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/markWord.hpp"
36 #include "oops/method.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/resolvedFieldEntry.hpp"
39 #include "oops/resolvedIndyEntry.hpp"
40 #include "oops/resolvedMethodEntry.hpp"
41 #include "prims/jvmtiExport.hpp"
42 #include "prims/jvmtiThreadState.hpp"
43 #include "runtime/basicLock.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/javaThread.hpp"
46 #include "runtime/safepointMechanism.hpp"
47 #include "runtime/sharedRuntime.hpp"
48 #include "utilities/powerOfTwo.hpp"
49
50 void InterpreterMacroAssembler::narrow(Register result) {
51
52 // Get method->_constMethod->_result_type
53 ldr(rscratch1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
54 ldr(rscratch1, Address(rscratch1, Method::const_offset()));
55 ldrb(rscratch1, Address(rscratch1, ConstMethod::result_type_offset()));
56
57 Label done, notBool, notByte, notChar;
58
59 // common case first
60 cmpw(rscratch1, T_INT);
61 br(Assembler::EQ, done);
62
63 // mask integer result to narrower return type.
64 cmpw(rscratch1, T_BOOLEAN);
65 br(Assembler::NE, notBool);
66 andw(result, result, 0x1);
67 b(done);
68
69 bind(notBool);
70 cmpw(rscratch1, T_BYTE);
71 br(Assembler::NE, notByte);
72 sbfx(result, result, 0, 8);
73 b(done);
74
75 bind(notByte);
76 cmpw(rscratch1, T_CHAR);
77 br(Assembler::NE, notChar);
78 ubfx(result, result, 0, 16); // truncate upper 16 bits
79 b(done);
80
81 bind(notChar);
82 sbfx(result, result, 0, 16); // sign-extend short
83
84 // Nothing to do for T_INT
85 bind(done);
86 }
87
88 void InterpreterMacroAssembler::jump_to_entry(address entry) {
89 assert(entry, "Entry must have been generated by now");
90 b(entry);
91 }
92
93 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
94 if (JvmtiExport::can_pop_frame()) {
95 Label L;
96 // Initiate popframe handling only if it is not already being
97 // processed. If the flag has the popframe_processing bit set, it
98 // means that this code is called *during* popframe handling - we
99 // don't want to reenter.
100 // This method is only called just after the call into the vm in
101 // call_VM_base, so the arg registers are available.
102 ldrw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
103 tbz(rscratch1, exact_log2(JavaThread::popframe_pending_bit), L);
104 tbnz(rscratch1, exact_log2(JavaThread::popframe_processing_bit), L);
105 // Call Interpreter::remove_activation_preserving_args_entry() to get the
106 // address of the same-named entrypoint in the generated interpreter code.
107 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
108 br(r0);
109 bind(L);
110 }
111 }
112
113
114 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
115 ldr(r2, Address(rthread, JavaThread::jvmti_thread_state_offset()));
116 const Address tos_addr(r2, JvmtiThreadState::earlyret_tos_offset());
117 const Address oop_addr(r2, JvmtiThreadState::earlyret_oop_offset());
118 const Address val_addr(r2, JvmtiThreadState::earlyret_value_offset());
119 switch (state) {
120 case atos: ldr(r0, oop_addr);
121 str(zr, oop_addr);
122 interp_verify_oop(r0, state); break;
123 case ltos: ldr(r0, val_addr); break;
124 case btos: // fall through
125 case ztos: // fall through
126 case ctos: // fall through
127 case stos: // fall through
128 case itos: ldrw(r0, val_addr); break;
129 case ftos: ldrs(v0, val_addr); break;
130 case dtos: ldrd(v0, val_addr); break;
131 case vtos: /* nothing to do */ break;
132 default : ShouldNotReachHere();
133 }
134 // Clean up tos value in the thread object
135 movw(rscratch1, (int) ilgl);
136 strw(rscratch1, tos_addr);
137 strw(zr, val_addr);
138 }
139
140
141 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
142 if (JvmtiExport::can_force_early_return()) {
143 Label L;
144 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
145 cbz(rscratch1, L); // if (thread->jvmti_thread_state() == nullptr) exit;
146
147 // Initiate earlyret handling only if it is not already being processed.
148 // If the flag has the earlyret_processing bit set, it means that this code
149 // is called *during* earlyret handling - we don't want to reenter.
150 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_state_offset()));
151 cmpw(rscratch1, JvmtiThreadState::earlyret_pending);
152 br(Assembler::NE, L);
153
154 // Call Interpreter::remove_activation_early_entry() to get the address of the
155 // same-named entrypoint in the generated interpreter code.
156 ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
157 ldrw(rscratch1, Address(rscratch1, JvmtiThreadState::earlyret_tos_offset()));
158 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), rscratch1);
159 br(r0);
160 bind(L);
161 }
162 }
163
164 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
165 Register reg,
166 int bcp_offset) {
167 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
168 ldrh(reg, Address(rbcp, bcp_offset));
169 rev16(reg, reg);
170 }
171
172 void InterpreterMacroAssembler::get_dispatch() {
173 uint64_t offset;
174 adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
175 // Use add() here after ARDP, rather than lea().
176 // lea() does not generate anything if its offset is zero.
177 // However, relocs expect to find either an ADD or a load/store
178 // insn after an ADRP. add() always generates an ADD insn, even
179 // for add(Rn, Rn, 0).
180 add(rdispatch, rdispatch, offset);
181 }
182
183 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
184 int bcp_offset,
185 size_t index_size) {
186 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
187 if (index_size == sizeof(u2)) {
188 load_unsigned_short(index, Address(rbcp, bcp_offset));
189 } else if (index_size == sizeof(u4)) {
190 // assert(EnableInvokeDynamic, "giant index used only for JSR 292");
191 ldrw(index, Address(rbcp, bcp_offset));
192 } else if (index_size == sizeof(u1)) {
193 load_unsigned_byte(index, Address(rbcp, bcp_offset));
194 } else {
195 ShouldNotReachHere();
196 }
197 }
198
199 void InterpreterMacroAssembler::get_method_counters(Register method,
200 Register mcs, Label& skip) {
201 Label has_counters;
202 ldr(mcs, Address(method, Method::method_counters_offset()));
203 cbnz(mcs, has_counters);
204 call_VM(noreg, CAST_FROM_FN_PTR(address,
205 InterpreterRuntime::build_method_counters), method);
206 ldr(mcs, Address(method, Method::method_counters_offset()));
207 cbz(mcs, skip); // No MethodCounters allocated, OutOfMemory
208 bind(has_counters);
209 }
210
211 // Load object from cpool->resolved_references(index)
212 void InterpreterMacroAssembler::load_resolved_reference_at_index(
213 Register result, Register index, Register tmp) {
214 assert_different_registers(result, index);
215
216 get_constant_pool(result);
217 // load pointer for resolved_references[] objArray
218 ldr(result, Address(result, ConstantPool::cache_offset()));
219 ldr(result, Address(result, ConstantPoolCache::resolved_references_offset()));
220 resolve_oop_handle(result, tmp, rscratch2);
221 // Add in the index
222 add(index, index, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
223 load_heap_oop(result, Address(result, index, Address::uxtw(LogBytesPerHeapOop)), tmp, rscratch2);
224 }
225
226 void InterpreterMacroAssembler::load_resolved_klass_at_offset(
227 Register cpool, Register index, Register klass, Register temp) {
228 add(temp, cpool, index, LSL, LogBytesPerWord);
229 ldrh(temp, Address(temp, sizeof(ConstantPool))); // temp = resolved_klass_index
230 ldr(klass, Address(cpool, ConstantPool::resolved_klasses_offset())); // klass = cpool->_resolved_klasses
231 add(klass, klass, temp, LSL, LogBytesPerWord);
232 ldr(klass, Address(klass, Array<Klass*>::base_offset_in_bytes()));
233 }
234
235 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
236 // subtype of super_klass.
237 //
238 // Args:
239 // r0: superklass
240 // Rsub_klass: subklass
241 //
242 // Kills:
243 // r2, r5
244 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
245 Label& ok_is_subtype) {
246 assert(Rsub_klass != r0, "r0 holds superklass");
247 assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
248 assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
249
250 // Profile the not-null value's klass.
251 profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
252
253 // Do the check.
254 check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
255 }
256
257 // Java Expression Stack
258
259 void InterpreterMacroAssembler::pop_ptr(Register r) {
260 ldr(r, post(esp, wordSize));
261 }
262
263 void InterpreterMacroAssembler::pop_i(Register r) {
264 ldrw(r, post(esp, wordSize));
265 }
266
267 void InterpreterMacroAssembler::pop_l(Register r) {
268 ldr(r, post(esp, 2 * Interpreter::stackElementSize));
269 }
270
271 void InterpreterMacroAssembler::push_ptr(Register r) {
272 str(r, pre(esp, -wordSize));
273 }
274
275 void InterpreterMacroAssembler::push_i(Register r) {
276 str(r, pre(esp, -wordSize));
277 }
278
279 void InterpreterMacroAssembler::push_l(Register r) {
280 str(zr, pre(esp, -wordSize));
281 str(r, pre(esp, - wordSize));
282 }
283
284 void InterpreterMacroAssembler::pop_f(FloatRegister r) {
285 ldrs(r, post(esp, wordSize));
286 }
287
288 void InterpreterMacroAssembler::pop_d(FloatRegister r) {
289 ldrd(r, post(esp, 2 * Interpreter::stackElementSize));
290 }
291
292 void InterpreterMacroAssembler::push_f(FloatRegister r) {
293 strs(r, pre(esp, -wordSize));
294 }
295
296 void InterpreterMacroAssembler::push_d(FloatRegister r) {
297 strd(r, pre(esp, 2* -wordSize));
298 }
299
300 void InterpreterMacroAssembler::pop(TosState state) {
301 switch (state) {
302 case atos: pop_ptr(); break;
303 case btos:
304 case ztos:
305 case ctos:
306 case stos:
307 case itos: pop_i(); break;
308 case ltos: pop_l(); break;
309 case ftos: pop_f(); break;
310 case dtos: pop_d(); break;
311 case vtos: /* nothing to do */ break;
312 default: ShouldNotReachHere();
313 }
314 interp_verify_oop(r0, state);
315 }
316
317 void InterpreterMacroAssembler::push(TosState state) {
318 interp_verify_oop(r0, state);
319 switch (state) {
320 case atos: push_ptr(); break;
321 case btos:
322 case ztos:
323 case ctos:
324 case stos:
325 case itos: push_i(); break;
326 case ltos: push_l(); break;
327 case ftos: push_f(); break;
328 case dtos: push_d(); break;
329 case vtos: /* nothing to do */ break;
330 default : ShouldNotReachHere();
331 }
332 }
333
334 // Helpers for swap and dup
335 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
336 ldr(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
337 }
338
339 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
340 str(val, Address(esp, Interpreter::expr_offset_in_bytes(n)));
341 }
342
343 void InterpreterMacroAssembler::load_float(Address src) {
344 ldrs(v0, src);
345 }
346
347 void InterpreterMacroAssembler::load_double(Address src) {
348 ldrd(v0, src);
349 }
350
351 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
352 // set sender sp
353 mov(r19_sender_sp, sp);
354 // record last_sp
355 sub(rscratch1, esp, rfp);
356 asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
357 str(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
358 }
359
360 // Jump to from_interpreted entry of a call unless single stepping is possible
361 // in this thread in which case we must call the i2i entry
362 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
363 prepare_to_jump_from_interpreted();
364
365 if (JvmtiExport::can_post_interpreter_events()) {
366 Label run_compiled_code;
367 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
368 // compiled code in threads for which the event is enabled. Check here for
369 // interp_only_mode if these events CAN be enabled.
370 ldrw(rscratch1, Address(rthread, JavaThread::interp_only_mode_offset()));
371 cbzw(rscratch1, run_compiled_code);
372 ldr(rscratch1, Address(method, Method::interpreter_entry_offset()));
373 br(rscratch1);
374 bind(run_compiled_code);
375 }
376
377 ldr(rscratch1, Address(method, Method::from_interpreted_offset()));
378 br(rscratch1);
379 }
380
381 // The following two routines provide a hook so that an implementation
382 // can schedule the dispatch in two parts. amd64 does not do this.
383 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
384 }
385
386 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
387 dispatch_next(state, step);
388 }
389
390 void InterpreterMacroAssembler::dispatch_base(TosState state,
391 address* table,
392 bool verifyoop,
393 bool generate_poll) {
394 if (VerifyActivationFrameSize) {
395 Label L;
396 sub(rscratch2, rfp, esp);
397 int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
398 subs(rscratch2, rscratch2, min_frame_size);
399 br(Assembler::GE, L);
400 stop("broken stack frame");
401 bind(L);
402 }
403 if (verifyoop) {
404 interp_verify_oop(r0, state);
405 }
406
407 Label safepoint;
408 address* const safepoint_table = Interpreter::safept_table(state);
409 bool needs_thread_local_poll = generate_poll && table != safepoint_table;
410
411 if (needs_thread_local_poll) {
412 NOT_PRODUCT(block_comment("Thread-local Safepoint poll"));
413 ldr(rscratch2, Address(rthread, JavaThread::polling_word_offset()));
414 tbnz(rscratch2, exact_log2(SafepointMechanism::poll_bit()), safepoint);
415 }
416
417 if (table == Interpreter::dispatch_table(state)) {
418 addw(rscratch2, rscratch1, Interpreter::distance_from_dispatch_table(state));
419 ldr(rscratch2, Address(rdispatch, rscratch2, Address::uxtw(3)));
420 } else {
421 mov(rscratch2, (address)table);
422 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
423 }
424 br(rscratch2);
425
426 if (needs_thread_local_poll) {
427 bind(safepoint);
428 lea(rscratch2, ExternalAddress((address)safepoint_table));
429 ldr(rscratch2, Address(rscratch2, rscratch1, Address::uxtw(3)));
430 br(rscratch2);
431 }
432 }
433
434 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) {
435 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll);
436 }
437
438 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
439 dispatch_base(state, Interpreter::normal_table(state));
440 }
441
442 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
443 dispatch_base(state, Interpreter::normal_table(state), false);
444 }
445
446
447 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) {
448 // load next bytecode
449 ldrb(rscratch1, Address(pre(rbcp, step)));
450 dispatch_base(state, Interpreter::dispatch_table(state), /*verifyoop*/true, generate_poll);
451 }
452
453 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
454 // load current bytecode
455 ldrb(rscratch1, Address(rbcp, 0));
456 dispatch_base(state, table);
457 }
458
459 // remove activation
460 //
461 // Unlock the receiver if this is a synchronized method.
462 // Unlock any Java monitors from synchronized blocks.
463 // Apply stack watermark barrier.
464 // Notify JVMTI.
465 // Remove the activation from the stack.
466 //
467 // If there are locked Java monitors
468 // If throw_monitor_exception
469 // throws IllegalMonitorStateException
470 // Else if install_monitor_exception
471 // installs IllegalMonitorStateException
472 // Else
473 // no error processing
474 void InterpreterMacroAssembler::remove_activation(TosState state,
475 bool throw_monitor_exception,
476 bool install_monitor_exception,
477 bool notify_jvmdi) {
478 // Note: Registers r3 xmm0 may be in use for the
479 // result check if synchronized method
480 Label unlocked, unlock, no_unlock;
481
482 // get the value of _do_not_unlock_if_synchronized into r3
483 const Address do_not_unlock_if_synchronized(rthread,
484 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
485 ldrb(r3, do_not_unlock_if_synchronized);
486 strb(zr, do_not_unlock_if_synchronized); // reset the flag
487
488 // get method access flags
489 ldr(r1, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
490 ldrh(r2, Address(r1, Method::access_flags_offset()));
491 tbz(r2, exact_log2(JVM_ACC_SYNCHRONIZED), unlocked);
492
493 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
494 // is set.
495 cbnz(r3, no_unlock);
496
497 // unlock monitor
498 push(state); // save result
499
500 // BasicObjectLock will be first in list, since this is a
501 // synchronized method. However, need to check that the object has
502 // not been unlocked by an explicit monitorexit bytecode.
503 const Address monitor(rfp, frame::interpreter_frame_initial_sp_offset *
504 wordSize - (int) sizeof(BasicObjectLock));
505 // We use c_rarg1 so that if we go slow path it will be the correct
506 // register for unlock_object to pass to VM directly
507 lea(c_rarg1, monitor); // address of first monitor
508
509 ldr(r0, Address(c_rarg1, BasicObjectLock::obj_offset()));
510 cbnz(r0, unlock);
511
512 pop(state);
513 if (throw_monitor_exception) {
514 // Entry already unlocked, need to throw exception
515 call_VM(noreg, CAST_FROM_FN_PTR(address,
516 InterpreterRuntime::throw_illegal_monitor_state_exception));
517 should_not_reach_here();
518 } else {
519 // Monitor already unlocked during a stack unroll. If requested,
520 // install an illegal_monitor_state_exception. Continue with
521 // stack unrolling.
522 if (install_monitor_exception) {
523 call_VM(noreg, CAST_FROM_FN_PTR(address,
524 InterpreterRuntime::new_illegal_monitor_state_exception));
525 }
526 b(unlocked);
527 }
528
529 bind(unlock);
530 unlock_object(c_rarg1);
531 pop(state);
532
533 // Check that for block-structured locking (i.e., that all locked
534 // objects has been unlocked)
535 bind(unlocked);
536
537 // r0: Might contain return value
538
539 // Check that all monitors are unlocked
540 {
541 Label loop, exception, entry, restart;
542 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
543 const Address monitor_block_top(
544 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
545 const Address monitor_block_bot(
546 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
547
548 bind(restart);
549 // We use c_rarg1 so that if we go slow path it will be the correct
550 // register for unlock_object to pass to VM directly
551 ldr(c_rarg1, monitor_block_top); // derelativize pointer
552 lea(c_rarg1, Address(rfp, c_rarg1, Address::lsl(Interpreter::logStackElementSize)));
553 // c_rarg1 points to current entry, starting with top-most entry
554
555 lea(r19, monitor_block_bot); // points to word before bottom of
556 // monitor block
557 b(entry);
558
559 // Entry already locked, need to throw exception
560 bind(exception);
561
562 if (throw_monitor_exception) {
563 // Throw exception
564 MacroAssembler::call_VM(noreg,
565 CAST_FROM_FN_PTR(address, InterpreterRuntime::
566 throw_illegal_monitor_state_exception));
567 should_not_reach_here();
568 } else {
569 // Stack unrolling. Unlock object and install illegal_monitor_exception.
570 // Unlock does not block, so don't have to worry about the frame.
571 // We don't have to preserve c_rarg1 since we are going to throw an exception.
572
573 push(state);
574 unlock_object(c_rarg1);
575 pop(state);
576
577 if (install_monitor_exception) {
578 call_VM(noreg, CAST_FROM_FN_PTR(address,
579 InterpreterRuntime::
580 new_illegal_monitor_state_exception));
581 }
582
583 b(restart);
584 }
585
586 bind(loop);
587 // check if current entry is used
588 ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset()));
589 cbnz(rscratch1, exception);
590
591 add(c_rarg1, c_rarg1, entry_size); // otherwise advance to next entry
592 bind(entry);
593 cmp(c_rarg1, r19); // check if bottom reached
594 br(Assembler::NE, loop); // if not at bottom then check this entry
595 }
596
597 bind(no_unlock);
598
599 JFR_ONLY(enter_jfr_critical_section();)
600
601 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily,
602 // that would normally not be safe to use. Such bad returns into unsafe territory of
603 // the stack, will call InterpreterRuntime::at_unwind.
604 Label slow_path;
605 Label fast_path;
606 safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
607 br(Assembler::AL, fast_path);
608 bind(slow_path);
609 push(state);
610 set_last_Java_frame(esp, rfp, pc(), rscratch1);
611 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
612 reset_last_Java_frame(true);
613 pop(state);
614 bind(fast_path);
615
616 // JVMTI support. Make sure the safepoint poll test is issued prior.
617 if (notify_jvmdi) {
618 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
619 } else {
620 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
621 }
622
623 // remove activation
624 // get sender esp
625 ldr(rscratch2,
626 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
627 if (StackReservedPages > 0) {
628 // testing if reserved zone needs to be re-enabled
629 Label no_reserved_zone_enabling;
630
631 // check if already enabled - if so no re-enabling needed
632 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size");
633 ldrw(rscratch1, Address(rthread, JavaThread::stack_guard_state_offset()));
634 cmpw(rscratch1, (u1)StackOverflow::stack_guard_enabled);
635 br(Assembler::EQ, no_reserved_zone_enabling);
636
637 // look for an overflow into the stack reserved zone, i.e.
638 // interpreter_frame_sender_sp <= JavaThread::reserved_stack_activation
639 ldr(rscratch1, Address(rthread, JavaThread::reserved_stack_activation_offset()));
640 cmp(rscratch2, rscratch1);
641 br(Assembler::LS, no_reserved_zone_enabling);
642
643 JFR_ONLY(leave_jfr_critical_section();)
644
645 call_VM_leaf(
646 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread);
647 call_VM(noreg, CAST_FROM_FN_PTR(address,
648 InterpreterRuntime::throw_delayed_StackOverflowError));
649 should_not_reach_here();
650
651 bind(no_reserved_zone_enabling);
652 }
653
654 // remove frame anchor
655 leave();
656
657 JFR_ONLY(leave_jfr_critical_section();)
658
659 // restore sender esp
660 mov(esp, rscratch2);
661
662 // If we're returning to interpreted code we will shortly be
663 // adjusting SP to allow some space for ESP. If we're returning to
664 // compiled code the saved sender SP was saved in sender_sp, so this
665 // restores it.
666 andr(sp, esp, -16);
667 }
668
669 #if INCLUDE_JFR
670 void InterpreterMacroAssembler::enter_jfr_critical_section() {
671 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
672 mov(rscratch1, true);
673 strb(rscratch1, sampling_critical_section);
674 }
675
676 void InterpreterMacroAssembler::leave_jfr_critical_section() {
677 const Address sampling_critical_section(rthread, in_bytes(SAMPLING_CRITICAL_SECTION_OFFSET_JFR));
678 strb(zr, sampling_critical_section);
679 }
680 #endif // INCLUDE_JFR
681
682 // Lock object
683 //
684 // Args:
685 // c_rarg1: BasicObjectLock to be used for locking
686 //
687 // Kills:
688 // r0
689 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, c_rarg4, .. (param regs)
690 // rscratch1, rscratch2 (scratch regs)
691 void InterpreterMacroAssembler::lock_object(Register lock_reg)
692 {
693 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
694
695 const Register tmp = c_rarg2;
696 const Register obj_reg = c_rarg3; // Will contain the oop
697 const Register tmp2 = c_rarg4;
698 const Register tmp3 = c_rarg5;
699
700 // Load object pointer into obj_reg %c_rarg3
701 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
702
703 Label slow_case, done;
704 lightweight_lock(lock_reg, obj_reg, tmp, tmp2, tmp3, slow_case);
705 b(done);
706
707 bind(slow_case);
708
709 // Call the runtime routine for slow case
710 call_VM_preemptable(noreg,
711 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
712 lock_reg);
713
714 bind(done);
715 }
716
717
718 // Unlocks an object. Used in monitorexit bytecode and
719 // remove_activation. Throws an IllegalMonitorException if object is
720 // not locked by current thread.
721 //
722 // Args:
723 // c_rarg1: BasicObjectLock for lock
724 //
725 // Kills:
726 // r0
727 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
728 // rscratch1, rscratch2 (scratch regs)
729 void InterpreterMacroAssembler::unlock_object(Register lock_reg)
730 {
731 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
732
733 const Register swap_reg = r0;
734 const Register header_reg = c_rarg2; // Will contain the old oopMark
735 const Register obj_reg = c_rarg3; // Will contain the oop
736 const Register tmp_reg = c_rarg4; // Temporary used by lightweight_unlock
737
738 save_bcp(); // Save in case of exception
739
740 // Load oop into obj_reg(%c_rarg3)
741 ldr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset()));
742
743 // Free entry
744 str(zr, Address(lock_reg, BasicObjectLock::obj_offset()));
745
746 Label slow_case, done;
747 lightweight_unlock(obj_reg, header_reg, swap_reg, tmp_reg, slow_case);
748 b(done);
749
750 bind(slow_case);
751 // Call the runtime routine for slow case.
752 str(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); // restore obj
753 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
754 bind(done);
755 restore_bcp();
756 }
757
758 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
759 Label& zero_continue) {
760 assert(ProfileInterpreter, "must be profiling interpreter");
761 ldr(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
762 cbz(mdp, zero_continue);
763 }
764
765 // Set the method data pointer for the current bcp.
766 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
767 assert(ProfileInterpreter, "must be profiling interpreter");
768 Label set_mdp;
769 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
770
771 // Test MDO to avoid the call if it is null.
772 ldr(r0, Address(rmethod, in_bytes(Method::method_data_offset())));
773 cbz(r0, set_mdp);
774 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rmethod, rbcp);
775 // r0: mdi
776 // mdo is guaranteed to be non-zero here, we checked for it before the call.
777 ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
778 lea(r1, Address(r1, in_bytes(MethodData::data_offset())));
779 add(r0, r1, r0);
780 str(r0, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
781 bind(set_mdp);
782 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
783 }
784
785 void InterpreterMacroAssembler::verify_method_data_pointer() {
786 assert(ProfileInterpreter, "must be profiling interpreter");
787 #ifdef ASSERT
788 Label verify_continue;
789 stp(r0, r1, Address(pre(sp, -2 * wordSize)));
790 stp(r2, r3, Address(pre(sp, -2 * wordSize)));
791 test_method_data_pointer(r3, verify_continue); // If mdp is zero, continue
792 get_method(r1);
793
794 // If the mdp is valid, it will point to a DataLayout header which is
795 // consistent with the bcp. The converse is highly probable also.
796 ldrsh(r2, Address(r3, in_bytes(DataLayout::bci_offset())));
797 ldr(rscratch1, Address(r1, Method::const_offset()));
798 add(r2, r2, rscratch1, Assembler::LSL);
799 lea(r2, Address(r2, ConstMethod::codes_offset()));
800 cmp(r2, rbcp);
801 br(Assembler::EQ, verify_continue);
802 // r1: method
803 // rbcp: bcp // rbcp == 22
804 // r3: mdp
805 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
806 r1, rbcp, r3);
807 bind(verify_continue);
808 ldp(r2, r3, Address(post(sp, 2 * wordSize)));
809 ldp(r0, r1, Address(post(sp, 2 * wordSize)));
810 #endif // ASSERT
811 }
812
813
814 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
815 int constant,
816 Register value) {
817 assert(ProfileInterpreter, "must be profiling interpreter");
818 Address data(mdp_in, constant);
819 str(value, data);
820 }
821
822
823 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
824 int constant) {
825 increment_mdp_data_at(mdp_in, noreg, constant);
826 }
827
828 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
829 Register index,
830 int constant) {
831 assert(ProfileInterpreter, "must be profiling interpreter");
832
833 assert_different_registers(rscratch2, rscratch1, mdp_in, index);
834
835 Address addr1(mdp_in, constant);
836 Address addr2(rscratch2, index, Address::lsl(0));
837 Address &addr = addr1;
838 if (index != noreg) {
839 lea(rscratch2, addr1);
840 addr = addr2;
841 }
842
843 increment(addr, DataLayout::counter_increment);
844 }
845
846 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
847 int flag_byte_constant) {
848 assert(ProfileInterpreter, "must be profiling interpreter");
849 int flags_offset = in_bytes(DataLayout::flags_offset());
850 // Set the flag
851 ldrb(rscratch1, Address(mdp_in, flags_offset));
852 orr(rscratch1, rscratch1, flag_byte_constant);
853 strb(rscratch1, Address(mdp_in, flags_offset));
854 }
855
856
857 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
858 int offset,
859 Register value,
860 Register test_value_out,
861 Label& not_equal_continue) {
862 assert(ProfileInterpreter, "must be profiling interpreter");
863 if (test_value_out == noreg) {
864 ldr(rscratch1, Address(mdp_in, offset));
865 cmp(value, rscratch1);
866 } else {
867 // Put the test value into a register, so caller can use it:
868 ldr(test_value_out, Address(mdp_in, offset));
869 cmp(value, test_value_out);
870 }
871 br(Assembler::NE, not_equal_continue);
872 }
873
874
875 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
876 int offset_of_disp) {
877 assert(ProfileInterpreter, "must be profiling interpreter");
878 ldr(rscratch1, Address(mdp_in, offset_of_disp));
879 add(mdp_in, mdp_in, rscratch1, LSL);
880 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
881 }
882
883
884 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
885 Register reg,
886 int offset_of_disp) {
887 assert(ProfileInterpreter, "must be profiling interpreter");
888 lea(rscratch1, Address(mdp_in, offset_of_disp));
889 ldr(rscratch1, Address(rscratch1, reg, Address::lsl(0)));
890 add(mdp_in, mdp_in, rscratch1, LSL);
891 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
892 }
893
894
895 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
896 int constant) {
897 assert(ProfileInterpreter, "must be profiling interpreter");
898 add(mdp_in, mdp_in, (unsigned)constant);
899 str(mdp_in, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
900 }
901
902
903 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
904 assert(ProfileInterpreter, "must be profiling interpreter");
905 // save/restore across call_VM
906 stp(zr, return_bci, Address(pre(sp, -2 * wordSize)));
907 call_VM(noreg,
908 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
909 return_bci);
910 ldp(zr, return_bci, Address(post(sp, 2 * wordSize)));
911 }
912
913
914 void InterpreterMacroAssembler::profile_taken_branch(Register mdp) {
915 if (ProfileInterpreter) {
916 Label profile_continue;
917
918 // If no method data exists, go to profile_continue.
919 test_method_data_pointer(mdp, profile_continue);
920
921 // We are taking a branch. Increment the taken count.
922 increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
923
924 // The method data pointer needs to be updated to reflect the new target.
925 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
926 bind(profile_continue);
927 }
928 }
929
930
931 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
932 if (ProfileInterpreter) {
933 Label profile_continue;
934
935 // If no method data exists, go to profile_continue.
936 test_method_data_pointer(mdp, profile_continue);
937
938 // We are not taking a branch. Increment the not taken count.
939 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
940
941 // The method data pointer needs to be updated to correspond to
942 // the next bytecode
943 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
944 bind(profile_continue);
945 }
946 }
947
948
949 void InterpreterMacroAssembler::profile_call(Register mdp) {
950 if (ProfileInterpreter) {
951 Label profile_continue;
952
953 // If no method data exists, go to profile_continue.
954 test_method_data_pointer(mdp, profile_continue);
955
956 // We are making a call. Increment the count.
957 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
958
959 // The method data pointer needs to be updated to reflect the new target.
960 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
961 bind(profile_continue);
962 }
963 }
964
965 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
966 if (ProfileInterpreter) {
967 Label profile_continue;
968
969 // If no method data exists, go to profile_continue.
970 test_method_data_pointer(mdp, profile_continue);
971
972 // We are making a call. Increment the count.
973 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
974
975 // The method data pointer needs to be updated to reflect the new target.
976 update_mdp_by_constant(mdp,
977 in_bytes(VirtualCallData::
978 virtual_call_data_size()));
979 bind(profile_continue);
980 }
981 }
982
983
984 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
985 Register mdp,
986 Register reg2,
987 bool receiver_can_be_null) {
988 if (ProfileInterpreter) {
989 Label profile_continue;
990
991 // If no method data exists, go to profile_continue.
992 test_method_data_pointer(mdp, profile_continue);
993
994 Label skip_receiver_profile;
995 if (receiver_can_be_null) {
996 Label not_null;
997 // We are making a call. Increment the count for null receiver.
998 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
999 b(skip_receiver_profile);
1000 bind(not_null);
1001 }
1002
1003 // Record the receiver type.
1004 record_klass_in_profile(receiver, mdp, reg2);
1005 bind(skip_receiver_profile);
1006
1007 // The method data pointer needs to be updated to reflect the new target.
1008 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
1009 bind(profile_continue);
1010 }
1011 }
1012
1013 // This routine creates a state machine for updating the multi-row
1014 // type profile at a virtual call site (or other type-sensitive bytecode).
1015 // The machine visits each row (of receiver/count) until the receiver type
1016 // is found, or until it runs out of rows. At the same time, it remembers
1017 // the location of the first empty row. (An empty row records null for its
1018 // receiver, and can be allocated for a newly-observed receiver type.)
1019 // Because there are two degrees of freedom in the state, a simple linear
1020 // search will not work; it must be a decision tree. Hence this helper
1021 // function is recursive, to generate the required tree structured code.
1022 // It's the interpreter, so we are trading off code space for speed.
1023 // See below for example code.
1024 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1025 Register receiver, Register mdp,
1026 Register reg2, int start_row,
1027 Label& done) {
1028 if (TypeProfileWidth == 0) {
1029 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1030 } else {
1031 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
1032 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
1033 }
1034 }
1035
1036 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
1037 Register reg2, int start_row, Label& done, int total_rows,
1038 OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
1039 int last_row = total_rows - 1;
1040 assert(start_row <= last_row, "must be work left to do");
1041 // Test this row for both the item and for null.
1042 // Take any of three different outcomes:
1043 // 1. found item => increment count and goto done
1044 // 2. found null => keep looking for case 1, maybe allocate this cell
1045 // 3. found something else => keep looking for cases 1 and 2
1046 // Case 3 is handled by a recursive call.
1047 for (int row = start_row; row <= last_row; row++) {
1048 Label next_test;
1049 bool test_for_null_also = (row == start_row);
1050
1051 // See if the item is item[n].
1052 int item_offset = in_bytes(item_offset_fn(row));
1053 test_mdp_data_at(mdp, item_offset, item,
1054 (test_for_null_also ? reg2 : noreg),
1055 next_test);
1056 // (Reg2 now contains the item from the CallData.)
1057
1058 // The item is item[n]. Increment count[n].
1059 int count_offset = in_bytes(item_count_offset_fn(row));
1060 increment_mdp_data_at(mdp, count_offset);
1061 b(done);
1062 bind(next_test);
1063
1064 if (test_for_null_also) {
1065 Label found_null;
1066 // Failed the equality check on item[n]... Test for null.
1067 if (start_row == last_row) {
1068 // The only thing left to do is handle the null case.
1069 cbz(reg2, found_null);
1070 // Item did not match any saved item and there is no empty row for it.
1071 // Increment total counter to indicate polymorphic case.
1072 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1073 b(done);
1074 bind(found_null);
1075 break;
1076 }
1077 // Since null is rare, make it be the branch-taken case.
1078 cbz(reg2, found_null);
1079
1080 // Put all the "Case 3" tests here.
1081 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
1082 item_offset_fn, item_count_offset_fn);
1083
1084 // Found a null. Keep searching for a matching item,
1085 // but remember that this is an empty (unused) slot.
1086 bind(found_null);
1087 }
1088 }
1089
1090 // In the fall-through case, we found no matching item, but we
1091 // observed the item[start_row] is null.
1092
1093 // Fill in the item field and increment the count.
1094 int item_offset = in_bytes(item_offset_fn(start_row));
1095 set_mdp_data_at(mdp, item_offset, item);
1096 int count_offset = in_bytes(item_count_offset_fn(start_row));
1097 mov(reg2, DataLayout::counter_increment);
1098 set_mdp_data_at(mdp, count_offset, reg2);
1099 if (start_row > 0) {
1100 b(done);
1101 }
1102 }
1103
1104 // Example state machine code for three profile rows:
1105 // // main copy of decision tree, rooted at row[1]
1106 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1107 // if (row[0].rec != nullptr) {
1108 // // inner copy of decision tree, rooted at row[1]
1109 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1110 // if (row[1].rec != nullptr) {
1111 // // degenerate decision tree, rooted at row[2]
1112 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1113 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
1114 // row[2].init(rec); goto done;
1115 // } else {
1116 // // remember row[1] is empty
1117 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1118 // row[1].init(rec); goto done;
1119 // }
1120 // } else {
1121 // // remember row[0] is empty
1122 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1123 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1124 // row[0].init(rec); goto done;
1125 // }
1126 // done:
1127
1128 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1129 Register mdp, Register reg2) {
1130 assert(ProfileInterpreter, "must be profiling");
1131 Label done;
1132
1133 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1134
1135 bind (done);
1136 }
1137
1138 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1139 Register mdp) {
1140 if (ProfileInterpreter) {
1141 Label profile_continue;
1142 uint row;
1143
1144 // If no method data exists, go to profile_continue.
1145 test_method_data_pointer(mdp, profile_continue);
1146
1147 // Update the total ret count.
1148 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1149
1150 for (row = 0; row < RetData::row_limit(); row++) {
1151 Label next_test;
1152
1153 // See if return_bci is equal to bci[n]:
1154 test_mdp_data_at(mdp,
1155 in_bytes(RetData::bci_offset(row)),
1156 return_bci, noreg,
1157 next_test);
1158
1159 // return_bci is equal to bci[n]. Increment the count.
1160 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1161
1162 // The method data pointer needs to be updated to reflect the new target.
1163 update_mdp_by_offset(mdp,
1164 in_bytes(RetData::bci_displacement_offset(row)));
1165 b(profile_continue);
1166 bind(next_test);
1167 }
1168
1169 update_mdp_for_ret(return_bci);
1170
1171 bind(profile_continue);
1172 }
1173 }
1174
1175 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1176 if (ProfileInterpreter) {
1177 Label profile_continue;
1178
1179 // If no method data exists, go to profile_continue.
1180 test_method_data_pointer(mdp, profile_continue);
1181
1182 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1183
1184 // The method data pointer needs to be updated.
1185 int mdp_delta = in_bytes(BitData::bit_data_size());
1186 if (TypeProfileCasts) {
1187 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1188 }
1189 update_mdp_by_constant(mdp, mdp_delta);
1190
1191 bind(profile_continue);
1192 }
1193 }
1194
1195 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1196 if (ProfileInterpreter) {
1197 Label profile_continue;
1198
1199 // If no method data exists, go to profile_continue.
1200 test_method_data_pointer(mdp, profile_continue);
1201
1202 // The method data pointer needs to be updated.
1203 int mdp_delta = in_bytes(BitData::bit_data_size());
1204 if (TypeProfileCasts) {
1205 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1206
1207 // Record the object type.
1208 record_klass_in_profile(klass, mdp, reg2);
1209 }
1210 update_mdp_by_constant(mdp, mdp_delta);
1211
1212 bind(profile_continue);
1213 }
1214 }
1215
1216 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1217 if (ProfileInterpreter) {
1218 Label profile_continue;
1219
1220 // If no method data exists, go to profile_continue.
1221 test_method_data_pointer(mdp, profile_continue);
1222
1223 // Update the default case count
1224 increment_mdp_data_at(mdp,
1225 in_bytes(MultiBranchData::default_count_offset()));
1226
1227 // The method data pointer needs to be updated.
1228 update_mdp_by_offset(mdp,
1229 in_bytes(MultiBranchData::
1230 default_displacement_offset()));
1231
1232 bind(profile_continue);
1233 }
1234 }
1235
1236 void InterpreterMacroAssembler::profile_switch_case(Register index,
1237 Register mdp,
1238 Register reg2) {
1239 if (ProfileInterpreter) {
1240 Label profile_continue;
1241
1242 // If no method data exists, go to profile_continue.
1243 test_method_data_pointer(mdp, profile_continue);
1244
1245 // Build the base (index * per_case_size_in_bytes()) +
1246 // case_array_offset_in_bytes()
1247 movw(reg2, in_bytes(MultiBranchData::per_case_size()));
1248 movw(rscratch1, in_bytes(MultiBranchData::case_array_offset()));
1249 Assembler::maddw(index, index, reg2, rscratch1);
1250
1251 // Update the case count
1252 increment_mdp_data_at(mdp,
1253 index,
1254 in_bytes(MultiBranchData::relative_count_offset()));
1255
1256 // The method data pointer needs to be updated.
1257 update_mdp_by_offset(mdp,
1258 index,
1259 in_bytes(MultiBranchData::
1260 relative_displacement_offset()));
1261
1262 bind(profile_continue);
1263 }
1264 }
1265
1266 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) {
1267 if (state == atos) {
1268 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line);
1269 }
1270 }
1271
1272 void InterpreterMacroAssembler::notify_method_entry() {
1273 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1274 // track stack depth. If it is possible to enter interp_only_mode we add
1275 // the code to check if the event should be sent.
1276 if (JvmtiExport::can_post_interpreter_events()) {
1277 Label L;
1278 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1279 cbzw(r3, L);
1280 call_VM(noreg, CAST_FROM_FN_PTR(address,
1281 InterpreterRuntime::post_method_entry));
1282 bind(L);
1283 }
1284
1285 if (DTraceMethodProbes) {
1286 get_method(c_rarg1);
1287 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1288 rthread, c_rarg1);
1289 }
1290
1291 // RedefineClasses() tracing support for obsolete method entry
1292 if (log_is_enabled(Trace, redefine, class, obsolete)) {
1293 get_method(c_rarg1);
1294 call_VM_leaf(
1295 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1296 rthread, c_rarg1);
1297 }
1298
1299 }
1300
1301
1302 void InterpreterMacroAssembler::notify_method_exit(
1303 TosState state, NotifyMethodExitMode mode) {
1304 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1305 // track stack depth. If it is possible to enter interp_only_mode we add
1306 // the code to check if the event should be sent.
1307 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1308 Label L;
1309 // Note: frame::interpreter_frame_result has a dependency on how the
1310 // method result is saved across the call to post_method_exit. If this
1311 // is changed then the interpreter_frame_result implementation will
1312 // need to be updated too.
1313
1314 // template interpreter will leave the result on the top of the stack.
1315 push(state);
1316 ldrw(r3, Address(rthread, JavaThread::interp_only_mode_offset()));
1317 cbz(r3, L);
1318 call_VM(noreg,
1319 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1320 bind(L);
1321 pop(state);
1322 }
1323
1324 if (DTraceMethodProbes) {
1325 push(state);
1326 get_method(c_rarg1);
1327 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1328 rthread, c_rarg1);
1329 pop(state);
1330 }
1331 }
1332
1333
1334 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1335 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1336 int increment, Address mask,
1337 Register scratch, Register scratch2,
1338 bool preloaded, Condition cond,
1339 Label* where) {
1340 if (!preloaded) {
1341 ldrw(scratch, counter_addr);
1342 }
1343 add(scratch, scratch, increment);
1344 strw(scratch, counter_addr);
1345 ldrw(scratch2, mask);
1346 ands(scratch, scratch, scratch2);
1347 br(cond, *where);
1348 }
1349
1350 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
1351 int number_of_arguments) {
1352 // interpreter specific
1353 //
1354 // Note: No need to save/restore rbcp & rlocals pointer since these
1355 // are callee saved registers and no blocking/ GC can happen
1356 // in leaf calls.
1357 #ifdef ASSERT
1358 {
1359 Label L;
1360 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1361 cbz(rscratch1, L);
1362 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
1363 " last_sp != nullptr");
1364 bind(L);
1365 }
1366 #endif /* ASSERT */
1367 // super call
1368 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
1369 }
1370
1371 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
1372 Register java_thread,
1373 Register last_java_sp,
1374 address entry_point,
1375 int number_of_arguments,
1376 bool check_exceptions) {
1377 // interpreter specific
1378 //
1379 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
1380 // really make a difference for these runtime calls, since they are
1381 // slow anyway. Btw., bcp must be saved/restored since it may change
1382 // due to GC.
1383 // assert(java_thread == noreg , "not expecting a precomputed java thread");
1384 save_bcp();
1385 #ifdef ASSERT
1386 {
1387 Label L;
1388 ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1389 cbz(rscratch1, L);
1390 stop("InterpreterMacroAssembler::call_VM_base:"
1391 " last_sp != nullptr");
1392 bind(L);
1393 }
1394 #endif /* ASSERT */
1395 // super call
1396 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
1397 entry_point, number_of_arguments,
1398 check_exceptions);
1399 // interpreter specific
1400 restore_bcp();
1401 restore_locals();
1402 }
1403
1404 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result,
1405 address entry_point,
1406 Register arg_1) {
1407 assert(arg_1 == c_rarg1, "");
1408 Label resume_pc, not_preempted;
1409
1410 #ifdef ASSERT
1411 {
1412 Label L;
1413 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1414 cbz(rscratch1, L);
1415 stop("Should not have alternate return address set");
1416 bind(L);
1417 }
1418 #endif /* ASSERT */
1419
1420 // Force freeze slow path.
1421 push_cont_fastpath();
1422
1423 // Make VM call. In case of preemption set last_pc to the one we want to resume to.
1424 adr(rscratch1, resume_pc);
1425 str(rscratch1, Address(rthread, JavaThread::last_Java_pc_offset()));
1426 call_VM_base(oop_result, noreg, noreg, entry_point, 1, false /*check_exceptions*/);
1427
1428 pop_cont_fastpath();
1429
1430 // Check if preempted.
1431 ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1432 cbz(rscratch1, not_preempted);
1433 str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1434 br(rscratch1);
1435
1436 // In case of preemption, this is where we will resume once we finally acquire the monitor.
1437 bind(resume_pc);
1438 restore_after_resume(false /* is_native */);
1439
1440 bind(not_preempted);
1441 }
1442
1443 void InterpreterMacroAssembler::restore_after_resume(bool is_native) {
1444 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter()));
1445 blr(rscratch1);
1446 if (is_native) {
1447 // On resume we need to set up stack as expected
1448 push(dtos);
1449 push(ltos);
1450 }
1451 }
1452
1453 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
1454 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index());
1455 Label update, next, none;
1456
1457 verify_oop(obj);
1458
1459 cbnz(obj, update);
1460 orptr(mdo_addr, TypeEntries::null_seen);
1461 b(next);
1462
1463 bind(update);
1464 load_klass(obj, obj);
1465
1466 ldr(rscratch1, mdo_addr);
1467 eor(obj, obj, rscratch1);
1468 tst(obj, TypeEntries::type_klass_mask);
1469 br(Assembler::EQ, next); // klass seen before, nothing to
1470 // do. The unknown bit may have been
1471 // set already but no need to check.
1472
1473 tbnz(obj, exact_log2(TypeEntries::type_unknown), next);
1474 // already unknown. Nothing to do anymore.
1475
1476 cbz(rscratch1, none);
1477 cmp(rscratch1, (u1)TypeEntries::null_seen);
1478 br(Assembler::EQ, none);
1479 // There is a chance that the checks above
1480 // fail if another thread has just set the
1481 // profiling to this obj's klass
1482 eor(obj, obj, rscratch1); // get back original value before XOR
1483 ldr(rscratch1, mdo_addr);
1484 eor(obj, obj, rscratch1);
1485 tst(obj, TypeEntries::type_klass_mask);
1486 br(Assembler::EQ, next);
1487
1488 // different than before. Cannot keep accurate profile.
1489 orptr(mdo_addr, TypeEntries::type_unknown);
1490 b(next);
1491
1492 bind(none);
1493 // first time here. Set profile type.
1494 str(obj, mdo_addr);
1495 #ifdef ASSERT
1496 andr(obj, obj, TypeEntries::type_mask);
1497 verify_klass_ptr(obj);
1498 #endif
1499
1500 bind(next);
1501 }
1502
1503 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
1504 if (!ProfileInterpreter) {
1505 return;
1506 }
1507
1508 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1509 Label profile_continue;
1510
1511 test_method_data_pointer(mdp, profile_continue);
1512
1513 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1514
1515 ldrb(rscratch1, Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start));
1516 cmp(rscratch1, u1(is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag));
1517 br(Assembler::NE, profile_continue);
1518
1519 if (MethodData::profile_arguments()) {
1520 Label done;
1521 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1522
1523 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1524 if (i > 0 || MethodData::profile_return()) {
1525 // If return value type is profiled we may have no argument to profile
1526 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1527 sub(tmp, tmp, i*TypeStackSlotEntries::per_arg_count());
1528 cmp(tmp, (u1)TypeStackSlotEntries::per_arg_count());
1529 add(rscratch1, mdp, off_to_args);
1530 br(Assembler::LT, done);
1531 }
1532 ldr(tmp, Address(callee, Method::const_offset()));
1533 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
1534 // stack offset o (zero based) from the start of the argument
1535 // list, for n arguments translates into offset n - o - 1 from
1536 // the end of the argument list
1537 ldr(rscratch1, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))));
1538 sub(tmp, tmp, rscratch1);
1539 sub(tmp, tmp, 1);
1540 Address arg_addr = argument_address(tmp);
1541 ldr(tmp, arg_addr);
1542
1543 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i)));
1544 profile_obj_type(tmp, mdo_arg_addr);
1545
1546 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1547 off_to_args += to_add;
1548 }
1549
1550 if (MethodData::profile_return()) {
1551 ldr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())));
1552 sub(tmp, tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
1553 }
1554
1555 add(rscratch1, mdp, off_to_args);
1556 bind(done);
1557 mov(mdp, rscratch1);
1558
1559 if (MethodData::profile_return()) {
1560 // We're right after the type profile for the last
1561 // argument. tmp is the number of cells left in the
1562 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1563 // if there's a return to profile.
1564 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
1565 add(mdp, mdp, tmp, LSL, exact_log2(DataLayout::cell_size));
1566 }
1567 str(mdp, Address(rfp, frame::interpreter_frame_mdp_offset * wordSize));
1568 } else {
1569 assert(MethodData::profile_return(), "either profile call args or call ret");
1570 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
1571 }
1572
1573 // mdp points right after the end of the
1574 // CallTypeData/VirtualCallTypeData, right after the cells for the
1575 // return value type if there's one
1576
1577 bind(profile_continue);
1578 }
1579 }
1580
1581 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
1582 assert_different_registers(mdp, ret, tmp, rbcp);
1583 if (ProfileInterpreter && MethodData::profile_return()) {
1584 Label profile_continue, done;
1585
1586 test_method_data_pointer(mdp, profile_continue);
1587
1588 if (MethodData::profile_return_jsr292_only()) {
1589 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2");
1590
1591 // If we don't profile all invoke bytecodes we must make sure
1592 // it's a bytecode we indeed profile. We can't go back to the
1593 // beginning of the ProfileData we intend to update to check its
1594 // type because we're right after it and we don't known its
1595 // length
1596 Label do_profile;
1597 ldrb(rscratch1, Address(rbcp, 0));
1598 cmp(rscratch1, (u1)Bytecodes::_invokedynamic);
1599 br(Assembler::EQ, do_profile);
1600 cmp(rscratch1, (u1)Bytecodes::_invokehandle);
1601 br(Assembler::EQ, do_profile);
1602 get_method(tmp);
1603 ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset()));
1604 subs(zr, rscratch1, static_cast<int>(vmIntrinsics::_compiledLambdaForm));
1605 br(Assembler::NE, profile_continue);
1606
1607 bind(do_profile);
1608 }
1609
1610 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
1611 mov(tmp, ret);
1612 profile_obj_type(tmp, mdo_ret_addr);
1613
1614 bind(profile_continue);
1615 }
1616 }
1617
1618 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
1619 assert_different_registers(rscratch1, rscratch2, mdp, tmp1, tmp2);
1620 if (ProfileInterpreter && MethodData::profile_parameters()) {
1621 Label profile_continue, done;
1622
1623 test_method_data_pointer(mdp, profile_continue);
1624
1625 // Load the offset of the area within the MDO used for
1626 // parameters. If it's negative we're not profiling any parameters
1627 ldrw(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
1628 tbnz(tmp1, 31, profile_continue); // i.e. sign bit set
1629
1630 // Compute a pointer to the area for parameters from the offset
1631 // and move the pointer to the slot for the last
1632 // parameters. Collect profiling from last parameter down.
1633 // mdo start + parameters offset + array length - 1
1634 add(mdp, mdp, tmp1);
1635 ldr(tmp1, Address(mdp, ArrayData::array_len_offset()));
1636 sub(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1637
1638 Label loop;
1639 bind(loop);
1640
1641 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
1642 int type_base = in_bytes(ParametersTypeData::type_offset(0));
1643 int per_arg_scale = exact_log2(DataLayout::cell_size);
1644 add(rscratch1, mdp, off_base);
1645 add(rscratch2, mdp, type_base);
1646
1647 Address arg_off(rscratch1, tmp1, Address::lsl(per_arg_scale));
1648 Address arg_type(rscratch2, tmp1, Address::lsl(per_arg_scale));
1649
1650 // load offset on the stack from the slot for this parameter
1651 ldr(tmp2, arg_off);
1652 neg(tmp2, tmp2);
1653 // read the parameter from the local area
1654 ldr(tmp2, Address(rlocals, tmp2, Address::lsl(Interpreter::logStackElementSize)));
1655
1656 // profile the parameter
1657 profile_obj_type(tmp2, arg_type);
1658
1659 // go to next parameter
1660 subs(tmp1, tmp1, TypeStackSlotEntries::per_arg_count());
1661 br(Assembler::GE, loop);
1662
1663 bind(profile_continue);
1664 }
1665 }
1666
1667 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) {
1668 // Get index out of bytecode pointer, get_cache_entry_pointer_at_bcp
1669 get_cache_index_at_bcp(index, 1, sizeof(u4));
1670 // Get address of invokedynamic array
1671 ldr(cache, Address(rcpool, in_bytes(ConstantPoolCache::invokedynamic_entries_offset())));
1672 // Scale the index to be the entry index * sizeof(ResolvedIndyEntry)
1673 lsl(index, index, log2i_exact(sizeof(ResolvedIndyEntry)));
1674 add(cache, cache, Array<ResolvedIndyEntry>::base_offset_in_bytes());
1675 lea(cache, Address(cache, index));
1676 }
1677
1678 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) {
1679 // Get index out of bytecode pointer
1680 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1681 // Take shortcut if the size is a power of 2
1682 if (is_power_of_2(sizeof(ResolvedFieldEntry))) {
1683 lsl(index, index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2
1684 } else {
1685 mov(cache, sizeof(ResolvedFieldEntry));
1686 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry)
1687 }
1688 // Get address of field entries array
1689 ldr(cache, Address(rcpool, ConstantPoolCache::field_entries_offset()));
1690 add(cache, cache, Array<ResolvedFieldEntry>::base_offset_in_bytes());
1691 lea(cache, Address(cache, index));
1692 // Prevents stale data from being read after the bytecode is patched to the fast bytecode
1693 membar(MacroAssembler::LoadLoad);
1694 }
1695
1696 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) {
1697 // Get index out of bytecode pointer
1698 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2));
1699 mov(cache, sizeof(ResolvedMethodEntry));
1700 mul(index, index, cache); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry)
1701
1702 // Get address of field entries array
1703 ldr(cache, Address(rcpool, ConstantPoolCache::method_entries_offset()));
1704 add(cache, cache, Array<ResolvedMethodEntry>::base_offset_in_bytes());
1705 lea(cache, Address(cache, index));
1706 }