1 /*
2 * Copyright (c) 2003, 2012, 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/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodDataOop.hpp"
34 #include "oops/methodOop.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/arguments.hpp"
39 #include "runtime/deoptimization.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/synchronizer.hpp"
44 #include "runtime/timer.hpp"
45 #include "runtime/vframeArray.hpp"
46 #include "utilities/debug.hpp"
47
48 #define __ _masm->
49
50 #ifndef CC_INTERP
51
52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
53 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
55
56 //-----------------------------------------------------------------------------
57
58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
59 address entry = __ pc();
60
61 #ifdef ASSERT
62 {
63 Label L;
64 __ lea(rax, Address(rbp,
65 frame::interpreter_frame_monitor_block_top_offset *
66 wordSize));
67 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
68 // grows negative)
69 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
70 __ stop ("interpreter frame not set up");
71 __ bind(L);
72 }
73 #endif // ASSERT
74 // Restore bcp under the assumption that the current frame is still
75 // interpreted
76 __ restore_bcp();
77
78 // expression stack must be empty before entering the VM if an
79 // exception happened
80 __ empty_expression_stack();
81 // throw exception
82 __ call_VM(noreg,
83 CAST_FROM_FN_PTR(address,
84 InterpreterRuntime::throw_StackOverflowError));
85 return entry;
86 }
87
88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
89 const char* name) {
90 address entry = __ pc();
91 // expression stack must be empty before entering the VM if an
92 // exception happened
93 __ empty_expression_stack();
94 // setup parameters
95 // ??? convention: expect aberrant index in register ebx
96 __ lea(c_rarg1, ExternalAddress((address)name));
97 __ call_VM(noreg,
98 CAST_FROM_FN_PTR(address,
99 InterpreterRuntime::
100 throw_ArrayIndexOutOfBoundsException),
101 c_rarg1, rbx);
102 return entry;
103 }
104
105 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
106 address entry = __ pc();
107
108 // object is at TOS
109 __ pop(c_rarg1);
110
111 // expression stack must be empty before entering the VM if an
112 // exception happened
113 __ empty_expression_stack();
114
115 __ call_VM(noreg,
116 CAST_FROM_FN_PTR(address,
117 InterpreterRuntime::
118 throw_ClassCastException),
119 c_rarg1);
120 return entry;
121 }
122
123 address TemplateInterpreterGenerator::generate_exception_handler_common(
124 const char* name, const char* message, bool pass_oop) {
125 assert(!pass_oop || message == NULL, "either oop or message but not both");
126 address entry = __ pc();
127 if (pass_oop) {
128 // object is at TOS
129 __ pop(c_rarg2);
130 }
131 // expression stack must be empty before entering the VM if an
132 // exception happened
133 __ empty_expression_stack();
134 // setup parameters
135 __ lea(c_rarg1, ExternalAddress((address)name));
136 if (pass_oop) {
137 __ call_VM(rax, CAST_FROM_FN_PTR(address,
138 InterpreterRuntime::
139 create_klass_exception),
140 c_rarg1, c_rarg2);
141 } else {
142 // kind of lame ExternalAddress can't take NULL because
143 // external_word_Relocation will assert.
144 if (message != NULL) {
145 __ lea(c_rarg2, ExternalAddress((address)message));
146 } else {
147 __ movptr(c_rarg2, NULL_WORD);
148 }
149 __ call_VM(rax,
150 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
151 c_rarg1, c_rarg2);
152 }
153 // throw exception
154 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
155 return entry;
156 }
157
158
159 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
160 address entry = __ pc();
161 // NULL last_sp until next java call
162 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
163 __ dispatch_next(state);
164 return entry;
165 }
166
167
168 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
169 address entry = __ pc();
170
171 // Restore stack bottom in case i2c adjusted stack
172 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
173 // and NULL it as marker that esp is now tos until next java call
174 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
175
176 __ restore_bcp();
177 __ restore_locals();
178
179 Label L_got_cache, L_giant_index;
180 if (EnableInvokeDynamic) {
181 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
182 __ jcc(Assembler::equal, L_giant_index);
183 }
184 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
185 __ bind(L_got_cache);
186 __ movl(rbx, Address(rbx, rcx,
187 Address::times_ptr,
188 in_bytes(constantPoolCacheOopDesc::base_offset()) +
189 3 * wordSize));
190 __ andl(rbx, 0xFF);
191 __ lea(rsp, Address(rsp, rbx, Address::times_8));
192 __ dispatch_next(state, step);
193
194 // out of the main line of code...
195 if (EnableInvokeDynamic) {
196 __ bind(L_giant_index);
197 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
198 __ jmp(L_got_cache);
199 }
200
201 return entry;
202 }
203
204
205 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
206 int step) {
207 address entry = __ pc();
208 // NULL last_sp until next java call
209 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
210 __ restore_bcp();
211 __ restore_locals();
212 // handle exceptions
213 {
214 Label L;
215 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
216 __ jcc(Assembler::zero, L);
217 __ call_VM(noreg,
218 CAST_FROM_FN_PTR(address,
219 InterpreterRuntime::throw_pending_exception));
220 __ should_not_reach_here();
221 __ bind(L);
222 }
223 __ dispatch_next(state, step);
224 return entry;
225 }
226
227 int AbstractInterpreter::BasicType_as_index(BasicType type) {
228 int i = 0;
229 switch (type) {
230 case T_BOOLEAN: i = 0; break;
231 case T_CHAR : i = 1; break;
232 case T_BYTE : i = 2; break;
233 case T_SHORT : i = 3; break;
234 case T_INT : i = 4; break;
235 case T_LONG : i = 5; break;
236 case T_VOID : i = 6; break;
237 case T_FLOAT : i = 7; break;
238 case T_DOUBLE : i = 8; break;
239 case T_OBJECT : i = 9; break;
240 case T_ARRAY : i = 9; break;
241 default : ShouldNotReachHere();
242 }
243 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
244 "index out of bounds");
245 return i;
246 }
247
248
249 address TemplateInterpreterGenerator::generate_result_handler_for(
250 BasicType type) {
251 address entry = __ pc();
252 switch (type) {
253 case T_BOOLEAN: __ c2bool(rax); break;
254 case T_CHAR : __ movzwl(rax, rax); break;
255 case T_BYTE : __ sign_extend_byte(rax); break;
256 case T_SHORT : __ sign_extend_short(rax); break;
257 case T_INT : /* nothing to do */ break;
258 case T_LONG : /* nothing to do */ break;
259 case T_VOID : /* nothing to do */ break;
260 case T_FLOAT : /* nothing to do */ break;
261 case T_DOUBLE : /* nothing to do */ break;
262 case T_OBJECT :
263 // retrieve result from frame
264 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
265 // and verify it
266 __ verify_oop(rax);
267 break;
268 default : ShouldNotReachHere();
269 }
270 __ ret(0); // return from result handler
271 return entry;
272 }
273
274 address TemplateInterpreterGenerator::generate_safept_entry_for(
275 TosState state,
276 address runtime_entry) {
277 address entry = __ pc();
278 __ push(state);
279 __ call_VM(noreg, runtime_entry);
280 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
281 return entry;
282 }
283
284
285
286 // Helpers for commoning out cases in the various type of method entries.
287 //
288
289
290 // increment invocation count & check for overflow
291 //
292 // Note: checking for negative value instead of overflow
293 // so we have a 'sticky' overflow test
294 //
295 // rbx: method
296 // ecx: invocation counter
297 //
298 void InterpreterGenerator::generate_counter_incr(
299 Label* overflow,
300 Label* profile_method,
301 Label* profile_method_continue) {
302 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
303 in_bytes(InvocationCounter::counter_offset()));
304 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
305 if (TieredCompilation) {
306 int increment = InvocationCounter::count_increment;
307 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
308 Label no_mdo, done;
309 if (ProfileInterpreter) {
310 // Are we profiling?
311 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
312 __ testptr(rax, rax);
313 __ jccb(Assembler::zero, no_mdo);
314 // Increment counter in the MDO
315 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
316 in_bytes(InvocationCounter::counter_offset()));
317 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
318 __ jmpb(done);
319 }
320 __ bind(no_mdo);
321 // Increment counter in methodOop (we don't need to load it, it's in ecx).
322 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
323 __ bind(done);
324 } else {
325 const Address backedge_counter(rbx,
326 methodOopDesc::backedge_counter_offset() +
327 InvocationCounter::counter_offset());
328
329 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
330 __ incrementl(Address(rbx,
331 methodOopDesc::interpreter_invocation_counter_offset()));
332 }
333 // Update standard invocation counters
334 __ movl(rax, backedge_counter); // load backedge counter
335
336 __ incrementl(rcx, InvocationCounter::count_increment);
337 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
338
339 __ movl(invocation_counter, rcx); // save invocation count
340 __ addl(rcx, rax); // add both counters
341
342 // profile_method is non-null only for interpreted method so
343 // profile_method != NULL == !native_call
344
345 if (ProfileInterpreter && profile_method != NULL) {
346 // Test to see if we should create a method data oop
347 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
348 __ jcc(Assembler::less, *profile_method_continue);
349
350 // if no method data exists, go to profile_method
351 __ test_method_data_pointer(rax, *profile_method);
352 }
353
354 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
355 __ jcc(Assembler::aboveEqual, *overflow);
356 }
357 }
358
359 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
360
361 // Asm interpreter on entry
362 // r14 - locals
363 // r13 - bcp
364 // rbx - method
365 // edx - cpool --- DOES NOT APPEAR TO BE TRUE
366 // rbp - interpreter frame
367
368 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
369 // Everything as it was on entry
370 // rdx is not restored. Doesn't appear to really be set.
371
372 const Address size_of_parameters(rbx,
373 methodOopDesc::size_of_parameters_offset());
374
375 // InterpreterRuntime::frequency_counter_overflow takes two
376 // arguments, the first (thread) is passed by call_VM, the second
377 // indicates if the counter overflow occurs at a backwards branch
378 // (NULL bcp). We pass zero for it. The call returns the address
379 // of the verified entry point for the method or NULL if the
380 // compilation did not complete (either went background or bailed
381 // out).
382 __ movl(c_rarg1, 0);
383 __ call_VM(noreg,
384 CAST_FROM_FN_PTR(address,
385 InterpreterRuntime::frequency_counter_overflow),
386 c_rarg1);
387
388 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
389 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
390 // and jump to the interpreted entry.
391 __ jmp(*do_continue, relocInfo::none);
392 }
393
394 // See if we've got enough room on the stack for locals plus overhead.
395 // The expression stack grows down incrementally, so the normal guard
396 // page mechanism will work for that.
397 //
398 // NOTE: Since the additional locals are also always pushed (wasn't
399 // obvious in generate_method_entry) so the guard should work for them
400 // too.
401 //
402 // Args:
403 // rdx: number of additional locals this frame needs (what we must check)
404 // rbx: methodOop
405 //
406 // Kills:
407 // rax
408 void InterpreterGenerator::generate_stack_overflow_check(void) {
409
410 // monitor entry size: see picture of stack set
411 // (generate_method_entry) and frame_amd64.hpp
412 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
413
414 // total overhead size: entry_size + (saved rbp through expr stack
415 // bottom). be sure to change this if you add/subtract anything
416 // to/from the overhead area
417 const int overhead_size =
418 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
419
420 const int page_size = os::vm_page_size();
421
422 Label after_frame_check;
423
424 // see if the frame is greater than one page in size. If so,
425 // then we need to verify there is enough stack space remaining
426 // for the additional locals.
427 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
428 __ jcc(Assembler::belowEqual, after_frame_check);
429
430 // compute rsp as if this were going to be the last frame on
431 // the stack before the red zone
432
433 const Address stack_base(r15_thread, Thread::stack_base_offset());
434 const Address stack_size(r15_thread, Thread::stack_size_offset());
435
436 // locals + overhead, in bytes
437 __ mov(rax, rdx);
438 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
439 __ addptr(rax, overhead_size);
440
441 #ifdef ASSERT
442 Label stack_base_okay, stack_size_okay;
443 // verify that thread stack base is non-zero
444 __ cmpptr(stack_base, (int32_t)NULL_WORD);
445 __ jcc(Assembler::notEqual, stack_base_okay);
446 __ stop("stack base is zero");
447 __ bind(stack_base_okay);
448 // verify that thread stack size is non-zero
449 __ cmpptr(stack_size, 0);
450 __ jcc(Assembler::notEqual, stack_size_okay);
451 __ stop("stack size is zero");
452 __ bind(stack_size_okay);
453 #endif
454
455 // Add stack base to locals and subtract stack size
456 __ addptr(rax, stack_base);
457 __ subptr(rax, stack_size);
458
459 // Use the maximum number of pages we might bang.
460 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
461 (StackRedPages+StackYellowPages);
462
463 // add in the red and yellow zone sizes
464 __ addptr(rax, max_pages * page_size);
465
466 // check against the current stack bottom
467 __ cmpptr(rsp, rax);
468 __ jcc(Assembler::above, after_frame_check);
469
470 // Restore sender's sp as SP. This is necessary if the sender's
471 // frame is an extended compiled frame (see gen_c2i_adapter())
472 // and safer anyway in case of JSR292 adaptations.
473
474 __ pop(rax); // return address must be moved if SP is changed
475 __ mov(rsp, r13);
476 __ push(rax);
477
478 // Note: the restored frame is not necessarily interpreted.
479 // Use the shared runtime version of the StackOverflowError.
480 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
481 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
482
483 // all done with frame size check
484 __ bind(after_frame_check);
485 }
486
487 // Allocate monitor and lock method (asm interpreter)
488 //
489 // Args:
490 // rbx: methodOop
491 // r14: locals
492 //
493 // Kills:
494 // rax
495 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
496 // rscratch1, rscratch2 (scratch regs)
497 void InterpreterGenerator::lock_method(void) {
498 // synchronize method
499 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
500 const Address monitor_block_top(
501 rbp,
502 frame::interpreter_frame_monitor_block_top_offset * wordSize);
503 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
504
505 #ifdef ASSERT
506 {
507 Label L;
508 __ movl(rax, access_flags);
509 __ testl(rax, JVM_ACC_SYNCHRONIZED);
510 __ jcc(Assembler::notZero, L);
511 __ stop("method doesn't need synchronization");
512 __ bind(L);
513 }
514 #endif // ASSERT
515
516 // get synchronization object
517 {
518 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
519 Label done;
520 __ movl(rax, access_flags);
521 __ testl(rax, JVM_ACC_STATIC);
522 // get receiver (assume this is frequent case)
523 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
524 __ jcc(Assembler::zero, done);
525 __ movptr(rax, Address(rbx, methodOopDesc::const_offset()));
526 __ movptr(rax, Address(rax, constMethodOopDesc::constants_offset()));
527 __ movptr(rax, Address(rax,
528 constantPoolOopDesc::pool_holder_offset_in_bytes()));
529 __ movptr(rax, Address(rax, mirror_offset));
530
531 #ifdef ASSERT
532 {
533 Label L;
534 __ testptr(rax, rax);
535 __ jcc(Assembler::notZero, L);
536 __ stop("synchronization object is NULL");
537 __ bind(L);
538 }
539 #endif // ASSERT
540
541 __ bind(done);
542 }
543
544 // add space for monitor & lock
545 __ subptr(rsp, entry_size); // add space for a monitor entry
546 __ movptr(monitor_block_top, rsp); // set new monitor block top
547 // store object
548 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
549 __ movptr(c_rarg1, rsp); // object address
550 __ lock_object(c_rarg1);
551 }
552
553 // Generate a fixed interpreter frame. This is identical setup for
554 // interpreted methods and for native methods hence the shared code.
555 //
556 // Args:
557 // rax: return address
558 // rbx: methodOop
559 // r14: pointer to locals
560 // r13: sender sp
561 // rdx: cp cache
562 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
563 // initialize fixed part of activation frame
564 __ push(rax); // save return address
565 __ enter(); // save old & set new rbp
566 __ push(r13); // set sender sp
567 __ push((int)NULL_WORD); // leave last_sp as null
568 __ movptr(r13, Address(rbx, methodOopDesc::const_offset())); // get constMethodOop
569 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
570 __ push(rbx); // save methodOop
571 if (ProfileInterpreter) {
572 Label method_data_continue;
573 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
574 __ testptr(rdx, rdx);
575 __ jcc(Assembler::zero, method_data_continue);
576 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
577 __ bind(method_data_continue);
578 __ push(rdx); // set the mdp (method data pointer)
579 } else {
580 __ push(0);
581 }
582
583 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
584 __ movptr(rdx, Address(rdx, constMethodOopDesc::constants_offset()));
585 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
586 __ push(rdx); // set constant pool cache
587 __ push(r14); // set locals pointer
588 if (native_call) {
589 __ push(0); // no bcp
590 } else {
591 __ push(r13); // set bcp
592 }
593 __ push(0); // reserve word for pointer to expression stack bottom
594 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
595 }
596
597 // End of helpers
598
599 // Various method entries
600 //------------------------------------------------------------------------------------------------------------------------
601 //
602 //
603
604 // Call an accessor method (assuming it is resolved, otherwise drop
605 // into vanilla (slow path) entry
606 address InterpreterGenerator::generate_accessor_entry(void) {
607 // rbx: methodOop
608
609 // r13: senderSP must preserver for slow path, set SP to it on fast path
610
611 address entry_point = __ pc();
612 Label xreturn_path;
613
614 // do fastpath for resolved accessor methods
615 if (UseFastAccessorMethods) {
616 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
617 // thereof; parameter size = 1
618 // Note: We can only use this code if the getfield has been resolved
619 // and if we don't have a null-pointer exception => check for
620 // these conditions first and use slow path if necessary.
621 Label slow_path;
622 // If we need a safepoint check, generate full interpreter entry.
623 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
624 SafepointSynchronize::_not_synchronized);
625
626 __ jcc(Assembler::notEqual, slow_path);
627 // rbx: method
628 __ movptr(rax, Address(rsp, wordSize));
629
630 // check if local 0 != NULL and read field
631 __ testptr(rax, rax);
632 __ jcc(Assembler::zero, slow_path);
633
634 // read first instruction word and extract bytecode @ 1 and index @ 2
635 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
636 __ movptr(rdi, Address(rdx, constMethodOopDesc::constants_offset()));
637 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
638 // Shift codes right to get the index on the right.
639 // The bytecode fetched looks like <index><0xb4><0x2a>
640 __ shrl(rdx, 2 * BitsPerByte);
641 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
642 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
643
644 // rax: local 0
645 // rbx: method
646 // rdx: constant pool cache index
647 // rdi: constant pool cache
648
649 // check if getfield has been resolved and read constant pool cache entry
650 // check the validity of the cache entry by testing whether _indices field
651 // contains Bytecode::_getfield in b1 byte.
652 assert(in_words(ConstantPoolCacheEntry::size()) == 4,
653 "adjust shift below");
654 __ movl(rcx,
655 Address(rdi,
656 rdx,
657 Address::times_8,
658 constantPoolCacheOopDesc::base_offset() +
659 ConstantPoolCacheEntry::indices_offset()));
660 __ shrl(rcx, 2 * BitsPerByte);
661 __ andl(rcx, 0xFF);
662 __ cmpl(rcx, Bytecodes::_getfield);
663 __ jcc(Assembler::notEqual, slow_path);
664
665 // Note: constant pool entry is not valid before bytecode is resolved
666 __ movptr(rcx,
667 Address(rdi,
668 rdx,
669 Address::times_8,
670 constantPoolCacheOopDesc::base_offset() +
671 ConstantPoolCacheEntry::f2_offset()));
672 // edx: flags
673 __ movl(rdx,
674 Address(rdi,
675 rdx,
676 Address::times_8,
677 constantPoolCacheOopDesc::base_offset() +
678 ConstantPoolCacheEntry::flags_offset()));
679
680 Label notObj, notInt, notByte, notShort;
681 const Address field_address(rax, rcx, Address::times_1);
682
683 // Need to differentiate between igetfield, agetfield, bgetfield etc.
684 // because they are different sizes.
685 // Use the type from the constant pool cache
686 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
687 // Make sure we don't need to mask edx for tosBits after the above shift
688 ConstantPoolCacheEntry::verify_tosBits();
689
690 __ cmpl(rdx, atos);
691 __ jcc(Assembler::notEqual, notObj);
692 // atos
693 __ load_heap_oop(rax, field_address);
694 __ jmp(xreturn_path);
695
696 __ bind(notObj);
697 __ cmpl(rdx, itos);
698 __ jcc(Assembler::notEqual, notInt);
699 // itos
700 __ movl(rax, field_address);
701 __ jmp(xreturn_path);
702
703 __ bind(notInt);
704 __ cmpl(rdx, btos);
705 __ jcc(Assembler::notEqual, notByte);
706 // btos
707 __ load_signed_byte(rax, field_address);
708 __ jmp(xreturn_path);
709
710 __ bind(notByte);
711 __ cmpl(rdx, stos);
712 __ jcc(Assembler::notEqual, notShort);
713 // stos
714 __ load_signed_short(rax, field_address);
715 __ jmp(xreturn_path);
716
717 __ bind(notShort);
718 #ifdef ASSERT
719 Label okay;
720 __ cmpl(rdx, ctos);
721 __ jcc(Assembler::equal, okay);
722 __ stop("what type is this?");
723 __ bind(okay);
724 #endif
725 // ctos
726 __ load_unsigned_short(rax, field_address);
727
728 __ bind(xreturn_path);
729
730 // _ireturn/_areturn
731 __ pop(rdi);
732 __ mov(rsp, r13);
733 __ jmp(rdi);
734 __ ret(0);
735
736 // generate a vanilla interpreter entry as the slow path
737 __ bind(slow_path);
738 (void) generate_normal_entry(false);
739 } else {
740 (void) generate_normal_entry(false);
741 }
742
743 return entry_point;
744 }
745
746 // Method entry for java.lang.ref.Reference.get.
747 address InterpreterGenerator::generate_Reference_get_entry(void) {
748 #ifndef SERIALGC
749 // Code: _aload_0, _getfield, _areturn
750 // parameter size = 1
751 //
752 // The code that gets generated by this routine is split into 2 parts:
753 // 1. The "intrinsified" code for G1 (or any SATB based GC),
754 // 2. The slow path - which is an expansion of the regular method entry.
755 //
756 // Notes:-
757 // * In the G1 code we do not check whether we need to block for
758 // a safepoint. If G1 is enabled then we must execute the specialized
759 // code for Reference.get (except when the Reference object is null)
760 // so that we can log the value in the referent field with an SATB
761 // update buffer.
762 // If the code for the getfield template is modified so that the
763 // G1 pre-barrier code is executed when the current method is
764 // Reference.get() then going through the normal method entry
765 // will be fine.
766 // * The G1 code can, however, check the receiver object (the instance
767 // of java.lang.Reference) and jump to the slow path if null. If the
768 // Reference object is null then we obviously cannot fetch the referent
769 // and so we don't need to call the G1 pre-barrier. Thus we can use the
770 // regular method entry code to generate the NPE.
771 //
772 // This code is based on generate_accessor_enty.
773 //
774 // rbx: methodOop
775
776 // r13: senderSP must preserve for slow path, set SP to it on fast path
777
778 address entry = __ pc();
779
780 const int referent_offset = java_lang_ref_Reference::referent_offset;
781 guarantee(referent_offset > 0, "referent offset not initialized");
782
783 if (UseG1GC) {
784 Label slow_path;
785 // rbx: method
786
787 // Check if local 0 != NULL
788 // If the receiver is null then it is OK to jump to the slow path.
789 __ movptr(rax, Address(rsp, wordSize));
790
791 __ testptr(rax, rax);
792 __ jcc(Assembler::zero, slow_path);
793
794 // rax: local 0
795 // rbx: method (but can be used as scratch now)
796 // rdx: scratch
797 // rdi: scratch
798
799 // Generate the G1 pre-barrier code to log the value of
800 // the referent field in an SATB buffer.
801
802 // Load the value of the referent field.
803 const Address field_address(rax, referent_offset);
804 __ load_heap_oop(rax, field_address);
805
806 // Generate the G1 pre-barrier code to log the value of
807 // the referent field in an SATB buffer.
808 __ g1_write_barrier_pre(noreg /* obj */,
809 rax /* pre_val */,
810 r15_thread /* thread */,
811 rbx /* tmp */,
812 true /* tosca_live */,
813 true /* expand_call */);
814
815 // _areturn
816 __ pop(rdi); // get return address
817 __ mov(rsp, r13); // set sp to sender sp
818 __ jmp(rdi);
819 __ ret(0);
820
821 // generate a vanilla interpreter entry as the slow path
822 __ bind(slow_path);
823 (void) generate_normal_entry(false);
824
825 return entry;
826 }
827 #endif // SERIALGC
828
829 // If G1 is not enabled then attempt to go through the accessor entry point
830 // Reference.get is an accessor
831 return generate_accessor_entry();
832 }
833
834
835 // Interpreter stub for calling a native method. (asm interpreter)
836 // This sets up a somewhat different looking stack for calling the
837 // native method than the typical interpreter frame setup.
838 address InterpreterGenerator::generate_native_entry(bool synchronized) {
839 // determine code generation flags
840 bool inc_counter = UseCompiler || CountCompiledCalls;
841
842 // rbx: methodOop
843 // r13: sender sp
844
845 address entry_point = __ pc();
846
847 const Address size_of_parameters(rbx, methodOopDesc::
848 size_of_parameters_offset());
849 const Address invocation_counter(rbx, methodOopDesc::
850 invocation_counter_offset() +
851 InvocationCounter::counter_offset());
852 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
853
854 // get parameter size (always needed)
855 __ load_unsigned_short(rcx, size_of_parameters);
856
857 // native calls don't need the stack size check since they have no
858 // expression stack and the arguments are already on the stack and
859 // we only add a handful of words to the stack
860
861 // rbx: methodOop
862 // rcx: size of parameters
863 // r13: sender sp
864 __ pop(rax); // get return address
865
866 // for natives the size of locals is zero
867
868 // compute beginning of parameters (r14)
869 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
870
871 // add 2 zero-initialized slots for native calls
872 // initialize result_handler slot
873 __ push((int) NULL_WORD);
874 // slot for oop temp
875 // (static native method holder mirror/jni oop result)
876 __ push((int) NULL_WORD);
877
878 if (inc_counter) {
879 __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
880 }
881
882 // initialize fixed part of activation frame
883 generate_fixed_frame(true);
884
885 // make sure method is native & not abstract
886 #ifdef ASSERT
887 __ movl(rax, access_flags);
888 {
889 Label L;
890 __ testl(rax, JVM_ACC_NATIVE);
891 __ jcc(Assembler::notZero, L);
892 __ stop("tried to execute non-native method as native");
893 __ bind(L);
894 }
895 {
896 Label L;
897 __ testl(rax, JVM_ACC_ABSTRACT);
898 __ jcc(Assembler::zero, L);
899 __ stop("tried to execute abstract method in interpreter");
900 __ bind(L);
901 }
902 #endif
903
904 // Since at this point in the method invocation the exception handler
905 // would try to exit the monitor of synchronized methods which hasn't
906 // been entered yet, we set the thread local variable
907 // _do_not_unlock_if_synchronized to true. The remove_activation will
908 // check this flag.
909
910 const Address do_not_unlock_if_synchronized(r15_thread,
911 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
912 __ movbool(do_not_unlock_if_synchronized, true);
913
914 // increment invocation count & check for overflow
915 Label invocation_counter_overflow;
916 if (inc_counter) {
917 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
918 }
919
920 Label continue_after_compile;
921 __ bind(continue_after_compile);
922
923 bang_stack_shadow_pages(true);
924
925 // reset the _do_not_unlock_if_synchronized flag
926 __ movbool(do_not_unlock_if_synchronized, false);
927
928 // check for synchronized methods
929 // Must happen AFTER invocation_counter check and stack overflow check,
930 // so method is not locked if overflows.
931 if (synchronized) {
932 lock_method();
933 } else {
934 // no synchronization necessary
935 #ifdef ASSERT
936 {
937 Label L;
938 __ movl(rax, access_flags);
939 __ testl(rax, JVM_ACC_SYNCHRONIZED);
940 __ jcc(Assembler::zero, L);
941 __ stop("method needs synchronization");
942 __ bind(L);
943 }
944 #endif
945 }
946
947 // start execution
948 #ifdef ASSERT
949 {
950 Label L;
951 const Address monitor_block_top(rbp,
952 frame::interpreter_frame_monitor_block_top_offset * wordSize);
953 __ movptr(rax, monitor_block_top);
954 __ cmpptr(rax, rsp);
955 __ jcc(Assembler::equal, L);
956 __ stop("broken stack frame setup in interpreter");
957 __ bind(L);
958 }
959 #endif
960
961 // jvmti support
962 __ notify_method_entry();
963
964 // work registers
965 const Register method = rbx;
966 const Register t = r11;
967
968 // allocate space for parameters
969 __ get_method(method);
970 __ verify_oop(method);
971 __ load_unsigned_short(t,
972 Address(method,
973 methodOopDesc::size_of_parameters_offset()));
974 __ shll(t, Interpreter::logStackElementSize);
975
976 __ subptr(rsp, t);
977 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
978 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
979
980 // get signature handler
981 {
982 Label L;
983 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
984 __ testptr(t, t);
985 __ jcc(Assembler::notZero, L);
986 __ call_VM(noreg,
987 CAST_FROM_FN_PTR(address,
988 InterpreterRuntime::prepare_native_call),
989 method);
990 __ get_method(method);
991 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
992 __ bind(L);
993 }
994
995 // call signature handler
996 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
997 "adjust this code");
998 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
999 "adjust this code");
1000 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1001 "adjust this code");
1002
1003 // The generated handlers do not touch RBX (the method oop).
1004 // However, large signatures cannot be cached and are generated
1005 // each time here. The slow-path generator can do a GC on return,
1006 // so we must reload it after the call.
1007 __ call(t);
1008 __ get_method(method); // slow path can do a GC, reload RBX
1009
1010
1011 // result handler is in rax
1012 // set result handler
1013 __ movptr(Address(rbp,
1014 (frame::interpreter_frame_result_handler_offset) * wordSize),
1015 rax);
1016
1017 // pass mirror handle if static call
1018 {
1019 Label L;
1020 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1021 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1022 __ testl(t, JVM_ACC_STATIC);
1023 __ jcc(Assembler::zero, L);
1024 // get mirror
1025 __ movptr(t, Address(method, methodOopDesc::const_offset()));
1026 __ movptr(t, Address(t, constMethodOopDesc::constants_offset()));
1027 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
1028 __ movptr(t, Address(t, mirror_offset));
1029 // copy mirror into activation frame
1030 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1031 t);
1032 // pass handle to mirror
1033 __ lea(c_rarg1,
1034 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1035 __ bind(L);
1036 }
1037
1038 // get native function entry point
1039 {
1040 Label L;
1041 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1042 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1043 __ movptr(rscratch2, unsatisfied.addr());
1044 __ cmpptr(rax, rscratch2);
1045 __ jcc(Assembler::notEqual, L);
1046 __ call_VM(noreg,
1047 CAST_FROM_FN_PTR(address,
1048 InterpreterRuntime::prepare_native_call),
1049 method);
1050 __ get_method(method);
1051 __ verify_oop(method);
1052 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1053 __ bind(L);
1054 }
1055
1056 // pass JNIEnv
1057 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1058
1059 // It is enough that the pc() points into the right code
1060 // segment. It does not have to be the correct return pc.
1061 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1062
1063 // change thread state
1064 #ifdef ASSERT
1065 {
1066 Label L;
1067 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1068 __ cmpl(t, _thread_in_Java);
1069 __ jcc(Assembler::equal, L);
1070 __ stop("Wrong thread state in native stub");
1071 __ bind(L);
1072 }
1073 #endif
1074
1075 // Change state to native
1076
1077 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1078 _thread_in_native);
1079
1080 // Call the native method.
1081 __ call(rax);
1082 // result potentially in rax or xmm0
1083
1084 // Depending on runtime options, either restore the MXCSR
1085 // register after returning from the JNI Call or verify that
1086 // it wasn't changed during -Xcheck:jni.
1087 if (RestoreMXCSROnJNICalls) {
1088 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
1089 }
1090 else if (CheckJNICalls) {
1091 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
1092 }
1093
1094 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1095 // in order to extract the result of a method call. If the order of these
1096 // pushes change or anything else is added to the stack then the code in
1097 // interpreter_frame_result must also change.
1098
1099 __ push(dtos);
1100 __ push(ltos);
1101
1102 // change thread state
1103 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1104 _thread_in_native_trans);
1105
1106 if (os::is_MP()) {
1107 if (UseMembar) {
1108 // Force this write out before the read below
1109 __ membar(Assembler::Membar_mask_bits(
1110 Assembler::LoadLoad | Assembler::LoadStore |
1111 Assembler::StoreLoad | Assembler::StoreStore));
1112 } else {
1113 // Write serialization page so VM thread can do a pseudo remote membar.
1114 // We use the current thread pointer to calculate a thread specific
1115 // offset to write to within the page. This minimizes bus traffic
1116 // due to cache line collision.
1117 __ serialize_memory(r15_thread, rscratch2);
1118 }
1119 }
1120
1121 // check for safepoint operation in progress and/or pending suspend requests
1122 {
1123 Label Continue;
1124 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1125 SafepointSynchronize::_not_synchronized);
1126
1127 Label L;
1128 __ jcc(Assembler::notEqual, L);
1129 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1130 __ jcc(Assembler::equal, Continue);
1131 __ bind(L);
1132
1133 // Don't use call_VM as it will see a possible pending exception
1134 // and forward it and never return here preventing us from
1135 // clearing _last_native_pc down below. Also can't use
1136 // call_VM_leaf either as it will check to see if r13 & r14 are
1137 // preserved and correspond to the bcp/locals pointers. So we do a
1138 // runtime call by hand.
1139 //
1140 __ mov(c_rarg0, r15_thread);
1141 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1142 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1143 __ andptr(rsp, -16); // align stack as required by ABI
1144 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1145 __ mov(rsp, r12); // restore sp
1146 __ reinit_heapbase();
1147 __ bind(Continue);
1148 }
1149
1150 // change thread state
1151 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1152
1153 // reset_last_Java_frame
1154 __ reset_last_Java_frame(true, true);
1155
1156 // reset handle block
1157 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1158 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1159
1160 // If result is an oop unbox and store it in frame where gc will see it
1161 // and result handler will pick it up
1162
1163 {
1164 Label no_oop, store_result;
1165 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1166 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1167 __ jcc(Assembler::notEqual, no_oop);
1168 // retrieve result
1169 __ pop(ltos);
1170 __ testptr(rax, rax);
1171 __ jcc(Assembler::zero, store_result);
1172 __ movptr(rax, Address(rax, 0));
1173 __ bind(store_result);
1174 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1175 // keep stack depth as expected by pushing oop which will eventually be discarde
1176 __ push(ltos);
1177 __ bind(no_oop);
1178 }
1179
1180
1181 {
1182 Label no_reguard;
1183 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1184 JavaThread::stack_guard_yellow_disabled);
1185 __ jcc(Assembler::notEqual, no_reguard);
1186
1187 __ pusha(); // XXX only save smashed registers
1188 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1189 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1190 __ andptr(rsp, -16); // align stack as required by ABI
1191 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1192 __ mov(rsp, r12); // restore sp
1193 __ popa(); // XXX only restore smashed registers
1194 __ reinit_heapbase();
1195
1196 __ bind(no_reguard);
1197 }
1198
1199
1200 // The method register is junk from after the thread_in_native transition
1201 // until here. Also can't call_VM until the bcp has been
1202 // restored. Need bcp for throwing exception below so get it now.
1203 __ get_method(method);
1204 __ verify_oop(method);
1205
1206 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1207 // r13 == code_base()
1208 __ movptr(r13, Address(method, methodOopDesc::const_offset())); // get constMethodOop
1209 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
1210 // handle exceptions (exception handling will handle unlocking!)
1211 {
1212 Label L;
1213 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1214 __ jcc(Assembler::zero, L);
1215 // Note: At some point we may want to unify this with the code
1216 // used in call_VM_base(); i.e., we should use the
1217 // StubRoutines::forward_exception code. For now this doesn't work
1218 // here because the rsp is not correctly set at this point.
1219 __ MacroAssembler::call_VM(noreg,
1220 CAST_FROM_FN_PTR(address,
1221 InterpreterRuntime::throw_pending_exception));
1222 __ should_not_reach_here();
1223 __ bind(L);
1224 }
1225
1226 // do unlocking if necessary
1227 {
1228 Label L;
1229 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1230 __ testl(t, JVM_ACC_SYNCHRONIZED);
1231 __ jcc(Assembler::zero, L);
1232 // the code below should be shared with interpreter macro
1233 // assembler implementation
1234 {
1235 Label unlock;
1236 // BasicObjectLock will be first in list, since this is a
1237 // synchronized method. However, need to check that the object
1238 // has not been unlocked by an explicit monitorexit bytecode.
1239 const Address monitor(rbp,
1240 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1241 wordSize - sizeof(BasicObjectLock)));
1242
1243 // monitor expect in c_rarg1 for slow unlock path
1244 __ lea(c_rarg1, monitor); // address of first monitor
1245
1246 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1247 __ testptr(t, t);
1248 __ jcc(Assembler::notZero, unlock);
1249
1250 // Entry already unlocked, need to throw exception
1251 __ MacroAssembler::call_VM(noreg,
1252 CAST_FROM_FN_PTR(address,
1253 InterpreterRuntime::throw_illegal_monitor_state_exception));
1254 __ should_not_reach_here();
1255
1256 __ bind(unlock);
1257 __ unlock_object(c_rarg1);
1258 }
1259 __ bind(L);
1260 }
1261
1262 // jvmti support
1263 // Note: This must happen _after_ handling/throwing any exceptions since
1264 // the exception handler code notifies the runtime of method exits
1265 // too. If this happens before, method entry/exit notifications are
1266 // not properly paired (was bug - gri 11/22/99).
1267 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1268
1269 // restore potential result in edx:eax, call result handler to
1270 // restore potential result in ST0 & handle result
1271
1272 __ pop(ltos);
1273 __ pop(dtos);
1274
1275 __ movptr(t, Address(rbp,
1276 (frame::interpreter_frame_result_handler_offset) * wordSize));
1277 __ call(t);
1278
1279 // remove activation
1280 __ movptr(t, Address(rbp,
1281 frame::interpreter_frame_sender_sp_offset *
1282 wordSize)); // get sender sp
1283 __ leave(); // remove frame anchor
1284 __ pop(rdi); // get return address
1285 __ mov(rsp, t); // set sp to sender sp
1286 __ jmp(rdi);
1287
1288 if (inc_counter) {
1289 // Handle overflow of counter and compile method
1290 __ bind(invocation_counter_overflow);
1291 generate_counter_overflow(&continue_after_compile);
1292 }
1293
1294 return entry_point;
1295 }
1296
1297 //
1298 // Generic interpreted method entry to (asm) interpreter
1299 //
1300 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1301 // determine code generation flags
1302 bool inc_counter = UseCompiler || CountCompiledCalls;
1303
1304 // ebx: methodOop
1305 // r13: sender sp
1306 address entry_point = __ pc();
1307
1308 const Address size_of_parameters(rbx,
1309 methodOopDesc::size_of_parameters_offset());
1310 const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1311 const Address invocation_counter(rbx,
1312 methodOopDesc::invocation_counter_offset() +
1313 InvocationCounter::counter_offset());
1314 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1315
1316 // get parameter size (always needed)
1317 __ load_unsigned_short(rcx, size_of_parameters);
1318
1319 // rbx: methodOop
1320 // rcx: size of parameters
1321 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1322
1323 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1324 __ subl(rdx, rcx); // rdx = no. of additional locals
1325
1326 // YYY
1327 // __ incrementl(rdx);
1328 // __ andl(rdx, -2);
1329
1330 // see if we've got enough room on the stack for locals plus overhead.
1331 generate_stack_overflow_check();
1332
1333 // get return address
1334 __ pop(rax);
1335
1336 // compute beginning of parameters (r14)
1337 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1338
1339 // rdx - # of additional locals
1340 // allocate space for locals
1341 // explicitly initialize locals
1342 {
1343 Label exit, loop;
1344 __ testl(rdx, rdx);
1345 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1346 __ bind(loop);
1347 __ push((int) NULL_WORD); // initialize local variables
1348 __ decrementl(rdx); // until everything initialized
1349 __ jcc(Assembler::greater, loop);
1350 __ bind(exit);
1351 }
1352
1353 // (pre-)fetch invocation count
1354 if (inc_counter) {
1355 __ movl(rcx, invocation_counter);
1356 }
1357 // initialize fixed part of activation frame
1358 generate_fixed_frame(false);
1359
1360 // make sure method is not native & not abstract
1361 #ifdef ASSERT
1362 __ movl(rax, access_flags);
1363 {
1364 Label L;
1365 __ testl(rax, JVM_ACC_NATIVE);
1366 __ jcc(Assembler::zero, L);
1367 __ stop("tried to execute native method as non-native");
1368 __ bind(L);
1369 }
1370 {
1371 Label L;
1372 __ testl(rax, JVM_ACC_ABSTRACT);
1373 __ jcc(Assembler::zero, L);
1374 __ stop("tried to execute abstract method in interpreter");
1375 __ bind(L);
1376 }
1377 #endif
1378
1379 // Since at this point in the method invocation the exception
1380 // handler would try to exit the monitor of synchronized methods
1381 // which hasn't been entered yet, we set the thread local variable
1382 // _do_not_unlock_if_synchronized to true. The remove_activation
1383 // will check this flag.
1384
1385 const Address do_not_unlock_if_synchronized(r15_thread,
1386 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1387 __ movbool(do_not_unlock_if_synchronized, true);
1388
1389 // increment invocation count & check for overflow
1390 Label invocation_counter_overflow;
1391 Label profile_method;
1392 Label profile_method_continue;
1393 if (inc_counter) {
1394 generate_counter_incr(&invocation_counter_overflow,
1395 &profile_method,
1396 &profile_method_continue);
1397 if (ProfileInterpreter) {
1398 __ bind(profile_method_continue);
1399 }
1400 }
1401
1402 Label continue_after_compile;
1403 __ bind(continue_after_compile);
1404
1405 // check for synchronized interpreted methods
1406 bang_stack_shadow_pages(false);
1407
1408 // reset the _do_not_unlock_if_synchronized flag
1409 __ movbool(do_not_unlock_if_synchronized, false);
1410
1411 // check for synchronized methods
1412 // Must happen AFTER invocation_counter check and stack overflow check,
1413 // so method is not locked if overflows.
1414 if (synchronized) {
1415 // Allocate monitor and lock method
1416 lock_method();
1417 } else {
1418 // no synchronization necessary
1419 #ifdef ASSERT
1420 {
1421 Label L;
1422 __ movl(rax, access_flags);
1423 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1424 __ jcc(Assembler::zero, L);
1425 __ stop("method needs synchronization");
1426 __ bind(L);
1427 }
1428 #endif
1429 }
1430
1431 // start execution
1432 #ifdef ASSERT
1433 {
1434 Label L;
1435 const Address monitor_block_top (rbp,
1436 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1437 __ movptr(rax, monitor_block_top);
1438 __ cmpptr(rax, rsp);
1439 __ jcc(Assembler::equal, L);
1440 __ stop("broken stack frame setup in interpreter");
1441 __ bind(L);
1442 }
1443 #endif
1444
1445 // jvmti support
1446 __ notify_method_entry();
1447
1448 __ dispatch_next(vtos);
1449
1450 // invocation counter overflow
1451 if (inc_counter) {
1452 if (ProfileInterpreter) {
1453 // We have decided to profile this method in the interpreter
1454 __ bind(profile_method);
1455 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1456 __ set_method_data_pointer_for_bcp();
1457 __ get_method(rbx);
1458 __ jmp(profile_method_continue);
1459 }
1460 // Handle overflow of counter and compile method
1461 __ bind(invocation_counter_overflow);
1462 generate_counter_overflow(&continue_after_compile);
1463 }
1464
1465 return entry_point;
1466 }
1467
1468 // Entry points
1469 //
1470 // Here we generate the various kind of entries into the interpreter.
1471 // The two main entry type are generic bytecode methods and native
1472 // call method. These both come in synchronized and non-synchronized
1473 // versions but the frame layout they create is very similar. The
1474 // other method entry types are really just special purpose entries
1475 // that are really entry and interpretation all in one. These are for
1476 // trivial methods like accessor, empty, or special math methods.
1477 //
1478 // When control flow reaches any of the entry types for the interpreter
1479 // the following holds ->
1480 //
1481 // Arguments:
1482 //
1483 // rbx: methodOop
1484 //
1485 // Stack layout immediately at entry
1486 //
1487 // [ return address ] <--- rsp
1488 // [ parameter n ]
1489 // ...
1490 // [ parameter 1 ]
1491 // [ expression stack ] (caller's java expression stack)
1492
1493 // Assuming that we don't go to one of the trivial specialized entries
1494 // the stack will look like below when we are ready to execute the
1495 // first bytecode (or call the native routine). The register usage
1496 // will be as the template based interpreter expects (see
1497 // interpreter_amd64.hpp).
1498 //
1499 // local variables follow incoming parameters immediately; i.e.
1500 // the return address is moved to the end of the locals).
1501 //
1502 // [ monitor entry ] <--- rsp
1503 // ...
1504 // [ monitor entry ]
1505 // [ expr. stack bottom ]
1506 // [ saved r13 ]
1507 // [ current r14 ]
1508 // [ methodOop ]
1509 // [ saved ebp ] <--- rbp
1510 // [ return address ]
1511 // [ local variable m ]
1512 // ...
1513 // [ local variable 1 ]
1514 // [ parameter n ]
1515 // ...
1516 // [ parameter 1 ] <--- r14
1517
1518 address AbstractInterpreterGenerator::generate_method_entry(
1519 AbstractInterpreter::MethodKind kind) {
1520 // determine code generation flags
1521 bool synchronized = false;
1522 address entry_point = NULL;
1523
1524 switch (kind) {
1525 case Interpreter::zerolocals : break;
1526 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1527 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1528 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1529 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1530 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1531 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1532 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1533
1534 case Interpreter::java_lang_math_sin : // fall thru
1535 case Interpreter::java_lang_math_cos : // fall thru
1536 case Interpreter::java_lang_math_tan : // fall thru
1537 case Interpreter::java_lang_math_abs : // fall thru
1538 case Interpreter::java_lang_math_log : // fall thru
1539 case Interpreter::java_lang_math_log10 : // fall thru
1540 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1541 case Interpreter::java_lang_ref_reference_get
1542 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1543 default : ShouldNotReachHere(); break;
1544 }
1545
1546 if (entry_point) {
1547 return entry_point;
1548 }
1549
1550 return ((InterpreterGenerator*) this)->
1551 generate_normal_entry(synchronized);
1552 }
1553
1554 // These should never be compiled since the interpreter will prefer
1555 // the compiled version to the intrinsic version.
1556 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1557 switch (method_kind(m)) {
1558 case Interpreter::java_lang_math_sin : // fall thru
1559 case Interpreter::java_lang_math_cos : // fall thru
1560 case Interpreter::java_lang_math_tan : // fall thru
1561 case Interpreter::java_lang_math_abs : // fall thru
1562 case Interpreter::java_lang_math_log : // fall thru
1563 case Interpreter::java_lang_math_log10 : // fall thru
1564 case Interpreter::java_lang_math_sqrt :
1565 return false;
1566 default:
1567 return true;
1568 }
1569 }
1570
1571 // How much stack a method activation needs in words.
1572 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1573 const int entry_size = frame::interpreter_frame_monitor_size();
1574
1575 // total overhead size: entry_size + (saved rbp thru expr stack
1576 // bottom). be sure to change this if you add/subtract anything
1577 // to/from the overhead area
1578 const int overhead_size =
1579 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1580
1581 const int stub_code = frame::entry_frame_after_call_words;
1582 const int extra_stack = methodOopDesc::extra_stack_entries();
1583 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1584 Interpreter::stackElementWords;
1585 return (overhead_size + method_stack + stub_code);
1586 }
1587
1588 int AbstractInterpreter::layout_activation(methodOop method,
1589 int tempcount,
1590 int popframe_extra_args,
1591 int moncount,
1592 int caller_actual_parameters,
1593 int callee_param_count,
1594 int callee_locals,
1595 frame* caller,
1596 frame* interpreter_frame,
1597 bool is_top_frame) {
1598 // Note: This calculation must exactly parallel the frame setup
1599 // in AbstractInterpreterGenerator::generate_method_entry.
1600 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1601 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1602 // right size, as determined by a previous call to this method.
1603 // It is also guaranteed to be walkable even though it is in a skeletal state
1604
1605 // fixed size of an interpreter frame:
1606 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1607 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1608 Interpreter::stackElementWords;
1609
1610 int overhead = frame::sender_sp_offset -
1611 frame::interpreter_frame_initial_sp_offset;
1612 // Our locals were accounted for by the caller (or last_frame_adjust
1613 // on the transistion) Since the callee parameters already account
1614 // for the callee's params we only need to account for the extra
1615 // locals.
1616 int size = overhead +
1617 (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1618 moncount * frame::interpreter_frame_monitor_size() +
1619 tempcount* Interpreter::stackElementWords + popframe_extra_args;
1620 if (interpreter_frame != NULL) {
1621 #ifdef ASSERT
1622 if (!EnableInvokeDynamic)
1623 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1624 // Probably, since deoptimization doesn't work yet.
1625 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1626 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1627 #endif
1628
1629 interpreter_frame->interpreter_frame_set_method(method);
1630 // NOTE the difference in using sender_sp and
1631 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1632 // the original sp of the caller (the unextended_sp) and
1633 // sender_sp is fp+16 XXX
1634 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1635
1636 #ifdef ASSERT
1637 if (caller->is_interpreted_frame()) {
1638 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1639 }
1640 #endif
1641
1642 interpreter_frame->interpreter_frame_set_locals(locals);
1643 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1644 BasicObjectLock* monbot = montop - moncount;
1645 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1646
1647 // Set last_sp
1648 intptr_t* esp = (intptr_t*) monbot -
1649 tempcount*Interpreter::stackElementWords -
1650 popframe_extra_args;
1651 interpreter_frame->interpreter_frame_set_last_sp(esp);
1652
1653 // All frames but the initial (oldest) interpreter frame we fill in have
1654 // a value for sender_sp that allows walking the stack but isn't
1655 // truly correct. Correct the value here.
1656 if (extra_locals != 0 &&
1657 interpreter_frame->sender_sp() ==
1658 interpreter_frame->interpreter_frame_sender_sp()) {
1659 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1660 extra_locals);
1661 }
1662 *interpreter_frame->interpreter_frame_cache_addr() =
1663 method->constants()->cache();
1664 }
1665 return size;
1666 }
1667
1668 //-----------------------------------------------------------------------------
1669 // Exceptions
1670
1671 void TemplateInterpreterGenerator::generate_throw_exception() {
1672 // Entry point in previous activation (i.e., if the caller was
1673 // interpreted)
1674 Interpreter::_rethrow_exception_entry = __ pc();
1675 // Restore sp to interpreter_frame_last_sp even though we are going
1676 // to empty the expression stack for the exception processing.
1677 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1678 // rax: exception
1679 // rdx: return address/pc that threw exception
1680 __ restore_bcp(); // r13 points to call/send
1681 __ restore_locals();
1682 __ reinit_heapbase(); // restore r12 as heapbase.
1683 // Entry point for exceptions thrown within interpreter code
1684 Interpreter::_throw_exception_entry = __ pc();
1685 // expression stack is undefined here
1686 // rax: exception
1687 // r13: exception bcp
1688 __ verify_oop(rax);
1689 __ mov(c_rarg1, rax);
1690
1691 // expression stack must be empty before entering the VM in case of
1692 // an exception
1693 __ empty_expression_stack();
1694 // find exception handler address and preserve exception oop
1695 __ call_VM(rdx,
1696 CAST_FROM_FN_PTR(address,
1697 InterpreterRuntime::exception_handler_for_exception),
1698 c_rarg1);
1699 // rax: exception handler entry point
1700 // rdx: preserved exception oop
1701 // r13: bcp for exception handler
1702 __ push_ptr(rdx); // push exception which is now the only value on the stack
1703 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1704
1705 // If the exception is not handled in the current frame the frame is
1706 // removed and the exception is rethrown (i.e. exception
1707 // continuation is _rethrow_exception).
1708 //
1709 // Note: At this point the bci is still the bxi for the instruction
1710 // which caused the exception and the expression stack is
1711 // empty. Thus, for any VM calls at this point, GC will find a legal
1712 // oop map (with empty expression stack).
1713
1714 // In current activation
1715 // tos: exception
1716 // esi: exception bcp
1717
1718 //
1719 // JVMTI PopFrame support
1720 //
1721
1722 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1723 __ empty_expression_stack();
1724 // Set the popframe_processing bit in pending_popframe_condition
1725 // indicating that we are currently handling popframe, so that
1726 // call_VMs that may happen later do not trigger new popframe
1727 // handling cycles.
1728 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1729 __ orl(rdx, JavaThread::popframe_processing_bit);
1730 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1731
1732 {
1733 // Check to see whether we are returning to a deoptimized frame.
1734 // (The PopFrame call ensures that the caller of the popped frame is
1735 // either interpreted or compiled and deoptimizes it if compiled.)
1736 // In this case, we can't call dispatch_next() after the frame is
1737 // popped, but instead must save the incoming arguments and restore
1738 // them after deoptimization has occurred.
1739 //
1740 // Note that we don't compare the return PC against the
1741 // deoptimization blob's unpack entry because of the presence of
1742 // adapter frames in C2.
1743 Label caller_not_deoptimized;
1744 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1745 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1746 InterpreterRuntime::interpreter_contains), c_rarg1);
1747 __ testl(rax, rax);
1748 __ jcc(Assembler::notZero, caller_not_deoptimized);
1749
1750 // Compute size of arguments for saving when returning to
1751 // deoptimized caller
1752 __ get_method(rax);
1753 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1754 size_of_parameters_offset())));
1755 __ shll(rax, Interpreter::logStackElementSize);
1756 __ restore_locals(); // XXX do we need this?
1757 __ subptr(r14, rax);
1758 __ addptr(r14, wordSize);
1759 // Save these arguments
1760 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1761 Deoptimization::
1762 popframe_preserve_args),
1763 r15_thread, rax, r14);
1764
1765 __ remove_activation(vtos, rdx,
1766 /* throw_monitor_exception */ false,
1767 /* install_monitor_exception */ false,
1768 /* notify_jvmdi */ false);
1769
1770 // Inform deoptimization that it is responsible for restoring
1771 // these arguments
1772 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1773 JavaThread::popframe_force_deopt_reexecution_bit);
1774
1775 // Continue in deoptimization handler
1776 __ jmp(rdx);
1777
1778 __ bind(caller_not_deoptimized);
1779 }
1780
1781 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1782 /* throw_monitor_exception */ false,
1783 /* install_monitor_exception */ false,
1784 /* notify_jvmdi */ false);
1785
1786 // Finish with popframe handling
1787 // A previous I2C followed by a deoptimization might have moved the
1788 // outgoing arguments further up the stack. PopFrame expects the
1789 // mutations to those outgoing arguments to be preserved and other
1790 // constraints basically require this frame to look exactly as
1791 // though it had previously invoked an interpreted activation with
1792 // no space between the top of the expression stack (current
1793 // last_sp) and the top of stack. Rather than force deopt to
1794 // maintain this kind of invariant all the time we call a small
1795 // fixup routine to move the mutated arguments onto the top of our
1796 // expression stack if necessary.
1797 __ mov(c_rarg1, rsp);
1798 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1799 // PC must point into interpreter here
1800 __ set_last_Java_frame(noreg, rbp, __ pc());
1801 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1802 __ reset_last_Java_frame(true, true);
1803 // Restore the last_sp and null it out
1804 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1805 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1806
1807 __ restore_bcp(); // XXX do we need this?
1808 __ restore_locals(); // XXX do we need this?
1809 // The method data pointer was incremented already during
1810 // call profiling. We have to restore the mdp for the current bcp.
1811 if (ProfileInterpreter) {
1812 __ set_method_data_pointer_for_bcp();
1813 }
1814
1815 // Clear the popframe condition flag
1816 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1817 JavaThread::popframe_inactive);
1818
1819 __ dispatch_next(vtos);
1820 // end of PopFrame support
1821
1822 Interpreter::_remove_activation_entry = __ pc();
1823
1824 // preserve exception over this code sequence
1825 __ pop_ptr(rax);
1826 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1827 // remove the activation (without doing throws on illegalMonitorExceptions)
1828 __ remove_activation(vtos, rdx, false, true, false);
1829 // restore exception
1830 __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1831 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1832 __ verify_oop(rax);
1833
1834 // In between activations - previous activation type unknown yet
1835 // compute continuation point - the continuation point expects the
1836 // following registers set up:
1837 //
1838 // rax: exception
1839 // rdx: return address/pc that threw exception
1840 // rsp: expression stack of caller
1841 // rbp: ebp of caller
1842 __ push(rax); // save exception
1843 __ push(rdx); // save return address
1844 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1845 SharedRuntime::exception_handler_for_return_address),
1846 r15_thread, rdx);
1847 __ mov(rbx, rax); // save exception handler
1848 __ pop(rdx); // restore return address
1849 __ pop(rax); // restore exception
1850 // Note that an "issuing PC" is actually the next PC after the call
1851 __ jmp(rbx); // jump to exception
1852 // handler of caller
1853 }
1854
1855
1856 //
1857 // JVMTI ForceEarlyReturn support
1858 //
1859 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1860 address entry = __ pc();
1861
1862 __ restore_bcp();
1863 __ restore_locals();
1864 __ empty_expression_stack();
1865 __ load_earlyret_value(state);
1866
1867 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1868 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1869
1870 // Clear the earlyret state
1871 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1872
1873 __ remove_activation(state, rsi,
1874 false, /* throw_monitor_exception */
1875 false, /* install_monitor_exception */
1876 true); /* notify_jvmdi */
1877 __ jmp(rsi);
1878
1879 return entry;
1880 } // end of ForceEarlyReturn support
1881
1882
1883 //-----------------------------------------------------------------------------
1884 // Helper for vtos entry point generation
1885
1886 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1887 address& bep,
1888 address& cep,
1889 address& sep,
1890 address& aep,
1891 address& iep,
1892 address& lep,
1893 address& fep,
1894 address& dep,
1895 address& vep) {
1896 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1897 Label L;
1898 aep = __ pc(); __ push_ptr(); __ jmp(L);
1899 fep = __ pc(); __ push_f(); __ jmp(L);
1900 dep = __ pc(); __ push_d(); __ jmp(L);
1901 lep = __ pc(); __ push_l(); __ jmp(L);
1902 bep = cep = sep =
1903 iep = __ pc(); __ push_i();
1904 vep = __ pc();
1905 __ bind(L);
1906 generate_and_dispatch(t);
1907 }
1908
1909
1910 //-----------------------------------------------------------------------------
1911 // Generation of individual instructions
1912
1913 // helpers for generate_and_dispatch
1914
1915
1916 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1917 : TemplateInterpreterGenerator(code) {
1918 generate_all(); // down here so it can be "virtual"
1919 }
1920
1921 //-----------------------------------------------------------------------------
1922
1923 // Non-product code
1924 #ifndef PRODUCT
1925 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1926 address entry = __ pc();
1927
1928 __ push(state);
1929 __ push(c_rarg0);
1930 __ push(c_rarg1);
1931 __ push(c_rarg2);
1932 __ push(c_rarg3);
1933 __ mov(c_rarg2, rax); // Pass itos
1934 #ifdef _WIN64
1935 __ movflt(xmm3, xmm0); // Pass ftos
1936 #endif
1937 __ call_VM(noreg,
1938 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1939 c_rarg1, c_rarg2, c_rarg3);
1940 __ pop(c_rarg3);
1941 __ pop(c_rarg2);
1942 __ pop(c_rarg1);
1943 __ pop(c_rarg0);
1944 __ pop(state);
1945 __ ret(0); // return from result handler
1946
1947 return entry;
1948 }
1949
1950 void TemplateInterpreterGenerator::count_bytecode() {
1951 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1952 }
1953
1954 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1955 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1956 }
1957
1958 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1959 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1960 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1961 __ orl(rbx,
1962 ((int) t->bytecode()) <<
1963 BytecodePairHistogram::log2_number_of_codes);
1964 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1965 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1966 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1967 }
1968
1969
1970 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1971 // Call a little run-time stub to avoid blow-up for each bytecode.
1972 // The run-time runtime saves the right registers, depending on
1973 // the tosca in-state for the given template.
1974
1975 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1976 "entry must have been generated");
1977 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1978 __ andptr(rsp, -16); // align stack as required by ABI
1979 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1980 __ mov(rsp, r12); // restore sp
1981 __ reinit_heapbase();
1982 }
1983
1984
1985 void TemplateInterpreterGenerator::stop_interpreter_at() {
1986 Label L;
1987 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1988 StopInterpreterAt);
1989 __ jcc(Assembler::notEqual, L);
1990 __ int3();
1991 __ bind(L);
1992 }
1993 #endif // !PRODUCT
1994 #endif // ! CC_INTERP