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 alloc_failed, 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);
234
235 // Grab the inline field klass
236 ldr(rscratch1, Address(entry, in_bytes(ResolvedFieldEntry::field_holder_offset())));
237 inline_layout_info(rscratch1, field_index, layout_info);
238
239 const Register field_klass = dst_temp;
240 ldr(field_klass, Address(layout_info, in_bytes(InlineLayoutInfo::klass_offset())));
241
242 // allocate buffer
243 push(obj); // save holder
244 allocate_instance(field_klass, obj, alloc_temp, rscratch2, false, alloc_failed);
245
246 // Have an oop instance buffer, copy into it
247 payload_address(obj, dst_temp, field_klass); // danger, uses rscratch1
248 pop(alloc_temp); // restore holder
249 lea(src, Address(alloc_temp, field_offset));
250 // call_VM_leaf, clobbers a few regs, save restore new obj
251 push(obj);
252 flat_field_copy(IS_DEST_UNINITIALIZED, src, dst_temp, layout_info);
253 pop(obj);
254 b(done);
255
256 bind(alloc_failed);
257 pop(obj);
258 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flat_field),
259 obj, entry);
260
261 bind(done);
262 membar(Assembler::StoreStore);
263 }
264
265 // Load object from cpool->resolved_references(index)
266 void InterpreterMacroAssembler::load_resolved_reference_at_index(
267 Register result, Register index, Register tmp) {
268 assert_different_registers(result, index);
269
270 get_constant_pool(result);
271 // load pointer for resolved_references[] objArray
272 ldr(result, Address(result, ConstantPool::cache_offset()));
273 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
274 resolve_oop_handle(result, tmp, rscratch2);
275 // Add in the index
276 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
277 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
278 }
279
280 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
281 Register cpool, Register index, Register klass, Register temp) {
282 add(temp, cpool, index, LSL, LogBytesPerWord);
283 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
284 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
285 add(klass, klass, temp, LSL, LogBytesPerWord);
286 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
287 }
288
289 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
290 // subtype of super_klass.
291 //
292 // Args:
293 // r0: superklass
294 // Rsub_klass: subklass
295 //
296 // Kills:
297 // r2, r5
298 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
299 Label& ok_is_subtype,
300 bool profile) {
301 assert(Rsub_klass != r0, "r0 holds superklass");
302 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
303 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
304
305 // Profile the not-null value's klass.
306 if (profile) {
307 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
308 }
309
310 // Do the check.
311 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
312 }
313
314 // Java Expression Stack
315
316 void InterpreterMacroAssembler::pop_ptr(Register r) {
317 ldr(r, post(esp, wordSize));
318 }
319
320 void InterpreterMacroAssembler::pop_i(Register r) {
321 ldrw(r, post(esp, wordSize));
322 }
323
324 void InterpreterMacroAssembler::pop_l(Register r) {
325 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
326 }
327
328 void InterpreterMacroAssembler::push_ptr(Register r) {
329 str(r, pre(esp, -wordSize));
330 }
331
332 void InterpreterMacroAssembler::push_i(Register r) {
333 str(r, pre(esp, -wordSize));
334 }
335
336 void InterpreterMacroAssembler::push_l(Register r) {
337 str(zr, pre(esp, -wordSize));
338 str(r, pre(esp, - wordSize));
339 }
340
341 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
342 ldrs(r, post(esp, wordSize));
343 }
344
345 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
346 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
347 }
348
349 void InterpreterMacroAssembler::push_f(FloatRegister r) {
350 strs(r, pre(esp, -wordSize));
351 }
352
353 void InterpreterMacroAssembler::push_d(FloatRegister r) {
354 strd(r, pre(esp, 2* -wordSize));
355 }
356
357 void InterpreterMacroAssembler::pop(TosState state) {
358 switch (state) {
359 case atos: pop_ptr(); break;
360 case btos:
361 case ztos:
362 case ctos:
363 case stos:
364 case itos: pop_i(); break;
365 case ltos: pop_l(); break;
366 case ftos: pop_f(); break;
367 case dtos: pop_d(); break;
368 case vtos: /* nothing to do */ break;
369 default: ShouldNotReachHere();
370 }
371 interp_verify_oop(r0, state);
372 }
373
374 void InterpreterMacroAssembler::push(TosState state) {
375 interp_verify_oop(r0, state);
376 switch (state) {
377 case atos: push_ptr(); break;
378 case btos:
379 case ztos:
380 case ctos:
381 case stos:
382 case itos: push_i(); break;
383 case ltos: push_l(); break;
384 case ftos: push_f(); break;
385 case dtos: push_d(); break;
386 case vtos: /* nothing to do */ break;
387 default : ShouldNotReachHere();
388 }
389 }
390
391 // Helpers for swap and dup
392 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
393 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
394 }
395
396 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
397 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
398 }
399
400 void InterpreterMacroAssembler::load_float(Address src) {
401 ldrs(v0, src);
402 }
403
404 void InterpreterMacroAssembler::load_double(Address src) {
405 ldrd(v0, src);
406 }
407
408 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
409 // set sender sp
410 mov(r19_sender_sp, sp);
411 // record last_sp
412 sub(rscratch1, esp, rfp);
413 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
414 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
415 }
416
417 // Jump to from_interpreted entry of a call unless single stepping is possible
418 // in this thread in which case we must call the i2i entry
419 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
420 prepare_to_jump_from_interpreted();
421
422 if (JvmtiExport::can_post_interpreter_events()) {
423 Label run_compiled_code;
424 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
425 // compiled code in threads for which the event is enabled. Check here for
426 // interp_only_mode if these events CAN be enabled.
427 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
428 cbzw(rscratch1, run_compiled_code);
429 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
430 br(rscratch1);
431 bind(run_compiled_code);
432 }
433
434 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
435 br(rscratch1);
436 }
437
438 // The following two routines provide a hook so that an implementation
439 // can schedule the dispatch in two parts. amd64 does not do this.
440 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
441 }
442
443 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
444 dispatch_next(state, step);
445 }
446
447 void InterpreterMacroAssembler::dispatch_base(TosState state,
448 address* table,
449 bool verifyoop,
450 bool generate_poll) {
451 if (VerifyActivationFrameSize) {
452 Label L;
453 sub(rscratch2, rfp, esp);
454 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
455 subs(rscratch2, rscratch2, min_frame_size);
456 br(Assembler::GE, L);
457 stop("broken stack frame");
458 bind(L);
459 }
460 if (verifyoop) {
461 interp_verify_oop(r0, state);
462 }
463
464 Label safepoint;
465 address* const safepoint_table = Interpreter::safept_table(state);
466 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
467
468 if (needs_thread_local_poll) {
469 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
470 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
471 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
472 }
473
474 if (table == Interpreter::dispatch_table(state)) {
475 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
476 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
477 } else {
478 mov(rscratch2, (address)table);
479 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
480 }
481 br(rscratch2);
482
483 if (needs_thread_local_poll) {
484 bind(safepoint);
485 lea(rscratch2, ExternalAddress((address)safepoint_table));
486 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
487 br(rscratch2);
488 }
489 }
490
491 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
492 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
493 }
494
495 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
496 dispatch_base(state, Interpreter::normal_table(state));
497 }
498
499 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
500 dispatch_base(state, Interpreter::normal_table(state), false);
501 }
502
503
504 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
505 // load next bytecode
506 ldrb(rscratch1, Address(pre(rbcp, step)));
507 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
508 }
509
510 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
511 // load current bytecode
512 ldrb(rscratch1, Address(rbcp, 0));
513 dispatch_base(state, table);
514 }
515
516 // remove activation
517 //
518 // Unlock the receiver if this is a synchronized method.
519 // Unlock any Java monitors from synchronized blocks.
520 // Apply stack watermark barrier.
521 // Notify JVMTI.
522 // Remove the activation from the stack.
523 //
524 // If there are locked Java monitors
525 // If throw_monitor_exception
526 // throws IllegalMonitorStateException
527 // Else if install_monitor_exception
528 // installs IllegalMonitorStateException
529 // Else
530 // no error processing
531 void InterpreterMacroAssembler::remove_activation(TosState state,
532 bool throw_monitor_exception,
533 bool install_monitor_exception,
534 bool notify_jvmdi) {
535 // Note: Registers r3 xmm0 may be in use for the
536 // result check if synchronized method
537 Label unlocked, unlock, no_unlock;
538
539 // get the value of _do_not_unlock_if_synchronized into r3
540 const Address do_not_unlock_if_synchronized(rthread,
541 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
542 ldrb(r3, do_not_unlock_if_synchronized);
543 strb(zr, do_not_unlock_if_synchronized); // reset the flag
544
545 // get method access flags
546 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
547 ldrh(r2, Address(r1, Method::access_flags_offset()));
548 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
549
550 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
551 // is set.
552 cbnz(r3, no_unlock);
553
554 // unlock monitor
555 push(state); // save result
556
557 // BasicObjectLock will be first in list, since this is a
558 // synchronized method. However, need to check that the object has
559 // not been unlocked by an explicit monitorexit bytecode.
560 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
561 wordSize - (int) sizeof(BasicObjectLock));
562 // We use c_rarg1 so that if we go slow path it will be the correct
563 // register for unlock_object to pass to VM directly
564 lea(c_rarg1, monitor); // address of first monitor
565
566 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
567 cbnz(r0, unlock);
568
569 pop(state);
570 if (throw_monitor_exception) {
571 // Entry already unlocked, need to throw exception
572 call_VM(noreg, CAST_FROM_FN_PTR(address,
573 InterpreterRuntime::throw_illegal_monitor_state_exception));
574 should_not_reach_here();
575 } else {
576 // Monitor already unlocked during a stack unroll. If requested,
577 // install an illegal_monitor_state_exception. Continue with
578 // stack unrolling.
579 if (install_monitor_exception) {
580 call_VM(noreg, CAST_FROM_FN_PTR(address,
581 InterpreterRuntime::new_illegal_monitor_state_exception));
582 }
583 b(unlocked);
584 }
585
586 bind(unlock);
587 unlock_object(c_rarg1);
588 pop(state);
589
590 // Check that for block-structured locking (i.e., that all locked
591 // objects has been unlocked)
592 bind(unlocked);
593
594 // r0: Might contain return value
595
596 // Check that all monitors are unlocked
597 {
598 Label loop, exception, entry, restart;
599 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
600 const Address monitor_block_top(
601 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
602 const Address monitor_block_bot(
603 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
604
605 bind(restart);
606 // We use c_rarg1 so that if we go slow path it will be the correct
607 // register for unlock_object to pass to VM directly
608 ldr(c_rarg1, monitor_block_top); // derelativize pointer
609 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
610 // c_rarg1 points to current entry, starting with top-most entry
611
612 lea(r19, monitor_block_bot); // points to word before bottom of
613 // monitor block
614 b(entry);
615
616 // Entry already locked, need to throw exception
617 bind(exception);
618
619 if (throw_monitor_exception) {
620 // Throw exception
621 MacroAssembler::call_VM(noreg,
622 CAST_FROM_FN_PTR(address, InterpreterRuntime::
623 throw_illegal_monitor_state_exception));
624 should_not_reach_here();
625 } else {
626 // Stack unrolling. Unlock object and install illegal_monitor_exception.
627 // Unlock does not block, so don't have to worry about the frame.
628 // We don't have to preserve c_rarg1 since we are going to throw an exception.
629
630 push(state);
631 unlock_object(c_rarg1);
632 pop(state);
633
634 if (install_monitor_exception) {
635 call_VM(noreg, CAST_FROM_FN_PTR(address,
636 InterpreterRuntime::
637 new_illegal_monitor_state_exception));
638 }
639
640 b(restart);
641 }
642
643 bind(loop);
644 // check if current entry is used
645 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
646 cbnz(rscratch1, exception);
647
648 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
649 bind(entry);
650 cmp(c_rarg1, r19); // check if bottom reached
651 br(Assembler::NE, loop); // if not at bottom then check this entry
652 }
653
654 bind(no_unlock);
655
656 JFR_ONLY(enter_jfr_critical_section();)
657
658 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
659 // that would normally not be safe to use. Such bad returns into unsafe territory of
660 // the stack, will call InterpreterRuntime::at_unwind.
661 Label slow_path;
662 Label fast_path;
663 safepoint_poll(slow_path, true /* at_return */, false /* acquire */, false /* in_nmethod */);
664 br(Assembler::AL, fast_path);
665 bind(slow_path);
666 push(state);
667 set_last_Java_frame(esp, rfp, pc(), rscratch1);
668 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
669 reset_last_Java_frame(true);
670 pop(state);
671 bind(fast_path);
672
673 // JVMTI support. Make sure the safepoint poll test is issued prior.
674 if (notify_jvmdi) {
675 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
676 } else {
677 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
678 }
679
680 // remove activation
681 // get sender esp
682 ldr(rscratch2,
683 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
684
685 if (StackReservedPages > 0) {
686 // testing if reserved zone needs to be re-enabled
687 Label no_reserved_zone_enabling;
688
689 // check if already enabled - if so no re-enabling needed
690 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
691 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
692 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
693 br(Assembler::EQ, no_reserved_zone_enabling);
694
695 // look for an overflow into the stack reserved zone, i.e.
696 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
697 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
698 cmp(rscratch2, rscratch1);
699 br(Assembler::LS, no_reserved_zone_enabling);
700
701 JFR_ONLY(leave_jfr_critical_section();)
702
703 call_VM_leaf(
704 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
705 call_VM(noreg, CAST_FROM_FN_PTR(address,
706 InterpreterRuntime::throw_delayed_StackOverflowError));
707 should_not_reach_here();
708
709 bind(no_reserved_zone_enabling);
710 }
711
712 if (state == atos && InlineTypeReturnedAsFields) {
713 Label skip;
714 Label not_null;
715 cbnz(r0, not_null);
716 // Returned value is null, zero all return registers because they may belong to oop fields
717 mov(j_rarg1, zr);
718 mov(j_rarg2, zr);
719 mov(j_rarg3, zr);
720 mov(j_rarg4, zr);
721 mov(j_rarg5, zr);
722 mov(j_rarg6, zr);
723 mov(j_rarg7, zr);
724 b(skip);
725 bind(not_null);
726
727 // Check if we are returning an non-null inline type and load its fields into registers
728 test_oop_is_not_inline_type(r0, rscratch2, skip, /* can_be_null= */ false);
729
730 // Load fields from a buffered value with an inline class specific handler
731 load_klass(rscratch1 /*dst*/, r0 /*src*/);
732 ldr(rscratch1, Address(rscratch1, InstanceKlass::adr_inlineklass_fixed_block_offset()));
733 ldr(rscratch1, Address(rscratch1, InlineKlass::unpack_handler_offset()));
734 // Unpack handler can be null if inline type is not scalarizable in returns
735 cbz(rscratch1, skip);
736
737 blr(rscratch1);
738 #ifdef ASSERT
739 // TODO 8284443 Enable
740 if (StressCallingConvention && false) {
741 Label skip_stress;
742 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
743 ldrw(rscratch1, Address(rscratch1, Method::flags_offset()));
744 tstw(rscratch1, MethodFlags::has_scalarized_return_flag());
745 br(Assembler::EQ, skip_stress);
746 load_klass(r0, r0);
747 orr(r0, r0, 1);
748 bind(skip_stress);
749 }
750 #endif
751 bind(skip);
752 // Check above kills sender esp in rscratch2. Reload it.
753 ldr(rscratch2, Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
754 }
755
756 // remove frame anchor
757 leave();
758
759 JFR_ONLY(leave_jfr_critical_section();)
760
761 // restore sender esp
762 mov(esp, rscratch2);
763
764 // If we're returning to interpreted code we will shortly be
765 // adjusting SP to allow some space for ESP. If we're returning to
766 // compiled code the saved sender SP was saved in sender_sp, so this
767 // restores it.
768 andr(sp, esp, -16);
769 }
770
771 #if INCLUDE_JFR
772 void InterpreterMacroAssembler::enter_jfr_critical_section() {
773 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
774 mov(rscratch1, true);
775 strb(rscratch1, sampling_critical_section);
776 }
777
778 void InterpreterMacroAssembler::leave_jfr_critical_section() {
779 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
780 strb(zr, sampling_critical_section);
781 }
782 #endif // INCLUDE_JFR
783
784 // Lock object
785 //
786 // Args:
787 // c_rarg1: BasicObjectLock to be used for locking
788 //
789 // Kills:
790 // r0
791 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
792 // rscratch1, rscratch2 (scratch regs)
793 void InterpreterMacroAssembler::lock_object(Register lock_reg)
794 {
795 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
796 if (LockingMode == LM_MONITOR) {
797 call_VM_preemptable(noreg,
798 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
799 lock_reg);
800 } else {
801 Label count, done;
802
803 const Register swap_reg = r0;
804 const Register tmp = c_rarg2;
805 const Register obj_reg = c_rarg3; // Will contain the oop
806 const Register tmp2 = c_rarg4;
807 const Register tmp3 = c_rarg5;
808
809 const int obj_offset = in_bytes(BasicObjectLock::obj_offset());
810 const int lock_offset = in_bytes(BasicObjectLock::lock_offset());
811 const int mark_offset = lock_offset +
812 BasicLock::displaced_header_offset_in_bytes();
813
814 Label slow_case;
815
816 // Load object pointer into obj_reg %c_rarg3
817 ldr(obj_reg, Address(lock_reg, obj_offset));
818
819 if (LockingMode == LM_LIGHTWEIGHT) {
820 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
821 b(done);
822 } else if (LockingMode == LM_LEGACY) {
823
824 if (DiagnoseSyncOnValueBasedClasses != 0) {
825 load_klass(tmp, obj_reg);
826 ldrb(tmp, Address(tmp, Klass::misc_flags_offset()));
827 tst(tmp, KlassFlags::_misc_is_value_based_class);
828 br(Assembler::NE, slow_case);
829 }
830
831 // Load (object->mark() | 1) into swap_reg
832 ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
833 orr(swap_reg, rscratch1, 1);
834 if (EnableValhalla) {
835 // Mask inline_type bit such that we go to the slow path if object is an inline type
836 andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
837 }
838
839 // Save (object->mark() | 1) into BasicLock's displaced header
840 str(swap_reg, Address(lock_reg, mark_offset));
841
842 assert(lock_offset == 0,
843 "displached header must be first word in BasicObjectLock");
844
845 Label fail;
846 cmpxchg_obj_header(swap_reg, lock_reg, obj_reg, rscratch1, count, /*fallthrough*/nullptr);
847
848 // Fast check for recursive lock.
849 //
850 // Can apply the optimization only if this is a stack lock
851 // allocated in this thread. For efficiency, we can focus on
852 // recently allocated stack locks (instead of reading the stack
853 // base and checking whether 'mark' points inside the current
854 // thread stack):
855 // 1) (mark & 7) == 0, and
856 // 2) sp <= mark < mark + os::pagesize()
857 //
858 // Warning: sp + os::pagesize can overflow the stack base. We must
859 // neither apply the optimization for an inflated lock allocated
860 // just above the thread stack (this is why condition 1 matters)
861 // nor apply the optimization if the stack lock is inside the stack
862 // of another thread. The latter is avoided even in case of overflow
863 // because we have guard pages at the end of all stacks. Hence, if
864 // we go over the stack base and hit the stack of another thread,
865 // this should not be in a writeable area that could contain a
866 // stack lock allocated by that thread. As a consequence, a stack
867 // lock less than page size away from sp is guaranteed to be
868 // owned by the current thread.
869 //
870 // These 3 tests can be done by evaluating the following
871 // expression: ((mark - sp) & (7 - os::vm_page_size())),
872 // assuming both stack pointer and pagesize have their
873 // least significant 3 bits clear.
874 // NOTE: the mark is in swap_reg %r0 as the result of cmpxchg
875 // NOTE2: aarch64 does not like to subtract sp from rn so take a
876 // copy
877 mov(rscratch1, sp);
878 sub(swap_reg, swap_reg, rscratch1);
879 ands(swap_reg, swap_reg, (uint64_t)(7 - (int)os::vm_page_size()));
880
881 // Save the test result, for recursive case, the result is zero
882 str(swap_reg, Address(lock_reg, mark_offset));
883 br(Assembler::NE, slow_case);
884
885 bind(count);
886 inc_held_monitor_count(rscratch1);
887 b(done);
888 }
889 bind(slow_case);
890
891 // Call the runtime routine for slow case
892 call_VM_preemptable(noreg,
893 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
894 lock_reg);
895
896 bind(done);
897 }
898 }
899
900
901 // Unlocks an object. Used in monitorexit bytecode and
902 // remove_activation. Throws an IllegalMonitorException if object is
903 // not locked by current thread.
904 //
905 // Args:
906 // c_rarg1: BasicObjectLock for lock
907 //
908 // Kills:
909 // r0
910 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
911 // rscratch1, rscratch2 (scratch regs)
912 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
913 {
914 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
915
916 if (LockingMode == LM_MONITOR) {
917 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
918 } else {
919 Label count, done;
920
921 const Register swap_reg = r0;
922 const Register header_reg = c_rarg2; // Will contain the old oopMark
923 const Register obj_reg = c_rarg3; // Will contain the oop
924 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
925
926 save_bcp(); // Save in case of exception
927
928 if (LockingMode != LM_LIGHTWEIGHT) {
929 // Convert from BasicObjectLock structure to object and BasicLock
930 // structure Store the BasicLock address into %r0
931 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset()));
932 }
933
934 // Load oop into obj_reg(%c_rarg3)
935 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
936
937 // Free entry
938 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
939
940 Label slow_case;
941 if (LockingMode == LM_LIGHTWEIGHT) {
942 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
943 b(done);
944 } else if (LockingMode == LM_LEGACY) {
945 // Load the old header from BasicLock structure
946 ldr(header_reg, Address(swap_reg,
947 BasicLock::displaced_header_offset_in_bytes()));
948
949 // Test for recursion
950 cbz(header_reg, count);
951
952 // Atomic swap back the old header
953 cmpxchg_obj_header(swap_reg, header_reg, obj_reg, rscratch1, count, &slow_case);
954
955 bind(count);
956 dec_held_monitor_count(rscratch1);
957 b(done);
958 }
959
960 bind(slow_case);
961 // Call the runtime routine for slow case.
962 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
963 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
964 bind(done);
965 restore_bcp();
966 }
967 }
968
969 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
970 Label& zero_continue) {
971 assert(ProfileInterpreter, "must be profiling interpreter");
972 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
973 cbz(mdp, zero_continue);
974 }
975
976 // Set the method data pointer for the current bcp.
977 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
978 assert(ProfileInterpreter, "must be profiling interpreter");
979 Label set_mdp;
980 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
981
982 // Test MDO to avoid the call if it is null.
983 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
984 cbz(r0, set_mdp);
985 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
986 // r0: mdi
987 // mdo is guaranteed to be non-zero here, we checked for it before the call.
988 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
989 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
990 add(r0, r1, r0);
991 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
992 bind(set_mdp);
993 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
994 }
995
996 void InterpreterMacroAssembler::verify_method_data_pointer() {
997 assert(ProfileInterpreter, "must be profiling interpreter");
998 #ifdef ASSERT
999 Label verify_continue;
1000 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
1001 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
1002 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
1003 get_method(r1);
1004
1005 // If the mdp is valid, it will point to a DataLayout header which is
1006 // consistent with the bcp. The converse is highly probable also.
1007 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
1008 ldr(rscratch1, Address(r1, Method::const_offset()));
1009 add(r2, r2, rscratch1, Assembler::LSL);
1010 lea(r2, Address(r2, ConstMethod::codes_offset()));
1011 cmp(r2, rbcp);
1012 br(Assembler::EQ, verify_continue);
1013 // r1: method
1014 // rbcp: bcp // rbcp == 22
1015 // r3: mdp
1016 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1017 r1, rbcp, r3);
1018 bind(verify_continue);
1019 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
1020 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
1021 #endif // ASSERT
1022 }
1023
1024
1025 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1026 int constant,
1027 Register value) {
1028 assert(ProfileInterpreter, "must be profiling interpreter");
1029 Address data(mdp_in, constant);
1030 str(value, data);
1031 }
1032
1033
1034 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1035 int constant,
1036 bool decrement) {
1037 increment_mdp_data_at(mdp_in, noreg, constant, decrement);
1038 }
1039
1040 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1041 Register reg,
1042 int constant,
1043 bool decrement) {
1044 assert(ProfileInterpreter, "must be profiling interpreter");
1045 // %%% this does 64bit counters at best it is wasting space
1046 // at worst it is a rare bug when counters overflow
1047
1048 assert_different_registers(rscratch2, rscratch1, mdp_in, reg);
1049
1050 Address addr1(mdp_in, constant);
1051 Address addr2(rscratch2, reg, Address::lsl(0));
1052 Address &addr = addr1;
1053 if (reg != noreg) {
1054 lea(rscratch2, addr1);
1055 addr = addr2;
1056 }
1057
1058 if (decrement) {
1059 // Decrement the register. Set condition codes.
1060 // Intel does this
1061 // addptr(data, (int32_t) -DataLayout::counter_increment);
1062 // If the decrement causes the counter to overflow, stay negative
1063 // Label L;
1064 // jcc(Assembler::negative, L);
1065 // addptr(data, (int32_t) DataLayout::counter_increment);
1066 // so we do this
1067 ldr(rscratch1, addr);
1068 subs(rscratch1, rscratch1, (unsigned)DataLayout::counter_increment);
1069 Label L;
1070 br(Assembler::LO, L); // skip store if counter underflow
1071 str(rscratch1, addr);
1072 bind(L);
1073 } else {
1074 assert(DataLayout::counter_increment == 1,
1075 "flow-free idiom only works with 1");
1076 // Intel does this
1077 // Increment the register. Set carry flag.
1078 // addptr(data, DataLayout::counter_increment);
1079 // If the increment causes the counter to overflow, pull back by 1.
1080 // sbbptr(data, (int32_t)0);
1081 // so we do this
1082 ldr(rscratch1, addr);
1083 adds(rscratch1, rscratch1, DataLayout::counter_increment);
1084 Label L;
1085 br(Assembler::CS, L); // skip store if counter overflow
1086 str(rscratch1, addr);
1087 bind(L);
1088 }
1089 }
1090
1091 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1092 int flag_byte_constant) {
1093 assert(ProfileInterpreter, "must be profiling interpreter");
1094 int flags_offset = in_bytes(DataLayout::flags_offset());
1095 // Set the flag
1096 ldrb(rscratch1, Address(mdp_in, flags_offset));
1097 orr(rscratch1, rscratch1, flag_byte_constant);
1098 strb(rscratch1, Address(mdp_in, flags_offset));
1099 }
1100
1101
1102 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1103 int offset,
1104 Register value,
1105 Register test_value_out,
1106 Label& not_equal_continue) {
1107 assert(ProfileInterpreter, "must be profiling interpreter");
1108 if (test_value_out == noreg) {
1109 ldr(rscratch1, Address(mdp_in, offset));
1110 cmp(value, rscratch1);
1111 } else {
1112 // Put the test value into a register, so caller can use it:
1113 ldr(test_value_out, Address(mdp_in, offset));
1114 cmp(value, test_value_out);
1115 }
1116 br(Assembler::NE, not_equal_continue);
1117 }
1118
1119
1120 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1121 int offset_of_disp) {
1122 assert(ProfileInterpreter, "must be profiling interpreter");
1123 ldr(rscratch1, Address(mdp_in, offset_of_disp));
1124 add(mdp_in, mdp_in, rscratch1, LSL);
1125 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1126 }
1127
1128
1129 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1130 Register reg,
1131 int offset_of_disp) {
1132 assert(ProfileInterpreter, "must be profiling interpreter");
1133 lea(rscratch1, Address(mdp_in, offset_of_disp));
1134 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
1135 add(mdp_in, mdp_in, rscratch1, LSL);
1136 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1137 }
1138
1139
1140 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1141 int constant) {
1142 assert(ProfileInterpreter, "must be profiling interpreter");
1143 add(mdp_in, mdp_in, (unsigned)constant);
1144 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1145 }
1146
1147
1148 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1149 assert(ProfileInterpreter, "must be profiling interpreter");
1150 // save/restore across call_VM
1151 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
1152 call_VM(noreg,
1153 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1154 return_bci);
1155 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
1156 }
1157
1158
1159 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1160 Register bumped_count) {
1161 if (ProfileInterpreter) {
1162 Label profile_continue;
1163
1164 // If no method data exists, go to profile_continue.
1165 // Otherwise, assign to mdp
1166 test_method_data_pointer(mdp, profile_continue);
1167
1168 // We are taking a branch. Increment the taken count.
1169 // We inline increment_mdp_data_at to return bumped_count in a register
1170 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1171 Address data(mdp, in_bytes(JumpData::taken_offset()));
1172 ldr(bumped_count, data);
1173 assert(DataLayout::counter_increment == 1,
1174 "flow-free idiom only works with 1");
1175 // Intel does this to catch overflow
1176 // addptr(bumped_count, DataLayout::counter_increment);
1177 // sbbptr(bumped_count, 0);
1178 // so we do this
1179 adds(bumped_count, bumped_count, DataLayout::counter_increment);
1180 Label L;
1181 br(Assembler::CS, L); // skip store if counter overflow
1182 str(bumped_count, data);
1183 bind(L);
1184 // The method data pointer needs to be updated to reflect the new target.
1185 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1186 bind(profile_continue);
1187 }
1188 }
1189
1190
1191 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp, bool acmp) {
1192 if (ProfileInterpreter) {
1193 Label profile_continue;
1194
1195 // If no method data exists, go to profile_continue.
1196 test_method_data_pointer(mdp, profile_continue);
1197
1198 // We are taking a branch. Increment the not taken count.
1199 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1200
1201 // The method data pointer needs to be updated to correspond to
1202 // the next bytecode
1203 update_mdp_by_constant(mdp, acmp ? in_bytes(ACmpData::acmp_data_size()) : in_bytes(BranchData::branch_data_size()));
1204 bind(profile_continue);
1205 }
1206 }
1207
1208
1209 void InterpreterMacroAssembler::profile_call(Register mdp) {
1210 if (ProfileInterpreter) {
1211 Label profile_continue;
1212
1213 // If no method data exists, go to profile_continue.
1214 test_method_data_pointer(mdp, profile_continue);
1215
1216 // We are making a call. Increment the count.
1217 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1218
1219 // The method data pointer needs to be updated to reflect the new target.
1220 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1221 bind(profile_continue);
1222 }
1223 }
1224
1225 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1226 if (ProfileInterpreter) {
1227 Label profile_continue;
1228
1229 // If no method data exists, go to profile_continue.
1230 test_method_data_pointer(mdp, profile_continue);
1231
1232 // We are making a call. Increment the count.
1233 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1234
1235 // The method data pointer needs to be updated to reflect the new target.
1236 update_mdp_by_constant(mdp,
1237 in_bytes(VirtualCallData::
1238 virtual_call_data_size()));
1239 bind(profile_continue);
1240 }
1241 }
1242
1243
1244 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1245 Register mdp,
1246 Register reg2,
1247 bool receiver_can_be_null) {
1248 if (ProfileInterpreter) {
1249 Label profile_continue;
1250
1251 // If no method data exists, go to profile_continue.
1252 test_method_data_pointer(mdp, profile_continue);
1253
1254 Label skip_receiver_profile;
1255 if (receiver_can_be_null) {
1256 Label not_null;
1257 // We are making a call. Increment the count for null receiver.
1258 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1259 b(skip_receiver_profile);
1260 bind(not_null);
1261 }
1262
1263 // Record the receiver type.
1264 record_klass_in_profile(receiver, mdp, reg2);
1265 bind(skip_receiver_profile);
1266
1267 // The method data pointer needs to be updated to reflect the new target.
1268 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1269 bind(profile_continue);
1270 }
1271 }
1272
1273 // This routine creates a state machine for updating the multi-row
1274 // type profile at a virtual call site (or other type-sensitive bytecode).
1275 // The machine visits each row (of receiver/count) until the receiver type
1276 // is found, or until it runs out of rows. At the same time, it remembers
1277 // the location of the first empty row. (An empty row records null for its
1278 // receiver, and can be allocated for a newly-observed receiver type.)
1279 // Because there are two degrees of freedom in the state, a simple linear
1280 // search will not work; it must be a decision tree. Hence this helper
1281 // function is recursive, to generate the required tree structured code.
1282 // It's the interpreter, so we are trading off code space for speed.
1283 // See below for example code.
1284 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1285 Register receiver, Register mdp,
1286 Register reg2, int start_row,
1287 Label& done) {
1288 if (TypeProfileWidth == 0) {
1289 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1290 } else {
1291 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1292 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1293 }
1294 }
1295
1296 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1297 Register reg2, int start_row, Label& done, int total_rows,
1298 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1299 int last_row = total_rows - 1;
1300 assert(start_row <= last_row, "must be work left to do");
1301 // Test this row for both the item and for null.
1302 // Take any of three different outcomes:
1303 // 1. found item => increment count and goto done
1304 // 2. found null => keep looking for case 1, maybe allocate this cell
1305 // 3. found something else => keep looking for cases 1 and 2
1306 // Case 3 is handled by a recursive call.
1307 for (int row = start_row; row <= last_row; row++) {
1308 Label next_test;
1309 bool test_for_null_also = (row == start_row);
1310
1311 // See if the item is item[n].
1312 int item_offset = in_bytes(item_offset_fn(row));
1313 test_mdp_data_at(mdp, item_offset, item,
1314 (test_for_null_also ? reg2 : noreg),
1315 next_test);
1316 // (Reg2 now contains the item from the CallData.)
1317
1318 // The item is item[n]. Increment count[n].
1319 int count_offset = in_bytes(item_count_offset_fn(row));
1320 increment_mdp_data_at(mdp, count_offset);
1321 b(done);
1322 bind(next_test);
1323
1324 if (test_for_null_also) {
1325 Label found_null;
1326 // Failed the equality check on item[n]... Test for null.
1327 if (start_row == last_row) {
1328 // The only thing left to do is handle the null case.
1329 cbz(reg2, found_null);
1330 // Item did not match any saved item and there is no empty row for it.
1331 // Increment total counter to indicate polymorphic case.
1332 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1333 b(done);
1334 bind(found_null);
1335 break;
1336 }
1337 // Since null is rare, make it be the branch-taken case.
1338 cbz(reg2, found_null);
1339
1340 // Put all the "Case 3" tests here.
1341 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1342 item_offset_fn, item_count_offset_fn);
1343
1344 // Found a null. Keep searching for a matching item,
1345 // but remember that this is an empty (unused) slot.
1346 bind(found_null);
1347 }
1348 }
1349
1350 // In the fall-through case, we found no matching item, but we
1351 // observed the item[start_row] is null.
1352
1353 // Fill in the item field and increment the count.
1354 int item_offset = in_bytes(item_offset_fn(start_row));
1355 set_mdp_data_at(mdp, item_offset, item);
1356 int count_offset = in_bytes(item_count_offset_fn(start_row));
1357 mov(reg2, DataLayout::counter_increment);
1358 set_mdp_data_at(mdp, count_offset, reg2);
1359 if (start_row > 0) {
1360 b(done);
1361 }
1362 }
1363
1364 // Example state machine code for three profile rows:
1365 // // main copy of decision tree, rooted at row[1]
1366 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1367 // if (row[0].rec != nullptr) {
1368 // // inner copy of decision tree, rooted at row[1]
1369 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1370 // if (row[1].rec != nullptr) {
1371 // // degenerate decision tree, rooted at row[2]
1372 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1373 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1374 // row[2].init(rec); goto done;
1375 // } else {
1376 // // remember row[1] is empty
1377 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1378 // row[1].init(rec); goto done;
1379 // }
1380 // } else {
1381 // // remember row[0] is empty
1382 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1383 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1384 // row[0].init(rec); goto done;
1385 // }
1386 // done:
1387
1388 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1389 Register mdp, Register reg2) {
1390 assert(ProfileInterpreter, "must be profiling");
1391 Label done;
1392
1393 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1394
1395 bind (done);
1396 }
1397
1398 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1399 Register mdp) {
1400 if (ProfileInterpreter) {
1401 Label profile_continue;
1402 uint row;
1403
1404 // If no method data exists, go to profile_continue.
1405 test_method_data_pointer(mdp, profile_continue);
1406
1407 // Update the total ret count.
1408 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1409
1410 for (row = 0; row < RetData::row_limit(); row++) {
1411 Label next_test;
1412
1413 // See if return_bci is equal to bci[n]:
1414 test_mdp_data_at(mdp,
1415 in_bytes(RetData::bci_offset(row)),
1416 return_bci, noreg,
1417 next_test);
1418
1419 // return_bci is equal to bci[n]. Increment the count.
1420 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1421
1422 // The method data pointer needs to be updated to reflect the new target.
1423 update_mdp_by_offset(mdp,
1424 in_bytes(RetData::bci_displacement_offset(row)));
1425 b(profile_continue);
1426 bind(next_test);
1427 }
1428
1429 update_mdp_for_ret(return_bci);
1430
1431 bind(profile_continue);
1432 }
1433 }
1434
1435 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1436 if (ProfileInterpreter) {
1437 Label profile_continue;
1438
1439 // If no method data exists, go to profile_continue.
1440 test_method_data_pointer(mdp, profile_continue);
1441
1442 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1443
1444 // The method data pointer needs to be updated.
1445 int mdp_delta = in_bytes(BitData::bit_data_size());
1446 if (TypeProfileCasts) {
1447 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1448 }
1449 update_mdp_by_constant(mdp, mdp_delta);
1450
1451 bind(profile_continue);
1452 }
1453 }
1454
1455 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1456 if (ProfileInterpreter) {
1457 Label profile_continue;
1458
1459 // If no method data exists, go to profile_continue.
1460 test_method_data_pointer(mdp, profile_continue);
1461
1462 // The method data pointer needs to be updated.
1463 int mdp_delta = in_bytes(BitData::bit_data_size());
1464 if (TypeProfileCasts) {
1465 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1466
1467 // Record the object type.
1468 record_klass_in_profile(klass, mdp, reg2);
1469 }
1470 update_mdp_by_constant(mdp, mdp_delta);
1471
1472 bind(profile_continue);
1473 }
1474 }
1475
1476 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1477 if (ProfileInterpreter) {
1478 Label profile_continue;
1479
1480 // If no method data exists, go to profile_continue.
1481 test_method_data_pointer(mdp, profile_continue);
1482
1483 // Update the default case count
1484 increment_mdp_data_at(mdp,
1485 in_bytes(MultiBranchData::default_count_offset()));
1486
1487 // The method data pointer needs to be updated.
1488 update_mdp_by_offset(mdp,
1489 in_bytes(MultiBranchData::
1490 default_displacement_offset()));
1491
1492 bind(profile_continue);
1493 }
1494 }
1495
1496 void InterpreterMacroAssembler::profile_switch_case(Register index,
1497 Register mdp,
1498 Register reg2) {
1499 if (ProfileInterpreter) {
1500 Label profile_continue;
1501
1502 // If no method data exists, go to profile_continue.
1503 test_method_data_pointer(mdp, profile_continue);
1504
1505 // Build the base (index * per_case_size_in_bytes()) +
1506 // case_array_offset_in_bytes()
1507 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1508 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1509 Assembler::maddw(index, index, reg2, rscratch1);
1510
1511 // Update the case count
1512 increment_mdp_data_at(mdp,
1513 index,
1514 in_bytes(MultiBranchData::relative_count_offset()));
1515
1516 // The method data pointer needs to be updated.
1517 update_mdp_by_offset(mdp,
1518 index,
1519 in_bytes(MultiBranchData::
1520 relative_displacement_offset()));
1521
1522 bind(profile_continue);
1523 }
1524 }
1525
1526 template <class ArrayData> void InterpreterMacroAssembler::profile_array_type(Register mdp,
1527 Register array,
1528 Register tmp) {
1529 if (ProfileInterpreter) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(mdp, profile_continue);
1534
1535 mov(tmp, array);
1536 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayData::array_offset())));
1537
1538 Label not_flat;
1539 test_non_flat_array_oop(array, tmp, not_flat);
1540
1541 set_mdp_flag_at(mdp, ArrayData::flat_array_byte_constant());
1542
1543 bind(not_flat);
1544
1545 Label not_null_free;
1546 test_non_null_free_array_oop(array, tmp, not_null_free);
1547
1548 set_mdp_flag_at(mdp, ArrayData::null_free_array_byte_constant());
1549
1550 bind(not_null_free);
1551
1552 bind(profile_continue);
1553 }
1554 }
1555
1556 template void InterpreterMacroAssembler::profile_array_type<ArrayLoadData>(Register mdp,
1557 Register array,
1558 Register tmp);
1559 template void InterpreterMacroAssembler::profile_array_type<ArrayStoreData>(Register mdp,
1560 Register array,
1561 Register tmp);
1562
1563 void InterpreterMacroAssembler::profile_multiple_element_types(Register mdp, Register element, Register tmp, const Register tmp2) {
1564 if (ProfileInterpreter) {
1565 Label profile_continue;
1566
1567 // If no method data exists, go to profile_continue.
1568 test_method_data_pointer(mdp, profile_continue);
1569
1570 Label done, update;
1571 cbnz(element, update);
1572 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1573 b(done);
1574
1575 bind(update);
1576 load_klass(tmp, element);
1577
1578 // Record the object type.
1579 record_klass_in_profile(tmp, mdp, tmp2);
1580
1581 bind(done);
1582
1583 // The method data pointer needs to be updated.
1584 update_mdp_by_constant(mdp, in_bytes(ArrayStoreData::array_store_data_size()));
1585
1586 bind(profile_continue);
1587 }
1588 }
1589
1590
1591 void InterpreterMacroAssembler::profile_element_type(Register mdp,
1592 Register element,
1593 Register tmp) {
1594 if (ProfileInterpreter) {
1595 Label profile_continue;
1596
1597 // If no method data exists, go to profile_continue.
1598 test_method_data_pointer(mdp, profile_continue);
1599
1600 mov(tmp, element);
1601 profile_obj_type(tmp, Address(mdp, in_bytes(ArrayLoadData::element_offset())));
1602
1603 // The method data pointer needs to be updated.
1604 update_mdp_by_constant(mdp, in_bytes(ArrayLoadData::array_load_data_size()));
1605
1606 bind(profile_continue);
1607 }
1608 }
1609
1610 void InterpreterMacroAssembler::profile_acmp(Register mdp,
1611 Register left,
1612 Register right,
1613 Register tmp) {
1614 if (ProfileInterpreter) {
1615 Label profile_continue;
1616
1617 // If no method data exists, go to profile_continue.
1618 test_method_data_pointer(mdp, profile_continue);
1619
1620 mov(tmp, left);
1621 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::left_offset())));
1622
1623 Label left_not_inline_type;
1624 test_oop_is_not_inline_type(left, tmp, left_not_inline_type);
1625 set_mdp_flag_at(mdp, ACmpData::left_inline_type_byte_constant());
1626 bind(left_not_inline_type);
1627
1628 mov(tmp, right);
1629 profile_obj_type(tmp, Address(mdp, in_bytes(ACmpData::right_offset())));
1630
1631 Label right_not_inline_type;
1632 test_oop_is_not_inline_type(right, tmp, right_not_inline_type);
1633 set_mdp_flag_at(mdp, ACmpData::right_inline_type_byte_constant());
1634 bind(right_not_inline_type);
1635
1636 bind(profile_continue);
1637 }
1638 }
1639
1640 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1641 if (state == atos) {
1642 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1643 }
1644 }
1645
1646 void InterpreterMacroAssembler::notify_method_entry() {
1647 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1648 // track stack depth. If it is possible to enter interp_only_mode we add
1649 // the code to check if the event should be sent.
1650 if (JvmtiExport::can_post_interpreter_events()) {
1651 Label L;
1652 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1653 cbzw(r3, L);
1654 call_VM(noreg, CAST_FROM_FN_PTR(address,
1655 InterpreterRuntime::post_method_entry));
1656 bind(L);
1657 }
1658
1659 if (DTraceMethodProbes) {
1660 get_method(c_rarg1);
1661 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1662 rthread, c_rarg1);
1663 }
1664
1665 // RedefineClasses() tracing support for obsolete method entry
1666 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1667 get_method(c_rarg1);
1668 call_VM_leaf(
1669 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1670 rthread, c_rarg1);
1671 }
1672
1673 }
1674
1675
1676 void InterpreterMacroAssembler::notify_method_exit(
1677 TosState state, NotifyMethodExitMode mode) {
1678 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1679 // track stack depth. If it is possible to enter interp_only_mode we add
1680 // the code to check if the event should be sent.
1681 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1682 Label L;
1683 // Note: frame::interpreter_frame_result has a dependency on how the
1684 // method result is saved across the call to post_method_exit. If this
1685 // is changed then the interpreter_frame_result implementation will
1686 // need to be updated too.
1687
1688 // template interpreter will leave the result on the top of the stack.
1689 push(state);
1690 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1691 cbz(r3, L);
1692 call_VM(noreg,
1693 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1694 bind(L);
1695 pop(state);
1696 }
1697
1698 if (DTraceMethodProbes) {
1699 push(state);
1700 get_method(c_rarg1);
1701 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1702 rthread, c_rarg1);
1703 pop(state);
1704 }
1705 }
1706
1707
1708 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1709 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1710 int increment, Address mask,
1711 Register scratch, Register scratch2,
1712 bool preloaded, Condition cond,
1713 Label* where) {
1714 if (!preloaded) {
1715 ldrw(scratch, counter_addr);
1716 }
1717 add(scratch, scratch, increment);
1718 strw(scratch, counter_addr);
1719 ldrw(scratch2, mask);
1720 ands(scratch, scratch, scratch2);
1721 br(cond, *where);
1722 }
1723
1724 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1725 int number_of_arguments) {
1726 // interpreter specific
1727 //
1728 // Note: No need to save/restore rbcp & rlocals pointer since these
1729 // are callee saved registers and no blocking/ GC can happen
1730 // in leaf calls.
1731 #ifdef ASSERT
1732 {
1733 Label L;
1734 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1735 cbz(rscratch1, L);
1736 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1737 " last_sp != nullptr");
1738 bind(L);
1739 }
1740 #endif /* ASSERT */
1741 // super call
1742 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1743 }
1744
1745 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1746 Register java_thread,
1747 Register last_java_sp,
1748 address entry_point,
1749 int number_of_arguments,
1750 bool check_exceptions) {
1751 // interpreter specific
1752 //
1753 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1754 // really make a difference for these runtime calls, since they are
1755 // slow anyway. Btw., bcp must be saved/restored since it may change
1756 // due to GC.
1757 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1758 save_bcp();
1759 #ifdef ASSERT
1760 {
1761 Label L;
1762 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1763 cbz(rscratch1, L);
1764 stop("InterpreterMacroAssembler::call_VM_base:"
1765 " last_sp != nullptr");
1766 bind(L);
1767 }
1768 #endif /* ASSERT */
1769 // super call
1770 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1771 entry_point, number_of_arguments,
1772 check_exceptions);
1773 // interpreter specific
1774 restore_bcp();
1775 restore_locals();
1776 }
1777
1778 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1779 address entry_point,
1780 Register arg_1) {
1781 assert(arg_1 == c_rarg1, "");
1782 Label resume_pc, not_preempted;
1783
1784 #ifdef ASSERT
1785 {
1786 Label L;
1787 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1788 cbz(rscratch1, L);
1789 stop("Should not have alternate return address set");
1790 bind(L);
1791 }
1792 #endif /* ASSERT */
1793
1794 // Force freeze slow path.
1795 push_cont_fastpath();
1796
1797 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1798 adr(rscratch1, resume_pc);
1799 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1800 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1801
1802 pop_cont_fastpath();
1803
1804 // Check if preempted.
1805 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1806 cbz(rscratch1, not_preempted);
1807 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1808 br(rscratch1);
1809
1810 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1811 bind(resume_pc);
1812 restore_after_resume(false /* is_native */);
1813
1814 bind(not_preempted);
1815 }
1816
1817 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1818 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1819 blr(rscratch1);
1820 if (is_native) {
1821 // On resume we need to set up stack as expected
1822 push(dtos);
1823 push(ltos);
1824 }
1825 }
1826
1827 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1828 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1829 Label update, next, none;
1830
1831 verify_oop(obj);
1832
1833 cbnz(obj, update);
1834 orptr(mdo_addr, TypeEntries::null_seen);
1835 b(next);
1836
1837 bind(update);
1838 load_klass(obj, obj);
1839
1840 ldr(rscratch1, mdo_addr);
1841 eor(obj, obj, rscratch1);
1842 tst(obj, TypeEntries::type_klass_mask);
1843 br(Assembler::EQ, next); // klass seen before, nothing to
1844 // do. The unknown bit may have been
1845 // set already but no need to check.
1846
1847 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1848 // already unknown. Nothing to do anymore.
1849
1850 cbz(rscratch1, none);
1851 cmp(rscratch1, (u1)TypeEntries::null_seen);
1852 br(Assembler::EQ, none);
1853 // There is a chance that the checks above
1854 // fail if another thread has just set the
1855 // profiling to this obj's klass
1856 eor(obj, obj, rscratch1); // get back original value before XOR
1857 ldr(rscratch1, mdo_addr);
1858 eor(obj, obj, rscratch1);
1859 tst(obj, TypeEntries::type_klass_mask);
1860 br(Assembler::EQ, next);
1861
1862 // different than before. Cannot keep accurate profile.
1863 orptr(mdo_addr, TypeEntries::type_unknown);
1864 b(next);
1865
1866 bind(none);
1867 // first time here. Set profile type.
1868 str(obj, mdo_addr);
1869 #ifdef ASSERT
1870 andr(obj, obj, TypeEntries::type_mask);
1871 verify_klass_ptr(obj);
1872 #endif
1873
1874 bind(next);
1875 }
1876
1877 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1878 if (!ProfileInterpreter) {
1879 return;
1880 }
1881
1882 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1883 Label profile_continue;
1884
1885 test_method_data_pointer(mdp, profile_continue);
1886
1887 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1888
1889 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1890 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1891 br(Assembler::NE, profile_continue);
1892
1893 if (MethodData::profile_arguments()) {
1894 Label done;
1895 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1896
1897 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1898 if (i > 0 || MethodData::profile_return()) {
1899 // If return value type is profiled we may have no argument to profile
1900 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1901 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1902 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1903 add(rscratch1, mdp, off_to_args);
1904 br(Assembler::LT, done);
1905 }
1906 ldr(tmp, Address(callee, Method::const_offset()));
1907 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1908 // stack offset o (zero based) from the start of the argument
1909 // list, for n arguments translates into offset n - o - 1 from
1910 // the end of the argument list
1911 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1912 sub(tmp, tmp, rscratch1);
1913 sub(tmp, tmp, 1);
1914 Address arg_addr = argument_address(tmp);
1915 ldr(tmp, arg_addr);
1916
1917 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1918 profile_obj_type(tmp, mdo_arg_addr);
1919
1920 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1921 off_to_args += to_add;
1922 }
1923
1924 if (MethodData::profile_return()) {
1925 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1926 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1927 }
1928
1929 add(rscratch1, mdp, off_to_args);
1930 bind(done);
1931 mov(mdp, rscratch1);
1932
1933 if (MethodData::profile_return()) {
1934 // We're right after the type profile for the last
1935 // argument. tmp is the number of cells left in the
1936 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1937 // if there's a return to profile.
1938 assert(SingleTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1939 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1940 }
1941 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1942 } else {
1943 assert(MethodData::profile_return(), "either profile call args or call ret");
1944 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1945 }
1946
1947 // mdp points right after the end of the
1948 // CallTypeData/VirtualCallTypeData, right after the cells for the
1949 // return value type if there's one
1950
1951 bind(profile_continue);
1952 }
1953 }
1954
1955 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1956 assert_different_registers(mdp, ret, tmp, rbcp);
1957 if (ProfileInterpreter && MethodData::profile_return()) {
1958 Label profile_continue, done;
1959
1960 test_method_data_pointer(mdp, profile_continue);
1961
1962 if (MethodData::profile_return_jsr292_only()) {
1963 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1964
1965 // If we don't profile all invoke bytecodes we must make sure
1966 // it's a bytecode we indeed profile. We can't go back to the
1967 // beginning of the ProfileData we intend to update to check its
1968 // type because we're right after it and we don't known its
1969 // length
1970 Label do_profile;
1971 ldrb(rscratch1, Address(rbcp, 0));
1972 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1973 br(Assembler::EQ, do_profile);
1974 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1975 br(Assembler::EQ, do_profile);
1976 get_method(tmp);
1977 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1978 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1979 br(Assembler::NE, profile_continue);
1980
1981 bind(do_profile);
1982 }
1983
1984 Address mdo_ret_addr(mdp, -in_bytes(SingleTypeEntry::size()));
1985 mov(tmp, ret);
1986 profile_obj_type(tmp, mdo_ret_addr);
1987
1988 bind(profile_continue);
1989 }
1990 }
1991
1992 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1993 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1994 if (ProfileInterpreter && MethodData::profile_parameters()) {
1995 Label profile_continue, done;
1996
1997 test_method_data_pointer(mdp, profile_continue);
1998
1999 // Load the offset of the area within the MDO used for
2000 // parameters. If it's negative we're not profiling any parameters
2001 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
2002 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
2003
2004 // Compute a pointer to the area for parameters from the offset
2005 // and move the pointer to the slot for the last
2006 // parameters. Collect profiling from last parameter down.
2007 // mdo start + parameters offset + array length - 1
2008 add(mdp, mdp, tmp1);
2009 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
2010 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
2011
2012 Label loop;
2013 bind(loop);
2014
2015 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2016 int type_base = in_bytes(ParametersTypeData::type_offset(0));
2017 int per_arg_scale = exact_log2(DataLayout::cell_size);
2018 add(rscratch1, mdp, off_base);
2019 add(rscratch2, mdp, type_base);
2020
2021 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
2022 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
2023
2024 // load offset on the stack from the slot for this parameter
2025 ldr(tmp2, arg_off);
2026 neg(tmp2, tmp2);
2027 // read the parameter from the local area
2028 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
2029
2030 // profile the parameter
2031 profile_obj_type(tmp2, arg_type);
2032
2033 // go to next parameter
2034 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
2035 br(Assembler::GE, loop);
2036
2037 bind(profile_continue);
2038 }
2039 }
2040
2041 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
2042 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
2043 get_cache_index_at_bcp(index, 1, sizeof(u4));
2044 // Get address of invokedynamic array
2045 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
2046 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
2047 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
2048 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
2049 lea(cache, Address(cache, index));
2050 }
2051
2052 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
2053 // Get index out of bytecode pointer
2054 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2055 // Take shortcut if the size is a power of 2
2056 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
2057 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
2058 } else {
2059 mov(cache, sizeof(ResolvedFieldEntry));
2060 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
2061 }
2062 // Get address of field entries array
2063 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
2064 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
2065 lea(cache, Address(cache, index));
2066 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
2067 membar(MacroAssembler::LoadLoad);
2068 }
2069
2070 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
2071 // Get index out of bytecode pointer
2072 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
2073 mov(cache, sizeof(ResolvedMethodEntry));
2074 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
2075
2076 // Get address of field entries array
2077 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
2078 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
2079 lea(cache, Address(cache, index));
2080 }