1 /*
2 * Copyright (c) 2003, 2020, 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/safepointMechanism.hpp"
44 #include "runtime/sharedRuntime.hpp"
45 #include "runtime/thread.inline.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 mvw(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 int32_t offset = 0;
186 la_patchable(xdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
187 addi(xdispatch, xdispatch, offset);
188 }
189
190 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
191 int bcp_offset,
192 size_t index_size) {
193 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
194 if (index_size == sizeof(u2)) {
195 load_unsigned_short(index, Address(xbcp, bcp_offset));
196 } else if (index_size == sizeof(u4)) {
197 lwu(index, Address(xbcp, bcp_offset));
198 // Check if the secondary index definition is still ~x, otherwise
199 // we have to change the following assembler code to calculate the
200 // plain index.
201 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
202 xori(index, index, -1);
203 addw(index, index, zr);
204 } else if (index_size == sizeof(u1)) {
205 load_unsigned_byte(index, Address(xbcp, bcp_offset));
206 } else {
207 ShouldNotReachHere();
208 }
209 }
210
211 // Return
212 // Rindex: index into constant pool
213 // Rcache: address of cache entry - ConstantPoolCache::base_offset()
214 //
215 // A caller must add ConstantPoolCache::base_offset() to Rcache to get
216 // the true address of the cache entry.
217 //
218 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
219 Register index,
220 int bcp_offset,
221 size_t index_size) {
222 assert_different_registers(cache, index);
223 assert_different_registers(cache, xcpool);
224 get_cache_index_at_bcp(index, bcp_offset, index_size);
225 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
226 // Convert from field index to ConstantPoolCacheEntry
227 // riscv already has the cache in xcpool so there is no need to
228 // install it in cache. Instead we pre-add the indexed offset to
229 // xcpool and return it in cache. All clients of this method need to
230 // be modified accordingly.
231 shadd(cache, index, xcpool, cache, 5);
232 }
233
234
235 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
236 Register index,
237 Register bytecode,
238 int byte_no,
239 int bcp_offset,
240 size_t index_size) {
241 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
242 // We use a 32-bit load here since the layout of 64-bit words on
243 // little-endian machines allow us that.
244 // n.b. unlike x86 cache already includes the index offset
245 la(bytecode, Address(cache,
246 ConstantPoolCache::base_offset() +
247 ConstantPoolCacheEntry::indices_offset()));
248 membar(MacroAssembler::AnyAny);
249 lwu(bytecode, bytecode);
250 membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
251 const int shift_count = (1 + byte_no) * BitsPerByte;
252 slli(bytecode, bytecode, XLEN - (shift_count + BitsPerByte));
253 srli(bytecode, bytecode, XLEN - BitsPerByte);
254 }
255
256 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
257 Register tmp,
258 int bcp_offset,
259 size_t index_size) {
260 assert(cache != tmp, "must use different register");
261 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
262 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
263 // Convert from field index to ConstantPoolCacheEntry index
264 // and from word offset to byte offset
265 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord,
266 "else change next line");
267 ld(cache, Address(fp, frame::interpreter_frame_cache_offset * wordSize));
268 // skip past the header
269 add(cache, cache, in_bytes(ConstantPoolCache::base_offset()));
270 // construct pointer to cache entry
271 shadd(cache, tmp, cache, tmp, 2 + LogBytesPerWord);
272 }
273
274 // Load object from cpool->resolved_references(index)
275 void InterpreterMacroAssembler::load_resolved_reference_at_index(
276 Register result, Register index, Register tmp) {
277 assert_different_registers(result, index);
278
279 get_constant_pool(result);
280 // Load pointer for resolved_references[] objArray
281 ld(result, Address(result, ConstantPool::cache_offset_in_bytes()));
282 ld(result, Address(result, ConstantPoolCache::resolved_references_offset_in_bytes()));
283 resolve_oop_handle(result, tmp);
284 // Add in the index
285 addi(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
286 shadd(result, index, result, index, LogBytesPerHeapOop);
287 load_heap_oop(result, Address(result, 0));
288 }
289
290 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
291 Register cpool, Register index, Register klass, Register temp) {
292 shadd(temp, index, cpool, temp, LogBytesPerWord);
293 lhu(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
294 ld(klass, Address(cpool, ConstantPool::resolved_klasses_offset_in_bytes())); // klass = cpool->_resolved_klasses
295 shadd(klass, temp, klass, temp, LogBytesPerWord);
296 ld(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
297 }
298
299 void InterpreterMacroAssembler::load_resolved_method_at_index(int byte_no,
300 Register method,
301 Register cache) {
302 const int method_offset = in_bytes(
303 ConstantPoolCache::base_offset() +
304 ((byte_no == TemplateTable::f2_byte)
305 ? ConstantPoolCacheEntry::f2_offset()
306 : ConstantPoolCacheEntry::f1_offset()));
307
308 ld(method, Address(cache, method_offset)); // get f1 Method*
309 }
310
311 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
312 // subtype of super_klass.
313 //
314 // Args:
315 // x10: superklass
316 // Rsub_klass: subklass
317 //
318 // Kills:
319 // x12, x15
320 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
321 Label& ok_is_subtype) {
322 assert(Rsub_klass != x10, "x10 holds superklass");
323 assert(Rsub_klass != x12, "x12 holds 2ndary super array length");
324 assert(Rsub_klass != x15, "x15 holds 2ndary super array scan ptr");
325
326 // Profile the not-null value's klass.
327 profile_typecheck(x12, Rsub_klass, x15); // blows x12, reloads x15
328
329 // Do the check.
330 check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12
331
332 // Profile the failure of the check.
333 profile_typecheck_failed(x12); // blows x12
334 }
335
336 // Java Expression Stack
337
338 void InterpreterMacroAssembler::pop_ptr(Register r) {
339 ld(r, Address(esp, 0));
340 addi(esp, esp, wordSize);
341 }
342
343 void InterpreterMacroAssembler::pop_i(Register r) {
344 lw(r, Address(esp, 0)); // lw do signed extended
345 addi(esp, esp, wordSize);
346 }
347
348 void InterpreterMacroAssembler::pop_l(Register r) {
349 ld(r, Address(esp, 0));
350 addi(esp, esp, 2 * Interpreter::stackElementSize);
351 }
352
353 void InterpreterMacroAssembler::push_ptr(Register r) {
354 addi(esp, esp, -wordSize);
355 sd(r, Address(esp, 0));
356 }
357
358 void InterpreterMacroAssembler::push_i(Register r) {
359 addi(esp, esp, -wordSize);
360 addw(r, r, zr); // signed extended
361 sd(r, Address(esp, 0));
362 }
363
364 void InterpreterMacroAssembler::push_l(Register r) {
365 addi(esp, esp, -2 * wordSize);
366 sd(zr, Address(esp, wordSize));
367 sd(r, Address(esp));
368 }
369
370 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
371 flw(r, esp, 0);
372 addi(esp, esp, wordSize);
373 }
374
375 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
376 fld(r, esp, 0);
377 addi(esp, esp, 2 * Interpreter::stackElementSize);
378 }
379
380 void InterpreterMacroAssembler::push_f(FloatRegister r) {
381 addi(esp, esp, -wordSize);
382 fsw(r, Address(esp, 0));
383 }
384
385 void InterpreterMacroAssembler::push_d(FloatRegister r) {
386 addi(esp, esp, -2 * wordSize);
387 fsd(r, Address(esp, 0));
388 }
389
390 void InterpreterMacroAssembler::pop(TosState state) {
391 switch (state) {
392 case atos:
393 pop_ptr();
394 verify_oop(x10);
395 break;
396 case btos: // fall through
397 case ztos: // fall through
398 case ctos: // fall through
399 case stos: // fall through
400 case itos:
401 pop_i();
402 break;
403 case ltos:
404 pop_l();
405 break;
406 case ftos:
407 pop_f();
408 break;
409 case dtos:
410 pop_d();
411 break;
412 case vtos:
413 /* nothing to do */
414 break;
415 default:
416 ShouldNotReachHere();
417 }
418 }
419
420 void InterpreterMacroAssembler::push(TosState state) {
421 switch (state) {
422 case atos:
423 verify_oop(x10);
424 push_ptr();
425 break;
426 case btos: // fall through
427 case ztos: // fall through
428 case ctos: // fall through
429 case stos: // fall through
430 case itos:
431 push_i();
432 break;
433 case ltos:
434 push_l();
435 break;
436 case ftos:
437 push_f();
438 break;
439 case dtos:
440 push_d();
441 break;
442 case vtos:
443 /* nothing to do */
444 break;
445 default:
446 ShouldNotReachHere();
447 }
448 }
449
450 // Helpers for swap and dup
451 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
452 ld(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
453 }
454
455 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
456 sd(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
457 }
458
459 void InterpreterMacroAssembler::load_float(Address src) {
460 flw(f10, src);
461 }
462
463 void InterpreterMacroAssembler::load_double(Address src) {
464 fld(f10, src);
465 }
466
467 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
468 // set sender sp
469 mv(x30, sp);
470 // record last_sp
471 sd(esp, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
472 }
473
474 // Jump to from_interpreted entry of a call unless single stepping is possible
475 // in this thread in which case we must call the i2i entry
476 void InterpreterMacroAssembler::jump_from_interpreted(Register method) {
477 prepare_to_jump_from_interpreted();
478 if (JvmtiExport::can_post_interpreter_events()) {
479 Label run_compiled_code;
480 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
481 // compiled code in threads for which the event is enabled. Check here for
482 // interp_only_mode if these events CAN be enabled.
483 lwu(t0, Address(xthread, JavaThread::interp_only_mode_offset()));
484 beqz(t0, run_compiled_code);
485 ld(t0, Address(method, Method::interpreter_entry_offset()));
486 jr(t0);
487 bind(run_compiled_code);
488 }
489
490 ld(t0, Address(method, Method::from_interpreted_offset()));
491 jr(t0);
492 }
493
494 // The following two routines provide a hook so that an implementation
495 // can schedule the dispatch in two parts. amd64 does not do this.
496 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
497 }
498
499 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
500 dispatch_next(state, step);
501 }
502
503 void InterpreterMacroAssembler::dispatch_base(TosState state,
504 address* table,
505 bool verifyoop,
506 bool generate_poll,
507 Register Rs) {
508 // Pay attention to the argument Rs, which is acquiesce in t0.
509 if (VerifyActivationFrameSize) {
510 Unimplemented();
511 }
512 if (verifyoop && state == atos) {
513 verify_oop(x10);
514 }
515
516 Label safepoint;
517 address* const safepoint_table = Interpreter::safept_table(state);
518 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
519
520 if (needs_thread_local_poll) {
521 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
522 ld(t1, Address(xthread, JavaThread::polling_word_offset()));
523 andi(t1, t1, SafepointMechanism::poll_bit());
524 bnez(t1, safepoint);
525 }
526 if (table == Interpreter::dispatch_table(state)) {
527 li(t1, Interpreter::distance_from_dispatch_table(state));
528 add(t1, Rs, t1);
529 shadd(t1, t1, xdispatch, t1, 3);
530 } else {
531 mv(t1, (address)table);
532 shadd(t1, Rs, t1, Rs, 3);
533 }
534 ld(t1, Address(t1));
535 jr(t1);
536
537 if (needs_thread_local_poll) {
538 bind(safepoint);
539 la(t1, ExternalAddress((address)safepoint_table));
540 shadd(t1, Rs, t1, Rs, 3);
541 ld(t1, Address(t1));
542 jr(t1);
543 }
544 }
545
546 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll, Register Rs) {
547 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll, Rs);
548 }
549
550 void InterpreterMacroAssembler::dispatch_only_normal(TosState state, Register Rs) {
551 dispatch_base(state, Interpreter::normal_table(state), Rs);
552 }
553
554 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state, Register Rs) {
555 dispatch_base(state, Interpreter::normal_table(state), false, Rs);
556 }
557
558 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
559 // load next bytecode
560 load_unsigned_byte(t0, Address(xbcp, step));
561 add(xbcp, xbcp, step);
562 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
563 }
564
565 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
566 // load current bytecode
567 lbu(t0, Address(xbcp, 0));
568 dispatch_base(state, table);
569 }
570
571 // remove activation
572 //
573 // Apply stack watermark barrier.
574 // Unlock the receiver if this is a synchronized method.
575 // Unlock any Java monitors from syncronized blocks.
576 // Remove the activation from the stack.
577 //
578 // If there are locked Java monitors
579 // If throw_monitor_exception
580 // throws IllegalMonitorStateException
581 // Else if install_monitor_exception
582 // installs IllegalMonitorStateException
583 // Else
584 // no error processing
585 void InterpreterMacroAssembler::remove_activation(
586 TosState state,
587 bool throw_monitor_exception,
588 bool install_monitor_exception,
589 bool notify_jvmdi) {
590 // Note: Registers x13 may be in use for the
591 // result check if synchronized method
592 Label unlocked, unlock, no_unlock;
593
594 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
595 // that would normally not be safe to use. Such bad returns into unsafe territory of
596 // the stack, will call InterpreterRuntime::at_unwind.
597 Label slow_path;
598 Label fast_path;
599 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
600 j(fast_path);
601
602 bind(slow_path);
603 push(state);
604 set_last_Java_frame(esp, fp, (address)pc(), t0);
605 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), xthread);
606 reset_last_Java_frame(true);
607 pop(state);
608
609 bind(fast_path);
610
611 // get the value of _do_not_unlock_if_synchronized into x13
612 const Address do_not_unlock_if_synchronized(xthread,
613 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
614 lbu(x13, do_not_unlock_if_synchronized);
615 sb(zr, do_not_unlock_if_synchronized); // reset the flag
616
617 // get method access flags
618 ld(x11, Address(fp, frame::interpreter_frame_method_offset * wordSize));
619 ld(x12, Address(x11, Method::access_flags_offset()));
620 andi(t0, x12, JVM_ACC_SYNCHRONIZED);
621 beqz(t0, unlocked);
622
623 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
624 // is set.
625 bnez(x13, no_unlock);
626
627 // unlock monitor
628 push(state); // save result
629
630 // BasicObjectLock will be first in list, since this is a
631 // synchronized method. However, need to check that the object has
632 // not been unlocked by an explicit monitorexit bytecode.
633 const Address monitor(fp, frame::interpreter_frame_initial_sp_offset *
634 wordSize - (int) sizeof(BasicObjectLock));
635 // We use c_rarg1 so that if we go slow path it will be the correct
636 // register for unlock_object to pass to VM directly
637 la(c_rarg1, monitor); // address of first monitor
638
639 ld(x10, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
640 bnez(x10, unlock);
641
642 pop(state);
643 if (throw_monitor_exception) {
644 // Entry already unlocked, need to throw exception
645 call_VM(noreg, CAST_FROM_FN_PTR(address,
646 InterpreterRuntime::throw_illegal_monitor_state_exception));
647 should_not_reach_here();
648 } else {
649 // Monitor already unlocked during a stack unroll. If requested,
650 // install an illegal_monitor_state_exception. Continue with
651 // stack unrolling.
652 if (install_monitor_exception) {
653 call_VM(noreg, CAST_FROM_FN_PTR(address,
654 InterpreterRuntime::new_illegal_monitor_state_exception));
655 }
656 j(unlocked);
657 }
658
659 bind(unlock);
660 unlock_object(c_rarg1);
661 pop(state);
662
663 // Check that for block-structured locking (i.e., that all locked
664 // objects has been unlocked)
665 bind(unlocked);
666
667 // x10: Might contain return value
668
669 // Check that all monitors are unlocked
670 {
671 Label loop, exception, entry, restart;
672 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
673 const Address monitor_block_top(
674 fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
675 const Address monitor_block_bot(
676 fp, frame::interpreter_frame_initial_sp_offset * wordSize);
677
678 bind(restart);
679 // We use c_rarg1 so that if we go slow path it will be the correct
680 // register for unlock_object to pass to VM directly
681 ld(c_rarg1, monitor_block_top); // points to current entry, starting
682 // with top-most entry
683 la(x9, monitor_block_bot); // points to word before bottom of
684 // monitor block
685
686 j(entry);
687
688 // Entry already locked, need to throw exception
689 bind(exception);
690
691 if (throw_monitor_exception) {
692 // Throw exception
693 MacroAssembler::call_VM(noreg,
694 CAST_FROM_FN_PTR(address, InterpreterRuntime::
695 throw_illegal_monitor_state_exception));
696
697 should_not_reach_here();
698 } else {
699 // Stack unrolling. Unlock object and install illegal_monitor_exception.
700 // Unlock does not block, so don't have to worry about the frame.
701 // We don't have to preserve c_rarg1 since we are going to throw an exception.
702
703 push(state);
704 unlock_object(c_rarg1);
705 pop(state);
706
707 if (install_monitor_exception) {
708 call_VM(noreg, CAST_FROM_FN_PTR(address,
709 InterpreterRuntime::
710 new_illegal_monitor_state_exception));
711 }
712
713 j(restart);
714 }
715
716 bind(loop);
717 // check if current entry is used
718 add(t0, c_rarg1, BasicObjectLock::obj_offset_in_bytes());
719 ld(t0, Address(t0, 0));
720 bnez(t0, exception);
721
722 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
723 bind(entry);
724 bne(c_rarg1, x9, loop); // check if bottom reached if not at bottom then check this entry
725 }
726
727 bind(no_unlock);
728
729 // jvmti support
730 if (notify_jvmdi) {
731 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
732
733 } else {
734 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
735 }
736
737 // remove activation
738 // get sender esp
739 ld(t1,
740 Address(fp, frame::interpreter_frame_sender_sp_offset * wordSize));
741 if (StackReservedPages > 0) {
742 // testing if reserved zone needs to be re-enabled
743 Label no_reserved_zone_enabling;
744
745 ld(t0, Address(xthread, JavaThread::reserved_stack_activation_offset()));
746 ble(t1, t0, no_reserved_zone_enabling);
747
748 call_VM_leaf(
749 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), xthread);
750 call_VM(noreg, CAST_FROM_FN_PTR(address,
751 InterpreterRuntime::throw_delayed_StackOverflowError));
752 should_not_reach_here();
753
754 bind(no_reserved_zone_enabling);
755 }
756
757 // restore sender esp
758 mv(esp, t1);
759
760 // remove frame anchor
761 leave();
762 // If we're returning to interpreted code we will shortly be
763 // adjusting SP to allow some space for ESP. If we're returning to
764 // compiled code the saved sender SP was saved in sender_sp, so this
765 // restores it.
766 andi(sp, esp, -16);
767 }
768
769 // Lock object
770 //
771 // Args:
772 // c_rarg1: BasicObjectLock to be used for locking
773 //
774 // Kills:
775 // x10
776 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
777 // t0, t1 (temp regs)
778 void InterpreterMacroAssembler::lock_object(Register lock_reg)
779 {
780 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
781 if (UseHeavyMonitors) {
782 call_VM(noreg,
783 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
784 lock_reg);
785 } else {
786 Label done;
787
788 const Register swap_reg = x10;
789 const Register tmp = c_rarg2;
790 const Register obj_reg = c_rarg3; // Will contain the oop
791
792 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
793 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
794 const int mark_offset = lock_offset +
795 BasicLock::displaced_header_offset_in_bytes();
796
797 Label slow_case;
798
799 // Load object pointer into obj_reg c_rarg3
800 ld(obj_reg, Address(lock_reg, obj_offset));
801
802 if (DiagnoseSyncOnValueBasedClasses != 0) {
803 load_klass(tmp, obj_reg);
804 lwu(tmp, Address(tmp, Klass::access_flags_offset()));
805 andi(tmp, tmp, JVM_ACC_IS_VALUE_BASED_CLASS);
806 bnez(tmp, slow_case);
807 }
808
809 if (UseBiasedLocking) {
810 biased_locking_enter(lock_reg, obj_reg, swap_reg, tmp, false, done, &slow_case);
811 }
812
813 // Load (object->mark() | 1) into swap_reg
814 ld(t0, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
815 ori(swap_reg, t0, 1);
816
817 // Save (object->mark() | 1) into BasicLock's displaced header
818 sd(swap_reg, Address(lock_reg, mark_offset));
819
820 assert(lock_offset == 0,
821 "displached header must be first word in BasicObjectLock");
822
823 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, t0, done, /*fallthrough*/NULL);
824
825 // Test if the oopMark is an obvious stack pointer, i.e.,
826 // 1) (mark & 7) == 0, and
827 // 2) sp <= mark < mark + os::pagesize()
828 //
829 // These 3 tests can be done by evaluating the following
830 // expression: ((mark - sp) & (7 - os::vm_page_size())),
831 // assuming both stack pointer and pagesize have their
832 // least significant 3 bits clear.
833 // NOTE: the oopMark is in swap_reg x10 as the result of cmpxchg
834 sub(swap_reg, swap_reg, sp);
835 li(t0, (int64_t)(7 - os::vm_page_size()));
836 andr(swap_reg, swap_reg, t0);
837
838 // Save the test result, for recursive case, the result is zero
839 sd(swap_reg, Address(lock_reg, mark_offset));
840 beqz(swap_reg, done);
841
842 bind(slow_case);
843
844 // Call the runtime routine for slow case
845 call_VM(noreg,
846 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
847 lock_reg);
848
849 bind(done);
850 }
851 }
852
853
854 // Unlocks an object. Used in monitorexit bytecode and
855 // remove_activation. Throws an IllegalMonitorException if object is
856 // not locked by current thread.
857 //
858 // Args:
859 // c_rarg1: BasicObjectLock for lock
860 //
861 // Kills:
862 // x10
863 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
864 // t0, t1 (temp regs)
865 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
866 {
867 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
868
869 if (UseHeavyMonitors) {
870 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
871 } else {
872 Label done;
873
874 const Register swap_reg = x10;
875 const Register header_reg = c_rarg2; // Will contain the old oopMark
876 const Register obj_reg = c_rarg3; // Will contain the oop
877
878 save_bcp(); // Save in case of exception
879
880 // Convert from BasicObjectLock structure to object and BasicLock
881 // structure Store the BasicLock address into x10
882 la(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
883
884 // Load oop into obj_reg(c_rarg3)
885 ld(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
886
887 // Free entry
888 sd(zr, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
889
890 if (UseBiasedLocking) {
891 biased_locking_exit(obj_reg, header_reg, done);
892 }
893
894 // Load the old header from BasicLock structure
895 ld(header_reg, Address(swap_reg,
896 BasicLock::displaced_header_offset_in_bytes()));
897
898 // Test for recursion
899 beqz(header_reg, done);
900
901 // Atomic swap back the old header
902 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, t0, done, /*fallthrough*/NULL);
903
904 // Call the runtime routine for slow case.
905 sd(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes())); // restore obj
906 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
907
908 bind(done);
909
910 restore_bcp();
911 }
912 }
913
914
915 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
916 Label& zero_continue) {
917 assert(ProfileInterpreter, "must be profiling interpreter");
918 ld(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
919 beqz(mdp, zero_continue);
920 }
921
922 // Set the method data pointer for the current bcp.
923 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
924 assert(ProfileInterpreter, "must be profiling interpreter");
925 Label set_mdp;
926 push_reg(0xc00, sp); // save x10, x11
927
928 // Test MDO to avoid the call if it is NULL.
929 ld(x10, Address(xmethod, in_bytes(Method::method_data_offset())));
930 beqz(x10, set_mdp);
931 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), xmethod, xbcp);
932 // x10: mdi
933 // mdo is guaranteed to be non-zero here, we checked for it before the call.
934 ld(x11, Address(xmethod, in_bytes(Method::method_data_offset())));
935 la(x11, Address(x11, in_bytes(MethodData::data_offset())));
936 add(x10, x11, x10);
937 sd(x10, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
938 bind(set_mdp);
939 pop_reg(0xc00, sp);
940 }
941
942 void InterpreterMacroAssembler::verify_method_data_pointer() {
943 assert(ProfileInterpreter, "must be profiling interpreter");
944 #ifdef ASSERT
945 Label verify_continue;
946 add(sp, sp, -4 * wordSize);
947 sd(x10, Address(sp, 0));
948 sd(x11, Address(sp, wordSize));
949 sd(x12, Address(sp, 2 * wordSize));
950 sd(x13, Address(sp, 3 * wordSize));
951 test_method_data_pointer(x13, verify_continue); // If mdp is zero, continue
952 get_method(x11);
953
954 // If the mdp is valid, it will point to a DataLayout header which is
955 // consistent with the bcp. The converse is highly probable also.
956 lh(x12, Address(x13, in_bytes(DataLayout::bci_offset())));
957 ld(t0, Address(x11, Method::const_offset()));
958 add(x12, x12, t0);
959 la(x12, Address(x12, ConstMethod::codes_offset()));
960 beq(x12, xbcp, verify_continue);
961 // x10: method
962 // xbcp: bcp // xbcp == 22
963 // x13: mdp
964 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
965 x11, xbcp, x13);
966 bind(verify_continue);
967 ld(x10, Address(sp, 0));
968 ld(x11, Address(sp, wordSize));
969 ld(x12, Address(sp, 2 * wordSize));
970 ld(x13, Address(sp, 3 * wordSize));
971 add(sp, sp, 4 * wordSize);
972 #endif // ASSERT
973 }
974
975
976 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
977 int constant,
978 Register value) {
979 assert(ProfileInterpreter, "must be profiling interpreter");
980 Address data(mdp_in, constant);
981 sd(value, data);
982 }
983
984
985 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
986 int constant,
987 bool decrement) {
988 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
989 }
990
991 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
992 Register reg,
993 int constant,
994 bool decrement) {
995 assert(ProfileInterpreter, "must be profiling interpreter");
996 // %%% this does 64bit counters at best it is wasting space
997 // at worst it is a rare bug when counters overflow
998
999 assert_different_registers(t1, t0, mdp_in, reg);
1000
1001 Address addr1(mdp_in, constant);
1002 Address addr2(t1, 0);
1003 Address &addr = addr1;
1004 if (reg != noreg) {
1005 la(t1, addr1);
1006 add(t1, t1, reg);
1007 addr = addr2;
1008 }
1009
1010 if (decrement) {
1011 ld(t0, addr);
1012 addi(t0, t0, -DataLayout::counter_increment);
1013 Label L;
1014 bltz(t0, L); // skip store if counter underflow
1015 sd(t0, addr);
1016 bind(L);
1017 } else {
1018 assert(DataLayout::counter_increment == 1,
1019 "flow-free idiom only works with 1");
1020 ld(t0, addr);
1021 addi(t0, t0, DataLayout::counter_increment);
1022 Label L;
1023 blez(t0, L); // skip store if counter overflow
1024 sd(t0, addr);
1025 bind(L);
1026 }
1027 }
1028
1029 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1030 int flag_byte_constant) {
1031 assert(ProfileInterpreter, "must be profiling interpreter");
1032 int flags_offset = in_bytes(DataLayout::flags_offset());
1033 // Set the flag
1034 lbu(t1, Address(mdp_in, flags_offset));
1035 ori(t1, t1, flag_byte_constant);
1036 sb(t1, Address(mdp_in, flags_offset));
1037 }
1038
1039
1040 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1041 int offset,
1042 Register value,
1043 Register test_value_out,
1044 Label& not_equal_continue) {
1045 assert(ProfileInterpreter, "must be profiling interpreter");
1046 if (test_value_out == noreg) {
1047 ld(t1, Address(mdp_in, offset));
1048 bne(value, t1, not_equal_continue);
1049 } else {
1050 // Put the test value into a register, so caller can use it:
1051 ld(test_value_out, Address(mdp_in, offset));
1052 bne(value, test_value_out, not_equal_continue);
1053 }
1054 }
1055
1056
1057 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1058 int offset_of_disp) {
1059 assert(ProfileInterpreter, "must be profiling interpreter");
1060 ld(t1, Address(mdp_in, offset_of_disp));
1061 add(mdp_in, mdp_in, t1);
1062 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1063 }
1064
1065 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1066 Register reg,
1067 int offset_of_disp) {
1068 assert(ProfileInterpreter, "must be profiling interpreter");
1069 add(t1, mdp_in, reg);
1070 ld(t1, Address(t1, offset_of_disp));
1071 add(mdp_in, mdp_in, t1);
1072 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1073 }
1074
1075
1076 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1077 int constant) {
1078 assert(ProfileInterpreter, "must be profiling interpreter");
1079 addi(mdp_in, mdp_in, (unsigned)constant);
1080 sd(mdp_in, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1081 }
1082
1083
1084 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1085 assert(ProfileInterpreter, "must be profiling interpreter");
1086
1087 // save/restore across call_VM
1088 addi(sp, sp, -2 * wordSize);
1089 sd(zr, Address(sp, 0));
1090 sd(return_bci, Address(sp, wordSize));
1091 call_VM(noreg,
1092 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1093 return_bci);
1094 ld(zr, Address(sp, 0));
1095 ld(return_bci, Address(sp, wordSize));
1096 addi(sp, sp, 2 * wordSize);
1097 }
1098
1099 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1100 Register bumped_count) {
1101 if (ProfileInterpreter) {
1102 Label profile_continue;
1103
1104 // If no method data exists, go to profile_continue.
1105 // Otherwise, assign to mdp
1106 test_method_data_pointer(mdp, profile_continue);
1107
1108 // We are taking a branch. Increment the taken count.
1109 Address data(mdp, in_bytes(JumpData::taken_offset()));
1110 ld(bumped_count, data);
1111 assert(DataLayout::counter_increment == 1,
1112 "flow-free idiom only works with 1");
1113 addi(bumped_count, bumped_count, DataLayout::counter_increment);
1114 Label L;
1115 // eg: bumped_count=0x7fff ffff ffff ffff + 1 < 0. so we use <= 0;
1116 blez(bumped_count, L); // skip store if counter overflow,
1117 sd(bumped_count, data);
1118 bind(L);
1119 // The method data pointer needs to be updated to reflect the new target.
1120 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1121 bind(profile_continue);
1122 }
1123 }
1124
1125 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1126 if (ProfileInterpreter) {
1127 Label profile_continue;
1128
1129 // If no method data exists, go to profile_continue.
1130 test_method_data_pointer(mdp, profile_continue);
1131
1132 // We are taking a branch. Increment the not taken count.
1133 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1134
1135 // The method data pointer needs to be updated to correspond to
1136 // the next bytecode
1137 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1138 bind(profile_continue);
1139 }
1140 }
1141
1142 void InterpreterMacroAssembler::profile_call(Register mdp) {
1143 if (ProfileInterpreter) {
1144 Label profile_continue;
1145
1146 // If no method data exists, go to profile_continue.
1147 test_method_data_pointer(mdp, profile_continue);
1148
1149 // We are making a call. Increment the count.
1150 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1151
1152 // The method data pointer needs to be updated to reflect the new target.
1153 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1154 bind(profile_continue);
1155 }
1156 }
1157
1158 void InterpreterMacroAssembler::profile_final_call(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 making a call. Increment the count.
1166 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1167
1168 // The method data pointer needs to be updated to reflect the new target.
1169 update_mdp_by_constant(mdp,
1170 in_bytes(VirtualCallData::
1171 virtual_call_data_size()));
1172 bind(profile_continue);
1173 }
1174 }
1175
1176
1177 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1178 Register mdp,
1179 Register reg2,
1180 bool receiver_can_be_null) {
1181 if (ProfileInterpreter) {
1182 Label profile_continue;
1183
1184 // If no method data exists, go to profile_continue.
1185 test_method_data_pointer(mdp, profile_continue);
1186
1187 Label skip_receiver_profile;
1188 if (receiver_can_be_null) {
1189 Label not_null;
1190 // We are making a call. Increment the count for null receiver.
1191 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1192 j(skip_receiver_profile);
1193 bind(not_null);
1194 }
1195
1196 // Record the receiver type.
1197 record_klass_in_profile(receiver, mdp, reg2, true);
1198 bind(skip_receiver_profile);
1199
1200 // The method data pointer needs to be updated to reflect the new target.
1201
1202 update_mdp_by_constant(mdp,
1203 in_bytes(VirtualCallData::
1204 virtual_call_data_size()));
1205 bind(profile_continue);
1206 }
1207 }
1208
1209 // This routine creates a state machine for updating the multi-row
1210 // type profile at a virtual call site (or other type-sensitive bytecode).
1211 // The machine visits each row (of receiver/count) until the receiver type
1212 // is found, or until it runs out of rows. At the same time, it remembers
1213 // the location of the first empty row. (An empty row records null for its
1214 // receiver, and can be allocated for a newly-observed receiver type.)
1215 // Because there are two degrees of freedom in the state, a simple linear
1216 // search will not work; it must be a decision tree. Hence this helper
1217 // function is recursive, to generate the required tree structured code.
1218 // It's the interpreter, so we are trading off code space for speed.
1219 // See below for example code.
1220 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1221 Register receiver, Register mdp,
1222 Register reg2,
1223 Label& done, bool is_virtual_call) {
1224 if (TypeProfileWidth == 0) {
1225 if (is_virtual_call) {
1226 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1227 }
1228
1229 } else {
1230 int non_profiled_offset = -1;
1231 if (is_virtual_call) {
1232 non_profiled_offset = in_bytes(CounterData::count_offset());
1233 }
1234
1235 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1236 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset, non_profiled_offset);
1237 }
1238 }
1239
1240 void InterpreterMacroAssembler::record_item_in_profile_helper(
1241 Register item, Register mdp, Register reg2, int start_row, Label& done, int total_rows,
1242 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn, int non_profiled_offset) {
1243 int last_row = total_rows - 1;
1244 assert(start_row <= last_row, "must be work left to do");
1245 // Test this row for both the item and for null.
1246 // Take any of three different outcomes:
1247 // 1. found item => increment count and goto done
1248 // 2. found null => keep looking for case 1, maybe allocate this cell
1249 // 3. found something else => keep looking for cases 1 and 2
1250 // Case 3 is handled by a recursive call.
1251 for (int row = start_row; row <= last_row; row++) {
1252 Label next_test;
1253 bool test_for_null_also = (row == start_row);
1254
1255 // See if the item is item[n].
1256 int item_offset = in_bytes(item_offset_fn(row));
1257 test_mdp_data_at(mdp, item_offset, item,
1258 (test_for_null_also ? reg2 : noreg),
1259 next_test);
1260 // (Reg2 now contains the item from the CallData.)
1261
1262 // The item is item[n]. Increment count[n].
1263 int count_offset = in_bytes(item_count_offset_fn(row));
1264 increment_mdp_data_at(mdp, count_offset);
1265 j(done);
1266 bind(next_test);
1267
1268 if (test_for_null_also) {
1269 Label found_null;
1270 // Failed the equality check on item[n]... Test for null.
1271 if (start_row == last_row) {
1272 // The only thing left to do is handle the null case.
1273 if (non_profiled_offset >= 0) {
1274 beqz(reg2, found_null);
1275 // Item did not match any saved item and there is no empty row for it.
1276 // Increment total counter to indicate polymorphic case.
1277 increment_mdp_data_at(mdp, non_profiled_offset);
1278 j(done);
1279 bind(found_null);
1280 } else {
1281 bnez(reg2, done);
1282 }
1283 break;
1284 }
1285 // Since null is rare, make it be the branch-taken case.
1286 beqz(reg2, found_null);
1287
1288 // Put all the "Case 3" tests here.
1289 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1290 item_offset_fn, item_count_offset_fn, non_profiled_offset);
1291
1292 // Found a null. Keep searching for a matching item,
1293 // but remember that this is an empty (unused) slot.
1294 bind(found_null);
1295 }
1296 }
1297
1298 // In the fall-through case, we found no matching item, but we
1299 // observed the item[start_row] is NULL.
1300 // Fill in the item field and increment the count.
1301 int item_offset = in_bytes(item_offset_fn(start_row));
1302 set_mdp_data_at(mdp, item_offset, item);
1303 int count_offset = in_bytes(item_count_offset_fn(start_row));
1304 mv(reg2, DataLayout::counter_increment);
1305 set_mdp_data_at(mdp, count_offset, reg2);
1306 if (start_row > 0) {
1307 j(done);
1308 }
1309 }
1310
1311 // Example state machine code for three profile rows:
1312 // # main copy of decision tree, rooted at row[1]
1313 // if (row[0].rec == rec) then [
1314 // row[0].incr()
1315 // goto done
1316 // ]
1317 // if (row[0].rec != NULL) then [
1318 // # inner copy of decision tree, rooted at row[1]
1319 // if (row[1].rec == rec) then [
1320 // row[1].incr()
1321 // goto done
1322 // ]
1323 // if (row[1].rec != NULL) then [
1324 // # degenerate decision tree, rooted at row[2]
1325 // if (row[2].rec == rec) then [
1326 // row[2].incr()
1327 // goto done
1328 // ]
1329 // if (row[2].rec != NULL) then [
1330 // count.incr()
1331 // goto done
1332 // ] # overflow
1333 // row[2].init(rec)
1334 // goto done
1335 // ] else [
1336 // # remember row[1] is empty
1337 // if (row[2].rec == rec) then [
1338 // row[2].incr()
1339 // goto done
1340 // ]
1341 // row[1].init(rec)
1342 // goto done
1343 // ]
1344 // else [
1345 // # remember row[0] is empty
1346 // if (row[1].rec == rec) then [
1347 // row[1].incr()
1348 // goto done
1349 // ]
1350 // if (row[2].rec == rec) then [
1351 // row[2].incr()
1352 // goto done
1353 // ]
1354 // row[0].init(rec)
1355 // goto done
1356 // ]
1357 // done:
1358
1359 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1360 Register mdp, Register reg2,
1361 bool is_virtual_call) {
1362 assert(ProfileInterpreter, "must be profiling");
1363 Label done;
1364
1365 record_klass_in_profile_helper(receiver, mdp, reg2, done, is_virtual_call);
1366
1367 bind(done);
1368 }
1369
1370 void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
1371 if (ProfileInterpreter) {
1372 Label profile_continue;
1373
1374 // If no method data exists, go to profile_continue.
1375 test_method_data_pointer(mdp, profile_continue);
1376
1377 // Update the total ret count.
1378 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1379
1380 for (uint row = 0; row < RetData::row_limit(); row++) {
1381 Label next_test;
1382
1383 // See if return_bci is equal to bci[n]:
1384 test_mdp_data_at(mdp,
1385 in_bytes(RetData::bci_offset(row)),
1386 return_bci, noreg,
1387 next_test);
1388
1389 // return_bci is equal to bci[n]. Increment the count.
1390 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1391
1392 // The method data pointer needs to be updated to reflect the new target.
1393 update_mdp_by_offset(mdp,
1394 in_bytes(RetData::bci_displacement_offset(row)));
1395 j(profile_continue);
1396 bind(next_test);
1397 }
1398
1399 update_mdp_for_ret(return_bci);
1400
1401 bind(profile_continue);
1402 }
1403 }
1404
1405 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1406 if (ProfileInterpreter) {
1407 Label profile_continue;
1408
1409 // If no method data exists, go to profile_continue.
1410 test_method_data_pointer(mdp, profile_continue);
1411
1412 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1413
1414 // The method data pointer needs to be updated.
1415 int mdp_delta = in_bytes(BitData::bit_data_size());
1416 if (TypeProfileCasts) {
1417 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1418 }
1419 update_mdp_by_constant(mdp, mdp_delta);
1420
1421 bind(profile_continue);
1422 }
1423 }
1424
1425 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1426 if (ProfileInterpreter && TypeProfileCasts) {
1427 Label profile_continue;
1428
1429 // If no method data exists, go to profile_continue.
1430 test_method_data_pointer(mdp, profile_continue);
1431
1432 int count_offset = in_bytes(CounterData::count_offset());
1433 // Back up the address, since we have already bumped the mdp.
1434 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1435
1436 // *Decrement* the counter. We expect to see zero or small negatives.
1437 increment_mdp_data_at(mdp, count_offset, true);
1438
1439 bind (profile_continue);
1440 }
1441 }
1442
1443 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1444 if (ProfileInterpreter) {
1445 Label profile_continue;
1446
1447 // If no method data exists, go to profile_continue.
1448 test_method_data_pointer(mdp, profile_continue);
1449
1450 // The method data pointer needs to be updated.
1451 int mdp_delta = in_bytes(BitData::bit_data_size());
1452 if (TypeProfileCasts) {
1453 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1454
1455 // Record the object type.
1456 record_klass_in_profile(klass, mdp, reg2, false);
1457 }
1458 update_mdp_by_constant(mdp, mdp_delta);
1459
1460 bind(profile_continue);
1461 }
1462 }
1463
1464 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1465 if (ProfileInterpreter) {
1466 Label profile_continue;
1467
1468 // If no method data exists, go to profile_continue.
1469 test_method_data_pointer(mdp, profile_continue);
1470
1471 // Update the default case count
1472 increment_mdp_data_at(mdp,
1473 in_bytes(MultiBranchData::default_count_offset()));
1474
1475 // The method data pointer needs to be updated.
1476 update_mdp_by_offset(mdp,
1477 in_bytes(MultiBranchData::
1478 default_displacement_offset()));
1479
1480 bind(profile_continue);
1481 }
1482 }
1483
1484 void InterpreterMacroAssembler::profile_switch_case(Register index,
1485 Register mdp,
1486 Register reg2) {
1487 if (ProfileInterpreter) {
1488 Label profile_continue;
1489
1490 // If no method data exists, go to profile_continue.
1491 test_method_data_pointer(mdp, profile_continue);
1492
1493 // Build the base (index * per_case_size_in_bytes()) +
1494 // case_array_offset_in_bytes()
1495 mvw(reg2, in_bytes(MultiBranchData::per_case_size()));
1496 mvw(t0, in_bytes(MultiBranchData::case_array_offset()));
1497 Assembler::mul(index, index, reg2);
1498 Assembler::add(index, index, t0);
1499
1500 // Update the case count
1501 increment_mdp_data_at(mdp,
1502 index,
1503 in_bytes(MultiBranchData::relative_count_offset()));
1504
1505 // The method data pointer need to be updated.
1506 update_mdp_by_offset(mdp,
1507 index,
1508 in_bytes(MultiBranchData::
1509 relative_displacement_offset()));
1510
1511 bind(profile_continue);
1512 }
1513 }
1514
1515 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { ; }
1516
1517 void InterpreterMacroAssembler::notify_method_entry() {
1518 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1519 // track stack depth. If it is possible to enter interp_only_mode we add
1520 // the code to check if the event should be sent.
1521 if (JvmtiExport::can_post_interpreter_events()) {
1522 Label L;
1523 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1524 beqz(x13, L);
1525 call_VM(noreg, CAST_FROM_FN_PTR(address,
1526 InterpreterRuntime::post_method_entry));
1527 bind(L);
1528 }
1529
1530 {
1531 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1532 get_method(c_rarg1);
1533 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1534 xthread, c_rarg1);
1535 }
1536
1537 // RedefineClasses() tracing support for obsolete method entry
1538 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1539 get_method(c_rarg1);
1540 call_VM_leaf(
1541 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1542 xthread, c_rarg1);
1543 }
1544 }
1545
1546
1547 void InterpreterMacroAssembler::notify_method_exit(
1548 TosState state, NotifyMethodExitMode mode) {
1549 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1550 // track stack depth. If it is possible to enter interp_only_mode we add
1551 // the code to check if the event should be sent.
1552 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1553 Label L;
1554 // Note: frame::interpreter_frame_result has a dependency on how the
1555 // method result is saved across the call to post_method_exit. If this
1556 // is changed then the interpreter_frame_result implementation will
1557 // need to be updated too.
1558
1559 // template interpreter will leave the result on the top of the stack.
1560 push(state);
1561 lwu(x13, Address(xthread, JavaThread::interp_only_mode_offset()));
1562 beqz(x13, L);
1563 call_VM(noreg,
1564 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1565 bind(L);
1566 pop(state);
1567 }
1568
1569 {
1570 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1571 push(state);
1572 get_method(c_rarg1);
1573 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1574 xthread, c_rarg1);
1575 pop(state);
1576 }
1577 }
1578
1579
1580 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1581 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1582 int increment, Address mask,
1583 Register tmp1, Register tmp2,
1584 bool preloaded, Label* where) {
1585 Label done;
1586 if (!preloaded) {
1587 lwu(tmp1, counter_addr);
1588 }
1589 add(tmp1, tmp1, increment);
1590 sw(tmp1, counter_addr);
1591 lwu(tmp2, mask);
1592 andr(tmp1, tmp1, tmp2);
1593 bnez(tmp1, done);
1594 j(*where); // offset is too large so we have to use j instead of beqz here
1595 bind(done);
1596 }
1597
1598 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1599 int number_of_arguments) {
1600 // interpreter specific
1601 //
1602 // Note: No need to save/restore rbcp & rlocals pointer since these
1603 // are callee saved registers and no blocking/ GC can happen
1604 // in leaf calls.
1605 #ifdef ASSERT
1606 {
1607 Label L;
1608 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1609 beqz(t0, L);
1610 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1611 " last_sp != NULL");
1612 bind(L);
1613 }
1614 #endif /* ASSERT */
1615 // super call
1616 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1617 }
1618
1619 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1620 Register java_thread,
1621 Register last_java_sp,
1622 address entry_point,
1623 int number_of_arguments,
1624 bool check_exceptions) {
1625 // interpreter specific
1626 //
1627 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1628 // really make a difference for these runtime calls, since they are
1629 // slow anyway. Btw., bcp must be saved/restored since it may change
1630 // due to GC.
1631 save_bcp();
1632 #ifdef ASSERT
1633 {
1634 Label L;
1635 ld(t0, Address(fp, frame::interpreter_frame_last_sp_offset * wordSize));
1636 beqz(t0, L);
1637 stop("InterpreterMacroAssembler::call_VM_base:"
1638 " last_sp != NULL");
1639 bind(L);
1640 }
1641 #endif /* ASSERT */
1642 // super call
1643 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1644 entry_point, number_of_arguments,
1645 check_exceptions);
1646 // interpreter specific
1647 restore_bcp();
1648 restore_locals();
1649 }
1650
1651 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1652 assert_different_registers(obj, tmp, t0, mdo_addr.base());
1653 Label update, next, none;
1654
1655 verify_oop(obj);
1656
1657 bnez(obj, update);
1658 orptr(mdo_addr, TypeEntries::null_seen, t0, tmp);
1659 j(next);
1660
1661 bind(update);
1662 load_klass(obj, obj);
1663
1664 ld(t0, mdo_addr);
1665 xorr(obj, obj, t0);
1666 andi(t0, obj, TypeEntries::type_klass_mask);
1667 beqz(t0, next); // klass seen before, nothing to
1668 // do. The unknown bit may have been
1669 // set already but no need to check.
1670
1671 andi(t0, obj, TypeEntries::type_unknown);
1672 bnez(t0, next);
1673 // already unknown. Nothing to do anymore.
1674
1675 ld(t0, mdo_addr);
1676 beqz(t0, none);
1677 li(tmp, (u1)TypeEntries::null_seen);
1678 beq(t0, tmp, none);
1679 // There is a chance that the checks above (re-reading profiling
1680 // data from memory) fail if another thread has just set the
1681 // profiling to this obj's klass
1682 ld(t0, mdo_addr);
1683 xorr(obj, obj, t0);
1684 andi(t0, obj, TypeEntries::type_klass_mask);
1685 beqz(t0, next);
1686
1687 // different than before. Cannot keep accurate profile.
1688 orptr(mdo_addr, TypeEntries::type_unknown, t0, tmp);
1689 j(next);
1690
1691 bind(none);
1692 // first time here. Set profile type.
1693 sd(obj, mdo_addr);
1694
1695 bind(next);
1696 }
1697
1698 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1699 if (!ProfileInterpreter) {
1700 return;
1701 }
1702
1703 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1704 Label profile_continue;
1705
1706 test_method_data_pointer(mdp, profile_continue);
1707
1708 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1709
1710 lbu(t0, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1711 if (is_virtual) {
1712 li(tmp, (u1)DataLayout::virtual_call_type_data_tag);
1713 bne(t0, tmp, profile_continue);
1714 } else {
1715 li(tmp, (u1)DataLayout::call_type_data_tag);
1716 bne(t0, tmp, profile_continue);
1717 }
1718
1719 // calculate slot step
1720 static int stack_slot_offset0 = in_bytes(TypeEntriesAtCall::stack_slot_offset(0));
1721 static int slot_step = in_bytes(TypeEntriesAtCall::stack_slot_offset(1)) - stack_slot_offset0;
1722
1723 // calculate type step
1724 static int argument_type_offset0 = in_bytes(TypeEntriesAtCall::argument_type_offset(0));
1725 static int type_step = in_bytes(TypeEntriesAtCall::argument_type_offset(1)) - argument_type_offset0;
1726
1727 if (MethodData::profile_arguments()) {
1728 Label done, loop, loopEnd, profileArgument, profileReturnType;
1729 RegSet pushed_registers;
1730 pushed_registers += x15;
1731 pushed_registers += x16;
1732 pushed_registers += x17;
1733 Register mdo_addr = x15;
1734 Register index = x16;
1735 Register off_to_args = x17;
1736 push_reg(pushed_registers, sp);
1737
1738 mv(off_to_args, in_bytes(TypeEntriesAtCall::args_data_offset()));
1739 mv(t0, TypeProfileArgsLimit);
1740 beqz(t0, loopEnd);
1741
1742 mv(index, zr); // index < TypeProfileArgsLimit
1743 bind(loop);
1744 bgtz(index, profileReturnType);
1745 li(t0, (int)MethodData::profile_return());
1746 beqz(t0, profileArgument); // (index > 0 || MethodData::profile_return()) == false
1747 bind(profileReturnType);
1748 // If return value type is profiled we may have no argument to profile
1749 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1750 mv(t1, - TypeStackSlotEntries::per_arg_count());
1751 mul(t1, index, t1);
1752 add(tmp, tmp, t1);
1753 li(t1, TypeStackSlotEntries::per_arg_count());
1754 add(t0, mdp, off_to_args);
1755 blt(tmp, t1, done);
1756
1757 bind(profileArgument);
1758
1759 ld(tmp, Address(callee, Method::const_offset()));
1760 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1761 // stack offset o (zero based) from the start of the argument
1762 // list, for n arguments translates into offset n - o - 1 from
1763 // the end of the argument list
1764 li(t0, stack_slot_offset0);
1765 li(t1, slot_step);
1766 mul(t1, index, t1);
1767 add(t0, t0, t1);
1768 add(t0, mdp, t0);
1769 ld(t0, Address(t0));
1770 sub(tmp, tmp, t0);
1771 addi(tmp, tmp, -1);
1772 Address arg_addr = argument_address(tmp);
1773 ld(tmp, arg_addr);
1774
1775 li(t0, argument_type_offset0);
1776 li(t1, type_step);
1777 mul(t1, index, t1);
1778 add(t0, t0, t1);
1779 add(mdo_addr, mdp, t0);
1780 Address mdo_arg_addr(mdo_addr, 0);
1781 profile_obj_type(tmp, mdo_arg_addr, t1);
1782
1783 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1784 addi(off_to_args, off_to_args, to_add);
1785
1786 // increment index by 1
1787 addi(index, index, 1);
1788 li(t1, TypeProfileArgsLimit);
1789 blt(index, t1, loop);
1790 bind(loopEnd);
1791
1792 if (MethodData::profile_return()) {
1793 ld(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1794 addi(tmp, tmp, -TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1795 }
1796
1797 add(t0, mdp, off_to_args);
1798 bind(done);
1799 mv(mdp, t0);
1800
1801 // unspill the clobbered registers
1802 pop_reg(pushed_registers, sp);
1803
1804 if (MethodData::profile_return()) {
1805 // We're right after the type profile for the last
1806 // argument. tmp is the number of cells left in the
1807 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1808 // if there's a return to profile.
1809 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1810 shadd(mdp, tmp, mdp, tmp, exact_log2(DataLayout::cell_size));
1811 }
1812 sd(mdp, Address(fp, frame::interpreter_frame_mdp_offset * wordSize));
1813 } else {
1814 assert(MethodData::profile_return(), "either profile call args or call ret");
1815 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1816 }
1817
1818 // mdp points right after the end of the
1819 // CallTypeData/VirtualCallTypeData, right after the cells for the
1820 // return value type if there's one
1821
1822 bind(profile_continue);
1823 }
1824 }
1825
1826 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1827 assert_different_registers(mdp, ret, tmp, xbcp, t0, t1);
1828 if (ProfileInterpreter && MethodData::profile_return()) {
1829 Label profile_continue, done;
1830
1831 test_method_data_pointer(mdp, profile_continue);
1832
1833 if (MethodData::profile_return_jsr292_only()) {
1834 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1835
1836 // If we don't profile all invoke bytecodes we must make sure
1837 // it's a bytecode we indeed profile. We can't go back to the
1838 // begining of the ProfileData we intend to update to check its
1839 // type because we're right after it and we don't known its
1840 // length
1841 Label do_profile;
1842 lbu(t0, Address(xbcp, 0));
1843 li(tmp, (u1)Bytecodes::_invokedynamic);
1844 beq(t0, tmp, do_profile);
1845 li(tmp, (u1)Bytecodes::_invokehandle);
1846 beq(t0, tmp, do_profile);
1847 get_method(tmp);
1848 lhu(t0, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
1849 li(t1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1850 bne(t0, t1, profile_continue);
1851 bind(do_profile);
1852 }
1853
1854 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1855 mv(tmp, ret);
1856 profile_obj_type(tmp, mdo_ret_addr, t1);
1857
1858 bind(profile_continue);
1859 }
1860 }
1861
1862 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2, Register tmp3) {
1863 assert_different_registers(t0, t1, mdp, tmp1, tmp2, tmp3);
1864 if (ProfileInterpreter && MethodData::profile_parameters()) {
1865 Label profile_continue, done;
1866
1867 test_method_data_pointer(mdp, profile_continue);
1868
1869 // Load the offset of the area within the MDO used for
1870 // parameters. If it's negative we're not profiling any parameters
1871 lwu(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1872 srli(tmp2, tmp1, 31);
1873 bnez(tmp2, profile_continue); // i.e. sign bit set
1874
1875 // Compute a pointer to the area for parameters from the offset
1876 // and move the pointer to the slot for the last
1877 // parameters. Collect profiling from last parameter down.
1878 // mdo start + parameters offset + array length - 1
1879 add(mdp, mdp, tmp1);
1880 ld(tmp1, Address(mdp, ArrayData::array_len_offset()));
1881 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1882
1883 Label loop;
1884 bind(loop);
1885
1886 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1887 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1888 int per_arg_scale = exact_log2(DataLayout::cell_size);
1889 add(t0, mdp, off_base);
1890 add(t1, mdp, type_base);
1891
1892 shadd(tmp2, tmp1, t0, tmp2, per_arg_scale);
1893 // load offset on the stack from the slot for this parameter
1894 ld(tmp2, Address(tmp2, 0));
1895 neg(tmp2, tmp2);
1896
1897 // read the parameter from the local area
1898 shadd(tmp2, tmp2, xlocals, tmp2, Interpreter::logStackElementSize);
1899 ld(tmp2, Address(tmp2, 0));
1900
1901 // profile the parameter
1902 shadd(t1, tmp1, t1, t0, per_arg_scale);
1903 Address arg_type(t1, 0);
1904 profile_obj_type(tmp2, arg_type, tmp3);
1905
1906 // go to next parameter
1907 add(tmp1, tmp1, - TypeStackSlotEntries::per_arg_count());
1908 bgez(tmp1, loop);
1909
1910 bind(profile_continue);
1911 }
1912 }
1913
1914 void InterpreterMacroAssembler::get_method_counters(Register method,
1915 Register mcs, Label& skip) {
1916 Label has_counters;
1917 ld(mcs, Address(method, Method::method_counters_offset()));
1918 bnez(mcs, has_counters);
1919 call_VM(noreg, CAST_FROM_FN_PTR(address,
1920 InterpreterRuntime::build_method_counters), method);
1921 ld(mcs, Address(method, Method::method_counters_offset()));
1922 beqz(mcs, skip); // No MethodCounters allocated, OutOfMemory
1923 bind(has_counters);
1924 }
1925
1926 #ifdef ASSERT
1927 void InterpreterMacroAssembler::verify_access_flags(Register access_flags, uint32_t flag_bits,
1928 const char* msg, bool stop_by_hit) {
1929 Label L;
1930 andi(t0, access_flags, flag_bits);
1931 if (stop_by_hit) {
1932 beqz(t0, L);
1933 } else {
1934 bnez(t0, L);
1935 }
1936 stop(msg);
1937 bind(L);
1938 }
1939
1940 void InterpreterMacroAssembler::verify_frame_setup() {
1941 Label L;
1942 const Address monitor_block_top(fp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1943 ld(t0, monitor_block_top);
1944 beq(esp, t0, L);
1945 stop("broken stack frame setup in interpreter");
1946 bind(L);
1947 }
1948 #endif