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 // Make sure there is room for the exception oop pushed in case method throws
848 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
849 __ sub(rscratch1, sp, 2 * wordSize);
850 __ stp(zr, rscratch1, Address(sp, 4 * wordSize));
851 __ mov(sp, rscratch1);
852 }
853 }
854
855 // End of helpers
856
857 // Various method entries
858 //------------------------------------------------------------------------------------------------------------------------
859 //
860 //
861
862 // Method entry for java.lang.ref.Reference.get.
863 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
864 // Code: _aload_0, _getfield, _areturn
865 // parameter size = 1
866 //
867 // The code that gets generated by this routine is split into 2 parts:
868 // 1. The "intrinsified" code for G1 (or any SATB based GC),
869 // 2. The slow path - which is an expansion of the regular method entry.
870 //
871 // Notes:-
872 // * In the G1 code we do not check whether we need to block for
873 // a safepoint. If G1 is enabled then we must execute the specialized
874 // code for Reference.get (except when the Reference object is null)
875 // so that we can log the value in the referent field with an SATB
876 // update buffer.
877 // If the code for the getfield template is modified so that the
878 // G1 pre-barrier code is executed when the current method is
879 // Reference.get() then going through the normal method entry
880 // will be fine.
881 // * The G1 code can, however, check the receiver object (the instance
882 // of java.lang.Reference) and jump to the slow path if null. If the
883 // Reference object is null then we obviously cannot fetch the referent
884 // and so we don't need to call the G1 pre-barrier. Thus we can use the
885 // regular method entry code to generate the NPE.
886 //
887 // This code is based on generate_accessor_entry.
888 //
889 // rmethod: Method*
890 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
891
892 // LR is live. It must be saved around calls.
893
894 address entry = __ pc();
895
896 const int referent_offset = java_lang_ref_Reference::referent_offset();
897
898 Label slow_path;
899 const Register local_0 = c_rarg0;
900 // Check if local 0 != NULL
901 // If the receiver is null then it is OK to jump to the slow path.
902 __ ldr(local_0, Address(esp, 0));
903 __ cbz(local_0, slow_path);
904
905 // Load the value of the referent field.
906 const Address field_address(local_0, referent_offset);
907 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
908 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
909
910 // areturn
911 __ andr(sp, r19_sender_sp, -16); // done with stack
912 __ ret(lr);
913
914 // generate a vanilla interpreter entry as the slow path
915 __ bind(slow_path);
916 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
917 return entry;
918
919 }
920
921 /**
922 * Method entry for static native methods:
923 * int java.util.zip.CRC32.update(int crc, int b)
924 */
925 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
926 if (UseCRC32Intrinsics) {
927 address entry = __ pc();
928
929 // rmethod: Method*
930 // r19_sender_sp: senderSP must preserved for slow path
931 // esp: args
932
933 Label slow_path;
934 // If we need a safepoint check, generate full interpreter entry.
935 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
936
937 // We don't generate local frame and don't align stack because
938 // we call stub code and there is no safepoint on this path.
939
940 // Load parameters
941 const Register crc = c_rarg0; // crc
942 const Register val = c_rarg1; // source java byte value
943 const Register tbl = c_rarg2; // scratch
944
945 // Arguments are reversed on java expression stack
946 __ ldrw(val, Address(esp, 0)); // byte value
947 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
948
949 uint64_t offset;
950 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
951 __ add(tbl, tbl, offset);
952
953 __ mvnw(crc, crc); // ~crc
954 __ update_byte_crc32(crc, val, tbl);
955 __ mvnw(crc, crc); // ~crc
956
957 // result in c_rarg0
958
959 __ andr(sp, r19_sender_sp, -16);
960 __ ret(lr);
961
962 // generate a vanilla native entry as the slow path
963 __ bind(slow_path);
964 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
965 return entry;
966 }
967 return NULL;
968 }
969
970 /**
971 * Method entry for static native methods:
972 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
973 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
974 */
975 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
976 if (UseCRC32Intrinsics) {
977 address entry = __ pc();
978
979 // rmethod,: Method*
980 // r19_sender_sp: senderSP must preserved for slow path
981
982 Label slow_path;
983 // If we need a safepoint check, generate full interpreter entry.
984 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
985
986 // We don't generate local frame and don't align stack because
987 // we call stub code and there is no safepoint on this path.
988
989 // Load parameters
990 const Register crc = c_rarg0; // crc
991 const Register buf = c_rarg1; // source java byte array address
992 const Register len = c_rarg2; // length
993 const Register off = len; // offset (never overlaps with 'len')
994
995 // Arguments are reversed on java expression stack
996 // Calculate address of start element
997 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
998 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
999 __ ldrw(off, Address(esp, wordSize)); // offset
1000 __ add(buf, buf, off); // + offset
1001 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1002 } else {
1003 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1004 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1005 __ ldrw(off, Address(esp, wordSize)); // offset
1006 __ add(buf, buf, off); // + offset
1007 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1008 }
1009 // Can now load 'len' since we're finished with 'off'
1010 __ ldrw(len, Address(esp, 0x0)); // Length
1011
1012 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1013
1014 // We are frameless so we can just jump to the stub.
1015 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1016
1017 // generate a vanilla native entry as the slow path
1018 __ bind(slow_path);
1019 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1020 return entry;
1021 }
1022 return NULL;
1023 }
1024
1025 /**
1026 * Method entry for intrinsic-candidate (non-native) methods:
1027 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1028 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1029 * Unlike CRC32, CRC32C does not have any methods marked as native
1030 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1031 */
1032 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1033 if (UseCRC32CIntrinsics) {
1034 address entry = __ pc();
1035
1036 // Prepare jump to stub using parameters from the stack
1037 const Register crc = c_rarg0; // initial crc
1038 const Register buf = c_rarg1; // source java byte array address
1039 const Register len = c_rarg2; // len argument to the kernel
1040
1041 const Register end = len; // index of last element to process
1042 const Register off = crc; // offset
1043
1044 __ ldrw(end, Address(esp)); // int end
1045 __ ldrw(off, Address(esp, wordSize)); // int offset
1046 __ sub(len, end, off);
1047 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1048 __ add(buf, buf, off); // + offset
1049 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1050 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1051 } else {
1052 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1053 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1054 }
1055
1056 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1057
1058 // Jump to the stub.
1059 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1060
1061 return entry;
1062 }
1063 return NULL;
1064 }
1065
1066 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1067 // See more discussion in stackOverflow.hpp.
1068
1069 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1070 const int page_size = os::vm_page_size();
1071 const int n_shadow_pages = shadow_zone_size / page_size;
1072
1073 #ifdef ASSERT
1074 Label L_good_limit;
1075 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1076 __ cbnz(rscratch1, L_good_limit);
1077 __ stop("shadow zone safe limit is not initialized");
1078 __ bind(L_good_limit);
1079
1080 Label L_good_watermark;
1081 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1082 __ cbnz(rscratch1, L_good_watermark);
1083 __ stop("shadow zone growth watermark is not initialized");
1084 __ bind(L_good_watermark);
1085 #endif
1086
1087 Label L_done;
1088
1089 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1090 __ cmp(sp, rscratch1);
1091 __ br(Assembler::HI, L_done);
1092
1093 for (int p = 1; p <= n_shadow_pages; p++) {
1094 __ sub(rscratch2, sp, p*page_size);
1095 __ str(zr, Address(rscratch2));
1096 }
1097
1098 // Record the new watermark, but only if the update is above the safe limit.
1099 // Otherwise, the next time around the check above would pass the safe limit.
1100 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1101 __ cmp(sp, rscratch1);
1102 __ br(Assembler::LS, L_done);
1103 __ mov(rscratch1, sp);
1104 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1105
1106 __ bind(L_done);
1107 }
1108
1109 // Interpreter stub for calling a native method. (asm interpreter)
1110 // This sets up a somewhat different looking stack for calling the
1111 // native method than the typical interpreter frame setup.
1112 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1113 // determine code generation flags
1114 bool inc_counter = UseCompiler || CountCompiledCalls;
1115
1116 // r1: Method*
1117 // rscratch1: sender sp
1118
1119 address entry_point = __ pc();
1120
1121 const Address constMethod (rmethod, Method::const_offset());
1122 const Address access_flags (rmethod, Method::access_flags_offset());
1123 const Address size_of_parameters(r2, ConstMethod::
1124 size_of_parameters_offset());
1125
1126 // get parameter size (always needed)
1127 __ ldr(r2, constMethod);
1128 __ load_unsigned_short(r2, size_of_parameters);
1129
1130 // Native calls don't need the stack size check since they have no
1131 // expression stack and the arguments are already on the stack and
1132 // we only add a handful of words to the stack.
1133
1134 // rmethod: Method*
1135 // r2: size of parameters
1136 // rscratch1: sender sp
1137
1138 // for natives the size of locals is zero
1139
1140 // compute beginning of parameters (rlocals)
1141 __ add(rlocals, esp, r2, ext::uxtx, 3);
1142 __ add(rlocals, rlocals, -wordSize);
1143
1144 // Pull SP back to minimum size: this avoids holes in the stack
1145 __ andr(sp, esp, -16);
1146
1147 // initialize fixed part of activation frame
1148 generate_fixed_frame(true);
1149
1150 // make sure method is native & not abstract
1151 #ifdef ASSERT
1152 __ ldrw(r0, access_flags);
1153 {
1154 Label L;
1155 __ tst(r0, JVM_ACC_NATIVE);
1156 __ br(Assembler::NE, L);
1157 __ stop("tried to execute non-native method as native");
1158 __ bind(L);
1159 }
1160 {
1161 Label L;
1162 __ tst(r0, JVM_ACC_ABSTRACT);
1163 __ br(Assembler::EQ, L);
1164 __ stop("tried to execute abstract method in interpreter");
1165 __ bind(L);
1166 }
1167 #endif
1168
1169 // Since at this point in the method invocation the exception
1170 // handler would try to exit the monitor of synchronized methods
1171 // which hasn't been entered yet, we set the thread local variable
1172 // _do_not_unlock_if_synchronized to true. The remove_activation
1173 // will check this flag.
1174
1175 const Address do_not_unlock_if_synchronized(rthread,
1176 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1177 __ mov(rscratch2, true);
1178 __ strb(rscratch2, do_not_unlock_if_synchronized);
1179
1180 // increment invocation count & check for overflow
1181 Label invocation_counter_overflow;
1182 if (inc_counter) {
1183 generate_counter_incr(&invocation_counter_overflow);
1184 }
1185
1186 Label continue_after_compile;
1187 __ bind(continue_after_compile);
1188
1189 bang_stack_shadow_pages(true);
1190
1191 // reset the _do_not_unlock_if_synchronized flag
1192 __ strb(zr, do_not_unlock_if_synchronized);
1193
1194 // check for synchronized methods
1195 // Must happen AFTER invocation_counter check and stack overflow check,
1196 // so method is not locked if overflows.
1197 if (synchronized) {
1198 lock_method();
1199 } else {
1200 // no synchronization necessary
1201 #ifdef ASSERT
1202 {
1203 Label L;
1204 __ ldrw(r0, access_flags);
1205 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1206 __ br(Assembler::EQ, L);
1207 __ stop("method needs synchronization");
1208 __ bind(L);
1209 }
1210 #endif
1211 }
1212
1213 // start execution
1214 #ifdef ASSERT
1215 {
1216 Label L;
1217 const Address monitor_block_top(rfp,
1218 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1219 __ ldr(rscratch1, monitor_block_top);
1220 __ cmp(esp, rscratch1);
1221 __ br(Assembler::EQ, L);
1222 __ stop("broken stack frame setup in interpreter 1");
1223 __ bind(L);
1224 }
1225 #endif
1226
1227 // jvmti support
1228 __ notify_method_entry();
1229
1230 // work registers
1231 const Register t = r17;
1232 const Register result_handler = r19;
1233
1234 // allocate space for parameters
1235 __ ldr(t, Address(rmethod, Method::const_offset()));
1236 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1237
1238 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1239 __ andr(sp, rscratch1, -16);
1240 __ mov(esp, rscratch1);
1241
1242 // get signature handler
1243 {
1244 Label L;
1245 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1246 __ cbnz(t, L);
1247 __ call_VM(noreg,
1248 CAST_FROM_FN_PTR(address,
1249 InterpreterRuntime::prepare_native_call),
1250 rmethod);
1251 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1252 __ bind(L);
1253 }
1254
1255 // call signature handler
1256 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1257 "adjust this code");
1258 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1259 "adjust this code");
1260 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1261 "adjust this code");
1262
1263 // The generated handlers do not touch rmethod (the method).
1264 // However, large signatures cannot be cached and are generated
1265 // each time here. The slow-path generator can do a GC on return,
1266 // so we must reload it after the call.
1267 __ blr(t);
1268 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1269
1270
1271 // result handler is in r0
1272 // set result handler
1273 __ mov(result_handler, r0);
1274 // pass mirror handle if static call
1275 {
1276 Label L;
1277 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1278 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1279 // get mirror
1280 __ load_mirror(t, rmethod, r10, rscratch2);
1281 // copy mirror into activation frame
1282 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1283 // pass handle to mirror
1284 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1285 __ bind(L);
1286 }
1287
1288 // get native function entry point in r10
1289 {
1290 Label L;
1291 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1292 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1293 __ mov(rscratch2, unsatisfied);
1294 __ ldr(rscratch2, rscratch2);
1295 __ cmp(r10, rscratch2);
1296 __ br(Assembler::NE, L);
1297 __ call_VM(noreg,
1298 CAST_FROM_FN_PTR(address,
1299 InterpreterRuntime::prepare_native_call),
1300 rmethod);
1301 __ get_method(rmethod);
1302 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1303 __ bind(L);
1304 }
1305
1306 // pass JNIEnv
1307 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1308
1309 // Set the last Java PC in the frame anchor to be the return address from
1310 // the call to the native method: this will allow the debugger to
1311 // generate an accurate stack trace.
1312 Label native_return;
1313 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1314
1315 // change thread state
1316 #ifdef ASSERT
1317 {
1318 Label L;
1319 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1320 __ cmp(t, (u1)_thread_in_Java);
1321 __ br(Assembler::EQ, L);
1322 __ stop("Wrong thread state in native stub");
1323 __ bind(L);
1324 }
1325 #endif
1326
1327 // Change state to native
1328 __ mov(rscratch1, _thread_in_native);
1329 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1330 __ stlrw(rscratch1, rscratch2);
1331
1332 // Call the native method.
1333 __ blr(r10);
1334 __ bind(native_return);
1335 __ get_method(rmethod);
1336 // result potentially in r0 or v0
1337
1338 // make room for the pushes we're about to do
1339 __ sub(rscratch1, esp, 4 * wordSize);
1340 __ andr(sp, rscratch1, -16);
1341
1342 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1343 // in order to extract the result of a method call. If the order of these
1344 // pushes change or anything else is added to the stack then the code in
1345 // interpreter_frame_result must also change.
1346 __ push(dtos);
1347 __ push(ltos);
1348
1349 __ verify_sve_vector_length();
1350
1351 // change thread state
1352 __ mov(rscratch1, _thread_in_native_trans);
1353 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1354 __ stlrw(rscratch1, rscratch2);
1355
1356 // Force this write out before the read below
1357 if (!UseSystemMemoryBarrier) {
1358 __ dmb(Assembler::ISH);
1359 }
1360
1361 // check for safepoint operation in progress and/or pending suspend requests
1362 {
1363 Label L, Continue;
1364
1365 // We need an acquire here to ensure that any subsequent load of the
1366 // global SafepointSynchronize::_state flag is ordered after this load
1367 // of the thread-local polling word. We don't want this poll to
1368 // return false (i.e. not safepointing) and a later poll of the global
1369 // SafepointSynchronize::_state spuriously to return true.
1370 //
1371 // This is to avoid a race when we're in a native->Java transition
1372 // racing the code which wakes up from a safepoint.
1373 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1374 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1375 __ cbz(rscratch2, Continue);
1376 __ bind(L);
1377
1378 // Don't use call_VM as it will see a possible pending exception
1379 // and forward it and never return here preventing us from
1380 // clearing _last_native_pc down below. So we do a runtime call by
1381 // hand.
1382 //
1383 __ mov(c_rarg0, rthread);
1384 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1385 __ blr(rscratch2);
1386 __ get_method(rmethod);
1387 __ reinit_heapbase();
1388 __ bind(Continue);
1389 }
1390
1391 // change thread state
1392 __ mov(rscratch1, _thread_in_Java);
1393 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1394 __ stlrw(rscratch1, rscratch2);
1395
1396 // reset_last_Java_frame
1397 __ reset_last_Java_frame(true);
1398
1399 if (CheckJNICalls) {
1400 // clear_pending_jni_exception_check
1401 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1402 }
1403
1404 // reset handle block
1405 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1406 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1407
1408 // If result is an oop unbox and store it in frame where gc will see it
1409 // and result handler will pick it up
1410
1411 {
1412 Label no_oop;
1413 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1414 __ cmp(t, result_handler);
1415 __ br(Assembler::NE, no_oop);
1416 // Unbox oop result, e.g. JNIHandles::resolve result.
1417 __ pop(ltos);
1418 __ resolve_jobject(r0, t, rscratch2);
1419 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1420 // keep stack depth as expected by pushing oop which will eventually be discarded
1421 __ push(ltos);
1422 __ bind(no_oop);
1423 }
1424
1425 {
1426 Label no_reguard;
1427 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1428 __ ldrw(rscratch1, Address(rscratch1));
1429 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1430 __ br(Assembler::NE, no_reguard);
1431
1432 __ push_call_clobbered_registers();
1433 __ mov(c_rarg0, rthread);
1434 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1435 __ blr(rscratch2);
1436 __ pop_call_clobbered_registers();
1437
1438 __ bind(no_reguard);
1439 }
1440
1441 // The method register is junk from after the thread_in_native transition
1442 // until here. Also can't call_VM until the bcp has been
1443 // restored. Need bcp for throwing exception below so get it now.
1444 __ get_method(rmethod);
1445
1446 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1447 // rbcp == code_base()
1448 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1449 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1450 // handle exceptions (exception handling will handle unlocking!)
1451 {
1452 Label L;
1453 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1454 __ cbz(rscratch1, L);
1455 // Note: At some point we may want to unify this with the code
1456 // used in call_VM_base(); i.e., we should use the
1457 // StubRoutines::forward_exception code. For now this doesn't work
1458 // here because the rsp is not correctly set at this point.
1459 __ MacroAssembler::call_VM(noreg,
1460 CAST_FROM_FN_PTR(address,
1461 InterpreterRuntime::throw_pending_exception));
1462 __ should_not_reach_here();
1463 __ bind(L);
1464 }
1465
1466 // do unlocking if necessary
1467 {
1468 Label L;
1469 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1470 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1471 // the code below should be shared with interpreter macro
1472 // assembler implementation
1473 {
1474 Label unlock;
1475 // BasicObjectLock will be first in list, since this is a
1476 // synchronized method. However, need to check that the object
1477 // has not been unlocked by an explicit monitorexit bytecode.
1478
1479 // monitor expect in c_rarg1 for slow unlock path
1480 __ lea (c_rarg1, Address(rfp, // address of first monitor
1481 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1482 wordSize - sizeof(BasicObjectLock))));
1483
1484 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1485 __ cbnz(t, unlock);
1486
1487 // Entry already unlocked, need to throw exception
1488 __ MacroAssembler::call_VM(noreg,
1489 CAST_FROM_FN_PTR(address,
1490 InterpreterRuntime::throw_illegal_monitor_state_exception));
1491 __ should_not_reach_here();
1492
1493 __ bind(unlock);
1494 __ unlock_object(c_rarg1);
1495 }
1496 __ bind(L);
1497 }
1498
1499 // jvmti support
1500 // Note: This must happen _after_ handling/throwing any exceptions since
1501 // the exception handler code notifies the runtime of method exits
1502 // too. If this happens before, method entry/exit notifications are
1503 // not properly paired (was bug - gri 11/22/99).
1504 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1505
1506 // restore potential result in r0:d0, call result handler to
1507 // restore potential result in ST0 & handle result
1508
1509 __ pop(ltos);
1510 __ pop(dtos);
1511
1512 __ blr(result_handler);
1513
1514 // remove activation
1515 __ ldr(esp, Address(rfp,
1516 frame::interpreter_frame_sender_sp_offset *
1517 wordSize)); // get sender sp
1518 // remove frame anchor
1519 __ leave();
1520
1521 // restore sender sp
1522 __ mov(sp, esp);
1523
1524 __ ret(lr);
1525
1526 if (inc_counter) {
1527 // Handle overflow of counter and compile method
1528 __ bind(invocation_counter_overflow);
1529 generate_counter_overflow(continue_after_compile);
1530 }
1531
1532 return entry_point;
1533 }
1534
1535 //
1536 // Generic interpreted method entry to (asm) interpreter
1537 //
1538 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1539 // determine code generation flags
1540 bool inc_counter = UseCompiler || CountCompiledCalls;
1541
1542 // rscratch1: sender sp
1543 address entry_point = __ pc();
1544
1545 const Address constMethod(rmethod, Method::const_offset());
1546 const Address access_flags(rmethod, Method::access_flags_offset());
1547 const Address size_of_parameters(r3,
1548 ConstMethod::size_of_parameters_offset());
1549 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1550
1551 // get parameter size (always needed)
1552 // need to load the const method first
1553 __ ldr(r3, constMethod);
1554 __ load_unsigned_short(r2, size_of_parameters);
1555
1556 // r2: size of parameters
1557
1558 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1559 __ sub(r3, r3, r2); // r3 = no. of additional locals
1560
1561 // see if we've got enough room on the stack for locals plus overhead.
1562 generate_stack_overflow_check();
1563
1564 // compute beginning of parameters (rlocals)
1565 __ add(rlocals, esp, r2, ext::uxtx, 3);
1566 __ sub(rlocals, rlocals, wordSize);
1567
1568 __ mov(rscratch1, esp);
1569
1570 // r3 - # of additional locals
1571 // allocate space for locals
1572 // explicitly initialize locals
1573 // Initializing memory allocated for locals in the same direction as
1574 // the stack grows to ensure page initialization order according
1575 // to windows-aarch64 stack page growth requirement (see
1576 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1577 {
1578 Label exit, loop;
1579 __ ands(zr, r3, r3);
1580 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1581 __ bind(loop);
1582 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1583 __ sub(r3, r3, 1); // until everything initialized
1584 __ cbnz(r3, loop);
1585 __ bind(exit);
1586 }
1587
1588 // Padding between locals and fixed part of activation frame to ensure
1589 // SP is always 16-byte aligned.
1590 __ andr(sp, rscratch1, -16);
1591
1592 // And the base dispatch table
1593 __ get_dispatch();
1594
1595 // initialize fixed part of activation frame
1596 generate_fixed_frame(false);
1597
1598 // make sure method is not native & not abstract
1599 #ifdef ASSERT
1600 __ ldrw(r0, access_flags);
1601 {
1602 Label L;
1603 __ tst(r0, JVM_ACC_NATIVE);
1604 __ br(Assembler::EQ, L);
1605 __ stop("tried to execute native method as non-native");
1606 __ bind(L);
1607 }
1608 {
1609 Label L;
1610 __ tst(r0, JVM_ACC_ABSTRACT);
1611 __ br(Assembler::EQ, L);
1612 __ stop("tried to execute abstract method in interpreter");
1613 __ bind(L);
1614 }
1615 #endif
1616
1617 // Since at this point in the method invocation the exception
1618 // handler would try to exit the monitor of synchronized methods
1619 // which hasn't been entered yet, we set the thread local variable
1620 // _do_not_unlock_if_synchronized to true. The remove_activation
1621 // will check this flag.
1622
1623 const Address do_not_unlock_if_synchronized(rthread,
1624 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1625 __ mov(rscratch2, true);
1626 __ strb(rscratch2, do_not_unlock_if_synchronized);
1627
1628 Register mdp = r3;
1629 __ profile_parameters_type(mdp, r1, r2);
1630
1631 // increment invocation count & check for overflow
1632 Label invocation_counter_overflow;
1633 if (inc_counter) {
1634 generate_counter_incr(&invocation_counter_overflow);
1635 }
1636
1637 Label continue_after_compile;
1638 __ bind(continue_after_compile);
1639
1640 bang_stack_shadow_pages(false);
1641
1642 // reset the _do_not_unlock_if_synchronized flag
1643 __ strb(zr, do_not_unlock_if_synchronized);
1644
1645 // check for synchronized methods
1646 // Must happen AFTER invocation_counter check and stack overflow check,
1647 // so method is not locked if overflows.
1648 if (synchronized) {
1649 // Allocate monitor and lock method
1650 lock_method();
1651 } else {
1652 // no synchronization necessary
1653 #ifdef ASSERT
1654 {
1655 Label L;
1656 __ ldrw(r0, access_flags);
1657 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1658 __ br(Assembler::EQ, L);
1659 __ stop("method needs synchronization");
1660 __ bind(L);
1661 }
1662 #endif
1663 }
1664
1665 // start execution
1666 #ifdef ASSERT
1667 {
1668 Label L;
1669 const Address monitor_block_top (rfp,
1670 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1671 __ ldr(rscratch1, monitor_block_top);
1672 __ cmp(esp, rscratch1);
1673 __ br(Assembler::EQ, L);
1674 __ stop("broken stack frame setup in interpreter 2");
1675 __ bind(L);
1676 }
1677 #endif
1678
1679 // jvmti support
1680 __ notify_method_entry();
1681
1682 __ dispatch_next(vtos);
1683
1684 // invocation counter overflow
1685 if (inc_counter) {
1686 // Handle overflow of counter and compile method
1687 __ bind(invocation_counter_overflow);
1688 generate_counter_overflow(continue_after_compile);
1689 }
1690
1691 return entry_point;
1692 }
1693
1694 // Method entry for java.lang.Thread.currentThread
1695 address TemplateInterpreterGenerator::generate_currentThread() {
1696 address entry_point = __ pc();
1697
1698 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1699 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1700 __ ret(lr);
1701
1702 return entry_point;
1703 }
1704
1705 //-----------------------------------------------------------------------------
1706 // Exceptions
1707
1708 void TemplateInterpreterGenerator::generate_throw_exception() {
1709 // Entry point in previous activation (i.e., if the caller was
1710 // interpreted)
1711 Interpreter::_rethrow_exception_entry = __ pc();
1712 // Restore sp to interpreter_frame_last_sp even though we are going
1713 // to empty the expression stack for the exception processing.
1714 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1715 // r0: exception
1716 // r3: return address/pc that threw exception
1717 __ restore_bcp(); // rbcp points to call/send
1718 __ restore_locals();
1719 __ restore_constant_pool_cache();
1720 __ reinit_heapbase(); // restore rheapbase as heapbase.
1721 __ get_dispatch();
1722
1723 // Entry point for exceptions thrown within interpreter code
1724 Interpreter::_throw_exception_entry = __ pc();
1725 // If we came here via a NullPointerException on the receiver of a
1726 // method, rmethod may be corrupt.
1727 __ get_method(rmethod);
1728 // expression stack is undefined here
1729 // r0: exception
1730 // rbcp: exception bcp
1731 __ verify_oop(r0);
1732 __ mov(c_rarg1, r0);
1733
1734 // expression stack must be empty before entering the VM in case of
1735 // an exception
1736 __ empty_expression_stack();
1737 // find exception handler address and preserve exception oop
1738 __ call_VM(r3,
1739 CAST_FROM_FN_PTR(address,
1740 InterpreterRuntime::exception_handler_for_exception),
1741 c_rarg1);
1742
1743 // Restore machine SP
1744 __ restore_sp_after_call();
1745
1746 // r0: exception handler entry point
1747 // r3: preserved exception oop
1748 // rbcp: bcp for exception handler
1749 __ push_ptr(r3); // push exception which is now the only value on the stack
1750 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1751
1752 // If the exception is not handled in the current frame the frame is
1753 // removed and the exception is rethrown (i.e. exception
1754 // continuation is _rethrow_exception).
1755 //
1756 // Note: At this point the bci is still the bxi for the instruction
1757 // which caused the exception and the expression stack is
1758 // empty. Thus, for any VM calls at this point, GC will find a legal
1759 // oop map (with empty expression stack).
1760
1761 //
1762 // JVMTI PopFrame support
1763 //
1764
1765 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1766 __ empty_expression_stack();
1767 // Set the popframe_processing bit in pending_popframe_condition
1768 // indicating that we are currently handling popframe, so that
1769 // call_VMs that may happen later do not trigger new popframe
1770 // handling cycles.
1771 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1772 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1773 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1774
1775 {
1776 // Check to see whether we are returning to a deoptimized frame.
1777 // (The PopFrame call ensures that the caller of the popped frame is
1778 // either interpreted or compiled and deoptimizes it if compiled.)
1779 // In this case, we can't call dispatch_next() after the frame is
1780 // popped, but instead must save the incoming arguments and restore
1781 // them after deoptimization has occurred.
1782 //
1783 // Note that we don't compare the return PC against the
1784 // deoptimization blob's unpack entry because of the presence of
1785 // adapter frames in C2.
1786 Label caller_not_deoptimized;
1787 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1788 // This is a return address, so requires authenticating for PAC.
1789 __ authenticate_return_address(c_rarg1, rscratch1);
1790 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1791 InterpreterRuntime::interpreter_contains), c_rarg1);
1792 __ cbnz(r0, caller_not_deoptimized);
1793
1794 // Compute size of arguments for saving when returning to
1795 // deoptimized caller
1796 __ get_method(r0);
1797 __ ldr(r0, Address(r0, Method::const_offset()));
1798 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1799 size_of_parameters_offset())));
1800 __ lsl(r0, r0, Interpreter::logStackElementSize);
1801 __ restore_locals(); // XXX do we need this?
1802 __ sub(rlocals, rlocals, r0);
1803 __ add(rlocals, rlocals, wordSize);
1804 // Save these arguments
1805 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1806 Deoptimization::
1807 popframe_preserve_args),
1808 rthread, r0, rlocals);
1809
1810 __ remove_activation(vtos,
1811 /* throw_monitor_exception */ false,
1812 /* install_monitor_exception */ false,
1813 /* notify_jvmdi */ false);
1814
1815 // Inform deoptimization that it is responsible for restoring
1816 // these arguments
1817 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1818 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1819
1820 // Continue in deoptimization handler
1821 __ ret(lr);
1822
1823 __ bind(caller_not_deoptimized);
1824 }
1825
1826 __ remove_activation(vtos,
1827 /* throw_monitor_exception */ false,
1828 /* install_monitor_exception */ false,
1829 /* notify_jvmdi */ false);
1830
1831 // Restore the last_sp and null it out
1832 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1833 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1834
1835 __ restore_bcp();
1836 __ restore_locals();
1837 __ restore_constant_pool_cache();
1838 __ get_method(rmethod);
1839 __ get_dispatch();
1840
1841 // The method data pointer was incremented already during
1842 // call profiling. We have to restore the mdp for the current bcp.
1843 if (ProfileInterpreter) {
1844 __ set_method_data_pointer_for_bcp();
1845 }
1846
1847 // Clear the popframe condition flag
1848 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1849 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1850
1851 #if INCLUDE_JVMTI
1852 {
1853 Label L_done;
1854
1855 __ ldrb(rscratch1, Address(rbcp, 0));
1856 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1857 __ br(Assembler::NE, L_done);
1858
1859 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1860 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1861
1862 __ ldr(c_rarg0, Address(rlocals, 0));
1863 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1864
1865 __ cbz(r0, L_done);
1866
1867 __ str(r0, Address(esp, 0));
1868 __ bind(L_done);
1869 }
1870 #endif // INCLUDE_JVMTI
1871
1872 // Restore machine SP
1873 __ restore_sp_after_call();
1874
1875 __ dispatch_next(vtos);
1876 // end of PopFrame support
1877
1878 Interpreter::_remove_activation_entry = __ pc();
1879
1880 // preserve exception over this code sequence
1881 __ pop_ptr(r0);
1882 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1883 // remove the activation (without doing throws on illegalMonitorExceptions)
1884 __ remove_activation(vtos, false, true, false);
1885 // restore exception
1886 __ get_vm_result(r0, rthread);
1887
1888 // In between activations - previous activation type unknown yet
1889 // compute continuation point - the continuation point expects the
1890 // following registers set up:
1891 //
1892 // r0: exception
1893 // lr: return address/pc that threw exception
1894 // esp: expression stack of caller
1895 // rfp: fp of caller
1896 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1897 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1898 SharedRuntime::exception_handler_for_return_address),
1899 rthread, lr);
1900 __ mov(r1, r0); // save exception handler
1901 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1902 // We might be returning to a deopt handler that expects r3 to
1903 // contain the exception pc
1904 __ mov(r3, lr);
1905 // Note that an "issuing PC" is actually the next PC after the call
1906 __ br(r1); // jump to exception
1907 // handler of caller
1908 }
1909
1910
1911 //
1912 // JVMTI ForceEarlyReturn support
1913 //
1914 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1915 address entry = __ pc();
1916
1917 __ restore_bcp();
1918 __ restore_locals();
1919 __ empty_expression_stack();
1920 __ load_earlyret_value(state);
1921
1922 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1923 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1924
1925 // Clear the earlyret state
1926 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1927 __ str(zr, cond_addr);
1928
1929 __ remove_activation(state,
1930 false, /* throw_monitor_exception */
1931 false, /* install_monitor_exception */
1932 true); /* notify_jvmdi */
1933 __ ret(lr);
1934
1935 return entry;
1936 } // end of ForceEarlyReturn support
1937
1938
1939
1940 //-----------------------------------------------------------------------------
1941 // Helper for vtos entry point generation
1942
1943 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1944 address& bep,
1945 address& cep,
1946 address& sep,
1947 address& aep,
1948 address& iep,
1949 address& lep,
1950 address& fep,
1951 address& dep,
1952 address& vep) {
1953 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1954 Label L;
1955 aep = __ pc(); __ push_ptr(); __ b(L);
1956 fep = __ pc(); __ push_f(); __ b(L);
1957 dep = __ pc(); __ push_d(); __ b(L);
1958 lep = __ pc(); __ push_l(); __ b(L);
1959 bep = cep = sep =
1960 iep = __ pc(); __ push_i();
1961 vep = __ pc();
1962 __ bind(L);
1963 generate_and_dispatch(t);
1964 }
1965
1966 //-----------------------------------------------------------------------------
1967
1968 // Non-product code
1969 #ifndef PRODUCT
1970 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1971 address entry = __ pc();
1972
1973 __ protect_return_address();
1974 __ push(lr);
1975 __ push(state);
1976 __ push(RegSet::range(r0, r15), sp);
1977 __ mov(c_rarg2, r0); // Pass itos
1978 __ call_VM(noreg,
1979 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
1980 c_rarg1, c_rarg2, c_rarg3);
1981 __ pop(RegSet::range(r0, r15), sp);
1982 __ pop(state);
1983 __ pop(lr);
1984 __ authenticate_return_address();
1985 __ ret(lr); // return from result handler
1986
1987 return entry;
1988 }
1989
1990 void TemplateInterpreterGenerator::count_bytecode() {
1991 __ mov(r10, (address) &BytecodeCounter::_counter_value);
1992 __ atomic_addw(noreg, 1, r10);
1993 }
1994
1995 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1996 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
1997 __ atomic_addw(noreg, 1, r10);
1998 }
1999
2000 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2001 // Calculate new index for counter:
2002 // _index = (_index >> log2_number_of_codes) |
2003 // (bytecode << log2_number_of_codes);
2004 Register index_addr = rscratch1;
2005 Register index = rscratch2;
2006 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2007 __ ldrw(index, index_addr);
2008 __ mov(r10,
2009 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2010 __ orrw(index, r10, index, Assembler::LSR,
2011 BytecodePairHistogram::log2_number_of_codes);
2012 __ strw(index, index_addr);
2013
2014 // Bump bucket contents:
2015 // _counters[_index] ++;
2016 Register counter_addr = rscratch1;
2017 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2018 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2019 __ atomic_addw(noreg, 1, counter_addr);
2020 }
2021
2022 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2023 // Call a little run-time stub to avoid blow-up for each bytecode.
2024 // The run-time runtime saves the right registers, depending on
2025 // the tosca in-state for the given template.
2026
2027 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2028 "entry must have been generated");
2029 __ bl(Interpreter::trace_code(t->tos_in()));
2030 __ reinit_heapbase();
2031 }
2032
2033
2034 void TemplateInterpreterGenerator::stop_interpreter_at() {
2035 Label L;
2036 __ push(rscratch1);
2037 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2038 __ ldr(rscratch1, Address(rscratch1));
2039 __ mov(rscratch2, StopInterpreterAt);
2040 __ cmpw(rscratch1, rscratch2);
2041 __ br(Assembler::NE, L);
2042 __ brk(0);
2043 __ bind(L);
2044 __ pop(rscratch1);
2045 }
2046
2047 #endif // !PRODUCT