1 /*
2 * Copyright (c) 2003, 2024, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "classfile/javaClasses.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "compiler/compiler_globals.hpp"
31 #include "gc/shared/barrierSetAssembler.hpp"
32 #include "interpreter/bytecodeHistogram.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "interpreter/interpreterRuntime.hpp"
35 #include "interpreter/interp_masm.hpp"
36 #include "interpreter/templateInterpreterGenerator.hpp"
37 #include "interpreter/templateTable.hpp"
38 #include "interpreter/bytecodeTracer.hpp"
39 #include "memory/resourceArea.hpp"
40 #include "oops/arrayOop.hpp"
41 #include "oops/method.hpp"
42 #include "oops/methodCounters.hpp"
43 #include "oops/methodData.hpp"
44 #include "oops/oop.inline.hpp"
45 #include "oops/inlineKlass.hpp"
46 #include "oops/resolvedIndyEntry.hpp"
47 #include "oops/resolvedMethodEntry.hpp"
48 #include "prims/jvmtiExport.hpp"
49 #include "prims/jvmtiThreadState.hpp"
50 #include "runtime/arguments.hpp"
51 #include "runtime/deoptimization.hpp"
52 #include "runtime/frame.inline.hpp"
53 #include "runtime/globals.hpp"
54 #include "runtime/jniHandles.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/synchronizer.hpp"
58 #include "runtime/timer.hpp"
59 #include "runtime/vframeArray.hpp"
60 #include "utilities/checkedCast.hpp"
61 #include "utilities/debug.hpp"
62 #include "utilities/powerOfTwo.hpp"
63 #include <sys/types.h>
64
65 // Size of interpreter code. Increase if too small. Interpreter will
66 // fail with a guarantee ("not enough space for interpreter generation");
67 // if too small.
68 // Run with +PrintInterpreter to get the VM to print out the size.
69 // Max size with JVMTI
70 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
71
72 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
73
74 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
75 address entry = __ pc();
76
77 __ andr(esp, esp, -16);
78 __ mov(c_rarg3, esp);
79 // rmethod
80 // rlocals
81 // c_rarg3: first stack arg - wordSize
82
83 // adjust sp
84 __ sub(sp, c_rarg3, 18 * wordSize);
85 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
86 __ call_VM(noreg,
87 CAST_FROM_FN_PTR(address,
88 InterpreterRuntime::slow_signature_handler),
89 rmethod, rlocals, c_rarg3);
90
91 // r0: result handler
92
93 // Stack layout:
94 // rsp: return address <- sp
95 // 1 garbage
96 // 8 integer args (if static first is unused)
97 // 1 float/double identifiers
98 // 8 double args
99 // stack args <- esp
100 // garbage
101 // expression stack bottom
102 // bcp (null)
103 // ...
104
105 // Restore LR
106 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
107
108 // Do FP first so we can use c_rarg3 as temp
109 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
110
111 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
112 const FloatRegister r = as_FloatRegister(i);
113
114 Label d, done;
115
116 __ tbnz(c_rarg3, i, d);
117 __ ldrs(r, Address(sp, (10 + i) * wordSize));
118 __ b(done);
119 __ bind(d);
120 __ ldrd(r, Address(sp, (10 + i) * wordSize));
121 __ bind(done);
122 }
123
124 // c_rarg0 contains the result from the call of
125 // InterpreterRuntime::slow_signature_handler so we don't touch it
126 // here. It will be loaded with the JNIEnv* later.
127 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
128 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
129 Register rm = as_Register(i), rn = as_Register(i+1);
130 __ ldp(rm, rn, Address(sp, i * wordSize));
131 }
132
133 __ add(sp, sp, 18 * wordSize);
134 __ ret(lr);
135
136 return entry;
137 }
138
139
140 //
141 // Various method entries
142 //
143
144 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
145 // rmethod: Method*
146 // r19_sender_sp: sender sp
147 // esp: args
148
149 // These don't need a safepoint check because they aren't virtually
150 // callable. We won't enter these intrinsics from compiled code.
151 // If in the future we added an intrinsic which was virtually callable
152 // we'd have to worry about how to safepoint so that this code is used.
153
154 // mathematical functions inlined by compiler
155 // (interpreter must provide identical implementation
156 // in order to avoid monotonicity bugs when switching
157 // from interpreter to compiler in the middle of some
158 // computation)
159 //
160 // stack:
161 // [ arg ] <-- esp
162 // [ arg ]
163 // retaddr in lr
164
165 address entry_point = nullptr;
166 Register continuation = lr;
167 switch (kind) {
168 case Interpreter::java_lang_math_abs:
169 entry_point = __ pc();
170 __ ldrd(v0, Address(esp));
171 __ fabsd(v0, v0);
172 __ mov(sp, r19_sender_sp); // Restore caller's SP
173 break;
174 case Interpreter::java_lang_math_sqrt:
175 entry_point = __ pc();
176 __ ldrd(v0, Address(esp));
177 __ fsqrtd(v0, v0);
178 __ mov(sp, r19_sender_sp);
179 break;
180 case Interpreter::java_lang_math_sin :
181 case Interpreter::java_lang_math_cos :
182 case Interpreter::java_lang_math_tan :
183 case Interpreter::java_lang_math_log :
184 case Interpreter::java_lang_math_log10 :
185 case Interpreter::java_lang_math_exp :
186 entry_point = __ pc();
187 __ ldrd(v0, Address(esp));
188 __ mov(sp, r19_sender_sp);
189 __ mov(r23, lr);
190 continuation = r23; // The first free callee-saved register
191 generate_transcendental_entry(kind, 1);
192 break;
193 case Interpreter::java_lang_math_pow :
194 entry_point = __ pc();
195 __ mov(r23, lr);
196 continuation = r23;
197 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
198 __ ldrd(v1, Address(esp));
199 __ mov(sp, r19_sender_sp);
200 generate_transcendental_entry(kind, 2);
201 break;
202 case Interpreter::java_lang_math_fmaD :
203 if (UseFMA) {
204 entry_point = __ pc();
205 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
206 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
207 __ ldrd(v2, Address(esp));
208 __ fmaddd(v0, v0, v1, v2);
209 __ mov(sp, r19_sender_sp); // Restore caller's SP
210 }
211 break;
212 case Interpreter::java_lang_math_fmaF :
213 if (UseFMA) {
214 entry_point = __ pc();
215 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
216 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
217 __ ldrs(v2, Address(esp));
218 __ fmadds(v0, v0, v1, v2);
219 __ mov(sp, r19_sender_sp); // Restore caller's SP
220 }
221 break;
222 default:
223 ;
224 }
225 if (entry_point) {
226 __ br(continuation);
227 }
228
229 return entry_point;
230 }
231
232 // double trigonometrics and transcendentals
233 // static jdouble dsin(jdouble x);
234 // static jdouble dcos(jdouble x);
235 // static jdouble dtan(jdouble x);
236 // static jdouble dlog(jdouble x);
237 // static jdouble dlog10(jdouble x);
238 // static jdouble dexp(jdouble x);
239 // static jdouble dpow(jdouble x, jdouble y);
240
241 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
242 address fn;
243 switch (kind) {
244 case Interpreter::java_lang_math_sin :
245 if (StubRoutines::dsin() == nullptr) {
246 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
247 } else {
248 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
249 }
250 break;
251 case Interpreter::java_lang_math_cos :
252 if (StubRoutines::dcos() == nullptr) {
253 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
254 } else {
255 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
256 }
257 break;
258 case Interpreter::java_lang_math_tan :
259 if (StubRoutines::dtan() == nullptr) {
260 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
261 } else {
262 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
263 }
264 break;
265 case Interpreter::java_lang_math_log :
266 if (StubRoutines::dlog() == nullptr) {
267 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
268 } else {
269 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
270 }
271 break;
272 case Interpreter::java_lang_math_log10 :
273 if (StubRoutines::dlog10() == nullptr) {
274 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
275 } else {
276 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
277 }
278 break;
279 case Interpreter::java_lang_math_exp :
280 if (StubRoutines::dexp() == nullptr) {
281 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
282 } else {
283 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
284 }
285 break;
286 case Interpreter::java_lang_math_pow :
287 if (StubRoutines::dpow() == nullptr) {
288 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
289 } else {
290 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
291 }
292 break;
293 default:
294 ShouldNotReachHere();
295 fn = nullptr; // unreachable
296 }
297 __ mov(rscratch1, fn);
298 __ blr(rscratch1);
299 }
300
301 address TemplateInterpreterGenerator::generate_Float_float16ToFloat_entry() {
302 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
303 // r19_sender_sp: sender sp
304 // stack:
305 // [ arg ] <-- esp
306 // [ arg ]
307 // retaddr in lr
308 // result in v0
309
310 address entry_point = __ pc();
311 __ ldrw(c_rarg0, Address(esp));
312 __ flt16_to_flt(v0, c_rarg0, v1);
313 __ mov(sp, r19_sender_sp); // Restore caller's SP
314 __ br(lr);
315 return entry_point;
316 }
317
318 address TemplateInterpreterGenerator::generate_Float_floatToFloat16_entry() {
319 assert(VM_Version::supports_float16(), "this intrinsic is not supported");
320 // r19_sender_sp: sender sp
321 // stack:
322 // [ arg ] <-- esp
323 // [ arg ]
324 // retaddr in lr
325 // result in c_rarg0
326
327 address entry_point = __ pc();
328 __ ldrs(v0, Address(esp));
329 __ flt_to_flt16(c_rarg0, v0, v1);
330 __ mov(sp, r19_sender_sp); // Restore caller's SP
331 __ br(lr);
332 return entry_point;
333 }
334
335 // Abstract method entry
336 // Attempt to execute abstract method. Throw exception
337 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
338 // rmethod: Method*
339 // r19_sender_sp: sender SP
340
341 address entry_point = __ pc();
342
343 // abstract method entry
344
345 // pop return address, reset last_sp to null
346 __ empty_expression_stack();
347 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
348 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
349
350 // throw exception
351 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
352 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
353 rmethod);
354 // the call_VM checks for exception, so we should never return here.
355 __ should_not_reach_here();
356
357 return entry_point;
358 }
359
360 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
361 address entry = __ pc();
362
363 #ifdef ASSERT
364 {
365 Label L;
366 __ ldr(rscratch1, Address(rfp,
367 frame::interpreter_frame_monitor_block_top_offset *
368 wordSize));
369 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
370 __ mov(rscratch2, sp);
371 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
372 // grows negative)
373 __ br(Assembler::HS, L); // check if frame is complete
374 __ stop ("interpreter frame not set up");
375 __ bind(L);
376 }
377 #endif // ASSERT
378 // Restore bcp under the assumption that the current frame is still
379 // interpreted
380 __ restore_bcp();
381
382 // expression stack must be empty before entering the VM if an
383 // exception happened
384 __ empty_expression_stack();
385 // throw exception
386 __ call_VM(noreg,
387 CAST_FROM_FN_PTR(address,
388 InterpreterRuntime::throw_StackOverflowError));
389 return entry;
390 }
391
392 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
393 address entry = __ pc();
394 // expression stack must be empty before entering the VM if an
395 // exception happened
396 __ empty_expression_stack();
397 // setup parameters
398
399 // ??? convention: expect aberrant index in register r1
400 __ movw(c_rarg2, r1);
401 // ??? convention: expect array in register r3
402 __ mov(c_rarg1, r3);
403 __ call_VM(noreg,
404 CAST_FROM_FN_PTR(address,
405 InterpreterRuntime::
406 throw_ArrayIndexOutOfBoundsException),
407 c_rarg1, c_rarg2);
408 return entry;
409 }
410
411 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
412 address entry = __ pc();
413
414 // object is at TOS
415 __ pop(c_rarg1);
416
417 // expression stack must be empty before entering the VM if an
418 // exception happened
419 __ empty_expression_stack();
420
421 __ call_VM(noreg,
422 CAST_FROM_FN_PTR(address,
423 InterpreterRuntime::
424 throw_ClassCastException),
425 c_rarg1);
426 return entry;
427 }
428
429 address TemplateInterpreterGenerator::generate_exception_handler_common(
430 const char* name, const char* message, bool pass_oop) {
431 assert(!pass_oop || message == nullptr, "either oop or message but not both");
432 address entry = __ pc();
433 if (pass_oop) {
434 // object is at TOS
435 __ pop(c_rarg2);
436 }
437 // expression stack must be empty before entering the VM if an
438 // exception happened
439 __ empty_expression_stack();
440 // setup parameters
441 __ lea(c_rarg1, Address((address)name));
442 if (pass_oop) {
443 __ call_VM(r0, CAST_FROM_FN_PTR(address,
444 InterpreterRuntime::
445 create_klass_exception),
446 c_rarg1, c_rarg2);
447 } else {
448 // kind of lame ExternalAddress can't take null because
449 // external_word_Relocation will assert.
450 if (message != nullptr) {
451 __ lea(c_rarg2, Address((address)message));
452 } else {
453 __ mov(c_rarg2, NULL_WORD);
454 }
455 __ call_VM(r0,
456 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
457 c_rarg1, c_rarg2);
458 }
459 // throw exception
460 __ b(address(Interpreter::throw_exception_entry()));
461 return entry;
462 }
463
464 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
465 address entry = __ pc();
466
467 // Restore stack bottom in case i2c adjusted stack
468 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
469 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
470 // and null it as marker that esp is now tos until next java call
471 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
472
473 if (state == atos && InlineTypeReturnedAsFields) {
474 __ store_inline_type_fields_to_buf(nullptr, true);
475 }
476
477 __ restore_bcp();
478 __ restore_locals();
479 __ restore_constant_pool_cache();
480 __ get_method(rmethod);
481
482 if (state == atos) {
483 Register obj = r0;
484 Register mdp = r1;
485 Register tmp = r2;
486 __ profile_return_type(mdp, obj, tmp);
487 }
488
489 const Register cache = r1;
490 const Register index = r2;
491
492 if (index_size == sizeof(u4)) {
493 __ load_resolved_indy_entry(cache, index);
494 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedIndyEntry::num_parameters_offset())));
495 __ add(esp, esp, cache, Assembler::LSL, 3);
496 } else {
497 // Pop N words from the stack
498 assert(index_size == sizeof(u2), "Can only be u2");
499 __ load_method_entry(cache, index);
500 __ load_unsigned_short(cache, Address(cache, in_bytes(ResolvedMethodEntry::num_parameters_offset())));
501 __ add(esp, esp, cache, Assembler::LSL, 3);
502 }
503
504 // Restore machine SP
505 __ restore_sp_after_call();
506
507 __ check_and_handle_popframe(rthread);
508 __ check_and_handle_earlyret(rthread);
509
510 __ get_dispatch();
511 __ dispatch_next(state, step);
512
513 return entry;
514 }
515
516 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
517 int step,
518 address continuation) {
519 address entry = __ pc();
520 __ restore_bcp();
521 __ restore_locals();
522 __ restore_constant_pool_cache();
523 __ get_method(rmethod);
524 __ get_dispatch();
525
526 __ restore_sp_after_call(); // Restore SP to extended SP
527
528 // Restore expression stack pointer
529 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
530 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
531 // null last_sp until next java call
532 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
533
534 #if INCLUDE_JVMCI
535 // Check if we need to take lock at entry of synchronized method. This can
536 // only occur on method entry so emit it only for vtos with step 0.
537 if (EnableJVMCI && state == vtos && step == 0) {
538 Label L;
539 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
540 __ cbz(rscratch1, L);
541 // Clear flag.
542 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
543 // Take lock.
544 lock_method();
545 __ bind(L);
546 } else {
547 #ifdef ASSERT
548 if (EnableJVMCI) {
549 Label L;
550 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
551 __ cbz(rscratch1, L);
552 __ stop("unexpected pending monitor in deopt entry");
553 __ bind(L);
554 }
555 #endif
556 }
557 #endif
558 // handle exceptions
559 {
560 Label L;
561 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
562 __ cbz(rscratch1, L);
563 __ call_VM(noreg,
564 CAST_FROM_FN_PTR(address,
565 InterpreterRuntime::throw_pending_exception));
566 __ should_not_reach_here();
567 __ bind(L);
568 }
569
570 if (continuation == nullptr) {
571 __ dispatch_next(state, step);
572 } else {
573 __ jump_to_entry(continuation);
574 }
575 return entry;
576 }
577
578 address TemplateInterpreterGenerator::generate_result_handler_for(
579 BasicType type) {
580 address entry = __ pc();
581 switch (type) {
582 case T_BOOLEAN: __ c2bool(r0); break;
583 case T_CHAR : __ uxth(r0, r0); break;
584 case T_BYTE : __ sxtb(r0, r0); break;
585 case T_SHORT : __ sxth(r0, r0); break;
586 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
587 case T_LONG : /* nothing to do */ break;
588 case T_VOID : /* nothing to do */ break;
589 case T_FLOAT : /* nothing to do */ break;
590 case T_DOUBLE : /* nothing to do */ break;
591 case T_OBJECT :
592 // retrieve result from frame
593 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
594 // and verify it
595 __ verify_oop(r0);
596 break;
597 default : ShouldNotReachHere();
598 }
599 __ ret(lr); // return from result handler
600 return entry;
601 }
602
603 address TemplateInterpreterGenerator::generate_safept_entry_for(
604 TosState state,
605 address runtime_entry) {
606 address entry = __ pc();
607 __ push(state);
608 __ push_cont_fastpath(rthread);
609 __ call_VM(noreg, runtime_entry);
610 __ pop_cont_fastpath(rthread);
611 __ membar(Assembler::AnyAny);
612 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
613 return entry;
614 }
615
616 // Helpers for commoning out cases in the various type of method entries.
617 //
618
619
620 // increment invocation count & check for overflow
621 //
622 // Note: checking for negative value instead of overflow
623 // so we have a 'sticky' overflow test
624 //
625 // rmethod: method
626 //
627 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow) {
628 Label done;
629 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
630 int increment = InvocationCounter::count_increment;
631 Label no_mdo;
632 if (ProfileInterpreter) {
633 // Are we profiling?
634 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
635 __ cbz(r0, no_mdo);
636 // Increment counter in the MDO
637 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
638 in_bytes(InvocationCounter::counter_offset()));
639 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
640 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
641 __ b(done);
642 }
643 __ bind(no_mdo);
644 // Increment counter in MethodCounters
645 const Address invocation_counter(rscratch2,
646 MethodCounters::invocation_counter_offset() +
647 InvocationCounter::counter_offset());
648 __ get_method_counters(rmethod, rscratch2, done);
649 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
650 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
651 __ bind(done);
652 }
653
654 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
655
656 // Asm interpreter on entry
657 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
658 // Everything as it was on entry
659
660 // InterpreterRuntime::frequency_counter_overflow takes two
661 // arguments, the first (thread) is passed by call_VM, the second
662 // indicates if the counter overflow occurs at a backwards branch
663 // (null bcp). We pass zero for it. The call returns the address
664 // of the verified entry point for the method or null if the
665 // compilation did not complete (either went background or bailed
666 // out).
667 __ mov(c_rarg1, 0);
668 __ call_VM(noreg,
669 CAST_FROM_FN_PTR(address,
670 InterpreterRuntime::frequency_counter_overflow),
671 c_rarg1);
672
673 __ b(do_continue);
674 }
675
676 // See if we've got enough room on the stack for locals plus overhead
677 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
678 // without going through the signal handler, i.e., reserved and yellow zones
679 // will not be made usable. The shadow zone must suffice to handle the
680 // overflow.
681 // The expression stack grows down incrementally, so the normal guard
682 // page mechanism will work for that.
683 //
684 // NOTE: Since the additional locals are also always pushed (wasn't
685 // obvious in generate_method_entry) so the guard should work for them
686 // too.
687 //
688 // Args:
689 // r3: number of additional locals this frame needs (what we must check)
690 // rmethod: Method*
691 //
692 // Kills:
693 // r0
694 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
695
696 // monitor entry size: see picture of stack set
697 // (generate_method_entry) and frame_amd64.hpp
698 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
699
700 // total overhead size: entry_size + (saved rbp through expr stack
701 // bottom). be sure to change this if you add/subtract anything
702 // to/from the overhead area
703 const int overhead_size =
704 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
705
706 const size_t page_size = os::vm_page_size();
707
708 Label after_frame_check;
709
710 // see if the frame is greater than one page in size. If so,
711 // then we need to verify there is enough stack space remaining
712 // for the additional locals.
713 //
714 // Note that we use SUBS rather than CMP here because the immediate
715 // field of this instruction may overflow. SUBS can cope with this
716 // because it is a macro that will expand to some number of MOV
717 // instructions and a register operation.
718 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
719 __ br(Assembler::LS, after_frame_check);
720
721 // compute rsp as if this were going to be the last frame on
722 // the stack before the red zone
723
724 // locals + overhead, in bytes
725 __ mov(r0, overhead_size);
726 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
727
728 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
729 __ ldr(rscratch1, stack_limit);
730
731 #ifdef ASSERT
732 Label limit_okay;
733 // Verify that thread stack limit is non-zero.
734 __ cbnz(rscratch1, limit_okay);
735 __ stop("stack overflow limit is zero");
736 __ bind(limit_okay);
737 #endif
738
739 // Add stack limit to locals.
740 __ add(r0, r0, rscratch1);
741
742 // Check against the current stack bottom.
743 __ cmp(sp, r0);
744 __ br(Assembler::HI, after_frame_check);
745
746 // Remove the incoming args, peeling the machine SP back to where it
747 // was in the caller. This is not strictly necessary, but unless we
748 // do so the stack frame may have a garbage FP; this ensures a
749 // correct call stack that we can always unwind. The ANDR should be
750 // unnecessary because the sender SP in r19 is always aligned, but
751 // it doesn't hurt.
752 __ andr(sp, r19_sender_sp, -16);
753
754 // Note: the restored frame is not necessarily interpreted.
755 // Use the shared runtime version of the StackOverflowError.
756 assert(SharedRuntime::throw_StackOverflowError_entry() != nullptr, "stub not yet generated");
757 __ far_jump(RuntimeAddress(SharedRuntime::throw_StackOverflowError_entry()));
758
759 // all done with frame size check
760 __ bind(after_frame_check);
761 }
762
763 // Allocate monitor and lock method (asm interpreter)
764 //
765 // Args:
766 // rmethod: Method*
767 // rlocals: locals
768 //
769 // Kills:
770 // r0
771 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
772 // rscratch1, rscratch2 (scratch regs)
773 void TemplateInterpreterGenerator::lock_method() {
774 // synchronize method
775 const Address access_flags(rmethod, Method::access_flags_offset());
776 const Address monitor_block_top(
777 rfp,
778 frame::interpreter_frame_monitor_block_top_offset * wordSize);
779 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes();
780
781 #ifdef ASSERT
782 {
783 Label L;
784 __ ldrw(r0, access_flags);
785 __ tst(r0, JVM_ACC_SYNCHRONIZED);
786 __ br(Assembler::NE, L);
787 __ stop("method doesn't need synchronization");
788 __ bind(L);
789 }
790 #endif // ASSERT
791
792 // get synchronization object
793 {
794 Label done;
795 __ ldrw(r0, access_flags);
796 __ tst(r0, JVM_ACC_STATIC);
797 // get receiver (assume this is frequent case)
798 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
799 __ br(Assembler::EQ, done);
800 __ load_mirror(r0, rmethod, r5, rscratch2);
801
802 #ifdef ASSERT
803 {
804 Label L;
805 __ cbnz(r0, L);
806 __ stop("synchronization object is null");
807 __ bind(L);
808 }
809 #endif // ASSERT
810
811 __ bind(done);
812 }
813
814 // add space for monitor & lock
815 __ check_extended_sp();
816 __ sub(sp, sp, entry_size); // add space for a monitor entry
817 __ sub(esp, esp, entry_size);
818 __ sub(rscratch1, sp, rfp);
819 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
820 __ str(rscratch1, Address(rfp, frame::interpreter_frame_extended_sp_offset * wordSize));
821 __ sub(rscratch1, esp, rfp);
822 __ asr(rscratch1, rscratch1, Interpreter::logStackElementSize);
823 __ str(rscratch1, monitor_block_top); // set new monitor block top
824
825 // store object
826 __ str(r0, Address(esp, BasicObjectLock::obj_offset()));
827 __ mov(c_rarg1, esp); // object address
828 __ lock_object(c_rarg1);
829 }
830
831 // Generate a fixed interpreter frame. This is identical setup for
832 // interpreted methods and for native methods hence the shared code.
833 //
834 // Args:
835 // lr: return address
836 // rmethod: Method*
837 // rlocals: pointer to locals
838 // rcpool: cp cache
839 // stack_pointer: previous sp
840 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
841 // initialize fixed part of activation frame
842 if (native_call) {
843 __ sub(esp, sp, 14 * wordSize);
844 __ mov(rbcp, zr);
845 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
846 __ stp(rscratch1, zr, Address(__ pre(sp, -14 * wordSize)));
847 // add 2 zero-initialized slots for native calls
848 __ stp(zr, zr, Address(sp, 12 * wordSize));
849 } else {
850 __ sub(esp, sp, 12 * wordSize);
851 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
852 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
853 __ mov(rscratch1, frame::interpreter_frame_initial_sp_offset);
854 __ stp(rscratch1, rbcp, Address(__ pre(sp, -12 * wordSize)));
855 }
856
857 if (ProfileInterpreter) {
858 Label method_data_continue;
859 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
860 __ cbz(rscratch1, method_data_continue);
861 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
862 __ bind(method_data_continue);
863 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
864 } else {
865 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
866 }
867
868 __ protect_return_address();
869 __ stp(rfp, lr, Address(sp, 10 * wordSize));
870 __ lea(rfp, Address(sp, 10 * wordSize));
871
872 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
873 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
874 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset()));
875 __ sub(rscratch1, rlocals, rfp);
876 __ lsr(rscratch1, rscratch1, Interpreter::logStackElementSize); // rscratch1 = rlocals - fp();
877 // Store relativized rlocals, see frame::interpreter_frame_locals().
878 __ stp(rscratch1, rcpool, Address(sp, 2 * wordSize));
879
880 // set sender sp
881 // leave last_sp as null
882 __ stp(zr, r19_sender_sp, Address(sp, 8 * wordSize));
883
884 // Get mirror
885 __ load_mirror(r10, rmethod, r5, rscratch2);
886 if (! native_call) {
887 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
888 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
889 __ add(rscratch1, rscratch1, MAX2(3, Method::extra_stack_entries()));
890 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
891 __ andr(rscratch1, rscratch1, -16);
892 __ sub(rscratch2, rscratch1, rfp);
893 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
894 // Store extended SP and mirror
895 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
896 // Move SP out of the way
897 __ mov(sp, rscratch1);
898 } else {
899 // Make sure there is room for the exception oop pushed in case method throws
900 // an exception (see TemplateInterpreterGenerator::generate_throw_exception())
901 __ sub(rscratch1, sp, 2 * wordSize);
902 __ sub(rscratch2, rscratch1, rfp);
903 __ asr(rscratch2, rscratch2, Interpreter::logStackElementSize);
904 __ stp(r10, rscratch2, Address(sp, 4 * wordSize));
905 __ mov(sp, rscratch1);
906 }
907 }
908
909 // End of helpers
910
911 // Various method entries
912 //------------------------------------------------------------------------------------------------------------------------
913 //
914 //
915
916 // Method entry for java.lang.ref.Reference.get.
917 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
918 // Code: _aload_0, _getfield, _areturn
919 // parameter size = 1
920 //
921 // The code that gets generated by this routine is split into 2 parts:
922 // 1. The "intrinsified" code for G1 (or any SATB based GC),
923 // 2. The slow path - which is an expansion of the regular method entry.
924 //
925 // Notes:-
926 // * In the G1 code we do not check whether we need to block for
927 // a safepoint. If G1 is enabled then we must execute the specialized
928 // code for Reference.get (except when the Reference object is null)
929 // so that we can log the value in the referent field with an SATB
930 // update buffer.
931 // If the code for the getfield template is modified so that the
932 // G1 pre-barrier code is executed when the current method is
933 // Reference.get() then going through the normal method entry
934 // will be fine.
935 // * The G1 code can, however, check the receiver object (the instance
936 // of java.lang.Reference) and jump to the slow path if null. If the
937 // Reference object is null then we obviously cannot fetch the referent
938 // and so we don't need to call the G1 pre-barrier. Thus we can use the
939 // regular method entry code to generate the NPE.
940 //
941 // This code is based on generate_accessor_entry.
942 //
943 // rmethod: Method*
944 // r19_sender_sp: senderSP must preserve for slow path, set SP to it on fast path
945
946 // LR is live. It must be saved around calls.
947
948 address entry = __ pc();
949
950 const int referent_offset = java_lang_ref_Reference::referent_offset();
951
952 Label slow_path;
953 const Register local_0 = c_rarg0;
954 // Check if local 0 != null
955 // If the receiver is null then it is OK to jump to the slow path.
956 __ ldr(local_0, Address(esp, 0));
957 __ cbz(local_0, slow_path);
958
959 // Load the value of the referent field.
960 const Address field_address(local_0, referent_offset);
961 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
962 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch1, /*tmp2*/ rscratch2);
963
964 // areturn
965 __ andr(sp, r19_sender_sp, -16); // done with stack
966 __ ret(lr);
967
968 // generate a vanilla interpreter entry as the slow path
969 __ bind(slow_path);
970 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
971 return entry;
972
973 }
974
975 /**
976 * Method entry for static native methods:
977 * int java.util.zip.CRC32.update(int crc, int b)
978 */
979 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
980 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
981 address entry = __ pc();
982
983 // rmethod: Method*
984 // r19_sender_sp: senderSP must preserved for slow path
985 // esp: args
986
987 Label slow_path;
988 // If we need a safepoint check, generate full interpreter entry.
989 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
990
991 // We don't generate local frame and don't align stack because
992 // we call stub code and there is no safepoint on this path.
993
994 // Load parameters
995 const Register crc = c_rarg0; // crc
996 const Register val = c_rarg1; // source java byte value
997 const Register tbl = c_rarg2; // scratch
998
999 // Arguments are reversed on java expression stack
1000 __ ldrw(val, Address(esp, 0)); // byte value
1001 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1002
1003 uint64_t offset;
1004 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1005 __ add(tbl, tbl, offset);
1006
1007 __ mvnw(crc, crc); // ~crc
1008 __ update_byte_crc32(crc, val, tbl);
1009 __ mvnw(crc, crc); // ~crc
1010
1011 // result in c_rarg0
1012
1013 __ andr(sp, r19_sender_sp, -16);
1014 __ ret(lr);
1015
1016 // generate a vanilla native entry as the slow path
1017 __ bind(slow_path);
1018 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1019 return entry;
1020 }
1021
1022 /**
1023 * Method entry for static native methods:
1024 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1025 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1026 */
1027 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1028 assert(UseCRC32Intrinsics, "this intrinsic is not supported");
1029 address entry = __ pc();
1030
1031 // rmethod,: Method*
1032 // r19_sender_sp: senderSP must preserved for slow path
1033
1034 Label slow_path;
1035 // If we need a safepoint check, generate full interpreter entry.
1036 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1037
1038 // We don't generate local frame and don't align stack because
1039 // we call stub code and there is no safepoint on this path.
1040
1041 // Load parameters
1042 const Register crc = c_rarg0; // crc
1043 const Register buf = c_rarg1; // source java byte array address
1044 const Register len = c_rarg2; // length
1045 const Register off = len; // offset (never overlaps with 'len')
1046
1047 // Arguments are reversed on java expression stack
1048 // Calculate address of start element
1049 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1050 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1051 __ ldrw(off, Address(esp, wordSize)); // offset
1052 __ add(buf, buf, off); // + offset
1053 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1054 } else {
1055 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1056 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1057 __ ldrw(off, Address(esp, wordSize)); // offset
1058 __ add(buf, buf, off); // + offset
1059 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1060 }
1061 // Can now load 'len' since we're finished with 'off'
1062 __ ldrw(len, Address(esp, 0x0)); // Length
1063
1064 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1065
1066 // We are frameless so we can just jump to the stub.
1067 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1068
1069 // generate a vanilla native entry as the slow path
1070 __ bind(slow_path);
1071 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1072 return entry;
1073 }
1074
1075 /**
1076 * Method entry for intrinsic-candidate (non-native) methods:
1077 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1078 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1079 * Unlike CRC32, CRC32C does not have any methods marked as native
1080 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1081 */
1082 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1083 assert(UseCRC32CIntrinsics, "this intrinsic is not supported");
1084 address entry = __ pc();
1085
1086 // Prepare jump to stub using parameters from the stack
1087 const Register crc = c_rarg0; // initial crc
1088 const Register buf = c_rarg1; // source java byte array address
1089 const Register len = c_rarg2; // len argument to the kernel
1090
1091 const Register end = len; // index of last element to process
1092 const Register off = crc; // offset
1093
1094 __ ldrw(end, Address(esp)); // int end
1095 __ ldrw(off, Address(esp, wordSize)); // int offset
1096 __ sub(len, end, off);
1097 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1098 __ add(buf, buf, off); // + offset
1099 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1100 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1101 } else {
1102 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1103 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1104 }
1105
1106 __ andr(sp, r19_sender_sp, -16); // Restore the caller's SP
1107
1108 // Jump to the stub.
1109 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1110
1111 return entry;
1112 }
1113
1114 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1115 // See more discussion in stackOverflow.hpp.
1116
1117 const int shadow_zone_size = checked_cast<int>(StackOverflow::stack_shadow_zone_size());
1118 const int page_size = (int)os::vm_page_size();
1119 const int n_shadow_pages = shadow_zone_size / page_size;
1120
1121 #ifdef ASSERT
1122 Label L_good_limit;
1123 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1124 __ cbnz(rscratch1, L_good_limit);
1125 __ stop("shadow zone safe limit is not initialized");
1126 __ bind(L_good_limit);
1127
1128 Label L_good_watermark;
1129 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1130 __ cbnz(rscratch1, L_good_watermark);
1131 __ stop("shadow zone growth watermark is not initialized");
1132 __ bind(L_good_watermark);
1133 #endif
1134
1135 Label L_done;
1136
1137 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1138 __ cmp(sp, rscratch1);
1139 __ br(Assembler::HI, L_done);
1140
1141 for (int p = 1; p <= n_shadow_pages; p++) {
1142 __ sub(rscratch2, sp, p*page_size);
1143 __ str(zr, Address(rscratch2));
1144 }
1145
1146 // Record the new watermark, but only if the update is above the safe limit.
1147 // Otherwise, the next time around the check above would pass the safe limit.
1148 __ ldr(rscratch1, Address(rthread, JavaThread::shadow_zone_safe_limit()));
1149 __ cmp(sp, rscratch1);
1150 __ br(Assembler::LS, L_done);
1151 __ mov(rscratch1, sp);
1152 __ str(rscratch1, Address(rthread, JavaThread::shadow_zone_growth_watermark()));
1153
1154 __ bind(L_done);
1155 }
1156
1157 // Interpreter stub for calling a native method. (asm interpreter)
1158 // This sets up a somewhat different looking stack for calling the
1159 // native method than the typical interpreter frame setup.
1160 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1161 // determine code generation flags
1162 bool inc_counter = UseCompiler || CountCompiledCalls;
1163
1164 // r1: Method*
1165 // rscratch1: sender sp
1166
1167 address entry_point = __ pc();
1168
1169 const Address constMethod (rmethod, Method::const_offset());
1170 const Address access_flags (rmethod, Method::access_flags_offset());
1171 const Address size_of_parameters(r2, ConstMethod::
1172 size_of_parameters_offset());
1173
1174 // get parameter size (always needed)
1175 __ ldr(r2, constMethod);
1176 __ load_unsigned_short(r2, size_of_parameters);
1177
1178 // Native calls don't need the stack size check since they have no
1179 // expression stack and the arguments are already on the stack and
1180 // we only add a handful of words to the stack.
1181
1182 // rmethod: Method*
1183 // r2: size of parameters
1184 // rscratch1: sender sp
1185
1186 // for natives the size of locals is zero
1187
1188 // compute beginning of parameters (rlocals)
1189 __ add(rlocals, esp, r2, ext::uxtx, 3);
1190 __ add(rlocals, rlocals, -wordSize);
1191
1192 // Pull SP back to minimum size: this avoids holes in the stack
1193 __ andr(sp, esp, -16);
1194
1195 // initialize fixed part of activation frame
1196 generate_fixed_frame(true);
1197
1198 // make sure method is native & not abstract
1199 #ifdef ASSERT
1200 __ ldrw(r0, access_flags);
1201 {
1202 Label L;
1203 __ tst(r0, JVM_ACC_NATIVE);
1204 __ br(Assembler::NE, L);
1205 __ stop("tried to execute non-native method as native");
1206 __ bind(L);
1207 }
1208 {
1209 Label L;
1210 __ tst(r0, JVM_ACC_ABSTRACT);
1211 __ br(Assembler::EQ, L);
1212 __ stop("tried to execute abstract method in interpreter");
1213 __ bind(L);
1214 }
1215 #endif
1216
1217 // Since at this point in the method invocation the exception
1218 // handler would try to exit the monitor of synchronized methods
1219 // which hasn't been entered yet, we set the thread local variable
1220 // _do_not_unlock_if_synchronized to true. The remove_activation
1221 // will check this flag.
1222
1223 const Address do_not_unlock_if_synchronized(rthread,
1224 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1225 __ mov(rscratch2, true);
1226 __ strb(rscratch2, do_not_unlock_if_synchronized);
1227
1228 // increment invocation count & check for overflow
1229 Label invocation_counter_overflow;
1230 if (inc_counter) {
1231 generate_counter_incr(&invocation_counter_overflow);
1232 }
1233
1234 Label continue_after_compile;
1235 __ bind(continue_after_compile);
1236
1237 bang_stack_shadow_pages(true);
1238
1239 // reset the _do_not_unlock_if_synchronized flag
1240 __ strb(zr, do_not_unlock_if_synchronized);
1241
1242 // check for synchronized methods
1243 // Must happen AFTER invocation_counter check and stack overflow check,
1244 // so method is not locked if overflows.
1245 if (synchronized) {
1246 lock_method();
1247 } else {
1248 // no synchronization necessary
1249 #ifdef ASSERT
1250 {
1251 Label L;
1252 __ ldrw(r0, access_flags);
1253 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1254 __ br(Assembler::EQ, L);
1255 __ stop("method needs synchronization");
1256 __ bind(L);
1257 }
1258 #endif
1259 }
1260
1261 // start execution
1262 #ifdef ASSERT
1263 {
1264 Label L;
1265 const Address monitor_block_top(rfp,
1266 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1267 __ ldr(rscratch1, monitor_block_top);
1268 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1269 __ cmp(esp, rscratch1);
1270 __ br(Assembler::EQ, L);
1271 __ stop("broken stack frame setup in interpreter 1");
1272 __ bind(L);
1273 }
1274 #endif
1275
1276 // jvmti support
1277 __ notify_method_entry();
1278
1279 // work registers
1280 const Register t = r17;
1281 const Register result_handler = r19;
1282
1283 // allocate space for parameters
1284 __ ldr(t, Address(rmethod, Method::const_offset()));
1285 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1286
1287 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1288 __ andr(sp, rscratch1, -16);
1289 __ mov(esp, rscratch1);
1290
1291 // get signature handler
1292 {
1293 Label L;
1294 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1295 __ cbnz(t, L);
1296 __ call_VM(noreg,
1297 CAST_FROM_FN_PTR(address,
1298 InterpreterRuntime::prepare_native_call),
1299 rmethod);
1300 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1301 __ bind(L);
1302 }
1303
1304 // call signature handler
1305 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1306 "adjust this code");
1307 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1308 "adjust this code");
1309 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1310 "adjust this code");
1311
1312 // The generated handlers do not touch rmethod (the method).
1313 // However, large signatures cannot be cached and are generated
1314 // each time here. The slow-path generator can do a GC on return,
1315 // so we must reload it after the call.
1316 __ blr(t);
1317 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1318
1319
1320 // result handler is in r0
1321 // set result handler
1322 __ mov(result_handler, r0);
1323 // pass mirror handle if static call
1324 {
1325 Label L;
1326 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1327 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1328 // get mirror
1329 __ load_mirror(t, rmethod, r10, rscratch2);
1330 // copy mirror into activation frame
1331 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1332 // pass handle to mirror
1333 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1334 __ bind(L);
1335 }
1336
1337 // get native function entry point in r10
1338 {
1339 Label L;
1340 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1341 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1342 __ lea(rscratch2, unsatisfied);
1343 __ ldr(rscratch2, rscratch2);
1344 __ cmp(r10, rscratch2);
1345 __ br(Assembler::NE, L);
1346 __ call_VM(noreg,
1347 CAST_FROM_FN_PTR(address,
1348 InterpreterRuntime::prepare_native_call),
1349 rmethod);
1350 __ get_method(rmethod);
1351 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1352 __ bind(L);
1353 }
1354
1355 // pass JNIEnv
1356 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1357
1358 // Set the last Java PC in the frame anchor to be the return address from
1359 // the call to the native method: this will allow the debugger to
1360 // generate an accurate stack trace.
1361 Label native_return;
1362 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1363
1364 // change thread state
1365 #ifdef ASSERT
1366 {
1367 Label L;
1368 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1369 __ cmp(t, (u1)_thread_in_Java);
1370 __ br(Assembler::EQ, L);
1371 __ stop("Wrong thread state in native stub");
1372 __ bind(L);
1373 }
1374 #endif
1375
1376 // Change state to native
1377 __ mov(rscratch1, _thread_in_native);
1378 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1379 __ stlrw(rscratch1, rscratch2);
1380
1381 // Call the native method.
1382 __ blr(r10);
1383 __ bind(native_return);
1384 __ get_method(rmethod);
1385 // result potentially in r0 or v0
1386
1387 // Restore cpu control state after JNI call
1388 __ restore_cpu_control_state_after_jni(rscratch1, rscratch2);
1389
1390 // make room for the pushes we're about to do
1391 __ sub(rscratch1, esp, 4 * wordSize);
1392 __ andr(sp, rscratch1, -16);
1393
1394 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1395 // in order to extract the result of a method call. If the order of these
1396 // pushes change or anything else is added to the stack then the code in
1397 // interpreter_frame_result must also change.
1398 __ push(dtos);
1399 __ push(ltos);
1400
1401 __ verify_sve_vector_length();
1402
1403 // change thread state
1404 __ mov(rscratch1, _thread_in_native_trans);
1405 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1406 __ stlrw(rscratch1, rscratch2);
1407
1408 // Force this write out before the read below
1409 if (!UseSystemMemoryBarrier) {
1410 __ dmb(Assembler::ISH);
1411 }
1412
1413 // check for safepoint operation in progress and/or pending suspend requests
1414 {
1415 Label L, Continue;
1416
1417 // No need for acquire as Java threads always disarm themselves.
1418 __ safepoint_poll(L, true /* at_return */, false /* acquire */, false /* in_nmethod */);
1419 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1420 __ cbz(rscratch2, Continue);
1421 __ bind(L);
1422
1423 // Don't use call_VM as it will see a possible pending exception
1424 // and forward it and never return here preventing us from
1425 // clearing _last_native_pc down below. So we do a runtime call by
1426 // hand.
1427 //
1428 __ mov(c_rarg0, rthread);
1429 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1430 __ blr(rscratch2);
1431 __ get_method(rmethod);
1432 __ reinit_heapbase();
1433 __ bind(Continue);
1434 }
1435
1436 // change thread state
1437 __ mov(rscratch1, _thread_in_Java);
1438 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1439 __ stlrw(rscratch1, rscratch2);
1440
1441 // reset_last_Java_frame
1442 __ reset_last_Java_frame(true);
1443
1444 if (CheckJNICalls) {
1445 // clear_pending_jni_exception_check
1446 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1447 }
1448
1449 // reset handle block
1450 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1451 __ str(zr, Address(t, JNIHandleBlock::top_offset()));
1452
1453 // If result is an oop unbox and store it in frame where gc will see it
1454 // and result handler will pick it up
1455
1456 {
1457 Label no_oop;
1458 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1459 __ cmp(t, result_handler);
1460 __ br(Assembler::NE, no_oop);
1461 // Unbox oop result, e.g. JNIHandles::resolve result.
1462 __ pop(ltos);
1463 __ resolve_jobject(r0, t, rscratch2);
1464 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1465 // keep stack depth as expected by pushing oop which will eventually be discarded
1466 __ push(ltos);
1467 __ bind(no_oop);
1468 }
1469
1470 {
1471 Label no_reguard;
1472 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1473 __ ldrw(rscratch1, Address(rscratch1));
1474 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1475 __ br(Assembler::NE, no_reguard);
1476
1477 __ push_call_clobbered_registers();
1478 __ mov(c_rarg0, rthread);
1479 __ lea(rscratch2, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1480 __ blr(rscratch2);
1481 __ pop_call_clobbered_registers();
1482
1483 __ bind(no_reguard);
1484 }
1485
1486 // The method register is junk from after the thread_in_native transition
1487 // until here. Also can't call_VM until the bcp has been
1488 // restored. Need bcp for throwing exception below so get it now.
1489 __ get_method(rmethod);
1490
1491 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1492 // rbcp == code_base()
1493 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1494 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1495 // handle exceptions (exception handling will handle unlocking!)
1496 {
1497 Label L;
1498 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1499 __ cbz(rscratch1, L);
1500 // Note: At some point we may want to unify this with the code
1501 // used in call_VM_base(); i.e., we should use the
1502 // StubRoutines::forward_exception code. For now this doesn't work
1503 // here because the rsp is not correctly set at this point.
1504 __ MacroAssembler::call_VM(noreg,
1505 CAST_FROM_FN_PTR(address,
1506 InterpreterRuntime::throw_pending_exception));
1507 __ should_not_reach_here();
1508 __ bind(L);
1509 }
1510
1511 // do unlocking if necessary
1512 {
1513 Label L;
1514 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1515 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1516 // the code below should be shared with interpreter macro
1517 // assembler implementation
1518 {
1519 Label unlock;
1520 // BasicObjectLock will be first in list, since this is a
1521 // synchronized method. However, need to check that the object
1522 // has not been unlocked by an explicit monitorexit bytecode.
1523
1524 // monitor expect in c_rarg1 for slow unlock path
1525 __ lea (c_rarg1, Address(rfp, // address of first monitor
1526 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1527 wordSize - sizeof(BasicObjectLock))));
1528
1529 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset()));
1530 __ cbnz(t, unlock);
1531
1532 // Entry already unlocked, need to throw exception
1533 __ MacroAssembler::call_VM(noreg,
1534 CAST_FROM_FN_PTR(address,
1535 InterpreterRuntime::throw_illegal_monitor_state_exception));
1536 __ should_not_reach_here();
1537
1538 __ bind(unlock);
1539 __ unlock_object(c_rarg1);
1540 }
1541 __ bind(L);
1542 }
1543
1544 // jvmti support
1545 // Note: This must happen _after_ handling/throwing any exceptions since
1546 // the exception handler code notifies the runtime of method exits
1547 // too. If this happens before, method entry/exit notifications are
1548 // not properly paired (was bug - gri 11/22/99).
1549 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1550
1551 // restore potential result in r0:d0, call result handler to
1552 // restore potential result in ST0 & handle result
1553
1554 __ pop(ltos);
1555 __ pop(dtos);
1556
1557 __ blr(result_handler);
1558
1559 // remove activation
1560 __ ldr(esp, Address(rfp,
1561 frame::interpreter_frame_sender_sp_offset *
1562 wordSize)); // get sender sp
1563 // remove frame anchor
1564 __ leave();
1565
1566 // restore sender sp
1567 __ mov(sp, esp);
1568
1569 __ ret(lr);
1570
1571 if (inc_counter) {
1572 // Handle overflow of counter and compile method
1573 __ bind(invocation_counter_overflow);
1574 generate_counter_overflow(continue_after_compile);
1575 }
1576
1577 return entry_point;
1578 }
1579
1580 //
1581 // Generic interpreted method entry to (asm) interpreter
1582 //
1583 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1584 // determine code generation flags
1585 bool inc_counter = UseCompiler || CountCompiledCalls;
1586
1587 // rscratch1: sender sp
1588 address entry_point = __ pc();
1589
1590 const Address constMethod(rmethod, Method::const_offset());
1591 const Address access_flags(rmethod, Method::access_flags_offset());
1592 const Address size_of_parameters(r3,
1593 ConstMethod::size_of_parameters_offset());
1594 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1595
1596 // get parameter size (always needed)
1597 // need to load the const method first
1598 __ ldr(r3, constMethod);
1599 __ load_unsigned_short(r2, size_of_parameters);
1600
1601 // r2: size of parameters
1602
1603 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1604 __ sub(r3, r3, r2); // r3 = no. of additional locals
1605
1606 // see if we've got enough room on the stack for locals plus overhead.
1607 generate_stack_overflow_check();
1608
1609 // compute beginning of parameters (rlocals)
1610 __ add(rlocals, esp, r2, ext::uxtx, 3);
1611 __ sub(rlocals, rlocals, wordSize);
1612
1613 __ mov(rscratch1, esp);
1614
1615 // r3 - # of additional locals
1616 // allocate space for locals
1617 // explicitly initialize locals
1618 // Initializing memory allocated for locals in the same direction as
1619 // the stack grows to ensure page initialization order according
1620 // to windows-aarch64 stack page growth requirement (see
1621 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1622 {
1623 Label exit, loop;
1624 __ ands(zr, r3, r3);
1625 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1626 __ bind(loop);
1627 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1628 __ sub(r3, r3, 1); // until everything initialized
1629 __ cbnz(r3, loop);
1630 __ bind(exit);
1631 }
1632
1633 // Padding between locals and fixed part of activation frame to ensure
1634 // SP is always 16-byte aligned.
1635 __ andr(sp, rscratch1, -16);
1636
1637 // And the base dispatch table
1638 __ get_dispatch();
1639
1640 // initialize fixed part of activation frame
1641 generate_fixed_frame(false);
1642
1643 // make sure method is not native & not abstract
1644 #ifdef ASSERT
1645 __ ldrw(r0, access_flags);
1646 {
1647 Label L;
1648 __ tst(r0, JVM_ACC_NATIVE);
1649 __ br(Assembler::EQ, L);
1650 __ stop("tried to execute native method as non-native");
1651 __ bind(L);
1652 }
1653 {
1654 Label L;
1655 __ tst(r0, JVM_ACC_ABSTRACT);
1656 __ br(Assembler::EQ, L);
1657 __ stop("tried to execute abstract method in interpreter");
1658 __ bind(L);
1659 }
1660 #endif
1661
1662 // Since at this point in the method invocation the exception
1663 // handler would try to exit the monitor of synchronized methods
1664 // which hasn't been entered yet, we set the thread local variable
1665 // _do_not_unlock_if_synchronized to true. The remove_activation
1666 // will check this flag.
1667
1668 const Address do_not_unlock_if_synchronized(rthread,
1669 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1670 __ mov(rscratch2, true);
1671 __ strb(rscratch2, do_not_unlock_if_synchronized);
1672
1673 Register mdp = r3;
1674 __ profile_parameters_type(mdp, r1, r2);
1675
1676 // increment invocation count & check for overflow
1677 Label invocation_counter_overflow;
1678 if (inc_counter) {
1679 generate_counter_incr(&invocation_counter_overflow);
1680 }
1681
1682 Label continue_after_compile;
1683 __ bind(continue_after_compile);
1684
1685 bang_stack_shadow_pages(false);
1686
1687 // reset the _do_not_unlock_if_synchronized flag
1688 __ strb(zr, do_not_unlock_if_synchronized);
1689
1690 // check for synchronized methods
1691 // Must happen AFTER invocation_counter check and stack overflow check,
1692 // so method is not locked if overflows.
1693 if (synchronized) {
1694 // Allocate monitor and lock method
1695 lock_method();
1696 } else {
1697 // no synchronization necessary
1698 #ifdef ASSERT
1699 {
1700 Label L;
1701 __ ldrw(r0, access_flags);
1702 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1703 __ br(Assembler::EQ, L);
1704 __ stop("method needs synchronization");
1705 __ bind(L);
1706 }
1707 #endif
1708 }
1709
1710 // start execution
1711 #ifdef ASSERT
1712 {
1713 Label L;
1714 const Address monitor_block_top (rfp,
1715 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1716 __ ldr(rscratch1, monitor_block_top);
1717 __ lea(rscratch1, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1718 __ cmp(esp, rscratch1);
1719 __ br(Assembler::EQ, L);
1720 __ stop("broken stack frame setup in interpreter 2");
1721 __ bind(L);
1722 }
1723 #endif
1724
1725 // jvmti support
1726 __ notify_method_entry();
1727
1728 __ dispatch_next(vtos);
1729
1730 // invocation counter overflow
1731 if (inc_counter) {
1732 // Handle overflow of counter and compile method
1733 __ bind(invocation_counter_overflow);
1734 generate_counter_overflow(continue_after_compile);
1735 }
1736
1737 return entry_point;
1738 }
1739
1740 // Method entry for java.lang.Thread.currentThread
1741 address TemplateInterpreterGenerator::generate_currentThread() {
1742 address entry_point = __ pc();
1743
1744 __ ldr(r0, Address(rthread, JavaThread::vthread_offset()));
1745 __ resolve_oop_handle(r0, rscratch1, rscratch2);
1746 __ ret(lr);
1747
1748 return entry_point;
1749 }
1750
1751 // Not supported
1752 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { return nullptr; }
1753 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { return nullptr; }
1754 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { return nullptr; }
1755 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { return nullptr; }
1756
1757 //-----------------------------------------------------------------------------
1758 // Exceptions
1759
1760 void TemplateInterpreterGenerator::generate_throw_exception() {
1761 // Entry point in previous activation (i.e., if the caller was
1762 // interpreted)
1763 Interpreter::_rethrow_exception_entry = __ pc();
1764 // Restore sp to interpreter_frame_last_sp even though we are going
1765 // to empty the expression stack for the exception processing.
1766 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1767 // r0: exception
1768 // r3: return address/pc that threw exception
1769 __ restore_bcp(); // rbcp points to call/send
1770 __ restore_locals();
1771 __ restore_constant_pool_cache();
1772 __ reinit_heapbase(); // restore rheapbase as heapbase.
1773 __ get_dispatch();
1774
1775 // Entry point for exceptions thrown within interpreter code
1776 Interpreter::_throw_exception_entry = __ pc();
1777 // If we came here via a NullPointerException on the receiver of a
1778 // method, rmethod may be corrupt.
1779 __ get_method(rmethod);
1780 // expression stack is undefined here
1781 // r0: exception
1782 // rbcp: exception bcp
1783 __ verify_oop(r0);
1784 __ mov(c_rarg1, r0);
1785
1786 // expression stack must be empty before entering the VM in case of
1787 // an exception
1788 __ empty_expression_stack();
1789 // find exception handler address and preserve exception oop
1790 __ call_VM(r3,
1791 CAST_FROM_FN_PTR(address,
1792 InterpreterRuntime::exception_handler_for_exception),
1793 c_rarg1);
1794
1795 // Restore machine SP
1796 __ restore_sp_after_call();
1797
1798 // r0: exception handler entry point
1799 // r3: preserved exception oop
1800 // rbcp: bcp for exception handler
1801 __ push_ptr(r3); // push exception which is now the only value on the stack
1802 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1803
1804 // If the exception is not handled in the current frame the frame is
1805 // removed and the exception is rethrown (i.e. exception
1806 // continuation is _rethrow_exception).
1807 //
1808 // Note: At this point the bci is still the bxi for the instruction
1809 // which caused the exception and the expression stack is
1810 // empty. Thus, for any VM calls at this point, GC will find a legal
1811 // oop map (with empty expression stack).
1812
1813 //
1814 // JVMTI PopFrame support
1815 //
1816
1817 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1818 __ empty_expression_stack();
1819 // Set the popframe_processing bit in pending_popframe_condition
1820 // indicating that we are currently handling popframe, so that
1821 // call_VMs that may happen later do not trigger new popframe
1822 // handling cycles.
1823 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1824 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1825 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1826
1827 {
1828 // Check to see whether we are returning to a deoptimized frame.
1829 // (The PopFrame call ensures that the caller of the popped frame is
1830 // either interpreted or compiled and deoptimizes it if compiled.)
1831 // In this case, we can't call dispatch_next() after the frame is
1832 // popped, but instead must save the incoming arguments and restore
1833 // them after deoptimization has occurred.
1834 //
1835 // Note that we don't compare the return PC against the
1836 // deoptimization blob's unpack entry because of the presence of
1837 // adapter frames in C2.
1838 Label caller_not_deoptimized;
1839 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1840 // This is a return address, so requires authenticating for PAC.
1841 __ authenticate_return_address(c_rarg1);
1842 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1843 InterpreterRuntime::interpreter_contains), c_rarg1);
1844 __ cbnz(r0, caller_not_deoptimized);
1845
1846 // Compute size of arguments for saving when returning to
1847 // deoptimized caller
1848 __ get_method(r0);
1849 __ ldr(r0, Address(r0, Method::const_offset()));
1850 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1851 size_of_parameters_offset())));
1852 __ lsl(r0, r0, Interpreter::logStackElementSize);
1853 __ restore_locals(); // XXX do we need this?
1854 __ sub(rlocals, rlocals, r0);
1855 __ add(rlocals, rlocals, wordSize);
1856 // Save these arguments
1857 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1858 Deoptimization::
1859 popframe_preserve_args),
1860 rthread, r0, rlocals);
1861
1862 __ remove_activation(vtos,
1863 /* throw_monitor_exception */ false,
1864 /* install_monitor_exception */ false,
1865 /* notify_jvmdi */ false);
1866
1867 // Inform deoptimization that it is responsible for restoring
1868 // these arguments
1869 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1870 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1871
1872 // Continue in deoptimization handler
1873 __ ret(lr);
1874
1875 __ bind(caller_not_deoptimized);
1876 }
1877
1878 __ remove_activation(vtos,
1879 /* throw_monitor_exception */ false,
1880 /* install_monitor_exception */ false,
1881 /* notify_jvmdi */ false);
1882
1883 // Restore the last_sp and null it out
1884 __ ldr(rscratch1, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1885 __ lea(esp, Address(rfp, rscratch1, Address::lsl(Interpreter::logStackElementSize)));
1886 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1887
1888 __ restore_bcp();
1889 __ restore_locals();
1890 __ restore_constant_pool_cache();
1891 __ get_method(rmethod);
1892 __ get_dispatch();
1893
1894 // The method data pointer was incremented already during
1895 // call profiling. We have to restore the mdp for the current bcp.
1896 if (ProfileInterpreter) {
1897 __ set_method_data_pointer_for_bcp();
1898 }
1899
1900 // Clear the popframe condition flag
1901 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1902 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1903
1904 #if INCLUDE_JVMTI
1905 {
1906 Label L_done;
1907
1908 __ ldrb(rscratch1, Address(rbcp, 0));
1909 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1910 __ br(Assembler::NE, L_done);
1911
1912 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1913 // Detect such a case in the InterpreterRuntime function and return the member name argument, or null.
1914
1915 __ ldr(c_rarg0, Address(rlocals, 0));
1916 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1917
1918 __ cbz(r0, L_done);
1919
1920 __ str(r0, Address(esp, 0));
1921 __ bind(L_done);
1922 }
1923 #endif // INCLUDE_JVMTI
1924
1925 // Restore machine SP
1926 __ restore_sp_after_call();
1927
1928 __ dispatch_next(vtos);
1929 // end of PopFrame support
1930
1931 Interpreter::_remove_activation_entry = __ pc();
1932
1933 // preserve exception over this code sequence
1934 __ pop_ptr(r0);
1935 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1936 // remove the activation (without doing throws on illegalMonitorExceptions)
1937 __ remove_activation(vtos, false, true, false);
1938 // restore exception
1939 __ get_vm_result(r0, rthread);
1940
1941 // In between activations - previous activation type unknown yet
1942 // compute continuation point - the continuation point expects the
1943 // following registers set up:
1944 //
1945 // r0: exception
1946 // lr: return address/pc that threw exception
1947 // esp: expression stack of caller
1948 // rfp: fp of caller
1949 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1950 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1951 SharedRuntime::exception_handler_for_return_address),
1952 rthread, lr);
1953 __ mov(r1, r0); // save exception handler
1954 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1955 // We might be returning to a deopt handler that expects r3 to
1956 // contain the exception pc
1957 __ mov(r3, lr);
1958 // Note that an "issuing PC" is actually the next PC after the call
1959 __ br(r1); // jump to exception
1960 // handler of caller
1961 }
1962
1963
1964 //
1965 // JVMTI ForceEarlyReturn support
1966 //
1967 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1968 address entry = __ pc();
1969
1970 __ restore_bcp();
1971 __ restore_locals();
1972 __ empty_expression_stack();
1973 __ load_earlyret_value(state);
1974
1975 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1976 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1977
1978 // Clear the earlyret state
1979 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1980 __ str(zr, cond_addr);
1981
1982 __ remove_activation(state,
1983 false, /* throw_monitor_exception */
1984 false, /* install_monitor_exception */
1985 true); /* notify_jvmdi */
1986 __ ret(lr);
1987
1988 return entry;
1989 } // end of ForceEarlyReturn support
1990
1991
1992
1993 //-----------------------------------------------------------------------------
1994 // Helper for vtos entry point generation
1995
1996 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1997 address& bep,
1998 address& cep,
1999 address& sep,
2000 address& aep,
2001 address& iep,
2002 address& lep,
2003 address& fep,
2004 address& dep,
2005 address& vep) {
2006 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2007 Label L;
2008 aep = __ pc(); // atos entry point
2009 __ push_ptr();
2010 __ b(L);
2011 fep = __ pc(); // ftos entry point
2012 __ push_f();
2013 __ b(L);
2014 dep = __ pc(); // dtos entry point
2015 __ push_d();
2016 __ b(L);
2017 lep = __ pc(); // ltos entry point
2018 __ push_l();
2019 __ b(L);
2020 bep = cep = sep = iep = __ pc(); // [bcsi]tos entry point
2021 __ push_i();
2022 vep = __ pc(); // vtos entry point
2023 __ bind(L);
2024 generate_and_dispatch(t);
2025 }
2026
2027 //-----------------------------------------------------------------------------
2028
2029 // Non-product code
2030 #ifndef PRODUCT
2031 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2032 address entry = __ pc();
2033
2034 __ protect_return_address();
2035 __ push(lr);
2036 __ push(state);
2037 __ push(RegSet::range(r0, r15), sp);
2038 __ mov(c_rarg2, r0); // Pass itos
2039 __ call_VM(noreg,
2040 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2041 c_rarg1, c_rarg2, c_rarg3);
2042 __ pop(RegSet::range(r0, r15), sp);
2043 __ pop(state);
2044 __ pop(lr);
2045 __ authenticate_return_address();
2046 __ ret(lr); // return from result handler
2047
2048 return entry;
2049 }
2050
2051 void TemplateInterpreterGenerator::count_bytecode() {
2052 __ mov(r10, (address) &BytecodeCounter::_counter_value);
2053 __ atomic_addw(noreg, 1, r10);
2054 }
2055
2056 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2057 __ mov(r10, (address) &BytecodeHistogram::_counters[t->bytecode()]);
2058 __ atomic_addw(noreg, 1, r10);
2059 }
2060
2061 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2062 // Calculate new index for counter:
2063 // _index = (_index >> log2_number_of_codes) |
2064 // (bytecode << log2_number_of_codes);
2065 Register index_addr = rscratch1;
2066 Register index = rscratch2;
2067 __ mov(index_addr, (address) &BytecodePairHistogram::_index);
2068 __ ldrw(index, index_addr);
2069 __ mov(r10,
2070 ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2071 __ orrw(index, r10, index, Assembler::LSR,
2072 BytecodePairHistogram::log2_number_of_codes);
2073 __ strw(index, index_addr);
2074
2075 // Bump bucket contents:
2076 // _counters[_index] ++;
2077 Register counter_addr = rscratch1;
2078 __ mov(r10, (address) &BytecodePairHistogram::_counters);
2079 __ lea(counter_addr, Address(r10, index, Address::lsl(LogBytesPerInt)));
2080 __ atomic_addw(noreg, 1, counter_addr);
2081 }
2082
2083 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2084 // Call a little run-time stub to avoid blow-up for each bytecode.
2085 // The run-time runtime saves the right registers, depending on
2086 // the tosca in-state for the given template.
2087
2088 assert(Interpreter::trace_code(t->tos_in()) != nullptr,
2089 "entry must have been generated");
2090 __ bl(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2091 __ reinit_heapbase();
2092 }
2093
2094
2095 void TemplateInterpreterGenerator::stop_interpreter_at() {
2096 Label L;
2097 __ push(rscratch1);
2098 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2099 __ ldr(rscratch1, Address(rscratch1));
2100 __ mov(rscratch2, StopInterpreterAt);
2101 __ cmpw(rscratch1, rscratch2);
2102 __ br(Assembler::NE, L);
2103 __ brk(0);
2104 __ bind(L);
2105 __ pop(rscratch1);
2106 }
2107
2108 #endif // !PRODUCT