1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interp_masm_riscv.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "logging/log.hpp"
35 #include "oops/arrayOop.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "prims/jvmtiExport.hpp"
40 #include "prims/jvmtiThreadState.hpp"
41 #include "runtime/basicLock.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/javaThread.hpp"
44 #include "runtime/safepointMechanism.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "utilities/powerOfTwo.hpp"
47
48 void InterpreterMacroAssembler::narrow(Register result) {
49 // Get method->_constMethod->_result_type
50 ld(t0, Address(fp, frame::interpreter_frame_method_offset * wordSize));
51 ld(t0, Address(t0, Method::const_offset()));
52 lbu(t0, Address(t0, ConstMethod::result_type_offset()));
53
54 Label done, notBool, notByte, notChar;
55
56 // common case first
57 mv(t1, T_INT);
58 beq(t0, t1, done);
59
60 // mask integer result to narrower return type.
61 mv(t1, T_BOOLEAN);
62 bne(t0, t1, notBool);
63
64 andi(result, result, 0x1);
65 j(done);
66
67 bind(notBool);
68 mv(t1, T_BYTE);
69 bne(t0, t1, notByte);
70 sign_extend(result, result, 8);
71 j(done);
72
73 bind(notByte);
74 mv(t1, T_CHAR);
75 bne(t0, t1, notChar);
76 zero_extend(result, result, 16);
77 j(done);
78
79 bind(notChar);
80 sign_extend(result, result, 16);
81
82 // Nothing to do for T_INT
83 bind(done);
84 addw(result, result, zr);
85 }
86
87 void InterpreterMacroAssembler::jump_to_entry(address entry) {
88 assert(entry != NULL, "Entry must have been generated by now");
89 j(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,
97 // it 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 lwu(t1, Address(xthread, JavaThread::popframe_condition_offset()));
102 andi(t0, t1, JavaThread::popframe_pending_bit);
103 beqz(t0, L);
104 andi(t0, t1, JavaThread::popframe_processing_bit);
105 bnez(t0, L);
106 // Call Interpreter::remove_activation_preserving_args_entry() to get the
107 // address of the same-named entrypoint in the generated interpreter code.
108 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
109 jr(x10);
110 bind(L);
111 }
112 }
113
114
115 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
116 ld(x12, Address(xthread, JavaThread::jvmti_thread_state_offset()));
117 const Address tos_addr(x12, JvmtiThreadState::earlyret_tos_offset());
118 const Address oop_addr(x12, JvmtiThreadState::earlyret_oop_offset());
119 const Address val_addr(x12, JvmtiThreadState::earlyret_value_offset());
120 switch (state) {
121 case atos:
122 ld(x10, oop_addr);
123 sd(zr, oop_addr);
124 verify_oop(x10);
125 break;
126 case ltos:
127 ld(x10, val_addr);
128 break;
129 case btos: // fall through
130 case ztos: // fall through
131 case ctos: // fall through
132 case stos: // fall through
133 case itos:
134 lwu(x10, val_addr);
135 break;
136 case ftos:
137 flw(f10, val_addr);
138 break;
139 case dtos:
140 fld(f10, val_addr);
141 break;
142 case vtos:
143 /* nothing to do */
144 break;
145 default:
146 ShouldNotReachHere();
147 }
148 // Clean up tos value in the thread object
149 mv(t0, (int)ilgl);
150 sw(t0, tos_addr);
151 sw(zr, val_addr);
152 }
153
154
155 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
156 if (JvmtiExport::can_force_early_return()) {
157 Label L;
158 ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
159 beqz(t0, L); // if [thread->jvmti_thread_state() == NULL] then exit
160
161 // Initiate earlyret handling only if it is not already being processed.
162 // If the flag has the earlyret_processing bit set, it means that this code
163 // is called *during* earlyret handling - we don't want to reenter.
164 lwu(t0, Address(t0, JvmtiThreadState::earlyret_state_offset()));
165 mv(t1, JvmtiThreadState::earlyret_pending);
166 bne(t0, t1, L);
167
168 // Call Interpreter::remove_activation_early_entry() to get the address of the
169 // same-named entrypoint in the generated interpreter code.
170 ld(t0, Address(xthread, JavaThread::jvmti_thread_state_offset()));
171 lwu(t0, Address(t0, JvmtiThreadState::earlyret_tos_offset()));
172 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), t0);
173 jr(x10);
174 bind(L);
175 }
176 }
177
178 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
179 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
180 lhu(reg, Address(xbcp, bcp_offset));
181 revb_h(reg, reg);
182 }
183
184 void InterpreterMacroAssembler::get_dispatch() {
185 ExternalAddress target((address)Interpreter::dispatch_table());
186 relocate(target.rspec(), [&] {
187 int32_t offset;
188 la_patchable(xdispatch, target, offset);
189 addi(xdispatch, xdispatch, offset);
190 });
191 }
192
193 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
194 int bcp_offset,
195 size_t index_size) {
196 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
197 if (index_size == sizeof(u2)) {
198 load_unsigned_short(index, Address(xbcp, bcp_offset));
199 } else if (index_size == sizeof(u4)) {
200 lwu(index, Address(xbcp, bcp_offset));
201 // Check if the secondary index definition is still ~x, otherwise
202 // we have to change the following assembler code to calculate the
203 // plain index.
204 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
205 xori(index, index, -1);
206 addw(index, index, zr);
207 } else if (index_size == sizeof(u1)) {
208 load_unsigned_byte(index, Address(xbcp, bcp_offset));
209 } else {
210 ShouldNotReachHere();
211 }
212 }
213
214 // Return
215 // Rindex: index into constant pool
216 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
217 //
218 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
219 // the true address of the cache entry.
220 //
221 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
222 Register index,
223 int bcp_offset,
224 size_t index_size) {
225 assert_different_registers(cache, index);
226 assert_different_registers(cache, xcpool);
227 get_cache_index_at_bcp(index, bcp_offset, index_size);
228 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
229 // Convert from field index to ConstantPoolCacheEntry
230 // riscv already has the cache in xcpool so there is no need to
231 // install it in cache. Instead we pre-add the indexed offset to
232 // xcpool and return it in cache. All clients of this method need to
233 // be modified accordingly.
234 shadd(cache, index, xcpool, cache, 5);
235 }
236
237
238 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
239 Register index,
240 Register bytecode,
241 int byte_no,
242 int bcp_offset,
243 size_t index_size) {
244 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
245 // We use a 32-bit load here since the layout of 64-bit words on
246 // little-endian machines allow us that.
247 // n.b. unlike x86 cache already includes the index offset
248 la(bytecode, Address(cache,
249 ConstantPoolCache::base_offset() +
250 ConstantPoolCacheEntry::indices_offset()));
251 membar(MacroAssembler::AnyAny);
252 lwu(bytecode, bytecode);
253 membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
254 const int shift_count = (1 + byte_no) * BitsPerByte;
255 slli(bytecode, bytecode, XLEN - (shift_count + BitsPerByte));
256 srli(bytecode, bytecode, XLEN - BitsPerByte);
257 }
258
259 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
260 Register tmp,
261 int bcp_offset,
262 size_t index_size) {
263 assert_different_registers(cache, tmp);
264 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
265 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
266 // Convert from field index to ConstantPoolCacheEntry index
267 // and from word offset to byte offset
268 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord,
269 "else change next line");
270 ld(cache, Address(fp, frame::interpreter_frame_cache_offset * wordSize));
271 // skip past the header
272 add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
273 // construct pointer to cache entry
274 shadd(cache, tmp, cache, tmp, 2 + LogBytesPerWord);
275 }
276
277 // Load object from cpool->resolved_references(index)
278 void InterpreterMacroAssembler::load_resolved_reference_at_index(
279 Register result, Register index, Register tmp) {
280 assert_different_registers(result, index);
281
282 get_constant_pool(result);
283 // Load pointer for resolved_references[] objArray
284 ld(result, Address(result, ConstantPool::cache_offset_in_bytes()));
285 ld(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
286 resolve_oop_handle(result, tmp, t1);
287 // Add in the index
288 addi(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
289 shadd(result, index, result, index, LogBytesPerHeapOop);
290 load_heap_oop(result, Address(result, 0), tmp, t1);
291 }
292
293 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
294 Register cpool, Register index, Register klass, Register temp) {
295 shadd(temp, index, cpool, temp, LogBytesPerWord);
296 lhu(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
297 ld(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
298 shadd(klass, temp, klass, temp, LogBytesPerWord);
299 ld(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
300 }
301
302 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
303 Register method,
304 Register cache) {
305 const int method_offset = in_bytes(
306 ConstantPoolCache::base_offset() +
307 ((byte_no == TemplateTable::f2_byte)
308 ? ConstantPoolCacheEntry::f2_offset()
309 : ConstantPoolCacheEntry::f1_offset()));
310
311 ld(method, Address(cache, method_offset)); // get f1 Method*
312 }
313
314 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
315 // subtype of super_klass.
316 //
317 // Args:
318 // x10: superklass
319 // Rsub_klass: subklass
320 //
321 // Kills:
322 // x12, x15
323 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
324 Label& ok_is_subtype) {
325 assert(Rsub_klass != x10, "x10 holds superklass");
326 assert(Rsub_klass != x12, "x12 holds 2ndary super array length");
327 assert(Rsub_klass != x15, "x15 holds 2ndary super array scan ptr");
328
329 // Profile the not-null value's klass.
330 profile_typecheck(x12, Rsub_klass, x15); // blows x12, reloads x15
331
332 // Do the check.
333 check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12
334
335 // Profile the failure of the check.
336 profile_typecheck_failed(x12); // blows x12
337 }
338
339 // Java Expression Stack
340
341 void InterpreterMacroAssembler::pop_ptr(Register r) {
342 ld(r, Address(esp, 0));
343 addi(esp, esp, wordSize);
344 }
345
346 void InterpreterMacroAssembler::pop_i(Register r) {
347 lw(r, Address(esp, 0)); // lw do signed extended
348 addi(esp, esp, wordSize);
349 }
350
351 void InterpreterMacroAssembler::pop_l(Register r) {
352 ld(r, Address(esp, 0));
353 addi(esp, esp, 2 * Interpreter::stackElementSize);
354 }
355
356 void InterpreterMacroAssembler::push_ptr(Register r) {
357 addi(esp, esp, -wordSize);
358 sd(r, Address(esp, 0));
359 }
360
361 void InterpreterMacroAssembler::push_i(Register r) {
362 addi(esp, esp, -wordSize);
363 addw(r, r, zr); // signed extended
364 sd(r, Address(esp, 0));
365 }
366
367 void InterpreterMacroAssembler::push_l(Register r) {
368 addi(esp, esp, -2 * wordSize);
369 sd(zr, Address(esp, wordSize));
370 sd(r, Address(esp));
371 }
372
373 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
374 flw(r, Address(esp, 0));
375 addi(esp, esp, wordSize);
376 }
377
378 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
379 fld(r, Address(esp, 0));
380 addi(esp, esp, 2 * Interpreter::stackElementSize);
381 }
382
383 void InterpreterMacroAssembler::push_f(FloatRegister r) {
384 addi(esp, esp, -wordSize);
385 fsw(r, Address(esp, 0));
386 }
387
388 void InterpreterMacroAssembler::push_d(FloatRegister r) {
389 addi(esp, esp, -2 * wordSize);
390 fsd(r, Address(esp, 0));
391 }
392
393 void InterpreterMacroAssembler::pop(TosState state) {
394 switch (state) {
395 case atos:
396 pop_ptr();
397 verify_oop(x10);
398 break;
399 case btos: // fall through
400 case ztos: // fall through
401 case ctos: // fall through
402 case stos: // fall through
403 case itos:
404 pop_i();
405 break;
406 case ltos:
407 pop_l();
408 break;
409 case ftos:
410 pop_f();
411 break;
412 case dtos:
413 pop_d();
414 break;
415 case vtos:
416 /* nothing to do */
417 break;
418 default:
419 ShouldNotReachHere();
420 }
421 }
422
423 void InterpreterMacroAssembler::push(TosState state) {
424 switch (state) {
425 case atos:
426 verify_oop(x10);
427 push_ptr();
428 break;
429 case btos: // fall through
430 case ztos: // fall through
431 case ctos: // fall through
432 case stos: // fall through
433 case itos:
434 push_i();
435 break;
436 case ltos:
437 push_l();
438 break;
439 case ftos:
440 push_f();
441 break;
442 case dtos:
443 push_d();
444 break;
445 case vtos:
446 /* nothing to do */
447 break;
448 default:
449 ShouldNotReachHere();
450 }
451 }
452
453 // Helpers for swap and dup
454 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
455 ld(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
456 }
457
458 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
459 sd(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
460 }
461
462 void InterpreterMacroAssembler::load_float(Address src) {
463 flw(f10, src);
464 }
465
466 void InterpreterMacroAssembler::load_double(Address src) {
467 fld(f10, src);
468 }
469
470 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
471 // set sender sp
472 mv(x19_sender_sp, sp);
473 // record last_sp
474 sd(esp, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
475 }
476
477 // Jump to from_interpreted entry of a call unless single stepping is possible
478 // in this thread in which case we must call the i2i entry
479 void InterpreterMacroAssembler::jump_from_interpreted(Register method) {
480 prepare_to_jump_from_interpreted();
481 if (JvmtiExport::can_post_interpreter_events()) {
482 Label run_compiled_code;
483 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
484 // compiled code in threads for which the event is enabled. Check here for
485 // interp_only_mode if these events CAN be enabled.
486 lwu(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
487 beqz(t0, run_compiled_code);
488 ld(t0, Address(method, Method::interpreter_entry_offset()));
489 jr(t0);
490 bind(run_compiled_code);
491 }
492
493 ld(t0, Address(method, Method::from_interpreted_offset()));
494 jr(t0);
495 }
496
497 // The following two routines provide a hook so that an implementation
498 // can schedule the dispatch in two parts. amd64 does not do this.
499 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
500 }
501
502 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
503 dispatch_next(state, step);
504 }
505
506 void InterpreterMacroAssembler::dispatch_base(TosState state,
507 address* table,
508 bool verifyoop,
509 bool generate_poll,
510 Register Rs) {
511 // Pay attention to the argument Rs, which is acquiesce in t0.
512 if (VerifyActivationFrameSize) {
513 Unimplemented();
514 }
515 if (verifyoop && state == atos) {
516 verify_oop(x10);
517 }
518
519 Label safepoint;
520 address* const safepoint_table = Interpreter::safept_table(state);
521 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
522
523 if (needs_thread_local_poll) {
524 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
525 ld(t1, Address(xthread, JavaThread::polling_word_offset()));
526 andi(t1, t1, SafepointMechanism::poll_bit());
527 bnez(t1, safepoint);
528 }
529 if (table == Interpreter::dispatch_table(state)) {
530 mv(t1, Interpreter::distance_from_dispatch_table(state));
531 add(t1, Rs, t1);
532 shadd(t1, t1, xdispatch, t1, 3);
533 } else {
534 mv(t1, (address)table);
535 shadd(t1, Rs, t1, Rs, 3);
536 }
537 ld(t1, Address(t1));
538 jr(t1);
539
540 if (needs_thread_local_poll) {
541 bind(safepoint);
542 la(t1, ExternalAddress((address)safepoint_table));
543 shadd(t1, Rs, t1, Rs, 3);
544 ld(t1, Address(t1));
545 jr(t1);
546 }
547 }
548
549 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll, Register Rs) {
550 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll, Rs);
551 }
552
553 void InterpreterMacroAssembler::dispatch_only_normal(TosState state, Register Rs) {
554 dispatch_base(state, Interpreter::normal_table(state), true, false, Rs);
555 }
556
557 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state, Register Rs) {
558 dispatch_base(state, Interpreter::normal_table(state), false, false, Rs);
559 }
560
561 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
562 // load next bytecode
563 load_unsigned_byte(t0, Address(xbcp, step));
564 add(xbcp, xbcp, step);
565 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
566 }
567
568 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
569 // load current bytecode
570 lbu(t0, Address(xbcp, 0));
571 dispatch_base(state, table);
572 }
573
574 // remove activation
575 //
576 // Apply stack watermark barrier.
577 // Unlock the receiver if this is a synchronized method.
578 // Unlock any Java monitors from synchronized blocks.
579 // Remove the activation from the stack.
580 //
581 // If there are locked Java monitors
582 // If throw_monitor_exception
583 // throws IllegalMonitorStateException
584 // Else if install_monitor_exception
585 // installs IllegalMonitorStateException
586 // Else
587 // no error processing
588 void InterpreterMacroAssembler::remove_activation(
589 TosState state,
590 bool throw_monitor_exception,
591 bool install_monitor_exception,
592 bool notify_jvmdi) {
593 // Note: Registers x13 may be in use for the
594 // result check if synchronized method
595 Label unlocked, unlock, no_unlock;
596
597 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
598 // that would normally not be safe to use. Such bad returns into unsafe territory of
599 // the stack, will call InterpreterRuntime::at_unwind.
600 Label slow_path;
601 Label fast_path;
602 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
603 j(fast_path);
604
605 bind(slow_path);
606 push(state);
607 set_last_Java_frame(esp, fp, (address)pc(), t0);
608 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), xthread);
609 reset_last_Java_frame(true);
610 pop(state);
611
612 bind(fast_path);
613
614 // get the value of _do_not_unlock_if_synchronized into x13
615 const Address do_not_unlock_if_synchronized(xthread,
616 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
617 lbu(x13, do_not_unlock_if_synchronized);
618 sb(zr, do_not_unlock_if_synchronized); // reset the flag
619
620 // get method access flags
621 ld(x11, Address(fp, frame::interpreter_frame_method_offset * wordSize));
622 ld(x12, Address(x11, Method::access_flags_offset()));
623 andi(t0, x12, JVM_ACC_SYNCHRONIZED);
624 beqz(t0, unlocked);
625
626 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
627 // is set.
628 bnez(x13, no_unlock);
629
630 // unlock monitor
631 push(state); // save result
632
633 // BasicObjectLock will be first in list, since this is a
634 // synchronized method. However, need to check that the object has
635 // not been unlocked by an explicit monitorexit bytecode.
636 const Address monitor(fp, frame::interpreter_frame_initial_sp_offset *
637 wordSize - (int) sizeof(BasicObjectLock));
638 // We use c_rarg1 so that if we go slow path it will be the correct
639 // register for unlock_object to pass to VM directly
640 la(c_rarg1, monitor); // address of first monitor
641
642 ld(x10, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
643 bnez(x10, unlock);
644
645 pop(state);
646 if (throw_monitor_exception) {
647 // Entry already unlocked, need to throw exception
648 call_VM(noreg, CAST_FROM_FN_PTR(address,
649 InterpreterRuntime::throw_illegal_monitor_state_exception));
650 should_not_reach_here();
651 } else {
652 // Monitor already unlocked during a stack unroll. If requested,
653 // install an illegal_monitor_state_exception. Continue with
654 // stack unrolling.
655 if (install_monitor_exception) {
656 call_VM(noreg, CAST_FROM_FN_PTR(address,
657 InterpreterRuntime::new_illegal_monitor_state_exception));
658 }
659 j(unlocked);
660 }
661
662 bind(unlock);
663 unlock_object(c_rarg1);
664 pop(state);
665
666 // Check that for block-structured locking (i.e., that all locked
667 // objects has been unlocked)
668 bind(unlocked);
669
670 // x10: Might contain return value
671
672 // Check that all monitors are unlocked
673 {
674 Label loop, exception, entry, restart;
675 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
676 const Address monitor_block_top(
677 fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
678 const Address monitor_block_bot(
679 fp, frame::interpreter_frame_initial_sp_offset * wordSize);
680
681 bind(restart);
682 // We use c_rarg1 so that if we go slow path it will be the correct
683 // register for unlock_object to pass to VM directly
684 ld(c_rarg1, monitor_block_top); // points to current entry, starting
685 // with top-most entry
686 la(x9, monitor_block_bot); // points to word before bottom of
687 // monitor block
688
689 j(entry);
690
691 // Entry already locked, need to throw exception
692 bind(exception);
693
694 if (throw_monitor_exception) {
695 // Throw exception
696 MacroAssembler::call_VM(noreg,
697 CAST_FROM_FN_PTR(address, InterpreterRuntime::
698 throw_illegal_monitor_state_exception));
699
700 should_not_reach_here();
701 } else {
702 // Stack unrolling. Unlock object and install illegal_monitor_exception.
703 // Unlock does not block, so don't have to worry about the frame.
704 // We don't have to preserve c_rarg1 since we are going to throw an exception.
705
706 push(state);
707 unlock_object(c_rarg1);
708 pop(state);
709
710 if (install_monitor_exception) {
711 call_VM(noreg, CAST_FROM_FN_PTR(address,
712 InterpreterRuntime::
713 new_illegal_monitor_state_exception));
714 }
715
716 j(restart);
717 }
718
719 bind(loop);
720 // check if current entry is used
721 add(t0, c_rarg1, BasicObjectLock::obj_offset_in_bytes());
722 ld(t0, Address(t0, 0));
723 bnez(t0, exception);
724
725 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
726 bind(entry);
727 bne(c_rarg1, x9, loop); // check if bottom reached if not at bottom then check this entry
728 }
729
730 bind(no_unlock);
731
732 // jvmti support
733 if (notify_jvmdi) {
734 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
735
736 } else {
737 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
738 }
739
740 // remove activation
741 // get sender esp
742 ld(t1,
743 Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize));
744 if (StackReservedPages > 0) {
745 // testing if reserved zone needs to be re-enabled
746 Label no_reserved_zone_enabling;
747
748 ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset()));
749 ble(t1, t0, no_reserved_zone_enabling);
750
751 call_VM_leaf(
752 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), xthread);
753 call_VM(noreg, CAST_FROM_FN_PTR(address,
754 InterpreterRuntime::throw_delayed_StackOverflowError));
755 should_not_reach_here();
756
757 bind(no_reserved_zone_enabling);
758 }
759
760 // restore sender esp
761 mv(esp, t1);
762
763 // remove frame anchor
764 leave();
765 // If we're returning to interpreted code we will shortly be
766 // adjusting SP to allow some space for ESP. If we're returning to
767 // compiled code the saved sender SP was saved in sender_sp, so this
768 // restores it.
769 andi(sp, esp, -16);
770 }
771
772 // Lock object
773 //
774 // Args:
775 // c_rarg1: BasicObjectLock to be used for locking
776 //
777 // Kills:
778 // x10
779 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
780 // t0, t1 (temp regs)
781 void InterpreterMacroAssembler::lock_object(Register lock_reg)
782 {
783 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
784 if (UseHeavyMonitors) {
785 call_VM(noreg,
786 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
787 lock_reg);
788 } else {
789 Label count, done;
790
791 const Register swap_reg = x10;
792 const Register tmp = c_rarg2;
793 const Register obj_reg = c_rarg3; // Will contain the oop
794
795 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
796 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
797 const int mark_offset = lock_offset +
798 BasicLock::displaced_header_offset_in_bytes();
799
800 Label slow_case;
801
802 // Load object pointer into obj_reg c_rarg3
803 ld(obj_reg, Address(lock_reg, obj_offset));
804
805 if (DiagnoseSyncOnValueBasedClasses != 0) {
806 load_klass(tmp, obj_reg);
807 lwu(tmp, Address(tmp, Klass::access_flags_offset()));
808 andi(tmp, tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
809 bnez(tmp, slow_case);
810 }
811
812 if (UseFastLocking) {
813 ld(tmp, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
814 fast_lock(obj_reg, tmp, t0, t1, slow_case);
815 j(count);
816 } else {
817 // Load (object->mark() | 1) into swap_reg
818 ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
819 ori(swap_reg, t0, 1);
820
821 // Save (object->mark() | 1) into BasicLock's displaced header
822 sd(swap_reg, Address(lock_reg, mark_offset));
823
824 assert(lock_offset == 0,
825 "displached header must be first word in BasicObjectLock");
826
827 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, t0, count, /*fallthrough*/NULL);
828
829 // Test if the oopMark is an obvious stack pointer, i.e.,
830 // 1) (mark & 7) == 0, and
831 // 2) sp <= mark < mark + os::pagesize()
832 //
833 // These 3 tests can be done by evaluating the following
834 // expression: ((mark - sp) & (7 - os::vm_page_size())),
835 // assuming both stack pointer and pagesize have their
836 // least significant 3 bits clear.
837 // NOTE: the oopMark is in swap_reg x10 as the result of cmpxchg
838 sub(swap_reg, swap_reg, sp);
839 mv(t0, (int64_t)(7 - (int)os::vm_page_size()));
840 andr(swap_reg, swap_reg, t0);
841
842 // Save the test result, for recursive case, the result is zero
843 sd(swap_reg, Address(lock_reg, mark_offset));
844 beqz(swap_reg, count);
845 }
846
847 bind(slow_case);
848
849 // Call the runtime routine for slow case
850 call_VM(noreg,
851 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
852 UseFastLocking ? obj_reg : lock_reg);
853
854 j(done);
855
856 bind(count);
857 increment(Address(xthread, JavaThread::held_monitor_count_offset()));
858
859 bind(done);
860 }
861 }
862
863
864 // Unlocks an object. Used in monitorexit bytecode and
865 // remove_activation. Throws an IllegalMonitorException if object is
866 // not locked by current thread.
867 //
868 // Args:
869 // c_rarg1: BasicObjectLock for lock
870 //
871 // Kills:
872 // x10
873 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
874 // t0, t1 (temp regs)
875 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
876 {
877 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
878
879 if (UseHeavyMonitors) {
880 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
881 } else {
882 Label count, done;
883
884 const Register swap_reg = x10;
885 const Register header_reg = c_rarg2; // Will contain the old oopMark
886 const Register obj_reg = c_rarg3; // Will contain the oop
887
888 save_bcp(); // Save in case of exception
889
890 if (UseFastLocking) {
891 Label slow_case;
892 // Load oop into obj_reg(c_rarg3)
893 ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
894
895 // Free entry
896 sd(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
897
898 // Check for non-symmetric locking. This is allowed by the spec and the interpreter
899 // must handle it.
900 Register tmp = header_reg;
901 ld(tmp, Address(xthread, JavaThread::lock_stack_current_offset()));
902 ld(tmp, Address(tmp, -oopSize));
903 bne(tmp, obj_reg, slow_case);
904
905 ld(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
906 fast_unlock(obj_reg, header_reg, swap_reg, t0, slow_case);
907 j(count);
908
909 bind(slow_case);
910 } else {
911 // Convert from BasicObjectLock structure to object and BasicLock
912 // structure Store the BasicLock address into x10
913 la(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
914
915 // Load oop into obj_reg(c_rarg3)
916 ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
917
918 // Free entry
919 sd(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
920
921 // Load the old header from BasicLock structure
922 ld(header_reg, Address(swap_reg,
923 BasicLock::displaced_header_offset_in_bytes()));
924
925 // Test for recursion
926 beqz(header_reg, count);
927
928 // Atomic swap back the old header
929 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, t0, count, /*fallthrough*/NULL);
930 }
931
932 // Call the runtime routine for slow case.
933 sd(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
934 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
935
936 j(done);
937
938 bind(count);
939 decrement(Address(xthread, JavaThread::held_monitor_count_offset()));
940
941 bind(done);
942
943 restore_bcp();
944 }
945 }
946
947
948 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
949 Label& zero_continue) {
950 assert(ProfileInterpreter, "must be profiling interpreter");
951 ld(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
952 beqz(mdp, zero_continue);
953 }
954
955 // Set the method data pointer for the current bcp.
956 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
957 assert(ProfileInterpreter, "must be profiling interpreter");
958 Label set_mdp;
959 push_reg(RegSet::of(x10, x11), sp); // save x10, x11
960
961 // Test MDO to avoid the call if it is NULL.
962 ld(x10, Address(xmethod, in_bytes(Method::method_data_offset())));
963 beqz(x10, set_mdp);
964 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), xmethod, xbcp);
965 // x10: mdi
966 // mdo is guaranteed to be non-zero here, we checked for it before the call.
967 ld(x11, Address(xmethod, in_bytes(Method::method_data_offset())));
968 la(x11, Address(x11, in_bytes(MethodData::data_offset())));
969 add(x10, x11, x10);
970 sd(x10, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
971 bind(set_mdp);
972 pop_reg(RegSet::of(x10, x11), sp);
973 }
974
975 void InterpreterMacroAssembler::verify_method_data_pointer() {
976 assert(ProfileInterpreter, "must be profiling interpreter");
977 #ifdef ASSERT
978 Label verify_continue;
979 add(sp, sp, -4 * wordSize);
980 sd(x10, Address(sp, 0));
981 sd(x11, Address(sp, wordSize));
982 sd(x12, Address(sp, 2 * wordSize));
983 sd(x13, Address(sp, 3 * wordSize));
984 test_method_data_pointer(x13, verify_continue); // If mdp is zero, continue
985 get_method(x11);
986
987 // If the mdp is valid, it will point to a DataLayout header which is
988 // consistent with the bcp. The converse is highly probable also.
989 lh(x12, Address(x13, in_bytes(DataLayout::bci_offset())));
990 ld(t0, Address(x11, Method::const_offset()));
991 add(x12, x12, t0);
992 la(x12, Address(x12, ConstMethod::codes_offset()));
993 beq(x12, xbcp, verify_continue);
994 // x10: method
995 // xbcp: bcp // xbcp == 22
996 // x13: mdp
997 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
998 x11, xbcp, x13);
999 bind(verify_continue);
1000 ld(x10, Address(sp, 0));
1001 ld(x11, Address(sp, wordSize));
1002 ld(x12, Address(sp, 2 * wordSize));
1003 ld(x13, Address(sp, 3 * wordSize));
1004 add(sp, sp, 4 * wordSize);
1005 #endif // ASSERT
1006 }
1007
1008
1009 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1010 int constant,
1011 Register value) {
1012 assert(ProfileInterpreter, "must be profiling interpreter");
1013 Address data(mdp_in, constant);
1014 sd(value, data);
1015 }
1016
1017
1018 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1019 int constant,
1020 bool decrement) {
1021 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1022 }
1023
1024 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1025 Register reg,
1026 int constant,
1027 bool decrement) {
1028 assert(ProfileInterpreter, "must be profiling interpreter");
1029 // %%% this does 64bit counters at best it is wasting space
1030 // at worst it is a rare bug when counters overflow
1031
1032 assert_different_registers(t1, t0, mdp_in, reg);
1033
1034 Address addr1(mdp_in, constant);
1035 Address addr2(t1, 0);
1036 Address &addr = addr1;
1037 if (reg != noreg) {
1038 la(t1, addr1);
1039 add(t1, t1, reg);
1040 addr = addr2;
1041 }
1042
1043 if (decrement) {
1044 ld(t0, addr);
1045 addi(t0, t0, -DataLayout::counter_increment);
1046 Label L;
1047 bltz(t0, L); // skip store if counter underflow
1048 sd(t0, addr);
1049 bind(L);
1050 } else {
1051 assert(DataLayout::counter_increment == 1,
1052 "flow-free idiom only works with 1");
1053 ld(t0, addr);
1054 addi(t0, t0, DataLayout::counter_increment);
1055 Label L;
1056 blez(t0, L); // skip store if counter overflow
1057 sd(t0, addr);
1058 bind(L);
1059 }
1060 }
1061
1062 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1063 int flag_byte_constant) {
1064 assert(ProfileInterpreter, "must be profiling interpreter");
1065 int flags_offset = in_bytes(DataLayout::flags_offset());
1066 // Set the flag
1067 lbu(t1, Address(mdp_in, flags_offset));
1068 ori(t1, t1, flag_byte_constant);
1069 sb(t1, Address(mdp_in, flags_offset));
1070 }
1071
1072
1073 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1074 int offset,
1075 Register value,
1076 Register test_value_out,
1077 Label& not_equal_continue) {
1078 assert(ProfileInterpreter, "must be profiling interpreter");
1079 if (test_value_out == noreg) {
1080 ld(t1, Address(mdp_in, offset));
1081 bne(value, t1, not_equal_continue);
1082 } else {
1083 // Put the test value into a register, so caller can use it:
1084 ld(test_value_out, Address(mdp_in, offset));
1085 bne(value, test_value_out, not_equal_continue);
1086 }
1087 }
1088
1089
1090 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1091 int offset_of_disp) {
1092 assert(ProfileInterpreter, "must be profiling interpreter");
1093 ld(t1, Address(mdp_in, offset_of_disp));
1094 add(mdp_in, mdp_in, t1);
1095 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1096 }
1097
1098 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1099 Register reg,
1100 int offset_of_disp) {
1101 assert(ProfileInterpreter, "must be profiling interpreter");
1102 add(t1, mdp_in, reg);
1103 ld(t1, Address(t1, offset_of_disp));
1104 add(mdp_in, mdp_in, t1);
1105 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1106 }
1107
1108
1109 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1110 int constant) {
1111 assert(ProfileInterpreter, "must be profiling interpreter");
1112 addi(mdp_in, mdp_in, (unsigned)constant);
1113 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1114 }
1115
1116
1117 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1118 assert(ProfileInterpreter, "must be profiling interpreter");
1119
1120 // save/restore across call_VM
1121 addi(sp, sp, -2 * wordSize);
1122 sd(zr, Address(sp, 0));
1123 sd(return_bci, Address(sp, wordSize));
1124 call_VM(noreg,
1125 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1126 return_bci);
1127 ld(zr, Address(sp, 0));
1128 ld(return_bci, Address(sp, wordSize));
1129 addi(sp, sp, 2 * wordSize);
1130 }
1131
1132 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1133 Register bumped_count) {
1134 if (ProfileInterpreter) {
1135 Label profile_continue;
1136
1137 // If no method data exists, go to profile_continue.
1138 // Otherwise, assign to mdp
1139 test_method_data_pointer(mdp, profile_continue);
1140
1141 // We are taking a branch. Increment the taken count.
1142 Address data(mdp, in_bytes(JumpData::taken_offset()));
1143 ld(bumped_count, data);
1144 assert(DataLayout::counter_increment == 1,
1145 "flow-free idiom only works with 1");
1146 addi(bumped_count, bumped_count, DataLayout::counter_increment);
1147 Label L;
1148 // eg: bumped_count=0x7fff ffff ffff ffff + 1 < 0. so we use <= 0;
1149 blez(bumped_count, L); // skip store if counter overflow,
1150 sd(bumped_count, data);
1151 bind(L);
1152 // The method data pointer needs to be updated to reflect the new target.
1153 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1154 bind(profile_continue);
1155 }
1156 }
1157
1158 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1159 if (ProfileInterpreter) {
1160 Label profile_continue;
1161
1162 // If no method data exists, go to profile_continue.
1163 test_method_data_pointer(mdp, profile_continue);
1164
1165 // We are taking a branch. Increment the not taken count.
1166 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1167
1168 // The method data pointer needs to be updated to correspond to
1169 // the next bytecode
1170 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1171 bind(profile_continue);
1172 }
1173 }
1174
1175 void InterpreterMacroAssembler::profile_call(Register mdp) {
1176 if (ProfileInterpreter) {
1177 Label profile_continue;
1178
1179 // If no method data exists, go to profile_continue.
1180 test_method_data_pointer(mdp, profile_continue);
1181
1182 // We are making a call. Increment the count.
1183 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1184
1185 // The method data pointer needs to be updated to reflect the new target.
1186 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1187 bind(profile_continue);
1188 }
1189 }
1190
1191 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1192 if (ProfileInterpreter) {
1193 Label profile_continue;
1194
1195 // If no method data exists, go to profile_continue.
1196 test_method_data_pointer(mdp, profile_continue);
1197
1198 // We are making a call. Increment the count.
1199 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1200
1201 // The method data pointer needs to be updated to reflect the new target.
1202 update_mdp_by_constant(mdp,
1203 in_bytes(VirtualCallData::
1204 virtual_call_data_size()));
1205 bind(profile_continue);
1206 }
1207 }
1208
1209
1210 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1211 Register mdp,
1212 Register reg2,
1213 bool receiver_can_be_null) {
1214 if (ProfileInterpreter) {
1215 Label profile_continue;
1216
1217 // If no method data exists, go to profile_continue.
1218 test_method_data_pointer(mdp, profile_continue);
1219
1220 Label skip_receiver_profile;
1221 if (receiver_can_be_null) {
1222 Label not_null;
1223 // We are making a call. Increment the count for null receiver.
1224 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1225 j(skip_receiver_profile);
1226 bind(not_null);
1227 }
1228
1229 // Record the receiver type.
1230 record_klass_in_profile(receiver, mdp, reg2, true);
1231 bind(skip_receiver_profile);
1232
1233 // The method data pointer needs to be updated to reflect the new target.
1234
1235 update_mdp_by_constant(mdp,
1236 in_bytes(VirtualCallData::
1237 virtual_call_data_size()));
1238 bind(profile_continue);
1239 }
1240 }
1241
1242 // This routine creates a state machine for updating the multi-row
1243 // type profile at a virtual call site (or other type-sensitive bytecode).
1244 // The machine visits each row (of receiver/count) until the receiver type
1245 // is found, or until it runs out of rows. At the same time, it remembers
1246 // the location of the first empty row. (An empty row records null for its
1247 // receiver, and can be allocated for a newly-observed receiver type.)
1248 // Because there are two degrees of freedom in the state, a simple linear
1249 // search will not work; it must be a decision tree. Hence this helper
1250 // function is recursive, to generate the required tree structured code.
1251 // It's the interpreter, so we are trading off code space for speed.
1252 // See below for example code.
1253 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1254 Register receiver, Register mdp,
1255 Register reg2,
1256 Label& done, bool is_virtual_call) {
1257 if (TypeProfileWidth == 0) {
1258 if (is_virtual_call) {
1259 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1260 }
1261
1262 } else {
1263 int non_profiled_offset = -1;
1264 if (is_virtual_call) {
1265 non_profiled_offset = in_bytes(CounterData::count_offset());
1266 }
1267
1268 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1269 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1270 }
1271 }
1272
1273 void InterpreterMacroAssembler::record_item_in_profile_helper(
1274 Register item, Register mdp, Register reg2, int start_row, Label& done, int total_rows,
1275 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, int non_profiled_offset) {
1276 int last_row = total_rows - 1;
1277 assert(start_row <= last_row, "must be work left to do");
1278 // Test this row for both the item and for null.
1279 // Take any of three different outcomes:
1280 // 1. found item => increment count and goto done
1281 // 2. found null => keep looking for case 1, maybe allocate this cell
1282 // 3. found something else => keep looking for cases 1 and 2
1283 // Case 3 is handled by a recursive call.
1284 for (int row = start_row; row <= last_row; row++) {
1285 Label next_test;
1286 bool test_for_null_also = (row == start_row);
1287
1288 // See if the item is item[n].
1289 int item_offset = in_bytes(item_offset_fn(row));
1290 test_mdp_data_at(mdp, item_offset, item,
1291 (test_for_null_also ? reg2 : noreg),
1292 next_test);
1293 // (Reg2 now contains the item from the CallData.)
1294
1295 // The item is item[n]. Increment count[n].
1296 int count_offset = in_bytes(item_count_offset_fn(row));
1297 increment_mdp_data_at(mdp, count_offset);
1298 j(done);
1299 bind(next_test);
1300
1301 if (test_for_null_also) {
1302 Label found_null;
1303 // Failed the equality check on item[n]... Test for null.
1304 if (start_row == last_row) {
1305 // The only thing left to do is handle the null case.
1306 if (non_profiled_offset >= 0) {
1307 beqz(reg2, found_null);
1308 // Item did not match any saved item and there is no empty row for it.
1309 // Increment total counter to indicate polymorphic case.
1310 increment_mdp_data_at(mdp, non_profiled_offset);
1311 j(done);
1312 bind(found_null);
1313 } else {
1314 bnez(reg2, done);
1315 }
1316 break;
1317 }
1318 // Since null is rare, make it be the branch-taken case.
1319 beqz(reg2, found_null);
1320
1321 // Put all the "Case 3" tests here.
1322 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1323 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1324
1325 // Found a null. Keep searching for a matching item,
1326 // but remember that this is an empty (unused) slot.
1327 bind(found_null);
1328 }
1329 }
1330
1331 // In the fall-through case, we found no matching item, but we
1332 // observed the item[start_row] is NULL.
1333 // Fill in the item field and increment the count.
1334 int item_offset = in_bytes(item_offset_fn(start_row));
1335 set_mdp_data_at(mdp, item_offset, item);
1336 int count_offset = in_bytes(item_count_offset_fn(start_row));
1337 mv(reg2, DataLayout::counter_increment);
1338 set_mdp_data_at(mdp, count_offset, reg2);
1339 if (start_row > 0) {
1340 j(done);
1341 }
1342 }
1343
1344 // Example state machine code for three profile rows:
1345 // # main copy of decision tree, rooted at row[1]
1346 // if (row[0].rec == rec) then [
1347 // row[0].incr()
1348 // goto done
1349 // ]
1350 // if (row[0].rec != NULL) then [
1351 // # inner copy of decision tree, rooted at row[1]
1352 // if (row[1].rec == rec) then [
1353 // row[1].incr()
1354 // goto done
1355 // ]
1356 // if (row[1].rec != NULL) then [
1357 // # degenerate decision tree, rooted at row[2]
1358 // if (row[2].rec == rec) then [
1359 // row[2].incr()
1360 // goto done
1361 // ]
1362 // if (row[2].rec != NULL) then [
1363 // count.incr()
1364 // goto done
1365 // ] # overflow
1366 // row[2].init(rec)
1367 // goto done
1368 // ] else [
1369 // # remember row[1] is empty
1370 // if (row[2].rec == rec) then [
1371 // row[2].incr()
1372 // goto done
1373 // ]
1374 // row[1].init(rec)
1375 // goto done
1376 // ]
1377 // else [
1378 // # remember row[0] is empty
1379 // if (row[1].rec == rec) then [
1380 // row[1].incr()
1381 // goto done
1382 // ]
1383 // if (row[2].rec == rec) then [
1384 // row[2].incr()
1385 // goto done
1386 // ]
1387 // row[0].init(rec)
1388 // goto done
1389 // ]
1390 // done:
1391
1392 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1393 Register mdp, Register reg2,
1394 bool is_virtual_call) {
1395 assert(ProfileInterpreter, "must be profiling");
1396 Label done;
1397
1398 record_klass_in_profile_helper(receiver, mdp, reg2, done, is_virtual_call);
1399
1400 bind(done);
1401 }
1402
1403 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1404 if (ProfileInterpreter) {
1405 Label profile_continue;
1406
1407 // If no method data exists, go to profile_continue.
1408 test_method_data_pointer(mdp, profile_continue);
1409
1410 // Update the total ret count.
1411 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1412
1413 for (uint row = 0; row < RetData::row_limit(); row++) {
1414 Label next_test;
1415
1416 // See if return_bci is equal to bci[n]:
1417 test_mdp_data_at(mdp,
1418 in_bytes(RetData::bci_offset(row)),
1419 return_bci, noreg,
1420 next_test);
1421
1422 // return_bci is equal to bci[n]. Increment the count.
1423 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1424
1425 // The method data pointer needs to be updated to reflect the new target.
1426 update_mdp_by_offset(mdp,
1427 in_bytes(RetData::bci_displacement_offset(row)));
1428 j(profile_continue);
1429 bind(next_test);
1430 }
1431
1432 update_mdp_for_ret(return_bci);
1433
1434 bind(profile_continue);
1435 }
1436 }
1437
1438 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1439 if (ProfileInterpreter) {
1440 Label profile_continue;
1441
1442 // If no method data exists, go to profile_continue.
1443 test_method_data_pointer(mdp, profile_continue);
1444
1445 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1446
1447 // The method data pointer needs to be updated.
1448 int mdp_delta = in_bytes(BitData::bit_data_size());
1449 if (TypeProfileCasts) {
1450 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1451 }
1452 update_mdp_by_constant(mdp, mdp_delta);
1453
1454 bind(profile_continue);
1455 }
1456 }
1457
1458 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1459 if (ProfileInterpreter && TypeProfileCasts) {
1460 Label profile_continue;
1461
1462 // If no method data exists, go to profile_continue.
1463 test_method_data_pointer(mdp, profile_continue);
1464
1465 int count_offset = in_bytes(CounterData::count_offset());
1466 // Back up the address, since we have already bumped the mdp.
1467 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1468
1469 // *Decrement* the counter. We expect to see zero or small negatives.
1470 increment_mdp_data_at(mdp, count_offset, true);
1471
1472 bind (profile_continue);
1473 }
1474 }
1475
1476 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1477 if (ProfileInterpreter) {
1478 Label profile_continue;
1479
1480 // If no method data exists, go to profile_continue.
1481 test_method_data_pointer(mdp, profile_continue);
1482
1483 // The method data pointer needs to be updated.
1484 int mdp_delta = in_bytes(BitData::bit_data_size());
1485 if (TypeProfileCasts) {
1486 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1487
1488 // Record the object type.
1489 record_klass_in_profile(klass, mdp, reg2, false);
1490 }
1491 update_mdp_by_constant(mdp, mdp_delta);
1492
1493 bind(profile_continue);
1494 }
1495 }
1496
1497 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1498 if (ProfileInterpreter) {
1499 Label profile_continue;
1500
1501 // If no method data exists, go to profile_continue.
1502 test_method_data_pointer(mdp, profile_continue);
1503
1504 // Update the default case count
1505 increment_mdp_data_at(mdp,
1506 in_bytes(MultiBranchData::default_count_offset()));
1507
1508 // The method data pointer needs to be updated.
1509 update_mdp_by_offset(mdp,
1510 in_bytes(MultiBranchData::
1511 default_displacement_offset()));
1512
1513 bind(profile_continue);
1514 }
1515 }
1516
1517 void InterpreterMacroAssembler::profile_switch_case(Register index,
1518 Register mdp,
1519 Register reg2) {
1520 if (ProfileInterpreter) {
1521 Label profile_continue;
1522
1523 // If no method data exists, go to profile_continue.
1524 test_method_data_pointer(mdp, profile_continue);
1525
1526 // Build the base (index * per_case_size_in_bytes()) +
1527 // case_array_offset_in_bytes()
1528 mv(reg2, in_bytes(MultiBranchData::per_case_size()));
1529 mv(t0, in_bytes(MultiBranchData::case_array_offset()));
1530 Assembler::mul(index, index, reg2);
1531 Assembler::add(index, index, t0);
1532
1533 // Update the case count
1534 increment_mdp_data_at(mdp,
1535 index,
1536 in_bytes(MultiBranchData::relative_count_offset()));
1537
1538 // The method data pointer need to be updated.
1539 update_mdp_by_offset(mdp,
1540 index,
1541 in_bytes(MultiBranchData::
1542 relative_displacement_offset()));
1543
1544 bind(profile_continue);
1545 }
1546 }
1547
1548 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1549
1550 void InterpreterMacroAssembler::notify_method_entry() {
1551 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1552 // track stack depth. If it is possible to enter interp_only_mode we add
1553 // the code to check if the event should be sent.
1554 if (JvmtiExport::can_post_interpreter_events()) {
1555 Label L;
1556 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1557 beqz(x13, L);
1558 call_VM(noreg, CAST_FROM_FN_PTR(address,
1559 InterpreterRuntime::post_method_entry));
1560 bind(L);
1561 }
1562
1563 {
1564 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1565 get_method(c_rarg1);
1566 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1567 xthread, c_rarg1);
1568 }
1569
1570 // RedefineClasses() tracing support for obsolete method entry
1571 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1572 get_method(c_rarg1);
1573 call_VM_leaf(
1574 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1575 xthread, c_rarg1);
1576 }
1577 }
1578
1579
1580 void InterpreterMacroAssembler::notify_method_exit(
1581 TosState state, NotifyMethodExitMode mode) {
1582 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1583 // track stack depth. If it is possible to enter interp_only_mode we add
1584 // the code to check if the event should be sent.
1585 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1586 Label L;
1587 // Note: frame::interpreter_frame_result has a dependency on how the
1588 // method result is saved across the call to post_method_exit. If this
1589 // is changed then the interpreter_frame_result implementation will
1590 // need to be updated too.
1591
1592 // template interpreter will leave the result on the top of the stack.
1593 push(state);
1594 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1595 beqz(x13, L);
1596 call_VM(noreg,
1597 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1598 bind(L);
1599 pop(state);
1600 }
1601
1602 {
1603 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1604 push(state);
1605 get_method(c_rarg1);
1606 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1607 xthread, c_rarg1);
1608 pop(state);
1609 }
1610 }
1611
1612
1613 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1614 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1615 int increment, Address mask,
1616 Register tmp1, Register tmp2,
1617 bool preloaded, Label* where) {
1618 Label done;
1619 if (!preloaded) {
1620 lwu(tmp1, counter_addr);
1621 }
1622 add(tmp1, tmp1, increment);
1623 sw(tmp1, counter_addr);
1624 lwu(tmp2, mask);
1625 andr(tmp1, tmp1, tmp2);
1626 bnez(tmp1, done);
1627 j(*where); // offset is too large so we have to use j instead of beqz here
1628 bind(done);
1629 }
1630
1631 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1632 int number_of_arguments) {
1633 // interpreter specific
1634 //
1635 // Note: No need to save/restore rbcp & rlocals pointer since these
1636 // are callee saved registers and no blocking/ GC can happen
1637 // in leaf calls.
1638 #ifdef ASSERT
1639 {
1640 Label L;
1641 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1642 beqz(t0, L);
1643 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1644 " last_sp != NULL");
1645 bind(L);
1646 }
1647 #endif /* ASSERT */
1648 // super call
1649 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1650 }
1651
1652 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1653 Register java_thread,
1654 Register last_java_sp,
1655 address entry_point,
1656 int number_of_arguments,
1657 bool check_exceptions) {
1658 // interpreter specific
1659 //
1660 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1661 // really make a difference for these runtime calls, since they are
1662 // slow anyway. Btw., bcp must be saved/restored since it may change
1663 // due to GC.
1664 save_bcp();
1665 #ifdef ASSERT
1666 {
1667 Label L;
1668 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1669 beqz(t0, L);
1670 stop("InterpreterMacroAssembler::call_VM_base:"
1671 " last_sp != NULL");
1672 bind(L);
1673 }
1674 #endif /* ASSERT */
1675 // super call
1676 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1677 entry_point, number_of_arguments,
1678 check_exceptions);
1679 // interpreter specific
1680 restore_bcp();
1681 restore_locals();
1682 }
1683
1684 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1685 assert_different_registers(obj, tmp, t0, mdo_addr.base());
1686 Label update, next, none;
1687
1688 verify_oop(obj);
1689
1690 bnez(obj, update);
1691 orptr(mdo_addr, TypeEntries::null_seen, t0, tmp);
1692 j(next);
1693
1694 bind(update);
1695 load_klass(obj, obj);
1696
1697 ld(t0, mdo_addr);
1698 xorr(obj, obj, t0);
1699 andi(t0, obj, TypeEntries::type_klass_mask);
1700 beqz(t0, next); // klass seen before, nothing to
1701 // do. The unknown bit may have been
1702 // set already but no need to check.
1703
1704 andi(t0, obj, TypeEntries::type_unknown);
1705 bnez(t0, next);
1706 // already unknown. Nothing to do anymore.
1707
1708 ld(t0, mdo_addr);
1709 beqz(t0, none);
1710 mv(tmp, (u1)TypeEntries::null_seen);
1711 beq(t0, tmp, none);
1712 // There is a chance that the checks above (re-reading profiling
1713 // data from memory) fail if another thread has just set the
1714 // profiling to this obj's klass
1715 ld(t0, mdo_addr);
1716 xorr(obj, obj, t0);
1717 andi(t0, obj, TypeEntries::type_klass_mask);
1718 beqz(t0, next);
1719
1720 // different than before. Cannot keep accurate profile.
1721 orptr(mdo_addr, TypeEntries::type_unknown, t0, tmp);
1722 j(next);
1723
1724 bind(none);
1725 // first time here. Set profile type.
1726 sd(obj, mdo_addr);
1727
1728 bind(next);
1729 }
1730
1731 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1732 if (!ProfileInterpreter) {
1733 return;
1734 }
1735
1736 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1737 Label profile_continue;
1738
1739 test_method_data_pointer(mdp, profile_continue);
1740
1741 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1742
1743 lbu(t0, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1744 if (is_virtual) {
1745 mv(tmp, (u1)DataLayout::virtual_call_type_data_tag);
1746 bne(t0, tmp, profile_continue);
1747 } else {
1748 mv(tmp, (u1)DataLayout::call_type_data_tag);
1749 bne(t0, tmp, profile_continue);
1750 }
1751
1752 // calculate slot step
1753 static int stack_slot_offset0 = in_bytes(TypeEntriesAtCall::stack_slot_offset(0));
1754 static int slot_step = in_bytes(TypeEntriesAtCall::stack_slot_offset(1)) - stack_slot_offset0;
1755
1756 // calculate type step
1757 static int argument_type_offset0 = in_bytes(TypeEntriesAtCall::argument_type_offset(0));
1758 static int type_step = in_bytes(TypeEntriesAtCall::argument_type_offset(1)) - argument_type_offset0;
1759
1760 if (MethodData::profile_arguments()) {
1761 Label done, loop, loopEnd, profileArgument, profileReturnType;
1762 RegSet pushed_registers;
1763 pushed_registers += x15;
1764 pushed_registers += x16;
1765 pushed_registers += x17;
1766 Register mdo_addr = x15;
1767 Register index = x16;
1768 Register off_to_args = x17;
1769 push_reg(pushed_registers, sp);
1770
1771 mv(off_to_args, in_bytes(TypeEntriesAtCall::args_data_offset()));
1772 mv(t0, TypeProfileArgsLimit);
1773 beqz(t0, loopEnd);
1774
1775 mv(index, zr); // index < TypeProfileArgsLimit
1776 bind(loop);
1777 bgtz(index, profileReturnType);
1778 mv(t0, (int)MethodData::profile_return());
1779 beqz(t0, profileArgument); // (index > 0 || MethodData::profile_return()) == false
1780 bind(profileReturnType);
1781 // If return value type is profiled we may have no argument to profile
1782 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1783 mv(t1, - TypeStackSlotEntries::per_arg_count());
1784 mul(t1, index, t1);
1785 add(tmp, tmp, t1);
1786 mv(t1, TypeStackSlotEntries::per_arg_count());
1787 add(t0, mdp, off_to_args);
1788 blt(tmp, t1, done);
1789
1790 bind(profileArgument);
1791
1792 ld(tmp, Address(callee, Method::const_offset()));
1793 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1794 // stack offset o (zero based) from the start of the argument
1795 // list, for n arguments translates into offset n - o - 1 from
1796 // the end of the argument list
1797 mv(t0, stack_slot_offset0);
1798 mv(t1, slot_step);
1799 mul(t1, index, t1);
1800 add(t0, t0, t1);
1801 add(t0, mdp, t0);
1802 ld(t0, Address(t0));
1803 sub(tmp, tmp, t0);
1804 addi(tmp, tmp, -1);
1805 Address arg_addr = argument_address(tmp);
1806 ld(tmp, arg_addr);
1807
1808 mv(t0, argument_type_offset0);
1809 mv(t1, type_step);
1810 mul(t1, index, t1);
1811 add(t0, t0, t1);
1812 add(mdo_addr, mdp, t0);
1813 Address mdo_arg_addr(mdo_addr, 0);
1814 profile_obj_type(tmp, mdo_arg_addr, t1);
1815
1816 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1817 addi(off_to_args, off_to_args, to_add);
1818
1819 // increment index by 1
1820 addi(index, index, 1);
1821 mv(t1, TypeProfileArgsLimit);
1822 blt(index, t1, loop);
1823 bind(loopEnd);
1824
1825 if (MethodData::profile_return()) {
1826 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1827 addi(tmp, tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1828 }
1829
1830 add(t0, mdp, off_to_args);
1831 bind(done);
1832 mv(mdp, t0);
1833
1834 // unspill the clobbered registers
1835 pop_reg(pushed_registers, sp);
1836
1837 if (MethodData::profile_return()) {
1838 // We're right after the type profile for the last
1839 // argument. tmp is the number of cells left in the
1840 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1841 // if there's a return to profile.
1842 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1843 shadd(mdp, tmp, mdp, tmp, exact_log2(DataLayout::cell_size));
1844 }
1845 sd(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1846 } else {
1847 assert(MethodData::profile_return(), "either profile call args or call ret");
1848 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1849 }
1850
1851 // mdp points right after the end of the
1852 // CallTypeData/VirtualCallTypeData, right after the cells for the
1853 // return value type if there's one
1854
1855 bind(profile_continue);
1856 }
1857 }
1858
1859 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1860 assert_different_registers(mdp, ret, tmp, xbcp, t0, t1);
1861 if (ProfileInterpreter && MethodData::profile_return()) {
1862 Label profile_continue, done;
1863
1864 test_method_data_pointer(mdp, profile_continue);
1865
1866 if (MethodData::profile_return_jsr292_only()) {
1867 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1868
1869 // If we don't profile all invoke bytecodes we must make sure
1870 // it's a bytecode we indeed profile. We can't go back to the
1871 // beginning of the ProfileData we intend to update to check its
1872 // type because we're right after it and we don't known its
1873 // length
1874 Label do_profile;
1875 lbu(t0, Address(xbcp, 0));
1876 mv(tmp, (u1)Bytecodes::_invokedynamic);
1877 beq(t0, tmp, do_profile);
1878 mv(tmp, (u1)Bytecodes::_invokehandle);
1879 beq(t0, tmp, do_profile);
1880 get_method(tmp);
1881 lhu(t0, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1882 mv(t1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1883 bne(t0, t1, profile_continue);
1884 bind(do_profile);
1885 }
1886
1887 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1888 mv(tmp, ret);
1889 profile_obj_type(tmp, mdo_ret_addr, t1);
1890
1891 bind(profile_continue);
1892 }
1893 }
1894
1895 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2, Register tmp3) {
1896 assert_different_registers(t0, t1, mdp, tmp1, tmp2, tmp3);
1897 if (ProfileInterpreter && MethodData::profile_parameters()) {
1898 Label profile_continue, done;
1899
1900 test_method_data_pointer(mdp, profile_continue);
1901
1902 // Load the offset of the area within the MDO used for
1903 // parameters. If it's negative we're not profiling any parameters
1904 lwu(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1905 srli(tmp2, tmp1, 31);
1906 bnez(tmp2, profile_continue); // i.e. sign bit set
1907
1908 // Compute a pointer to the area for parameters from the offset
1909 // and move the pointer to the slot for the last
1910 // parameters. Collect profiling from last parameter down.
1911 // mdo start + parameters offset + array length - 1
1912 add(mdp, mdp, tmp1);
1913 ld(tmp1, Address(mdp, ArrayData::array_len_offset()));
1914 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1915
1916 Label loop;
1917 bind(loop);
1918
1919 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1920 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1921 int per_arg_scale = exact_log2(DataLayout::cell_size);
1922 add(t0, mdp, off_base);
1923 add(t1, mdp, type_base);
1924
1925 shadd(tmp2, tmp1, t0, tmp2, per_arg_scale);
1926 // load offset on the stack from the slot for this parameter
1927 ld(tmp2, Address(tmp2, 0));
1928 neg(tmp2, tmp2);
1929
1930 // read the parameter from the local area
1931 shadd(tmp2, tmp2, xlocals, tmp2, Interpreter::logStackElementSize);
1932 ld(tmp2, Address(tmp2, 0));
1933
1934 // profile the parameter
1935 shadd(t1, tmp1, t1, t0, per_arg_scale);
1936 Address arg_type(t1, 0);
1937 profile_obj_type(tmp2, arg_type, tmp3);
1938
1939 // go to next parameter
1940 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1941 bgez(tmp1, loop);
1942
1943 bind(profile_continue);
1944 }
1945 }
1946
1947 void InterpreterMacroAssembler::get_method_counters(Register method,
1948 Register mcs, Label& skip) {
1949 Label has_counters;
1950 ld(mcs, Address(method, Method::method_counters_offset()));
1951 bnez(mcs, has_counters);
1952 call_VM(noreg, CAST_FROM_FN_PTR(address,
1953 InterpreterRuntime::build_method_counters), method);
1954 ld(mcs, Address(method, Method::method_counters_offset()));
1955 beqz(mcs, skip); // No MethodCounters allocated, OutOfMemory
1956 bind(has_counters);
1957 }
1958
1959 #ifdef ASSERT
1960 void InterpreterMacroAssembler::verify_access_flags(Register access_flags, uint32_t flag_bits,
1961 const char* msg, bool stop_by_hit) {
1962 Label L;
1963 andi(t0, access_flags, flag_bits);
1964 if (stop_by_hit) {
1965 beqz(t0, L);
1966 } else {
1967 bnez(t0, L);
1968 }
1969 stop(msg);
1970 bind(L);
1971 }
1972
1973 void InterpreterMacroAssembler::verify_frame_setup() {
1974 Label L;
1975 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1976 ld(t0, monitor_block_top);
1977 beq(esp, t0, L);
1978 stop("broken stack frame setup in interpreter");
1979 bind(L);
1980 }
1981 #endif