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