1 /*
2 * Copyright (c) 2003, 2023, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "compiler/compiler_globals.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "interpreter/templateInterpreterGenerator.hpp"
36 #include "interpreter/templateTable.hpp"
37 #include "interpreter/bytecodeTracer.hpp"
38 #include "memory/resourceArea.hpp"
39 #include "oops/arrayOop.hpp"
40 #include "oops/method.hpp"
41 #include "oops/methodCounters.hpp"
42 #include "oops/methodData.hpp"
43 #include "oops/oop.inline.hpp"
44 #include "oops/resolvedIndyEntry.hpp"
45 #include "oops/resolvedMethodEntry.hpp"
46 #include "prims/jvmtiExport.hpp"
47 #include "prims/jvmtiThreadState.hpp"
48 #include "runtime/arguments.hpp"
49 #include "runtime/deoptimization.hpp"
50 #include "runtime/frame.inline.hpp"
51 #include "runtime/globals.hpp"
52 #include "runtime/jniHandles.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/synchronizer.hpp"
56 #include "runtime/timer.hpp"
57 #include "runtime/vframeArray.hpp"
58 #include "utilities/checkedCast.hpp"
59 #include "utilities/debug.hpp"
60 #include "utilities/powerOfTwo.hpp"
61 #include <sys/types.h>
62
63 // Size of interpreter code. Increase if too small. Interpreter will
64 // fail with a guarantee ("not enough space for interpreter generation");
65 // if too small.
66 // Run with +PrintInterpreter to get the VM to print out the size.
67 // Max size with JVMTI
68 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
69
70 #define __ _masm->
71
72 //-----------------------------------------------------------------------------
73
74 extern "C" void entry(CodeBuffer*);
75
76 //-----------------------------------------------------------------------------
77
78 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
79 address entry = __ pc();
80
81 __ andr(esp, esp, -16);
82 __ mov(c_rarg3, esp);
83 // rmethod
84 // rlocals
85 // c_rarg3: first stack arg - wordSize
86
87 // adjust sp
88 __ sub(sp, c_rarg3, 18 * wordSize);
89 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
90 __ call_VM(noreg,
91 CAST_FROM_FN_PTR(address,
92 InterpreterRuntime::slow_signature_handler),
93 rmethod, rlocals, c_rarg3);
94
95 // r0: result handler
96
97 // Stack layout:
98 // rsp: return address <- sp
99 // 1 garbage
100 // 8 integer args (if static first is unused)
101 // 1 float/double identifiers
102 // 8 double args
103 // stack args <- esp
104 // garbage
105 // expression stack bottom
106 // bcp (null)
107 // ...
108
109 // Restore LR
110 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
111
112 // Do FP first so we can use c_rarg3 as temp
113 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
114
115 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
116 const FloatRegister r = as_FloatRegister(i);
117
118 Label d, done;
119
120 __ tbnz(c_rarg3, i, d);
121 __ ldrs(r, Address(sp, (10 + i) * wordSize));
122 __ b(done);
123 __ bind(d);
124 __ ldrd(r, Address(sp, (10 + i) * wordSize));
125 __ bind(done);
126 }
127
128 // c_rarg0 contains the result from the call of
129 // InterpreterRuntime::slow_signature_handler so we don't touch it
130 // here. It will be loaded with the JNIEnv* later.
131 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
132 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
133 Register rm = as_Register(i), rn = as_Register(i+1);
134 __ ldp(rm, rn, Address(sp, i * wordSize));
135 }
136
137 __ add(sp, sp, 18 * wordSize);
138 __ ret(lr);
139
140 return entry;
141 }
142
143
144 //
145 // Various method entries
146 //
147
148 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
149 // rmethod: Method*
150 // r19_sender_sp: sender sp
151 // esp: args
152
153 // These don't need a safepoint check because they aren't virtually
154 // callable. We won't enter these intrinsics from compiled code.
155 // If in the future we added an intrinsic which was virtually callable
156 // we'd have to worry about how to safepoint so that this code is used.
157
158 // mathematical functions inlined by compiler
159 // (interpreter must provide identical implementation
160 // in order to avoid monotonicity bugs when switching
161 // from interpreter to compiler in the middle of some
162 // computation)
163 //
164 // stack:
165 // [ arg ] <-- esp
166 // [ arg ]
167 // retaddr in lr
168
169 address entry_point = nullptr;
170 Register continuation = lr;
171 switch (kind) {
172 case Interpreter::java_lang_math_abs:
173 entry_point = __ pc();
174 __ ldrd(v0, Address(esp));
175 __ fabsd(v0, v0);
176 __ mov(sp, r19_sender_sp); // Restore caller's SP
177 break;
178 case Interpreter::java_lang_math_sqrt:
179 entry_point = __ pc();
180 __ ldrd(v0, Address(esp));
181 __ fsqrtd(v0, v0);
182 __ mov(sp, r19_sender_sp);
183 break;
184 case Interpreter::java_lang_math_sin :
185 case Interpreter::java_lang_math_cos :
186 case Interpreter::java_lang_math_tan :
187 case Interpreter::java_lang_math_log :
188 case Interpreter::java_lang_math_log10 :
189 case Interpreter::java_lang_math_exp :
190 entry_point = __ pc();
191 __ ldrd(v0, Address(esp));
192 __ mov(sp, r19_sender_sp);
193 __ mov(r23, lr);
194 continuation = r23; // The first free callee-saved register
195 generate_transcendental_entry(kind, 1);
196 break;
197 case Interpreter::java_lang_math_pow :
198 entry_point = __ pc();
199 __ mov(r23, lr);
200 continuation = r23;
201 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
202 __ ldrd(v1, Address(esp));
203 __ mov(sp, r19_sender_sp);
204 generate_transcendental_entry(kind, 2);
205 break;
206 case Interpreter::java_lang_math_fmaD :
207 if (UseFMA) {
208 entry_point = __ pc();
209 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
210 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
211 __ ldrd(v2, Address(esp));
212 __ fmaddd(v0, v0, v1, v2);
213 __ mov(sp, r19_sender_sp); // Restore caller's SP
214 }
215 break;
216 case Interpreter::java_lang_math_fmaF :
217 if (UseFMA) {
218 entry_point = __ pc();
219 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
220 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
221 __ ldrs(v2, Address(esp));
222 __ fmadds(v0, v0, v1, v2);
223 __ mov(sp, r19_sender_sp); // Restore caller's SP
224 }
225 break;
226 default:
227 ;
228 }
229 if (entry_point) {
230 __ br(continuation);
231 }
232
233 return entry_point;
234 }
235
236 // double trigonometrics and transcendentals
237 // static jdouble dsin(jdouble x);
238 // static jdouble dcos(jdouble x);
239 // static jdouble dtan(jdouble x);
240 // static jdouble dlog(jdouble x);
241 // static jdouble dlog10(jdouble x);
242 // static jdouble dexp(jdouble x);
243 // static jdouble dpow(jdouble x, jdouble y);
244
245 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
246 address fn;
247 switch (kind) {
248 case Interpreter::java_lang_math_sin :
249 if (StubRoutines::dsin() == nullptr) {
250 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
251 } else {
252 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
253 }
254 break;
255 case Interpreter::java_lang_math_cos :
256 if (StubRoutines::dcos() == nullptr) {
257 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
258 } else {
259 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
260 }
261 break;
262 case Interpreter::java_lang_math_tan :
263 if (StubRoutines::dtan() == nullptr) {
264 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
265 } else {
266 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
267 }
268 break;
269 case Interpreter::java_lang_math_log :
270 if (StubRoutines::dlog() == nullptr) {
271 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
272 } else {
273 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
274 }
275 break;
276 case Interpreter::java_lang_math_log10 :
277 if (StubRoutines::dlog10() == nullptr) {
278 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
279 } else {
280 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
281 }
282 break;
283 case Interpreter::java_lang_math_exp :
284 if (StubRoutines::dexp() == nullptr) {
285 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
286 } else {
287 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
288 }
289 break;
290 case Interpreter::java_lang_math_pow :
291 if (StubRoutines::dpow() == nullptr) {
292 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
293 } else {
294 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
295 }
296 break;
297 default:
298 ShouldNotReachHere();
299 fn = nullptr; // unreachable
300 }
301 __ mov(rscratch1, fn);
302 __ blr(rscratch1);
303 }
304
305 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
306 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
307 // r19_sender_sp: sender sp
308 // stack:
309 // [ arg ] <-- esp
310 // [ arg ]
311 // retaddr in lr
312 // result in v0
313
314 address entry_point = __ pc();
315 __ ldrw(c_rarg0, Address(esp));
316 __ flt16_to_flt(v0, c_rarg0, v1);
317 __ mov(sp, r19_sender_sp); // Restore caller's SP
318 __ br(lr);
319 return entry_point;
320 }
321
322 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
323 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
324 // r19_sender_sp: sender sp
325 // stack:
326 // [ arg ] <-- esp
327 // [ arg ]
328 // retaddr in lr
329 // result in c_rarg0
330
331 address entry_point = __ pc();
332 __ ldrs(v0, Address(esp));
333 __ flt_to_flt16(c_rarg0, v0, v1);
334 __ mov(sp, r19_sender_sp); // Restore caller's SP
335 __ br(lr);
336 return entry_point;
337 }
338
339 // Abstract method entry
340 // Attempt to execute abstract method. Throw exception
341 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
342 // rmethod: Method*
343 // r19_sender_sp: sender SP
344
345 address entry_point = __ pc();
346
347 // abstract method entry
348
349 // pop return address, reset last_sp to null
350 __ empty_expression_stack();
351 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
352 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
353
354 // throw exception
355 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
356 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
357 rmethod);
358 // the call_VM checks for exception, so we should never return here.
359 __ should_not_reach_here();
360
361 return entry_point;
362 }
363
364 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
365 address entry = __ pc();
366
367 #ifdef ASSERT
368 {
369 Label L;
370 __ ldr(rscratch1, Address(rfp,
371 frame::interpreter_frame_monitor_block_top_offset *
372 wordSize));
373 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
374 __ mov(rscratch2, sp);
375 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
376 // grows negative)
377 __ br(Assembler::HS, L); // check if frame is complete
378 __ stop ("interpreter frame not set up");
379 __ bind(L);
380 }
381 #endif // ASSERT
382 // Restore bcp under the assumption that the current frame is still
383 // interpreted
384 __ restore_bcp();
385
386 // expression stack must be empty before entering the VM if an
387 // exception happened
388 __ empty_expression_stack();
389 // throw exception
390 __ call_VM(noreg,
391 CAST_FROM_FN_PTR(address,
392 InterpreterRuntime::throw_StackOverflowError));
393 return entry;
394 }
395
396 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
397 address entry = __ pc();
398 // expression stack must be empty before entering the VM if an
399 // exception happened
400 __ empty_expression_stack();
401 // setup parameters
402
403 // ??? convention: expect aberrant index in register r1
404 __ movw(c_rarg2, r1);
405 // ??? convention: expect array in register r3
406 __ mov(c_rarg1, r3);
407 __ call_VM(noreg,
408 CAST_FROM_FN_PTR(address,
409 InterpreterRuntime::
410 throw_ArrayIndexOutOfBoundsException),
411 c_rarg1, c_rarg2);
412 return entry;
413 }
414
415 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
416 address entry = __ pc();
417
418 // object is at TOS
419 __ pop(c_rarg1);
420
421 // expression stack must be empty before entering the VM if an
422 // exception happened
423 __ empty_expression_stack();
424
425 __ call_VM(noreg,
426 CAST_FROM_FN_PTR(address,
427 InterpreterRuntime::
428 throw_ClassCastException),
429 c_rarg1);
430 return entry;
431 }
432
433 address TemplateInterpreterGenerator::generate_exception_handler_common(
434 const char* name, const char* message, bool pass_oop) {
435 assert(!pass_oop || message == nullptr, "either oop or message but not both");
436 address entry = __ pc();
437 if (pass_oop) {
438 // object is at TOS
439 __ pop(c_rarg2);
440 }
441 // expression stack must be empty before entering the VM if an
442 // exception happened
443 __ empty_expression_stack();
444 // setup parameters
445 __ lea(c_rarg1, Address((address)name));
446 if (pass_oop) {
447 __ call_VM(r0, CAST_FROM_FN_PTR(address,
448 InterpreterRuntime::
449 create_klass_exception),
450 c_rarg1, c_rarg2);
451 } else {
452 // kind of lame ExternalAddress can't take null because
453 // external_word_Relocation will assert.
454 if (message != nullptr) {
455 __ lea(c_rarg2, Address((address)message));
456 } else {
457 __ mov(c_rarg2, NULL_WORD);
458 }
459 __ call_VM(r0,
460 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
461 c_rarg1, c_rarg2);
462 }
463 // throw exception
464 __ b(address(Interpreter::throw_exception_entry()));
465 return entry;
466 }
467
468 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
469 address entry = __ pc();
470
471 // Restore stack bottom in case i2c adjusted stack
472 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
473 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
474 // and null it as marker that esp is now tos until next java call
475 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
476 __ restore_bcp();
477 __ restore_locals();
478 __ restore_constant_pool_cache();
479 __ get_method(rmethod);
480
481 if (state == atos) {
482 Register obj = r0;
483 Register mdp = r1;
484 Register tmp = r2;
485 __ profile_return_type(mdp, obj, tmp);
486 }
487
488 const Register cache = r1;
489 const Register index = r2;
490
491 if (index_size == sizeof(u4)) {
492 __ load_resolved_indy_entry(cache, index);
493 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
494 __ add(esp, esp, cache, Assembler::LSL, 3);
495 } else {
496 // Pop N words from the stack
497 assert(index_size == sizeof(u2), "Can only be u2");
498 __ load_method_entry(cache, index);
499 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
500 __ add(esp, esp, cache, Assembler::LSL, 3);
501 }
502
503 // Restore machine SP
504 __ restore_sp_after_call();
505
506 __ check_and_handle_popframe(rthread);
507 __ check_and_handle_earlyret(rthread);
508
509 __ get_dispatch();
510 __ dispatch_next(state, step);
511
512 return entry;
513 }
514
515 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
516 int step,
517 address continuation) {
518 address entry = __ pc();
519 __ restore_bcp();
520 __ restore_locals();
521 __ restore_constant_pool_cache();
522 __ get_method(rmethod);
523 __ get_dispatch();
524
525 __ restore_sp_after_call(); // Restore SP to extended SP
526
527 // Restore expression stack pointer
528 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
529 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
530 // null last_sp until next java call
531 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
532
533 #if INCLUDE_JVMCI
534 // Check if we need to take lock at entry of synchronized method. This can
535 // only occur on method entry so emit it only for vtos with step 0.
536 if (EnableJVMCI && state == vtos && step == 0) {
537 Label L;
538 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
539 __ cbz(rscratch1, L);
540 // Clear flag.
541 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
542 // Take lock.
543 lock_method();
544 __ bind(L);
545 } else {
546 #ifdef ASSERT
547 if (EnableJVMCI) {
548 Label L;
549 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
550 __ cbz(rscratch1, L);
551 __ stop("unexpected pending monitor in deopt entry");
552 __ bind(L);
553 }
554 #endif
555 }
556 #endif
557 // handle exceptions
558 {
559 Label L;
560 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
561 __ cbz(rscratch1, L);
562 __ call_VM(noreg,
563 CAST_FROM_FN_PTR(address,
564 InterpreterRuntime::throw_pending_exception));
565 __ should_not_reach_here();
566 __ bind(L);
567 }
568
569 if (continuation == nullptr) {
570 __ dispatch_next(state, step);
571 } else {
572 __ jump_to_entry(continuation);
573 }
574 return entry;
575 }
576
577 address TemplateInterpreterGenerator::generate_result_handler_for(
578 BasicType type) {
579 address entry = __ pc();
580 switch (type) {
581 case T_BOOLEAN: __ c2bool(r0); break;
582 case T_CHAR : __ uxth(r0, r0); break;
583 case T_BYTE : __ sxtb(r0, r0); break;
584 case T_SHORT : __ sxth(r0, r0); break;
585 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
586 case T_LONG : /* nothing to do */ break;
587 case T_VOID : /* nothing to do */ break;
588 case T_FLOAT : /* nothing to do */ break;
589 case T_DOUBLE : /* nothing to do */ break;
590 case T_OBJECT :
591 // retrieve result from frame
592 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
593 // and verify it
594 __ verify_oop(r0);
595 break;
596 default : ShouldNotReachHere();
597 }
598 __ ret(lr); // return from result handler
599 return entry;
600 }
601
602 address TemplateInterpreterGenerator::generate_safept_entry_for(
603 TosState state,
604 address runtime_entry) {
605 address entry = __ pc();
606 __ push(state);
607 __ push_cont_fastpath(rthread);
608 __ call_VM(noreg, runtime_entry);
609 __ pop_cont_fastpath(rthread);
610 __ membar(Assembler::AnyAny);
611 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
612 return entry;
613 }
614
615 // Helpers for commoning out cases in the various type of method entries.
616 //
617
618
619 // increment invocation count & check for overflow
620 //
621 // Note: checking for negative value instead of overflow
622 // so we have a 'sticky' overflow test
623 //
624 // rmethod: method
625 //
626 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
627 Label done;
628 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
629 int increment = InvocationCounter::count_increment;
630 Label no_mdo;
631 if (ProfileInterpreter) {
632 // Are we profiling?
633 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
634 __ cbz(r0, no_mdo);
635 // Increment counter in the MDO
636 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
637 in_bytes(InvocationCounter::counter_offset()));
638 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
639 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
640 __ b(done);
641 }
642 __ bind(no_mdo);
643 // Increment counter in MethodCounters
644 const Address invocation_counter(rscratch2,
645 MethodCounters::invocation_counter_offset() +
646 InvocationCounter::counter_offset());
647 __ get_method_counters(rmethod, rscratch2, done);
648 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
649 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
650 __ bind(done);
651 }
652
653 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
654
655 // Asm interpreter on entry
656 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
657 // Everything as it was on entry
658
659 // InterpreterRuntime::frequency_counter_overflow takes two
660 // arguments, the first (thread) is passed by call_VM, the second
661 // indicates if the counter overflow occurs at a backwards branch
662 // (null bcp). We pass zero for it. The call returns the address
663 // of the verified entry point for the method or null if the
664 // compilation did not complete (either went background or bailed
665 // out).
666 __ mov(c_rarg1, 0);
667 __ call_VM(noreg,
668 CAST_FROM_FN_PTR(address,
669 InterpreterRuntime::frequency_counter_overflow),
670 c_rarg1);
671
672 __ b(do_continue);
673 }
674
675 // See if we've got enough room on the stack for locals plus overhead
676 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
677 // without going through the signal handler, i.e., reserved and yellow zones
678 // will not be made usable. The shadow zone must suffice to handle the
679 // overflow.
680 // The expression stack grows down incrementally, so the normal guard
681 // page mechanism will work for that.
682 //
683 // NOTE: Since the additional locals are also always pushed (wasn't
684 // obvious in generate_method_entry) so the guard should work for them
685 // too.
686 //
687 // Args:
688 // r3: number of additional locals this frame needs (what we must check)
689 // rmethod: Method*
690 //
691 // Kills:
692 // r0
693 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
694
695 // monitor entry size: see picture of stack set
696 // (generate_method_entry) and frame_amd64.hpp
697 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
698
699 // total overhead size: entry_size + (saved rbp through expr stack
700 // bottom). be sure to change this if you add/subtract anything
701 // to/from the overhead area
702 const int overhead_size =
703 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
704
705 const size_t page_size = os::vm_page_size();
706
707 Label after_frame_check;
708
709 // see if the frame is greater than one page in size. If so,
710 // then we need to verify there is enough stack space remaining
711 // for the additional locals.
712 //
713 // Note that we use SUBS rather than CMP here because the immediate
714 // field of this instruction may overflow. SUBS can cope with this
715 // because it is a macro that will expand to some number of MOV
716 // instructions and a register operation.
717 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
718 __ br(Assembler::LS, after_frame_check);
719
720 // compute rsp as if this were going to be the last frame on
721 // the stack before the red zone
722
723 // locals + overhead, in bytes
724 __ mov(r0, overhead_size);
725 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
726
727 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
728 __ ldr(rscratch1, stack_limit);
729
730 #ifdef ASSERT
731 Label limit_okay;
732 // Verify that thread stack limit is non-zero.
733 __ cbnz(rscratch1, limit_okay);
734 __ stop("stack overflow limit is zero");
735 __ bind(limit_okay);
736 #endif
737
738 // Add stack limit to locals.
739 __ add(r0, r0, rscratch1);
740
741 // Check against the current stack bottom.
742 __ cmp(sp, r0);
743 __ br(Assembler::HI, after_frame_check);
744
745 // Remove the incoming args, peeling the machine SP back to where it
746 // was in the caller. This is not strictly necessary, but unless we
747 // do so the stack frame may have a garbage FP; this ensures a
748 // correct call stack that we can always unwind. The ANDR should be
749 // unnecessary because the sender SP in r19 is always aligned, but
750 // it doesn't hurt.
751 __ andr(sp, r19_sender_sp, -16);
752
753 // Note: the restored frame is not necessarily interpreted.
754 // Use the shared runtime version of the StackOverflowError.
755 assert(StubRoutines::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
756 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
757
758 // all done with frame size check
759 __ bind(after_frame_check);
760 }
761
762 // Allocate monitor and lock method (asm interpreter)
763 //
764 // Args:
765 // rmethod: Method*
766 // rlocals: locals
767 //
768 // Kills:
769 // r0
770 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
771 // rscratch1, rscratch2 (scratch regs)
772 void TemplateInterpreterGenerator::lock_method() {
773 // synchronize method
774 const Address access_flags(rmethod, Method::access_flags_offset());
775 const Address monitor_block_top(
776 rfp,
777 frame::interpreter_frame_monitor_block_top_offset * wordSize);
778 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
779
780 #ifdef ASSERT
781 {
782 Label L;
783 __ ldrw(r0, access_flags);
784 __ tst(r0, JVM_ACC_SYNCHRONIZED);
785 __ br(Assembler::NE, L);
786 __ stop("method doesn't need synchronization");
787 __ bind(L);
788 }
789 #endif // ASSERT
790
791 // get synchronization object
792 {
793 Label done;
794 __ ldrw(r0, access_flags);
795 __ tst(r0, JVM_ACC_STATIC);
796 // get receiver (assume this is frequent case)
797 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
798 __ br(Assembler::EQ, done);
799 __ load_mirror(r0, rmethod, r5, rscratch2);
800
801 #ifdef ASSERT
802 {
803 Label L;
804 __ cbnz(r0, L);
805 __ stop("synchronization object is null");
806 __ bind(L);
807 }
808 #endif // ASSERT
809
810 __ bind(done);
811 }
812
813 // add space for monitor & lock
814 __ check_extended_sp();
815 __ sub(sp, sp, entry_size); // add space for a monitor entry
816 __ sub(esp, esp, entry_size);
817 __ sub(rscratch1, sp, rfp);
818 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
819 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
820 __ sub(rscratch1, esp, rfp);
821 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
822 __ str(rscratch1, monitor_block_top); // set new monitor block top
823
824 // store object
825 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
826 __ mov(c_rarg1, esp); // object address
827 __ lock_object(c_rarg1);
828 }
829
830 // Generate a fixed interpreter frame. This is identical setup for
831 // interpreted methods and for native methods hence the shared code.
832 //
833 // Args:
834 // lr: return address
835 // rmethod: Method*
836 // rlocals: pointer to locals
837 // rcpool: cp cache
838 // stack_pointer: previous sp
839 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
840 // initialize fixed part of activation frame
841 if (native_call) {
842 __ sub(esp, sp, 14 * wordSize);
843 __ mov(rbcp, zr);
844 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
845 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
846 // add 2 zero-initialized slots for native calls
847 __ stp(zr, zr, Address(sp, 12 * wordSize));
848 } else {
849 __ sub(esp, sp, 12 * wordSize);
850 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
851 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
852 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
853 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
854 }
855
856 if (ProfileInterpreter) {
857 Label method_data_continue;
858 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
859 __ cbz(rscratch1, method_data_continue);
860 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
861 __ bind(method_data_continue);
862 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
863 } else {
864 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
865 }
866
867 __ protect_return_address();
868 __ stp(rfp, lr, Address(sp, 10 * wordSize));
869 __ lea(rfp, Address(sp, 10 * wordSize));
870
871 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
872 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
873 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
874 __ sub(rscratch1, rlocals, rfp);
875 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
876 // Store relativized rlocals, see frame::interpreter_frame_locals().
877 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
878
879 // set sender sp
880 // leave last_sp as null
881 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
882
883 // Get mirror
884 __ load_mirror(r10, rmethod, r5, rscratch2);
885 if (! native_call) {
886 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
887 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
888 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
889 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
890 __ andr(rscratch1, rscratch1, -16);
891 __ sub(rscratch2, rscratch1, rfp);
892 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
893 // Store extended SP and mirror
894 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
895 // Move SP out of the way
896 __ mov(sp, rscratch1);
897 } else {
898 // Make sure there is room for the exception oop pushed in case method throws
899 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
900 __ sub(rscratch1, sp, 2 * wordSize);
901 __ sub(rscratch2, rscratch1, rfp);
902 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
903 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
904 __ mov(sp, rscratch1);
905 }
906 }
907
908 // End of helpers
909
910 // Various method entries
911 //------------------------------------------------------------------------------------------------------------------------
912 //
913 //
914
915 // Method entry for java.lang.ref.Reference.get.
916 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
917 // Code: _aload_0, _getfield, _areturn
918 // parameter size = 1
919 //
920 // The code that gets generated by this routine is split into 2 parts:
921 // 1. The "intrinsified" code for G1 (or any SATB based GC),
922 // 2. The slow path - which is an expansion of the regular method entry.
923 //
924 // Notes:-
925 // * In the G1 code we do not check whether we need to block for
926 // a safepoint. If G1 is enabled then we must execute the specialized
927 // code for Reference.get (except when the Reference object is null)
928 // so that we can log the value in the referent field with an SATB
929 // update buffer.
930 // If the code for the getfield template is modified so that the
931 // G1 pre-barrier code is executed when the current method is
932 // Reference.get() then going through the normal method entry
933 // will be fine.
934 // * The G1 code can, however, check the receiver object (the instance
935 // of java.lang.Reference) and jump to the slow path if null. If the
936 // Reference object is null then we obviously cannot fetch the referent
937 // and so we don't need to call the G1 pre-barrier. Thus we can use the
938 // regular method entry code to generate the NPE.
939 //
940 // This code is based on generate_accessor_entry.
941 //
942 // rmethod: Method*
943 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
944
945 // LR is live. It must be saved around calls.
946
947 address entry = __ pc();
948
949 const int referent_offset = java_lang_ref_Reference::referent_offset();
950
951 Label slow_path;
952 const Register local_0 = c_rarg0;
953 // Check if local 0 != null
954 // If the receiver is null then it is OK to jump to the slow path.
955 __ ldr(local_0, Address(esp, 0));
956 __ cbz(local_0, slow_path);
957
958 // Load the value of the referent field.
959 const Address field_address(local_0, referent_offset);
960 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
961 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
962
963 // areturn
964 __ andr(sp, r19_sender_sp, -16); // done with stack
965 __ ret(lr);
966
967 // generate a vanilla interpreter entry as the slow path
968 __ bind(slow_path);
969 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
970 return entry;
971
972 }
973
974 /**
975 * Method entry for static native methods:
976 * int java.util.zip.CRC32.update(int crc, int b)
977 */
978 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
979 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
980 address entry = __ pc();
981
982 // rmethod: Method*
983 // r19_sender_sp: senderSP must preserved for slow path
984 // esp: args
985
986 Label slow_path;
987 // If we need a safepoint check, generate full interpreter entry.
988 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
989
990 // We don't generate local frame and don't align stack because
991 // we call stub code and there is no safepoint on this path.
992
993 // Load parameters
994 const Register crc = c_rarg0; // crc
995 const Register val = c_rarg1; // source java byte value
996 const Register tbl = c_rarg2; // scratch
997
998 // Arguments are reversed on java expression stack
999 __ ldrw(val, Address(esp, 0)); // byte value
1000 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1001
1002 uint64_t offset;
1003 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1004 __ add(tbl, tbl, offset);
1005
1006 __ mvnw(crc, crc); // ~crc
1007 __ update_byte_crc32(crc, val, tbl);
1008 __ mvnw(crc, crc); // ~crc
1009
1010 // result in c_rarg0
1011
1012 __ andr(sp, r19_sender_sp, -16);
1013 __ ret(lr);
1014
1015 // generate a vanilla native entry as the slow path
1016 __ bind(slow_path);
1017 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1018 return entry;
1019 }
1020
1021 /**
1022 * Method entry for static native methods:
1023 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1024 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1025 */
1026 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1027 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1028 address entry = __ pc();
1029
1030 // rmethod,: Method*
1031 // r19_sender_sp: senderSP must preserved for slow path
1032
1033 Label slow_path;
1034 // If we need a safepoint check, generate full interpreter entry.
1035 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1036
1037 // We don't generate local frame and don't align stack because
1038 // we call stub code and there is no safepoint on this path.
1039
1040 // Load parameters
1041 const Register crc = c_rarg0; // crc
1042 const Register buf = c_rarg1; // source java byte array address
1043 const Register len = c_rarg2; // length
1044 const Register off = len; // offset (never overlaps with 'len')
1045
1046 // Arguments are reversed on java expression stack
1047 // Calculate address of start element
1048 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1049 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1050 __ ldrw(off, Address(esp, wordSize)); // offset
1051 __ add(buf, buf, off); // + offset
1052 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1053 } else {
1054 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1055 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1056 __ ldrw(off, Address(esp, wordSize)); // offset
1057 __ add(buf, buf, off); // + offset
1058 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1059 }
1060 // Can now load 'len' since we're finished with 'off'
1061 __ ldrw(len, Address(esp, 0x0)); // Length
1062
1063 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1064
1065 // We are frameless so we can just jump to the stub.
1066 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1067
1068 // generate a vanilla native entry as the slow path
1069 __ bind(slow_path);
1070 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1071 return entry;
1072 }
1073
1074 /**
1075 * Method entry for intrinsic-candidate (non-native) methods:
1076 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1077 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1078 * Unlike CRC32, CRC32C does not have any methods marked as native
1079 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1080 */
1081 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1082 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1083 address entry = __ pc();
1084
1085 // Prepare jump to stub using parameters from the stack
1086 const Register crc = c_rarg0; // initial crc
1087 const Register buf = c_rarg1; // source java byte array address
1088 const Register len = c_rarg2; // len argument to the kernel
1089
1090 const Register end = len; // index of last element to process
1091 const Register off = crc; // offset
1092
1093 __ ldrw(end, Address(esp)); // int end
1094 __ ldrw(off, Address(esp, wordSize)); // int offset
1095 __ sub(len, end, off);
1096 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1097 __ add(buf, buf, off); // + offset
1098 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1099 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1100 } else {
1101 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1102 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1103 }
1104
1105 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1106
1107 // Jump to the stub.
1108 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1109
1110 return entry;
1111 }
1112
1113 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1114 // See more discussion in stackOverflow.hpp.
1115
1116 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1117 const int page_size = (int)os::vm_page_size();
1118 const int n_shadow_pages = shadow_zone_size / page_size;
1119
1120 #ifdef ASSERT
1121 Label L_good_limit;
1122 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1123 __ cbnz(rscratch1, L_good_limit);
1124 __ stop("shadow zone safe limit is not initialized");
1125 __ bind(L_good_limit);
1126
1127 Label L_good_watermark;
1128 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1129 __ cbnz(rscratch1, L_good_watermark);
1130 __ stop("shadow zone growth watermark is not initialized");
1131 __ bind(L_good_watermark);
1132 #endif
1133
1134 Label L_done;
1135
1136 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1137 __ cmp(sp, rscratch1);
1138 __ br(Assembler::HI, L_done);
1139
1140 for (int p = 1; p <= n_shadow_pages; p++) {
1141 __ sub(rscratch2, sp, p*page_size);
1142 __ str(zr, Address(rscratch2));
1143 }
1144
1145 // Record the new watermark, but only if the update is above the safe limit.
1146 // Otherwise, the next time around the check above would pass the safe limit.
1147 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1148 __ cmp(sp, rscratch1);
1149 __ br(Assembler::LS, L_done);
1150 __ mov(rscratch1, sp);
1151 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1152
1153 __ bind(L_done);
1154 }
1155
1156 // Interpreter stub for calling a native method. (asm interpreter)
1157 // This sets up a somewhat different looking stack for calling the
1158 // native method than the typical interpreter frame setup.
1159 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1160 // determine code generation flags
1161 bool inc_counter = UseCompiler || CountCompiledCalls;
1162
1163 // r1: Method*
1164 // rscratch1: sender sp
1165
1166 address entry_point = __ pc();
1167
1168 const Address constMethod (rmethod, Method::const_offset());
1169 const Address access_flags (rmethod, Method::access_flags_offset());
1170 const Address size_of_parameters(r2, ConstMethod::
1171 size_of_parameters_offset());
1172
1173 // get parameter size (always needed)
1174 __ ldr(r2, constMethod);
1175 __ load_unsigned_short(r2, size_of_parameters);
1176
1177 // Native calls don't need the stack size check since they have no
1178 // expression stack and the arguments are already on the stack and
1179 // we only add a handful of words to the stack.
1180
1181 // rmethod: Method*
1182 // r2: size of parameters
1183 // rscratch1: sender sp
1184
1185 // for natives the size of locals is zero
1186
1187 // compute beginning of parameters (rlocals)
1188 __ add(rlocals, esp, r2, ext::uxtx, 3);
1189 __ add(rlocals, rlocals, -wordSize);
1190
1191 // Pull SP back to minimum size: this avoids holes in the stack
1192 __ andr(sp, esp, -16);
1193
1194 // initialize fixed part of activation frame
1195 generate_fixed_frame(true);
1196
1197 // make sure method is native & not abstract
1198 #ifdef ASSERT
1199 __ ldrw(r0, access_flags);
1200 {
1201 Label L;
1202 __ tst(r0, JVM_ACC_NATIVE);
1203 __ br(Assembler::NE, L);
1204 __ stop("tried to execute non-native method as native");
1205 __ bind(L);
1206 }
1207 {
1208 Label L;
1209 __ tst(r0, JVM_ACC_ABSTRACT);
1210 __ br(Assembler::EQ, L);
1211 __ stop("tried to execute abstract method in interpreter");
1212 __ bind(L);
1213 }
1214 #endif
1215
1216 // Since at this point in the method invocation the exception
1217 // handler would try to exit the monitor of synchronized methods
1218 // which hasn't been entered yet, we set the thread local variable
1219 // _do_not_unlock_if_synchronized to true. The remove_activation
1220 // will check this flag.
1221
1222 const Address do_not_unlock_if_synchronized(rthread,
1223 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1224 __ mov(rscratch2, true);
1225 __ strb(rscratch2, do_not_unlock_if_synchronized);
1226
1227 // increment invocation count & check for overflow
1228 Label invocation_counter_overflow;
1229 if (inc_counter) {
1230 generate_counter_incr(&invocation_counter_overflow);
1231 }
1232
1233 Label continue_after_compile;
1234 __ bind(continue_after_compile);
1235
1236 bang_stack_shadow_pages(true);
1237
1238 // reset the _do_not_unlock_if_synchronized flag
1239 __ strb(zr, do_not_unlock_if_synchronized);
1240
1241 // check for synchronized methods
1242 // Must happen AFTER invocation_counter check and stack overflow check,
1243 // so method is not locked if overflows.
1244 if (synchronized) {
1245 lock_method();
1246 } else {
1247 // no synchronization necessary
1248 #ifdef ASSERT
1249 {
1250 Label L;
1251 __ ldrw(r0, access_flags);
1252 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1253 __ br(Assembler::EQ, L);
1254 __ stop("method needs synchronization");
1255 __ bind(L);
1256 }
1257 #endif
1258 }
1259
1260 // start execution
1261 #ifdef ASSERT
1262 {
1263 Label L;
1264 const Address monitor_block_top(rfp,
1265 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1266 __ ldr(rscratch1, monitor_block_top);
1267 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1268 __ cmp(esp, rscratch1);
1269 __ br(Assembler::EQ, L);
1270 __ stop("broken stack frame setup in interpreter 1");
1271 __ bind(L);
1272 }
1273 #endif
1274
1275 // jvmti support
1276 __ notify_method_entry();
1277
1278 // work registers
1279 const Register t = r17;
1280 const Register result_handler = r19;
1281
1282 // allocate space for parameters
1283 __ ldr(t, Address(rmethod, Method::const_offset()));
1284 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1285
1286 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1287 __ andr(sp, rscratch1, -16);
1288 __ mov(esp, rscratch1);
1289
1290 // get signature handler
1291 {
1292 Label L;
1293 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1294 __ cbnz(t, L);
1295 __ call_VM(noreg,
1296 CAST_FROM_FN_PTR(address,
1297 InterpreterRuntime::prepare_native_call),
1298 rmethod);
1299 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1300 __ bind(L);
1301 }
1302
1303 // call signature handler
1304 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1305 "adjust this code");
1306 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1307 "adjust this code");
1308 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1309 "adjust this code");
1310
1311 // The generated handlers do not touch rmethod (the method).
1312 // However, large signatures cannot be cached and are generated
1313 // each time here. The slow-path generator can do a GC on return,
1314 // so we must reload it after the call.
1315 __ blr(t);
1316 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1317
1318
1319 // result handler is in r0
1320 // set result handler
1321 __ mov(result_handler, r0);
1322 // pass mirror handle if static call
1323 {
1324 Label L;
1325 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1326 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1327 // get mirror
1328 __ load_mirror(t, rmethod, r10, rscratch2);
1329 // copy mirror into activation frame
1330 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1331 // pass handle to mirror
1332 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1333 __ bind(L);
1334 }
1335
1336 // get native function entry point in r10
1337 {
1338 Label L;
1339 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1340 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1341 __ mov(rscratch2, unsatisfied);
1342 __ ldr(rscratch2, rscratch2);
1343 __ cmp(r10, rscratch2);
1344 __ br(Assembler::NE, L);
1345 __ call_VM(noreg,
1346 CAST_FROM_FN_PTR(address,
1347 InterpreterRuntime::prepare_native_call),
1348 rmethod);
1349 __ get_method(rmethod);
1350 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1351 __ bind(L);
1352 }
1353
1354 // pass JNIEnv
1355 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1356
1357 // Set the last Java PC in the frame anchor to be the return address from
1358 // the call to the native method: this will allow the debugger to
1359 // generate an accurate stack trace.
1360 Label native_return;
1361 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1362
1363 // change thread state
1364 #ifdef ASSERT
1365 {
1366 Label L;
1367 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1368 __ cmp(t, (u1)_thread_in_Java);
1369 __ br(Assembler::EQ, L);
1370 __ stop("Wrong thread state in native stub");
1371 __ bind(L);
1372 }
1373 #endif
1374
1375 // Change state to native
1376 __ mov(rscratch1, _thread_in_native);
1377 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1378 __ stlrw(rscratch1, rscratch2);
1379
1380 // Call the native method.
1381 __ blr(r10);
1382 __ bind(native_return);
1383 __ get_method(rmethod);
1384 // result potentially in r0 or v0
1385
1386 // Restore cpu control state after JNI call
1387 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1388
1389 // make room for the pushes we're about to do
1390 __ sub(rscratch1, esp, 4 * wordSize);
1391 __ andr(sp, rscratch1, -16);
1392
1393 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1394 // in order to extract the result of a method call. If the order of these
1395 // pushes change or anything else is added to the stack then the code in
1396 // interpreter_frame_result must also change.
1397 __ push(dtos);
1398 __ push(ltos);
1399
1400 __ verify_sve_vector_length();
1401
1402 // change thread state
1403 __ mov(rscratch1, _thread_in_native_trans);
1404 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1405 __ stlrw(rscratch1, rscratch2);
1406
1407 // Force this write out before the read below
1408 if (!UseSystemMemoryBarrier) {
1409 __ dmb(Assembler::ISH);
1410 }
1411
1412 // check for safepoint operation in progress and/or pending suspend requests
1413 {
1414 Label L, Continue;
1415
1416 // We need an acquire here to ensure that any subsequent load of the
1417 // global SafepointSynchronize::_state flag is ordered after this load
1418 // of the thread-local polling word. We don't want this poll to
1419 // return false (i.e. not safepointing) and a later poll of the global
1420 // SafepointSynchronize::_state spuriously to return true.
1421 //
1422 // This is to avoid a race when we're in a native->Java transition
1423 // racing the code which wakes up from a safepoint.
1424 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1425 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1426 __ cbz(rscratch2, Continue);
1427 __ bind(L);
1428
1429 // Don't use call_VM as it will see a possible pending exception
1430 // and forward it and never return here preventing us from
1431 // clearing _last_native_pc down below. So we do a runtime call by
1432 // hand.
1433 //
1434 __ mov(c_rarg0, rthread);
1435 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1436 __ blr(rscratch2);
1437 __ get_method(rmethod);
1438 __ reinit_heapbase();
1439 __ bind(Continue);
1440 }
1441
1442 // change thread state
1443 __ mov(rscratch1, _thread_in_Java);
1444 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1445 __ stlrw(rscratch1, rscratch2);
1446
1447 // reset_last_Java_frame
1448 __ reset_last_Java_frame(true);
1449
1450 if (CheckJNICalls) {
1451 // clear_pending_jni_exception_check
1452 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1453 }
1454
1455 // reset handle block
1456 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1457 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1458
1459 // If result is an oop unbox and store it in frame where gc will see it
1460 // and result handler will pick it up
1461
1462 {
1463 Label no_oop;
1464 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1465 __ cmp(t, result_handler);
1466 __ br(Assembler::NE, no_oop);
1467 // Unbox oop result, e.g. JNIHandles::resolve result.
1468 __ pop(ltos);
1469 __ resolve_jobject(r0, t, rscratch2);
1470 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1471 // keep stack depth as expected by pushing oop which will eventually be discarded
1472 __ push(ltos);
1473 __ bind(no_oop);
1474 }
1475
1476 {
1477 Label no_reguard;
1478 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1479 __ ldrw(rscratch1, Address(rscratch1));
1480 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1481 __ br(Assembler::NE, no_reguard);
1482
1483 __ push_call_clobbered_registers();
1484 __ mov(c_rarg0, rthread);
1485 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1486 __ blr(rscratch2);
1487 __ pop_call_clobbered_registers();
1488
1489 __ bind(no_reguard);
1490 }
1491
1492 // The method register is junk from after the thread_in_native transition
1493 // until here. Also can't call_VM until the bcp has been
1494 // restored. Need bcp for throwing exception below so get it now.
1495 __ get_method(rmethod);
1496
1497 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1498 // rbcp == code_base()
1499 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1500 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1501 // handle exceptions (exception handling will handle unlocking!)
1502 {
1503 Label L;
1504 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1505 __ cbz(rscratch1, L);
1506 // Note: At some point we may want to unify this with the code
1507 // used in call_VM_base(); i.e., we should use the
1508 // StubRoutines::forward_exception code. For now this doesn't work
1509 // here because the rsp is not correctly set at this point.
1510 __ MacroAssembler::call_VM(noreg,
1511 CAST_FROM_FN_PTR(address,
1512 InterpreterRuntime::throw_pending_exception));
1513 __ should_not_reach_here();
1514 __ bind(L);
1515 }
1516
1517 // do unlocking if necessary
1518 {
1519 Label L;
1520 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1521 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1522 // the code below should be shared with interpreter macro
1523 // assembler implementation
1524 {
1525 Label unlock;
1526 // BasicObjectLock will be first in list, since this is a
1527 // synchronized method. However, need to check that the object
1528 // has not been unlocked by an explicit monitorexit bytecode.
1529
1530 // monitor expect in c_rarg1 for slow unlock path
1531 __ lea (c_rarg1, Address(rfp, // address of first monitor
1532 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1533 wordSize - sizeof(BasicObjectLock))));
1534
1535 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1536 __ cbnz(t, unlock);
1537
1538 // Entry already unlocked, need to throw exception
1539 __ MacroAssembler::call_VM(noreg,
1540 CAST_FROM_FN_PTR(address,
1541 InterpreterRuntime::throw_illegal_monitor_state_exception));
1542 __ should_not_reach_here();
1543
1544 __ bind(unlock);
1545 __ unlock_object(c_rarg1);
1546 }
1547 __ bind(L);
1548 }
1549
1550 // jvmti support
1551 // Note: This must happen _after_ handling/throwing any exceptions since
1552 // the exception handler code notifies the runtime of method exits
1553 // too. If this happens before, method entry/exit notifications are
1554 // not properly paired (was bug - gri 11/22/99).
1555 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1556
1557 // restore potential result in r0:d0, call result handler to
1558 // restore potential result in ST0 & handle result
1559
1560 __ pop(ltos);
1561 __ pop(dtos);
1562
1563 __ blr(result_handler);
1564
1565 // remove activation
1566 __ ldr(esp, Address(rfp,
1567 frame::interpreter_frame_sender_sp_offset *
1568 wordSize)); // get sender sp
1569 // remove frame anchor
1570 __ leave();
1571
1572 // restore sender sp
1573 __ mov(sp, esp);
1574
1575 __ ret(lr);
1576
1577 if (inc_counter) {
1578 // Handle overflow of counter and compile method
1579 __ bind(invocation_counter_overflow);
1580 generate_counter_overflow(continue_after_compile);
1581 }
1582
1583 return entry_point;
1584 }
1585
1586 //
1587 // Generic interpreted method entry to (asm) interpreter
1588 //
1589 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1590 // determine code generation flags
1591 bool inc_counter = UseCompiler || CountCompiledCalls;
1592
1593 // rscratch1: sender sp
1594 address entry_point = __ pc();
1595
1596 const Address constMethod(rmethod, Method::const_offset());
1597 const Address access_flags(rmethod, Method::access_flags_offset());
1598 const Address size_of_parameters(r3,
1599 ConstMethod::size_of_parameters_offset());
1600 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1601
1602 // get parameter size (always needed)
1603 // need to load the const method first
1604 __ ldr(r3, constMethod);
1605 __ load_unsigned_short(r2, size_of_parameters);
1606
1607 // r2: size of parameters
1608
1609 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1610 __ sub(r3, r3, r2); // r3 = no. of additional locals
1611
1612 // see if we've got enough room on the stack for locals plus overhead.
1613 generate_stack_overflow_check();
1614
1615 // compute beginning of parameters (rlocals)
1616 __ add(rlocals, esp, r2, ext::uxtx, 3);
1617 __ sub(rlocals, rlocals, wordSize);
1618
1619 __ mov(rscratch1, esp);
1620
1621 // r3 - # of additional locals
1622 // allocate space for locals
1623 // explicitly initialize locals
1624 // Initializing memory allocated for locals in the same direction as
1625 // the stack grows to ensure page initialization order according
1626 // to windows-aarch64 stack page growth requirement (see
1627 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1628 {
1629 Label exit, loop;
1630 __ ands(zr, r3, r3);
1631 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1632 __ bind(loop);
1633 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1634 __ sub(r3, r3, 1); // until everything initialized
1635 __ cbnz(r3, loop);
1636 __ bind(exit);
1637 }
1638
1639 // Padding between locals and fixed part of activation frame to ensure
1640 // SP is always 16-byte aligned.
1641 __ andr(sp, rscratch1, -16);
1642
1643 // And the base dispatch table
1644 __ get_dispatch();
1645
1646 // initialize fixed part of activation frame
1647 generate_fixed_frame(false);
1648
1649 // make sure method is not native & not abstract
1650 #ifdef ASSERT
1651 __ ldrw(r0, access_flags);
1652 {
1653 Label L;
1654 __ tst(r0, JVM_ACC_NATIVE);
1655 __ br(Assembler::EQ, L);
1656 __ stop("tried to execute native method as non-native");
1657 __ bind(L);
1658 }
1659 {
1660 Label L;
1661 __ tst(r0, JVM_ACC_ABSTRACT);
1662 __ br(Assembler::EQ, L);
1663 __ stop("tried to execute abstract method in interpreter");
1664 __ bind(L);
1665 }
1666 #endif
1667
1668 // Since at this point in the method invocation the exception
1669 // handler would try to exit the monitor of synchronized methods
1670 // which hasn't been entered yet, we set the thread local variable
1671 // _do_not_unlock_if_synchronized to true. The remove_activation
1672 // will check this flag.
1673
1674 const Address do_not_unlock_if_synchronized(rthread,
1675 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1676 __ mov(rscratch2, true);
1677 __ strb(rscratch2, do_not_unlock_if_synchronized);
1678
1679 Register mdp = r3;
1680 __ profile_parameters_type(mdp, r1, r2);
1681
1682 // increment invocation count & check for overflow
1683 Label invocation_counter_overflow;
1684 if (inc_counter) {
1685 generate_counter_incr(&invocation_counter_overflow);
1686 }
1687
1688 Label continue_after_compile;
1689 __ bind(continue_after_compile);
1690
1691 bang_stack_shadow_pages(false);
1692
1693 // reset the _do_not_unlock_if_synchronized flag
1694 __ strb(zr, do_not_unlock_if_synchronized);
1695
1696 // check for synchronized methods
1697 // Must happen AFTER invocation_counter check and stack overflow check,
1698 // so method is not locked if overflows.
1699 if (synchronized) {
1700 // Allocate monitor and lock method
1701 lock_method();
1702 } else {
1703 // no synchronization necessary
1704 #ifdef ASSERT
1705 {
1706 Label L;
1707 __ ldrw(r0, access_flags);
1708 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1709 __ br(Assembler::EQ, L);
1710 __ stop("method needs synchronization");
1711 __ bind(L);
1712 }
1713 #endif
1714 }
1715
1716 // start execution
1717 #ifdef ASSERT
1718 {
1719 Label L;
1720 const Address monitor_block_top (rfp,
1721 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1722 __ ldr(rscratch1, monitor_block_top);
1723 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1724 __ cmp(esp, rscratch1);
1725 __ br(Assembler::EQ, L);
1726 __ stop("broken stack frame setup in interpreter 2");
1727 __ bind(L);
1728 }
1729 #endif
1730
1731 // jvmti support
1732 __ notify_method_entry();
1733
1734 __ dispatch_next(vtos);
1735
1736 // invocation counter overflow
1737 if (inc_counter) {
1738 // Handle overflow of counter and compile method
1739 __ bind(invocation_counter_overflow);
1740 generate_counter_overflow(continue_after_compile);
1741 }
1742
1743 return entry_point;
1744 }
1745
1746 // Method entry for java.lang.Thread.currentThread
1747 address TemplateInterpreterGenerator::generate_currentThread() {
1748 address entry_point = __ pc();
1749
1750 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1751 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1752 __ ret(lr);
1753
1754 return entry_point;
1755 }
1756
1757 // Not supported
1758 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1759 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1760 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1761 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1762
1763 //-----------------------------------------------------------------------------
1764 // Exceptions
1765
1766 void TemplateInterpreterGenerator::generate_throw_exception() {
1767 // Entry point in previous activation (i.e., if the caller was
1768 // interpreted)
1769 Interpreter::_rethrow_exception_entry = __ pc();
1770 // Restore sp to interpreter_frame_last_sp even though we are going
1771 // to empty the expression stack for the exception processing.
1772 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1773 // r0: exception
1774 // r3: return address/pc that threw exception
1775 __ restore_bcp(); // rbcp points to call/send
1776 __ restore_locals();
1777 __ restore_constant_pool_cache();
1778 __ reinit_heapbase(); // restore rheapbase as heapbase.
1779 __ get_dispatch();
1780
1781 // Entry point for exceptions thrown within interpreter code
1782 Interpreter::_throw_exception_entry = __ pc();
1783 // If we came here via a NullPointerException on the receiver of a
1784 // method, rmethod may be corrupt.
1785 __ get_method(rmethod);
1786 // expression stack is undefined here
1787 // r0: exception
1788 // rbcp: exception bcp
1789 __ verify_oop(r0);
1790 __ mov(c_rarg1, r0);
1791
1792 // expression stack must be empty before entering the VM in case of
1793 // an exception
1794 __ empty_expression_stack();
1795 // find exception handler address and preserve exception oop
1796 __ call_VM(r3,
1797 CAST_FROM_FN_PTR(address,
1798 InterpreterRuntime::exception_handler_for_exception),
1799 c_rarg1);
1800
1801 // Restore machine SP
1802 __ restore_sp_after_call();
1803
1804 // r0: exception handler entry point
1805 // r3: preserved exception oop
1806 // rbcp: bcp for exception handler
1807 __ push_ptr(r3); // push exception which is now the only value on the stack
1808 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1809
1810 // If the exception is not handled in the current frame the frame is
1811 // removed and the exception is rethrown (i.e. exception
1812 // continuation is _rethrow_exception).
1813 //
1814 // Note: At this point the bci is still the bxi for the instruction
1815 // which caused the exception and the expression stack is
1816 // empty. Thus, for any VM calls at this point, GC will find a legal
1817 // oop map (with empty expression stack).
1818
1819 //
1820 // JVMTI PopFrame support
1821 //
1822
1823 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1824 __ empty_expression_stack();
1825 // Set the popframe_processing bit in pending_popframe_condition
1826 // indicating that we are currently handling popframe, so that
1827 // call_VMs that may happen later do not trigger new popframe
1828 // handling cycles.
1829 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1830 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1831 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1832
1833 {
1834 // Check to see whether we are returning to a deoptimized frame.
1835 // (The PopFrame call ensures that the caller of the popped frame is
1836 // either interpreted or compiled and deoptimizes it if compiled.)
1837 // In this case, we can't call dispatch_next() after the frame is
1838 // popped, but instead must save the incoming arguments and restore
1839 // them after deoptimization has occurred.
1840 //
1841 // Note that we don't compare the return PC against the
1842 // deoptimization blob's unpack entry because of the presence of
1843 // adapter frames in C2.
1844 Label caller_not_deoptimized;
1845 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1846 // This is a return address, so requires authenticating for PAC.
1847 __ authenticate_return_address(c_rarg1);
1848 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1849 InterpreterRuntime::interpreter_contains), c_rarg1);
1850 __ cbnz(r0, caller_not_deoptimized);
1851
1852 // Compute size of arguments for saving when returning to
1853 // deoptimized caller
1854 __ get_method(r0);
1855 __ ldr(r0, Address(r0, Method::const_offset()));
1856 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1857 size_of_parameters_offset())));
1858 __ lsl(r0, r0, Interpreter::logStackElementSize);
1859 __ restore_locals(); // XXX do we need this?
1860 __ sub(rlocals, rlocals, r0);
1861 __ add(rlocals, rlocals, wordSize);
1862 // Save these arguments
1863 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1864 Deoptimization::
1865 popframe_preserve_args),
1866 rthread, r0, rlocals);
1867
1868 __ remove_activation(vtos,
1869 /* throw_monitor_exception */ false,
1870 /* install_monitor_exception */ false,
1871 /* notify_jvmdi */ false);
1872
1873 // Inform deoptimization that it is responsible for restoring
1874 // these arguments
1875 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1876 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1877
1878 // Continue in deoptimization handler
1879 __ ret(lr);
1880
1881 __ bind(caller_not_deoptimized);
1882 }
1883
1884 __ remove_activation(vtos,
1885 /* throw_monitor_exception */ false,
1886 /* install_monitor_exception */ false,
1887 /* notify_jvmdi */ false);
1888
1889 // Restore the last_sp and null it out
1890 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1891 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1892 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1893
1894 __ restore_bcp();
1895 __ restore_locals();
1896 __ restore_constant_pool_cache();
1897 __ get_method(rmethod);
1898 __ get_dispatch();
1899
1900 // The method data pointer was incremented already during
1901 // call profiling. We have to restore the mdp for the current bcp.
1902 if (ProfileInterpreter) {
1903 __ set_method_data_pointer_for_bcp();
1904 }
1905
1906 // Clear the popframe condition flag
1907 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1908 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1909
1910 #if INCLUDE_JVMTI
1911 {
1912 Label L_done;
1913
1914 __ ldrb(rscratch1, Address(rbcp, 0));
1915 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1916 __ br(Assembler::NE, L_done);
1917
1918 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1919 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1920
1921 __ ldr(c_rarg0, Address(rlocals, 0));
1922 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1923
1924 __ cbz(r0, L_done);
1925
1926 __ str(r0, Address(esp, 0));
1927 __ bind(L_done);
1928 }
1929 #endif // INCLUDE_JVMTI
1930
1931 // Restore machine SP
1932 __ restore_sp_after_call();
1933
1934 __ dispatch_next(vtos);
1935 // end of PopFrame support
1936
1937 Interpreter::_remove_activation_entry = __ pc();
1938
1939 // preserve exception over this code sequence
1940 __ pop_ptr(r0);
1941 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1942 // remove the activation (without doing throws on illegalMonitorExceptions)
1943 __ remove_activation(vtos, false, true, false);
1944 // restore exception
1945 __ get_vm_result(r0, rthread);
1946
1947 // In between activations - previous activation type unknown yet
1948 // compute continuation point - the continuation point expects the
1949 // following registers set up:
1950 //
1951 // r0: exception
1952 // lr: return address/pc that threw exception
1953 // esp: expression stack of caller
1954 // rfp: fp of caller
1955 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1956 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1957 SharedRuntime::exception_handler_for_return_address),
1958 rthread, lr);
1959 __ mov(r1, r0); // save exception handler
1960 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1961 // We might be returning to a deopt handler that expects r3 to
1962 // contain the exception pc
1963 __ mov(r3, lr);
1964 // Note that an "issuing PC" is actually the next PC after the call
1965 __ br(r1); // jump to exception
1966 // handler of caller
1967 }
1968
1969
1970 //
1971 // JVMTI ForceEarlyReturn support
1972 //
1973 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1974 address entry = __ pc();
1975
1976 __ restore_bcp();
1977 __ restore_locals();
1978 __ empty_expression_stack();
1979 __ load_earlyret_value(state);
1980
1981 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1982 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1983
1984 // Clear the earlyret state
1985 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1986 __ str(zr, cond_addr);
1987
1988 __ remove_activation(state,
1989 false, /* throw_monitor_exception */
1990 false, /* install_monitor_exception */
1991 true); /* notify_jvmdi */
1992 __ ret(lr);
1993
1994 return entry;
1995 } // end of ForceEarlyReturn support
1996
1997
1998
1999 //-----------------------------------------------------------------------------
2000 // Helper for vtos entry point generation
2001
2002 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2003 address& bep,
2004 address& cep,
2005 address& sep,
2006 address& aep,
2007 address& iep,
2008 address& lep,
2009 address& fep,
2010 address& dep,
2011 address& vep) {
2012 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2013 Label L;
2014 aep = __ pc(); __ push_ptr(); __ b(L);
2015 fep = __ pc(); __ push_f(); __ b(L);
2016 dep = __ pc(); __ push_d(); __ b(L);
2017 lep = __ pc(); __ push_l(); __ b(L);
2018 bep = cep = sep =
2019 iep = __ pc(); __ push_i();
2020 vep = __ pc();
2021 __ bind(L);
2022 generate_and_dispatch(t);
2023 }
2024
2025 //-----------------------------------------------------------------------------
2026
2027 // Non-product code
2028 #ifndef PRODUCT
2029 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2030 address entry = __ pc();
2031
2032 __ protect_return_address();
2033 __ push(lr);
2034 __ push(state);
2035 __ push(RegSet::range(r0, r15), sp);
2036 __ mov(c_rarg2, r0); // Pass itos
2037 __ call_VM(noreg,
2038 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2039 c_rarg1, c_rarg2, c_rarg3);
2040 __ pop(RegSet::range(r0, r15), sp);
2041 __ pop(state);
2042 __ pop(lr);
2043 __ authenticate_return_address();
2044 __ ret(lr); // return from result handler
2045
2046 return entry;
2047 }
2048
2049 void TemplateInterpreterGenerator::count_bytecode() {
2050 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2051 __ atomic_addw(noreg, 1, r10);
2052 }
2053
2054 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2055 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2056 __ atomic_addw(noreg, 1, r10);
2057 }
2058
2059 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2060 // Calculate new index for counter:
2061 // _index = (_index >> log2_number_of_codes) |
2062 // (bytecode << log2_number_of_codes);
2063 Register index_addr = rscratch1;
2064 Register index = rscratch2;
2065 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2066 __ ldrw(index, index_addr);
2067 __ mov(r10,
2068 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2069 __ orrw(index, r10, index, Assembler::LSR,
2070 BytecodePairHistogram::log2_number_of_codes);
2071 __ strw(index, index_addr);
2072
2073 // Bump bucket contents:
2074 // _counters[_index] ++;
2075 Register counter_addr = rscratch1;
2076 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2077 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2078 __ atomic_addw(noreg, 1, counter_addr);
2079 }
2080
2081 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2082 // Call a little run-time stub to avoid blow-up for each bytecode.
2083 // The run-time runtime saves the right registers, depending on
2084 // the tosca in-state for the given template.
2085
2086 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2087 "entry must have been generated");
2088 __ bl(Interpreter::trace_code(t->tos_in()));
2089 __ reinit_heapbase();
2090 }
2091
2092
2093 void TemplateInterpreterGenerator::stop_interpreter_at() {
2094 Label L;
2095 __ push(rscratch1);
2096 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2097 __ ldr(rscratch1, Address(rscratch1));
2098 __ mov(rscratch2, StopInterpreterAt);
2099 __ cmpw(rscratch1, rscratch2);
2100 __ br(Assembler::NE, L);
2101 __ brk(0);
2102 __ bind(L);
2103 __ pop(rscratch1);
2104 }
2105
2106 #endif // !PRODUCT