytes(DataLayout::flags_offset());
1060 // Set the flag
1061 ldrb(rscratch1, Address(mdp_in, flags_offset));
1062 orr(rscratch1, rscratch1, flag_byte_constant);
1063 strb(rscratch1, Address(mdp_in, flags_offset));
1064 }
1065
1066
1067 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1068 int offset,
1069 Register value,
1070 Register test_value_out,
1071 Label& not_equal_continue) {
1072 assert(ProfileInterpreter, "must be profiling interpreter");
1073 if (test_value_out == noreg) {
1074 ldr(rscratch1, Address(mdp_in, offset));
1075 cmp(value, rscratch1);
1076 } else {
1077 // Put the test value into a register, so caller can use it:
1078 ldr(test_value_out, Address(mdp_in, offset));
1079 cmp(value, test_value_out);
1080 }
1081 br(Assembler::NE, not_equal_continue);
1082 }
1083
1084
1085 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1086 int offset_of_disp) {
1087 assert(ProfileInterpreter, "must be profiling interpreter");
1088 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1089 add(mdp_in, mdp_in, rscratch1, LSL);
1090 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1091 }
1092
1093
1094 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1095 Register reg,
1096 int offset_of_disp) {
1097 assert(ProfileInterpreter, "must be profiling interpreter");
1098 lea(rscratch1, Address(mdp_in, offset_of_disp));
1099 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1100 add(mdp_in, mdp_in, rscratch1, LSL);
1101 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1102 }
1103
1104
1105 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1106 int constant) {
1107 assert(ProfileInterpreter, "must be profiling interpreter");
1108 add(mdp_in, mdp_in, (unsigned)constant);
1109 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1110 }
1111
1112
1113 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1114 assert(ProfileInterpreter, "must be profiling interpreter");
1115 // save/restore across call_VM
1116 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1117 call_VM(noreg,
1118 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1119 return_bci);
1120 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1121 }
1122
1123
1124 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1125 Register bumped_count) {
1126 if (ProfileInterpreter) {
1127 Label profile_continue;
1128
1129 // If no method data exists, go to profile_continue.
1130 // Otherwise, assign to mdp
1131 test_method_data_pointer(mdp, profile_continue);
1132
1133 // We are taking a branch. Increment the taken count.
1134 // We inline increment_mdp_data_at to return bumped_count in a register
1135 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1136 Address data(mdp, in_bytes(JumpData::taken_offset()));
1137 ldr(bumped_count, data);
1138 assert(DataLayout::counter_increment == 1,
1139 "flow-free idiom only works with 1");
1140 // Intel does this to catch overflow
1141 // addptr(bumped_count, DataLayout::counter_increment);
1142 // sbbptr(bumped_count, 0);
1143 // so we do this
1144 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1145 Label L;
1146 br(Assembler::CS, L); // skip store if counter overflow
1147 str(bumped_count, data);
1148 bind(L);
1149 // The method data pointer needs to be updated to reflect the new target.
1150 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1151 bind(profile_continue);
1152 }
1153 }
1154
1155
1156 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1157 if (ProfileInterpreter) {
1158 Label profile_continue;
1159
1160 // If no method data exists, go to profile_continue.
1161 test_method_data_pointer(mdp, profile_continue);
1162
1163 // We are taking a branch. Increment the not taken count.
1164 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1165
1166 // The method data pointer needs to be updated to correspond to
1167 // the next bytecode
1168 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1169 bind(profile_continue);
1170 }
1171 }
1172
1173
1174 void InterpreterMacroAssembler::profile_call(Register mdp) {
1175 if (ProfileInterpreter) {
1176 Label profile_continue;
1177
1178 // If no method data exists, go to profile_continue.
1179 test_method_data_pointer(mdp, profile_continue);
1180
1181 // We are making a call. Increment the count.
1182 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1183
1184 // The method data pointer needs to be updated to reflect the new target.
1185 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1186 bind(profile_continue);
1187 }
1188 }
1189
1190 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1191 if (ProfileInterpreter) {
1192 Label profile_continue;
1193
1194 // If no method data exists, go to profile_continue.
1195 test_method_data_pointer(mdp, profile_continue);
1196
1197 // We are making a call. Increment the count.
1198 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1199
1200 // The method data pointer needs to be updated to reflect the new target.
1201 update_mdp_by_constant(mdp,
1202 in_bytes(VirtualCallData::
1203 virtual_call_data_size()));
1204 bind(profile_continue);
1205 }
1206 }
1207
1208
1209 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1210 Register mdp,
1211 Register reg2,
1212 bool receiver_can_be_null) {
1213 if (ProfileInterpreter) {
1214 Label profile_continue;
1215
1216 // If no method data exists, go to profile_continue.
1217 test_method_data_pointer(mdp, profile_continue);
1218
1219 Label skip_receiver_profile;
1220 if (receiver_can_be_null) {
1221 Label not_null;
1222 // We are making a call. Increment the count for null receiver.
1223 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1224 b(skip_receiver_profile);
1225 bind(not_null);
1226 }
1227
1228 // Record the receiver type.
1229 record_klass_in_profile(receiver, mdp, reg2);
1230 bind(skip_receiver_profile);
1231
1232 // The method data pointer needs to be updated to reflect the new target.
1233 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1234 bind(profile_continue);
1235 }
1236 }
1237
1238 // This routine creates a state machine for updating the multi-row
1239 // type profile at a virtual call site (or other type-sensitive bytecode).
1240 // The machine visits each row (of receiver/count) until the receiver type
1241 // is found, or until it runs out of rows. At the same time, it remembers
1242 // the location of the first empty row. (An empty row records null for its
1243 // receiver, and can be allocated for a newly-observed receiver type.)
1244 // Because there are two degrees of freedom in the state, a simple linear
1245 // search will not work; it must be a decision tree. Hence this helper
1246 // function is recursive, to generate the required tree structured code.
1247 // It's the interpreter, so we are trading off code space for speed.
1248 // See below for example code.
1249 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1250 Register receiver, Register mdp,
1251 Register reg2, int start_row,
1252 Label& done) {
1253 if (TypeProfileWidth == 0) {
1254 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1255 } else {
1256 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1257 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1258 }
1259 }
1260
1261 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1262 Register reg2, int start_row, Label& done, int total_rows,
1263 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1264 int last_row = total_rows - 1;
1265 assert(start_row <= last_row, "must be work left to do");
1266 // Test this row for both the item and for null.
1267 // Take any of three different outcomes:
1268 // 1. found item => increment count and goto done
1269 // 2. found null => keep looking for case 1, maybe allocate this cell
1270 // 3. found something else => keep looking for cases 1 and 2
1271 // Case 3 is handled by a recursive call.
1272 for (int row = start_row; row <= last_row; row++) {
1273 Label next_test;
1274 bool test_for_null_also = (row == start_row);
1275
1276 // See if the item is item[n].
1277 int item_offset = in_bytes(item_offset_fn(row));
1278 test_mdp_data_at(mdp, item_offset, item,
1279 (test_for_null_also ? reg2 : noreg),
1280 next_test);
1281 // (Reg2 now contains the item from the CallData.)
1282
1283 // The item is item[n]. Increment count[n].
1284 int count_offset = in_bytes(item_count_offset_fn(row));
1285 increment_mdp_data_at(mdp, count_offset);
1286 b(done);
1287 bind(next_test);
1288
1289 if (test_for_null_also) {
1290 Label found_null;
1291 // Failed the equality check on item[n]... Test for null.
1292 if (start_row == last_row) {
1293 // The only thing left to do is handle the null case.
1294 cbz(reg2, found_null);
1295 // Item did not match any saved item and there is no empty row for it.
1296 // Increment total counter to indicate polymorphic case.
1297 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1298 b(done);
1299 bind(found_null);
1300 break;
1301 }
1302 // Since null is rare, make it be the branch-taken case.
1303 cbz(reg2, found_null);
1304
1305 // Put all the "Case 3" tests here.
1306 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1307 item_offset_fn, item_count_offset_fn);
1308
1309 // Found a null. Keep searching for a matching item,
1310 // but remember that this is an empty (unused) slot.
1311 bind(found_null);
1312 }
1313 }
1314
1315 // In the fall-through case, we found no matching item, but we
1316 // observed the item[start_row] is null.
1317
1318 // Fill in the item field and increment the count.
1319 int item_offset = in_bytes(item_offset_fn(start_row));
1320 set_mdp_data_at(mdp, item_offset, item);
1321 int count_offset = in_bytes(item_count_offset_fn(start_row));
1322 mov(reg2, DataLayout::counter_increment);
1323 set_mdp_data_at(mdp, count_offset, reg2);
1324 if (start_row > 0) {
1325 b(done);
1326 }
1327 }
1328
1329 // Example state machine code for three profile rows:
1330 // // main copy of decision tree, rooted at row[1]
1331 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1332 // if (row[0].rec != nullptr) {
1333 // // inner copy of decision tree, rooted at row[1]
1334 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1335 // if (row[1].rec != nullptr) {
1336 // // degenerate decision tree, rooted at row[2]
1337 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1338 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1339 // row[2].init(rec); goto done;
1340 // } else {
1341 // // remember row[1] is empty
1342 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1343 // row[1].init(rec); goto done;
1344 // }
1345 // } else {
1346 // // remember row[0] is empty
1347 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1348 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1349 // row[0].init(rec); goto done;
1350 // }
1351 // done:
1352
1353 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1354 Register mdp, Register reg2) {
1355 assert(ProfileInterpreter, "must be profiling");
1356 Label done;
1357
1358 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1359
1360 bind (done);
1361 }
1362
1363 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1364 Register mdp) {
1365 if (ProfileInterpreter) {
1366 Label profile_continue;
1367 uint row;
1368
1369 // If no method data exists, go to profile_continue.
1370 test_method_data_pointer(mdp, profile_continue);
1371
1372 // Update the total ret count.
1373 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1374
1375 for (row = 0; row < RetData::row_limit(); row++) {
1376 Label next_test;
1377
1378 // See if return_bci is equal to bci[n]:
1379 test_mdp_data_at(mdp,
1380 in_bytes(RetData::bci_offset(row)),
1381 return_bci, noreg,
1382 next_test);
1383
1384 // return_bci is equal to bci[n]. Increment the count.
1385 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1386
1387 // The method data pointer needs to be updated to reflect the new target.
1388 update_mdp_by_offset(mdp,
1389 in_bytes(RetData::bci_displacement_offset(row)));
1390 b(profile_continue);
1391 bind(next_test);
1392 }
1393
1394 update_mdp_for_ret(return_bci);
1395
1396 bind(profile_continue);
1397 }
1398 }
1399
1400 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1401 if (ProfileInterpreter) {
1402 Label profile_continue;
1403
1404 // If no method data exists, go to profile_continue.
1405 test_method_data_pointer(mdp, profile_continue);
1406
1407 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1408
1409 // The method data pointer needs to be updated.
1410 int mdp_delta = in_bytes(BitData::bit_data_size());
1411 if (TypeProfileCasts) {
1412 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1413 }
1414 update_mdp_by_constant(mdp, mdp_delta);
1415
1416 bind(profile_continue);
1417 }
1418 }
1419
1420 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1421 if (ProfileInterpreter) {
1422 Label profile_continue;
1423
1424 // If no method data exists, go to profile_continue.
1425 test_method_data_pointer(mdp, profile_continue);
1426
1427 // The method data pointer needs to be updated.
1428 int mdp_delta = in_bytes(BitData::bit_data_size());
1429 if (TypeProfileCasts) {
1430 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1431
1432 // Record the object type.
1433 record_klass_in_profile(klass, mdp, reg2);
1434 }
1435 update_mdp_by_constant(mdp, mdp_delta);
1436
1437 bind(profile_continue);
1438 }
1439 }
1440
1441 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1442 if (ProfileInterpreter) {
1443 Label profile_continue;
1444
1445 // If no method data exists, go to profile_continue.
1446 test_method_data_pointer(mdp, profile_continue);
1447
1448 // Update the default case count
1449 increment_mdp_data_at(mdp,
1450 in_bytes(MultiBranchData::default_count_offset()));
1451
1452 // The method data pointer needs to be updated.
1453 update_mdp_by_offset(mdp,
1454 in_bytes(MultiBranchData::
1455 default_displacement_offset()));
1456
1457 bind(profile_continue);
1458 }
1459 }
1460
1461 void InterpreterMacroAssembler::profile_switch_case(Register index,
1462 Register mdp,
1463 Register reg2) {
1464 if (ProfileInterpreter) {
1465 Label profile_continue;
1466
1467 // If no method data exists, go to profile_continue.
1468 test_method_data_pointer(mdp, profile_continue);
1469
1470 // Build the base (index * per_case_size_in_bytes()) +
1471 // case_array_offset_in_bytes()
1472 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1473 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1474 Assembler::maddw(index, index, reg2, rscratch1);
1475
1476 // Update the case count
1477 increment_mdp_data_at(mdp,
1478 index,
1479 in_bytes(MultiBranchData::relative_count_offset()));
1480
1481 // The method data pointer needs to be updated.
1482 update_mdp_by_offset(mdp,
1483 index,
1484 in_bytes(MultiBranchData::
1485 relative_displacement_offset()));
1486
1487 bind(profile_continue);
1488 }
1489 }
1490
1491 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1492 Register array,
1493 Register tmp) {
1494 if (ProfileInterpreter) {
1495 Label profile_continue;
1496
1497 // If no method data exists, go to profile_continue.
1498 test_method_data_pointer(mdp, profile_continue);
1499
1500 mov(tmp, array);
1501 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1502
1503 Label not_flat;
1504 test_non_flat_array_oop(array, tmp, not_flat);
1505
1506 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1507
1508 bind(not_flat);
1509
1510 Label not_null_free;
1511 test_non_null_free_array_oop(array, tmp, not_null_free);
1512
1513 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1514
1515 bind(not_null_free);
1516
1517 bind(profile_continue);
1518 }
1519 }
1520
1521 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1522 Register array,
1523 Register tmp);
1524 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1525 Register array,
1526 Register tmp);
1527
1528 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1529 if (ProfileInterpreter) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(mdp, profile_continue);
1534
1535 Label done, update;
1536 cbnz(element, update);
1537 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1538 b(done);
1539
1540 bind(update);
1541 load_klass(tmp, element);
1542
1543 // Record the object type.
1544 record_klass_in_profile(tmp, mdp, tmp2);
1545
1546 bind(done);
1547
1548 // The method data pointer needs to be updated.
1549 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1550
1551 bind(profile_continue);
1552 }
1553 }
1554
1555
1556 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1557 Register element,
1558 Register tmp) {
1559 if (ProfileInterpreter) {
1560 Label profile_continue;
1561
1562 // If no method data exists, go to profile_continue.
1563 test_method_data_pointer(mdp, profile_continue);
1564
1565 mov(tmp, element);
1566 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1567
1568 // The method data pointer needs to be updated.
1569 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1570
1571 bind(profile_continue);
1572 }
1573 }
1574
1575 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1576 Register left,
1577 Register right,
1578 Register tmp) {
1579 if (ProfileInterpreter) {
1580 Label profile_continue;
1581
1582 // If no method data exists, go to profile_continue.
1583 test_method_data_pointer(mdp, profile_continue);
1584
1585 mov(tmp, left);
1586 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1587
1588 Label left_not_inline_type;
1589 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1590 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1591 bind(left_not_inline_type);
1592
1593 mov(tmp, right);
1594 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1595
1596 Label right_not_inline_type;
1597 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1598 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1599 bind(right_not_inline_type);
1600
1601 bind(profile_continue);
1602 }
1603 }
1604
1605 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1606 if (state == atos) {
1607 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1608 }
1609 }
1610
1611 void InterpreterMacroAssembler::notify_method_entry() {
1612 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1613 // track stack depth. If it is possible to enter interp_only_mode we add
1614 // the code to check if the event should be sent.
1615 if (JvmtiExport::can_post_interpreter_events()) {
1616 Label L;
1617 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1618 cbzw(r3, L);
1619 call_VM(noreg, CAST_FROM_FN_PTR(address,
1620 InterpreterRuntime::post_method_entry));
1621 bind(L);
1622 }
1623
1624 if (DTraceMethodProbes) {
1625 get_method(c_rarg1);
1626 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1627 rthread, c_rarg1);
1628 }
1629
1630 // RedefineClasses() tracing support for obsolete method entry
1631 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1632 get_method(c_rarg1);
1633 call_VM_leaf(
1634 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1635 rthread, c_rarg1);
1636 }
1637
1638 }
1639
1640
1641 void InterpreterMacroAssembler::notify_method_exit(
1642 TosState state, NotifyMethodExitMode mode) {
1643 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1644 // track stack depth. If it is possible to enter interp_only_mode we add
1645 // the code to check if the event should be sent.
1646 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1647 Label L;
1648 // Note: frame::interpreter_frame_result has a dependency on how the
1649 // method result is saved across the call to post_method_exit. If this
1650 // is changed then the interpreter_frame_result implementation will
1651 // need to be updated too.
1652
1653 // template interpreter will leave the result on the top of the stack.
1654 push(state);
1655 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1656 cbz(r3, L);
1657 call_VM(noreg,
1658 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1659 bind(L);
1660 pop(state);
1661 }
1662
1663 if (DTraceMethodProbes) {
1664 push(state);
1665 get_method(c_rarg1);
1666 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1667 rthread, c_rarg1);
1668 pop(state);
1669 }
1670 }
1671
1672
1673 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1674 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1675 int increment, Address mask,
1676 Register scratch, Register scratch2,
1677 bool preloaded, Condition cond,
1678 Label* where) {
1679 if (!preloaded) {
1680 ldrw(scratch, counter_addr);
1681 }
1682 add(scratch, scratch, increment);
1683 strw(scratch, counter_addr);
1684 ldrw(scratch2, mask);
1685 ands(scratch, scratch, scratch2);
1686 br(cond, *where);
1687 }
1688
1689 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1690 int number_of_arguments) {
1691 // interpreter specific
1692 //
1693 // Note: No need to save/restore rbcp & rlocals pointer since these
1694 // are callee saved registers and no blocking/ GC can happen
1695 // in leaf calls.
1696 #ifdef ASSERT
1697 {
1698 Label L;
1699 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1700 cbz(rscratch1, L);
1701 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1702 " last_sp != nullptr");
1703 bind(L);
1704 }
1705 #endif /* ASSERT */
1706 // super call
1707 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1708 }
1709
1710 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1711 Register java_thread,
1712 Register last_java_sp,
1713 address entry_point,
1714 int number_of_arguments,
1715 bool check_exceptions) {
1716 // interpreter specific
1717 //
1718 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1719 // really make a difference for these runtime calls, since they are
1720 // slow anyway. Btw., bcp must be saved/restored since it may change
1721 // due to GC.
1722 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1723 save_bcp();
1724 #ifdef ASSERT
1725 {
1726 Label L;
1727 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1728 cbz(rscratch1, L);
1729 stop("InterpreterMacroAssembler::call_VM_base:"
1730 " last_sp != nullptr");
1731 bind(L);
1732 }
1733 #endif /* ASSERT */
1734 // super call
1735 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1736 entry_point, number_of_arguments,
1737 check_exceptions);
1738 // interpreter specific
1739 restore_bcp();
1740 restore_locals();
1741 }
1742
1743 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1744 address entry_point,
1745 Register arg_1) {
1746 assert(arg_1 == c_rarg1, "");
1747 Label resume_pc, not_preempted;
1748
1749 #ifdef ASSERT
1750 {
1751 Label L;
1752 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1753 cbz(rscratch1, L);
1754 stop("Should not have alternate return address set");
1755 bind(L);
1756 }
1757 #endif /* ASSERT */
1758
1759 // Force freeze slow path.
1760 push_cont_fastpath();
1761
1762 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1763 adr(rscratch1, resume_pc);
1764 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1765 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1766
1767 pop_cont_fastpath();
1768
1769 // Check if preempted.
1770 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1771 cbz(rscratch1, not_preempted);
1772 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1773 br(rscratch1);
1774
1775 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1776 bind(resume_pc);
1777 restore_after_resume(false /* is_native */);
1778
1779 bind(not_preempted);
1780 }
1781
1782 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1783 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1784 blr(rscratch1);
1785 if (is_native) {
1786 // On resume we need to set up stack as expected
1787 push(dtos);
1788 push(ltos);
1789 }
1790 }
1791
1792 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1793 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1794 Label update, next, none;
1795
1796 verify_oop(obj);
1797
1798 cbnz(obj, update);
1799 orptr(mdo_addr, TypeEntries::null_seen);
1800 b(next);
1801
1802 bind(update);
1803 load_klass(obj, obj);
1804
1805 ldr(rscratch1, mdo_addr);
1806 eor(obj, obj, rscratch1);
1807 tst(obj, TypeEntries::type_klass_mask);
1808 br(Assembler::EQ, next); // klass seen before, nothing to
1809 // do. The unknown bit may have been
1810 // set already but no need to check.
1811
1812 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1813 // already unknown. Nothing to do anymore.
1814
1815 cbz(rscratch1, none);
1816 cmp(rscratch1, (u1)TypeEntries::null_seen);
1817 br(Assembler::EQ, none);
1818 // There is a chance that the checks above
1819 // fail if another thread has just set the
1820 // profiling to this obj's klass
1821 eor(obj, obj, rscratch1); // get back original value before XOR
1822 ldr(rscratch1, mdo_addr);
1823 eor(obj, obj, rscratch1);
1824 tst(obj, TypeEntries::type_klass_mask);
1825 br(Assembler::EQ, next);
1826
1827 // different than before. Cannot keep accurate profile.
1828 orptr(mdo_addr, TypeEntries::type_unknown);
1829 b(next);
1830
1831 bind(none);
1832 // first time here. Set profile type.
1833 str(obj, mdo_addr);
1834 #ifdef ASSERT
1835 andr(obj, obj, TypeEntries::type_mask);
1836 verify_klass_ptr(obj);
1837 #endif
1838
1839 bind(next);
1840 }
1841
1842 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1843 if (!ProfileInterpreter) {
1844 return;
1845 }
1846
1847 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1848 Label profile_continue;
1849
1850 test_method_data_pointer(mdp, profile_continue);
1851
1852 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1853
1854 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1855 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1856 br(Assembler::NE, profile_continue);
1857
1858 if (MethodData::profile_arguments()) {
1859 Label done;
1860 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1861
1862 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1863 if (i > 0 || MethodData::profile_return()) {
1864 // If return value type is profiled we may have no argument to profile
1865 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1866 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1867 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1868 add(rscratch1, mdp, off_to_args);
1869 br(Assembler::LT, done);
1870 }
1871 ldr(tmp, Address(callee, Method::const_offset()));
1872 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1873 // stack offset o (zero based) from the start of the argument
1874 // list, for n arguments translates into offset n - o - 1 from
1875 // the end of the argument list
1876 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1877 sub(tmp, tmp, rscratch1);
1878 sub(tmp, tmp, 1);
1879 Address arg_addr = argument_address(tmp);
1880 ldr(tmp, arg_addr);
1881
1882 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1883 profile_obj_type(tmp, mdo_arg_addr);
1884
1885 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1886 off_to_args += to_add;
1887 }
1888
1889 if (MethodData::profile_return()) {
1890 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1891 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1892 }
1893
1894 add(rscratch1, mdp, off_to_args);
1895 bind(done);
1896 mov(mdp, rscratch1);
1897
1898 if (MethodData::profile_return()) {
1899 // We're right after the type profile for the last
1900 // argument. tmp is the number of cells left in the
1901 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1902 // if there's a return to profile.
1903 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1904 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1905 }
1906 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1907 } else {
1908 assert(MethodData::profile_return(), "either profile call args or call ret");
1909 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1910 }
1911
1912 // mdp points right after the end of the
1913 // CallTypeData/VirtualCallTypeData, right after the cells for the
1914 // return value type if there's one
1915
1916 bind(profile_continue);
1917 }
1918 }
1919
1920 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1921 assert_different_registers(mdp, ret, tmp, rbcp);
1922 if (ProfileInterpreter && MethodData::profile_return()) {
1923 Label profile_continue, done;
1924
1925 test_method_data_pointer(mdp, profile_continue);
1926
1927 if (MethodData::profile_return_jsr292_only()) {
1928 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1929
1930 // If we don't profile all invoke bytecodes we must make sure
1931 // it's a bytecode we indeed profile. We can't go back to the
1932 // beginning of the ProfileData we intend to update to check its
1933 // type because we're right after it and we don't known its
1934 // length
1935 Label do_profile;
1936 ldrb(rscratch1, Address(rbcp, 0));
1937 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1938 br(Assembler::EQ, do_profile);
1939 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1940 br(Assembler::EQ, do_profile);
1941 get_method(tmp);
1942 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1943 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1944 br(Assembler::NE, profile_continue);
1945
1946 bind(do_profile);
1947 }
1948
1949 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1950 mov(tmp, ret);
1951 profile_obj_type(tmp, mdo_ret_addr);
1952
1953 bind(profile_continue);
1954 }
1955 }
1956
1957 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1958 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1959 if (ProfileInterpreter && MethodData::profile_parameters()) {
1960 Label profile_continue, done;
1961
1962 test_method_data_pointer(mdp, profile_continue);
1963
1964 // Load the offset of the area within the MDO used for
1965 // parameters. If it's negative we're not profiling any parameters
1966 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1967 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1968
1969 // Compute a pointer to the area for parameters from the offset
1970 // and move the pointer to the slot for the last
1971 // parameters. Collect profiling from last parameter down.
1972 // mdo start + parameters offset + array length - 1
1973 add(mdp, mdp, tmp1);
1974 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1975 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1976
1977 Label loop;
1978 bind(loop);
1979
1980 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1981 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1982 int per_arg_scale = exact_log2(DataLayout::cell_size);
1983 add(rscratch1, mdp, off_base);
1984 add(rscratch2, mdp, type_base);
1985
1986 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1987 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1988
1989 // load offset on the stack from the slot for this parameter
1990 ldr(tmp2, arg_off);
1991 neg(tmp2, tmp2);
1992 // read the parameter from the local area
1993 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1994
1995 // profile the parameter
1996 profile_obj_type(tmp2, arg_type);
1997
1998 // go to next parameter
1999 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
2000 br(Assembler::GE, loop);
2001
2002 bind(profile_continue);
2003 }
2004 }
2005
2006 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2007 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
2008 get_cache_index_at_bcp(index, 1, sizeof(u4));
2009 // Get address of invokedynamic array
2010 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2011 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2012 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
2013 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
2014 lea(cache, Address(cache, index));
2015 }
2016
2017 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2018 // Get index out of bytecode pointer
2019 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2020 // Take shortcut if the size is a power of 2
2021 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2022 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2023 } else {
2024 mov(cache, sizeof(ResolvedFieldEntry));
2025 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2026 }
2027 // Get address of field entries array
2028 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
2029 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
2030 lea(cache, Address(cache, index));
2031 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
2032 membar(MacroAssembler::LoadLoad);
2033 }
2034
2035 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2036 // Get index out of bytecode pointer
2037 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2038 mov(cache, sizeof(ResolvedMethodEntry));
2039 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2040
2041 // Get address of field entries array
2042 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
2043 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2044 lea(cache, Address(cache, index));
2045 }