1 /*
2 * Copyright (c) 2003, 2025, 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 "asm/macroAssembler.inline.hpp"
27 #include "classfile/javaClasses.hpp"
28 #include "compiler/disassembler.hpp"
29 #include "compiler/compiler_globals.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "interpreter/bytecodeHistogram.hpp"
32 #include "interpreter/interpreter.hpp"
33 #include "interpreter/interpreterRuntime.hpp"
34 #include "interpreter/interp_masm.hpp"
35 #include "interpreter/templateInterpreterGenerator.hpp"
36 #include "interpreter/templateTable.hpp"
37 #include "interpreter/bytecodeTracer.hpp"
38 #include "memory/resourceArea.hpp"
39 #include "oops/arrayOop.hpp"
40 #include "oops/method.hpp"
41 #include "oops/methodCounters.hpp"
42 #include "oops/methodData.hpp"
43 #include "oops/oop.inline.hpp"
44 #include "oops/resolvedIndyEntry.hpp"
45 #include "oops/resolvedMethodEntry.hpp"
46 #include "prims/jvmtiExport.hpp"
47 #include "prims/jvmtiThreadState.hpp"
48 #include "runtime/arguments.hpp"
49 #include "runtime/deoptimization.hpp"
50 #include "runtime/frame.inline.hpp"
51 #include "runtime/globals.hpp"
52 #include "runtime/jniHandles.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/stubRoutines.hpp"
55 #include "runtime/synchronizer.hpp"
56 #include "runtime/timer.hpp"
57 #include "runtime/vframeArray.hpp"
58 #include "utilities/checkedCast.hpp"
59 #include "utilities/debug.hpp"
60 #include "utilities/powerOfTwo.hpp"
61 #include <sys/types.h>
62
63 // Size of interpreter code. Increase if too small. Interpreter will
64 // fail with a guarantee ("not enough space for interpreter generation");
65 // if too small.
66 // Run with +PrintInterpreter to get the VM to print out the size.
67 // Max size with JVMTI
68 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
69
70 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
71
72 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
73 address entry = __ pc();
74
75 __ andr(esp, esp, -16);
76 __ mov(c_rarg3, esp);
77 // rmethod
78 // rlocals
79 // c_rarg3: first stack arg - wordSize
80
81 // adjust sp
82 __ sub(sp, c_rarg3, 18 * wordSize);
83 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
84 __ call_VM(noreg,
85 CAST_FROM_FN_PTR(address,
86 InterpreterRuntime::slow_signature_handler),
87 rmethod, rlocals, c_rarg3);
88
89 // r0: result handler
90
91 // Stack layout:
92 // rsp: return address <- sp
93 // 1 garbage
94 // 8 integer args (if static first is unused)
95 // 1 float/double identifiers
96 // 8 double args
97 // stack args <- esp
98 // garbage
99 // expression stack bottom
100 // bcp (null)
101 // ...
102
103 // Restore LR
104 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
105
106 // Do FP first so we can use c_rarg3 as temp
107 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
108
109 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
110 const FloatRegister r = as_FloatRegister(i);
111
112 Label d, done;
113
114 __ tbnz(c_rarg3, i, d);
115 __ ldrs(r, Address(sp, (10 + i) * wordSize));
116 __ b(done);
117 __ bind(d);
118 __ ldrd(r, Address(sp, (10 + i) * wordSize));
119 __ bind(done);
120 }
121
122 // c_rarg0 contains the result from the call of
123 // InterpreterRuntime::slow_signature_handler so we don't touch it
124 // here. It will be loaded with the JNIEnv* later.
125 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
126 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
127 Register rm = as_Register(i), rn = as_Register(i+1);
128 __ ldp(rm, rn, Address(sp, i * wordSize));
129 }
130
131 __ add(sp, sp, 18 * wordSize);
132 __ ret(lr);
133
134 return entry;
135 }
136
137
138 //
139 // Various method entries
140 //
141
142 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
143 // rmethod: Method*
144 // r19_sender_sp: sender sp
145 // esp: args
146
147 // These don't need a safepoint check because they aren't virtually
148 // callable. We won't enter these intrinsics from compiled code.
149 // If in the future we added an intrinsic which was virtually callable
150 // we'd have to worry about how to safepoint so that this code is used.
151
152 // mathematical functions inlined by compiler
153 // (interpreter must provide identical implementation
154 // in order to avoid monotonicity bugs when switching
155 // from interpreter to compiler in the middle of some
156 // computation)
157 //
158 // stack:
159 // [ arg ] <-- esp
160 // [ arg ]
161 // retaddr in lr
162
163 address entry_point = nullptr;
164 Register continuation = lr;
165 switch (kind) {
166 case Interpreter::java_lang_math_abs:
167 entry_point = __ pc();
168 __ ldrd(v0, Address(esp));
169 __ fabsd(v0, v0);
170 __ mov(sp, r19_sender_sp); // Restore caller's SP
171 break;
172 case Interpreter::java_lang_math_sqrt:
173 entry_point = __ pc();
174 __ ldrd(v0, Address(esp));
175 __ fsqrtd(v0, v0);
176 __ mov(sp, r19_sender_sp);
177 break;
178 case Interpreter::java_lang_math_sin :
179 case Interpreter::java_lang_math_cos :
180 case Interpreter::java_lang_math_tan :
181 case Interpreter::java_lang_math_log :
182 case Interpreter::java_lang_math_log10 :
183 case Interpreter::java_lang_math_exp :
184 entry_point = __ pc();
185 __ ldrd(v0, Address(esp));
186 __ mov(sp, r19_sender_sp);
187 __ mov(r23, lr);
188 continuation = r23; // The first free callee-saved register
189 generate_transcendental_entry(kind, 1);
190 break;
191 case Interpreter::java_lang_math_pow :
192 entry_point = __ pc();
193 __ mov(r23, lr);
194 continuation = r23;
195 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
196 __ ldrd(v1, Address(esp));
197 __ mov(sp, r19_sender_sp);
198 generate_transcendental_entry(kind, 2);
199 break;
200 case Interpreter::java_lang_math_fmaD :
201 if (UseFMA) {
202 entry_point = __ pc();
203 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
204 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
205 __ ldrd(v2, Address(esp));
206 __ fmaddd(v0, v0, v1, v2);
207 __ mov(sp, r19_sender_sp); // Restore caller's SP
208 }
209 break;
210 case Interpreter::java_lang_math_fmaF :
211 if (UseFMA) {
212 entry_point = __ pc();
213 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
214 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
215 __ ldrs(v2, Address(esp));
216 __ fmadds(v0, v0, v1, v2);
217 __ mov(sp, r19_sender_sp); // Restore caller's SP
218 }
219 break;
220 default:
221 ;
222 }
223 if (entry_point) {
224 __ br(continuation);
225 }
226
227 return entry_point;
228 }
229
230 // double trigonometrics and transcendentals
231 // static jdouble dsin(jdouble x);
232 // static jdouble dcos(jdouble x);
233 // static jdouble dtan(jdouble x);
234 // static jdouble dlog(jdouble x);
235 // static jdouble dlog10(jdouble x);
236 // static jdouble dexp(jdouble x);
237 // static jdouble dpow(jdouble x, jdouble y);
238
239 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
240 address fn;
241 switch (kind) {
242 case Interpreter::java_lang_math_sin :
243 if (StubRoutines::dsin() == nullptr) {
244 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
245 } else {
246 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
247 }
248 break;
249 case Interpreter::java_lang_math_cos :
250 if (StubRoutines::dcos() == nullptr) {
251 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
252 } else {
253 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
254 }
255 break;
256 case Interpreter::java_lang_math_tan :
257 if (StubRoutines::dtan() == nullptr) {
258 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
259 } else {
260 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
261 }
262 break;
263 case Interpreter::java_lang_math_log :
264 if (StubRoutines::dlog() == nullptr) {
265 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
266 } else {
267 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
268 }
269 break;
270 case Interpreter::java_lang_math_log10 :
271 if (StubRoutines::dlog10() == nullptr) {
272 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
273 } else {
274 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
275 }
276 break;
277 case Interpreter::java_lang_math_exp :
278 if (StubRoutines::dexp() == nullptr) {
279 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
280 } else {
281 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
282 }
283 break;
284 case Interpreter::java_lang_math_pow :
285 if (StubRoutines::dpow() == nullptr) {
286 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
287 } else {
288 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
289 }
290 break;
291 default:
292 ShouldNotReachHere();
293 fn = nullptr; // unreachable
294 }
295 __ mov(rscratch1, fn);
296 __ blr(rscratch1);
297 }
298
299 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
300 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
301 // r19_sender_sp: sender sp
302 // stack:
303 // [ arg ] <-- esp
304 // [ arg ]
305 // retaddr in lr
306 // result in v0
307
308 address entry_point = __ pc();
309 __ ldrw(c_rarg0, Address(esp));
310 __ flt16_to_flt(v0, c_rarg0, v1);
311 __ mov(sp, r19_sender_sp); // Restore caller's SP
312 __ br(lr);
313 return entry_point;
314 }
315
316 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
317 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
318 // r19_sender_sp: sender sp
319 // stack:
320 // [ arg ] <-- esp
321 // [ arg ]
322 // retaddr in lr
323 // result in c_rarg0
324
325 address entry_point = __ pc();
326 __ ldrs(v0, Address(esp));
327 __ flt_to_flt16(c_rarg0, v0, v1);
328 __ mov(sp, r19_sender_sp); // Restore caller's SP
329 __ br(lr);
330 return entry_point;
331 }
332
333 // Abstract method entry
334 // Attempt to execute abstract method. Throw exception
335 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
336 // rmethod: Method*
337 // r19_sender_sp: sender SP
338
339 address entry_point = __ pc();
340
341 // abstract method entry
342
343 // pop return address, reset last_sp to null
344 __ empty_expression_stack();
345 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
346 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
347
348 // throw exception
349 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
350 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
351 rmethod);
352 // the call_VM checks for exception, so we should never return here.
353 __ should_not_reach_here();
354
355 return entry_point;
356 }
357
358 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
359 address entry = __ pc();
360
361 #ifdef ASSERT
362 {
363 Label L;
364 __ ldr(rscratch1, Address(rfp,
365 frame::interpreter_frame_monitor_block_top_offset *
366 wordSize));
367 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
368 __ mov(rscratch2, sp);
369 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
370 // grows negative)
371 __ br(Assembler::HS, L); // check if frame is complete
372 __ stop ("interpreter frame not set up");
373 __ bind(L);
374 }
375 #endif // ASSERT
376 // Restore bcp under the assumption that the current frame is still
377 // interpreted
378 __ restore_bcp();
379
380 // expression stack must be empty before entering the VM if an
381 // exception happened
382 __ empty_expression_stack();
383 // throw exception
384 __ call_VM(noreg,
385 CAST_FROM_FN_PTR(address,
386 InterpreterRuntime::throw_StackOverflowError));
387 return entry;
388 }
389
390 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
391 address entry = __ pc();
392 // expression stack must be empty before entering the VM if an
393 // exception happened
394 __ empty_expression_stack();
395 // setup parameters
396
397 // ??? convention: expect aberrant index in register r1
398 __ movw(c_rarg2, r1);
399 // ??? convention: expect array in register r3
400 __ mov(c_rarg1, r3);
401 __ call_VM(noreg,
402 CAST_FROM_FN_PTR(address,
403 InterpreterRuntime::
404 throw_ArrayIndexOutOfBoundsException),
405 c_rarg1, c_rarg2);
406 return entry;
407 }
408
409 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
410 address entry = __ pc();
411
412 // object is at TOS
413 __ pop(c_rarg1);
414
415 // expression stack must be empty before entering the VM if an
416 // exception happened
417 __ empty_expression_stack();
418
419 __ call_VM(noreg,
420 CAST_FROM_FN_PTR(address,
421 InterpreterRuntime::
422 throw_ClassCastException),
423 c_rarg1);
424 return entry;
425 }
426
427 address TemplateInterpreterGenerator::generate_exception_handler_common(
428 const char* name, const char* message, bool pass_oop) {
429 assert(!pass_oop || message == nullptr, "either oop or message but not both");
430 address entry = __ pc();
431 if (pass_oop) {
432 // object is at TOS
433 __ pop(c_rarg2);
434 }
435 // expression stack must be empty before entering the VM if an
436 // exception happened
437 __ empty_expression_stack();
438 // setup parameters
439 __ lea(c_rarg1, Address((address)name));
440 if (pass_oop) {
441 __ call_VM(r0, CAST_FROM_FN_PTR(address,
442 InterpreterRuntime::
443 create_klass_exception),
444 c_rarg1, c_rarg2);
445 } else {
446 // kind of lame ExternalAddress can't take null because
447 // external_word_Relocation will assert.
448 if (message != nullptr) {
449 __ lea(c_rarg2, Address((address)message));
450 } else {
451 __ mov(c_rarg2, NULL_WORD);
452 }
453 __ call_VM(r0,
454 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
455 c_rarg1, c_rarg2);
456 }
457 // throw exception
458 __ b(address(Interpreter::throw_exception_entry()));
459 return entry;
460 }
461
462 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
463 address entry = __ pc();
464
465 // Restore stack bottom in case i2c adjusted stack
466 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
467 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
468 // and null it as marker that esp is now tos until next java call
469 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
470 __ restore_bcp();
471 __ restore_locals();
472 __ restore_constant_pool_cache();
473 __ get_method(rmethod);
474
475 if (state == atos) {
476 Register obj = r0;
477 Register mdp = r1;
478 Register tmp = r2;
479 __ profile_return_type(mdp, obj, tmp);
480 }
481
482 const Register cache = r1;
483 const Register index = r2;
484
485 if (index_size == sizeof(u4)) {
486 __ load_resolved_indy_entry(cache, index);
487 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
488 __ add(esp, esp, cache, Assembler::LSL, 3);
489 } else {
490 // Pop N words from the stack
491 assert(index_size == sizeof(u2), "Can only be u2");
492 __ load_method_entry(cache, index);
493 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
494 __ add(esp, esp, cache, Assembler::LSL, 3);
495 }
496
497 // Restore machine SP
498 __ restore_sp_after_call();
499
500 __ check_and_handle_popframe(rthread);
501 __ check_and_handle_earlyret(rthread);
502
503 __ get_dispatch();
504 __ dispatch_next(state, step);
505
506 return entry;
507 }
508
509 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
510 int step,
511 address continuation) {
512 address entry = __ pc();
513 __ restore_bcp();
514 __ restore_locals();
515 __ restore_constant_pool_cache();
516 __ get_method(rmethod);
517 __ get_dispatch();
518
519 __ restore_sp_after_call(); // Restore SP to extended SP
520
521 // Restore expression stack pointer
522 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
523 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
524 // null last_sp until next java call
525 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
526
527 #if INCLUDE_JVMCI
528 // Check if we need to take lock at entry of synchronized method. This can
529 // only occur on method entry so emit it only for vtos with step 0.
530 if (EnableJVMCI && state == vtos && step == 0) {
531 Label L;
532 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
533 __ cbz(rscratch1, L);
534 // Clear flag.
535 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
536 // Take lock.
537 lock_method();
538 __ bind(L);
539 } else {
540 #ifdef ASSERT
541 if (EnableJVMCI) {
542 Label L;
543 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
544 __ cbz(rscratch1, L);
545 __ stop("unexpected pending monitor in deopt entry");
546 __ bind(L);
547 }
548 #endif
549 }
550 #endif
551 // handle exceptions
552 {
553 Label L;
554 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
555 __ cbz(rscratch1, L);
556 __ call_VM(noreg,
557 CAST_FROM_FN_PTR(address,
558 InterpreterRuntime::throw_pending_exception));
559 __ should_not_reach_here();
560 __ bind(L);
561 }
562
563 if (continuation == nullptr) {
564 __ dispatch_next(state, step);
565 } else {
566 __ jump_to_entry(continuation);
567 }
568 return entry;
569 }
570
571 address TemplateInterpreterGenerator::generate_result_handler_for(
572 BasicType type) {
573 address entry = __ pc();
574 switch (type) {
575 case T_BOOLEAN: __ c2bool(r0); break;
576 case T_CHAR : __ uxth(r0, r0); break;
577 case T_BYTE : __ sxtb(r0, r0); break;
578 case T_SHORT : __ sxth(r0, r0); break;
579 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
580 case T_LONG : /* nothing to do */ break;
581 case T_VOID : /* nothing to do */ break;
582 case T_FLOAT : /* nothing to do */ break;
583 case T_DOUBLE : /* nothing to do */ break;
584 case T_OBJECT :
585 // retrieve result from frame
586 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
587 // and verify it
588 __ verify_oop(r0);
589 break;
590 default : ShouldNotReachHere();
591 }
592 __ ret(lr); // return from result handler
593 return entry;
594 }
595
596 address TemplateInterpreterGenerator::generate_safept_entry_for(
597 TosState state,
598 address runtime_entry) {
599 address entry = __ pc();
600 __ push(state);
601 __ push_cont_fastpath(rthread);
602 __ call_VM(noreg, runtime_entry);
603 __ pop_cont_fastpath(rthread);
604 __ membar(Assembler::AnyAny);
605 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
606 return entry;
607 }
608
609 address TemplateInterpreterGenerator::generate_cont_resume_interpreter_adapter() {
610 if (!Continuations::enabled()) return nullptr;
611 address start = __ pc();
612
613 __ restore_bcp();
614 __ restore_locals();
615
616 // Restore constant pool cache
617 __ ldr(rcpool, Address(rfp, frame::interpreter_frame_cache_offset * wordSize));
618
619 // Restore Java expression stack pointer
620 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
621 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
622 // and null it as marker that esp is now tos until next java call
623 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
624
625 // Restore machine SP
626 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
627 __ lea(sp, Address(rfp, rscratch1, Address::lsl(LogBytesPerWord)));
628
629 // Restore method
630 __ ldr(rmethod, Address(rfp, frame::interpreter_frame_method_offset * wordSize));
631
632 // Restore dispatch
633 uint64_t offset;
634 __ adrp(rdispatch, ExternalAddress((address)Interpreter::dispatch_table()), offset);
635 __ add(rdispatch, rdispatch, offset);
636
637 __ ret(lr);
638
639 return start;
640 }
641
642
643 // Helpers for commoning out cases in the various type of method entries.
644 //
645
646
647 // increment invocation count & check for overflow
648 //
649 // Note: checking for negative value instead of overflow
650 // so we have a 'sticky' overflow test
651 //
652 // rmethod: method
653 //
654 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
655 Label done;
656 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
657 int increment = InvocationCounter::count_increment;
658 Label no_mdo;
659 if (ProfileInterpreter) {
660 // Are we profiling?
661 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
662 __ cbz(r0, no_mdo);
663 // Increment counter in the MDO
664 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
665 in_bytes(InvocationCounter::counter_offset()));
666 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
667 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
668 __ b(done);
669 }
670 __ bind(no_mdo);
671 // Increment counter in MethodCounters
672 const Address invocation_counter(rscratch2,
673 MethodCounters::invocation_counter_offset() +
674 InvocationCounter::counter_offset());
675 __ get_method_counters(rmethod, rscratch2, done);
676 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
677 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
678 __ bind(done);
679 }
680
681 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
682
683 // Asm interpreter on entry
684 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
685 // Everything as it was on entry
686
687 // InterpreterRuntime::frequency_counter_overflow takes two
688 // arguments, the first (thread) is passed by call_VM, the second
689 // indicates if the counter overflow occurs at a backwards branch
690 // (null bcp). We pass zero for it. The call returns the address
691 // of the verified entry point for the method or null if the
692 // compilation did not complete (either went background or bailed
693 // out).
694 __ mov(c_rarg1, 0);
695 __ call_VM(noreg,
696 CAST_FROM_FN_PTR(address,
697 InterpreterRuntime::frequency_counter_overflow),
698 c_rarg1);
699
700 __ b(do_continue);
701 }
702
703 // See if we've got enough room on the stack for locals plus overhead
704 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
705 // without going through the signal handler, i.e., reserved and yellow zones
706 // will not be made usable. The shadow zone must suffice to handle the
707 // overflow.
708 // The expression stack grows down incrementally, so the normal guard
709 // page mechanism will work for that.
710 //
711 // NOTE: Since the additional locals are also always pushed (wasn't
712 // obvious in generate_method_entry) so the guard should work for them
713 // too.
714 //
715 // Args:
716 // r3: number of additional locals this frame needs (what we must check)
717 // rmethod: Method*
718 //
719 // Kills:
720 // r0
721 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
722
723 // monitor entry size: see picture of stack set
724 // (generate_method_entry) and frame_amd64.hpp
725 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
726
727 // total overhead size: entry_size + (saved rbp through expr stack
728 // bottom). be sure to change this if you add/subtract anything
729 // to/from the overhead area
730 const int overhead_size =
731 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
732
733 const size_t page_size = os::vm_page_size();
734
735 Label after_frame_check;
736
737 // see if the frame is greater than one page in size. If so,
738 // then we need to verify there is enough stack space remaining
739 // for the additional locals.
740 //
741 // Note that we use SUBS rather than CMP here because the immediate
742 // field of this instruction may overflow. SUBS can cope with this
743 // because it is a macro that will expand to some number of MOV
744 // instructions and a register operation.
745 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
746 __ br(Assembler::LS, after_frame_check);
747
748 // compute rsp as if this were going to be the last frame on
749 // the stack before the red zone
750
751 // locals + overhead, in bytes
752 __ mov(r0, overhead_size);
753 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
754
755 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
756 __ ldr(rscratch1, stack_limit);
757
758 #ifdef ASSERT
759 Label limit_okay;
760 // Verify that thread stack limit is non-zero.
761 __ cbnz(rscratch1, limit_okay);
762 __ stop("stack overflow limit is zero");
763 __ bind(limit_okay);
764 #endif
765
766 // Add stack limit to locals.
767 __ add(r0, r0, rscratch1);
768
769 // Check against the current stack bottom.
770 __ cmp(sp, r0);
771 __ br(Assembler::HI, after_frame_check);
772
773 // Remove the incoming args, peeling the machine SP back to where it
774 // was in the caller. This is not strictly necessary, but unless we
775 // do so the stack frame may have a garbage FP; this ensures a
776 // correct call stack that we can always unwind. The ANDR should be
777 // unnecessary because the sender SP in r19 is always aligned, but
778 // it doesn't hurt.
779 __ andr(sp, r19_sender_sp, -16);
780
781 // Note: the restored frame is not necessarily interpreted.
782 // Use the shared runtime version of the StackOverflowError.
783 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
784 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
785
786 // all done with frame size check
787 __ bind(after_frame_check);
788 }
789
790 // Allocate monitor and lock method (asm interpreter)
791 //
792 // Args:
793 // rmethod: Method*
794 // rlocals: locals
795 //
796 // Kills:
797 // r0
798 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
799 // rscratch1, rscratch2 (scratch regs)
800 void TemplateInterpreterGenerator::lock_method() {
801 // synchronize method
802 const Address access_flags(rmethod, Method::access_flags_offset());
803 const Address monitor_block_top(
804 rfp,
805 frame::interpreter_frame_monitor_block_top_offset * wordSize);
806 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
807
808 #ifdef ASSERT
809 {
810 Label L;
811 __ ldrh(r0, access_flags);
812 __ tst(r0, JVM_ACC_SYNCHRONIZED);
813 __ br(Assembler::NE, L);
814 __ stop("method doesn't need synchronization");
815 __ bind(L);
816 }
817 #endif // ASSERT
818
819 // get synchronization object
820 {
821 Label done;
822 __ ldrh(r0, access_flags);
823 __ tst(r0, JVM_ACC_STATIC);
824 // get receiver (assume this is frequent case)
825 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
826 __ br(Assembler::EQ, done);
827 __ load_mirror(r0, rmethod, r5, rscratch2);
828
829 #ifdef ASSERT
830 {
831 Label L;
832 __ cbnz(r0, L);
833 __ stop("synchronization object is null");
834 __ bind(L);
835 }
836 #endif // ASSERT
837
838 __ bind(done);
839 }
840
841 // add space for monitor & lock
842 __ check_extended_sp();
843 __ sub(sp, sp, entry_size); // add space for a monitor entry
844 __ sub(esp, esp, entry_size);
845 __ sub(rscratch1, sp, rfp);
846 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
847 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
848 __ sub(rscratch1, esp, rfp);
849 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
850 __ str(rscratch1, monitor_block_top); // set new monitor block top
851
852 // store object
853 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
854 __ mov(c_rarg1, esp); // object address
855 __ lock_object(c_rarg1);
856 }
857
858 // Generate a fixed interpreter frame. This is identical setup for
859 // interpreted methods and for native methods hence the shared code.
860 //
861 // Args:
862 // lr: return address
863 // rmethod: Method*
864 // rlocals: pointer to locals
865 // rcpool: cp cache
866 // stack_pointer: previous sp
867 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
868 // initialize fixed part of activation frame
869 if (native_call) {
870 __ sub(esp, sp, 14 * wordSize);
871 __ mov(rbcp, zr);
872 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
873 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
874 // add 2 zero-initialized slots for native calls
875 __ stp(zr, zr, Address(sp, 12 * wordSize));
876 } else {
877 __ sub(esp, sp, 12 * wordSize);
878 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
879 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
880 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
881 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
882 }
883
884 if (ProfileInterpreter) {
885 Label method_data_continue;
886 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
887 __ cbz(rscratch1, method_data_continue);
888 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
889 __ bind(method_data_continue);
890 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
891 } else {
892 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
893 }
894
895 __ protect_return_address();
896 __ stp(rfp, lr, Address(sp, 10 * wordSize));
897 __ lea(rfp, Address(sp, 10 * wordSize));
898
899 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
900 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
901 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
902 __ sub(rscratch1, rlocals, rfp);
903 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
904 // Store relativized rlocals, see frame::interpreter_frame_locals().
905 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
906
907 // set sender sp
908 // leave last_sp as null
909 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
910
911 // Get mirror
912 __ load_mirror(r10, rmethod, r5, rscratch2);
913 if (! native_call) {
914 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
915 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
916 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
917 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
918 __ andr(rscratch1, rscratch1, -16);
919 __ sub(rscratch2, rscratch1, rfp);
920 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
921 // Store extended SP and mirror
922 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
923 // Move SP out of the way
924 __ mov(sp, rscratch1);
925 } else {
926 // Make sure there is room for the exception oop pushed in case method throws
927 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
928 __ sub(rscratch1, sp, 2 * wordSize);
929 __ sub(rscratch2, rscratch1, rfp);
930 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
931 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
932 __ mov(sp, rscratch1);
933 }
934 }
935
936 // End of helpers
937
938 // Various method entries
939 //------------------------------------------------------------------------------------------------------------------------
940 //
941 //
942
943 // Method entry for java.lang.ref.Reference.get.
944 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
945 // Code: _aload_0, _getfield, _areturn
946 // parameter size = 1
947 //
948 // The code that gets generated by this routine is split into 2 parts:
949 // 1. The "intrinsified" code for G1 (or any SATB based GC),
950 // 2. The slow path - which is an expansion of the regular method entry.
951 //
952 // Notes:-
953 // * In the G1 code we do not check whether we need to block for
954 // a safepoint. If G1 is enabled then we must execute the specialized
955 // code for Reference.get (except when the Reference object is null)
956 // so that we can log the value in the referent field with an SATB
957 // update buffer.
958 // If the code for the getfield template is modified so that the
959 // G1 pre-barrier code is executed when the current method is
960 // Reference.get() then going through the normal method entry
961 // will be fine.
962 // * The G1 code can, however, check the receiver object (the instance
963 // of java.lang.Reference) and jump to the slow path if null. If the
964 // Reference object is null then we obviously cannot fetch the referent
965 // and so we don't need to call the G1 pre-barrier. Thus we can use the
966 // regular method entry code to generate the NPE.
967 //
968 // This code is based on generate_accessor_entry.
969 //
970 // rmethod: Method*
971 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
972
973 // LR is live. It must be saved around calls.
974
975 address entry = __ pc();
976
977 const int referent_offset = java_lang_ref_Reference::referent_offset();
978
979 Label slow_path;
980 const Register local_0 = c_rarg0;
981 // Check if local 0 != null
982 // If the receiver is null then it is OK to jump to the slow path.
983 __ ldr(local_0, Address(esp, 0));
984 __ cbz(local_0, slow_path);
985
986 // Load the value of the referent field.
987 const Address field_address(local_0, referent_offset);
988 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
989 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
990
991 // areturn
992 __ andr(sp, r19_sender_sp, -16); // done with stack
993 __ ret(lr);
994
995 // generate a vanilla interpreter entry as the slow path
996 __ bind(slow_path);
997 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
998 return entry;
999
1000 }
1001
1002 /**
1003 * Method entry for static native methods:
1004 * int java.util.zip.CRC32.update(int crc, int b)
1005 */
1006 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
1007 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1008 address entry = __ pc();
1009
1010 // rmethod: Method*
1011 // r19_sender_sp: senderSP must preserved for slow path
1012 // esp: args
1013
1014 Label slow_path;
1015 // If we need a safepoint check, generate full interpreter entry.
1016 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1017
1018 // We don't generate local frame and don't align stack because
1019 // we call stub code and there is no safepoint on this path.
1020
1021 // Load parameters
1022 const Register crc = c_rarg0; // crc
1023 const Register val = c_rarg1; // source java byte value
1024 const Register tbl = c_rarg2; // scratch
1025
1026 // Arguments are reversed on java expression stack
1027 __ ldrw(val, Address(esp, 0)); // byte value
1028 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1029
1030 uint64_t offset;
1031 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1032 __ add(tbl, tbl, offset);
1033
1034 __ mvnw(crc, crc); // ~crc
1035 __ update_byte_crc32(crc, val, tbl);
1036 __ mvnw(crc, crc); // ~crc
1037
1038 // result in c_rarg0
1039
1040 __ andr(sp, r19_sender_sp, -16);
1041 __ ret(lr);
1042
1043 // generate a vanilla native entry as the slow path
1044 __ bind(slow_path);
1045 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1046 return entry;
1047 }
1048
1049 /**
1050 * Method entry for static native methods:
1051 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1052 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1053 */
1054 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1055 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1056 address entry = __ pc();
1057
1058 // rmethod,: Method*
1059 // r19_sender_sp: senderSP must preserved for slow path
1060
1061 Label slow_path;
1062 // If we need a safepoint check, generate full interpreter entry.
1063 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1064
1065 // We don't generate local frame and don't align stack because
1066 // we call stub code and there is no safepoint on this path.
1067
1068 // Load parameters
1069 const Register crc = c_rarg0; // crc
1070 const Register buf = c_rarg1; // source java byte array address
1071 const Register len = c_rarg2; // length
1072 const Register off = len; // offset (never overlaps with 'len')
1073
1074 // Arguments are reversed on java expression stack
1075 // Calculate address of start element
1076 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1077 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1078 __ ldrw(off, Address(esp, wordSize)); // offset
1079 __ add(buf, buf, off); // + offset
1080 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1081 } else {
1082 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1083 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1084 __ ldrw(off, Address(esp, wordSize)); // offset
1085 __ add(buf, buf, off); // + offset
1086 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1087 }
1088 // Can now load 'len' since we're finished with 'off'
1089 __ ldrw(len, Address(esp, 0x0)); // Length
1090
1091 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1092
1093 // We are frameless so we can just jump to the stub.
1094 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1095
1096 // generate a vanilla native entry as the slow path
1097 __ bind(slow_path);
1098 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1099 return entry;
1100 }
1101
1102 /**
1103 * Method entry for intrinsic-candidate (non-native) methods:
1104 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1105 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1106 * Unlike CRC32, CRC32C does not have any methods marked as native
1107 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1108 */
1109 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1110 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1111 address entry = __ pc();
1112
1113 // Prepare jump to stub using parameters from the stack
1114 const Register crc = c_rarg0; // initial crc
1115 const Register buf = c_rarg1; // source java byte array address
1116 const Register len = c_rarg2; // len argument to the kernel
1117
1118 const Register end = len; // index of last element to process
1119 const Register off = crc; // offset
1120
1121 __ ldrw(end, Address(esp)); // int end
1122 __ ldrw(off, Address(esp, wordSize)); // int offset
1123 __ sub(len, end, off);
1124 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1125 __ add(buf, buf, off); // + offset
1126 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1127 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1128 } else {
1129 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1130 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1131 }
1132
1133 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1134
1135 // Jump to the stub.
1136 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1137
1138 return entry;
1139 }
1140
1141 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1142 // See more discussion in stackOverflow.hpp.
1143
1144 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1145 const int page_size = (int)os::vm_page_size();
1146 const int n_shadow_pages = shadow_zone_size / page_size;
1147
1148 #ifdef ASSERT
1149 Label L_good_limit;
1150 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1151 __ cbnz(rscratch1, L_good_limit);
1152 __ stop("shadow zone safe limit is not initialized");
1153 __ bind(L_good_limit);
1154
1155 Label L_good_watermark;
1156 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1157 __ cbnz(rscratch1, L_good_watermark);
1158 __ stop("shadow zone growth watermark is not initialized");
1159 __ bind(L_good_watermark);
1160 #endif
1161
1162 Label L_done;
1163
1164 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1165 __ cmp(sp, rscratch1);
1166 __ br(Assembler::HI, L_done);
1167
1168 for (int p = 1; p <= n_shadow_pages; p++) {
1169 __ sub(rscratch2, sp, p*page_size);
1170 __ str(zr, Address(rscratch2));
1171 }
1172
1173 // Record the new watermark, but only if the update is above the safe limit.
1174 // Otherwise, the next time around the check above would pass the safe limit.
1175 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1176 __ cmp(sp, rscratch1);
1177 __ br(Assembler::LS, L_done);
1178 __ mov(rscratch1, sp);
1179 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1180
1181 __ bind(L_done);
1182 }
1183
1184 // Interpreter stub for calling a native method. (asm interpreter)
1185 // This sets up a somewhat different looking stack for calling the
1186 // native method than the typical interpreter frame setup.
1187 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized, bool runtime_upcalls) {
1188 // determine code generation flags
1189 bool inc_counter = (UseCompiler || CountCompiledCalls) && !PreloadOnly;
1190
1191 // r1: Method*
1192 // rscratch1: sender sp
1193
1194 address entry_point = __ pc();
1195
1196 const Address constMethod (rmethod, Method::const_offset());
1197 const Address access_flags (rmethod, Method::access_flags_offset());
1198 const Address size_of_parameters(r2, ConstMethod::
1199 size_of_parameters_offset());
1200
1201 // get parameter size (always needed)
1202 __ ldr(r2, constMethod);
1203 __ load_unsigned_short(r2, size_of_parameters);
1204
1205 // Native calls don't need the stack size check since they have no
1206 // expression stack and the arguments are already on the stack and
1207 // we only add a handful of words to the stack.
1208
1209 // rmethod: Method*
1210 // r2: size of parameters
1211 // rscratch1: sender sp
1212
1213 // for natives the size of locals is zero
1214
1215 // compute beginning of parameters (rlocals)
1216 __ add(rlocals, esp, r2, ext::uxtx, 3);
1217 __ add(rlocals, rlocals, -wordSize);
1218
1219 // Pull SP back to minimum size: this avoids holes in the stack
1220 __ andr(sp, esp, -16);
1221
1222 // initialize fixed part of activation frame
1223 generate_fixed_frame(true);
1224
1225 // make sure method is native & not abstract
1226 #ifdef ASSERT
1227 __ ldrh(r0, access_flags);
1228 {
1229 Label L;
1230 __ tst(r0, JVM_ACC_NATIVE);
1231 __ br(Assembler::NE, L);
1232 __ stop("tried to execute non-native method as native");
1233 __ bind(L);
1234 }
1235 {
1236 Label L;
1237 __ tst(r0, JVM_ACC_ABSTRACT);
1238 __ br(Assembler::EQ, L);
1239 __ stop("tried to execute abstract method in interpreter");
1240 __ bind(L);
1241 }
1242 #endif
1243
1244 // Since at this point in the method invocation the exception
1245 // handler would try to exit the monitor of synchronized methods
1246 // which hasn't been entered yet, we set the thread local variable
1247 // _do_not_unlock_if_synchronized to true. The remove_activation
1248 // will check this flag.
1249
1250 const Address do_not_unlock_if_synchronized(rthread,
1251 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1252 __ mov(rscratch2, true);
1253 __ strb(rscratch2, do_not_unlock_if_synchronized);
1254
1255 // increment invocation count & check for overflow
1256 Label invocation_counter_overflow;
1257 if (inc_counter) {
1258 generate_counter_incr(&invocation_counter_overflow);
1259 }
1260
1261 Label continue_after_compile;
1262 __ bind(continue_after_compile);
1263
1264 bang_stack_shadow_pages(true);
1265
1266 // reset the _do_not_unlock_if_synchronized flag
1267 __ strb(zr, do_not_unlock_if_synchronized);
1268
1269 // check for synchronized methods
1270 // Must happen AFTER invocation_counter check and stack overflow check,
1271 // so method is not locked if overflows.
1272 if (synchronized) {
1273 lock_method();
1274 } else {
1275 // no synchronization necessary
1276 #ifdef ASSERT
1277 {
1278 Label L;
1279 __ ldrh(r0, access_flags);
1280 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1281 __ br(Assembler::EQ, L);
1282 __ stop("method needs synchronization");
1283 __ bind(L);
1284 }
1285 #endif
1286 }
1287
1288 // start execution
1289 #ifdef ASSERT
1290 {
1291 Label L;
1292 const Address monitor_block_top(rfp,
1293 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1294 __ ldr(rscratch1, monitor_block_top);
1295 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1296 __ cmp(esp, rscratch1);
1297 __ br(Assembler::EQ, L);
1298 __ stop("broken stack frame setup in interpreter 1");
1299 __ bind(L);
1300 }
1301 #endif
1302
1303 // jvmti support
1304 __ notify_method_entry();
1305
1306 if (runtime_upcalls) {
1307 __ generate_runtime_upcalls_on_method_entry();
1308 }
1309
1310 // work registers
1311 const Register t = r17;
1312 const Register result_handler = r19;
1313
1314 // allocate space for parameters
1315 __ ldr(t, Address(rmethod, Method::const_offset()));
1316 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1317
1318 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1319 __ andr(sp, rscratch1, -16);
1320 __ mov(esp, rscratch1);
1321
1322 // get signature handler
1323 {
1324 Label L;
1325 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1326 __ cbnz(t, L);
1327 __ call_VM(noreg,
1328 CAST_FROM_FN_PTR(address,
1329 InterpreterRuntime::prepare_native_call),
1330 rmethod);
1331 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1332 __ bind(L);
1333 }
1334
1335 // call signature handler
1336 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1337 "adjust this code");
1338 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1339 "adjust this code");
1340 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1341 "adjust this code");
1342
1343 // The generated handlers do not touch rmethod (the method).
1344 // However, large signatures cannot be cached and are generated
1345 // each time here. The slow-path generator can do a GC on return,
1346 // so we must reload it after the call.
1347 __ blr(t);
1348 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1349
1350
1351 // result handler is in r0
1352 // set result handler
1353 __ mov(result_handler, r0);
1354 // Save it in the frame in case of preemption; we cannot rely on callee saved registers.
1355 __ str(r0, Address(rfp, frame::interpreter_frame_result_handler_offset * wordSize));
1356
1357 // pass mirror handle if static call
1358 {
1359 Label L;
1360 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1361 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1362 // get mirror
1363 __ load_mirror(t, rmethod, r10, rscratch2);
1364 // copy mirror into activation frame
1365 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1366 // pass handle to mirror
1367 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1368 __ bind(L);
1369 }
1370
1371 // get native function entry point in r10
1372 {
1373 Label L;
1374 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1375 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1376 __ lea(rscratch2, unsatisfied);
1377 __ ldr(rscratch2, rscratch2);
1378 __ cmp(r10, rscratch2);
1379 __ br(Assembler::NE, L);
1380 __ call_VM(noreg,
1381 CAST_FROM_FN_PTR(address,
1382 InterpreterRuntime::prepare_native_call),
1383 rmethod);
1384 __ get_method(rmethod);
1385 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1386 __ bind(L);
1387 }
1388
1389 // pass JNIEnv
1390 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1391
1392 // It is enough that the pc() points into the right code
1393 // segment. It does not have to be the correct return pc.
1394 // For convenience we use the pc we want to resume to in
1395 // case of preemption on Object.wait.
1396 Label native_return;
1397 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1398
1399 // change thread state
1400 #ifdef ASSERT
1401 {
1402 Label L;
1403 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1404 __ cmp(t, (u1)_thread_in_Java);
1405 __ br(Assembler::EQ, L);
1406 __ stop("Wrong thread state in native stub");
1407 __ bind(L);
1408 }
1409 #endif
1410
1411 // Change state to native
1412 __ mov(rscratch1, _thread_in_native);
1413 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1414 __ stlrw(rscratch1, rscratch2);
1415
1416 __ push_cont_fastpath();
1417
1418 // Call the native method.
1419 __ blr(r10);
1420
1421 __ pop_cont_fastpath();
1422
1423 __ get_method(rmethod);
1424 // result potentially in r0 or v0
1425
1426 // Restore cpu control state after JNI call
1427 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1428
1429 // make room for the pushes we're about to do
1430 __ sub(rscratch1, esp, 4 * wordSize);
1431 __ andr(sp, rscratch1, -16);
1432
1433 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1434 // in order to extract the result of a method call. If the order of these
1435 // pushes change or anything else is added to the stack then the code in
1436 // interpreter_frame_result must also change.
1437 __ push(dtos);
1438 __ push(ltos);
1439
1440 __ verify_sve_vector_length();
1441
1442 // change thread state
1443 __ mov(rscratch1, _thread_in_native_trans);
1444 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1445 __ stlrw(rscratch1, rscratch2);
1446
1447 // Force this write out before the read below
1448 if (!UseSystemMemoryBarrier) {
1449 __ dmb(Assembler::ISH);
1450 }
1451
1452 // check for safepoint operation in progress and/or pending suspend requests
1453 {
1454 Label L, Continue;
1455
1456 // No need for acquire as Java threads always disarm themselves.
1457 __ safepoint_poll(L, true /* at_return */, false /* acquire */, false /* in_nmethod */);
1458 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1459 __ cbz(rscratch2, Continue);
1460 __ bind(L);
1461
1462 // Don't use call_VM as it will see a possible pending exception
1463 // and forward it and never return here preventing us from
1464 // clearing _last_native_pc down below. So we do a runtime call by
1465 // hand.
1466 //
1467 __ mov(c_rarg0, rthread);
1468 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1469 __ blr(rscratch2);
1470 __ get_method(rmethod);
1471 __ reinit_heapbase();
1472 __ bind(Continue);
1473 }
1474
1475 // change thread state
1476 __ mov(rscratch1, _thread_in_Java);
1477 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1478 __ stlrw(rscratch1, rscratch2);
1479
1480 if (LockingMode != LM_LEGACY) {
1481 // Check preemption for Object.wait()
1482 Label not_preempted;
1483 __ ldr(rscratch1, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1484 __ cbz(rscratch1, not_preempted);
1485 __ str(zr, Address(rthread, JavaThread::preempt_alternate_return_offset()));
1486 __ br(rscratch1);
1487 __ bind(native_return);
1488 __ restore_after_resume(true /* is_native */);
1489 // reload result_handler
1490 __ ldr(result_handler, Address(rfp, frame::interpreter_frame_result_handler_offset*wordSize));
1491 __ bind(not_preempted);
1492 } else {
1493 // any pc will do so just use this one for LM_LEGACY to keep code together.
1494 __ bind(native_return);
1495 }
1496
1497 // reset_last_Java_frame
1498 __ reset_last_Java_frame(true);
1499
1500 if (CheckJNICalls) {
1501 // clear_pending_jni_exception_check
1502 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1503 }
1504
1505 // reset handle block
1506 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1507 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1508
1509 // If result is an oop unbox and store it in frame where gc will see it
1510 // and result handler will pick it up
1511
1512 {
1513 Label no_oop;
1514 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1515 __ cmp(t, result_handler);
1516 __ br(Assembler::NE, no_oop);
1517 // Unbox oop result, e.g. JNIHandles::resolve result.
1518 __ pop(ltos);
1519 __ resolve_jobject(r0, t, rscratch2);
1520 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1521 // keep stack depth as expected by pushing oop which will eventually be discarded
1522 __ push(ltos);
1523 __ bind(no_oop);
1524 }
1525
1526 {
1527 Label no_reguard;
1528 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1529 __ ldrw(rscratch1, Address(rscratch1));
1530 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1531 __ br(Assembler::NE, no_reguard);
1532
1533 __ push_call_clobbered_registers();
1534 __ mov(c_rarg0, rthread);
1535 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1536 __ blr(rscratch2);
1537 __ pop_call_clobbered_registers();
1538
1539 __ bind(no_reguard);
1540 }
1541
1542 // The method register is junk from after the thread_in_native transition
1543 // until here. Also can't call_VM until the bcp has been
1544 // restored. Need bcp for throwing exception below so get it now.
1545 __ get_method(rmethod);
1546
1547 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1548 // rbcp == code_base()
1549 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1550 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1551 // handle exceptions (exception handling will handle unlocking!)
1552 {
1553 Label L;
1554 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1555 __ cbz(rscratch1, L);
1556 // Note: At some point we may want to unify this with the code
1557 // used in call_VM_base(); i.e., we should use the
1558 // StubRoutines::forward_exception code. For now this doesn't work
1559 // here because the rsp is not correctly set at this point.
1560 __ MacroAssembler::call_VM(noreg,
1561 CAST_FROM_FN_PTR(address,
1562 InterpreterRuntime::throw_pending_exception));
1563 __ should_not_reach_here();
1564 __ bind(L);
1565 }
1566
1567 // do unlocking if necessary
1568 {
1569 Label L;
1570 __ ldrh(t, Address(rmethod, Method::access_flags_offset()));
1571 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1572 // the code below should be shared with interpreter macro
1573 // assembler implementation
1574 {
1575 Label unlock;
1576 // BasicObjectLock will be first in list, since this is a
1577 // synchronized method. However, need to check that the object
1578 // has not been unlocked by an explicit monitorexit bytecode.
1579
1580 // monitor expect in c_rarg1 for slow unlock path
1581 __ lea (c_rarg1, Address(rfp, // address of first monitor
1582 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1583 wordSize - sizeof(BasicObjectLock))));
1584
1585 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1586 __ cbnz(t, unlock);
1587
1588 // Entry already unlocked, need to throw exception
1589 __ MacroAssembler::call_VM(noreg,
1590 CAST_FROM_FN_PTR(address,
1591 InterpreterRuntime::throw_illegal_monitor_state_exception));
1592 __ should_not_reach_here();
1593
1594 __ bind(unlock);
1595 __ unlock_object(c_rarg1);
1596 }
1597 __ bind(L);
1598 }
1599
1600 // jvmti support
1601 // Note: This must happen _after_ handling/throwing any exceptions since
1602 // the exception handler code notifies the runtime of method exits
1603 // too. If this happens before, method entry/exit notifications are
1604 // not properly paired (was bug - gri 11/22/99).
1605 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1606
1607 // restore potential result in r0:d0, call result handler to
1608 // restore potential result in ST0 & handle result
1609
1610 __ pop(ltos);
1611 __ pop(dtos);
1612
1613 __ blr(result_handler);
1614
1615 // remove activation
1616 __ ldr(esp, Address(rfp,
1617 frame::interpreter_frame_sender_sp_offset *
1618 wordSize)); // get sender sp
1619 // remove frame anchor
1620 __ leave();
1621
1622 // restore sender sp
1623 __ mov(sp, esp);
1624
1625 __ ret(lr);
1626
1627 if (inc_counter) {
1628 // Handle overflow of counter and compile method
1629 __ bind(invocation_counter_overflow);
1630 generate_counter_overflow(continue_after_compile);
1631 }
1632
1633 return entry_point;
1634 }
1635
1636 //
1637 // Generic interpreted method entry to (asm) interpreter
1638 //
1639 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized, bool runtime_upcalls) {
1640 // determine code generation flags
1641 bool inc_counter = (UseCompiler || CountCompiledCalls) && !PreloadOnly;
1642
1643 // rscratch1: sender sp
1644 address entry_point = __ pc();
1645
1646 const Address constMethod(rmethod, Method::const_offset());
1647 const Address access_flags(rmethod, Method::access_flags_offset());
1648 const Address size_of_parameters(r3,
1649 ConstMethod::size_of_parameters_offset());
1650 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1651
1652 // get parameter size (always needed)
1653 // need to load the const method first
1654 __ ldr(r3, constMethod);
1655 __ load_unsigned_short(r2, size_of_parameters);
1656
1657 // r2: size of parameters
1658
1659 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1660 __ sub(r3, r3, r2); // r3 = no. of additional locals
1661
1662 // see if we've got enough room on the stack for locals plus overhead.
1663 generate_stack_overflow_check();
1664
1665 // compute beginning of parameters (rlocals)
1666 __ add(rlocals, esp, r2, ext::uxtx, 3);
1667 __ sub(rlocals, rlocals, wordSize);
1668
1669 __ mov(rscratch1, esp);
1670
1671 // r3 - # of additional locals
1672 // allocate space for locals
1673 // explicitly initialize locals
1674 // Initializing memory allocated for locals in the same direction as
1675 // the stack grows to ensure page initialization order according
1676 // to windows-aarch64 stack page growth requirement (see
1677 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1678 {
1679 Label exit, loop;
1680 __ ands(zr, r3, r3);
1681 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1682 __ bind(loop);
1683 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1684 __ sub(r3, r3, 1); // until everything initialized
1685 __ cbnz(r3, loop);
1686 __ bind(exit);
1687 }
1688
1689 // Padding between locals and fixed part of activation frame to ensure
1690 // SP is always 16-byte aligned.
1691 __ andr(sp, rscratch1, -16);
1692
1693 // And the base dispatch table
1694 __ get_dispatch();
1695
1696 // initialize fixed part of activation frame
1697 generate_fixed_frame(false);
1698
1699 // make sure method is not native & not abstract
1700 #ifdef ASSERT
1701 __ ldrh(r0, access_flags);
1702 {
1703 Label L;
1704 __ tst(r0, JVM_ACC_NATIVE);
1705 __ br(Assembler::EQ, L);
1706 __ stop("tried to execute native method as non-native");
1707 __ bind(L);
1708 }
1709 {
1710 Label L;
1711 __ tst(r0, JVM_ACC_ABSTRACT);
1712 __ br(Assembler::EQ, L);
1713 __ stop("tried to execute abstract method in interpreter");
1714 __ bind(L);
1715 }
1716 #endif
1717
1718 // Since at this point in the method invocation the exception
1719 // handler would try to exit the monitor of synchronized methods
1720 // which hasn't been entered yet, we set the thread local variable
1721 // _do_not_unlock_if_synchronized to true. The remove_activation
1722 // will check this flag.
1723
1724 const Address do_not_unlock_if_synchronized(rthread,
1725 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1726 __ mov(rscratch2, true);
1727 __ strb(rscratch2, do_not_unlock_if_synchronized);
1728
1729 Register mdp = r3;
1730 __ profile_parameters_type(mdp, r1, r2);
1731
1732 // increment invocation count & check for overflow
1733 Label invocation_counter_overflow;
1734 if (inc_counter) {
1735 generate_counter_incr(&invocation_counter_overflow);
1736 }
1737
1738 Label continue_after_compile;
1739 __ bind(continue_after_compile);
1740
1741 bang_stack_shadow_pages(false);
1742
1743 // reset the _do_not_unlock_if_synchronized flag
1744 __ strb(zr, do_not_unlock_if_synchronized);
1745
1746 // check for synchronized methods
1747 // Must happen AFTER invocation_counter check and stack overflow check,
1748 // so method is not locked if overflows.
1749 if (synchronized) {
1750 // Allocate monitor and lock method
1751 lock_method();
1752 } else {
1753 // no synchronization necessary
1754 #ifdef ASSERT
1755 {
1756 Label L;
1757 __ ldrh(r0, access_flags);
1758 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1759 __ br(Assembler::EQ, L);
1760 __ stop("method needs synchronization");
1761 __ bind(L);
1762 }
1763 #endif
1764 }
1765
1766 // start execution
1767 #ifdef ASSERT
1768 {
1769 Label L;
1770 const Address monitor_block_top (rfp,
1771 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1772 __ ldr(rscratch1, monitor_block_top);
1773 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1774 __ cmp(esp, rscratch1);
1775 __ br(Assembler::EQ, L);
1776 __ stop("broken stack frame setup in interpreter 2");
1777 __ bind(L);
1778 }
1779 #endif
1780
1781 // jvmti support
1782 __ notify_method_entry();
1783
1784 // runtime upcalls
1785 if (runtime_upcalls) {
1786 __ generate_runtime_upcalls_on_method_entry();
1787 }
1788
1789 __ dispatch_next(vtos);
1790
1791 // invocation counter overflow
1792 if (inc_counter) {
1793 // Handle overflow of counter and compile method
1794 __ bind(invocation_counter_overflow);
1795 generate_counter_overflow(continue_after_compile);
1796 }
1797
1798 return entry_point;
1799 }
1800
1801 // Method entry for java.lang.Thread.currentThread
1802 address TemplateInterpreterGenerator::generate_currentThread() {
1803 address entry_point = __ pc();
1804
1805 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1806 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1807 __ ret(lr);
1808
1809 return entry_point;
1810 }
1811
1812 // Not supported
1813 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1814 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1815 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1816 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1817
1818 //-----------------------------------------------------------------------------
1819 // Exceptions
1820
1821 void TemplateInterpreterGenerator::generate_throw_exception() {
1822 // Entry point in previous activation (i.e., if the caller was
1823 // interpreted)
1824 Interpreter::_rethrow_exception_entry = __ pc();
1825 // Restore sp to interpreter_frame_last_sp even though we are going
1826 // to empty the expression stack for the exception processing.
1827 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1828 // r0: exception
1829 // r3: return address/pc that threw exception
1830 __ restore_bcp(); // rbcp points to call/send
1831 __ restore_locals();
1832 __ restore_constant_pool_cache();
1833 __ reinit_heapbase(); // restore rheapbase as heapbase.
1834 __ get_dispatch();
1835
1836 // Entry point for exceptions thrown within interpreter code
1837 Interpreter::_throw_exception_entry = __ pc();
1838 // If we came here via a NullPointerException on the receiver of a
1839 // method, rmethod may be corrupt.
1840 __ get_method(rmethod);
1841 // expression stack is undefined here
1842 // r0: exception
1843 // rbcp: exception bcp
1844 __ verify_oop(r0);
1845 __ mov(c_rarg1, r0);
1846
1847 // expression stack must be empty before entering the VM in case of
1848 // an exception
1849 __ empty_expression_stack();
1850 // find exception handler address and preserve exception oop
1851 __ call_VM(r3,
1852 CAST_FROM_FN_PTR(address,
1853 InterpreterRuntime::exception_handler_for_exception),
1854 c_rarg1);
1855
1856 // Restore machine SP
1857 __ restore_sp_after_call();
1858
1859 // r0: exception handler entry point
1860 // r3: preserved exception oop
1861 // rbcp: bcp for exception handler
1862 __ push_ptr(r3); // push exception which is now the only value on the stack
1863 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1864
1865 // If the exception is not handled in the current frame the frame is
1866 // removed and the exception is rethrown (i.e. exception
1867 // continuation is _rethrow_exception).
1868 //
1869 // Note: At this point the bci is still the bxi for the instruction
1870 // which caused the exception and the expression stack is
1871 // empty. Thus, for any VM calls at this point, GC will find a legal
1872 // oop map (with empty expression stack).
1873
1874 //
1875 // JVMTI PopFrame support
1876 //
1877
1878 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1879 __ empty_expression_stack();
1880 // Set the popframe_processing bit in pending_popframe_condition
1881 // indicating that we are currently handling popframe, so that
1882 // call_VMs that may happen later do not trigger new popframe
1883 // handling cycles.
1884 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1885 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1886 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1887
1888 {
1889 // Check to see whether we are returning to a deoptimized frame.
1890 // (The PopFrame call ensures that the caller of the popped frame is
1891 // either interpreted or compiled and deoptimizes it if compiled.)
1892 // In this case, we can't call dispatch_next() after the frame is
1893 // popped, but instead must save the incoming arguments and restore
1894 // them after deoptimization has occurred.
1895 //
1896 // Note that we don't compare the return PC against the
1897 // deoptimization blob's unpack entry because of the presence of
1898 // adapter frames in C2.
1899 Label caller_not_deoptimized;
1900 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1901 // This is a return address, so requires authenticating for PAC.
1902 __ authenticate_return_address(c_rarg1);
1903 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1904 InterpreterRuntime::interpreter_contains), c_rarg1);
1905 __ cbnz(r0, caller_not_deoptimized);
1906
1907 // Compute size of arguments for saving when returning to
1908 // deoptimized caller
1909 __ get_method(r0);
1910 __ ldr(r0, Address(r0, Method::const_offset()));
1911 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1912 size_of_parameters_offset())));
1913 __ lsl(r0, r0, Interpreter::logStackElementSize);
1914 __ restore_locals(); // XXX do we need this?
1915 __ sub(rlocals, rlocals, r0);
1916 __ add(rlocals, rlocals, wordSize);
1917 // Save these arguments
1918 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1919 Deoptimization::
1920 popframe_preserve_args),
1921 rthread, r0, rlocals);
1922
1923 __ remove_activation(vtos,
1924 /* throw_monitor_exception */ false,
1925 /* install_monitor_exception */ false,
1926 /* notify_jvmdi */ false);
1927
1928 // Inform deoptimization that it is responsible for restoring
1929 // these arguments
1930 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1931 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1932
1933 // Continue in deoptimization handler
1934 __ ret(lr);
1935
1936 __ bind(caller_not_deoptimized);
1937 }
1938
1939 __ remove_activation(vtos,
1940 /* throw_monitor_exception */ false,
1941 /* install_monitor_exception */ false,
1942 /* notify_jvmdi */ false);
1943
1944 // Restore the last_sp and null it out
1945 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1946 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1947 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1948
1949 __ restore_bcp();
1950 __ restore_locals();
1951 __ restore_constant_pool_cache();
1952 __ get_method(rmethod);
1953 __ get_dispatch();
1954
1955 // The method data pointer was incremented already during
1956 // call profiling. We have to restore the mdp for the current bcp.
1957 if (ProfileInterpreter) {
1958 __ set_method_data_pointer_for_bcp();
1959 }
1960
1961 // Clear the popframe condition flag
1962 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1963 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1964
1965 #if INCLUDE_JVMTI
1966 {
1967 Label L_done;
1968
1969 __ ldrb(rscratch1, Address(rbcp, 0));
1970 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1971 __ br(Assembler::NE, L_done);
1972
1973 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1974 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1975
1976 __ ldr(c_rarg0, Address(rlocals, 0));
1977 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1978
1979 __ cbz(r0, L_done);
1980
1981 __ str(r0, Address(esp, 0));
1982 __ bind(L_done);
1983 }
1984 #endif // INCLUDE_JVMTI
1985
1986 // Restore machine SP
1987 __ restore_sp_after_call();
1988
1989 __ dispatch_next(vtos);
1990 // end of PopFrame support
1991
1992 Interpreter::_remove_activation_entry = __ pc();
1993
1994 // preserve exception over this code sequence
1995 __ pop_ptr(r0);
1996 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1997 // remove the activation (without doing throws on illegalMonitorExceptions)
1998 __ remove_activation(vtos, false, true, false);
1999 // restore exception
2000 __ get_vm_result(r0, rthread);
2001
2002 // In between activations - previous activation type unknown yet
2003 // compute continuation point - the continuation point expects the
2004 // following registers set up:
2005 //
2006 // r0: exception
2007 // lr: return address/pc that threw exception
2008 // esp: expression stack of caller
2009 // rfp: fp of caller
2010 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
2011 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
2012 SharedRuntime::exception_handler_for_return_address),
2013 rthread, lr);
2014 __ mov(r1, r0); // save exception handler
2015 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
2016 // We might be returning to a deopt handler that expects r3 to
2017 // contain the exception pc
2018 __ mov(r3, lr);
2019 // Note that an "issuing PC" is actually the next PC after the call
2020 __ br(r1); // jump to exception
2021 // handler of caller
2022 }
2023
2024
2025 //
2026 // JVMTI ForceEarlyReturn support
2027 //
2028 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2029 address entry = __ pc();
2030
2031 __ restore_bcp();
2032 __ restore_locals();
2033 __ empty_expression_stack();
2034 __ load_earlyret_value(state);
2035
2036 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
2037 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
2038
2039 // Clear the earlyret state
2040 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
2041 __ str(zr, cond_addr);
2042
2043 __ remove_activation(state,
2044 false, /* throw_monitor_exception */
2045 false, /* install_monitor_exception */
2046 true); /* notify_jvmdi */
2047 __ ret(lr);
2048
2049 return entry;
2050 } // end of ForceEarlyReturn support
2051
2052
2053
2054 //-----------------------------------------------------------------------------
2055 // Helper for vtos entry point generation
2056
2057 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2058 address& bep,
2059 address& cep,
2060 address& sep,
2061 address& aep,
2062 address& iep,
2063 address& lep,
2064 address& fep,
2065 address& dep,
2066 address& vep) {
2067 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2068 Label L;
2069 aep = __ pc(); // atos entry point
2070 __ push_ptr();
2071 __ b(L);
2072 fep = __ pc(); // ftos entry point
2073 __ push_f();
2074 __ b(L);
2075 dep = __ pc(); // dtos entry point
2076 __ push_d();
2077 __ b(L);
2078 lep = __ pc(); // ltos entry point
2079 __ push_l();
2080 __ b(L);
2081 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
2082 __ push_i();
2083 vep = __ pc(); // vtos entry point
2084 __ bind(L);
2085 generate_and_dispatch(t);
2086 }
2087
2088 //-----------------------------------------------------------------------------
2089
2090 void TemplateInterpreterGenerator::count_bytecode() {
2091 if (CountBytecodesPerThread) {
2092 Address bc_counter_addr(rthread, Thread::bc_counter_offset());
2093 __ ldr(r10, bc_counter_addr);
2094 __ add(r10, r10, 1);
2095 __ str(r10, bc_counter_addr);
2096 }
2097 if (CountBytecodes || TraceBytecodes || StopInterpreterAt > 0) {
2098 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2099 __ atomic_add(noreg, 1, r10);
2100 }
2101 }
2102
2103 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2104 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2105 __ atomic_addw(noreg, 1, r10);
2106 }
2107
2108 // Non-product code
2109 #ifndef PRODUCT
2110 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2111 address entry = __ pc();
2112
2113 __ protect_return_address();
2114 __ push(lr);
2115 __ push(state);
2116 __ push(RegSet::range(r0, r15), sp);
2117 __ mov(c_rarg2, r0); // Pass itos
2118 __ call_VM(noreg,
2119 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2120 c_rarg1, c_rarg2, c_rarg3);
2121 __ pop(RegSet::range(r0, r15), sp);
2122 __ pop(state);
2123 __ pop(lr);
2124 __ authenticate_return_address();
2125 __ ret(lr); // return from result handler
2126
2127 return entry;
2128 }
2129
2130 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2131 // Calculate new index for counter:
2132 // _index = (_index >> log2_number_of_codes) |
2133 // (bytecode << log2_number_of_codes);
2134 Register index_addr = rscratch1;
2135 Register index = rscratch2;
2136 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2137 __ ldrw(index, index_addr);
2138 __ mov(r10,
2139 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2140 __ orrw(index, r10, index, Assembler::LSR,
2141 BytecodePairHistogram::log2_number_of_codes);
2142 __ strw(index, index_addr);
2143
2144 // Bump bucket contents:
2145 // _counters[_index] ++;
2146 Register counter_addr = rscratch1;
2147 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2148 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2149 __ atomic_addw(noreg, 1, counter_addr);
2150 }
2151
2152 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2153 // Call a little run-time stub to avoid blow-up for each bytecode.
2154 // The run-time runtime saves the right registers, depending on
2155 // the tosca in-state for the given template.
2156
2157 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2158 "entry must have been generated");
2159 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2160 __ reinit_heapbase();
2161 }
2162
2163
2164 void TemplateInterpreterGenerator::stop_interpreter_at() {
2165 Label L;
2166 __ push(rscratch1);
2167 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2168 __ ldr(rscratch1, Address(rscratch1));
2169 __ mov(rscratch2, StopInterpreterAt);
2170 __ cmpw(rscratch1, rscratch2);
2171 __ br(Assembler::NE, L);
2172 __ brk(0);
2173 __ bind(L);
2174 __ pop(rscratch1);
2175 }
2176
2177 #endif // !PRODUCT