1 /*
2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 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 "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interp_masm_aarch64.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "logging/log.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/markOop.hpp"
36 #include "oops/method.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/valueKlass.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
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 unsigned long 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 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
293 // subtype of super_klass.
294 //
295 // Args:
296 // r0: superklass
297 // Rsub_klass: subklass
298 //
299 // Kills:
300 // r2, r5
301 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
302 Label& ok_is_subtype) {
303 assert(Rsub_klass != r0, "r0 holds superklass");
304 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
305 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
306
307 // Profile the not-null value's klass.
308 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
309
310 // Do the check.
311 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
312
313 // Profile the failure of the check.
314 profile_typecheck_failed(r2); // blows r2
315 }
316
317 // Java Expression Stack
318
319 void InterpreterMacroAssembler::pop_ptr(Register r) {
320 ldr(r, post(esp, wordSize));
321 }
322
323 void InterpreterMacroAssembler::pop_i(Register r) {
324 ldrw(r, post(esp, wordSize));
325 }
326
327 void InterpreterMacroAssembler::pop_l(Register r) {
328 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
329 }
330
331 void InterpreterMacroAssembler::push_ptr(Register r) {
332 str(r, pre(esp, -wordSize));
333 }
334
335 void InterpreterMacroAssembler::push_i(Register r) {
336 str(r, pre(esp, -wordSize));
337 }
338
339 void InterpreterMacroAssembler::push_l(Register r) {
340 str(zr, pre(esp, -wordSize));
341 str(r, pre(esp, - wordSize));
342 }
343
344 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
345 ldrs(r, post(esp, wordSize));
346 }
347
348 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
349 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
350 }
351
352 void InterpreterMacroAssembler::push_f(FloatRegister r) {
353 strs(r, pre(esp, -wordSize));
354 }
355
356 void InterpreterMacroAssembler::push_d(FloatRegister r) {
357 strd(r, pre(esp, 2* -wordSize));
358 }
359
360 void InterpreterMacroAssembler::pop(TosState state) {
361 switch (state) {
362 case atos: pop_ptr(); break;
363 case btos:
364 case ztos:
365 case ctos:
366 case stos:
367 case itos: pop_i(); break;
368 case ltos: pop_l(); break;
369 case ftos: pop_f(); break;
370 case dtos: pop_d(); break;
371 case vtos: /* nothing to do */ break;
372 default: ShouldNotReachHere();
373 }
374 verify_oop(r0, state);
375 }
376
377 void InterpreterMacroAssembler::push(TosState state) {
378 verify_oop(r0, state);
379 switch (state) {
380 case atos: push_ptr(); break;
381 case btos:
382 case ztos:
383 case ctos:
384 case stos:
385 case itos: push_i(); break;
386 case ltos: push_l(); break;
387 case ftos: push_f(); break;
388 case dtos: push_d(); break;
389 case vtos: /* nothing to do */ break;
390 default : ShouldNotReachHere();
391 }
392 }
393
394 // Helpers for swap and dup
395 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
396 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
397 }
398
399 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
400 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
401 }
402
403 void InterpreterMacroAssembler::load_float(Address src) {
404 ldrs(v0, src);
405 }
406
407 void InterpreterMacroAssembler::load_double(Address src) {
408 ldrd(v0, src);
409 }
410
411 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
412 // set sender sp
413 mov(r13, sp);
414 // record last_sp
415 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
416 }
417
418 // Jump to from_interpreted entry of a call unless single stepping is possible
419 // in this thread in which case we must call the i2i entry
420 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
421 prepare_to_jump_from_interpreted();
422
423 if (JvmtiExport::can_post_interpreter_events()) {
424 Label run_compiled_code;
425 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
426 // compiled code in threads for which the event is enabled. Check here for
427 // interp_only_mode if these events CAN be enabled.
428 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
429 cbzw(rscratch1, run_compiled_code);
430 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
431 br(rscratch1);
432 bind(run_compiled_code);
433 }
434
435 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
436 br(rscratch1);
437 }
438
439 // The following two routines provide a hook so that an implementation
440 // can schedule the dispatch in two parts. amd64 does not do this.
441 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
442 }
443
444 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
445 dispatch_next(state, step);
446 }
447
448 void InterpreterMacroAssembler::dispatch_base(TosState state,
449 address* table,
450 bool verifyoop,
451 bool generate_poll) {
452 if (VerifyActivationFrameSize) {
453 Unimplemented();
454 }
455 if (verifyoop) {
456 verify_oop(r0, state);
457 }
458
459 Label safepoint;
460 address* const safepoint_table = Interpreter::safept_table(state);
461 bool needs_thread_local_poll = generate_poll &&
462 SafepointMechanism::uses_thread_local_poll() && table != safepoint_table;
463
464 if (needs_thread_local_poll) {
465 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
466 ldr(rscratch2, Address(rthread, Thread::polling_page_offset()));
467 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
468 }
469
470 if (table == Interpreter::dispatch_table(state)) {
471 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
472 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
473 } else {
474 mov(rscratch2, (address)table);
475 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
476 }
477 br(rscratch2);
478
479 if (needs_thread_local_poll) {
480 bind(safepoint);
481 lea(rscratch2, ExternalAddress((address)safepoint_table));
482 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
483 br(rscratch2);
484 }
485 }
486
487 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
488 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
489 }
490
491 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
492 dispatch_base(state, Interpreter::normal_table(state));
493 }
494
495 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
496 dispatch_base(state, Interpreter::normal_table(state), false);
497 }
498
499
500 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
501 // load next bytecode
502 ldrb(rscratch1, Address(pre(rbcp, step)));
503 dispatch_base(state, Interpreter::dispatch_table(state), generate_poll);
504 }
505
506 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
507 // load current bytecode
508 ldrb(rscratch1, Address(rbcp, 0));
509 dispatch_base(state, table);
510 }
511
512 // remove activation
513 //
514 // Unlock the receiver if this is a synchronized method.
515 // Unlock any Java monitors from syncronized blocks.
516 // Remove the activation from the stack.
517 //
518 // If there are locked Java monitors
519 // If throw_monitor_exception
520 // throws IllegalMonitorStateException
521 // Else if install_monitor_exception
522 // installs IllegalMonitorStateException
523 // Else
524 // no error processing
525 void InterpreterMacroAssembler::remove_activation(
526 TosState state,
527 bool throw_monitor_exception,
528 bool install_monitor_exception,
529 bool notify_jvmdi) {
530 // Note: Registers r3 xmm0 may be in use for the
531 // result check if synchronized method
532 Label unlocked, unlock, no_unlock;
533
534 // get the value of _do_not_unlock_if_synchronized into r3
535 const Address do_not_unlock_if_synchronized(rthread,
536 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
537 ldrb(r3, do_not_unlock_if_synchronized);
538 strb(zr, do_not_unlock_if_synchronized); // reset the flag
539
540 // get method access flags
541 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
542 ldr(r2, Address(r1, Method::access_flags_offset()));
543 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
544
545 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
546 // is set.
547 cbnz(r3, no_unlock);
548
549 // unlock monitor
550 push(state); // save result
551
552 // BasicObjectLock will be first in list, since this is a
553 // synchronized method. However, need to check that the object has
554 // not been unlocked by an explicit monitorexit bytecode.
555 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
556 wordSize - (int) sizeof(BasicObjectLock));
557 // We use c_rarg1 so that if we go slow path it will be the correct
558 // register for unlock_object to pass to VM directly
559 lea(c_rarg1, monitor); // address of first monitor
560
561 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
562 cbnz(r0, unlock);
563
564 pop(state);
565 if (throw_monitor_exception) {
566 // Entry already unlocked, need to throw exception
567 call_VM(noreg, CAST_FROM_FN_PTR(address,
568 InterpreterRuntime::throw_illegal_monitor_state_exception));
569 should_not_reach_here();
570 } else {
571 // Monitor already unlocked during a stack unroll. If requested,
572 // install an illegal_monitor_state_exception. Continue with
573 // stack unrolling.
574 if (install_monitor_exception) {
575 call_VM(noreg, CAST_FROM_FN_PTR(address,
576 InterpreterRuntime::new_illegal_monitor_state_exception));
577 }
578 b(unlocked);
579 }
580
581 bind(unlock);
582 unlock_object(c_rarg1);
583 pop(state);
584
585 // Check that for block-structured locking (i.e., that all locked
586 // objects has been unlocked)
587 bind(unlocked);
588
589 // r0: Might contain return value
590
591 // Check that all monitors are unlocked
592 {
593 Label loop, exception, entry, restart;
594 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
595 const Address monitor_block_top(
596 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
597 const Address monitor_block_bot(
598 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
599
600 bind(restart);
601 // We use c_rarg1 so that if we go slow path it will be the correct
602 // register for unlock_object to pass to VM directly
603 ldr(c_rarg1, monitor_block_top); // points to current entry, starting
604 // with top-most entry
605 lea(r19, monitor_block_bot); // points to word before bottom of
606 // monitor block
607 b(entry);
608
609 // Entry already locked, need to throw exception
610 bind(exception);
611
612 if (throw_monitor_exception) {
613 // Throw exception
614 MacroAssembler::call_VM(noreg,
615 CAST_FROM_FN_PTR(address, InterpreterRuntime::
616 throw_illegal_monitor_state_exception));
617 should_not_reach_here();
618 } else {
619 // Stack unrolling. Unlock object and install illegal_monitor_exception.
620 // Unlock does not block, so don't have to worry about the frame.
621 // We don't have to preserve c_rarg1 since we are going to throw an exception.
622
623 push(state);
624 unlock_object(c_rarg1);
625 pop(state);
626
627 if (install_monitor_exception) {
628 call_VM(noreg, CAST_FROM_FN_PTR(address,
629 InterpreterRuntime::
630 new_illegal_monitor_state_exception));
631 }
632
633 b(restart);
634 }
635
636 bind(loop);
637 // check if current entry is used
638 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
639 cbnz(rscratch1, exception);
640
641 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
642 bind(entry);
643 cmp(c_rarg1, r19); // check if bottom reached
644 br(Assembler::NE, loop); // if not at bottom then check this entry
645 }
646
647 bind(no_unlock);
648
649 // jvmti support
650 if (notify_jvmdi) {
651 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
652 } else {
653 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
654 }
655
656 // remove activation
657 // get sender esp
658 ldr(esp,
659 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
660
661 if (StackReservedPages > 0) {
662 // testing if reserved zone needs to be re-enabled
663 Label no_reserved_zone_enabling;
664
665 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
666 cmp(esp, rscratch1);
667 br(Assembler::LS, no_reserved_zone_enabling);
668
669 call_VM_leaf(
670 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
671 call_VM(noreg, CAST_FROM_FN_PTR(address,
672 InterpreterRuntime::throw_delayed_StackOverflowError));
673 should_not_reach_here();
674
675 bind(no_reserved_zone_enabling);
676 }
677
678 if (state == atos && ValueTypeReturnedAsFields) {
679 Label skip;
680
681 // Test if the return type is a value type
682 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
683 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
684 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
685 cmpw(rscratch1, (u1) T_VALUETYPE);
686 br(Assembler::NE, skip);
687
688 // We are returning a value type, load its fields into registers
689 // Load fields from a buffered value with a value class specific handler
690
691 load_klass(rscratch1 /*dst*/, r0 /*src*/);
692 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_valueklass_fixed_block_offset()));
693 ldr(rscratch1, Address(rscratch1, ValueKlass::unpack_handler_offset()));
694 cbz(rscratch1, skip);
695
696 blrt(rscratch1, 1, 0, 0);
697
698 bind(skip);
699 }
700
701 // remove frame anchor
702 leave();
703 // If we're returning to interpreted code we will shortly be
704 // adjusting SP to allow some space for ESP. If we're returning to
705 // compiled code the saved sender SP was saved in sender_sp, so this
706 // restores it.
707 andr(sp, esp, -16);
708 }
709
710 // Lock object
711 //
712 // Args:
713 // c_rarg1: BasicObjectLock to be used for locking
714 //
715 // Kills:
716 // r0
717 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
718 // rscratch1, rscratch2 (scratch regs)
719 void InterpreterMacroAssembler::lock_object(Register lock_reg)
720 {
721 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
722 if (UseHeavyMonitors) {
723 call_VM(noreg,
724 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
725 lock_reg);
726 } else {
727 Label done;
728
729 const Register swap_reg = r0;
730 const Register tmp = c_rarg2;
731 const Register obj_reg = c_rarg3; // Will contain the oop
732
733 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
734 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
735 const int mark_offset = lock_offset +
736 BasicLock::displaced_header_offset_in_bytes();
737
738 Label slow_case;
739
740 // Load object pointer into obj_reg %c_rarg3
741 ldr(obj_reg, Address(lock_reg, obj_offset));
742
743 if (UseBiasedLocking) {
744 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
745 }
746
747 // Load (object->mark() | 1) into swap_reg
748 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
749 orr(swap_reg, rscratch1, 1);
750
751 // Save (object->mark() | 1) into BasicLock's displaced header
752 str(swap_reg, Address(lock_reg, mark_offset));
753
754 if (EnableValhalla && !UseBiasedLocking) {
755 // For slow path is_always_locked, using biased, which is never natural for !UseBiasLocking
756 andr(swap_reg, swap_reg, ~markOopDesc::biased_lock_bit_in_place);
757 }
758
759 assert(lock_offset == 0,
760 "displached header must be first word in BasicObjectLock");
761
762 Label fail;
763 if (PrintBiasedLockingStatistics) {
764 Label fast;
765 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail);
766 bind(fast);
767 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
768 rscratch2, rscratch1, tmp);
769 b(done);
770 bind(fail);
771 } else {
772 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
773 }
774
775 // Test if the oopMark is an obvious stack pointer, i.e.,
776 // 1) (mark & 7) == 0, and
777 // 2) rsp <= mark < mark + os::pagesize()
778 //
779 // These 3 tests can be done by evaluating the following
780 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
781 // assuming both stack pointer and pagesize have their
782 // least significant 3 bits clear.
783 // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg
784 // NOTE2: aarch64 does not like to subtract sp from rn so take a
785 // copy
786 mov(rscratch1, sp);
787 sub(swap_reg, swap_reg, rscratch1);
788 ands(swap_reg, swap_reg, (unsigned long)(7 - os::vm_page_size()));
789
790 // Save the test result, for recursive case, the result is zero
791 str(swap_reg, Address(lock_reg, mark_offset));
792
793 if (PrintBiasedLockingStatistics) {
794 br(Assembler::NE, slow_case);
795 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
796 rscratch2, rscratch1, tmp);
797 }
798 br(Assembler::EQ, done);
799
800 bind(slow_case);
801
802 // Call the runtime routine for slow case
803 call_VM(noreg,
804 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
805 lock_reg);
806
807 bind(done);
808 }
809 }
810
811
812 // Unlocks an object. Used in monitorexit bytecode and
813 // remove_activation. Throws an IllegalMonitorException if object is
814 // not locked by current thread.
815 //
816 // Args:
817 // c_rarg1: BasicObjectLock for lock
818 //
819 // Kills:
820 // r0
821 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
822 // rscratch1, rscratch2 (scratch regs)
823 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
824 {
825 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
826
827 if (UseHeavyMonitors) {
828 call_VM(noreg,
829 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
830 lock_reg);
831 } else {
832 Label done;
833
834 const Register swap_reg = r0;
835 const Register header_reg = c_rarg2; // Will contain the old oopMark
836 const Register obj_reg = c_rarg3; // Will contain the oop
837
838 save_bcp(); // Save in case of exception
839
840 // Convert from BasicObjectLock structure to object and BasicLock
841 // structure Store the BasicLock address into %r0
842 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
843
844 // Load oop into obj_reg(%c_rarg3)
845 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
846
847 // Free entry
848 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
849
850 if (UseBiasedLocking) {
851 biased_locking_exit(obj_reg, header_reg, done);
852 }
853
854 // Load the old header from BasicLock structure
855 ldr(header_reg, Address(swap_reg,
856 BasicLock::displaced_header_offset_in_bytes()));
857
858 // Test for recursion
859 cbz(header_reg, done);
860
861 // Atomic swap back the old header
862 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
863
864 // Call the runtime routine for slow case.
865 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
866 call_VM(noreg,
867 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
868 lock_reg);
869
870 bind(done);
871
872 restore_bcp();
873 }
874 }
875
876 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
877 Label& zero_continue) {
878 assert(ProfileInterpreter, "must be profiling interpreter");
879 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
880 cbz(mdp, zero_continue);
881 }
882
883 // Set the method data pointer for the current bcp.
884 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
885 assert(ProfileInterpreter, "must be profiling interpreter");
886 Label set_mdp;
887 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
888
889 // Test MDO to avoid the call if it is NULL.
890 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
891 cbz(r0, set_mdp);
892 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
893 // r0: mdi
894 // mdo is guaranteed to be non-zero here, we checked for it before the call.
895 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
896 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
897 add(r0, r1, r0);
898 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
899 bind(set_mdp);
900 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
901 }
902
903 void InterpreterMacroAssembler::verify_method_data_pointer() {
904 assert(ProfileInterpreter, "must be profiling interpreter");
905 #ifdef ASSERT
906 Label verify_continue;
907 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
908 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
909 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
910 get_method(r1);
911
912 // If the mdp is valid, it will point to a DataLayout header which is
913 // consistent with the bcp. The converse is highly probable also.
914 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
915 ldr(rscratch1, Address(r1, Method::const_offset()));
916 add(r2, r2, rscratch1, Assembler::LSL);
917 lea(r2, Address(r2, ConstMethod::codes_offset()));
918 cmp(r2, rbcp);
919 br(Assembler::EQ, verify_continue);
920 // r1: method
921 // rbcp: bcp // rbcp == 22
922 // r3: mdp
923 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
924 r1, rbcp, r3);
925 bind(verify_continue);
926 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
927 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
928 #endif // ASSERT
929 }
930
931
932 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
933 int constant,
934 Register value) {
935 assert(ProfileInterpreter, "must be profiling interpreter");
936 Address data(mdp_in, constant);
937 str(value, data);
938 }
939
940
941 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
942 int constant,
943 bool decrement) {
944 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
945 }
946
947 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
948 Register reg,
949 int constant,
950 bool decrement) {
951 assert(ProfileInterpreter, "must be profiling interpreter");
952 // %%% this does 64bit counters at best it is wasting space
953 // at worst it is a rare bug when counters overflow
954
955 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
956
957 Address addr1(mdp_in, constant);
958 Address addr2(rscratch2, reg, Address::lsl(0));
959 Address &addr = addr1;
960 if (reg != noreg) {
961 lea(rscratch2, addr1);
962 addr = addr2;
963 }
964
965 if (decrement) {
966 // Decrement the register. Set condition codes.
967 // Intel does this
968 // addptr(data, (int32_t) -DataLayout::counter_increment);
969 // If the decrement causes the counter to overflow, stay negative
970 // Label L;
971 // jcc(Assembler::negative, L);
972 // addptr(data, (int32_t) DataLayout::counter_increment);
973 // so we do this
974 ldr(rscratch1, addr);
975 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
976 Label L;
977 br(Assembler::LO, L); // skip store if counter underflow
978 str(rscratch1, addr);
979 bind(L);
980 } else {
981 assert(DataLayout::counter_increment == 1,
982 "flow-free idiom only works with 1");
983 // Intel does this
984 // Increment the register. Set carry flag.
985 // addptr(data, DataLayout::counter_increment);
986 // If the increment causes the counter to overflow, pull back by 1.
987 // sbbptr(data, (int32_t)0);
988 // so we do this
989 ldr(rscratch1, addr);
990 adds(rscratch1, rscratch1, DataLayout::counter_increment);
991 Label L;
992 br(Assembler::CS, L); // skip store if counter overflow
993 str(rscratch1, addr);
994 bind(L);
995 }
996 }
997
998 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
999 int flag_byte_constant) {
1000 assert(ProfileInterpreter, "must be profiling interpreter");
1001 int flags_offset = in_bytes(DataLayout::flags_offset());
1002 // Set the flag
1003 ldrb(rscratch1, Address(mdp_in, flags_offset));
1004 orr(rscratch1, rscratch1, flag_byte_constant);
1005 strb(rscratch1, Address(mdp_in, flags_offset));
1006 }
1007
1008
1009 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1010 int offset,
1011 Register value,
1012 Register test_value_out,
1013 Label& not_equal_continue) {
1014 assert(ProfileInterpreter, "must be profiling interpreter");
1015 if (test_value_out == noreg) {
1016 ldr(rscratch1, Address(mdp_in, offset));
1017 cmp(value, rscratch1);
1018 } else {
1019 // Put the test value into a register, so caller can use it:
1020 ldr(test_value_out, Address(mdp_in, offset));
1021 cmp(value, test_value_out);
1022 }
1023 br(Assembler::NE, not_equal_continue);
1024 }
1025
1026
1027 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1028 int offset_of_disp) {
1029 assert(ProfileInterpreter, "must be profiling interpreter");
1030 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1031 add(mdp_in, mdp_in, rscratch1, LSL);
1032 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1033 }
1034
1035
1036 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1037 Register reg,
1038 int offset_of_disp) {
1039 assert(ProfileInterpreter, "must be profiling interpreter");
1040 lea(rscratch1, Address(mdp_in, offset_of_disp));
1041 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1042 add(mdp_in, mdp_in, rscratch1, LSL);
1043 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1044 }
1045
1046
1047 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1048 int constant) {
1049 assert(ProfileInterpreter, "must be profiling interpreter");
1050 add(mdp_in, mdp_in, (unsigned)constant);
1051 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1052 }
1053
1054
1055 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1056 assert(ProfileInterpreter, "must be profiling interpreter");
1057 // save/restore across call_VM
1058 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1059 call_VM(noreg,
1060 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1061 return_bci);
1062 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1063 }
1064
1065
1066 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1067 Register bumped_count) {
1068 if (ProfileInterpreter) {
1069 Label profile_continue;
1070
1071 // If no method data exists, go to profile_continue.
1072 // Otherwise, assign to mdp
1073 test_method_data_pointer(mdp, profile_continue);
1074
1075 // We are taking a branch. Increment the taken count.
1076 // We inline increment_mdp_data_at to return bumped_count in a register
1077 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1078 Address data(mdp, in_bytes(JumpData::taken_offset()));
1079 ldr(bumped_count, data);
1080 assert(DataLayout::counter_increment == 1,
1081 "flow-free idiom only works with 1");
1082 // Intel does this to catch overflow
1083 // addptr(bumped_count, DataLayout::counter_increment);
1084 // sbbptr(bumped_count, 0);
1085 // so we do this
1086 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1087 Label L;
1088 br(Assembler::CS, L); // skip store if counter overflow
1089 str(bumped_count, data);
1090 bind(L);
1091 // The method data pointer needs to be updated to reflect the new target.
1092 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1093 bind(profile_continue);
1094 }
1095 }
1096
1097
1098 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1099 if (ProfileInterpreter) {
1100 Label profile_continue;
1101
1102 // If no method data exists, go to profile_continue.
1103 test_method_data_pointer(mdp, profile_continue);
1104
1105 // We are taking a branch. Increment the not taken count.
1106 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1107
1108 // The method data pointer needs to be updated to correspond to
1109 // the next bytecode
1110 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1111 bind(profile_continue);
1112 }
1113 }
1114
1115
1116 void InterpreterMacroAssembler::profile_call(Register mdp) {
1117 if (ProfileInterpreter) {
1118 Label profile_continue;
1119
1120 // If no method data exists, go to profile_continue.
1121 test_method_data_pointer(mdp, profile_continue);
1122
1123 // We are making a call. Increment the count.
1124 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1125
1126 // The method data pointer needs to be updated to reflect the new target.
1127 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1128 bind(profile_continue);
1129 }
1130 }
1131
1132 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1133 if (ProfileInterpreter) {
1134 Label profile_continue;
1135
1136 // If no method data exists, go to profile_continue.
1137 test_method_data_pointer(mdp, profile_continue);
1138
1139 // We are making a call. Increment the count.
1140 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1141
1142 // The method data pointer needs to be updated to reflect the new target.
1143 update_mdp_by_constant(mdp,
1144 in_bytes(VirtualCallData::
1145 virtual_call_data_size()));
1146 bind(profile_continue);
1147 }
1148 }
1149
1150
1151 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1152 Register mdp,
1153 Register reg2,
1154 bool receiver_can_be_null) {
1155 if (ProfileInterpreter) {
1156 Label profile_continue;
1157
1158 // If no method data exists, go to profile_continue.
1159 test_method_data_pointer(mdp, profile_continue);
1160
1161 Label skip_receiver_profile;
1162 if (receiver_can_be_null) {
1163 Label not_null;
1164 // We are making a call. Increment the count for null receiver.
1165 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1166 b(skip_receiver_profile);
1167 bind(not_null);
1168 }
1169
1170 // Record the receiver type.
1171 record_klass_in_profile(receiver, mdp, reg2, true);
1172 bind(skip_receiver_profile);
1173
1174 // The method data pointer needs to be updated to reflect the new target.
1175 #if INCLUDE_JVMCI
1176 if (MethodProfileWidth == 0) {
1177 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1178 }
1179 #else // INCLUDE_JVMCI
1180 update_mdp_by_constant(mdp,
1181 in_bytes(VirtualCallData::
1182 virtual_call_data_size()));
1183 #endif // INCLUDE_JVMCI
1184 bind(profile_continue);
1185 }
1186 }
1187
1188 #if INCLUDE_JVMCI
1189 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1190 assert_different_registers(method, mdp, reg2);
1191 if (ProfileInterpreter && MethodProfileWidth > 0) {
1192 Label profile_continue;
1193
1194 // If no method data exists, go to profile_continue.
1195 test_method_data_pointer(mdp, profile_continue);
1196
1197 Label done;
1198 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1199 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1200 bind(done);
1201
1202 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1203 bind(profile_continue);
1204 }
1205 }
1206 #endif // INCLUDE_JVMCI
1207
1208 // This routine creates a state machine for updating the multi-row
1209 // type profile at a virtual call site (or other type-sensitive bytecode).
1210 // The machine visits each row (of receiver/count) until the receiver type
1211 // is found, or until it runs out of rows. At the same time, it remembers
1212 // the location of the first empty row. (An empty row records null for its
1213 // receiver, and can be allocated for a newly-observed receiver type.)
1214 // Because there are two degrees of freedom in the state, a simple linear
1215 // search will not work; it must be a decision tree. Hence this helper
1216 // function is recursive, to generate the required tree structured code.
1217 // It's the interpreter, so we are trading off code space for speed.
1218 // See below for example code.
1219 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1220 Register receiver, Register mdp,
1221 Register reg2, int start_row,
1222 Label& done, bool is_virtual_call) {
1223 if (TypeProfileWidth == 0) {
1224 if (is_virtual_call) {
1225 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1226 }
1227 #if INCLUDE_JVMCI
1228 else if (EnableJVMCI) {
1229 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1230 }
1231 #endif // INCLUDE_JVMCI
1232 } else {
1233 int non_profiled_offset = -1;
1234 if (is_virtual_call) {
1235 non_profiled_offset = in_bytes(CounterData::count_offset());
1236 }
1237 #if INCLUDE_JVMCI
1238 else if (EnableJVMCI) {
1239 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1240 }
1241 #endif // INCLUDE_JVMCI
1242
1243 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1244 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1245 }
1246 }
1247
1248 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1249 Register reg2, int start_row, Label& done, int total_rows,
1250 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1251 int non_profiled_offset) {
1252 int last_row = total_rows - 1;
1253 assert(start_row <= last_row, "must be work left to do");
1254 // Test this row for both the item and for null.
1255 // Take any of three different outcomes:
1256 // 1. found item => increment count and goto done
1257 // 2. found null => keep looking for case 1, maybe allocate this cell
1258 // 3. found something else => keep looking for cases 1 and 2
1259 // Case 3 is handled by a recursive call.
1260 for (int row = start_row; row <= last_row; row++) {
1261 Label next_test;
1262 bool test_for_null_also = (row == start_row);
1263
1264 // See if the item is item[n].
1265 int item_offset = in_bytes(item_offset_fn(row));
1266 test_mdp_data_at(mdp, item_offset, item,
1267 (test_for_null_also ? reg2 : noreg),
1268 next_test);
1269 // (Reg2 now contains the item from the CallData.)
1270
1271 // The item is item[n]. Increment count[n].
1272 int count_offset = in_bytes(item_count_offset_fn(row));
1273 increment_mdp_data_at(mdp, count_offset);
1274 b(done);
1275 bind(next_test);
1276
1277 if (test_for_null_also) {
1278 Label found_null;
1279 // Failed the equality check on item[n]... Test for null.
1280 if (start_row == last_row) {
1281 // The only thing left to do is handle the null case.
1282 if (non_profiled_offset >= 0) {
1283 cbz(reg2, found_null);
1284 // Item did not match any saved item and there is no empty row for it.
1285 // Increment total counter to indicate polymorphic case.
1286 increment_mdp_data_at(mdp, non_profiled_offset);
1287 b(done);
1288 bind(found_null);
1289 } else {
1290 cbnz(reg2, done);
1291 }
1292 break;
1293 }
1294 // Since null is rare, make it be the branch-taken case.
1295 cbz(reg2, found_null);
1296
1297 // Put all the "Case 3" tests here.
1298 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1299 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1300
1301 // Found a null. Keep searching for a matching item,
1302 // but remember that this is an empty (unused) slot.
1303 bind(found_null);
1304 }
1305 }
1306
1307 // In the fall-through case, we found no matching item, but we
1308 // observed the item[start_row] is NULL.
1309
1310 // Fill in the item field and increment the count.
1311 int item_offset = in_bytes(item_offset_fn(start_row));
1312 set_mdp_data_at(mdp, item_offset, item);
1313 int count_offset = in_bytes(item_count_offset_fn(start_row));
1314 mov(reg2, DataLayout::counter_increment);
1315 set_mdp_data_at(mdp, count_offset, reg2);
1316 if (start_row > 0) {
1317 b(done);
1318 }
1319 }
1320
1321 // Example state machine code for three profile rows:
1322 // // main copy of decision tree, rooted at row[1]
1323 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1324 // if (row[0].rec != NULL) {
1325 // // inner copy of decision tree, rooted at row[1]
1326 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1327 // if (row[1].rec != NULL) {
1328 // // degenerate decision tree, rooted at row[2]
1329 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1330 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1331 // row[2].init(rec); goto done;
1332 // } else {
1333 // // remember row[1] is empty
1334 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1335 // row[1].init(rec); goto done;
1336 // }
1337 // } else {
1338 // // remember row[0] is empty
1339 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1340 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1341 // row[0].init(rec); goto done;
1342 // }
1343 // done:
1344
1345 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1346 Register mdp, Register reg2,
1347 bool is_virtual_call) {
1348 assert(ProfileInterpreter, "must be profiling");
1349 Label done;
1350
1351 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1352
1353 bind (done);
1354 }
1355
1356 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1357 Register mdp) {
1358 if (ProfileInterpreter) {
1359 Label profile_continue;
1360 uint row;
1361
1362 // If no method data exists, go to profile_continue.
1363 test_method_data_pointer(mdp, profile_continue);
1364
1365 // Update the total ret count.
1366 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1367
1368 for (row = 0; row < RetData::row_limit(); row++) {
1369 Label next_test;
1370
1371 // See if return_bci is equal to bci[n]:
1372 test_mdp_data_at(mdp,
1373 in_bytes(RetData::bci_offset(row)),
1374 return_bci, noreg,
1375 next_test);
1376
1377 // return_bci is equal to bci[n]. Increment the count.
1378 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1379
1380 // The method data pointer needs to be updated to reflect the new target.
1381 update_mdp_by_offset(mdp,
1382 in_bytes(RetData::bci_displacement_offset(row)));
1383 b(profile_continue);
1384 bind(next_test);
1385 }
1386
1387 update_mdp_for_ret(return_bci);
1388
1389 bind(profile_continue);
1390 }
1391 }
1392
1393 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1394 if (ProfileInterpreter) {
1395 Label profile_continue;
1396
1397 // If no method data exists, go to profile_continue.
1398 test_method_data_pointer(mdp, profile_continue);
1399
1400 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1401
1402 // The method data pointer needs to be updated.
1403 int mdp_delta = in_bytes(BitData::bit_data_size());
1404 if (TypeProfileCasts) {
1405 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1406 }
1407 update_mdp_by_constant(mdp, mdp_delta);
1408
1409 bind(profile_continue);
1410 }
1411 }
1412
1413 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1414 if (ProfileInterpreter && TypeProfileCasts) {
1415 Label profile_continue;
1416
1417 // If no method data exists, go to profile_continue.
1418 test_method_data_pointer(mdp, profile_continue);
1419
1420 int count_offset = in_bytes(CounterData::count_offset());
1421 // Back up the address, since we have already bumped the mdp.
1422 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1423
1424 // *Decrement* the counter. We expect to see zero or small negatives.
1425 increment_mdp_data_at(mdp, count_offset, true);
1426
1427 bind (profile_continue);
1428 }
1429 }
1430
1431 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1432 if (ProfileInterpreter) {
1433 Label profile_continue;
1434
1435 // If no method data exists, go to profile_continue.
1436 test_method_data_pointer(mdp, profile_continue);
1437
1438 // The method data pointer needs to be updated.
1439 int mdp_delta = in_bytes(BitData::bit_data_size());
1440 if (TypeProfileCasts) {
1441 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1442
1443 // Record the object type.
1444 record_klass_in_profile(klass, mdp, reg2, false);
1445 }
1446 update_mdp_by_constant(mdp, mdp_delta);
1447
1448 bind(profile_continue);
1449 }
1450 }
1451
1452 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1453 if (ProfileInterpreter) {
1454 Label profile_continue;
1455
1456 // If no method data exists, go to profile_continue.
1457 test_method_data_pointer(mdp, profile_continue);
1458
1459 // Update the default case count
1460 increment_mdp_data_at(mdp,
1461 in_bytes(MultiBranchData::default_count_offset()));
1462
1463 // The method data pointer needs to be updated.
1464 update_mdp_by_offset(mdp,
1465 in_bytes(MultiBranchData::
1466 default_displacement_offset()));
1467
1468 bind(profile_continue);
1469 }
1470 }
1471
1472 void InterpreterMacroAssembler::profile_switch_case(Register index,
1473 Register mdp,
1474 Register reg2) {
1475 if (ProfileInterpreter) {
1476 Label profile_continue;
1477
1478 // If no method data exists, go to profile_continue.
1479 test_method_data_pointer(mdp, profile_continue);
1480
1481 // Build the base (index * per_case_size_in_bytes()) +
1482 // case_array_offset_in_bytes()
1483 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1484 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1485 Assembler::maddw(index, index, reg2, rscratch1);
1486
1487 // Update the case count
1488 increment_mdp_data_at(mdp,
1489 index,
1490 in_bytes(MultiBranchData::relative_count_offset()));
1491
1492 // The method data pointer needs to be updated.
1493 update_mdp_by_offset(mdp,
1494 index,
1495 in_bytes(MultiBranchData::
1496 relative_displacement_offset()));
1497
1498 bind(profile_continue);
1499 }
1500 }
1501
1502 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1503 if (state == atos) {
1504 MacroAssembler::verify_oop(reg);
1505 }
1506 }
1507
1508 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1509
1510
1511 void InterpreterMacroAssembler::notify_method_entry() {
1512 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1513 // track stack depth. If it is possible to enter interp_only_mode we add
1514 // the code to check if the event should be sent.
1515 if (JvmtiExport::can_post_interpreter_events()) {
1516 Label L;
1517 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1518 cbzw(r3, L);
1519 call_VM(noreg, CAST_FROM_FN_PTR(address,
1520 InterpreterRuntime::post_method_entry));
1521 bind(L);
1522 }
1523
1524 {
1525 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1526 get_method(c_rarg1);
1527 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1528 rthread, c_rarg1);
1529 }
1530
1531 // RedefineClasses() tracing support for obsolete method entry
1532 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1533 get_method(c_rarg1);
1534 call_VM_leaf(
1535 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1536 rthread, c_rarg1);
1537 }
1538
1539 }
1540
1541
1542 void InterpreterMacroAssembler::notify_method_exit(
1543 TosState state, NotifyMethodExitMode mode) {
1544 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1545 // track stack depth. If it is possible to enter interp_only_mode we add
1546 // the code to check if the event should be sent.
1547 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1548 Label L;
1549 // Note: frame::interpreter_frame_result has a dependency on how the
1550 // method result is saved across the call to post_method_exit. If this
1551 // is changed then the interpreter_frame_result implementation will
1552 // need to be updated too.
1553
1554 // template interpreter will leave the result on the top of the stack.
1555 push(state);
1556 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1557 cbz(r3, L);
1558 call_VM(noreg,
1559 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1560 bind(L);
1561 pop(state);
1562 }
1563
1564 {
1565 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1566 push(state);
1567 get_method(c_rarg1);
1568 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1569 rthread, c_rarg1);
1570 pop(state);
1571 }
1572 }
1573
1574
1575 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1576 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1577 int increment, Address mask,
1578 Register scratch, Register scratch2,
1579 bool preloaded, Condition cond,
1580 Label* where) {
1581 if (!preloaded) {
1582 ldrw(scratch, counter_addr);
1583 }
1584 add(scratch, scratch, increment);
1585 strw(scratch, counter_addr);
1586 ldrw(scratch2, mask);
1587 ands(scratch, scratch, scratch2);
1588 br(cond, *where);
1589 }
1590
1591 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1592 int number_of_arguments) {
1593 // interpreter specific
1594 //
1595 // Note: No need to save/restore rbcp & rlocals pointer since these
1596 // are callee saved registers and no blocking/ GC can happen
1597 // in leaf calls.
1598 #ifdef ASSERT
1599 {
1600 Label L;
1601 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1602 cbz(rscratch1, L);
1603 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1604 " last_sp != NULL");
1605 bind(L);
1606 }
1607 #endif /* ASSERT */
1608 // super call
1609 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1610 }
1611
1612 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1613 Register java_thread,
1614 Register last_java_sp,
1615 address entry_point,
1616 int number_of_arguments,
1617 bool check_exceptions) {
1618 // interpreter specific
1619 //
1620 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1621 // really make a difference for these runtime calls, since they are
1622 // slow anyway. Btw., bcp must be saved/restored since it may change
1623 // due to GC.
1624 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1625 save_bcp();
1626 #ifdef ASSERT
1627 {
1628 Label L;
1629 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1630 cbz(rscratch1, L);
1631 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1632 " last_sp != NULL");
1633 bind(L);
1634 }
1635 #endif /* ASSERT */
1636 // super call
1637 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1638 entry_point, number_of_arguments,
1639 check_exceptions);
1640 // interpreter specific
1641 restore_bcp();
1642 restore_locals();
1643 }
1644
1645 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1646 assert_different_registers(obj, rscratch1);
1647 Label update, next, none;
1648
1649 verify_oop(obj);
1650
1651 cbnz(obj, update);
1652 orptr(mdo_addr, TypeEntries::null_seen);
1653 b(next);
1654
1655 bind(update);
1656 load_klass(obj, obj);
1657
1658 ldr(rscratch1, mdo_addr);
1659 eor(obj, obj, rscratch1);
1660 tst(obj, TypeEntries::type_klass_mask);
1661 br(Assembler::EQ, next); // klass seen before, nothing to
1662 // do. The unknown bit may have been
1663 // set already but no need to check.
1664
1665 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1666 // already unknown. Nothing to do anymore.
1667
1668 ldr(rscratch1, mdo_addr);
1669 cbz(rscratch1, none);
1670 cmp(rscratch1, (u1)TypeEntries::null_seen);
1671 br(Assembler::EQ, none);
1672 // There is a chance that the checks above (re-reading profiling
1673 // data from memory) fail if another thread has just set the
1674 // profiling to this obj's klass
1675 ldr(rscratch1, mdo_addr);
1676 eor(obj, obj, rscratch1);
1677 tst(obj, TypeEntries::type_klass_mask);
1678 br(Assembler::EQ, next);
1679
1680 // different than before. Cannot keep accurate profile.
1681 orptr(mdo_addr, TypeEntries::type_unknown);
1682 b(next);
1683
1684 bind(none);
1685 // first time here. Set profile type.
1686 str(obj, mdo_addr);
1687
1688 bind(next);
1689 }
1690
1691 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1692 if (!ProfileInterpreter) {
1693 return;
1694 }
1695
1696 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1697 Label profile_continue;
1698
1699 test_method_data_pointer(mdp, profile_continue);
1700
1701 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1702
1703 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1704 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1705 br(Assembler::NE, profile_continue);
1706
1707 if (MethodData::profile_arguments()) {
1708 Label done;
1709 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1710
1711 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1712 if (i > 0 || MethodData::profile_return()) {
1713 // If return value type is profiled we may have no argument to profile
1714 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1715 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1716 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1717 add(rscratch1, mdp, off_to_args);
1718 br(Assembler::LT, done);
1719 }
1720 ldr(tmp, Address(callee, Method::const_offset()));
1721 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1722 // stack offset o (zero based) from the start of the argument
1723 // list, for n arguments translates into offset n - o - 1 from
1724 // the end of the argument list
1725 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1726 sub(tmp, tmp, rscratch1);
1727 sub(tmp, tmp, 1);
1728 Address arg_addr = argument_address(tmp);
1729 ldr(tmp, arg_addr);
1730
1731 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1732 profile_obj_type(tmp, mdo_arg_addr);
1733
1734 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1735 off_to_args += to_add;
1736 }
1737
1738 if (MethodData::profile_return()) {
1739 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1740 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1741 }
1742
1743 add(rscratch1, mdp, off_to_args);
1744 bind(done);
1745 mov(mdp, rscratch1);
1746
1747 if (MethodData::profile_return()) {
1748 // We're right after the type profile for the last
1749 // argument. tmp is the number of cells left in the
1750 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1751 // if there's a return to profile.
1752 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1753 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1754 }
1755 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1756 } else {
1757 assert(MethodData::profile_return(), "either profile call args or call ret");
1758 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1759 }
1760
1761 // mdp points right after the end of the
1762 // CallTypeData/VirtualCallTypeData, right after the cells for the
1763 // return value type if there's one
1764
1765 bind(profile_continue);
1766 }
1767 }
1768
1769 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1770 assert_different_registers(mdp, ret, tmp, rbcp);
1771 if (ProfileInterpreter && MethodData::profile_return()) {
1772 Label profile_continue, done;
1773
1774 test_method_data_pointer(mdp, profile_continue);
1775
1776 if (MethodData::profile_return_jsr292_only()) {
1777 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1778
1779 // If we don't profile all invoke bytecodes we must make sure
1780 // it's a bytecode we indeed profile. We can't go back to the
1781 // begining of the ProfileData we intend to update to check its
1782 // type because we're right after it and we don't known its
1783 // length
1784 Label do_profile;
1785 ldrb(rscratch1, Address(rbcp, 0));
1786 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1787 br(Assembler::EQ, do_profile);
1788 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1789 br(Assembler::EQ, do_profile);
1790 get_method(tmp);
1791 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1792 subs(zr, rscratch1, vmIntrinsics::_compiledLambdaForm);
1793 br(Assembler::NE, profile_continue);
1794
1795 bind(do_profile);
1796 }
1797
1798 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1799 mov(tmp, ret);
1800 profile_obj_type(tmp, mdo_ret_addr);
1801
1802 bind(profile_continue);
1803 }
1804 }
1805
1806 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1807 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1808 if (ProfileInterpreter && MethodData::profile_parameters()) {
1809 Label profile_continue, done;
1810
1811 test_method_data_pointer(mdp, profile_continue);
1812
1813 // Load the offset of the area within the MDO used for
1814 // parameters. If it's negative we're not profiling any parameters
1815 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1816 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1817
1818 // Compute a pointer to the area for parameters from the offset
1819 // and move the pointer to the slot for the last
1820 // parameters. Collect profiling from last parameter down.
1821 // mdo start + parameters offset + array length - 1
1822 add(mdp, mdp, tmp1);
1823 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1824 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1825
1826 Label loop;
1827 bind(loop);
1828
1829 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1830 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1831 int per_arg_scale = exact_log2(DataLayout::cell_size);
1832 add(rscratch1, mdp, off_base);
1833 add(rscratch2, mdp, type_base);
1834
1835 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1836 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1837
1838 // load offset on the stack from the slot for this parameter
1839 ldr(tmp2, arg_off);
1840 neg(tmp2, tmp2);
1841 // read the parameter from the local area
1842 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1843
1844 // profile the parameter
1845 profile_obj_type(tmp2, arg_type);
1846
1847 // go to next parameter
1848 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1849 br(Assembler::GE, loop);
1850
1851 bind(profile_continue);
1852 }
1853 }