1 /*
2 * Copyright (c) 2003, 2022, 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/methodData.hpp"
41 #include "oops/method.hpp"
42 #include "oops/oop.inline.hpp"
43 #include "oops/inlineKlass.hpp"
44 #include "prims/jvmtiExport.hpp"
45 #include "prims/jvmtiThreadState.hpp"
46 #include "runtime/arguments.hpp"
47 #include "runtime/deoptimization.hpp"
48 #include "runtime/frame.inline.hpp"
49 #include "runtime/globals.hpp"
50 #include "runtime/jniHandles.hpp"
51 #include "runtime/sharedRuntime.hpp"
52 #include "runtime/stubRoutines.hpp"
53 #include "runtime/synchronizer.hpp"
54 #include "runtime/timer.hpp"
55 #include "runtime/vframeArray.hpp"
56 #include "utilities/debug.hpp"
57 #include "utilities/powerOfTwo.hpp"
58 #include <sys/types.h>
59
60 // Size of interpreter code. Increase if too small. Interpreter will
61 // fail with a guarantee ("not enough space for interpreter generation");
62 // if too small.
63 // Run with +PrintInterpreter to get the VM to print out the size.
64 // Max size with JVMTI
65 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
66
67 #define __ _masm->
68
69 //-----------------------------------------------------------------------------
70
71 extern "C" void entry(CodeBuffer*);
72
73 //-----------------------------------------------------------------------------
74
75 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
76 address entry = __ pc();
77
78 __ andr(esp, esp, -16);
79 __ mov(c_rarg3, esp);
80 // rmethod
81 // rlocals
82 // c_rarg3: first stack arg - wordSize
83
84 // adjust sp
85 __ sub(sp, c_rarg3, 18 * wordSize);
86 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
87 __ call_VM(noreg,
88 CAST_FROM_FN_PTR(address,
89 InterpreterRuntime::slow_signature_handler),
90 rmethod, rlocals, c_rarg3);
91
92 // r0: result handler
93
94 // Stack layout:
95 // rsp: return address <- sp
96 // 1 garbage
97 // 8 integer args (if static first is unused)
98 // 1 float/double identifiers
99 // 8 double args
100 // stack args <- esp
101 // garbage
102 // expression stack bottom
103 // bcp (NULL)
104 // ...
105
106 // Restore LR
107 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
108
109 // Do FP first so we can use c_rarg3 as temp
110 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
111
112 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
113 const FloatRegister r = as_FloatRegister(i);
114
115 Label d, done;
116
117 __ tbnz(c_rarg3, i, d);
118 __ ldrs(r, Address(sp, (10 + i) * wordSize));
119 __ b(done);
120 __ bind(d);
121 __ ldrd(r, Address(sp, (10 + i) * wordSize));
122 __ bind(done);
123 }
124
125 // c_rarg0 contains the result from the call of
126 // InterpreterRuntime::slow_signature_handler so we don't touch it
127 // here. It will be loaded with the JNIEnv* later.
128 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
129 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
130 Register rm = as_Register(i), rn = as_Register(i+1);
131 __ ldp(rm, rn, Address(sp, i * wordSize));
132 }
133
134 __ add(sp, sp, 18 * wordSize);
135 __ ret(lr);
136
137 return entry;
138 }
139
140
141 //
142 // Various method entries
143 //
144
145 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
146 // rmethod: Method*
147 // r19_sender_sp: sender sp
148 // esp: args
149
150 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
151
152 // These don't need a safepoint check because they aren't virtually
153 // callable. We won't enter these intrinsics from compiled code.
154 // If in the future we added an intrinsic which was virtually callable
155 // we'd have to worry about how to safepoint so that this code is used.
156
157 // mathematical functions inlined by compiler
158 // (interpreter must provide identical implementation
159 // in order to avoid monotonicity bugs when switching
160 // from interpreter to compiler in the middle of some
161 // computation)
162 //
163 // stack:
164 // [ arg ] <-- esp
165 // [ arg ]
166 // retaddr in lr
167
168 address entry_point = NULL;
169 Register continuation = lr;
170 switch (kind) {
171 case Interpreter::java_lang_math_abs:
172 entry_point = __ pc();
173 __ ldrd(v0, Address(esp));
174 __ fabsd(v0, v0);
175 __ mov(sp, r19_sender_sp); // Restore caller's SP
176 break;
177 case Interpreter::java_lang_math_sqrt:
178 entry_point = __ pc();
179 __ ldrd(v0, Address(esp));
180 __ fsqrtd(v0, v0);
181 __ mov(sp, r19_sender_sp);
182 break;
183 case Interpreter::java_lang_math_sin :
184 case Interpreter::java_lang_math_cos :
185 case Interpreter::java_lang_math_tan :
186 case Interpreter::java_lang_math_log :
187 case Interpreter::java_lang_math_log10 :
188 case Interpreter::java_lang_math_exp :
189 entry_point = __ pc();
190 __ ldrd(v0, Address(esp));
191 __ mov(sp, r19_sender_sp);
192 __ mov(r23, lr);
193 continuation = r23; // The first free callee-saved register
194 generate_transcendental_entry(kind, 1);
195 break;
196 case Interpreter::java_lang_math_pow :
197 entry_point = __ pc();
198 __ mov(r23, lr);
199 continuation = r23;
200 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
201 __ ldrd(v1, Address(esp));
202 __ mov(sp, r19_sender_sp);
203 generate_transcendental_entry(kind, 2);
204 break;
205 case Interpreter::java_lang_math_fmaD :
206 if (UseFMA) {
207 entry_point = __ pc();
208 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
209 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
210 __ ldrd(v2, Address(esp));
211 __ fmaddd(v0, v0, v1, v2);
212 __ mov(sp, r19_sender_sp); // Restore caller's SP
213 }
214 break;
215 case Interpreter::java_lang_math_fmaF :
216 if (UseFMA) {
217 entry_point = __ pc();
218 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
219 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
220 __ ldrs(v2, Address(esp));
221 __ fmadds(v0, v0, v1, v2);
222 __ mov(sp, r19_sender_sp); // Restore caller's SP
223 }
224 break;
225 default:
226 ;
227 }
228 if (entry_point) {
229 __ br(continuation);
230 }
231
232 return entry_point;
233 }
234
235 // double trigonometrics and transcendentals
236 // static jdouble dsin(jdouble x);
237 // static jdouble dcos(jdouble x);
238 // static jdouble dtan(jdouble x);
239 // static jdouble dlog(jdouble x);
240 // static jdouble dlog10(jdouble x);
241 // static jdouble dexp(jdouble x);
242 // static jdouble dpow(jdouble x, jdouble y);
243
244 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
245 address fn;
246 switch (kind) {
247 case Interpreter::java_lang_math_sin :
248 if (StubRoutines::dsin() == NULL) {
249 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
250 } else {
251 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
252 }
253 break;
254 case Interpreter::java_lang_math_cos :
255 if (StubRoutines::dcos() == NULL) {
256 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
257 } else {
258 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
259 }
260 break;
261 case Interpreter::java_lang_math_tan :
262 if (StubRoutines::dtan() == NULL) {
263 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
264 } else {
265 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
266 }
267 break;
268 case Interpreter::java_lang_math_log :
269 if (StubRoutines::dlog() == NULL) {
270 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
271 } else {
272 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
273 }
274 break;
275 case Interpreter::java_lang_math_log10 :
276 if (StubRoutines::dlog10() == NULL) {
277 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
278 } else {
279 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
280 }
281 break;
282 case Interpreter::java_lang_math_exp :
283 if (StubRoutines::dexp() == NULL) {
284 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
285 } else {
286 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
287 }
288 break;
289 case Interpreter::java_lang_math_pow :
290 if (StubRoutines::dpow() == NULL) {
291 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
292 } else {
293 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
294 }
295 break;
296 default:
297 ShouldNotReachHere();
298 fn = NULL; // unreachable
299 }
300 __ mov(rscratch1, fn);
301 __ blr(rscratch1);
302 }
303
304 // Abstract method entry
305 // Attempt to execute abstract method. Throw exception
306 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
307 // rmethod: Method*
308 // r19_sender_sp: sender SP
309
310 address entry_point = __ pc();
311
312 // abstract method entry
313
314 // pop return address, reset last_sp to NULL
315 __ empty_expression_stack();
316 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
317 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
318
319 // throw exception
320 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
321 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
322 rmethod);
323 // the call_VM checks for exception, so we should never return here.
324 __ should_not_reach_here();
325
326 return entry_point;
327 }
328
329 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
330 address entry = __ pc();
331
332 #ifdef ASSERT
333 {
334 Label L;
335 __ ldr(rscratch1, Address(rfp,
336 frame::interpreter_frame_monitor_block_top_offset *
337 wordSize));
338 __ mov(rscratch2, sp);
339 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
340 // grows negative)
341 __ br(Assembler::HS, L); // check if frame is complete
342 __ stop ("interpreter frame not set up");
343 __ bind(L);
344 }
345 #endif // ASSERT
346 // Restore bcp under the assumption that the current frame is still
347 // interpreted
348 __ restore_bcp();
349
350 // expression stack must be empty before entering the VM if an
351 // exception happened
352 __ empty_expression_stack();
353 // throw exception
354 __ call_VM(noreg,
355 CAST_FROM_FN_PTR(address,
356 InterpreterRuntime::throw_StackOverflowError));
357 return entry;
358 }
359
360 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
361 address entry = __ pc();
362 // expression stack must be empty before entering the VM if an
363 // exception happened
364 __ empty_expression_stack();
365 // setup parameters
366
367 // ??? convention: expect aberrant index in register r1
368 __ movw(c_rarg2, r1);
369 // ??? convention: expect array in register r3
370 __ mov(c_rarg1, r3);
371 __ call_VM(noreg,
372 CAST_FROM_FN_PTR(address,
373 InterpreterRuntime::
374 throw_ArrayIndexOutOfBoundsException),
375 c_rarg1, c_rarg2);
376 return entry;
377 }
378
379 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
380 address entry = __ pc();
381
382 // object is at TOS
383 __ pop(c_rarg1);
384
385 // expression stack must be empty before entering the VM if an
386 // exception happened
387 __ empty_expression_stack();
388
389 __ call_VM(noreg,
390 CAST_FROM_FN_PTR(address,
391 InterpreterRuntime::
392 throw_ClassCastException),
393 c_rarg1);
394 return entry;
395 }
396
397 address TemplateInterpreterGenerator::generate_exception_handler_common(
398 const char* name, const char* message, bool pass_oop) {
399 assert(!pass_oop || message == NULL, "either oop or message but not both");
400 address entry = __ pc();
401 if (pass_oop) {
402 // object is at TOS
403 __ pop(c_rarg2);
404 }
405 // expression stack must be empty before entering the VM if an
406 // exception happened
407 __ empty_expression_stack();
408 // setup parameters
409 __ lea(c_rarg1, Address((address)name));
410 if (pass_oop) {
411 __ call_VM(r0, CAST_FROM_FN_PTR(address,
412 InterpreterRuntime::
413 create_klass_exception),
414 c_rarg1, c_rarg2);
415 } else {
416 // kind of lame ExternalAddress can't take NULL because
417 // external_word_Relocation will assert.
418 if (message != NULL) {
419 __ lea(c_rarg2, Address((address)message));
420 } else {
421 __ mov(c_rarg2, NULL_WORD);
422 }
423 __ call_VM(r0,
424 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
425 c_rarg1, c_rarg2);
426 }
427 // throw exception
428 __ b(address(Interpreter::throw_exception_entry()));
429 return entry;
430 }
431
432 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
433 address entry = __ pc();
434
435 // Restore stack bottom in case i2c adjusted stack
436 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
437 // and NULL it as marker that esp is now tos until next java call
438 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
439
440 if (state == atos && InlineTypeReturnedAsFields) {
441 __ store_inline_type_fields_to_buf(NULL, true);
442 }
443
444 __ restore_bcp();
445 __ restore_locals();
446 __ restore_constant_pool_cache();
447 __ get_method(rmethod);
448
449 if (state == atos) {
450 Register obj = r0;
451 Register mdp = r1;
452 Register tmp = r2;
453 __ profile_return_type(mdp, obj, tmp);
454 }
455
456 // Pop N words from the stack
457 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
458 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
459 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
460
461 __ add(esp, esp, r1, Assembler::LSL, 3);
462
463 // Restore machine SP
464 __ restore_sp_after_call();
465
466 __ check_and_handle_popframe(rthread);
467 __ check_and_handle_earlyret(rthread);
468
469 __ get_dispatch();
470 __ dispatch_next(state, step);
471
472 return entry;
473 }
474
475 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
476 int step,
477 address continuation) {
478 address entry = __ pc();
479 __ restore_bcp();
480 __ restore_locals();
481 __ restore_constant_pool_cache();
482 __ get_method(rmethod);
483 __ get_dispatch();
484
485 __ restore_sp_after_call(); // Restore SP to extended SP
486
487 // Restore expression stack pointer
488 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
489 // NULL last_sp until next java call
490 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
491
492 #if INCLUDE_JVMCI
493 // Check if we need to take lock at entry of synchronized method. This can
494 // only occur on method entry so emit it only for vtos with step 0.
495 if (EnableJVMCI && state == vtos && step == 0) {
496 Label L;
497 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
498 __ cbz(rscratch1, L);
499 // Clear flag.
500 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
501 // Take lock.
502 lock_method();
503 __ bind(L);
504 } else {
505 #ifdef ASSERT
506 if (EnableJVMCI) {
507 Label L;
508 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
509 __ cbz(rscratch1, L);
510 __ stop("unexpected pending monitor in deopt entry");
511 __ bind(L);
512 }
513 #endif
514 }
515 #endif
516 // handle exceptions
517 {
518 Label L;
519 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
520 __ cbz(rscratch1, L);
521 __ call_VM(noreg,
522 CAST_FROM_FN_PTR(address,
523 InterpreterRuntime::throw_pending_exception));
524 __ should_not_reach_here();
525 __ bind(L);
526 }
527
528 if (continuation == NULL) {
529 __ dispatch_next(state, step);
530 } else {
531 __ jump_to_entry(continuation);
532 }
533 return entry;
534 }
535
536 address TemplateInterpreterGenerator::generate_result_handler_for(
537 BasicType type) {
538 address entry = __ pc();
539 switch (type) {
540 case T_BOOLEAN: __ c2bool(r0); break;
541 case T_CHAR : __ uxth(r0, r0); break;
542 case T_BYTE : __ sxtb(r0, r0); break;
543 case T_SHORT : __ sxth(r0, r0); break;
544 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
545 case T_LONG : /* nothing to do */ break;
546 case T_VOID : /* nothing to do */ break;
547 case T_FLOAT : /* nothing to do */ break;
548 case T_DOUBLE : /* nothing to do */ break;
549 case T_PRIMITIVE_OBJECT: // fall through (value types are handled with oops)
550 case T_OBJECT :
551 // retrieve result from frame
552 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
553 // and verify it
554 __ verify_oop(r0);
555 break;
556 default : ShouldNotReachHere();
557 }
558 __ ret(lr); // return from result handler
559 return entry;
560 }
561
562 address TemplateInterpreterGenerator::generate_safept_entry_for(
563 TosState state,
564 address runtime_entry) {
565 address entry = __ pc();
566 __ push(state);
567 __ push_cont_fastpath(rthread);
568 __ call_VM(noreg, runtime_entry);
569 __ pop_cont_fastpath(rthread);
570 __ membar(Assembler::AnyAny);
571 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
572 return entry;
573 }
574
575 // Helpers for commoning out cases in the various type of method entries.
576 //
577
578
579 // increment invocation count & check for overflow
580 //
581 // Note: checking for negative value instead of overflow
582 // so we have a 'sticky' overflow test
583 //
584 // rmethod: method
585 //
586 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
587 Label done;
588 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
589 int increment = InvocationCounter::count_increment;
590 Label no_mdo;
591 if (ProfileInterpreter) {
592 // Are we profiling?
593 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
594 __ cbz(r0, no_mdo);
595 // Increment counter in the MDO
596 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
597 in_bytes(InvocationCounter::counter_offset()));
598 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
599 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
600 __ b(done);
601 }
602 __ bind(no_mdo);
603 // Increment counter in MethodCounters
604 const Address invocation_counter(rscratch2,
605 MethodCounters::invocation_counter_offset() +
606 InvocationCounter::counter_offset());
607 __ get_method_counters(rmethod, rscratch2, done);
608 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
609 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
610 __ bind(done);
611 }
612
613 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
614
615 // Asm interpreter on entry
616 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
617 // Everything as it was on entry
618
619 // InterpreterRuntime::frequency_counter_overflow takes two
620 // arguments, the first (thread) is passed by call_VM, the second
621 // indicates if the counter overflow occurs at a backwards branch
622 // (NULL bcp). We pass zero for it. The call returns the address
623 // of the verified entry point for the method or NULL if the
624 // compilation did not complete (either went background or bailed
625 // out).
626 __ mov(c_rarg1, 0);
627 __ call_VM(noreg,
628 CAST_FROM_FN_PTR(address,
629 InterpreterRuntime::frequency_counter_overflow),
630 c_rarg1);
631
632 __ b(do_continue);
633 }
634
635 // See if we've got enough room on the stack for locals plus overhead
636 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
637 // without going through the signal handler, i.e., reserved and yellow zones
638 // will not be made usable. The shadow zone must suffice to handle the
639 // overflow.
640 // The expression stack grows down incrementally, so the normal guard
641 // page mechanism will work for that.
642 //
643 // NOTE: Since the additional locals are also always pushed (wasn't
644 // obvious in generate_method_entry) so the guard should work for them
645 // too.
646 //
647 // Args:
648 // r3: number of additional locals this frame needs (what we must check)
649 // rmethod: Method*
650 //
651 // Kills:
652 // r0
653 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
654
655 // monitor entry size: see picture of stack set
656 // (generate_method_entry) and frame_amd64.hpp
657 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
658
659 // total overhead size: entry_size + (saved rbp through expr stack
660 // bottom). be sure to change this if you add/subtract anything
661 // to/from the overhead area
662 const int overhead_size =
663 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
664
665 const int page_size = os::vm_page_size();
666
667 Label after_frame_check;
668
669 // see if the frame is greater than one page in size. If so,
670 // then we need to verify there is enough stack space remaining
671 // for the additional locals.
672 //
673 // Note that we use SUBS rather than CMP here because the immediate
674 // field of this instruction may overflow. SUBS can cope with this
675 // because it is a macro that will expand to some number of MOV
676 // instructions and a register operation.
677 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
678 __ br(Assembler::LS, after_frame_check);
679
680 // compute rsp as if this were going to be the last frame on
681 // the stack before the red zone
682
683 // locals + overhead, in bytes
684 __ mov(r0, overhead_size);
685 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
686
687 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
688 __ ldr(rscratch1, stack_limit);
689
690 #ifdef ASSERT
691 Label limit_okay;
692 // Verify that thread stack limit is non-zero.
693 __ cbnz(rscratch1, limit_okay);
694 __ stop("stack overflow limit is zero");
695 __ bind(limit_okay);
696 #endif
697
698 // Add stack limit to locals.
699 __ add(r0, r0, rscratch1);
700
701 // Check against the current stack bottom.
702 __ cmp(sp, r0);
703 __ br(Assembler::HI, after_frame_check);
704
705 // Remove the incoming args, peeling the machine SP back to where it
706 // was in the caller. This is not strictly necessary, but unless we
707 // do so the stack frame may have a garbage FP; this ensures a
708 // correct call stack that we can always unwind. The ANDR should be
709 // unnecessary because the sender SP in r19 is always aligned, but
710 // it doesn't hurt.
711 __ andr(sp, r19_sender_sp, -16);
712
713 // Note: the restored frame is not necessarily interpreted.
714 // Use the shared runtime version of the StackOverflowError.
715 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
716 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
717
718 // all done with frame size check
719 __ bind(after_frame_check);
720 }
721
722 // Allocate monitor and lock method (asm interpreter)
723 //
724 // Args:
725 // rmethod: Method*
726 // rlocals: locals
727 //
728 // Kills:
729 // r0
730 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
731 // rscratch1, rscratch2 (scratch regs)
732 void TemplateInterpreterGenerator::lock_method() {
733 // synchronize method
734 const Address access_flags(rmethod, Method::access_flags_offset());
735 const Address monitor_block_top(
736 rfp,
737 frame::interpreter_frame_monitor_block_top_offset * wordSize);
738 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
739
740 #ifdef ASSERT
741 {
742 Label L;
743 __ ldrw(r0, access_flags);
744 __ tst(r0, JVM_ACC_SYNCHRONIZED);
745 __ br(Assembler::NE, L);
746 __ stop("method doesn't need synchronization");
747 __ bind(L);
748 }
749 #endif // ASSERT
750
751 // get synchronization object
752 {
753 Label done;
754 __ ldrw(r0, access_flags);
755 __ tst(r0, JVM_ACC_STATIC);
756 // get receiver (assume this is frequent case)
757 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
758 __ br(Assembler::EQ, done);
759 __ load_mirror(r0, rmethod, r5, rscratch2);
760
761 #ifdef ASSERT
762 {
763 Label L;
764 __ cbnz(r0, L);
765 __ stop("synchronization object is NULL");
766 __ bind(L);
767 }
768 #endif // ASSERT
769
770 __ bind(done);
771 }
772
773 // add space for monitor & lock
774 __ check_extended_sp();
775 __ sub(sp, sp, entry_size); // add space for a monitor entry
776 __ sub(esp, esp, entry_size);
777 __ mov(rscratch1, sp);
778 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
779 __ str(esp, monitor_block_top); // set new monitor block top
780 // store object
781 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
782 __ mov(c_rarg1, esp); // object address
783 __ lock_object(c_rarg1);
784 }
785
786 // Generate a fixed interpreter frame. This is identical setup for
787 // interpreted methods and for native methods hence the shared code.
788 //
789 // Args:
790 // lr: return address
791 // rmethod: Method*
792 // rlocals: pointer to locals
793 // rcpool: cp cache
794 // stack_pointer: previous sp
795 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
796 // initialize fixed part of activation frame
797 if (native_call) {
798 __ sub(esp, sp, 14 * wordSize);
799 __ mov(rbcp, zr);
800 __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
801 // add 2 zero-initialized slots for native calls
802 __ stp(zr, zr, Address(sp, 12 * wordSize));
803 } else {
804 __ sub(esp, sp, 12 * wordSize);
805 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
806 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
807 __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
808 }
809
810 if (ProfileInterpreter) {
811 Label method_data_continue;
812 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
813 __ cbz(rscratch1, method_data_continue);
814 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
815 __ bind(method_data_continue);
816 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
817 } else {
818 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
819 }
820
821 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
822 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
823 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
824 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
825
826 __ protect_return_address();
827 __ stp(rfp, lr, Address(sp, 10 * wordSize));
828 __ lea(rfp, Address(sp, 10 * wordSize));
829
830 // set sender sp
831 // leave last_sp as null
832 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
833
834 // Get mirror
835 __ load_mirror(r10, rmethod, r5, rscratch2);
836 if (! native_call) {
837 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
838 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
839 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
840 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
841 __ andr(rscratch1, rscratch1, -16);
842 // Store extended SP and mirror
843 __ stp(r10, rscratch1, Address(sp, 4 * wordSize));
844 // Move SP out of the way
845 __ mov(sp, rscratch1);
846 } else {
847 __ mov(rscratch1, sp);
848 __ stp(zr, rscratch1, Address(sp, 4 * wordSize));
849 }
850 }
851
852 // End of helpers
853
854 // Various method entries
855 //------------------------------------------------------------------------------------------------------------------------
856 //
857 //
858
859 // Method entry for java.lang.ref.Reference.get.
860 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
861 // Code: _aload_0, _getfield, _areturn
862 // parameter size = 1
863 //
864 // The code that gets generated by this routine is split into 2 parts:
865 // 1. The "intrinsified" code for G1 (or any SATB based GC),
866 // 2. The slow path - which is an expansion of the regular method entry.
867 //
868 // Notes:-
869 // * In the G1 code we do not check whether we need to block for
870 // a safepoint. If G1 is enabled then we must execute the specialized
871 // code for Reference.get (except when the Reference object is null)
872 // so that we can log the value in the referent field with an SATB
873 // update buffer.
874 // If the code for the getfield template is modified so that the
875 // G1 pre-barrier code is executed when the current method is
876 // Reference.get() then going through the normal method entry
877 // will be fine.
878 // * The G1 code can, however, check the receiver object (the instance
879 // of java.lang.Reference) and jump to the slow path if null. If the
880 // Reference object is null then we obviously cannot fetch the referent
881 // and so we don't need to call the G1 pre-barrier. Thus we can use the
882 // regular method entry code to generate the NPE.
883 //
884 // This code is based on generate_accessor_entry.
885 //
886 // rmethod: Method*
887 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
888
889 // LR is live. It must be saved around calls.
890
891 address entry = __ pc();
892
893 const int referent_offset = java_lang_ref_Reference::referent_offset();
894
895 Label slow_path;
896 const Register local_0 = c_rarg0;
897 // Check if local 0 != NULL
898 // If the receiver is null then it is OK to jump to the slow path.
899 __ ldr(local_0, Address(esp, 0));
900 __ cbz(local_0, slow_path);
901
902 // Load the value of the referent field.
903 const Address field_address(local_0, referent_offset);
904 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
905 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
906
907 // areturn
908 __ andr(sp, r19_sender_sp, -16); // done with stack
909 __ ret(lr);
910
911 // generate a vanilla interpreter entry as the slow path
912 __ bind(slow_path);
913 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
914 return entry;
915
916 }
917
918 /**
919 * Method entry for static native methods:
920 * int java.util.zip.CRC32.update(int crc, int b)
921 */
922 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
923 if (UseCRC32Intrinsics) {
924 address entry = __ pc();
925
926 // rmethod: Method*
927 // r19_sender_sp: senderSP must preserved for slow path
928 // esp: args
929
930 Label slow_path;
931 // If we need a safepoint check, generate full interpreter entry.
932 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
933
934 // We don't generate local frame and don't align stack because
935 // we call stub code and there is no safepoint on this path.
936
937 // Load parameters
938 const Register crc = c_rarg0; // crc
939 const Register val = c_rarg1; // source java byte value
940 const Register tbl = c_rarg2; // scratch
941
942 // Arguments are reversed on java expression stack
943 __ ldrw(val, Address(esp, 0)); // byte value
944 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
945
946 uint64_t offset;
947 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
948 __ add(tbl, tbl, offset);
949
950 __ mvnw(crc, crc); // ~crc
951 __ update_byte_crc32(crc, val, tbl);
952 __ mvnw(crc, crc); // ~crc
953
954 // result in c_rarg0
955
956 __ andr(sp, r19_sender_sp, -16);
957 __ ret(lr);
958
959 // generate a vanilla native entry as the slow path
960 __ bind(slow_path);
961 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
962 return entry;
963 }
964 return NULL;
965 }
966
967 /**
968 * Method entry for static native methods:
969 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
970 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
971 */
972 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
973 if (UseCRC32Intrinsics) {
974 address entry = __ pc();
975
976 // rmethod,: Method*
977 // r19_sender_sp: senderSP must preserved for slow path
978
979 Label slow_path;
980 // If we need a safepoint check, generate full interpreter entry.
981 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
982
983 // We don't generate local frame and don't align stack because
984 // we call stub code and there is no safepoint on this path.
985
986 // Load parameters
987 const Register crc = c_rarg0; // crc
988 const Register buf = c_rarg1; // source java byte array address
989 const Register len = c_rarg2; // length
990 const Register off = len; // offset (never overlaps with 'len')
991
992 // Arguments are reversed on java expression stack
993 // Calculate address of start element
994 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
995 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
996 __ ldrw(off, Address(esp, wordSize)); // offset
997 __ add(buf, buf, off); // + offset
998 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
999 } else {
1000 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1001 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1002 __ ldrw(off, Address(esp, wordSize)); // offset
1003 __ add(buf, buf, off); // + offset
1004 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1005 }
1006 // Can now load 'len' since we're finished with 'off'
1007 __ ldrw(len, Address(esp, 0x0)); // Length
1008
1009 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1010
1011 // We are frameless so we can just jump to the stub.
1012 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1013
1014 // generate a vanilla native entry as the slow path
1015 __ bind(slow_path);
1016 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1017 return entry;
1018 }
1019 return NULL;
1020 }
1021
1022 /**
1023 * Method entry for intrinsic-candidate (non-native) methods:
1024 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1025 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1026 * Unlike CRC32, CRC32C does not have any methods marked as native
1027 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1028 */
1029 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1030 if (UseCRC32CIntrinsics) {
1031 address entry = __ pc();
1032
1033 // Prepare jump to stub using parameters from the stack
1034 const Register crc = c_rarg0; // initial crc
1035 const Register buf = c_rarg1; // source java byte array address
1036 const Register len = c_rarg2; // len argument to the kernel
1037
1038 const Register end = len; // index of last element to process
1039 const Register off = crc; // offset
1040
1041 __ ldrw(end, Address(esp)); // int end
1042 __ ldrw(off, Address(esp, wordSize)); // int offset
1043 __ sub(len, end, off);
1044 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1045 __ add(buf, buf, off); // + offset
1046 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1047 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1048 } else {
1049 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1050 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1051 }
1052
1053 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1054
1055 // Jump to the stub.
1056 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1057
1058 return entry;
1059 }
1060 return NULL;
1061 }
1062
1063 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1064 // See more discussion in stackOverflow.hpp.
1065
1066 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1067 const int page_size = os::vm_page_size();
1068 const int n_shadow_pages = shadow_zone_size / page_size;
1069
1070 #ifdef ASSERT
1071 Label L_good_limit;
1072 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1073 __ cbnz(rscratch1, L_good_limit);
1074 __ stop("shadow zone safe limit is not initialized");
1075 __ bind(L_good_limit);
1076
1077 Label L_good_watermark;
1078 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1079 __ cbnz(rscratch1, L_good_watermark);
1080 __ stop("shadow zone growth watermark is not initialized");
1081 __ bind(L_good_watermark);
1082 #endif
1083
1084 Label L_done;
1085
1086 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1087 __ cmp(sp, rscratch1);
1088 __ br(Assembler::HI, L_done);
1089
1090 for (int p = 1; p <= n_shadow_pages; p++) {
1091 __ sub(rscratch2, sp, p*page_size);
1092 __ str(zr, Address(rscratch2));
1093 }
1094
1095 // Record the new watermark, but only if the update is above the safe limit.
1096 // Otherwise, the next time around the check above would pass the safe limit.
1097 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1098 __ cmp(sp, rscratch1);
1099 __ br(Assembler::LS, L_done);
1100 __ mov(rscratch1, sp);
1101 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1102
1103 __ bind(L_done);
1104 }
1105
1106 // Interpreter stub for calling a native method. (asm interpreter)
1107 // This sets up a somewhat different looking stack for calling the
1108 // native method than the typical interpreter frame setup.
1109 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1110 // determine code generation flags
1111 bool inc_counter = UseCompiler || CountCompiledCalls;
1112
1113 // r1: Method*
1114 // rscratch1: sender sp
1115
1116 address entry_point = __ pc();
1117
1118 const Address constMethod (rmethod, Method::const_offset());
1119 const Address access_flags (rmethod, Method::access_flags_offset());
1120 const Address size_of_parameters(r2, ConstMethod::
1121 size_of_parameters_offset());
1122
1123 // get parameter size (always needed)
1124 __ ldr(r2, constMethod);
1125 __ load_unsigned_short(r2, size_of_parameters);
1126
1127 // Native calls don't need the stack size check since they have no
1128 // expression stack and the arguments are already on the stack and
1129 // we only add a handful of words to the stack.
1130
1131 // rmethod: Method*
1132 // r2: size of parameters
1133 // rscratch1: sender sp
1134
1135 // for natives the size of locals is zero
1136
1137 // compute beginning of parameters (rlocals)
1138 __ add(rlocals, esp, r2, ext::uxtx, 3);
1139 __ add(rlocals, rlocals, -wordSize);
1140
1141 // Pull SP back to minimum size: this avoids holes in the stack
1142 __ andr(sp, esp, -16);
1143
1144 // initialize fixed part of activation frame
1145 generate_fixed_frame(true);
1146
1147 // make sure method is native & not abstract
1148 #ifdef ASSERT
1149 __ ldrw(r0, access_flags);
1150 {
1151 Label L;
1152 __ tst(r0, JVM_ACC_NATIVE);
1153 __ br(Assembler::NE, L);
1154 __ stop("tried to execute non-native method as native");
1155 __ bind(L);
1156 }
1157 {
1158 Label L;
1159 __ tst(r0, JVM_ACC_ABSTRACT);
1160 __ br(Assembler::EQ, L);
1161 __ stop("tried to execute abstract method in interpreter");
1162 __ bind(L);
1163 }
1164 #endif
1165
1166 // Since at this point in the method invocation the exception
1167 // handler would try to exit the monitor of synchronized methods
1168 // which hasn't been entered yet, we set the thread local variable
1169 // _do_not_unlock_if_synchronized to true. The remove_activation
1170 // will check this flag.
1171
1172 const Address do_not_unlock_if_synchronized(rthread,
1173 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1174 __ mov(rscratch2, true);
1175 __ strb(rscratch2, do_not_unlock_if_synchronized);
1176
1177 // increment invocation count & check for overflow
1178 Label invocation_counter_overflow;
1179 if (inc_counter) {
1180 generate_counter_incr(&invocation_counter_overflow);
1181 }
1182
1183 Label continue_after_compile;
1184 __ bind(continue_after_compile);
1185
1186 bang_stack_shadow_pages(true);
1187
1188 // reset the _do_not_unlock_if_synchronized flag
1189 __ strb(zr, do_not_unlock_if_synchronized);
1190
1191 // check for synchronized methods
1192 // Must happen AFTER invocation_counter check and stack overflow check,
1193 // so method is not locked if overflows.
1194 if (synchronized) {
1195 lock_method();
1196 } else {
1197 // no synchronization necessary
1198 #ifdef ASSERT
1199 {
1200 Label L;
1201 __ ldrw(r0, access_flags);
1202 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1203 __ br(Assembler::EQ, L);
1204 __ stop("method needs synchronization");
1205 __ bind(L);
1206 }
1207 #endif
1208 }
1209
1210 // start execution
1211 #ifdef ASSERT
1212 {
1213 Label L;
1214 const Address monitor_block_top(rfp,
1215 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1216 __ ldr(rscratch1, monitor_block_top);
1217 __ cmp(esp, rscratch1);
1218 __ br(Assembler::EQ, L);
1219 __ stop("broken stack frame setup in interpreter 1");
1220 __ bind(L);
1221 }
1222 #endif
1223
1224 // jvmti support
1225 __ notify_method_entry();
1226
1227 // work registers
1228 const Register t = r17;
1229 const Register result_handler = r19;
1230
1231 // allocate space for parameters
1232 __ ldr(t, Address(rmethod, Method::const_offset()));
1233 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1234
1235 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1236 __ andr(sp, rscratch1, -16);
1237 __ mov(esp, rscratch1);
1238
1239 // get signature handler
1240 {
1241 Label L;
1242 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1243 __ cbnz(t, L);
1244 __ call_VM(noreg,
1245 CAST_FROM_FN_PTR(address,
1246 InterpreterRuntime::prepare_native_call),
1247 rmethod);
1248 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1249 __ bind(L);
1250 }
1251
1252 // call signature handler
1253 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1254 "adjust this code");
1255 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1256 "adjust this code");
1257 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1258 "adjust this code");
1259
1260 // The generated handlers do not touch rmethod (the method).
1261 // However, large signatures cannot be cached and are generated
1262 // each time here. The slow-path generator can do a GC on return,
1263 // so we must reload it after the call.
1264 __ blr(t);
1265 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1266
1267
1268 // result handler is in r0
1269 // set result handler
1270 __ mov(result_handler, r0);
1271 // pass mirror handle if static call
1272 {
1273 Label L;
1274 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1275 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1276 // get mirror
1277 __ load_mirror(t, rmethod, r10, rscratch2);
1278 // copy mirror into activation frame
1279 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1280 // pass handle to mirror
1281 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1282 __ bind(L);
1283 }
1284
1285 // get native function entry point in r10
1286 {
1287 Label L;
1288 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1289 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1290 __ mov(rscratch2, unsatisfied);
1291 __ ldr(rscratch2, rscratch2);
1292 __ cmp(r10, rscratch2);
1293 __ br(Assembler::NE, L);
1294 __ call_VM(noreg,
1295 CAST_FROM_FN_PTR(address,
1296 InterpreterRuntime::prepare_native_call),
1297 rmethod);
1298 __ get_method(rmethod);
1299 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1300 __ bind(L);
1301 }
1302
1303 // pass JNIEnv
1304 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1305
1306 // Set the last Java PC in the frame anchor to be the return address from
1307 // the call to the native method: this will allow the debugger to
1308 // generate an accurate stack trace.
1309 Label native_return;
1310 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1311
1312 // change thread state
1313 #ifdef ASSERT
1314 {
1315 Label L;
1316 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1317 __ cmp(t, (u1)_thread_in_Java);
1318 __ br(Assembler::EQ, L);
1319 __ stop("Wrong thread state in native stub");
1320 __ bind(L);
1321 }
1322 #endif
1323
1324 // Change state to native
1325 __ mov(rscratch1, _thread_in_native);
1326 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1327 __ stlrw(rscratch1, rscratch2);
1328
1329 // Call the native method.
1330 __ blr(r10);
1331 __ bind(native_return);
1332 __ get_method(rmethod);
1333 // result potentially in r0 or v0
1334
1335 // make room for the pushes we're about to do
1336 __ sub(rscratch1, esp, 4 * wordSize);
1337 __ andr(sp, rscratch1, -16);
1338
1339 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1340 // in order to extract the result of a method call. If the order of these
1341 // pushes change or anything else is added to the stack then the code in
1342 // interpreter_frame_result must also change.
1343 __ push(dtos);
1344 __ push(ltos);
1345
1346 __ verify_sve_vector_length();
1347
1348 // change thread state
1349 __ mov(rscratch1, _thread_in_native_trans);
1350 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1351 __ stlrw(rscratch1, rscratch2);
1352
1353 // Force this write out before the read below
1354 if (!UseSystemMemoryBarrier) {
1355 __ dmb(Assembler::ISH);
1356 }
1357
1358 // check for safepoint operation in progress and/or pending suspend requests
1359 {
1360 Label L, Continue;
1361
1362 // We need an acquire here to ensure that any subsequent load of the
1363 // global SafepointSynchronize::_state flag is ordered after this load
1364 // of the thread-local polling word. We don't want this poll to
1365 // return false (i.e. not safepointing) and a later poll of the global
1366 // SafepointSynchronize::_state spuriously to return true.
1367 //
1368 // This is to avoid a race when we're in a native->Java transition
1369 // racing the code which wakes up from a safepoint.
1370 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1371 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1372 __ cbz(rscratch2, Continue);
1373 __ bind(L);
1374
1375 // Don't use call_VM as it will see a possible pending exception
1376 // and forward it and never return here preventing us from
1377 // clearing _last_native_pc down below. So we do a runtime call by
1378 // hand.
1379 //
1380 __ mov(c_rarg0, rthread);
1381 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1382 __ blr(rscratch2);
1383 __ get_method(rmethod);
1384 __ reinit_heapbase();
1385 __ bind(Continue);
1386 }
1387
1388 // change thread state
1389 __ mov(rscratch1, _thread_in_Java);
1390 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1391 __ stlrw(rscratch1, rscratch2);
1392
1393 // reset_last_Java_frame
1394 __ reset_last_Java_frame(true);
1395
1396 if (CheckJNICalls) {
1397 // clear_pending_jni_exception_check
1398 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1399 }
1400
1401 // reset handle block
1402 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1403 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1404
1405 // If result is an oop unbox and store it in frame where gc will see it
1406 // and result handler will pick it up
1407
1408 {
1409 Label no_oop;
1410 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1411 __ cmp(t, result_handler);
1412 __ br(Assembler::NE, no_oop);
1413 // Unbox oop result, e.g. JNIHandles::resolve result.
1414 __ pop(ltos);
1415 __ resolve_jobject(r0, t, rscratch2);
1416 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1417 // keep stack depth as expected by pushing oop which will eventually be discarded
1418 __ push(ltos);
1419 __ bind(no_oop);
1420 }
1421
1422 {
1423 Label no_reguard;
1424 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1425 __ ldrw(rscratch1, Address(rscratch1));
1426 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1427 __ br(Assembler::NE, no_reguard);
1428
1429 __ push_call_clobbered_registers();
1430 __ mov(c_rarg0, rthread);
1431 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1432 __ blr(rscratch2);
1433 __ pop_call_clobbered_registers();
1434
1435 __ bind(no_reguard);
1436 }
1437
1438 // The method register is junk from after the thread_in_native transition
1439 // until here. Also can't call_VM until the bcp has been
1440 // restored. Need bcp for throwing exception below so get it now.
1441 __ get_method(rmethod);
1442
1443 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1444 // rbcp == code_base()
1445 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1446 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1447 // handle exceptions (exception handling will handle unlocking!)
1448 {
1449 Label L;
1450 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1451 __ cbz(rscratch1, L);
1452 // Note: At some point we may want to unify this with the code
1453 // used in call_VM_base(); i.e., we should use the
1454 // StubRoutines::forward_exception code. For now this doesn't work
1455 // here because the rsp is not correctly set at this point.
1456 __ MacroAssembler::call_VM(noreg,
1457 CAST_FROM_FN_PTR(address,
1458 InterpreterRuntime::throw_pending_exception));
1459 __ should_not_reach_here();
1460 __ bind(L);
1461 }
1462
1463 // do unlocking if necessary
1464 {
1465 Label L;
1466 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1467 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1468 // the code below should be shared with interpreter macro
1469 // assembler implementation
1470 {
1471 Label unlock;
1472 // BasicObjectLock will be first in list, since this is a
1473 // synchronized method. However, need to check that the object
1474 // has not been unlocked by an explicit monitorexit bytecode.
1475
1476 // monitor expect in c_rarg1 for slow unlock path
1477 __ lea (c_rarg1, Address(rfp, // address of first monitor
1478 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1479 wordSize - sizeof(BasicObjectLock))));
1480
1481 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1482 __ cbnz(t, unlock);
1483
1484 // Entry already unlocked, need to throw exception
1485 __ MacroAssembler::call_VM(noreg,
1486 CAST_FROM_FN_PTR(address,
1487 InterpreterRuntime::throw_illegal_monitor_state_exception));
1488 __ should_not_reach_here();
1489
1490 __ bind(unlock);
1491 __ unlock_object(c_rarg1);
1492 }
1493 __ bind(L);
1494 }
1495
1496 // jvmti support
1497 // Note: This must happen _after_ handling/throwing any exceptions since
1498 // the exception handler code notifies the runtime of method exits
1499 // too. If this happens before, method entry/exit notifications are
1500 // not properly paired (was bug - gri 11/22/99).
1501 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1502
1503 // restore potential result in r0:d0, call result handler to
1504 // restore potential result in ST0 & handle result
1505
1506 __ pop(ltos);
1507 __ pop(dtos);
1508
1509 __ blr(result_handler);
1510
1511 // remove activation
1512 __ ldr(esp, Address(rfp,
1513 frame::interpreter_frame_sender_sp_offset *
1514 wordSize)); // get sender sp
1515 // remove frame anchor
1516 __ leave();
1517
1518 // restore sender sp
1519 __ mov(sp, esp);
1520
1521 __ ret(lr);
1522
1523 if (inc_counter) {
1524 // Handle overflow of counter and compile method
1525 __ bind(invocation_counter_overflow);
1526 generate_counter_overflow(continue_after_compile);
1527 }
1528
1529 return entry_point;
1530 }
1531
1532 //
1533 // Generic interpreted method entry to (asm) interpreter
1534 //
1535 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1536 // determine code generation flags
1537 bool inc_counter = UseCompiler || CountCompiledCalls;
1538
1539 // rscratch1: sender sp
1540 address entry_point = __ pc();
1541
1542 const Address constMethod(rmethod, Method::const_offset());
1543 const Address access_flags(rmethod, Method::access_flags_offset());
1544 const Address size_of_parameters(r3,
1545 ConstMethod::size_of_parameters_offset());
1546 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1547
1548 // get parameter size (always needed)
1549 // need to load the const method first
1550 __ ldr(r3, constMethod);
1551 __ load_unsigned_short(r2, size_of_parameters);
1552
1553 // r2: size of parameters
1554
1555 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1556 __ sub(r3, r3, r2); // r3 = no. of additional locals
1557
1558 // see if we've got enough room on the stack for locals plus overhead.
1559 generate_stack_overflow_check();
1560
1561 // compute beginning of parameters (rlocals)
1562 __ add(rlocals, esp, r2, ext::uxtx, 3);
1563 __ sub(rlocals, rlocals, wordSize);
1564
1565 __ mov(rscratch1, esp);
1566
1567 // r3 - # of additional locals
1568 // allocate space for locals
1569 // explicitly initialize locals
1570 // Initializing memory allocated for locals in the same direction as
1571 // the stack grows to ensure page initialization order according
1572 // to windows-aarch64 stack page growth requirement (see
1573 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1574 {
1575 Label exit, loop;
1576 __ ands(zr, r3, r3);
1577 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1578 __ bind(loop);
1579 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1580 __ sub(r3, r3, 1); // until everything initialized
1581 __ cbnz(r3, loop);
1582 __ bind(exit);
1583 }
1584
1585 // Padding between locals and fixed part of activation frame to ensure
1586 // SP is always 16-byte aligned.
1587 __ andr(sp, rscratch1, -16);
1588
1589 // And the base dispatch table
1590 __ get_dispatch();
1591
1592 // initialize fixed part of activation frame
1593 generate_fixed_frame(false);
1594
1595 // make sure method is not native & not abstract
1596 #ifdef ASSERT
1597 __ ldrw(r0, access_flags);
1598 {
1599 Label L;
1600 __ tst(r0, JVM_ACC_NATIVE);
1601 __ br(Assembler::EQ, L);
1602 __ stop("tried to execute native method as non-native");
1603 __ bind(L);
1604 }
1605 {
1606 Label L;
1607 __ tst(r0, JVM_ACC_ABSTRACT);
1608 __ br(Assembler::EQ, L);
1609 __ stop("tried to execute abstract method in interpreter");
1610 __ bind(L);
1611 }
1612 #endif
1613
1614 // Since at this point in the method invocation the exception
1615 // handler would try to exit the monitor of synchronized methods
1616 // which hasn't been entered yet, we set the thread local variable
1617 // _do_not_unlock_if_synchronized to true. The remove_activation
1618 // will check this flag.
1619
1620 const Address do_not_unlock_if_synchronized(rthread,
1621 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1622 __ mov(rscratch2, true);
1623 __ strb(rscratch2, do_not_unlock_if_synchronized);
1624
1625 Register mdp = r3;
1626 __ profile_parameters_type(mdp, r1, r2);
1627
1628 // increment invocation count & check for overflow
1629 Label invocation_counter_overflow;
1630 if (inc_counter) {
1631 generate_counter_incr(&invocation_counter_overflow);
1632 }
1633
1634 Label continue_after_compile;
1635 __ bind(continue_after_compile);
1636
1637 bang_stack_shadow_pages(false);
1638
1639 // reset the _do_not_unlock_if_synchronized flag
1640 __ strb(zr, do_not_unlock_if_synchronized);
1641
1642 // check for synchronized methods
1643 // Must happen AFTER invocation_counter check and stack overflow check,
1644 // so method is not locked if overflows.
1645 if (synchronized) {
1646 // Allocate monitor and lock method
1647 lock_method();
1648 } else {
1649 // no synchronization necessary
1650 #ifdef ASSERT
1651 {
1652 Label L;
1653 __ ldrw(r0, access_flags);
1654 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1655 __ br(Assembler::EQ, L);
1656 __ stop("method needs synchronization");
1657 __ bind(L);
1658 }
1659 #endif
1660 }
1661
1662 // start execution
1663 #ifdef ASSERT
1664 {
1665 Label L;
1666 const Address monitor_block_top (rfp,
1667 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1668 __ ldr(rscratch1, monitor_block_top);
1669 __ cmp(esp, rscratch1);
1670 __ br(Assembler::EQ, L);
1671 __ stop("broken stack frame setup in interpreter 2");
1672 __ bind(L);
1673 }
1674 #endif
1675
1676 // jvmti support
1677 __ notify_method_entry();
1678
1679 __ dispatch_next(vtos);
1680
1681 // invocation counter overflow
1682 if (inc_counter) {
1683 // Handle overflow of counter and compile method
1684 __ bind(invocation_counter_overflow);
1685 generate_counter_overflow(continue_after_compile);
1686 }
1687
1688 return entry_point;
1689 }
1690
1691 //-----------------------------------------------------------------------------
1692 // Exceptions
1693
1694 void TemplateInterpreterGenerator::generate_throw_exception() {
1695 // Entry point in previous activation (i.e., if the caller was
1696 // interpreted)
1697 Interpreter::_rethrow_exception_entry = __ pc();
1698 // Restore sp to interpreter_frame_last_sp even though we are going
1699 // to empty the expression stack for the exception processing.
1700 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1701 // r0: exception
1702 // r3: return address/pc that threw exception
1703 __ restore_bcp(); // rbcp points to call/send
1704 __ restore_locals();
1705 __ restore_constant_pool_cache();
1706 __ reinit_heapbase(); // restore rheapbase as heapbase.
1707 __ get_dispatch();
1708
1709 // Entry point for exceptions thrown within interpreter code
1710 Interpreter::_throw_exception_entry = __ pc();
1711 // If we came here via a NullPointerException on the receiver of a
1712 // method, rmethod may be corrupt.
1713 __ get_method(rmethod);
1714 // expression stack is undefined here
1715 // r0: exception
1716 // rbcp: exception bcp
1717 __ verify_oop(r0);
1718 __ mov(c_rarg1, r0);
1719
1720 // expression stack must be empty before entering the VM in case of
1721 // an exception
1722 __ empty_expression_stack();
1723 // find exception handler address and preserve exception oop
1724 __ call_VM(r3,
1725 CAST_FROM_FN_PTR(address,
1726 InterpreterRuntime::exception_handler_for_exception),
1727 c_rarg1);
1728
1729 // Restore machine SP
1730 __ restore_sp_after_call();
1731
1732 // r0: exception handler entry point
1733 // r3: preserved exception oop
1734 // rbcp: bcp for exception handler
1735 __ push_ptr(r3); // push exception which is now the only value on the stack
1736 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1737
1738 // If the exception is not handled in the current frame the frame is
1739 // removed and the exception is rethrown (i.e. exception
1740 // continuation is _rethrow_exception).
1741 //
1742 // Note: At this point the bci is still the bxi for the instruction
1743 // which caused the exception and the expression stack is
1744 // empty. Thus, for any VM calls at this point, GC will find a legal
1745 // oop map (with empty expression stack).
1746
1747 //
1748 // JVMTI PopFrame support
1749 //
1750
1751 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1752 __ empty_expression_stack();
1753 // Set the popframe_processing bit in pending_popframe_condition
1754 // indicating that we are currently handling popframe, so that
1755 // call_VMs that may happen later do not trigger new popframe
1756 // handling cycles.
1757 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1758 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1759 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1760
1761 {
1762 // Check to see whether we are returning to a deoptimized frame.
1763 // (The PopFrame call ensures that the caller of the popped frame is
1764 // either interpreted or compiled and deoptimizes it if compiled.)
1765 // In this case, we can't call dispatch_next() after the frame is
1766 // popped, but instead must save the incoming arguments and restore
1767 // them after deoptimization has occurred.
1768 //
1769 // Note that we don't compare the return PC against the
1770 // deoptimization blob's unpack entry because of the presence of
1771 // adapter frames in C2.
1772 Label caller_not_deoptimized;
1773 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1774 // This is a return address, so requires authenticating for PAC.
1775 __ authenticate_return_address(c_rarg1, rscratch1);
1776 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1777 InterpreterRuntime::interpreter_contains), c_rarg1);
1778 __ cbnz(r0, caller_not_deoptimized);
1779
1780 // Compute size of arguments for saving when returning to
1781 // deoptimized caller
1782 __ get_method(r0);
1783 __ ldr(r0, Address(r0, Method::const_offset()));
1784 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1785 size_of_parameters_offset())));
1786 __ lsl(r0, r0, Interpreter::logStackElementSize);
1787 __ restore_locals(); // XXX do we need this?
1788 __ sub(rlocals, rlocals, r0);
1789 __ add(rlocals, rlocals, wordSize);
1790 // Save these arguments
1791 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1792 Deoptimization::
1793 popframe_preserve_args),
1794 rthread, r0, rlocals);
1795
1796 __ remove_activation(vtos,
1797 /* throw_monitor_exception */ false,
1798 /* install_monitor_exception */ false,
1799 /* notify_jvmdi */ false);
1800
1801 // Inform deoptimization that it is responsible for restoring
1802 // these arguments
1803 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1804 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1805
1806 // Continue in deoptimization handler
1807 __ ret(lr);
1808
1809 __ bind(caller_not_deoptimized);
1810 }
1811
1812 __ remove_activation(vtos,
1813 /* throw_monitor_exception */ false,
1814 /* install_monitor_exception */ false,
1815 /* notify_jvmdi */ false);
1816
1817 // Restore the last_sp and null it out
1818 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1819 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1820
1821 __ restore_bcp();
1822 __ restore_locals();
1823 __ restore_constant_pool_cache();
1824 __ get_method(rmethod);
1825 __ get_dispatch();
1826
1827 // The method data pointer was incremented already during
1828 // call profiling. We have to restore the mdp for the current bcp.
1829 if (ProfileInterpreter) {
1830 __ set_method_data_pointer_for_bcp();
1831 }
1832
1833 // Clear the popframe condition flag
1834 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1835 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1836
1837 #if INCLUDE_JVMTI
1838 {
1839 Label L_done;
1840
1841 __ ldrb(rscratch1, Address(rbcp, 0));
1842 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1843 __ br(Assembler::NE, L_done);
1844
1845 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1846 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1847
1848 __ ldr(c_rarg0, Address(rlocals, 0));
1849 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1850
1851 __ cbz(r0, L_done);
1852
1853 __ str(r0, Address(esp, 0));
1854 __ bind(L_done);
1855 }
1856 #endif // INCLUDE_JVMTI
1857
1858 // Restore machine SP
1859 __ restore_sp_after_call();
1860
1861 __ dispatch_next(vtos);
1862 // end of PopFrame support
1863
1864 Interpreter::_remove_activation_entry = __ pc();
1865
1866 // preserve exception over this code sequence
1867 __ pop_ptr(r0);
1868 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1869 // remove the activation (without doing throws on illegalMonitorExceptions)
1870 __ remove_activation(vtos, false, true, false);
1871 // restore exception
1872 __ get_vm_result(r0, rthread);
1873
1874 // In between activations - previous activation type unknown yet
1875 // compute continuation point - the continuation point expects the
1876 // following registers set up:
1877 //
1878 // r0: exception
1879 // lr: return address/pc that threw exception
1880 // esp: expression stack of caller
1881 // rfp: fp of caller
1882 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1883 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1884 SharedRuntime::exception_handler_for_return_address),
1885 rthread, lr);
1886 __ mov(r1, r0); // save exception handler
1887 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1888 // We might be returning to a deopt handler that expects r3 to
1889 // contain the exception pc
1890 __ mov(r3, lr);
1891 // Note that an "issuing PC" is actually the next PC after the call
1892 __ br(r1); // jump to exception
1893 // handler of caller
1894 }
1895
1896
1897 //
1898 // JVMTI ForceEarlyReturn support
1899 //
1900 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1901 address entry = __ pc();
1902
1903 __ restore_bcp();
1904 __ restore_locals();
1905 __ empty_expression_stack();
1906 __ load_earlyret_value(state);
1907
1908 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1909 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1910
1911 // Clear the earlyret state
1912 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1913 __ str(zr, cond_addr);
1914
1915 __ remove_activation(state,
1916 false, /* throw_monitor_exception */
1917 false, /* install_monitor_exception */
1918 true); /* notify_jvmdi */
1919 __ ret(lr);
1920
1921 return entry;
1922 } // end of ForceEarlyReturn support
1923
1924
1925
1926 //-----------------------------------------------------------------------------
1927 // Helper for vtos entry point generation
1928
1929 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1930 address& bep,
1931 address& cep,
1932 address& sep,
1933 address& aep,
1934 address& iep,
1935 address& lep,
1936 address& fep,
1937 address& dep,
1938 address& vep) {
1939 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1940 Label L;
1941 aep = __ pc(); __ push_ptr(); __ b(L);
1942 fep = __ pc(); __ push_f(); __ b(L);
1943 dep = __ pc(); __ push_d(); __ b(L);
1944 lep = __ pc(); __ push_l(); __ b(L);
1945 bep = cep = sep =
1946 iep = __ pc(); __ push_i();
1947 vep = __ pc();
1948 __ bind(L);
1949 generate_and_dispatch(t);
1950 }
1951
1952 //-----------------------------------------------------------------------------
1953
1954 // Non-product code
1955 #ifndef PRODUCT
1956 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1957 address entry = __ pc();
1958
1959 __ protect_return_address();
1960 __ push(lr);
1961 __ push(state);
1962 __ push(RegSet::range(r0, r15), sp);
1963 __ mov(c_rarg2, r0); // Pass itos
1964 __ call_VM(noreg,
1965 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1966 c_rarg1, c_rarg2, c_rarg3);
1967 __ pop(RegSet::range(r0, r15), sp);
1968 __ pop(state);
1969 __ pop(lr);
1970 __ authenticate_return_address();
1971 __ ret(lr); // return from result handler
1972
1973 return entry;
1974 }
1975
1976 void TemplateInterpreterGenerator::count_bytecode() {
1977 Register rscratch3 = r0;
1978 __ push(rscratch1);
1979 __ push(rscratch2);
1980 __ push(rscratch3);
1981 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
1982 __ atomic_add(noreg, 1, rscratch3);
1983 __ pop(rscratch3);
1984 __ pop(rscratch2);
1985 __ pop(rscratch1);
1986 }
1987
1988 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
1989
1990 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
1991
1992
1993 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1994 // Call a little run-time stub to avoid blow-up for each bytecode.
1995 // The run-time runtime saves the right registers, depending on
1996 // the tosca in-state for the given template.
1997
1998 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1999 "entry must have been generated");
2000 __ bl(Interpreter::trace_code(t->tos_in()));
2001 __ reinit_heapbase();
2002 }
2003
2004
2005 void TemplateInterpreterGenerator::stop_interpreter_at() {
2006 Label L;
2007 __ push(rscratch1);
2008 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2009 __ ldr(rscratch1, Address(rscratch1));
2010 __ mov(rscratch2, StopInterpreterAt);
2011 __ cmpw(rscratch1, rscratch2);
2012 __ br(Assembler::NE, L);
2013 __ brk(0);
2014 __ bind(L);
2015 __ pop(rscratch1);
2016 }
2017
2018 #endif // !PRODUCT