1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * 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 "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "compiler/compiler_globals.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "interpreter/templateInterpreterGenerator.hpp"
36 #include "interpreter/templateTable.hpp"
37 #include "interpreter/bytecodeTracer.hpp"
38 #include "memory/resourceArea.hpp"
39 #include "oops/arrayOop.hpp"
40 #include "oops/method.hpp"
41 #include "oops/methodCounters.hpp"
42 #include "oops/methodData.hpp"
43 #include "oops/oop.inline.hpp"
44 #include "oops/resolvedIndyEntry.hpp"
45 #include "oops/resolvedMethodEntry.hpp"
46 #include "prims/jvmtiExport.hpp"
47 #include "prims/jvmtiThreadState.hpp"
48 #include "runtime/arguments.hpp"
49 #include "runtime/deoptimization.hpp"
50 #include "runtime/frame.inline.hpp"
51 #include "runtime/globals.hpp"
52 #include "runtime/jniHandles.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/synchronizer.hpp"
56 #include "runtime/timer.hpp"
57 #include "runtime/vframeArray.hpp"
58 #include "utilities/checkedCast.hpp"
59 #include "utilities/debug.hpp"
60 #include "utilities/powerOfTwo.hpp"
61 #include <sys/types.h>
62
63 // Size of interpreter code. Increase if too small. Interpreter will
64 // fail with a guarantee ("not enough space for interpreter generation");
65 // if too small.
66 // Run with +PrintInterpreter to get the VM to print out the size.
67 // Max size with JVMTI
68 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
69
70 #define __ _masm->
71
72 //-----------------------------------------------------------------------------
73
74 extern "C" void entry(CodeBuffer*);
75
76 //-----------------------------------------------------------------------------
77
78 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
79 address entry = __ pc();
80
81 __ andr(esp, esp, -16);
82 __ mov(c_rarg3, esp);
83 // rmethod
84 // rlocals
85 // c_rarg3: first stack arg - wordSize
86
87 // adjust sp
88 __ sub(sp, c_rarg3, 18 * wordSize);
89 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
90 __ call_VM(noreg,
91 CAST_FROM_FN_PTR(address,
92 InterpreterRuntime::slow_signature_handler),
93 rmethod, rlocals, c_rarg3);
94
95 // r0: result handler
96
97 // Stack layout:
98 // rsp: return address <- sp
99 // 1 garbage
100 // 8 integer args (if static first is unused)
101 // 1 float/double identifiers
102 // 8 double args
103 // stack args <- esp
104 // garbage
105 // expression stack bottom
106 // bcp (null)
107 // ...
108
109 // Restore LR
110 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
111
112 // Do FP first so we can use c_rarg3 as temp
113 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
114
115 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
116 const FloatRegister r = as_FloatRegister(i);
117
118 Label d, done;
119
120 __ tbnz(c_rarg3, i, d);
121 __ ldrs(r, Address(sp, (10 + i) * wordSize));
122 __ b(done);
123 __ bind(d);
124 __ ldrd(r, Address(sp, (10 + i) * wordSize));
125 __ bind(done);
126 }
127
128 // c_rarg0 contains the result from the call of
129 // InterpreterRuntime::slow_signature_handler so we don't touch it
130 // here. It will be loaded with the JNIEnv* later.
131 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
132 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
133 Register rm = as_Register(i), rn = as_Register(i+1);
134 __ ldp(rm, rn, Address(sp, i * wordSize));
135 }
136
137 __ add(sp, sp, 18 * wordSize);
138 __ ret(lr);
139
140 return entry;
141 }
142
143
144 //
145 // Various method entries
146 //
147
148 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
149 // rmethod: Method*
150 // r19_sender_sp: sender sp
151 // esp: args
152
153 // These don't need a safepoint check because they aren't virtually
154 // callable. We won't enter these intrinsics from compiled code.
155 // If in the future we added an intrinsic which was virtually callable
156 // we'd have to worry about how to safepoint so that this code is used.
157
158 // mathematical functions inlined by compiler
159 // (interpreter must provide identical implementation
160 // in order to avoid monotonicity bugs when switching
161 // from interpreter to compiler in the middle of some
162 // computation)
163 //
164 // stack:
165 // [ arg ] <-- esp
166 // [ arg ]
167 // retaddr in lr
168
169 address entry_point = nullptr;
170 Register continuation = lr;
171 switch (kind) {
172 case Interpreter::java_lang_math_abs:
173 entry_point = __ pc();
174 __ ldrd(v0, Address(esp));
175 __ fabsd(v0, v0);
176 __ mov(sp, r19_sender_sp); // Restore caller's SP
177 break;
178 case Interpreter::java_lang_math_sqrt:
179 entry_point = __ pc();
180 __ ldrd(v0, Address(esp));
181 __ fsqrtd(v0, v0);
182 __ mov(sp, r19_sender_sp);
183 break;
184 case Interpreter::java_lang_math_sin :
185 case Interpreter::java_lang_math_cos :
186 case Interpreter::java_lang_math_tan :
187 case Interpreter::java_lang_math_log :
188 case Interpreter::java_lang_math_log10 :
189 case Interpreter::java_lang_math_exp :
190 entry_point = __ pc();
191 __ ldrd(v0, Address(esp));
192 __ mov(sp, r19_sender_sp);
193 __ mov(r23, lr);
194 continuation = r23; // The first free callee-saved register
195 generate_transcendental_entry(kind, 1);
196 break;
197 case Interpreter::java_lang_math_pow :
198 entry_point = __ pc();
199 __ mov(r23, lr);
200 continuation = r23;
201 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
202 __ ldrd(v1, Address(esp));
203 __ mov(sp, r19_sender_sp);
204 generate_transcendental_entry(kind, 2);
205 break;
206 case Interpreter::java_lang_math_fmaD :
207 if (UseFMA) {
208 entry_point = __ pc();
209 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
210 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
211 __ ldrd(v2, Address(esp));
212 __ fmaddd(v0, v0, v1, v2);
213 __ mov(sp, r19_sender_sp); // Restore caller's SP
214 }
215 break;
216 case Interpreter::java_lang_math_fmaF :
217 if (UseFMA) {
218 entry_point = __ pc();
219 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
220 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
221 __ ldrs(v2, Address(esp));
222 __ fmadds(v0, v0, v1, v2);
223 __ mov(sp, r19_sender_sp); // Restore caller's SP
224 }
225 break;
226 default:
227 ;
228 }
229 if (entry_point) {
230 __ br(continuation);
231 }
232
233 return entry_point;
234 }
235
236 // double trigonometrics and transcendentals
237 // static jdouble dsin(jdouble x);
238 // static jdouble dcos(jdouble x);
239 // static jdouble dtan(jdouble x);
240 // static jdouble dlog(jdouble x);
241 // static jdouble dlog10(jdouble x);
242 // static jdouble dexp(jdouble x);
243 // static jdouble dpow(jdouble x, jdouble y);
244
245 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
246 address fn;
247 switch (kind) {
248 case Interpreter::java_lang_math_sin :
249 if (StubRoutines::dsin() == nullptr) {
250 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
251 } else {
252 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
253 }
254 break;
255 case Interpreter::java_lang_math_cos :
256 if (StubRoutines::dcos() == nullptr) {
257 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
258 } else {
259 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
260 }
261 break;
262 case Interpreter::java_lang_math_tan :
263 if (StubRoutines::dtan() == nullptr) {
264 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
265 } else {
266 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
267 }
268 break;
269 case Interpreter::java_lang_math_log :
270 if (StubRoutines::dlog() == nullptr) {
271 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
272 } else {
273 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
274 }
275 break;
276 case Interpreter::java_lang_math_log10 :
277 if (StubRoutines::dlog10() == nullptr) {
278 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
279 } else {
280 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
281 }
282 break;
283 case Interpreter::java_lang_math_exp :
284 if (StubRoutines::dexp() == nullptr) {
285 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
286 } else {
287 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
288 }
289 break;
290 case Interpreter::java_lang_math_pow :
291 if (StubRoutines::dpow() == nullptr) {
292 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
293 } else {
294 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
295 }
296 break;
297 default:
298 ShouldNotReachHere();
299 fn = nullptr; // unreachable
300 }
301 __ mov(rscratch1, fn);
302 __ blr(rscratch1);
303 }
304
305 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
306 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
307 // r19_sender_sp: sender sp
308 // stack:
309 // [ arg ] <-- esp
310 // [ arg ]
311 // retaddr in lr
312 // result in v0
313
314 address entry_point = __ pc();
315 __ ldrw(c_rarg0, Address(esp));
316 __ flt16_to_flt(v0, c_rarg0, v1);
317 __ mov(sp, r19_sender_sp); // Restore caller's SP
318 __ br(lr);
319 return entry_point;
320 }
321
322 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
323 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
324 // r19_sender_sp: sender sp
325 // stack:
326 // [ arg ] <-- esp
327 // [ arg ]
328 // retaddr in lr
329 // result in c_rarg0
330
331 address entry_point = __ pc();
332 __ ldrs(v0, Address(esp));
333 __ flt_to_flt16(c_rarg0, v0, v1);
334 __ mov(sp, r19_sender_sp); // Restore caller's SP
335 __ br(lr);
336 return entry_point;
337 }
338
339 // Abstract method entry
340 // Attempt to execute abstract method. Throw exception
341 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
342 // rmethod: Method*
343 // r19_sender_sp: sender SP
344
345 address entry_point = __ pc();
346
347 // abstract method entry
348
349 // pop return address, reset last_sp to null
350 __ empty_expression_stack();
351 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
352 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
353
354 // throw exception
355 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
356 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
357 rmethod);
358 // the call_VM checks for exception, so we should never return here.
359 __ should_not_reach_here();
360
361 return entry_point;
362 }
363
364 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
365 address entry = __ pc();
366
367 #ifdef ASSERT
368 {
369 Label L;
370 __ ldr(rscratch1, Address(rfp,
371 frame::interpreter_frame_monitor_block_top_offset *
372 wordSize));
373 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
374 __ mov(rscratch2, sp);
375 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
376 // grows negative)
377 __ br(Assembler::HS, L); // check if frame is complete
378 __ stop ("interpreter frame not set up");
379 __ bind(L);
380 }
381 #endif // ASSERT
382 // Restore bcp under the assumption that the current frame is still
383 // interpreted
384 __ restore_bcp();
385
386 // expression stack must be empty before entering the VM if an
387 // exception happened
388 __ empty_expression_stack();
389 // throw exception
390 __ call_VM(noreg,
391 CAST_FROM_FN_PTR(address,
392 InterpreterRuntime::throw_StackOverflowError));
393 return entry;
394 }
395
396 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
397 address entry = __ pc();
398 // expression stack must be empty before entering the VM if an
399 // exception happened
400 __ empty_expression_stack();
401 // setup parameters
402
403 // ??? convention: expect aberrant index in register r1
404 __ movw(c_rarg2, r1);
405 // ??? convention: expect array in register r3
406 __ mov(c_rarg1, r3);
407 __ call_VM(noreg,
408 CAST_FROM_FN_PTR(address,
409 InterpreterRuntime::
410 throw_ArrayIndexOutOfBoundsException),
411 c_rarg1, c_rarg2);
412 return entry;
413 }
414
415 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
416 address entry = __ pc();
417
418 // object is at TOS
419 __ pop(c_rarg1);
420
421 // expression stack must be empty before entering the VM if an
422 // exception happened
423 __ empty_expression_stack();
424
425 __ call_VM(noreg,
426 CAST_FROM_FN_PTR(address,
427 InterpreterRuntime::
428 throw_ClassCastException),
429 c_rarg1);
430 return entry;
431 }
432
433 address TemplateInterpreterGenerator::generate_exception_handler_common(
434 const char* name, const char* message, bool pass_oop) {
435 assert(!pass_oop || message == nullptr, "either oop or message but not both");
436 address entry = __ pc();
437 if (pass_oop) {
438 // object is at TOS
439 __ pop(c_rarg2);
440 }
441 // expression stack must be empty before entering the VM if an
442 // exception happened
443 __ empty_expression_stack();
444 // setup parameters
445 __ lea(c_rarg1, Address((address)name));
446 if (pass_oop) {
447 __ call_VM(r0, CAST_FROM_FN_PTR(address,
448 InterpreterRuntime::
449 create_klass_exception),
450 c_rarg1, c_rarg2);
451 } else {
452 // kind of lame ExternalAddress can't take null because
453 // external_word_Relocation will assert.
454 if (message != nullptr) {
455 __ lea(c_rarg2, Address((address)message));
456 } else {
457 __ mov(c_rarg2, NULL_WORD);
458 }
459 __ call_VM(r0,
460 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
461 c_rarg1, c_rarg2);
462 }
463 // throw exception
464 __ b(address(Interpreter::throw_exception_entry()));
465 return entry;
466 }
467
468 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
469 address entry = __ pc();
470
471 // Restore stack bottom in case i2c adjusted stack
472 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
473 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
474 // and null it as marker that esp is now tos until next java call
475 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
476 __ restore_bcp();
477 __ restore_locals();
478 __ restore_constant_pool_cache();
479 __ get_method(rmethod);
480
481 if (state == atos) {
482 Register obj = r0;
483 Register mdp = r1;
484 Register tmp = r2;
485 __ profile_return_type(mdp, obj, tmp);
486 }
487
488 const Register cache = r1;
489 const Register index = r2;
490
491 if (index_size == sizeof(u4)) {
492 __ load_resolved_indy_entry(cache, index);
493 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
494 __ add(esp, esp, cache, Assembler::LSL, 3);
495 } else {
496 // Pop N words from the stack
497 assert(index_size == sizeof(u2), "Can only be u2");
498 __ load_method_entry(cache, index);
499 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
500 __ add(esp, esp, cache, Assembler::LSL, 3);
501 }
502
503 // Restore machine SP
504 __ restore_sp_after_call();
505
506 __ check_and_handle_popframe(rthread);
507 __ check_and_handle_earlyret(rthread);
508
509 __ get_dispatch();
510 __ dispatch_next(state, step);
511
512 return entry;
513 }
514
515 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
516 int step,
517 address continuation) {
518 address entry = __ pc();
519 __ restore_bcp();
520 __ restore_locals();
521 __ restore_constant_pool_cache();
522 __ get_method(rmethod);
523 __ get_dispatch();
524
525 __ restore_sp_after_call(); // Restore SP to extended SP
526
527 // Restore expression stack pointer
528 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
529 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
530 // null last_sp until next java call
531 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
532
533 #if INCLUDE_JVMCI
534 // Check if we need to take lock at entry of synchronized method. This can
535 // only occur on method entry so emit it only for vtos with step 0.
536 if (EnableJVMCI && state == vtos && step == 0) {
537 Label L;
538 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
539 __ cbz(rscratch1, L);
540 // Clear flag.
541 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
542 // Take lock.
543 lock_method();
544 __ bind(L);
545 } else {
546 #ifdef ASSERT
547 if (EnableJVMCI) {
548 Label L;
549 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
550 __ cbz(rscratch1, L);
551 __ stop("unexpected pending monitor in deopt entry");
552 __ bind(L);
553 }
554 #endif
555 }
556 #endif
557 // handle exceptions
558 {
559 Label L;
560 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
561 __ cbz(rscratch1, L);
562 __ call_VM(noreg,
563 CAST_FROM_FN_PTR(address,
564 InterpreterRuntime::throw_pending_exception));
565 __ should_not_reach_here();
566 __ bind(L);
567 }
568
569 if (continuation == nullptr) {
570 __ dispatch_next(state, step);
571 } else {
572 __ jump_to_entry(continuation);
573 }
574 return entry;
575 }
576
577 address TemplateInterpreterGenerator::generate_result_handler_for(
578 BasicType type) {
579 address entry = __ pc();
580 switch (type) {
581 case T_BOOLEAN: __ c2bool(r0); break;
582 case T_CHAR : __ uxth(r0, r0); break;
583 case T_BYTE : __ sxtb(r0, r0); break;
584 case T_SHORT : __ sxth(r0, r0); break;
585 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
586 case T_LONG : /* nothing to do */ break;
587 case T_VOID : /* nothing to do */ break;
588 case T_FLOAT : /* nothing to do */ break;
589 case T_DOUBLE : /* nothing to do */ break;
590 case T_OBJECT :
591 // retrieve result from frame
592 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
593 // and verify it
594 __ verify_oop(r0);
595 break;
596 default : ShouldNotReachHere();
597 }
598 __ ret(lr); // return from result handler
599 return entry;
600 }
601
602 address TemplateInterpreterGenerator::generate_safept_entry_for(
603 TosState state,
604 address runtime_entry) {
605 address entry = __ pc();
606 __ push(state);
607 __ push_cont_fastpath(rthread);
608 __ call_VM(noreg, runtime_entry);
609 __ pop_cont_fastpath(rthread);
610 __ membar(Assembler::AnyAny);
611 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
612 return entry;
613 }
614
615 address TemplateInterpreterGenerator::generate_cont_preempt_rerun_interpreter_adapter() {
616 if (!Continuations::enabled()) return nullptr;
617 address start = __ pc();
618
619 // Restore rfp first since we need it to restore rest of registers
620 __ leave();
621
622 // Restore constant pool cache
623 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
624
625 // Restore Java expression stack pointer
626 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
627 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
628 // and NULL it as marker that esp is now tos until next java call
629 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
630
631 // Restore machine SP
632 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
633 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord)));
634
635 // Prepare for adjustment on return to call_VM_leaf_base()
636 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
637 __ stp(rscratch1, rmethod, Address(__ pre(sp, -2 * wordSize)));
638
639 // Restore dispatch
640 uint64_t offset;
641 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
642 __ add(rdispatch, rdispatch, offset);
643
644 __ ret(lr);
645
646 return start;
647 }
648
649
650 // Helpers for commoning out cases in the various type of method entries.
651 //
652
653
654 // increment invocation count & check for overflow
655 //
656 // Note: checking for negative value instead of overflow
657 // so we have a 'sticky' overflow test
658 //
659 // rmethod: method
660 //
661 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
662 Label done;
663 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
664 int increment = InvocationCounter::count_increment;
665 Label no_mdo;
666 if (ProfileInterpreter) {
667 // Are we profiling?
668 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
669 __ cbz(r0, no_mdo);
670 // Increment counter in the MDO
671 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
672 in_bytes(InvocationCounter::counter_offset()));
673 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
674 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
675 __ b(done);
676 }
677 __ bind(no_mdo);
678 // Increment counter in MethodCounters
679 const Address invocation_counter(rscratch2,
680 MethodCounters::invocation_counter_offset() +
681 InvocationCounter::counter_offset());
682 __ get_method_counters(rmethod, rscratch2, done);
683 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
684 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
685 __ bind(done);
686 }
687
688 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
689
690 // Asm interpreter on entry
691 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
692 // Everything as it was on entry
693
694 // InterpreterRuntime::frequency_counter_overflow takes two
695 // arguments, the first (thread) is passed by call_VM, the second
696 // indicates if the counter overflow occurs at a backwards branch
697 // (null bcp). We pass zero for it. The call returns the address
698 // of the verified entry point for the method or null if the
699 // compilation did not complete (either went background or bailed
700 // out).
701 __ mov(c_rarg1, 0);
702 __ call_VM(noreg,
703 CAST_FROM_FN_PTR(address,
704 InterpreterRuntime::frequency_counter_overflow),
705 c_rarg1);
706
707 __ b(do_continue);
708 }
709
710 // See if we've got enough room on the stack for locals plus overhead
711 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
712 // without going through the signal handler, i.e., reserved and yellow zones
713 // will not be made usable. The shadow zone must suffice to handle the
714 // overflow.
715 // The expression stack grows down incrementally, so the normal guard
716 // page mechanism will work for that.
717 //
718 // NOTE: Since the additional locals are also always pushed (wasn't
719 // obvious in generate_method_entry) so the guard should work for them
720 // too.
721 //
722 // Args:
723 // r3: number of additional locals this frame needs (what we must check)
724 // rmethod: Method*
725 //
726 // Kills:
727 // r0
728 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
729
730 // monitor entry size: see picture of stack set
731 // (generate_method_entry) and frame_amd64.hpp
732 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
733
734 // total overhead size: entry_size + (saved rbp through expr stack
735 // bottom). be sure to change this if you add/subtract anything
736 // to/from the overhead area
737 const int overhead_size =
738 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
739
740 const size_t page_size = os::vm_page_size();
741
742 Label after_frame_check;
743
744 // see if the frame is greater than one page in size. If so,
745 // then we need to verify there is enough stack space remaining
746 // for the additional locals.
747 //
748 // Note that we use SUBS rather than CMP here because the immediate
749 // field of this instruction may overflow. SUBS can cope with this
750 // because it is a macro that will expand to some number of MOV
751 // instructions and a register operation.
752 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
753 __ br(Assembler::LS, after_frame_check);
754
755 // compute rsp as if this were going to be the last frame on
756 // the stack before the red zone
757
758 // locals + overhead, in bytes
759 __ mov(r0, overhead_size);
760 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
761
762 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
763 __ ldr(rscratch1, stack_limit);
764
765 #ifdef ASSERT
766 Label limit_okay;
767 // Verify that thread stack limit is non-zero.
768 __ cbnz(rscratch1, limit_okay);
769 __ stop("stack overflow limit is zero");
770 __ bind(limit_okay);
771 #endif
772
773 // Add stack limit to locals.
774 __ add(r0, r0, rscratch1);
775
776 // Check against the current stack bottom.
777 __ cmp(sp, r0);
778 __ br(Assembler::HI, after_frame_check);
779
780 // Remove the incoming args, peeling the machine SP back to where it
781 // was in the caller. This is not strictly necessary, but unless we
782 // do so the stack frame may have a garbage FP; this ensures a
783 // correct call stack that we can always unwind. The ANDR should be
784 // unnecessary because the sender SP in r19 is always aligned, but
785 // it doesn't hurt.
786 __ andr(sp, r19_sender_sp, -16);
787
788 // Note: the restored frame is not necessarily interpreted.
789 // Use the shared runtime version of the StackOverflowError.
790 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
791 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
792
793 // all done with frame size check
794 __ bind(after_frame_check);
795 }
796
797 // Allocate monitor and lock method (asm interpreter)
798 //
799 // Args:
800 // rmethod: Method*
801 // rlocals: locals
802 //
803 // Kills:
804 // r0
805 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
806 // rscratch1, rscratch2 (scratch regs)
807 void TemplateInterpreterGenerator::lock_method() {
808 // synchronize method
809 const Address access_flags(rmethod, Method::access_flags_offset());
810 const Address monitor_block_top(
811 rfp,
812 frame::interpreter_frame_monitor_block_top_offset * wordSize);
813 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
814
815 #ifdef ASSERT
816 {
817 Label L;
818 __ ldrw(r0, access_flags);
819 __ tst(r0, JVM_ACC_SYNCHRONIZED);
820 __ br(Assembler::NE, L);
821 __ stop("method doesn't need synchronization");
822 __ bind(L);
823 }
824 #endif // ASSERT
825
826 // get synchronization object
827 {
828 Label done;
829 __ ldrw(r0, access_flags);
830 __ tst(r0, JVM_ACC_STATIC);
831 // get receiver (assume this is frequent case)
832 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
833 __ br(Assembler::EQ, done);
834 __ load_mirror(r0, rmethod, r5, rscratch2);
835
836 #ifdef ASSERT
837 {
838 Label L;
839 __ cbnz(r0, L);
840 __ stop("synchronization object is null");
841 __ bind(L);
842 }
843 #endif // ASSERT
844
845 __ bind(done);
846 }
847
848 // add space for monitor & lock
849 __ check_extended_sp();
850 __ sub(sp, sp, entry_size); // add space for a monitor entry
851 __ sub(esp, esp, entry_size);
852 __ sub(rscratch1, sp, rfp);
853 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
854 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
855 __ sub(rscratch1, esp, rfp);
856 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
857 __ str(rscratch1, monitor_block_top); // set new monitor block top
858
859 // store object
860 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
861 __ mov(c_rarg1, esp); // object address
862 __ lock_object(c_rarg1);
863 }
864
865 // Generate a fixed interpreter frame. This is identical setup for
866 // interpreted methods and for native methods hence the shared code.
867 //
868 // Args:
869 // lr: return address
870 // rmethod: Method*
871 // rlocals: pointer to locals
872 // rcpool: cp cache
873 // stack_pointer: previous sp
874 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
875 // initialize fixed part of activation frame
876 if (native_call) {
877 __ sub(esp, sp, 14 * wordSize);
878 __ mov(rbcp, zr);
879 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
880 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
881 // add 2 zero-initialized slots for native calls
882 __ stp(zr, zr, Address(sp, 12 * wordSize));
883 } else {
884 __ sub(esp, sp, 12 * wordSize);
885 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
886 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
887 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
888 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
889 }
890
891 if (ProfileInterpreter) {
892 Label method_data_continue;
893 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
894 __ cbz(rscratch1, method_data_continue);
895 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
896 __ bind(method_data_continue);
897 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
898 } else {
899 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
900 }
901
902 __ protect_return_address();
903 __ stp(rfp, lr, Address(sp, 10 * wordSize));
904 __ lea(rfp, Address(sp, 10 * wordSize));
905
906 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
907 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
908 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
909 __ sub(rscratch1, rlocals, rfp);
910 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
911 // Store relativized rlocals, see frame::interpreter_frame_locals().
912 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
913
914 // set sender sp
915 // leave last_sp as null
916 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
917
918 // Get mirror
919 __ load_mirror(r10, rmethod, r5, rscratch2);
920 if (! native_call) {
921 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
922 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
923 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
924 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
925 __ andr(rscratch1, rscratch1, -16);
926 __ sub(rscratch2, rscratch1, rfp);
927 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
928 // Store extended SP and mirror
929 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
930 // Move SP out of the way
931 __ mov(sp, rscratch1);
932 } else {
933 // Make sure there is room for the exception oop pushed in case method throws
934 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
935 __ sub(rscratch1, sp, 2 * wordSize);
936 __ sub(rscratch2, rscratch1, rfp);
937 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
938 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
939 __ mov(sp, rscratch1);
940 }
941 }
942
943 // End of helpers
944
945 // Various method entries
946 //------------------------------------------------------------------------------------------------------------------------
947 //
948 //
949
950 // Method entry for java.lang.ref.Reference.get.
951 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
952 // Code: _aload_0, _getfield, _areturn
953 // parameter size = 1
954 //
955 // The code that gets generated by this routine is split into 2 parts:
956 // 1. The "intrinsified" code for G1 (or any SATB based GC),
957 // 2. The slow path - which is an expansion of the regular method entry.
958 //
959 // Notes:-
960 // * In the G1 code we do not check whether we need to block for
961 // a safepoint. If G1 is enabled then we must execute the specialized
962 // code for Reference.get (except when the Reference object is null)
963 // so that we can log the value in the referent field with an SATB
964 // update buffer.
965 // If the code for the getfield template is modified so that the
966 // G1 pre-barrier code is executed when the current method is
967 // Reference.get() then going through the normal method entry
968 // will be fine.
969 // * The G1 code can, however, check the receiver object (the instance
970 // of java.lang.Reference) and jump to the slow path if null. If the
971 // Reference object is null then we obviously cannot fetch the referent
972 // and so we don't need to call the G1 pre-barrier. Thus we can use the
973 // regular method entry code to generate the NPE.
974 //
975 // This code is based on generate_accessor_entry.
976 //
977 // rmethod: Method*
978 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
979
980 // LR is live. It must be saved around calls.
981
982 address entry = __ pc();
983
984 const int referent_offset = java_lang_ref_Reference::referent_offset();
985
986 Label slow_path;
987 const Register local_0 = c_rarg0;
988 // Check if local 0 != null
989 // If the receiver is null then it is OK to jump to the slow path.
990 __ ldr(local_0, Address(esp, 0));
991 __ cbz(local_0, slow_path);
992
993 // Load the value of the referent field.
994 const Address field_address(local_0, referent_offset);
995 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
996 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
997
998 // areturn
999 __ andr(sp, r19_sender_sp, -16); // done with stack
1000 __ ret(lr);
1001
1002 // generate a vanilla interpreter entry as the slow path
1003 __ bind(slow_path);
1004 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
1005 return entry;
1006
1007 }
1008
1009 /**
1010 * Method entry for static native methods:
1011 * int java.util.zip.CRC32.update(int crc, int b)
1012 */
1013 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1014 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1015 address entry = __ pc();
1016
1017 // rmethod: Method*
1018 // r19_sender_sp: senderSP must preserved for slow path
1019 // esp: args
1020
1021 Label slow_path;
1022 // If we need a safepoint check, generate full interpreter entry.
1023 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1024
1025 // We don't generate local frame and don't align stack because
1026 // we call stub code and there is no safepoint on this path.
1027
1028 // Load parameters
1029 const Register crc = c_rarg0; // crc
1030 const Register val = c_rarg1; // source java byte value
1031 const Register tbl = c_rarg2; // scratch
1032
1033 // Arguments are reversed on java expression stack
1034 __ ldrw(val, Address(esp, 0)); // byte value
1035 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1036
1037 uint64_t offset;
1038 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1039 __ add(tbl, tbl, offset);
1040
1041 __ mvnw(crc, crc); // ~crc
1042 __ update_byte_crc32(crc, val, tbl);
1043 __ mvnw(crc, crc); // ~crc
1044
1045 // result in c_rarg0
1046
1047 __ andr(sp, r19_sender_sp, -16);
1048 __ ret(lr);
1049
1050 // generate a vanilla native entry as the slow path
1051 __ bind(slow_path);
1052 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1053 return entry;
1054 }
1055
1056 /**
1057 * Method entry for static native methods:
1058 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1059 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1060 */
1061 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1062 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1063 address entry = __ pc();
1064
1065 // rmethod,: Method*
1066 // r19_sender_sp: senderSP must preserved for slow path
1067
1068 Label slow_path;
1069 // If we need a safepoint check, generate full interpreter entry.
1070 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1071
1072 // We don't generate local frame and don't align stack because
1073 // we call stub code and there is no safepoint on this path.
1074
1075 // Load parameters
1076 const Register crc = c_rarg0; // crc
1077 const Register buf = c_rarg1; // source java byte array address
1078 const Register len = c_rarg2; // length
1079 const Register off = len; // offset (never overlaps with 'len')
1080
1081 // Arguments are reversed on java expression stack
1082 // Calculate address of start element
1083 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1084 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1085 __ ldrw(off, Address(esp, wordSize)); // offset
1086 __ add(buf, buf, off); // + offset
1087 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1088 } else {
1089 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1090 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1091 __ ldrw(off, Address(esp, wordSize)); // offset
1092 __ add(buf, buf, off); // + offset
1093 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1094 }
1095 // Can now load 'len' since we're finished with 'off'
1096 __ ldrw(len, Address(esp, 0x0)); // Length
1097
1098 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1099
1100 // We are frameless so we can just jump to the stub.
1101 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1102
1103 // generate a vanilla native entry as the slow path
1104 __ bind(slow_path);
1105 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1106 return entry;
1107 }
1108
1109 /**
1110 * Method entry for intrinsic-candidate (non-native) methods:
1111 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1112 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1113 * Unlike CRC32, CRC32C does not have any methods marked as native
1114 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1115 */
1116 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1117 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1118 address entry = __ pc();
1119
1120 // Prepare jump to stub using parameters from the stack
1121 const Register crc = c_rarg0; // initial crc
1122 const Register buf = c_rarg1; // source java byte array address
1123 const Register len = c_rarg2; // len argument to the kernel
1124
1125 const Register end = len; // index of last element to process
1126 const Register off = crc; // offset
1127
1128 __ ldrw(end, Address(esp)); // int end
1129 __ ldrw(off, Address(esp, wordSize)); // int offset
1130 __ sub(len, end, off);
1131 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1132 __ add(buf, buf, off); // + offset
1133 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1134 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1135 } else {
1136 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1137 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1138 }
1139
1140 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1141
1142 // Jump to the stub.
1143 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1144
1145 return entry;
1146 }
1147
1148 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1149 // See more discussion in stackOverflow.hpp.
1150
1151 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1152 const int page_size = (int)os::vm_page_size();
1153 const int n_shadow_pages = shadow_zone_size / page_size;
1154
1155 #ifdef ASSERT
1156 Label L_good_limit;
1157 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1158 __ cbnz(rscratch1, L_good_limit);
1159 __ stop("shadow zone safe limit is not initialized");
1160 __ bind(L_good_limit);
1161
1162 Label L_good_watermark;
1163 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1164 __ cbnz(rscratch1, L_good_watermark);
1165 __ stop("shadow zone growth watermark is not initialized");
1166 __ bind(L_good_watermark);
1167 #endif
1168
1169 Label L_done;
1170
1171 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1172 __ cmp(sp, rscratch1);
1173 __ br(Assembler::HI, L_done);
1174
1175 for (int p = 1; p <= n_shadow_pages; p++) {
1176 __ sub(rscratch2, sp, p*page_size);
1177 __ str(zr, Address(rscratch2));
1178 }
1179
1180 // Record the new watermark, but only if the update is above the safe limit.
1181 // Otherwise, the next time around the check above would pass the safe limit.
1182 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1183 __ cmp(sp, rscratch1);
1184 __ br(Assembler::LS, L_done);
1185 __ mov(rscratch1, sp);
1186 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1187
1188 __ bind(L_done);
1189 }
1190
1191 // Interpreter stub for calling a native method. (asm interpreter)
1192 // This sets up a somewhat different looking stack for calling the
1193 // native method than the typical interpreter frame setup.
1194 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1195 // determine code generation flags
1196 bool inc_counter = UseCompiler || CountCompiledCalls;
1197
1198 // r1: Method*
1199 // rscratch1: sender sp
1200
1201 address entry_point = __ pc();
1202
1203 const Address constMethod (rmethod, Method::const_offset());
1204 const Address access_flags (rmethod, Method::access_flags_offset());
1205 const Address size_of_parameters(r2, ConstMethod::
1206 size_of_parameters_offset());
1207
1208 // get parameter size (always needed)
1209 __ ldr(r2, constMethod);
1210 __ load_unsigned_short(r2, size_of_parameters);
1211
1212 // Native calls don't need the stack size check since they have no
1213 // expression stack and the arguments are already on the stack and
1214 // we only add a handful of words to the stack.
1215
1216 // rmethod: Method*
1217 // r2: size of parameters
1218 // rscratch1: sender sp
1219
1220 // for natives the size of locals is zero
1221
1222 // compute beginning of parameters (rlocals)
1223 __ add(rlocals, esp, r2, ext::uxtx, 3);
1224 __ add(rlocals, rlocals, -wordSize);
1225
1226 // Pull SP back to minimum size: this avoids holes in the stack
1227 __ andr(sp, esp, -16);
1228
1229 // initialize fixed part of activation frame
1230 generate_fixed_frame(true);
1231
1232 // make sure method is native & not abstract
1233 #ifdef ASSERT
1234 __ ldrw(r0, access_flags);
1235 {
1236 Label L;
1237 __ tst(r0, JVM_ACC_NATIVE);
1238 __ br(Assembler::NE, L);
1239 __ stop("tried to execute non-native method as native");
1240 __ bind(L);
1241 }
1242 {
1243 Label L;
1244 __ tst(r0, JVM_ACC_ABSTRACT);
1245 __ br(Assembler::EQ, L);
1246 __ stop("tried to execute abstract method in interpreter");
1247 __ bind(L);
1248 }
1249 #endif
1250
1251 // Since at this point in the method invocation the exception
1252 // handler would try to exit the monitor of synchronized methods
1253 // which hasn't been entered yet, we set the thread local variable
1254 // _do_not_unlock_if_synchronized to true. The remove_activation
1255 // will check this flag.
1256
1257 const Address do_not_unlock_if_synchronized(rthread,
1258 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1259 __ mov(rscratch2, true);
1260 __ strb(rscratch2, do_not_unlock_if_synchronized);
1261
1262 // increment invocation count & check for overflow
1263 Label invocation_counter_overflow;
1264 if (inc_counter) {
1265 generate_counter_incr(&invocation_counter_overflow);
1266 }
1267
1268 Label continue_after_compile;
1269 __ bind(continue_after_compile);
1270
1271 bang_stack_shadow_pages(true);
1272
1273 // reset the _do_not_unlock_if_synchronized flag
1274 __ strb(zr, do_not_unlock_if_synchronized);
1275
1276 // check for synchronized methods
1277 // Must happen AFTER invocation_counter check and stack overflow check,
1278 // so method is not locked if overflows.
1279 if (synchronized) {
1280 lock_method();
1281 } else {
1282 // no synchronization necessary
1283 #ifdef ASSERT
1284 {
1285 Label L;
1286 __ ldrw(r0, access_flags);
1287 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1288 __ br(Assembler::EQ, L);
1289 __ stop("method needs synchronization");
1290 __ bind(L);
1291 }
1292 #endif
1293 }
1294
1295 // start execution
1296 #ifdef ASSERT
1297 {
1298 Label L;
1299 const Address monitor_block_top(rfp,
1300 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1301 __ ldr(rscratch1, monitor_block_top);
1302 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1303 __ cmp(esp, rscratch1);
1304 __ br(Assembler::EQ, L);
1305 __ stop("broken stack frame setup in interpreter 1");
1306 __ bind(L);
1307 }
1308 #endif
1309
1310 // jvmti support
1311 __ notify_method_entry();
1312
1313 // work registers
1314 const Register t = r17;
1315 const Register result_handler = r19;
1316
1317 // allocate space for parameters
1318 __ ldr(t, Address(rmethod, Method::const_offset()));
1319 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1320
1321 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1322 __ andr(sp, rscratch1, -16);
1323 __ mov(esp, rscratch1);
1324
1325 // get signature handler
1326 {
1327 Label L;
1328 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1329 __ cbnz(t, L);
1330 __ call_VM(noreg,
1331 CAST_FROM_FN_PTR(address,
1332 InterpreterRuntime::prepare_native_call),
1333 rmethod);
1334 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1335 __ bind(L);
1336 }
1337
1338 // call signature handler
1339 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1340 "adjust this code");
1341 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1342 "adjust this code");
1343 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1344 "adjust this code");
1345
1346 // The generated handlers do not touch rmethod (the method).
1347 // However, large signatures cannot be cached and are generated
1348 // each time here. The slow-path generator can do a GC on return,
1349 // so we must reload it after the call.
1350 __ blr(t);
1351 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1352
1353
1354 // result handler is in r0
1355 // set result handler
1356 __ mov(result_handler, r0);
1357 // pass mirror handle if static call
1358 {
1359 Label L;
1360 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1361 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1362 // get mirror
1363 __ load_mirror(t, rmethod, r10, rscratch2);
1364 // copy mirror into activation frame
1365 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1366 // pass handle to mirror
1367 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1368 __ bind(L);
1369 }
1370
1371 // get native function entry point in r10
1372 {
1373 Label L;
1374 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1375 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1376 __ mov(rscratch2, unsatisfied);
1377 __ ldr(rscratch2, rscratch2);
1378 __ cmp(r10, rscratch2);
1379 __ br(Assembler::NE, L);
1380 __ call_VM(noreg,
1381 CAST_FROM_FN_PTR(address,
1382 InterpreterRuntime::prepare_native_call),
1383 rmethod);
1384 __ get_method(rmethod);
1385 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1386 __ bind(L);
1387 }
1388
1389 // pass JNIEnv
1390 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1391
1392 // Set the last Java PC in the frame anchor to be the return address from
1393 // the call to the native method: this will allow the debugger to
1394 // generate an accurate stack trace.
1395 Label native_return;
1396 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1397
1398 // change thread state
1399 #ifdef ASSERT
1400 {
1401 Label L;
1402 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1403 __ cmp(t, (u1)_thread_in_Java);
1404 __ br(Assembler::EQ, L);
1405 __ stop("Wrong thread state in native stub");
1406 __ bind(L);
1407 }
1408 #endif
1409
1410 // Change state to native
1411 __ mov(rscratch1, _thread_in_native);
1412 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1413 __ stlrw(rscratch1, rscratch2);
1414
1415 // Call the native method.
1416 __ blr(r10);
1417 __ bind(native_return);
1418 __ get_method(rmethod);
1419 // result potentially in r0 or v0
1420
1421 // Restore cpu control state after JNI call
1422 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1423
1424 // make room for the pushes we're about to do
1425 __ sub(rscratch1, esp, 4 * wordSize);
1426 __ andr(sp, rscratch1, -16);
1427
1428 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1429 // in order to extract the result of a method call. If the order of these
1430 // pushes change or anything else is added to the stack then the code in
1431 // interpreter_frame_result must also change.
1432 __ push(dtos);
1433 __ push(ltos);
1434
1435 __ verify_sve_vector_length();
1436
1437 // change thread state
1438 __ mov(rscratch1, _thread_in_native_trans);
1439 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1440 __ stlrw(rscratch1, rscratch2);
1441
1442 // Force this write out before the read below
1443 if (!UseSystemMemoryBarrier) {
1444 __ dmb(Assembler::ISH);
1445 }
1446
1447 // check for safepoint operation in progress and/or pending suspend requests
1448 {
1449 Label L, Continue;
1450
1451 // We need an acquire here to ensure that any subsequent load of the
1452 // global SafepointSynchronize::_state flag is ordered after this load
1453 // of the thread-local polling word. We don't want this poll to
1454 // return false (i.e. not safepointing) and a later poll of the global
1455 // SafepointSynchronize::_state spuriously to return true.
1456 //
1457 // This is to avoid a race when we're in a native->Java transition
1458 // racing the code which wakes up from a safepoint.
1459 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1460 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1461 __ cbz(rscratch2, Continue);
1462 __ bind(L);
1463
1464 // Don't use call_VM as it will see a possible pending exception
1465 // and forward it and never return here preventing us from
1466 // clearing _last_native_pc down below. So we do a runtime call by
1467 // hand.
1468 //
1469 __ mov(c_rarg0, rthread);
1470 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1471 __ blr(rscratch2);
1472 __ get_method(rmethod);
1473 __ reinit_heapbase();
1474 __ bind(Continue);
1475 }
1476
1477 // change thread state
1478 __ mov(rscratch1, _thread_in_Java);
1479 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1480 __ stlrw(rscratch1, rscratch2);
1481
1482 // reset_last_Java_frame
1483 __ reset_last_Java_frame(true);
1484
1485 if (CheckJNICalls) {
1486 // clear_pending_jni_exception_check
1487 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1488 }
1489
1490 // reset handle block
1491 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1492 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1493
1494 // If result is an oop unbox and store it in frame where gc will see it
1495 // and result handler will pick it up
1496
1497 {
1498 Label no_oop;
1499 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1500 __ cmp(t, result_handler);
1501 __ br(Assembler::NE, no_oop);
1502 // Unbox oop result, e.g. JNIHandles::resolve result.
1503 __ pop(ltos);
1504 __ resolve_jobject(r0, t, rscratch2);
1505 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1506 // keep stack depth as expected by pushing oop which will eventually be discarded
1507 __ push(ltos);
1508 __ bind(no_oop);
1509 }
1510
1511 {
1512 Label no_reguard;
1513 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1514 __ ldrw(rscratch1, Address(rscratch1));
1515 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1516 __ br(Assembler::NE, no_reguard);
1517
1518 __ push_call_clobbered_registers();
1519 __ mov(c_rarg0, rthread);
1520 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1521 __ blr(rscratch2);
1522 __ pop_call_clobbered_registers();
1523
1524 __ bind(no_reguard);
1525 }
1526
1527 // The method register is junk from after the thread_in_native transition
1528 // until here. Also can't call_VM until the bcp has been
1529 // restored. Need bcp for throwing exception below so get it now.
1530 __ get_method(rmethod);
1531
1532 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1533 // rbcp == code_base()
1534 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1535 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1536 // handle exceptions (exception handling will handle unlocking!)
1537 {
1538 Label L;
1539 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1540 __ cbz(rscratch1, L);
1541 // Note: At some point we may want to unify this with the code
1542 // used in call_VM_base(); i.e., we should use the
1543 // StubRoutines::forward_exception code. For now this doesn't work
1544 // here because the rsp is not correctly set at this point.
1545 __ MacroAssembler::call_VM(noreg,
1546 CAST_FROM_FN_PTR(address,
1547 InterpreterRuntime::throw_pending_exception));
1548 __ should_not_reach_here();
1549 __ bind(L);
1550 }
1551
1552 // do unlocking if necessary
1553 {
1554 Label L;
1555 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1556 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1557 // the code below should be shared with interpreter macro
1558 // assembler implementation
1559 {
1560 Label unlock;
1561 // BasicObjectLock will be first in list, since this is a
1562 // synchronized method. However, need to check that the object
1563 // has not been unlocked by an explicit monitorexit bytecode.
1564
1565 // monitor expect in c_rarg1 for slow unlock path
1566 __ lea (c_rarg1, Address(rfp, // address of first monitor
1567 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1568 wordSize - sizeof(BasicObjectLock))));
1569
1570 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1571 __ cbnz(t, unlock);
1572
1573 // Entry already unlocked, need to throw exception
1574 __ MacroAssembler::call_VM(noreg,
1575 CAST_FROM_FN_PTR(address,
1576 InterpreterRuntime::throw_illegal_monitor_state_exception));
1577 __ should_not_reach_here();
1578
1579 __ bind(unlock);
1580 __ unlock_object(c_rarg1);
1581 }
1582 __ bind(L);
1583 }
1584
1585 // jvmti support
1586 // Note: This must happen _after_ handling/throwing any exceptions since
1587 // the exception handler code notifies the runtime of method exits
1588 // too. If this happens before, method entry/exit notifications are
1589 // not properly paired (was bug - gri 11/22/99).
1590 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1591
1592 // restore potential result in r0:d0, call result handler to
1593 // restore potential result in ST0 & handle result
1594
1595 __ pop(ltos);
1596 __ pop(dtos);
1597
1598 __ blr(result_handler);
1599
1600 // remove activation
1601 __ ldr(esp, Address(rfp,
1602 frame::interpreter_frame_sender_sp_offset *
1603 wordSize)); // get sender sp
1604 // remove frame anchor
1605 __ leave();
1606
1607 // restore sender sp
1608 __ mov(sp, esp);
1609
1610 __ ret(lr);
1611
1612 if (inc_counter) {
1613 // Handle overflow of counter and compile method
1614 __ bind(invocation_counter_overflow);
1615 generate_counter_overflow(continue_after_compile);
1616 }
1617
1618 return entry_point;
1619 }
1620
1621 //
1622 // Generic interpreted method entry to (asm) interpreter
1623 //
1624 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1625 // determine code generation flags
1626 bool inc_counter = UseCompiler || CountCompiledCalls;
1627
1628 // rscratch1: sender sp
1629 address entry_point = __ pc();
1630
1631 const Address constMethod(rmethod, Method::const_offset());
1632 const Address access_flags(rmethod, Method::access_flags_offset());
1633 const Address size_of_parameters(r3,
1634 ConstMethod::size_of_parameters_offset());
1635 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1636
1637 // get parameter size (always needed)
1638 // need to load the const method first
1639 __ ldr(r3, constMethod);
1640 __ load_unsigned_short(r2, size_of_parameters);
1641
1642 // r2: size of parameters
1643
1644 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1645 __ sub(r3, r3, r2); // r3 = no. of additional locals
1646
1647 // see if we've got enough room on the stack for locals plus overhead.
1648 generate_stack_overflow_check();
1649
1650 // compute beginning of parameters (rlocals)
1651 __ add(rlocals, esp, r2, ext::uxtx, 3);
1652 __ sub(rlocals, rlocals, wordSize);
1653
1654 __ mov(rscratch1, esp);
1655
1656 // r3 - # of additional locals
1657 // allocate space for locals
1658 // explicitly initialize locals
1659 // Initializing memory allocated for locals in the same direction as
1660 // the stack grows to ensure page initialization order according
1661 // to windows-aarch64 stack page growth requirement (see
1662 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1663 {
1664 Label exit, loop;
1665 __ ands(zr, r3, r3);
1666 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1667 __ bind(loop);
1668 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1669 __ sub(r3, r3, 1); // until everything initialized
1670 __ cbnz(r3, loop);
1671 __ bind(exit);
1672 }
1673
1674 // Padding between locals and fixed part of activation frame to ensure
1675 // SP is always 16-byte aligned.
1676 __ andr(sp, rscratch1, -16);
1677
1678 // And the base dispatch table
1679 __ get_dispatch();
1680
1681 // initialize fixed part of activation frame
1682 generate_fixed_frame(false);
1683
1684 // make sure method is not native & not abstract
1685 #ifdef ASSERT
1686 __ ldrw(r0, access_flags);
1687 {
1688 Label L;
1689 __ tst(r0, JVM_ACC_NATIVE);
1690 __ br(Assembler::EQ, L);
1691 __ stop("tried to execute native method as non-native");
1692 __ bind(L);
1693 }
1694 {
1695 Label L;
1696 __ tst(r0, JVM_ACC_ABSTRACT);
1697 __ br(Assembler::EQ, L);
1698 __ stop("tried to execute abstract method in interpreter");
1699 __ bind(L);
1700 }
1701 #endif
1702
1703 // Since at this point in the method invocation the exception
1704 // handler would try to exit the monitor of synchronized methods
1705 // which hasn't been entered yet, we set the thread local variable
1706 // _do_not_unlock_if_synchronized to true. The remove_activation
1707 // will check this flag.
1708
1709 const Address do_not_unlock_if_synchronized(rthread,
1710 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1711 __ mov(rscratch2, true);
1712 __ strb(rscratch2, do_not_unlock_if_synchronized);
1713
1714 Register mdp = r3;
1715 __ profile_parameters_type(mdp, r1, r2);
1716
1717 // increment invocation count & check for overflow
1718 Label invocation_counter_overflow;
1719 if (inc_counter) {
1720 generate_counter_incr(&invocation_counter_overflow);
1721 }
1722
1723 Label continue_after_compile;
1724 __ bind(continue_after_compile);
1725
1726 bang_stack_shadow_pages(false);
1727
1728 // reset the _do_not_unlock_if_synchronized flag
1729 __ strb(zr, do_not_unlock_if_synchronized);
1730
1731 // check for synchronized methods
1732 // Must happen AFTER invocation_counter check and stack overflow check,
1733 // so method is not locked if overflows.
1734 if (synchronized) {
1735 // Allocate monitor and lock method
1736 lock_method();
1737 } else {
1738 // no synchronization necessary
1739 #ifdef ASSERT
1740 {
1741 Label L;
1742 __ ldrw(r0, access_flags);
1743 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1744 __ br(Assembler::EQ, L);
1745 __ stop("method needs synchronization");
1746 __ bind(L);
1747 }
1748 #endif
1749 }
1750
1751 // start execution
1752 #ifdef ASSERT
1753 {
1754 Label L;
1755 const Address monitor_block_top (rfp,
1756 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1757 __ ldr(rscratch1, monitor_block_top);
1758 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1759 __ cmp(esp, rscratch1);
1760 __ br(Assembler::EQ, L);
1761 __ stop("broken stack frame setup in interpreter 2");
1762 __ bind(L);
1763 }
1764 #endif
1765
1766 // jvmti support
1767 __ notify_method_entry();
1768
1769 __ dispatch_next(vtos);
1770
1771 // invocation counter overflow
1772 if (inc_counter) {
1773 // Handle overflow of counter and compile method
1774 __ bind(invocation_counter_overflow);
1775 generate_counter_overflow(continue_after_compile);
1776 }
1777
1778 return entry_point;
1779 }
1780
1781 // Method entry for java.lang.Thread.currentThread
1782 address TemplateInterpreterGenerator::generate_currentThread() {
1783 address entry_point = __ pc();
1784
1785 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1786 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1787 __ ret(lr);
1788
1789 return entry_point;
1790 }
1791
1792 // Not supported
1793 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1794 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1795 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1796 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1797
1798 //-----------------------------------------------------------------------------
1799 // Exceptions
1800
1801 void TemplateInterpreterGenerator::generate_throw_exception() {
1802 // Entry point in previous activation (i.e., if the caller was
1803 // interpreted)
1804 Interpreter::_rethrow_exception_entry = __ pc();
1805 // Restore sp to interpreter_frame_last_sp even though we are going
1806 // to empty the expression stack for the exception processing.
1807 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1808 // r0: exception
1809 // r3: return address/pc that threw exception
1810 __ restore_bcp(); // rbcp points to call/send
1811 __ restore_locals();
1812 __ restore_constant_pool_cache();
1813 __ reinit_heapbase(); // restore rheapbase as heapbase.
1814 __ get_dispatch();
1815
1816 // Entry point for exceptions thrown within interpreter code
1817 Interpreter::_throw_exception_entry = __ pc();
1818 // If we came here via a NullPointerException on the receiver of a
1819 // method, rmethod may be corrupt.
1820 __ get_method(rmethod);
1821 // expression stack is undefined here
1822 // r0: exception
1823 // rbcp: exception bcp
1824 __ verify_oop(r0);
1825 __ mov(c_rarg1, r0);
1826
1827 // expression stack must be empty before entering the VM in case of
1828 // an exception
1829 __ empty_expression_stack();
1830 // find exception handler address and preserve exception oop
1831 __ call_VM(r3,
1832 CAST_FROM_FN_PTR(address,
1833 InterpreterRuntime::exception_handler_for_exception),
1834 c_rarg1);
1835
1836 // Restore machine SP
1837 __ restore_sp_after_call();
1838
1839 // r0: exception handler entry point
1840 // r3: preserved exception oop
1841 // rbcp: bcp for exception handler
1842 __ push_ptr(r3); // push exception which is now the only value on the stack
1843 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1844
1845 // If the exception is not handled in the current frame the frame is
1846 // removed and the exception is rethrown (i.e. exception
1847 // continuation is _rethrow_exception).
1848 //
1849 // Note: At this point the bci is still the bxi for the instruction
1850 // which caused the exception and the expression stack is
1851 // empty. Thus, for any VM calls at this point, GC will find a legal
1852 // oop map (with empty expression stack).
1853
1854 //
1855 // JVMTI PopFrame support
1856 //
1857
1858 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1859 __ empty_expression_stack();
1860 // Set the popframe_processing bit in pending_popframe_condition
1861 // indicating that we are currently handling popframe, so that
1862 // call_VMs that may happen later do not trigger new popframe
1863 // handling cycles.
1864 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1865 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1866 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1867
1868 {
1869 // Check to see whether we are returning to a deoptimized frame.
1870 // (The PopFrame call ensures that the caller of the popped frame is
1871 // either interpreted or compiled and deoptimizes it if compiled.)
1872 // In this case, we can't call dispatch_next() after the frame is
1873 // popped, but instead must save the incoming arguments and restore
1874 // them after deoptimization has occurred.
1875 //
1876 // Note that we don't compare the return PC against the
1877 // deoptimization blob's unpack entry because of the presence of
1878 // adapter frames in C2.
1879 Label caller_not_deoptimized;
1880 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1881 // This is a return address, so requires authenticating for PAC.
1882 __ authenticate_return_address(c_rarg1);
1883 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1884 InterpreterRuntime::interpreter_contains), c_rarg1);
1885 __ cbnz(r0, caller_not_deoptimized);
1886
1887 // Compute size of arguments for saving when returning to
1888 // deoptimized caller
1889 __ get_method(r0);
1890 __ ldr(r0, Address(r0, Method::const_offset()));
1891 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1892 size_of_parameters_offset())));
1893 __ lsl(r0, r0, Interpreter::logStackElementSize);
1894 __ restore_locals(); // XXX do we need this?
1895 __ sub(rlocals, rlocals, r0);
1896 __ add(rlocals, rlocals, wordSize);
1897 // Save these arguments
1898 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1899 Deoptimization::
1900 popframe_preserve_args),
1901 rthread, r0, rlocals);
1902
1903 __ remove_activation(vtos,
1904 /* throw_monitor_exception */ false,
1905 /* install_monitor_exception */ false,
1906 /* notify_jvmdi */ false);
1907
1908 // Inform deoptimization that it is responsible for restoring
1909 // these arguments
1910 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1911 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1912
1913 // Continue in deoptimization handler
1914 __ ret(lr);
1915
1916 __ bind(caller_not_deoptimized);
1917 }
1918
1919 __ remove_activation(vtos,
1920 /* throw_monitor_exception */ false,
1921 /* install_monitor_exception */ false,
1922 /* notify_jvmdi */ false);
1923
1924 // Restore the last_sp and null it out
1925 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1926 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1927 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1928
1929 __ restore_bcp();
1930 __ restore_locals();
1931 __ restore_constant_pool_cache();
1932 __ get_method(rmethod);
1933 __ get_dispatch();
1934
1935 // The method data pointer was incremented already during
1936 // call profiling. We have to restore the mdp for the current bcp.
1937 if (ProfileInterpreter) {
1938 __ set_method_data_pointer_for_bcp();
1939 }
1940
1941 // Clear the popframe condition flag
1942 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1943 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1944
1945 #if INCLUDE_JVMTI
1946 {
1947 Label L_done;
1948
1949 __ ldrb(rscratch1, Address(rbcp, 0));
1950 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1951 __ br(Assembler::NE, L_done);
1952
1953 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1954 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1955
1956 __ ldr(c_rarg0, Address(rlocals, 0));
1957 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1958
1959 __ cbz(r0, L_done);
1960
1961 __ str(r0, Address(esp, 0));
1962 __ bind(L_done);
1963 }
1964 #endif // INCLUDE_JVMTI
1965
1966 // Restore machine SP
1967 __ restore_sp_after_call();
1968
1969 __ dispatch_next(vtos);
1970 // end of PopFrame support
1971
1972 Interpreter::_remove_activation_entry = __ pc();
1973
1974 // preserve exception over this code sequence
1975 __ pop_ptr(r0);
1976 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1977 // remove the activation (without doing throws on illegalMonitorExceptions)
1978 __ remove_activation(vtos, false, true, false);
1979 // restore exception
1980 __ get_vm_result(r0, rthread);
1981
1982 // In between activations - previous activation type unknown yet
1983 // compute continuation point - the continuation point expects the
1984 // following registers set up:
1985 //
1986 // r0: exception
1987 // lr: return address/pc that threw exception
1988 // esp: expression stack of caller
1989 // rfp: fp of caller
1990 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1991 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1992 SharedRuntime::exception_handler_for_return_address),
1993 rthread, lr);
1994 __ mov(r1, r0); // save exception handler
1995 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1996 // We might be returning to a deopt handler that expects r3 to
1997 // contain the exception pc
1998 __ mov(r3, lr);
1999 // Note that an "issuing PC" is actually the next PC after the call
2000 __ br(r1); // jump to exception
2001 // handler of caller
2002 }
2003
2004
2005 //
2006 // JVMTI ForceEarlyReturn support
2007 //
2008 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2009 address entry = __ pc();
2010
2011 __ restore_bcp();
2012 __ restore_locals();
2013 __ empty_expression_stack();
2014 __ load_earlyret_value(state);
2015
2016 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
2017 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
2018
2019 // Clear the earlyret state
2020 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
2021 __ str(zr, cond_addr);
2022
2023 __ remove_activation(state,
2024 false, /* throw_monitor_exception */
2025 false, /* install_monitor_exception */
2026 true); /* notify_jvmdi */
2027 __ ret(lr);
2028
2029 return entry;
2030 } // end of ForceEarlyReturn support
2031
2032
2033
2034 //-----------------------------------------------------------------------------
2035 // Helper for vtos entry point generation
2036
2037 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2038 address& bep,
2039 address& cep,
2040 address& sep,
2041 address& aep,
2042 address& iep,
2043 address& lep,
2044 address& fep,
2045 address& dep,
2046 address& vep) {
2047 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2048 Label L;
2049 aep = __ pc(); __ push_ptr(); __ b(L);
2050 fep = __ pc(); __ push_f(); __ b(L);
2051 dep = __ pc(); __ push_d(); __ b(L);
2052 lep = __ pc(); __ push_l(); __ b(L);
2053 bep = cep = sep =
2054 iep = __ pc(); __ push_i();
2055 vep = __ pc();
2056 __ bind(L);
2057 generate_and_dispatch(t);
2058 }
2059
2060 //-----------------------------------------------------------------------------
2061
2062 // Non-product code
2063 #ifndef PRODUCT
2064 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2065 address entry = __ pc();
2066
2067 __ protect_return_address();
2068 __ push(lr);
2069 __ push(state);
2070 __ push(RegSet::range(r0, r15), sp);
2071 __ mov(c_rarg2, r0); // Pass itos
2072 __ call_VM(noreg,
2073 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2074 c_rarg1, c_rarg2, c_rarg3);
2075 __ pop(RegSet::range(r0, r15), sp);
2076 __ pop(state);
2077 __ pop(lr);
2078 __ authenticate_return_address();
2079 __ ret(lr); // return from result handler
2080
2081 return entry;
2082 }
2083
2084 void TemplateInterpreterGenerator::count_bytecode() {
2085 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2086 __ atomic_addw(noreg, 1, r10);
2087 }
2088
2089 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2090 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2091 __ atomic_addw(noreg, 1, r10);
2092 }
2093
2094 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2095 // Calculate new index for counter:
2096 // _index = (_index >> log2_number_of_codes) |
2097 // (bytecode << log2_number_of_codes);
2098 Register index_addr = rscratch1;
2099 Register index = rscratch2;
2100 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2101 __ ldrw(index, index_addr);
2102 __ mov(r10,
2103 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2104 __ orrw(index, r10, index, Assembler::LSR,
2105 BytecodePairHistogram::log2_number_of_codes);
2106 __ strw(index, index_addr);
2107
2108 // Bump bucket contents:
2109 // _counters[_index] ++;
2110 Register counter_addr = rscratch1;
2111 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2112 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2113 __ atomic_addw(noreg, 1, counter_addr);
2114 }
2115
2116 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2117 // Call a little run-time stub to avoid blow-up for each bytecode.
2118 // The run-time runtime saves the right registers, depending on
2119 // the tosca in-state for the given template.
2120
2121 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2122 "entry must have been generated");
2123 __ bl(Interpreter::trace_code(t->tos_in()));
2124 __ reinit_heapbase();
2125 }
2126
2127
2128 void TemplateInterpreterGenerator::stop_interpreter_at() {
2129 Label L;
2130 __ push(rscratch1);
2131 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2132 __ ldr(rscratch1, Address(rscratch1));
2133 __ mov(rscratch2, StopInterpreterAt);
2134 __ cmpw(rscratch1, rscratch2);
2135 __ br(Assembler::NE, L);
2136 __ brk(0);
2137 __ bind(L);
2138 __ pop(rscratch1);
2139 }
2140
2141 #endif // !PRODUCT