1 /*
2 * Copyright (c) 2003, 2024, 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 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
73 address entry = __ pc();
74
75 __ andr(esp, esp, -16);
76 __ mov(c_rarg3, esp);
77 // rmethod
78 // rlocals
79 // c_rarg3: first stack arg - wordSize
80
81 // adjust sp
82 __ sub(sp, c_rarg3, 18 * wordSize);
83 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
84 __ call_VM(noreg,
85 CAST_FROM_FN_PTR(address,
86 InterpreterRuntime::slow_signature_handler),
87 rmethod, rlocals, c_rarg3);
88
89 // r0: result handler
90
91 // Stack layout:
92 // rsp: return address <- sp
93 // 1 garbage
94 // 8 integer args (if static first is unused)
95 // 1 float/double identifiers
96 // 8 double args
97 // stack args <- esp
98 // garbage
99 // expression stack bottom
100 // bcp (null)
101 // ...
102
103 // Restore LR
104 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
105
106 // Do FP first so we can use c_rarg3 as temp
107 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
108
109 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
110 const FloatRegister r = as_FloatRegister(i);
111
112 Label d, done;
113
114 __ tbnz(c_rarg3, i, d);
115 __ ldrs(r, Address(sp, (10 + i) * wordSize));
116 __ b(done);
117 __ bind(d);
118 __ ldrd(r, Address(sp, (10 + i) * wordSize));
119 __ bind(done);
120 }
121
122 // c_rarg0 contains the result from the call of
123 // InterpreterRuntime::slow_signature_handler so we don't touch it
124 // here. It will be loaded with the JNIEnv* later.
125 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
126 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
127 Register rm = as_Register(i), rn = as_Register(i+1);
128 __ ldp(rm, rn, Address(sp, i * wordSize));
129 }
130
131 __ add(sp, sp, 18 * wordSize);
132 __ ret(lr);
133
134 return entry;
135 }
136
137
138 //
139 // Various method entries
140 //
141
142 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
143 // rmethod: Method*
144 // r19_sender_sp: sender sp
145 // esp: args
146
147 // These don't need a safepoint check because they aren't virtually
148 // callable. We won't enter these intrinsics from compiled code.
149 // If in the future we added an intrinsic which was virtually callable
150 // we'd have to worry about how to safepoint so that this code is used.
151
152 // mathematical functions inlined by compiler
153 // (interpreter must provide identical implementation
154 // in order to avoid monotonicity bugs when switching
155 // from interpreter to compiler in the middle of some
156 // computation)
157 //
158 // stack:
159 // [ arg ] <-- esp
160 // [ arg ]
161 // retaddr in lr
162
163 address entry_point = nullptr;
164 Register continuation = lr;
165 switch (kind) {
166 case Interpreter::java_lang_math_abs:
167 entry_point = __ pc();
168 __ ldrd(v0, Address(esp));
169 __ fabsd(v0, v0);
170 __ mov(sp, r19_sender_sp); // Restore caller's SP
171 break;
172 case Interpreter::java_lang_math_sqrt:
173 entry_point = __ pc();
174 __ ldrd(v0, Address(esp));
175 __ fsqrtd(v0, v0);
176 __ mov(sp, r19_sender_sp);
177 break;
178 case Interpreter::java_lang_math_sin :
179 case Interpreter::java_lang_math_cos :
180 case Interpreter::java_lang_math_tan :
181 case Interpreter::java_lang_math_log :
182 case Interpreter::java_lang_math_log10 :
183 case Interpreter::java_lang_math_exp :
184 entry_point = __ pc();
185 __ ldrd(v0, Address(esp));
186 __ mov(sp, r19_sender_sp);
187 __ mov(r23, lr);
188 continuation = r23; // The first free callee-saved register
189 generate_transcendental_entry(kind, 1);
190 break;
191 case Interpreter::java_lang_math_pow :
192 entry_point = __ pc();
193 __ mov(r23, lr);
194 continuation = r23;
195 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
196 __ ldrd(v1, Address(esp));
197 __ mov(sp, r19_sender_sp);
198 generate_transcendental_entry(kind, 2);
199 break;
200 case Interpreter::java_lang_math_fmaD :
201 if (UseFMA) {
202 entry_point = __ pc();
203 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
204 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
205 __ ldrd(v2, Address(esp));
206 __ fmaddd(v0, v0, v1, v2);
207 __ mov(sp, r19_sender_sp); // Restore caller's SP
208 }
209 break;
210 case Interpreter::java_lang_math_fmaF :
211 if (UseFMA) {
212 entry_point = __ pc();
213 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
214 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
215 __ ldrs(v2, Address(esp));
216 __ fmadds(v0, v0, v1, v2);
217 __ mov(sp, r19_sender_sp); // Restore caller's SP
218 }
219 break;
220 default:
221 ;
222 }
223 if (entry_point) {
224 __ br(continuation);
225 }
226
227 return entry_point;
228 }
229
230 // double trigonometrics and transcendentals
231 // static jdouble dsin(jdouble x);
232 // static jdouble dcos(jdouble x);
233 // static jdouble dtan(jdouble x);
234 // static jdouble dlog(jdouble x);
235 // static jdouble dlog10(jdouble x);
236 // static jdouble dexp(jdouble x);
237 // static jdouble dpow(jdouble x, jdouble y);
238
239 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
240 address fn;
241 switch (kind) {
242 case Interpreter::java_lang_math_sin :
243 if (StubRoutines::dsin() == nullptr) {
244 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
245 } else {
246 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
247 }
248 break;
249 case Interpreter::java_lang_math_cos :
250 if (StubRoutines::dcos() == nullptr) {
251 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
252 } else {
253 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
254 }
255 break;
256 case Interpreter::java_lang_math_tan :
257 if (StubRoutines::dtan() == nullptr) {
258 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
259 } else {
260 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
261 }
262 break;
263 case Interpreter::java_lang_math_log :
264 if (StubRoutines::dlog() == nullptr) {
265 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
266 } else {
267 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
268 }
269 break;
270 case Interpreter::java_lang_math_log10 :
271 if (StubRoutines::dlog10() == nullptr) {
272 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
273 } else {
274 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
275 }
276 break;
277 case Interpreter::java_lang_math_exp :
278 if (StubRoutines::dexp() == nullptr) {
279 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
280 } else {
281 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
282 }
283 break;
284 case Interpreter::java_lang_math_pow :
285 if (StubRoutines::dpow() == nullptr) {
286 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
287 } else {
288 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
289 }
290 break;
291 default:
292 ShouldNotReachHere();
293 fn = nullptr; // unreachable
294 }
295 __ mov(rscratch1, fn);
296 __ blr(rscratch1);
297 }
298
299 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
300 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
301 // r19_sender_sp: sender sp
302 // stack:
303 // [ arg ] <-- esp
304 // [ arg ]
305 // retaddr in lr
306 // result in v0
307
308 address entry_point = __ pc();
309 __ ldrw(c_rarg0, Address(esp));
310 __ flt16_to_flt(v0, c_rarg0, v1);
311 __ mov(sp, r19_sender_sp); // Restore caller's SP
312 __ br(lr);
313 return entry_point;
314 }
315
316 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
317 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
318 // r19_sender_sp: sender sp
319 // stack:
320 // [ arg ] <-- esp
321 // [ arg ]
322 // retaddr in lr
323 // result in c_rarg0
324
325 address entry_point = __ pc();
326 __ ldrs(v0, Address(esp));
327 __ flt_to_flt16(c_rarg0, v0, v1);
328 __ mov(sp, r19_sender_sp); // Restore caller's SP
329 __ br(lr);
330 return entry_point;
331 }
332
333 // Abstract method entry
334 // Attempt to execute abstract method. Throw exception
335 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
336 // rmethod: Method*
337 // r19_sender_sp: sender SP
338
339 address entry_point = __ pc();
340
341 // abstract method entry
342
343 // pop return address, reset last_sp to null
344 __ empty_expression_stack();
345 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
346 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
347
348 // throw exception
349 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
350 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
351 rmethod);
352 // the call_VM checks for exception, so we should never return here.
353 __ should_not_reach_here();
354
355 return entry_point;
356 }
357
358 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
359 address entry = __ pc();
360
361 #ifdef ASSERT
362 {
363 Label L;
364 __ ldr(rscratch1, Address(rfp,
365 frame::interpreter_frame_monitor_block_top_offset *
366 wordSize));
367 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
368 __ mov(rscratch2, sp);
369 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
370 // grows negative)
371 __ br(Assembler::HS, L); // check if frame is complete
372 __ stop ("interpreter frame not set up");
373 __ bind(L);
374 }
375 #endif // ASSERT
376 // Restore bcp under the assumption that the current frame is still
377 // interpreted
378 __ restore_bcp();
379
380 // expression stack must be empty before entering the VM if an
381 // exception happened
382 __ empty_expression_stack();
383 // throw exception
384 __ call_VM(noreg,
385 CAST_FROM_FN_PTR(address,
386 InterpreterRuntime::throw_StackOverflowError));
387 return entry;
388 }
389
390 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
391 address entry = __ pc();
392 // expression stack must be empty before entering the VM if an
393 // exception happened
394 __ empty_expression_stack();
395 // setup parameters
396
397 // ??? convention: expect aberrant index in register r1
398 __ movw(c_rarg2, r1);
399 // ??? convention: expect array in register r3
400 __ mov(c_rarg1, r3);
401 __ call_VM(noreg,
402 CAST_FROM_FN_PTR(address,
403 InterpreterRuntime::
404 throw_ArrayIndexOutOfBoundsException),
405 c_rarg1, c_rarg2);
406 return entry;
407 }
408
409 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
410 address entry = __ pc();
411
412 // object is at TOS
413 __ pop(c_rarg1);
414
415 // expression stack must be empty before entering the VM if an
416 // exception happened
417 __ empty_expression_stack();
418
419 __ call_VM(noreg,
420 CAST_FROM_FN_PTR(address,
421 InterpreterRuntime::
422 throw_ClassCastException),
423 c_rarg1);
424 return entry;
425 }
426
427 address TemplateInterpreterGenerator::generate_exception_handler_common(
428 const char* name, const char* message, bool pass_oop) {
429 assert(!pass_oop || message == nullptr, "either oop or message but not both");
430 address entry = __ pc();
431 if (pass_oop) {
432 // object is at TOS
433 __ pop(c_rarg2);
434 }
435 // expression stack must be empty before entering the VM if an
436 // exception happened
437 __ empty_expression_stack();
438 // setup parameters
439 __ lea(c_rarg1, Address((address)name));
440 if (pass_oop) {
441 __ call_VM(r0, CAST_FROM_FN_PTR(address,
442 InterpreterRuntime::
443 create_klass_exception),
444 c_rarg1, c_rarg2);
445 } else {
446 // kind of lame ExternalAddress can't take null because
447 // external_word_Relocation will assert.
448 if (message != nullptr) {
449 __ lea(c_rarg2, Address((address)message));
450 } else {
451 __ mov(c_rarg2, NULL_WORD);
452 }
453 __ call_VM(r0,
454 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
455 c_rarg1, c_rarg2);
456 }
457 // throw exception
458 __ b(address(Interpreter::throw_exception_entry()));
459 return entry;
460 }
461
462 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
463 address entry = __ pc();
464
465 // Restore stack bottom in case i2c adjusted stack
466 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
467 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
468 // and null it as marker that esp is now tos until next java call
469 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
470 __ restore_bcp();
471 __ restore_locals();
472 __ restore_constant_pool_cache();
473 __ get_method(rmethod);
474
475 if (state == atos) {
476 Register obj = r0;
477 Register mdp = r1;
478 Register tmp = r2;
479 __ profile_return_type(mdp, obj, tmp);
480 }
481
482 const Register cache = r1;
483 const Register index = r2;
484
485 if (index_size == sizeof(u4)) {
486 __ load_resolved_indy_entry(cache, index);
487 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
488 __ add(esp, esp, cache, Assembler::LSL, 3);
489 } else {
490 // Pop N words from the stack
491 assert(index_size == sizeof(u2), "Can only be u2");
492 __ load_method_entry(cache, index);
493 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
494 __ add(esp, esp, cache, Assembler::LSL, 3);
495 }
496
497 // Restore machine SP
498 __ restore_sp_after_call();
499
500 __ check_and_handle_popframe(rthread);
501 __ check_and_handle_earlyret(rthread);
502
503 __ get_dispatch();
504 __ dispatch_next(state, step);
505
506 return entry;
507 }
508
509 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
510 int step,
511 address continuation) {
512 address entry = __ pc();
513 __ restore_bcp();
514 __ restore_locals();
515 __ restore_constant_pool_cache();
516 __ get_method(rmethod);
517 __ get_dispatch();
518
519 __ restore_sp_after_call(); // Restore SP to extended SP
520
521 // Restore expression stack pointer
522 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
523 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
524 // null last_sp until next java call
525 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
526
527 #if INCLUDE_JVMCI
528 // Check if we need to take lock at entry of synchronized method. This can
529 // only occur on method entry so emit it only for vtos with step 0.
530 if (EnableJVMCI && state == vtos && step == 0) {
531 Label L;
532 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
533 __ cbz(rscratch1, L);
534 // Clear flag.
535 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
536 // Take lock.
537 lock_method();
538 __ bind(L);
539 } else {
540 #ifdef ASSERT
541 if (EnableJVMCI) {
542 Label L;
543 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
544 __ cbz(rscratch1, L);
545 __ stop("unexpected pending monitor in deopt entry");
546 __ bind(L);
547 }
548 #endif
549 }
550 #endif
551 // handle exceptions
552 {
553 Label L;
554 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
555 __ cbz(rscratch1, L);
556 __ call_VM(noreg,
557 CAST_FROM_FN_PTR(address,
558 InterpreterRuntime::throw_pending_exception));
559 __ should_not_reach_here();
560 __ bind(L);
561 }
562
563 if (continuation == nullptr) {
564 __ dispatch_next(state, step);
565 } else {
566 __ jump_to_entry(continuation);
567 }
568 return entry;
569 }
570
571 address TemplateInterpreterGenerator::generate_result_handler_for(
572 BasicType type) {
573 address entry = __ pc();
574 switch (type) {
575 case T_BOOLEAN: __ c2bool(r0); break;
576 case T_CHAR : __ uxth(r0, r0); break;
577 case T_BYTE : __ sxtb(r0, r0); break;
578 case T_SHORT : __ sxth(r0, r0); break;
579 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
580 case T_LONG : /* nothing to do */ break;
581 case T_VOID : /* nothing to do */ break;
582 case T_FLOAT : /* nothing to do */ break;
583 case T_DOUBLE : /* nothing to do */ break;
584 case T_OBJECT :
585 // retrieve result from frame
586 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
587 // and verify it
588 __ verify_oop(r0);
589 break;
590 default : ShouldNotReachHere();
591 }
592 __ ret(lr); // return from result handler
593 return entry;
594 }
595
596 address TemplateInterpreterGenerator::generate_safept_entry_for(
597 TosState state,
598 address runtime_entry) {
599 address entry = __ pc();
600 __ push(state);
601 __ push_cont_fastpath(rthread);
602 __ call_VM(noreg, runtime_entry);
603 __ pop_cont_fastpath(rthread);
604 __ membar(Assembler::AnyAny);
605 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
606 return entry;
607 }
608
609 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
610 if (!Continuations::enabled()) return nullptr;
611 address start = __ pc();
612
613 // Restore rfp first since we need it to restore rest of registers
614 __ leave();
615
616 // Restore constant pool cache
617 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
618
619 // Restore Java expression stack pointer
620 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
621 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
622 // and NULL it as marker that esp is now tos until next java call
623 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
624
625 // Restore machine SP
626 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
627 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord)));
628
629 // Prepare for adjustment on return to call_VM_leaf_base()
630 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
631 __ stp(rscratch1, rmethod, Address(__ pre(sp, -2 * wordSize)));
632
633 // Restore dispatch
634 uint64_t offset;
635 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
636 __ add(rdispatch, rdispatch, offset);
637
638 __ ret(lr);
639
640 return start;
641 }
642
643
644 // Helpers for commoning out cases in the various type of method entries.
645 //
646
647
648 // increment invocation count & check for overflow
649 //
650 // Note: checking for negative value instead of overflow
651 // so we have a 'sticky' overflow test
652 //
653 // rmethod: method
654 //
655 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
656 Label done;
657 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
658 int increment = InvocationCounter::count_increment;
659 Label no_mdo;
660 if (ProfileInterpreter) {
661 // Are we profiling?
662 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
663 __ cbz(r0, no_mdo);
664 // Increment counter in the MDO
665 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
666 in_bytes(InvocationCounter::counter_offset()));
667 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
668 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
669 __ b(done);
670 }
671 __ bind(no_mdo);
672 // Increment counter in MethodCounters
673 const Address invocation_counter(rscratch2,
674 MethodCounters::invocation_counter_offset() +
675 InvocationCounter::counter_offset());
676 __ get_method_counters(rmethod, rscratch2, done);
677 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
678 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
679 __ bind(done);
680 }
681
682 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
683
684 // Asm interpreter on entry
685 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
686 // Everything as it was on entry
687
688 // InterpreterRuntime::frequency_counter_overflow takes two
689 // arguments, the first (thread) is passed by call_VM, the second
690 // indicates if the counter overflow occurs at a backwards branch
691 // (null bcp). We pass zero for it. The call returns the address
692 // of the verified entry point for the method or null if the
693 // compilation did not complete (either went background or bailed
694 // out).
695 __ mov(c_rarg1, 0);
696 __ call_VM(noreg,
697 CAST_FROM_FN_PTR(address,
698 InterpreterRuntime::frequency_counter_overflow),
699 c_rarg1);
700
701 __ b(do_continue);
702 }
703
704 // See if we've got enough room on the stack for locals plus overhead
705 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
706 // without going through the signal handler, i.e., reserved and yellow zones
707 // will not be made usable. The shadow zone must suffice to handle the
708 // overflow.
709 // The expression stack grows down incrementally, so the normal guard
710 // page mechanism will work for that.
711 //
712 // NOTE: Since the additional locals are also always pushed (wasn't
713 // obvious in generate_method_entry) so the guard should work for them
714 // too.
715 //
716 // Args:
717 // r3: number of additional locals this frame needs (what we must check)
718 // rmethod: Method*
719 //
720 // Kills:
721 // r0
722 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
723
724 // monitor entry size: see picture of stack set
725 // (generate_method_entry) and frame_amd64.hpp
726 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
727
728 // total overhead size: entry_size + (saved rbp through expr stack
729 // bottom). be sure to change this if you add/subtract anything
730 // to/from the overhead area
731 const int overhead_size =
732 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
733
734 const size_t page_size = os::vm_page_size();
735
736 Label after_frame_check;
737
738 // see if the frame is greater than one page in size. If so,
739 // then we need to verify there is enough stack space remaining
740 // for the additional locals.
741 //
742 // Note that we use SUBS rather than CMP here because the immediate
743 // field of this instruction may overflow. SUBS can cope with this
744 // because it is a macro that will expand to some number of MOV
745 // instructions and a register operation.
746 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
747 __ br(Assembler::LS, after_frame_check);
748
749 // compute rsp as if this were going to be the last frame on
750 // the stack before the red zone
751
752 // locals + overhead, in bytes
753 __ mov(r0, overhead_size);
754 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
755
756 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
757 __ ldr(rscratch1, stack_limit);
758
759 #ifdef ASSERT
760 Label limit_okay;
761 // Verify that thread stack limit is non-zero.
762 __ cbnz(rscratch1, limit_okay);
763 __ stop("stack overflow limit is zero");
764 __ bind(limit_okay);
765 #endif
766
767 // Add stack limit to locals.
768 __ add(r0, r0, rscratch1);
769
770 // Check against the current stack bottom.
771 __ cmp(sp, r0);
772 __ br(Assembler::HI, after_frame_check);
773
774 // Remove the incoming args, peeling the machine SP back to where it
775 // was in the caller. This is not strictly necessary, but unless we
776 // do so the stack frame may have a garbage FP; this ensures a
777 // correct call stack that we can always unwind. The ANDR should be
778 // unnecessary because the sender SP in r19 is always aligned, but
779 // it doesn't hurt.
780 __ andr(sp, r19_sender_sp, -16);
781
782 // Note: the restored frame is not necessarily interpreted.
783 // Use the shared runtime version of the StackOverflowError.
784 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
785 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
786
787 // all done with frame size check
788 __ bind(after_frame_check);
789 }
790
791 // Allocate monitor and lock method (asm interpreter)
792 //
793 // Args:
794 // rmethod: Method*
795 // rlocals: locals
796 //
797 // Kills:
798 // r0
799 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
800 // rscratch1, rscratch2 (scratch regs)
801 void TemplateInterpreterGenerator::lock_method() {
802 // synchronize method
803 const Address access_flags(rmethod, Method::access_flags_offset());
804 const Address monitor_block_top(
805 rfp,
806 frame::interpreter_frame_monitor_block_top_offset * wordSize);
807 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
808
809 #ifdef ASSERT
810 {
811 Label L;
812 __ ldrw(r0, access_flags);
813 __ tst(r0, JVM_ACC_SYNCHRONIZED);
814 __ br(Assembler::NE, L);
815 __ stop("method doesn't need synchronization");
816 __ bind(L);
817 }
818 #endif // ASSERT
819
820 // get synchronization object
821 {
822 Label done;
823 __ ldrw(r0, access_flags);
824 __ tst(r0, JVM_ACC_STATIC);
825 // get receiver (assume this is frequent case)
826 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
827 __ br(Assembler::EQ, done);
828 __ load_mirror(r0, rmethod, r5, rscratch2);
829
830 #ifdef ASSERT
831 {
832 Label L;
833 __ cbnz(r0, L);
834 __ stop("synchronization object is null");
835 __ bind(L);
836 }
837 #endif // ASSERT
838
839 __ bind(done);
840 }
841
842 // add space for monitor & lock
843 __ check_extended_sp();
844 __ sub(sp, sp, entry_size); // add space for a monitor entry
845 __ sub(esp, esp, entry_size);
846 __ sub(rscratch1, sp, rfp);
847 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
848 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
849 __ sub(rscratch1, esp, rfp);
850 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
851 __ str(rscratch1, monitor_block_top); // set new monitor block top
852
853 // store object
854 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
855 __ mov(c_rarg1, esp); // object address
856 __ lock_object(c_rarg1);
857 }
858
859 // Generate a fixed interpreter frame. This is identical setup for
860 // interpreted methods and for native methods hence the shared code.
861 //
862 // Args:
863 // lr: return address
864 // rmethod: Method*
865 // rlocals: pointer to locals
866 // rcpool: cp cache
867 // stack_pointer: previous sp
868 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
869 // initialize fixed part of activation frame
870 if (native_call) {
871 __ sub(esp, sp, 14 * wordSize);
872 __ mov(rbcp, zr);
873 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
874 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
875 // add 2 zero-initialized slots for native calls
876 __ stp(zr, zr, Address(sp, 12 * wordSize));
877 } else {
878 __ sub(esp, sp, 12 * wordSize);
879 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
880 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
881 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
882 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
883 }
884
885 if (ProfileInterpreter) {
886 Label method_data_continue;
887 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
888 __ cbz(rscratch1, method_data_continue);
889 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
890 __ bind(method_data_continue);
891 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
892 } else {
893 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
894 }
895
896 __ protect_return_address();
897 __ stp(rfp, lr, Address(sp, 10 * wordSize));
898 __ lea(rfp, Address(sp, 10 * wordSize));
899
900 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
901 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
902 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
903 __ sub(rscratch1, rlocals, rfp);
904 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
905 // Store relativized rlocals, see frame::interpreter_frame_locals().
906 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
907
908 // set sender sp
909 // leave last_sp as null
910 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
911
912 // Get mirror
913 __ load_mirror(r10, rmethod, r5, rscratch2);
914 if (! native_call) {
915 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
916 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
917 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
918 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
919 __ andr(rscratch1, rscratch1, -16);
920 __ sub(rscratch2, rscratch1, rfp);
921 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
922 // Store extended SP and mirror
923 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
924 // Move SP out of the way
925 __ mov(sp, rscratch1);
926 } else {
927 // Make sure there is room for the exception oop pushed in case method throws
928 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
929 __ sub(rscratch1, sp, 2 * wordSize);
930 __ sub(rscratch2, rscratch1, rfp);
931 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
932 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
933 __ mov(sp, rscratch1);
934 }
935 }
936
937 // End of helpers
938
939 // Various method entries
940 //------------------------------------------------------------------------------------------------------------------------
941 //
942 //
943
944 // Method entry for java.lang.ref.Reference.get.
945 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
946 // Code: _aload_0, _getfield, _areturn
947 // parameter size = 1
948 //
949 // The code that gets generated by this routine is split into 2 parts:
950 // 1. The "intrinsified" code for G1 (or any SATB based GC),
951 // 2. The slow path - which is an expansion of the regular method entry.
952 //
953 // Notes:-
954 // * In the G1 code we do not check whether we need to block for
955 // a safepoint. If G1 is enabled then we must execute the specialized
956 // code for Reference.get (except when the Reference object is null)
957 // so that we can log the value in the referent field with an SATB
958 // update buffer.
959 // If the code for the getfield template is modified so that the
960 // G1 pre-barrier code is executed when the current method is
961 // Reference.get() then going through the normal method entry
962 // will be fine.
963 // * The G1 code can, however, check the receiver object (the instance
964 // of java.lang.Reference) and jump to the slow path if null. If the
965 // Reference object is null then we obviously cannot fetch the referent
966 // and so we don't need to call the G1 pre-barrier. Thus we can use the
967 // regular method entry code to generate the NPE.
968 //
969 // This code is based on generate_accessor_entry.
970 //
971 // rmethod: Method*
972 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
973
974 // LR is live. It must be saved around calls.
975
976 address entry = __ pc();
977
978 const int referent_offset = java_lang_ref_Reference::referent_offset();
979
980 Label slow_path;
981 const Register local_0 = c_rarg0;
982 // Check if local 0 != null
983 // If the receiver is null then it is OK to jump to the slow path.
984 __ ldr(local_0, Address(esp, 0));
985 __ cbz(local_0, slow_path);
986
987 // Load the value of the referent field.
988 const Address field_address(local_0, referent_offset);
989 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
990 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
991
992 // areturn
993 __ andr(sp, r19_sender_sp, -16); // done with stack
994 __ ret(lr);
995
996 // generate a vanilla interpreter entry as the slow path
997 __ bind(slow_path);
998 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
999 return entry;
1000
1001 }
1002
1003 /**
1004 * Method entry for static native methods:
1005 * int java.util.zip.CRC32.update(int crc, int b)
1006 */
1007 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1008 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1009 address entry = __ pc();
1010
1011 // rmethod: Method*
1012 // r19_sender_sp: senderSP must preserved for slow path
1013 // esp: args
1014
1015 Label slow_path;
1016 // If we need a safepoint check, generate full interpreter entry.
1017 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1018
1019 // We don't generate local frame and don't align stack because
1020 // we call stub code and there is no safepoint on this path.
1021
1022 // Load parameters
1023 const Register crc = c_rarg0; // crc
1024 const Register val = c_rarg1; // source java byte value
1025 const Register tbl = c_rarg2; // scratch
1026
1027 // Arguments are reversed on java expression stack
1028 __ ldrw(val, Address(esp, 0)); // byte value
1029 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1030
1031 uint64_t offset;
1032 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1033 __ add(tbl, tbl, offset);
1034
1035 __ mvnw(crc, crc); // ~crc
1036 __ update_byte_crc32(crc, val, tbl);
1037 __ mvnw(crc, crc); // ~crc
1038
1039 // result in c_rarg0
1040
1041 __ andr(sp, r19_sender_sp, -16);
1042 __ ret(lr);
1043
1044 // generate a vanilla native entry as the slow path
1045 __ bind(slow_path);
1046 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1047 return entry;
1048 }
1049
1050 /**
1051 * Method entry for static native methods:
1052 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1053 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1054 */
1055 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1056 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1057 address entry = __ pc();
1058
1059 // rmethod,: Method*
1060 // r19_sender_sp: senderSP must preserved for slow path
1061
1062 Label slow_path;
1063 // If we need a safepoint check, generate full interpreter entry.
1064 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1065
1066 // We don't generate local frame and don't align stack because
1067 // we call stub code and there is no safepoint on this path.
1068
1069 // Load parameters
1070 const Register crc = c_rarg0; // crc
1071 const Register buf = c_rarg1; // source java byte array address
1072 const Register len = c_rarg2; // length
1073 const Register off = len; // offset (never overlaps with 'len')
1074
1075 // Arguments are reversed on java expression stack
1076 // Calculate address of start element
1077 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1078 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1079 __ ldrw(off, Address(esp, wordSize)); // offset
1080 __ add(buf, buf, off); // + offset
1081 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1082 } else {
1083 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1084 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1085 __ ldrw(off, Address(esp, wordSize)); // offset
1086 __ add(buf, buf, off); // + offset
1087 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1088 }
1089 // Can now load 'len' since we're finished with 'off'
1090 __ ldrw(len, Address(esp, 0x0)); // Length
1091
1092 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1093
1094 // We are frameless so we can just jump to the stub.
1095 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1096
1097 // generate a vanilla native entry as the slow path
1098 __ bind(slow_path);
1099 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1100 return entry;
1101 }
1102
1103 /**
1104 * Method entry for intrinsic-candidate (non-native) methods:
1105 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1106 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1107 * Unlike CRC32, CRC32C does not have any methods marked as native
1108 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1109 */
1110 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1111 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1112 address entry = __ pc();
1113
1114 // Prepare jump to stub using parameters from the stack
1115 const Register crc = c_rarg0; // initial crc
1116 const Register buf = c_rarg1; // source java byte array address
1117 const Register len = c_rarg2; // len argument to the kernel
1118
1119 const Register end = len; // index of last element to process
1120 const Register off = crc; // offset
1121
1122 __ ldrw(end, Address(esp)); // int end
1123 __ ldrw(off, Address(esp, wordSize)); // int offset
1124 __ sub(len, end, off);
1125 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1126 __ add(buf, buf, off); // + offset
1127 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1128 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1129 } else {
1130 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1131 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1132 }
1133
1134 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1135
1136 // Jump to the stub.
1137 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1138
1139 return entry;
1140 }
1141
1142 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1143 // See more discussion in stackOverflow.hpp.
1144
1145 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1146 const int page_size = (int)os::vm_page_size();
1147 const int n_shadow_pages = shadow_zone_size / page_size;
1148
1149 #ifdef ASSERT
1150 Label L_good_limit;
1151 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1152 __ cbnz(rscratch1, L_good_limit);
1153 __ stop("shadow zone safe limit is not initialized");
1154 __ bind(L_good_limit);
1155
1156 Label L_good_watermark;
1157 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1158 __ cbnz(rscratch1, L_good_watermark);
1159 __ stop("shadow zone growth watermark is not initialized");
1160 __ bind(L_good_watermark);
1161 #endif
1162
1163 Label L_done;
1164
1165 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1166 __ cmp(sp, rscratch1);
1167 __ br(Assembler::HI, L_done);
1168
1169 for (int p = 1; p <= n_shadow_pages; p++) {
1170 __ sub(rscratch2, sp, p*page_size);
1171 __ str(zr, Address(rscratch2));
1172 }
1173
1174 // Record the new watermark, but only if the update is above the safe limit.
1175 // Otherwise, the next time around the check above would pass the safe limit.
1176 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1177 __ cmp(sp, rscratch1);
1178 __ br(Assembler::LS, L_done);
1179 __ mov(rscratch1, sp);
1180 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1181
1182 __ bind(L_done);
1183 }
1184
1185 // Interpreter stub for calling a native method. (asm interpreter)
1186 // This sets up a somewhat different looking stack for calling the
1187 // native method than the typical interpreter frame setup.
1188 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1189 // determine code generation flags
1190 bool inc_counter = UseCompiler || CountCompiledCalls;
1191
1192 // r1: Method*
1193 // rscratch1: sender sp
1194
1195 address entry_point = __ pc();
1196
1197 const Address constMethod (rmethod, Method::const_offset());
1198 const Address access_flags (rmethod, Method::access_flags_offset());
1199 const Address size_of_parameters(r2, ConstMethod::
1200 size_of_parameters_offset());
1201
1202 // get parameter size (always needed)
1203 __ ldr(r2, constMethod);
1204 __ load_unsigned_short(r2, size_of_parameters);
1205
1206 // Native calls don't need the stack size check since they have no
1207 // expression stack and the arguments are already on the stack and
1208 // we only add a handful of words to the stack.
1209
1210 // rmethod: Method*
1211 // r2: size of parameters
1212 // rscratch1: sender sp
1213
1214 // for natives the size of locals is zero
1215
1216 // compute beginning of parameters (rlocals)
1217 __ add(rlocals, esp, r2, ext::uxtx, 3);
1218 __ add(rlocals, rlocals, -wordSize);
1219
1220 // Pull SP back to minimum size: this avoids holes in the stack
1221 __ andr(sp, esp, -16);
1222
1223 // initialize fixed part of activation frame
1224 generate_fixed_frame(true);
1225
1226 // make sure method is native & not abstract
1227 #ifdef ASSERT
1228 __ ldrw(r0, access_flags);
1229 {
1230 Label L;
1231 __ tst(r0, JVM_ACC_NATIVE);
1232 __ br(Assembler::NE, L);
1233 __ stop("tried to execute non-native method as native");
1234 __ bind(L);
1235 }
1236 {
1237 Label L;
1238 __ tst(r0, JVM_ACC_ABSTRACT);
1239 __ br(Assembler::EQ, L);
1240 __ stop("tried to execute abstract method in interpreter");
1241 __ bind(L);
1242 }
1243 #endif
1244
1245 // Since at this point in the method invocation the exception
1246 // handler would try to exit the monitor of synchronized methods
1247 // which hasn't been entered yet, we set the thread local variable
1248 // _do_not_unlock_if_synchronized to true. The remove_activation
1249 // will check this flag.
1250
1251 const Address do_not_unlock_if_synchronized(rthread,
1252 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1253 __ mov(rscratch2, true);
1254 __ strb(rscratch2, do_not_unlock_if_synchronized);
1255
1256 // increment invocation count & check for overflow
1257 Label invocation_counter_overflow;
1258 if (inc_counter) {
1259 generate_counter_incr(&invocation_counter_overflow);
1260 }
1261
1262 Label continue_after_compile;
1263 __ bind(continue_after_compile);
1264
1265 bang_stack_shadow_pages(true);
1266
1267 // reset the _do_not_unlock_if_synchronized flag
1268 __ strb(zr, do_not_unlock_if_synchronized);
1269
1270 // check for synchronized methods
1271 // Must happen AFTER invocation_counter check and stack overflow check,
1272 // so method is not locked if overflows.
1273 if (synchronized) {
1274 lock_method();
1275 } else {
1276 // no synchronization necessary
1277 #ifdef ASSERT
1278 {
1279 Label L;
1280 __ ldrw(r0, access_flags);
1281 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1282 __ br(Assembler::EQ, L);
1283 __ stop("method needs synchronization");
1284 __ bind(L);
1285 }
1286 #endif
1287 }
1288
1289 // start execution
1290 #ifdef ASSERT
1291 {
1292 Label L;
1293 const Address monitor_block_top(rfp,
1294 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1295 __ ldr(rscratch1, monitor_block_top);
1296 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1297 __ cmp(esp, rscratch1);
1298 __ br(Assembler::EQ, L);
1299 __ stop("broken stack frame setup in interpreter 1");
1300 __ bind(L);
1301 }
1302 #endif
1303
1304 // jvmti support
1305 __ notify_method_entry();
1306
1307 // work registers
1308 const Register t = r17;
1309 const Register result_handler = r19;
1310
1311 // allocate space for parameters
1312 __ ldr(t, Address(rmethod, Method::const_offset()));
1313 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1314
1315 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1316 __ andr(sp, rscratch1, -16);
1317 __ mov(esp, rscratch1);
1318
1319 // get signature handler
1320 {
1321 Label L;
1322 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1323 __ cbnz(t, L);
1324 __ call_VM(noreg,
1325 CAST_FROM_FN_PTR(address,
1326 InterpreterRuntime::prepare_native_call),
1327 rmethod);
1328 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1329 __ bind(L);
1330 }
1331
1332 // call signature handler
1333 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1334 "adjust this code");
1335 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1336 "adjust this code");
1337 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1338 "adjust this code");
1339
1340 // The generated handlers do not touch rmethod (the method).
1341 // However, large signatures cannot be cached and are generated
1342 // each time here. The slow-path generator can do a GC on return,
1343 // so we must reload it after the call.
1344 __ blr(t);
1345 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1346
1347
1348 // result handler is in r0
1349 // set result handler
1350 __ mov(result_handler, r0);
1351 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize));
1352
1353 // pass mirror handle if static call
1354 {
1355 Label L;
1356 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1357 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1358 // get mirror
1359 __ load_mirror(t, rmethod, r10, rscratch2);
1360 // copy mirror into activation frame
1361 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1362 // pass handle to mirror
1363 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1364 __ bind(L);
1365 }
1366
1367 // get native function entry point in r10
1368 {
1369 Label L;
1370 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1371 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1372 __ lea(rscratch2, unsatisfied);
1373 __ ldr(rscratch2, rscratch2);
1374 __ cmp(r10, rscratch2);
1375 __ br(Assembler::NE, L);
1376 __ call_VM(noreg,
1377 CAST_FROM_FN_PTR(address,
1378 InterpreterRuntime::prepare_native_call),
1379 rmethod);
1380 __ get_method(rmethod);
1381 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1382 __ bind(L);
1383 }
1384
1385 // pass JNIEnv
1386 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1387
1388 // Set the last Java PC in the frame anchor to be the return address from
1389 // the call to the native method: this will allow the debugger to
1390 // generate an accurate stack trace.
1391 Label native_return;
1392 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1393
1394 // change thread state
1395 #ifdef ASSERT
1396 {
1397 Label L;
1398 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1399 __ cmp(t, (u1)_thread_in_Java);
1400 __ br(Assembler::EQ, L);
1401 __ stop("Wrong thread state in native stub");
1402 __ bind(L);
1403 }
1404 #endif
1405
1406 // Change state to native
1407 __ mov(rscratch1, _thread_in_native);
1408 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1409 __ stlrw(rscratch1, rscratch2);
1410
1411 __ push_cont_fastpath();
1412
1413 // Call the native method.
1414 __ blr(r10);
1415
1416 __ pop_cont_fastpath();
1417
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 // No need for acquire as Java threads always disarm themselves.
1452 __ safepoint_poll(L, true /* at_return */, false /* acquire */, false /* in_nmethod */);
1453 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1454 __ cbz(rscratch2, Continue);
1455 __ bind(L);
1456
1457 // Don't use call_VM as it will see a possible pending exception
1458 // and forward it and never return here preventing us from
1459 // clearing _last_native_pc down below. So we do a runtime call by
1460 // hand.
1461 //
1462 __ mov(c_rarg0, rthread);
1463 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1464 __ blr(rscratch2);
1465 __ get_method(rmethod);
1466 __ reinit_heapbase();
1467 __ bind(Continue);
1468 }
1469
1470 // change thread state
1471 __ mov(rscratch1, _thread_in_Java);
1472 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1473 __ stlrw(rscratch1, rscratch2);
1474
1475 if (LockingMode != LM_LEGACY) {
1476 // Check preemption for Object.wait()
1477 Label not_preempted;
1478 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1479 __ cbz(rscratch1, not_preempted);
1480 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1481 __ br(rscratch1);
1482 __ bind(native_return);
1483 // On resume we need to set up stack as expected
1484 __ push(dtos);
1485 __ push(ltos);
1486 __ bind(not_preempted);
1487 } else {
1488 // any pc will do so just use this one for LM_LEGACY to keep code together.
1489 __ bind(native_return);
1490 }
1491
1492 // reset_last_Java_frame
1493 __ reset_last_Java_frame(true);
1494
1495 if (CheckJNICalls) {
1496 // clear_pending_jni_exception_check
1497 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1498 }
1499
1500 // reset handle block
1501 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1502 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1503
1504 // If result is an oop unbox and store it in frame where gc will see it
1505 // and result handler will pick it up
1506
1507 {
1508 Label no_oop;
1509 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1510 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize));
1511 __ cmp(t, result_handler);
1512 __ br(Assembler::NE, no_oop);
1513 // Unbox oop result, e.g. JNIHandles::resolve result.
1514 __ pop(ltos);
1515 __ resolve_jobject(r0, t, rscratch2);
1516 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1517 // keep stack depth as expected by pushing oop which will eventually be discarded
1518 __ push(ltos);
1519 __ bind(no_oop);
1520 }
1521
1522 {
1523 Label no_reguard;
1524 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1525 __ ldrw(rscratch1, Address(rscratch1));
1526 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1527 __ br(Assembler::NE, no_reguard);
1528
1529 __ push_call_clobbered_registers();
1530 __ mov(c_rarg0, rthread);
1531 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1532 __ blr(rscratch2);
1533 __ pop_call_clobbered_registers();
1534
1535 __ bind(no_reguard);
1536 }
1537
1538 // The method register is junk from after the thread_in_native transition
1539 // until here. Also can't call_VM until the bcp has been
1540 // restored. Need bcp for throwing exception below so get it now.
1541 __ get_method(rmethod);
1542
1543 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1544 // rbcp == code_base()
1545 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1546 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1547 // handle exceptions (exception handling will handle unlocking!)
1548 {
1549 Label L;
1550 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1551 __ cbz(rscratch1, L);
1552 // Note: At some point we may want to unify this with the code
1553 // used in call_VM_base(); i.e., we should use the
1554 // StubRoutines::forward_exception code. For now this doesn't work
1555 // here because the rsp is not correctly set at this point.
1556 __ MacroAssembler::call_VM(noreg,
1557 CAST_FROM_FN_PTR(address,
1558 InterpreterRuntime::throw_pending_exception));
1559 __ should_not_reach_here();
1560 __ bind(L);
1561 }
1562
1563 // do unlocking if necessary
1564 {
1565 Label L;
1566 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1567 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1568 // the code below should be shared with interpreter macro
1569 // assembler implementation
1570 {
1571 Label unlock;
1572 // BasicObjectLock will be first in list, since this is a
1573 // synchronized method. However, need to check that the object
1574 // has not been unlocked by an explicit monitorexit bytecode.
1575
1576 // monitor expect in c_rarg1 for slow unlock path
1577 __ lea (c_rarg1, Address(rfp, // address of first monitor
1578 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1579 wordSize - sizeof(BasicObjectLock))));
1580
1581 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1582 __ cbnz(t, unlock);
1583
1584 // Entry already unlocked, need to throw exception
1585 __ MacroAssembler::call_VM(noreg,
1586 CAST_FROM_FN_PTR(address,
1587 InterpreterRuntime::throw_illegal_monitor_state_exception));
1588 __ should_not_reach_here();
1589
1590 __ bind(unlock);
1591 __ unlock_object(c_rarg1);
1592 }
1593 __ bind(L);
1594 }
1595
1596 // jvmti support
1597 // Note: This must happen _after_ handling/throwing any exceptions since
1598 // the exception handler code notifies the runtime of method exits
1599 // too. If this happens before, method entry/exit notifications are
1600 // not properly paired (was bug - gri 11/22/99).
1601 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1602
1603 // restore potential result in r0:d0, call result handler to
1604 // restore potential result in ST0 & handle result
1605
1606 __ pop(ltos);
1607 __ pop(dtos);
1608
1609 __ blr(result_handler);
1610
1611 // remove activation
1612 __ ldr(esp, Address(rfp,
1613 frame::interpreter_frame_sender_sp_offset *
1614 wordSize)); // get sender sp
1615 // remove frame anchor
1616 __ leave();
1617
1618 // restore sender sp
1619 __ mov(sp, esp);
1620
1621 __ ret(lr);
1622
1623 if (inc_counter) {
1624 // Handle overflow of counter and compile method
1625 __ bind(invocation_counter_overflow);
1626 generate_counter_overflow(continue_after_compile);
1627 }
1628
1629 return entry_point;
1630 }
1631
1632 //
1633 // Generic interpreted method entry to (asm) interpreter
1634 //
1635 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1636 // determine code generation flags
1637 bool inc_counter = UseCompiler || CountCompiledCalls;
1638
1639 // rscratch1: sender sp
1640 address entry_point = __ pc();
1641
1642 const Address constMethod(rmethod, Method::const_offset());
1643 const Address access_flags(rmethod, Method::access_flags_offset());
1644 const Address size_of_parameters(r3,
1645 ConstMethod::size_of_parameters_offset());
1646 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1647
1648 // get parameter size (always needed)
1649 // need to load the const method first
1650 __ ldr(r3, constMethod);
1651 __ load_unsigned_short(r2, size_of_parameters);
1652
1653 // r2: size of parameters
1654
1655 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1656 __ sub(r3, r3, r2); // r3 = no. of additional locals
1657
1658 // see if we've got enough room on the stack for locals plus overhead.
1659 generate_stack_overflow_check();
1660
1661 // compute beginning of parameters (rlocals)
1662 __ add(rlocals, esp, r2, ext::uxtx, 3);
1663 __ sub(rlocals, rlocals, wordSize);
1664
1665 __ mov(rscratch1, esp);
1666
1667 // r3 - # of additional locals
1668 // allocate space for locals
1669 // explicitly initialize locals
1670 // Initializing memory allocated for locals in the same direction as
1671 // the stack grows to ensure page initialization order according
1672 // to windows-aarch64 stack page growth requirement (see
1673 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1674 {
1675 Label exit, loop;
1676 __ ands(zr, r3, r3);
1677 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1678 __ bind(loop);
1679 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1680 __ sub(r3, r3, 1); // until everything initialized
1681 __ cbnz(r3, loop);
1682 __ bind(exit);
1683 }
1684
1685 // Padding between locals and fixed part of activation frame to ensure
1686 // SP is always 16-byte aligned.
1687 __ andr(sp, rscratch1, -16);
1688
1689 // And the base dispatch table
1690 __ get_dispatch();
1691
1692 // initialize fixed part of activation frame
1693 generate_fixed_frame(false);
1694
1695 // make sure method is not native & not abstract
1696 #ifdef ASSERT
1697 __ ldrw(r0, access_flags);
1698 {
1699 Label L;
1700 __ tst(r0, JVM_ACC_NATIVE);
1701 __ br(Assembler::EQ, L);
1702 __ stop("tried to execute native method as non-native");
1703 __ bind(L);
1704 }
1705 {
1706 Label L;
1707 __ tst(r0, JVM_ACC_ABSTRACT);
1708 __ br(Assembler::EQ, L);
1709 __ stop("tried to execute abstract method in interpreter");
1710 __ bind(L);
1711 }
1712 #endif
1713
1714 // Since at this point in the method invocation the exception
1715 // handler would try to exit the monitor of synchronized methods
1716 // which hasn't been entered yet, we set the thread local variable
1717 // _do_not_unlock_if_synchronized to true. The remove_activation
1718 // will check this flag.
1719
1720 const Address do_not_unlock_if_synchronized(rthread,
1721 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1722 __ mov(rscratch2, true);
1723 __ strb(rscratch2, do_not_unlock_if_synchronized);
1724
1725 Register mdp = r3;
1726 __ profile_parameters_type(mdp, r1, r2);
1727
1728 // increment invocation count & check for overflow
1729 Label invocation_counter_overflow;
1730 if (inc_counter) {
1731 generate_counter_incr(&invocation_counter_overflow);
1732 }
1733
1734 Label continue_after_compile;
1735 __ bind(continue_after_compile);
1736
1737 bang_stack_shadow_pages(false);
1738
1739 // reset the _do_not_unlock_if_synchronized flag
1740 __ strb(zr, do_not_unlock_if_synchronized);
1741
1742 // check for synchronized methods
1743 // Must happen AFTER invocation_counter check and stack overflow check,
1744 // so method is not locked if overflows.
1745 if (synchronized) {
1746 // Allocate monitor and lock method
1747 lock_method();
1748 } else {
1749 // no synchronization necessary
1750 #ifdef ASSERT
1751 {
1752 Label L;
1753 __ ldrw(r0, access_flags);
1754 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1755 __ br(Assembler::EQ, L);
1756 __ stop("method needs synchronization");
1757 __ bind(L);
1758 }
1759 #endif
1760 }
1761
1762 // start execution
1763 #ifdef ASSERT
1764 {
1765 Label L;
1766 const Address monitor_block_top (rfp,
1767 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1768 __ ldr(rscratch1, monitor_block_top);
1769 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1770 __ cmp(esp, rscratch1);
1771 __ br(Assembler::EQ, L);
1772 __ stop("broken stack frame setup in interpreter 2");
1773 __ bind(L);
1774 }
1775 #endif
1776
1777 // jvmti support
1778 __ notify_method_entry();
1779
1780 __ dispatch_next(vtos);
1781
1782 // invocation counter overflow
1783 if (inc_counter) {
1784 // Handle overflow of counter and compile method
1785 __ bind(invocation_counter_overflow);
1786 generate_counter_overflow(continue_after_compile);
1787 }
1788
1789 return entry_point;
1790 }
1791
1792 // Method entry for java.lang.Thread.currentThread
1793 address TemplateInterpreterGenerator::generate_currentThread() {
1794 address entry_point = __ pc();
1795
1796 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1797 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1798 __ ret(lr);
1799
1800 return entry_point;
1801 }
1802
1803 // Not supported
1804 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1805 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1806 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1807 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1808
1809 //-----------------------------------------------------------------------------
1810 // Exceptions
1811
1812 void TemplateInterpreterGenerator::generate_throw_exception() {
1813 // Entry point in previous activation (i.e., if the caller was
1814 // interpreted)
1815 Interpreter::_rethrow_exception_entry = __ pc();
1816 // Restore sp to interpreter_frame_last_sp even though we are going
1817 // to empty the expression stack for the exception processing.
1818 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1819 // r0: exception
1820 // r3: return address/pc that threw exception
1821 __ restore_bcp(); // rbcp points to call/send
1822 __ restore_locals();
1823 __ restore_constant_pool_cache();
1824 __ reinit_heapbase(); // restore rheapbase as heapbase.
1825 __ get_dispatch();
1826
1827 // Entry point for exceptions thrown within interpreter code
1828 Interpreter::_throw_exception_entry = __ pc();
1829 // If we came here via a NullPointerException on the receiver of a
1830 // method, rmethod may be corrupt.
1831 __ get_method(rmethod);
1832 // expression stack is undefined here
1833 // r0: exception
1834 // rbcp: exception bcp
1835 __ verify_oop(r0);
1836 __ mov(c_rarg1, r0);
1837
1838 // expression stack must be empty before entering the VM in case of
1839 // an exception
1840 __ empty_expression_stack();
1841 // find exception handler address and preserve exception oop
1842 __ call_VM(r3,
1843 CAST_FROM_FN_PTR(address,
1844 InterpreterRuntime::exception_handler_for_exception),
1845 c_rarg1);
1846
1847 // Restore machine SP
1848 __ restore_sp_after_call();
1849
1850 // r0: exception handler entry point
1851 // r3: preserved exception oop
1852 // rbcp: bcp for exception handler
1853 __ push_ptr(r3); // push exception which is now the only value on the stack
1854 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1855
1856 // If the exception is not handled in the current frame the frame is
1857 // removed and the exception is rethrown (i.e. exception
1858 // continuation is _rethrow_exception).
1859 //
1860 // Note: At this point the bci is still the bxi for the instruction
1861 // which caused the exception and the expression stack is
1862 // empty. Thus, for any VM calls at this point, GC will find a legal
1863 // oop map (with empty expression stack).
1864
1865 //
1866 // JVMTI PopFrame support
1867 //
1868
1869 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1870 __ empty_expression_stack();
1871 // Set the popframe_processing bit in pending_popframe_condition
1872 // indicating that we are currently handling popframe, so that
1873 // call_VMs that may happen later do not trigger new popframe
1874 // handling cycles.
1875 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1876 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1877 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1878
1879 {
1880 // Check to see whether we are returning to a deoptimized frame.
1881 // (The PopFrame call ensures that the caller of the popped frame is
1882 // either interpreted or compiled and deoptimizes it if compiled.)
1883 // In this case, we can't call dispatch_next() after the frame is
1884 // popped, but instead must save the incoming arguments and restore
1885 // them after deoptimization has occurred.
1886 //
1887 // Note that we don't compare the return PC against the
1888 // deoptimization blob's unpack entry because of the presence of
1889 // adapter frames in C2.
1890 Label caller_not_deoptimized;
1891 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1892 // This is a return address, so requires authenticating for PAC.
1893 __ authenticate_return_address(c_rarg1);
1894 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1895 InterpreterRuntime::interpreter_contains), c_rarg1);
1896 __ cbnz(r0, caller_not_deoptimized);
1897
1898 // Compute size of arguments for saving when returning to
1899 // deoptimized caller
1900 __ get_method(r0);
1901 __ ldr(r0, Address(r0, Method::const_offset()));
1902 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1903 size_of_parameters_offset())));
1904 __ lsl(r0, r0, Interpreter::logStackElementSize);
1905 __ restore_locals(); // XXX do we need this?
1906 __ sub(rlocals, rlocals, r0);
1907 __ add(rlocals, rlocals, wordSize);
1908 // Save these arguments
1909 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1910 Deoptimization::
1911 popframe_preserve_args),
1912 rthread, r0, rlocals);
1913
1914 __ remove_activation(vtos,
1915 /* throw_monitor_exception */ false,
1916 /* install_monitor_exception */ false,
1917 /* notify_jvmdi */ false);
1918
1919 // Inform deoptimization that it is responsible for restoring
1920 // these arguments
1921 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1922 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1923
1924 // Continue in deoptimization handler
1925 __ ret(lr);
1926
1927 __ bind(caller_not_deoptimized);
1928 }
1929
1930 __ remove_activation(vtos,
1931 /* throw_monitor_exception */ false,
1932 /* install_monitor_exception */ false,
1933 /* notify_jvmdi */ false);
1934
1935 // Restore the last_sp and null it out
1936 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1937 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1938 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1939
1940 __ restore_bcp();
1941 __ restore_locals();
1942 __ restore_constant_pool_cache();
1943 __ get_method(rmethod);
1944 __ get_dispatch();
1945
1946 // The method data pointer was incremented already during
1947 // call profiling. We have to restore the mdp for the current bcp.
1948 if (ProfileInterpreter) {
1949 __ set_method_data_pointer_for_bcp();
1950 }
1951
1952 // Clear the popframe condition flag
1953 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1954 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1955
1956 #if INCLUDE_JVMTI
1957 {
1958 Label L_done;
1959
1960 __ ldrb(rscratch1, Address(rbcp, 0));
1961 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1962 __ br(Assembler::NE, L_done);
1963
1964 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1965 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1966
1967 __ ldr(c_rarg0, Address(rlocals, 0));
1968 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1969
1970 __ cbz(r0, L_done);
1971
1972 __ str(r0, Address(esp, 0));
1973 __ bind(L_done);
1974 }
1975 #endif // INCLUDE_JVMTI
1976
1977 // Restore machine SP
1978 __ restore_sp_after_call();
1979
1980 __ dispatch_next(vtos);
1981 // end of PopFrame support
1982
1983 Interpreter::_remove_activation_entry = __ pc();
1984
1985 // preserve exception over this code sequence
1986 __ pop_ptr(r0);
1987 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1988 // remove the activation (without doing throws on illegalMonitorExceptions)
1989 __ remove_activation(vtos, false, true, false);
1990 // restore exception
1991 __ get_vm_result(r0, rthread);
1992
1993 // In between activations - previous activation type unknown yet
1994 // compute continuation point - the continuation point expects the
1995 // following registers set up:
1996 //
1997 // r0: exception
1998 // lr: return address/pc that threw exception
1999 // esp: expression stack of caller
2000 // rfp: fp of caller
2001 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
2002 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2003 SharedRuntime::exception_handler_for_return_address),
2004 rthread, lr);
2005 __ mov(r1, r0); // save exception handler
2006 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
2007 // We might be returning to a deopt handler that expects r3 to
2008 // contain the exception pc
2009 __ mov(r3, lr);
2010 // Note that an "issuing PC" is actually the next PC after the call
2011 __ br(r1); // jump to exception
2012 // handler of caller
2013 }
2014
2015
2016 //
2017 // JVMTI ForceEarlyReturn support
2018 //
2019 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2020 address entry = __ pc();
2021
2022 __ restore_bcp();
2023 __ restore_locals();
2024 __ empty_expression_stack();
2025 __ load_earlyret_value(state);
2026
2027 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
2028 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
2029
2030 // Clear the earlyret state
2031 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
2032 __ str(zr, cond_addr);
2033
2034 __ remove_activation(state,
2035 false, /* throw_monitor_exception */
2036 false, /* install_monitor_exception */
2037 true); /* notify_jvmdi */
2038 __ ret(lr);
2039
2040 return entry;
2041 } // end of ForceEarlyReturn support
2042
2043
2044
2045 //-----------------------------------------------------------------------------
2046 // Helper for vtos entry point generation
2047
2048 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2049 address& bep,
2050 address& cep,
2051 address& sep,
2052 address& aep,
2053 address& iep,
2054 address& lep,
2055 address& fep,
2056 address& dep,
2057 address& vep) {
2058 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2059 Label L;
2060 aep = __ pc(); __ push_ptr(); __ b(L);
2061 fep = __ pc(); __ push_f(); __ b(L);
2062 dep = __ pc(); __ push_d(); __ b(L);
2063 lep = __ pc(); __ push_l(); __ b(L);
2064 bep = cep = sep =
2065 iep = __ pc(); __ push_i();
2066 vep = __ pc();
2067 __ bind(L);
2068 generate_and_dispatch(t);
2069 }
2070
2071 //-----------------------------------------------------------------------------
2072
2073 // Non-product code
2074 #ifndef PRODUCT
2075 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2076 address entry = __ pc();
2077
2078 __ protect_return_address();
2079 __ push(lr);
2080 __ push(state);
2081 __ push(RegSet::range(r0, r15), sp);
2082 __ mov(c_rarg2, r0); // Pass itos
2083 __ call_VM(noreg,
2084 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2085 c_rarg1, c_rarg2, c_rarg3);
2086 __ pop(RegSet::range(r0, r15), sp);
2087 __ pop(state);
2088 __ pop(lr);
2089 __ authenticate_return_address();
2090 __ ret(lr); // return from result handler
2091
2092 return entry;
2093 }
2094
2095 void TemplateInterpreterGenerator::count_bytecode() {
2096 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2097 __ atomic_addw(noreg, 1, r10);
2098 }
2099
2100 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2101 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2102 __ atomic_addw(noreg, 1, r10);
2103 }
2104
2105 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2106 // Calculate new index for counter:
2107 // _index = (_index >> log2_number_of_codes) |
2108 // (bytecode << log2_number_of_codes);
2109 Register index_addr = rscratch1;
2110 Register index = rscratch2;
2111 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2112 __ ldrw(index, index_addr);
2113 __ mov(r10,
2114 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2115 __ orrw(index, r10, index, Assembler::LSR,
2116 BytecodePairHistogram::log2_number_of_codes);
2117 __ strw(index, index_addr);
2118
2119 // Bump bucket contents:
2120 // _counters[_index] ++;
2121 Register counter_addr = rscratch1;
2122 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2123 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2124 __ atomic_addw(noreg, 1, counter_addr);
2125 }
2126
2127 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2128 // Call a little run-time stub to avoid blow-up for each bytecode.
2129 // The run-time runtime saves the right registers, depending on
2130 // the tosca in-state for the given template.
2131
2132 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2133 "entry must have been generated");
2134 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2135 __ reinit_heapbase();
2136 }
2137
2138
2139 void TemplateInterpreterGenerator::stop_interpreter_at() {
2140 Label L;
2141 __ push(rscratch1);
2142 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2143 __ ldr(rscratch1, Address(rscratch1));
2144 __ mov(rscratch2, StopInterpreterAt);
2145 __ cmpw(rscratch1, rscratch2);
2146 __ br(Assembler::NE, L);
2147 __ brk(0);
2148 __ bind(L);
2149 __ pop(rscratch1);
2150 }
2151
2152 #endif // !PRODUCT