1 /*
2 * Copyright (c) 2003, 2017, 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 "interp_masm_aarch64.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "logging/log.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "prims/jvmtiExport.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/biasedLocking.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/thread.inline.hpp"
41
42
43 void InterpreterMacroAssembler::narrow(Register result) {
44
45 // Get method->_constMethod->_result_type
46 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
47 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
48 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
49
50 Label done, notBool, notByte, notChar;
51
52 // common case first
53 cmpw(rscratch1, T_INT);
54 br(Assembler::EQ, done);
55
56 // mask integer result to narrower return type.
57 cmpw(rscratch1, T_BOOLEAN);
58 br(Assembler::NE, notBool);
59 andw(result, result, 0x1);
60 b(done);
61
62 bind(notBool);
63 cmpw(rscratch1, T_BYTE);
64 br(Assembler::NE, notByte);
65 sbfx(result, result, 0, 8);
66 b(done);
67
68 bind(notByte);
69 cmpw(rscratch1, T_CHAR);
70 br(Assembler::NE, notChar);
71 ubfx(result, result, 0, 16); // truncate upper 16 bits
72 b(done);
73
74 bind(notChar);
75 sbfx(result, result, 0, 16); // sign-extend short
76
77 // Nothing to do for T_INT
78 bind(done);
79 }
80
81 void InterpreterMacroAssembler::jump_to_entry(address entry) {
82 assert(entry, "Entry must have been generated by now");
83 b(entry);
84 }
85
86 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
87 if (JvmtiExport::can_pop_frame()) {
88 Label L;
89 // Initiate popframe handling only if it is not already being
90 // processed. If the flag has the popframe_processing bit set, it
91 // means that this code is called *during* popframe handling - we
92 // don't want to reenter.
93 // This method is only called just after the call into the vm in
94 // call_VM_base, so the arg registers are available.
95 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
96 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
97 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
98 // Call Interpreter::remove_activation_preserving_args_entry() to get the
99 // address of the same-named entrypoint in the generated interpreter code.
100 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
101 br(r0);
102 bind(L);
103 }
104 }
105
106
107 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
108 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
109 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
110 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
111 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
112 switch (state) {
113 case atos: ldr(r0, oop_addr);
114 str(zr, oop_addr);
115 verify_oop(r0, state); break;
116 case ltos: ldr(r0, val_addr); break;
117 case btos: // fall through
118 case ztos: // fall through
119 case ctos: // fall through
120 case stos: // fall through
121 case itos: ldrw(r0, val_addr); break;
122 case ftos: ldrs(v0, val_addr); break;
123 case dtos: ldrd(v0, val_addr); break;
124 case vtos: /* nothing to do */ break;
125 default : ShouldNotReachHere();
126 }
127 // Clean up tos value in the thread object
128 movw(rscratch1, (int) ilgl);
129 strw(rscratch1, tos_addr);
130 strw(zr, val_addr);
131 }
132
133
134 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
135 if (JvmtiExport::can_force_early_return()) {
136 Label L;
137 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
138 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == NULL) exit;
139
140 // Initiate earlyret handling only if it is not already being processed.
141 // If the flag has the earlyret_processing bit set, it means that this code
142 // is called *during* earlyret handling - we don't want to reenter.
143 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
144 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
145 br(Assembler::NE, L);
146
147 // Call Interpreter::remove_activation_early_entry() to get the address of the
148 // same-named entrypoint in the generated interpreter code.
149 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
150 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
151 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
152 br(r0);
153 bind(L);
154 }
155 }
156
157 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
158 Register reg,
159 int bcp_offset) {
160 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
161 ldrh(reg, Address(rbcp, bcp_offset));
162 rev16(reg, reg);
163 }
164
165 void InterpreterMacroAssembler::get_dispatch() {
166 unsigned long offset;
167 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
168 lea(rdispatch, Address(rdispatch, offset));
169 }
170
171 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
172 int bcp_offset,
173 size_t index_size) {
174 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
175 if (index_size == sizeof(u2)) {
176 load_unsigned_short(index, Address(rbcp, bcp_offset));
177 } else if (index_size == sizeof(u4)) {
178 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
179 ldrw(index, Address(rbcp, bcp_offset));
180 // Check if the secondary index definition is still ~x, otherwise
181 // we have to change the following assembler code to calculate the
182 // plain index.
183 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
184 eonw(index, index, zr); // convert to plain index
185 } else if (index_size == sizeof(u1)) {
186 load_unsigned_byte(index, Address(rbcp, bcp_offset));
187 } else {
188 ShouldNotReachHere();
189 }
190 }
191
192 // Return
193 // Rindex: index into constant pool
194 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
195 //
196 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
197 // the true address of the cache entry.
198 //
199 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
200 Register index,
201 int bcp_offset,
202 size_t index_size) {
203 assert_different_registers(cache, index);
204 assert_different_registers(cache, rcpool);
205 get_cache_index_at_bcp(index, bcp_offset, index_size);
206 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
207 // convert from field index to ConstantPoolCacheEntry
208 // aarch64 already has the cache in rcpool so there is no need to
209 // install it in cache. instead we pre-add the indexed offset to
210 // rcpool and return it in cache. All clients of this method need to
211 // be modified accordingly.
212 add(cache, rcpool, index, Assembler::LSL, 5);
213 }
214
215
216 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
217 Register index,
218 Register bytecode,
219 int byte_no,
220 int bcp_offset,
221 size_t index_size) {
222 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
223 // We use a 32-bit load here since the layout of 64-bit words on
224 // little-endian machines allow us that.
225 // n.b. unlike x86 cache already includes the index offset
226 lea(bytecode, Address(cache,
227 ConstantPoolCache::base_offset()
228 + ConstantPoolCacheEntry::indices_offset()));
229 ldarw(bytecode, bytecode);
230 const int shift_count = (1 + byte_no) * BitsPerByte;
231 ubfx(bytecode, bytecode, shift_count, BitsPerByte);
232 }
233
234 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
235 Register tmp,
236 int bcp_offset,
237 size_t index_size) {
238 assert(cache != tmp, "must use different register");
239 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
240 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
241 // convert from field index to ConstantPoolCacheEntry index
242 // and from word offset to byte offset
243 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
244 ldr(cache, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
245 // skip past the header
246 add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
247 add(cache, cache, tmp, Assembler::LSL, 2 + LogBytesPerWord); // construct pointer to cache entry
248 }
249
250 void InterpreterMacroAssembler::get_method_counters(Register method,
251 Register mcs, Label& skip) {
252 Label has_counters;
253 ldr(mcs, Address(method, Method::method_counters_offset()));
254 cbnz(mcs, has_counters);
255 call_VM(noreg, CAST_FROM_FN_PTR(address,
256 InterpreterRuntime::build_method_counters), method);
257 ldr(mcs, Address(method, Method::method_counters_offset()));
258 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
259 bind(has_counters);
260 }
261
262 // Load object from cpool->resolved_references(index)
263 void InterpreterMacroAssembler::load_resolved_reference_at_index(
264 Register result, Register index) {
265 assert_different_registers(result, index);
266 // convert from field index to resolved_references() index and from
267 // word index to byte offset. Since this is a java object, it can be compressed
268 Register tmp = index; // reuse
269 lslw(tmp, tmp, LogBytesPerHeapOop);
270
271 get_constant_pool(result);
272 // load pointer for resolved_references[] objArray
273 ldr(result, Address(result, ConstantPool::cache_offset_in_bytes()));
274 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
275 // JNIHandles::resolve(obj);
276 ldr(result, Address(result, 0));
277 // Add in the index
278 add(result, result, tmp);
279 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
280 }
281
282 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
283 Register cpool, Register index, Register klass, Register temp) {
284 add(temp, cpool, index, LSL, LogBytesPerWord);
285 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
286 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
287 add(klass, klass, temp, LSL, LogBytesPerWord);
288 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
289 }
290
291 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
292 // subtype of super_klass.
293 //
294 // Args:
295 // r0: superklass
296 // Rsub_klass: subklass
297 //
298 // Kills:
299 // r2, r5
300 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
301 Label& ok_is_subtype) {
302 assert(Rsub_klass != r0, "r0 holds superklass");
303 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
304 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
305
306 // Profile the not-null value's klass.
307 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
308
309 // Do the check.
310 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
311
312 // Profile the failure of the check.
313 profile_typecheck_failed(r2); // blows r2
314 }
315
316 // Java Expression Stack
317
318 void InterpreterMacroAssembler::pop_ptr(Register r) {
319 ldr(r, post(esp, wordSize));
320 }
321
322 void InterpreterMacroAssembler::pop_i(Register r) {
323 ldrw(r, post(esp, wordSize));
324 }
325
326 void InterpreterMacroAssembler::pop_l(Register r) {
327 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
328 }
329
330 void InterpreterMacroAssembler::push_ptr(Register r) {
331 str(r, pre(esp, -wordSize));
332 }
333
334 void InterpreterMacroAssembler::push_i(Register r) {
335 str(r, pre(esp, -wordSize));
336 }
337
338 void InterpreterMacroAssembler::push_l(Register r) {
339 str(zr, pre(esp, -wordSize));
340 str(r, pre(esp, - wordSize));
341 }
342
343 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
344 ldrs(r, post(esp, wordSize));
345 }
346
347 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
348 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
349 }
350
351 void InterpreterMacroAssembler::push_f(FloatRegister r) {
352 strs(r, pre(esp, -wordSize));
353 }
354
355 void InterpreterMacroAssembler::push_d(FloatRegister r) {
356 strd(r, pre(esp, 2* -wordSize));
357 }
358
359 void InterpreterMacroAssembler::pop(TosState state) {
360 switch (state) {
361 case atos: pop_ptr(); break;
362 case btos:
363 case ztos:
364 case ctos:
365 case stos:
366 case itos: pop_i(); break;
367 case ltos: pop_l(); break;
368 case ftos: pop_f(); break;
369 case dtos: pop_d(); break;
370 case vtos: /* nothing to do */ break;
371 default: ShouldNotReachHere();
372 }
373 verify_oop(r0, state);
374 }
375
376 void InterpreterMacroAssembler::push(TosState state) {
377 verify_oop(r0, state);
378 switch (state) {
379 case atos: push_ptr(); break;
380 case btos:
381 case ztos:
382 case ctos:
383 case stos:
384 case itos: push_i(); break;
385 case ltos: push_l(); break;
386 case ftos: push_f(); break;
387 case dtos: push_d(); break;
388 case vtos: /* nothing to do */ break;
389 default : ShouldNotReachHere();
390 }
391 }
392
393 // Helpers for swap and dup
394 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
395 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
396 }
397
398 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
399 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
400 }
401
402
403 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
404 // set sender sp
405 mov(r13, sp);
406 // record last_sp
407 str(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
408 }
409
410 // Jump to from_interpreted entry of a call unless single stepping is possible
411 // in this thread in which case we must call the i2i entry
412 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
413 prepare_to_jump_from_interpreted();
414
415 if (JvmtiExport::can_post_interpreter_events()) {
416 Label run_compiled_code;
417 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
418 // compiled code in threads for which the event is enabled. Check here for
419 // interp_only_mode if these events CAN be enabled.
420 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
421 cbzw(rscratch1, run_compiled_code);
422 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
423 br(rscratch1);
424 bind(run_compiled_code);
425 }
426
427 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
428 br(rscratch1);
429 }
430
431 // The following two routines provide a hook so that an implementation
432 // can schedule the dispatch in two parts. amd64 does not do this.
433 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
434 }
435
436 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
437 dispatch_next(state, step);
438 }
439
440 void InterpreterMacroAssembler::dispatch_base(TosState state,
441 address* table,
442 bool verifyoop) {
443 if (VerifyActivationFrameSize) {
444 Unimplemented();
445 }
446 if (verifyoop) {
447 verify_oop(r0, state);
448 }
449 if (table == Interpreter::dispatch_table(state)) {
450 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
451 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
452 } else {
453 mov(rscratch2, (address)table);
454 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
455 }
456 br(rscratch2);
457 }
458
459 void InterpreterMacroAssembler::dispatch_only(TosState state) {
460 dispatch_base(state, Interpreter::dispatch_table(state));
461 }
462
463 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
464 dispatch_base(state, Interpreter::normal_table(state));
465 }
466
467 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
468 dispatch_base(state, Interpreter::normal_table(state), false);
469 }
470
471
472 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
473 // load next bytecode
474 ldrb(rscratch1, Address(pre(rbcp, step)));
475 dispatch_base(state, Interpreter::dispatch_table(state));
476 }
477
478 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
479 // load current bytecode
480 ldrb(rscratch1, Address(rbcp, 0));
481 dispatch_base(state, table);
482 }
483
484 // remove activation
485 //
486 // Unlock the receiver if this is a synchronized method.
487 // Unlock any Java monitors from syncronized blocks.
488 // Remove the activation from the stack.
489 //
490 // If there are locked Java monitors
491 // If throw_monitor_exception
492 // throws IllegalMonitorStateException
493 // Else if install_monitor_exception
494 // installs IllegalMonitorStateException
495 // Else
496 // no error processing
497 void InterpreterMacroAssembler::remove_activation(
498 TosState state,
499 bool throw_monitor_exception,
500 bool install_monitor_exception,
501 bool notify_jvmdi) {
502 // Note: Registers r3 xmm0 may be in use for the
503 // result check if synchronized method
504 Label unlocked, unlock, no_unlock;
505
506 // get the value of _do_not_unlock_if_synchronized into r3
507 const Address do_not_unlock_if_synchronized(rthread,
508 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
509 ldrb(r3, do_not_unlock_if_synchronized);
510 strb(zr, do_not_unlock_if_synchronized); // reset the flag
511
512 // get method access flags
513 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
514 ldr(r2, Address(r1, Method::access_flags_offset()));
515 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
516
517 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
518 // is set.
519 cbnz(r3, no_unlock);
520
521 // unlock monitor
522 push(state); // save result
523
524 // BasicObjectLock will be first in list, since this is a
525 // synchronized method. However, need to check that the object has
526 // not been unlocked by an explicit monitorexit bytecode.
527 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
528 wordSize - (int) sizeof(BasicObjectLock));
529 // We use c_rarg1 so that if we go slow path it will be the correct
530 // register for unlock_object to pass to VM directly
531 lea(c_rarg1, monitor); // address of first monitor
532
533 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
534 cbnz(r0, unlock);
535
536 pop(state);
537 if (throw_monitor_exception) {
538 // Entry already unlocked, need to throw exception
539 call_VM(noreg, CAST_FROM_FN_PTR(address,
540 InterpreterRuntime::throw_illegal_monitor_state_exception));
541 should_not_reach_here();
542 } else {
543 // Monitor already unlocked during a stack unroll. If requested,
544 // install an illegal_monitor_state_exception. Continue with
545 // stack unrolling.
546 if (install_monitor_exception) {
547 call_VM(noreg, CAST_FROM_FN_PTR(address,
548 InterpreterRuntime::new_illegal_monitor_state_exception));
549 }
550 b(unlocked);
551 }
552
553 bind(unlock);
554 unlock_object(c_rarg1);
555 pop(state);
556
557 // Check that for block-structured locking (i.e., that all locked
558 // objects has been unlocked)
559 bind(unlocked);
560
561 // r0: Might contain return value
562
563 // Check that all monitors are unlocked
564 {
565 Label loop, exception, entry, restart;
566 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
567 const Address monitor_block_top(
568 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
569 const Address monitor_block_bot(
570 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
571
572 bind(restart);
573 // We use c_rarg1 so that if we go slow path it will be the correct
574 // register for unlock_object to pass to VM directly
575 ldr(c_rarg1, monitor_block_top); // points to current entry, starting
576 // with top-most entry
577 lea(r19, monitor_block_bot); // points to word before bottom of
578 // monitor block
579 b(entry);
580
581 // Entry already locked, need to throw exception
582 bind(exception);
583
584 if (throw_monitor_exception) {
585 // Throw exception
586 MacroAssembler::call_VM(noreg,
587 CAST_FROM_FN_PTR(address, InterpreterRuntime::
588 throw_illegal_monitor_state_exception));
589 should_not_reach_here();
590 } else {
591 // Stack unrolling. Unlock object and install illegal_monitor_exception.
592 // Unlock does not block, so don't have to worry about the frame.
593 // We don't have to preserve c_rarg1 since we are going to throw an exception.
594
595 push(state);
596 unlock_object(c_rarg1);
597 pop(state);
598
599 if (install_monitor_exception) {
600 call_VM(noreg, CAST_FROM_FN_PTR(address,
601 InterpreterRuntime::
602 new_illegal_monitor_state_exception));
603 }
604
605 b(restart);
606 }
607
608 bind(loop);
609 // check if current entry is used
610 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
611 cbnz(rscratch1, exception);
612
613 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
614 bind(entry);
615 cmp(c_rarg1, r19); // check if bottom reached
616 br(Assembler::NE, loop); // if not at bottom then check this entry
617 }
618
619 bind(no_unlock);
620
621 // jvmti support
622 if (notify_jvmdi) {
623 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
624 } else {
625 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
626 }
627
628 // remove activation
629 // get sender esp
630 ldr(esp,
631 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
632 if (StackReservedPages > 0) {
633 // testing if reserved zone needs to be re-enabled
634 Label no_reserved_zone_enabling;
635
636 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
637 cmp(esp, rscratch1);
638 br(Assembler::LS, no_reserved_zone_enabling);
639
640 call_VM_leaf(
641 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
642 call_VM(noreg, CAST_FROM_FN_PTR(address,
643 InterpreterRuntime::throw_delayed_StackOverflowError));
644 should_not_reach_here();
645
646 bind(no_reserved_zone_enabling);
647 }
648 // remove frame anchor
649 leave();
650 // If we're returning to interpreted code we will shortly be
651 // adjusting SP to allow some space for ESP. If we're returning to
652 // compiled code the saved sender SP was saved in sender_sp, so this
653 // restores it.
654 andr(sp, esp, -16);
655 }
656
657 // Lock object
658 //
659 // Args:
660 // c_rarg1: BasicObjectLock to be used for locking
661 //
662 // Kills:
663 // r0
664 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
665 // rscratch1, rscratch2 (scratch regs)
666 void InterpreterMacroAssembler::lock_object(Register lock_reg)
667 {
668 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
669 if (UseHeavyMonitors) {
670 call_VM(noreg,
671 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
672 lock_reg);
673 } else {
674 Label done;
675
676 const Register swap_reg = r0;
677 const Register tmp = c_rarg2;
678 const Register obj_reg = c_rarg3; // Will contain the oop
679
680 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
681 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
682 const int mark_offset = lock_offset +
683 BasicLock::displaced_header_offset_in_bytes();
684
685 Label slow_case;
686
687 // Load object pointer into obj_reg %c_rarg3
688 ldr(obj_reg, Address(lock_reg, obj_offset));
689
690 if (UseBiasedLocking) {
691 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
692 }
693
694 // Load (object->mark() | 1) into swap_reg
695 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
696 orr(swap_reg, rscratch1, 1);
697
698 // Save (object->mark() | 1) into BasicLock's displaced header
699 str(swap_reg, Address(lock_reg, mark_offset));
700
701 assert(lock_offset == 0,
702 "displached header must be first word in BasicObjectLock");
703
704 Label fail;
705 if (PrintBiasedLockingStatistics) {
706 Label fast;
707 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, fast, &fail);
708 bind(fast);
709 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
710 rscratch2, rscratch1, tmp);
711 b(done);
712 bind(fail);
713 } else {
714 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
715 }
716
717 // Test if the oopMark is an obvious stack pointer, i.e.,
718 // 1) (mark & 7) == 0, and
719 // 2) rsp <= mark < mark + os::pagesize()
720 //
721 // These 3 tests can be done by evaluating the following
722 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
723 // assuming both stack pointer and pagesize have their
724 // least significant 3 bits clear.
725 // NOTE: the oopMark is in swap_reg %r0 as the result of cmpxchg
726 // NOTE2: aarch64 does not like to subtract sp from rn so take a
727 // copy
728 mov(rscratch1, sp);
729 sub(swap_reg, swap_reg, rscratch1);
730 ands(swap_reg, swap_reg, (unsigned long)(7 - os::vm_page_size()));
731
732 // Save the test result, for recursive case, the result is zero
733 str(swap_reg, Address(lock_reg, mark_offset));
734
735 if (PrintBiasedLockingStatistics) {
736 br(Assembler::NE, slow_case);
737 atomic_incw(Address((address)BiasedLocking::fast_path_entry_count_addr()),
738 rscratch2, rscratch1, tmp);
739 }
740 br(Assembler::EQ, done);
741
742 bind(slow_case);
743
744 // Call the runtime routine for slow case
745 call_VM(noreg,
746 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
747 lock_reg);
748
749 bind(done);
750 }
751 }
752
753
754 // Unlocks an object. Used in monitorexit bytecode and
755 // remove_activation. Throws an IllegalMonitorException if object is
756 // not locked by current thread.
757 //
758 // Args:
759 // c_rarg1: BasicObjectLock for lock
760 //
761 // Kills:
762 // r0
763 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
764 // rscratch1, rscratch2 (scratch regs)
765 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
766 {
767 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
768
769 if (UseHeavyMonitors) {
770 call_VM(noreg,
771 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
772 lock_reg);
773 } else {
774 Label done;
775
776 const Register swap_reg = r0;
777 const Register header_reg = c_rarg2; // Will contain the old oopMark
778 const Register obj_reg = c_rarg3; // Will contain the oop
779
780 save_bcp(); // Save in case of exception
781
782 // Convert from BasicObjectLock structure to object and BasicLock
783 // structure Store the BasicLock address into %r0
784 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
785
786 // Load oop into obj_reg(%c_rarg3)
787 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
788
789 // Free entry
790 str(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
791
792 if (UseBiasedLocking) {
793 biased_locking_exit(obj_reg, header_reg, done);
794 }
795
796 // Load the old header from BasicLock structure
797 ldr(header_reg, Address(swap_reg,
798 BasicLock::displaced_header_offset_in_bytes()));
799
800 // Test for recursion
801 cbz(header_reg, done);
802
803 // Atomic swap back the old header
804 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, done, /*fallthrough*/NULL);
805
806 // Call the runtime routine for slow case.
807 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
808 call_VM(noreg,
809 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
810 lock_reg);
811
812 bind(done);
813
814 restore_bcp();
815 }
816 }
817
818 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
819 Label& zero_continue) {
820 assert(ProfileInterpreter, "must be profiling interpreter");
821 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
822 cbz(mdp, zero_continue);
823 }
824
825 // Set the method data pointer for the current bcp.
826 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
827 assert(ProfileInterpreter, "must be profiling interpreter");
828 Label set_mdp;
829 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
830
831 // Test MDO to avoid the call if it is NULL.
832 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
833 cbz(r0, set_mdp);
834 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
835 // r0: mdi
836 // mdo is guaranteed to be non-zero here, we checked for it before the call.
837 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
838 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
839 add(r0, r1, r0);
840 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
841 bind(set_mdp);
842 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
843 }
844
845 void InterpreterMacroAssembler::verify_method_data_pointer() {
846 assert(ProfileInterpreter, "must be profiling interpreter");
847 #ifdef ASSERT
848 Label verify_continue;
849 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
850 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
851 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
852 get_method(r1);
853
854 // If the mdp is valid, it will point to a DataLayout header which is
855 // consistent with the bcp. The converse is highly probable also.
856 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
857 ldr(rscratch1, Address(r1, Method::const_offset()));
858 add(r2, r2, rscratch1, Assembler::LSL);
859 lea(r2, Address(r2, ConstMethod::codes_offset()));
860 cmp(r2, rbcp);
861 br(Assembler::EQ, verify_continue);
862 // r1: method
863 // rbcp: bcp // rbcp == 22
864 // r3: mdp
865 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
866 r1, rbcp, r3);
867 bind(verify_continue);
868 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
869 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
870 #endif // ASSERT
871 }
872
873
874 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
875 int constant,
876 Register value) {
877 assert(ProfileInterpreter, "must be profiling interpreter");
878 Address data(mdp_in, constant);
879 str(value, data);
880 }
881
882
883 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
884 int constant,
885 bool decrement) {
886 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
887 }
888
889 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
890 Register reg,
891 int constant,
892 bool decrement) {
893 assert(ProfileInterpreter, "must be profiling interpreter");
894 // %%% this does 64bit counters at best it is wasting space
895 // at worst it is a rare bug when counters overflow
896
897 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
898
899 Address addr1(mdp_in, constant);
900 Address addr2(rscratch2, reg, Address::lsl(0));
901 Address &addr = addr1;
902 if (reg != noreg) {
903 lea(rscratch2, addr1);
904 addr = addr2;
905 }
906
907 if (decrement) {
908 // Decrement the register. Set condition codes.
909 // Intel does this
910 // addptr(data, (int32_t) -DataLayout::counter_increment);
911 // If the decrement causes the counter to overflow, stay negative
912 // Label L;
913 // jcc(Assembler::negative, L);
914 // addptr(data, (int32_t) DataLayout::counter_increment);
915 // so we do this
916 ldr(rscratch1, addr);
917 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
918 Label L;
919 br(Assembler::LO, L); // skip store if counter underflow
920 str(rscratch1, addr);
921 bind(L);
922 } else {
923 assert(DataLayout::counter_increment == 1,
924 "flow-free idiom only works with 1");
925 // Intel does this
926 // Increment the register. Set carry flag.
927 // addptr(data, DataLayout::counter_increment);
928 // If the increment causes the counter to overflow, pull back by 1.
929 // sbbptr(data, (int32_t)0);
930 // so we do this
931 ldr(rscratch1, addr);
932 adds(rscratch1, rscratch1, DataLayout::counter_increment);
933 Label L;
934 br(Assembler::CS, L); // skip store if counter overflow
935 str(rscratch1, addr);
936 bind(L);
937 }
938 }
939
940 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
941 int flag_byte_constant) {
942 assert(ProfileInterpreter, "must be profiling interpreter");
943 int header_offset = in_bytes(DataLayout::header_offset());
944 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
945 // Set the flag
946 ldr(rscratch1, Address(mdp_in, header_offset));
947 orr(rscratch1, rscratch1, header_bits);
948 str(rscratch1, Address(mdp_in, header_offset));
949 }
950
951
952 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
953 int offset,
954 Register value,
955 Register test_value_out,
956 Label& not_equal_continue) {
957 assert(ProfileInterpreter, "must be profiling interpreter");
958 if (test_value_out == noreg) {
959 ldr(rscratch1, Address(mdp_in, offset));
960 cmp(value, rscratch1);
961 } else {
962 // Put the test value into a register, so caller can use it:
963 ldr(test_value_out, Address(mdp_in, offset));
964 cmp(value, test_value_out);
965 }
966 br(Assembler::NE, not_equal_continue);
967 }
968
969
970 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
971 int offset_of_disp) {
972 assert(ProfileInterpreter, "must be profiling interpreter");
973 ldr(rscratch1, Address(mdp_in, offset_of_disp));
974 add(mdp_in, mdp_in, rscratch1, LSL);
975 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
976 }
977
978
979 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
980 Register reg,
981 int offset_of_disp) {
982 assert(ProfileInterpreter, "must be profiling interpreter");
983 lea(rscratch1, Address(mdp_in, offset_of_disp));
984 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
985 add(mdp_in, mdp_in, rscratch1, LSL);
986 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
987 }
988
989
990 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
991 int constant) {
992 assert(ProfileInterpreter, "must be profiling interpreter");
993 add(mdp_in, mdp_in, (unsigned)constant);
994 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
995 }
996
997
998 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
999 assert(ProfileInterpreter, "must be profiling interpreter");
1000 // save/restore across call_VM
1001 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1002 call_VM(noreg,
1003 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1004 return_bci);
1005 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1006 }
1007
1008
1009 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1010 Register bumped_count) {
1011 if (ProfileInterpreter) {
1012 Label profile_continue;
1013
1014 // If no method data exists, go to profile_continue.
1015 // Otherwise, assign to mdp
1016 test_method_data_pointer(mdp, profile_continue);
1017
1018 // We are taking a branch. Increment the taken count.
1019 // We inline increment_mdp_data_at to return bumped_count in a register
1020 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1021 Address data(mdp, in_bytes(JumpData::taken_offset()));
1022 ldr(bumped_count, data);
1023 assert(DataLayout::counter_increment == 1,
1024 "flow-free idiom only works with 1");
1025 // Intel does this to catch overflow
1026 // addptr(bumped_count, DataLayout::counter_increment);
1027 // sbbptr(bumped_count, 0);
1028 // so we do this
1029 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1030 Label L;
1031 br(Assembler::CS, L); // skip store if counter overflow
1032 str(bumped_count, data);
1033 bind(L);
1034 // The method data pointer needs to be updated to reflect the new target.
1035 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1036 bind(profile_continue);
1037 }
1038 }
1039
1040
1041 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1042 if (ProfileInterpreter) {
1043 Label profile_continue;
1044
1045 // If no method data exists, go to profile_continue.
1046 test_method_data_pointer(mdp, profile_continue);
1047
1048 // We are taking a branch. Increment the not taken count.
1049 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1050
1051 // The method data pointer needs to be updated to correspond to
1052 // the next bytecode
1053 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1054 bind(profile_continue);
1055 }
1056 }
1057
1058
1059 void InterpreterMacroAssembler::profile_call(Register mdp) {
1060 if (ProfileInterpreter) {
1061 Label profile_continue;
1062
1063 // If no method data exists, go to profile_continue.
1064 test_method_data_pointer(mdp, profile_continue);
1065
1066 // We are making a call. Increment the count.
1067 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1068
1069 // The method data pointer needs to be updated to reflect the new target.
1070 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1071 bind(profile_continue);
1072 }
1073 }
1074
1075 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1076 if (ProfileInterpreter) {
1077 Label profile_continue;
1078
1079 // If no method data exists, go to profile_continue.
1080 test_method_data_pointer(mdp, profile_continue);
1081
1082 // We are making a call. Increment the count.
1083 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1084
1085 // The method data pointer needs to be updated to reflect the new target.
1086 update_mdp_by_constant(mdp,
1087 in_bytes(VirtualCallData::
1088 virtual_call_data_size()));
1089 bind(profile_continue);
1090 }
1091 }
1092
1093
1094 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1095 Register mdp,
1096 Register reg2,
1097 bool receiver_can_be_null) {
1098 if (ProfileInterpreter) {
1099 Label profile_continue;
1100
1101 // If no method data exists, go to profile_continue.
1102 test_method_data_pointer(mdp, profile_continue);
1103
1104 Label skip_receiver_profile;
1105 if (receiver_can_be_null) {
1106 Label not_null;
1107 // We are making a call. Increment the count for null receiver.
1108 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1109 b(skip_receiver_profile);
1110 bind(not_null);
1111 }
1112
1113 // Record the receiver type.
1114 record_klass_in_profile(receiver, mdp, reg2, true);
1115 bind(skip_receiver_profile);
1116
1117 // The method data pointer needs to be updated to reflect the new target.
1118 #if INCLUDE_JVMCI
1119 if (MethodProfileWidth == 0) {
1120 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1121 }
1122 #else // INCLUDE_JVMCI
1123 update_mdp_by_constant(mdp,
1124 in_bytes(VirtualCallData::
1125 virtual_call_data_size()));
1126 #endif // INCLUDE_JVMCI
1127 bind(profile_continue);
1128 }
1129 }
1130
1131 #if INCLUDE_JVMCI
1132 void InterpreterMacroAssembler::profile_called_method(Register method, Register mdp, Register reg2) {
1133 assert_different_registers(method, mdp, reg2);
1134 if (ProfileInterpreter && MethodProfileWidth > 0) {
1135 Label profile_continue;
1136
1137 // If no method data exists, go to profile_continue.
1138 test_method_data_pointer(mdp, profile_continue);
1139
1140 Label done;
1141 record_item_in_profile_helper(method, mdp, reg2, 0, done, MethodProfileWidth,
1142 &VirtualCallData::method_offset, &VirtualCallData::method_count_offset, in_bytes(VirtualCallData::nonprofiled_receiver_count_offset()));
1143 bind(done);
1144
1145 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1146 bind(profile_continue);
1147 }
1148 }
1149 #endif // INCLUDE_JVMCI
1150
1151 // This routine creates a state machine for updating the multi-row
1152 // type profile at a virtual call site (or other type-sensitive bytecode).
1153 // The machine visits each row (of receiver/count) until the receiver type
1154 // is found, or until it runs out of rows. At the same time, it remembers
1155 // the location of the first empty row. (An empty row records null for its
1156 // receiver, and can be allocated for a newly-observed receiver type.)
1157 // Because there are two degrees of freedom in the state, a simple linear
1158 // search will not work; it must be a decision tree. Hence this helper
1159 // function is recursive, to generate the required tree structured code.
1160 // It's the interpreter, so we are trading off code space for speed.
1161 // See below for example code.
1162 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1163 Register receiver, Register mdp,
1164 Register reg2, int start_row,
1165 Label& done, bool is_virtual_call) {
1166 if (TypeProfileWidth == 0) {
1167 if (is_virtual_call) {
1168 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1169 }
1170 #if INCLUDE_JVMCI
1171 else if (EnableJVMCI) {
1172 increment_mdp_data_at(mdp, in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset()));
1173 }
1174 #endif // INCLUDE_JVMCI
1175 } else {
1176 int non_profiled_offset = -1;
1177 if (is_virtual_call) {
1178 non_profiled_offset = in_bytes(CounterData::count_offset());
1179 }
1180 #if INCLUDE_JVMCI
1181 else if (EnableJVMCI) {
1182 non_profiled_offset = in_bytes(ReceiverTypeData::nonprofiled_receiver_count_offset());
1183 }
1184 #endif // INCLUDE_JVMCI
1185
1186 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1187 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1188 }
1189 }
1190
1191 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1192 Register reg2, int start_row, Label& done, int total_rows,
1193 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn,
1194 int non_profiled_offset) {
1195 int last_row = total_rows - 1;
1196 assert(start_row <= last_row, "must be work left to do");
1197 // Test this row for both the item and for null.
1198 // Take any of three different outcomes:
1199 // 1. found item => increment count and goto done
1200 // 2. found null => keep looking for case 1, maybe allocate this cell
1201 // 3. found something else => keep looking for cases 1 and 2
1202 // Case 3 is handled by a recursive call.
1203 for (int row = start_row; row <= last_row; row++) {
1204 Label next_test;
1205 bool test_for_null_also = (row == start_row);
1206
1207 // See if the item is item[n].
1208 int item_offset = in_bytes(item_offset_fn(row));
1209 test_mdp_data_at(mdp, item_offset, item,
1210 (test_for_null_also ? reg2 : noreg),
1211 next_test);
1212 // (Reg2 now contains the item from the CallData.)
1213
1214 // The item is item[n]. Increment count[n].
1215 int count_offset = in_bytes(item_count_offset_fn(row));
1216 increment_mdp_data_at(mdp, count_offset);
1217 b(done);
1218 bind(next_test);
1219
1220 if (test_for_null_also) {
1221 Label found_null;
1222 // Failed the equality check on item[n]... Test for null.
1223 if (start_row == last_row) {
1224 // The only thing left to do is handle the null case.
1225 if (non_profiled_offset >= 0) {
1226 cbz(reg2, found_null);
1227 // Item did not match any saved item and there is no empty row for it.
1228 // Increment total counter to indicate polymorphic case.
1229 increment_mdp_data_at(mdp, non_profiled_offset);
1230 b(done);
1231 bind(found_null);
1232 } else {
1233 cbnz(reg2, done);
1234 }
1235 break;
1236 }
1237 // Since null is rare, make it be the branch-taken case.
1238 cbz(reg2, found_null);
1239
1240 // Put all the "Case 3" tests here.
1241 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1242 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1243
1244 // Found a null. Keep searching for a matching item,
1245 // but remember that this is an empty (unused) slot.
1246 bind(found_null);
1247 }
1248 }
1249
1250 // In the fall-through case, we found no matching item, but we
1251 // observed the item[start_row] is NULL.
1252
1253 // Fill in the item field and increment the count.
1254 int item_offset = in_bytes(item_offset_fn(start_row));
1255 set_mdp_data_at(mdp, item_offset, item);
1256 int count_offset = in_bytes(item_count_offset_fn(start_row));
1257 mov(reg2, DataLayout::counter_increment);
1258 set_mdp_data_at(mdp, count_offset, reg2);
1259 if (start_row > 0) {
1260 b(done);
1261 }
1262 }
1263
1264 // Example state machine code for three profile rows:
1265 // // main copy of decision tree, rooted at row[1]
1266 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1267 // if (row[0].rec != NULL) {
1268 // // inner copy of decision tree, rooted at row[1]
1269 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1270 // if (row[1].rec != NULL) {
1271 // // degenerate decision tree, rooted at row[2]
1272 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1273 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1274 // row[2].init(rec); goto done;
1275 // } else {
1276 // // remember row[1] is empty
1277 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1278 // row[1].init(rec); goto done;
1279 // }
1280 // } else {
1281 // // remember row[0] is empty
1282 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1283 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1284 // row[0].init(rec); goto done;
1285 // }
1286 // done:
1287
1288 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1289 Register mdp, Register reg2,
1290 bool is_virtual_call) {
1291 assert(ProfileInterpreter, "must be profiling");
1292 Label done;
1293
1294 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1295
1296 bind (done);
1297 }
1298
1299 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1300 Register mdp) {
1301 if (ProfileInterpreter) {
1302 Label profile_continue;
1303 uint row;
1304
1305 // If no method data exists, go to profile_continue.
1306 test_method_data_pointer(mdp, profile_continue);
1307
1308 // Update the total ret count.
1309 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1310
1311 for (row = 0; row < RetData::row_limit(); row++) {
1312 Label next_test;
1313
1314 // See if return_bci is equal to bci[n]:
1315 test_mdp_data_at(mdp,
1316 in_bytes(RetData::bci_offset(row)),
1317 return_bci, noreg,
1318 next_test);
1319
1320 // return_bci is equal to bci[n]. Increment the count.
1321 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1322
1323 // The method data pointer needs to be updated to reflect the new target.
1324 update_mdp_by_offset(mdp,
1325 in_bytes(RetData::bci_displacement_offset(row)));
1326 b(profile_continue);
1327 bind(next_test);
1328 }
1329
1330 update_mdp_for_ret(return_bci);
1331
1332 bind(profile_continue);
1333 }
1334 }
1335
1336 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1337 if (ProfileInterpreter) {
1338 Label profile_continue;
1339
1340 // If no method data exists, go to profile_continue.
1341 test_method_data_pointer(mdp, profile_continue);
1342
1343 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1344
1345 // The method data pointer needs to be updated.
1346 int mdp_delta = in_bytes(BitData::bit_data_size());
1347 if (TypeProfileCasts) {
1348 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1349 }
1350 update_mdp_by_constant(mdp, mdp_delta);
1351
1352 bind(profile_continue);
1353 }
1354 }
1355
1356 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1357 if (ProfileInterpreter && TypeProfileCasts) {
1358 Label profile_continue;
1359
1360 // If no method data exists, go to profile_continue.
1361 test_method_data_pointer(mdp, profile_continue);
1362
1363 int count_offset = in_bytes(CounterData::count_offset());
1364 // Back up the address, since we have already bumped the mdp.
1365 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1366
1367 // *Decrement* the counter. We expect to see zero or small negatives.
1368 increment_mdp_data_at(mdp, count_offset, true);
1369
1370 bind (profile_continue);
1371 }
1372 }
1373
1374 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1375 if (ProfileInterpreter) {
1376 Label profile_continue;
1377
1378 // If no method data exists, go to profile_continue.
1379 test_method_data_pointer(mdp, profile_continue);
1380
1381 // The method data pointer needs to be updated.
1382 int mdp_delta = in_bytes(BitData::bit_data_size());
1383 if (TypeProfileCasts) {
1384 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1385
1386 // Record the object type.
1387 record_klass_in_profile(klass, mdp, reg2, false);
1388 }
1389 update_mdp_by_constant(mdp, mdp_delta);
1390
1391 bind(profile_continue);
1392 }
1393 }
1394
1395 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1396 if (ProfileInterpreter) {
1397 Label profile_continue;
1398
1399 // If no method data exists, go to profile_continue.
1400 test_method_data_pointer(mdp, profile_continue);
1401
1402 // Update the default case count
1403 increment_mdp_data_at(mdp,
1404 in_bytes(MultiBranchData::default_count_offset()));
1405
1406 // The method data pointer needs to be updated.
1407 update_mdp_by_offset(mdp,
1408 in_bytes(MultiBranchData::
1409 default_displacement_offset()));
1410
1411 bind(profile_continue);
1412 }
1413 }
1414
1415 void InterpreterMacroAssembler::profile_switch_case(Register index,
1416 Register mdp,
1417 Register reg2) {
1418 if (ProfileInterpreter) {
1419 Label profile_continue;
1420
1421 // If no method data exists, go to profile_continue.
1422 test_method_data_pointer(mdp, profile_continue);
1423
1424 // Build the base (index * per_case_size_in_bytes()) +
1425 // case_array_offset_in_bytes()
1426 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1427 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1428 Assembler::maddw(index, index, reg2, rscratch1);
1429
1430 // Update the case count
1431 increment_mdp_data_at(mdp,
1432 index,
1433 in_bytes(MultiBranchData::relative_count_offset()));
1434
1435 // The method data pointer needs to be updated.
1436 update_mdp_by_offset(mdp,
1437 index,
1438 in_bytes(MultiBranchData::
1439 relative_displacement_offset()));
1440
1441 bind(profile_continue);
1442 }
1443 }
1444
1445 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1446 if (state == atos) {
1447 MacroAssembler::verify_oop(reg);
1448 }
1449 }
1450
1451 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1452
1453
1454 void InterpreterMacroAssembler::notify_method_entry() {
1455 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1456 // track stack depth. If it is possible to enter interp_only_mode we add
1457 // the code to check if the event should be sent.
1458 if (JvmtiExport::can_post_interpreter_events()) {
1459 Label L;
1460 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1461 cbzw(r3, L);
1462 call_VM(noreg, CAST_FROM_FN_PTR(address,
1463 InterpreterRuntime::post_method_entry));
1464 bind(L);
1465 }
1466
1467 {
1468 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1469 get_method(c_rarg1);
1470 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1471 rthread, c_rarg1);
1472 }
1473
1474 // RedefineClasses() tracing support for obsolete method entry
1475 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1476 get_method(c_rarg1);
1477 call_VM_leaf(
1478 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1479 rthread, c_rarg1);
1480 }
1481
1482 }
1483
1484
1485 void InterpreterMacroAssembler::notify_method_exit(
1486 TosState state, NotifyMethodExitMode mode) {
1487 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1488 // track stack depth. If it is possible to enter interp_only_mode we add
1489 // the code to check if the event should be sent.
1490 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1491 Label L;
1492 // Note: frame::interpreter_frame_result has a dependency on how the
1493 // method result is saved across the call to post_method_exit. If this
1494 // is changed then the interpreter_frame_result implementation will
1495 // need to be updated too.
1496
1497 // template interpreter will leave the result on the top of the stack.
1498 push(state);
1499 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1500 cbz(r3, L);
1501 call_VM(noreg,
1502 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1503 bind(L);
1504 pop(state);
1505 }
1506
1507 {
1508 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1509 push(state);
1510 get_method(c_rarg1);
1511 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1512 rthread, c_rarg1);
1513 pop(state);
1514 }
1515 }
1516
1517
1518 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1519 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1520 int increment, Address mask,
1521 Register scratch, Register scratch2,
1522 bool preloaded, Condition cond,
1523 Label* where) {
1524 if (!preloaded) {
1525 ldrw(scratch, counter_addr);
1526 }
1527 add(scratch, scratch, increment);
1528 strw(scratch, counter_addr);
1529 ldrw(scratch2, mask);
1530 ands(scratch, scratch, scratch2);
1531 br(cond, *where);
1532 }
1533
1534 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1535 int number_of_arguments) {
1536 // interpreter specific
1537 //
1538 // Note: No need to save/restore rbcp & rlocals pointer since these
1539 // are callee saved registers and no blocking/ GC can happen
1540 // in leaf calls.
1541 #ifdef ASSERT
1542 {
1543 Label L;
1544 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1545 cbz(rscratch1, L);
1546 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1547 " last_sp != NULL");
1548 bind(L);
1549 }
1550 #endif /* ASSERT */
1551 // super call
1552 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1553 }
1554
1555 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1556 Register java_thread,
1557 Register last_java_sp,
1558 address entry_point,
1559 int number_of_arguments,
1560 bool check_exceptions) {
1561 // interpreter specific
1562 //
1563 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1564 // really make a difference for these runtime calls, since they are
1565 // slow anyway. Btw., bcp must be saved/restored since it may change
1566 // due to GC.
1567 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1568 save_bcp();
1569 #ifdef ASSERT
1570 {
1571 Label L;
1572 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1573 cbz(rscratch1, L);
1574 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1575 " last_sp != NULL");
1576 bind(L);
1577 }
1578 #endif /* ASSERT */
1579 // super call
1580 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1581 entry_point, number_of_arguments,
1582 check_exceptions);
1583 // interpreter specific
1584 restore_bcp();
1585 restore_locals();
1586 }
1587
1588 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1589 Label update, next, none;
1590
1591 verify_oop(obj);
1592
1593 cbnz(obj, update);
1594 orptr(mdo_addr, TypeEntries::null_seen);
1595 b(next);
1596
1597 bind(update);
1598 load_klass(obj, obj);
1599
1600 ldr(rscratch1, mdo_addr);
1601 eor(obj, obj, rscratch1);
1602 tst(obj, TypeEntries::type_klass_mask);
1603 br(Assembler::EQ, next); // klass seen before, nothing to
1604 // do. The unknown bit may have been
1605 // set already but no need to check.
1606
1607 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1608 // already unknown. Nothing to do anymore.
1609
1610 ldr(rscratch1, mdo_addr);
1611 cbz(rscratch1, none);
1612 cmp(rscratch1, TypeEntries::null_seen);
1613 br(Assembler::EQ, none);
1614 // There is a chance that the checks above (re-reading profiling
1615 // data from memory) fail if another thread has just set the
1616 // profiling to this obj's klass
1617 ldr(rscratch1, mdo_addr);
1618 eor(obj, obj, rscratch1);
1619 tst(obj, TypeEntries::type_klass_mask);
1620 br(Assembler::EQ, next);
1621
1622 // different than before. Cannot keep accurate profile.
1623 orptr(mdo_addr, TypeEntries::type_unknown);
1624 b(next);
1625
1626 bind(none);
1627 // first time here. Set profile type.
1628 str(obj, mdo_addr);
1629
1630 bind(next);
1631 }
1632
1633 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1634 if (!ProfileInterpreter) {
1635 return;
1636 }
1637
1638 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1639 Label profile_continue;
1640
1641 test_method_data_pointer(mdp, profile_continue);
1642
1643 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1644
1645 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1646 cmp(rscratch1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
1647 br(Assembler::NE, profile_continue);
1648
1649 if (MethodData::profile_arguments()) {
1650 Label done;
1651 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1652
1653 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1654 if (i > 0 || MethodData::profile_return()) {
1655 // If return value type is profiled we may have no argument to profile
1656 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1657 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1658 cmp(tmp, TypeStackSlotEntries::per_arg_count());
1659 add(rscratch1, mdp, off_to_args);
1660 br(Assembler::LT, done);
1661 }
1662 ldr(tmp, Address(callee, Method::const_offset()));
1663 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1664 // stack offset o (zero based) from the start of the argument
1665 // list, for n arguments translates into offset n - o - 1 from
1666 // the end of the argument list
1667 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1668 sub(tmp, tmp, rscratch1);
1669 sub(tmp, tmp, 1);
1670 Address arg_addr = argument_address(tmp);
1671 ldr(tmp, arg_addr);
1672
1673 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1674 profile_obj_type(tmp, mdo_arg_addr);
1675
1676 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1677 off_to_args += to_add;
1678 }
1679
1680 if (MethodData::profile_return()) {
1681 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1682 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1683 }
1684
1685 add(rscratch1, mdp, off_to_args);
1686 bind(done);
1687 mov(mdp, rscratch1);
1688
1689 if (MethodData::profile_return()) {
1690 // We're right after the type profile for the last
1691 // argument. tmp is the number of cells left in the
1692 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1693 // if there's a return to profile.
1694 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1695 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1696 }
1697 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1698 } else {
1699 assert(MethodData::profile_return(), "either profile call args or call ret");
1700 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1701 }
1702
1703 // mdp points right after the end of the
1704 // CallTypeData/VirtualCallTypeData, right after the cells for the
1705 // return value type if there's one
1706
1707 bind(profile_continue);
1708 }
1709 }
1710
1711 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1712 assert_different_registers(mdp, ret, tmp, rbcp);
1713 if (ProfileInterpreter && MethodData::profile_return()) {
1714 Label profile_continue, done;
1715
1716 test_method_data_pointer(mdp, profile_continue);
1717
1718 if (MethodData::profile_return_jsr292_only()) {
1719 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1720
1721 // If we don't profile all invoke bytecodes we must make sure
1722 // it's a bytecode we indeed profile. We can't go back to the
1723 // begining of the ProfileData we intend to update to check its
1724 // type because we're right after it and we don't known its
1725 // length
1726 Label do_profile;
1727 ldrb(rscratch1, Address(rbcp, 0));
1728 cmp(rscratch1, Bytecodes::_invokedynamic);
1729 br(Assembler::EQ, do_profile);
1730 cmp(rscratch1, Bytecodes::_invokehandle);
1731 br(Assembler::EQ, do_profile);
1732 get_method(tmp);
1733 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1734 cmp(rscratch1, vmIntrinsics::_compiledLambdaForm);
1735 br(Assembler::NE, profile_continue);
1736
1737 bind(do_profile);
1738 }
1739
1740 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1741 mov(tmp, ret);
1742 profile_obj_type(tmp, mdo_ret_addr);
1743
1744 bind(profile_continue);
1745 }
1746 }
1747
1748 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1749 if (ProfileInterpreter && MethodData::profile_parameters()) {
1750 Label profile_continue, done;
1751
1752 test_method_data_pointer(mdp, profile_continue);
1753
1754 // Load the offset of the area within the MDO used for
1755 // parameters. If it's negative we're not profiling any parameters
1756 ldr(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1757 cmp(tmp1, 0u);
1758 br(Assembler::LT, profile_continue);
1759
1760 // Compute a pointer to the area for parameters from the offset
1761 // and move the pointer to the slot for the last
1762 // parameters. Collect profiling from last parameter down.
1763 // mdo start + parameters offset + array length - 1
1764 add(mdp, mdp, tmp1);
1765 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1766 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1767
1768 Label loop;
1769 bind(loop);
1770
1771 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1772 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1773 int per_arg_scale = exact_log2(DataLayout::cell_size);
1774 add(rscratch1, mdp, off_base);
1775 add(rscratch2, mdp, type_base);
1776
1777 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1778 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1779
1780 // load offset on the stack from the slot for this parameter
1781 ldr(tmp2, arg_off);
1782 neg(tmp2, tmp2);
1783 // read the parameter from the local area
1784 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1785
1786 // profile the parameter
1787 profile_obj_type(tmp2, arg_type);
1788
1789 // go to next parameter
1790 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1791 br(Assembler::GE, loop);
1792
1793 bind(profile_continue);
1794 }
1795 }