1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "compiler/compiler_globals.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_aarch64.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/basicLock.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/javaThread.hpp"
44 #include "runtime/safepointMechanism.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "utilities/powerOfTwo.hpp"
47
48 void InterpreterMacroAssembler::narrow(Register result) {
49
50 // Get method->_constMethod->_result_type
51 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
52 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
53 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
54
55 Label done, notBool, notByte, notChar;
56
57 // common case first
58 cmpw(rscratch1, T_INT);
59 br(Assembler::EQ, done);
60
61 // mask integer result to narrower return type.
62 cmpw(rscratch1, T_BOOLEAN);
63 br(Assembler::NE, notBool);
64 andw(result, result, 0x1);
65 b(done);
66
67 bind(notBool);
68 cmpw(rscratch1, T_BYTE);
69 br(Assembler::NE, notByte);
70 sbfx(result, result, 0, 8);
71 b(done);
72
73 bind(notByte);
74 cmpw(rscratch1, T_CHAR);
75 br(Assembler::NE, notChar);
76 ubfx(result, result, 0, 16); // truncate upper 16 bits
77 b(done);
78
79 bind(notChar);
80 sbfx(result, result, 0, 16); // sign-extend short
81
82 // Nothing to do for T_INT
83 bind(done);
84 }
85
86 void InterpreterMacroAssembler::jump_to_entry(address entry) {
87 assert(entry, "Entry must have been generated by now");
88 b(entry);
89 }
90
91 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
92 if (JvmtiExport::can_pop_frame()) {
93 Label L;
94 // Initiate popframe handling only if it is not already being
95 // processed. If the flag has the popframe_processing bit set, it
96 // means that this code is called *during* popframe handling - we
97 // don't want to reenter.
98 // This method is only called just after the call into the vm in
99 // call_VM_base, so the arg registers are available.
100 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
101 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
102 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
103 // Call Interpreter::remove_activation_preserving_args_entry() to get the
104 // address of the same-named entrypoint in the generated interpreter code.
105 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
106 br(r0);
107 bind(L);
108 }
109 }
110
111
112 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
113 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
114 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
115 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
116 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
117 switch (state) {
118 case atos: ldr(r0, oop_addr);
119 str(zr, oop_addr);
120 interp_verify_oop(r0, state); break;
121 case ltos: ldr(r0, val_addr); break;
122 case btos: // fall through
123 case ztos: // fall through
124 case ctos: // fall through
125 case stos: // fall through
126 case itos: ldrw(r0, val_addr); break;
127 case ftos: ldrs(v0, val_addr); break;
128 case dtos: ldrd(v0, val_addr); break;
129 case vtos: /* nothing to do */ break;
130 default : ShouldNotReachHere();
131 }
132 // Clean up tos value in the thread object
133 movw(rscratch1, (int) ilgl);
134 strw(rscratch1, tos_addr);
135 strw(zr, val_addr);
136 }
137
138
139 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
140 if (JvmtiExport::can_force_early_return()) {
141 Label L;
142 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
143 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
144
145 // Initiate earlyret handling only if it is not already being processed.
146 // If the flag has the earlyret_processing bit set, it means that this code
147 // is called *during* earlyret handling - we don't want to reenter.
148 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
149 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
150 br(Assembler::NE, L);
151
152 // Call Interpreter::remove_activation_early_entry() to get the address of the
153 // same-named entrypoint in the generated interpreter code.
154 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
155 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
156 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
157 br(r0);
158 bind(L);
159 }
160 }
161
162 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
163 Register reg,
164 int bcp_offset) {
165 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
166 ldrh(reg, Address(rbcp, bcp_offset));
167 rev16(reg, reg);
168 }
169
170 void InterpreterMacroAssembler::get_dispatch() {
171 uint64_t offset;
172 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
173 // Use add() here after ARDP, rather than lea().
174 // lea() does not generate anything if its offset is zero.
175 // However, relocs expect to find either an ADD or a load/store
176 // insn after an ADRP. add() always generates an ADD insn, even
177 // for add(Rn, Rn, 0).
178 add(rdispatch, rdispatch, offset);
179 }
180
181 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
182 int bcp_offset,
183 size_t index_size) {
184 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
185 if (index_size == sizeof(u2)) {
186 load_unsigned_short(index, Address(rbcp, bcp_offset));
187 } else if (index_size == sizeof(u4)) {
188 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
189 ldrw(index, Address(rbcp, bcp_offset));
190 // Check if the secondary index definition is still ~x, otherwise
191 // we have to change the following assembler code to calculate the
192 // plain index.
193 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
194 eonw(index, index, zr); // convert to plain index
195 } else if (index_size == sizeof(u1)) {
196 load_unsigned_byte(index, Address(rbcp, bcp_offset));
197 } else {
198 ShouldNotReachHere();
199 }
200 }
201
202 // Return
203 // Rindex: index into constant pool
204 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
205 //
206 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
207 // the true address of the cache entry.
208 //
209 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
210 Register index,
211 int bcp_offset,
212 size_t index_size) {
213 assert_different_registers(cache, index);
214 assert_different_registers(cache, rcpool);
215 get_cache_index_at_bcp(index, bcp_offset, index_size);
216 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
217 // convert from field index to ConstantPoolCacheEntry
218 // aarch64 already has the cache in rcpool so there is no need to
219 // install it in cache. instead we pre-add the indexed offset to
220 // rcpool and return it in cache. All clients of this method need to
221 // be modified accordingly.
222 add(cache, rcpool, index, Assembler::LSL, 5);
223 }
224
225
226 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
227 Register index,
228 Register bytecode,
229 int byte_no,
230 int bcp_offset,
231 size_t index_size) {
232 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
233 // We use a 32-bit load here since the layout of 64-bit words on
234 // little-endian machines allow us that.
235 // n.b. unlike x86 cache already includes the index offset
236 lea(bytecode, Address(cache,
237 ConstantPoolCache::base_offset()
238 + ConstantPoolCacheEntry::indices_offset()));
239 ldarw(bytecode, bytecode);
240 const int shift_count = (1 + byte_no) * BitsPerByte;
241 ubfx(bytecode, bytecode, shift_count, BitsPerByte);
242 }
243
244 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
245 Register tmp,
246 int bcp_offset,
247 size_t index_size) {
248 assert(cache != tmp, "must use different register");
249 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
250 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
251 // convert from field index to ConstantPoolCacheEntry index
252 // and from word offset to byte offset
253 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
254 ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
255 // skip past the header
256 add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
257 add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord); // construct pointer to cache entry
258 }
259
260 void InterpreterMacroAssembler::get_method_counters(Register method,
261 Register mcs, Label& skip) {
262 Label has_counters;
263 ldr(mcs, Address(method, Method::method_counters_offset()));
264 cbnz(mcs, has_counters);
265 call_VM(noreg, CAST_FROM_FN_PTR(address,
266 InterpreterRuntime::build_method_counters), method);
267 ldr(mcs, Address(method, Method::method_counters_offset()));
268 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
269 bind(has_counters);
270 }
271
272 // Load object from cpool->resolved_references(index)
273 void InterpreterMacroAssembler::load_resolved_reference_at_index(
274 Register result, Register index, Register tmp) {
275 assert_different_registers(result, index);
276
277 get_constant_pool(result);
278 // load pointer for resolved_references[] objArray
279 ldr(result, Address(result, ConstantPool::cache_offset()));
280 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
281 resolve_oop_handle(result, tmp, rscratch2);
282 // Add in the index
283 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
284 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
285 }
286
287 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
288 Register cpool, Register index, Register klass, Register temp) {
289 add(temp, cpool, index, LSL, LogBytesPerWord);
290 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
291 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
292 add(klass, klass, temp, LSL, LogBytesPerWord);
293 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
294 }
295
296 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
297 Register method,
298 Register cache) {
299 const int method_offset = in_bytes(
300 ConstantPoolCache::base_offset() +
301 ((byte_no == TemplateTable::f2_byte)
302 ? ConstantPoolCacheEntry::f2_offset()
303 : ConstantPoolCacheEntry::f1_offset()));
304
305 ldr(method, Address(cache, method_offset)); // get f1 Method*
306 }
307
308 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
309 // subtype of super_klass.
310 //
311 // Args:
312 // r0: superklass
313 // Rsub_klass: subklass
314 //
315 // Kills:
316 // r2, r5
317 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
318 Label& ok_is_subtype) {
319 assert(Rsub_klass != r0, "r0 holds superklass");
320 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
321 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
322
323 // Profile the not-null value's klass.
324 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
325
326 // Do the check.
327 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
328
329 // Profile the failure of the check.
330 profile_typecheck_failed(r2); // blows r2
331 }
332
333 // Java Expression Stack
334
335 void InterpreterMacroAssembler::pop_ptr(Register r) {
336 ldr(r, post(esp, wordSize));
337 }
338
339 void InterpreterMacroAssembler::pop_i(Register r) {
340 ldrw(r, post(esp, wordSize));
341 }
342
343 void InterpreterMacroAssembler::pop_l(Register r) {
344 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
345 }
346
347 void InterpreterMacroAssembler::push_ptr(Register r) {
348 str(r, pre(esp, -wordSize));
349 }
350
351 void InterpreterMacroAssembler::push_i(Register r) {
352 str(r, pre(esp, -wordSize));
353 }
354
355 void InterpreterMacroAssembler::push_l(Register r) {
356 str(zr, pre(esp, -wordSize));
357 str(r, pre(esp, - wordSize));
358 }
359
360 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
361 ldrs(r, post(esp, wordSize));
362 }
363
364 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
365 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
366 }
367
368 void InterpreterMacroAssembler::push_f(FloatRegister r) {
369 strs(r, pre(esp, -wordSize));
370 }
371
372 void InterpreterMacroAssembler::push_d(FloatRegister r) {
373 strd(r, pre(esp, 2* -wordSize));
374 }
375
376 void InterpreterMacroAssembler::pop(TosState state) {
377 switch (state) {
378 case atos: pop_ptr(); break;
379 case btos:
380 case ztos:
381 case ctos:
382 case stos:
383 case itos: pop_i(); break;
384 case ltos: pop_l(); break;
385 case ftos: pop_f(); break;
386 case dtos: pop_d(); break;
387 case vtos: /* nothing to do */ break;
388 default: ShouldNotReachHere();
389 }
390 interp_verify_oop(r0, state);
391 }
392
393 void InterpreterMacroAssembler::push(TosState state) {
394 interp_verify_oop(r0, state);
395 switch (state) {
396 case atos: push_ptr(); break;
397 case btos:
398 case ztos:
399 case ctos:
400 case stos:
401 case itos: push_i(); break;
402 case ltos: push_l(); break;
403 case ftos: push_f(); break;
404 case dtos: push_d(); break;
405 case vtos: /* nothing to do */ break;
406 default : ShouldNotReachHere();
407 }
408 }
409
410 // Helpers for swap and dup
411 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
412 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
413 }
414
415 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
416 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
417 }
418
419 void InterpreterMacroAssembler::load_float(Address src) {
420 ldrs(v0, src);
421 }
422
423 void InterpreterMacroAssembler::load_double(Address src) {
424 ldrd(v0, src);
425 }
426
427 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
428 // set sender sp
429 mov(r19_sender_sp, sp);
430 // record last_sp
431 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
432 }
433
434 // Jump to from_interpreted entry of a call unless single stepping is possible
435 // in this thread in which case we must call the i2i entry
436 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
437 prepare_to_jump_from_interpreted();
438
439 if (JvmtiExport::can_post_interpreter_events()) {
440 Label run_compiled_code;
441 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
442 // compiled code in threads for which the event is enabled. Check here for
443 // interp_only_mode if these events CAN be enabled.
444 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
445 cbzw(rscratch1, run_compiled_code);
446 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
447 br(rscratch1);
448 bind(run_compiled_code);
449 }
450
451 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
452 br(rscratch1);
453 }
454
455 // The following two routines provide a hook so that an implementation
456 // can schedule the dispatch in two parts. amd64 does not do this.
457 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
458 }
459
460 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
461 dispatch_next(state, step);
462 }
463
464 void InterpreterMacroAssembler::dispatch_base(TosState state,
465 address* table,
466 bool verifyoop,
467 bool generate_poll) {
468 if (VerifyActivationFrameSize) {
469 Unimplemented();
470 }
471 if (verifyoop) {
472 interp_verify_oop(r0, state);
473 }
474
475 Label safepoint;
476 address* const safepoint_table = Interpreter::safept_table(state);
477 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
478
479 if (needs_thread_local_poll) {
480 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
481 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
482 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
483 }
484
485 if (table == Interpreter::dispatch_table(state)) {
486 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
487 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
488 } else {
489 mov(rscratch2, (address)table);
490 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
491 }
492 br(rscratch2);
493
494 if (needs_thread_local_poll) {
495 bind(safepoint);
496 lea(rscratch2, ExternalAddress((address)safepoint_table));
497 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
498 br(rscratch2);
499 }
500 }
501
502 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
503 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
504 }
505
506 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
507 dispatch_base(state, Interpreter::normal_table(state));
508 }
509
510 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
511 dispatch_base(state, Interpreter::normal_table(state), false);
512 }
513
514
515 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
516 // load next bytecode
517 ldrb(rscratch1, Address(pre(rbcp, step)));
518 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
519 }
520
521 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
522 // load current bytecode
523 ldrb(rscratch1, Address(rbcp, 0));
524 dispatch_base(state, table);
525 }
526
527 // remove activation
528 //
529 // Apply stack watermark barrier.
530 // Unlock the receiver if this is a synchronized method.
531 // Unlock any Java monitors from synchronized blocks.
532 // Remove the activation from the stack.
533 //
534 // If there are locked Java monitors
535 // If throw_monitor_exception
536 // throws IllegalMonitorStateException
537 // Else if install_monitor_exception
538 // installs IllegalMonitorStateException
539 // Else
540 // no error processing
541 void InterpreterMacroAssembler::remove_activation(
542 TosState state,
543 bool throw_monitor_exception,
544 bool install_monitor_exception,
545 bool notify_jvmdi) {
546 // Note: Registers r3 xmm0 may be in use for the
547 // result check if synchronized method
548 Label unlocked, unlock, no_unlock;
549
550 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
551 // that would normally not be safe to use. Such bad returns into unsafe territory of
552 // the stack, will call InterpreterRuntime::at_unwind.
553 Label slow_path;
554 Label fast_path;
555 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
556 br(Assembler::AL, fast_path);
557 bind(slow_path);
558 push(state);
559 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
560 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
561 reset_last_Java_frame(true);
562 pop(state);
563 bind(fast_path);
564
565 // get the value of _do_not_unlock_if_synchronized into r3
566 const Address do_not_unlock_if_synchronized(rthread,
567 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
568 ldrb(r3, do_not_unlock_if_synchronized);
569 strb(zr, do_not_unlock_if_synchronized); // reset the flag
570
571 // get method access flags
572 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
573 ldr(r2, Address(r1, Method::access_flags_offset()));
574 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
575
576 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
577 // is set.
578 cbnz(r3, no_unlock);
579
580 // unlock monitor
581 push(state); // save result
582
583 // BasicObjectLock will be first in list, since this is a
584 // synchronized method. However, need to check that the object has
585 // not been unlocked by an explicit monitorexit bytecode.
586 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
587 wordSize - (int) sizeof(BasicObjectLock));
588 // We use c_rarg1 so that if we go slow path it will be the correct
589 // register for unlock_object to pass to VM directly
590 lea(c_rarg1, monitor); // address of first monitor
591
592 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
593 cbnz(r0, unlock);
594
595 pop(state);
596 if (throw_monitor_exception) {
597 // Entry already unlocked, need to throw exception
598 call_VM(noreg, CAST_FROM_FN_PTR(address,
599 InterpreterRuntime::throw_illegal_monitor_state_exception));
600 should_not_reach_here();
601 } else {
602 // Monitor already unlocked during a stack unroll. If requested,
603 // install an illegal_monitor_state_exception. Continue with
604 // stack unrolling.
605 if (install_monitor_exception) {
606 call_VM(noreg, CAST_FROM_FN_PTR(address,
607 InterpreterRuntime::new_illegal_monitor_state_exception));
608 }
609 b(unlocked);
610 }
611
612 bind(unlock);
613 unlock_object(c_rarg1);
614 pop(state);
615
616 // Check that for block-structured locking (i.e., that all locked
617 // objects has been unlocked)
618 bind(unlocked);
619
620 // r0: Might contain return value
621
622 // Check that all monitors are unlocked
623 {
624 Label loop, exception, entry, restart;
625 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
626 const Address monitor_block_top(
627 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
628 const Address monitor_block_bot(
629 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
630
631 bind(restart);
632 // We use c_rarg1 so that if we go slow path it will be the correct
633 // register for unlock_object to pass to VM directly
634 ldr(c_rarg1, monitor_block_top); // points to current entry, starting
635 // with top-most entry
636 lea(r19, monitor_block_bot); // points to word before bottom of
637 // monitor block
638 b(entry);
639
640 // Entry already locked, need to throw exception
641 bind(exception);
642
643 if (throw_monitor_exception) {
644 // Throw exception
645 MacroAssembler::call_VM(noreg,
646 CAST_FROM_FN_PTR(address, InterpreterRuntime::
647 throw_illegal_monitor_state_exception));
648 should_not_reach_here();
649 } else {
650 // Stack unrolling. Unlock object and install illegal_monitor_exception.
651 // Unlock does not block, so don't have to worry about the frame.
652 // We don't have to preserve c_rarg1 since we are going to throw an exception.
653
654 push(state);
655 unlock_object(c_rarg1);
656 pop(state);
657
658 if (install_monitor_exception) {
659 call_VM(noreg, CAST_FROM_FN_PTR(address,
660 InterpreterRuntime::
661 new_illegal_monitor_state_exception));
662 }
663
664 b(restart);
665 }
666
667 bind(loop);
668 // check if current entry is used
669 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
670 cbnz(rscratch1, exception);
671
672 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
673 bind(entry);
674 cmp(c_rarg1, r19); // check if bottom reached
675 br(Assembler::NE, loop); // if not at bottom then check this entry
676 }
677
678 bind(no_unlock);
679
680 // jvmti support
681 if (notify_jvmdi) {
682 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
683 } else {
684 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
685 }
686
687 // remove activation
688 // get sender esp
689 ldr(rscratch2,
690 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
691 if (StackReservedPages > 0) {
692 // testing if reserved zone needs to be re-enabled
693 Label no_reserved_zone_enabling;
694
695 // check if already enabled - if so no re-enabling needed
696 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
697 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
698 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
699 br(Assembler::EQ, no_reserved_zone_enabling);
700
701 // look for an overflow into the stack reserved zone, i.e.
702 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
703 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
704 cmp(rscratch2, rscratch1);
705 br(Assembler::LS, no_reserved_zone_enabling);
706
707 call_VM_leaf(
708 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
709 call_VM(noreg, CAST_FROM_FN_PTR(address,
710 InterpreterRuntime::throw_delayed_StackOverflowError));
711 should_not_reach_here();
712
713 bind(no_reserved_zone_enabling);
714 }
715
716 // restore sender esp
717 mov(esp, rscratch2);
718 // remove frame anchor
719 leave();
720 // If we're returning to interpreted code we will shortly be
721 // adjusting SP to allow some space for ESP. If we're returning to
722 // compiled code the saved sender SP was saved in sender_sp, so this
723 // restores it.
724 andr(sp, esp, -16);
725 }
726
727 // Lock object
728 //
729 // Args:
730 // c_rarg1: BasicObjectLock to be used for locking
731 //
732 // Kills:
733 // r0
734 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
735 // rscratch1, rscratch2 (scratch regs)
736 void InterpreterMacroAssembler::lock_object(Register lock_reg)
737 {
738 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
739 if (LockingMode == LM_MONITOR) {
740 call_VM(noreg,
741 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
742 lock_reg);
743 } else {
744 Label count, done;
745
746 const Register swap_reg = r0;
747 const Register tmp = c_rarg2;
748 const Register obj_reg = c_rarg3; // Will contain the oop
749 const Register tmp2 = c_rarg4;
750 const Register tmp3 = c_rarg5;
751
752 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
753 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
754 const int mark_offset = lock_offset +
755 BasicLock::displaced_header_offset_in_bytes();
756
757 Label slow_case;
758
759 // Load object pointer into obj_reg %c_rarg3
760 ldr(obj_reg, Address(lock_reg, obj_offset));
761
762 if (DiagnoseSyncOnValueBasedClasses != 0) {
763 load_klass(tmp, obj_reg);
764 ldrw(tmp, Address(tmp, Klass::access_flags_offset()));
765 tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
766 br(Assembler::NE, slow_case);
767 }
768
769 if (LockingMode == LM_LIGHTWEIGHT) {
770 lightweight_lock(obj_reg, tmp, tmp2, tmp3, slow_case);
771 b(count);
772 } else if (LockingMode == LM_LEGACY) {
773 // Load (object->mark() | 1) into swap_reg
774 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
775 orr(swap_reg, rscratch1, 1);
776
777 // Save (object->mark() | 1) into BasicLock's displaced header
778 str(swap_reg, Address(lock_reg, mark_offset));
779
780 assert(lock_offset == 0,
781 "displached header must be first word in BasicObjectLock");
782
783 Label fail;
784 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
785
786 // Fast check for recursive lock.
787 //
788 // Can apply the optimization only if this is a stack lock
789 // allocated in this thread. For efficiency, we can focus on
790 // recently allocated stack locks (instead of reading the stack
791 // base and checking whether 'mark' points inside the current
792 // thread stack):
793 // 1) (mark & 7) == 0, and
794 // 2) sp <= mark < mark + os::pagesize()
795 //
796 // Warning: sp + os::pagesize can overflow the stack base. We must
797 // neither apply the optimization for an inflated lock allocated
798 // just above the thread stack (this is why condition 1 matters)
799 // nor apply the optimization if the stack lock is inside the stack
800 // of another thread. The latter is avoided even in case of overflow
801 // because we have guard pages at the end of all stacks. Hence, if
802 // we go over the stack base and hit the stack of another thread,
803 // this should not be in a writeable area that could contain a
804 // stack lock allocated by that thread. As a consequence, a stack
805 // lock less than page size away from sp is guaranteed to be
806 // owned by the current thread.
807 //
808 // These 3 tests can be done by evaluating the following
809 // expression: ((mark - sp) & (7 - os::vm_page_size())),
810 // assuming both stack pointer and pagesize have their
811 // least significant 3 bits clear.
812 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
813 // NOTE2: aarch64 does not like to subtract sp from rn so take a
814 // copy
815 mov(rscratch1, sp);
816 sub(swap_reg, swap_reg, rscratch1);
817 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
818
819 // Save the test result, for recursive case, the result is zero
820 str(swap_reg, Address(lock_reg, mark_offset));
821 br(Assembler::EQ, count);
822 }
823 bind(slow_case);
824
825 // Call the runtime routine for slow case
826 if (LockingMode == LM_LIGHTWEIGHT) {
827 call_VM(noreg,
828 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
829 obj_reg);
830 } else {
831 call_VM(noreg,
832 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
833 lock_reg);
834 }
835 b(done);
836
837 bind(count);
838 increment(Address(rthread, JavaThread::held_monitor_count_offset()));
839
840 bind(done);
841 }
842 }
843
844
845 // Unlocks an object. Used in monitorexit bytecode and
846 // remove_activation. Throws an IllegalMonitorException if object is
847 // not locked by current thread.
848 //
849 // Args:
850 // c_rarg1: BasicObjectLock for lock
851 //
852 // Kills:
853 // r0
854 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
855 // rscratch1, rscratch2 (scratch regs)
856 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
857 {
858 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
859
860 if (LockingMode == LM_MONITOR) {
861 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
862 } else {
863 Label count, done;
864
865 const Register swap_reg = r0;
866 const Register header_reg = c_rarg2; // Will contain the old oopMark
867 const Register obj_reg = c_rarg3; // Will contain the oop
868 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
869
870 save_bcp(); // Save in case of exception
871
872 if (LockingMode != LM_LIGHTWEIGHT) {
873 // Convert from BasicObjectLock structure to object and BasicLock
874 // structure Store the BasicLock address into %r0
875 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
876 }
877
878 // Load oop into obj_reg(%c_rarg3)
879 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
880
881 // Free entry
882 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
883
884 if (LockingMode == LM_LIGHTWEIGHT) {
885 Label slow_case;
886 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
887 b(count);
888 bind(slow_case);
889 } else if (LockingMode == LM_LEGACY) {
890 // Load the old header from BasicLock structure
891 ldr(header_reg, Address(swap_reg,
892 BasicLock::displaced_header_offset_in_bytes()));
893
894 // Test for recursion
895 cbz(header_reg, count);
896
897 // Atomic swap back the old header
898 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
899 }
900 // Call the runtime routine for slow case.
901 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
902 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
903 b(done);
904
905 bind(count);
906 decrement(Address(rthread, JavaThread::held_monitor_count_offset()));
907
908 bind(done);
909 restore_bcp();
910 }
911 }
912
913 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
914 Label& zero_continue) {
915 assert(ProfileInterpreter, "must be profiling interpreter");
916 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
917 cbz(mdp, zero_continue);
918 }
919
920 // Set the method data pointer for the current bcp.
921 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
922 assert(ProfileInterpreter, "must be profiling interpreter");
923 Label set_mdp;
924 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
925
926 // Test MDO to avoid the call if it is null.
927 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
928 cbz(r0, set_mdp);
929 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
930 // r0: mdi
931 // mdo is guaranteed to be non-zero here, we checked for it before the call.
932 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
933 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
934 add(r0, r1, r0);
935 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
936 bind(set_mdp);
937 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
938 }
939
940 void InterpreterMacroAssembler::verify_method_data_pointer() {
941 assert(ProfileInterpreter, "must be profiling interpreter");
942 #ifdef ASSERT
943 Label verify_continue;
944 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
945 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
946 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
947 get_method(r1);
948
949 // If the mdp is valid, it will point to a DataLayout header which is
950 // consistent with the bcp. The converse is highly probable also.
951 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
952 ldr(rscratch1, Address(r1, Method::const_offset()));
953 add(r2, r2, rscratch1, Assembler::LSL);
954 lea(r2, Address(r2, ConstMethod::codes_offset()));
955 cmp(r2, rbcp);
956 br(Assembler::EQ, verify_continue);
957 // r1: method
958 // rbcp: bcp // rbcp == 22
959 // r3: mdp
960 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
961 r1, rbcp, r3);
962 bind(verify_continue);
963 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
964 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
965 #endif // ASSERT
966 }
967
968
969 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
970 int constant,
971 Register value) {
972 assert(ProfileInterpreter, "must be profiling interpreter");
973 Address data(mdp_in, constant);
974 str(value, data);
975 }
976
977
978 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
979 int constant,
980 bool decrement) {
981 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
982 }
983
984 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
985 Register reg,
986 int constant,
987 bool decrement) {
988 assert(ProfileInterpreter, "must be profiling interpreter");
989 // %%% this does 64bit counters at best it is wasting space
990 // at worst it is a rare bug when counters overflow
991
992 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
993
994 Address addr1(mdp_in, constant);
995 Address addr2(rscratch2, reg, Address::lsl(0));
996 Address &addr = addr1;
997 if (reg != noreg) {
998 lea(rscratch2, addr1);
999 addr = addr2;
1000 }
1001
1002 if (decrement) {
1003 // Decrement the register. Set condition codes.
1004 // Intel does this
1005 // addptr(data, (int32_t) -DataLayout::counter_increment);
1006 // If the decrement causes the counter to overflow, stay negative
1007 // Label L;
1008 // jcc(Assembler::negative, L);
1009 // addptr(data, (int32_t) DataLayout::counter_increment);
1010 // so we do this
1011 ldr(rscratch1, addr);
1012 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1013 Label L;
1014 br(Assembler::LO, L); // skip store if counter underflow
1015 str(rscratch1, addr);
1016 bind(L);
1017 } else {
1018 assert(DataLayout::counter_increment == 1,
1019 "flow-free idiom only works with 1");
1020 // Intel does this
1021 // Increment the register. Set carry flag.
1022 // addptr(data, DataLayout::counter_increment);
1023 // If the increment causes the counter to overflow, pull back by 1.
1024 // sbbptr(data, (int32_t)0);
1025 // so we do this
1026 ldr(rscratch1, addr);
1027 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1028 Label L;
1029 br(Assembler::CS, L); // skip store if counter overflow
1030 str(rscratch1, addr);
1031 bind(L);
1032 }
1033 }
1034
1035 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1036 int flag_byte_constant) {
1037 assert(ProfileInterpreter, "must be profiling interpreter");
1038 int flags_offset = in_bytes(DataLayout::flags_offset());
1039 // Set the flag
1040 ldrb(rscratch1, Address(mdp_in, flags_offset));
1041 orr(rscratch1, rscratch1, flag_byte_constant);
1042 strb(rscratch1, Address(mdp_in, flags_offset));
1043 }
1044
1045
1046 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1047 int offset,
1048 Register value,
1049 Register test_value_out,
1050 Label& not_equal_continue) {
1051 assert(ProfileInterpreter, "must be profiling interpreter");
1052 if (test_value_out == noreg) {
1053 ldr(rscratch1, Address(mdp_in, offset));
1054 cmp(value, rscratch1);
1055 } else {
1056 // Put the test value into a register, so caller can use it:
1057 ldr(test_value_out, Address(mdp_in, offset));
1058 cmp(value, test_value_out);
1059 }
1060 br(Assembler::NE, not_equal_continue);
1061 }
1062
1063
1064 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1065 int offset_of_disp) {
1066 assert(ProfileInterpreter, "must be profiling interpreter");
1067 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1068 add(mdp_in, mdp_in, rscratch1, LSL);
1069 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1070 }
1071
1072
1073 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1074 Register reg,
1075 int offset_of_disp) {
1076 assert(ProfileInterpreter, "must be profiling interpreter");
1077 lea(rscratch1, Address(mdp_in, offset_of_disp));
1078 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1079 add(mdp_in, mdp_in, rscratch1, LSL);
1080 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1081 }
1082
1083
1084 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1085 int constant) {
1086 assert(ProfileInterpreter, "must be profiling interpreter");
1087 add(mdp_in, mdp_in, (unsigned)constant);
1088 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1089 }
1090
1091
1092 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1093 assert(ProfileInterpreter, "must be profiling interpreter");
1094 // save/restore across call_VM
1095 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1096 call_VM(noreg,
1097 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1098 return_bci);
1099 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1100 }
1101
1102
1103 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1104 Register bumped_count) {
1105 if (ProfileInterpreter) {
1106 Label profile_continue;
1107
1108 // If no method data exists, go to profile_continue.
1109 // Otherwise, assign to mdp
1110 test_method_data_pointer(mdp, profile_continue);
1111
1112 // We are taking a branch. Increment the taken count.
1113 // We inline increment_mdp_data_at to return bumped_count in a register
1114 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1115 Address data(mdp, in_bytes(JumpData::taken_offset()));
1116 ldr(bumped_count, data);
1117 assert(DataLayout::counter_increment == 1,
1118 "flow-free idiom only works with 1");
1119 // Intel does this to catch overflow
1120 // addptr(bumped_count, DataLayout::counter_increment);
1121 // sbbptr(bumped_count, 0);
1122 // so we do this
1123 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1124 Label L;
1125 br(Assembler::CS, L); // skip store if counter overflow
1126 str(bumped_count, data);
1127 bind(L);
1128 // The method data pointer needs to be updated to reflect the new target.
1129 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1130 bind(profile_continue);
1131 }
1132 }
1133
1134
1135 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1136 if (ProfileInterpreter) {
1137 Label profile_continue;
1138
1139 // If no method data exists, go to profile_continue.
1140 test_method_data_pointer(mdp, profile_continue);
1141
1142 // We are taking a branch. Increment the not taken count.
1143 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1144
1145 // The method data pointer needs to be updated to correspond to
1146 // the next bytecode
1147 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1148 bind(profile_continue);
1149 }
1150 }
1151
1152
1153 void InterpreterMacroAssembler::profile_call(Register mdp) {
1154 if (ProfileInterpreter) {
1155 Label profile_continue;
1156
1157 // If no method data exists, go to profile_continue.
1158 test_method_data_pointer(mdp, profile_continue);
1159
1160 // We are making a call. Increment the count.
1161 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1162
1163 // The method data pointer needs to be updated to reflect the new target.
1164 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1165 bind(profile_continue);
1166 }
1167 }
1168
1169 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1170 if (ProfileInterpreter) {
1171 Label profile_continue;
1172
1173 // If no method data exists, go to profile_continue.
1174 test_method_data_pointer(mdp, profile_continue);
1175
1176 // We are making a call. Increment the count.
1177 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1178
1179 // The method data pointer needs to be updated to reflect the new target.
1180 update_mdp_by_constant(mdp,
1181 in_bytes(VirtualCallData::
1182 virtual_call_data_size()));
1183 bind(profile_continue);
1184 }
1185 }
1186
1187
1188 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1189 Register mdp,
1190 Register reg2,
1191 bool receiver_can_be_null) {
1192 if (ProfileInterpreter) {
1193 Label profile_continue;
1194
1195 // If no method data exists, go to profile_continue.
1196 test_method_data_pointer(mdp, profile_continue);
1197
1198 Label skip_receiver_profile;
1199 if (receiver_can_be_null) {
1200 Label not_null;
1201 // We are making a call. Increment the count for null receiver.
1202 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1203 b(skip_receiver_profile);
1204 bind(not_null);
1205 }
1206
1207 // Record the receiver type.
1208 record_klass_in_profile(receiver, mdp, reg2, true);
1209 bind(skip_receiver_profile);
1210
1211 // The method data pointer needs to be updated to reflect the new target.
1212 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1213 bind(profile_continue);
1214 }
1215 }
1216
1217 // This routine creates a state machine for updating the multi-row
1218 // type profile at a virtual call site (or other type-sensitive bytecode).
1219 // The machine visits each row (of receiver/count) until the receiver type
1220 // is found, or until it runs out of rows. At the same time, it remembers
1221 // the location of the first empty row. (An empty row records null for its
1222 // receiver, and can be allocated for a newly-observed receiver type.)
1223 // Because there are two degrees of freedom in the state, a simple linear
1224 // search will not work; it must be a decision tree. Hence this helper
1225 // function is recursive, to generate the required tree structured code.
1226 // It's the interpreter, so we are trading off code space for speed.
1227 // See below for example code.
1228 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1229 Register receiver, Register mdp,
1230 Register reg2, int start_row,
1231 Label& done, bool is_virtual_call) {
1232 if (TypeProfileWidth == 0) {
1233 if (is_virtual_call) {
1234 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1235 }
1236 #if INCLUDE_JVMCI
1237 else if (EnableJVMCI) {
1238 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1239 }
1240 #endif // INCLUDE_JVMCI
1241 } else {
1242 int non_profiled_offset = -1;
1243 if (is_virtual_call) {
1244 non_profiled_offset = in_bytes(CounterData::count_offset());
1245 }
1246 #if INCLUDE_JVMCI
1247 else if (EnableJVMCI) {
1248 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1249 }
1250 #endif // INCLUDE_JVMCI
1251
1252 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1253 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1254 }
1255 }
1256
1257 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1258 Register reg2, int start_row, Label& done, int total_rows,
1259 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1260 int non_profiled_offset) {
1261 int last_row = total_rows - 1;
1262 assert(start_row <= last_row, "must be work left to do");
1263 // Test this row for both the item and for null.
1264 // Take any of three different outcomes:
1265 // 1. found item => increment count and goto done
1266 // 2. found null => keep looking for case 1, maybe allocate this cell
1267 // 3. found something else => keep looking for cases 1 and 2
1268 // Case 3 is handled by a recursive call.
1269 for (int row = start_row; row <= last_row; row++) {
1270 Label next_test;
1271 bool test_for_null_also = (row == start_row);
1272
1273 // See if the item is item[n].
1274 int item_offset = in_bytes(item_offset_fn(row));
1275 test_mdp_data_at(mdp, item_offset, item,
1276 (test_for_null_also ? reg2 : noreg),
1277 next_test);
1278 // (Reg2 now contains the item from the CallData.)
1279
1280 // The item is item[n]. Increment count[n].
1281 int count_offset = in_bytes(item_count_offset_fn(row));
1282 increment_mdp_data_at(mdp, count_offset);
1283 b(done);
1284 bind(next_test);
1285
1286 if (test_for_null_also) {
1287 Label found_null;
1288 // Failed the equality check on item[n]... Test for null.
1289 if (start_row == last_row) {
1290 // The only thing left to do is handle the null case.
1291 if (non_profiled_offset >= 0) {
1292 cbz(reg2, found_null);
1293 // Item did not match any saved item and there is no empty row for it.
1294 // Increment total counter to indicate polymorphic case.
1295 increment_mdp_data_at(mdp, non_profiled_offset);
1296 b(done);
1297 bind(found_null);
1298 } else {
1299 cbnz(reg2, done);
1300 }
1301 break;
1302 }
1303 // Since null is rare, make it be the branch-taken case.
1304 cbz(reg2, found_null);
1305
1306 // Put all the "Case 3" tests here.
1307 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1308 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1309
1310 // Found a null. Keep searching for a matching item,
1311 // but remember that this is an empty (unused) slot.
1312 bind(found_null);
1313 }
1314 }
1315
1316 // In the fall-through case, we found no matching item, but we
1317 // observed the item[start_row] is null.
1318
1319 // Fill in the item field and increment the count.
1320 int item_offset = in_bytes(item_offset_fn(start_row));
1321 set_mdp_data_at(mdp, item_offset, item);
1322 int count_offset = in_bytes(item_count_offset_fn(start_row));
1323 mov(reg2, DataLayout::counter_increment);
1324 set_mdp_data_at(mdp, count_offset, reg2);
1325 if (start_row > 0) {
1326 b(done);
1327 }
1328 }
1329
1330 // Example state machine code for three profile rows:
1331 // // main copy of decision tree, rooted at row[1]
1332 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1333 // if (row[0].rec != nullptr) {
1334 // // inner copy of decision tree, rooted at row[1]
1335 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1336 // if (row[1].rec != nullptr) {
1337 // // degenerate decision tree, rooted at row[2]
1338 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1339 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1340 // row[2].init(rec); goto done;
1341 // } else {
1342 // // remember row[1] is empty
1343 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1344 // row[1].init(rec); goto done;
1345 // }
1346 // } else {
1347 // // remember row[0] is empty
1348 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1349 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1350 // row[0].init(rec); goto done;
1351 // }
1352 // done:
1353
1354 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1355 Register mdp, Register reg2,
1356 bool is_virtual_call) {
1357 assert(ProfileInterpreter, "must be profiling");
1358 Label done;
1359
1360 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1361
1362 bind (done);
1363 }
1364
1365 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1366 Register mdp) {
1367 if (ProfileInterpreter) {
1368 Label profile_continue;
1369 uint row;
1370
1371 // If no method data exists, go to profile_continue.
1372 test_method_data_pointer(mdp, profile_continue);
1373
1374 // Update the total ret count.
1375 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1376
1377 for (row = 0; row < RetData::row_limit(); row++) {
1378 Label next_test;
1379
1380 // See if return_bci is equal to bci[n]:
1381 test_mdp_data_at(mdp,
1382 in_bytes(RetData::bci_offset(row)),
1383 return_bci, noreg,
1384 next_test);
1385
1386 // return_bci is equal to bci[n]. Increment the count.
1387 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1388
1389 // The method data pointer needs to be updated to reflect the new target.
1390 update_mdp_by_offset(mdp,
1391 in_bytes(RetData::bci_displacement_offset(row)));
1392 b(profile_continue);
1393 bind(next_test);
1394 }
1395
1396 update_mdp_for_ret(return_bci);
1397
1398 bind(profile_continue);
1399 }
1400 }
1401
1402 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1403 if (ProfileInterpreter) {
1404 Label profile_continue;
1405
1406 // If no method data exists, go to profile_continue.
1407 test_method_data_pointer(mdp, profile_continue);
1408
1409 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1410
1411 // The method data pointer needs to be updated.
1412 int mdp_delta = in_bytes(BitData::bit_data_size());
1413 if (TypeProfileCasts) {
1414 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1415 }
1416 update_mdp_by_constant(mdp, mdp_delta);
1417
1418 bind(profile_continue);
1419 }
1420 }
1421
1422 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1423 if (ProfileInterpreter && TypeProfileCasts) {
1424 Label profile_continue;
1425
1426 // If no method data exists, go to profile_continue.
1427 test_method_data_pointer(mdp, profile_continue);
1428
1429 int count_offset = in_bytes(CounterData::count_offset());
1430 // Back up the address, since we have already bumped the mdp.
1431 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1432
1433 // *Decrement* the counter. We expect to see zero or small negatives.
1434 increment_mdp_data_at(mdp, count_offset, true);
1435
1436 bind (profile_continue);
1437 }
1438 }
1439
1440 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1441 if (ProfileInterpreter) {
1442 Label profile_continue;
1443
1444 // If no method data exists, go to profile_continue.
1445 test_method_data_pointer(mdp, profile_continue);
1446
1447 // The method data pointer needs to be updated.
1448 int mdp_delta = in_bytes(BitData::bit_data_size());
1449 if (TypeProfileCasts) {
1450 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1451
1452 // Record the object type.
1453 record_klass_in_profile(klass, mdp, reg2, false);
1454 }
1455 update_mdp_by_constant(mdp, mdp_delta);
1456
1457 bind(profile_continue);
1458 }
1459 }
1460
1461 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1462 if (ProfileInterpreter) {
1463 Label profile_continue;
1464
1465 // If no method data exists, go to profile_continue.
1466 test_method_data_pointer(mdp, profile_continue);
1467
1468 // Update the default case count
1469 increment_mdp_data_at(mdp,
1470 in_bytes(MultiBranchData::default_count_offset()));
1471
1472 // The method data pointer needs to be updated.
1473 update_mdp_by_offset(mdp,
1474 in_bytes(MultiBranchData::
1475 default_displacement_offset()));
1476
1477 bind(profile_continue);
1478 }
1479 }
1480
1481 void InterpreterMacroAssembler::profile_switch_case(Register index,
1482 Register mdp,
1483 Register reg2) {
1484 if (ProfileInterpreter) {
1485 Label profile_continue;
1486
1487 // If no method data exists, go to profile_continue.
1488 test_method_data_pointer(mdp, profile_continue);
1489
1490 // Build the base (index * per_case_size_in_bytes()) +
1491 // case_array_offset_in_bytes()
1492 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1493 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1494 Assembler::maddw(index, index, reg2, rscratch1);
1495
1496 // Update the case count
1497 increment_mdp_data_at(mdp,
1498 index,
1499 in_bytes(MultiBranchData::relative_count_offset()));
1500
1501 // The method data pointer needs to be updated.
1502 update_mdp_by_offset(mdp,
1503 index,
1504 in_bytes(MultiBranchData::
1505 relative_displacement_offset()));
1506
1507 bind(profile_continue);
1508 }
1509 }
1510
1511 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1512 if (state == atos) {
1513 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1514 }
1515 }
1516
1517 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1518
1519
1520 void InterpreterMacroAssembler::notify_method_entry() {
1521 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1522 // track stack depth. If it is possible to enter interp_only_mode we add
1523 // the code to check if the event should be sent.
1524 if (JvmtiExport::can_post_interpreter_events()) {
1525 Label L;
1526 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1527 cbzw(r3, L);
1528 call_VM(noreg, CAST_FROM_FN_PTR(address,
1529 InterpreterRuntime::post_method_entry));
1530 bind(L);
1531 }
1532
1533 {
1534 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1535 get_method(c_rarg1);
1536 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1537 rthread, c_rarg1);
1538 }
1539
1540 // RedefineClasses() tracing support for obsolete method entry
1541 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1542 get_method(c_rarg1);
1543 call_VM_leaf(
1544 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1545 rthread, c_rarg1);
1546 }
1547
1548 }
1549
1550
1551 void InterpreterMacroAssembler::notify_method_exit(
1552 TosState state, NotifyMethodExitMode mode) {
1553 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1554 // track stack depth. If it is possible to enter interp_only_mode we add
1555 // the code to check if the event should be sent.
1556 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1557 Label L;
1558 // Note: frame::interpreter_frame_result has a dependency on how the
1559 // method result is saved across the call to post_method_exit. If this
1560 // is changed then the interpreter_frame_result implementation will
1561 // need to be updated too.
1562
1563 // template interpreter will leave the result on the top of the stack.
1564 push(state);
1565 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1566 cbz(r3, L);
1567 call_VM(noreg,
1568 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1569 bind(L);
1570 pop(state);
1571 }
1572
1573 {
1574 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1575 push(state);
1576 get_method(c_rarg1);
1577 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1578 rthread, c_rarg1);
1579 pop(state);
1580 }
1581 }
1582
1583
1584 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1585 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1586 int increment, Address mask,
1587 Register scratch, Register scratch2,
1588 bool preloaded, Condition cond,
1589 Label* where) {
1590 if (!preloaded) {
1591 ldrw(scratch, counter_addr);
1592 }
1593 add(scratch, scratch, increment);
1594 strw(scratch, counter_addr);
1595 ldrw(scratch2, mask);
1596 ands(scratch, scratch, scratch2);
1597 br(cond, *where);
1598 }
1599
1600 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1601 int number_of_arguments) {
1602 // interpreter specific
1603 //
1604 // Note: No need to save/restore rbcp & rlocals pointer since these
1605 // are callee saved registers and no blocking/ GC can happen
1606 // in leaf calls.
1607 #ifdef ASSERT
1608 {
1609 Label L;
1610 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1611 cbz(rscratch1, L);
1612 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1613 " last_sp != nullptr");
1614 bind(L);
1615 }
1616 #endif /* ASSERT */
1617 // super call
1618 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1619 }
1620
1621 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1622 Register java_thread,
1623 Register last_java_sp,
1624 address entry_point,
1625 int number_of_arguments,
1626 bool check_exceptions) {
1627 // interpreter specific
1628 //
1629 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1630 // really make a difference for these runtime calls, since they are
1631 // slow anyway. Btw., bcp must be saved/restored since it may change
1632 // due to GC.
1633 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1634 save_bcp();
1635 #ifdef ASSERT
1636 {
1637 Label L;
1638 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1639 cbz(rscratch1, L);
1640 stop("InterpreterMacroAssembler::call_VM_base:"
1641 " last_sp != nullptr");
1642 bind(L);
1643 }
1644 #endif /* ASSERT */
1645 // super call
1646 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1647 entry_point, number_of_arguments,
1648 check_exceptions);
1649 // interpreter specific
1650 restore_bcp();
1651 restore_locals();
1652 }
1653
1654 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1655 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1656 Label update, next, none;
1657
1658 verify_oop(obj);
1659
1660 cbnz(obj, update);
1661 orptr(mdo_addr, TypeEntries::null_seen);
1662 b(next);
1663
1664 bind(update);
1665 load_klass(obj, obj);
1666
1667 ldr(rscratch1, mdo_addr);
1668 eor(obj, obj, rscratch1);
1669 tst(obj, TypeEntries::type_klass_mask);
1670 br(Assembler::EQ, next); // klass seen before, nothing to
1671 // do. The unknown bit may have been
1672 // set already but no need to check.
1673
1674 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1675 // already unknown. Nothing to do anymore.
1676
1677 cbz(rscratch1, none);
1678 cmp(rscratch1, (u1)TypeEntries::null_seen);
1679 br(Assembler::EQ, none);
1680 // There is a chance that the checks above
1681 // fail if another thread has just set the
1682 // profiling to this obj's klass
1683 eor(obj, obj, rscratch1); // get back original value before XOR
1684 ldr(rscratch1, mdo_addr);
1685 eor(obj, obj, rscratch1);
1686 tst(obj, TypeEntries::type_klass_mask);
1687 br(Assembler::EQ, next);
1688
1689 // different than before. Cannot keep accurate profile.
1690 orptr(mdo_addr, TypeEntries::type_unknown);
1691 b(next);
1692
1693 bind(none);
1694 // first time here. Set profile type.
1695 str(obj, mdo_addr);
1696 #ifdef ASSERT
1697 andr(obj, obj, TypeEntries::type_mask);
1698 verify_klass_ptr(obj);
1699 #endif
1700
1701 bind(next);
1702 }
1703
1704 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1705 if (!ProfileInterpreter) {
1706 return;
1707 }
1708
1709 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1710 Label profile_continue;
1711
1712 test_method_data_pointer(mdp, profile_continue);
1713
1714 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1715
1716 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1717 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1718 br(Assembler::NE, profile_continue);
1719
1720 if (MethodData::profile_arguments()) {
1721 Label done;
1722 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1723
1724 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1725 if (i > 0 || MethodData::profile_return()) {
1726 // If return value type is profiled we may have no argument to profile
1727 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1728 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1729 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1730 add(rscratch1, mdp, off_to_args);
1731 br(Assembler::LT, done);
1732 }
1733 ldr(tmp, Address(callee, Method::const_offset()));
1734 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1735 // stack offset o (zero based) from the start of the argument
1736 // list, for n arguments translates into offset n - o - 1 from
1737 // the end of the argument list
1738 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1739 sub(tmp, tmp, rscratch1);
1740 sub(tmp, tmp, 1);
1741 Address arg_addr = argument_address(tmp);
1742 ldr(tmp, arg_addr);
1743
1744 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1745 profile_obj_type(tmp, mdo_arg_addr);
1746
1747 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1748 off_to_args += to_add;
1749 }
1750
1751 if (MethodData::profile_return()) {
1752 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1753 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1754 }
1755
1756 add(rscratch1, mdp, off_to_args);
1757 bind(done);
1758 mov(mdp, rscratch1);
1759
1760 if (MethodData::profile_return()) {
1761 // We're right after the type profile for the last
1762 // argument. tmp is the number of cells left in the
1763 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1764 // if there's a return to profile.
1765 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1766 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1767 }
1768 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1769 } else {
1770 assert(MethodData::profile_return(), "either profile call args or call ret");
1771 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1772 }
1773
1774 // mdp points right after the end of the
1775 // CallTypeData/VirtualCallTypeData, right after the cells for the
1776 // return value type if there's one
1777
1778 bind(profile_continue);
1779 }
1780 }
1781
1782 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1783 assert_different_registers(mdp, ret, tmp, rbcp);
1784 if (ProfileInterpreter && MethodData::profile_return()) {
1785 Label profile_continue, done;
1786
1787 test_method_data_pointer(mdp, profile_continue);
1788
1789 if (MethodData::profile_return_jsr292_only()) {
1790 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1791
1792 // If we don't profile all invoke bytecodes we must make sure
1793 // it's a bytecode we indeed profile. We can't go back to the
1794 // beginning of the ProfileData we intend to update to check its
1795 // type because we're right after it and we don't known its
1796 // length
1797 Label do_profile;
1798 ldrb(rscratch1, Address(rbcp, 0));
1799 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1800 br(Assembler::EQ, do_profile);
1801 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1802 br(Assembler::EQ, do_profile);
1803 get_method(tmp);
1804 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1805 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1806 br(Assembler::NE, profile_continue);
1807
1808 bind(do_profile);
1809 }
1810
1811 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1812 mov(tmp, ret);
1813 profile_obj_type(tmp, mdo_ret_addr);
1814
1815 bind(profile_continue);
1816 }
1817 }
1818
1819 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1820 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1821 if (ProfileInterpreter && MethodData::profile_parameters()) {
1822 Label profile_continue, done;
1823
1824 test_method_data_pointer(mdp, profile_continue);
1825
1826 // Load the offset of the area within the MDO used for
1827 // parameters. If it's negative we're not profiling any parameters
1828 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1829 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1830
1831 // Compute a pointer to the area for parameters from the offset
1832 // and move the pointer to the slot for the last
1833 // parameters. Collect profiling from last parameter down.
1834 // mdo start + parameters offset + array length - 1
1835 add(mdp, mdp, tmp1);
1836 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1837 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1838
1839 Label loop;
1840 bind(loop);
1841
1842 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1843 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1844 int per_arg_scale = exact_log2(DataLayout::cell_size);
1845 add(rscratch1, mdp, off_base);
1846 add(rscratch2, mdp, type_base);
1847
1848 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1849 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1850
1851 // load offset on the stack from the slot for this parameter
1852 ldr(tmp2, arg_off);
1853 neg(tmp2, tmp2);
1854 // read the parameter from the local area
1855 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1856
1857 // profile the parameter
1858 profile_obj_type(tmp2, arg_type);
1859
1860 // go to next parameter
1861 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1862 br(Assembler::GE, loop);
1863
1864 bind(profile_continue);
1865 }
1866 }
1867
1868 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1869 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1870 get_cache_index_at_bcp(index, 1, sizeof(u4));
1871 // Get address of invokedynamic array
1872 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1873 // Scale the index to be the entry index * sizeof(ResolvedInvokeDynamicInfo)
1874 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1875 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1876 lea(cache, Address(cache, index));
1877 }