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/biasedLocking.hpp"
43 #include "runtime/frame.inline.hpp"
44 #include "runtime/safepointMechanism.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/thread.inline.hpp"
47 #include "utilities/powerOfTwo.hpp"
48
49 void InterpreterMacroAssembler::narrow(Register result) {
50
51 // Get method->_constMethod->_result_type
52 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
53 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
54 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
55
56 Label done, notBool, notByte, notChar;
57
58 // common case first
59 cmpw(rscratch1, T_INT);
60 br(Assembler::EQ, done);
61
62 // mask integer result to narrower return type.
63 cmpw(rscratch1, T_BOOLEAN);
64 br(Assembler::NE, notBool);
65 andw(result, result, 0x1);
66 b(done);
67
68 bind(notBool);
69 cmpw(rscratch1, T_BYTE);
70 br(Assembler::NE, notByte);
71 sbfx(result, result, 0, 8);
72 b(done);
73
74 bind(notByte);
75 cmpw(rscratch1, T_CHAR);
76 br(Assembler::NE, notChar);
77 ubfx(result, result, 0, 16); // truncate upper 16 bits
78 b(done);
79
80 bind(notChar);
81 sbfx(result, result, 0, 16); // sign-extend short
82
83 // Nothing to do for T_INT
84 bind(done);
85 }
86
87 void InterpreterMacroAssembler::jump_to_entry(address entry) {
88 assert(entry, "Entry must have been generated by now");
89 b(entry);
90 }
91
92 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
93 if (JvmtiExport::can_pop_frame()) {
94 Label L;
95 // Initiate popframe handling only if it is not already being
96 // processed. If the flag has the popframe_processing bit set, it
97 // means that this code is called *during* popframe handling - we
98 // don't want to reenter.
99 // This method is only called just after the call into the vm in
100 // call_VM_base, so the arg registers are available.
101 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
102 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
103 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
104 // Call Interpreter::remove_activation_preserving_args_entry() to get the
105 // address of the same-named entrypoint in the generated interpreter code.
106 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
107 br(r0);
108 bind(L);
109 }
110 }
111
112
113 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
114 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
115 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
116 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
117 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
118 switch (state) {
119 case atos: ldr(r0, oop_addr);
120 str(zr, oop_addr);
121 verify_oop(r0, state); break;
122 case ltos: ldr(r0, val_addr); break;
123 case btos: // fall through
124 case ztos: // fall through
125 case ctos: // fall through
126 case stos: // fall through
127 case itos: ldrw(r0, val_addr); break;
128 case ftos: ldrs(v0, val_addr); break;
129 case dtos: ldrd(v0, val_addr); break;
130 case vtos: /* nothing to do */ break;
131 default : ShouldNotReachHere();
132 }
133 // Clean up tos value in the thread object
134 movw(rscratch1, (int) ilgl);
135 strw(rscratch1, tos_addr);
136 strw(zr, val_addr);
137 }
138
139
140 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
141 if (JvmtiExport::can_force_early_return()) {
142 Label L;
143 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
144 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit;
145
146 // Initiate earlyret handling only if it is not already being processed.
147 // If the flag has the earlyret_processing bit set, it means that this code
148 // is called *during* earlyret handling - we don't want to reenter.
149 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
150 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
151 br(Assembler::NE, L);
152
153 // Call Interpreter::remove_activation_early_entry() to get the address of the
154 // same-named entrypoint in the generated interpreter code.
155 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
156 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
157 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
158 br(r0);
159 bind(L);
160 }
161 }
162
163 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
164 Register reg,
165 int bcp_offset) {
166 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
167 ldrh(reg, Address(rbcp, bcp_offset));
168 rev16(reg, reg);
169 }
170
171 void InterpreterMacroAssembler::get_dispatch() {
172 uint64_t offset;
173 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
174 lea(rdispatch, Address(rdispatch, offset));
175 }
176
177 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
178 int bcp_offset,
179 size_t index_size) {
180 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
181 if (index_size == sizeof(u2)) {
182 load_unsigned_short(index, Address(rbcp, bcp_offset));
183 } else if (index_size == sizeof(u4)) {
184 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
185 ldrw(index, Address(rbcp, bcp_offset));
186 // Check if the secondary index definition is still ~x, otherwise
187 // we have to change the following assembler code to calculate the
188 // plain index.
189 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
190 eonw(index, index, zr); // convert to plain index
191 } else if (index_size == sizeof(u1)) {
192 load_unsigned_byte(index, Address(rbcp, bcp_offset));
193 } else {
194 ShouldNotReachHere();
195 }
196 }
197
198 // Return
199 // Rindex: index into constant pool
200 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
201 //
202 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
203 // the true address of the cache entry.
204 //
205 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
206 Register index,
207 int bcp_offset,
208 size_t index_size) {
209 assert_different_registers(cache, index);
210 assert_different_registers(cache, rcpool);
211 get_cache_index_at_bcp(index, bcp_offset, index_size);
212 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
213 // convert from field index to ConstantPoolCacheEntry
214 // aarch64 already has the cache in rcpool so there is no need to
215 // install it in cache. instead we pre-add the indexed offset to
216 // rcpool and return it in cache. All clients of this method need to
217 // be modified accordingly.
218 add(cache, rcpool, index, Assembler::LSL, 5);
219 }
220
221
222 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
223 Register index,
224 Register bytecode,
225 int byte_no,
226 int bcp_offset,
227 size_t index_size) {
228 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
229 // We use a 32-bit load here since the layout of 64-bit words on
230 // little-endian machines allow us that.
231 // n.b. unlike x86 cache already includes the index offset
232 lea(bytecode, Address(cache,
233 ConstantPoolCache::base_offset()
234 + ConstantPoolCacheEntry::indices_offset()));
235 ldarw(bytecode, bytecode);
236 const int shift_count = (1 + byte_no) * BitsPerByte;
237 ubfx(bytecode, bytecode, shift_count, BitsPerByte);
238 }
239
240 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
241 Register tmp,
242 int bcp_offset,
243 size_t index_size) {
244 assert(cache != tmp, "must use different register");
245 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
246 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
247 // convert from field index to ConstantPoolCacheEntry index
248 // and from word offset to byte offset
249 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
250 ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
251 // skip past the header
252 add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
253 add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord); // construct pointer to cache entry
254 }
255
256 void InterpreterMacroAssembler::get_method_counters(Register method,
257 Register mcs, Label& skip) {
258 Label has_counters;
259 ldr(mcs, Address(method, Method::method_counters_offset()));
260 cbnz(mcs, has_counters);
261 call_VM(noreg, CAST_FROM_FN_PTR(address,
262 InterpreterRuntime::build_method_counters), method);
263 ldr(mcs, Address(method, Method::method_counters_offset()));
264 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
265 bind(has_counters);
266 }
267
268 // Load object from cpool->resolved_references(index)
269 void InterpreterMacroAssembler::load_resolved_reference_at_index(
270 Register result, Register index, Register tmp) {
271 assert_different_registers(result, index);
272
273 get_constant_pool(result);
274 // load pointer for resolved_references[] objArray
275 ldr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
276 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
277 resolve_oop_handle(result, tmp);
278 // Add in the index
279 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
280 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)));
281 }
282
283 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
284 Register cpool, Register index, Register klass, Register temp) {
285 add(temp, cpool, index, LSL, LogBytesPerWord);
286 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
287 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
288 add(klass, klass, temp, LSL, LogBytesPerWord);
289 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
290 }
291
292 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
293 Register method,
294 Register cache) {
295 const int method_offset = in_bytes(
296 ConstantPoolCache::base_offset() +
297 ((byte_no == TemplateTable::f2_byte)
298 ? ConstantPoolCacheEntry::f2_offset()
299 : ConstantPoolCacheEntry::f1_offset()));
300
301 ldr(method, Address(cache, method_offset)); // get f1 Method*
302 }
303
304 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
305 // subtype of super_klass.
306 //
307 // Args:
308 // r0: superklass
309 // Rsub_klass: subklass
310 //
311 // Kills:
312 // r2, r5
313 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
314 Label& ok_is_subtype) {
315 assert(Rsub_klass != r0, "r0 holds superklass");
316 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
317 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
318
319 // Profile the not-null value's klass.
320 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
321
322 // Do the check.
323 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
324
325 // Profile the failure of the check.
326 profile_typecheck_failed(r2); // blows r2
327 }
328
329 // Java Expression Stack
330
331 void InterpreterMacroAssembler::pop_ptr(Register r) {
332 ldr(r, post(esp, wordSize));
333 }
334
335 void InterpreterMacroAssembler::pop_i(Register r) {
336 ldrw(r, post(esp, wordSize));
337 }
338
339 void InterpreterMacroAssembler::pop_l(Register r) {
340 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
341 }
342
343 void InterpreterMacroAssembler::push_ptr(Register r) {
344 str(r, pre(esp, -wordSize));
345 }
346
347 void InterpreterMacroAssembler::push_i(Register r) {
348 str(r, pre(esp, -wordSize));
349 }
350
351 void InterpreterMacroAssembler::push_l(Register r) {
352 str(zr, pre(esp, -wordSize));
353 str(r, pre(esp, - wordSize));
354 }
355
356 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
357 ldrs(r, post(esp, wordSize));
358 }
359
360 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
361 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
362 }
363
364 void InterpreterMacroAssembler::push_f(FloatRegister r) {
365 strs(r, pre(esp, -wordSize));
366 }
367
368 void InterpreterMacroAssembler::push_d(FloatRegister r) {
369 strd(r, pre(esp, 2* -wordSize));
370 }
371
372 void InterpreterMacroAssembler::pop(TosState state) {
373 switch (state) {
374 case atos: pop_ptr(); break;
375 case btos:
376 case ztos:
377 case ctos:
378 case stos:
379 case itos: pop_i(); break;
380 case ltos: pop_l(); break;
381 case ftos: pop_f(); break;
382 case dtos: pop_d(); break;
383 case vtos: /* nothing to do */ break;
384 default: ShouldNotReachHere();
385 }
386 verify_oop(r0, state);
387 }
388
389 void InterpreterMacroAssembler::push(TosState state) {
390 verify_oop(r0, state);
391 switch (state) {
392 case atos: push_ptr(); break;
393 case btos:
394 case ztos:
395 case ctos:
396 case stos:
397 case itos: push_i(); break;
398 case ltos: push_l(); break;
399 case ftos: push_f(); break;
400 case dtos: push_d(); break;
401 case vtos: /* nothing to do */ break;
402 default : ShouldNotReachHere();
403 }
404 }
405
406 // Helpers for swap and dup
407 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
408 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
409 }
410
411 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
412 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
413 }
414
415 void InterpreterMacroAssembler::load_float(Address src) {
416 ldrs(v0, src);
417 }
418
419 void InterpreterMacroAssembler::load_double(Address src) {
420 ldrd(v0, src);
421 }
422
423 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
424 // set sender sp
425 mov(r13, sp);
426 // record last_sp
427 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
428 }
429
430 // Jump to from_interpreted entry of a call unless single stepping is possible
431 // in this thread in which case we must call the i2i entry
432 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
433 prepare_to_jump_from_interpreted();
434
435 if (JvmtiExport::can_post_interpreter_events()) {
436 Label run_compiled_code;
437 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
438 // compiled code in threads for which the event is enabled. Check here for
439 // interp_only_mode if these events CAN be enabled.
440 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
441 cbzw(rscratch1, run_compiled_code);
442 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
443 br(rscratch1);
444 bind(run_compiled_code);
445 }
446
447 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
448 br(rscratch1);
449 }
450
451 // The following two routines provide a hook so that an implementation
452 // can schedule the dispatch in two parts. amd64 does not do this.
453 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
454 }
455
456 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
457 dispatch_next(state, step);
458 }
459
460 void InterpreterMacroAssembler::dispatch_base(TosState state,
461 address* table,
462 bool verifyoop,
463 bool generate_poll) {
464 if (VerifyActivationFrameSize) {
465 Unimplemented();
466 }
467 if (verifyoop) {
468 verify_oop(r0, state);
469 }
470
471 Label safepoint;
472 address* const safepoint_table = Interpreter::safept_table(state);
473 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
474
475 if (needs_thread_local_poll) {
476 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
477 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
478 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
479 }
480
481 if (table == Interpreter::dispatch_table(state)) {
482 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
483 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
484 } else {
485 mov(rscratch2, (address)table);
486 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
487 }
488 br(rscratch2);
489
490 if (needs_thread_local_poll) {
491 bind(safepoint);
492 lea(rscratch2, ExternalAddress((address)safepoint_table));
493 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
494 br(rscratch2);
495 }
496 }
497
498 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
499 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
500 }
501
502 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
503 dispatch_base(state, Interpreter::normal_table(state));
504 }
505
506 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
507 dispatch_base(state, Interpreter::normal_table(state), false);
508 }
509
510
511 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
512 // load next bytecode
513 ldrb(rscratch1, Address(pre(rbcp, step)));
514 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
515 }
516
517 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
518 // load current bytecode
519 ldrb(rscratch1, Address(rbcp, 0));
520 dispatch_base(state, table);
521 }
522
523 // remove activation
524 //
525 // Apply stack watermark barrier.
526 // Unlock the receiver if this is a synchronized method.
527 // Unlock any Java monitors from syncronized blocks.
528 // Remove the activation from the stack.
529 //
530 // If there are locked Java monitors
531 // If throw_monitor_exception
532 // throws IllegalMonitorStateException
533 // Else if install_monitor_exception
534 // installs IllegalMonitorStateException
535 // Else
536 // no error processing
537 void InterpreterMacroAssembler::remove_activation(
538 TosState state,
539 bool throw_monitor_exception,
540 bool install_monitor_exception,
541 bool notify_jvmdi) {
542 // Note: Registers r3 xmm0 may be in use for the
543 // result check if synchronized method
544 Label unlocked, unlock, no_unlock;
545
546 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
547 // that would normally not be safe to use. Such bad returns into unsafe territory of
548 // the stack, will call InterpreterRuntime::at_unwind.
549 Label slow_path;
550 Label fast_path;
551 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
552 br(Assembler::AL, fast_path);
553 bind(slow_path);
554 push(state);
555 set_last_Java_frame(esp, rfp, (address)pc(), rscratch1);
556 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
557 reset_last_Java_frame(true);
558 pop(state);
559 bind(fast_path);
560
561 // get the value of _do_not_unlock_if_synchronized into r3
562 const Address do_not_unlock_if_synchronized(rthread,
563 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
564 ldrb(r3, do_not_unlock_if_synchronized);
565 strb(zr, do_not_unlock_if_synchronized); // reset the flag
566
567 // get method access flags
568 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
569 ldr(r2, Address(r1, Method::access_flags_offset()));
570 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
571
572 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
573 // is set.
574 cbnz(r3, no_unlock);
575
576 // unlock monitor
577 push(state); // save result
578
579 // BasicObjectLock will be first in list, since this is a
580 // synchronized method. However, need to check that the object has
581 // not been unlocked by an explicit monitorexit bytecode.
582 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
583 wordSize - (int) sizeof(BasicObjectLock));
584 // We use c_rarg1 so that if we go slow path it will be the correct
585 // register for unlock_object to pass to VM directly
586 lea(c_rarg1, monitor); // address of first monitor
587
588 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
589 cbnz(r0, unlock);
590
591 pop(state);
592 if (throw_monitor_exception) {
593 // Entry already unlocked, need to throw exception
594 call_VM(noreg, CAST_FROM_FN_PTR(address,
595 InterpreterRuntime::throw_illegal_monitor_state_exception));
596 should_not_reach_here();
597 } else {
598 // Monitor already unlocked during a stack unroll. If requested,
599 // install an illegal_monitor_state_exception. Continue with
600 // stack unrolling.
601 if (install_monitor_exception) {
602 call_VM(noreg, CAST_FROM_FN_PTR(address,
603 InterpreterRuntime::new_illegal_monitor_state_exception));
604 }
605 b(unlocked);
606 }
607
608 bind(unlock);
609 unlock_object(c_rarg1);
610 pop(state);
611
612 // Check that for block-structured locking (i.e., that all locked
613 // objects has been unlocked)
614 bind(unlocked);
615
616 // r0: Might contain return value
617
618 // Check that all monitors are unlocked
619 {
620 Label loop, exception, entry, restart;
621 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
622 const Address monitor_block_top(
623 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
624 const Address monitor_block_bot(
625 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
626
627 bind(restart);
628 // We use c_rarg1 so that if we go slow path it will be the correct
629 // register for unlock_object to pass to VM directly
630 ldr(c_rarg1, monitor_block_top); // points to current entry, starting
631 // with top-most entry
632 lea(r19, monitor_block_bot); // points to word before bottom of
633 // monitor block
634 b(entry);
635
636 // Entry already locked, need to throw exception
637 bind(exception);
638
639 if (throw_monitor_exception) {
640 // Throw exception
641 MacroAssembler::call_VM(noreg,
642 CAST_FROM_FN_PTR(address, InterpreterRuntime::
643 throw_illegal_monitor_state_exception));
644 should_not_reach_here();
645 } else {
646 // Stack unrolling. Unlock object and install illegal_monitor_exception.
647 // Unlock does not block, so don't have to worry about the frame.
648 // We don't have to preserve c_rarg1 since we are going to throw an exception.
649
650 push(state);
651 unlock_object(c_rarg1);
652 pop(state);
653
654 if (install_monitor_exception) {
655 call_VM(noreg, CAST_FROM_FN_PTR(address,
656 InterpreterRuntime::
657 new_illegal_monitor_state_exception));
658 }
659
660 b(restart);
661 }
662
663 bind(loop);
664 // check if current entry is used
665 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
666 cbnz(rscratch1, exception);
667
668 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
669 bind(entry);
670 cmp(c_rarg1, r19); // check if bottom reached
671 br(Assembler::NE, loop); // if not at bottom then check this entry
672 }
673
674 bind(no_unlock);
675
676 // jvmti support
677 if (notify_jvmdi) {
678 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
679 } else {
680 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
681 }
682
683 // remove activation
684 // get sender esp
685 ldr(rscratch2,
686 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
687 if (StackReservedPages > 0) {
688 // testing if reserved zone needs to be re-enabled
689 Label no_reserved_zone_enabling;
690
691 // look for an overflow into the stack reserved zone, i.e.
692 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
693 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
694 cmp(rscratch2, rscratch1);
695 br(Assembler::LS, no_reserved_zone_enabling);
696
697 call_VM_leaf(
698 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
699 call_VM(noreg, CAST_FROM_FN_PTR(address,
700 InterpreterRuntime::throw_delayed_StackOverflowError));
701 should_not_reach_here();
702
703 bind(no_reserved_zone_enabling);
704 }
705
706 // restore sender esp
707 mov(esp, rscratch2);
708 // remove frame anchor
709 leave();
710 // If we're returning to interpreted code we will shortly be
711 // adjusting SP to allow some space for ESP. If we're returning to
712 // compiled code the saved sender SP was saved in sender_sp, so this
713 // restores it.
714 andr(sp, esp, -16);
715 }
716
717 // Lock object
718 //
719 // Args:
720 // c_rarg1: BasicObjectLock to be used for locking
721 //
722 // Kills:
723 // r0
724 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
725 // rscratch1, rscratch2 (scratch regs)
726 void InterpreterMacroAssembler::lock_object(Register lock_reg)
727 {
728 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
729 if (LockingMode == LM_MONITOR) {
730 call_VM(noreg,
731 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
732 lock_reg);
733 } else {
734 Label done;
735
736 const Register swap_reg = r0;
737 const Register tmp = c_rarg2;
738 const Register obj_reg = c_rarg3; // Will contain the oop
739 const Register tmp2 = c_rarg4;
740 const Register tmp3 = c_rarg5;
741
742 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
743 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
744 const int mark_offset = lock_offset +
745 BasicLock::displaced_header_offset_in_bytes();
746
747 Label slow_case;
748
749 // Load object pointer into obj_reg %c_rarg3
750 ldr(obj_reg, Address(lock_reg, obj_offset));
751
752 if (DiagnoseSyncOnValueBasedClasses != 0) {
753 load_klass(tmp, obj_reg);
754 ldrw(tmp, Address(tmp, Klass::access_flags_offset()));
755 tstw(tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
756 br(Assembler::NE, slow_case);
757 }
758
759 if (LockingMode == LM_LIGHTWEIGHT) {
760 lightweight_lock(obj_reg, tmp, tmp2, tmp3, slow_case);
761 b(done);
762 } else {
763 if (UseBiasedLocking) {
764 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
765 }
766
767 // Load (object->mark() | 1) into swap_reg
768 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
769 orr(swap_reg, rscratch1, 1);
770
771 // Save (object->mark() | 1) into BasicLock's displaced header
772 str(swap_reg, Address(lock_reg, mark_offset));
773
774 assert(lock_offset == 0,
775 "displached header must be first word in BasicObjectLock");
776
777 Label fail;
778 if (PrintBiasedLockingStatistics) {
779 Label fast;
780 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail);
781 bind(fast);
782 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
783 rscratch2, rscratch1, tmp);
784 b(done);
785 bind(fail);
786 } else {
787 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
788 }
789
790 // Fast check for recursive lock.
791 //
792 // Can apply the optimization only if this is a stack lock
793 // allocated in this thread. For efficiency, we can focus on
794 // recently allocated stack locks (instead of reading the stack
795 // base and checking whether 'mark' points inside the current
796 // thread stack):
797 // 1) (mark & 7) == 0, and
798 // 2) sp <= mark < mark + os::pagesize()
799 //
800 // Warning: sp + os::pagesize can overflow the stack base. We must
801 // neither apply the optimization for an inflated lock allocated
802 // just above the thread stack (this is why condition 1 matters)
803 // nor apply the optimization if the stack lock is inside the stack
804 // of another thread. The latter is avoided even in case of overflow
805 // because we have guard pages at the end of all stacks. Hence, if
806 // we go over the stack base and hit the stack of another thread,
807 // this should not be in a writeable area that could contain a
808 // stack lock allocated by that thread. As a consequence, a stack
809 // lock less than page size away from sp is guaranteed to be
810 // owned by the current thread.
811 //
812 // These 3 tests can be done by evaluating the following
813 // expression: ((mark - sp) & (7 - os::vm_page_size())),
814 // assuming both stack pointer and pagesize have their
815 // least significant 3 bits clear.
816 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
817 // NOTE2: aarch64 does not like to subtract sp from rn so take a
818 // copy
819 mov(rscratch1, sp);
820 sub(swap_reg, swap_reg, rscratch1);
821 ands(swap_reg, swap_reg, (uint64_t)(7 - os::vm_page_size()));
822
823 // Save the test result, for recursive case, the result is zero
824 str(swap_reg, Address(lock_reg, mark_offset));
825
826 if (PrintBiasedLockingStatistics) {
827 br(Assembler::NE, slow_case);
828 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
829 rscratch2, rscratch1, tmp);
830 }
831 br(Assembler::EQ, done);
832 }
833 bind(slow_case);
834
835 // Call the runtime routine for slow case
836 if (LockingMode == LM_LIGHTWEIGHT) {
837 call_VM(noreg,
838 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter_obj),
839 obj_reg);
840 } else {
841 call_VM(noreg,
842 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
843 lock_reg);
844 }
845
846 bind(done);
847 }
848 }
849
850
851 // Unlocks an object. Used in monitorexit bytecode and
852 // remove_activation. Throws an IllegalMonitorException if object is
853 // not locked by current thread.
854 //
855 // Args:
856 // c_rarg1: BasicObjectLock for lock
857 //
858 // Kills:
859 // r0
860 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
861 // rscratch1, rscratch2 (scratch regs)
862 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
863 {
864 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
865
866 if (LockingMode == LM_MONITOR) {
867 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
868 } else {
869 Label done;
870
871 const Register swap_reg = r0;
872 const Register header_reg = c_rarg2; // Will contain the old oopMark
873 const Register obj_reg = c_rarg3; // Will contain the oop
874 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
875
876 save_bcp(); // Save in case of exception
877
878 if (LockingMode != LM_LIGHTWEIGHT) {
879 // Convert from BasicObjectLock structure to object and BasicLock
880 // structure Store the BasicLock address into %r0
881 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
882 }
883
884 // Load oop into obj_reg(%c_rarg3)
885 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
886
887 // Free entry
888 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
889
890 if (LockingMode == LM_LIGHTWEIGHT) {
891 Label slow_case;
892 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
893 b(done);
894 bind(slow_case);
895 } else {
896 if (UseBiasedLocking) {
897 biased_locking_exit(obj_reg, header_reg, done);
898 }
899
900 // Load the old header from BasicLock structure
901 ldr(header_reg, Address(swap_reg,
902 BasicLock::displaced_header_offset_in_bytes()));
903
904 // Test for recursion
905 cbz(header_reg, done);
906
907 // Atomic swap back the old header
908 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
909 }
910 // Call the runtime routine for slow case.
911 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
912 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
913
914 bind(done);
915
916 restore_bcp();
917 }
918 }
919
920 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
921 Label& zero_continue) {
922 assert(ProfileInterpreter, "must be profiling interpreter");
923 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
924 cbz(mdp, zero_continue);
925 }
926
927 // Set the method data pointer for the current bcp.
928 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
929 assert(ProfileInterpreter, "must be profiling interpreter");
930 Label set_mdp;
931 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
932
933 // Test MDO to avoid the call if it is NULL.
934 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
935 cbz(r0, set_mdp);
936 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
937 // r0: mdi
938 // mdo is guaranteed to be non-zero here, we checked for it before the call.
939 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
940 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
941 add(r0, r1, r0);
942 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
943 bind(set_mdp);
944 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
945 }
946
947 void InterpreterMacroAssembler::verify_method_data_pointer() {
948 assert(ProfileInterpreter, "must be profiling interpreter");
949 #ifdef ASSERT
950 Label verify_continue;
951 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
952 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
953 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
954 get_method(r1);
955
956 // If the mdp is valid, it will point to a DataLayout header which is
957 // consistent with the bcp. The converse is highly probable also.
958 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
959 ldr(rscratch1, Address(r1, Method::const_offset()));
960 add(r2, r2, rscratch1, Assembler::LSL);
961 lea(r2, Address(r2, ConstMethod::codes_offset()));
962 cmp(r2, rbcp);
963 br(Assembler::EQ, verify_continue);
964 // r1: method
965 // rbcp: bcp // rbcp == 22
966 // r3: mdp
967 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
968 r1, rbcp, r3);
969 bind(verify_continue);
970 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
971 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
972 #endif // ASSERT
973 }
974
975
976 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
977 int constant,
978 Register value) {
979 assert(ProfileInterpreter, "must be profiling interpreter");
980 Address data(mdp_in, constant);
981 str(value, data);
982 }
983
984
985 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
986 int constant,
987 bool decrement) {
988 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
989 }
990
991 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
992 Register reg,
993 int constant,
994 bool decrement) {
995 assert(ProfileInterpreter, "must be profiling interpreter");
996 // %%% this does 64bit counters at best it is wasting space
997 // at worst it is a rare bug when counters overflow
998
999 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1000
1001 Address addr1(mdp_in, constant);
1002 Address addr2(rscratch2, reg, Address::lsl(0));
1003 Address &addr = addr1;
1004 if (reg != noreg) {
1005 lea(rscratch2, addr1);
1006 addr = addr2;
1007 }
1008
1009 if (decrement) {
1010 // Decrement the register. Set condition codes.
1011 // Intel does this
1012 // addptr(data, (int32_t) -DataLayout::counter_increment);
1013 // If the decrement causes the counter to overflow, stay negative
1014 // Label L;
1015 // jcc(Assembler::negative, L);
1016 // addptr(data, (int32_t) DataLayout::counter_increment);
1017 // so we do this
1018 ldr(rscratch1, addr);
1019 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1020 Label L;
1021 br(Assembler::LO, L); // skip store if counter underflow
1022 str(rscratch1, addr);
1023 bind(L);
1024 } else {
1025 assert(DataLayout::counter_increment == 1,
1026 "flow-free idiom only works with 1");
1027 // Intel does this
1028 // Increment the register. Set carry flag.
1029 // addptr(data, DataLayout::counter_increment);
1030 // If the increment causes the counter to overflow, pull back by 1.
1031 // sbbptr(data, (int32_t)0);
1032 // so we do this
1033 ldr(rscratch1, addr);
1034 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1035 Label L;
1036 br(Assembler::CS, L); // skip store if counter overflow
1037 str(rscratch1, addr);
1038 bind(L);
1039 }
1040 }
1041
1042 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1043 int flag_byte_constant) {
1044 assert(ProfileInterpreter, "must be profiling interpreter");
1045 int flags_offset = in_bytes(DataLayout::flags_offset());
1046 // Set the flag
1047 ldrb(rscratch1, Address(mdp_in, flags_offset));
1048 orr(rscratch1, rscratch1, flag_byte_constant);
1049 strb(rscratch1, Address(mdp_in, flags_offset));
1050 }
1051
1052
1053 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1054 int offset,
1055 Register value,
1056 Register test_value_out,
1057 Label& not_equal_continue) {
1058 assert(ProfileInterpreter, "must be profiling interpreter");
1059 if (test_value_out == noreg) {
1060 ldr(rscratch1, Address(mdp_in, offset));
1061 cmp(value, rscratch1);
1062 } else {
1063 // Put the test value into a register, so caller can use it:
1064 ldr(test_value_out, Address(mdp_in, offset));
1065 cmp(value, test_value_out);
1066 }
1067 br(Assembler::NE, not_equal_continue);
1068 }
1069
1070
1071 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1072 int offset_of_disp) {
1073 assert(ProfileInterpreter, "must be profiling interpreter");
1074 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1075 add(mdp_in, mdp_in, rscratch1, LSL);
1076 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1077 }
1078
1079
1080 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1081 Register reg,
1082 int offset_of_disp) {
1083 assert(ProfileInterpreter, "must be profiling interpreter");
1084 lea(rscratch1, Address(mdp_in, offset_of_disp));
1085 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1086 add(mdp_in, mdp_in, rscratch1, LSL);
1087 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1088 }
1089
1090
1091 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1092 int constant) {
1093 assert(ProfileInterpreter, "must be profiling interpreter");
1094 add(mdp_in, mdp_in, (unsigned)constant);
1095 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1096 }
1097
1098
1099 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1100 assert(ProfileInterpreter, "must be profiling interpreter");
1101 // save/restore across call_VM
1102 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1103 call_VM(noreg,
1104 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1105 return_bci);
1106 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1107 }
1108
1109
1110 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1111 Register bumped_count) {
1112 if (ProfileInterpreter) {
1113 Label profile_continue;
1114
1115 // If no method data exists, go to profile_continue.
1116 // Otherwise, assign to mdp
1117 test_method_data_pointer(mdp, profile_continue);
1118
1119 // We are taking a branch. Increment the taken count.
1120 // We inline increment_mdp_data_at to return bumped_count in a register
1121 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1122 Address data(mdp, in_bytes(JumpData::taken_offset()));
1123 ldr(bumped_count, data);
1124 assert(DataLayout::counter_increment == 1,
1125 "flow-free idiom only works with 1");
1126 // Intel does this to catch overflow
1127 // addptr(bumped_count, DataLayout::counter_increment);
1128 // sbbptr(bumped_count, 0);
1129 // so we do this
1130 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1131 Label L;
1132 br(Assembler::CS, L); // skip store if counter overflow
1133 str(bumped_count, data);
1134 bind(L);
1135 // The method data pointer needs to be updated to reflect the new target.
1136 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1137 bind(profile_continue);
1138 }
1139 }
1140
1141
1142 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1143 if (ProfileInterpreter) {
1144 Label profile_continue;
1145
1146 // If no method data exists, go to profile_continue.
1147 test_method_data_pointer(mdp, profile_continue);
1148
1149 // We are taking a branch. Increment the not taken count.
1150 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1151
1152 // The method data pointer needs to be updated to correspond to
1153 // the next bytecode
1154 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1155 bind(profile_continue);
1156 }
1157 }
1158
1159
1160 void InterpreterMacroAssembler::profile_call(Register mdp) {
1161 if (ProfileInterpreter) {
1162 Label profile_continue;
1163
1164 // If no method data exists, go to profile_continue.
1165 test_method_data_pointer(mdp, profile_continue);
1166
1167 // We are making a call. Increment the count.
1168 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1169
1170 // The method data pointer needs to be updated to reflect the new target.
1171 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1172 bind(profile_continue);
1173 }
1174 }
1175
1176 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1177 if (ProfileInterpreter) {
1178 Label profile_continue;
1179
1180 // If no method data exists, go to profile_continue.
1181 test_method_data_pointer(mdp, profile_continue);
1182
1183 // We are making a call. Increment the count.
1184 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1185
1186 // The method data pointer needs to be updated to reflect the new target.
1187 update_mdp_by_constant(mdp,
1188 in_bytes(VirtualCallData::
1189 virtual_call_data_size()));
1190 bind(profile_continue);
1191 }
1192 }
1193
1194
1195 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1196 Register mdp,
1197 Register reg2,
1198 bool receiver_can_be_null) {
1199 if (ProfileInterpreter) {
1200 Label profile_continue;
1201
1202 // If no method data exists, go to profile_continue.
1203 test_method_data_pointer(mdp, profile_continue);
1204
1205 Label skip_receiver_profile;
1206 if (receiver_can_be_null) {
1207 Label not_null;
1208 // We are making a call. Increment the count for null receiver.
1209 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1210 b(skip_receiver_profile);
1211 bind(not_null);
1212 }
1213
1214 // Record the receiver type.
1215 record_klass_in_profile(receiver, mdp, reg2, true);
1216 bind(skip_receiver_profile);
1217
1218 // The method data pointer needs to be updated to reflect the new target.
1219 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1220 bind(profile_continue);
1221 }
1222 }
1223
1224 // This routine creates a state machine for updating the multi-row
1225 // type profile at a virtual call site (or other type-sensitive bytecode).
1226 // The machine visits each row (of receiver/count) until the receiver type
1227 // is found, or until it runs out of rows. At the same time, it remembers
1228 // the location of the first empty row. (An empty row records null for its
1229 // receiver, and can be allocated for a newly-observed receiver type.)
1230 // Because there are two degrees of freedom in the state, a simple linear
1231 // search will not work; it must be a decision tree. Hence this helper
1232 // function is recursive, to generate the required tree structured code.
1233 // It's the interpreter, so we are trading off code space for speed.
1234 // See below for example code.
1235 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1236 Register receiver, Register mdp,
1237 Register reg2, int start_row,
1238 Label& done, bool is_virtual_call) {
1239 if (TypeProfileWidth == 0) {
1240 if (is_virtual_call) {
1241 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1242 }
1243 #if INCLUDE_JVMCI
1244 else if (EnableJVMCI) {
1245 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1246 }
1247 #endif // INCLUDE_JVMCI
1248 } else {
1249 int non_profiled_offset = -1;
1250 if (is_virtual_call) {
1251 non_profiled_offset = in_bytes(CounterData::count_offset());
1252 }
1253 #if INCLUDE_JVMCI
1254 else if (EnableJVMCI) {
1255 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1256 }
1257 #endif // INCLUDE_JVMCI
1258
1259 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1260 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_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 non_profiled_offset) {
1268 int last_row = total_rows - 1;
1269 assert(start_row <= last_row, "must be work left to do");
1270 // Test this row for both the item and for null.
1271 // Take any of three different outcomes:
1272 // 1. found item => increment count and goto done
1273 // 2. found null => keep looking for case 1, maybe allocate this cell
1274 // 3. found something else => keep looking for cases 1 and 2
1275 // Case 3 is handled by a recursive call.
1276 for (int row = start_row; row <= last_row; row++) {
1277 Label next_test;
1278 bool test_for_null_also = (row == start_row);
1279
1280 // See if the item is item[n].
1281 int item_offset = in_bytes(item_offset_fn(row));
1282 test_mdp_data_at(mdp, item_offset, item,
1283 (test_for_null_also ? reg2 : noreg),
1284 next_test);
1285 // (Reg2 now contains the item from the CallData.)
1286
1287 // The item is item[n]. Increment count[n].
1288 int count_offset = in_bytes(item_count_offset_fn(row));
1289 increment_mdp_data_at(mdp, count_offset);
1290 b(done);
1291 bind(next_test);
1292
1293 if (test_for_null_also) {
1294 Label found_null;
1295 // Failed the equality check on item[n]... Test for null.
1296 if (start_row == last_row) {
1297 // The only thing left to do is handle the null case.
1298 if (non_profiled_offset >= 0) {
1299 cbz(reg2, found_null);
1300 // Item did not match any saved item and there is no empty row for it.
1301 // Increment total counter to indicate polymorphic case.
1302 increment_mdp_data_at(mdp, non_profiled_offset);
1303 b(done);
1304 bind(found_null);
1305 } else {
1306 cbnz(reg2, done);
1307 }
1308 break;
1309 }
1310 // Since null is rare, make it be the branch-taken case.
1311 cbz(reg2, found_null);
1312
1313 // Put all the "Case 3" tests here.
1314 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1315 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1316
1317 // Found a null. Keep searching for a matching item,
1318 // but remember that this is an empty (unused) slot.
1319 bind(found_null);
1320 }
1321 }
1322
1323 // In the fall-through case, we found no matching item, but we
1324 // observed the item[start_row] is NULL.
1325
1326 // Fill in the item field and increment the count.
1327 int item_offset = in_bytes(item_offset_fn(start_row));
1328 set_mdp_data_at(mdp, item_offset, item);
1329 int count_offset = in_bytes(item_count_offset_fn(start_row));
1330 mov(reg2, DataLayout::counter_increment);
1331 set_mdp_data_at(mdp, count_offset, reg2);
1332 if (start_row > 0) {
1333 b(done);
1334 }
1335 }
1336
1337 // Example state machine code for three profile rows:
1338 // // main copy of decision tree, rooted at row[1]
1339 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1340 // if (row[0].rec != NULL) {
1341 // // inner copy of decision tree, rooted at row[1]
1342 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1343 // if (row[1].rec != NULL) {
1344 // // degenerate decision tree, rooted at row[2]
1345 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1346 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1347 // row[2].init(rec); goto done;
1348 // } else {
1349 // // remember row[1] is empty
1350 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1351 // row[1].init(rec); goto done;
1352 // }
1353 // } else {
1354 // // remember row[0] is empty
1355 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1356 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1357 // row[0].init(rec); goto done;
1358 // }
1359 // done:
1360
1361 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1362 Register mdp, Register reg2,
1363 bool is_virtual_call) {
1364 assert(ProfileInterpreter, "must be profiling");
1365 Label done;
1366
1367 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1368
1369 bind (done);
1370 }
1371
1372 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1373 Register mdp) {
1374 if (ProfileInterpreter) {
1375 Label profile_continue;
1376 uint row;
1377
1378 // If no method data exists, go to profile_continue.
1379 test_method_data_pointer(mdp, profile_continue);
1380
1381 // Update the total ret count.
1382 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1383
1384 for (row = 0; row < RetData::row_limit(); row++) {
1385 Label next_test;
1386
1387 // See if return_bci is equal to bci[n]:
1388 test_mdp_data_at(mdp,
1389 in_bytes(RetData::bci_offset(row)),
1390 return_bci, noreg,
1391 next_test);
1392
1393 // return_bci is equal to bci[n]. Increment the count.
1394 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1395
1396 // The method data pointer needs to be updated to reflect the new target.
1397 update_mdp_by_offset(mdp,
1398 in_bytes(RetData::bci_displacement_offset(row)));
1399 b(profile_continue);
1400 bind(next_test);
1401 }
1402
1403 update_mdp_for_ret(return_bci);
1404
1405 bind(profile_continue);
1406 }
1407 }
1408
1409 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1410 if (ProfileInterpreter) {
1411 Label profile_continue;
1412
1413 // If no method data exists, go to profile_continue.
1414 test_method_data_pointer(mdp, profile_continue);
1415
1416 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1417
1418 // The method data pointer needs to be updated.
1419 int mdp_delta = in_bytes(BitData::bit_data_size());
1420 if (TypeProfileCasts) {
1421 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1422 }
1423 update_mdp_by_constant(mdp, mdp_delta);
1424
1425 bind(profile_continue);
1426 }
1427 }
1428
1429 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1430 if (ProfileInterpreter && TypeProfileCasts) {
1431 Label profile_continue;
1432
1433 // If no method data exists, go to profile_continue.
1434 test_method_data_pointer(mdp, profile_continue);
1435
1436 int count_offset = in_bytes(CounterData::count_offset());
1437 // Back up the address, since we have already bumped the mdp.
1438 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1439
1440 // *Decrement* the counter. We expect to see zero or small negatives.
1441 increment_mdp_data_at(mdp, count_offset, true);
1442
1443 bind (profile_continue);
1444 }
1445 }
1446
1447 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1448 if (ProfileInterpreter) {
1449 Label profile_continue;
1450
1451 // If no method data exists, go to profile_continue.
1452 test_method_data_pointer(mdp, profile_continue);
1453
1454 // The method data pointer needs to be updated.
1455 int mdp_delta = in_bytes(BitData::bit_data_size());
1456 if (TypeProfileCasts) {
1457 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1458
1459 // Record the object type.
1460 record_klass_in_profile(klass, mdp, reg2, false);
1461 }
1462 update_mdp_by_constant(mdp, mdp_delta);
1463
1464 bind(profile_continue);
1465 }
1466 }
1467
1468 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1469 if (ProfileInterpreter) {
1470 Label profile_continue;
1471
1472 // If no method data exists, go to profile_continue.
1473 test_method_data_pointer(mdp, profile_continue);
1474
1475 // Update the default case count
1476 increment_mdp_data_at(mdp,
1477 in_bytes(MultiBranchData::default_count_offset()));
1478
1479 // The method data pointer needs to be updated.
1480 update_mdp_by_offset(mdp,
1481 in_bytes(MultiBranchData::
1482 default_displacement_offset()));
1483
1484 bind(profile_continue);
1485 }
1486 }
1487
1488 void InterpreterMacroAssembler::profile_switch_case(Register index,
1489 Register mdp,
1490 Register reg2) {
1491 if (ProfileInterpreter) {
1492 Label profile_continue;
1493
1494 // If no method data exists, go to profile_continue.
1495 test_method_data_pointer(mdp, profile_continue);
1496
1497 // Build the base (index * per_case_size_in_bytes()) +
1498 // case_array_offset_in_bytes()
1499 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1500 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1501 Assembler::maddw(index, index, reg2, rscratch1);
1502
1503 // Update the case count
1504 increment_mdp_data_at(mdp,
1505 index,
1506 in_bytes(MultiBranchData::relative_count_offset()));
1507
1508 // The method data pointer needs to be updated.
1509 update_mdp_by_offset(mdp,
1510 index,
1511 in_bytes(MultiBranchData::
1512 relative_displacement_offset()));
1513
1514 bind(profile_continue);
1515 }
1516 }
1517
1518 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1519 if (state == atos) {
1520 MacroAssembler::verify_oop(reg);
1521 }
1522 }
1523
1524 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1525
1526
1527 void InterpreterMacroAssembler::notify_method_entry() {
1528 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1529 // track stack depth. If it is possible to enter interp_only_mode we add
1530 // the code to check if the event should be sent.
1531 if (JvmtiExport::can_post_interpreter_events()) {
1532 Label L;
1533 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1534 cbzw(r3, L);
1535 call_VM(noreg, CAST_FROM_FN_PTR(address,
1536 InterpreterRuntime::post_method_entry));
1537 bind(L);
1538 }
1539
1540 {
1541 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1542 get_method(c_rarg1);
1543 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1544 rthread, c_rarg1);
1545 }
1546
1547 // RedefineClasses() tracing support for obsolete method entry
1548 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1549 get_method(c_rarg1);
1550 call_VM_leaf(
1551 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1552 rthread, c_rarg1);
1553 }
1554
1555 }
1556
1557
1558 void InterpreterMacroAssembler::notify_method_exit(
1559 TosState state, NotifyMethodExitMode mode) {
1560 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1561 // track stack depth. If it is possible to enter interp_only_mode we add
1562 // the code to check if the event should be sent.
1563 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1564 Label L;
1565 // Note: frame::interpreter_frame_result has a dependency on how the
1566 // method result is saved across the call to post_method_exit. If this
1567 // is changed then the interpreter_frame_result implementation will
1568 // need to be updated too.
1569
1570 // template interpreter will leave the result on the top of the stack.
1571 push(state);
1572 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1573 cbz(r3, L);
1574 call_VM(noreg,
1575 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1576 bind(L);
1577 pop(state);
1578 }
1579
1580 {
1581 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1582 push(state);
1583 get_method(c_rarg1);
1584 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1585 rthread, c_rarg1);
1586 pop(state);
1587 }
1588 }
1589
1590
1591 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1592 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1593 int increment, Address mask,
1594 Register scratch, Register scratch2,
1595 bool preloaded, Condition cond,
1596 Label* where) {
1597 if (!preloaded) {
1598 ldrw(scratch, counter_addr);
1599 }
1600 add(scratch, scratch, increment);
1601 strw(scratch, counter_addr);
1602 ldrw(scratch2, mask);
1603 ands(scratch, scratch, scratch2);
1604 br(cond, *where);
1605 }
1606
1607 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1608 int number_of_arguments) {
1609 // interpreter specific
1610 //
1611 // Note: No need to save/restore rbcp & rlocals pointer since these
1612 // are callee saved registers and no blocking/ GC can happen
1613 // in leaf calls.
1614 #ifdef ASSERT
1615 {
1616 Label L;
1617 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1618 cbz(rscratch1, L);
1619 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1620 " last_sp != NULL");
1621 bind(L);
1622 }
1623 #endif /* ASSERT */
1624 // super call
1625 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1626 }
1627
1628 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1629 Register java_thread,
1630 Register last_java_sp,
1631 address entry_point,
1632 int number_of_arguments,
1633 bool check_exceptions) {
1634 // interpreter specific
1635 //
1636 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1637 // really make a difference for these runtime calls, since they are
1638 // slow anyway. Btw., bcp must be saved/restored since it may change
1639 // due to GC.
1640 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1641 save_bcp();
1642 #ifdef ASSERT
1643 {
1644 Label L;
1645 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1646 cbz(rscratch1, L);
1647 stop("InterpreterMacroAssembler::call_VM_base:"
1648 " last_sp != NULL");
1649 bind(L);
1650 }
1651 #endif /* ASSERT */
1652 // super call
1653 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1654 entry_point, number_of_arguments,
1655 check_exceptions);
1656 // interpreter specific
1657 restore_bcp();
1658 restore_locals();
1659 }
1660
1661 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1662 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1663 Label update, next, none;
1664
1665 verify_oop(obj);
1666
1667 cbnz(obj, update);
1668 orptr(mdo_addr, TypeEntries::null_seen);
1669 b(next);
1670
1671 bind(update);
1672 load_klass(obj, obj);
1673
1674 ldr(rscratch1, mdo_addr);
1675 eor(obj, obj, rscratch1);
1676 tst(obj, TypeEntries::type_klass_mask);
1677 br(Assembler::EQ, next); // klass seen before, nothing to
1678 // do. The unknown bit may have been
1679 // set already but no need to check.
1680
1681 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1682 // already unknown. Nothing to do anymore.
1683
1684 cbz(rscratch1, none);
1685 cmp(rscratch1, (u1)TypeEntries::null_seen);
1686 br(Assembler::EQ, none);
1687 // There is a chance that the checks above
1688 // fail if another thread has just set the
1689 // profiling to this obj's klass
1690 eor(obj, obj, rscratch1); // get back original value before XOR
1691 ldr(rscratch1, mdo_addr);
1692 eor(obj, obj, rscratch1);
1693 tst(obj, TypeEntries::type_klass_mask);
1694 br(Assembler::EQ, next);
1695
1696 // different than before. Cannot keep accurate profile.
1697 orptr(mdo_addr, TypeEntries::type_unknown);
1698 b(next);
1699
1700 bind(none);
1701 // first time here. Set profile type.
1702 str(obj, mdo_addr);
1703 #ifdef ASSERT
1704 andr(obj, obj, TypeEntries::type_mask);
1705 verify_klass_ptr(obj);
1706 #endif
1707
1708 bind(next);
1709 }
1710
1711 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1712 if (!ProfileInterpreter) {
1713 return;
1714 }
1715
1716 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1717 Label profile_continue;
1718
1719 test_method_data_pointer(mdp, profile_continue);
1720
1721 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1722
1723 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1724 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1725 br(Assembler::NE, profile_continue);
1726
1727 if (MethodData::profile_arguments()) {
1728 Label done;
1729 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1730
1731 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1732 if (i > 0 || MethodData::profile_return()) {
1733 // If return value type is profiled we may have no argument to profile
1734 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1735 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1736 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1737 add(rscratch1, mdp, off_to_args);
1738 br(Assembler::LT, done);
1739 }
1740 ldr(tmp, Address(callee, Method::const_offset()));
1741 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1742 // stack offset o (zero based) from the start of the argument
1743 // list, for n arguments translates into offset n - o - 1 from
1744 // the end of the argument list
1745 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1746 sub(tmp, tmp, rscratch1);
1747 sub(tmp, tmp, 1);
1748 Address arg_addr = argument_address(tmp);
1749 ldr(tmp, arg_addr);
1750
1751 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1752 profile_obj_type(tmp, mdo_arg_addr);
1753
1754 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1755 off_to_args += to_add;
1756 }
1757
1758 if (MethodData::profile_return()) {
1759 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1760 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1761 }
1762
1763 add(rscratch1, mdp, off_to_args);
1764 bind(done);
1765 mov(mdp, rscratch1);
1766
1767 if (MethodData::profile_return()) {
1768 // We're right after the type profile for the last
1769 // argument. tmp is the number of cells left in the
1770 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1771 // if there's a return to profile.
1772 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1773 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1774 }
1775 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1776 } else {
1777 assert(MethodData::profile_return(), "either profile call args or call ret");
1778 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1779 }
1780
1781 // mdp points right after the end of the
1782 // CallTypeData/VirtualCallTypeData, right after the cells for the
1783 // return value type if there's one
1784
1785 bind(profile_continue);
1786 }
1787 }
1788
1789 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1790 assert_different_registers(mdp, ret, tmp, rbcp);
1791 if (ProfileInterpreter && MethodData::profile_return()) {
1792 Label profile_continue, done;
1793
1794 test_method_data_pointer(mdp, profile_continue);
1795
1796 if (MethodData::profile_return_jsr292_only()) {
1797 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1798
1799 // If we don't profile all invoke bytecodes we must make sure
1800 // it's a bytecode we indeed profile. We can't go back to the
1801 // begining of the ProfileData we intend to update to check its
1802 // type because we're right after it and we don't known its
1803 // length
1804 Label do_profile;
1805 ldrb(rscratch1, Address(rbcp, 0));
1806 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1807 br(Assembler::EQ, do_profile);
1808 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1809 br(Assembler::EQ, do_profile);
1810 get_method(tmp);
1811 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1812 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1813 br(Assembler::NE, profile_continue);
1814
1815 bind(do_profile);
1816 }
1817
1818 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1819 mov(tmp, ret);
1820 profile_obj_type(tmp, mdo_ret_addr);
1821
1822 bind(profile_continue);
1823 }
1824 }
1825
1826 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1827 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1828 if (ProfileInterpreter && MethodData::profile_parameters()) {
1829 Label profile_continue, done;
1830
1831 test_method_data_pointer(mdp, profile_continue);
1832
1833 // Load the offset of the area within the MDO used for
1834 // parameters. If it's negative we're not profiling any parameters
1835 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1836 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1837
1838 // Compute a pointer to the area for parameters from the offset
1839 // and move the pointer to the slot for the last
1840 // parameters. Collect profiling from last parameter down.
1841 // mdo start + parameters offset + array length - 1
1842 add(mdp, mdp, tmp1);
1843 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1844 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1845
1846 Label loop;
1847 bind(loop);
1848
1849 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1850 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1851 int per_arg_scale = exact_log2(DataLayout::cell_size);
1852 add(rscratch1, mdp, off_base);
1853 add(rscratch2, mdp, type_base);
1854
1855 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1856 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1857
1858 // load offset on the stack from the slot for this parameter
1859 ldr(tmp2, arg_off);
1860 neg(tmp2, tmp2);
1861 // read the parameter from the local area
1862 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1863
1864 // profile the parameter
1865 profile_obj_type(tmp2, arg_type);
1866
1867 // go to next parameter
1868 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1869 br(Assembler::GE, loop);
1870
1871 bind(profile_continue);
1872 }
1873 }