1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/macroAssembler.inline.hpp"
27 #include "compiler/compiler_globals.hpp"
28 #include "gc/shared/barrierSet.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interp_masm_aarch64.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "logging/log.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/constMethodFlags.hpp"
36 #include "oops/markWord.hpp"
37 #include "oops/method.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/inlineKlass.hpp"
40 #include "oops/resolvedFieldEntry.hpp"
41 #include "oops/resolvedIndyEntry.hpp"
42 #include "oops/resolvedMethodEntry.hpp"
43 #include "prims/jvmtiExport.hpp"
44 #include "prims/jvmtiThreadState.hpp"
45 #include "runtime/basicLock.hpp"
46 #include "runtime/frame.inline.hpp"
47 #include "runtime/javaThread.hpp"
48 #include "runtime/safepointMechanism.hpp"
49 #include "runtime/sharedRuntime.hpp"
50 #include "utilities/powerOfTwo.hpp"
51
52 void InterpreterMacroAssembler::narrow(Register result) {
53
54 // Get method->_constMethod->_result_type
55 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
56 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
57 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
58
59 Label done, notBool, notByte, notChar;
60
61 // common case first
62 cmpw(rscratch1, T_INT);
63 br(Assembler::EQ, done);
64
65 // mask integer result to narrower return type.
66 cmpw(rscratch1, T_BOOLEAN);
67 br(Assembler::NE, notBool);
68 andw(result, result, 0x1);
69 b(done);
70
71 bind(notBool);
72 cmpw(rscratch1, T_BYTE);
73 br(Assembler::NE, notByte);
74 sbfx(result, result, 0, 8);
75 b(done);
76
77 bind(notByte);
78 cmpw(rscratch1, T_CHAR);
79 br(Assembler::NE, notChar);
80 ubfx(result, result, 0, 16); // truncate upper 16 bits
81 b(done);
82
83 bind(notChar);
84 sbfx(result, result, 0, 16); // sign-extend short
85
86 // Nothing to do for T_INT
87 bind(done);
88 }
89
90 void InterpreterMacroAssembler::jump_to_entry(address entry) {
91 assert(entry, "Entry must have been generated by now");
92 b(entry);
93 }
94
95 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
96 if (JvmtiExport::can_pop_frame()) {
97 Label L;
98 // Initiate popframe handling only if it is not already being
99 // processed. If the flag has the popframe_processing bit set, it
100 // means that this code is called *during* popframe handling - we
101 // don't want to reenter.
102 // This method is only called just after the call into the vm in
103 // call_VM_base, so the arg registers are available.
104 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
105 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
106 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
107 // Call Interpreter::remove_activation_preserving_args_entry() to get the
108 // address of the same-named entrypoint in the generated interpreter code.
109 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
110 br(r0);
111 bind(L);
112 }
113 }
114
115
116 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
117 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
118 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
119 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
120 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
121 switch (state) {
122 case atos: ldr(r0, oop_addr);
123 str(zr, oop_addr);
124 interp_verify_oop(r0, state); break;
125 case ltos: ldr(r0, val_addr); break;
126 case btos: // fall through
127 case ztos: // fall through
128 case ctos: // fall through
129 case stos: // fall through
130 case itos: ldrw(r0, val_addr); break;
131 case ftos: ldrs(v0, val_addr); break;
132 case dtos: ldrd(v0, val_addr); break;
133 case vtos: /* nothing to do */ break;
134 default : ShouldNotReachHere();
135 }
136 // Clean up tos value in the thread object
137 movw(rscratch1, (int) ilgl);
138 strw(rscratch1, tos_addr);
139 strw(zr, val_addr);
140 }
141
142
143 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
144 if (JvmtiExport::can_force_early_return()) {
145 Label L;
146 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
147 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
148
149 // Initiate earlyret handling only if it is not already being processed.
150 // If the flag has the earlyret_processing bit set, it means that this code
151 // is called *during* earlyret handling - we don't want to reenter.
152 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
153 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
154 br(Assembler::NE, L);
155
156 // Call Interpreter::remove_activation_early_entry() to get the address of the
157 // same-named entrypoint in the generated interpreter code.
158 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
159 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
160 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
161 br(r0);
162 bind(L);
163 }
164 }
165
166 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
167 Register reg,
168 int bcp_offset) {
169 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
170 ldrh(reg, Address(rbcp, bcp_offset));
171 rev16(reg, reg);
172 }
173
174 void InterpreterMacroAssembler::get_dispatch() {
175 uint64_t offset;
176 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
177 // Use add() here after ARDP, rather than lea().
178 // lea() does not generate anything if its offset is zero.
179 // However, relocs expect to find either an ADD or a load/store
180 // insn after an ADRP. add() always generates an ADD insn, even
181 // for add(Rn, Rn, 0).
182 add(rdispatch, rdispatch, offset);
183 }
184
185 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
186 int bcp_offset,
187 size_t index_size) {
188 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
189 if (index_size == sizeof(u2)) {
190 load_unsigned_short(index, Address(rbcp, bcp_offset));
191 } else if (index_size == sizeof(u4)) {
192 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
193 ldrw(index, Address(rbcp, bcp_offset));
194 } else if (index_size == sizeof(u1)) {
195 load_unsigned_byte(index, Address(rbcp, bcp_offset));
196 } else {
197 ShouldNotReachHere();
198 }
199 }
200
201 void InterpreterMacroAssembler::get_method_counters(Register method,
202 Register mcs, Label& skip) {
203 Label has_counters;
204 ldr(mcs, Address(method, Method::method_counters_offset()));
205 cbnz(mcs, has_counters);
206 call_VM(noreg, CAST_FROM_FN_PTR(address,
207 InterpreterRuntime::build_method_counters), method);
208 ldr(mcs, Address(method, Method::method_counters_offset()));
209 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
210 bind(has_counters);
211 }
212
213 void InterpreterMacroAssembler::allocate_instance(Register klass, Register new_obj,
214 Register t1, Register t2,
215 bool clear_fields, Label& alloc_failed) {
216 MacroAssembler::allocate_instance(klass, new_obj, t1, t2, clear_fields, alloc_failed);
217 if (DTraceAllocProbes) {
218 // Trigger dtrace event for fastpath
219 push(atos);
220 call_VM_leaf(CAST_FROM_FN_PTR(address, static_cast<int (*)(oopDesc*)>(SharedRuntime::dtrace_object_alloc)), new_obj);
221 pop(atos);
222 }
223 }
224
225 void InterpreterMacroAssembler::read_flat_field(Register entry,
226 Register field_index, Register field_offset,
227 Register temp, Register obj) {
228 Label failed_alloc, slow_path, done;
229 const Register src = field_offset;
230 const Register alloc_temp = r10;
231 const Register dst_temp = field_index;
232 const Register layout_info = temp;
233 assert_different_registers(obj, entry, field_index, field_offset, temp, alloc_temp, rscratch1);
234
235 load_unsigned_byte(temp, Address(entry, in_bytes(ResolvedFieldEntry::flags_offset())));
236 // If the field is nullable, jump to slow path
237 tbz(temp, ResolvedFieldEntry::is_null_free_inline_type_shift, slow_path);
238
239 // Grab the inline field klass
240 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
241 inline_layout_info(rscratch1, field_index, layout_info);
242
243 const Register field_klass = dst_temp;
244 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
245
246 // allocate buffer
247 push(obj); // save holder
248 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, failed_alloc);
249
250 // Have an oop instance buffer, copy into it
251 payload_address(obj, dst_temp, field_klass); // danger, uses rscratch1
252 pop(alloc_temp); // restore holder
253 lea(src, Address(alloc_temp, field_offset));
254 // call_VM_leaf, clobbers a few regs, save restore new obj
255 push(obj);
256 flat_field_copy(IS_DEST_UNINITIALIZED, src, dst_temp, layout_info);
257 pop(obj);
258 b(done);
259
260 bind(failed_alloc);
261 pop(obj);
262 bind(slow_path);
263 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
264 obj, entry);
265
266 bind(done);
267 membar(Assembler::StoreStore);
268 }
269
270 void InterpreterMacroAssembler::write_flat_field(Register entry, Register field_offset,
271 Register tmp1, Register tmp2,
272 Register obj) {
273 assert_different_registers(entry, field_offset, tmp1, tmp2, obj);
274 Label slow_path, done;
275
276 load_unsigned_byte(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::flags_offset())));
277 test_field_is_not_null_free_inline_type(tmp1, noreg /* temp */, slow_path);
278
279 null_check(r0); // FIXME JDK-8341120
280
281 add(obj, obj, field_offset);
282
283 load_klass(tmp1, r0);
284 payload_address(r0, r0, tmp1);
285
286 Register layout_info = field_offset;
287 load_unsigned_short(tmp1, Address(entry, in_bytes(ResolvedFieldEntry::field_index_offset())));
288 ldr(tmp2, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
289 inline_layout_info(tmp2, tmp1, layout_info);
290
291 flat_field_copy(IN_HEAP, r0, obj, layout_info);
292 b(done);
293
294 bind(slow_path);
295 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flat_field), obj, r0, entry);
296 bind(done);
297 }
298
299 // Load object from cpool->resolved_references(index)
300 void InterpreterMacroAssembler::load_resolved_reference_at_index(
301 Register result, Register index, Register tmp) {
302 assert_different_registers(result, index);
303
304 get_constant_pool(result);
305 // load pointer for resolved_references[] objArray
306 ldr(result, Address(result, ConstantPool::cache_offset()));
307 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
308 resolve_oop_handle(result, tmp, rscratch2);
309 // Add in the index
310 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
311 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
312 }
313
314 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
315 Register cpool, Register index, Register klass, Register temp) {
316 add(temp, cpool, index, LSL, LogBytesPerWord);
317 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
318 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
319 add(klass, klass, temp, LSL, LogBytesPerWord);
320 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
321 }
322
323 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
324 // subtype of super_klass.
325 //
326 // Args:
327 // r0: superklass
328 // Rsub_klass: subklass
329 //
330 // Kills:
331 // r2, r5
332 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
333 Label& ok_is_subtype,
334 bool profile) {
335 assert(Rsub_klass != r0, "r0 holds superklass");
336 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
337 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
338
339 // Profile the not-null value's klass.
340 if (profile) {
341 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
342 }
343
344 // Do the check.
345 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
346 }
347
348 // Java Expression Stack
349
350 void InterpreterMacroAssembler::pop_ptr(Register r) {
351 ldr(r, post(esp, wordSize));
352 }
353
354 void InterpreterMacroAssembler::pop_i(Register r) {
355 ldrw(r, post(esp, wordSize));
356 }
357
358 void InterpreterMacroAssembler::pop_l(Register r) {
359 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
360 }
361
362 void InterpreterMacroAssembler::push_ptr(Register r) {
363 str(r, pre(esp, -wordSize));
364 }
365
366 void InterpreterMacroAssembler::push_i(Register r) {
367 str(r, pre(esp, -wordSize));
368 }
369
370 void InterpreterMacroAssembler::push_l(Register r) {
371 str(zr, pre(esp, -wordSize));
372 str(r, pre(esp, - wordSize));
373 }
374
375 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
376 ldrs(r, post(esp, wordSize));
377 }
378
379 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
380 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
381 }
382
383 void InterpreterMacroAssembler::push_f(FloatRegister r) {
384 strs(r, pre(esp, -wordSize));
385 }
386
387 void InterpreterMacroAssembler::push_d(FloatRegister r) {
388 strd(r, pre(esp, 2* -wordSize));
389 }
390
391 void InterpreterMacroAssembler::pop(TosState state) {
392 switch (state) {
393 case atos: pop_ptr(); break;
394 case btos:
395 case ztos:
396 case ctos:
397 case stos:
398 case itos: pop_i(); break;
399 case ltos: pop_l(); break;
400 case ftos: pop_f(); break;
401 case dtos: pop_d(); break;
402 case vtos: /* nothing to do */ break;
403 default: ShouldNotReachHere();
404 }
405 interp_verify_oop(r0, state);
406 }
407
408 void InterpreterMacroAssembler::push(TosState state) {
409 interp_verify_oop(r0, state);
410 switch (state) {
411 case atos: push_ptr(); break;
412 case btos:
413 case ztos:
414 case ctos:
415 case stos:
416 case itos: push_i(); break;
417 case ltos: push_l(); break;
418 case ftos: push_f(); break;
419 case dtos: push_d(); break;
420 case vtos: /* nothing to do */ break;
421 default : ShouldNotReachHere();
422 }
423 }
424
425 // Helpers for swap and dup
426 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
427 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
428 }
429
430 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
431 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
432 }
433
434 void InterpreterMacroAssembler::load_float(Address src) {
435 ldrs(v0, src);
436 }
437
438 void InterpreterMacroAssembler::load_double(Address src) {
439 ldrd(v0, src);
440 }
441
442 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
443 // set sender sp
444 mov(r19_sender_sp, sp);
445 // record last_sp
446 sub(rscratch1, esp, rfp);
447 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
448 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
449 }
450
451 // Jump to from_interpreted entry of a call unless single stepping is possible
452 // in this thread in which case we must call the i2i entry
453 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
454 prepare_to_jump_from_interpreted();
455
456 if (JvmtiExport::can_post_interpreter_events()) {
457 Label run_compiled_code;
458 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
459 // compiled code in threads for which the event is enabled. Check here for
460 // interp_only_mode if these events CAN be enabled.
461 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
462 cbzw(rscratch1, run_compiled_code);
463 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
464 br(rscratch1);
465 bind(run_compiled_code);
466 }
467
468 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
469 br(rscratch1);
470 }
471
472 // The following two routines provide a hook so that an implementation
473 // can schedule the dispatch in two parts. amd64 does not do this.
474 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
475 }
476
477 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
478 dispatch_next(state, step);
479 }
480
481 void InterpreterMacroAssembler::dispatch_base(TosState state,
482 address* table,
483 bool verifyoop,
484 bool generate_poll) {
485 if (VerifyActivationFrameSize) {
486 Label L;
487 sub(rscratch2, rfp, esp);
488 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
489 subs(rscratch2, rscratch2, min_frame_size);
490 br(Assembler::GE, L);
491 stop("broken stack frame");
492 bind(L);
493 }
494 if (verifyoop) {
495 interp_verify_oop(r0, state);
496 }
497
498 Label safepoint;
499 address* const safepoint_table = Interpreter::safept_table(state);
500 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
501
502 if (needs_thread_local_poll) {
503 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
504 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
505 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
506 }
507
508 if (table == Interpreter::dispatch_table(state)) {
509 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
510 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
511 } else {
512 mov(rscratch2, (address)table);
513 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
514 }
515 br(rscratch2);
516
517 if (needs_thread_local_poll) {
518 bind(safepoint);
519 lea(rscratch2, ExternalAddress((address)safepoint_table));
520 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
521 br(rscratch2);
522 }
523 }
524
525 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
526 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
527 }
528
529 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
530 dispatch_base(state, Interpreter::normal_table(state));
531 }
532
533 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
534 dispatch_base(state, Interpreter::normal_table(state), false);
535 }
536
537
538 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
539 // load next bytecode
540 ldrb(rscratch1, Address(pre(rbcp, step)));
541 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
542 }
543
544 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
545 // load current bytecode
546 ldrb(rscratch1, Address(rbcp, 0));
547 dispatch_base(state, table);
548 }
549
550 // remove activation
551 //
552 // Unlock the receiver if this is a synchronized method.
553 // Unlock any Java monitors from synchronized blocks.
554 // Apply stack watermark barrier.
555 // Notify JVMTI.
556 // Remove the activation from the stack.
557 //
558 // If there are locked Java monitors
559 // If throw_monitor_exception
560 // throws IllegalMonitorStateException
561 // Else if install_monitor_exception
562 // installs IllegalMonitorStateException
563 // Else
564 // no error processing
565 void InterpreterMacroAssembler::remove_activation(TosState state,
566 bool throw_monitor_exception,
567 bool install_monitor_exception,
568 bool notify_jvmdi) {
569 // Note: Registers r3 xmm0 may be in use for the
570 // result check if synchronized method
571 Label unlocked, unlock, no_unlock;
572
573 #ifdef ASSERT
574 Label not_preempted;
575 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
576 cbz(rscratch1, not_preempted);
577 stop("remove_activation: should not have alternate return address set");
578 bind(not_preempted);
579 #endif /* ASSERT */
580
581 // get the value of _do_not_unlock_if_synchronized into r3
582 const Address do_not_unlock_if_synchronized(rthread,
583 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
584 ldrb(r3, do_not_unlock_if_synchronized);
585 strb(zr, do_not_unlock_if_synchronized); // reset the flag
586
587 // get method access flags
588 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
589 ldrh(r2, Address(r1, Method::access_flags_offset()));
590 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
591
592 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
593 // is set.
594 cbnz(r3, no_unlock);
595
596 // unlock monitor
597 push(state); // save result
598
599 // BasicObjectLock will be first in list, since this is a
600 // synchronized method. However, need to check that the object has
601 // not been unlocked by an explicit monitorexit bytecode.
602 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
603 wordSize - (int) sizeof(BasicObjectLock));
604 // We use c_rarg1 so that if we go slow path it will be the correct
605 // register for unlock_object to pass to VM directly
606 lea(c_rarg1, monitor); // address of first monitor
607
608 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
609 cbnz(r0, unlock);
610
611 pop(state);
612 if (throw_monitor_exception) {
613 // Entry already unlocked, need to throw exception
614 call_VM(noreg, CAST_FROM_FN_PTR(address,
615 InterpreterRuntime::throw_illegal_monitor_state_exception));
616 should_not_reach_here();
617 } else {
618 // Monitor already unlocked during a stack unroll. If requested,
619 // install an illegal_monitor_state_exception. Continue with
620 // stack unrolling.
621 if (install_monitor_exception) {
622 call_VM(noreg, CAST_FROM_FN_PTR(address,
623 InterpreterRuntime::new_illegal_monitor_state_exception));
624 }
625 b(unlocked);
626 }
627
628 bind(unlock);
629 unlock_object(c_rarg1);
630 pop(state);
631
632 // Check that for block-structured locking (i.e., that all locked
633 // objects has been unlocked)
634 bind(unlocked);
635
636 // r0: Might contain return value
637
638 // Check that all monitors are unlocked
639 {
640 Label loop, exception, entry, restart;
641 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
642 const Address monitor_block_top(
643 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
644 const Address monitor_block_bot(
645 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
646
647 bind(restart);
648 // We use c_rarg1 so that if we go slow path it will be the correct
649 // register for unlock_object to pass to VM directly
650 ldr(c_rarg1, monitor_block_top); // derelativize pointer
651 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
652 // c_rarg1 points to current entry, starting with top-most entry
653
654 lea(r19, monitor_block_bot); // points to word before bottom of
655 // monitor block
656 b(entry);
657
658 // Entry already locked, need to throw exception
659 bind(exception);
660
661 if (throw_monitor_exception) {
662 // Throw exception
663 MacroAssembler::call_VM(noreg,
664 CAST_FROM_FN_PTR(address, InterpreterRuntime::
665 throw_illegal_monitor_state_exception));
666 should_not_reach_here();
667 } else {
668 // Stack unrolling. Unlock object and install illegal_monitor_exception.
669 // Unlock does not block, so don't have to worry about the frame.
670 // We don't have to preserve c_rarg1 since we are going to throw an exception.
671
672 push(state);
673 unlock_object(c_rarg1);
674 pop(state);
675
676 if (install_monitor_exception) {
677 call_VM(noreg, CAST_FROM_FN_PTR(address,
678 InterpreterRuntime::
679 new_illegal_monitor_state_exception));
680 }
681
682 b(restart);
683 }
684
685 bind(loop);
686 // check if current entry is used
687 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
688 cbnz(rscratch1, exception);
689
690 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
691 bind(entry);
692 cmp(c_rarg1, r19); // check if bottom reached
693 br(Assembler::NE, loop); // if not at bottom then check this entry
694 }
695
696 bind(no_unlock);
697
698 JFR_ONLY(enter_jfr_critical_section();)
699
700 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
701 // that would normally not be safe to use. Such bad returns into unsafe territory of
702 // the stack, will call InterpreterRuntime::at_unwind.
703 Label slow_path;
704 Label fast_path;
705 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
706 br(Assembler::AL, fast_path);
707 bind(slow_path);
708 push(state);
709 set_last_Java_frame(esp, rfp, pc(), rscratch1);
710 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
711 reset_last_Java_frame(true);
712 pop(state);
713 bind(fast_path);
714
715 // JVMTI support. Make sure the safepoint poll test is issued prior.
716 if (notify_jvmdi) {
717 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
718 } else {
719 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
720 }
721
722 // remove activation
723 // get sender esp
724 ldr(rscratch2,
725 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
726
727 if (StackReservedPages > 0) {
728 // testing if reserved zone needs to be re-enabled
729 Label no_reserved_zone_enabling;
730
731 // check if already enabled - if so no re-enabling needed
732 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
733 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
734 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
735 br(Assembler::EQ, no_reserved_zone_enabling);
736
737 // look for an overflow into the stack reserved zone, i.e.
738 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
739 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
740 cmp(rscratch2, rscratch1);
741 br(Assembler::LS, no_reserved_zone_enabling);
742
743 JFR_ONLY(leave_jfr_critical_section();)
744
745 call_VM_leaf(
746 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
747 call_VM(noreg, CAST_FROM_FN_PTR(address,
748 InterpreterRuntime::throw_delayed_StackOverflowError));
749 should_not_reach_here();
750
751 bind(no_reserved_zone_enabling);
752 }
753
754 if (state == atos && InlineTypeReturnedAsFields) {
755 Label skip;
756 Label not_null;
757 cbnz(r0, not_null);
758 // Returned value is null, zero all return registers because they may belong to oop fields
759 mov(j_rarg1, zr);
760 mov(j_rarg2, zr);
761 mov(j_rarg3, zr);
762 mov(j_rarg4, zr);
763 mov(j_rarg5, zr);
764 mov(j_rarg6, zr);
765 mov(j_rarg7, zr);
766 b(skip);
767 bind(not_null);
768
769 // Check if we are returning an non-null inline type and load its fields into registers
770 test_oop_is_not_inline_type(r0, rscratch2, skip, /* can_be_null= */ false);
771
772 // Load fields from a buffered value with an inline class specific handler
773 load_klass(rscratch1 /*dst*/, r0 /*src*/);
774 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
775 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
776 // Unpack handler can be null if inline type is not scalarizable in returns
777 cbz(rscratch1, skip);
778
779 blr(rscratch1);
780 #ifdef ASSERT
781 // TODO 8284443 Enable
782 if (StressCallingConvention && false) {
783 Label skip_stress;
784 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
785 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
786 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
787 br(Assembler::EQ, skip_stress);
788 load_klass(r0, r0);
789 orr(r0, r0, 1);
790 bind(skip_stress);
791 }
792 #endif
793 bind(skip);
794 // Check above kills sender esp in rscratch2. Reload it.
795 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
796 }
797
798 // remove frame anchor
799 leave();
800
801 JFR_ONLY(leave_jfr_critical_section();)
802
803 // restore sender esp
804 mov(esp, rscratch2);
805
806 // If we're returning to interpreted code we will shortly be
807 // adjusting SP to allow some space for ESP. If we're returning to
808 // compiled code the saved sender SP was saved in sender_sp, so this
809 // restores it.
810 andr(sp, esp, -16);
811 }
812
813 #if INCLUDE_JFR
814 void InterpreterMacroAssembler::enter_jfr_critical_section() {
815 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
816 mov(rscratch1, true);
817 strb(rscratch1, sampling_critical_section);
818 }
819
820 void InterpreterMacroAssembler::leave_jfr_critical_section() {
821 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
822 strb(zr, sampling_critical_section);
823 }
824 #endif // INCLUDE_JFR
825
826 // Lock object
827 //
828 // Args:
829 // c_rarg1: BasicObjectLock to be used for locking
830 //
831 // Kills:
832 // r0
833 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
834 // rscratch1, rscratch2 (scratch regs)
835 void InterpreterMacroAssembler::lock_object(Register lock_reg)
836 {
837 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
838
839 const Register tmp = c_rarg2;
840 const Register obj_reg = c_rarg3; // Will contain the oop
841 const Register tmp2 = c_rarg4;
842 const Register tmp3 = c_rarg5;
843
844 // Load object pointer into obj_reg %c_rarg3
845 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
846
847 Label slow_case, done;
848 fast_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
849 b(done);
850
851 bind(slow_case);
852
853 // Call the runtime routine for slow case
854 call_VM_preemptable(noreg,
855 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
856 lock_reg);
857
858 bind(done);
859 }
860
861
862 // Unlocks an object. Used in monitorexit bytecode and
863 // remove_activation. Throws an IllegalMonitorException if object is
864 // not locked by current thread.
865 //
866 // Args:
867 // c_rarg1: BasicObjectLock for lock
868 //
869 // Kills:
870 // r0
871 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
872 // rscratch1, rscratch2 (scratch regs)
873 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
874 {
875 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
876
877 const Register swap_reg = r0;
878 const Register header_reg = c_rarg2; // Will contain the old oopMark
879 const Register obj_reg = c_rarg3; // Will contain the oop
880 const Register tmp_reg = c_rarg4; // Temporary used by fast_unlock
881
882 save_bcp(); // Save in case of exception
883
884 // Load oop into obj_reg(%c_rarg3)
885 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
886
887 // Free entry
888 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
889
890 Label slow_case, done;
891 fast_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
892 b(done);
893
894 bind(slow_case);
895 // Call the runtime routine for slow case.
896 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
897 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
898 bind(done);
899 restore_bcp();
900 }
901
902 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
903 Label& zero_continue) {
904 assert(ProfileInterpreter, "must be profiling interpreter");
905 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
906 cbz(mdp, zero_continue);
907 }
908
909 // Set the method data pointer for the current bcp.
910 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
911 assert(ProfileInterpreter, "must be profiling interpreter");
912 Label set_mdp;
913 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
914
915 // Test MDO to avoid the call if it is null.
916 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
917 cbz(r0, set_mdp);
918 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
919 // r0: mdi
920 // mdo is guaranteed to be non-zero here, we checked for it before the call.
921 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
922 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
923 add(r0, r1, r0);
924 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
925 bind(set_mdp);
926 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
927 }
928
929 void InterpreterMacroAssembler::verify_method_data_pointer() {
930 assert(ProfileInterpreter, "must be profiling interpreter");
931 #ifdef ASSERT
932 Label verify_continue;
933 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
934 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
935 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
936 get_method(r1);
937
938 // If the mdp is valid, it will point to a DataLayout header which is
939 // consistent with the bcp. The converse is highly probable also.
940 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
941 ldr(rscratch1, Address(r1, Method::const_offset()));
942 add(r2, r2, rscratch1, Assembler::LSL);
943 lea(r2, Address(r2, ConstMethod::codes_offset()));
944 cmp(r2, rbcp);
945 br(Assembler::EQ, verify_continue);
946 // r1: method
947 // rbcp: bcp // rbcp == 22
948 // r3: mdp
949 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
950 r1, rbcp, r3);
951 bind(verify_continue);
952 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
953 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
954 #endif // ASSERT
955 }
956
957
958 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
959 int constant,
960 Register value) {
961 assert(ProfileInterpreter, "must be profiling interpreter");
962 Address data(mdp_in, constant);
963 str(value, data);
964 }
965
966
967 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
968 int constant) {
969 increment_mdp_data_at(mdp_in, noreg, constant);
970 }
971
972 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
973 Register index,
974 int constant) {
975 assert(ProfileInterpreter, "must be profiling interpreter");
976
977 assert_different_registers(rscratch2, rscratch1, mdp_in, index);
978
979 Address addr1(mdp_in, constant);
980 Address addr2(rscratch2, index, Address::lsl(0));
981 Address &addr = addr1;
982 if (index != noreg) {
983 lea(rscratch2, addr1);
984 addr = addr2;
985 }
986
987 increment(addr, DataLayout::counter_increment);
988 }
989
990 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
991 int flag_byte_constant) {
992 assert(ProfileInterpreter, "must be profiling interpreter");
993 int flags_offset = in_bytes(DataLayout::flags_offset());
994 // Set the flag
995 ldrb(rscratch1, Address(mdp_in, flags_offset));
996 orr(rscratch1, rscratch1, flag_byte_constant);
997 strb(rscratch1, Address(mdp_in, flags_offset));
998 }
999
1000
1001 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1002 int offset,
1003 Register value,
1004 Register test_value_out,
1005 Label& not_equal_continue) {
1006 assert(ProfileInterpreter, "must be profiling interpreter");
1007 if (test_value_out == noreg) {
1008 ldr(rscratch1, Address(mdp_in, offset));
1009 cmp(value, rscratch1);
1010 } else {
1011 // Put the test value into a register, so caller can use it:
1012 ldr(test_value_out, Address(mdp_in, offset));
1013 cmp(value, test_value_out);
1014 }
1015 br(Assembler::NE, not_equal_continue);
1016 }
1017
1018
1019 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1020 int offset_of_disp) {
1021 assert(ProfileInterpreter, "must be profiling interpreter");
1022 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1023 add(mdp_in, mdp_in, rscratch1, LSL);
1024 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1025 }
1026
1027
1028 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1029 Register reg,
1030 int offset_of_disp) {
1031 assert(ProfileInterpreter, "must be profiling interpreter");
1032 lea(rscratch1, Address(mdp_in, offset_of_disp));
1033 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1034 add(mdp_in, mdp_in, rscratch1, LSL);
1035 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1036 }
1037
1038
1039 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1040 int constant) {
1041 assert(ProfileInterpreter, "must be profiling interpreter");
1042 add(mdp_in, mdp_in, (unsigned)constant);
1043 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1044 }
1045
1046
1047 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1048 assert(ProfileInterpreter, "must be profiling interpreter");
1049 // save/restore across call_VM
1050 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1051 call_VM(noreg,
1052 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1053 return_bci);
1054 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1055 }
1056
1057
1058 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
1059 if (ProfileInterpreter) {
1060 Label profile_continue;
1061
1062 // If no method data exists, go to profile_continue.
1063 test_method_data_pointer(mdp, profile_continue);
1064
1065 // We are taking a branch. Increment the taken count.
1066 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1067
1068 // The method data pointer needs to be updated to reflect the new target.
1069 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1070 bind(profile_continue);
1071 }
1072 }
1073
1074
1075 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1076 if (ProfileInterpreter) {
1077 Label profile_continue;
1078
1079 // If no method data exists, go to profile_continue.
1080 test_method_data_pointer(mdp, profile_continue);
1081
1082 // We are not taking a branch. Increment the not taken count.
1083 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1084
1085 // The method data pointer needs to be updated to correspond to
1086 // the next bytecode
1087 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1088 bind(profile_continue);
1089 }
1090 }
1091
1092
1093 void InterpreterMacroAssembler::profile_call(Register mdp) {
1094 if (ProfileInterpreter) {
1095 Label profile_continue;
1096
1097 // If no method data exists, go to profile_continue.
1098 test_method_data_pointer(mdp, profile_continue);
1099
1100 // We are making a call. Increment the count.
1101 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1102
1103 // The method data pointer needs to be updated to reflect the new target.
1104 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1105 bind(profile_continue);
1106 }
1107 }
1108
1109 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1110 if (ProfileInterpreter) {
1111 Label profile_continue;
1112
1113 // If no method data exists, go to profile_continue.
1114 test_method_data_pointer(mdp, profile_continue);
1115
1116 // We are making a call. Increment the count.
1117 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1118
1119 // The method data pointer needs to be updated to reflect the new target.
1120 update_mdp_by_constant(mdp,
1121 in_bytes(VirtualCallData::
1122 virtual_call_data_size()));
1123 bind(profile_continue);
1124 }
1125 }
1126
1127
1128 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1129 Register mdp,
1130 Register reg2,
1131 bool receiver_can_be_null) {
1132 if (ProfileInterpreter) {
1133 Label profile_continue;
1134
1135 // If no method data exists, go to profile_continue.
1136 test_method_data_pointer(mdp, profile_continue);
1137
1138 Label skip_receiver_profile;
1139 if (receiver_can_be_null) {
1140 Label not_null;
1141 // We are making a call. Increment the count for null receiver.
1142 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1143 b(skip_receiver_profile);
1144 bind(not_null);
1145 }
1146
1147 // Record the receiver type.
1148 record_klass_in_profile(receiver, mdp, reg2);
1149 bind(skip_receiver_profile);
1150
1151 // The method data pointer needs to be updated to reflect the new target.
1152 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1153 bind(profile_continue);
1154 }
1155 }
1156
1157 // This routine creates a state machine for updating the multi-row
1158 // type profile at a virtual call site (or other type-sensitive bytecode).
1159 // The machine visits each row (of receiver/count) until the receiver type
1160 // is found, or until it runs out of rows. At the same time, it remembers
1161 // the location of the first empty row. (An empty row records null for its
1162 // receiver, and can be allocated for a newly-observed receiver type.)
1163 // Because there are two degrees of freedom in the state, a simple linear
1164 // search will not work; it must be a decision tree. Hence this helper
1165 // function is recursive, to generate the required tree structured code.
1166 // It's the interpreter, so we are trading off code space for speed.
1167 // See below for example code.
1168 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1169 Register receiver, Register mdp,
1170 Register reg2, int start_row,
1171 Label& done) {
1172 if (TypeProfileWidth == 0) {
1173 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1174 } else {
1175 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1176 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1177 }
1178 }
1179
1180 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1181 Register reg2, int start_row, Label& done, int total_rows,
1182 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1183 int last_row = total_rows - 1;
1184 assert(start_row <= last_row, "must be work left to do");
1185 // Test this row for both the item and for null.
1186 // Take any of three different outcomes:
1187 // 1. found item => increment count and goto done
1188 // 2. found null => keep looking for case 1, maybe allocate this cell
1189 // 3. found something else => keep looking for cases 1 and 2
1190 // Case 3 is handled by a recursive call.
1191 for (int row = start_row; row <= last_row; row++) {
1192 Label next_test;
1193 bool test_for_null_also = (row == start_row);
1194
1195 // See if the item is item[n].
1196 int item_offset = in_bytes(item_offset_fn(row));
1197 test_mdp_data_at(mdp, item_offset, item,
1198 (test_for_null_also ? reg2 : noreg),
1199 next_test);
1200 // (Reg2 now contains the item from the CallData.)
1201
1202 // The item is item[n]. Increment count[n].
1203 int count_offset = in_bytes(item_count_offset_fn(row));
1204 increment_mdp_data_at(mdp, count_offset);
1205 b(done);
1206 bind(next_test);
1207
1208 if (test_for_null_also) {
1209 Label found_null;
1210 // Failed the equality check on item[n]... Test for null.
1211 if (start_row == last_row) {
1212 // The only thing left to do is handle the null case.
1213 cbz(reg2, found_null);
1214 // Item did not match any saved item and there is no empty row for it.
1215 // Increment total counter to indicate polymorphic case.
1216 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1217 b(done);
1218 bind(found_null);
1219 break;
1220 }
1221 // Since null is rare, make it be the branch-taken case.
1222 cbz(reg2, found_null);
1223
1224 // Put all the "Case 3" tests here.
1225 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1226 item_offset_fn, item_count_offset_fn);
1227
1228 // Found a null. Keep searching for a matching item,
1229 // but remember that this is an empty (unused) slot.
1230 bind(found_null);
1231 }
1232 }
1233
1234 // In the fall-through case, we found no matching item, but we
1235 // observed the item[start_row] is null.
1236
1237 // Fill in the item field and increment the count.
1238 int item_offset = in_bytes(item_offset_fn(start_row));
1239 set_mdp_data_at(mdp, item_offset, item);
1240 int count_offset = in_bytes(item_count_offset_fn(start_row));
1241 mov(reg2, DataLayout::counter_increment);
1242 set_mdp_data_at(mdp, count_offset, reg2);
1243 if (start_row > 0) {
1244 b(done);
1245 }
1246 }
1247
1248 // Example state machine code for three profile rows:
1249 // // main copy of decision tree, rooted at row[1]
1250 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1251 // if (row[0].rec != nullptr) {
1252 // // inner copy of decision tree, rooted at row[1]
1253 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1254 // if (row[1].rec != nullptr) {
1255 // // degenerate decision tree, rooted at row[2]
1256 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1257 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1258 // row[2].init(rec); goto done;
1259 // } else {
1260 // // remember row[1] is empty
1261 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1262 // row[1].init(rec); goto done;
1263 // }
1264 // } else {
1265 // // remember row[0] is empty
1266 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1267 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1268 // row[0].init(rec); goto done;
1269 // }
1270 // done:
1271
1272 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1273 Register mdp, Register reg2) {
1274 assert(ProfileInterpreter, "must be profiling");
1275 Label done;
1276
1277 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1278
1279 bind (done);
1280 }
1281
1282 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1283 Register mdp) {
1284 if (ProfileInterpreter) {
1285 Label profile_continue;
1286 uint row;
1287
1288 // If no method data exists, go to profile_continue.
1289 test_method_data_pointer(mdp, profile_continue);
1290
1291 // Update the total ret count.
1292 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1293
1294 for (row = 0; row < RetData::row_limit(); row++) {
1295 Label next_test;
1296
1297 // See if return_bci is equal to bci[n]:
1298 test_mdp_data_at(mdp,
1299 in_bytes(RetData::bci_offset(row)),
1300 return_bci, noreg,
1301 next_test);
1302
1303 // return_bci is equal to bci[n]. Increment the count.
1304 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1305
1306 // The method data pointer needs to be updated to reflect the new target.
1307 update_mdp_by_offset(mdp,
1308 in_bytes(RetData::bci_displacement_offset(row)));
1309 b(profile_continue);
1310 bind(next_test);
1311 }
1312
1313 update_mdp_for_ret(return_bci);
1314
1315 bind(profile_continue);
1316 }
1317 }
1318
1319 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1320 if (ProfileInterpreter) {
1321 Label profile_continue;
1322
1323 // If no method data exists, go to profile_continue.
1324 test_method_data_pointer(mdp, profile_continue);
1325
1326 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1327
1328 // The method data pointer needs to be updated.
1329 int mdp_delta = in_bytes(BitData::bit_data_size());
1330 if (TypeProfileCasts) {
1331 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1332 }
1333 update_mdp_by_constant(mdp, mdp_delta);
1334
1335 bind(profile_continue);
1336 }
1337 }
1338
1339 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1340 if (ProfileInterpreter) {
1341 Label profile_continue;
1342
1343 // If no method data exists, go to profile_continue.
1344 test_method_data_pointer(mdp, profile_continue);
1345
1346 // The method data pointer needs to be updated.
1347 int mdp_delta = in_bytes(BitData::bit_data_size());
1348 if (TypeProfileCasts) {
1349 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1350
1351 // Record the object type.
1352 record_klass_in_profile(klass, mdp, reg2);
1353 }
1354 update_mdp_by_constant(mdp, mdp_delta);
1355
1356 bind(profile_continue);
1357 }
1358 }
1359
1360 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1361 if (ProfileInterpreter) {
1362 Label profile_continue;
1363
1364 // If no method data exists, go to profile_continue.
1365 test_method_data_pointer(mdp, profile_continue);
1366
1367 // Update the default case count
1368 increment_mdp_data_at(mdp,
1369 in_bytes(MultiBranchData::default_count_offset()));
1370
1371 // The method data pointer needs to be updated.
1372 update_mdp_by_offset(mdp,
1373 in_bytes(MultiBranchData::
1374 default_displacement_offset()));
1375
1376 bind(profile_continue);
1377 }
1378 }
1379
1380 void InterpreterMacroAssembler::profile_switch_case(Register index,
1381 Register mdp,
1382 Register reg2) {
1383 if (ProfileInterpreter) {
1384 Label profile_continue;
1385
1386 // If no method data exists, go to profile_continue.
1387 test_method_data_pointer(mdp, profile_continue);
1388
1389 // Build the base (index * per_case_size_in_bytes()) +
1390 // case_array_offset_in_bytes()
1391 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1392 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1393 Assembler::maddw(index, index, reg2, rscratch1);
1394
1395 // Update the case count
1396 increment_mdp_data_at(mdp,
1397 index,
1398 in_bytes(MultiBranchData::relative_count_offset()));
1399
1400 // The method data pointer needs to be updated.
1401 update_mdp_by_offset(mdp,
1402 index,
1403 in_bytes(MultiBranchData::
1404 relative_displacement_offset()));
1405
1406 bind(profile_continue);
1407 }
1408 }
1409
1410 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1411 Register array,
1412 Register tmp) {
1413 if (ProfileInterpreter) {
1414 Label profile_continue;
1415
1416 // If no method data exists, go to profile_continue.
1417 test_method_data_pointer(mdp, profile_continue);
1418
1419 mov(tmp, array);
1420 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1421
1422 Label not_flat;
1423 test_non_flat_array_oop(array, tmp, not_flat);
1424
1425 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1426
1427 bind(not_flat);
1428
1429 Label not_null_free;
1430 test_non_null_free_array_oop(array, tmp, not_null_free);
1431
1432 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1433
1434 bind(not_null_free);
1435
1436 bind(profile_continue);
1437 }
1438 }
1439
1440 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1441 Register array,
1442 Register tmp);
1443 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1444 Register array,
1445 Register tmp);
1446
1447 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1448 if (ProfileInterpreter) {
1449 Label profile_continue;
1450
1451 // If no method data exists, go to profile_continue.
1452 test_method_data_pointer(mdp, profile_continue);
1453
1454 Label done, update;
1455 cbnz(element, update);
1456 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1457 b(done);
1458
1459 bind(update);
1460 load_klass(tmp, element);
1461
1462 // Record the object type.
1463 record_klass_in_profile(tmp, mdp, tmp2);
1464
1465 bind(done);
1466
1467 // The method data pointer needs to be updated.
1468 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1469
1470 bind(profile_continue);
1471 }
1472 }
1473
1474
1475 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1476 Register element,
1477 Register tmp) {
1478 if (ProfileInterpreter) {
1479 Label profile_continue;
1480
1481 // If no method data exists, go to profile_continue.
1482 test_method_data_pointer(mdp, profile_continue);
1483
1484 mov(tmp, element);
1485 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1486
1487 // The method data pointer needs to be updated.
1488 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1489
1490 bind(profile_continue);
1491 }
1492 }
1493
1494 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1495 Register left,
1496 Register right,
1497 Register tmp) {
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 mov(tmp, left);
1505 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1506
1507 Label left_not_inline_type;
1508 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1509 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1510 bind(left_not_inline_type);
1511
1512 mov(tmp, right);
1513 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1514
1515 Label right_not_inline_type;
1516 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1517 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1518 bind(right_not_inline_type);
1519
1520 bind(profile_continue);
1521 }
1522 }
1523
1524 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1525 if (state == atos) {
1526 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1527 }
1528 }
1529
1530 void InterpreterMacroAssembler::notify_method_entry() {
1531 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1532 // track stack depth. If it is possible to enter interp_only_mode we add
1533 // the code to check if the event should be sent.
1534 if (JvmtiExport::can_post_interpreter_events()) {
1535 Label L;
1536 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1537 cbzw(r3, L);
1538 call_VM(noreg, CAST_FROM_FN_PTR(address,
1539 InterpreterRuntime::post_method_entry));
1540 bind(L);
1541 }
1542
1543 if (DTraceMethodProbes) {
1544 get_method(c_rarg1);
1545 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1546 rthread, c_rarg1);
1547 }
1548
1549 // RedefineClasses() tracing support for obsolete method entry
1550 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1551 get_method(c_rarg1);
1552 call_VM_leaf(
1553 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1554 rthread, c_rarg1);
1555 }
1556
1557 }
1558
1559
1560 void InterpreterMacroAssembler::notify_method_exit(
1561 TosState state, NotifyMethodExitMode mode) {
1562 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1563 // track stack depth. If it is possible to enter interp_only_mode we add
1564 // the code to check if the event should be sent.
1565 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1566 Label L;
1567 // Note: frame::interpreter_frame_result has a dependency on how the
1568 // method result is saved across the call to post_method_exit. If this
1569 // is changed then the interpreter_frame_result implementation will
1570 // need to be updated too.
1571
1572 // template interpreter will leave the result on the top of the stack.
1573 push(state);
1574 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1575 cbz(r3, L);
1576 call_VM(noreg,
1577 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1578 bind(L);
1579 pop(state);
1580 }
1581
1582 if (DTraceMethodProbes) {
1583 push(state);
1584 get_method(c_rarg1);
1585 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1586 rthread, c_rarg1);
1587 pop(state);
1588 }
1589 }
1590
1591
1592 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1593 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1594 int increment, Address mask,
1595 Register scratch, Register scratch2,
1596 bool preloaded, Condition cond,
1597 Label* where) {
1598 if (!preloaded) {
1599 ldrw(scratch, counter_addr);
1600 }
1601 add(scratch, scratch, increment);
1602 strw(scratch, counter_addr);
1603 ldrw(scratch2, mask);
1604 ands(scratch, scratch, scratch2);
1605 br(cond, *where);
1606 }
1607
1608 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1609 int number_of_arguments) {
1610 // interpreter specific
1611 //
1612 // Note: No need to save/restore rbcp & rlocals pointer since these
1613 // are callee saved registers and no blocking/ GC can happen
1614 // in leaf calls.
1615 #ifdef ASSERT
1616 {
1617 Label L;
1618 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1619 cbz(rscratch1, L);
1620 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1621 " last_sp != nullptr");
1622 bind(L);
1623 }
1624 #endif /* ASSERT */
1625 // super call
1626 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1627 }
1628
1629 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1630 Register java_thread,
1631 Register last_java_sp,
1632 Label* return_pc,
1633 address entry_point,
1634 int number_of_arguments,
1635 bool check_exceptions) {
1636 // interpreter specific
1637 //
1638 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1639 // really make a difference for these runtime calls, since they are
1640 // slow anyway. Btw., bcp must be saved/restored since it may change
1641 // due to GC.
1642 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1643 save_bcp();
1644 #ifdef ASSERT
1645 {
1646 Label L;
1647 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1648 cbz(rscratch1, L);
1649 stop("InterpreterMacroAssembler::call_VM_base:"
1650 " last_sp != nullptr");
1651 bind(L);
1652 }
1653 #endif /* ASSERT */
1654 // super call
1655 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1656 return_pc, entry_point,
1657 number_of_arguments, check_exceptions);
1658 // interpreter specific
1659 restore_bcp();
1660 restore_locals();
1661 }
1662
1663 void InterpreterMacroAssembler::call_VM_preemptable_helper(Register oop_result,
1664 address entry_point,
1665 int number_of_arguments,
1666 bool check_exceptions) {
1667 assert(InterpreterRuntime::is_preemptable_call(entry_point), "VM call not preemptable, should use call_VM()");
1668 Label resume_pc, not_preempted;
1669
1670 #ifdef ASSERT
1671 {
1672 Label L1, L2;
1673 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1674 cbz(rscratch1, L1);
1675 stop("call_VM_preemptable_helper: Should not have alternate return address set");
1676 bind(L1);
1677 // We check this counter in patch_return_pc_with_preempt_stub() during freeze.
1678 incrementw(Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1679 ldrw(rscratch1, Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1680 cmpw(rscratch1, 0);
1681 br(Assembler::GT, L2);
1682 stop("call_VM_preemptable_helper: should be > 0");
1683 bind(L2);
1684 }
1685 #endif /* ASSERT */
1686
1687 // Force freeze slow path.
1688 push_cont_fastpath();
1689
1690 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1691 // Note: call_VM_base will use resume_pc label to set last_Java_pc.
1692 call_VM_base(noreg, noreg, noreg, &resume_pc, entry_point, number_of_arguments, false /*check_exceptions*/);
1693
1694 pop_cont_fastpath();
1695
1696 #ifdef ASSERT
1697 {
1698 Label L;
1699 decrementw(Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1700 ldrw(rscratch1, Address(rthread, JavaThread::interp_at_preemptable_vmcall_cnt_offset()));
1701 cmpw(rscratch1, 0);
1702 br(Assembler::GE, L);
1703 stop("call_VM_preemptable_helper: should be >= 0");
1704 bind(L);
1705 }
1706 #endif /* ASSERT */
1707
1708 // Check if preempted.
1709 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1710 cbz(rscratch1, not_preempted);
1711 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1712 br(rscratch1);
1713
1714 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1715 bind(resume_pc);
1716 restore_after_resume(false /* is_native */);
1717
1718 bind(not_preempted);
1719 if (check_exceptions) {
1720 // check for pending exceptions
1721 ldr(rscratch1, Address(rthread, in_bytes(Thread::pending_exception_offset())));
1722 Label ok;
1723 cbz(rscratch1, ok);
1724 lea(rscratch1, RuntimeAddress(StubRoutines::forward_exception_entry()));
1725 br(rscratch1);
1726 bind(ok);
1727 }
1728
1729 // get oop result if there is one and reset the value in the thread
1730 if (oop_result->is_valid()) {
1731 get_vm_result_oop(oop_result, rthread);
1732 }
1733 }
1734
1735 static void pass_arg1(MacroAssembler* masm, Register arg) {
1736 if (c_rarg1 != arg ) {
1737 masm->mov(c_rarg1, arg);
1738 }
1739 }
1740
1741 static void pass_arg2(MacroAssembler* masm, Register arg) {
1742 if (c_rarg2 != arg ) {
1743 masm->mov(c_rarg2, arg);
1744 }
1745 }
1746
1747 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1748 address entry_point,
1749 Register arg_1,
1750 bool check_exceptions) {
1751 pass_arg1(this, arg_1);
1752 call_VM_preemptable_helper(oop_result, entry_point, 1, check_exceptions);
1753 }
1754
1755 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1756 address entry_point,
1757 Register arg_1,
1758 Register arg_2,
1759 bool check_exceptions) {
1760 LP64_ONLY(assert_different_registers(arg_1, c_rarg2));
1761 pass_arg2(this, arg_2);
1762 pass_arg1(this, arg_1);
1763 call_VM_preemptable_helper(oop_result, entry_point, 2, check_exceptions);
1764 }
1765
1766 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1767 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1768 blr(rscratch1);
1769 if (is_native) {
1770 // On resume we need to set up stack as expected
1771 push(dtos);
1772 push(ltos);
1773 }
1774 }
1775
1776 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1777 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1778 Label update, next, none;
1779
1780 verify_oop(obj);
1781
1782 cbnz(obj, update);
1783 orptr(mdo_addr, TypeEntries::null_seen);
1784 b(next);
1785
1786 bind(update);
1787 load_klass(obj, obj);
1788
1789 ldr(rscratch1, mdo_addr);
1790 eor(obj, obj, rscratch1);
1791 tst(obj, TypeEntries::type_klass_mask);
1792 br(Assembler::EQ, next); // klass seen before, nothing to
1793 // do. The unknown bit may have been
1794 // set already but no need to check.
1795
1796 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1797 // already unknown. Nothing to do anymore.
1798
1799 cbz(rscratch1, none);
1800 cmp(rscratch1, (u1)TypeEntries::null_seen);
1801 br(Assembler::EQ, none);
1802 // There is a chance that the checks above
1803 // fail if another thread has just set the
1804 // profiling to this obj's klass
1805 eor(obj, obj, rscratch1); // get back original value before XOR
1806 ldr(rscratch1, mdo_addr);
1807 eor(obj, obj, rscratch1);
1808 tst(obj, TypeEntries::type_klass_mask);
1809 br(Assembler::EQ, next);
1810
1811 // different than before. Cannot keep accurate profile.
1812 orptr(mdo_addr, TypeEntries::type_unknown);
1813 b(next);
1814
1815 bind(none);
1816 // first time here. Set profile type.
1817 str(obj, mdo_addr);
1818 #ifdef ASSERT
1819 andr(obj, obj, TypeEntries::type_mask);
1820 verify_klass_ptr(obj);
1821 #endif
1822
1823 bind(next);
1824 }
1825
1826 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1827 if (!ProfileInterpreter) {
1828 return;
1829 }
1830
1831 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1832 Label profile_continue;
1833
1834 test_method_data_pointer(mdp, profile_continue);
1835
1836 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1837
1838 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1839 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1840 br(Assembler::NE, profile_continue);
1841
1842 if (MethodData::profile_arguments()) {
1843 Label done;
1844 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1845
1846 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1847 if (i > 0 || MethodData::profile_return()) {
1848 // If return value type is profiled we may have no argument to profile
1849 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1850 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1851 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1852 add(rscratch1, mdp, off_to_args);
1853 br(Assembler::LT, done);
1854 }
1855 ldr(tmp, Address(callee, Method::const_offset()));
1856 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1857 // stack offset o (zero based) from the start of the argument
1858 // list, for n arguments translates into offset n - o - 1 from
1859 // the end of the argument list
1860 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1861 sub(tmp, tmp, rscratch1);
1862 sub(tmp, tmp, 1);
1863 Address arg_addr = argument_address(tmp);
1864 ldr(tmp, arg_addr);
1865
1866 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1867 profile_obj_type(tmp, mdo_arg_addr);
1868
1869 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1870 off_to_args += to_add;
1871 }
1872
1873 if (MethodData::profile_return()) {
1874 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1875 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1876 }
1877
1878 add(rscratch1, mdp, off_to_args);
1879 bind(done);
1880 mov(mdp, rscratch1);
1881
1882 if (MethodData::profile_return()) {
1883 // We're right after the type profile for the last
1884 // argument. tmp is the number of cells left in the
1885 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1886 // if there's a return to profile.
1887 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1888 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1889 }
1890 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1891 } else {
1892 assert(MethodData::profile_return(), "either profile call args or call ret");
1893 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1894 }
1895
1896 // mdp points right after the end of the
1897 // CallTypeData/VirtualCallTypeData, right after the cells for the
1898 // return value type if there's one
1899
1900 bind(profile_continue);
1901 }
1902 }
1903
1904 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1905 assert_different_registers(mdp, ret, tmp, rbcp);
1906 if (ProfileInterpreter && MethodData::profile_return()) {
1907 Label profile_continue, done;
1908
1909 test_method_data_pointer(mdp, profile_continue);
1910
1911 if (MethodData::profile_return_jsr292_only()) {
1912 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1913
1914 // If we don't profile all invoke bytecodes we must make sure
1915 // it's a bytecode we indeed profile. We can't go back to the
1916 // beginning of the ProfileData we intend to update to check its
1917 // type because we're right after it and we don't known its
1918 // length
1919 Label do_profile;
1920 ldrb(rscratch1, Address(rbcp, 0));
1921 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1922 br(Assembler::EQ, do_profile);
1923 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1924 br(Assembler::EQ, do_profile);
1925 get_method(tmp);
1926 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1927 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1928 br(Assembler::NE, profile_continue);
1929
1930 bind(do_profile);
1931 }
1932
1933 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1934 mov(tmp, ret);
1935 profile_obj_type(tmp, mdo_ret_addr);
1936
1937 bind(profile_continue);
1938 }
1939 }
1940
1941 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1942 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1943 if (ProfileInterpreter && MethodData::profile_parameters()) {
1944 Label profile_continue, done;
1945
1946 test_method_data_pointer(mdp, profile_continue);
1947
1948 // Load the offset of the area within the MDO used for
1949 // parameters. If it's negative we're not profiling any parameters
1950 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1951 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1952
1953 // Compute a pointer to the area for parameters from the offset
1954 // and move the pointer to the slot for the last
1955 // parameters. Collect profiling from last parameter down.
1956 // mdo start + parameters offset + array length - 1
1957 add(mdp, mdp, tmp1);
1958 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1959 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1960
1961 Label loop;
1962 bind(loop);
1963
1964 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1965 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1966 int per_arg_scale = exact_log2(DataLayout::cell_size);
1967 add(rscratch1, mdp, off_base);
1968 add(rscratch2, mdp, type_base);
1969
1970 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1971 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1972
1973 // load offset on the stack from the slot for this parameter
1974 ldr(tmp2, arg_off);
1975 neg(tmp2, tmp2);
1976 // read the parameter from the local area
1977 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1978
1979 // profile the parameter
1980 profile_obj_type(tmp2, arg_type);
1981
1982 // go to next parameter
1983 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1984 br(Assembler::GE, loop);
1985
1986 bind(profile_continue);
1987 }
1988 }
1989
1990 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1991 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1992 get_cache_index_at_bcp(index, 1, sizeof(u4));
1993 // Get address of invokedynamic array
1994 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1995 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1996 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1997 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1998 lea(cache, Address(cache, index));
1999 }
2000
2001 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2002 // Get index out of bytecode pointer
2003 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2004 // Take shortcut if the size is a power of 2
2005 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2006 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2007 } else {
2008 mov(cache, sizeof(ResolvedFieldEntry));
2009 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2010 }
2011 // Get address of field entries array
2012 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
2013 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
2014 lea(cache, Address(cache, index));
2015 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
2016 membar(MacroAssembler::LoadLoad);
2017 }
2018
2019 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2020 // Get index out of bytecode pointer
2021 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2022 mov(cache, sizeof(ResolvedMethodEntry));
2023 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2024
2025 // Get address of field entries array
2026 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
2027 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2028 lea(cache, Address(cache, index));
2029 }
2030
2031 #ifdef ASSERT
2032 void InterpreterMacroAssembler::verify_field_offset(Register reg) {
2033 // Verify the field offset is not in the header, implicitly checks for 0
2034 Label L;
2035 subs(zr, reg, oopDesc::base_offset_in_bytes());
2036 br(Assembler::GE, L);
2037 stop("bad field offset");
2038 bind(L);
2039 }
2040 #endif