1 /*
2 * Copyright (c) 2003, 2017, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, 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.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateInterpreterGenerator.hpp"
33 #include "interpreter/templateTable.hpp"
34 #include "interpreter/bytecodeTracer.hpp"
35 #include "memory/resourceArea.hpp"
36 #include "oops/arrayOop.hpp"
37 #include "oops/methodData.hpp"
38 #include "oops/method.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "prims/jvmtiExport.hpp"
41 #include "prims/jvmtiThreadState.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/deoptimization.hpp"
44 #include "runtime/frame.inline.hpp"
45 #include "runtime/sharedRuntime.hpp"
46 #include "runtime/stubRoutines.hpp"
47 #include "runtime/synchronizer.hpp"
48 #include "runtime/timer.hpp"
49 #include "runtime/vframeArray.hpp"
50 #include "utilities/debug.hpp"
51 #include <sys/types.h>
52
53 #ifndef PRODUCT
54 #include "oops/method.hpp"
55 #endif // !PRODUCT
56
57 #ifdef BUILTIN_SIM
58 #include "../../../../../../simulator/simulator.hpp"
59 #endif
60
61 // Size of interpreter code. Increase if too small. Interpreter will
62 // fail with a guarantee ("not enough space for interpreter generation");
63 // if too small.
64 // Run with +PrintInterpreter to get the VM to print out the size.
65 // Max size with JVMTI
66 int TemplateInterpreter::InterpreterCodeSize = 200 * 1024;
67
68 #define __ _masm->
69
70 //-----------------------------------------------------------------------------
71
72 extern "C" void entry(CodeBuffer*);
73
74 //-----------------------------------------------------------------------------
75
76 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
77 address entry = __ pc();
78
79 __ andr(esp, esp, -16);
80 __ mov(c_rarg3, esp);
81 // rmethod
82 // rlocals
83 // c_rarg3: first stack arg - wordSize
84
85 // adjust sp
86 __ sub(sp, c_rarg3, 18 * wordSize);
87 __ str(lr, Address(__ pre(sp, -2 * wordSize)));
88 __ call_VM(noreg,
89 CAST_FROM_FN_PTR(address,
90 InterpreterRuntime::slow_signature_handler),
91 rmethod, rlocals, c_rarg3);
92
93 // r0: result handler
94
95 // Stack layout:
96 // rsp: return address <- sp
97 // 1 garbage
98 // 8 integer args (if static first is unused)
99 // 1 float/double identifiers
100 // 8 double args
101 // stack args <- esp
102 // garbage
103 // expression stack bottom
104 // bcp (NULL)
105 // ...
106
107 // Restore LR
108 __ ldr(lr, Address(__ post(sp, 2 * wordSize)));
109
110 // Do FP first so we can use c_rarg3 as temp
111 __ ldrw(c_rarg3, Address(sp, 9 * wordSize)); // float/double identifiers
112
113 for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
114 const FloatRegister r = as_FloatRegister(i);
115
116 Label d, done;
117
118 __ tbnz(c_rarg3, i, d);
119 __ ldrs(r, Address(sp, (10 + i) * wordSize));
120 __ b(done);
121 __ bind(d);
122 __ ldrd(r, Address(sp, (10 + i) * wordSize));
123 __ bind(done);
124 }
125
126 // c_rarg0 contains the result from the call of
127 // InterpreterRuntime::slow_signature_handler so we don't touch it
128 // here. It will be loaded with the JNIEnv* later.
129 __ ldr(c_rarg1, Address(sp, 1 * wordSize));
130 for (int i = c_rarg2->encoding(); i <= c_rarg7->encoding(); i += 2) {
131 Register rm = as_Register(i), rn = as_Register(i+1);
132 __ ldp(rm, rn, Address(sp, i * wordSize));
133 }
134
135 __ add(sp, sp, 18 * wordSize);
136 __ ret(lr);
137
138 return entry;
139 }
140
141
142 //
143 // Various method entries
144 //
145
146 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
147 // rmethod: Method*
148 // r13: sender sp
149 // esp: args
150
151 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
152
153 // These don't need a safepoint check because they aren't virtually
154 // callable. We won't enter these intrinsics from compiled code.
155 // If in the future we added an intrinsic which was virtually callable
156 // we'd have to worry about how to safepoint so that this code is used.
157
158 // mathematical functions inlined by compiler
159 // (interpreter must provide identical implementation
160 // in order to avoid monotonicity bugs when switching
161 // from interpreter to compiler in the middle of some
162 // computation)
163 //
164 // stack:
165 // [ arg ] <-- esp
166 // [ arg ]
167 // retaddr in lr
168
169 address entry_point = NULL;
170 Register continuation = lr;
171 switch (kind) {
172 case Interpreter::java_lang_math_abs:
173 entry_point = __ pc();
174 __ ldrd(v0, Address(esp));
175 __ fabsd(v0, v0);
176 __ mov(sp, r13); // Restore caller's SP
177 break;
178 case Interpreter::java_lang_math_sqrt:
179 entry_point = __ pc();
180 __ ldrd(v0, Address(esp));
181 __ fsqrtd(v0, v0);
182 __ mov(sp, r13);
183 break;
184 case Interpreter::java_lang_math_sin :
185 case Interpreter::java_lang_math_cos :
186 case Interpreter::java_lang_math_tan :
187 case Interpreter::java_lang_math_log :
188 case Interpreter::java_lang_math_log10 :
189 case Interpreter::java_lang_math_exp :
190 entry_point = __ pc();
191 __ ldrd(v0, Address(esp));
192 __ mov(sp, r13);
193 __ mov(r19, lr);
194 continuation = r19; // The first callee-saved register
195 generate_transcendental_entry(kind, 1);
196 break;
197 case Interpreter::java_lang_math_pow :
198 entry_point = __ pc();
199 __ mov(r19, lr);
200 continuation = r19;
201 __ ldrd(v0, Address(esp, 2 * Interpreter::stackElementSize));
202 __ ldrd(v1, Address(esp));
203 __ mov(sp, r13);
204 generate_transcendental_entry(kind, 2);
205 break;
206 case Interpreter::java_lang_math_fmaD :
207 if (UseFMA) {
208 entry_point = __ pc();
209 __ ldrd(v0, Address(esp, 4 * Interpreter::stackElementSize));
210 __ ldrd(v1, Address(esp, 2 * Interpreter::stackElementSize));
211 __ ldrd(v2, Address(esp));
212 __ fmaddd(v0, v0, v1, v2);
213 __ mov(sp, r13); // Restore caller's SP
214 }
215 break;
216 case Interpreter::java_lang_math_fmaF :
217 if (UseFMA) {
218 entry_point = __ pc();
219 __ ldrs(v0, Address(esp, 2 * Interpreter::stackElementSize));
220 __ ldrs(v1, Address(esp, Interpreter::stackElementSize));
221 __ ldrs(v2, Address(esp));
222 __ fmadds(v0, v0, v1, v2);
223 __ mov(sp, r13); // Restore caller's SP
224 }
225 break;
226 default:
227 ;
228 }
229 if (entry_point) {
230 __ br(continuation);
231 }
232
233 return entry_point;
234 }
235
236 // double trigonometrics and transcendentals
237 // static jdouble dsin(jdouble x);
238 // static jdouble dcos(jdouble x);
239 // static jdouble dtan(jdouble x);
240 // static jdouble dlog(jdouble x);
241 // static jdouble dlog10(jdouble x);
242 // static jdouble dexp(jdouble x);
243 // static jdouble dpow(jdouble x, jdouble y);
244
245 void TemplateInterpreterGenerator::generate_transcendental_entry(AbstractInterpreter::MethodKind kind, int fpargs) {
246 address fn;
247 switch (kind) {
248 case Interpreter::java_lang_math_sin :
249 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dsin);
250 break;
251 case Interpreter::java_lang_math_cos :
252 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dcos);
253 break;
254 case Interpreter::java_lang_math_tan :
255 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dtan);
256 break;
257 case Interpreter::java_lang_math_log :
258 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog);
259 break;
260 case Interpreter::java_lang_math_log10 :
261 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10);
262 break;
263 case Interpreter::java_lang_math_exp :
264 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dexp);
265 break;
266 case Interpreter::java_lang_math_pow :
267 fpargs = 2;
268 fn = CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
269 break;
270 default:
271 ShouldNotReachHere();
272 fn = NULL; // unreachable
273 }
274 const int gpargs = 0, rtype = 3;
275 __ mov(rscratch1, fn);
276 __ blrt(rscratch1, gpargs, fpargs, rtype);
277 }
278
279 // Abstract method entry
280 // Attempt to execute abstract method. Throw exception
281 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
282 // rmethod: Method*
283 // r13: sender SP
284
285 address entry_point = __ pc();
286
287 // abstract method entry
288
289 // pop return address, reset last_sp to NULL
290 __ empty_expression_stack();
291 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
292 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
293
294 // throw exception
295 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
296 InterpreterRuntime::throw_AbstractMethodError));
297 // the call_VM checks for exception, so we should never return here.
298 __ should_not_reach_here();
299
300 return entry_point;
301 }
302
303 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
304 address entry = __ pc();
305
306 #ifdef ASSERT
307 {
308 Label L;
309 __ ldr(rscratch1, Address(rfp,
310 frame::interpreter_frame_monitor_block_top_offset *
311 wordSize));
312 __ mov(rscratch2, sp);
313 __ cmp(rscratch1, rscratch2); // maximal rsp for current rfp (stack
314 // grows negative)
315 __ br(Assembler::HS, L); // check if frame is complete
316 __ stop ("interpreter frame not set up");
317 __ bind(L);
318 }
319 #endif // ASSERT
320 // Restore bcp under the assumption that the current frame is still
321 // interpreted
322 __ restore_bcp();
323
324 // expression stack must be empty before entering the VM if an
325 // exception happened
326 __ empty_expression_stack();
327 // throw exception
328 __ call_VM(noreg,
329 CAST_FROM_FN_PTR(address,
330 InterpreterRuntime::throw_StackOverflowError));
331 return entry;
332 }
333
334 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
335 const char* name) {
336 address entry = __ pc();
337 // expression stack must be empty before entering the VM if an
338 // exception happened
339 __ empty_expression_stack();
340 // setup parameters
341 // ??? convention: expect aberrant index in register r1
342 __ movw(c_rarg2, r1);
343 __ mov(c_rarg1, (address)name);
344 __ call_VM(noreg,
345 CAST_FROM_FN_PTR(address,
346 InterpreterRuntime::
347 throw_ArrayIndexOutOfBoundsException),
348 c_rarg1, c_rarg2);
349 return entry;
350 }
351
352 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
353 address entry = __ pc();
354
355 // object is at TOS
356 __ pop(c_rarg1);
357
358 // expression stack must be empty before entering the VM if an
359 // exception happened
360 __ empty_expression_stack();
361
362 __ call_VM(noreg,
363 CAST_FROM_FN_PTR(address,
364 InterpreterRuntime::
365 throw_ClassCastException),
366 c_rarg1);
367 return entry;
368 }
369
370 address TemplateInterpreterGenerator::generate_exception_handler_common(
371 const char* name, const char* message, bool pass_oop) {
372 assert(!pass_oop || message == NULL, "either oop or message but not both");
373 address entry = __ pc();
374 if (pass_oop) {
375 // object is at TOS
376 __ pop(c_rarg2);
377 }
378 // expression stack must be empty before entering the VM if an
379 // exception happened
380 __ empty_expression_stack();
381 // setup parameters
382 __ lea(c_rarg1, Address((address)name));
383 if (pass_oop) {
384 __ call_VM(r0, CAST_FROM_FN_PTR(address,
385 InterpreterRuntime::
386 create_klass_exception),
387 c_rarg1, c_rarg2);
388 } else {
389 // kind of lame ExternalAddress can't take NULL because
390 // external_word_Relocation will assert.
391 if (message != NULL) {
392 __ lea(c_rarg2, Address((address)message));
393 } else {
394 __ mov(c_rarg2, NULL_WORD);
395 }
396 __ call_VM(r0,
397 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
398 c_rarg1, c_rarg2);
399 }
400 // throw exception
401 __ b(address(Interpreter::throw_exception_entry()));
402 return entry;
403 }
404
405 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
406 address entry = __ pc();
407
408 // Restore stack bottom in case i2c adjusted stack
409 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
410 // and NULL it as marker that esp is now tos until next java call
411 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
412 __ restore_bcp();
413 __ restore_locals();
414 __ restore_constant_pool_cache();
415 __ get_method(rmethod);
416
417 // Pop N words from the stack
418 __ get_cache_and_index_at_bcp(r1, r2, 1, index_size);
419 __ ldr(r1, Address(r1, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
420 __ andr(r1, r1, ConstantPoolCacheEntry::parameter_size_mask);
421
422 __ add(esp, esp, r1, Assembler::LSL, 3);
423
424 // Restore machine SP
425 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
426 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
427 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
428 __ ldr(rscratch2,
429 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
430 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
431 __ andr(sp, rscratch1, -16);
432
433 #ifndef PRODUCT
434 // tell the simulator that the method has been reentered
435 if (NotifySimulator) {
436 __ notify(Assembler::method_reentry);
437 }
438 #endif
439
440 __ check_and_handle_popframe(rthread);
441 __ check_and_handle_earlyret(rthread);
442
443 __ get_dispatch();
444 __ dispatch_next(state, step);
445
446 return entry;
447 }
448
449 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
450 int step) {
451 address entry = __ pc();
452 __ restore_bcp();
453 __ restore_locals();
454 __ restore_constant_pool_cache();
455 __ get_method(rmethod);
456 __ get_dispatch();
457
458 // Calculate stack limit
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::uxtx, 3);
465 __ andr(sp, rscratch1, -16);
466
467 // Restore expression stack pointer
468 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
469 // NULL last_sp until next java call
470 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
471
472 #if INCLUDE_JVMCI
473 // Check if we need to take lock at entry of synchronized method. This can
474 // only occur on method entry so emit it only for vtos with step 0.
475 if (UseJVMCICompiler && state == vtos && step == 0) {
476 Label L;
477 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
478 __ cbz(rscratch1, L);
479 // Clear flag.
480 __ strb(zr, Address(rthread, JavaThread::pending_monitorenter_offset()));
481 // Take lock.
482 lock_method();
483 __ bind(L);
484 } else {
485 #ifdef ASSERT
486 if (UseJVMCICompiler) {
487 Label L;
488 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
489 __ cbz(rscratch1, L);
490 __ stop("unexpected pending monitor in deopt entry");
491 __ bind(L);
492 }
493 #endif
494 }
495 #endif
496 // handle exceptions
497 {
498 Label L;
499 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
500 __ cbz(rscratch1, L);
501 __ call_VM(noreg,
502 CAST_FROM_FN_PTR(address,
503 InterpreterRuntime::throw_pending_exception));
504 __ should_not_reach_here();
505 __ bind(L);
506 }
507
508 __ dispatch_next(state, step);
509 return entry;
510 }
511
512 address TemplateInterpreterGenerator::generate_result_handler_for(
513 BasicType type) {
514 address entry = __ pc();
515 switch (type) {
516 case T_BOOLEAN: __ uxtb(r0, r0); break;
517 case T_CHAR : __ uxth(r0, r0); break;
518 case T_BYTE : __ sxtb(r0, r0); break;
519 case T_SHORT : __ sxth(r0, r0); break;
520 case T_INT : __ uxtw(r0, r0); break; // FIXME: We almost certainly don't need this
521 case T_LONG : /* nothing to do */ break;
522 case T_VOID : /* nothing to do */ break;
523 case T_FLOAT : /* nothing to do */ break;
524 case T_DOUBLE : /* nothing to do */ break;
525 case T_OBJECT :
526 // retrieve result from frame
527 __ ldr(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
528 // and verify it
529 __ verify_oop(r0);
530 break;
531 default : ShouldNotReachHere();
532 }
533 __ ret(lr); // return from result handler
534 return entry;
535 }
536
537 address TemplateInterpreterGenerator::generate_safept_entry_for(
538 TosState state,
539 address runtime_entry) {
540 address entry = __ pc();
541 __ push(state);
542 __ call_VM(noreg, runtime_entry);
543 __ membar(Assembler::AnyAny);
544 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
545 return entry;
546 }
547
548 // Helpers for commoning out cases in the various type of method entries.
549 //
550
551
552 // increment invocation count & check for overflow
553 //
554 // Note: checking for negative value instead of overflow
555 // so we have a 'sticky' overflow test
556 //
557 // rmethod: method
558 //
559 void TemplateInterpreterGenerator::generate_counter_incr(
560 Label* overflow,
561 Label* profile_method,
562 Label* profile_method_continue) {
563 Label done;
564 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
565 if (TieredCompilation) {
566 int increment = InvocationCounter::count_increment;
567 Label no_mdo;
568 if (ProfileInterpreter) {
569 // Are we profiling?
570 __ ldr(r0, Address(rmethod, Method::method_data_offset()));
571 __ cbz(r0, no_mdo);
572 // Increment counter in the MDO
573 const Address mdo_invocation_counter(r0, in_bytes(MethodData::invocation_counter_offset()) +
574 in_bytes(InvocationCounter::counter_offset()));
575 const Address mask(r0, in_bytes(MethodData::invoke_mask_offset()));
576 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rscratch1, rscratch2, false, Assembler::EQ, overflow);
577 __ b(done);
578 }
579 __ bind(no_mdo);
580 // Increment counter in MethodCounters
581 const Address invocation_counter(rscratch2,
582 MethodCounters::invocation_counter_offset() +
583 InvocationCounter::counter_offset());
584 __ get_method_counters(rmethod, rscratch2, done);
585 const Address mask(rscratch2, in_bytes(MethodCounters::invoke_mask_offset()));
586 __ increment_mask_and_jump(invocation_counter, increment, mask, rscratch1, r1, false, Assembler::EQ, overflow);
587 __ bind(done);
588 } else { // not TieredCompilation
589 const Address backedge_counter(rscratch2,
590 MethodCounters::backedge_counter_offset() +
591 InvocationCounter::counter_offset());
592 const Address invocation_counter(rscratch2,
593 MethodCounters::invocation_counter_offset() +
594 InvocationCounter::counter_offset());
595
596 __ get_method_counters(rmethod, rscratch2, done);
597
598 if (ProfileInterpreter) { // %%% Merge this into MethodData*
599 __ ldrw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
600 __ addw(r1, r1, 1);
601 __ strw(r1, Address(rscratch2, MethodCounters::interpreter_invocation_counter_offset()));
602 }
603 // Update standard invocation counters
604 __ ldrw(r1, invocation_counter);
605 __ ldrw(r0, backedge_counter);
606
607 __ addw(r1, r1, InvocationCounter::count_increment);
608 __ andw(r0, r0, InvocationCounter::count_mask_value);
609
610 __ strw(r1, invocation_counter);
611 __ addw(r0, r0, r1); // add both counters
612
613 // profile_method is non-null only for interpreted method so
614 // profile_method != NULL == !native_call
615
616 if (ProfileInterpreter && profile_method != NULL) {
617 // Test to see if we should create a method data oop
618 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
619 __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
620 __ cmpw(r0, rscratch2);
621 __ br(Assembler::LT, *profile_method_continue);
622
623 // if no method data exists, go to profile_method
624 __ test_method_data_pointer(rscratch2, *profile_method);
625 }
626
627 {
628 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
629 __ ldrw(rscratch2, Address(rscratch2, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
630 __ cmpw(r0, rscratch2);
631 __ br(Assembler::HS, *overflow);
632 }
633 __ bind(done);
634 }
635 }
636
637 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
638
639 // Asm interpreter on entry
640 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
641 // Everything as it was on entry
642
643 // InterpreterRuntime::frequency_counter_overflow takes two
644 // arguments, the first (thread) is passed by call_VM, the second
645 // indicates if the counter overflow occurs at a backwards branch
646 // (NULL bcp). We pass zero for it. The call returns the address
647 // of the verified entry point for the method or NULL if the
648 // compilation did not complete (either went background or bailed
649 // out).
650 __ mov(c_rarg1, 0);
651 __ call_VM(noreg,
652 CAST_FROM_FN_PTR(address,
653 InterpreterRuntime::frequency_counter_overflow),
654 c_rarg1);
655
656 __ b(do_continue);
657 }
658
659 // See if we've got enough room on the stack for locals plus overhead
660 // below JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError
661 // without going through the signal handler, i.e., reserved and yellow zones
662 // will not be made usable. The shadow zone must suffice to handle the
663 // overflow.
664 // The expression stack grows down incrementally, so the normal guard
665 // page mechanism will work for that.
666 //
667 // NOTE: Since the additional locals are also always pushed (wasn't
668 // obvious in generate_method_entry) so the guard should work for them
669 // too.
670 //
671 // Args:
672 // r3: number of additional locals this frame needs (what we must check)
673 // rmethod: Method*
674 //
675 // Kills:
676 // r0
677 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
678
679 // monitor entry size: see picture of stack set
680 // (generate_method_entry) and frame_amd64.hpp
681 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
682
683 // total overhead size: entry_size + (saved rbp through expr stack
684 // bottom). be sure to change this if you add/subtract anything
685 // to/from the overhead area
686 const int overhead_size =
687 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
688
689 const int page_size = os::vm_page_size();
690
691 Label after_frame_check;
692
693 // see if the frame is greater than one page in size. If so,
694 // then we need to verify there is enough stack space remaining
695 // for the additional locals.
696 //
697 // Note that we use SUBS rather than CMP here because the immediate
698 // field of this instruction may overflow. SUBS can cope with this
699 // because it is a macro that will expand to some number of MOV
700 // instructions and a register operation.
701 __ subs(rscratch1, r3, (page_size - overhead_size) / Interpreter::stackElementSize);
702 __ br(Assembler::LS, after_frame_check);
703
704 // compute rsp as if this were going to be the last frame on
705 // the stack before the red zone
706
707 // locals + overhead, in bytes
708 __ mov(r0, overhead_size);
709 __ add(r0, r0, r3, Assembler::LSL, Interpreter::logStackElementSize); // 2 slots per parameter.
710
711 const Address stack_limit(rthread, JavaThread::stack_overflow_limit_offset());
712 __ ldr(rscratch1, stack_limit);
713
714 #ifdef ASSERT
715 Label limit_okay;
716 // Verify that thread stack limit is non-zero.
717 __ cbnz(rscratch1, limit_okay);
718 __ stop("stack overflow limit is zero");
719 __ bind(limit_okay);
720 #endif
721
722 // Add stack limit to locals.
723 __ add(r0, r0, rscratch1);
724
725 // Check against the current stack bottom.
726 __ cmp(sp, r0);
727 __ br(Assembler::HI, after_frame_check);
728
729 // Remove the incoming args, peeling the machine SP back to where it
730 // was in the caller. This is not strictly necessary, but unless we
731 // do so the stack frame may have a garbage FP; this ensures a
732 // correct call stack that we can always unwind. The ANDR should be
733 // unnecessary because the sender SP in r13 is always aligned, but
734 // it doesn't hurt.
735 __ andr(sp, r13, -16);
736
737 // Note: the restored frame is not necessarily interpreted.
738 // Use the shared runtime version of the StackOverflowError.
739 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
740 __ far_jump(RuntimeAddress(StubRoutines::throw_StackOverflowError_entry()));
741
742 // all done with frame size check
743 __ bind(after_frame_check);
744 }
745
746 // Allocate monitor and lock method (asm interpreter)
747 //
748 // Args:
749 // rmethod: Method*
750 // rlocals: locals
751 //
752 // Kills:
753 // r0
754 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
755 // rscratch1, rscratch2 (scratch regs)
756 void TemplateInterpreterGenerator::lock_method() {
757 // synchronize method
758 const Address access_flags(rmethod, Method::access_flags_offset());
759 const Address monitor_block_top(
760 rfp,
761 frame::interpreter_frame_monitor_block_top_offset * wordSize);
762 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
763
764 #ifdef ASSERT
765 {
766 Label L;
767 __ ldrw(r0, access_flags);
768 __ tst(r0, JVM_ACC_SYNCHRONIZED);
769 __ br(Assembler::NE, L);
770 __ stop("method doesn't need synchronization");
771 __ bind(L);
772 }
773 #endif // ASSERT
774
775 // get synchronization object
776 {
777 Label done;
778 __ ldrw(r0, access_flags);
779 __ tst(r0, JVM_ACC_STATIC);
780 // get receiver (assume this is frequent case)
781 __ ldr(r0, Address(rlocals, Interpreter::local_offset_in_bytes(0)));
782 __ br(Assembler::EQ, done);
783 __ load_mirror(r0, rmethod);
784
785 #ifdef ASSERT
786 {
787 Label L;
788 __ cbnz(r0, L);
789 __ stop("synchronization object is NULL");
790 __ bind(L);
791 }
792 #endif // ASSERT
793
794 __ bind(done);
795 }
796
797 // add space for monitor & lock
798 __ sub(sp, sp, entry_size); // add space for a monitor entry
799 __ sub(esp, esp, entry_size);
800 __ mov(rscratch1, esp);
801 __ str(rscratch1, monitor_block_top); // set new monitor block top
802 // store object
803 __ str(r0, Address(esp, BasicObjectLock::obj_offset_in_bytes()));
804 __ mov(c_rarg1, esp); // object address
805 __ lock_object(c_rarg1);
806 }
807
808 // Generate a fixed interpreter frame. This is identical setup for
809 // interpreted methods and for native methods hence the shared code.
810 //
811 // Args:
812 // lr: return address
813 // rmethod: Method*
814 // rlocals: pointer to locals
815 // rcpool: cp cache
816 // stack_pointer: previous sp
817 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
818 // initialize fixed part of activation frame
819 if (native_call) {
820 __ sub(esp, sp, 14 * wordSize);
821 __ mov(rbcp, zr);
822 __ stp(esp, zr, Address(__ pre(sp, -14 * wordSize)));
823 // add 2 zero-initialized slots for native calls
824 __ stp(zr, zr, Address(sp, 12 * wordSize));
825 } else {
826 __ sub(esp, sp, 12 * wordSize);
827 __ ldr(rscratch1, Address(rmethod, Method::const_offset())); // get ConstMethod
828 __ add(rbcp, rscratch1, in_bytes(ConstMethod::codes_offset())); // get codebase
829 __ stp(esp, rbcp, Address(__ pre(sp, -12 * wordSize)));
830 }
831
832 if (ProfileInterpreter) {
833 Label method_data_continue;
834 __ ldr(rscratch1, Address(rmethod, Method::method_data_offset()));
835 __ cbz(rscratch1, method_data_continue);
836 __ lea(rscratch1, Address(rscratch1, in_bytes(MethodData::data_offset())));
837 __ bind(method_data_continue);
838 __ stp(rscratch1, rmethod, Address(sp, 6 * wordSize)); // save Method* and mdp (method data pointer)
839 } else {
840 __ stp(zr, rmethod, Address(sp, 6 * wordSize)); // save Method* (no mdp)
841 }
842
843 // Get mirror and store it in the frame as GC root for this Method*
844 __ load_mirror(rscratch1, rmethod);
845 __ stp(rscratch1, zr, Address(sp, 4 * wordSize));
846
847 __ ldr(rcpool, Address(rmethod, Method::const_offset()));
848 __ ldr(rcpool, Address(rcpool, ConstMethod::constants_offset()));
849 __ ldr(rcpool, Address(rcpool, ConstantPool::cache_offset_in_bytes()));
850 __ stp(rlocals, rcpool, Address(sp, 2 * wordSize));
851
852 __ stp(rfp, lr, Address(sp, 10 * wordSize));
853 __ lea(rfp, Address(sp, 10 * wordSize));
854
855 // set sender sp
856 // leave last_sp as null
857 __ stp(zr, r13, Address(sp, 8 * wordSize));
858
859 // Move SP out of the way
860 if (! native_call) {
861 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
862 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
863 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 2);
864 __ sub(rscratch1, sp, rscratch1, ext::uxtw, 3);
865 __ andr(sp, rscratch1, -16);
866 }
867 }
868
869 // End of helpers
870
871 // Various method entries
872 //------------------------------------------------------------------------------------------------------------------------
873 //
874 //
875
876 // Method entry for java.lang.ref.Reference.get.
877 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
878 #if INCLUDE_ALL_GCS
879 // Code: _aload_0, _getfield, _areturn
880 // parameter size = 1
881 //
882 // The code that gets generated by this routine is split into 2 parts:
883 // 1. The "intrinsified" code for G1 (or any SATB based GC),
884 // 2. The slow path - which is an expansion of the regular method entry.
885 //
886 // Notes:-
887 // * In the G1 code we do not check whether we need to block for
888 // a safepoint. If G1 is enabled then we must execute the specialized
889 // code for Reference.get (except when the Reference object is null)
890 // so that we can log the value in the referent field with an SATB
891 // update buffer.
892 // If the code for the getfield template is modified so that the
893 // G1 pre-barrier code is executed when the current method is
894 // Reference.get() then going through the normal method entry
895 // will be fine.
896 // * The G1 code can, however, check the receiver object (the instance
897 // of java.lang.Reference) and jump to the slow path if null. If the
898 // Reference object is null then we obviously cannot fetch the referent
899 // and so we don't need to call the G1 pre-barrier. Thus we can use the
900 // regular method entry code to generate the NPE.
901 //
902 // This code is based on generate_accessor_entry.
903 //
904 // rmethod: Method*
905 // r13: senderSP must preserve for slow path, set SP to it on fast path
906
907 address entry = __ pc();
908
909 const int referent_offset = java_lang_ref_Reference::referent_offset;
910 guarantee(referent_offset > 0, "referent offset not initialized");
911
912 if (UseG1GC) {
913 Label slow_path;
914 const Register local_0 = c_rarg0;
915 // Check if local 0 != NULL
916 // If the receiver is null then it is OK to jump to the slow path.
917 __ ldr(local_0, Address(esp, 0));
918 __ cbz(local_0, slow_path);
919
920 // Load the value of the referent field.
921 const Address field_address(local_0, referent_offset);
922 __ load_heap_oop(local_0, field_address);
923
924 __ mov(r19, r13); // Move senderSP to a callee-saved register
925 // Generate the G1 pre-barrier code to log the value of
926 // the referent field in an SATB buffer.
927 __ enter(); // g1_write may call runtime
928 __ g1_write_barrier_pre(noreg /* obj */,
929 local_0 /* pre_val */,
930 rthread /* thread */,
931 rscratch2 /* tmp */,
932 true /* tosca_live */,
933 true /* expand_call */);
934 __ leave();
935 // areturn
936 __ andr(sp, r19, -16); // done with stack
937 __ ret(lr);
938
939 // generate a vanilla interpreter entry as the slow path
940 __ bind(slow_path);
941 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
942 return entry;
943 }
944 #endif // INCLUDE_ALL_GCS
945
946 // If G1 is not enabled then attempt to go through the accessor entry point
947 // Reference.get is an accessor
948 return NULL;
949 }
950
951 /**
952 * Method entry for static native methods:
953 * int java.util.zip.CRC32.update(int crc, int b)
954 */
955 address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
956 if (UseCRC32Intrinsics) {
957 address entry = __ pc();
958
959 // rmethod: Method*
960 // r13: senderSP must preserved for slow path
961 // esp: args
962
963 Label slow_path;
964 // If we need a safepoint check, generate full interpreter entry.
965 ExternalAddress state(SafepointSynchronize::address_of_state());
966 unsigned long offset;
967 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
968 __ ldrw(rscratch1, Address(rscratch1, offset));
969 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
970 __ cbnz(rscratch1, slow_path);
971
972 // We don't generate local frame and don't align stack because
973 // we call stub code and there is no safepoint on this path.
974
975 // Load parameters
976 const Register crc = c_rarg0; // crc
977 const Register val = c_rarg1; // source java byte value
978 const Register tbl = c_rarg2; // scratch
979
980 // Arguments are reversed on java expression stack
981 __ ldrw(val, Address(esp, 0)); // byte value
982 __ ldrw(crc, Address(esp, wordSize)); // Initial CRC
983
984 __ adrp(tbl, ExternalAddress(StubRoutines::crc_table_addr()), offset);
985 __ add(tbl, tbl, offset);
986
987 __ ornw(crc, zr, crc); // ~crc
988 __ update_byte_crc32(crc, val, tbl);
989 __ ornw(crc, zr, crc); // ~crc
990
991 // result in c_rarg0
992
993 __ andr(sp, r13, -16);
994 __ ret(lr);
995
996 // generate a vanilla native entry as the slow path
997 __ bind(slow_path);
998 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
999 return entry;
1000 }
1001 return NULL;
1002 }
1003
1004 /**
1005 * Method entry for static native methods:
1006 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
1007 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
1008 */
1009 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1010 if (UseCRC32Intrinsics) {
1011 address entry = __ pc();
1012
1013 // rmethod,: Method*
1014 // r13: senderSP must preserved for slow path
1015
1016 Label slow_path;
1017 // If we need a safepoint check, generate full interpreter entry.
1018 ExternalAddress state(SafepointSynchronize::address_of_state());
1019 unsigned long offset;
1020 __ adrp(rscratch1, ExternalAddress(SafepointSynchronize::address_of_state()), offset);
1021 __ ldrw(rscratch1, Address(rscratch1, offset));
1022 assert(SafepointSynchronize::_not_synchronized == 0, "rewrite this code");
1023 __ cbnz(rscratch1, slow_path);
1024
1025 // We don't generate local frame and don't align stack because
1026 // we call stub code and there is no safepoint on this path.
1027
1028 // Load parameters
1029 const Register crc = c_rarg0; // crc
1030 const Register buf = c_rarg1; // source java byte array address
1031 const Register len = c_rarg2; // length
1032 const Register off = len; // offset (never overlaps with 'len')
1033
1034 // Arguments are reversed on java expression stack
1035 // Calculate address of start element
1036 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
1037 __ ldr(buf, Address(esp, 2*wordSize)); // long buf
1038 __ ldrw(off, Address(esp, wordSize)); // offset
1039 __ add(buf, buf, off); // + offset
1040 __ ldrw(crc, Address(esp, 4*wordSize)); // Initial CRC
1041 } else {
1042 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] array
1043 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1044 __ ldrw(off, Address(esp, wordSize)); // offset
1045 __ add(buf, buf, off); // + offset
1046 __ ldrw(crc, Address(esp, 3*wordSize)); // Initial CRC
1047 }
1048 // Can now load 'len' since we're finished with 'off'
1049 __ ldrw(len, Address(esp, 0x0)); // Length
1050
1051 __ andr(sp, r13, -16); // Restore the caller's SP
1052
1053 // We are frameless so we can just jump to the stub.
1054 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()));
1055
1056 // generate a vanilla native entry as the slow path
1057 __ bind(slow_path);
1058 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
1059 return entry;
1060 }
1061 return NULL;
1062 }
1063
1064 /**
1065 * Method entry for intrinsic-candidate (non-native) methods:
1066 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
1067 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
1068 * Unlike CRC32, CRC32C does not have any methods marked as native
1069 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses
1070 */
1071 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
1072 if (UseCRC32Intrinsics) {
1073 address entry = __ pc();
1074
1075 // Prepare jump to stub using parameters from the stack
1076 const Register crc = c_rarg0; // initial crc
1077 const Register buf = c_rarg1; // source java byte array address
1078 const Register len = c_rarg2; // len argument to the kernel
1079
1080 const Register end = len; // index of last element to process
1081 const Register off = crc; // offset
1082
1083 __ ldrw(end, Address(esp)); // int end
1084 __ ldrw(off, Address(esp, wordSize)); // int offset
1085 __ sub(len, end, off);
1086 __ ldr(buf, Address(esp, 2*wordSize)); // byte[] buf | long buf
1087 __ add(buf, buf, off); // + offset
1088 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
1089 __ ldrw(crc, Address(esp, 4*wordSize)); // long crc
1090 } else {
1091 __ add(buf, buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
1092 __ ldrw(crc, Address(esp, 3*wordSize)); // long crc
1093 }
1094
1095 __ andr(sp, r13, -16); // Restore the caller's SP
1096
1097 // Jump to the stub.
1098 __ b(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()));
1099
1100 return entry;
1101 }
1102 return NULL;
1103 }
1104
1105 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) {
1106 // Bang each page in the shadow zone. We can't assume it's been done for
1107 // an interpreter frame with greater than a page of locals, so each page
1108 // needs to be checked. Only true for non-native.
1109 if (UseStackBanging) {
1110 const int n_shadow_pages = JavaThread::stack_shadow_zone_size() / os::vm_page_size();
1111 const int start_page = native_call ? n_shadow_pages : 1;
1112 const int page_size = os::vm_page_size();
1113 for (int pages = start_page; pages <= n_shadow_pages ; pages++) {
1114 __ sub(rscratch2, sp, pages*page_size);
1115 __ str(zr, Address(rscratch2));
1116 }
1117 }
1118 }
1119
1120
1121 // Interpreter stub for calling a native method. (asm interpreter)
1122 // This sets up a somewhat different looking stack for calling the
1123 // native method than the typical interpreter frame setup.
1124 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
1125 // determine code generation flags
1126 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1127
1128 // r1: Method*
1129 // rscratch1: sender sp
1130
1131 address entry_point = __ pc();
1132
1133 const Address constMethod (rmethod, Method::const_offset());
1134 const Address access_flags (rmethod, Method::access_flags_offset());
1135 const Address size_of_parameters(r2, ConstMethod::
1136 size_of_parameters_offset());
1137
1138 // get parameter size (always needed)
1139 __ ldr(r2, constMethod);
1140 __ load_unsigned_short(r2, size_of_parameters);
1141
1142 // Native calls don't need the stack size check since they have no
1143 // expression stack and the arguments are already on the stack and
1144 // we only add a handful of words to the stack.
1145
1146 // rmethod: Method*
1147 // r2: size of parameters
1148 // rscratch1: sender sp
1149
1150 // for natives the size of locals is zero
1151
1152 // compute beginning of parameters (rlocals)
1153 __ add(rlocals, esp, r2, ext::uxtx, 3);
1154 __ add(rlocals, rlocals, -wordSize);
1155
1156 // Pull SP back to minimum size: this avoids holes in the stack
1157 __ andr(sp, esp, -16);
1158
1159 // initialize fixed part of activation frame
1160 generate_fixed_frame(true);
1161 #ifndef PRODUCT
1162 // tell the simulator that a method has been entered
1163 if (NotifySimulator) {
1164 __ notify(Assembler::method_entry);
1165 }
1166 #endif
1167
1168 // make sure method is native & not abstract
1169 #ifdef ASSERT
1170 __ ldrw(r0, access_flags);
1171 {
1172 Label L;
1173 __ tst(r0, JVM_ACC_NATIVE);
1174 __ br(Assembler::NE, L);
1175 __ stop("tried to execute non-native method as native");
1176 __ bind(L);
1177 }
1178 {
1179 Label L;
1180 __ tst(r0, JVM_ACC_ABSTRACT);
1181 __ br(Assembler::EQ, L);
1182 __ stop("tried to execute abstract method in interpreter");
1183 __ bind(L);
1184 }
1185 #endif
1186
1187 // Since at this point in the method invocation the exception
1188 // handler would try to exit the monitor of synchronized methods
1189 // which hasn't been entered yet, we set the thread local variable
1190 // _do_not_unlock_if_synchronized to true. The remove_activation
1191 // will check this flag.
1192
1193 const Address do_not_unlock_if_synchronized(rthread,
1194 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1195 __ mov(rscratch2, true);
1196 __ strb(rscratch2, do_not_unlock_if_synchronized);
1197
1198 // increment invocation count & check for overflow
1199 Label invocation_counter_overflow;
1200 if (inc_counter) {
1201 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1202 }
1203
1204 Label continue_after_compile;
1205 __ bind(continue_after_compile);
1206
1207 bang_stack_shadow_pages(true);
1208
1209 // reset the _do_not_unlock_if_synchronized flag
1210 __ strb(zr, do_not_unlock_if_synchronized);
1211
1212 // check for synchronized methods
1213 // Must happen AFTER invocation_counter check and stack overflow check,
1214 // so method is not locked if overflows.
1215 if (synchronized) {
1216 lock_method();
1217 } else {
1218 // no synchronization necessary
1219 #ifdef ASSERT
1220 {
1221 Label L;
1222 __ ldrw(r0, access_flags);
1223 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1224 __ br(Assembler::EQ, L);
1225 __ stop("method needs synchronization");
1226 __ bind(L);
1227 }
1228 #endif
1229 }
1230
1231 // start execution
1232 #ifdef ASSERT
1233 {
1234 Label L;
1235 const Address monitor_block_top(rfp,
1236 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1237 __ ldr(rscratch1, monitor_block_top);
1238 __ cmp(esp, rscratch1);
1239 __ br(Assembler::EQ, L);
1240 __ stop("broken stack frame setup in interpreter");
1241 __ bind(L);
1242 }
1243 #endif
1244
1245 // jvmti support
1246 __ notify_method_entry();
1247
1248 // work registers
1249 const Register t = r17;
1250 const Register result_handler = r19;
1251
1252 // allocate space for parameters
1253 __ ldr(t, Address(rmethod, Method::const_offset()));
1254 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1255
1256 __ sub(rscratch1, esp, t, ext::uxtx, Interpreter::logStackElementSize);
1257 __ andr(sp, rscratch1, -16);
1258 __ mov(esp, rscratch1);
1259
1260 // get signature handler
1261 {
1262 Label L;
1263 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1264 __ cbnz(t, L);
1265 __ call_VM(noreg,
1266 CAST_FROM_FN_PTR(address,
1267 InterpreterRuntime::prepare_native_call),
1268 rmethod);
1269 __ ldr(t, Address(rmethod, Method::signature_handler_offset()));
1270 __ bind(L);
1271 }
1272
1273 // call signature handler
1274 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rlocals,
1275 "adjust this code");
1276 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == sp,
1277 "adjust this code");
1278 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1279 "adjust this code");
1280
1281 // The generated handlers do not touch rmethod (the method).
1282 // However, large signatures cannot be cached and are generated
1283 // each time here. The slow-path generator can do a GC on return,
1284 // so we must reload it after the call.
1285 __ blr(t);
1286 __ get_method(rmethod); // slow path can do a GC, reload rmethod
1287
1288
1289 // result handler is in r0
1290 // set result handler
1291 __ mov(result_handler, r0);
1292 // pass mirror handle if static call
1293 {
1294 Label L;
1295 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1296 __ tbz(t, exact_log2(JVM_ACC_STATIC), L);
1297 // get mirror
1298 __ load_mirror(t, rmethod);
1299 // copy mirror into activation frame
1300 __ str(t, Address(rfp, frame::interpreter_frame_oop_temp_offset * wordSize));
1301 // pass handle to mirror
1302 __ add(c_rarg1, rfp, frame::interpreter_frame_oop_temp_offset * wordSize);
1303 __ bind(L);
1304 }
1305
1306 // get native function entry point in r10
1307 {
1308 Label L;
1309 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1310 address unsatisfied = (SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1311 __ mov(rscratch2, unsatisfied);
1312 __ ldr(rscratch2, rscratch2);
1313 __ cmp(r10, rscratch2);
1314 __ br(Assembler::NE, L);
1315 __ call_VM(noreg,
1316 CAST_FROM_FN_PTR(address,
1317 InterpreterRuntime::prepare_native_call),
1318 rmethod);
1319 __ get_method(rmethod);
1320 __ ldr(r10, Address(rmethod, Method::native_function_offset()));
1321 __ bind(L);
1322 }
1323
1324 // pass JNIEnv
1325 __ add(c_rarg0, rthread, in_bytes(JavaThread::jni_environment_offset()));
1326
1327 // It is enough that the pc() points into the right code
1328 // segment. It does not have to be the correct return pc.
1329 __ set_last_Java_frame(esp, rfp, (address)NULL, rscratch1);
1330
1331 // change thread state
1332 #ifdef ASSERT
1333 {
1334 Label L;
1335 __ ldrw(t, Address(rthread, JavaThread::thread_state_offset()));
1336 __ cmp(t, _thread_in_Java);
1337 __ br(Assembler::EQ, L);
1338 __ stop("Wrong thread state in native stub");
1339 __ bind(L);
1340 }
1341 #endif
1342
1343 // Change state to native
1344 __ mov(rscratch1, _thread_in_native);
1345 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1346 __ stlrw(rscratch1, rscratch2);
1347
1348 // Call the native method.
1349 __ blrt(r10, rscratch1);
1350 __ maybe_isb();
1351 __ get_method(rmethod);
1352 // result potentially in r0 or v0
1353
1354 // make room for the pushes we're about to do
1355 __ sub(rscratch1, esp, 4 * wordSize);
1356 __ andr(sp, rscratch1, -16);
1357
1358 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1359 // in order to extract the result of a method call. If the order of these
1360 // pushes change or anything else is added to the stack then the code in
1361 // interpreter_frame_result must also change.
1362 __ push(dtos);
1363 __ push(ltos);
1364
1365 // change thread state
1366 __ mov(rscratch1, _thread_in_native_trans);
1367 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1368 __ stlrw(rscratch1, rscratch2);
1369
1370 if (os::is_MP()) {
1371 if (UseMembar) {
1372 // Force this write out before the read below
1373 __ dsb(Assembler::SY);
1374 } else {
1375 // Write serialization page so VM thread can do a pseudo remote membar.
1376 // We use the current thread pointer to calculate a thread specific
1377 // offset to write to within the page. This minimizes bus traffic
1378 // due to cache line collision.
1379 __ serialize_memory(rthread, rscratch2);
1380 }
1381 }
1382
1383 // check for safepoint operation in progress and/or pending suspend requests
1384 {
1385 Label Continue;
1386 {
1387 unsigned long offset;
1388 __ adrp(rscratch2, SafepointSynchronize::address_of_state(), offset);
1389 __ ldrw(rscratch2, Address(rscratch2, offset));
1390 }
1391 assert(SafepointSynchronize::_not_synchronized == 0,
1392 "SafepointSynchronize::_not_synchronized");
1393 Label L;
1394 __ cbnz(rscratch2, L);
1395 __ ldrw(rscratch2, Address(rthread, JavaThread::suspend_flags_offset()));
1396 __ cbz(rscratch2, Continue);
1397 __ bind(L);
1398
1399 // Don't use call_VM as it will see a possible pending exception
1400 // and forward it and never return here preventing us from
1401 // clearing _last_native_pc down below. So we do a runtime call by
1402 // hand.
1403 //
1404 __ mov(c_rarg0, rthread);
1405 __ mov(rscratch2, CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans));
1406 __ blrt(rscratch2, 1, 0, 0);
1407 __ maybe_isb();
1408 __ get_method(rmethod);
1409 __ reinit_heapbase();
1410 __ bind(Continue);
1411 }
1412
1413 // change thread state
1414 __ mov(rscratch1, _thread_in_Java);
1415 __ lea(rscratch2, Address(rthread, JavaThread::thread_state_offset()));
1416 __ stlrw(rscratch1, rscratch2);
1417
1418 // reset_last_Java_frame
1419 __ reset_last_Java_frame(true);
1420
1421 if (CheckJNICalls) {
1422 // clear_pending_jni_exception_check
1423 __ str(zr, Address(rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1424 }
1425
1426 // reset handle block
1427 __ ldr(t, Address(rthread, JavaThread::active_handles_offset()));
1428 __ str(zr, Address(t, JNIHandleBlock::top_offset_in_bytes()));
1429
1430 // If result is an oop unbox and store it in frame where gc will see it
1431 // and result handler will pick it up
1432
1433 {
1434 Label no_oop, not_weak, store_result;
1435 __ adr(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1436 __ cmp(t, result_handler);
1437 __ br(Assembler::NE, no_oop);
1438 // Unbox oop result, e.g. JNIHandles::resolve result.
1439 __ pop(ltos);
1440 __ cbz(r0, store_result); // Use NULL as-is.
1441 STATIC_ASSERT(JNIHandles::weak_tag_mask == 1u);
1442 __ tbz(r0, 0, not_weak); // Test for jweak tag.
1443 // Resolve jweak.
1444 __ ldr(r0, Address(r0, -JNIHandles::weak_tag_value));
1445 #if INCLUDE_ALL_GCS
1446 if (UseG1GC) {
1447 __ enter(); // Barrier may call runtime.
1448 __ g1_write_barrier_pre(noreg /* obj */,
1449 r0 /* pre_val */,
1450 rthread /* thread */,
1451 t /* tmp */,
1452 true /* tosca_live */,
1453 true /* expand_call */);
1454 __ leave();
1455 }
1456 #endif // INCLUDE_ALL_GCS
1457 __ b(store_result);
1458 __ bind(not_weak);
1459 // Resolve (untagged) jobject.
1460 __ ldr(r0, Address(r0, 0));
1461 __ bind(store_result);
1462 __ str(r0, Address(rfp, frame::interpreter_frame_oop_temp_offset*wordSize));
1463 // keep stack depth as expected by pushing oop which will eventually be discarded
1464 __ push(ltos);
1465 __ bind(no_oop);
1466 }
1467
1468 {
1469 Label no_reguard;
1470 __ lea(rscratch1, Address(rthread, in_bytes(JavaThread::stack_guard_state_offset())));
1471 __ ldrw(rscratch1, Address(rscratch1));
1472 __ cmp(rscratch1, JavaThread::stack_guard_yellow_reserved_disabled);
1473 __ br(Assembler::NE, no_reguard);
1474
1475 __ pusha(); // XXX only save smashed registers
1476 __ mov(c_rarg0, rthread);
1477 __ mov(rscratch2, CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages));
1478 __ blrt(rscratch2, 0, 0, 0);
1479 __ popa(); // XXX only restore smashed registers
1480 __ bind(no_reguard);
1481 }
1482
1483 // The method register is junk from after the thread_in_native transition
1484 // until here. Also can't call_VM until the bcp has been
1485 // restored. Need bcp for throwing exception below so get it now.
1486 __ get_method(rmethod);
1487
1488 // restore bcp to have legal interpreter frame, i.e., bci == 0 <=>
1489 // rbcp == code_base()
1490 __ ldr(rbcp, Address(rmethod, Method::const_offset())); // get ConstMethod*
1491 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset())); // get codebase
1492 // handle exceptions (exception handling will handle unlocking!)
1493 {
1494 Label L;
1495 __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
1496 __ cbz(rscratch1, L);
1497 // Note: At some point we may want to unify this with the code
1498 // used in call_VM_base(); i.e., we should use the
1499 // StubRoutines::forward_exception code. For now this doesn't work
1500 // here because the rsp is not correctly set at this point.
1501 __ MacroAssembler::call_VM(noreg,
1502 CAST_FROM_FN_PTR(address,
1503 InterpreterRuntime::throw_pending_exception));
1504 __ should_not_reach_here();
1505 __ bind(L);
1506 }
1507
1508 // do unlocking if necessary
1509 {
1510 Label L;
1511 __ ldrw(t, Address(rmethod, Method::access_flags_offset()));
1512 __ tbz(t, exact_log2(JVM_ACC_SYNCHRONIZED), L);
1513 // the code below should be shared with interpreter macro
1514 // assembler implementation
1515 {
1516 Label unlock;
1517 // BasicObjectLock will be first in list, since this is a
1518 // synchronized method. However, need to check that the object
1519 // has not been unlocked by an explicit monitorexit bytecode.
1520
1521 // monitor expect in c_rarg1 for slow unlock path
1522 __ lea (c_rarg1, Address(rfp, // address of first monitor
1523 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1524 wordSize - sizeof(BasicObjectLock))));
1525
1526 __ ldr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1527 __ cbnz(t, unlock);
1528
1529 // Entry already unlocked, need to throw exception
1530 __ MacroAssembler::call_VM(noreg,
1531 CAST_FROM_FN_PTR(address,
1532 InterpreterRuntime::throw_illegal_monitor_state_exception));
1533 __ should_not_reach_here();
1534
1535 __ bind(unlock);
1536 __ unlock_object(c_rarg1);
1537 }
1538 __ bind(L);
1539 }
1540
1541 // jvmti support
1542 // Note: This must happen _after_ handling/throwing any exceptions since
1543 // the exception handler code notifies the runtime of method exits
1544 // too. If this happens before, method entry/exit notifications are
1545 // not properly paired (was bug - gri 11/22/99).
1546 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1547
1548 // restore potential result in r0:d0, call result handler to
1549 // restore potential result in ST0 & handle result
1550
1551 __ pop(ltos);
1552 __ pop(dtos);
1553
1554 __ blr(result_handler);
1555
1556 // remove activation
1557 __ ldr(esp, Address(rfp,
1558 frame::interpreter_frame_sender_sp_offset *
1559 wordSize)); // get sender sp
1560 // remove frame anchor
1561 __ leave();
1562
1563 // resture sender sp
1564 __ mov(sp, esp);
1565
1566 __ ret(lr);
1567
1568 if (inc_counter) {
1569 // Handle overflow of counter and compile method
1570 __ bind(invocation_counter_overflow);
1571 generate_counter_overflow(continue_after_compile);
1572 }
1573
1574 return entry_point;
1575 }
1576
1577 //
1578 // Generic interpreted method entry to (asm) interpreter
1579 //
1580 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1581 // determine code generation flags
1582 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1583
1584 // rscratch1: sender sp
1585 address entry_point = __ pc();
1586
1587 const Address constMethod(rmethod, Method::const_offset());
1588 const Address access_flags(rmethod, Method::access_flags_offset());
1589 const Address size_of_parameters(r3,
1590 ConstMethod::size_of_parameters_offset());
1591 const Address size_of_locals(r3, ConstMethod::size_of_locals_offset());
1592
1593 // get parameter size (always needed)
1594 // need to load the const method first
1595 __ ldr(r3, constMethod);
1596 __ load_unsigned_short(r2, size_of_parameters);
1597
1598 // r2: size of parameters
1599
1600 __ load_unsigned_short(r3, size_of_locals); // get size of locals in words
1601 __ sub(r3, r3, r2); // r3 = no. of additional locals
1602
1603 // see if we've got enough room on the stack for locals plus overhead.
1604 generate_stack_overflow_check();
1605
1606 // compute beginning of parameters (rlocals)
1607 __ add(rlocals, esp, r2, ext::uxtx, 3);
1608 __ sub(rlocals, rlocals, wordSize);
1609
1610 // Make room for locals
1611 __ sub(rscratch1, esp, r3, ext::uxtx, 3);
1612 __ andr(sp, rscratch1, -16);
1613
1614 // r3 - # of additional locals
1615 // allocate space for locals
1616 // explicitly initialize locals
1617 {
1618 Label exit, loop;
1619 __ ands(zr, r3, r3);
1620 __ br(Assembler::LE, exit); // do nothing if r3 <= 0
1621 __ bind(loop);
1622 __ str(zr, Address(__ post(rscratch1, wordSize)));
1623 __ sub(r3, r3, 1); // until everything initialized
1624 __ cbnz(r3, loop);
1625 __ bind(exit);
1626 }
1627
1628 // And the base dispatch table
1629 __ get_dispatch();
1630
1631 // initialize fixed part of activation frame
1632 generate_fixed_frame(false);
1633 #ifndef PRODUCT
1634 // tell the simulator that a method has been entered
1635 if (NotifySimulator) {
1636 __ notify(Assembler::method_entry);
1637 }
1638 #endif
1639 // make sure method is not native & not abstract
1640 #ifdef ASSERT
1641 __ ldrw(r0, access_flags);
1642 {
1643 Label L;
1644 __ tst(r0, JVM_ACC_NATIVE);
1645 __ br(Assembler::EQ, L);
1646 __ stop("tried to execute native method as non-native");
1647 __ bind(L);
1648 }
1649 {
1650 Label L;
1651 __ tst(r0, JVM_ACC_ABSTRACT);
1652 __ br(Assembler::EQ, L);
1653 __ stop("tried to execute abstract method in interpreter");
1654 __ bind(L);
1655 }
1656 #endif
1657
1658 // Since at this point in the method invocation the exception
1659 // handler would try to exit the monitor of synchronized methods
1660 // which hasn't been entered yet, we set the thread local variable
1661 // _do_not_unlock_if_synchronized to true. The remove_activation
1662 // will check this flag.
1663
1664 const Address do_not_unlock_if_synchronized(rthread,
1665 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1666 __ mov(rscratch2, true);
1667 __ strb(rscratch2, do_not_unlock_if_synchronized);
1668
1669 // increment invocation count & check for overflow
1670 Label invocation_counter_overflow;
1671 Label profile_method;
1672 Label profile_method_continue;
1673 if (inc_counter) {
1674 generate_counter_incr(&invocation_counter_overflow,
1675 &profile_method,
1676 &profile_method_continue);
1677 if (ProfileInterpreter) {
1678 __ bind(profile_method_continue);
1679 }
1680 }
1681
1682 Label continue_after_compile;
1683 __ bind(continue_after_compile);
1684
1685 bang_stack_shadow_pages(false);
1686
1687 // reset the _do_not_unlock_if_synchronized flag
1688 __ strb(zr, do_not_unlock_if_synchronized);
1689
1690 // check for synchronized methods
1691 // Must happen AFTER invocation_counter check and stack overflow check,
1692 // so method is not locked if overflows.
1693 if (synchronized) {
1694 // Allocate monitor and lock method
1695 lock_method();
1696 } else {
1697 // no synchronization necessary
1698 #ifdef ASSERT
1699 {
1700 Label L;
1701 __ ldrw(r0, access_flags);
1702 __ tst(r0, JVM_ACC_SYNCHRONIZED);
1703 __ br(Assembler::EQ, L);
1704 __ stop("method needs synchronization");
1705 __ bind(L);
1706 }
1707 #endif
1708 }
1709
1710 // start execution
1711 #ifdef ASSERT
1712 {
1713 Label L;
1714 const Address monitor_block_top (rfp,
1715 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1716 __ ldr(rscratch1, monitor_block_top);
1717 __ cmp(esp, rscratch1);
1718 __ br(Assembler::EQ, L);
1719 __ stop("broken stack frame setup in interpreter");
1720 __ bind(L);
1721 }
1722 #endif
1723
1724 // jvmti support
1725 __ notify_method_entry();
1726
1727 __ dispatch_next(vtos);
1728
1729 // invocation counter overflow
1730 if (inc_counter) {
1731 if (ProfileInterpreter) {
1732 // We have decided to profile this method in the interpreter
1733 __ bind(profile_method);
1734 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1735 __ set_method_data_pointer_for_bcp();
1736 // don't think we need this
1737 __ get_method(r1);
1738 __ b(profile_method_continue);
1739 }
1740 // Handle overflow of counter and compile method
1741 __ bind(invocation_counter_overflow);
1742 generate_counter_overflow(continue_after_compile);
1743 }
1744
1745 return entry_point;
1746 }
1747
1748 //-----------------------------------------------------------------------------
1749 // Exceptions
1750
1751 void TemplateInterpreterGenerator::generate_throw_exception() {
1752 // Entry point in previous activation (i.e., if the caller was
1753 // interpreted)
1754 Interpreter::_rethrow_exception_entry = __ pc();
1755 // Restore sp to interpreter_frame_last_sp even though we are going
1756 // to empty the expression stack for the exception processing.
1757 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1758 // r0: exception
1759 // r3: return address/pc that threw exception
1760 __ restore_bcp(); // rbcp points to call/send
1761 __ restore_locals();
1762 __ restore_constant_pool_cache();
1763 __ reinit_heapbase(); // restore rheapbase as heapbase.
1764 __ get_dispatch();
1765
1766 #ifndef PRODUCT
1767 // tell the simulator that the caller method has been reentered
1768 if (NotifySimulator) {
1769 __ get_method(rmethod);
1770 __ notify(Assembler::method_reentry);
1771 }
1772 #endif
1773 // Entry point for exceptions thrown within interpreter code
1774 Interpreter::_throw_exception_entry = __ pc();
1775 // If we came here via a NullPointerException on the receiver of a
1776 // method, rmethod may be corrupt.
1777 __ get_method(rmethod);
1778 // expression stack is undefined here
1779 // r0: exception
1780 // rbcp: exception bcp
1781 __ verify_oop(r0);
1782 __ mov(c_rarg1, r0);
1783
1784 // expression stack must be empty before entering the VM in case of
1785 // an exception
1786 __ empty_expression_stack();
1787 // find exception handler address and preserve exception oop
1788 __ call_VM(r3,
1789 CAST_FROM_FN_PTR(address,
1790 InterpreterRuntime::exception_handler_for_exception),
1791 c_rarg1);
1792
1793 // Calculate stack limit
1794 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1795 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1796 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1797 __ ldr(rscratch2,
1798 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1799 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtx, 3);
1800 __ andr(sp, rscratch1, -16);
1801
1802 // r0: exception handler entry point
1803 // r3: preserved exception oop
1804 // rbcp: bcp for exception handler
1805 __ push_ptr(r3); // push exception which is now the only value on the stack
1806 __ br(r0); // jump to exception handler (may be _remove_activation_entry!)
1807
1808 // If the exception is not handled in the current frame the frame is
1809 // removed and the exception is rethrown (i.e. exception
1810 // continuation is _rethrow_exception).
1811 //
1812 // Note: At this point the bci is still the bxi for the instruction
1813 // which caused the exception and the expression stack is
1814 // empty. Thus, for any VM calls at this point, GC will find a legal
1815 // oop map (with empty expression stack).
1816
1817 //
1818 // JVMTI PopFrame support
1819 //
1820
1821 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1822 __ empty_expression_stack();
1823 // Set the popframe_processing bit in pending_popframe_condition
1824 // indicating that we are currently handling popframe, so that
1825 // call_VMs that may happen later do not trigger new popframe
1826 // handling cycles.
1827 __ ldrw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1828 __ orr(r3, r3, JavaThread::popframe_processing_bit);
1829 __ strw(r3, Address(rthread, JavaThread::popframe_condition_offset()));
1830
1831 {
1832 // Check to see whether we are returning to a deoptimized frame.
1833 // (The PopFrame call ensures that the caller of the popped frame is
1834 // either interpreted or compiled and deoptimizes it if compiled.)
1835 // In this case, we can't call dispatch_next() after the frame is
1836 // popped, but instead must save the incoming arguments and restore
1837 // them after deoptimization has occurred.
1838 //
1839 // Note that we don't compare the return PC against the
1840 // deoptimization blob's unpack entry because of the presence of
1841 // adapter frames in C2.
1842 Label caller_not_deoptimized;
1843 __ ldr(c_rarg1, Address(rfp, frame::return_addr_offset * wordSize));
1844 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1845 InterpreterRuntime::interpreter_contains), c_rarg1);
1846 __ cbnz(r0, caller_not_deoptimized);
1847
1848 // Compute size of arguments for saving when returning to
1849 // deoptimized caller
1850 __ get_method(r0);
1851 __ ldr(r0, Address(r0, Method::const_offset()));
1852 __ load_unsigned_short(r0, Address(r0, in_bytes(ConstMethod::
1853 size_of_parameters_offset())));
1854 __ lsl(r0, r0, Interpreter::logStackElementSize);
1855 __ restore_locals(); // XXX do we need this?
1856 __ sub(rlocals, rlocals, r0);
1857 __ add(rlocals, rlocals, wordSize);
1858 // Save these arguments
1859 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1860 Deoptimization::
1861 popframe_preserve_args),
1862 rthread, r0, rlocals);
1863
1864 __ remove_activation(vtos,
1865 /* throw_monitor_exception */ false,
1866 /* install_monitor_exception */ false,
1867 /* notify_jvmdi */ false);
1868
1869 // Inform deoptimization that it is responsible for restoring
1870 // these arguments
1871 __ mov(rscratch1, JavaThread::popframe_force_deopt_reexecution_bit);
1872 __ strw(rscratch1, Address(rthread, JavaThread::popframe_condition_offset()));
1873
1874 // Continue in deoptimization handler
1875 __ ret(lr);
1876
1877 __ bind(caller_not_deoptimized);
1878 }
1879
1880 __ remove_activation(vtos,
1881 /* throw_monitor_exception */ false,
1882 /* install_monitor_exception */ false,
1883 /* notify_jvmdi */ false);
1884
1885 // Restore the last_sp and null it out
1886 __ ldr(esp, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1887 __ str(zr, Address(rfp, frame::interpreter_frame_last_sp_offset * wordSize));
1888
1889 __ restore_bcp();
1890 __ restore_locals();
1891 __ restore_constant_pool_cache();
1892 __ get_method(rmethod);
1893
1894 // The method data pointer was incremented already during
1895 // call profiling. We have to restore the mdp for the current bcp.
1896 if (ProfileInterpreter) {
1897 __ set_method_data_pointer_for_bcp();
1898 }
1899
1900 // Clear the popframe condition flag
1901 __ strw(zr, Address(rthread, JavaThread::popframe_condition_offset()));
1902 assert(JavaThread::popframe_inactive == 0, "fix popframe_inactive");
1903
1904 #if INCLUDE_JVMTI
1905 {
1906 Label L_done;
1907
1908 __ ldrb(rscratch1, Address(rbcp, 0));
1909 __ cmpw(r1, Bytecodes::_invokestatic);
1910 __ br(Assembler::EQ, L_done);
1911
1912 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1913 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1914
1915 __ ldr(c_rarg0, Address(rlocals, 0));
1916 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), c_rarg0, rmethod, rbcp);
1917
1918 __ cbz(r0, L_done);
1919
1920 __ str(r0, Address(esp, 0));
1921 __ bind(L_done);
1922 }
1923 #endif // INCLUDE_JVMTI
1924
1925 // Restore machine SP
1926 __ ldr(rscratch1, Address(rmethod, Method::const_offset()));
1927 __ ldrh(rscratch1, Address(rscratch1, ConstMethod::max_stack_offset()));
1928 __ add(rscratch1, rscratch1, frame::interpreter_frame_monitor_size() + 4);
1929 __ ldr(rscratch2,
1930 Address(rfp, frame::interpreter_frame_initial_sp_offset * wordSize));
1931 __ sub(rscratch1, rscratch2, rscratch1, ext::uxtw, 3);
1932 __ andr(sp, rscratch1, -16);
1933
1934 __ dispatch_next(vtos);
1935 // end of PopFrame support
1936
1937 Interpreter::_remove_activation_entry = __ pc();
1938
1939 // preserve exception over this code sequence
1940 __ pop_ptr(r0);
1941 __ str(r0, Address(rthread, JavaThread::vm_result_offset()));
1942 // remove the activation (without doing throws on illegalMonitorExceptions)
1943 __ remove_activation(vtos, false, true, false);
1944 // restore exception
1945 // restore exception
1946 __ get_vm_result(r0, rthread);
1947
1948 // In between activations - previous activation type unknown yet
1949 // compute continuation point - the continuation point expects the
1950 // following registers set up:
1951 //
1952 // r0: exception
1953 // lr: return address/pc that threw exception
1954 // rsp: expression stack of caller
1955 // rfp: fp of caller
1956 // FIXME: There's no point saving LR here because VM calls don't trash it
1957 __ stp(r0, lr, Address(__ pre(sp, -2 * wordSize))); // save exception & return address
1958 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1959 SharedRuntime::exception_handler_for_return_address),
1960 rthread, lr);
1961 __ mov(r1, r0); // save exception handler
1962 __ ldp(r0, lr, Address(__ post(sp, 2 * wordSize))); // restore exception & return address
1963 // We might be returning to a deopt handler that expects r3 to
1964 // contain the exception pc
1965 __ mov(r3, lr);
1966 // Note that an "issuing PC" is actually the next PC after the call
1967 __ br(r1); // jump to exception
1968 // handler of caller
1969 }
1970
1971
1972 //
1973 // JVMTI ForceEarlyReturn support
1974 //
1975 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1976 address entry = __ pc();
1977
1978 __ restore_bcp();
1979 __ restore_locals();
1980 __ empty_expression_stack();
1981 __ load_earlyret_value(state);
1982
1983 __ ldr(rscratch1, Address(rthread, JavaThread::jvmti_thread_state_offset()));
1984 Address cond_addr(rscratch1, JvmtiThreadState::earlyret_state_offset());
1985
1986 // Clear the earlyret state
1987 assert(JvmtiThreadState::earlyret_inactive == 0, "should be");
1988 __ str(zr, cond_addr);
1989
1990 __ remove_activation(state,
1991 false, /* throw_monitor_exception */
1992 false, /* install_monitor_exception */
1993 true); /* notify_jvmdi */
1994 __ ret(lr);
1995
1996 return entry;
1997 } // end of ForceEarlyReturn support
1998
1999
2000
2001 //-----------------------------------------------------------------------------
2002 // Helper for vtos entry point generation
2003
2004 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2005 address& bep,
2006 address& cep,
2007 address& sep,
2008 address& aep,
2009 address& iep,
2010 address& lep,
2011 address& fep,
2012 address& dep,
2013 address& vep) {
2014 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2015 Label L;
2016 aep = __ pc(); __ push_ptr(); __ b(L);
2017 fep = __ pc(); __ push_f(); __ b(L);
2018 dep = __ pc(); __ push_d(); __ b(L);
2019 lep = __ pc(); __ push_l(); __ b(L);
2020 bep = cep = sep =
2021 iep = __ pc(); __ push_i();
2022 vep = __ pc();
2023 __ bind(L);
2024 generate_and_dispatch(t);
2025 }
2026
2027 //-----------------------------------------------------------------------------
2028
2029 // Non-product code
2030 #ifndef PRODUCT
2031 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2032 address entry = __ pc();
2033
2034 __ push(lr);
2035 __ push(state);
2036 __ push(RegSet::range(r0, r15), sp);
2037 __ mov(c_rarg2, r0); // Pass itos
2038 __ call_VM(noreg,
2039 CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode),
2040 c_rarg1, c_rarg2, c_rarg3);
2041 __ pop(RegSet::range(r0, r15), sp);
2042 __ pop(state);
2043 __ pop(lr);
2044 __ ret(lr); // return from result handler
2045
2046 return entry;
2047 }
2048
2049 void TemplateInterpreterGenerator::count_bytecode() {
2050 Register rscratch3 = r0;
2051 __ push(rscratch1);
2052 __ push(rscratch2);
2053 __ push(rscratch3);
2054 __ mov(rscratch3, (address) &BytecodeCounter::_counter_value);
2055 __ atomic_add(noreg, 1, rscratch3);
2056 __ pop(rscratch3);
2057 __ pop(rscratch2);
2058 __ pop(rscratch1);
2059 }
2060
2061 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { ; }
2062
2063 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { ; }
2064
2065
2066 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2067 // Call a little run-time stub to avoid blow-up for each bytecode.
2068 // The run-time runtime saves the right registers, depending on
2069 // the tosca in-state for the given template.
2070
2071 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2072 "entry must have been generated");
2073 __ bl(Interpreter::trace_code(t->tos_in()));
2074 __ reinit_heapbase();
2075 }
2076
2077
2078 void TemplateInterpreterGenerator::stop_interpreter_at() {
2079 Label L;
2080 __ push(rscratch1);
2081 __ mov(rscratch1, (address) &BytecodeCounter::_counter_value);
2082 __ ldr(rscratch1, Address(rscratch1));
2083 __ mov(rscratch2, StopInterpreterAt);
2084 __ cmpw(rscratch1, rscratch2);
2085 __ br(Assembler::NE, L);
2086 __ brk(0);
2087 __ bind(L);
2088 __ pop(rscratch1);
2089 }
2090
2091 #ifdef BUILTIN_SIM
2092
2093 #include <sys/mman.h>
2094 #include <unistd.h>
2095
2096 extern "C" {
2097 static int PAGESIZE = getpagesize();
2098 int is_mapped_address(u_int64_t address)
2099 {
2100 address = (address & ~((u_int64_t)PAGESIZE - 1));
2101 if (msync((void *)address, PAGESIZE, MS_ASYNC) == 0) {
2102 return true;
2103 }
2104 if (errno != ENOMEM) {
2105 return true;
2106 }
2107 return false;
2108 }
2109
2110 void bccheck1(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2111 {
2112 if (method != 0) {
2113 method[0] = '\0';
2114 }
2115 if (bcidx != 0) {
2116 *bcidx = -2;
2117 }
2118 if (decode != 0) {
2119 decode[0] = 0;
2120 }
2121
2122 if (framesize != 0) {
2123 *framesize = -1;
2124 }
2125
2126 if (Interpreter::contains((address)pc)) {
2127 AArch64Simulator *sim = AArch64Simulator::get_current(UseSimulatorCache, DisableBCCheck);
2128 Method* meth;
2129 address bcp;
2130 if (fp) {
2131 #define FRAME_SLOT_METHOD 3
2132 #define FRAME_SLOT_BCP 7
2133 meth = (Method*)sim->getMemory()->loadU64(fp - (FRAME_SLOT_METHOD << 3));
2134 bcp = (address)sim->getMemory()->loadU64(fp - (FRAME_SLOT_BCP << 3));
2135 #undef FRAME_SLOT_METHOD
2136 #undef FRAME_SLOT_BCP
2137 } else {
2138 meth = (Method*)sim->getCPUState().xreg(RMETHOD, 0);
2139 bcp = (address)sim->getCPUState().xreg(RBCP, 0);
2140 }
2141 if (meth->is_native()) {
2142 return;
2143 }
2144 if(method && meth->is_method()) {
2145 ResourceMark rm;
2146 method[0] = 'I';
2147 method[1] = ' ';
2148 meth->name_and_sig_as_C_string(method + 2, 398);
2149 }
2150 if (bcidx) {
2151 if (meth->contains(bcp)) {
2152 *bcidx = meth->bci_from(bcp);
2153 } else {
2154 *bcidx = -2;
2155 }
2156 }
2157 if (decode) {
2158 if (!BytecodeTracer::closure()) {
2159 BytecodeTracer::set_closure(BytecodeTracer::std_closure());
2160 }
2161 stringStream str(decode, 400);
2162 BytecodeTracer::trace(meth, bcp, &str);
2163 }
2164 } else {
2165 if (method) {
2166 CodeBlob *cb = CodeCache::find_blob((address)pc);
2167 if (cb != NULL) {
2168 if (cb->is_nmethod()) {
2169 ResourceMark rm;
2170 nmethod* nm = (nmethod*)cb;
2171 method[0] = 'C';
2172 method[1] = ' ';
2173 nm->method()->name_and_sig_as_C_string(method + 2, 398);
2174 } else if (cb->is_adapter_blob()) {
2175 strcpy(method, "B adapter blob");
2176 } else if (cb->is_runtime_stub()) {
2177 strcpy(method, "B runtime stub");
2178 } else if (cb->is_exception_stub()) {
2179 strcpy(method, "B exception stub");
2180 } else if (cb->is_deoptimization_stub()) {
2181 strcpy(method, "B deoptimization stub");
2182 } else if (cb->is_safepoint_stub()) {
2183 strcpy(method, "B safepoint stub");
2184 } else if (cb->is_uncommon_trap_stub()) {
2185 strcpy(method, "B uncommon trap stub");
2186 } else if (cb->contains((address)StubRoutines::call_stub())) {
2187 strcpy(method, "B call stub");
2188 } else {
2189 strcpy(method, "B unknown blob : ");
2190 strcat(method, cb->name());
2191 }
2192 if (framesize != NULL) {
2193 *framesize = cb->frame_size();
2194 }
2195 }
2196 }
2197 }
2198 }
2199
2200
2201 JNIEXPORT void bccheck(u_int64_t pc, u_int64_t fp, char *method, int *bcidx, int *framesize, char *decode)
2202 {
2203 bccheck1(pc, fp, method, bcidx, framesize, decode);
2204 }
2205 }
2206
2207 #endif // BUILTIN_SIM
2208 #endif // !PRODUCT