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 "classfile/javaClasses.hpp"
29 #include "gc/shared/barrierSetAssembler.hpp"
30 #include "interpreter/bytecodeHistogram.hpp"
31 #include "interpreter/interpreter.hpp"
32 #include "interpreter/interpreterRuntime.hpp"
33 #include "interpreter/interp_masm.hpp"
34 #include "interpreter/templateInterpreterGenerator.hpp"
35 #include "interpreter/templateTable.hpp"
36 #include "interpreter/bytecodeTracer.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/arrayOop.hpp"
39 #include "oops/methodData.hpp"
40 #include "oops/method.hpp"
41 #include "oops/oop.inline.hpp"
42 #include "prims/jvmtiExport.hpp"
43 #include "prims/jvmtiThreadState.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/deoptimization.hpp"
46 #include "runtime/frame.inline.hpp"
47 #include "runtime/sharedRuntime.hpp"
48 #include "runtime/stubRoutines.hpp"
49 #include "runtime/synchronizer.hpp"
50 #include "runtime/timer.hpp"
51 #include "runtime/vframeArray.hpp"
52 #include "utilities/debug.hpp"
53 #include "utilities/powerOfTwo.hpp"
54 #include <sys/types.h>
55
56 #ifndef PRODUCT
57 #include "oops/method.hpp"
58 #endif // !PRODUCT
59
60 // Size of interpreter code. Increase if too small. Interpreter will
61 // fail with a guarantee ("not enough space for interpreter generation");
62 // if too small.
63 // Run with +PrintInterpreter to get the VM to print out the size.
64 // Max size with JVMTI
65 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
66
67 #define __ _masm->
68
69 //-----------------------------------------------------------------------------
70
71 extern "C" void entry(CodeBuffer*);
72
73 //-----------------------------------------------------------------------------
74
generate_slow_signature_handler()75 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
76 address entry = __ pc();
77
78 __ andr(esp, esp, -16);
79 __ mov(c_rarg3, esp);
80 // rmethod
81 // rlocals
82 // c_rarg3: first stack arg - wordSize
83
84 // adjust sp
85 __ sub(sp, c_rarg3, 18 * wordSize);
86 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
87 __ call_VM(noreg,
88 CAST_FROM_FN_PTR(address,
89 InterpreterRuntime::slow_signature_handler),
90 rmethod, rlocals, c_rarg3);
91
92 // r0: result handler
93
94 // Stack layout:
95 // rsp: return address <- sp
96 // 1 garbage
97 // 8 integer args (if static first is unused)
98 // 1 float/double identifiers
99 // 8 double args
100 // stack args <- esp
101 // garbage
102 // expression stack bottom
103 // bcp (NULL)
104 // ...
105
106 // Restore LR
107 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
108
109 // Do FP first so we can use c_rarg3 as temp
110 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
111
112 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
113 const FloatRegister r = as_FloatRegister(i);
114
115 Label d, done;
116
117 __ tbnz(c_rarg3, i, d);
118 __ ldrs(r, Address(sp, (10 + i) * wordSize));
119 __ b(done);
120 __ bind(d);
121 __ ldrd(r, Address(sp, (10 + i) * wordSize));
122 __ bind(done);
123 }
124
125 // c_rarg0 contains the result from the call of
126 // InterpreterRuntime::slow_signature_handler so we don't touch it
127 // here. It will be loaded with the JNIEnv* later.
128 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
129 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
130 Register rm = as_Register(i), rn = as_Register(i+1);
131 __ ldp(rm, rn, Address(sp, i * wordSize));
132 }
133
134 __ add(sp, sp, 18 * wordSize);
135 __ ret(lr);
136
137 return entry;
138 }
139
140
141 //
142 // Various method entries
143 //
144
generate_math_entry(AbstractInterpreter::MethodKind kind)145 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
146 // rmethod: Method*
147 // r13: sender sp
148 // esp: args
149
150 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
151
152 // These don't need a safepoint check because they aren't virtually
153 // callable. We won't enter these intrinsics from compiled code.
154 // If in the future we added an intrinsic which was virtually callable
155 // we'd have to worry about how to safepoint so that this code is used.
156
157 // mathematical functions inlined by compiler
158 // (interpreter must provide identical implementation
159 // in order to avoid monotonicity bugs when switching
160 // from interpreter to compiler in the middle of some
161 // computation)
162 //
163 // stack:
164 // [ arg ] <-- esp
165 // [ arg ]
166 // retaddr in lr
167
168 address entry_point = NULL;
169 Register continuation = lr;
170 switch (kind) {
171 case Interpreter::java_lang_math_abs:
172 entry_point = __ pc();
173 __ ldrd(v0, Address(esp));
174 __ fabsd(v0, v0);
175 __ mov(sp, r13); // Restore caller's SP
176 break;
177 case Interpreter::java_lang_math_sqrt:
178 entry_point = __ pc();
179 __ ldrd(v0, Address(esp));
180 __ fsqrtd(v0, v0);
181 __ mov(sp, r13);
182 break;
183 case Interpreter::java_lang_math_sin :
184 case Interpreter::java_lang_math_cos :
185 case Interpreter::java_lang_math_tan :
186 case Interpreter::java_lang_math_log :
187 case Interpreter::java_lang_math_log10 :
188 case Interpreter::java_lang_math_exp :
189 entry_point = __ pc();
190 __ ldrd(v0, Address(esp));
191 __ mov(sp, r13);
192 __ mov(r19, lr);
193 continuation = r19; // The first callee-saved register
194 generate_transcendental_entry(kind, 1);
195 break;
196 case Interpreter::java_lang_math_pow :
197 entry_point = __ pc();
198 __ mov(r19, lr);
199 continuation = r19;
200 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
201 __ ldrd(v1, Address(esp));
202 __ mov(sp, r13);
203 generate_transcendental_entry(kind, 2);
204 break;
205 case Interpreter::java_lang_math_fmaD :
206 if (UseFMA) {
207 entry_point = __ pc();
208 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
209 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
210 __ ldrd(v2, Address(esp));
211 __ fmaddd(v0, v0, v1, v2);
212 __ mov(sp, r13); // Restore caller's SP
213 }
214 break;
215 case Interpreter::java_lang_math_fmaF :
216 if (UseFMA) {
217 entry_point = __ pc();
218 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
219 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
220 __ ldrs(v2, Address(esp));
221 __ fmadds(v0, v0, v1, v2);
222 __ mov(sp, r13); // Restore caller's SP
223 }
224 break;
225 default:
226 ;
227 }
228 if (entry_point) {
229 __ br(continuation);
230 }
231
232 return entry_point;
233 }
234
235 // double trigonometrics and transcendentals
236 // static jdouble dsin(jdouble x);
237 // static jdouble dcos(jdouble x);
238 // static jdouble dtan(jdouble x);
239 // static jdouble dlog(jdouble x);
240 // static jdouble dlog10(jdouble x);
241 // static jdouble dexp(jdouble x);
242 // static jdouble dpow(jdouble x, jdouble y);
243
generate_transcendental_entry(AbstractInterpreter::MethodKind kind,int fpargs)244 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
245 address fn;
246 switch (kind) {
247 case Interpreter::java_lang_math_sin :
248 if (StubRoutines::dsin() == NULL) {
249 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
250 } else {
251 fn = CAST_FROM_FN_PTR(address, StubRoutines::dsin());
252 }
253 break;
254 case Interpreter::java_lang_math_cos :
255 if (StubRoutines::dcos() == NULL) {
256 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
257 } else {
258 fn = CAST_FROM_FN_PTR(address, StubRoutines::dcos());
259 }
260 break;
261 case Interpreter::java_lang_math_tan :
262 if (StubRoutines::dtan() == NULL) {
263 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
264 } else {
265 fn = CAST_FROM_FN_PTR(address, StubRoutines::dtan());
266 }
267 break;
268 case Interpreter::java_lang_math_log :
269 if (StubRoutines::dlog() == NULL) {
270 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
271 } else {
272 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog());
273 }
274 break;
275 case Interpreter::java_lang_math_log10 :
276 if (StubRoutines::dlog10() == NULL) {
277 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
278 } else {
279 fn = CAST_FROM_FN_PTR(address, StubRoutines::dlog10());
280 }
281 break;
282 case Interpreter::java_lang_math_exp :
283 if (StubRoutines::dexp() == NULL) {
284 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
285 } else {
286 fn = CAST_FROM_FN_PTR(address, StubRoutines::dexp());
287 }
288 break;
289 case Interpreter::java_lang_math_pow :
290 if (StubRoutines::dpow() == NULL) {
291 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
292 } else {
293 fn = CAST_FROM_FN_PTR(address, StubRoutines::dpow());
294 }
295 break;
296 default:
297 ShouldNotReachHere();
298 fn = NULL; // unreachable
299 }
300 __ mov(rscratch1, fn);
301 __ blr(rscratch1);
302 }
303
304 // Abstract method entry
305 // Attempt to execute abstract method. Throw exception
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 __ resolve(IS_NOT_NULL, r0);
828 }
829
830 // add space for monitor & lock
831 __ sub(sp, sp, entry_size); // add space for a monitor entry
832 __ sub(esp, esp, entry_size);
833 __ mov(rscratch1, esp);
834 __ str(rscratch1, monitor_block_top); // set new monitor block top
835 // store object
836 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
837 __ mov(c_rarg1, esp); // object address
838 __ lock_object(c_rarg1);
839 }
840
841 // Generate a fixed interpreter frame. This is identical setup for
842 // interpreted methods and for native methods hence the shared code.
843 //
844 // Args:
845 // lr: return address
846 // rmethod: Method*
847 // rlocals: pointer to locals
848 // rcpool: cp cache
849 // stack_pointer: previous sp
generate_fixed_frame(bool native_call)850 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
851 // initialize fixed part of activation frame
852 if (native_call) {
853 __ sub(esp, sp, 14 * wordSize);
854 __ mov(rbcp, zr);
855 __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
856 // add 2 zero-initialized slots for native calls
857 __ stp(zr, zr, Address(sp, 12 * wordSize));
858 } else {
859 __ sub(esp, sp, 12 * wordSize);
860 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
861 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
862 __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
863 }
864
865 if (ProfileInterpreter) {
866 Label method_data_continue;
867 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
868 __ cbz(rscratch1, method_data_continue);
869 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
870 __ bind(method_data_continue);
871 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
872 } else {
873 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
874 }
875
876 // Get mirror and store it in the frame as GC root for this Method*
877 __ load_mirror(r10, rmethod);
878 __ stp(r10, zr, Address(sp, 4 * wordSize));
879
880 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
881 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
882 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
883 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
884
885 __ stp(rfp, lr, Address(sp, 10 * wordSize));
886 __ lea(rfp, Address(sp, 10 * wordSize));
887
888 // set sender sp
889 // leave last_sp as null
890 __ stp(zr, r13, Address(sp, 8 * wordSize));
891
892 // Move SP out of the way
893 if (! native_call) {
894 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
895 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
896 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
897 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
898 __ andr(sp, rscratch1, -16);
899 }
900 }
901
902 // End of helpers
903
904 // Various method entries
905 //------------------------------------------------------------------------------------------------------------------------
906 //
907 //
908
909 // Method entry for java.lang.ref.Reference.get.
generate_Reference_get_entry(void)910 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
911 // Code: _aload_0, _getfield, _areturn
912 // parameter size = 1
913 //
914 // The code that gets generated by this routine is split into 2 parts:
915 // 1. The "intrinsified" code for G1 (or any SATB based GC),
916 // 2. The slow path - which is an expansion of the regular method entry.
917 //
918 // Notes:-
919 // * In the G1 code we do not check whether we need to block for
920 // a safepoint. If G1 is enabled then we must execute the specialized
921 // code for Reference.get (except when the Reference object is null)
922 // so that we can log the value in the referent field with an SATB
923 // update buffer.
924 // If the code for the getfield template is modified so that the
925 // G1 pre-barrier code is executed when the current method is
926 // Reference.get() then going through the normal method entry
927 // will be fine.
928 // * The G1 code can, however, check the receiver object (the instance
929 // of java.lang.Reference) and jump to the slow path if null. If the
930 // Reference object is null then we obviously cannot fetch the referent
931 // and so we don't need to call the G1 pre-barrier. Thus we can use the
932 // regular method entry code to generate the NPE.
933 //
934 // This code is based on generate_accessor_entry.
935 //
936 // rmethod: Method*
937 // r13: senderSP must preserve for slow path, set SP to it on fast path
938
939 // LR is live. It must be saved around calls.
940
941 address entry = __ pc();
942
943 const int referent_offset = java_lang_ref_Reference::referent_offset();
944
945 Label slow_path;
946 const Register local_0 = c_rarg0;
947 // Check if local 0 != NULL
948 // If the receiver is null then it is OK to jump to the slow path.
949 __ ldr(local_0, Address(esp, 0));
950 __ cbz(local_0, slow_path);
951
952 __ mov(r19, r13); // Move senderSP to a callee-saved register
953
954 // Load the value of the referent field.
955 const Address field_address(local_0, referent_offset);
956 BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
957 bs->load_at(_masm, IN_HEAP | ON_WEAK_OOP_REF, T_OBJECT, local_0, field_address, /*tmp1*/ rscratch2, /*tmp2*/ rscratch1);
958
959 // areturn
960 __ andr(sp, r19, -16); // done with stack
961 __ ret(lr);
962
963 // generate a vanilla interpreter entry as the slow path
964 __ bind(slow_path);
965 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
966 return entry;
967
968 }
969
970 /**
971 * Method entry for static native methods:
972 * int java.util.zip.CRC32.update(int crc, int b)
973 */
generate_CRC32_update_entry()974 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
975 if (UseCRC32Intrinsics) {
976 address entry = __ pc();
977
978 // rmethod: Method*
979 // r13: senderSP must preserved for slow path
980 // esp: args
981
982 Label slow_path;
983 // If we need a safepoint check, generate full interpreter entry.
984 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
985
986 // We don't generate local frame and don't align stack because
987 // we call stub code and there is no safepoint on this path.
988
989 // Load parameters
990 const Register crc = c_rarg0; // crc
991 const Register val = c_rarg1; // source java byte value
992 const Register tbl = c_rarg2; // scratch
993
994 // Arguments are reversed on java expression stack
995 __ ldrw(val, Address(esp, 0)); // byte value
996 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
997
998 uint64_t offset;
999 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
1000 __ add(tbl, tbl, offset);
1001
1002 __ mvnw(crc, crc); // ~crc
1003 __ update_byte_crc32(crc, val, tbl);
1004 __ mvnw(crc, crc); // ~crc
1005
1006 // result in c_rarg0
1007
1008 __ andr(sp, r13, -16);
1009 __ ret(lr);
1010
1011 // generate a vanilla native entry as the slow path
1012 __ bind(slow_path);
1013 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1014 return entry;
1015 }
1016 return NULL;
1017 }
1018
1019 /**
1020 * Method entry for static native methods:
1021 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1022 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1023 */
generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind)1024 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1025 if (UseCRC32Intrinsics) {
1026 address entry = __ pc();
1027
1028 // rmethod,: Method*
1029 // r13: senderSP must preserved for slow path
1030
1031 Label slow_path;
1032 // If we need a safepoint check, generate full interpreter entry.
1033 __ safepoint_poll(slow_path, false /* at_return */, false /* acquire */, false /* in_nmethod */);
1034
1035 // We don't generate local frame and don't align stack because
1036 // we call stub code and there is no safepoint on this path.
1037
1038 // Load parameters
1039 const Register crc = c_rarg0; // crc
1040 const Register buf = c_rarg1; // source java byte array address
1041 const Register len = c_rarg2; // length
1042 const Register off = len; // offset (never overlaps with 'len')
1043
1044 // Arguments are reversed on java expression stack
1045 // Calculate address of start element
1046 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1047 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1048 __ ldrw(off, Address(esp, wordSize)); // offset
1049 __ add(buf, buf, off); // + offset
1050 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1051 } else {
1052 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1053 __ resolve(IS_NOT_NULL | ACCESS_READ, buf);
1054 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1055 __ ldrw(off, Address(esp, wordSize)); // offset
1056 __ add(buf, buf, off); // + offset
1057 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1058 }
1059 // Can now load 'len' since we're finished with 'off'
1060 __ ldrw(len, Address(esp, 0x0)); // Length
1061
1062 __ andr(sp, r13, -16); // Restore the caller's SP
1063
1064 // We are frameless so we can just jump to the stub.
1065 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1066
1067 // generate a vanilla native entry as the slow path
1068 __ bind(slow_path);
1069 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1070 return entry;
1071 }
1072 return NULL;
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 */
generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind)1082 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1083 if (UseCRC32CIntrinsics) {
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 if (kind == Interpreter::java_util_zip_CRC32C_updateBytes) {
1099 __ resolve(IS_NOT_NULL | ACCESS_READ, buf);
1100 }
1101 __ add(buf, buf, off); // + offset
1102 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1103 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1104 } else {
1105 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1106 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1107 }
1108
1109 __ andr(sp, r13, -16); // Restore the caller's SP
1110
1111 // Jump to the stub.
1112 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1113
1114 return entry;
1115 }
1116 return NULL;
1117 }
1118
bang_stack_shadow_pages(bool native_call)1119 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1120 // Bang each page in the shadow zone. We can't assume it's been done for
1121 // an interpreter frame with greater than a page of locals, so each page
1122 // needs to be checked. Only true for non-native.
1123 if (UseStackBanging) {
1124 const int n_shadow_pages = (int)(StackOverflow::stack_shadow_zone_size() / os::vm_page_size());
1125 const int start_page = native_call ? n_shadow_pages : 1;
1126 const int page_size = os::vm_page_size();
1127 for (int pages = start_page; pages <= n_shadow_pages ; pages++) {
1128 __ sub(rscratch2, sp, pages*page_size);
1129 __ str(zr, Address(rscratch2));
1130 }
1131 }
1132 }
1133
1134
1135 // Interpreter stub for calling a native method. (asm interpreter)
1136 // This sets up a somewhat different looking stack for calling the
1137 // native method than the typical interpreter frame setup.
generate_native_entry(bool synchronized)1138 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1139 // determine code generation flags
1140 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1141
1142 // r1: Method*
1143 // rscratch1: sender sp
1144
1145 address entry_point = __ pc();
1146
1147 const Address constMethod (rmethod, Method::const_offset());
1148 const Address access_flags (rmethod, Method::access_flags_offset());
1149 const Address size_of_parameters(r2, ConstMethod::
1150 size_of_parameters_offset());
1151
1152 // get parameter size (always needed)
1153 __ ldr(r2, constMethod);
1154 __ load_unsigned_short(r2, size_of_parameters);
1155
1156 // Native calls don't need the stack size check since they have no
1157 // expression stack and the arguments are already on the stack and
1158 // we only add a handful of words to the stack.
1159
1160 // rmethod: Method*
1161 // r2: size of parameters
1162 // rscratch1: sender sp
1163
1164 // for natives the size of locals is zero
1165
1166 // compute beginning of parameters (rlocals)
1167 __ add(rlocals, esp, r2, ext::uxtx, 3);
1168 __ add(rlocals, rlocals, -wordSize);
1169
1170 // Pull SP back to minimum size: this avoids holes in the stack
1171 __ andr(sp, esp, -16);
1172
1173 // initialize fixed part of activation frame
1174 generate_fixed_frame(true);
1175
1176 // make sure method is native & not abstract
1177 #ifdef ASSERT
1178 __ ldrw(r0, access_flags);
1179 {
1180 Label L;
1181 __ tst(r0, JVM_ACC_NATIVE);
1182 __ br(Assembler::NE, L);
1183 __ stop("tried to execute non-native method as native");
1184 __ bind(L);
1185 }
1186 {
1187 Label L;
1188 __ tst(r0, JVM_ACC_ABSTRACT);
1189 __ br(Assembler::EQ, L);
1190 __ stop("tried to execute abstract method in interpreter");
1191 __ bind(L);
1192 }
1193 #endif
1194
1195 // Since at this point in the method invocation the exception
1196 // handler would try to exit the monitor of synchronized methods
1197 // which hasn't been entered yet, we set the thread local variable
1198 // _do_not_unlock_if_synchronized to true. The remove_activation
1199 // will check this flag.
1200
1201 const Address do_not_unlock_if_synchronized(rthread,
1202 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1203 __ mov(rscratch2, true);
1204 __ strb(rscratch2, do_not_unlock_if_synchronized);
1205
1206 // increment invocation count & check for overflow
1207 Label invocation_counter_overflow;
1208 if (inc_counter) {
1209 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1210 }
1211
1212 Label continue_after_compile;
1213 __ bind(continue_after_compile);
1214
1215 bang_stack_shadow_pages(true);
1216
1217 // reset the _do_not_unlock_if_synchronized flag
1218 __ strb(zr, do_not_unlock_if_synchronized);
1219
1220 // check for synchronized methods
1221 // Must happen AFTER invocation_counter check and stack overflow check,
1222 // so method is not locked if overflows.
1223 if (synchronized) {
1224 lock_method();
1225 } else {
1226 // no synchronization necessary
1227 #ifdef ASSERT
1228 {
1229 Label L;
1230 __ ldrw(r0, access_flags);
1231 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1232 __ br(Assembler::EQ, L);
1233 __ stop("method needs synchronization");
1234 __ bind(L);
1235 }
1236 #endif
1237 }
1238
1239 // start execution
1240 #ifdef ASSERT
1241 {
1242 Label L;
1243 const Address monitor_block_top(rfp,
1244 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1245 __ ldr(rscratch1, monitor_block_top);
1246 __ cmp(esp, rscratch1);
1247 __ br(Assembler::EQ, L);
1248 __ stop("broken stack frame setup in interpreter");
1249 __ bind(L);
1250 }
1251 #endif
1252
1253 // jvmti support
1254 __ notify_method_entry();
1255
1256 // work registers
1257 const Register t = r17;
1258 const Register result_handler = r19;
1259
1260 // allocate space for parameters
1261 __ ldr(t, Address(rmethod, Method::const_offset()));
1262 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1263
1264 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1265 __ andr(sp, rscratch1, -16);
1266 __ mov(esp, rscratch1);
1267
1268 // get signature handler
1269 {
1270 Label L;
1271 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1272 __ cbnz(t, L);
1273 __ call_VM(noreg,
1274 CAST_FROM_FN_PTR(address,
1275 InterpreterRuntime::prepare_native_call),
1276 rmethod);
1277 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1278 __ bind(L);
1279 }
1280
1281 // call signature handler
1282 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1283 "adjust this code");
1284 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1285 "adjust this code");
1286 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1287 "adjust this code");
1288
1289 // The generated handlers do not touch rmethod (the method).
1290 // However, large signatures cannot be cached and are generated
1291 // each time here. The slow-path generator can do a GC on return,
1292 // so we must reload it after the call.
1293 __ blr(t);
1294 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1295
1296
1297 // result handler is in r0
1298 // set result handler
1299 __ mov(result_handler, r0);
1300 // pass mirror handle if static call
1301 {
1302 Label L;
1303 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1304 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1305 // get mirror
1306 __ load_mirror(t, rmethod);
1307 // copy mirror into activation frame
1308 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1309 // pass handle to mirror
1310 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1311 __ bind(L);
1312 }
1313
1314 // get native function entry point in r10
1315 {
1316 Label L;
1317 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1318 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1319 __ mov(rscratch2, unsatisfied);
1320 __ ldr(rscratch2, rscratch2);
1321 __ cmp(r10, rscratch2);
1322 __ br(Assembler::NE, L);
1323 __ call_VM(noreg,
1324 CAST_FROM_FN_PTR(address,
1325 InterpreterRuntime::prepare_native_call),
1326 rmethod);
1327 __ get_method(rmethod);
1328 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1329 __ bind(L);
1330 }
1331
1332 // pass JNIEnv
1333 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1334
1335 // Set the last Java PC in the frame anchor to be the return address from
1336 // the call to the native method: this will allow the debugger to
1337 // generate an accurate stack trace.
1338 Label native_return;
1339 __ set_last_Java_frame(esp, rfp, native_return, rscratch1);
1340
1341 // change thread state
1342 #ifdef ASSERT
1343 {
1344 Label L;
1345 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1346 __ cmp(t, (u1)_thread_in_Java);
1347 __ br(Assembler::EQ, L);
1348 __ stop("Wrong thread state in native stub");
1349 __ bind(L);
1350 }
1351 #endif
1352
1353 // Change state to native
1354 __ mov(rscratch1, _thread_in_native);
1355 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1356 __ stlrw(rscratch1, rscratch2);
1357
1358 // Call the native method.
1359 __ blr(r10);
1360 __ bind(native_return);
1361 __ get_method(rmethod);
1362 // result potentially in r0 or v0
1363
1364 // make room for the pushes we're about to do
1365 __ sub(rscratch1, esp, 4 * wordSize);
1366 __ andr(sp, rscratch1, -16);
1367
1368 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1369 // in order to extract the result of a method call. If the order of these
1370 // pushes change or anything else is added to the stack then the code in
1371 // interpreter_frame_result must also change.
1372 __ push(dtos);
1373 __ push(ltos);
1374
1375 if (UseSVE > 0) {
1376 // Make sure that jni code does not change SVE vector length.
1377 __ verify_sve_vector_length();
1378 }
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 // Force this write out before the read below
1386 __ dmb(Assembler::ISH);
1387
1388 // check for safepoint operation in progress and/or pending suspend requests
1389 {
1390 Label L, Continue;
1391
1392 // We need an acquire here to ensure that any subsequent load of the
1393 // global SafepointSynchronize::_state flag is ordered after this load
1394 // of the thread-local polling word. We don't want this poll to
1395 // return false (i.e. not safepointing) and a later poll of the global
1396 // SafepointSynchronize::_state spuriously to return true.
1397 //
1398 // This is to avoid a race when we're in a native->Java transition
1399 // racing the code which wakes up from a safepoint.
1400 __ safepoint_poll(L, true /* at_return */, true /* acquire */, false /* in_nmethod */);
1401 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1402 __ cbz(rscratch2, Continue);
1403 __ bind(L);
1404
1405 // Don't use call_VM as it will see a possible pending exception
1406 // and forward it and never return here preventing us from
1407 // clearing _last_native_pc down below. So we do a runtime call by
1408 // hand.
1409 //
1410 __ mov(c_rarg0, rthread);
1411 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1412 __ blr(rscratch2);
1413 __ get_method(rmethod);
1414 __ reinit_heapbase();
1415 __ bind(Continue);
1416 }
1417
1418 // change thread state
1419 __ mov(rscratch1, _thread_in_Java);
1420 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1421 __ stlrw(rscratch1, rscratch2);
1422
1423 // reset_last_Java_frame
1424 __ reset_last_Java_frame(true);
1425
1426 if (CheckJNICalls) {
1427 // clear_pending_jni_exception_check
1428 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1429 }
1430
1431 // reset handle block
1432 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1433 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1434
1435 // If result is an oop unbox and store it in frame where gc will see it
1436 // and result handler will pick it up
1437
1438 {
1439 Label no_oop;
1440 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1441 __ cmp(t, result_handler);
1442 __ br(Assembler::NE, no_oop);
1443 // Unbox oop result, e.g. JNIHandles::resolve result.
1444 __ pop(ltos);
1445 __ resolve_jobject(r0, rthread, t);
1446 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1447 // keep stack depth as expected by pushing oop which will eventually be discarded
1448 __ push(ltos);
1449 __ bind(no_oop);
1450 }
1451
1452 {
1453 Label no_reguard;
1454 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1455 __ ldrw(rscratch1, Address(rscratch1));
1456 __ cmp(rscratch1, (u1)StackOverflow::stack_guard_yellow_reserved_disabled);
1457 __ br(Assembler::NE, no_reguard);
1458
1459 __ pusha(); // XXX only save smashed registers
1460 __ mov(c_rarg0, rthread);
1461 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1462 __ blr(rscratch2);
1463 __ popa(); // XXX only restore smashed registers
1464 __ bind(no_reguard);
1465 }
1466
1467 // The method register is junk from after the thread_in_native transition
1468 // until here. Also can't call_VM until the bcp has been
1469 // restored. Need bcp for throwing exception below so get it now.
1470 __ get_method(rmethod);
1471
1472 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1473 // rbcp == code_base()
1474 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1475 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1476 // handle exceptions (exception handling will handle unlocking!)
1477 {
1478 Label L;
1479 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1480 __ cbz(rscratch1, L);
1481 // Note: At some point we may want to unify this with the code
1482 // used in call_VM_base(); i.e., we should use the
1483 // StubRoutines::forward_exception code. For now this doesn't work
1484 // here because the rsp is not correctly set at this point.
1485 __ MacroAssembler::call_VM(noreg,
1486 CAST_FROM_FN_PTR(address,
1487 InterpreterRuntime::throw_pending_exception));
1488 __ should_not_reach_here();
1489 __ bind(L);
1490 }
1491
1492 // do unlocking if necessary
1493 {
1494 Label L;
1495 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1496 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1497 // the code below should be shared with interpreter macro
1498 // assembler implementation
1499 {
1500 Label unlock;
1501 // BasicObjectLock will be first in list, since this is a
1502 // synchronized method. However, need to check that the object
1503 // has not been unlocked by an explicit monitorexit bytecode.
1504
1505 // monitor expect in c_rarg1 for slow unlock path
1506 __ lea (c_rarg1, Address(rfp, // address of first monitor
1507 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1508 wordSize - sizeof(BasicObjectLock))));
1509
1510 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1511 __ cbnz(t, unlock);
1512
1513 // Entry already unlocked, need to throw exception
1514 __ MacroAssembler::call_VM(noreg,
1515 CAST_FROM_FN_PTR(address,
1516 InterpreterRuntime::throw_illegal_monitor_state_exception));
1517 __ should_not_reach_here();
1518
1519 __ bind(unlock);
1520 __ unlock_object(c_rarg1);
1521 }
1522 __ bind(L);
1523 }
1524
1525 // jvmti support
1526 // Note: This must happen _after_ handling/throwing any exceptions since
1527 // the exception handler code notifies the runtime of method exits
1528 // too. If this happens before, method entry/exit notifications are
1529 // not properly paired (was bug - gri 11/22/99).
1530 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1531
1532 // restore potential result in r0:d0, call result handler to
1533 // restore potential result in ST0 & handle result
1534
1535 __ pop(ltos);
1536 __ pop(dtos);
1537
1538 __ blr(result_handler);
1539
1540 // remove activation
1541 __ ldr(esp, Address(rfp,
1542 frame::interpreter_frame_sender_sp_offset *
1543 wordSize)); // get sender sp
1544 // remove frame anchor
1545 __ leave();
1546
1547 // resture sender sp
1548 __ mov(sp, esp);
1549
1550 __ ret(lr);
1551
1552 if (inc_counter) {
1553 // Handle overflow of counter and compile method
1554 __ bind(invocation_counter_overflow);
1555 generate_counter_overflow(continue_after_compile);
1556 }
1557
1558 return entry_point;
1559 }
1560
1561 //
1562 // Generic interpreted method entry to (asm) interpreter
1563 //
generate_normal_entry(bool synchronized)1564 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1565 // determine code generation flags
1566 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1567
1568 // rscratch1: sender sp
1569 address entry_point = __ pc();
1570
1571 const Address constMethod(rmethod, Method::const_offset());
1572 const Address access_flags(rmethod, Method::access_flags_offset());
1573 const Address size_of_parameters(r3,
1574 ConstMethod::size_of_parameters_offset());
1575 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1576
1577 // get parameter size (always needed)
1578 // need to load the const method first
1579 __ ldr(r3, constMethod);
1580 __ load_unsigned_short(r2, size_of_parameters);
1581
1582 // r2: size of parameters
1583
1584 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1585 __ sub(r3, r3, r2); // r3 = no. of additional locals
1586
1587 // see if we've got enough room on the stack for locals plus overhead.
1588 generate_stack_overflow_check();
1589
1590 // compute beginning of parameters (rlocals)
1591 __ add(rlocals, esp, r2, ext::uxtx, 3);
1592 __ sub(rlocals, rlocals, wordSize);
1593
1594 // Make room for locals
1595 __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1596
1597 // Padding between locals and fixed part of activation frame to ensure
1598 // SP is always 16-byte aligned.
1599 __ andr(sp, rscratch1, -16);
1600
1601 // r3 - # of additional locals
1602 // allocate space for locals
1603 // explicitly initialize locals
1604 {
1605 Label exit, loop;
1606 __ ands(zr, r3, r3);
1607 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1608 __ bind(loop);
1609 __ str(zr, Address(__ post(rscratch1, wordSize)));
1610 __ sub(r3, r3, 1); // until everything initialized
1611 __ cbnz(r3, loop);
1612 __ bind(exit);
1613 }
1614
1615 // And the base dispatch table
1616 __ get_dispatch();
1617
1618 // initialize fixed part of activation frame
1619 generate_fixed_frame(false);
1620
1621 // make sure method is not native & not abstract
1622 #ifdef ASSERT
1623 __ ldrw(r0, access_flags);
1624 {
1625 Label L;
1626 __ tst(r0, JVM_ACC_NATIVE);
1627 __ br(Assembler::EQ, L);
1628 __ stop("tried to execute native method as non-native");
1629 __ bind(L);
1630 }
1631 {
1632 Label L;
1633 __ tst(r0, JVM_ACC_ABSTRACT);
1634 __ br(Assembler::EQ, L);
1635 __ stop("tried to execute abstract method in interpreter");
1636 __ bind(L);
1637 }
1638 #endif
1639
1640 // Since at this point in the method invocation the exception
1641 // handler would try to exit the monitor of synchronized methods
1642 // which hasn't been entered yet, we set the thread local variable
1643 // _do_not_unlock_if_synchronized to true. The remove_activation
1644 // will check this flag.
1645
1646 const Address do_not_unlock_if_synchronized(rthread,
1647 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1648 __ mov(rscratch2, true);
1649 __ strb(rscratch2, do_not_unlock_if_synchronized);
1650
1651 Register mdp = r3;
1652 __ profile_parameters_type(mdp, r1, r2);
1653
1654 // increment invocation count & check for overflow
1655 Label invocation_counter_overflow;
1656 Label profile_method;
1657 Label profile_method_continue;
1658 if (inc_counter) {
1659 generate_counter_incr(&invocation_counter_overflow,
1660 &profile_method,
1661 &profile_method_continue);
1662 if (ProfileInterpreter) {
1663 __ bind(profile_method_continue);
1664 }
1665 }
1666
1667 Label continue_after_compile;
1668 __ bind(continue_after_compile);
1669
1670 bang_stack_shadow_pages(false);
1671
1672 // reset the _do_not_unlock_if_synchronized flag
1673 __ strb(zr, do_not_unlock_if_synchronized);
1674
1675 // check for synchronized methods
1676 // Must happen AFTER invocation_counter check and stack overflow check,
1677 // so method is not locked if overflows.
1678 if (synchronized) {
1679 // Allocate monitor and lock method
1680 lock_method();
1681 } else {
1682 // no synchronization necessary
1683 #ifdef ASSERT
1684 {
1685 Label L;
1686 __ ldrw(r0, access_flags);
1687 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1688 __ br(Assembler::EQ, L);
1689 __ stop("method needs synchronization");
1690 __ bind(L);
1691 }
1692 #endif
1693 }
1694
1695 // start execution
1696 #ifdef ASSERT
1697 {
1698 Label L;
1699 const Address monitor_block_top (rfp,
1700 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1701 __ ldr(rscratch1, monitor_block_top);
1702 __ cmp(esp, rscratch1);
1703 __ br(Assembler::EQ, L);
1704 __ stop("broken stack frame setup in interpreter");
1705 __ bind(L);
1706 }
1707 #endif
1708
1709 // jvmti support
1710 __ notify_method_entry();
1711
1712 __ dispatch_next(vtos);
1713
1714 // invocation counter overflow
1715 if (inc_counter) {
1716 if (ProfileInterpreter) {
1717 // We have decided to profile this method in the interpreter
1718 __ bind(profile_method);
1719 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1720 __ set_method_data_pointer_for_bcp();
1721 // don't think we need this
1722 __ get_method(r1);
1723 __ b(profile_method_continue);
1724 }
1725 // Handle overflow of counter and compile method
1726 __ bind(invocation_counter_overflow);
1727 generate_counter_overflow(continue_after_compile);
1728 }
1729
1730 return entry_point;
1731 }
1732
1733 //-----------------------------------------------------------------------------
1734 // Exceptions
1735
generate_throw_exception()1736 void TemplateInterpreterGenerator::generate_throw_exception() {
1737 // Entry point in previous activation (i.e., if the caller was
1738 // interpreted)
1739 Interpreter::_rethrow_exception_entry = __ pc();
1740 // Restore sp to interpreter_frame_last_sp even though we are going
1741 // to empty the expression stack for the exception processing.
1742 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1743 // r0: exception
1744 // r3: return address/pc that threw exception
1745 __ restore_bcp(); // rbcp points to call/send
1746 __ restore_locals();
1747 __ restore_constant_pool_cache();
1748 __ reinit_heapbase(); // restore rheapbase as heapbase.
1749 __ get_dispatch();
1750
1751 // Entry point for exceptions thrown within interpreter code
1752 Interpreter::_throw_exception_entry = __ pc();
1753 // If we came here via a NullPointerException on the receiver of a
1754 // method, rmethod may be corrupt.
1755 __ get_method(rmethod);
1756 // expression stack is undefined here
1757 // r0: exception
1758 // rbcp: exception bcp
1759 __ verify_oop(r0);
1760 __ mov(c_rarg1, r0);
1761
1762 // expression stack must be empty before entering the VM in case of
1763 // an exception
1764 __ empty_expression_stack();
1765 // find exception handler address and preserve exception oop
1766 __ call_VM(r3,
1767 CAST_FROM_FN_PTR(address,
1768 InterpreterRuntime::exception_handler_for_exception),
1769 c_rarg1);
1770
1771 // Calculate stack limit
1772 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1773 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1774 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1775 __ ldr(rscratch2,
1776 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1777 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1778 __ andr(sp, rscratch1, -16);
1779
1780 // r0: exception handler entry point
1781 // r3: preserved exception oop
1782 // rbcp: bcp for exception handler
1783 __ push_ptr(r3); // push exception which is now the only value on the stack
1784 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1785
1786 // If the exception is not handled in the current frame the frame is
1787 // removed and the exception is rethrown (i.e. exception
1788 // continuation is _rethrow_exception).
1789 //
1790 // Note: At this point the bci is still the bxi for the instruction
1791 // which caused the exception and the expression stack is
1792 // empty. Thus, for any VM calls at this point, GC will find a legal
1793 // oop map (with empty expression stack).
1794
1795 //
1796 // JVMTI PopFrame support
1797 //
1798
1799 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1800 __ empty_expression_stack();
1801 // Set the popframe_processing bit in pending_popframe_condition
1802 // indicating that we are currently handling popframe, so that
1803 // call_VMs that may happen later do not trigger new popframe
1804 // handling cycles.
1805 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1806 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1807 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1808
1809 {
1810 // Check to see whether we are returning to a deoptimized frame.
1811 // (The PopFrame call ensures that the caller of the popped frame is
1812 // either interpreted or compiled and deoptimizes it if compiled.)
1813 // In this case, we can't call dispatch_next() after the frame is
1814 // popped, but instead must save the incoming arguments and restore
1815 // them after deoptimization has occurred.
1816 //
1817 // Note that we don't compare the return PC against the
1818 // deoptimization blob's unpack entry because of the presence of
1819 // adapter frames in C2.
1820 Label caller_not_deoptimized;
1821 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1822 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1823 InterpreterRuntime::interpreter_contains), c_rarg1);
1824 __ cbnz(r0, caller_not_deoptimized);
1825
1826 // Compute size of arguments for saving when returning to
1827 // deoptimized caller
1828 __ get_method(r0);
1829 __ ldr(r0, Address(r0, Method::const_offset()));
1830 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1831 size_of_parameters_offset())));
1832 __ lsl(r0, r0, Interpreter::logStackElementSize);
1833 __ restore_locals(); // XXX do we need this?
1834 __ sub(rlocals, rlocals, r0);
1835 __ add(rlocals, rlocals, wordSize);
1836 // Save these arguments
1837 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1838 Deoptimization::
1839 popframe_preserve_args),
1840 rthread, r0, rlocals);
1841
1842 __ remove_activation(vtos,
1843 /* throw_monitor_exception */ false,
1844 /* install_monitor_exception */ false,
1845 /* notify_jvmdi */ false);
1846
1847 // Inform deoptimization that it is responsible for restoring
1848 // these arguments
1849 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1850 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1851
1852 // Continue in deoptimization handler
1853 __ ret(lr);
1854
1855 __ bind(caller_not_deoptimized);
1856 }
1857
1858 __ remove_activation(vtos,
1859 /* throw_monitor_exception */ false,
1860 /* install_monitor_exception */ false,
1861 /* notify_jvmdi */ false);
1862
1863 // Restore the last_sp and null it out
1864 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1865 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1866
1867 __ restore_bcp();
1868 __ restore_locals();
1869 __ restore_constant_pool_cache();
1870 __ get_method(rmethod);
1871 __ get_dispatch();
1872
1873 // The method data pointer was incremented already during
1874 // call profiling. We have to restore the mdp for the current bcp.
1875 if (ProfileInterpreter) {
1876 __ set_method_data_pointer_for_bcp();
1877 }
1878
1879 // Clear the popframe condition flag
1880 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1881 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1882
1883 #if INCLUDE_JVMTI
1884 {
1885 Label L_done;
1886
1887 __ ldrb(rscratch1, Address(rbcp, 0));
1888 __ cmpw(rscratch1, Bytecodes::_invokestatic);
1889 __ br(Assembler::NE, L_done);
1890
1891 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1892 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1893
1894 __ ldr(c_rarg0, Address(rlocals, 0));
1895 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1896
1897 __ cbz(r0, L_done);
1898
1899 __ str(r0, Address(esp, 0));
1900 __ bind(L_done);
1901 }
1902 #endif // INCLUDE_JVMTI
1903
1904 // Restore machine SP
1905 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1906 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1907 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1908 __ ldr(rscratch2,
1909 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1910 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1911 __ andr(sp, rscratch1, -16);
1912
1913 __ dispatch_next(vtos);
1914 // end of PopFrame support
1915
1916 Interpreter::_remove_activation_entry = __ pc();
1917
1918 // preserve exception over this code sequence
1919 __ pop_ptr(r0);
1920 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1921 // remove the activation (without doing throws on illegalMonitorExceptions)
1922 __ remove_activation(vtos, false, true, false);
1923 // restore exception
1924 __ get_vm_result(r0, rthread);
1925
1926 // In between activations - previous activation type unknown yet
1927 // compute continuation point - the continuation point expects the
1928 // following registers set up:
1929 //
1930 // r0: exception
1931 // lr: return address/pc that threw exception
1932 // esp: expression stack of caller
1933 // rfp: fp of caller
1934 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1935 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1936 SharedRuntime::exception_handler_for_return_address),
1937 rthread, lr);
1938 __ mov(r1, r0); // save exception handler
1939 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1940 // We might be returning to a deopt handler that expects r3 to
1941 // contain the exception pc
1942 __ mov(r3, lr);
1943 // Note that an "issuing PC" is actually the next PC after the call
1944 __ br(r1); // jump to exception
1945 // handler of caller
1946 }
1947
1948
1949 //
1950 // JVMTI ForceEarlyReturn support
1951 //
generate_earlyret_entry_for(TosState state)1952 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1953 address entry = __ pc();
1954
1955 __ restore_bcp();
1956 __ restore_locals();
1957 __ empty_expression_stack();
1958 __ load_earlyret_value(state);
1959
1960 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1961 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1962
1963 // Clear the earlyret state
1964 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1965 __ str(zr, cond_addr);
1966
1967 __ remove_activation(state,
1968 false, /* throw_monitor_exception */
1969 false, /* install_monitor_exception */
1970 true); /* notify_jvmdi */
1971 __ ret(lr);
1972
1973 return entry;
1974 } // end of ForceEarlyReturn support
1975
1976
1977
1978 //-----------------------------------------------------------------------------
1979 // Helper for vtos entry point generation
1980
set_vtos_entry_points(Template * t,address & bep,address & cep,address & sep,address & aep,address & iep,address & lep,address & fep,address & dep,address & vep)1981 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1982 address& bep,
1983 address& cep,
1984 address& sep,
1985 address& aep,
1986 address& iep,
1987 address& lep,
1988 address& fep,
1989 address& dep,
1990 address& vep) {
1991 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1992 Label L;
1993 aep = __ pc(); __ push_ptr(); __ b(L);
1994 fep = __ pc(); __ push_f(); __ b(L);
1995 dep = __ pc(); __ push_d(); __ b(L);
1996 lep = __ pc(); __ push_l(); __ b(L);
1997 bep = cep = sep =
1998 iep = __ pc(); __ push_i();
1999 vep = __ pc();
2000 __ bind(L);
2001 generate_and_dispatch(t);
2002 }
2003
2004 //-----------------------------------------------------------------------------
2005
2006 // Non-product code
2007 #ifndef PRODUCT
generate_trace_code(TosState state)2008 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2009 address entry = __ pc();
2010
2011 __ push(lr);
2012 __ push(state);
2013 __ push(RegSet::range(r0, r15), sp);
2014 __ mov(c_rarg2, r0); // Pass itos
2015 __ call_VM(noreg,
2016 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2017 c_rarg1, c_rarg2, c_rarg3);
2018 __ pop(RegSet::range(r0, r15), sp);
2019 __ pop(state);
2020 __ pop(lr);
2021 __ ret(lr); // return from result handler
2022
2023 return entry;
2024 }
2025
count_bytecode()2026 void TemplateInterpreterGenerator::count_bytecode() {
2027 Register rscratch3 = r0;
2028 __ push(rscratch1);
2029 __ push(rscratch2);
2030 __ push(rscratch3);
2031 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2032 __ atomic_add(noreg, 1, rscratch3);
2033 __ pop(rscratch3);
2034 __ pop(rscratch2);
2035 __ pop(rscratch1);
2036 }
2037
histogram_bytecode(Template * t)2038 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2039
histogram_bytecode_pair(Template * t)2040 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2041
2042
trace_bytecode(Template * t)2043 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2044 // Call a little run-time stub to avoid blow-up for each bytecode.
2045 // The run-time runtime saves the right registers, depending on
2046 // the tosca in-state for the given template.
2047
2048 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2049 "entry must have been generated");
2050 __ bl(Interpreter::trace_code(t->tos_in()));
2051 __ reinit_heapbase();
2052 }
2053
2054
stop_interpreter_at()2055 void TemplateInterpreterGenerator::stop_interpreter_at() {
2056 Label L;
2057 __ push(rscratch1);
2058 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2059 __ ldr(rscratch1, Address(rscratch1));
2060 __ mov(rscratch2, StopInterpreterAt);
2061 __ cmpw(rscratch1, rscratch2);
2062 __ br(Assembler::NE, L);
2063 __ brk(0);
2064 __ bind(L);
2065 __ pop(rscratch1);
2066 }
2067
2068 #endif // !PRODUCT
2069