1 /*
2 * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "asm/macroAssembler.inline.hpp"
27 #include "classfile/javaClasses.hpp"
28 #include "compiler/disassembler.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/inlineKlass.hpp"
45 #include "oops/resolvedIndyEntry.hpp"
46 #include "oops/resolvedMethodEntry.hpp"
47 #include "prims/jvmtiExport.hpp"
48 #include "prims/jvmtiThreadState.hpp"
49 #include "runtime/arguments.hpp"
50 #include "runtime/deoptimization.hpp"
51 #include "runtime/frame.inline.hpp"
52 #include "runtime/globals.hpp"
53 #include "runtime/jniHandles.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/stubRoutines.hpp"
56 #include "runtime/synchronizer.hpp"
57 #include "runtime/timer.hpp"
58 #include "runtime/vframeArray.hpp"
59 #include "utilities/checkedCast.hpp"
60 #include "utilities/debug.hpp"
61 #include "utilities/powerOfTwo.hpp"
62 #include <sys/types.h>
63
64 // Size of interpreter code. Increase if too small. Interpreter will
65 // fail with a guarantee ("not enough space for interpreter generation");
66 // if too small.
67 // Run with +PrintInterpreter to get the VM to print out the size.
68 // Max size with JVMTI
69 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
70
71 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
72
73 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
74 address entry = __ pc();
75
76 __ andr(esp, esp, -16);
77 __ mov(c_rarg3, esp);
78 // rmethod
79 // rlocals
80 // c_rarg3: first stack arg - wordSize
81
82 // adjust sp
83 __ sub(sp, c_rarg3, 18 * wordSize);
84 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
85 __ call_VM(noreg,
86 CAST_FROM_FN_PTR(address,
87 InterpreterRuntime::slow_signature_handler),
88 rmethod, rlocals, c_rarg3);
89
90 // r0: result handler
91
92 // Stack layout:
93 // rsp: return address <- sp
94 // 1 garbage
95 // 8 integer args (if static first is unused)
96 // 1 float/double identifiers
97 // 8 double args
98 // stack args <- esp
99 // garbage
100 // expression stack bottom
101 // bcp (null)
102 // ...
103
104 // Restore LR
105 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
106
107 // Do FP first so we can use c_rarg3 as temp
108 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
109
110 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
111 const FloatRegister r = as_FloatRegister(i);
112
113 Label d, done;
114
115 __ tbnz(c_rarg3, i, d);
116 __ ldrs(r, Address(sp, (10 + i) * wordSize));
117 __ b(done);
118 __ bind(d);
119 __ ldrd(r, Address(sp, (10 + i) * wordSize));
120 __ bind(done);
121 }
122
123 // c_rarg0 contains the result from the call of
124 // InterpreterRuntime::slow_signature_handler so we don't touch it
125 // here. It will be loaded with the JNIEnv* later.
126 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
127 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
128 Register rm = as_Register(i), rn = as_Register(i+1);
129 __ ldp(rm, rn, Address(sp, i * wordSize));
130 }
131
132 __ add(sp, sp, 18 * wordSize);
133 __ ret(lr);
134
135 return entry;
136 }
137
138
139 //
140 // Various method entries
141 //
142
143 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
144 // rmethod: Method*
145 // r19_sender_sp: sender sp
146 // esp: args
147
148 // These don't need a safepoint check because they aren't virtually
149 // callable. We won't enter these intrinsics from compiled code.
150 // If in the future we added an intrinsic which was virtually callable
151 // we'd have to worry about how to safepoint so that this code is used.
152
153 // mathematical functions inlined by compiler
154 // (interpreter must provide identical implementation
155 // in order to avoid monotonicity bugs when switching
156 // from interpreter to compiler in the middle of some
157 // computation)
158 //
159 // stack:
160 // [ arg ] <-- esp
161 // [ arg ]
162 // retaddr in lr
163
164 address entry_point = nullptr;
165 Register continuation = lr;
166 switch (kind) {
167 case Interpreter::java_lang_math_abs:
168 entry_point = __ pc();
169 __ ldrd(v0, Address(esp));
170 __ fabsd(v0, v0);
171 __ mov(sp, r19_sender_sp); // Restore caller's SP
172 break;
173 case Interpreter::java_lang_math_sqrt:
174 entry_point = __ pc();
175 __ ldrd(v0, Address(esp));
176 __ fsqrtd(v0, v0);
177 __ mov(sp, r19_sender_sp);
178 break;
179 case Interpreter::java_lang_math_sin :
180 case Interpreter::java_lang_math_cos :
181 case Interpreter::java_lang_math_tan :
182 case Interpreter::java_lang_math_log :
183 case Interpreter::java_lang_math_log10 :
184 case Interpreter::java_lang_math_exp :
185 entry_point = __ pc();
186 __ ldrd(v0, Address(esp));
187 __ mov(sp, r19_sender_sp);
188 __ mov(r23, lr);
189 continuation = r23; // The first free callee-saved register
190 generate_transcendental_entry(kind, 1);
191 break;
192 case Interpreter::java_lang_math_pow :
193 entry_point = __ pc();
194 __ mov(r23, lr);
195 continuation = r23;
196 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
197 __ ldrd(v1, Address(esp));
198 __ mov(sp, r19_sender_sp);
199 generate_transcendental_entry(kind, 2);
200 break;
201 case Interpreter::java_lang_math_fmaD :
202 if (UseFMA) {
203 entry_point = __ pc();
204 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
205 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
206 __ ldrd(v2, Address(esp));
207 __ fmaddd(v0, v0, v1, v2);
208 __ mov(sp, r19_sender_sp); // Restore caller's SP
209 }
210 break;
211 case Interpreter::java_lang_math_fmaF :
212 if (UseFMA) {
213 entry_point = __ pc();
214 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
215 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
216 __ ldrs(v2, Address(esp));
217 __ fmadds(v0, v0, v1, v2);
218 __ mov(sp, r19_sender_sp); // Restore caller's SP
219 }
220 break;
221 default:
222 ;
223 }
224 if (entry_point) {
225 __ br(continuation);
226 }
227
228 return entry_point;
229 }
230
231 // double trigonometrics and transcendentals
232 // static jdouble dsin(jdouble x);
233 // static jdouble dcos(jdouble x);
234 // static jdouble dtan(jdouble x);
235 // static jdouble dlog(jdouble x);
236 // static jdouble dlog10(jdouble x);
237 // static jdouble dexp(jdouble x);
238 // static jdouble dpow(jdouble x, jdouble y);
239
240 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
241 address fn;
242 switch (kind) {
243 case Interpreter::java_lang_math_sin :
244 if (StubRoutines::dsin() == nullptr) {
245 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
246 } else {
247 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
248 }
249 break;
250 case Interpreter::java_lang_math_cos :
251 if (StubRoutines::dcos() == nullptr) {
252 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
253 } else {
254 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
255 }
256 break;
257 case Interpreter::java_lang_math_tan :
258 if (StubRoutines::dtan() == nullptr) {
259 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
260 } else {
261 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
262 }
263 break;
264 case Interpreter::java_lang_math_log :
265 if (StubRoutines::dlog() == nullptr) {
266 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
267 } else {
268 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
269 }
270 break;
271 case Interpreter::java_lang_math_log10 :
272 if (StubRoutines::dlog10() == nullptr) {
273 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
274 } else {
275 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
276 }
277 break;
278 case Interpreter::java_lang_math_exp :
279 if (StubRoutines::dexp() == nullptr) {
280 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
281 } else {
282 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
283 }
284 break;
285 case Interpreter::java_lang_math_pow :
286 if (StubRoutines::dpow() == nullptr) {
287 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
288 } else {
289 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
290 }
291 break;
292 default:
293 ShouldNotReachHere();
294 fn = nullptr; // unreachable
295 }
296 __ mov(rscratch1, fn);
297 __ blr(rscratch1);
298 }
299
300 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
301 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
302 // r19_sender_sp: sender sp
303 // stack:
304 // [ arg ] <-- esp
305 // [ arg ]
306 // retaddr in lr
307 // result in v0
308
309 address entry_point = __ pc();
310 __ ldrw(c_rarg0, Address(esp));
311 __ flt16_to_flt(v0, c_rarg0, v1);
312 __ mov(sp, r19_sender_sp); // Restore caller's SP
313 __ br(lr);
314 return entry_point;
315 }
316
317 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
318 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
319 // r19_sender_sp: sender sp
320 // stack:
321 // [ arg ] <-- esp
322 // [ arg ]
323 // retaddr in lr
324 // result in c_rarg0
325
326 address entry_point = __ pc();
327 __ ldrs(v0, Address(esp));
328 __ flt_to_flt16(c_rarg0, v0, v1);
329 __ mov(sp, r19_sender_sp); // Restore caller's SP
330 __ br(lr);
331 return entry_point;
332 }
333
334 // Abstract method entry
335 // Attempt to execute abstract method. Throw exception
336 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
337 // rmethod: Method*
338 // r19_sender_sp: sender SP
339
340 address entry_point = __ pc();
341
342 // abstract method entry
343
344 // pop return address, reset last_sp to null
345 __ empty_expression_stack();
346 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
347 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
348
349 // throw exception
350 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
351 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
352 rmethod);
353 // the call_VM checks for exception, so we should never return here.
354 __ should_not_reach_here();
355
356 return entry_point;
357 }
358
359 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
360 address entry = __ pc();
361
362 #ifdef ASSERT
363 {
364 Label L;
365 __ ldr(rscratch1, Address(rfp,
366 frame::interpreter_frame_monitor_block_top_offset *
367 wordSize));
368 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
369 __ mov(rscratch2, sp);
370 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
371 // grows negative)
372 __ br(Assembler::HS, L); // check if frame is complete
373 __ stop ("interpreter frame not set up");
374 __ bind(L);
375 }
376 #endif // ASSERT
377 // Restore bcp under the assumption that the current frame is still
378 // interpreted
379 __ restore_bcp();
380
381 // expression stack must be empty before entering the VM if an
382 // exception happened
383 __ empty_expression_stack();
384 // throw exception
385 __ call_VM(noreg,
386 CAST_FROM_FN_PTR(address,
387 InterpreterRuntime::throw_StackOverflowError));
388 return entry;
389 }
390
391 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
392 address entry = __ pc();
393 // expression stack must be empty before entering the VM if an
394 // exception happened
395 __ empty_expression_stack();
396 // setup parameters
397
398 // ??? convention: expect aberrant index in register r1
399 __ movw(c_rarg2, r1);
400 // ??? convention: expect array in register r3
401 __ mov(c_rarg1, r3);
402 __ call_VM(noreg,
403 CAST_FROM_FN_PTR(address,
404 InterpreterRuntime::
405 throw_ArrayIndexOutOfBoundsException),
406 c_rarg1, c_rarg2);
407 return entry;
408 }
409
410 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
411 address entry = __ pc();
412
413 // object is at TOS
414 __ pop(c_rarg1);
415
416 // expression stack must be empty before entering the VM if an
417 // exception happened
418 __ empty_expression_stack();
419
420 __ call_VM(noreg,
421 CAST_FROM_FN_PTR(address,
422 InterpreterRuntime::
423 throw_ClassCastException),
424 c_rarg1);
425 return entry;
426 }
427
428 address TemplateInterpreterGenerator::generate_exception_handler_common(
429 const char* name, const char* message, bool pass_oop) {
430 assert(!pass_oop || message == nullptr, "either oop or message but not both");
431 address entry = __ pc();
432 if (pass_oop) {
433 // object is at TOS
434 __ pop(c_rarg2);
435 }
436 // expression stack must be empty before entering the VM if an
437 // exception happened
438 __ empty_expression_stack();
439 // setup parameters
440 __ lea(c_rarg1, Address((address)name));
441 if (pass_oop) {
442 __ call_VM(r0, CAST_FROM_FN_PTR(address,
443 InterpreterRuntime::
444 create_klass_exception),
445 c_rarg1, c_rarg2);
446 } else {
447 // kind of lame ExternalAddress can't take null because
448 // external_word_Relocation will assert.
449 if (message != nullptr) {
450 __ lea(c_rarg2, Address((address)message));
451 } else {
452 __ mov(c_rarg2, NULL_WORD);
453 }
454 __ call_VM(r0,
455 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
456 c_rarg1, c_rarg2);
457 }
458 // throw exception
459 __ b(address(Interpreter::throw_exception_entry()));
460 return entry;
461 }
462
463 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
464 address entry = __ pc();
465
466 // Restore stack bottom in case i2c adjusted stack
467 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
468 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
469 // and null it as marker that esp is now tos until next java call
470 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
471
472 if (state == atos && InlineTypeReturnedAsFields) {
473 __ store_inline_type_fields_to_buf(nullptr, true);
474 }
475
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_resume_interpreter_adapter() {
616 if (!Continuations::enabled()) return nullptr;
617 address start = __ pc();
618
619 __ restore_bcp();
620 __ restore_locals();
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 // Restore method
636 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
637
638 // Restore dispatch
639 uint64_t offset;
640 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
641 __ add(rdispatch, rdispatch, offset);
642
643 __ ret(lr);
644
645 return start;
646 }
647
648
649 // Helpers for commoning out cases in the various type of method entries.
650 //
651
652
653 // increment invocation count & check for overflow
654 //
655 // Note: checking for negative value instead of overflow
656 // so we have a 'sticky' overflow test
657 //
658 // rmethod: method
659 //
660 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
661 Label done;
662 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
663 int increment = InvocationCounter::count_increment;
664 Label no_mdo;
665 if (ProfileInterpreter) {
666 // Are we profiling?
667 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
668 __ cbz(r0, no_mdo);
669 // Increment counter in the MDO
670 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
671 in_bytes(InvocationCounter::counter_offset()));
672 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
673 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
674 __ b(done);
675 }
676 __ bind(no_mdo);
677 // Increment counter in MethodCounters
678 const Address invocation_counter(rscratch2,
679 MethodCounters::invocation_counter_offset() +
680 InvocationCounter::counter_offset());
681 __ get_method_counters(rmethod, rscratch2, done);
682 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
683 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
684 __ bind(done);
685 }
686
687 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
688
689 // Asm interpreter on entry
690 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
691 // Everything as it was on entry
692
693 // InterpreterRuntime::frequency_counter_overflow takes two
694 // arguments, the first (thread) is passed by call_VM, the second
695 // indicates if the counter overflow occurs at a backwards branch
696 // (null bcp). We pass zero for it. The call returns the address
697 // of the verified entry point for the method or null if the
698 // compilation did not complete (either went background or bailed
699 // out).
700 __ mov(c_rarg1, 0);
701 __ call_VM(noreg,
702 CAST_FROM_FN_PTR(address,
703 InterpreterRuntime::frequency_counter_overflow),
704 c_rarg1);
705
706 __ b(do_continue);
707 }
708
709 // See if we've got enough room on the stack for locals plus overhead
710 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
711 // without going through the signal handler, i.e., reserved and yellow zones
712 // will not be made usable. The shadow zone must suffice to handle the
713 // overflow.
714 // The expression stack grows down incrementally, so the normal guard
715 // page mechanism will work for that.
716 //
717 // NOTE: Since the additional locals are also always pushed (wasn't
718 // obvious in generate_method_entry) so the guard should work for them
719 // too.
720 //
721 // Args:
722 // r3: number of additional locals this frame needs (what we must check)
723 // rmethod: Method*
724 //
725 // Kills:
726 // r0
727 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
728
729 // monitor entry size: see picture of stack set
730 // (generate_method_entry) and frame_amd64.hpp
731 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
732
733 // total overhead size: entry_size + (saved rbp through expr stack
734 // bottom). be sure to change this if you add/subtract anything
735 // to/from the overhead area
736 const int overhead_size =
737 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
738
739 const size_t page_size = os::vm_page_size();
740
741 Label after_frame_check;
742
743 // see if the frame is greater than one page in size. If so,
744 // then we need to verify there is enough stack space remaining
745 // for the additional locals.
746 //
747 // Note that we use SUBS rather than CMP here because the immediate
748 // field of this instruction may overflow. SUBS can cope with this
749 // because it is a macro that will expand to some number of MOV
750 // instructions and a register operation.
751 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
752 __ br(Assembler::LS, after_frame_check);
753
754 // compute rsp as if this were going to be the last frame on
755 // the stack before the red zone
756
757 // locals + overhead, in bytes
758 __ mov(r0, overhead_size);
759 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
760
761 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
762 __ ldr(rscratch1, stack_limit);
763
764 #ifdef ASSERT
765 Label limit_okay;
766 // Verify that thread stack limit is non-zero.
767 __ cbnz(rscratch1, limit_okay);
768 __ stop("stack overflow limit is zero");
769 __ bind(limit_okay);
770 #endif
771
772 // Add stack limit to locals.
773 __ add(r0, r0, rscratch1);
774
775 // Check against the current stack bottom.
776 __ cmp(sp, r0);
777 __ br(Assembler::HI, after_frame_check);
778
779 // Remove the incoming args, peeling the machine SP back to where it
780 // was in the caller. This is not strictly necessary, but unless we
781 // do so the stack frame may have a garbage FP; this ensures a
782 // correct call stack that we can always unwind. The ANDR should be
783 // unnecessary because the sender SP in r19 is always aligned, but
784 // it doesn't hurt.
785 __ andr(sp, r19_sender_sp, -16);
786
787 // Note: the restored frame is not necessarily interpreted.
788 // Use the shared runtime version of the StackOverflowError.
789 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
790 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
791
792 // all done with frame size check
793 __ bind(after_frame_check);
794 }
795
796 // Allocate monitor and lock method (asm interpreter)
797 //
798 // Args:
799 // rmethod: Method*
800 // rlocals: locals
801 //
802 // Kills:
803 // r0
804 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
805 // rscratch1, rscratch2 (scratch regs)
806 void TemplateInterpreterGenerator::lock_method() {
807 // synchronize method
808 const Address access_flags(rmethod, Method::access_flags_offset());
809 const Address monitor_block_top(
810 rfp,
811 frame::interpreter_frame_monitor_block_top_offset * wordSize);
812 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
813
814 #ifdef ASSERT
815 {
816 Label L;
817 __ ldrh(r0, access_flags);
818 __ tst(r0, JVM_ACC_SYNCHRONIZED);
819 __ br(Assembler::NE, L);
820 __ stop("method doesn't need synchronization");
821 __ bind(L);
822 }
823 #endif // ASSERT
824
825 // get synchronization object
826 {
827 Label done;
828 __ ldrh(r0, access_flags);
829 __ tst(r0, JVM_ACC_STATIC);
830 // get receiver (assume this is frequent case)
831 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
832 __ br(Assembler::EQ, done);
833 __ load_mirror(r0, rmethod, r5, rscratch2);
834
835 #ifdef ASSERT
836 {
837 Label L;
838 __ cbnz(r0, L);
839 __ stop("synchronization object is null");
840 __ bind(L);
841 }
842 #endif // ASSERT
843
844 __ bind(done);
845 }
846
847 // add space for monitor & lock
848 __ check_extended_sp();
849 __ sub(sp, sp, entry_size); // add space for a monitor entry
850 __ sub(esp, esp, entry_size);
851 __ sub(rscratch1, sp, rfp);
852 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
853 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
854 __ sub(rscratch1, esp, rfp);
855 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
856 __ str(rscratch1, monitor_block_top); // set new monitor block top
857
858 // store object
859 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
860 __ mov(c_rarg1, esp); // object address
861 __ lock_object(c_rarg1);
862 }
863
864 // Generate a fixed interpreter frame. This is identical setup for
865 // interpreted methods and for native methods hence the shared code.
866 //
867 // Args:
868 // lr: return address
869 // rmethod: Method*
870 // rlocals: pointer to locals
871 // rcpool: cp cache
872 // stack_pointer: previous sp
873 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
874 // initialize fixed part of activation frame
875 if (native_call) {
876 __ sub(esp, sp, 14 * wordSize);
877 __ mov(rbcp, zr);
878 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
879 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
880 // add 2 zero-initialized slots for native calls
881 __ stp(zr, zr, Address(sp, 12 * wordSize));
882 } else {
883 __ sub(esp, sp, 12 * wordSize);
884 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
885 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
886 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
887 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
888 }
889
890 if (ProfileInterpreter) {
891 Label method_data_continue;
892 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
893 __ cbz(rscratch1, method_data_continue);
894 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
895 __ bind(method_data_continue);
896 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
897 } else {
898 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
899 }
900
901 __ protect_return_address();
902 __ stp(rfp, lr, Address(sp, 10 * wordSize));
903 __ lea(rfp, Address(sp, 10 * wordSize));
904
905 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
906 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
907 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
908 __ sub(rscratch1, rlocals, rfp);
909 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
910 // Store relativized rlocals, see frame::interpreter_frame_locals().
911 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
912
913 // set sender sp
914 // leave last_sp as null
915 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
916
917 // Get mirror
918 __ load_mirror(r10, rmethod, r5, rscratch2);
919 if (! native_call) {
920 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
921 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
922 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
923 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
924 __ andr(rscratch1, rscratch1, -16);
925 __ sub(rscratch2, rscratch1, rfp);
926 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
927 // Store extended SP and mirror
928 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
929 // Move SP out of the way
930 __ mov(sp, rscratch1);
931 } else {
932 // Make sure there is room for the exception oop pushed in case method throws
933 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
934 __ sub(rscratch1, sp, 2 * wordSize);
935 __ sub(rscratch2, rscratch1, rfp);
936 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
937 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
938 __ mov(sp, rscratch1);
939 }
940 }
941
942 // End of helpers
943
944 // Various method entries
945 //------------------------------------------------------------------------------------------------------------------------
946 //
947 //
948
949 // Method entry for java.lang.ref.Reference.get.
950 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
951 // Code: _aload_0, _getfield, _areturn
952 // parameter size = 1
953 //
954 // The code that gets generated by this routine is split into 2 parts:
955 // 1. The "intrinsified" code for G1 (or any SATB based GC),
956 // 2. The slow path - which is an expansion of the regular method entry.
957 //
958 // Notes:-
959 // * In the G1 code we do not check whether we need to block for
960 // a safepoint. If G1 is enabled then we must execute the specialized
961 // code for Reference.get (except when the Reference object is null)
962 // so that we can log the value in the referent field with an SATB
963 // update buffer.
964 // If the code for the getfield template is modified so that the
965 // G1 pre-barrier code is executed when the current method is
966 // Reference.get() then going through the normal method entry
967 // will be fine.
968 // * The G1 code can, however, check the receiver object (the instance
969 // of java.lang.Reference) and jump to the slow path if null. If the
970 // Reference object is null then we obviously cannot fetch the referent
971 // and so we don't need to call the G1 pre-barrier. Thus we can use the
972 // regular method entry code to generate the NPE.
973 //
974 // This code is based on generate_accessor_entry.
975 //
976 // rmethod: Method*
977 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
978
979 // LR is live. It must be saved around calls.
980
981 address entry = __ pc();
982
983 const int referent_offset = java_lang_ref_Reference::referent_offset();
984
985 Label slow_path;
986 const Register local_0 = c_rarg0;
987 // Check if local 0 != null
988 // If the receiver is null then it is OK to jump to the slow path.
989 __ ldr(local_0, Address(esp, 0));
990 __ cbz(local_0, slow_path);
991
992 // Load the value of the referent field.
993 const Address field_address(local_0, referent_offset);
994 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
995 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
996
997 // areturn
998 __ andr(sp, r19_sender_sp, -16); // done with stack
999 __ ret(lr);
1000
1001 // generate a vanilla interpreter entry as the slow path
1002 __ bind(slow_path);
1003 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
1004 return entry;
1005
1006 }
1007
1008 /**
1009 * Method entry for static native methods:
1010 * int java.util.zip.CRC32.update(int crc, int b)
1011 */
1012 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1013 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1014 address entry = __ pc();
1015
1016 // rmethod: Method*
1017 // r19_sender_sp: senderSP must preserved for slow path
1018 // esp: args
1019
1020 Label slow_path;
1021 // If we need a safepoint check, generate full interpreter entry.
1022 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1023
1024 // We don't generate local frame and don't align stack because
1025 // we call stub code and there is no safepoint on this path.
1026
1027 // Load parameters
1028 const Register crc = c_rarg0; // crc
1029 const Register val = c_rarg1; // source java byte value
1030 const Register tbl = c_rarg2; // scratch
1031
1032 // Arguments are reversed on java expression stack
1033 __ ldrw(val, Address(esp, 0)); // byte value
1034 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1035
1036 uint64_t offset;
1037 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1038 __ add(tbl, tbl, offset);
1039
1040 __ mvnw(crc, crc); // ~crc
1041 __ update_byte_crc32(crc, val, tbl);
1042 __ mvnw(crc, crc); // ~crc
1043
1044 // result in c_rarg0
1045
1046 __ andr(sp, r19_sender_sp, -16);
1047 __ ret(lr);
1048
1049 // generate a vanilla native entry as the slow path
1050 __ bind(slow_path);
1051 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1052 return entry;
1053 }
1054
1055 /**
1056 * Method entry for static native methods:
1057 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1058 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1059 */
1060 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1061 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1062 address entry = __ pc();
1063
1064 // rmethod,: Method*
1065 // r19_sender_sp: senderSP must preserved for slow path
1066
1067 Label slow_path;
1068 // If we need a safepoint check, generate full interpreter entry.
1069 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1070
1071 // We don't generate local frame and don't align stack because
1072 // we call stub code and there is no safepoint on this path.
1073
1074 // Load parameters
1075 const Register crc = c_rarg0; // crc
1076 const Register buf = c_rarg1; // source java byte array address
1077 const Register len = c_rarg2; // length
1078 const Register off = len; // offset (never overlaps with 'len')
1079
1080 // Arguments are reversed on java expression stack
1081 // Calculate address of start element
1082 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1083 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1084 __ ldrw(off, Address(esp, wordSize)); // offset
1085 __ add(buf, buf, off); // + offset
1086 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1087 } else {
1088 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1089 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1090 __ ldrw(off, Address(esp, wordSize)); // offset
1091 __ add(buf, buf, off); // + offset
1092 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1093 }
1094 // Can now load 'len' since we're finished with 'off'
1095 __ ldrw(len, Address(esp, 0x0)); // Length
1096
1097 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1098
1099 // We are frameless so we can just jump to the stub.
1100 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1101
1102 // generate a vanilla native entry as the slow path
1103 __ bind(slow_path);
1104 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1105 return entry;
1106 }
1107
1108 /**
1109 * Method entry for intrinsic-candidate (non-native) methods:
1110 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1111 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1112 * Unlike CRC32, CRC32C does not have any methods marked as native
1113 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1114 */
1115 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1116 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1117 address entry = __ pc();
1118
1119 // Prepare jump to stub using parameters from the stack
1120 const Register crc = c_rarg0; // initial crc
1121 const Register buf = c_rarg1; // source java byte array address
1122 const Register len = c_rarg2; // len argument to the kernel
1123
1124 const Register end = len; // index of last element to process
1125 const Register off = crc; // offset
1126
1127 __ ldrw(end, Address(esp)); // int end
1128 __ ldrw(off, Address(esp, wordSize)); // int offset
1129 __ sub(len, end, off);
1130 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1131 __ add(buf, buf, off); // + offset
1132 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1133 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1134 } else {
1135 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1136 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1137 }
1138
1139 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1140
1141 // Jump to the stub.
1142 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1143
1144 return entry;
1145 }
1146
1147 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1148 // See more discussion in stackOverflow.hpp.
1149
1150 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1151 const int page_size = (int)os::vm_page_size();
1152 const int n_shadow_pages = shadow_zone_size / page_size;
1153
1154 #ifdef ASSERT
1155 Label L_good_limit;
1156 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1157 __ cbnz(rscratch1, L_good_limit);
1158 __ stop("shadow zone safe limit is not initialized");
1159 __ bind(L_good_limit);
1160
1161 Label L_good_watermark;
1162 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1163 __ cbnz(rscratch1, L_good_watermark);
1164 __ stop("shadow zone growth watermark is not initialized");
1165 __ bind(L_good_watermark);
1166 #endif
1167
1168 Label L_done;
1169
1170 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1171 __ cmp(sp, rscratch1);
1172 __ br(Assembler::HI, L_done);
1173
1174 for (int p = 1; p <= n_shadow_pages; p++) {
1175 __ sub(rscratch2, sp, p*page_size);
1176 __ str(zr, Address(rscratch2));
1177 }
1178
1179 // Record the new watermark, but only if the update is above the safe limit.
1180 // Otherwise, the next time around the check above would pass the safe limit.
1181 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1182 __ cmp(sp, rscratch1);
1183 __ br(Assembler::LS, L_done);
1184 __ mov(rscratch1, sp);
1185 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1186
1187 __ bind(L_done);
1188 }
1189
1190 // Interpreter stub for calling a native method. (asm interpreter)
1191 // This sets up a somewhat different looking stack for calling the
1192 // native method than the typical interpreter frame setup.
1193 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1194 // determine code generation flags
1195 bool inc_counter = UseCompiler || CountCompiledCalls;
1196
1197 // r1: Method*
1198 // rscratch1: sender sp
1199
1200 address entry_point = __ pc();
1201
1202 const Address constMethod (rmethod, Method::const_offset());
1203 const Address access_flags (rmethod, Method::access_flags_offset());
1204 const Address size_of_parameters(r2, ConstMethod::
1205 size_of_parameters_offset());
1206
1207 // get parameter size (always needed)
1208 __ ldr(r2, constMethod);
1209 __ load_unsigned_short(r2, size_of_parameters);
1210
1211 // Native calls don't need the stack size check since they have no
1212 // expression stack and the arguments are already on the stack and
1213 // we only add a handful of words to the stack.
1214
1215 // rmethod: Method*
1216 // r2: size of parameters
1217 // rscratch1: sender sp
1218
1219 // for natives the size of locals is zero
1220
1221 // compute beginning of parameters (rlocals)
1222 __ add(rlocals, esp, r2, ext::uxtx, 3);
1223 __ add(rlocals, rlocals, -wordSize);
1224
1225 // Pull SP back to minimum size: this avoids holes in the stack
1226 __ andr(sp, esp, -16);
1227
1228 // initialize fixed part of activation frame
1229 generate_fixed_frame(true);
1230
1231 // make sure method is native & not abstract
1232 #ifdef ASSERT
1233 __ ldrh(r0, access_flags);
1234 {
1235 Label L;
1236 __ tst(r0, JVM_ACC_NATIVE);
1237 __ br(Assembler::NE, L);
1238 __ stop("tried to execute non-native method as native");
1239 __ bind(L);
1240 }
1241 {
1242 Label L;
1243 __ tst(r0, JVM_ACC_ABSTRACT);
1244 __ br(Assembler::EQ, L);
1245 __ stop("tried to execute abstract method in interpreter");
1246 __ bind(L);
1247 }
1248 #endif
1249
1250 // Since at this point in the method invocation the exception
1251 // handler would try to exit the monitor of synchronized methods
1252 // which hasn't been entered yet, we set the thread local variable
1253 // _do_not_unlock_if_synchronized to true. The remove_activation
1254 // will check this flag.
1255
1256 const Address do_not_unlock_if_synchronized(rthread,
1257 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1258 __ mov(rscratch2, true);
1259 __ strb(rscratch2, do_not_unlock_if_synchronized);
1260
1261 // increment invocation count & check for overflow
1262 Label invocation_counter_overflow;
1263 if (inc_counter) {
1264 generate_counter_incr(&invocation_counter_overflow);
1265 }
1266
1267 Label continue_after_compile;
1268 __ bind(continue_after_compile);
1269
1270 bang_stack_shadow_pages(true);
1271
1272 // reset the _do_not_unlock_if_synchronized flag
1273 __ strb(zr, do_not_unlock_if_synchronized);
1274
1275 // check for synchronized methods
1276 // Must happen AFTER invocation_counter check and stack overflow check,
1277 // so method is not locked if overflows.
1278 if (synchronized) {
1279 lock_method();
1280 } else {
1281 // no synchronization necessary
1282 #ifdef ASSERT
1283 {
1284 Label L;
1285 __ ldrh(r0, access_flags);
1286 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1287 __ br(Assembler::EQ, L);
1288 __ stop("method needs synchronization");
1289 __ bind(L);
1290 }
1291 #endif
1292 }
1293
1294 // start execution
1295 #ifdef ASSERT
1296 {
1297 Label L;
1298 const Address monitor_block_top(rfp,
1299 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1300 __ ldr(rscratch1, monitor_block_top);
1301 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1302 __ cmp(esp, rscratch1);
1303 __ br(Assembler::EQ, L);
1304 __ stop("broken stack frame setup in interpreter 1");
1305 __ bind(L);
1306 }
1307 #endif
1308
1309 // jvmti support
1310 __ notify_method_entry();
1311
1312 // work registers
1313 const Register t = r17;
1314 const Register result_handler = r19;
1315
1316 // allocate space for parameters
1317 __ ldr(t, Address(rmethod, Method::const_offset()));
1318 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1319
1320 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1321 __ andr(sp, rscratch1, -16);
1322 __ mov(esp, rscratch1);
1323
1324 // get signature handler
1325 {
1326 Label L;
1327 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1328 __ cbnz(t, L);
1329 __ call_VM(noreg,
1330 CAST_FROM_FN_PTR(address,
1331 InterpreterRuntime::prepare_native_call),
1332 rmethod);
1333 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1334 __ bind(L);
1335 }
1336
1337 // call signature handler
1338 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1339 "adjust this code");
1340 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1341 "adjust this code");
1342 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1343 "adjust this code");
1344
1345 // The generated handlers do not touch rmethod (the method).
1346 // However, large signatures cannot be cached and are generated
1347 // each time here. The slow-path generator can do a GC on return,
1348 // so we must reload it after the call.
1349 __ blr(t);
1350 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1351
1352
1353 // result handler is in r0
1354 // set result handler
1355 __ mov(result_handler, r0);
1356 // Save it in the frame in case of preemption; we cannot rely on callee saved registers.
1357 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize));
1358
1359 // pass mirror handle if static call
1360 {
1361 Label L;
1362 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1363 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1364 // get mirror
1365 __ load_mirror(t, rmethod, r10, rscratch2);
1366 // copy mirror into activation frame
1367 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1368 // pass handle to mirror
1369 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1370 __ bind(L);
1371 }
1372
1373 // get native function entry point in r10
1374 {
1375 Label L;
1376 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1377 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1378 __ lea(rscratch2, unsatisfied);
1379 __ ldr(rscratch2, rscratch2);
1380 __ cmp(r10, rscratch2);
1381 __ br(Assembler::NE, L);
1382 __ call_VM(noreg,
1383 CAST_FROM_FN_PTR(address,
1384 InterpreterRuntime::prepare_native_call),
1385 rmethod);
1386 __ get_method(rmethod);
1387 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1388 __ bind(L);
1389 }
1390
1391 // pass JNIEnv
1392 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1393
1394 // It is enough that the pc() points into the right code
1395 // segment. It does not have to be the correct return pc.
1396 // For convenience we use the pc we want to resume to in
1397 // case of preemption on Object.wait.
1398 Label native_return;
1399 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1400
1401 // change thread state
1402 #ifdef ASSERT
1403 {
1404 Label L;
1405 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1406 __ cmp(t, (u1)_thread_in_Java);
1407 __ br(Assembler::EQ, L);
1408 __ stop("Wrong thread state in native stub");
1409 __ bind(L);
1410 }
1411 #endif
1412
1413 // Change state to native
1414 __ mov(rscratch1, _thread_in_native);
1415 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1416 __ stlrw(rscratch1, rscratch2);
1417
1418 __ push_cont_fastpath();
1419
1420 // Call the native method.
1421 __ blr(r10);
1422
1423 __ pop_cont_fastpath();
1424
1425 __ get_method(rmethod);
1426 // result potentially in r0 or v0
1427
1428 // Restore cpu control state after JNI call
1429 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1430
1431 // make room for the pushes we're about to do
1432 __ sub(rscratch1, esp, 4 * wordSize);
1433 __ andr(sp, rscratch1, -16);
1434
1435 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1436 // in order to extract the result of a method call. If the order of these
1437 // pushes change or anything else is added to the stack then the code in
1438 // interpreter_frame_result must also change.
1439 __ push(dtos);
1440 __ push(ltos);
1441
1442 __ verify_sve_vector_length();
1443
1444 // change thread state
1445 __ mov(rscratch1, _thread_in_native_trans);
1446 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1447 __ stlrw(rscratch1, rscratch2);
1448
1449 // Force this write out before the read below
1450 if (!UseSystemMemoryBarrier) {
1451 __ dmb(Assembler::ISH);
1452 }
1453
1454 // check for safepoint operation in progress and/or pending suspend requests
1455 {
1456 Label L, Continue;
1457
1458 // No need for acquire as Java threads always disarm themselves.
1459 __ safepoint_poll(L, true /* at_return */, false /* 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 __ lea(rscratch2, RuntimeAddress(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 if (LockingMode != LM_LEGACY) {
1483 // Check preemption for Object.wait()
1484 Label not_preempted;
1485 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1486 __ cbz(rscratch1, not_preempted);
1487 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1488 __ br(rscratch1);
1489 __ bind(native_return);
1490 __ restore_after_resume(true /* is_native */);
1491 // reload result_handler
1492 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize));
1493 __ bind(not_preempted);
1494 } else {
1495 // any pc will do so just use this one for LM_LEGACY to keep code together.
1496 __ bind(native_return);
1497 }
1498
1499 // reset_last_Java_frame
1500 __ reset_last_Java_frame(true);
1501
1502 if (CheckJNICalls) {
1503 // clear_pending_jni_exception_check
1504 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1505 }
1506
1507 // reset handle block
1508 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1509 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1510
1511 // If result is an oop unbox and store it in frame where gc will see it
1512 // and result handler will pick it up
1513
1514 {
1515 Label no_oop;
1516 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1517 __ cmp(t, result_handler);
1518 __ br(Assembler::NE, no_oop);
1519 // Unbox oop result, e.g. JNIHandles::resolve result.
1520 __ pop(ltos);
1521 __ resolve_jobject(r0, t, rscratch2);
1522 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1523 // keep stack depth as expected by pushing oop which will eventually be discarded
1524 __ push(ltos);
1525 __ bind(no_oop);
1526 }
1527
1528 {
1529 Label no_reguard;
1530 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1531 __ ldrw(rscratch1, Address(rscratch1));
1532 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1533 __ br(Assembler::NE, no_reguard);
1534
1535 __ push_call_clobbered_registers();
1536 __ mov(c_rarg0, rthread);
1537 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1538 __ blr(rscratch2);
1539 __ pop_call_clobbered_registers();
1540
1541 __ bind(no_reguard);
1542 }
1543
1544 // The method register is junk from after the thread_in_native transition
1545 // until here. Also can't call_VM until the bcp has been
1546 // restored. Need bcp for throwing exception below so get it now.
1547 __ get_method(rmethod);
1548
1549 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1550 // rbcp == code_base()
1551 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1552 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1553 // handle exceptions (exception handling will handle unlocking!)
1554 {
1555 Label L;
1556 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1557 __ cbz(rscratch1, L);
1558 // Note: At some point we may want to unify this with the code
1559 // used in call_VM_base(); i.e., we should use the
1560 // StubRoutines::forward_exception code. For now this doesn't work
1561 // here because the rsp is not correctly set at this point.
1562 __ MacroAssembler::call_VM(noreg,
1563 CAST_FROM_FN_PTR(address,
1564 InterpreterRuntime::throw_pending_exception));
1565 __ should_not_reach_here();
1566 __ bind(L);
1567 }
1568
1569 // do unlocking if necessary
1570 {
1571 Label L;
1572 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1573 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1574 // the code below should be shared with interpreter macro
1575 // assembler implementation
1576 {
1577 Label unlock;
1578 // BasicObjectLock will be first in list, since this is a
1579 // synchronized method. However, need to check that the object
1580 // has not been unlocked by an explicit monitorexit bytecode.
1581
1582 // monitor expect in c_rarg1 for slow unlock path
1583 __ lea (c_rarg1, Address(rfp, // address of first monitor
1584 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1585 wordSize - sizeof(BasicObjectLock))));
1586
1587 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1588 __ cbnz(t, unlock);
1589
1590 // Entry already unlocked, need to throw exception
1591 __ MacroAssembler::call_VM(noreg,
1592 CAST_FROM_FN_PTR(address,
1593 InterpreterRuntime::throw_illegal_monitor_state_exception));
1594 __ should_not_reach_here();
1595
1596 __ bind(unlock);
1597 __ unlock_object(c_rarg1);
1598 }
1599 __ bind(L);
1600 }
1601
1602 // jvmti support
1603 // Note: This must happen _after_ handling/throwing any exceptions since
1604 // the exception handler code notifies the runtime of method exits
1605 // too. If this happens before, method entry/exit notifications are
1606 // not properly paired (was bug - gri 11/22/99).
1607 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1608
1609 // restore potential result in r0:d0, call result handler to
1610 // restore potential result in ST0 & handle result
1611
1612 __ pop(ltos);
1613 __ pop(dtos);
1614
1615 __ blr(result_handler);
1616
1617 // remove activation
1618 __ ldr(esp, Address(rfp,
1619 frame::interpreter_frame_sender_sp_offset *
1620 wordSize)); // get sender sp
1621 // remove frame anchor
1622 __ leave();
1623
1624 // restore sender sp
1625 __ mov(sp, esp);
1626
1627 __ ret(lr);
1628
1629 if (inc_counter) {
1630 // Handle overflow of counter and compile method
1631 __ bind(invocation_counter_overflow);
1632 generate_counter_overflow(continue_after_compile);
1633 }
1634
1635 return entry_point;
1636 }
1637
1638 //
1639 // Generic interpreted method entry to (asm) interpreter
1640 //
1641 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1642 // determine code generation flags
1643 bool inc_counter = UseCompiler || CountCompiledCalls;
1644
1645 // rscratch1: sender sp
1646 address entry_point = __ pc();
1647
1648 const Address constMethod(rmethod, Method::const_offset());
1649 const Address access_flags(rmethod, Method::access_flags_offset());
1650 const Address size_of_parameters(r3,
1651 ConstMethod::size_of_parameters_offset());
1652 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1653
1654 // get parameter size (always needed)
1655 // need to load the const method first
1656 __ ldr(r3, constMethod);
1657 __ load_unsigned_short(r2, size_of_parameters);
1658
1659 // r2: size of parameters
1660
1661 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1662 __ sub(r3, r3, r2); // r3 = no. of additional locals
1663
1664 // see if we've got enough room on the stack for locals plus overhead.
1665 generate_stack_overflow_check();
1666
1667 // compute beginning of parameters (rlocals)
1668 __ add(rlocals, esp, r2, ext::uxtx, 3);
1669 __ sub(rlocals, rlocals, wordSize);
1670
1671 __ mov(rscratch1, esp);
1672
1673 // r3 - # of additional locals
1674 // allocate space for locals
1675 // explicitly initialize locals
1676 // Initializing memory allocated for locals in the same direction as
1677 // the stack grows to ensure page initialization order according
1678 // to windows-aarch64 stack page growth requirement (see
1679 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1680 {
1681 Label exit, loop;
1682 __ ands(zr, r3, r3);
1683 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1684 __ bind(loop);
1685 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1686 __ sub(r3, r3, 1); // until everything initialized
1687 __ cbnz(r3, loop);
1688 __ bind(exit);
1689 }
1690
1691 // Padding between locals and fixed part of activation frame to ensure
1692 // SP is always 16-byte aligned.
1693 __ andr(sp, rscratch1, -16);
1694
1695 // And the base dispatch table
1696 __ get_dispatch();
1697
1698 // initialize fixed part of activation frame
1699 generate_fixed_frame(false);
1700
1701 // make sure method is not native & not abstract
1702 #ifdef ASSERT
1703 __ ldrh(r0, access_flags);
1704 {
1705 Label L;
1706 __ tst(r0, JVM_ACC_NATIVE);
1707 __ br(Assembler::EQ, L);
1708 __ stop("tried to execute native method as non-native");
1709 __ bind(L);
1710 }
1711 {
1712 Label L;
1713 __ tst(r0, JVM_ACC_ABSTRACT);
1714 __ br(Assembler::EQ, L);
1715 __ stop("tried to execute abstract method in interpreter");
1716 __ bind(L);
1717 }
1718 #endif
1719
1720 // Since at this point in the method invocation the exception
1721 // handler would try to exit the monitor of synchronized methods
1722 // which hasn't been entered yet, we set the thread local variable
1723 // _do_not_unlock_if_synchronized to true. The remove_activation
1724 // will check this flag.
1725
1726 const Address do_not_unlock_if_synchronized(rthread,
1727 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1728 __ mov(rscratch2, true);
1729 __ strb(rscratch2, do_not_unlock_if_synchronized);
1730
1731 Register mdp = r3;
1732 __ profile_parameters_type(mdp, r1, r2);
1733
1734 // increment invocation count & check for overflow
1735 Label invocation_counter_overflow;
1736 if (inc_counter) {
1737 generate_counter_incr(&invocation_counter_overflow);
1738 }
1739
1740 Label continue_after_compile;
1741 __ bind(continue_after_compile);
1742
1743 bang_stack_shadow_pages(false);
1744
1745 // reset the _do_not_unlock_if_synchronized flag
1746 __ strb(zr, do_not_unlock_if_synchronized);
1747
1748 // check for synchronized methods
1749 // Must happen AFTER invocation_counter check and stack overflow check,
1750 // so method is not locked if overflows.
1751 if (synchronized) {
1752 // Allocate monitor and lock method
1753 lock_method();
1754 } else {
1755 // no synchronization necessary
1756 #ifdef ASSERT
1757 {
1758 Label L;
1759 __ ldrh(r0, access_flags);
1760 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1761 __ br(Assembler::EQ, L);
1762 __ stop("method needs synchronization");
1763 __ bind(L);
1764 }
1765 #endif
1766 }
1767
1768 // start execution
1769 #ifdef ASSERT
1770 {
1771 Label L;
1772 const Address monitor_block_top (rfp,
1773 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1774 __ ldr(rscratch1, monitor_block_top);
1775 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1776 __ cmp(esp, rscratch1);
1777 __ br(Assembler::EQ, L);
1778 __ stop("broken stack frame setup in interpreter 2");
1779 __ bind(L);
1780 }
1781 #endif
1782
1783 // jvmti support
1784 __ notify_method_entry();
1785
1786 __ dispatch_next(vtos);
1787
1788 // invocation counter overflow
1789 if (inc_counter) {
1790 // Handle overflow of counter and compile method
1791 __ bind(invocation_counter_overflow);
1792 generate_counter_overflow(continue_after_compile);
1793 }
1794
1795 return entry_point;
1796 }
1797
1798 // Method entry for java.lang.Thread.currentThread
1799 address TemplateInterpreterGenerator::generate_currentThread() {
1800 address entry_point = __ pc();
1801
1802 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1803 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1804 __ ret(lr);
1805
1806 return entry_point;
1807 }
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_oop_offset()));
1988 // remove the activation (without doing throws on illegalMonitorExceptions)
1989 __ remove_activation(vtos, false, true, false);
1990 // restore exception
1991 __ get_vm_result_oop(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(); // atos entry point
2061 __ push_ptr();
2062 __ b(L);
2063 fep = __ pc(); // ftos entry point
2064 __ push_f();
2065 __ b(L);
2066 dep = __ pc(); // dtos entry point
2067 __ push_d();
2068 __ b(L);
2069 lep = __ pc(); // ltos entry point
2070 __ push_l();
2071 __ b(L);
2072 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
2073 __ push_i();
2074 vep = __ pc(); // vtos entry point
2075 __ bind(L);
2076 generate_and_dispatch(t);
2077 }
2078
2079 //-----------------------------------------------------------------------------
2080
2081 // Non-product code
2082 #ifndef PRODUCT
2083 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2084 address entry = __ pc();
2085
2086 __ protect_return_address();
2087 __ push(lr);
2088 __ push(state);
2089 __ push(RegSet::range(r0, r15), sp);
2090 __ mov(c_rarg2, r0); // Pass itos
2091 __ call_VM(noreg,
2092 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2093 c_rarg1, c_rarg2, c_rarg3);
2094 __ pop(RegSet::range(r0, r15), sp);
2095 __ pop(state);
2096 __ pop(lr);
2097 __ authenticate_return_address();
2098 __ ret(lr); // return from result handler
2099
2100 return entry;
2101 }
2102
2103 void TemplateInterpreterGenerator::count_bytecode() {
2104 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2105 __ atomic_add(noreg, 1, r10);
2106 }
2107
2108 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2109 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2110 __ atomic_addw(noreg, 1, r10);
2111 }
2112
2113 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2114 // Calculate new index for counter:
2115 // _index = (_index >> log2_number_of_codes) |
2116 // (bytecode << log2_number_of_codes);
2117 Register index_addr = rscratch1;
2118 Register index = rscratch2;
2119 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2120 __ ldrw(index, index_addr);
2121 __ mov(r10,
2122 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2123 __ orrw(index, r10, index, Assembler::LSR,
2124 BytecodePairHistogram::log2_number_of_codes);
2125 __ strw(index, index_addr);
2126
2127 // Bump bucket contents:
2128 // _counters[_index] ++;
2129 Register counter_addr = rscratch1;
2130 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2131 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2132 __ atomic_addw(noreg, 1, counter_addr);
2133 }
2134
2135 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2136 // Call a little run-time stub to avoid blow-up for each bytecode.
2137 // The run-time runtime saves the right registers, depending on
2138 // the tosca in-state for the given template.
2139
2140 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2141 "entry must have been generated");
2142 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2143 __ reinit_heapbase();
2144 }
2145
2146
2147 void TemplateInterpreterGenerator::stop_interpreter_at() {
2148 Label L;
2149 __ push(rscratch1);
2150 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2151 __ ldr(rscratch1, Address(rscratch1));
2152 __ mov(rscratch2, StopInterpreterAt);
2153 __ cmp(rscratch1, rscratch2);
2154 __ br(Assembler::NE, L);
2155 __ brk(0);
2156 __ bind(L);
2157 __ pop(rscratch1);
2158 }
2159
2160 #endif // !PRODUCT