1 /*
2 * Copyright (c) 2008, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interp_masm.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateInterpreterGenerator.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "prims/methodHandles.hpp"
40 #include "runtime/arguments.hpp"
41 #include "runtime/deoptimization.hpp"
42 #include "runtime/frame.inline.hpp"
43 #include "runtime/sharedRuntime.hpp"
44 #include "runtime/stubRoutines.hpp"
45 #include "runtime/synchronizer.hpp"
46 #include "runtime/timer.hpp"
47 #include "runtime/vframeArray.hpp"
48 #include "utilities/debug.hpp"
49 #include "utilities/macros.hpp"
50
51 // Size of interpreter code. Increase if too small. Interpreter will
52 // fail with a guarantee ("not enough space for interpreter generation");
53 // if too small.
54 // Run with +PrintInterpreter to get the VM to print out the size.
55 // Max size with JVMTI
56 int TemplateInterpreter::InterpreterCodeSize = 180 * 1024;
57
58 #define __ _masm->
59
60 //------------------------------------------------------------------------------------------------------------------------
61
62 address TemplateInterpreterGenerator::generate_slow_signature_handler() {
63 address entry = __ pc();
64
65 // callee-save register for saving LR, shared with generate_native_entry
66 const Register Rsaved_ret_addr = AARCH64_ONLY(R21) NOT_AARCH64(Rtmp_save0);
67
68 __ mov(Rsaved_ret_addr, LR);
69
70 __ mov(R1, Rmethod);
71 __ mov(R2, Rlocals);
72 __ mov(R3, SP);
73
74 #ifdef AARCH64
75 // expand expr. stack and extended SP to avoid cutting SP in call_VM
76 __ mov(Rstack_top, SP);
77 __ str(Rstack_top, Address(FP, frame::interpreter_frame_extended_sp_offset * wordSize));
78 __ check_stack_top();
79
80 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), R1, R2, R3, false);
81
82 __ ldp(ZR, c_rarg1, Address(SP, 2*wordSize, post_indexed));
83 __ ldp(c_rarg2, c_rarg3, Address(SP, 2*wordSize, post_indexed));
84 __ ldp(c_rarg4, c_rarg5, Address(SP, 2*wordSize, post_indexed));
85 __ ldp(c_rarg6, c_rarg7, Address(SP, 2*wordSize, post_indexed));
86
87 __ ldp_d(V0, V1, Address(SP, 2*wordSize, post_indexed));
88 __ ldp_d(V2, V3, Address(SP, 2*wordSize, post_indexed));
89 __ ldp_d(V4, V5, Address(SP, 2*wordSize, post_indexed));
90 __ ldp_d(V6, V7, Address(SP, 2*wordSize, post_indexed));
91 #else
92
93 // Safer to save R9 (when scratched) since callers may have been
94 // written assuming R9 survives. This is suboptimal but
95 // probably not important for this slow case call site.
96 // Note for R9 saving: slow_signature_handler may copy register
97 // arguments above the current SP (passed as R3). It is safe for
98 // call_VM to use push and pop to protect additional values on the
99 // stack if needed.
100 __ call_VM(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), true /* save R9 if needed*/);
101 __ add(SP, SP, wordSize); // Skip R0
102 __ pop(RegisterSet(R1, R3)); // Load arguments passed in registers
103 #ifdef __ABI_HARD__
104 // Few alternatives to an always-load-FP-registers approach:
105 // - parse method signature to detect FP arguments
106 // - keep a counter/flag on a stack indicationg number of FP arguments in the method.
107 // The later has been originally implemented and tested but a conditional path could
108 // eliminate any gain imposed by avoiding 8 double word loads.
109 __ fldmiad(SP, FloatRegisterSet(D0, 8), writeback);
110 #endif // __ABI_HARD__
111 #endif // AARCH64
112
113 __ ret(Rsaved_ret_addr);
114
115 return entry;
116 }
117
118
119 //
120 // Various method entries (that c++ and asm interpreter agree upon)
121 //------------------------------------------------------------------------------------------------------------------------
122 //
123 //
124
125 // Abstract method entry
126 // Attempt to execute abstract method. Throw exception
127 address TemplateInterpreterGenerator::generate_abstract_entry(void) {
128 address entry_point = __ pc();
129
130 #ifdef AARCH64
131 __ restore_sp_after_call(Rtemp);
132 __ restore_stack_top();
133 #endif
134
135 __ empty_expression_stack();
136
137 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
138
139 DEBUG_ONLY(STOP("generate_abstract_entry");) // Should not reach here
140 return entry_point;
141 }
142
143 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
144 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
145
146 // TODO: ARM
147 return NULL;
148
149 address entry_point = __ pc();
150 STOP("generate_math_entry");
151 return entry_point;
152 }
153
154 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
155 address entry = __ pc();
156
157 // Note: There should be a minimal interpreter frame set up when stack
158 // overflow occurs since we check explicitly for it now.
159 //
160 #ifdef ASSERT
161 { Label L;
162 __ sub(Rtemp, FP, - frame::interpreter_frame_monitor_block_top_offset * wordSize);
163 __ cmp(SP, Rtemp); // Rtemp = maximal SP for current FP,
164 // (stack grows negative)
165 __ b(L, ls); // check if frame is complete
166 __ stop ("interpreter frame not set up");
167 __ bind(L);
168 }
169 #endif // ASSERT
170
171 // Restore bcp under the assumption that the current frame is still
172 // interpreted
173 __ restore_bcp();
174
175 // expression stack must be empty before entering the VM if an exception
176 // happened
177 __ empty_expression_stack();
178
179 // throw exception
180 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
181
182 __ should_not_reach_here();
183
184 return entry;
185 }
186
187 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
188 address entry = __ pc();
189
190 // index is in R4_ArrayIndexOutOfBounds_index
191
192 InlinedString Lname(name);
193
194 // expression stack must be empty before entering the VM if an exception happened
195 __ empty_expression_stack();
196
197 // setup parameters
198 __ ldr_literal(R1, Lname);
199 __ mov(R2, R4_ArrayIndexOutOfBounds_index);
200
201 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), R1, R2);
202
203 __ nop(); // to avoid filling CPU pipeline with invalid instructions
204 __ nop();
205 __ should_not_reach_here();
206 __ bind_literal(Lname);
207
208 return entry;
209 }
210
211 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
212 address entry = __ pc();
213
214 // object is in R2_ClassCastException_obj
215
216 // expression stack must be empty before entering the VM if an exception
217 // happened
218 __ empty_expression_stack();
219
220 __ mov(R1, R2_ClassCastException_obj);
221 __ call_VM(noreg,
222 CAST_FROM_FN_PTR(address,
223 InterpreterRuntime::throw_ClassCastException),
224 R1);
225
226 __ should_not_reach_here();
227
228 return entry;
229 }
230
231 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
232 assert(!pass_oop || message == NULL, "either oop or message but not both");
233 address entry = __ pc();
234
235 InlinedString Lname(name);
236 InlinedString Lmessage(message);
237
238 if (pass_oop) {
239 // object is at TOS
240 __ pop_ptr(R2);
241 }
242
243 // expression stack must be empty before entering the VM if an exception happened
244 __ empty_expression_stack();
245
246 // setup parameters
247 __ ldr_literal(R1, Lname);
248
249 if (pass_oop) {
250 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), R1, R2);
251 } else {
252 if (message != NULL) {
253 __ ldr_literal(R2, Lmessage);
254 } else {
255 __ mov(R2, 0);
256 }
257 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), R1, R2);
258 }
259
260 // throw exception
261 __ b(Interpreter::throw_exception_entry());
262
263 __ nop(); // to avoid filling CPU pipeline with invalid instructions
264 __ nop();
265 __ bind_literal(Lname);
266 if (!pass_oop && (message != NULL)) {
267 __ bind_literal(Lmessage);
268 }
269
270 return entry;
271 }
272
273 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
274 // Not used.
275 STOP("generate_continuation_for");
276 return NULL;
277 }
278
279 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
280 address entry = __ pc();
281
282 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
283
284 #ifdef AARCH64
285 __ restore_sp_after_call(Rtemp); // Restore SP to extended SP
286 __ restore_stack_top();
287 #else
288 // Restore stack bottom in case i2c adjusted stack
289 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
290 // and NULL it as marker that SP is now tos until next java call
291 __ mov(Rtemp, (int)NULL_WORD);
292 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
293 #endif // AARCH64
294
295 __ restore_method();
296 __ restore_bcp();
297 __ restore_dispatch();
298 __ restore_locals();
299
300 const Register Rcache = R2_tmp;
301 const Register Rindex = R3_tmp;
302 __ get_cache_and_index_at_bcp(Rcache, Rindex, 1, index_size);
303
304 __ add(Rtemp, Rcache, AsmOperand(Rindex, lsl, LogBytesPerWord));
305 __ ldrb(Rtemp, Address(Rtemp, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
306 __ check_stack_top();
307 __ add(Rstack_top, Rstack_top, AsmOperand(Rtemp, lsl, Interpreter::logStackElementSize));
308
309 #ifndef AARCH64
310 __ convert_retval_to_tos(state);
311 #endif // !AARCH64
312
313 __ dispatch_next(state, step);
314
315 return entry;
316 }
317
318
319 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
320 address entry = __ pc();
321
322 __ interp_verify_oop(R0_tos, state, __FILE__, __LINE__);
323
324 #ifdef AARCH64
325 __ restore_sp_after_call(Rtemp); // Restore SP to extended SP
326 __ restore_stack_top();
327 #else
328 // The stack is not extended by deopt but we must NULL last_sp as this
329 // entry is like a "return".
330 __ mov(Rtemp, 0);
331 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
332 #endif // AARCH64
333
334 __ restore_method();
335 __ restore_bcp();
336 __ restore_dispatch();
337 __ restore_locals();
338
339 // handle exceptions
340 { Label L;
341 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset()));
342 __ cbz(Rtemp, L);
343 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
344 __ should_not_reach_here();
345 __ bind(L);
346 }
347
348 __ dispatch_next(state, step);
349
350 return entry;
351 }
352
353 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
354 #ifdef AARCH64
355 address entry = __ pc();
356 switch (type) {
357 case T_BOOLEAN:
358 __ tst(R0, 0xff);
359 __ cset(R0, ne);
360 break;
361 case T_CHAR : __ zero_extend(R0, R0, 16); break;
362 case T_BYTE : __ sign_extend(R0, R0, 8); break;
363 case T_SHORT : __ sign_extend(R0, R0, 16); break;
364 case T_INT : // fall through
365 case T_LONG : // fall through
366 case T_VOID : // fall through
367 case T_FLOAT : // fall through
368 case T_DOUBLE : /* nothing to do */ break;
369 case T_OBJECT :
370 // retrieve result from frame
371 __ ldr(R0, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize));
372 // and verify it
373 __ verify_oop(R0);
374 break;
375 default : ShouldNotReachHere();
376 }
377 __ ret();
378 return entry;
379 #else
380 // Result handlers are not used on 32-bit ARM
381 // since the returned value is already in appropriate format.
382 __ should_not_reach_here(); // to avoid empty code block
383
384 // The result handler non-zero indicates an object is returned and this is
385 // used in the native entry code.
386 return type == T_OBJECT ? (address)(-1) : NULL;
387 #endif // AARCH64
388 }
389
390 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
391 address entry = __ pc();
392 __ push(state);
393 __ call_VM(noreg, runtime_entry);
394
395 // load current bytecode
396 __ ldrb(R3_bytecode, Address(Rbcp));
397 __ dispatch_only_normal(vtos);
398 return entry;
399 }
400
401
402 // Helpers for commoning out cases in the various type of method entries.
403 //
404
405 // increment invocation count & check for overflow
406 //
407 // Note: checking for negative value instead of overflow
408 // so we have a 'sticky' overflow test
409 //
410 // In: Rmethod.
411 //
412 // Uses R0, R1, Rtemp.
413 //
414 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow,
415 Label* profile_method,
416 Label* profile_method_continue) {
417 Label done;
418 const Register Rcounters = Rtemp;
419 const Address invocation_counter(Rcounters,
420 MethodCounters::invocation_counter_offset() +
421 InvocationCounter::counter_offset());
422
423 // Note: In tiered we increment either counters in MethodCounters* or
424 // in MDO depending if we're profiling or not.
425 if (TieredCompilation) {
426 int increment = InvocationCounter::count_increment;
427 Label no_mdo;
428 if (ProfileInterpreter) {
429 // Are we profiling?
430 __ ldr(R1_tmp, Address(Rmethod, Method::method_data_offset()));
431 __ cbz(R1_tmp, no_mdo);
432 // Increment counter in the MDO
433 const Address mdo_invocation_counter(R1_tmp,
434 in_bytes(MethodData::invocation_counter_offset()) +
435 in_bytes(InvocationCounter::counter_offset()));
436 const Address mask(R1_tmp, in_bytes(MethodData::invoke_mask_offset()));
437 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, R0_tmp, Rtemp, eq, overflow);
438 __ b(done);
439 }
440 __ bind(no_mdo);
441 __ get_method_counters(Rmethod, Rcounters, done);
442 const Address mask(Rcounters, in_bytes(MethodCounters::invoke_mask_offset()));
443 __ increment_mask_and_jump(invocation_counter, increment, mask, R0_tmp, R1_tmp, eq, overflow);
444 __ bind(done);
445 } else { // not TieredCompilation
446 const Address backedge_counter(Rcounters,
447 MethodCounters::backedge_counter_offset() +
448 InvocationCounter::counter_offset());
449
450 const Register Ricnt = R0_tmp; // invocation counter
451 const Register Rbcnt = R1_tmp; // backedge counter
452
453 __ get_method_counters(Rmethod, Rcounters, done);
454
455 if (ProfileInterpreter) {
456 const Register Riic = R1_tmp;
457 __ ldr_s32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset()));
458 __ add(Riic, Riic, 1);
459 __ str_32(Riic, Address(Rcounters, MethodCounters::interpreter_invocation_counter_offset()));
460 }
461
462 // Update standard invocation counters
463
464 __ ldr_u32(Ricnt, invocation_counter);
465 __ ldr_u32(Rbcnt, backedge_counter);
466
467 __ add(Ricnt, Ricnt, InvocationCounter::count_increment);
468
469 #ifdef AARCH64
470 __ andr(Rbcnt, Rbcnt, (unsigned int)InvocationCounter::count_mask_value); // mask out the status bits
471 #else
472 __ bic(Rbcnt, Rbcnt, ~InvocationCounter::count_mask_value); // mask out the status bits
473 #endif // AARCH64
474
475 __ str_32(Ricnt, invocation_counter); // save invocation count
476 __ add(Ricnt, Ricnt, Rbcnt); // add both counters
477
478 // profile_method is non-null only for interpreted method so
479 // profile_method != NULL == !native_call
480 // BytecodeInterpreter only calls for native so code is elided.
481
482 if (ProfileInterpreter && profile_method != NULL) {
483 assert(profile_method_continue != NULL, "should be non-null");
484
485 // Test to see if we should create a method data oop
486 // Reuse R1_tmp as we don't need backedge counters anymore.
487 Address profile_limit(Rcounters, in_bytes(MethodCounters::interpreter_profile_limit_offset()));
488 __ ldr_s32(R1_tmp, profile_limit);
489 __ cmp_32(Ricnt, R1_tmp);
490 __ b(*profile_method_continue, lt);
491
492 // if no method data exists, go to profile_method
493 __ test_method_data_pointer(R1_tmp, *profile_method);
494 }
495
496 Address invoke_limit(Rcounters, in_bytes(MethodCounters::interpreter_invocation_limit_offset()));
497 __ ldr_s32(R1_tmp, invoke_limit);
498 __ cmp_32(Ricnt, R1_tmp);
499 __ b(*overflow, hs);
500 __ bind(done);
501 }
502 }
503
504 void TemplateInterpreterGenerator::generate_counter_overflow(Label& do_continue) {
505 // InterpreterRuntime::frequency_counter_overflow takes one argument
506 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
507 // The call returns the address of the verified entry point for the method or NULL
508 // if the compilation did not complete (either went background or bailed out).
509 __ mov(R1, (int)false);
510 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R1);
511
512 // jump to the interpreted entry.
513 __ b(do_continue);
514 }
515
516 void TemplateInterpreterGenerator::generate_stack_overflow_check(void) {
517 // Check if we've got enough room on the stack for
518 // - overhead;
519 // - locals;
520 // - expression stack.
521 //
522 // Registers on entry:
523 //
524 // R3 = number of additional locals
525 // R11 = max expression stack slots (AArch64 only)
526 // Rthread
527 // Rmethod
528 // Registers used: R0, R1, R2, Rtemp.
529
530 const Register Radditional_locals = R3;
531 const Register RmaxStack = AARCH64_ONLY(R11) NOT_AARCH64(R2);
532
533 // monitor entry size
534 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
535
536 // total overhead size: entry_size + (saved registers, thru expr stack bottom).
537 // be sure to change this if you add/subtract anything to/from the overhead area
538 const int overhead_size = (frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset)*wordSize + entry_size;
539
540 // Pages reserved for VM runtime calls and subsequent Java calls.
541 const int reserved_pages = JavaThread::stack_shadow_zone_size();
542
543 // Thread::stack_size() includes guard pages, and they should not be touched.
544 const int guard_pages = JavaThread::stack_guard_zone_size();
545
546 __ ldr(R0, Address(Rthread, Thread::stack_base_offset()));
547 __ ldr(R1, Address(Rthread, Thread::stack_size_offset()));
548 #ifndef AARCH64
549 __ ldr(Rtemp, Address(Rmethod, Method::const_offset()));
550 __ ldrh(RmaxStack, Address(Rtemp, ConstMethod::max_stack_offset()));
551 #endif // !AARCH64
552 __ sub_slow(Rtemp, SP, overhead_size + reserved_pages + guard_pages + Method::extra_stack_words());
553
554 // reserve space for additional locals
555 __ sub(Rtemp, Rtemp, AsmOperand(Radditional_locals, lsl, Interpreter::logStackElementSize));
556
557 // stack size
558 __ sub(R0, R0, R1);
559
560 // reserve space for expression stack
561 __ sub(Rtemp, Rtemp, AsmOperand(RmaxStack, lsl, Interpreter::logStackElementSize));
562
563 __ cmp(Rtemp, R0);
564
565 #ifdef AARCH64
566 Label L;
567 __ b(L, hi);
568 __ mov(SP, Rsender_sp); // restore SP
569 __ b(StubRoutines::throw_StackOverflowError_entry());
570 __ bind(L);
571 #else
572 __ mov(SP, Rsender_sp, ls); // restore SP
573 __ b(StubRoutines::throw_StackOverflowError_entry(), ls);
574 #endif // AARCH64
575 }
576
577
578 // Allocate monitor and lock method (asm interpreter)
579 //
580 void TemplateInterpreterGenerator::lock_method() {
581 // synchronize method
582
583 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
584 assert ((entry_size % StackAlignmentInBytes) == 0, "should keep stack alignment");
585
586 #ifdef ASSERT
587 { Label L;
588 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
589 __ tbnz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
590 __ stop("method doesn't need synchronization");
591 __ bind(L);
592 }
593 #endif // ASSERT
594
595 // get synchronization object
596 { Label done;
597 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
598 #ifdef AARCH64
599 __ ldr(R0, Address(Rlocals, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
600 __ tbz(Rtemp, JVM_ACC_STATIC_BIT, done);
601 #else
602 __ tst(Rtemp, JVM_ACC_STATIC);
603 __ ldr(R0, Address(Rlocals, Interpreter::local_offset_in_bytes(0)), eq); // get receiver (assume this is frequent case)
604 __ b(done, eq);
605 #endif // AARCH64
606 __ load_mirror(R0, Rmethod, Rtemp);
607 __ bind(done);
608 }
609
610 // add space for monitor & lock
611
612 #ifdef AARCH64
613 __ check_extended_sp(Rtemp);
614 __ sub(SP, SP, entry_size); // adjust extended SP
615 __ mov(Rtemp, SP);
616 __ str(Rtemp, Address(FP, frame::interpreter_frame_extended_sp_offset * wordSize));
617 #endif // AARCH64
618
619 __ sub(Rstack_top, Rstack_top, entry_size);
620 __ check_stack_top_on_expansion();
621 // add space for a monitor entry
622 __ str(Rstack_top, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
623 // set new monitor block top
624 __ str(R0, Address(Rstack_top, BasicObjectLock::obj_offset_in_bytes()));
625 // store object
626 __ mov(R1, Rstack_top); // monitor entry address
627 __ lock_object(R1);
628 }
629
630 #ifdef AARCH64
631
632 //
633 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
634 // and for native methods hence the shared code.
635 //
636 // On entry:
637 // R10 = ConstMethod
638 // R11 = max expr. stack (in slots), if !native_call
639 //
640 // On exit:
641 // Rbcp, Rstack_top are initialized, SP is extended
642 //
643 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
644 // Incoming registers
645 const Register RconstMethod = R10;
646 const Register RmaxStack = R11;
647 // Temporary registers
648 const Register RextendedSP = R0;
649 const Register Rcache = R1;
650 const Register Rmdp = ProfileInterpreter ? R2 : ZR;
651
652 // Generates the following stack layout (stack grows up in this picture):
653 //
654 // [ expr. stack bottom ]
655 // [ saved Rbcp ]
656 // [ current Rlocals ]
657 // [ cache ]
658 // [ mdx ]
659 // [ mirror ]
660 // [ Method* ]
661 // [ extended SP ]
662 // [ expr. stack top ]
663 // [ sender_sp ]
664 // [ saved FP ] <--- FP
665 // [ saved LR ]
666
667 // initialize fixed part of activation frame
668 __ stp(FP, LR, Address(SP, -2*wordSize, pre_indexed));
669 __ mov(FP, SP); // establish new FP
670
671 // setup Rbcp
672 if (native_call) {
673 __ mov(Rbcp, ZR); // bcp = 0 for native calls
674 } else {
675 __ add(Rbcp, RconstMethod, in_bytes(ConstMethod::codes_offset())); // get codebase
676 }
677
678 // Rstack_top & RextendedSP
679 __ sub(Rstack_top, SP, 10*wordSize);
680 if (native_call) {
681 __ sub(RextendedSP, Rstack_top, round_to(wordSize, StackAlignmentInBytes)); // reserve 1 slot for exception handling
682 } else {
683 __ sub(RextendedSP, Rstack_top, AsmOperand(RmaxStack, lsl, Interpreter::logStackElementSize));
684 __ align_reg(RextendedSP, RextendedSP, StackAlignmentInBytes);
685 }
686 __ mov(SP, RextendedSP);
687 __ check_stack_top();
688
689 // Load Rmdp
690 if (ProfileInterpreter) {
691 __ ldr(Rtemp, Address(Rmethod, Method::method_data_offset()));
692 __ tst(Rtemp, Rtemp);
693 __ add(Rtemp, Rtemp, in_bytes(MethodData::data_offset()));
694 __ csel(Rmdp, ZR, Rtemp, eq);
695 }
696
697 // Load Rcache
698 __ ldr(Rtemp, Address(RconstMethod, ConstMethod::constants_offset()));
699 __ ldr(Rcache, Address(Rtemp, ConstantPool::cache_offset_in_bytes()));
700 // Get mirror and store it in the frame as GC root for this Method*
701 __ load_mirror(Rtemp, Rmethod, Rtemp);
702
703 // Build fixed frame
704 __ stp(Rstack_top, Rbcp, Address(FP, -10*wordSize));
705 __ stp(Rlocals, Rcache, Address(FP, -8*wordSize));
706 __ stp(Rmdp, Rtemp, Address(FP, -6*wordSize));
707 __ stp(Rmethod, RextendedSP, Address(FP, -4*wordSize));
708 __ stp(ZR, Rsender_sp, Address(FP, -2*wordSize));
709 assert(frame::interpreter_frame_initial_sp_offset == -10, "interpreter frame broken");
710 assert(frame::interpreter_frame_stack_top_offset == -2, "stack top broken");
711 }
712
713 #else // AARCH64
714
715 //
716 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
717 // and for native methods hence the shared code.
718
719 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
720 // Generates the following stack layout:
721 //
722 // [ expr. stack bottom ]
723 // [ saved Rbcp ]
724 // [ current Rlocals ]
725 // [ cache ]
726 // [ mdx ]
727 // [ Method* ]
728 // [ last_sp ]
729 // [ sender_sp ]
730 // [ saved FP ] <--- FP
731 // [ saved LR ]
732
733 // initialize fixed part of activation frame
734 __ push(LR); // save return address
735 __ push(FP); // save FP
736 __ mov(FP, SP); // establish new FP
737
738 __ push(Rsender_sp);
739
740 __ mov(R0, 0);
741 __ push(R0); // leave last_sp as null
742
743 // setup Rbcp
744 if (native_call) {
745 __ mov(Rbcp, 0); // bcp = 0 for native calls
746 } else {
747 __ ldr(Rtemp, Address(Rmethod, Method::const_offset())); // get ConstMethod*
748 __ add(Rbcp, Rtemp, ConstMethod::codes_offset()); // get codebase
749 }
750
751 __ push(Rmethod); // save Method*
752 // Get mirror and store it in the frame as GC root for this Method*
753 __ load_mirror(Rtemp, Rmethod, Rtemp);
754 __ push(Rtemp);
755
756 if (ProfileInterpreter) {
757 __ ldr(Rtemp, Address(Rmethod, Method::method_data_offset()));
758 __ tst(Rtemp, Rtemp);
759 __ add(Rtemp, Rtemp, in_bytes(MethodData::data_offset()), ne);
760 __ push(Rtemp); // set the mdp (method data pointer)
761 } else {
762 __ push(R0);
763 }
764
765 __ ldr(Rtemp, Address(Rmethod, Method::const_offset()));
766 __ ldr(Rtemp, Address(Rtemp, ConstMethod::constants_offset()));
767 __ ldr(Rtemp, Address(Rtemp, ConstantPool::cache_offset_in_bytes()));
768 __ push(Rtemp); // set constant pool cache
769 __ push(Rlocals); // set locals pointer
770 __ push(Rbcp); // set bcp
771 __ push(R0); // reserve word for pointer to expression stack bottom
772 __ str(SP, Address(SP, 0)); // set expression stack bottom
773 }
774
775 #endif // AARCH64
776
777 // End of helpers
778
779 //------------------------------------------------------------------------------------------------------------------------
780 // Entry points
781 //
782 // Here we generate the various kind of entries into the interpreter.
783 // The two main entry type are generic bytecode methods and native call method.
784 // These both come in synchronized and non-synchronized versions but the
785 // frame layout they create is very similar. The other method entry
786 // types are really just special purpose entries that are really entry
787 // and interpretation all in one. These are for trivial methods like
788 // accessor, empty, or special math methods.
789 //
790 // When control flow reaches any of the entry types for the interpreter
791 // the following holds ->
792 //
793 // Arguments:
794 //
795 // Rmethod: Method*
796 // Rthread: thread
797 // Rsender_sp: sender sp
798 // Rparams (SP on 32-bit ARM): pointer to method parameters
799 //
800 // LR: return address
801 //
802 // Stack layout immediately at entry
803 //
804 // [ optional padding(*)] <--- SP (AArch64)
805 // [ parameter n ] <--- Rparams (SP on 32-bit ARM)
806 // ...
807 // [ parameter 1 ]
808 // [ expression stack ] (caller's java expression stack)
809
810 // Assuming that we don't go to one of the trivial specialized
811 // entries the stack will look like below when we are ready to execute
812 // the first bytecode (or call the native routine). The register usage
813 // will be as the template based interpreter expects.
814 //
815 // local variables follow incoming parameters immediately; i.e.
816 // the return address is saved at the end of the locals.
817 //
818 // [ reserved stack (*) ] <--- SP (AArch64)
819 // [ expr. stack ] <--- Rstack_top (SP on 32-bit ARM)
820 // [ monitor entry ]
821 // ...
822 // [ monitor entry ]
823 // [ expr. stack bottom ]
824 // [ saved Rbcp ]
825 // [ current Rlocals ]
826 // [ cache ]
827 // [ mdx ]
828 // [ mirror ]
829 // [ Method* ]
830 //
831 // 32-bit ARM:
832 // [ last_sp ]
833 //
834 // AArch64:
835 // [ extended SP (*) ]
836 // [ stack top (*) ]
837 //
838 // [ sender_sp ]
839 // [ saved FP ] <--- FP
840 // [ saved LR ]
841 // [ optional padding(*)]
842 // [ local variable m ]
843 // ...
844 // [ local variable 1 ]
845 // [ parameter n ]
846 // ...
847 // [ parameter 1 ] <--- Rlocals
848 //
849 // (*) - AArch64 only
850 //
851
852 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) {
853 #if INCLUDE_ALL_GCS
854 if (UseG1GC) {
855 // Code: _aload_0, _getfield, _areturn
856 // parameter size = 1
857 //
858 // The code that gets generated by this routine is split into 2 parts:
859 // 1. The "intrinsified" code for G1 (or any SATB based GC),
860 // 2. The slow path - which is an expansion of the regular method entry.
861 //
862 // Notes:-
863 // * In the G1 code we do not check whether we need to block for
864 // a safepoint. If G1 is enabled then we must execute the specialized
865 // code for Reference.get (except when the Reference object is null)
866 // so that we can log the value in the referent field with an SATB
867 // update buffer.
868 // If the code for the getfield template is modified so that the
869 // G1 pre-barrier code is executed when the current method is
870 // Reference.get() then going through the normal method entry
871 // will be fine.
872 // * The G1 code can, however, check the receiver object (the instance
873 // of java.lang.Reference) and jump to the slow path if null. If the
874 // Reference object is null then we obviously cannot fetch the referent
875 // and so we don't need to call the G1 pre-barrier. Thus we can use the
876 // regular method entry code to generate the NPE.
877 //
878 // This code is based on generate_accessor_enty.
879 //
880 // Rmethod: Method*
881 // Rthread: thread
882 // Rsender_sp: sender sp, must be preserved for slow path, set SP to it on fast path
883 // Rparams: parameters
884
885 address entry = __ pc();
886 Label slow_path;
887 const Register Rthis = R0;
888 const Register Rret_addr = Rtmp_save1;
889 assert_different_registers(Rthis, Rret_addr, Rsender_sp);
890
891 const int referent_offset = java_lang_ref_Reference::referent_offset;
892 guarantee(referent_offset > 0, "referent offset not initialized");
893
894 // Check if local 0 != NULL
895 // If the receiver is null then it is OK to jump to the slow path.
896 __ ldr(Rthis, Address(Rparams));
897 __ cbz(Rthis, slow_path);
898
899 // Generate the G1 pre-barrier code to log the value of
900 // the referent field in an SATB buffer.
901
902 // Load the value of the referent field.
903 __ load_heap_oop(R0, Address(Rthis, referent_offset));
904
905 // Preserve LR
906 __ mov(Rret_addr, LR);
907
908 __ g1_write_barrier_pre(noreg, // store_addr
909 noreg, // new_val
910 R0, // pre_val
911 Rtemp, // tmp1
912 R1_tmp); // tmp2
913
914 // _areturn
915 __ mov(SP, Rsender_sp);
916 __ ret(Rret_addr);
917
918 // generate a vanilla interpreter entry as the slow path
919 __ bind(slow_path);
920 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
921 return entry;
922 }
923 #endif // INCLUDE_ALL_GCS
924
925 // If G1 is not enabled then attempt to go through the normal entry point
926 return NULL;
927 }
928
929 // Not supported
930 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { return NULL; }
931 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; }
932 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { return NULL; }
933
934 //
935 // Interpreter stub for calling a native method. (asm interpreter)
936 // This sets up a somewhat different looking stack for calling the native method
937 // than the typical interpreter frame setup.
938 //
939
940 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) {
941 // determine code generation flags
942 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
943
944 // Incoming registers:
945 //
946 // Rmethod: Method*
947 // Rthread: thread
948 // Rsender_sp: sender sp
949 // Rparams: parameters
950
951 address entry_point = __ pc();
952
953 // Register allocation
954 const Register Rsize_of_params = AARCH64_ONLY(R20) NOT_AARCH64(R6);
955 const Register Rsig_handler = AARCH64_ONLY(R21) NOT_AARCH64(Rtmp_save0 /* R4 */);
956 const Register Rnative_code = AARCH64_ONLY(R22) NOT_AARCH64(Rtmp_save1 /* R5 */);
957 const Register Rresult_handler = AARCH64_ONLY(Rsig_handler) NOT_AARCH64(R6);
958
959 #ifdef AARCH64
960 const Register RconstMethod = R10; // also used in generate_fixed_frame (should match)
961 const Register Rsaved_result = Rnative_code;
962 const FloatRegister Dsaved_result = V8;
963 #else
964 const Register Rsaved_result_lo = Rtmp_save0; // R4
965 const Register Rsaved_result_hi = Rtmp_save1; // R5
966 FloatRegister saved_result_fp;
967 #endif // AARCH64
968
969
970 #ifdef AARCH64
971 __ ldr(RconstMethod, Address(Rmethod, Method::const_offset()));
972 __ ldrh(Rsize_of_params, Address(RconstMethod, ConstMethod::size_of_parameters_offset()));
973 #else
974 __ ldr(Rsize_of_params, Address(Rmethod, Method::const_offset()));
975 __ ldrh(Rsize_of_params, Address(Rsize_of_params, ConstMethod::size_of_parameters_offset()));
976 #endif // AARCH64
977
978 // native calls don't need the stack size check since they have no expression stack
979 // and the arguments are already on the stack and we only add a handful of words
980 // to the stack
981
982 // compute beginning of parameters (Rlocals)
983 __ sub(Rlocals, Rparams, wordSize);
984 __ add(Rlocals, Rlocals, AsmOperand(Rsize_of_params, lsl, Interpreter::logStackElementSize));
985
986 #ifdef AARCH64
987 int extra_stack_reserve = 2*wordSize; // extra space for oop_temp
988 if(__ can_post_interpreter_events()) {
989 // extra space for saved results
990 extra_stack_reserve += 2*wordSize;
991 }
992 // reserve extra stack space and nullify oop_temp slot
993 __ stp(ZR, ZR, Address(SP, -extra_stack_reserve, pre_indexed));
994 #else
995 // reserve stack space for oop_temp
996 __ mov(R0, 0);
997 __ push(R0);
998 #endif // AARCH64
999
1000 generate_fixed_frame(true); // Note: R9 is now saved in the frame
1001
1002 // make sure method is native & not abstract
1003 #ifdef ASSERT
1004 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1005 {
1006 Label L;
1007 __ tbnz(Rtemp, JVM_ACC_NATIVE_BIT, L);
1008 __ stop("tried to execute non-native method as native");
1009 __ bind(L);
1010 }
1011 { Label L;
1012 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L);
1013 __ stop("tried to execute abstract method in interpreter");
1014 __ bind(L);
1015 }
1016 #endif
1017
1018 // increment invocation count & check for overflow
1019 Label invocation_counter_overflow;
1020 if (inc_counter) {
1021 if (synchronized) {
1022 // Avoid unlocking method's monitor in case of exception, as it has not
1023 // been locked yet.
1024 __ set_do_not_unlock_if_synchronized(true, Rtemp);
1025 }
1026 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1027 }
1028
1029 Label continue_after_compile;
1030 __ bind(continue_after_compile);
1031
1032 if (inc_counter && synchronized) {
1033 __ set_do_not_unlock_if_synchronized(false, Rtemp);
1034 }
1035
1036 // check for synchronized methods
1037 // Must happen AFTER invocation_counter check and stack overflow check,
1038 // so method is not locked if overflows.
1039 //
1040 if (synchronized) {
1041 lock_method();
1042 } else {
1043 // no synchronization necessary
1044 #ifdef ASSERT
1045 { Label L;
1046 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1047 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
1048 __ stop("method needs synchronization");
1049 __ bind(L);
1050 }
1051 #endif
1052 }
1053
1054 // start execution
1055 #ifdef ASSERT
1056 { Label L;
1057 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
1058 __ cmp(Rtemp, Rstack_top);
1059 __ b(L, eq);
1060 __ stop("broken stack frame setup in interpreter");
1061 __ bind(L);
1062 }
1063 #endif
1064 __ check_extended_sp(Rtemp);
1065
1066 // jvmti/dtrace support
1067 __ notify_method_entry();
1068 #if R9_IS_SCRATCHED
1069 __ restore_method();
1070 #endif
1071
1072 {
1073 Label L;
1074 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset()));
1075 __ cbnz(Rsig_handler, L);
1076 __ mov(R1, Rmethod);
1077 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1, true);
1078 __ ldr(Rsig_handler, Address(Rmethod, Method::signature_handler_offset()));
1079 __ bind(L);
1080 }
1081
1082 {
1083 Label L;
1084 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset()));
1085 __ cbnz(Rnative_code, L);
1086 __ mov(R1, Rmethod);
1087 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R1);
1088 __ ldr(Rnative_code, Address(Rmethod, Method::native_function_offset()));
1089 __ bind(L);
1090 }
1091
1092 // Allocate stack space for arguments
1093
1094 #ifdef AARCH64
1095 __ sub(Rtemp, SP, Rsize_of_params, ex_uxtw, LogBytesPerWord);
1096 __ align_reg(SP, Rtemp, StackAlignmentInBytes);
1097
1098 // Allocate more stack space to accomodate all arguments passed on GP and FP registers:
1099 // 8 * wordSize for GPRs
1100 // 8 * wordSize for FPRs
1101 int reg_arguments = round_to(8*wordSize + 8*wordSize, StackAlignmentInBytes);
1102 #else
1103
1104 // C functions need aligned stack
1105 __ bic(SP, SP, StackAlignmentInBytes - 1);
1106 // Multiply by BytesPerLong instead of BytesPerWord, because calling convention
1107 // may require empty slots due to long alignment, e.g. func(int, jlong, int, jlong)
1108 __ sub(SP, SP, AsmOperand(Rsize_of_params, lsl, LogBytesPerLong));
1109
1110 #ifdef __ABI_HARD__
1111 // Allocate more stack space to accomodate all GP as well as FP registers:
1112 // 4 * wordSize
1113 // 8 * BytesPerLong
1114 int reg_arguments = round_to((4*wordSize) + (8*BytesPerLong), StackAlignmentInBytes);
1115 #else
1116 // Reserve at least 4 words on the stack for loading
1117 // of parameters passed on registers (R0-R3).
1118 // See generate_slow_signature_handler().
1119 // It is also used for JNIEnv & class additional parameters.
1120 int reg_arguments = 4 * wordSize;
1121 #endif // __ABI_HARD__
1122 #endif // AARCH64
1123
1124 __ sub(SP, SP, reg_arguments);
1125
1126
1127 // Note: signature handler blows R4 (32-bit ARM) or R21 (AArch64) besides all scratch registers.
1128 // See AbstractInterpreterGenerator::generate_slow_signature_handler().
1129 __ call(Rsig_handler);
1130 #if R9_IS_SCRATCHED
1131 __ restore_method();
1132 #endif
1133 __ mov(Rresult_handler, R0);
1134
1135 // Pass JNIEnv and mirror for static methods
1136 {
1137 Label L;
1138 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1139 __ add(R0, Rthread, in_bytes(JavaThread::jni_environment_offset()));
1140 __ tbz(Rtemp, JVM_ACC_STATIC_BIT, L);
1141 __ load_mirror(Rtemp, Rmethod, Rtemp);
1142 __ add(R1, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1143 __ str(Rtemp, Address(R1, 0));
1144 __ bind(L);
1145 }
1146
1147 __ set_last_Java_frame(SP, FP, true, Rtemp);
1148
1149 // Changing state to _thread_in_native must be the last thing to do
1150 // before the jump to native code. At this moment stack must be
1151 // safepoint-safe and completely prepared for stack walking.
1152 #ifdef ASSERT
1153 {
1154 Label L;
1155 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
1156 __ cmp_32(Rtemp, _thread_in_Java);
1157 __ b(L, eq);
1158 __ stop("invalid thread state");
1159 __ bind(L);
1160 }
1161 #endif
1162
1163 #ifdef AARCH64
1164 __ mov(Rtemp, _thread_in_native);
1165 __ add(Rtemp2, Rthread, in_bytes(JavaThread::thread_state_offset()));
1166 // STLR is used to force all preceding writes to be observed prior to thread state change
1167 __ stlr_w(Rtemp, Rtemp2);
1168 #else
1169 // Force all preceding writes to be observed prior to thread state change
1170 __ membar(MacroAssembler::StoreStore, Rtemp);
1171
1172 __ mov(Rtemp, _thread_in_native);
1173 __ str(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
1174 #endif // AARCH64
1175
1176 __ call(Rnative_code);
1177 #if R9_IS_SCRATCHED
1178 __ restore_method();
1179 #endif
1180
1181 // Set FPSCR/FPCR to a known state
1182 if (AlwaysRestoreFPU) {
1183 __ restore_default_fp_mode();
1184 }
1185
1186 // Do safepoint check
1187 __ mov(Rtemp, _thread_in_native_trans);
1188 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
1189
1190 // Force this write out before the read below
1191 __ membar(MacroAssembler::StoreLoad, Rtemp);
1192
1193 __ ldr_global_s32(Rtemp, SafepointSynchronize::address_of_state());
1194
1195 // Protect the return value in the interleaved code: save it to callee-save registers.
1196 #ifdef AARCH64
1197 __ mov(Rsaved_result, R0);
1198 __ fmov_d(Dsaved_result, D0);
1199 #else
1200 __ mov(Rsaved_result_lo, R0);
1201 __ mov(Rsaved_result_hi, R1);
1202 #ifdef __ABI_HARD__
1203 // preserve native FP result in a callee-saved register
1204 saved_result_fp = D8;
1205 __ fcpyd(saved_result_fp, D0);
1206 #else
1207 saved_result_fp = fnoreg;
1208 #endif // __ABI_HARD__
1209 #endif // AARCH64
1210
1211 {
1212 __ ldr_u32(R3, Address(Rthread, JavaThread::suspend_flags_offset()));
1213 __ cmp(Rtemp, SafepointSynchronize::_not_synchronized);
1214 __ cond_cmp(R3, 0, eq);
1215
1216 #ifdef AARCH64
1217 Label L;
1218 __ b(L, eq);
1219 __ mov(R0, Rthread);
1220 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::none);
1221 __ bind(L);
1222 #else
1223 __ mov(R0, Rthread, ne);
1224 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), relocInfo::none, ne);
1225 #if R9_IS_SCRATCHED
1226 __ restore_method();
1227 #endif
1228 #endif // AARCH64
1229 }
1230
1231 // Perform Native->Java thread transition
1232 __ mov(Rtemp, _thread_in_Java);
1233 __ str_32(Rtemp, Address(Rthread, JavaThread::thread_state_offset()));
1234
1235 // Zero handles and last_java_sp
1236 __ reset_last_Java_frame(Rtemp);
1237 __ ldr(R3, Address(Rthread, JavaThread::active_handles_offset()));
1238 __ str_32(__ zero_register(Rtemp), Address(R3, JNIHandleBlock::top_offset_in_bytes()));
1239 if (CheckJNICalls) {
1240 __ str(__ zero_register(Rtemp), Address(Rthread, JavaThread::pending_jni_exception_check_fn_offset()));
1241 }
1242
1243 // Unbox if the result is non-zero object
1244 #ifdef AARCH64
1245 {
1246 Label L, Lnull;
1247 __ mov_slow(Rtemp, AbstractInterpreter::result_handler(T_OBJECT));
1248 __ cmp(Rresult_handler, Rtemp);
1249 __ b(L, ne);
1250 __ cbz(Rsaved_result, Lnull);
1251 __ ldr(Rsaved_result, Address(Rsaved_result));
1252 __ bind(Lnull);
1253 // Store oop on the stack for GC
1254 __ str(Rsaved_result, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize));
1255 __ bind(L);
1256 }
1257 #else
1258 __ tst(Rsaved_result_lo, Rresult_handler);
1259 __ ldr(Rsaved_result_lo, Address(Rsaved_result_lo), ne);
1260
1261 // Store oop on the stack for GC
1262 __ cmp(Rresult_handler, 0);
1263 __ str(Rsaved_result_lo, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize), ne);
1264 #endif // AARCH64
1265
1266 #ifdef AARCH64
1267 // Restore SP (drop native parameters area), to keep SP in sync with extended_sp in frame
1268 __ restore_sp_after_call(Rtemp);
1269 __ check_stack_top();
1270 #endif // AARCH64
1271
1272 // reguard stack if StackOverflow exception happened while in native.
1273 {
1274 __ ldr_u32(Rtemp, Address(Rthread, JavaThread::stack_guard_state_offset()));
1275 __ cmp_32(Rtemp, JavaThread::stack_guard_yellow_reserved_disabled);
1276 #ifdef AARCH64
1277 Label L;
1278 __ b(L, ne);
1279 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::none);
1280 __ bind(L);
1281 #else
1282 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::none, eq);
1283 #if R9_IS_SCRATCHED
1284 __ restore_method();
1285 #endif
1286 #endif // AARCH64
1287 }
1288
1289 // check pending exceptions
1290 {
1291 __ ldr(Rtemp, Address(Rthread, Thread::pending_exception_offset()));
1292 #ifdef AARCH64
1293 Label L;
1294 __ cbz(Rtemp, L);
1295 __ mov_pc_to(Rexception_pc);
1296 __ b(StubRoutines::forward_exception_entry());
1297 __ bind(L);
1298 #else
1299 __ cmp(Rtemp, 0);
1300 __ mov(Rexception_pc, PC, ne);
1301 __ b(StubRoutines::forward_exception_entry(), ne);
1302 #endif // AARCH64
1303 }
1304
1305 if (synchronized) {
1306 // address of first monitor
1307 __ sub(R1, FP, - (frame::interpreter_frame_monitor_block_bottom_offset - frame::interpreter_frame_monitor_size()) * wordSize);
1308 __ unlock_object(R1);
1309 }
1310
1311 // jvmti/dtrace support
1312 // Note: This must happen _after_ handling/throwing any exceptions since
1313 // the exception handler code notifies the runtime of method exits
1314 // too. If this happens before, method entry/exit notifications are
1315 // not properly paired (was bug - gri 11/22/99).
1316 #ifdef AARCH64
1317 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI, true, Rsaved_result, noreg, Dsaved_result);
1318 #else
1319 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI, true, Rsaved_result_lo, Rsaved_result_hi, saved_result_fp);
1320 #endif // AARCH64
1321
1322 // Restore the result. Oop result is restored from the stack.
1323 #ifdef AARCH64
1324 __ mov(R0, Rsaved_result);
1325 __ fmov_d(D0, Dsaved_result);
1326
1327 __ blr(Rresult_handler);
1328 #else
1329 __ cmp(Rresult_handler, 0);
1330 __ ldr(R0, Address(FP, frame::interpreter_frame_oop_temp_offset * wordSize), ne);
1331 __ mov(R0, Rsaved_result_lo, eq);
1332 __ mov(R1, Rsaved_result_hi);
1333
1334 #ifdef __ABI_HARD__
1335 // reload native FP result
1336 __ fcpyd(D0, D8);
1337 #endif // __ABI_HARD__
1338
1339 #ifdef ASSERT
1340 if (VerifyOops) {
1341 Label L;
1342 __ cmp(Rresult_handler, 0);
1343 __ b(L, eq);
1344 __ verify_oop(R0);
1345 __ bind(L);
1346 }
1347 #endif // ASSERT
1348 #endif // AARCH64
1349
1350 // Restore FP/LR, sender_sp and return
1351 #ifdef AARCH64
1352 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_sender_sp_offset * wordSize));
1353 __ ldp(FP, LR, Address(FP));
1354 __ mov(SP, Rtemp);
1355 #else
1356 __ mov(Rtemp, FP);
1357 __ ldmia(FP, RegisterSet(FP) | RegisterSet(LR));
1358 __ ldr(SP, Address(Rtemp, frame::interpreter_frame_sender_sp_offset * wordSize));
1359 #endif // AARCH64
1360
1361 __ ret();
1362
1363 if (inc_counter) {
1364 // Handle overflow of counter and compile method
1365 __ bind(invocation_counter_overflow);
1366 generate_counter_overflow(continue_after_compile);
1367 }
1368
1369 return entry_point;
1370 }
1371
1372 //
1373 // Generic interpreted method entry to (asm) interpreter
1374 //
1375 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) {
1376 // determine code generation flags
1377 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1378
1379 // Rmethod: Method*
1380 // Rthread: thread
1381 // Rsender_sp: sender sp (could differ from SP if we were called via c2i)
1382 // Rparams: pointer to the last parameter in the stack
1383
1384 address entry_point = __ pc();
1385
1386 const Register RconstMethod = AARCH64_ONLY(R10) NOT_AARCH64(R3);
1387
1388 #ifdef AARCH64
1389 const Register RmaxStack = R11;
1390 const Register RlocalsBase = R12;
1391 #endif // AARCH64
1392
1393 __ ldr(RconstMethod, Address(Rmethod, Method::const_offset()));
1394
1395 __ ldrh(R2, Address(RconstMethod, ConstMethod::size_of_parameters_offset()));
1396 __ ldrh(R3, Address(RconstMethod, ConstMethod::size_of_locals_offset()));
1397
1398 // setup Rlocals
1399 __ sub(Rlocals, Rparams, wordSize);
1400 __ add(Rlocals, Rlocals, AsmOperand(R2, lsl, Interpreter::logStackElementSize));
1401
1402 __ sub(R3, R3, R2); // number of additional locals
1403
1404 #ifdef AARCH64
1405 // setup RmaxStack
1406 __ ldrh(RmaxStack, Address(RconstMethod, ConstMethod::max_stack_offset()));
1407 __ add(RmaxStack, RmaxStack, MAX2(1, Method::extra_stack_entries())); // reserve slots for exception handler and JSR292 appendix argument
1408 #endif // AARCH64
1409
1410 // see if we've got enough room on the stack for locals plus overhead.
1411 generate_stack_overflow_check();
1412
1413 #ifdef AARCH64
1414
1415 // allocate space for locals
1416 {
1417 __ sub(RlocalsBase, Rparams, AsmOperand(R3, lsl, Interpreter::logStackElementSize));
1418 __ align_reg(SP, RlocalsBase, StackAlignmentInBytes);
1419 }
1420
1421 // explicitly initialize locals
1422 {
1423 Label zero_loop, done;
1424 __ cbz(R3, done);
1425
1426 __ tbz(R3, 0, zero_loop);
1427 __ subs(R3, R3, 1);
1428 __ str(ZR, Address(RlocalsBase, wordSize, post_indexed));
1429 __ b(done, eq);
1430
1431 __ bind(zero_loop);
1432 __ subs(R3, R3, 2);
1433 __ stp(ZR, ZR, Address(RlocalsBase, 2*wordSize, post_indexed));
1434 __ b(zero_loop, ne);
1435
1436 __ bind(done);
1437 }
1438
1439 #else
1440 // allocate space for locals
1441 // explicitly initialize locals
1442
1443 // Loop is unrolled 4 times
1444 Label loop;
1445 __ mov(R0, 0);
1446 __ bind(loop);
1447
1448 // #1
1449 __ subs(R3, R3, 1);
1450 __ push(R0, ge);
1451
1452 // #2
1453 __ subs(R3, R3, 1, ge);
1454 __ push(R0, ge);
1455
1456 // #3
1457 __ subs(R3, R3, 1, ge);
1458 __ push(R0, ge);
1459
1460 // #4
1461 __ subs(R3, R3, 1, ge);
1462 __ push(R0, ge);
1463
1464 __ b(loop, gt);
1465 #endif // AARCH64
1466
1467 // initialize fixed part of activation frame
1468 generate_fixed_frame(false);
1469
1470 __ restore_dispatch();
1471
1472 // make sure method is not native & not abstract
1473 #ifdef ASSERT
1474 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1475 {
1476 Label L;
1477 __ tbz(Rtemp, JVM_ACC_NATIVE_BIT, L);
1478 __ stop("tried to execute native method as non-native");
1479 __ bind(L);
1480 }
1481 { Label L;
1482 __ tbz(Rtemp, JVM_ACC_ABSTRACT_BIT, L);
1483 __ stop("tried to execute abstract method in interpreter");
1484 __ bind(L);
1485 }
1486 #endif
1487
1488 // increment invocation count & check for overflow
1489 Label invocation_counter_overflow;
1490 Label profile_method;
1491 Label profile_method_continue;
1492 if (inc_counter) {
1493 if (synchronized) {
1494 // Avoid unlocking method's monitor in case of exception, as it has not
1495 // been locked yet.
1496 __ set_do_not_unlock_if_synchronized(true, Rtemp);
1497 }
1498 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1499 if (ProfileInterpreter) {
1500 __ bind(profile_method_continue);
1501 }
1502 }
1503 Label continue_after_compile;
1504 __ bind(continue_after_compile);
1505
1506 if (inc_counter && synchronized) {
1507 __ set_do_not_unlock_if_synchronized(false, Rtemp);
1508 }
1509 #if R9_IS_SCRATCHED
1510 __ restore_method();
1511 #endif
1512
1513 // check for synchronized methods
1514 // Must happen AFTER invocation_counter check and stack overflow check,
1515 // so method is not locked if overflows.
1516 //
1517 if (synchronized) {
1518 // Allocate monitor and lock method
1519 lock_method();
1520 } else {
1521 // no synchronization necessary
1522 #ifdef ASSERT
1523 { Label L;
1524 __ ldr_u32(Rtemp, Address(Rmethod, Method::access_flags_offset()));
1525 __ tbz(Rtemp, JVM_ACC_SYNCHRONIZED_BIT, L);
1526 __ stop("method needs synchronization");
1527 __ bind(L);
1528 }
1529 #endif
1530 }
1531
1532 // start execution
1533 #ifdef ASSERT
1534 { Label L;
1535 __ ldr(Rtemp, Address(FP, frame::interpreter_frame_monitor_block_top_offset * wordSize));
1536 __ cmp(Rtemp, Rstack_top);
1537 __ b(L, eq);
1538 __ stop("broken stack frame setup in interpreter");
1539 __ bind(L);
1540 }
1541 #endif
1542 __ check_extended_sp(Rtemp);
1543
1544 // jvmti support
1545 __ notify_method_entry();
1546 #if R9_IS_SCRATCHED
1547 __ restore_method();
1548 #endif
1549
1550 __ dispatch_next(vtos);
1551
1552 // invocation counter overflow
1553 if (inc_counter) {
1554 if (ProfileInterpreter) {
1555 // We have decided to profile this method in the interpreter
1556 __ bind(profile_method);
1557
1558 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1559 __ set_method_data_pointer_for_bcp();
1560
1561 __ b(profile_method_continue);
1562 }
1563
1564 // Handle overflow of counter and compile method
1565 __ bind(invocation_counter_overflow);
1566 generate_counter_overflow(continue_after_compile);
1567 }
1568
1569 return entry_point;
1570 }
1571
1572 //------------------------------------------------------------------------------------------------------------------------
1573 // Exceptions
1574
1575 void TemplateInterpreterGenerator::generate_throw_exception() {
1576 // Entry point in previous activation (i.e., if the caller was interpreted)
1577 Interpreter::_rethrow_exception_entry = __ pc();
1578 // Rexception_obj: exception
1579
1580 #ifndef AARCH64
1581 // Clear interpreter_frame_last_sp.
1582 __ mov(Rtemp, 0);
1583 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1584 #endif // !AARCH64
1585
1586 #if R9_IS_SCRATCHED
1587 __ restore_method();
1588 #endif
1589 __ restore_bcp();
1590 __ restore_dispatch();
1591 __ restore_locals();
1592
1593 #ifdef AARCH64
1594 __ restore_sp_after_call(Rtemp);
1595 #endif // AARCH64
1596
1597 // Entry point for exceptions thrown within interpreter code
1598 Interpreter::_throw_exception_entry = __ pc();
1599
1600 // expression stack is undefined here
1601 // Rexception_obj: exception
1602 // Rbcp: exception bcp
1603 __ verify_oop(Rexception_obj);
1604
1605 // expression stack must be empty before entering the VM in case of an exception
1606 __ empty_expression_stack();
1607 // find exception handler address and preserve exception oop
1608 __ mov(R1, Rexception_obj);
1609 __ call_VM(Rexception_obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), R1);
1610 // R0: exception handler entry point
1611 // Rexception_obj: preserved exception oop
1612 // Rbcp: bcp for exception handler
1613 __ push_ptr(Rexception_obj); // push exception which is now the only value on the stack
1614 __ jump(R0); // jump to exception handler (may be _remove_activation_entry!)
1615
1616 // If the exception is not handled in the current frame the frame is removed and
1617 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1618 //
1619 // Note: At this point the bci is still the bxi for the instruction which caused
1620 // the exception and the expression stack is empty. Thus, for any VM calls
1621 // at this point, GC will find a legal oop map (with empty expression stack).
1622
1623 // In current activation
1624 // tos: exception
1625 // Rbcp: exception bcp
1626
1627 //
1628 // JVMTI PopFrame support
1629 //
1630 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1631
1632 #ifdef AARCH64
1633 __ restore_sp_after_call(Rtemp); // restore SP to extended SP
1634 #endif // AARCH64
1635
1636 __ empty_expression_stack();
1637
1638 // Set the popframe_processing bit in _popframe_condition indicating that we are
1639 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1640 // popframe handling cycles.
1641
1642 __ ldr_s32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1643 __ orr(Rtemp, Rtemp, (unsigned)JavaThread::popframe_processing_bit);
1644 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1645
1646 {
1647 // Check to see whether we are returning to a deoptimized frame.
1648 // (The PopFrame call ensures that the caller of the popped frame is
1649 // either interpreted or compiled and deoptimizes it if compiled.)
1650 // In this case, we can't call dispatch_next() after the frame is
1651 // popped, but instead must save the incoming arguments and restore
1652 // them after deoptimization has occurred.
1653 //
1654 // Note that we don't compare the return PC against the
1655 // deoptimization blob's unpack entry because of the presence of
1656 // adapter frames in C2.
1657 Label caller_not_deoptimized;
1658 __ ldr(R0, Address(FP, frame::return_addr_offset * wordSize));
1659 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), R0);
1660 __ cbnz_32(R0, caller_not_deoptimized);
1661 #ifdef AARCH64
1662 __ NOT_TESTED();
1663 #endif
1664
1665 // Compute size of arguments for saving when returning to deoptimized caller
1666 __ restore_method();
1667 __ ldr(R0, Address(Rmethod, Method::const_offset()));
1668 __ ldrh(R0, Address(R0, ConstMethod::size_of_parameters_offset()));
1669
1670 __ logical_shift_left(R1, R0, Interpreter::logStackElementSize);
1671 // Save these arguments
1672 __ restore_locals();
1673 __ sub(R2, Rlocals, R1);
1674 __ add(R2, R2, wordSize);
1675 __ mov(R0, Rthread);
1676 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R0, R1, R2);
1677
1678 __ remove_activation(vtos, LR,
1679 /* throw_monitor_exception */ false,
1680 /* install_monitor_exception */ false,
1681 /* notify_jvmdi */ false);
1682
1683 // Inform deoptimization that it is responsible for restoring these arguments
1684 __ mov(Rtemp, JavaThread::popframe_force_deopt_reexecution_bit);
1685 __ str_32(Rtemp, Address(Rthread, JavaThread::popframe_condition_offset()));
1686
1687 // Continue in deoptimization handler
1688 __ ret();
1689
1690 __ bind(caller_not_deoptimized);
1691 }
1692
1693 __ remove_activation(vtos, R4,
1694 /* throw_monitor_exception */ false,
1695 /* install_monitor_exception */ false,
1696 /* notify_jvmdi */ false);
1697
1698 #ifndef AARCH64
1699 // Finish with popframe handling
1700 // A previous I2C followed by a deoptimization might have moved the
1701 // outgoing arguments further up the stack. PopFrame expects the
1702 // mutations to those outgoing arguments to be preserved and other
1703 // constraints basically require this frame to look exactly as
1704 // though it had previously invoked an interpreted activation with
1705 // no space between the top of the expression stack (current
1706 // last_sp) and the top of stack. Rather than force deopt to
1707 // maintain this kind of invariant all the time we call a small
1708 // fixup routine to move the mutated arguments onto the top of our
1709 // expression stack if necessary.
1710 __ mov(R1, SP);
1711 __ ldr(R2, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1712 // PC must point into interpreter here
1713 __ set_last_Java_frame(SP, FP, true, Rtemp);
1714 __ mov(R0, Rthread);
1715 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), R0, R1, R2);
1716 __ reset_last_Java_frame(Rtemp);
1717 #endif // !AARCH64
1718
1719 #ifdef AARCH64
1720 __ restore_sp_after_call(Rtemp);
1721 __ restore_stack_top();
1722 #else
1723 // Restore the last_sp and null it out
1724 __ ldr(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1725 __ mov(Rtemp, (int)NULL_WORD);
1726 __ str(Rtemp, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
1727 #endif // AARCH64
1728
1729 __ restore_bcp();
1730 __ restore_dispatch();
1731 __ restore_locals();
1732 __ restore_method();
1733
1734 // The method data pointer was incremented already during
1735 // call profiling. We have to restore the mdp for the current bcp.
1736 if (ProfileInterpreter) {
1737 __ set_method_data_pointer_for_bcp();
1738 }
1739
1740 // Clear the popframe condition flag
1741 assert(JavaThread::popframe_inactive == 0, "adjust this code");
1742 __ str_32(__ zero_register(Rtemp), Address(Rthread, JavaThread::popframe_condition_offset()));
1743
1744 #if INCLUDE_JVMTI
1745 {
1746 Label L_done;
1747
1748 __ ldrb(Rtemp, Address(Rbcp, 0));
1749 __ cmp(Rtemp, Bytecodes::_invokestatic);
1750 __ b(L_done, ne);
1751
1752 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1753 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1754
1755 // get local0
1756 __ ldr(R1, Address(Rlocals, 0));
1757 __ mov(R2, Rmethod);
1758 __ mov(R3, Rbcp);
1759 __ call_VM(R0, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R1, R2, R3);
1760
1761 __ cbz(R0, L_done);
1762
1763 __ str(R0, Address(Rstack_top));
1764 __ bind(L_done);
1765 }
1766 #endif // INCLUDE_JVMTI
1767
1768 __ dispatch_next(vtos);
1769 // end of PopFrame support
1770
1771 Interpreter::_remove_activation_entry = __ pc();
1772
1773 // preserve exception over this code sequence
1774 __ pop_ptr(R0_tos);
1775 __ str(R0_tos, Address(Rthread, JavaThread::vm_result_offset()));
1776 // remove the activation (without doing throws on illegalMonitorExceptions)
1777 __ remove_activation(vtos, Rexception_pc, false, true, false);
1778 // restore exception
1779 __ get_vm_result(Rexception_obj, Rtemp);
1780
1781 // Inbetween activations - previous activation type unknown yet
1782 // compute continuation point - the continuation point expects
1783 // the following registers set up:
1784 //
1785 // Rexception_obj: exception
1786 // Rexception_pc: return address/pc that threw exception
1787 // SP: expression stack of caller
1788 // FP: frame pointer of caller
1789 __ mov(c_rarg0, Rthread);
1790 __ mov(c_rarg1, Rexception_pc);
1791 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), c_rarg0, c_rarg1);
1792 // Note that an "issuing PC" is actually the next PC after the call
1793
1794 __ jump(R0); // jump to exception handler of caller
1795 }
1796
1797
1798 //
1799 // JVMTI ForceEarlyReturn support
1800 //
1801 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1802 address entry = __ pc();
1803
1804 #ifdef AARCH64
1805 __ restore_sp_after_call(Rtemp); // restore SP to extended SP
1806 #endif // AARCH64
1807
1808 __ restore_bcp();
1809 __ restore_dispatch();
1810 __ restore_locals();
1811
1812 __ empty_expression_stack();
1813
1814 __ load_earlyret_value(state);
1815
1816 // Clear the earlyret state
1817 __ ldr(Rtemp, Address(Rthread, JavaThread::jvmti_thread_state_offset()));
1818
1819 assert(JvmtiThreadState::earlyret_inactive == 0, "adjust this code");
1820 __ str_32(__ zero_register(R2), Address(Rtemp, JvmtiThreadState::earlyret_state_offset()));
1821
1822 __ remove_activation(state, LR,
1823 false, /* throw_monitor_exception */
1824 false, /* install_monitor_exception */
1825 true); /* notify_jvmdi */
1826
1827 #ifndef AARCH64
1828 // According to interpreter calling conventions, result is returned in R0/R1,
1829 // so ftos (S0) and dtos (D0) are moved to R0/R1.
1830 // This conversion should be done after remove_activation, as it uses
1831 // push(state) & pop(state) to preserve return value.
1832 __ convert_tos_to_retval(state);
1833 #endif // !AARCH64
1834 __ ret();
1835
1836 return entry;
1837 } // end of ForceEarlyReturn support
1838
1839
1840 //------------------------------------------------------------------------------------------------------------------------
1841 // Helper for vtos entry point generation
1842
1843 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1844 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1845 Label L;
1846
1847 #ifdef __SOFTFP__
1848 dep = __ pc(); // fall through
1849 #else
1850 fep = __ pc(); __ push(ftos); __ b(L);
1851 dep = __ pc(); __ push(dtos); __ b(L);
1852 #endif // __SOFTFP__
1853
1854 lep = __ pc(); __ push(ltos); __ b(L);
1855
1856 if (AARCH64_ONLY(true) NOT_AARCH64(VerifyOops)) { // can't share atos entry with itos on AArch64 or if VerifyOops
1857 aep = __ pc(); __ push(atos); __ b(L);
1858 } else {
1859 aep = __ pc(); // fall through
1860 }
1861
1862 #ifdef __SOFTFP__
1863 fep = __ pc(); // fall through
1864 #endif // __SOFTFP__
1865
1866 bep = cep = sep = // fall through
1867 iep = __ pc(); __ push(itos); // fall through
1868 vep = __ pc(); __ bind(L); // fall through
1869 generate_and_dispatch(t);
1870 }
1871
1872 //------------------------------------------------------------------------------------------------------------------------
1873
1874 // Non-product code
1875 #ifndef PRODUCT
1876 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1877 address entry = __ pc();
1878
1879 // prepare expression stack
1880 __ push(state); // save tosca
1881
1882 // pass tosca registers as arguments
1883 __ mov(R2, R0_tos);
1884 #ifdef AARCH64
1885 __ mov(R3, ZR);
1886 #else
1887 __ mov(R3, R1_tos_hi);
1888 #endif // AARCH64
1889 __ mov(R1, LR); // save return address
1890
1891 // call tracer
1892 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), R1, R2, R3);
1893
1894 __ mov(LR, R0); // restore return address
1895 __ pop(state); // restore tosca
1896
1897 // return
1898 __ ret();
1899
1900 return entry;
1901 }
1902
1903
1904 void TemplateInterpreterGenerator::count_bytecode() {
1905 __ inc_global_counter((address) &BytecodeCounter::_counter_value, 0, Rtemp, R2_tmp, true);
1906 }
1907
1908
1909 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1910 __ inc_global_counter((address)&BytecodeHistogram::_counters[0], sizeof(BytecodeHistogram::_counters[0]) * t->bytecode(), Rtemp, R2_tmp, true);
1911 }
1912
1913
1914 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1915 const Register Rindex_addr = R2_tmp;
1916 Label Lcontinue;
1917 InlinedAddress Lcounters((address)BytecodePairHistogram::_counters);
1918 InlinedAddress Lindex((address)&BytecodePairHistogram::_index);
1919 const Register Rcounters_addr = R2_tmp;
1920 const Register Rindex = R4_tmp;
1921
1922 // calculate new index for counter:
1923 // index = (_index >> log2_number_of_codes) | (bytecode << log2_number_of_codes).
1924 // (_index >> log2_number_of_codes) is previous bytecode
1925
1926 __ ldr_literal(Rindex_addr, Lindex);
1927 __ ldr_s32(Rindex, Address(Rindex_addr));
1928 __ mov_slow(Rtemp, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1929 __ orr(Rindex, Rtemp, AsmOperand(Rindex, lsr, BytecodePairHistogram::log2_number_of_codes));
1930 __ str_32(Rindex, Address(Rindex_addr));
1931
1932 // Rindex (R4) contains index of counter
1933
1934 __ ldr_literal(Rcounters_addr, Lcounters);
1935 __ ldr_s32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex));
1936 __ adds_32(Rtemp, Rtemp, 1);
1937 __ b(Lcontinue, mi); // avoid overflow
1938 __ str_32(Rtemp, Address::indexed_32(Rcounters_addr, Rindex));
1939
1940 __ b(Lcontinue);
1941
1942 __ bind_literal(Lindex);
1943 __ bind_literal(Lcounters);
1944
1945 __ bind(Lcontinue);
1946 }
1947
1948
1949 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1950 // Call a little run-time stub to avoid blow-up for each bytecode.
1951 // The run-time runtime saves the right registers, depending on
1952 // the tosca in-state for the given template.
1953 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1954 "entry must have been generated");
1955 address trace_entry = Interpreter::trace_code(t->tos_in());
1956 __ call(trace_entry, relocInfo::none);
1957 }
1958
1959
1960 void TemplateInterpreterGenerator::stop_interpreter_at() {
1961 Label Lcontinue;
1962 const Register stop_at = R2_tmp;
1963
1964 __ ldr_global_s32(Rtemp, (address) &BytecodeCounter::_counter_value);
1965 __ mov_slow(stop_at, StopInterpreterAt);
1966
1967 // test bytecode counter
1968 __ cmp(Rtemp, stop_at);
1969 __ b(Lcontinue, ne);
1970
1971 __ trace_state("stop_interpreter_at");
1972 __ breakpoint();
1973
1974 __ bind(Lcontinue);
1975 }
1976 #endif // !PRODUCT