1 /*
2 * Copyright (c) 2003, 2020, 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 "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/bytecodeHistogram.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "interpreter/interpreterRuntime.hpp"
32 #include "interpreter/interp_masm.hpp"
33 #include "interpreter/templateInterpreterGenerator.hpp"
34 #include "interpreter/templateTable.hpp"
35 #include "interpreter/bytecodeTracer.hpp"
36 #include "memory/resourceArea.hpp"
37 #include "oops/arrayOop.hpp"
38 #include "oops/methodData.hpp"
39 #include "oops/method.hpp"
40 #include "oops/oop.inline.hpp"
41 #include "prims/jvmtiExport.hpp"
42 #include "prims/jvmtiThreadState.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/deoptimization.hpp"
45 #include "runtime/frame.inline.hpp"
46 #include "runtime/sharedRuntime.hpp"
47 #include "runtime/stubRoutines.hpp"
48 #include "runtime/synchronizer.hpp"
49 #include "runtime/timer.hpp"
50 #include "runtime/vframeArray.hpp"
51 #include "utilities/debug.hpp"
52 #include "utilities/macros.hpp"
53 #include <sys/types.h>
54
55 #ifndef PRODUCT
56 #include "oops/method.hpp"
57 #endif // !PRODUCT
58
59 // Size of interpreter code. Increase if too small. Interpreter will
60 // fail with a guarantee ("not enough space for interpreter generation");
61 // if too small.
62 // Run with +PrintInterpreter to get the VM to print out the size.
63 // Max size with JVMTI
64 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
65
66 #define __ _masm->
67
68 //-----------------------------------------------------------------------------
69
70 extern "C" void entry(CodeBuffer*);
71
72 //-----------------------------------------------------------------------------
73
generate_slow_signature_handler()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
generate_math_entry(AbstractInterpreter::MethodKind kind)144 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
145 // rmethod: Method*
146 // r13: sender sp
147 // esp: args
148
149 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
150
151 // These don't need a safepoint check because they aren't virtually
152 // callable. We won't enter these intrinsics from compiled code.
153 // If in the future we added an intrinsic which was virtually callable
154 // we'd have to worry about how to safepoint so that this code is used.
155
156 // mathematical functions inlined by compiler
157 // (interpreter must provide identical implementation
158 // in order to avoid monotonicity bugs when switching
159 // from interpreter to compiler in the middle of some
160 // computation)
161 //
162 // stack:
163 // [ arg ] <-- esp
164 // [ arg ]
165 // retaddr in lr
166
167 address entry_point = NULL;
168 Register continuation = lr;
169 switch (kind) {
170 case Interpreter::java_lang_math_abs:
171 entry_point = __ pc();
172 __ ldrd(v0, Address(esp));
173 __ fabsd(v0, v0);
174 __ mov(sp, r13); // Restore caller's SP
175 break;
176 case Interpreter::java_lang_math_sqrt:
177 entry_point = __ pc();
178 __ ldrd(v0, Address(esp));
179 __ fsqrtd(v0, v0);
180 __ mov(sp, r13);
181 break;
182 case Interpreter::java_lang_math_sin :
183 case Interpreter::java_lang_math_cos :
184 case Interpreter::java_lang_math_tan :
185 case Interpreter::java_lang_math_log :
186 case Interpreter::java_lang_math_log10 :
187 case Interpreter::java_lang_math_exp :
188 entry_point = __ pc();
189 __ ldrd(v0, Address(esp));
190 __ mov(sp, r13);
191 __ mov(r19, lr);
192 continuation = r19; // The first callee-saved register
193 generate_transcendental_entry(kind, 1);
194 break;
195 case Interpreter::java_lang_math_pow :
196 entry_point = __ pc();
197 __ mov(r19, lr);
198 continuation = r19;
199 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
200 __ ldrd(v1, Address(esp));
201 __ mov(sp, r13);
202 generate_transcendental_entry(kind, 2);
203 break;
204 case Interpreter::java_lang_math_fmaD :
205 if (UseFMA) {
206 entry_point = __ pc();
207 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
208 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
209 __ ldrd(v2, Address(esp));
210 __ fmaddd(v0, v0, v1, v2);
211 __ mov(sp, r13); // Restore caller's SP
212 }
213 break;
214 case Interpreter::java_lang_math_fmaF :
215 if (UseFMA) {
216 entry_point = __ pc();
217 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
218 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
219 __ ldrs(v2, Address(esp));
220 __ fmadds(v0, v0, v1, v2);
221 __ mov(sp, r13); // Restore caller's SP
222 }
223 break;
224 default:
225 ;
226 }
227 if (entry_point) {
228 __ br(continuation);
229 }
230
231 return entry_point;
232 }
233
234 // double trigonometrics and transcendentals
235 // static jdouble dsin(jdouble x);
236 // static jdouble dcos(jdouble x);
237 // static jdouble dtan(jdouble x);
238 // static jdouble dlog(jdouble x);
239 // static jdouble dlog10(jdouble x);
240 // static jdouble dexp(jdouble x);
241 // static jdouble dpow(jdouble x, jdouble y);
242
generate_transcendental_entry(AbstractInterpreter::MethodKind kind,int fpargs)243 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
244 address fn;
245 switch (kind) {
246 case Interpreter::java_lang_math_sin :
247 if (StubRoutines::dsin() == NULL) {
248 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
249 } else {
250 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
251 }
252 break;
253 case Interpreter::java_lang_math_cos :
254 if (StubRoutines::dcos() == NULL) {
255 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
256 } else {
257 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
258 }
259 break;
260 case Interpreter::java_lang_math_tan :
261 if (StubRoutines::dtan() == NULL) {
262 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
263 } else {
264 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
265 }
266 break;
267 case Interpreter::java_lang_math_log :
268 if (StubRoutines::dlog() == NULL) {
269 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
270 } else {
271 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
272 }
273 break;
274 case Interpreter::java_lang_math_log10 :
275 if (StubRoutines::dlog10() == NULL) {
276 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
277 } else {
278 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
279 }
280 break;
281 case Interpreter::java_lang_math_exp :
282 if (StubRoutines::dexp() == NULL) {
283 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
284 } else {
285 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
286 }
287 break;
288 case Interpreter::java_lang_math_pow :
289 fpargs = 2;
290 if (StubRoutines::dpow() == NULL) {
291 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
292 } else {
293 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
294 }
295 break;
296 default:
297 ShouldNotReachHere();
298 fn = NULL; // unreachable
299 }
300 __ mov(rscratch1, fn);
301 __ blr(rscratch1);
302 }
303
304 // Abstract method entry
305 // Attempt to execute abstract method. Throw exception
generate_abstract_entry(void)306 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
307 // rmethod: Method*
308 // r13: sender SP
309
310 address entry_point = __ pc();
311
312 // abstract method entry
313
314 // pop return address, reset last_sp to NULL
315 __ empty_expression_stack();
316 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
317 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
318
319 // throw exception
320 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
321 InterpreterRuntime::throw_AbstractMethodErrorWithMethod),
322 rmethod);
323 // the call_VM checks for exception, so we should never return here.
324 __ should_not_reach_here();
325
326 return entry_point;
327 }
328
generate_StackOverflowError_handler()329 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
330 address entry = __ pc();
331
332 #ifdef ASSERT
333 {
334 Label L;
335 __ ldr(rscratch1, Address(rfp,
336 frame::interpreter_frame_monitor_block_top_offset *
337 wordSize));
338 __ mov(rscratch2, sp);
339 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
340 // grows negative)
341 __ br(Assembler::HS, L); // check if frame is complete
342 __ stop ("interpreter frame not set up");
343 __ bind(L);
344 }
345 #endif // ASSERT
346 // Restore bcp under the assumption that the current frame is still
347 // interpreted
348 __ restore_bcp();
349
350 // expression stack must be empty before entering the VM if an
351 // exception happened
352 __ empty_expression_stack();
353 // throw exception
354 __ call_VM(noreg,
355 CAST_FROM_FN_PTR(address,
356 InterpreterRuntime::throw_StackOverflowError));
357 return entry;
358 }
359
generate_ArrayIndexOutOfBounds_handler()360 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler() {
361 address entry = __ pc();
362 // expression stack must be empty before entering the VM if an
363 // exception happened
364 __ empty_expression_stack();
365 // setup parameters
366
367 // ??? convention: expect aberrant index in register r1
368 __ movw(c_rarg2, r1);
369 // ??? convention: expect array in register r3
370 __ mov(c_rarg1, r3);
371 __ call_VM(noreg,
372 CAST_FROM_FN_PTR(address,
373 InterpreterRuntime::
374 throw_ArrayIndexOutOfBoundsException),
375 c_rarg1, c_rarg2);
376 return entry;
377 }
378
generate_ClassCastException_handler()379 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
380 address entry = __ pc();
381
382 // object is at TOS
383 __ pop(c_rarg1);
384
385 // expression stack must be empty before entering the VM if an
386 // exception happened
387 __ empty_expression_stack();
388
389 __ call_VM(noreg,
390 CAST_FROM_FN_PTR(address,
391 InterpreterRuntime::
392 throw_ClassCastException),
393 c_rarg1);
394 return entry;
395 }
396
generate_exception_handler_common(const char * name,const char * message,bool pass_oop)397 address TemplateInterpreterGenerator::generate_exception_handler_common(
398 const char* name, const char* message, bool pass_oop) {
399 assert(!pass_oop || message == NULL, "either oop or message but not both");
400 address entry = __ pc();
401 if (pass_oop) {
402 // object is at TOS
403 __ pop(c_rarg2);
404 }
405 // expression stack must be empty before entering the VM if an
406 // exception happened
407 __ empty_expression_stack();
408 // setup parameters
409 __ lea(c_rarg1, Address((address)name));
410 if (pass_oop) {
411 __ call_VM(r0, CAST_FROM_FN_PTR(address,
412 InterpreterRuntime::
413 create_klass_exception),
414 c_rarg1, c_rarg2);
415 } else {
416 // kind of lame ExternalAddress can't take NULL because
417 // external_word_Relocation will assert.
418 if (message != NULL) {
419 __ lea(c_rarg2, Address((address)message));
420 } else {
421 __ mov(c_rarg2, NULL_WORD);
422 }
423 __ call_VM(r0,
424 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
425 c_rarg1, c_rarg2);
426 }
427 // throw exception
428 __ b(address(Interpreter::throw_exception_entry()));
429 return entry;
430 }
431
generate_return_entry_for(TosState state,int step,size_t index_size)432 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
433 address entry = __ pc();
434
435 // Restore stack bottom in case i2c adjusted stack
436 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
437 // and NULL it as marker that esp is now tos until next java call
438 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
439 __ restore_bcp();
440 __ restore_locals();
441 __ restore_constant_pool_cache();
442 __ get_method(rmethod);
443
444 if (state == atos) {
445 Register obj = r0;
446 Register mdp = r1;
447 Register tmp = r2;
448 __ profile_return_type(mdp, obj, tmp);
449 }
450
451 // Pop N words from the stack
452 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
453 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
454 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
455
456 __ add(esp, esp, r1, Assembler::LSL, 3);
457
458 // Restore machine SP
459 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
460 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
461 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
462 __ ldr(rscratch2,
463 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
464 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
465 __ andr(sp, rscratch1, -16);
466
467 __ check_and_handle_popframe(rthread);
468 __ check_and_handle_earlyret(rthread);
469
470 __ get_dispatch();
471 __ dispatch_next(state, step);
472
473 return entry;
474 }
475
generate_deopt_entry_for(TosState state,int step,address continuation)476 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
477 int step,
478 address continuation) {
479 address entry = __ pc();
480 __ restore_bcp();
481 __ restore_locals();
482 __ restore_constant_pool_cache();
483 __ get_method(rmethod);
484 __ get_dispatch();
485
486 // Calculate stack limit
487 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
488 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
489 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
490 __ ldr(rscratch2,
491 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
492 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
493 __ andr(sp, rscratch1, -16);
494
495 // Restore expression stack pointer
496 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
497 // NULL last_sp until next java call
498 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
499
500 #if INCLUDE_JVMCI
501 // Check if we need to take lock at entry of synchronized method. This can
502 // only occur on method entry so emit it only for vtos with step 0.
503 if ((EnableJVMCI || UseAOT) && state == vtos && step == 0) {
504 Label L;
505 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
506 __ cbz(rscratch1, L);
507 // Clear flag.
508 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
509 // Take lock.
510 lock_method();
511 __ bind(L);
512 } else {
513 #ifdef ASSERT
514 if (EnableJVMCI) {
515 Label L;
516 __ ldrb(rscratch1, Address(rthread, JavaThread::pending_monitorenter_offset()));
517 __ cbz(rscratch1, L);
518 __ stop("unexpected pending monitor in deopt entry");
519 __ bind(L);
520 }
521 #endif
522 }
523 #endif
524 // handle exceptions
525 {
526 Label L;
527 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
528 __ cbz(rscratch1, L);
529 __ call_VM(noreg,
530 CAST_FROM_FN_PTR(address,
531 InterpreterRuntime::throw_pending_exception));
532 __ should_not_reach_here();
533 __ bind(L);
534 }
535
536 if (continuation == NULL) {
537 __ dispatch_next(state, step);
538 } else {
539 __ jump_to_entry(continuation);
540 }
541 return entry;
542 }
543
generate_result_handler_for(BasicType type)544 address TemplateInterpreterGenerator::generate_result_handler_for(
545 BasicType type) {
546 address entry = __ pc();
547 switch (type) {
548 case T_BOOLEAN: __ c2bool(r0); break;
549 case T_CHAR : __ uxth(r0, r0); break;
550 case T_BYTE : __ sxtb(r0, r0); break;
551 case T_SHORT : __ sxth(r0, r0); break;
552 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
553 case T_LONG : /* nothing to do */ break;
554 case T_VOID : /* nothing to do */ break;
555 case T_FLOAT : /* nothing to do */ break;
556 case T_DOUBLE : /* nothing to do */ break;
557 case T_OBJECT :
558 // retrieve result from frame
559 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
560 // and verify it
561 __ verify_oop(r0);
562 break;
563 default : ShouldNotReachHere();
564 }
565 __ ret(lr); // return from result handler
566 return entry;
567 }
568
generate_safept_entry_for(TosState state,address runtime_entry)569 address TemplateInterpreterGenerator::generate_safept_entry_for(
570 TosState state,
571 address runtime_entry) {
572 address entry = __ pc();
573 __ push(state);
574 __ call_VM(noreg, runtime_entry);
575 __ membar(Assembler::AnyAny);
576 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
577 return entry;
578 }
579
580 // Helpers for commoning out cases in the various type of method entries.
581 //
582
583
584 // increment invocation count & check for overflow
585 //
586 // Note: checking for negative value instead of overflow
587 // so we have a 'sticky' overflow test
588 //
589 // rmethod: method
590 //
generate_counter_incr(Label * overflow,Label * profile_method,Label * profile_method_continue)591 void TemplateInterpreterGenerator::generate_counter_incr(
592 Label* overflow,
593 Label* profile_method,
594 Label* profile_method_continue) {
595 Label done;
596 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
597 if (TieredCompilation) {
598 int increment = InvocationCounter::count_increment;
599 Label no_mdo;
600 if (ProfileInterpreter) {
601 // Are we profiling?
602 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
603 __ cbz(r0, no_mdo);
604 // Increment counter in the MDO
605 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
606 in_bytes(InvocationCounter::counter_offset()));
607 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
608 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
609 __ b(done);
610 }
611 __ bind(no_mdo);
612 // Increment counter in MethodCounters
613 const Address invocation_counter(rscratch2,
614 MethodCounters::invocation_counter_offset() +
615 InvocationCounter::counter_offset());
616 __ get_method_counters(rmethod, rscratch2, done);
617 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
618 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
619 __ bind(done);
620 } else { // not TieredCompilation
621 const Address backedge_counter(rscratch2,
622 MethodCounters::backedge_counter_offset() +
623 InvocationCounter::counter_offset());
624 const Address invocation_counter(rscratch2,
625 MethodCounters::invocation_counter_offset() +
626 InvocationCounter::counter_offset());
627
628 __ get_method_counters(rmethod, rscratch2, done);
629
630 if (ProfileInterpreter) { // %%% Merge this into MethodData*
631 __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
632 __ addw(r1, r1, 1);
633 __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
634 }
635 // Update standard invocation counters
636 __ ldrw(r1, invocation_counter);
637 __ ldrw(r0, backedge_counter);
638
639 __ addw(r1, r1, InvocationCounter::count_increment);
640 __ andw(r0, r0, InvocationCounter::count_mask_value);
641
642 __ strw(r1, invocation_counter);
643 __ addw(r0, r0, r1); // add both counters
644
645 // profile_method is non-null only for interpreted method so
646 // profile_method != NULL == !native_call
647
648 if (ProfileInterpreter && profile_method != NULL) {
649 // Test to see if we should create a method data oop
650 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
651 __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
652 __ cmpw(r0, rscratch2);
653 __ br(Assembler::LT, *profile_method_continue);
654
655 // if no method data exists, go to profile_method
656 __ test_method_data_pointer(rscratch2, *profile_method);
657 }
658
659 {
660 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
661 __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
662 __ cmpw(r0, rscratch2);
663 __ br(Assembler::HS, *overflow);
664 }
665 __ bind(done);
666 }
667 }
668
generate_counter_overflow(Label & do_continue)669 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
670
671 // Asm interpreter on entry
672 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
673 // Everything as it was on entry
674
675 // InterpreterRuntime::frequency_counter_overflow takes two
676 // arguments, the first (thread) is passed by call_VM, the second
677 // indicates if the counter overflow occurs at a backwards branch
678 // (NULL bcp). We pass zero for it. The call returns the address
679 // of the verified entry point for the method or NULL if the
680 // compilation did not complete (either went background or bailed
681 // out).
682 __ mov(c_rarg1, 0);
683 __ call_VM(noreg,
684 CAST_FROM_FN_PTR(address,
685 InterpreterRuntime::frequency_counter_overflow),
686 c_rarg1);
687
688 __ b(do_continue);
689 }
690
691 // See if we've got enough room on the stack for locals plus overhead
692 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
693 // without going through the signal handler, i.e., reserved and yellow zones
694 // will not be made usable. The shadow zone must suffice to handle the
695 // overflow.
696 // The expression stack grows down incrementally, so the normal guard
697 // page mechanism will work for that.
698 //
699 // NOTE: Since the additional locals are also always pushed (wasn't
700 // obvious in generate_method_entry) so the guard should work for them
701 // too.
702 //
703 // Args:
704 // r3: number of additional locals this frame needs (what we must check)
705 // rmethod: Method*
706 //
707 // Kills:
708 // r0
generate_stack_overflow_check(void)709 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
710
711 // monitor entry size: see picture of stack set
712 // (generate_method_entry) and frame_amd64.hpp
713 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
714
715 // total overhead size: entry_size + (saved rbp through expr stack
716 // bottom). be sure to change this if you add/subtract anything
717 // to/from the overhead area
718 const int overhead_size =
719 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
720
721 const int page_size = os::vm_page_size();
722
723 Label after_frame_check;
724
725 // see if the frame is greater than one page in size. If so,
726 // then we need to verify there is enough stack space remaining
727 // for the additional locals.
728 //
729 // Note that we use SUBS rather than CMP here because the immediate
730 // field of this instruction may overflow. SUBS can cope with this
731 // because it is a macro that will expand to some number of MOV
732 // instructions and a register operation.
733 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
734 __ br(Assembler::LS, after_frame_check);
735
736 // compute rsp as if this were going to be the last frame on
737 // the stack before the red zone
738
739 // locals + overhead, in bytes
740 __ mov(r0, overhead_size);
741 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
742
743 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
744 __ ldr(rscratch1, stack_limit);
745
746 #ifdef ASSERT
747 Label limit_okay;
748 // Verify that thread stack limit is non-zero.
749 __ cbnz(rscratch1, limit_okay);
750 __ stop("stack overflow limit is zero");
751 __ bind(limit_okay);
752 #endif
753
754 // Add stack limit to locals.
755 __ add(r0, r0, rscratch1);
756
757 // Check against the current stack bottom.
758 __ cmp(sp, r0);
759 __ br(Assembler::HI, after_frame_check);
760
761 // Remove the incoming args, peeling the machine SP back to where it
762 // was in the caller. This is not strictly necessary, but unless we
763 // do so the stack frame may have a garbage FP; this ensures a
764 // correct call stack that we can always unwind. The ANDR should be
765 // unnecessary because the sender SP in r13 is always aligned, but
766 // it doesn't hurt.
767 __ andr(sp, r13, -16);
768
769 // Note: the restored frame is not necessarily interpreted.
770 // Use the shared runtime version of the StackOverflowError.
771 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
772 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
773
774 // all done with frame size check
775 __ bind(after_frame_check);
776 }
777
778 // Allocate monitor and lock method (asm interpreter)
779 //
780 // Args:
781 // rmethod: Method*
782 // rlocals: locals
783 //
784 // Kills:
785 // r0
786 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
787 // rscratch1, rscratch2 (scratch regs)
lock_method()788 void TemplateInterpreterGenerator::lock_method() {
789 // synchronize method
790 const Address access_flags(rmethod, Method::access_flags_offset());
791 const Address monitor_block_top(
792 rfp,
793 frame::interpreter_frame_monitor_block_top_offset * wordSize);
794 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
795
796 #ifdef ASSERT
797 {
798 Label L;
799 __ ldrw(r0, access_flags);
800 __ tst(r0, JVM_ACC_SYNCHRONIZED);
801 __ br(Assembler::NE, L);
802 __ stop("method doesn't need synchronization");
803 __ bind(L);
804 }
805 #endif // ASSERT
806
807 // get synchronization object
808 {
809 Label done;
810 __ ldrw(r0, access_flags);
811 __ tst(r0, JVM_ACC_STATIC);
812 // get receiver (assume this is frequent case)
813 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
814 __ br(Assembler::EQ, done);
815 __ load_mirror(r0, rmethod);
816
817 #ifdef ASSERT
818 {
819 Label L;
820 __ cbnz(r0, L);
821 __ stop("synchronization object is NULL");
822 __ bind(L);
823 }
824 #endif // ASSERT
825
826 __ bind(done);
827 }
828
829 // add space for monitor & lock
830 __ sub(sp, sp, entry_size); // add space for a monitor entry
831 __ sub(esp, esp, entry_size);
832 __ mov(rscratch1, esp);
833 __ str(rscratch1, monitor_block_top); // set new monitor block top
834 // store object
835 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
836 __ mov(c_rarg1, esp); // object address
837 __ lock_object(c_rarg1);
838 }
839
840 // Generate a fixed interpreter frame. This is identical setup for
841 // interpreted methods and for native methods hence the shared code.
842 //
843 // Args:
844 // lr: return address
845 // rmethod: Method*
846 // rlocals: pointer to locals
847 // rcpool: cp cache
848 // stack_pointer: previous sp
generate_fixed_frame(bool native_call)849 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
850 // initialize fixed part of activation frame
851 if (native_call) {
852 __ sub(esp, sp, 14 * wordSize);
853 __ mov(rbcp, zr);
854 __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
855 // add 2 zero-initialized slots for native calls
856 __ stp(zr, zr, Address(sp, 12 * wordSize));
857 } else {
858 __ sub(esp, sp, 12 * wordSize);
859 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
860 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
861 __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
862 }
863
864 if (ProfileInterpreter) {
865 Label method_data_continue;
866 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
867 __ cbz(rscratch1, method_data_continue);
868 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
869 __ bind(method_data_continue);
870 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
871 } else {
872 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
873 }
874
875 // Get mirror and store it in the frame as GC root for this Method*
876 #if INCLUDE_SHENANDOAHGC
877 if (UseShenandoahGC) {
878 __ load_mirror(r10, rmethod);
879 __ stp(r10, zr, Address(sp, 4 * wordSize));
880 } else
881 #endif
882 {
883 __ load_mirror(rscratch1, rmethod);
884 __ stp(rscratch1, zr, Address(sp, 4 * wordSize));
885 }
886
887 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
888 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
889 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
890 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
891
892 __ stp(rfp, lr, Address(sp, 10 * wordSize));
893 __ lea(rfp, Address(sp, 10 * wordSize));
894
895 // set sender sp
896 // leave last_sp as null
897 __ stp(zr, r13, Address(sp, 8 * wordSize));
898
899 // Move SP out of the way
900 if (! native_call) {
901 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
902 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
903 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
904 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
905 __ andr(sp, rscratch1, -16);
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.
generate_Reference_get_entry(void)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 // r13: 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 guarantee(referent_offset > 0, "referent offset not initialized");
952
953 Label slow_path;
954 const Register local_0 = c_rarg0;
955 // Check if local 0 != NULL
956 // If the receiver is null then it is OK to jump to the slow path.
957 __ ldr(local_0, Address(esp, 0));
958 __ cbz(local_0, slow_path);
959
960 __ mov(r19, r13); // Move senderSP to a callee-saved register
961
962 // Load the value of the referent field.
963 const Address field_address(local_0, referent_offset);
964 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
965 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch2, /*tmp2*/ rscratch1);
966
967 // areturn
968 __ andr(sp, r19, -16); // done with stack
969 __ ret(lr);
970
971 // generate a vanilla interpreter entry as the slow path
972 __ bind(slow_path);
973 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
974 return entry;
975
976 }
977
978 /**
979 * Method entry for static native methods:
980 * int java.util.zip.CRC32.update(int crc, int b)
981 */
generate_CRC32_update_entry()982 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
983 if (UseCRC32Intrinsics) {
984 address entry = __ pc();
985
986 // rmethod: Method*
987 // r13: senderSP must preserved for slow path
988 // esp: args
989
990 Label slow_path;
991 // If we need a safepoint check, generate full interpreter entry.
992 __ safepoint_poll(slow_path);
993
994 // We don't generate local frame and don't align stack because
995 // we call stub code and there is no safepoint on this path.
996
997 // Load parameters
998 const Register crc = c_rarg0; // crc
999 const Register val = c_rarg1; // source java byte value
1000 const Register tbl = c_rarg2; // scratch
1001
1002 // Arguments are reversed on java expression stack
1003 __ ldrw(val, Address(esp, 0)); // byte value
1004 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
1005
1006 uint64_t offset;
1007 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1008 __ add(tbl, tbl, offset);
1009
1010 __ mvnw(crc, crc); // ~crc
1011 __ update_byte_crc32(crc, val, tbl);
1012 __ mvnw(crc, crc); // ~crc
1013
1014 // result in c_rarg0
1015
1016 __ andr(sp, r13, -16);
1017 __ ret(lr);
1018
1019 // generate a vanilla native entry as the slow path
1020 __ bind(slow_path);
1021 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1022 return entry;
1023 }
1024 return NULL;
1025 }
1026
1027 /**
1028 * Method entry for static native methods:
1029 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1030 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1031 */
generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind)1032 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1033 if (UseCRC32Intrinsics) {
1034 address entry = __ pc();
1035
1036 // rmethod,: Method*
1037 // r13: senderSP must preserved for slow path
1038
1039 Label slow_path;
1040 // If we need a safepoint check, generate full interpreter entry.
1041 __ safepoint_poll(slow_path);
1042
1043 // We don't generate local frame and don't align stack because
1044 // we call stub code and there is no safepoint on this path.
1045
1046 // Load parameters
1047 const Register crc = c_rarg0; // crc
1048 const Register buf = c_rarg1; // source java byte array address
1049 const Register len = c_rarg2; // length
1050 const Register off = len; // offset (never overlaps with 'len')
1051
1052 // Arguments are reversed on java expression stack
1053 // Calculate address of start element
1054 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1055 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1056 __ ldrw(off, Address(esp, wordSize)); // offset
1057 __ add(buf, buf, off); // + offset
1058 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1059 } else {
1060 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1061 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1062 __ ldrw(off, Address(esp, wordSize)); // offset
1063 __ add(buf, buf, off); // + offset
1064 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1065 }
1066 // Can now load 'len' since we're finished with 'off'
1067 __ ldrw(len, Address(esp, 0x0)); // Length
1068
1069 __ andr(sp, r13, -16); // Restore the caller's SP
1070
1071 // We are frameless so we can just jump to the stub.
1072 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1073
1074 // generate a vanilla native entry as the slow path
1075 __ bind(slow_path);
1076 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1077 return entry;
1078 }
1079 return NULL;
1080 }
1081
1082 /**
1083 * Method entry for intrinsic-candidate (non-native) methods:
1084 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1085 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1086 * Unlike CRC32, CRC32C does not have any methods marked as native
1087 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1088 */
generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind)1089 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1090 if (UseCRC32CIntrinsics) {
1091 address entry = __ pc();
1092
1093 // Prepare jump to stub using parameters from the stack
1094 const Register crc = c_rarg0; // initial crc
1095 const Register buf = c_rarg1; // source java byte array address
1096 const Register len = c_rarg2; // len argument to the kernel
1097
1098 const Register end = len; // index of last element to process
1099 const Register off = crc; // offset
1100
1101 __ ldrw(end, Address(esp)); // int end
1102 __ ldrw(off, Address(esp, wordSize)); // int offset
1103 __ sub(len, end, off);
1104 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1105 __ add(buf, buf, off); // + offset
1106 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1107 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1108 } else {
1109 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1110 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1111 }
1112
1113 __ andr(sp, r13, -16); // Restore the caller's SP
1114
1115 // Jump to the stub.
1116 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1117
1118 return entry;
1119 }
1120 return NULL;
1121 }
1122
bang_stack_shadow_pages(bool native_call)1123 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1124 // Bang each page in the shadow zone. We can't assume it's been done for
1125 // an interpreter frame with greater than a page of locals, so each page
1126 // needs to be checked. Only true for non-native.
1127 if (UseStackBanging) {
1128 const int n_shadow_pages = JavaThread::stack_shadow_zone_size() / os::vm_page_size();
1129 const int start_page = native_call ? n_shadow_pages : 1;
1130 const int page_size = os::vm_page_size();
1131 for (int pages = start_page; pages <= n_shadow_pages ; pages++) {
1132 __ sub(rscratch2, sp, pages*page_size);
1133 __ str(zr, Address(rscratch2));
1134 }
1135 }
1136 }
1137
1138
1139 // Interpreter stub for calling a native method. (asm interpreter)
1140 // This sets up a somewhat different looking stack for calling the
1141 // native method than the typical interpreter frame setup.
generate_native_entry(bool synchronized)1142 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1143 // determine code generation flags
1144 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1145
1146 // r1: Method*
1147 // rscratch1: sender sp
1148
1149 address entry_point = __ pc();
1150
1151 const Address constMethod (rmethod, Method::const_offset());
1152 const Address access_flags (rmethod, Method::access_flags_offset());
1153 const Address size_of_parameters(r2, ConstMethod::
1154 size_of_parameters_offset());
1155
1156 // get parameter size (always needed)
1157 __ ldr(r2, constMethod);
1158 __ load_unsigned_short(r2, size_of_parameters);
1159
1160 // Native calls don't need the stack size check since they have no
1161 // expression stack and the arguments are already on the stack and
1162 // we only add a handful of words to the stack.
1163
1164 // rmethod: Method*
1165 // r2: size of parameters
1166 // rscratch1: sender sp
1167
1168 // for natives the size of locals is zero
1169
1170 // compute beginning of parameters (rlocals)
1171 __ add(rlocals, esp, r2, ext::uxtx, 3);
1172 __ add(rlocals, rlocals, -wordSize);
1173
1174 // Pull SP back to minimum size: this avoids holes in the stack
1175 __ andr(sp, esp, -16);
1176
1177 // initialize fixed part of activation frame
1178 generate_fixed_frame(true);
1179
1180 // make sure method is native & not abstract
1181 #ifdef ASSERT
1182 __ ldrw(r0, access_flags);
1183 {
1184 Label L;
1185 __ tst(r0, JVM_ACC_NATIVE);
1186 __ br(Assembler::NE, L);
1187 __ stop("tried to execute non-native method as native");
1188 __ bind(L);
1189 }
1190 {
1191 Label L;
1192 __ tst(r0, JVM_ACC_ABSTRACT);
1193 __ br(Assembler::EQ, L);
1194 __ stop("tried to execute abstract method in interpreter");
1195 __ bind(L);
1196 }
1197 #endif
1198
1199 // Since at this point in the method invocation the exception
1200 // handler would try to exit the monitor of synchronized methods
1201 // which hasn't been entered yet, we set the thread local variable
1202 // _do_not_unlock_if_synchronized to true. The remove_activation
1203 // will check this flag.
1204
1205 const Address do_not_unlock_if_synchronized(rthread,
1206 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1207 __ mov(rscratch2, true);
1208 __ strb(rscratch2, do_not_unlock_if_synchronized);
1209
1210 // increment invocation count & check for overflow
1211 Label invocation_counter_overflow;
1212 if (inc_counter) {
1213 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1214 }
1215
1216 Label continue_after_compile;
1217 __ bind(continue_after_compile);
1218
1219 bang_stack_shadow_pages(true);
1220
1221 // reset the _do_not_unlock_if_synchronized flag
1222 __ strb(zr, do_not_unlock_if_synchronized);
1223
1224 // check for synchronized methods
1225 // Must happen AFTER invocation_counter check and stack overflow check,
1226 // so method is not locked if overflows.
1227 if (synchronized) {
1228 lock_method();
1229 } else {
1230 // no synchronization necessary
1231 #ifdef ASSERT
1232 {
1233 Label L;
1234 __ ldrw(r0, access_flags);
1235 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1236 __ br(Assembler::EQ, L);
1237 __ stop("method needs synchronization");
1238 __ bind(L);
1239 }
1240 #endif
1241 }
1242
1243 // start execution
1244 #ifdef ASSERT
1245 {
1246 Label L;
1247 const Address monitor_block_top(rfp,
1248 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1249 __ ldr(rscratch1, monitor_block_top);
1250 __ cmp(esp, rscratch1);
1251 __ br(Assembler::EQ, L);
1252 __ stop("broken stack frame setup in interpreter");
1253 __ bind(L);
1254 }
1255 #endif
1256
1257 // jvmti support
1258 __ notify_method_entry();
1259
1260 // work registers
1261 const Register t = r17;
1262 const Register result_handler = r19;
1263
1264 // allocate space for parameters
1265 __ ldr(t, Address(rmethod, Method::const_offset()));
1266 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1267
1268 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1269 __ andr(sp, rscratch1, -16);
1270 __ mov(esp, rscratch1);
1271
1272 // get signature handler
1273 {
1274 Label L;
1275 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1276 __ cbnz(t, L);
1277 __ call_VM(noreg,
1278 CAST_FROM_FN_PTR(address,
1279 InterpreterRuntime::prepare_native_call),
1280 rmethod);
1281 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1282 __ bind(L);
1283 }
1284
1285 // call signature handler
1286 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1287 "adjust this code");
1288 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1289 "adjust this code");
1290 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1291 "adjust this code");
1292
1293 // The generated handlers do not touch rmethod (the method).
1294 // However, large signatures cannot be cached and are generated
1295 // each time here. The slow-path generator can do a GC on return,
1296 // so we must reload it after the call.
1297 __ blr(t);
1298 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1299
1300
1301 // result handler is in r0
1302 // set result handler
1303 __ mov(result_handler, r0);
1304 // pass mirror handle if static call
1305 {
1306 Label L;
1307 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1308 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1309 // get mirror
1310 __ load_mirror(t, rmethod);
1311 // copy mirror into activation frame
1312 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1313 // pass handle to mirror
1314 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1315 __ bind(L);
1316 }
1317
1318 // get native function entry point in r10
1319 {
1320 Label L;
1321 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1322 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1323 __ mov(rscratch2, unsatisfied);
1324 __ ldr(rscratch2, rscratch2);
1325 __ cmp(r10, rscratch2);
1326 __ br(Assembler::NE, L);
1327 __ call_VM(noreg,
1328 CAST_FROM_FN_PTR(address,
1329 InterpreterRuntime::prepare_native_call),
1330 rmethod);
1331 __ get_method(rmethod);
1332 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1333 __ bind(L);
1334 }
1335
1336 // pass JNIEnv
1337 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1338
1339 // Set the last Java PC in the frame anchor to be the return address from
1340 // the call to the native method: this will allow the debugger to
1341 // generate an accurate stack trace.
1342 Label native_return;
1343 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1344
1345 // change thread state
1346 #ifdef ASSERT
1347 {
1348 Label L;
1349 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1350 __ cmp(t, _thread_in_Java);
1351 __ br(Assembler::EQ, L);
1352 __ stop("Wrong thread state in native stub");
1353 __ bind(L);
1354 }
1355 #endif
1356
1357 // Change state to native
1358 __ mov(rscratch1, _thread_in_native);
1359 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1360 __ stlrw(rscratch1, rscratch2);
1361
1362 // Call the native method.
1363 __ blr(r10);
1364 __ bind(native_return);
1365 __ maybe_isb();
1366 __ get_method(rmethod);
1367 // result potentially in r0 or v0
1368
1369 // make room for the pushes we're about to do
1370 __ sub(rscratch1, esp, 4 * wordSize);
1371 __ andr(sp, rscratch1, -16);
1372
1373 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1374 // in order to extract the result of a method call. If the order of these
1375 // pushes change or anything else is added to the stack then the code in
1376 // interpreter_frame_result must also change.
1377 __ push(dtos);
1378 __ push(ltos);
1379
1380 // change thread state
1381 __ mov(rscratch1, _thread_in_native_trans);
1382 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1383 __ stlrw(rscratch1, rscratch2);
1384
1385 if (os::is_MP()) {
1386 if (UseMembar) {
1387 // Force this write out before the read below
1388 __ dmb(Assembler::ISH);
1389 } else {
1390 // Write serialization page so VM thread can do a pseudo remote membar.
1391 // We use the current thread pointer to calculate a thread specific
1392 // offset to write to within the page. This minimizes bus traffic
1393 // due to cache line collision.
1394 __ serialize_memory(rthread, rscratch2);
1395 }
1396 }
1397
1398 // check for safepoint operation in progress and/or pending suspend requests
1399 {
1400 Label L, Continue;
1401 __ safepoint_poll_acquire(L);
1402 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1403 __ cbz(rscratch2, Continue);
1404 __ bind(L);
1405
1406 // Don't use call_VM as it will see a possible pending exception
1407 // and forward it and never return here preventing us from
1408 // clearing _last_native_pc down below. So we do a runtime call by
1409 // hand.
1410 //
1411 __ mov(c_rarg0, rthread);
1412 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1413 __ blr(rscratch2);
1414 __ maybe_isb();
1415 __ get_method(rmethod);
1416 __ reinit_heapbase();
1417 __ bind(Continue);
1418 }
1419
1420 // change thread state
1421 __ mov(rscratch1, _thread_in_Java);
1422 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1423 __ stlrw(rscratch1, rscratch2);
1424
1425 // reset_last_Java_frame
1426 __ reset_last_Java_frame(true);
1427
1428 if (CheckJNICalls) {
1429 // clear_pending_jni_exception_check
1430 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1431 }
1432
1433 // reset handle block
1434 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1435 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1436
1437 // If result is an oop unbox and store it in frame where gc will see it
1438 // and result handler will pick it up
1439
1440 {
1441 Label no_oop, not_weak, store_result;
1442 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1443 __ cmp(t, result_handler);
1444 __ br(Assembler::NE, no_oop);
1445 // Unbox oop result, e.g. JNIHandles::resolve result.
1446 __ pop(ltos);
1447 __ resolve_jobject(r0, rthread, t);
1448 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1449 // keep stack depth as expected by pushing oop which will eventually be discarded
1450 __ push(ltos);
1451 __ bind(no_oop);
1452 }
1453
1454 {
1455 Label no_reguard;
1456 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1457 __ ldrw(rscratch1, Address(rscratch1));
1458 __ cmp(rscratch1, JavaThread::stack_guard_yellow_reserved_disabled);
1459 __ br(Assembler::NE, no_reguard);
1460
1461 __ pusha(); // XXX only save smashed registers
1462 __ mov(c_rarg0, rthread);
1463 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1464 __ blr(rscratch2);
1465 __ popa(); // XXX only restore smashed registers
1466 __ bind(no_reguard);
1467 }
1468
1469 // The method register is junk from after the thread_in_native transition
1470 // until here. Also can't call_VM until the bcp has been
1471 // restored. Need bcp for throwing exception below so get it now.
1472 __ get_method(rmethod);
1473
1474 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1475 // rbcp == code_base()
1476 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1477 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1478 // handle exceptions (exception handling will handle unlocking!)
1479 {
1480 Label L;
1481 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1482 __ cbz(rscratch1, L);
1483 // Note: At some point we may want to unify this with the code
1484 // used in call_VM_base(); i.e., we should use the
1485 // StubRoutines::forward_exception code. For now this doesn't work
1486 // here because the rsp is not correctly set at this point.
1487 __ MacroAssembler::call_VM(noreg,
1488 CAST_FROM_FN_PTR(address,
1489 InterpreterRuntime::throw_pending_exception));
1490 __ should_not_reach_here();
1491 __ bind(L);
1492 }
1493
1494 // do unlocking if necessary
1495 {
1496 Label L;
1497 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1498 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1499 // the code below should be shared with interpreter macro
1500 // assembler implementation
1501 {
1502 Label unlock;
1503 // BasicObjectLock will be first in list, since this is a
1504 // synchronized method. However, need to check that the object
1505 // has not been unlocked by an explicit monitorexit bytecode.
1506
1507 // monitor expect in c_rarg1 for slow unlock path
1508 __ lea (c_rarg1, Address(rfp, // address of first monitor
1509 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1510 wordSize - sizeof(BasicObjectLock))));
1511
1512 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1513 __ cbnz(t, unlock);
1514
1515 // Entry already unlocked, need to throw exception
1516 __ MacroAssembler::call_VM(noreg,
1517 CAST_FROM_FN_PTR(address,
1518 InterpreterRuntime::throw_illegal_monitor_state_exception));
1519 __ should_not_reach_here();
1520
1521 __ bind(unlock);
1522 __ unlock_object(c_rarg1);
1523 }
1524 __ bind(L);
1525 }
1526
1527 // jvmti support
1528 // Note: This must happen _after_ handling/throwing any exceptions since
1529 // the exception handler code notifies the runtime of method exits
1530 // too. If this happens before, method entry/exit notifications are
1531 // not properly paired (was bug - gri 11/22/99).
1532 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1533
1534 // restore potential result in r0:d0, call result handler to
1535 // restore potential result in ST0 & handle result
1536
1537 __ pop(ltos);
1538 __ pop(dtos);
1539
1540 __ blr(result_handler);
1541
1542 // remove activation
1543 __ ldr(esp, Address(rfp,
1544 frame::interpreter_frame_sender_sp_offset *
1545 wordSize)); // get sender sp
1546 // remove frame anchor
1547 __ leave();
1548
1549 // resture sender sp
1550 __ mov(sp, esp);
1551
1552 __ ret(lr);
1553
1554 if (inc_counter) {
1555 // Handle overflow of counter and compile method
1556 __ bind(invocation_counter_overflow);
1557 generate_counter_overflow(continue_after_compile);
1558 }
1559
1560 return entry_point;
1561 }
1562
1563 //
1564 // Generic interpreted method entry to (asm) interpreter
1565 //
generate_normal_entry(bool synchronized)1566 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1567 // determine code generation flags
1568 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1569
1570 // rscratch1: sender sp
1571 address entry_point = __ pc();
1572
1573 const Address constMethod(rmethod, Method::const_offset());
1574 const Address access_flags(rmethod, Method::access_flags_offset());
1575 const Address size_of_parameters(r3,
1576 ConstMethod::size_of_parameters_offset());
1577 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1578
1579 // get parameter size (always needed)
1580 // need to load the const method first
1581 __ ldr(r3, constMethod);
1582 __ load_unsigned_short(r2, size_of_parameters);
1583
1584 // r2: size of parameters
1585
1586 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1587 __ sub(r3, r3, r2); // r3 = no. of additional locals
1588
1589 // see if we've got enough room on the stack for locals plus overhead.
1590 generate_stack_overflow_check();
1591
1592 // compute beginning of parameters (rlocals)
1593 __ add(rlocals, esp, r2, ext::uxtx, 3);
1594 __ sub(rlocals, rlocals, wordSize);
1595
1596 __ mov(rscratch1, esp);
1597
1598 // r3 - # of additional locals
1599 // allocate space for locals
1600 // explicitly initialize locals
1601 // Initializing memory allocated for locals in the same direction as
1602 // the stack grows to ensure page initialization order according
1603 // to windows-aarch64 stack page growth requirement (see
1604 // https://docs.microsoft.com/en-us/cpp/build/arm64-windows-abi-conventions?view=msvc-160#stack)
1605 {
1606 Label exit, loop;
1607 __ ands(zr, r3, r3);
1608 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1609 __ bind(loop);
1610 __ str(zr, Address(__ pre(rscratch1, -wordSize)));
1611 __ sub(r3, r3, 1); // until everything initialized
1612 __ cbnz(r3, loop);
1613 __ bind(exit);
1614 }
1615
1616 // Padding between locals and fixed part of activation frame to ensure
1617 // SP is always 16-byte aligned.
1618 __ andr(sp, rscratch1, -16);
1619
1620 // And the base dispatch table
1621 __ get_dispatch();
1622
1623 // initialize fixed part of activation frame
1624 generate_fixed_frame(false);
1625
1626 // make sure method is not native & not abstract
1627 #ifdef ASSERT
1628 __ ldrw(r0, access_flags);
1629 {
1630 Label L;
1631 __ tst(r0, JVM_ACC_NATIVE);
1632 __ br(Assembler::EQ, L);
1633 __ stop("tried to execute native method as non-native");
1634 __ bind(L);
1635 }
1636 {
1637 Label L;
1638 __ tst(r0, JVM_ACC_ABSTRACT);
1639 __ br(Assembler::EQ, L);
1640 __ stop("tried to execute abstract method in interpreter");
1641 __ bind(L);
1642 }
1643 #endif
1644
1645 // Since at this point in the method invocation the exception
1646 // handler would try to exit the monitor of synchronized methods
1647 // which hasn't been entered yet, we set the thread local variable
1648 // _do_not_unlock_if_synchronized to true. The remove_activation
1649 // will check this flag.
1650
1651 const Address do_not_unlock_if_synchronized(rthread,
1652 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1653 __ mov(rscratch2, true);
1654 __ strb(rscratch2, do_not_unlock_if_synchronized);
1655
1656 Register mdp = r3;
1657 __ profile_parameters_type(mdp, r1, r2);
1658
1659 // increment invocation count & check for overflow
1660 Label invocation_counter_overflow;
1661 Label profile_method;
1662 Label profile_method_continue;
1663 if (inc_counter) {
1664 generate_counter_incr(&invocation_counter_overflow,
1665 &profile_method,
1666 &profile_method_continue);
1667 if (ProfileInterpreter) {
1668 __ bind(profile_method_continue);
1669 }
1670 }
1671
1672 Label continue_after_compile;
1673 __ bind(continue_after_compile);
1674
1675 bang_stack_shadow_pages(false);
1676
1677 // reset the _do_not_unlock_if_synchronized flag
1678 __ strb(zr, do_not_unlock_if_synchronized);
1679
1680 // check for synchronized methods
1681 // Must happen AFTER invocation_counter check and stack overflow check,
1682 // so method is not locked if overflows.
1683 if (synchronized) {
1684 // Allocate monitor and lock method
1685 lock_method();
1686 } else {
1687 // no synchronization necessary
1688 #ifdef ASSERT
1689 {
1690 Label L;
1691 __ ldrw(r0, access_flags);
1692 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1693 __ br(Assembler::EQ, L);
1694 __ stop("method needs synchronization");
1695 __ bind(L);
1696 }
1697 #endif
1698 }
1699
1700 // start execution
1701 #ifdef ASSERT
1702 {
1703 Label L;
1704 const Address monitor_block_top (rfp,
1705 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1706 __ ldr(rscratch1, monitor_block_top);
1707 __ cmp(esp, rscratch1);
1708 __ br(Assembler::EQ, L);
1709 __ stop("broken stack frame setup in interpreter");
1710 __ bind(L);
1711 }
1712 #endif
1713
1714 // jvmti support
1715 __ notify_method_entry();
1716
1717 __ dispatch_next(vtos);
1718
1719 // invocation counter overflow
1720 if (inc_counter) {
1721 if (ProfileInterpreter) {
1722 // We have decided to profile this method in the interpreter
1723 __ bind(profile_method);
1724 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1725 __ set_method_data_pointer_for_bcp();
1726 // don't think we need this
1727 __ get_method(r1);
1728 __ b(profile_method_continue);
1729 }
1730 // Handle overflow of counter and compile method
1731 __ bind(invocation_counter_overflow);
1732 generate_counter_overflow(continue_after_compile);
1733 }
1734
1735 return entry_point;
1736 }
1737
1738 //-----------------------------------------------------------------------------
1739 // Exceptions
1740
generate_throw_exception()1741 void TemplateInterpreterGenerator::generate_throw_exception() {
1742 // Entry point in previous activation (i.e., if the caller was
1743 // interpreted)
1744 Interpreter::_rethrow_exception_entry = __ pc();
1745 // Restore sp to interpreter_frame_last_sp even though we are going
1746 // to empty the expression stack for the exception processing.
1747 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1748 // r0: exception
1749 // r3: return address/pc that threw exception
1750 __ restore_bcp(); // rbcp points to call/send
1751 __ restore_locals();
1752 __ restore_constant_pool_cache();
1753 __ reinit_heapbase(); // restore rheapbase as heapbase.
1754 __ get_dispatch();
1755
1756 // Entry point for exceptions thrown within interpreter code
1757 Interpreter::_throw_exception_entry = __ pc();
1758 // If we came here via a NullPointerException on the receiver of a
1759 // method, rmethod may be corrupt.
1760 __ get_method(rmethod);
1761 // expression stack is undefined here
1762 // r0: exception
1763 // rbcp: exception bcp
1764 __ verify_oop(r0);
1765 __ mov(c_rarg1, r0);
1766
1767 // expression stack must be empty before entering the VM in case of
1768 // an exception
1769 __ empty_expression_stack();
1770 // find exception handler address and preserve exception oop
1771 __ call_VM(r3,
1772 CAST_FROM_FN_PTR(address,
1773 InterpreterRuntime::exception_handler_for_exception),
1774 c_rarg1);
1775
1776 // Calculate stack limit
1777 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1778 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1779 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1780 __ ldr(rscratch2,
1781 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1782 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1783 __ andr(sp, rscratch1, -16);
1784
1785 // r0: exception handler entry point
1786 // r3: preserved exception oop
1787 // rbcp: bcp for exception handler
1788 __ push_ptr(r3); // push exception which is now the only value on the stack
1789 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1790
1791 // If the exception is not handled in the current frame the frame is
1792 // removed and the exception is rethrown (i.e. exception
1793 // continuation is _rethrow_exception).
1794 //
1795 // Note: At this point the bci is still the bxi for the instruction
1796 // which caused the exception and the expression stack is
1797 // empty. Thus, for any VM calls at this point, GC will find a legal
1798 // oop map (with empty expression stack).
1799
1800 //
1801 // JVMTI PopFrame support
1802 //
1803
1804 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1805 __ empty_expression_stack();
1806 // Set the popframe_processing bit in pending_popframe_condition
1807 // indicating that we are currently handling popframe, so that
1808 // call_VMs that may happen later do not trigger new popframe
1809 // handling cycles.
1810 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1811 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1812 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1813
1814 {
1815 // Check to see whether we are returning to a deoptimized frame.
1816 // (The PopFrame call ensures that the caller of the popped frame is
1817 // either interpreted or compiled and deoptimizes it if compiled.)
1818 // In this case, we can't call dispatch_next() after the frame is
1819 // popped, but instead must save the incoming arguments and restore
1820 // them after deoptimization has occurred.
1821 //
1822 // Note that we don't compare the return PC against the
1823 // deoptimization blob's unpack entry because of the presence of
1824 // adapter frames in C2.
1825 Label caller_not_deoptimized;
1826 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1827 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1828 InterpreterRuntime::interpreter_contains), c_rarg1);
1829 __ cbnz(r0, caller_not_deoptimized);
1830
1831 // Compute size of arguments for saving when returning to
1832 // deoptimized caller
1833 __ get_method(r0);
1834 __ ldr(r0, Address(r0, Method::const_offset()));
1835 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1836 size_of_parameters_offset())));
1837 __ lsl(r0, r0, Interpreter::logStackElementSize);
1838 __ restore_locals(); // XXX do we need this?
1839 __ sub(rlocals, rlocals, r0);
1840 __ add(rlocals, rlocals, wordSize);
1841 // Save these arguments
1842 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1843 Deoptimization::
1844 popframe_preserve_args),
1845 rthread, r0, rlocals);
1846
1847 __ remove_activation(vtos,
1848 /* throw_monitor_exception */ false,
1849 /* install_monitor_exception */ false,
1850 /* notify_jvmdi */ false);
1851
1852 // Inform deoptimization that it is responsible for restoring
1853 // these arguments
1854 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1855 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1856
1857 // Continue in deoptimization handler
1858 __ ret(lr);
1859
1860 __ bind(caller_not_deoptimized);
1861 }
1862
1863 __ remove_activation(vtos,
1864 /* throw_monitor_exception */ false,
1865 /* install_monitor_exception */ false,
1866 /* notify_jvmdi */ false);
1867
1868 // Restore the last_sp and null it out
1869 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1870 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1871
1872 __ restore_bcp();
1873 __ restore_locals();
1874 __ restore_constant_pool_cache();
1875 __ get_method(rmethod);
1876 __ get_dispatch();
1877
1878 // The method data pointer was incremented already during
1879 // call profiling. We have to restore the mdp for the current bcp.
1880 if (ProfileInterpreter) {
1881 __ set_method_data_pointer_for_bcp();
1882 }
1883
1884 // Clear the popframe condition flag
1885 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1886 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1887
1888 #if INCLUDE_JVMTI
1889 {
1890 Label L_done;
1891
1892 __ ldrb(rscratch1, Address(rbcp, 0));
1893 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1894 __ br(Assembler::NE, L_done);
1895
1896 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1897 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1898
1899 __ ldr(c_rarg0, Address(rlocals, 0));
1900 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1901
1902 __ cbz(r0, L_done);
1903
1904 __ str(r0, Address(esp, 0));
1905 __ bind(L_done);
1906 }
1907 #endif // INCLUDE_JVMTI
1908
1909 // Restore machine SP
1910 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1911 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1912 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1913 __ ldr(rscratch2,
1914 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1915 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1916 __ andr(sp, rscratch1, -16);
1917
1918 __ dispatch_next(vtos);
1919 // end of PopFrame support
1920
1921 Interpreter::_remove_activation_entry = __ pc();
1922
1923 // preserve exception over this code sequence
1924 __ pop_ptr(r0);
1925 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1926 // remove the activation (without doing throws on illegalMonitorExceptions)
1927 __ remove_activation(vtos, false, true, false);
1928 // restore exception
1929 __ get_vm_result(r0, rthread);
1930
1931 // In between activations - previous activation type unknown yet
1932 // compute continuation point - the continuation point expects the
1933 // following registers set up:
1934 //
1935 // r0: exception
1936 // lr: return address/pc that threw exception
1937 // esp: expression stack of caller
1938 // rfp: fp of caller
1939 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1940 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1941 SharedRuntime::exception_handler_for_return_address),
1942 rthread, lr);
1943 __ mov(r1, r0); // save exception handler
1944 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1945 // We might be returning to a deopt handler that expects r3 to
1946 // contain the exception pc
1947 __ mov(r3, lr);
1948 // Note that an "issuing PC" is actually the next PC after the call
1949 __ br(r1); // jump to exception
1950 // handler of caller
1951 }
1952
1953
1954 //
1955 // JVMTI ForceEarlyReturn support
1956 //
generate_earlyret_entry_for(TosState state)1957 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1958 address entry = __ pc();
1959
1960 __ restore_bcp();
1961 __ restore_locals();
1962 __ empty_expression_stack();
1963 __ load_earlyret_value(state);
1964
1965 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1966 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1967
1968 // Clear the earlyret state
1969 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1970 __ str(zr, cond_addr);
1971
1972 __ remove_activation(state,
1973 false, /* throw_monitor_exception */
1974 false, /* install_monitor_exception */
1975 true); /* notify_jvmdi */
1976 __ ret(lr);
1977
1978 return entry;
1979 } // end of ForceEarlyReturn support
1980
1981
1982
1983 //-----------------------------------------------------------------------------
1984 // Helper for vtos entry point generation
1985
set_vtos_entry_points(Template * t,address & bep,address & cep,address & sep,address & aep,address & iep,address & lep,address & fep,address & dep,address & vep)1986 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1987 address& bep,
1988 address& cep,
1989 address& sep,
1990 address& aep,
1991 address& iep,
1992 address& lep,
1993 address& fep,
1994 address& dep,
1995 address& vep) {
1996 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1997 Label L;
1998 aep = __ pc(); __ push_ptr(); __ b(L);
1999 fep = __ pc(); __ push_f(); __ b(L);
2000 dep = __ pc(); __ push_d(); __ b(L);
2001 lep = __ pc(); __ push_l(); __ b(L);
2002 bep = cep = sep =
2003 iep = __ pc(); __ push_i();
2004 vep = __ pc();
2005 __ bind(L);
2006 generate_and_dispatch(t);
2007 }
2008
2009 //-----------------------------------------------------------------------------
2010
2011 // Non-product code
2012 #ifndef PRODUCT
generate_trace_code(TosState state)2013 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2014 address entry = __ pc();
2015
2016 __ push(lr);
2017 __ push(state);
2018 __ push(RegSet::range(r0, r15), sp);
2019 __ mov(c_rarg2, r0); // Pass itos
2020 __ call_VM(noreg,
2021 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2022 c_rarg1, c_rarg2, c_rarg3);
2023 __ pop(RegSet::range(r0, r15), sp);
2024 __ pop(state);
2025 __ pop(lr);
2026 __ ret(lr); // return from result handler
2027
2028 return entry;
2029 }
2030
count_bytecode()2031 void TemplateInterpreterGenerator::count_bytecode() {
2032 Register rscratch3 = r0;
2033 __ push(rscratch1);
2034 __ push(rscratch2);
2035 __ push(rscratch3);
2036 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2037 __ atomic_add(noreg, 1, rscratch3);
2038 __ pop(rscratch3);
2039 __ pop(rscratch2);
2040 __ pop(rscratch1);
2041 }
2042
histogram_bytecode(Template * t)2043 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2044
histogram_bytecode_pair(Template * t)2045 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2046
2047
trace_bytecode(Template * t)2048 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2049 // Call a little run-time stub to avoid blow-up for each bytecode.
2050 // The run-time runtime saves the right registers, depending on
2051 // the tosca in-state for the given template.
2052
2053 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2054 "entry must have been generated");
2055 __ bl(Interpreter::trace_code(t->tos_in()));
2056 __ reinit_heapbase();
2057 }
2058
2059
stop_interpreter_at()2060 void TemplateInterpreterGenerator::stop_interpreter_at() {
2061 Label L;
2062 __ push(rscratch1);
2063 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2064 __ ldr(rscratch1, Address(rscratch1));
2065 __ mov(rscratch2, StopInterpreterAt);
2066 __ cmpw(rscratch1, rscratch2);
2067 __ br(Assembler::NE, L);
2068 __ brk(0);
2069 __ bind(L);
2070 __ pop(rscratch1);
2071 }
2072
2073 #endif // !PRODUCT
2074