1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_aarch64.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/constMethodFlags.hpp"
37 #include "oops/markWord.hpp"
38 #include "oops/method.hpp"
39 #include "oops/methodData.hpp"
40 #include "oops/inlineKlass.hpp"
41 #include "oops/resolvedFieldEntry.hpp"
42 #include "oops/resolvedIndyEntry.hpp"
43 #include "oops/resolvedMethodEntry.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/jvmtiThreadState.hpp"
46 #include "runtime/basicLock.hpp"
47 #include "runtime/frame.inline.hpp"
48 #include "runtime/javaThread.hpp"
49 #include "runtime/safepointMechanism.hpp"
50 #include "runtime/sharedRuntime.hpp"
51 #include "utilities/powerOfTwo.hpp"
52
53 void InterpreterMacroAssembler::narrow(Register result) {
54
55 // Get method->_constMethod->_result_type
56 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
57 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
58 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
59
60 Label done, notBool, notByte, notChar;
61
62 // common case first
63 cmpw(rscratch1, T_INT);
64 br(Assembler::EQ, done);
65
66 // mask integer result to narrower return type.
67 cmpw(rscratch1, T_BOOLEAN);
68 br(Assembler::NE, notBool);
69 andw(result, result, 0x1);
70 b(done);
71
72 bind(notBool);
73 cmpw(rscratch1, T_BYTE);
74 br(Assembler::NE, notByte);
75 sbfx(result, result, 0, 8);
76 b(done);
77
78 bind(notByte);
79 cmpw(rscratch1, T_CHAR);
80 br(Assembler::NE, notChar);
81 ubfx(result, result, 0, 16); // truncate upper 16 bits
82 b(done);
83
84 bind(notChar);
85 sbfx(result, result, 0, 16); // sign-extend short
86
87 // Nothing to do for T_INT
88 bind(done);
89 }
90
91 void InterpreterMacroAssembler::jump_to_entry(address entry) {
92 assert(entry, "Entry must have been generated by now");
93 b(entry);
94 }
95
96 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
97 if (JvmtiExport::can_pop_frame()) {
98 Label L;
99 // Initiate popframe handling only if it is not already being
100 // processed. If the flag has the popframe_processing bit set, it
101 // means that this code is called *during* popframe handling - we
102 // don't want to reenter.
103 // This method is only called just after the call into the vm in
104 // call_VM_base, so the arg registers are available.
105 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
106 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
107 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
108 // Call Interpreter::remove_activation_preserving_args_entry() to get the
109 // address of the same-named entrypoint in the generated interpreter code.
110 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
111 br(r0);
112 bind(L);
113 }
114 }
115
116
117 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
118 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
119 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
120 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
121 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
122 switch (state) {
123 case atos: ldr(r0, oop_addr);
124 str(zr, oop_addr);
125 interp_verify_oop(r0, state); break;
126 case ltos: ldr(r0, val_addr); break;
127 case btos: // fall through
128 case ztos: // fall through
129 case ctos: // fall through
130 case stos: // fall through
131 case itos: ldrw(r0, val_addr); break;
132 case ftos: ldrs(v0, val_addr); break;
133 case dtos: ldrd(v0, val_addr); break;
134 case vtos: /* nothing to do */ break;
135 default : ShouldNotReachHere();
136 }
137 // Clean up tos value in the thread object
138 movw(rscratch1, (int) ilgl);
139 strw(rscratch1, tos_addr);
140 strw(zr, val_addr);
141 }
142
143
144 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
145 if (JvmtiExport::can_force_early_return()) {
146 Label L;
147 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
148 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
149
150 // Initiate earlyret handling only if it is not already being processed.
151 // If the flag has the earlyret_processing bit set, it means that this code
152 // is called *during* earlyret handling - we don't want to reenter.
153 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
154 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
155 br(Assembler::NE, L);
156
157 // Call Interpreter::remove_activation_early_entry() to get the address of the
158 // same-named entrypoint in the generated interpreter code.
159 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
160 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
161 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
162 br(r0);
163 bind(L);
164 }
165 }
166
167 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
168 Register reg,
169 int bcp_offset) {
170 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
171 ldrh(reg, Address(rbcp, bcp_offset));
172 rev16(reg, reg);
173 }
174
175 void InterpreterMacroAssembler::get_dispatch() {
176 uint64_t offset;
177 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
178 // Use add() here after ARDP, rather than lea().
179 // lea() does not generate anything if its offset is zero.
180 // However, relocs expect to find either an ADD or a load/store
181 // insn after an ADRP. add() always generates an ADD insn, even
182 // for add(Rn, Rn, 0).
183 add(rdispatch, rdispatch, offset);
184 }
185
186 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
187 int bcp_offset,
188 size_t index_size) {
189 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
190 if (index_size == sizeof(u2)) {
191 load_unsigned_short(index, Address(rbcp, bcp_offset));
192 } else if (index_size == sizeof(u4)) {
193 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
194 ldrw(index, Address(rbcp, bcp_offset));
195 } else if (index_size == sizeof(u1)) {
196 load_unsigned_byte(index, Address(rbcp, bcp_offset));
197 } else {
198 ShouldNotReachHere();
199 }
200 }
201
202 void InterpreterMacroAssembler::get_method_counters(Register method,
203 Register mcs, Label& skip) {
204 Label has_counters;
205 ldr(mcs, Address(method, Method::method_counters_offset()));
206 cbnz(mcs, has_counters);
207 call_VM(noreg, CAST_FROM_FN_PTR(address,
208 InterpreterRuntime::build_method_counters), method);
209 ldr(mcs, Address(method, Method::method_counters_offset()));
210 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
211 bind(has_counters);
212 }
213
214 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
215 Register t1, Register t2,
216 bool clear_fields, Label& alloc_failed) {
217 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
218 {
219 SkipIfEqual skip_if(this, &DTraceAllocProbes, 0);
220 // Trigger dtrace event for fastpath
221 push(atos);
222 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
223 pop(atos);
224 }
225 }
226
227 void InterpreterMacroAssembler::read_flat_field(Register entry,
228 Register field_index, Register field_offset,
229 Register temp, Register obj) {
230 Label alloc_failed, empty_value, done;
231 const Register src = field_offset;
232 const Register alloc_temp = r10;
233 const Register dst_temp = field_index;
234 const Register layout_info = temp;
235 assert_different_registers(obj, entry, field_index, field_offset, temp, alloc_temp);
236
237 // Grab the inline field klass
238 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
239 inline_layout_info(rscratch1, field_index, layout_info);
240
241 const Register field_klass = dst_temp;
242 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
243
244 // check for empty value klass
245 test_klass_is_empty_inline_type(field_klass, rscratch1, empty_value);
246
247 // allocate buffer
248 push(obj); // save holder
249 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, alloc_failed);
250
251 // Have an oop instance buffer, copy into it
252 data_for_oop(obj, dst_temp, field_klass); // danger, uses rscratch1
253 pop(alloc_temp); // restore holder
254 lea(src, Address(alloc_temp, field_offset));
255 // call_VM_leaf, clobbers a few regs, save restore new obj
256 push(obj);
257 flat_field_copy(IS_DEST_UNINITIALIZED, src, dst_temp, layout_info);
258 pop(obj);
259 b(done);
260
261 bind(empty_value);
262 get_empty_inline_type_oop(field_klass, alloc_temp, obj);
263 b(done);
264
265 bind(alloc_failed);
266 pop(obj);
267 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
268 obj, entry);
269
270 bind(done);
271 membar(Assembler::StoreStore);
272 }
273
274 // Load object from cpool->resolved_references(index)
275 void InterpreterMacroAssembler::load_resolved_reference_at_index(
276 Register result, Register index, Register tmp) {
277 assert_different_registers(result, index);
278
279 get_constant_pool(result);
280 // load pointer for resolved_references[] objArray
281 ldr(result, Address(result, ConstantPool::cache_offset()));
282 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
283 resolve_oop_handle(result, tmp, rscratch2);
284 // Add in the index
285 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
286 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
287 }
288
289 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
290 Register cpool, Register index, Register klass, Register temp) {
291 add(temp, cpool, index, LSL, LogBytesPerWord);
292 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
293 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
294 add(klass, klass, temp, LSL, LogBytesPerWord);
295 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
296 }
297
298 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
299 // subtype of super_klass.
300 //
301 // Args:
302 // r0: superklass
303 // Rsub_klass: subklass
304 //
305 // Kills:
306 // r2, r5
307 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
308 Label& ok_is_subtype,
309 bool profile) {
310 assert(Rsub_klass != r0, "r0 holds superklass");
311 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
312 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
313
314 // Profile the not-null value's klass.
315 if (profile) {
316 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
317 }
318
319 // Do the check.
320 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
321 }
322
323 // Java Expression Stack
324
325 void InterpreterMacroAssembler::pop_ptr(Register r) {
326 ldr(r, post(esp, wordSize));
327 }
328
329 void InterpreterMacroAssembler::pop_i(Register r) {
330 ldrw(r, post(esp, wordSize));
331 }
332
333 void InterpreterMacroAssembler::pop_l(Register r) {
334 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
335 }
336
337 void InterpreterMacroAssembler::push_ptr(Register r) {
338 str(r, pre(esp, -wordSize));
339 }
340
341 void InterpreterMacroAssembler::push_i(Register r) {
342 str(r, pre(esp, -wordSize));
343 }
344
345 void InterpreterMacroAssembler::push_l(Register r) {
346 str(zr, pre(esp, -wordSize));
347 str(r, pre(esp, - wordSize));
348 }
349
350 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
351 ldrs(r, post(esp, wordSize));
352 }
353
354 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
355 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
356 }
357
358 void InterpreterMacroAssembler::push_f(FloatRegister r) {
359 strs(r, pre(esp, -wordSize));
360 }
361
362 void InterpreterMacroAssembler::push_d(FloatRegister r) {
363 strd(r, pre(esp, 2* -wordSize));
364 }
365
366 void InterpreterMacroAssembler::pop(TosState state) {
367 switch (state) {
368 case atos: pop_ptr(); break;
369 case btos:
370 case ztos:
371 case ctos:
372 case stos:
373 case itos: pop_i(); break;
374 case ltos: pop_l(); break;
375 case ftos: pop_f(); break;
376 case dtos: pop_d(); break;
377 case vtos: /* nothing to do */ break;
378 default: ShouldNotReachHere();
379 }
380 interp_verify_oop(r0, state);
381 }
382
383 void InterpreterMacroAssembler::push(TosState state) {
384 interp_verify_oop(r0, state);
385 switch (state) {
386 case atos: push_ptr(); break;
387 case btos:
388 case ztos:
389 case ctos:
390 case stos:
391 case itos: push_i(); break;
392 case ltos: push_l(); break;
393 case ftos: push_f(); break;
394 case dtos: push_d(); break;
395 case vtos: /* nothing to do */ break;
396 default : ShouldNotReachHere();
397 }
398 }
399
400 // Helpers for swap and dup
401 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
402 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
403 }
404
405 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
406 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
407 }
408
409 void InterpreterMacroAssembler::load_float(Address src) {
410 ldrs(v0, src);
411 }
412
413 void InterpreterMacroAssembler::load_double(Address src) {
414 ldrd(v0, src);
415 }
416
417 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
418 // set sender sp
419 mov(r19_sender_sp, sp);
420 // record last_sp
421 sub(rscratch1, esp, rfp);
422 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
423 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
424 }
425
426 // Jump to from_interpreted entry of a call unless single stepping is possible
427 // in this thread in which case we must call the i2i entry
428 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
429 prepare_to_jump_from_interpreted();
430
431 if (JvmtiExport::can_post_interpreter_events()) {
432 Label run_compiled_code;
433 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
434 // compiled code in threads for which the event is enabled. Check here for
435 // interp_only_mode if these events CAN be enabled.
436 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
437 cbzw(rscratch1, run_compiled_code);
438 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
439 br(rscratch1);
440 bind(run_compiled_code);
441 }
442
443 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
444 br(rscratch1);
445 }
446
447 // The following two routines provide a hook so that an implementation
448 // can schedule the dispatch in two parts. amd64 does not do this.
449 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
450 }
451
452 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
453 dispatch_next(state, step);
454 }
455
456 void InterpreterMacroAssembler::dispatch_base(TosState state,
457 address* table,
458 bool verifyoop,
459 bool generate_poll) {
460 if (VerifyActivationFrameSize) {
461 Unimplemented();
462 }
463 if (verifyoop) {
464 interp_verify_oop(r0, state);
465 }
466
467 Label safepoint;
468 address* const safepoint_table = Interpreter::safept_table(state);
469 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
470
471 if (needs_thread_local_poll) {
472 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
473 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
474 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
475 }
476
477 if (table == Interpreter::dispatch_table(state)) {
478 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
479 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
480 } else {
481 mov(rscratch2, (address)table);
482 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
483 }
484 br(rscratch2);
485
486 if (needs_thread_local_poll) {
487 bind(safepoint);
488 lea(rscratch2, ExternalAddress((address)safepoint_table));
489 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
490 br(rscratch2);
491 }
492 }
493
494 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
495 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
496 }
497
498 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
499 dispatch_base(state, Interpreter::normal_table(state));
500 }
501
502 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
503 dispatch_base(state, Interpreter::normal_table(state), false);
504 }
505
506
507 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
508 // load next bytecode
509 ldrb(rscratch1, Address(pre(rbcp, step)));
510 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
511 }
512
513 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
514 // load current bytecode
515 ldrb(rscratch1, Address(rbcp, 0));
516 dispatch_base(state, table);
517 }
518
519 // remove activation
520 //
521 // Apply stack watermark barrier.
522 // Unlock the receiver if this is a synchronized method.
523 // Unlock any Java monitors from synchronized blocks.
524 // Remove the activation from the stack.
525 //
526 // If there are locked Java monitors
527 // If throw_monitor_exception
528 // throws IllegalMonitorStateException
529 // Else if install_monitor_exception
530 // installs IllegalMonitorStateException
531 // Else
532 // no error processing
533 void InterpreterMacroAssembler::remove_activation(
534 TosState state,
535 bool throw_monitor_exception,
536 bool install_monitor_exception,
537 bool notify_jvmdi) {
538 // Note: Registers r3 xmm0 may be in use for the
539 // result check if synchronized method
540 Label unlocked, unlock, no_unlock;
541
542 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
543 // that would normally not be safe to use. Such bad returns into unsafe territory of
544 // the stack, will call InterpreterRuntime::at_unwind.
545 Label slow_path;
546 Label fast_path;
547 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
548 br(Assembler::AL, fast_path);
549 bind(slow_path);
550 push(state);
551 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
552 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
553 reset_last_Java_frame(true);
554 pop(state);
555 bind(fast_path);
556
557 // get the value of _do_not_unlock_if_synchronized into r3
558 const Address do_not_unlock_if_synchronized(rthread,
559 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
560 ldrb(r3, do_not_unlock_if_synchronized);
561 strb(zr, do_not_unlock_if_synchronized); // reset the flag
562
563 // get method access flags
564 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
565 ldr(r2, Address(r1, Method::access_flags_offset()));
566 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
567
568 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
569 // is set.
570 cbnz(r3, no_unlock);
571
572 // unlock monitor
573 push(state); // save result
574
575 // BasicObjectLock will be first in list, since this is a
576 // synchronized method. However, need to check that the object has
577 // not been unlocked by an explicit monitorexit bytecode.
578 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
579 wordSize - (int) sizeof(BasicObjectLock));
580 // We use c_rarg1 so that if we go slow path it will be the correct
581 // register for unlock_object to pass to VM directly
582 lea(c_rarg1, monitor); // address of first monitor
583
584 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
585 cbnz(r0, unlock);
586
587 pop(state);
588 if (throw_monitor_exception) {
589 // Entry already unlocked, need to throw exception
590 call_VM(noreg, CAST_FROM_FN_PTR(address,
591 InterpreterRuntime::throw_illegal_monitor_state_exception));
592 should_not_reach_here();
593 } else {
594 // Monitor already unlocked during a stack unroll. If requested,
595 // install an illegal_monitor_state_exception. Continue with
596 // stack unrolling.
597 if (install_monitor_exception) {
598 call_VM(noreg, CAST_FROM_FN_PTR(address,
599 InterpreterRuntime::new_illegal_monitor_state_exception));
600 }
601 b(unlocked);
602 }
603
604 bind(unlock);
605 unlock_object(c_rarg1);
606 pop(state);
607
608 // Check that for block-structured locking (i.e., that all locked
609 // objects has been unlocked)
610 bind(unlocked);
611
612 // r0: Might contain return value
613
614 // Check that all monitors are unlocked
615 {
616 Label loop, exception, entry, restart;
617 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
618 const Address monitor_block_top(
619 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
620 const Address monitor_block_bot(
621 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
622
623 bind(restart);
624 // We use c_rarg1 so that if we go slow path it will be the correct
625 // register for unlock_object to pass to VM directly
626 ldr(c_rarg1, monitor_block_top); // derelativize pointer
627 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
628 // c_rarg1 points to current entry, starting with top-most entry
629
630 lea(r19, monitor_block_bot); // points to word before bottom of
631 // monitor block
632 b(entry);
633
634 // Entry already locked, need to throw exception
635 bind(exception);
636
637 if (throw_monitor_exception) {
638 // Throw exception
639 MacroAssembler::call_VM(noreg,
640 CAST_FROM_FN_PTR(address, InterpreterRuntime::
641 throw_illegal_monitor_state_exception));
642 should_not_reach_here();
643 } else {
644 // Stack unrolling. Unlock object and install illegal_monitor_exception.
645 // Unlock does not block, so don't have to worry about the frame.
646 // We don't have to preserve c_rarg1 since we are going to throw an exception.
647
648 push(state);
649 unlock_object(c_rarg1);
650 pop(state);
651
652 if (install_monitor_exception) {
653 call_VM(noreg, CAST_FROM_FN_PTR(address,
654 InterpreterRuntime::
655 new_illegal_monitor_state_exception));
656 }
657
658 b(restart);
659 }
660
661 bind(loop);
662 // check if current entry is used
663 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
664 cbnz(rscratch1, exception);
665
666 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
667 bind(entry);
668 cmp(c_rarg1, r19); // check if bottom reached
669 br(Assembler::NE, loop); // if not at bottom then check this entry
670 }
671
672 bind(no_unlock);
673
674 // jvmti support
675 if (notify_jvmdi) {
676 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
677 } else {
678 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
679 }
680
681 // remove activation
682 // get sender esp
683 ldr(rscratch2,
684 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
685
686 if (StackReservedPages > 0) {
687 // testing if reserved zone needs to be re-enabled
688 Label no_reserved_zone_enabling;
689
690 // check if already enabled - if so no re-enabling needed
691 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
692 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
693 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
694 br(Assembler::EQ, no_reserved_zone_enabling);
695
696 // look for an overflow into the stack reserved zone, i.e.
697 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
698 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
699 cmp(rscratch2, rscratch1);
700 br(Assembler::LS, no_reserved_zone_enabling);
701
702 call_VM_leaf(
703 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
704 call_VM(noreg, CAST_FROM_FN_PTR(address,
705 InterpreterRuntime::throw_delayed_StackOverflowError));
706 should_not_reach_here();
707
708 bind(no_reserved_zone_enabling);
709 }
710
711 if (state == atos && InlineTypeReturnedAsFields) {
712 // Check if we are returning an non-null inline type and load its fields into registers
713 Label skip;
714 test_oop_is_not_inline_type(r0, rscratch2, skip);
715
716 // Load fields from a buffered value with an inline class specific handler
717 load_klass(rscratch1 /*dst*/, r0 /*src*/);
718 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
719 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
720 // Unpack handler can be null if inline type is not scalarizable in returns
721 cbz(rscratch1, skip);
722
723 blr(rscratch1);
724 #ifdef ASSERT
725 // TODO 8284443 Enable
726 if (StressCallingConvention && false) {
727 Label skip_stress;
728 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
729 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
730 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
731 br(Assembler::EQ, skip_stress);
732 load_klass(r0, r0);
733 orr(r0, r0, 1);
734 bind(skip_stress);
735 }
736 #endif
737 bind(skip);
738 // Check above kills sender esp in rscratch2. Reload it.
739 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
740 }
741
742 // restore sender esp
743 mov(esp, rscratch2);
744 // remove frame anchor
745 leave();
746 // If we're returning to interpreted code we will shortly be
747 // adjusting SP to allow some space for ESP. If we're returning to
748 // compiled code the saved sender SP was saved in sender_sp, so this
749 // restores it.
750 andr(sp, esp, -16);
751 }
752
753 // Lock object
754 //
755 // Args:
756 // c_rarg1: BasicObjectLock to be used for locking
757 //
758 // Kills:
759 // r0
760 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
761 // rscratch1, rscratch2 (scratch regs)
762 void InterpreterMacroAssembler::lock_object(Register lock_reg)
763 {
764 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
765 if (LockingMode == LM_MONITOR) {
766 call_VM(noreg,
767 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
768 lock_reg);
769 } else {
770 Label count, done;
771
772 const Register swap_reg = r0;
773 const Register tmp = c_rarg2;
774 const Register obj_reg = c_rarg3; // Will contain the oop
775 const Register tmp2 = c_rarg4;
776 const Register tmp3 = c_rarg5;
777
778 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
779 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
780 const int mark_offset = lock_offset +
781 BasicLock::displaced_header_offset_in_bytes();
782
783 Label slow_case;
784
785 // Load object pointer into obj_reg %c_rarg3
786 ldr(obj_reg, Address(lock_reg, obj_offset));
787
788 if (DiagnoseSyncOnValueBasedClasses != 0) {
789 load_klass(tmp, obj_reg);
790 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
791 tst(tmp, KlassFlags::_misc_is_value_based_class);
792 br(Assembler::NE, slow_case);
793 }
794
795 if (LockingMode == LM_LIGHTWEIGHT) {
796 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
797 b(count);
798 } else if (LockingMode == LM_LEGACY) {
799 // Load (object->mark() | 1) into swap_reg
800 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
801 orr(swap_reg, rscratch1, 1);
802 if (EnableValhalla) {
803 // Mask inline_type bit such that we go to the slow path if object is an inline type
804 andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
805 }
806
807 // Save (object->mark() | 1) into BasicLock's displaced header
808 str(swap_reg, Address(lock_reg, mark_offset));
809
810 assert(lock_offset == 0,
811 "displached header must be first word in BasicObjectLock");
812
813 Label fail;
814 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
815
816 // Fast check for recursive lock.
817 //
818 // Can apply the optimization only if this is a stack lock
819 // allocated in this thread. For efficiency, we can focus on
820 // recently allocated stack locks (instead of reading the stack
821 // base and checking whether 'mark' points inside the current
822 // thread stack):
823 // 1) (mark & 7) == 0, and
824 // 2) sp <= mark < mark + os::pagesize()
825 //
826 // Warning: sp + os::pagesize can overflow the stack base. We must
827 // neither apply the optimization for an inflated lock allocated
828 // just above the thread stack (this is why condition 1 matters)
829 // nor apply the optimization if the stack lock is inside the stack
830 // of another thread. The latter is avoided even in case of overflow
831 // because we have guard pages at the end of all stacks. Hence, if
832 // we go over the stack base and hit the stack of another thread,
833 // this should not be in a writeable area that could contain a
834 // stack lock allocated by that thread. As a consequence, a stack
835 // lock less than page size away from sp is guaranteed to be
836 // owned by the current thread.
837 //
838 // These 3 tests can be done by evaluating the following
839 // expression: ((mark - sp) & (7 - os::vm_page_size())),
840 // assuming both stack pointer and pagesize have their
841 // least significant 3 bits clear.
842 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
843 // NOTE2: aarch64 does not like to subtract sp from rn so take a
844 // copy
845 mov(rscratch1, sp);
846 sub(swap_reg, swap_reg, rscratch1);
847 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
848
849 // Save the test result, for recursive case, the result is zero
850 str(swap_reg, Address(lock_reg, mark_offset));
851 br(Assembler::EQ, count);
852 }
853 bind(slow_case);
854
855 // Call the runtime routine for slow case
856 call_VM(noreg,
857 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
858 lock_reg);
859 b(done);
860
861 bind(count);
862 increment(Address(rthread, JavaThread::held_monitor_count_offset()));
863
864 bind(done);
865 }
866 }
867
868
869 // Unlocks an object. Used in monitorexit bytecode and
870 // remove_activation. Throws an IllegalMonitorException if object is
871 // not locked by current thread.
872 //
873 // Args:
874 // c_rarg1: BasicObjectLock for lock
875 //
876 // Kills:
877 // r0
878 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
879 // rscratch1, rscratch2 (scratch regs)
880 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
881 {
882 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
883
884 if (LockingMode == LM_MONITOR) {
885 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
886 } else {
887 Label count, done;
888
889 const Register swap_reg = r0;
890 const Register header_reg = c_rarg2; // Will contain the old oopMark
891 const Register obj_reg = c_rarg3; // Will contain the oop
892 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
893
894 save_bcp(); // Save in case of exception
895
896 if (LockingMode != LM_LIGHTWEIGHT) {
897 // Convert from BasicObjectLock structure to object and BasicLock
898 // structure Store the BasicLock address into %r0
899 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
900 }
901
902 // Load oop into obj_reg(%c_rarg3)
903 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
904
905 // Free entry
906 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
907
908 if (LockingMode == LM_LIGHTWEIGHT) {
909 Label slow_case;
910 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
911 b(count);
912 bind(slow_case);
913 } else if (LockingMode == LM_LEGACY) {
914 // Load the old header from BasicLock structure
915 ldr(header_reg, Address(swap_reg,
916 BasicLock::displaced_header_offset_in_bytes()));
917
918 // Test for recursion
919 cbz(header_reg, count);
920
921 // Atomic swap back the old header
922 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
923 }
924 // Call the runtime routine for slow case.
925 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
926 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
927 b(done);
928
929 bind(count);
930 decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
931
932 bind(done);
933 restore_bcp();
934 }
935 }
936
937 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
938 Label& zero_continue) {
939 assert(ProfileInterpreter, "must be profiling interpreter");
940 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
941 cbz(mdp, zero_continue);
942 }
943
944 // Set the method data pointer for the current bcp.
945 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
946 assert(ProfileInterpreter, "must be profiling interpreter");
947 Label set_mdp;
948 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
949
950 // Test MDO to avoid the call if it is null.
951 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
952 cbz(r0, set_mdp);
953 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
954 // r0: mdi
955 // mdo is guaranteed to be non-zero here, we checked for it before the call.
956 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
957 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
958 add(r0, r1, r0);
959 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
960 bind(set_mdp);
961 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
962 }
963
964 void InterpreterMacroAssembler::verify_method_data_pointer() {
965 assert(ProfileInterpreter, "must be profiling interpreter");
966 #ifdef ASSERT
967 Label verify_continue;
968 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
969 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
970 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
971 get_method(r1);
972
973 // If the mdp is valid, it will point to a DataLayout header which is
974 // consistent with the bcp. The converse is highly probable also.
975 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
976 ldr(rscratch1, Address(r1, Method::const_offset()));
977 add(r2, r2, rscratch1, Assembler::LSL);
978 lea(r2, Address(r2, ConstMethod::codes_offset()));
979 cmp(r2, rbcp);
980 br(Assembler::EQ, verify_continue);
981 // r1: method
982 // rbcp: bcp // rbcp == 22
983 // r3: mdp
984 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
985 r1, rbcp, r3);
986 bind(verify_continue);
987 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
988 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
989 #endif // ASSERT
990 }
991
992
993 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
994 int constant,
995 Register value) {
996 assert(ProfileInterpreter, "must be profiling interpreter");
997 Address data(mdp_in, constant);
998 str(value, data);
999 }
1000
1001
1002 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1003 int constant,
1004 bool decrement) {
1005 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1006 }
1007
1008 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1009 Register reg,
1010 int constant,
1011 bool decrement) {
1012 assert(ProfileInterpreter, "must be profiling interpreter");
1013 // %%% this does 64bit counters at best it is wasting space
1014 // at worst it is a rare bug when counters overflow
1015
1016 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1017
1018 Address addr1(mdp_in, constant);
1019 Address addr2(rscratch2, reg, Address::lsl(0));
1020 Address &addr = addr1;
1021 if (reg != noreg) {
1022 lea(rscratch2, addr1);
1023 addr = addr2;
1024 }
1025
1026 if (decrement) {
1027 // Decrement the register. Set condition codes.
1028 // Intel does this
1029 // addptr(data, (int32_t) -DataLayout::counter_increment);
1030 // If the decrement causes the counter to overflow, stay negative
1031 // Label L;
1032 // jcc(Assembler::negative, L);
1033 // addptr(data, (int32_t) DataLayout::counter_increment);
1034 // so we do this
1035 ldr(rscratch1, addr);
1036 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1037 Label L;
1038 br(Assembler::LO, L); // skip store if counter underflow
1039 str(rscratch1, addr);
1040 bind(L);
1041 } else {
1042 assert(DataLayout::counter_increment == 1,
1043 "flow-free idiom only works with 1");
1044 // Intel does this
1045 // Increment the register. Set carry flag.
1046 // addptr(data, DataLayout::counter_increment);
1047 // If the increment causes the counter to overflow, pull back by 1.
1048 // sbbptr(data, (int32_t)0);
1049 // so we do this
1050 ldr(rscratch1, addr);
1051 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1052 Label L;
1053 br(Assembler::CS, L); // skip store if counter overflow
1054 str(rscratch1, addr);
1055 bind(L);
1056 }
1057 }
1058
1059 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1060 int flag_byte_constant) {
1061 assert(ProfileInterpreter, "must be profiling interpreter");
1062 int flags_offset = in_bytes(DataLayout::flags_offset());
1063 // Set the flag
1064 ldrb(rscratch1, Address(mdp_in, flags_offset));
1065 orr(rscratch1, rscratch1, flag_byte_constant);
1066 strb(rscratch1, Address(mdp_in, flags_offset));
1067 }
1068
1069
1070 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1071 int offset,
1072 Register value,
1073 Register test_value_out,
1074 Label& not_equal_continue) {
1075 assert(ProfileInterpreter, "must be profiling interpreter");
1076 if (test_value_out == noreg) {
1077 ldr(rscratch1, Address(mdp_in, offset));
1078 cmp(value, rscratch1);
1079 } else {
1080 // Put the test value into a register, so caller can use it:
1081 ldr(test_value_out, Address(mdp_in, offset));
1082 cmp(value, test_value_out);
1083 }
1084 br(Assembler::NE, not_equal_continue);
1085 }
1086
1087
1088 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1089 int offset_of_disp) {
1090 assert(ProfileInterpreter, "must be profiling interpreter");
1091 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1092 add(mdp_in, mdp_in, rscratch1, LSL);
1093 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1094 }
1095
1096
1097 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1098 Register reg,
1099 int offset_of_disp) {
1100 assert(ProfileInterpreter, "must be profiling interpreter");
1101 lea(rscratch1, Address(mdp_in, offset_of_disp));
1102 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1103 add(mdp_in, mdp_in, rscratch1, LSL);
1104 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1105 }
1106
1107
1108 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1109 int constant) {
1110 assert(ProfileInterpreter, "must be profiling interpreter");
1111 add(mdp_in, mdp_in, (unsigned)constant);
1112 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1113 }
1114
1115
1116 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1117 assert(ProfileInterpreter, "must be profiling interpreter");
1118 // save/restore across call_VM
1119 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1120 call_VM(noreg,
1121 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1122 return_bci);
1123 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1124 }
1125
1126
1127 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1128 Register bumped_count) {
1129 if (ProfileInterpreter) {
1130 Label profile_continue;
1131
1132 // If no method data exists, go to profile_continue.
1133 // Otherwise, assign to mdp
1134 test_method_data_pointer(mdp, profile_continue);
1135
1136 // We are taking a branch. Increment the taken count.
1137 // We inline increment_mdp_data_at to return bumped_count in a register
1138 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1139 Address data(mdp, in_bytes(JumpData::taken_offset()));
1140 ldr(bumped_count, data);
1141 assert(DataLayout::counter_increment == 1,
1142 "flow-free idiom only works with 1");
1143 // Intel does this to catch overflow
1144 // addptr(bumped_count, DataLayout::counter_increment);
1145 // sbbptr(bumped_count, 0);
1146 // so we do this
1147 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1148 Label L;
1149 br(Assembler::CS, L); // skip store if counter overflow
1150 str(bumped_count, data);
1151 bind(L);
1152 // The method data pointer needs to be updated to reflect the new target.
1153 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1154 bind(profile_continue);
1155 }
1156 }
1157
1158
1159 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1160 if (ProfileInterpreter) {
1161 Label profile_continue;
1162
1163 // If no method data exists, go to profile_continue.
1164 test_method_data_pointer(mdp, profile_continue);
1165
1166 // We are taking a branch. Increment the not taken count.
1167 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1168
1169 // The method data pointer needs to be updated to correspond to
1170 // the next bytecode
1171 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1172 bind(profile_continue);
1173 }
1174 }
1175
1176
1177 void InterpreterMacroAssembler::profile_call(Register mdp) {
1178 if (ProfileInterpreter) {
1179 Label profile_continue;
1180
1181 // If no method data exists, go to profile_continue.
1182 test_method_data_pointer(mdp, profile_continue);
1183
1184 // We are making a call. Increment the count.
1185 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1186
1187 // The method data pointer needs to be updated to reflect the new target.
1188 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1189 bind(profile_continue);
1190 }
1191 }
1192
1193 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1194 if (ProfileInterpreter) {
1195 Label profile_continue;
1196
1197 // If no method data exists, go to profile_continue.
1198 test_method_data_pointer(mdp, profile_continue);
1199
1200 // We are making a call. Increment the count.
1201 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1202
1203 // The method data pointer needs to be updated to reflect the new target.
1204 update_mdp_by_constant(mdp,
1205 in_bytes(VirtualCallData::
1206 virtual_call_data_size()));
1207 bind(profile_continue);
1208 }
1209 }
1210
1211
1212 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1213 Register mdp,
1214 Register reg2,
1215 bool receiver_can_be_null) {
1216 if (ProfileInterpreter) {
1217 Label profile_continue;
1218
1219 // If no method data exists, go to profile_continue.
1220 test_method_data_pointer(mdp, profile_continue);
1221
1222 Label skip_receiver_profile;
1223 if (receiver_can_be_null) {
1224 Label not_null;
1225 // We are making a call. Increment the count for null receiver.
1226 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1227 b(skip_receiver_profile);
1228 bind(not_null);
1229 }
1230
1231 // Record the receiver type.
1232 record_klass_in_profile(receiver, mdp, reg2);
1233 bind(skip_receiver_profile);
1234
1235 // The method data pointer needs to be updated to reflect the new target.
1236 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1237 bind(profile_continue);
1238 }
1239 }
1240
1241 // This routine creates a state machine for updating the multi-row
1242 // type profile at a virtual call site (or other type-sensitive bytecode).
1243 // The machine visits each row (of receiver/count) until the receiver type
1244 // is found, or until it runs out of rows. At the same time, it remembers
1245 // the location of the first empty row. (An empty row records null for its
1246 // receiver, and can be allocated for a newly-observed receiver type.)
1247 // Because there are two degrees of freedom in the state, a simple linear
1248 // search will not work; it must be a decision tree. Hence this helper
1249 // function is recursive, to generate the required tree structured code.
1250 // It's the interpreter, so we are trading off code space for speed.
1251 // See below for example code.
1252 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1253 Register receiver, Register mdp,
1254 Register reg2, int start_row,
1255 Label& done) {
1256 if (TypeProfileWidth == 0) {
1257 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1258 } else {
1259 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1260 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1261 }
1262 }
1263
1264 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1265 Register reg2, int start_row, Label& done, int total_rows,
1266 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1267 int last_row = total_rows - 1;
1268 assert(start_row <= last_row, "must be work left to do");
1269 // Test this row for both the item and for null.
1270 // Take any of three different outcomes:
1271 // 1. found item => increment count and goto done
1272 // 2. found null => keep looking for case 1, maybe allocate this cell
1273 // 3. found something else => keep looking for cases 1 and 2
1274 // Case 3 is handled by a recursive call.
1275 for (int row = start_row; row <= last_row; row++) {
1276 Label next_test;
1277 bool test_for_null_also = (row == start_row);
1278
1279 // See if the item is item[n].
1280 int item_offset = in_bytes(item_offset_fn(row));
1281 test_mdp_data_at(mdp, item_offset, item,
1282 (test_for_null_also ? reg2 : noreg),
1283 next_test);
1284 // (Reg2 now contains the item from the CallData.)
1285
1286 // The item is item[n]. Increment count[n].
1287 int count_offset = in_bytes(item_count_offset_fn(row));
1288 increment_mdp_data_at(mdp, count_offset);
1289 b(done);
1290 bind(next_test);
1291
1292 if (test_for_null_also) {
1293 Label found_null;
1294 // Failed the equality check on item[n]... Test for null.
1295 if (start_row == last_row) {
1296 // The only thing left to do is handle the null case.
1297 cbz(reg2, found_null);
1298 // Item did not match any saved item and there is no empty row for it.
1299 // Increment total counter to indicate polymorphic case.
1300 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1301 b(done);
1302 bind(found_null);
1303 break;
1304 }
1305 // Since null is rare, make it be the branch-taken case.
1306 cbz(reg2, found_null);
1307
1308 // Put all the "Case 3" tests here.
1309 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1310 item_offset_fn, item_count_offset_fn);
1311
1312 // Found a null. Keep searching for a matching item,
1313 // but remember that this is an empty (unused) slot.
1314 bind(found_null);
1315 }
1316 }
1317
1318 // In the fall-through case, we found no matching item, but we
1319 // observed the item[start_row] is null.
1320
1321 // Fill in the item field and increment the count.
1322 int item_offset = in_bytes(item_offset_fn(start_row));
1323 set_mdp_data_at(mdp, item_offset, item);
1324 int count_offset = in_bytes(item_count_offset_fn(start_row));
1325 mov(reg2, DataLayout::counter_increment);
1326 set_mdp_data_at(mdp, count_offset, reg2);
1327 if (start_row > 0) {
1328 b(done);
1329 }
1330 }
1331
1332 // Example state machine code for three profile rows:
1333 // // main copy of decision tree, rooted at row[1]
1334 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1335 // if (row[0].rec != nullptr) {
1336 // // inner copy of decision tree, rooted at row[1]
1337 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1338 // if (row[1].rec != nullptr) {
1339 // // degenerate decision tree, rooted at row[2]
1340 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1341 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1342 // row[2].init(rec); goto done;
1343 // } else {
1344 // // remember row[1] is empty
1345 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1346 // row[1].init(rec); goto done;
1347 // }
1348 // } else {
1349 // // remember row[0] is empty
1350 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1351 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1352 // row[0].init(rec); goto done;
1353 // }
1354 // done:
1355
1356 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1357 Register mdp, Register reg2) {
1358 assert(ProfileInterpreter, "must be profiling");
1359 Label done;
1360
1361 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1362
1363 bind (done);
1364 }
1365
1366 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1367 Register mdp) {
1368 if (ProfileInterpreter) {
1369 Label profile_continue;
1370 uint row;
1371
1372 // If no method data exists, go to profile_continue.
1373 test_method_data_pointer(mdp, profile_continue);
1374
1375 // Update the total ret count.
1376 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1377
1378 for (row = 0; row < RetData::row_limit(); row++) {
1379 Label next_test;
1380
1381 // See if return_bci is equal to bci[n]:
1382 test_mdp_data_at(mdp,
1383 in_bytes(RetData::bci_offset(row)),
1384 return_bci, noreg,
1385 next_test);
1386
1387 // return_bci is equal to bci[n]. Increment the count.
1388 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1389
1390 // The method data pointer needs to be updated to reflect the new target.
1391 update_mdp_by_offset(mdp,
1392 in_bytes(RetData::bci_displacement_offset(row)));
1393 b(profile_continue);
1394 bind(next_test);
1395 }
1396
1397 update_mdp_for_ret(return_bci);
1398
1399 bind(profile_continue);
1400 }
1401 }
1402
1403 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1404 if (ProfileInterpreter) {
1405 Label profile_continue;
1406
1407 // If no method data exists, go to profile_continue.
1408 test_method_data_pointer(mdp, profile_continue);
1409
1410 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1411
1412 // The method data pointer needs to be updated.
1413 int mdp_delta = in_bytes(BitData::bit_data_size());
1414 if (TypeProfileCasts) {
1415 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1416 }
1417 update_mdp_by_constant(mdp, mdp_delta);
1418
1419 bind(profile_continue);
1420 }
1421 }
1422
1423 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1424 if (ProfileInterpreter) {
1425 Label profile_continue;
1426
1427 // If no method data exists, go to profile_continue.
1428 test_method_data_pointer(mdp, profile_continue);
1429
1430 // The method data pointer needs to be updated.
1431 int mdp_delta = in_bytes(BitData::bit_data_size());
1432 if (TypeProfileCasts) {
1433 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1434
1435 // Record the object type.
1436 record_klass_in_profile(klass, mdp, reg2);
1437 }
1438 update_mdp_by_constant(mdp, mdp_delta);
1439
1440 bind(profile_continue);
1441 }
1442 }
1443
1444 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1445 if (ProfileInterpreter) {
1446 Label profile_continue;
1447
1448 // If no method data exists, go to profile_continue.
1449 test_method_data_pointer(mdp, profile_continue);
1450
1451 // Update the default case count
1452 increment_mdp_data_at(mdp,
1453 in_bytes(MultiBranchData::default_count_offset()));
1454
1455 // The method data pointer needs to be updated.
1456 update_mdp_by_offset(mdp,
1457 in_bytes(MultiBranchData::
1458 default_displacement_offset()));
1459
1460 bind(profile_continue);
1461 }
1462 }
1463
1464 void InterpreterMacroAssembler::profile_switch_case(Register index,
1465 Register mdp,
1466 Register reg2) {
1467 if (ProfileInterpreter) {
1468 Label profile_continue;
1469
1470 // If no method data exists, go to profile_continue.
1471 test_method_data_pointer(mdp, profile_continue);
1472
1473 // Build the base (index * per_case_size_in_bytes()) +
1474 // case_array_offset_in_bytes()
1475 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1476 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1477 Assembler::maddw(index, index, reg2, rscratch1);
1478
1479 // Update the case count
1480 increment_mdp_data_at(mdp,
1481 index,
1482 in_bytes(MultiBranchData::relative_count_offset()));
1483
1484 // The method data pointer needs to be updated.
1485 update_mdp_by_offset(mdp,
1486 index,
1487 in_bytes(MultiBranchData::
1488 relative_displacement_offset()));
1489
1490 bind(profile_continue);
1491 }
1492 }
1493
1494 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1495 Register array,
1496 Register tmp) {
1497 if (ProfileInterpreter) {
1498 Label profile_continue;
1499
1500 // If no method data exists, go to profile_continue.
1501 test_method_data_pointer(mdp, profile_continue);
1502
1503 mov(tmp, array);
1504 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1505
1506 Label not_flat;
1507 test_non_flat_array_oop(array, tmp, not_flat);
1508
1509 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1510
1511 bind(not_flat);
1512
1513 Label not_null_free;
1514 test_non_null_free_array_oop(array, tmp, not_null_free);
1515
1516 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1517
1518 bind(not_null_free);
1519
1520 bind(profile_continue);
1521 }
1522 }
1523
1524 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1525 Register array,
1526 Register tmp);
1527 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1528 Register array,
1529 Register tmp);
1530
1531 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1532 if (ProfileInterpreter) {
1533 Label profile_continue;
1534
1535 // If no method data exists, go to profile_continue.
1536 test_method_data_pointer(mdp, profile_continue);
1537
1538 Label done, update;
1539 cbnz(element, update);
1540 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1541 b(done);
1542
1543 bind(update);
1544 load_klass(tmp, element);
1545
1546 // Record the object type.
1547 record_klass_in_profile(tmp, mdp, tmp2);
1548
1549 bind(done);
1550
1551 // The method data pointer needs to be updated.
1552 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1553
1554 bind(profile_continue);
1555 }
1556 }
1557
1558
1559 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1560 Register element,
1561 Register tmp) {
1562 if (ProfileInterpreter) {
1563 Label profile_continue;
1564
1565 // If no method data exists, go to profile_continue.
1566 test_method_data_pointer(mdp, profile_continue);
1567
1568 mov(tmp, element);
1569 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1570
1571 // The method data pointer needs to be updated.
1572 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1573
1574 bind(profile_continue);
1575 }
1576 }
1577
1578 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1579 Register left,
1580 Register right,
1581 Register tmp) {
1582 if (ProfileInterpreter) {
1583 Label profile_continue;
1584
1585 // If no method data exists, go to profile_continue.
1586 test_method_data_pointer(mdp, profile_continue);
1587
1588 mov(tmp, left);
1589 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1590
1591 Label left_not_inline_type;
1592 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1593 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1594 bind(left_not_inline_type);
1595
1596 mov(tmp, right);
1597 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1598
1599 Label right_not_inline_type;
1600 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1601 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1602 bind(right_not_inline_type);
1603
1604 bind(profile_continue);
1605 }
1606 }
1607
1608 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1609 if (state == atos) {
1610 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1611 }
1612 }
1613
1614 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1615
1616
1617 void InterpreterMacroAssembler::notify_method_entry() {
1618 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1619 // track stack depth. If it is possible to enter interp_only_mode we add
1620 // the code to check if the event should be sent.
1621 if (JvmtiExport::can_post_interpreter_events()) {
1622 Label L;
1623 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1624 cbzw(r3, L);
1625 call_VM(noreg, CAST_FROM_FN_PTR(address,
1626 InterpreterRuntime::post_method_entry));
1627 bind(L);
1628 }
1629
1630 if (DTraceMethodProbes) {
1631 get_method(c_rarg1);
1632 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1633 rthread, c_rarg1);
1634 }
1635
1636 // RedefineClasses() tracing support for obsolete method entry
1637 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1638 get_method(c_rarg1);
1639 call_VM_leaf(
1640 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1641 rthread, c_rarg1);
1642 }
1643
1644 }
1645
1646
1647 void InterpreterMacroAssembler::notify_method_exit(
1648 TosState state, NotifyMethodExitMode mode) {
1649 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1650 // track stack depth. If it is possible to enter interp_only_mode we add
1651 // the code to check if the event should be sent.
1652 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1653 Label L;
1654 // Note: frame::interpreter_frame_result has a dependency on how the
1655 // method result is saved across the call to post_method_exit. If this
1656 // is changed then the interpreter_frame_result implementation will
1657 // need to be updated too.
1658
1659 // template interpreter will leave the result on the top of the stack.
1660 push(state);
1661 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1662 cbz(r3, L);
1663 call_VM(noreg,
1664 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1665 bind(L);
1666 pop(state);
1667 }
1668
1669 if (DTraceMethodProbes) {
1670 push(state);
1671 get_method(c_rarg1);
1672 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1673 rthread, c_rarg1);
1674 pop(state);
1675 }
1676 }
1677
1678
1679 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1680 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1681 int increment, Address mask,
1682 Register scratch, Register scratch2,
1683 bool preloaded, Condition cond,
1684 Label* where) {
1685 if (!preloaded) {
1686 ldrw(scratch, counter_addr);
1687 }
1688 add(scratch, scratch, increment);
1689 strw(scratch, counter_addr);
1690 ldrw(scratch2, mask);
1691 ands(scratch, scratch, scratch2);
1692 br(cond, *where);
1693 }
1694
1695 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1696 int number_of_arguments) {
1697 // interpreter specific
1698 //
1699 // Note: No need to save/restore rbcp & rlocals pointer since these
1700 // are callee saved registers and no blocking/ GC can happen
1701 // in leaf calls.
1702 #ifdef ASSERT
1703 {
1704 Label L;
1705 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1706 cbz(rscratch1, L);
1707 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1708 " last_sp != nullptr");
1709 bind(L);
1710 }
1711 #endif /* ASSERT */
1712 // super call
1713 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1714 }
1715
1716 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1717 Register java_thread,
1718 Register last_java_sp,
1719 address entry_point,
1720 int number_of_arguments,
1721 bool check_exceptions) {
1722 // interpreter specific
1723 //
1724 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1725 // really make a difference for these runtime calls, since they are
1726 // slow anyway. Btw., bcp must be saved/restored since it may change
1727 // due to GC.
1728 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1729 save_bcp();
1730 #ifdef ASSERT
1731 {
1732 Label L;
1733 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1734 cbz(rscratch1, L);
1735 stop("InterpreterMacroAssembler::call_VM_base:"
1736 " last_sp != nullptr");
1737 bind(L);
1738 }
1739 #endif /* ASSERT */
1740 // super call
1741 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1742 entry_point, number_of_arguments,
1743 check_exceptions);
1744 // interpreter specific
1745 restore_bcp();
1746 restore_locals();
1747 }
1748
1749 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1750 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1751 Label update, next, none;
1752
1753 verify_oop(obj);
1754
1755 cbnz(obj, update);
1756 orptr(mdo_addr, TypeEntries::null_seen);
1757 b(next);
1758
1759 bind(update);
1760 load_klass(obj, obj);
1761
1762 ldr(rscratch1, mdo_addr);
1763 eor(obj, obj, rscratch1);
1764 tst(obj, TypeEntries::type_klass_mask);
1765 br(Assembler::EQ, next); // klass seen before, nothing to
1766 // do. The unknown bit may have been
1767 // set already but no need to check.
1768
1769 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1770 // already unknown. Nothing to do anymore.
1771
1772 cbz(rscratch1, none);
1773 cmp(rscratch1, (u1)TypeEntries::null_seen);
1774 br(Assembler::EQ, none);
1775 // There is a chance that the checks above
1776 // fail if another thread has just set the
1777 // profiling to this obj's klass
1778 eor(obj, obj, rscratch1); // get back original value before XOR
1779 ldr(rscratch1, mdo_addr);
1780 eor(obj, obj, rscratch1);
1781 tst(obj, TypeEntries::type_klass_mask);
1782 br(Assembler::EQ, next);
1783
1784 // different than before. Cannot keep accurate profile.
1785 orptr(mdo_addr, TypeEntries::type_unknown);
1786 b(next);
1787
1788 bind(none);
1789 // first time here. Set profile type.
1790 str(obj, mdo_addr);
1791 #ifdef ASSERT
1792 andr(obj, obj, TypeEntries::type_mask);
1793 verify_klass_ptr(obj);
1794 #endif
1795
1796 bind(next);
1797 }
1798
1799 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1800 if (!ProfileInterpreter) {
1801 return;
1802 }
1803
1804 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1805 Label profile_continue;
1806
1807 test_method_data_pointer(mdp, profile_continue);
1808
1809 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1810
1811 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1812 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1813 br(Assembler::NE, profile_continue);
1814
1815 if (MethodData::profile_arguments()) {
1816 Label done;
1817 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1818
1819 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1820 if (i > 0 || MethodData::profile_return()) {
1821 // If return value type is profiled we may have no argument to profile
1822 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1823 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1824 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1825 add(rscratch1, mdp, off_to_args);
1826 br(Assembler::LT, done);
1827 }
1828 ldr(tmp, Address(callee, Method::const_offset()));
1829 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1830 // stack offset o (zero based) from the start of the argument
1831 // list, for n arguments translates into offset n - o - 1 from
1832 // the end of the argument list
1833 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1834 sub(tmp, tmp, rscratch1);
1835 sub(tmp, tmp, 1);
1836 Address arg_addr = argument_address(tmp);
1837 ldr(tmp, arg_addr);
1838
1839 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1840 profile_obj_type(tmp, mdo_arg_addr);
1841
1842 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1843 off_to_args += to_add;
1844 }
1845
1846 if (MethodData::profile_return()) {
1847 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1848 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1849 }
1850
1851 add(rscratch1, mdp, off_to_args);
1852 bind(done);
1853 mov(mdp, rscratch1);
1854
1855 if (MethodData::profile_return()) {
1856 // We're right after the type profile for the last
1857 // argument. tmp is the number of cells left in the
1858 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1859 // if there's a return to profile.
1860 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1861 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1862 }
1863 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1864 } else {
1865 assert(MethodData::profile_return(), "either profile call args or call ret");
1866 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1867 }
1868
1869 // mdp points right after the end of the
1870 // CallTypeData/VirtualCallTypeData, right after the cells for the
1871 // return value type if there's one
1872
1873 bind(profile_continue);
1874 }
1875 }
1876
1877 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1878 assert_different_registers(mdp, ret, tmp, rbcp);
1879 if (ProfileInterpreter && MethodData::profile_return()) {
1880 Label profile_continue, done;
1881
1882 test_method_data_pointer(mdp, profile_continue);
1883
1884 if (MethodData::profile_return_jsr292_only()) {
1885 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1886
1887 // If we don't profile all invoke bytecodes we must make sure
1888 // it's a bytecode we indeed profile. We can't go back to the
1889 // beginning of the ProfileData we intend to update to check its
1890 // type because we're right after it and we don't known its
1891 // length
1892 Label do_profile;
1893 ldrb(rscratch1, Address(rbcp, 0));
1894 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1895 br(Assembler::EQ, do_profile);
1896 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1897 br(Assembler::EQ, do_profile);
1898 get_method(tmp);
1899 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1900 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1901 br(Assembler::NE, profile_continue);
1902
1903 bind(do_profile);
1904 }
1905
1906 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1907 mov(tmp, ret);
1908 profile_obj_type(tmp, mdo_ret_addr);
1909
1910 bind(profile_continue);
1911 }
1912 }
1913
1914 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1915 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1916 if (ProfileInterpreter && MethodData::profile_parameters()) {
1917 Label profile_continue, done;
1918
1919 test_method_data_pointer(mdp, profile_continue);
1920
1921 // Load the offset of the area within the MDO used for
1922 // parameters. If it's negative we're not profiling any parameters
1923 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1924 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1925
1926 // Compute a pointer to the area for parameters from the offset
1927 // and move the pointer to the slot for the last
1928 // parameters. Collect profiling from last parameter down.
1929 // mdo start + parameters offset + array length - 1
1930 add(mdp, mdp, tmp1);
1931 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1932 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1933
1934 Label loop;
1935 bind(loop);
1936
1937 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1938 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1939 int per_arg_scale = exact_log2(DataLayout::cell_size);
1940 add(rscratch1, mdp, off_base);
1941 add(rscratch2, mdp, type_base);
1942
1943 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1944 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1945
1946 // load offset on the stack from the slot for this parameter
1947 ldr(tmp2, arg_off);
1948 neg(tmp2, tmp2);
1949 // read the parameter from the local area
1950 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1951
1952 // profile the parameter
1953 profile_obj_type(tmp2, arg_type);
1954
1955 // go to next parameter
1956 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1957 br(Assembler::GE, loop);
1958
1959 bind(profile_continue);
1960 }
1961 }
1962
1963 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1964 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1965 get_cache_index_at_bcp(index, 1, sizeof(u4));
1966 // Get address of invokedynamic array
1967 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1968 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1969 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1970 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1971 lea(cache, Address(cache, index));
1972 }
1973
1974 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1975 // Get index out of bytecode pointer
1976 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1977 // Take shortcut if the size is a power of 2
1978 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1979 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1980 } else {
1981 mov(cache, sizeof(ResolvedFieldEntry));
1982 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1983 }
1984 // Get address of field entries array
1985 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1986 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1987 lea(cache, Address(cache, index));
1988 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1989 membar(MacroAssembler::LoadLoad);
1990 }
1991
1992 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1993 // Get index out of bytecode pointer
1994 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1995 mov(cache, sizeof(ResolvedMethodEntry));
1996 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1997
1998 // Get address of field entries array
1999 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
2000 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2001 lea(cache, Address(cache, index));
2002 }