1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "compiler/compiler_globals.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "interpreter/bytecodeHistogram.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "interpreter/interpreterRuntime.hpp"
35 #include "interpreter/interp_masm.hpp"
36 #include "interpreter/templateInterpreterGenerator.hpp"
37 #include "interpreter/templateTable.hpp"
38 #include "interpreter/bytecodeTracer.hpp"
39 #include "memory/resourceArea.hpp"
40 #include "oops/arrayOop.hpp"
41 #include "oops/method.hpp"
42 #include "oops/methodCounters.hpp"
43 #include "oops/methodData.hpp"
44 #include "oops/oop.inline.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 __ restore_bcp();
472 __ restore_locals();
473 __ restore_constant_pool_cache();
474 __ get_method(rmethod);
475
476 if (state == atos) {
477 Register obj = r0;
478 Register mdp = r1;
479 Register tmp = r2;
480 __ profile_return_type(mdp, obj, tmp);
481 }
482
483 const Register cache = r1;
484 const Register index = r2;
485
486 if (index_size == sizeof(u4)) {
487 __ load_resolved_indy_entry(cache, index);
488 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
489 __ add(esp, esp, cache, Assembler::LSL, 3);
490 } else {
491 // Pop N words from the stack
492 assert(index_size == sizeof(u2), "Can only be u2");
493 __ load_method_entry(cache, index);
494 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
495 __ add(esp, esp, cache, Assembler::LSL, 3);
496 }
497
498 // Restore machine SP
499 __ restore_sp_after_call();
500
501 __ check_and_handle_popframe(rthread);
502 __ check_and_handle_earlyret(rthread);
503
504 __ get_dispatch();
505 __ dispatch_next(state, step);
506
507 return entry;
508 }
509
510 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
511 int step,
512 address continuation) {
513 address entry = __ pc();
514 __ restore_bcp();
515 __ restore_locals();
516 __ restore_constant_pool_cache();
517 __ get_method(rmethod);
518 __ get_dispatch();
519
520 __ restore_sp_after_call(); // Restore SP to extended SP
521
522 // Restore expression stack pointer
523 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
524 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
525 // null last_sp until next java call
526 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
527
528 #if INCLUDE_JVMCI
529 // Check if we need to take lock at entry of synchronized method. This can
530 // only occur on method entry so emit it only for vtos with step 0.
531 if (EnableJVMCI && state == vtos && step == 0) {
532 Label L;
533 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
534 __ cbz(rscratch1, L);
535 // Clear flag.
536 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
537 // Take lock.
538 lock_method();
539 __ bind(L);
540 } else {
541 #ifdef ASSERT
542 if (EnableJVMCI) {
543 Label L;
544 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
545 __ cbz(rscratch1, L);
546 __ stop("unexpected pending monitor in deopt entry");
547 __ bind(L);
548 }
549 #endif
550 }
551 #endif
552 // handle exceptions
553 {
554 Label L;
555 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
556 __ cbz(rscratch1, L);
557 __ call_VM(noreg,
558 CAST_FROM_FN_PTR(address,
559 InterpreterRuntime::throw_pending_exception));
560 __ should_not_reach_here();
561 __ bind(L);
562 }
563
564 if (continuation == nullptr) {
565 __ dispatch_next(state, step);
566 } else {
567 __ jump_to_entry(continuation);
568 }
569 return entry;
570 }
571
572 address TemplateInterpreterGenerator::generate_result_handler_for(
573 BasicType type) {
574 address entry = __ pc();
575 switch (type) {
576 case T_BOOLEAN: __ c2bool(r0); break;
577 case T_CHAR : __ uxth(r0, r0); break;
578 case T_BYTE : __ sxtb(r0, r0); break;
579 case T_SHORT : __ sxth(r0, r0); break;
580 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
581 case T_LONG : /* nothing to do */ break;
582 case T_VOID : /* nothing to do */ break;
583 case T_FLOAT : /* nothing to do */ break;
584 case T_DOUBLE : /* nothing to do */ break;
585 case T_OBJECT :
586 // retrieve result from frame
587 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
588 // and verify it
589 __ verify_oop(r0);
590 break;
591 default : ShouldNotReachHere();
592 }
593 __ ret(lr); // return from result handler
594 return entry;
595 }
596
597 address TemplateInterpreterGenerator::generate_safept_entry_for(
598 TosState state,
599 address runtime_entry) {
600 address entry = __ pc();
601 __ push(state);
602 __ push_cont_fastpath(rthread);
603 __ call_VM(noreg, runtime_entry);
604 __ pop_cont_fastpath(rthread);
605 __ membar(Assembler::AnyAny);
606 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
607 return entry;
608 }
609
610 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
611 if (!Continuations::enabled()) return nullptr;
612 address start = __ pc();
613
614 __ restore_bcp();
615 __ restore_locals();
616
617 // Restore constant pool cache
618 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
619
620 // Restore Java expression stack pointer
621 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
622 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
623 // and NULL it as marker that esp is now tos until next java call
624 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
625
626 // Restore machine SP
627 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
628 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord)));
629
630 // Restore method
631 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
632
633 // Restore dispatch
634 uint64_t offset;
635 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
636 __ add(rdispatch, rdispatch, offset);
637
638 __ ret(lr);
639
640 return start;
641 }
642
643
644 // Helpers for commoning out cases in the various type of method entries.
645 //
646
647
648 // increment invocation count & check for overflow
649 //
650 // Note: checking for negative value instead of overflow
651 // so we have a 'sticky' overflow test
652 //
653 // rmethod: method
654 //
655 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
656 Label done;
657 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
658 int increment = InvocationCounter::count_increment;
659 Label no_mdo;
660 if (ProfileInterpreter) {
661 // Are we profiling?
662 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
663 __ cbz(r0, no_mdo);
664 // Increment counter in the MDO
665 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
666 in_bytes(InvocationCounter::counter_offset()));
667 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
668 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
669 __ b(done);
670 }
671 __ bind(no_mdo);
672 // Increment counter in MethodCounters
673 const Address invocation_counter(rscratch2,
674 MethodCounters::invocation_counter_offset() +
675 InvocationCounter::counter_offset());
676 __ get_method_counters(rmethod, rscratch2, done);
677 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
678 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
679 __ bind(done);
680 }
681
682 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
683
684 // Asm interpreter on entry
685 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
686 // Everything as it was on entry
687
688 // InterpreterRuntime::frequency_counter_overflow takes two
689 // arguments, the first (thread) is passed by call_VM, the second
690 // indicates if the counter overflow occurs at a backwards branch
691 // (null bcp). We pass zero for it. The call returns the address
692 // of the verified entry point for the method or null if the
693 // compilation did not complete (either went background or bailed
694 // out).
695 __ mov(c_rarg1, 0);
696 __ call_VM(noreg,
697 CAST_FROM_FN_PTR(address,
698 InterpreterRuntime::frequency_counter_overflow),
699 c_rarg1);
700
701 __ b(do_continue);
702 }
703
704 // See if we've got enough room on the stack for locals plus overhead
705 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
706 // without going through the signal handler, i.e., reserved and yellow zones
707 // will not be made usable. The shadow zone must suffice to handle the
708 // overflow.
709 // The expression stack grows down incrementally, so the normal guard
710 // page mechanism will work for that.
711 //
712 // NOTE: Since the additional locals are also always pushed (wasn't
713 // obvious in generate_method_entry) so the guard should work for them
714 // too.
715 //
716 // Args:
717 // r3: number of additional locals this frame needs (what we must check)
718 // rmethod: Method*
719 //
720 // Kills:
721 // r0
722 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
723
724 // monitor entry size: see picture of stack set
725 // (generate_method_entry) and frame_amd64.hpp
726 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
727
728 // total overhead size: entry_size + (saved rbp through expr stack
729 // bottom). be sure to change this if you add/subtract anything
730 // to/from the overhead area
731 const int overhead_size =
732 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
733
734 const size_t page_size = os::vm_page_size();
735
736 Label after_frame_check;
737
738 // see if the frame is greater than one page in size. If so,
739 // then we need to verify there is enough stack space remaining
740 // for the additional locals.
741 //
742 // Note that we use SUBS rather than CMP here because the immediate
743 // field of this instruction may overflow. SUBS can cope with this
744 // because it is a macro that will expand to some number of MOV
745 // instructions and a register operation.
746 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
747 __ br(Assembler::LS, after_frame_check);
748
749 // compute rsp as if this were going to be the last frame on
750 // the stack before the red zone
751
752 // locals + overhead, in bytes
753 __ mov(r0, overhead_size);
754 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
755
756 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
757 __ ldr(rscratch1, stack_limit);
758
759 #ifdef ASSERT
760 Label limit_okay;
761 // Verify that thread stack limit is non-zero.
762 __ cbnz(rscratch1, limit_okay);
763 __ stop("stack overflow limit is zero");
764 __ bind(limit_okay);
765 #endif
766
767 // Add stack limit to locals.
768 __ add(r0, r0, rscratch1);
769
770 // Check against the current stack bottom.
771 __ cmp(sp, r0);
772 __ br(Assembler::HI, after_frame_check);
773
774 // Remove the incoming args, peeling the machine SP back to where it
775 // was in the caller. This is not strictly necessary, but unless we
776 // do so the stack frame may have a garbage FP; this ensures a
777 // correct call stack that we can always unwind. The ANDR should be
778 // unnecessary because the sender SP in r19 is always aligned, but
779 // it doesn't hurt.
780 __ andr(sp, r19_sender_sp, -16);
781
782 // Note: the restored frame is not necessarily interpreted.
783 // Use the shared runtime version of the StackOverflowError.
784 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
785 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
786
787 // all done with frame size check
788 __ bind(after_frame_check);
789 }
790
791 // Allocate monitor and lock method (asm interpreter)
792 //
793 // Args:
794 // rmethod: Method*
795 // rlocals: locals
796 //
797 // Kills:
798 // r0
799 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
800 // rscratch1, rscratch2 (scratch regs)
801 void TemplateInterpreterGenerator::lock_method() {
802 // synchronize method
803 const Address access_flags(rmethod, Method::access_flags_offset());
804 const Address monitor_block_top(
805 rfp,
806 frame::interpreter_frame_monitor_block_top_offset * wordSize);
807 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
808
809 #ifdef ASSERT
810 {
811 Label L;
812 __ ldrw(r0, access_flags);
813 __ tst(r0, JVM_ACC_SYNCHRONIZED);
814 __ br(Assembler::NE, L);
815 __ stop("method doesn't need synchronization");
816 __ bind(L);
817 }
818 #endif // ASSERT
819
820 // get synchronization object
821 {
822 Label done;
823 __ ldrw(r0, access_flags);
824 __ tst(r0, JVM_ACC_STATIC);
825 // get receiver (assume this is frequent case)
826 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
827 __ br(Assembler::EQ, done);
828 __ load_mirror(r0, rmethod, r5, rscratch2);
829
830 #ifdef ASSERT
831 {
832 Label L;
833 __ cbnz(r0, L);
834 __ stop("synchronization object is null");
835 __ bind(L);
836 }
837 #endif // ASSERT
838
839 __ bind(done);
840 }
841
842 // add space for monitor & lock
843 __ check_extended_sp();
844 __ sub(sp, sp, entry_size); // add space for a monitor entry
845 __ sub(esp, esp, entry_size);
846 __ sub(rscratch1, sp, rfp);
847 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
848 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
849 __ sub(rscratch1, esp, rfp);
850 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
851 __ str(rscratch1, monitor_block_top); // set new monitor block top
852
853 // store object
854 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
855 __ mov(c_rarg1, esp); // object address
856 __ lock_object(c_rarg1);
857 }
858
859 // Generate a fixed interpreter frame. This is identical setup for
860 // interpreted methods and for native methods hence the shared code.
861 //
862 // Args:
863 // lr: return address
864 // rmethod: Method*
865 // rlocals: pointer to locals
866 // rcpool: cp cache
867 // stack_pointer: previous sp
868 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
869 // initialize fixed part of activation frame
870 if (native_call) {
871 __ sub(esp, sp, 14 * wordSize);
872 __ mov(rbcp, zr);
873 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
874 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
875 // add 2 zero-initialized slots for native calls
876 __ stp(zr, zr, Address(sp, 12 * wordSize));
877 } else {
878 __ sub(esp, sp, 12 * wordSize);
879 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
880 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
881 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
882 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
883 }
884
885 if (ProfileInterpreter) {
886 Label method_data_continue;
887 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
888 __ cbz(rscratch1, method_data_continue);
889 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
890 __ bind(method_data_continue);
891 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
892 } else {
893 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
894 }
895
896 __ protect_return_address();
897 __ stp(rfp, lr, Address(sp, 10 * wordSize));
898 __ lea(rfp, Address(sp, 10 * wordSize));
899
900 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
901 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
902 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
903 __ sub(rscratch1, rlocals, rfp);
904 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
905 // Store relativized rlocals, see frame::interpreter_frame_locals().
906 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
907
908 // set sender sp
909 // leave last_sp as null
910 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
911
912 // Get mirror
913 __ load_mirror(r10, rmethod, r5, rscratch2);
914 if (! native_call) {
915 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
916 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
917 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
918 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
919 __ andr(rscratch1, rscratch1, -16);
920 __ sub(rscratch2, rscratch1, rfp);
921 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
922 // Store extended SP and mirror
923 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
924 // Move SP out of the way
925 __ mov(sp, rscratch1);
926 } else {
927 // Make sure there is room for the exception oop pushed in case method throws
928 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
929 __ sub(rscratch1, sp, 2 * wordSize);
930 __ sub(rscratch2, rscratch1, rfp);
931 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
932 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
933 __ mov(sp, rscratch1);
934 }
935 }
936
937 // End of helpers
938
939 // Various method entries
940 //------------------------------------------------------------------------------------------------------------------------
941 //
942 //
943
944 // Method entry for java.lang.ref.Reference.get.
945 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
946 // Code: _aload_0, _getfield, _areturn
947 // parameter size = 1
948 //
949 // The code that gets generated by this routine is split into 2 parts:
950 // 1. The "intrinsified" code for G1 (or any SATB based GC),
951 // 2. The slow path - which is an expansion of the regular method entry.
952 //
953 // Notes:-
954 // * In the G1 code we do not check whether we need to block for
955 // a safepoint. If G1 is enabled then we must execute the specialized
956 // code for Reference.get (except when the Reference object is null)
957 // so that we can log the value in the referent field with an SATB
958 // update buffer.
959 // If the code for the getfield template is modified so that the
960 // G1 pre-barrier code is executed when the current method is
961 // Reference.get() then going through the normal method entry
962 // will be fine.
963 // * The G1 code can, however, check the receiver object (the instance
964 // of java.lang.Reference) and jump to the slow path if null. If the
965 // Reference object is null then we obviously cannot fetch the referent
966 // and so we don't need to call the G1 pre-barrier. Thus we can use the
967 // regular method entry code to generate the NPE.
968 //
969 // This code is based on generate_accessor_entry.
970 //
971 // rmethod: Method*
972 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
973
974 // LR is live. It must be saved around calls.
975
976 address entry = __ pc();
977
978 const int referent_offset = java_lang_ref_Reference::referent_offset();
979
980 Label slow_path;
981 const Register local_0 = c_rarg0;
982 // Check if local 0 != null
983 // If the receiver is null then it is OK to jump to the slow path.
984 __ ldr(local_0, Address(esp, 0));
985 __ cbz(local_0, slow_path);
986
987 // Load the value of the referent field.
988 const Address field_address(local_0, referent_offset);
989 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
990 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
991
992 // areturn
993 __ andr(sp, r19_sender_sp, -16); // done with stack
994 __ ret(lr);
995
996 // generate a vanilla interpreter entry as the slow path
997 __ bind(slow_path);
998 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
999 return entry;
1000
1001 }
1002
1003 /**
1004 * Method entry for static native methods:
1005 * int java.util.zip.CRC32.update(int crc, int b)
1006 */
1007 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1008 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1009 address entry = __ pc();
1010
1011 // rmethod: Method*
1012 // r19_sender_sp: senderSP must preserved for slow path
1013 // esp: args
1014
1015 Label slow_path;
1016 // If we need a safepoint check, generate full interpreter entry.
1017 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1018
1019 // We don't generate local frame and don't align stack because
1020 // we call stub code and there is no safepoint on this path.
1021
1022 // Load parameters
1023 const Register crc = c_rarg0; // crc
1024 const Register val = c_rarg1; // source java byte value
1025 const Register tbl = c_rarg2; // scratch
1026
1027 // Arguments are reversed on java expression stack
1028 __ ldrw(val, Address(esp, 0)); // byte value
1029 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1030
1031 uint64_t offset;
1032 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1033 __ add(tbl, tbl, offset);
1034
1035 __ mvnw(crc, crc); // ~crc
1036 __ update_byte_crc32(crc, val, tbl);
1037 __ mvnw(crc, crc); // ~crc
1038
1039 // result in c_rarg0
1040
1041 __ andr(sp, r19_sender_sp, -16);
1042 __ ret(lr);
1043
1044 // generate a vanilla native entry as the slow path
1045 __ bind(slow_path);
1046 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1047 return entry;
1048 }
1049
1050 /**
1051 * Method entry for static native methods:
1052 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1053 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1054 */
1055 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1056 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1057 address entry = __ pc();
1058
1059 // rmethod,: Method*
1060 // r19_sender_sp: senderSP must preserved for slow path
1061
1062 Label slow_path;
1063 // If we need a safepoint check, generate full interpreter entry.
1064 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1065
1066 // We don't generate local frame and don't align stack because
1067 // we call stub code and there is no safepoint on this path.
1068
1069 // Load parameters
1070 const Register crc = c_rarg0; // crc
1071 const Register buf = c_rarg1; // source java byte array address
1072 const Register len = c_rarg2; // length
1073 const Register off = len; // offset (never overlaps with 'len')
1074
1075 // Arguments are reversed on java expression stack
1076 // Calculate address of start element
1077 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1078 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1079 __ ldrw(off, Address(esp, wordSize)); // offset
1080 __ add(buf, buf, off); // + offset
1081 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1082 } else {
1083 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1084 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1085 __ ldrw(off, Address(esp, wordSize)); // offset
1086 __ add(buf, buf, off); // + offset
1087 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1088 }
1089 // Can now load 'len' since we're finished with 'off'
1090 __ ldrw(len, Address(esp, 0x0)); // Length
1091
1092 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1093
1094 // We are frameless so we can just jump to the stub.
1095 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1096
1097 // generate a vanilla native entry as the slow path
1098 __ bind(slow_path);
1099 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1100 return entry;
1101 }
1102
1103 /**
1104 * Method entry for intrinsic-candidate (non-native) methods:
1105 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1106 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1107 * Unlike CRC32, CRC32C does not have any methods marked as native
1108 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1109 */
1110 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1111 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1112 address entry = __ pc();
1113
1114 // Prepare jump to stub using parameters from the stack
1115 const Register crc = c_rarg0; // initial crc
1116 const Register buf = c_rarg1; // source java byte array address
1117 const Register len = c_rarg2; // len argument to the kernel
1118
1119 const Register end = len; // index of last element to process
1120 const Register off = crc; // offset
1121
1122 __ ldrw(end, Address(esp)); // int end
1123 __ ldrw(off, Address(esp, wordSize)); // int offset
1124 __ sub(len, end, off);
1125 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1126 __ add(buf, buf, off); // + offset
1127 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1128 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1129 } else {
1130 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1131 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1132 }
1133
1134 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1135
1136 // Jump to the stub.
1137 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1138
1139 return entry;
1140 }
1141
1142 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1143 // See more discussion in stackOverflow.hpp.
1144
1145 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1146 const int page_size = (int)os::vm_page_size();
1147 const int n_shadow_pages = shadow_zone_size / page_size;
1148
1149 #ifdef ASSERT
1150 Label L_good_limit;
1151 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1152 __ cbnz(rscratch1, L_good_limit);
1153 __ stop("shadow zone safe limit is not initialized");
1154 __ bind(L_good_limit);
1155
1156 Label L_good_watermark;
1157 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1158 __ cbnz(rscratch1, L_good_watermark);
1159 __ stop("shadow zone growth watermark is not initialized");
1160 __ bind(L_good_watermark);
1161 #endif
1162
1163 Label L_done;
1164
1165 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1166 __ cmp(sp, rscratch1);
1167 __ br(Assembler::HI, L_done);
1168
1169 for (int p = 1; p <= n_shadow_pages; p++) {
1170 __ sub(rscratch2, sp, p*page_size);
1171 __ str(zr, Address(rscratch2));
1172 }
1173
1174 // Record the new watermark, but only if the update is above the safe limit.
1175 // Otherwise, the next time around the check above would pass the safe limit.
1176 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1177 __ cmp(sp, rscratch1);
1178 __ br(Assembler::LS, L_done);
1179 __ mov(rscratch1, sp);
1180 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1181
1182 __ bind(L_done);
1183 }
1184
1185 // Interpreter stub for calling a native method. (asm interpreter)
1186 // This sets up a somewhat different looking stack for calling the
1187 // native method than the typical interpreter frame setup.
1188 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
1189 // determine code generation flags
1190 bool inc_counter = UseCompiler || CountCompiledCalls;
1191
1192 // r1: Method*
1193 // rscratch1: sender sp
1194
1195 address entry_point = __ pc();
1196
1197 const Address constMethod (rmethod, Method::const_offset());
1198 const Address access_flags (rmethod, Method::access_flags_offset());
1199 const Address size_of_parameters(r2, ConstMethod::
1200 size_of_parameters_offset());
1201
1202 // get parameter size (always needed)
1203 __ ldr(r2, constMethod);
1204 __ load_unsigned_short(r2, size_of_parameters);
1205
1206 // Native calls don't need the stack size check since they have no
1207 // expression stack and the arguments are already on the stack and
1208 // we only add a handful of words to the stack.
1209
1210 // rmethod: Method*
1211 // r2: size of parameters
1212 // rscratch1: sender sp
1213
1214 // for natives the size of locals is zero
1215
1216 // compute beginning of parameters (rlocals)
1217 __ add(rlocals, esp, r2, ext::uxtx, 3);
1218 __ add(rlocals, rlocals, -wordSize);
1219
1220 // Pull SP back to minimum size: this avoids holes in the stack
1221 __ andr(sp, esp, -16);
1222
1223 // initialize fixed part of activation frame
1224 generate_fixed_frame(true);
1225
1226 // make sure method is native & not abstract
1227 #ifdef ASSERT
1228 __ ldrw(r0, access_flags);
1229 {
1230 Label L;
1231 __ tst(r0, JVM_ACC_NATIVE);
1232 __ br(Assembler::NE, L);
1233 __ stop("tried to execute non-native method as native");
1234 __ bind(L);
1235 }
1236 {
1237 Label L;
1238 __ tst(r0, JVM_ACC_ABSTRACT);
1239 __ br(Assembler::EQ, L);
1240 __ stop("tried to execute abstract method in interpreter");
1241 __ bind(L);
1242 }
1243 #endif
1244
1245 // Since at this point in the method invocation the exception
1246 // handler would try to exit the monitor of synchronized methods
1247 // which hasn't been entered yet, we set the thread local variable
1248 // _do_not_unlock_if_synchronized to true. The remove_activation
1249 // will check this flag.
1250
1251 const Address do_not_unlock_if_synchronized(rthread,
1252 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1253 __ mov(rscratch2, true);
1254 __ strb(rscratch2, do_not_unlock_if_synchronized);
1255
1256 // increment invocation count & check for overflow
1257 Label invocation_counter_overflow;
1258 if (inc_counter) {
1259 generate_counter_incr(&invocation_counter_overflow);
1260 }
1261
1262 Label continue_after_compile;
1263 __ bind(continue_after_compile);
1264
1265 bang_stack_shadow_pages(true);
1266
1267 // reset the _do_not_unlock_if_synchronized flag
1268 __ strb(zr, do_not_unlock_if_synchronized);
1269
1270 // check for synchronized methods
1271 // Must happen AFTER invocation_counter check and stack overflow check,
1272 // so method is not locked if overflows.
1273 if (synchronized) {
1274 lock_method();
1275 } else {
1276 // no synchronization necessary
1277 #ifdef ASSERT
1278 {
1279 Label L;
1280 __ ldrw(r0, access_flags);
1281 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1282 __ br(Assembler::EQ, L);
1283 __ stop("method needs synchronization");
1284 __ bind(L);
1285 }
1286 #endif
1287 }
1288
1289 // start execution
1290 #ifdef ASSERT
1291 {
1292 Label L;
1293 const Address monitor_block_top(rfp,
1294 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1295 __ ldr(rscratch1, monitor_block_top);
1296 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1297 __ cmp(esp, rscratch1);
1298 __ br(Assembler::EQ, L);
1299 __ stop("broken stack frame setup in interpreter 1");
1300 __ bind(L);
1301 }
1302 #endif
1303
1304 // jvmti support
1305 __ notify_method_entry();
1306
1307 if (runtime_upcalls) {
1308 __ generate_runtime_upcalls_on_method_entry();
1309 }
1310
1311 // work registers
1312 const Register t = r17;
1313 const Register result_handler = r19;
1314
1315 // allocate space for parameters
1316 __ ldr(t, Address(rmethod, Method::const_offset()));
1317 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1318
1319 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1320 __ andr(sp, rscratch1, -16);
1321 __ mov(esp, rscratch1);
1322
1323 // get signature handler
1324 {
1325 Label L;
1326 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1327 __ cbnz(t, L);
1328 __ call_VM(noreg,
1329 CAST_FROM_FN_PTR(address,
1330 InterpreterRuntime::prepare_native_call),
1331 rmethod);
1332 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1333 __ bind(L);
1334 }
1335
1336 // call signature handler
1337 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1338 "adjust this code");
1339 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1340 "adjust this code");
1341 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1342 "adjust this code");
1343
1344 // The generated handlers do not touch rmethod (the method).
1345 // However, large signatures cannot be cached and are generated
1346 // each time here. The slow-path generator can do a GC on return,
1347 // so we must reload it after the call.
1348 __ blr(t);
1349 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1350
1351
1352 // result handler is in r0
1353 // set result handler
1354 __ mov(result_handler, r0);
1355 // Save it in the frame in case of preemption; we cannot rely on callee saved registers.
1356 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize));
1357
1358 // pass mirror handle if static call
1359 {
1360 Label L;
1361 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1362 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1363 // get mirror
1364 __ load_mirror(t, rmethod, r10, rscratch2);
1365 // copy mirror into activation frame
1366 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1367 // pass handle to mirror
1368 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1369 __ bind(L);
1370 }
1371
1372 // get native function entry point in r10
1373 {
1374 Label L;
1375 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1376 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1377 __ lea(rscratch2, unsatisfied);
1378 __ ldr(rscratch2, rscratch2);
1379 __ cmp(r10, rscratch2);
1380 __ br(Assembler::NE, L);
1381 __ call_VM(noreg,
1382 CAST_FROM_FN_PTR(address,
1383 InterpreterRuntime::prepare_native_call),
1384 rmethod);
1385 __ get_method(rmethod);
1386 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1387 __ bind(L);
1388 }
1389
1390 // pass JNIEnv
1391 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1392
1393 // It is enough that the pc() points into the right code
1394 // segment. It does not have to be the correct return pc.
1395 // For convenience we use the pc we want to resume to in
1396 // case of preemption on Object.wait.
1397 Label native_return;
1398 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1399
1400 // change thread state
1401 #ifdef ASSERT
1402 {
1403 Label L;
1404 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1405 __ cmp(t, (u1)_thread_in_Java);
1406 __ br(Assembler::EQ, L);
1407 __ stop("Wrong thread state in native stub");
1408 __ bind(L);
1409 }
1410 #endif
1411
1412 // Change state to native
1413 __ mov(rscratch1, _thread_in_native);
1414 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1415 __ stlrw(rscratch1, rscratch2);
1416
1417 __ push_cont_fastpath();
1418
1419 // Call the native method.
1420 __ blr(r10);
1421
1422 __ pop_cont_fastpath();
1423
1424 __ get_method(rmethod);
1425 // result potentially in r0 or v0
1426
1427 // Restore cpu control state after JNI call
1428 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1429
1430 // make room for the pushes we're about to do
1431 __ sub(rscratch1, esp, 4 * wordSize);
1432 __ andr(sp, rscratch1, -16);
1433
1434 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1435 // in order to extract the result of a method call. If the order of these
1436 // pushes change or anything else is added to the stack then the code in
1437 // interpreter_frame_result must also change.
1438 __ push(dtos);
1439 __ push(ltos);
1440
1441 __ verify_sve_vector_length();
1442
1443 // change thread state
1444 __ mov(rscratch1, _thread_in_native_trans);
1445 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1446 __ stlrw(rscratch1, rscratch2);
1447
1448 // Force this write out before the read below
1449 if (!UseSystemMemoryBarrier) {
1450 __ dmb(Assembler::ISH);
1451 }
1452
1453 // check for safepoint operation in progress and/or pending suspend requests
1454 {
1455 Label L, Continue;
1456
1457 // No need for acquire as Java threads always disarm themselves.
1458 __ safepoint_poll(L, true /* at_return */, false /* acquire */, false /* in_nmethod */);
1459 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1460 __ cbz(rscratch2, Continue);
1461 __ bind(L);
1462
1463 // Don't use call_VM as it will see a possible pending exception
1464 // and forward it and never return here preventing us from
1465 // clearing _last_native_pc down below. So we do a runtime call by
1466 // hand.
1467 //
1468 __ mov(c_rarg0, rthread);
1469 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1470 __ blr(rscratch2);
1471 __ get_method(rmethod);
1472 __ reinit_heapbase();
1473 __ bind(Continue);
1474 }
1475
1476 // change thread state
1477 __ mov(rscratch1, _thread_in_Java);
1478 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1479 __ stlrw(rscratch1, rscratch2);
1480
1481 if (LockingMode != LM_LEGACY) {
1482 // Check preemption for Object.wait()
1483 Label not_preempted;
1484 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1485 __ cbz(rscratch1, not_preempted);
1486 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1487 __ br(rscratch1);
1488 __ bind(native_return);
1489 __ restore_after_resume(true /* is_native */);
1490 // reload result_handler
1491 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize));
1492 __ bind(not_preempted);
1493 } else {
1494 // any pc will do so just use this one for LM_LEGACY to keep code together.
1495 __ bind(native_return);
1496 }
1497
1498 // reset_last_Java_frame
1499 __ reset_last_Java_frame(true);
1500
1501 if (CheckJNICalls) {
1502 // clear_pending_jni_exception_check
1503 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1504 }
1505
1506 // reset handle block
1507 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1508 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1509
1510 // If result is an oop unbox and store it in frame where gc will see it
1511 // and result handler will pick it up
1512
1513 {
1514 Label no_oop;
1515 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1516 __ cmp(t, result_handler);
1517 __ br(Assembler::NE, no_oop);
1518 // Unbox oop result, e.g. JNIHandles::resolve result.
1519 __ pop(ltos);
1520 __ resolve_jobject(r0, t, rscratch2);
1521 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1522 // keep stack depth as expected by pushing oop which will eventually be discarded
1523 __ push(ltos);
1524 __ bind(no_oop);
1525 }
1526
1527 {
1528 Label no_reguard;
1529 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1530 __ ldrw(rscratch1, Address(rscratch1));
1531 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1532 __ br(Assembler::NE, no_reguard);
1533
1534 __ push_call_clobbered_registers();
1535 __ mov(c_rarg0, rthread);
1536 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1537 __ blr(rscratch2);
1538 __ pop_call_clobbered_registers();
1539
1540 __ bind(no_reguard);
1541 }
1542
1543 // The method register is junk from after the thread_in_native transition
1544 // until here. Also can't call_VM until the bcp has been
1545 // restored. Need bcp for throwing exception below so get it now.
1546 __ get_method(rmethod);
1547
1548 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1549 // rbcp == code_base()
1550 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1551 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1552 // handle exceptions (exception handling will handle unlocking!)
1553 {
1554 Label L;
1555 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1556 __ cbz(rscratch1, L);
1557 // Note: At some point we may want to unify this with the code
1558 // used in call_VM_base(); i.e., we should use the
1559 // StubRoutines::forward_exception code. For now this doesn't work
1560 // here because the rsp is not correctly set at this point.
1561 __ MacroAssembler::call_VM(noreg,
1562 CAST_FROM_FN_PTR(address,
1563 InterpreterRuntime::throw_pending_exception));
1564 __ should_not_reach_here();
1565 __ bind(L);
1566 }
1567
1568 // do unlocking if necessary
1569 {
1570 Label L;
1571 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1572 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1573 // the code below should be shared with interpreter macro
1574 // assembler implementation
1575 {
1576 Label unlock;
1577 // BasicObjectLock will be first in list, since this is a
1578 // synchronized method. However, need to check that the object
1579 // has not been unlocked by an explicit monitorexit bytecode.
1580
1581 // monitor expect in c_rarg1 for slow unlock path
1582 __ lea (c_rarg1, Address(rfp, // address of first monitor
1583 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1584 wordSize - sizeof(BasicObjectLock))));
1585
1586 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1587 __ cbnz(t, unlock);
1588
1589 // Entry already unlocked, need to throw exception
1590 __ MacroAssembler::call_VM(noreg,
1591 CAST_FROM_FN_PTR(address,
1592 InterpreterRuntime::throw_illegal_monitor_state_exception));
1593 __ should_not_reach_here();
1594
1595 __ bind(unlock);
1596 __ unlock_object(c_rarg1);
1597 }
1598 __ bind(L);
1599 }
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 // restore sender sp
1624 __ mov(sp, esp);
1625
1626 __ ret(lr);
1627
1628 if (inc_counter) {
1629 // Handle overflow of counter and compile method
1630 __ bind(invocation_counter_overflow);
1631 generate_counter_overflow(continue_after_compile);
1632 }
1633
1634 return entry_point;
1635 }
1636
1637 //
1638 // Generic interpreted method entry to (asm) interpreter
1639 //
1640 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1641 // determine code generation flags
1642 bool inc_counter = UseCompiler || CountCompiledCalls;
1643
1644 // rscratch1: sender sp
1645 address entry_point = __ pc();
1646
1647 const Address constMethod(rmethod, Method::const_offset());
1648 const Address access_flags(rmethod, Method::access_flags_offset());
1649 const Address size_of_parameters(r3,
1650 ConstMethod::size_of_parameters_offset());
1651 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1652
1653 // get parameter size (always needed)
1654 // need to load the const method first
1655 __ ldr(r3, constMethod);
1656 __ load_unsigned_short(r2, size_of_parameters);
1657
1658 // r2: size of parameters
1659
1660 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1661 __ sub(r3, r3, r2); // r3 = no. of additional locals
1662
1663 // see if we've got enough room on the stack for locals plus overhead.
1664 generate_stack_overflow_check();
1665
1666 // compute beginning of parameters (rlocals)
1667 __ add(rlocals, esp, r2, ext::uxtx, 3);
1668 __ sub(rlocals, rlocals, wordSize);
1669
1670 __ mov(rscratch1, esp);
1671
1672 // r3 - # of additional locals
1673 // allocate space for locals
1674 // explicitly initialize locals
1675 // Initializing memory allocated for locals in the same direction as
1676 // the stack grows to ensure page initialization order according
1677 // to windows-aarch64 stack page growth requirement (see
1678 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1679 {
1680 Label exit, loop;
1681 __ ands(zr, r3, r3);
1682 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1683 __ bind(loop);
1684 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1685 __ sub(r3, r3, 1); // until everything initialized
1686 __ cbnz(r3, loop);
1687 __ bind(exit);
1688 }
1689
1690 // Padding between locals and fixed part of activation frame to ensure
1691 // SP is always 16-byte aligned.
1692 __ andr(sp, rscratch1, -16);
1693
1694 // And the base dispatch table
1695 __ get_dispatch();
1696
1697 // initialize fixed part of activation frame
1698 generate_fixed_frame(false);
1699
1700 // make sure method is not native & not abstract
1701 #ifdef ASSERT
1702 __ ldrw(r0, access_flags);
1703 {
1704 Label L;
1705 __ tst(r0, JVM_ACC_NATIVE);
1706 __ br(Assembler::EQ, L);
1707 __ stop("tried to execute native method as non-native");
1708 __ bind(L);
1709 }
1710 {
1711 Label L;
1712 __ tst(r0, JVM_ACC_ABSTRACT);
1713 __ br(Assembler::EQ, L);
1714 __ stop("tried to execute abstract method in interpreter");
1715 __ bind(L);
1716 }
1717 #endif
1718
1719 // Since at this point in the method invocation the exception
1720 // handler would try to exit the monitor of synchronized methods
1721 // which hasn't been entered yet, we set the thread local variable
1722 // _do_not_unlock_if_synchronized to true. The remove_activation
1723 // will check this flag.
1724
1725 const Address do_not_unlock_if_synchronized(rthread,
1726 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1727 __ mov(rscratch2, true);
1728 __ strb(rscratch2, do_not_unlock_if_synchronized);
1729
1730 Register mdp = r3;
1731 __ profile_parameters_type(mdp, r1, r2);
1732
1733 // increment invocation count & check for overflow
1734 Label invocation_counter_overflow;
1735 if (inc_counter) {
1736 generate_counter_incr(&invocation_counter_overflow);
1737 }
1738
1739 Label continue_after_compile;
1740 __ bind(continue_after_compile);
1741
1742 bang_stack_shadow_pages(false);
1743
1744 // reset the _do_not_unlock_if_synchronized flag
1745 __ strb(zr, do_not_unlock_if_synchronized);
1746
1747 // check for synchronized methods
1748 // Must happen AFTER invocation_counter check and stack overflow check,
1749 // so method is not locked if overflows.
1750 if (synchronized) {
1751 // Allocate monitor and lock method
1752 lock_method();
1753 } else {
1754 // no synchronization necessary
1755 #ifdef ASSERT
1756 {
1757 Label L;
1758 __ ldrw(r0, access_flags);
1759 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1760 __ br(Assembler::EQ, L);
1761 __ stop("method needs synchronization");
1762 __ bind(L);
1763 }
1764 #endif
1765 }
1766
1767 // start execution
1768 #ifdef ASSERT
1769 {
1770 Label L;
1771 const Address monitor_block_top (rfp,
1772 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1773 __ ldr(rscratch1, monitor_block_top);
1774 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1775 __ cmp(esp, rscratch1);
1776 __ br(Assembler::EQ, L);
1777 __ stop("broken stack frame setup in interpreter 2");
1778 __ bind(L);
1779 }
1780 #endif
1781
1782 // jvmti support
1783 __ notify_method_entry();
1784
1785 // runtime upcalls
1786 if (runtime_upcalls) {
1787 __ generate_runtime_upcalls_on_method_entry();
1788 }
1789
1790 __ dispatch_next(vtos);
1791
1792 // invocation counter overflow
1793 if (inc_counter) {
1794 // Handle overflow of counter and compile method
1795 __ bind(invocation_counter_overflow);
1796 generate_counter_overflow(continue_after_compile);
1797 }
1798
1799 return entry_point;
1800 }
1801
1802 // Method entry for java.lang.Thread.currentThread
1803 address TemplateInterpreterGenerator::generate_currentThread() {
1804 address entry_point = __ pc();
1805
1806 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1807 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1808 __ ret(lr);
1809
1810 return entry_point;
1811 }
1812
1813 // Not supported
1814 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1815 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1816 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1817 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1818
1819 //-----------------------------------------------------------------------------
1820 // Exceptions
1821
1822 void TemplateInterpreterGenerator::generate_throw_exception() {
1823 // Entry point in previous activation (i.e., if the caller was
1824 // interpreted)
1825 Interpreter::_rethrow_exception_entry = __ pc();
1826 // Restore sp to interpreter_frame_last_sp even though we are going
1827 // to empty the expression stack for the exception processing.
1828 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1829 // r0: exception
1830 // r3: return address/pc that threw exception
1831 __ restore_bcp(); // rbcp points to call/send
1832 __ restore_locals();
1833 __ restore_constant_pool_cache();
1834 __ reinit_heapbase(); // restore rheapbase as heapbase.
1835 __ get_dispatch();
1836
1837 // Entry point for exceptions thrown within interpreter code
1838 Interpreter::_throw_exception_entry = __ pc();
1839 // If we came here via a NullPointerException on the receiver of a
1840 // method, rmethod may be corrupt.
1841 __ get_method(rmethod);
1842 // expression stack is undefined here
1843 // r0: exception
1844 // rbcp: exception bcp
1845 __ verify_oop(r0);
1846 __ mov(c_rarg1, r0);
1847
1848 // expression stack must be empty before entering the VM in case of
1849 // an exception
1850 __ empty_expression_stack();
1851 // find exception handler address and preserve exception oop
1852 __ call_VM(r3,
1853 CAST_FROM_FN_PTR(address,
1854 InterpreterRuntime::exception_handler_for_exception),
1855 c_rarg1);
1856
1857 // Restore machine SP
1858 __ restore_sp_after_call();
1859
1860 // r0: exception handler entry point
1861 // r3: preserved exception oop
1862 // rbcp: bcp for exception handler
1863 __ push_ptr(r3); // push exception which is now the only value on the stack
1864 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1865
1866 // If the exception is not handled in the current frame the frame is
1867 // removed and the exception is rethrown (i.e. exception
1868 // continuation is _rethrow_exception).
1869 //
1870 // Note: At this point the bci is still the bxi for the instruction
1871 // which caused the exception and the expression stack is
1872 // empty. Thus, for any VM calls at this point, GC will find a legal
1873 // oop map (with empty expression stack).
1874
1875 //
1876 // JVMTI PopFrame support
1877 //
1878
1879 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1880 __ empty_expression_stack();
1881 // Set the popframe_processing bit in pending_popframe_condition
1882 // indicating that we are currently handling popframe, so that
1883 // call_VMs that may happen later do not trigger new popframe
1884 // handling cycles.
1885 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1886 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1887 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1888
1889 {
1890 // Check to see whether we are returning to a deoptimized frame.
1891 // (The PopFrame call ensures that the caller of the popped frame is
1892 // either interpreted or compiled and deoptimizes it if compiled.)
1893 // In this case, we can't call dispatch_next() after the frame is
1894 // popped, but instead must save the incoming arguments and restore
1895 // them after deoptimization has occurred.
1896 //
1897 // Note that we don't compare the return PC against the
1898 // deoptimization blob's unpack entry because of the presence of
1899 // adapter frames in C2.
1900 Label caller_not_deoptimized;
1901 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1902 // This is a return address, so requires authenticating for PAC.
1903 __ authenticate_return_address(c_rarg1);
1904 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1905 InterpreterRuntime::interpreter_contains), c_rarg1);
1906 __ cbnz(r0, caller_not_deoptimized);
1907
1908 // Compute size of arguments for saving when returning to
1909 // deoptimized caller
1910 __ get_method(r0);
1911 __ ldr(r0, Address(r0, Method::const_offset()));
1912 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1913 size_of_parameters_offset())));
1914 __ lsl(r0, r0, Interpreter::logStackElementSize);
1915 __ restore_locals(); // XXX do we need this?
1916 __ sub(rlocals, rlocals, r0);
1917 __ add(rlocals, rlocals, wordSize);
1918 // Save these arguments
1919 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1920 Deoptimization::
1921 popframe_preserve_args),
1922 rthread, r0, rlocals);
1923
1924 __ remove_activation(vtos,
1925 /* throw_monitor_exception */ false,
1926 /* install_monitor_exception */ false,
1927 /* notify_jvmdi */ false);
1928
1929 // Inform deoptimization that it is responsible for restoring
1930 // these arguments
1931 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1932 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1933
1934 // Continue in deoptimization handler
1935 __ ret(lr);
1936
1937 __ bind(caller_not_deoptimized);
1938 }
1939
1940 __ remove_activation(vtos,
1941 /* throw_monitor_exception */ false,
1942 /* install_monitor_exception */ false,
1943 /* notify_jvmdi */ false);
1944
1945 // Restore the last_sp and null it out
1946 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1947 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1948 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1949
1950 __ restore_bcp();
1951 __ restore_locals();
1952 __ restore_constant_pool_cache();
1953 __ get_method(rmethod);
1954 __ get_dispatch();
1955
1956 // The method data pointer was incremented already during
1957 // call profiling. We have to restore the mdp for the current bcp.
1958 if (ProfileInterpreter) {
1959 __ set_method_data_pointer_for_bcp();
1960 }
1961
1962 // Clear the popframe condition flag
1963 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1964 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1965
1966 #if INCLUDE_JVMTI
1967 {
1968 Label L_done;
1969
1970 __ ldrb(rscratch1, Address(rbcp, 0));
1971 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1972 __ br(Assembler::NE, L_done);
1973
1974 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1975 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1976
1977 __ ldr(c_rarg0, Address(rlocals, 0));
1978 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1979
1980 __ cbz(r0, L_done);
1981
1982 __ str(r0, Address(esp, 0));
1983 __ bind(L_done);
1984 }
1985 #endif // INCLUDE_JVMTI
1986
1987 // Restore machine SP
1988 __ restore_sp_after_call();
1989
1990 __ dispatch_next(vtos);
1991 // end of PopFrame support
1992
1993 Interpreter::_remove_activation_entry = __ pc();
1994
1995 // preserve exception over this code sequence
1996 __ pop_ptr(r0);
1997 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1998 // remove the activation (without doing throws on illegalMonitorExceptions)
1999 __ remove_activation(vtos, false, true, false);
2000 // restore exception
2001 __ get_vm_result(r0, rthread);
2002
2003 // In between activations - previous activation type unknown yet
2004 // compute continuation point - the continuation point expects the
2005 // following registers set up:
2006 //
2007 // r0: exception
2008 // lr: return address/pc that threw exception
2009 // esp: expression stack of caller
2010 // rfp: fp of caller
2011 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
2012 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2013 SharedRuntime::exception_handler_for_return_address),
2014 rthread, lr);
2015 __ mov(r1, r0); // save exception handler
2016 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
2017 // We might be returning to a deopt handler that expects r3 to
2018 // contain the exception pc
2019 __ mov(r3, lr);
2020 // Note that an "issuing PC" is actually the next PC after the call
2021 __ br(r1); // jump to exception
2022 // handler of caller
2023 }
2024
2025
2026 //
2027 // JVMTI ForceEarlyReturn support
2028 //
2029 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2030 address entry = __ pc();
2031
2032 __ restore_bcp();
2033 __ restore_locals();
2034 __ empty_expression_stack();
2035 __ load_earlyret_value(state);
2036
2037 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
2038 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
2039
2040 // Clear the earlyret state
2041 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
2042 __ str(zr, cond_addr);
2043
2044 __ remove_activation(state,
2045 false, /* throw_monitor_exception */
2046 false, /* install_monitor_exception */
2047 true); /* notify_jvmdi */
2048 __ ret(lr);
2049
2050 return entry;
2051 } // end of ForceEarlyReturn support
2052
2053
2054
2055 //-----------------------------------------------------------------------------
2056 // Helper for vtos entry point generation
2057
2058 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2059 address& bep,
2060 address& cep,
2061 address& sep,
2062 address& aep,
2063 address& iep,
2064 address& lep,
2065 address& fep,
2066 address& dep,
2067 address& vep) {
2068 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2069 Label L;
2070 aep = __ pc(); // atos entry point
2071 __ push_ptr();
2072 __ b(L);
2073 fep = __ pc(); // ftos entry point
2074 __ push_f();
2075 __ b(L);
2076 dep = __ pc(); // dtos entry point
2077 __ push_d();
2078 __ b(L);
2079 lep = __ pc(); // ltos entry point
2080 __ push_l();
2081 __ b(L);
2082 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
2083 __ push_i();
2084 vep = __ pc(); // vtos entry point
2085 __ bind(L);
2086 generate_and_dispatch(t);
2087 }
2088
2089 //-----------------------------------------------------------------------------
2090
2091 void TemplateInterpreterGenerator::count_bytecode() {
2092 if (CountBytecodesPerThread) {
2093 Address bc_counter_addr(rthread, Thread::bc_counter_offset());
2094 __ ldr(r10, bc_counter_addr);
2095 __ add(r10, r10, 1);
2096 __ str(r10, bc_counter_addr);
2097 }
2098 if (CountBytecodes || TraceBytecodes || StopInterpreterAt > 0) {
2099 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2100 __ atomic_add(noreg, 1, r10);
2101 }
2102 }
2103
2104 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2105 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2106 __ atomic_addw(noreg, 1, r10);
2107 }
2108
2109 // Non-product code
2110 #ifndef PRODUCT
2111 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2112 address entry = __ pc();
2113
2114 __ protect_return_address();
2115 __ push(lr);
2116 __ push(state);
2117 __ push(RegSet::range(r0, r15), sp);
2118 __ mov(c_rarg2, r0); // Pass itos
2119 __ call_VM(noreg,
2120 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2121 c_rarg1, c_rarg2, c_rarg3);
2122 __ pop(RegSet::range(r0, r15), sp);
2123 __ pop(state);
2124 __ pop(lr);
2125 __ authenticate_return_address();
2126 __ ret(lr); // return from result handler
2127
2128 return entry;
2129 }
2130
2131 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2132 // Calculate new index for counter:
2133 // _index = (_index >> log2_number_of_codes) |
2134 // (bytecode << log2_number_of_codes);
2135 Register index_addr = rscratch1;
2136 Register index = rscratch2;
2137 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2138 __ ldrw(index, index_addr);
2139 __ mov(r10,
2140 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2141 __ orrw(index, r10, index, Assembler::LSR,
2142 BytecodePairHistogram::log2_number_of_codes);
2143 __ strw(index, index_addr);
2144
2145 // Bump bucket contents:
2146 // _counters[_index] ++;
2147 Register counter_addr = rscratch1;
2148 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2149 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2150 __ atomic_addw(noreg, 1, counter_addr);
2151 }
2152
2153 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2154 // Call a little run-time stub to avoid blow-up for each bytecode.
2155 // The run-time runtime saves the right registers, depending on
2156 // the tosca in-state for the given template.
2157
2158 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2159 "entry must have been generated");
2160 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2161 __ reinit_heapbase();
2162 }
2163
2164
2165 void TemplateInterpreterGenerator::stop_interpreter_at() {
2166 Label L;
2167 __ push(rscratch1);
2168 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2169 __ ldr(rscratch1, Address(rscratch1));
2170 __ mov(rscratch2, StopInterpreterAt);
2171 __ cmpw(rscratch1, rscratch2);
2172 __ br(Assembler::NE, L);
2173 __ brk(0);
2174 __ bind(L);
2175 __ pop(rscratch1);
2176 }
2177
2178 #endif // !PRODUCT