1 /*
2 * Copyright (c) 2003, 2026, 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 // handle exceptions
534 {
535 Label L;
536 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
537 __ cbz(rscratch1, L);
538 __ call_VM(noreg,
539 CAST_FROM_FN_PTR(address,
540 InterpreterRuntime::throw_pending_exception));
541 __ should_not_reach_here();
542 __ bind(L);
543 }
544
545 if (continuation == nullptr) {
546 __ dispatch_next(state, step);
547 } else {
548 __ jump_to_entry(continuation);
549 }
550 return entry;
551 }
552
553 address TemplateInterpreterGenerator::generate_result_handler_for(
554 BasicType type) {
555 address entry = __ pc();
556 switch (type) {
557 case T_BOOLEAN: __ c2bool(r0); break;
558 case T_CHAR : __ uxth(r0, r0); break;
559 case T_BYTE : __ sxtb(r0, r0); break;
560 case T_SHORT : __ sxth(r0, r0); break;
561 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
562 case T_LONG : /* nothing to do */ break;
563 case T_VOID : /* nothing to do */ break;
564 case T_FLOAT : /* nothing to do */ break;
565 case T_DOUBLE : /* nothing to do */ break;
566 case T_OBJECT :
567 // retrieve result from frame
568 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
569 // and verify it
570 __ verify_oop(r0);
571 break;
572 default : ShouldNotReachHere();
573 }
574 __ ret(lr); // return from result handler
575 return entry;
576 }
577
578 address TemplateInterpreterGenerator::generate_safept_entry_for(
579 TosState state,
580 address runtime_entry) {
581 address entry = __ pc();
582 __ push(state);
583 __ push_cont_fastpath(rthread);
584 __ call_VM(noreg, runtime_entry);
585 __ pop_cont_fastpath(rthread);
586 __ membar(Assembler::AnyAny);
587 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
588 return entry;
589 }
590
591 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
592 if (!Continuations::enabled()) return nullptr;
593 address start = __ pc();
594
595 __ restore_bcp();
596 __ restore_locals();
597
598 // Restore constant pool cache
599 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
600
601 // Restore Java expression stack pointer
602 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
603 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
604 // and null it as marker that esp is now tos until next java call
605 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
606
607 // Restore machine SP
608 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
609 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord)));
610
611 // Restore method
612 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
613
614 // Restore dispatch
615 uint64_t offset;
616 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
617 __ add(rdispatch, rdispatch, offset);
618
619 __ ret(lr);
620
621 return start;
622 }
623
624
625 // Helpers for commoning out cases in the various type of method entries.
626 //
627
628
629 // increment invocation count & check for overflow
630 //
631 // Note: checking for negative value instead of overflow
632 // so we have a 'sticky' overflow test
633 //
634 // rmethod: method
635 //
636 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
637 Label done;
638 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
639 int increment = InvocationCounter::count_increment;
640 Label no_mdo;
641 if (ProfileInterpreter) {
642 // Are we profiling?
643 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
644 __ cbz(r0, no_mdo);
645 // Increment counter in the MDO
646 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
647 in_bytes(InvocationCounter::counter_offset()));
648 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
649 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
650 __ b(done);
651 }
652 __ bind(no_mdo);
653 // Increment counter in MethodCounters
654 const Address invocation_counter(rscratch2,
655 MethodCounters::invocation_counter_offset() +
656 InvocationCounter::counter_offset());
657 __ get_method_counters(rmethod, rscratch2, done);
658 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
659 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
660 __ bind(done);
661 }
662
663 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
664
665 // Asm interpreter on entry
666 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
667 // Everything as it was on entry
668
669 // InterpreterRuntime::frequency_counter_overflow takes two
670 // arguments, the first (thread) is passed by call_VM, the second
671 // indicates if the counter overflow occurs at a backwards branch
672 // (null bcp). We pass zero for it. The call returns the address
673 // of the verified entry point for the method or null if the
674 // compilation did not complete (either went background or bailed
675 // out).
676 __ mov(c_rarg1, 0);
677 __ call_VM(noreg,
678 CAST_FROM_FN_PTR(address,
679 InterpreterRuntime::frequency_counter_overflow),
680 c_rarg1);
681
682 __ b(do_continue);
683 }
684
685 // See if we've got enough room on the stack for locals plus overhead
686 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
687 // without going through the signal handler, i.e., reserved and yellow zones
688 // will not be made usable. The shadow zone must suffice to handle the
689 // overflow.
690 // The expression stack grows down incrementally, so the normal guard
691 // page mechanism will work for that.
692 //
693 // NOTE: Since the additional locals are also always pushed (wasn't
694 // obvious in generate_method_entry) so the guard should work for them
695 // too.
696 //
697 // Args:
698 // r3: number of additional locals this frame needs (what we must check)
699 // rmethod: Method*
700 //
701 // Kills:
702 // r0
703 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
704
705 // monitor entry size: see picture of stack set
706 // (generate_method_entry) and frame_amd64.hpp
707 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
708
709 // total overhead size: entry_size + (saved rbp through expr stack
710 // bottom). be sure to change this if you add/subtract anything
711 // to/from the overhead area
712 const int overhead_size =
713 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
714
715 const size_t page_size = os::vm_page_size();
716
717 Label after_frame_check;
718
719 // see if the frame is greater than one page in size. If so,
720 // then we need to verify there is enough stack space remaining
721 // for the additional locals.
722 //
723 // Note that we use SUBS rather than CMP here because the immediate
724 // field of this instruction may overflow. SUBS can cope with this
725 // because it is a macro that will expand to some number of MOV
726 // instructions and a register operation.
727 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
728 __ br(Assembler::LS, after_frame_check);
729
730 // compute rsp as if this were going to be the last frame on
731 // the stack before the red zone
732
733 // locals + overhead, in bytes
734 __ mov(r0, overhead_size);
735 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
736
737 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
738 __ ldr(rscratch1, stack_limit);
739
740 #ifdef ASSERT
741 Label limit_okay;
742 // Verify that thread stack limit is non-zero.
743 __ cbnz(rscratch1, limit_okay);
744 __ stop("stack overflow limit is zero");
745 __ bind(limit_okay);
746 #endif
747
748 // Add stack limit to locals.
749 __ add(r0, r0, rscratch1);
750
751 // Check against the current stack bottom.
752 __ cmp(sp, r0);
753 __ br(Assembler::HI, after_frame_check);
754
755 // Remove the incoming args, peeling the machine SP back to where it
756 // was in the caller. This is not strictly necessary, but unless we
757 // do so the stack frame may have a garbage FP; this ensures a
758 // correct call stack that we can always unwind. The ANDR should be
759 // unnecessary because the sender SP in r19 is always aligned, but
760 // it doesn't hurt.
761 __ andr(sp, r19_sender_sp, -16);
762
763 // Note: the restored frame is not necessarily interpreted.
764 // Use the shared runtime version of the StackOverflowError.
765 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
766 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
767
768 // all done with frame size check
769 __ bind(after_frame_check);
770 }
771
772 // Allocate monitor and lock method (asm interpreter)
773 //
774 // Args:
775 // rmethod: Method*
776 // rlocals: locals
777 //
778 // Kills:
779 // r0
780 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
781 // rscratch1, rscratch2 (scratch regs)
782 void TemplateInterpreterGenerator::lock_method() {
783 // synchronize method
784 const Address access_flags(rmethod, Method::access_flags_offset());
785 const Address monitor_block_top(
786 rfp,
787 frame::interpreter_frame_monitor_block_top_offset * wordSize);
788 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
789
790 #ifdef ASSERT
791 {
792 Label L;
793 __ ldrh(r0, access_flags);
794 __ tst(r0, JVM_ACC_SYNCHRONIZED);
795 __ br(Assembler::NE, L);
796 __ stop("method doesn't need synchronization");
797 __ bind(L);
798 }
799 #endif // ASSERT
800
801 // get synchronization object
802 {
803 Label done;
804 __ ldrh(r0, access_flags);
805 __ tst(r0, JVM_ACC_STATIC);
806 // get receiver (assume this is frequent case)
807 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
808 __ br(Assembler::EQ, done);
809 __ load_mirror(r0, rmethod, r5, rscratch2);
810
811 #ifdef ASSERT
812 {
813 Label L;
814 __ cbnz(r0, L);
815 __ stop("synchronization object is null");
816 __ bind(L);
817 }
818 #endif // ASSERT
819
820 __ bind(done);
821 }
822
823 // add space for monitor & lock
824 __ check_extended_sp();
825 __ sub(sp, sp, entry_size); // add space for a monitor entry
826 __ sub(esp, esp, entry_size);
827 __ sub(rscratch1, sp, rfp);
828 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
829 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
830 __ sub(rscratch1, esp, rfp);
831 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
832 __ str(rscratch1, monitor_block_top); // set new monitor block top
833
834 // store object
835 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
836 __ mov(c_rarg1, esp); // object address
837 __ lock_object(c_rarg1);
838 }
839
840 // Generate a fixed interpreter frame. This is identical setup for
841 // interpreted methods and for native methods hence the shared code.
842 //
843 // Args:
844 // lr: return address
845 // rmethod: Method*
846 // rlocals: pointer to locals
847 // rcpool: cp cache
848 // stack_pointer: previous sp
849 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
850 // Save ConstMethod* in r5_const_method for later use to avoid loading multiple times
851 Register r5_const_method = r5;
852 __ ldr(r5_const_method, Address(rmethod, Method::const_offset()));
853
854 // initialize fixed part of activation frame
855 if (native_call) {
856 __ sub(esp, sp, 14 * wordSize);
857 __ mov(rbcp, zr);
858 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
859 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
860 // add 2 zero-initialized slots for native calls
861 __ stp(zr, zr, Address(sp, 12 * wordSize));
862 } else {
863 __ sub(esp, sp, 12 * wordSize);
864 __ add(rbcp, r5_const_method, in_bytes(ConstMethod::codes_offset())); // get codebase
865 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
866 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
867 }
868
869 if (ProfileInterpreter) {
870 Label method_data_continue;
871 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
872 __ cbz(rscratch1, method_data_continue);
873 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
874 __ bind(method_data_continue);
875 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
876 } else {
877 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
878 }
879
880 __ protect_return_address();
881 __ stp(rfp, lr, Address(sp, 10 * wordSize));
882 __ lea(rfp, Address(sp, 10 * wordSize));
883
884 // Save ConstantPool* in r11_constants for later use to avoid loading multiple times
885 Register r11_constants = r11;
886 __ ldr(r11_constants, Address(r5_const_method, ConstMethod::constants_offset()));
887 __ ldr(rcpool, Address(r11_constants, ConstantPool::cache_offset()));
888 __ sub(rscratch1, rlocals, rfp);
889 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
890 // Store relativized rlocals, see frame::interpreter_frame_locals().
891 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
892
893 // set sender sp
894 // leave last_sp as null
895 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
896
897 // Get mirror. Resolve ConstantPool* -> InstanceKlass* -> Java mirror.
898 __ ldr(r10, Address(r11_constants, ConstantPool::pool_holder_offset()));
899 __ ldr(r10, Address(r10, in_bytes(Klass::java_mirror_offset())));
900 __ resolve_oop_handle(r10, rscratch1, rscratch2);
901 if (! native_call) {
902 __ ldrh(rscratch1, Address(r5_const_method, ConstMethod::max_stack_offset()));
903 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
904 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
905 __ andr(rscratch1, rscratch1, -16);
906 __ sub(rscratch2, rscratch1, rfp);
907 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
908 // Store extended SP and mirror
909 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
910 // Move SP out of the way
911 __ mov(sp, rscratch1);
912 } else {
913 // Make sure there is room for the exception oop pushed in case method throws
914 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
915 __ sub(rscratch1, sp, 2 * wordSize);
916 __ sub(rscratch2, rscratch1, rfp);
917 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
918 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
919 __ mov(sp, rscratch1);
920 }
921 }
922
923 // End of helpers
924
925 // Various method entries
926 //------------------------------------------------------------------------------------------------------------------------
927 //
928 //
929
930 // Method entry for java.lang.ref.Reference.get.
931 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
932 // Code: _aload_0, _getfield, _areturn
933 // parameter size = 1
934 //
935 // The code that gets generated by this routine is split into 2 parts:
936 // 1. The "intrinsified" code for G1 (or any SATB based GC),
937 // 2. The slow path - which is an expansion of the regular method entry.
938 //
939 // Notes:-
940 // * In the G1 code we do not check whether we need to block for
941 // a safepoint. If G1 is enabled then we must execute the specialized
942 // code for Reference.get (except when the Reference object is null)
943 // so that we can log the value in the referent field with an SATB
944 // update buffer.
945 // If the code for the getfield template is modified so that the
946 // G1 pre-barrier code is executed when the current method is
947 // Reference.get() then going through the normal method entry
948 // will be fine.
949 // * The G1 code can, however, check the receiver object (the instance
950 // of java.lang.Reference) and jump to the slow path if null. If the
951 // Reference object is null then we obviously cannot fetch the referent
952 // and so we don't need to call the G1 pre-barrier. Thus we can use the
953 // regular method entry code to generate the NPE.
954 //
955 // This code is based on generate_accessor_entry.
956 //
957 // rmethod: Method*
958 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
959
960 // LR is live. It must be saved around calls.
961
962 address entry = __ pc();
963
964 const int referent_offset = java_lang_ref_Reference::referent_offset();
965
966 Label slow_path;
967 const Register local_0 = c_rarg0;
968 // Check if local 0 != null
969 // If the receiver is null then it is OK to jump to the slow path.
970 __ ldr(local_0, Address(esp, 0));
971 __ cbz(local_0, slow_path);
972
973 // Load the value of the referent field.
974 const Address field_address(local_0, referent_offset);
975 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
976 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
977
978 // areturn
979 __ andr(sp, r19_sender_sp, -16); // done with stack
980 __ ret(lr);
981
982 // generate a vanilla interpreter entry as the slow path
983 __ bind(slow_path);
984 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
985 return entry;
986
987 }
988
989 /**
990 * Method entry for static native methods:
991 * int java.util.zip.CRC32.update(int crc, int b)
992 */
993 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
994 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
995 address entry = __ pc();
996
997 // rmethod: Method*
998 // r19_sender_sp: senderSP must preserved for slow path
999 // esp: args
1000
1001 Label slow_path;
1002 // If we need a safepoint check, generate full interpreter entry.
1003 __ safepoint_poll(slow_path, false /* at_return */, false /* in_nmethod */);
1004
1005 // We don't generate local frame and don't align stack because
1006 // we call stub code and there is no safepoint on this path.
1007
1008 // Load parameters
1009 const Register crc = c_rarg0; // crc
1010 const Register val = c_rarg1; // source java byte value
1011 const Register tbl = c_rarg2; // scratch
1012
1013 // Arguments are reversed on java expression stack
1014 __ ldrw(val, Address(esp, 0)); // byte value
1015 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1016
1017 uint64_t offset;
1018 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1019 __ add(tbl, tbl, offset);
1020
1021 __ mvnw(crc, crc); // ~crc
1022 __ update_byte_crc32(crc, val, tbl);
1023 __ mvnw(crc, crc); // ~crc
1024
1025 // result in c_rarg0
1026
1027 __ andr(sp, r19_sender_sp, -16);
1028 __ ret(lr);
1029
1030 // generate a vanilla native entry as the slow path
1031 __ bind(slow_path);
1032 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1033 return entry;
1034 }
1035
1036 /**
1037 * Method entry for static native methods:
1038 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1039 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1040 */
1041 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1042 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1043 address entry = __ pc();
1044
1045 // rmethod,: Method*
1046 // r19_sender_sp: senderSP must preserved for slow path
1047
1048 Label slow_path;
1049 // If we need a safepoint check, generate full interpreter entry.
1050 __ safepoint_poll(slow_path, false /* at_return */, false /* in_nmethod */);
1051
1052 // We don't generate local frame and don't align stack because
1053 // we call stub code and there is no safepoint on this path.
1054
1055 // Load parameters
1056 const Register crc = c_rarg0; // crc
1057 const Register buf = c_rarg1; // source java byte array address
1058 const Register len = c_rarg2; // length
1059 const Register off = len; // offset (never overlaps with 'len')
1060
1061 // Arguments are reversed on java expression stack
1062 // Calculate address of start element
1063 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1064 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1065 __ ldrw(off, Address(esp, wordSize)); // offset
1066 __ add(buf, buf, off); // + offset
1067 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1068 } else {
1069 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1070 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1071 __ ldrw(off, Address(esp, wordSize)); // offset
1072 __ add(buf, buf, off); // + offset
1073 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1074 }
1075 // Can now load 'len' since we're finished with 'off'
1076 __ ldrw(len, Address(esp, 0x0)); // Length
1077
1078 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1079
1080 // We are frameless so we can just jump to the stub.
1081 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1082
1083 // generate a vanilla native entry as the slow path
1084 __ bind(slow_path);
1085 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1086 return entry;
1087 }
1088
1089 /**
1090 * Method entry for intrinsic-candidate (non-native) methods:
1091 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1092 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1093 * Unlike CRC32, CRC32C does not have any methods marked as native
1094 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1095 */
1096 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1097 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1098 address entry = __ pc();
1099
1100 // Prepare jump to stub using parameters from the stack
1101 const Register crc = c_rarg0; // initial crc
1102 const Register buf = c_rarg1; // source java byte array address
1103 const Register len = c_rarg2; // len argument to the kernel
1104
1105 const Register end = len; // index of last element to process
1106 const Register off = crc; // offset
1107
1108 __ ldrw(end, Address(esp)); // int end
1109 __ ldrw(off, Address(esp, wordSize)); // int offset
1110 __ sub(len, end, off);
1111 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1112 __ add(buf, buf, off); // + offset
1113 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1114 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1115 } else {
1116 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1117 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1118 }
1119
1120 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1121
1122 // Jump to the stub.
1123 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1124
1125 return entry;
1126 }
1127
1128 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1129 // See more discussion in stackOverflow.hpp.
1130
1131 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1132 const int page_size = (int)os::vm_page_size();
1133 const int n_shadow_pages = shadow_zone_size / page_size;
1134
1135 #ifdef ASSERT
1136 Label L_good_limit;
1137 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1138 __ cbnz(rscratch1, L_good_limit);
1139 __ stop("shadow zone safe limit is not initialized");
1140 __ bind(L_good_limit);
1141
1142 Label L_good_watermark;
1143 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1144 __ cbnz(rscratch1, L_good_watermark);
1145 __ stop("shadow zone growth watermark is not initialized");
1146 __ bind(L_good_watermark);
1147 #endif
1148
1149 Label L_done;
1150
1151 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1152 __ cmp(sp, rscratch1);
1153 __ br(Assembler::HI, L_done);
1154
1155 for (int p = 1; p <= n_shadow_pages; p++) {
1156 __ sub(rscratch2, sp, p*page_size);
1157 __ str(zr, Address(rscratch2));
1158 }
1159
1160 // Record the new watermark, but only if the update is above the safe limit.
1161 // Otherwise, the next time around the check above would pass the safe limit.
1162 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1163 __ cmp(sp, rscratch1);
1164 __ br(Assembler::LS, L_done);
1165 __ mov(rscratch1, sp);
1166 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1167
1168 __ bind(L_done);
1169 }
1170
1171 // Interpreter stub for calling a native method. (asm interpreter)
1172 // This sets up a somewhat different looking stack for calling the
1173 // native method than the typical interpreter frame setup.
1174 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1175 // determine code generation flags
1176 bool inc_counter = UseCompiler || CountCompiledCalls;
1177
1178 // r1: Method*
1179 // rscratch1: sender sp
1180
1181 address entry_point = __ pc();
1182
1183 const Address constMethod (rmethod, Method::const_offset());
1184 const Address access_flags (rmethod, Method::access_flags_offset());
1185 const Address size_of_parameters(r2, ConstMethod::
1186 size_of_parameters_offset());
1187
1188 // get parameter size (always needed)
1189 __ ldr(r2, constMethod);
1190 __ load_unsigned_short(r2, size_of_parameters);
1191
1192 // Native calls don't need the stack size check since they have no
1193 // expression stack and the arguments are already on the stack and
1194 // we only add a handful of words to the stack.
1195
1196 // rmethod: Method*
1197 // r2: size of parameters
1198 // rscratch1: sender sp
1199
1200 // for natives the size of locals is zero
1201
1202 // compute beginning of parameters (rlocals)
1203 __ add(rlocals, esp, r2, ext::uxtx, 3);
1204 __ add(rlocals, rlocals, -wordSize);
1205
1206 // Pull SP back to minimum size: this avoids holes in the stack
1207 __ andr(sp, esp, -16);
1208
1209 // initialize fixed part of activation frame
1210 generate_fixed_frame(true);
1211
1212 // make sure method is native & not abstract
1213 #ifdef ASSERT
1214 __ ldrh(r0, access_flags);
1215 {
1216 Label L;
1217 __ tst(r0, JVM_ACC_NATIVE);
1218 __ br(Assembler::NE, L);
1219 __ stop("tried to execute non-native method as native");
1220 __ bind(L);
1221 }
1222 {
1223 Label L;
1224 __ tst(r0, JVM_ACC_ABSTRACT);
1225 __ br(Assembler::EQ, L);
1226 __ stop("tried to execute abstract method in interpreter");
1227 __ bind(L);
1228 }
1229 #endif
1230
1231 // Since at this point in the method invocation the exception
1232 // handler would try to exit the monitor of synchronized methods
1233 // which hasn't been entered yet, we set the thread local variable
1234 // _do_not_unlock_if_synchronized to true. The remove_activation
1235 // will check this flag.
1236
1237 const Address do_not_unlock_if_synchronized(rthread,
1238 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1239 __ mov(rscratch2, true);
1240 __ strb(rscratch2, do_not_unlock_if_synchronized);
1241
1242 // increment invocation count & check for overflow
1243 Label invocation_counter_overflow;
1244 if (inc_counter) {
1245 generate_counter_incr(&invocation_counter_overflow);
1246 }
1247
1248 Label continue_after_compile;
1249 __ bind(continue_after_compile);
1250
1251 bang_stack_shadow_pages(true);
1252
1253 // reset the _do_not_unlock_if_synchronized flag
1254 __ strb(zr, do_not_unlock_if_synchronized);
1255
1256 // check for synchronized methods
1257 // Must happen AFTER invocation_counter check and stack overflow check,
1258 // so method is not locked if overflows.
1259 if (synchronized) {
1260 lock_method();
1261 } else {
1262 // no synchronization necessary
1263 #ifdef ASSERT
1264 {
1265 Label L;
1266 __ ldrh(r0, access_flags);
1267 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1268 __ br(Assembler::EQ, L);
1269 __ stop("method needs synchronization");
1270 __ bind(L);
1271 }
1272 #endif
1273 }
1274
1275 // start execution
1276 #ifdef ASSERT
1277 {
1278 Label L;
1279 const Address monitor_block_top(rfp,
1280 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1281 __ ldr(rscratch1, monitor_block_top);
1282 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1283 __ cmp(esp, rscratch1);
1284 __ br(Assembler::EQ, L);
1285 __ stop("broken stack frame setup in interpreter 1");
1286 __ bind(L);
1287 }
1288 #endif
1289
1290 // jvmti support
1291 __ notify_method_entry();
1292
1293 // work registers
1294 const Register t = r17;
1295 const Register result_handler = r19;
1296
1297 // allocate space for parameters
1298 __ ldr(t, Address(rmethod, Method::const_offset()));
1299 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1300
1301 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1302 __ andr(sp, rscratch1, -16);
1303 __ mov(esp, rscratch1);
1304
1305 // get signature handler
1306 {
1307 Label L;
1308 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1309 __ cbnz(t, L);
1310 __ call_VM(noreg,
1311 CAST_FROM_FN_PTR(address,
1312 InterpreterRuntime::prepare_native_call),
1313 rmethod);
1314 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1315 __ bind(L);
1316 }
1317
1318 // call signature handler
1319 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1320 "adjust this code");
1321 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1322 "adjust this code");
1323 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1324 "adjust this code");
1325
1326 // The generated handlers do not touch rmethod (the method).
1327 // However, large signatures cannot be cached and are generated
1328 // each time here. The slow-path generator can do a GC on return,
1329 // so we must reload it after the call.
1330 __ blr(t);
1331 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1332
1333
1334 // result handler is in r0
1335 // set result handler
1336 __ mov(result_handler, r0);
1337 // Save it in the frame in case of preemption; we cannot rely on callee saved registers.
1338 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize));
1339
1340 // pass mirror handle if static call
1341 {
1342 Label L;
1343 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1344 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1345 // get mirror
1346 __ load_mirror(t, rmethod, r10, rscratch2);
1347 // copy mirror into activation frame
1348 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1349 // pass handle to mirror
1350 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1351 __ bind(L);
1352 }
1353
1354 // get native function entry point in r10
1355 {
1356 Label L;
1357 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1358 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1359 __ lea(rscratch2, unsatisfied);
1360 __ cmp(r10, rscratch2);
1361 __ br(Assembler::NE, L);
1362 __ call_VM(noreg,
1363 CAST_FROM_FN_PTR(address,
1364 InterpreterRuntime::prepare_native_call),
1365 rmethod);
1366 __ get_method(rmethod);
1367 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1368 __ bind(L);
1369 }
1370
1371 // pass JNIEnv
1372 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1373
1374 // It is enough that the pc() points into the right code
1375 // segment. It does not have to be the correct return pc.
1376 // For convenience we use the pc we want to resume to in
1377 // case of preemption on Object.wait.
1378 Label native_return;
1379 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1380
1381 // change thread state
1382 #ifdef ASSERT
1383 {
1384 Label L;
1385 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1386 __ cmp(t, (u1)_thread_in_Java);
1387 __ br(Assembler::EQ, L);
1388 __ stop("Wrong thread state in native stub");
1389 __ bind(L);
1390 }
1391 #endif
1392
1393 // Change state to native
1394 __ mov(rscratch1, _thread_in_native);
1395 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1396 __ stlrw(rscratch1, rscratch2);
1397
1398 __ push_cont_fastpath();
1399
1400 // Call the native method.
1401 __ blr(r10);
1402
1403 __ pop_cont_fastpath();
1404
1405 __ get_method(rmethod);
1406 // result potentially in r0 or v0
1407
1408 // Restore cpu control state after JNI call
1409 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1410
1411 // make room for the pushes we're about to do
1412 __ sub(rscratch1, esp, 4 * wordSize);
1413 __ andr(sp, rscratch1, -16);
1414
1415 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1416 // in order to extract the result of a method call. If the order of these
1417 // pushes change or anything else is added to the stack then the code in
1418 // interpreter_frame_result must also change.
1419 __ push(dtos);
1420 __ push(ltos);
1421
1422 __ verify_sve_vector_length();
1423
1424 // change thread state
1425 __ mov(rscratch1, _thread_in_native_trans);
1426 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1427 __ stlrw(rscratch1, rscratch2);
1428
1429 // Force this write out before the read below
1430 if (!UseSystemMemoryBarrier) {
1431 __ dmb(Assembler::ISH);
1432 }
1433
1434 // check for safepoint operation in progress and/or pending suspend requests
1435 {
1436 Label L, Continue;
1437
1438 // No need for acquire as Java threads always disarm themselves.
1439 __ safepoint_poll(L, true /* at_return */, false /* in_nmethod */);
1440 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1441 __ cbz(rscratch2, Continue);
1442 __ bind(L);
1443
1444 // Don't use call_VM as it will see a possible pending exception
1445 // and forward it and never return here preventing us from
1446 // clearing _last_native_pc down below. So we do a runtime call by
1447 // hand.
1448 //
1449 __ mov(c_rarg0, rthread);
1450 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1451 __ blr(rscratch2);
1452 __ get_method(rmethod);
1453 __ reinit_heapbase();
1454 __ bind(Continue);
1455 }
1456
1457 // change thread state
1458 __ mov(rscratch1, _thread_in_Java);
1459 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1460 __ stlrw(rscratch1, rscratch2);
1461
1462 // Check preemption for Object.wait()
1463 Label not_preempted;
1464 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1465 __ cbz(rscratch1, not_preempted);
1466 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1467 __ br(rscratch1);
1468 __ bind(native_return);
1469 __ restore_after_resume(true /* is_native */);
1470 // reload result_handler
1471 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize));
1472 __ bind(not_preempted);
1473
1474 // reset_last_Java_frame
1475 __ reset_last_Java_frame(true);
1476
1477 if (CheckJNICalls) {
1478 // clear_pending_jni_exception_check
1479 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1480 }
1481
1482 // reset handle block
1483 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1484 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1485
1486 // If result is an oop unbox and store it in frame where gc will see it
1487 // and result handler will pick it up
1488
1489 {
1490 Label no_oop;
1491 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1492 __ cmp(t, result_handler);
1493 __ br(Assembler::NE, no_oop);
1494 // Unbox oop result, e.g. JNIHandles::resolve result.
1495 __ pop(ltos);
1496 __ resolve_jobject(r0, t, rscratch2);
1497 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1498 // keep stack depth as expected by pushing oop which will eventually be discarded
1499 __ push(ltos);
1500 __ bind(no_oop);
1501 }
1502
1503 {
1504 Label no_reguard;
1505 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1506 __ ldrw(rscratch1, Address(rscratch1));
1507 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1508 __ br(Assembler::NE, no_reguard);
1509
1510 __ push_call_clobbered_registers();
1511 __ mov(c_rarg0, rthread);
1512 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1513 __ blr(rscratch2);
1514 __ pop_call_clobbered_registers();
1515
1516 __ bind(no_reguard);
1517 }
1518
1519 // The method register is junk from after the thread_in_native transition
1520 // until here. Also can't call_VM until the bcp has been
1521 // restored. Need bcp for throwing exception below so get it now.
1522 __ get_method(rmethod);
1523
1524 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1525 // rbcp == code_base()
1526 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1527 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1528 // handle exceptions (exception handling will handle unlocking!)
1529 {
1530 Label L;
1531 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1532 __ cbz(rscratch1, L);
1533 // Note: At some point we may want to unify this with the code
1534 // used in call_VM_base(); i.e., we should use the
1535 // StubRoutines::forward_exception code. For now this doesn't work
1536 // here because the rsp is not correctly set at this point.
1537 __ MacroAssembler::call_VM(noreg,
1538 CAST_FROM_FN_PTR(address,
1539 InterpreterRuntime::throw_pending_exception));
1540 __ should_not_reach_here();
1541 __ bind(L);
1542 }
1543
1544 // do unlocking if necessary
1545 {
1546 Label L;
1547 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1548 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1549 // the code below should be shared with interpreter macro
1550 // assembler implementation
1551 {
1552 Label unlock;
1553 // BasicObjectLock will be first in list, since this is a
1554 // synchronized method. However, need to check that the object
1555 // has not been unlocked by an explicit monitorexit bytecode.
1556
1557 // monitor expect in c_rarg1 for slow unlock path
1558 __ lea (c_rarg1, Address(rfp, // address of first monitor
1559 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1560 wordSize - sizeof(BasicObjectLock))));
1561
1562 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1563 __ cbnz(t, unlock);
1564
1565 // Entry already unlocked, need to throw exception
1566 __ MacroAssembler::call_VM(noreg,
1567 CAST_FROM_FN_PTR(address,
1568 InterpreterRuntime::throw_illegal_monitor_state_exception));
1569 __ should_not_reach_here();
1570
1571 __ bind(unlock);
1572 __ unlock_object(c_rarg1);
1573 }
1574 __ bind(L);
1575 }
1576
1577 #if INCLUDE_JFR
1578 __ enter_jfr_critical_section();
1579
1580 // This poll test is to uphold the invariant that a JFR sampled frame
1581 // must not return to its caller without a prior safepoint poll check.
1582 // The earlier poll check in this routine is insufficient for this purpose
1583 // because the thread has transitioned back to Java.
1584
1585 Label slow_path;
1586 Label fast_path;
1587 __ safepoint_poll(slow_path, true /* at_return */, false /* in_nmethod */);
1588 __ br(Assembler::AL, fast_path);
1589 __ bind(slow_path);
1590 __ push(dtos);
1591 __ push(ltos);
1592 __ set_last_Java_frame(esp, rfp, __ pc(), rscratch1);
1593 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread);
1594 __ reset_last_Java_frame(true);
1595 __ pop(ltos);
1596 __ pop(dtos);
1597 __ bind(fast_path);
1598
1599 #endif // INCLUDE_JFR
1600
1601 // jvmti support
1602 // Note: This must happen _after_ handling/throwing any exceptions since
1603 // the exception handler code notifies the runtime of method exits
1604 // too. If this happens before, method entry/exit notifications are
1605 // not properly paired (was bug - gri 11/22/99).
1606 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1607
1608 // restore potential result in r0:d0, call result handler to
1609 // restore potential result in ST0 & handle result
1610
1611 __ pop(ltos);
1612 __ pop(dtos);
1613
1614 __ blr(result_handler);
1615
1616 // remove activation
1617 __ ldr(esp, Address(rfp,
1618 frame::interpreter_frame_sender_sp_offset *
1619 wordSize)); // get sender sp
1620 // remove frame anchor
1621 __ leave();
1622
1623 JFR_ONLY(__ leave_jfr_critical_section();)
1624
1625 // restore sender sp
1626 __ mov(sp, esp);
1627
1628 __ ret(lr);
1629
1630 if (inc_counter) {
1631 // Handle overflow of counter and compile method
1632 __ bind(invocation_counter_overflow);
1633 generate_counter_overflow(continue_after_compile);
1634 }
1635
1636 return entry_point;
1637 }
1638
1639 //
1640 // Generic interpreted method entry to (asm) interpreter
1641 //
1642 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool object_init) {
1643 // determine code generation flags
1644 bool inc_counter = UseCompiler || CountCompiledCalls;
1645
1646 // rscratch1: sender sp
1647 address entry_point = __ pc();
1648
1649 const Address constMethod(rmethod, Method::const_offset());
1650 const Address access_flags(rmethod, Method::access_flags_offset());
1651 const Address size_of_parameters(r3,
1652 ConstMethod::size_of_parameters_offset());
1653 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1654
1655 // get parameter size (always needed)
1656 // need to load the const method first
1657 __ ldr(r3, constMethod);
1658 __ load_unsigned_short(r2, size_of_parameters);
1659
1660 // r2: size of parameters
1661
1662 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1663 __ sub(r3, r3, r2); // r3 = no. of additional locals
1664
1665 // see if we've got enough room on the stack for locals plus overhead.
1666 generate_stack_overflow_check();
1667
1668 // compute beginning of parameters (rlocals)
1669 __ add(rlocals, esp, r2, ext::uxtx, 3);
1670 __ sub(rlocals, rlocals, wordSize);
1671
1672 __ mov(rscratch1, esp);
1673
1674 // r3 - # of additional locals
1675 // allocate space for locals
1676 // explicitly initialize locals
1677 // Initializing memory allocated for locals in the same direction as
1678 // the stack grows to ensure page initialization order according
1679 // to windows-aarch64 stack page growth requirement (see
1680 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1681 {
1682 Label exit, loop;
1683 __ ands(zr, r3, r3);
1684 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1685 __ bind(loop);
1686 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1687 __ sub(r3, r3, 1); // until everything initialized
1688 __ cbnz(r3, loop);
1689 __ bind(exit);
1690 }
1691
1692 // Padding between locals and fixed part of activation frame to ensure
1693 // SP is always 16-byte aligned.
1694 __ andr(sp, rscratch1, -16);
1695
1696 // And the base dispatch table
1697 __ get_dispatch();
1698
1699 // initialize fixed part of activation frame
1700 generate_fixed_frame(false);
1701
1702 // make sure method is not native & not abstract
1703 #ifdef ASSERT
1704 __ ldrh(r0, access_flags);
1705 {
1706 Label L;
1707 __ tst(r0, JVM_ACC_NATIVE);
1708 __ br(Assembler::EQ, L);
1709 __ stop("tried to execute native method as non-native");
1710 __ bind(L);
1711 }
1712 {
1713 Label L;
1714 __ tst(r0, JVM_ACC_ABSTRACT);
1715 __ br(Assembler::EQ, L);
1716 __ stop("tried to execute abstract method in interpreter");
1717 __ bind(L);
1718 }
1719 #endif
1720
1721 // Since at this point in the method invocation the exception
1722 // handler would try to exit the monitor of synchronized methods
1723 // which hasn't been entered yet, we set the thread local variable
1724 // _do_not_unlock_if_synchronized to true. The remove_activation
1725 // will check this flag.
1726
1727 const Address do_not_unlock_if_synchronized(rthread,
1728 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1729 __ mov(rscratch2, true);
1730 __ strb(rscratch2, do_not_unlock_if_synchronized);
1731
1732 Register mdp = r3;
1733 __ profile_parameters_type(mdp, r1, r2);
1734
1735 // increment invocation count & check for overflow
1736 Label invocation_counter_overflow;
1737 if (inc_counter) {
1738 generate_counter_incr(&invocation_counter_overflow);
1739 }
1740
1741 Label continue_after_compile;
1742 __ bind(continue_after_compile);
1743
1744 bang_stack_shadow_pages(false);
1745
1746 // reset the _do_not_unlock_if_synchronized flag
1747 __ strb(zr, do_not_unlock_if_synchronized);
1748
1749 // check for synchronized methods
1750 // Must happen AFTER invocation_counter check and stack overflow check,
1751 // so method is not locked if overflows.
1752 if (synchronized) {
1753 // Allocate monitor and lock method
1754 lock_method();
1755 } else {
1756 // no synchronization necessary
1757 #ifdef ASSERT
1758 {
1759 Label L;
1760 __ ldrh(r0, access_flags);
1761 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1762 __ br(Assembler::EQ, L);
1763 __ stop("method needs synchronization");
1764 __ bind(L);
1765 }
1766 #endif
1767 }
1768
1769 // Issue a StoreStore barrier on entry to Object_init if the
1770 // class has strict field fields. Be lazy, always do it.
1771 if (object_init) {
1772 __ membar(MacroAssembler::StoreStore);
1773 }
1774
1775 // start execution
1776 #ifdef ASSERT
1777 {
1778 Label L;
1779 const Address monitor_block_top (rfp,
1780 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1781 __ ldr(rscratch1, monitor_block_top);
1782 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1783 __ cmp(esp, rscratch1);
1784 __ br(Assembler::EQ, L);
1785 __ stop("broken stack frame setup in interpreter 2");
1786 __ bind(L);
1787 }
1788 #endif
1789
1790 // jvmti support
1791 __ notify_method_entry();
1792
1793 __ dispatch_next(vtos);
1794
1795 // invocation counter overflow
1796 if (inc_counter) {
1797 // Handle overflow of counter and compile method
1798 __ bind(invocation_counter_overflow);
1799 generate_counter_overflow(continue_after_compile);
1800 }
1801
1802 return entry_point;
1803 }
1804
1805 // Method entry for java.lang.Thread.currentThread
1806 address TemplateInterpreterGenerator::generate_currentThread() {
1807 address entry_point = __ pc();
1808
1809 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1810 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1811 __ ret(lr);
1812
1813 return entry_point;
1814 }
1815
1816 //-----------------------------------------------------------------------------
1817 // Exceptions
1818
1819 void TemplateInterpreterGenerator::generate_throw_exception() {
1820 // Entry point in previous activation (i.e., if the caller was
1821 // interpreted)
1822 Interpreter::_rethrow_exception_entry = __ pc();
1823 // Restore sp to interpreter_frame_last_sp even though we are going
1824 // to empty the expression stack for the exception processing.
1825 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1826 // r0: exception
1827 // r3: return address/pc that threw exception
1828 __ restore_bcp(); // rbcp points to call/send
1829 __ restore_locals();
1830 __ restore_constant_pool_cache();
1831 __ reinit_heapbase(); // restore rheapbase as heapbase.
1832 __ get_dispatch();
1833
1834 // Entry point for exceptions thrown within interpreter code
1835 Interpreter::_throw_exception_entry = __ pc();
1836 // If we came here via a NullPointerException on the receiver of a
1837 // method, rmethod may be corrupt.
1838 __ get_method(rmethod);
1839 // expression stack is undefined here
1840 // r0: exception
1841 // rbcp: exception bcp
1842 __ verify_oop(r0);
1843 __ mov(c_rarg1, r0);
1844
1845 // expression stack must be empty before entering the VM in case of
1846 // an exception
1847 __ empty_expression_stack();
1848 // find exception handler address and preserve exception oop
1849 __ call_VM(r3,
1850 CAST_FROM_FN_PTR(address,
1851 InterpreterRuntime::exception_handler_for_exception),
1852 c_rarg1);
1853
1854 // Restore machine SP
1855 __ restore_sp_after_call();
1856
1857 // r0: exception handler entry point
1858 // r3: preserved exception oop
1859 // rbcp: bcp for exception handler
1860 __ push_ptr(r3); // push exception which is now the only value on the stack
1861 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1862
1863 // If the exception is not handled in the current frame the frame is
1864 // removed and the exception is rethrown (i.e. exception
1865 // continuation is _rethrow_exception).
1866 //
1867 // Note: At this point the bci is still the bxi for the instruction
1868 // which caused the exception and the expression stack is
1869 // empty. Thus, for any VM calls at this point, GC will find a legal
1870 // oop map (with empty expression stack).
1871
1872 //
1873 // JVMTI PopFrame support
1874 //
1875
1876 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1877 __ empty_expression_stack();
1878 __ restore_bcp(); // We could have returned from deoptimizing this frame, so restore rbcp.
1879 // Set the popframe_processing bit in pending_popframe_condition
1880 // indicating that we are currently handling popframe, so that
1881 // call_VMs that may happen later do not trigger new popframe
1882 // handling cycles.
1883 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1884 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1885 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1886
1887 {
1888 // Check to see whether we are returning to a deoptimized frame.
1889 // (The PopFrame call ensures that the caller of the popped frame is
1890 // either interpreted or compiled and deoptimizes it if compiled.)
1891 // In this case, we can't call dispatch_next() after the frame is
1892 // popped, but instead must save the incoming arguments and restore
1893 // them after deoptimization has occurred.
1894 //
1895 // Note that we don't compare the return PC against the
1896 // deoptimization blob's unpack entry because of the presence of
1897 // adapter frames in C2.
1898 Label caller_not_deoptimized;
1899 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1900 // This is a return address, so requires authenticating for PAC.
1901 __ authenticate_return_address(c_rarg1);
1902 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1903 InterpreterRuntime::interpreter_contains), c_rarg1);
1904 __ cbnz(r0, caller_not_deoptimized);
1905
1906 // Compute size of arguments for saving when returning to
1907 // deoptimized caller
1908 __ get_method(r0);
1909 __ ldr(r0, Address(r0, Method::const_offset()));
1910 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1911 size_of_parameters_offset())));
1912 __ lsl(r0, r0, Interpreter::logStackElementSize);
1913 __ restore_locals(); // XXX do we need this?
1914 __ sub(rlocals, rlocals, r0);
1915 __ add(rlocals, rlocals, wordSize);
1916 // Save these arguments
1917 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1918 Deoptimization::
1919 popframe_preserve_args),
1920 rthread, r0, rlocals);
1921
1922 __ remove_activation(vtos,
1923 /* throw_monitor_exception */ false,
1924 /* install_monitor_exception */ false,
1925 /* notify_jvmdi */ false);
1926
1927 // Inform deoptimization that it is responsible for restoring
1928 // these arguments
1929 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1930 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1931
1932 // Continue in deoptimization handler
1933 __ ret(lr);
1934
1935 __ bind(caller_not_deoptimized);
1936 }
1937
1938 __ remove_activation(vtos,
1939 /* throw_monitor_exception */ false,
1940 /* install_monitor_exception */ false,
1941 /* notify_jvmdi */ false);
1942
1943 // Restore the last_sp and null it out
1944 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1945 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1946 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1947
1948 __ restore_bcp();
1949 __ restore_locals();
1950 __ restore_constant_pool_cache();
1951 __ get_method(rmethod);
1952 __ get_dispatch();
1953
1954 // The method data pointer was incremented already during
1955 // call profiling. We have to restore the mdp for the current bcp.
1956 if (ProfileInterpreter) {
1957 __ set_method_data_pointer_for_bcp();
1958 }
1959
1960 // Clear the popframe condition flag
1961 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1962 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1963
1964 #if INCLUDE_JVMTI
1965 {
1966 Label L_done;
1967
1968 __ ldrb(rscratch1, Address(rbcp, 0));
1969 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1970 __ br(Assembler::NE, L_done);
1971
1972 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1973 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1974
1975 __ ldr(c_rarg0, Address(rlocals, 0));
1976 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1977
1978 __ cbz(r0, L_done);
1979
1980 __ str(r0, Address(esp, 0));
1981 __ bind(L_done);
1982 }
1983 #endif // INCLUDE_JVMTI
1984
1985 // Restore machine SP
1986 __ restore_sp_after_call();
1987
1988 __ dispatch_next(vtos);
1989 // end of PopFrame support
1990
1991 Interpreter::_remove_activation_entry = __ pc();
1992
1993 // preserve exception over this code sequence
1994 __ pop_ptr(r0);
1995 __ str(r0, Address(rthread, JavaThread::vm_result_oop_offset()));
1996 // remove the activation (without doing throws on illegalMonitorExceptions)
1997 __ remove_activation(vtos, false, true, false);
1998 // restore exception
1999 __ get_vm_result_oop(r0, rthread);
2000
2001 // In between activations - previous activation type unknown yet
2002 // compute continuation point - the continuation point expects the
2003 // following registers set up:
2004 //
2005 // r0: exception
2006 // lr: return address/pc that threw exception
2007 // esp: expression stack of caller
2008 // rfp: fp of caller
2009 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
2010 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2011 SharedRuntime::exception_handler_for_return_address),
2012 rthread, lr);
2013 __ mov(r1, r0); // save exception handler
2014 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
2015 // We might be returning to a deopt handler that expects r3 to
2016 // contain the exception pc
2017 __ mov(r3, lr);
2018 // Note that an "issuing PC" is actually the next PC after the call
2019 __ br(r1); // jump to exception
2020 // handler of caller
2021 }
2022
2023
2024 //
2025 // JVMTI ForceEarlyReturn support
2026 //
2027 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2028 address entry = __ pc();
2029
2030 __ restore_bcp();
2031 __ restore_locals();
2032 __ empty_expression_stack();
2033 __ load_earlyret_value(state);
2034
2035 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
2036 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
2037
2038 // Clear the earlyret state
2039 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
2040 __ str(zr, cond_addr);
2041
2042 __ remove_activation(state,
2043 false, /* throw_monitor_exception */
2044 false, /* install_monitor_exception */
2045 true); /* notify_jvmdi */
2046 __ ret(lr);
2047
2048 return entry;
2049 } // end of ForceEarlyReturn support
2050
2051
2052
2053 //-----------------------------------------------------------------------------
2054 // Helper for vtos entry point generation
2055
2056 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2057 address& bep,
2058 address& cep,
2059 address& sep,
2060 address& aep,
2061 address& iep,
2062 address& lep,
2063 address& fep,
2064 address& dep,
2065 address& vep) {
2066 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2067 Label L;
2068 aep = __ pc(); // atos entry point
2069 __ push_ptr();
2070 __ b(L);
2071 fep = __ pc(); // ftos entry point
2072 __ push_f();
2073 __ b(L);
2074 dep = __ pc(); // dtos entry point
2075 __ push_d();
2076 __ b(L);
2077 lep = __ pc(); // ltos entry point
2078 __ push_l();
2079 __ b(L);
2080 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
2081 __ push_i();
2082 vep = __ pc(); // vtos entry point
2083 __ bind(L);
2084 generate_and_dispatch(t);
2085 }
2086
2087 //-----------------------------------------------------------------------------
2088
2089 // Non-product code
2090 #ifndef PRODUCT
2091 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2092 address entry = __ pc();
2093
2094 __ protect_return_address();
2095 __ push(lr);
2096 __ push(state);
2097 __ push(RegSet::range(r0, r15), sp);
2098 __ mov(c_rarg2, r0); // Pass itos
2099 __ call_VM(noreg,
2100 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2101 c_rarg1, c_rarg2, c_rarg3);
2102 __ pop(RegSet::range(r0, r15), sp);
2103 __ pop(state);
2104 __ pop(lr);
2105 __ authenticate_return_address();
2106 __ ret(lr); // return from result handler
2107
2108 return entry;
2109 }
2110
2111 void TemplateInterpreterGenerator::count_bytecode() {
2112 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2113 __ atomic_add(noreg, 1, r10);
2114 }
2115
2116 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2117 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2118 __ atomic_addw(noreg, 1, r10);
2119 }
2120
2121 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2122 // Calculate new index for counter:
2123 // _index = (_index >> log2_number_of_codes) |
2124 // (bytecode << log2_number_of_codes);
2125 Register index_addr = rscratch1;
2126 Register index = rscratch2;
2127 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2128 __ ldrw(index, index_addr);
2129 __ mov(r10,
2130 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2131 __ orrw(index, r10, index, Assembler::LSR,
2132 BytecodePairHistogram::log2_number_of_codes);
2133 __ strw(index, index_addr);
2134
2135 // Bump bucket contents:
2136 // _counters[_index] ++;
2137 Register counter_addr = rscratch1;
2138 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2139 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2140 __ atomic_addw(noreg, 1, counter_addr);
2141 }
2142
2143 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2144 // Call a little run-time stub to avoid blow-up for each bytecode.
2145 // The run-time runtime saves the right registers, depending on
2146 // the tosca in-state for the given template.
2147
2148 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2149 "entry must have been generated");
2150 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2151 __ reinit_heapbase();
2152 }
2153
2154
2155 void TemplateInterpreterGenerator::stop_interpreter_at() {
2156 Label L;
2157 __ push(rscratch1);
2158 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2159 __ ldr(rscratch1, Address(rscratch1));
2160 __ mov(rscratch2, StopInterpreterAt);
2161 __ cmp(rscratch1, rscratch2);
2162 __ br(Assembler::NE, L);
2163 __ brk(0);
2164 __ bind(L);
2165 __ pop(rscratch1);
2166 }
2167
2168 #endif // !PRODUCT