1 /*
2 * Copyright (c) 2003, 2013, 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/macroAssembler.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/methodData.hpp"
34 #include "oops/method.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 #include "utilities/macros.hpp"
48
49 #define __ _masm->
50
51 #ifndef CC_INTERP
52
53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
56
57 //-----------------------------------------------------------------------------
58
59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
60 address entry = __ pc();
61
62 #ifdef ASSERT
63 {
64 Label L;
65 __ lea(rax, Address(rbp,
66 frame::interpreter_frame_monitor_block_top_offset *
67 wordSize));
68 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
69 // grows negative)
70 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
71 __ stop ("interpreter frame not set up");
72 __ bind(L);
73 }
74 #endif // ASSERT
75 // Restore bcp under the assumption that the current frame is still
76 // interpreted
77 __ restore_bcp();
78
79 // expression stack must be empty before entering the VM if an
80 // exception happened
81 __ empty_expression_stack();
82 // throw exception
83 __ call_VM(noreg,
84 CAST_FROM_FN_PTR(address,
85 InterpreterRuntime::throw_StackOverflowError));
86 return entry;
87 }
88
89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
90 const char* name) {
91 address entry = __ pc();
92 // expression stack must be empty before entering the VM if an
93 // exception happened
94 __ empty_expression_stack();
95 // setup parameters
96 // ??? convention: expect aberrant index in register ebx
97 __ lea(c_rarg1, ExternalAddress((address)name));
98 __ call_VM(noreg,
99 CAST_FROM_FN_PTR(address,
100 InterpreterRuntime::
101 throw_ArrayIndexOutOfBoundsException),
102 c_rarg1, rbx);
103 return entry;
104 }
105
106 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
107 address entry = __ pc();
108
109 // object is at TOS
110 __ pop(c_rarg1);
111
112 // expression stack must be empty before entering the VM if an
113 // exception happened
114 __ empty_expression_stack();
115
116 __ call_VM(noreg,
117 CAST_FROM_FN_PTR(address,
118 InterpreterRuntime::
119 throw_ClassCastException),
120 c_rarg1);
121 return entry;
122 }
123
124 address TemplateInterpreterGenerator::generate_exception_handler_common(
125 const char* name, const char* message, bool pass_oop) {
126 assert(!pass_oop || message == NULL, "either oop or message but not both");
127 address entry = __ pc();
128 if (pass_oop) {
129 // object is at TOS
130 __ pop(c_rarg2);
131 }
132 // expression stack must be empty before entering the VM if an
133 // exception happened
134 __ empty_expression_stack();
135 // setup parameters
136 __ lea(c_rarg1, ExternalAddress((address)name));
137 if (pass_oop) {
138 __ call_VM(rax, CAST_FROM_FN_PTR(address,
139 InterpreterRuntime::
140 create_klass_exception),
141 c_rarg1, c_rarg2);
142 } else {
143 // kind of lame ExternalAddress can't take NULL because
144 // external_word_Relocation will assert.
145 if (message != NULL) {
146 __ lea(c_rarg2, ExternalAddress((address)message));
147 } else {
148 __ movptr(c_rarg2, NULL_WORD);
149 }
150 __ call_VM(rax,
151 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
152 c_rarg1, c_rarg2);
153 }
154 // throw exception
155 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
156 return entry;
157 }
158
159
160 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
161 address entry = __ pc();
162 // NULL last_sp until next java call
163 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
164 __ dispatch_next(state);
165 return entry;
166 }
167
168
169 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
170 address entry = __ pc();
171
172 // Restore stack bottom in case i2c adjusted stack
173 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
174 // and NULL it as marker that esp is now tos until next java call
175 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
176
177 __ restore_bcp();
178 __ restore_locals();
179
180 Label L_got_cache, L_giant_index;
181 if (EnableInvokeDynamic) {
182 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
183 __ jcc(Assembler::equal, L_giant_index);
184 }
185 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
186 __ bind(L_got_cache);
187 __ movl(rbx, Address(rbx, rcx,
188 Address::times_ptr,
189 in_bytes(ConstantPoolCache::base_offset()) +
190 3 * wordSize));
191 __ andl(rbx, 0xFF);
192 __ lea(rsp, Address(rsp, rbx, Address::times_8));
193 __ dispatch_next(state, step);
194
195 // out of the main line of code...
196 if (EnableInvokeDynamic) {
197 __ bind(L_giant_index);
198 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
199 __ jmp(L_got_cache);
200 }
201
202 return entry;
203 }
204
205
206 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
207 int step) {
208 address entry = __ pc();
209 // NULL last_sp until next java call
210 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
211 __ restore_bcp();
212 __ restore_locals();
213 // handle exceptions
214 {
215 Label L;
216 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
217 __ jcc(Assembler::zero, L);
218 __ call_VM(noreg,
219 CAST_FROM_FN_PTR(address,
220 InterpreterRuntime::throw_pending_exception));
221 __ should_not_reach_here();
222 __ bind(L);
223 }
224 __ dispatch_next(state, step);
225 return entry;
226 }
227
228 int AbstractInterpreter::BasicType_as_index(BasicType type) {
229 int i = 0;
230 switch (type) {
231 case T_BOOLEAN: i = 0; break;
232 case T_CHAR : i = 1; break;
233 case T_BYTE : i = 2; break;
234 case T_SHORT : i = 3; break;
235 case T_INT : i = 4; break;
236 case T_LONG : i = 5; break;
237 case T_VOID : i = 6; break;
238 case T_FLOAT : i = 7; break;
239 case T_DOUBLE : i = 8; break;
240 case T_OBJECT : i = 9; break;
241 case T_ARRAY : i = 9; break;
242 default : ShouldNotReachHere();
243 }
244 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
245 "index out of bounds");
246 return i;
247 }
248
249
250 address TemplateInterpreterGenerator::generate_result_handler_for(
251 BasicType type) {
252 address entry = __ pc();
253 switch (type) {
254 case T_BOOLEAN: __ c2bool(rax); break;
255 case T_CHAR : __ movzwl(rax, rax); break;
256 case T_BYTE : __ sign_extend_byte(rax); break;
257 case T_SHORT : __ sign_extend_short(rax); break;
258 case T_INT : /* nothing to do */ break;
259 case T_LONG : /* nothing to do */ break;
260 case T_VOID : /* nothing to do */ break;
261 case T_FLOAT : /* nothing to do */ break;
262 case T_DOUBLE : /* nothing to do */ break;
263 case T_OBJECT :
264 // retrieve result from frame
265 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
266 // and verify it
267 __ verify_oop(rax);
268 break;
269 default : ShouldNotReachHere();
270 }
271 __ ret(0); // return from result handler
272 return entry;
273 }
274
275 address TemplateInterpreterGenerator::generate_safept_entry_for(
276 TosState state,
277 address runtime_entry) {
278 address entry = __ pc();
279 __ push(state);
280 __ call_VM(noreg, runtime_entry);
281 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
282 return entry;
283 }
284
285
286
287 // Helpers for commoning out cases in the various type of method entries.
288 //
289
290
291 // increment invocation count & check for overflow
292 //
293 // Note: checking for negative value instead of overflow
294 // so we have a 'sticky' overflow test
295 //
296 // rbx: method
297 // ecx: invocation counter
298 //
299 void InterpreterGenerator::generate_counter_incr(
300 Label* overflow,
301 Label* profile_method,
302 Label* profile_method_continue) {
303 Label done;
304 // Note: In tiered we increment either counters in Method* 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;
309 if (ProfileInterpreter) {
310 // Are we profiling?
311 __ movptr(rax, Address(rbx, Method::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(MethodData::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 __ jmp(done);
319 }
320 __ bind(no_mdo);
321 // Increment counter in MethodCounters
322 const Address invocation_counter(rax,
323 MethodCounters::invocation_counter_offset() +
324 InvocationCounter::counter_offset());
325 __ get_method_counters(rbx, rax, done);
326 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
327 false, Assembler::zero, overflow);
328 __ bind(done);
329 } else {
330 const Address backedge_counter(rax,
331 MethodCounters::backedge_counter_offset() +
332 InvocationCounter::counter_offset());
333 const Address invocation_counter(rax,
334 MethodCounters::invocation_counter_offset() +
335 InvocationCounter::counter_offset());
336
337 __ get_method_counters(rbx, rax, done);
338
339 if (ProfileInterpreter) {
340 __ incrementl(Address(rax,
341 MethodCounters::interpreter_invocation_counter_offset()));
342 }
343 // Update standard invocation counters
344 __ movl(rcx, invocation_counter);
345 __ incrementl(rcx, InvocationCounter::count_increment);
346 __ movl(invocation_counter, rcx); // save invocation count
347
348 __ movl(rax, backedge_counter); // load backedge counter
349 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
350
351 __ addl(rcx, rax); // add both counters
352
353 // profile_method is non-null only for interpreted method so
354 // profile_method != NULL == !native_call
355
356 if (ProfileInterpreter && profile_method != NULL) {
357 // Test to see if we should create a method data oop
358 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
359 __ jcc(Assembler::less, *profile_method_continue);
360
361 // if no method data exists, go to profile_method
362 __ test_method_data_pointer(rax, *profile_method);
363 }
364
365 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
366 __ jcc(Assembler::aboveEqual, *overflow);
367 __ bind(done);
368 }
369 }
370
371 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
372
373 // Asm interpreter on entry
374 // r14 - locals
375 // r13 - bcp
376 // rbx - method
377 // edx - cpool --- DOES NOT APPEAR TO BE TRUE
378 // rbp - interpreter frame
379
380 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
381 // Everything as it was on entry
382 // rdx is not restored. Doesn't appear to really be set.
383
384 // InterpreterRuntime::frequency_counter_overflow takes two
385 // arguments, the first (thread) is passed by call_VM, the second
386 // indicates if the counter overflow occurs at a backwards branch
387 // (NULL bcp). We pass zero for it. The call returns the address
388 // of the verified entry point for the method or NULL if the
389 // compilation did not complete (either went background or bailed
390 // out).
391 __ movl(c_rarg1, 0);
392 __ call_VM(noreg,
393 CAST_FROM_FN_PTR(address,
394 InterpreterRuntime::frequency_counter_overflow),
395 c_rarg1);
396
397 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
398 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
399 // and jump to the interpreted entry.
400 __ jmp(*do_continue, relocInfo::none);
401 }
402
403 // See if we've got enough room on the stack for locals plus overhead.
404 // The expression stack grows down incrementally, so the normal guard
405 // page mechanism will work for that.
406 //
407 // NOTE: Since the additional locals are also always pushed (wasn't
408 // obvious in generate_method_entry) so the guard should work for them
409 // too.
410 //
411 // Args:
412 // rdx: number of additional locals this frame needs (what we must check)
413 // rbx: Method*
414 //
415 // Kills:
416 // rax
417 void InterpreterGenerator::generate_stack_overflow_check(void) {
418
419 // monitor entry size: see picture of stack set
420 // (generate_method_entry) and frame_amd64.hpp
421 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
422
423 // total overhead size: entry_size + (saved rbp through expr stack
424 // bottom). be sure to change this if you add/subtract anything
425 // to/from the overhead area
426 const int overhead_size =
427 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
428
429 const int page_size = os::vm_page_size();
430
431 Label after_frame_check;
432
433 // see if the frame is greater than one page in size. If so,
434 // then we need to verify there is enough stack space remaining
435 // for the additional locals.
436 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
437 __ jcc(Assembler::belowEqual, after_frame_check);
438
439 // compute rsp as if this were going to be the last frame on
440 // the stack before the red zone
441
442 const Address stack_base(r15_thread, Thread::stack_base_offset());
443 const Address stack_size(r15_thread, Thread::stack_size_offset());
444
445 // locals + overhead, in bytes
446 __ mov(rax, rdx);
447 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
448 __ addptr(rax, overhead_size);
449
450 #ifdef ASSERT
451 Label stack_base_okay, stack_size_okay;
452 // verify that thread stack base is non-zero
453 __ cmpptr(stack_base, (int32_t)NULL_WORD);
454 __ jcc(Assembler::notEqual, stack_base_okay);
455 __ stop("stack base is zero");
456 __ bind(stack_base_okay);
457 // verify that thread stack size is non-zero
458 __ cmpptr(stack_size, 0);
459 __ jcc(Assembler::notEqual, stack_size_okay);
460 __ stop("stack size is zero");
461 __ bind(stack_size_okay);
462 #endif
463
464 // Add stack base to locals and subtract stack size
465 __ addptr(rax, stack_base);
466 __ subptr(rax, stack_size);
467
468 // Use the maximum number of pages we might bang.
469 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
470 (StackRedPages+StackYellowPages);
471
472 // add in the red and yellow zone sizes
473 __ addptr(rax, max_pages * page_size);
474
475 // check against the current stack bottom
476 __ cmpptr(rsp, rax);
477 __ jcc(Assembler::above, after_frame_check);
478
479 // Restore sender's sp as SP. This is necessary if the sender's
480 // frame is an extended compiled frame (see gen_c2i_adapter())
481 // and safer anyway in case of JSR292 adaptations.
482
483 __ pop(rax); // return address must be moved if SP is changed
484 __ mov(rsp, r13);
485 __ push(rax);
486
487 // Note: the restored frame is not necessarily interpreted.
488 // Use the shared runtime version of the StackOverflowError.
489 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
490 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
491
492 // all done with frame size check
493 __ bind(after_frame_check);
494 }
495
496 // Allocate monitor and lock method (asm interpreter)
497 //
498 // Args:
499 // rbx: Method*
500 // r14: locals
501 //
502 // Kills:
503 // rax
504 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
505 // rscratch1, rscratch2 (scratch regs)
506 void InterpreterGenerator::lock_method(void) {
507 // synchronize method
508 const Address access_flags(rbx, Method::access_flags_offset());
509 const Address monitor_block_top(
510 rbp,
511 frame::interpreter_frame_monitor_block_top_offset * wordSize);
512 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
513
514 #ifdef ASSERT
515 {
516 Label L;
517 __ movl(rax, access_flags);
518 __ testl(rax, JVM_ACC_SYNCHRONIZED);
519 __ jcc(Assembler::notZero, L);
520 __ stop("method doesn't need synchronization");
521 __ bind(L);
522 }
523 #endif // ASSERT
524
525 // get synchronization object
526 {
527 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
528 Label done;
529 __ movl(rax, access_flags);
530 __ testl(rax, JVM_ACC_STATIC);
531 // get receiver (assume this is frequent case)
532 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
533 __ jcc(Assembler::zero, done);
534 __ movptr(rax, Address(rbx, Method::const_offset()));
535 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
536 __ movptr(rax, Address(rax,
537 ConstantPool::pool_holder_offset_in_bytes()));
538 __ movptr(rax, Address(rax, mirror_offset));
539
540 #ifdef ASSERT
541 {
542 Label L;
543 __ testptr(rax, rax);
544 __ jcc(Assembler::notZero, L);
545 __ stop("synchronization object is NULL");
546 __ bind(L);
547 }
548 #endif // ASSERT
549
550 __ bind(done);
551 }
552
553 // add space for monitor & lock
554 __ subptr(rsp, entry_size); // add space for a monitor entry
555 __ movptr(monitor_block_top, rsp); // set new monitor block top
556 // store object
557 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
558 __ movptr(c_rarg1, rsp); // object address
559 __ lock_object(c_rarg1);
560 }
561
562 // Generate a fixed interpreter frame. This is identical setup for
563 // interpreted methods and for native methods hence the shared code.
564 //
565 // Args:
566 // rax: return address
567 // rbx: Method*
568 // r14: pointer to locals
569 // r13: sender sp
570 // rdx: cp cache
571 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
572 // initialize fixed part of activation frame
573 __ push(rax); // save return address
574 __ enter(); // save old & set new rbp
575 __ push(r13); // set sender sp
576 __ push((int)NULL_WORD); // leave last_sp as null
577 __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod*
578 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
579 __ push(rbx); // save Method*
580 if (ProfileInterpreter) {
581 Label method_data_continue;
582 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
583 __ testptr(rdx, rdx);
584 __ jcc(Assembler::zero, method_data_continue);
585 __ addptr(rdx, in_bytes(MethodData::data_offset()));
586 __ bind(method_data_continue);
587 __ push(rdx); // set the mdp (method data pointer)
588 } else {
589 __ push(0);
590 }
591
592 __ movptr(rdx, Address(rbx, Method::const_offset()));
593 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
594 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
595 __ push(rdx); // set constant pool cache
596 __ push(r14); // set locals pointer
597 if (native_call) {
598 __ push(0); // no bcp
599 } else {
600 __ push(r13); // set bcp
601 }
602 __ push(0); // reserve word for pointer to expression stack bottom
603 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
604 }
605
606 // End of helpers
607
608 // Various method entries
609 //------------------------------------------------------------------------------------------------------------------------
610 //
611 //
612
613 // Call an accessor method (assuming it is resolved, otherwise drop
614 // into vanilla (slow path) entry
615 address InterpreterGenerator::generate_accessor_entry(void) {
616 // rbx: Method*
617
618 // r13: senderSP must preserver for slow path, set SP to it on fast path
619
620 address entry_point = __ pc();
621 Label xreturn_path;
622
623 // do fastpath for resolved accessor methods
624 if (UseFastAccessorMethods) {
625 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
626 // thereof; parameter size = 1
627 // Note: We can only use this code if the getfield has been resolved
628 // and if we don't have a null-pointer exception => check for
629 // these conditions first and use slow path if necessary.
630 Label slow_path;
631 // If we need a safepoint check, generate full interpreter entry.
632 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
633 SafepointSynchronize::_not_synchronized);
634
635 __ jcc(Assembler::notEqual, slow_path);
636 // rbx: method
637 __ movptr(rax, Address(rsp, wordSize));
638
639 // check if local 0 != NULL and read field
640 __ testptr(rax, rax);
641 __ jcc(Assembler::zero, slow_path);
642
643 // read first instruction word and extract bytecode @ 1 and index @ 2
644 __ movptr(rdx, Address(rbx, Method::const_offset()));
645 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
646 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
647 // Shift codes right to get the index on the right.
648 // The bytecode fetched looks like <index><0xb4><0x2a>
649 __ shrl(rdx, 2 * BitsPerByte);
650 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
651 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
652
653 // rax: local 0
654 // rbx: method
655 // rdx: constant pool cache index
656 // rdi: constant pool cache
657
658 // check if getfield has been resolved and read constant pool cache entry
659 // check the validity of the cache entry by testing whether _indices field
660 // contains Bytecode::_getfield in b1 byte.
661 assert(in_words(ConstantPoolCacheEntry::size()) == 4,
662 "adjust shift below");
663 __ movl(rcx,
664 Address(rdi,
665 rdx,
666 Address::times_8,
667 ConstantPoolCache::base_offset() +
668 ConstantPoolCacheEntry::indices_offset()));
669 __ shrl(rcx, 2 * BitsPerByte);
670 __ andl(rcx, 0xFF);
671 __ cmpl(rcx, Bytecodes::_getfield);
672 __ jcc(Assembler::notEqual, slow_path);
673
674 // Note: constant pool entry is not valid before bytecode is resolved
675 __ movptr(rcx,
676 Address(rdi,
677 rdx,
678 Address::times_8,
679 ConstantPoolCache::base_offset() +
680 ConstantPoolCacheEntry::f2_offset()));
681 // edx: flags
682 __ movl(rdx,
683 Address(rdi,
684 rdx,
685 Address::times_8,
686 ConstantPoolCache::base_offset() +
687 ConstantPoolCacheEntry::flags_offset()));
688
689 Label notObj, notInt, notByte, notShort;
690 const Address field_address(rax, rcx, Address::times_1);
691
692 // Need to differentiate between igetfield, agetfield, bgetfield etc.
693 // because they are different sizes.
694 // Use the type from the constant pool cache
695 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
696 // Make sure we don't need to mask edx after the above shift
697 ConstantPoolCacheEntry::verify_tos_state_shift();
698
699 __ cmpl(rdx, atos);
700 __ jcc(Assembler::notEqual, notObj);
701 // atos
702 __ load_heap_oop(rax, field_address);
703 __ jmp(xreturn_path);
704
705 __ bind(notObj);
706 __ cmpl(rdx, itos);
707 __ jcc(Assembler::notEqual, notInt);
708 // itos
709 __ movl(rax, field_address);
710 __ jmp(xreturn_path);
711
712 __ bind(notInt);
713 __ cmpl(rdx, btos);
714 __ jcc(Assembler::notEqual, notByte);
715 // btos
716 __ load_signed_byte(rax, field_address);
717 __ jmp(xreturn_path);
718
719 __ bind(notByte);
720 __ cmpl(rdx, stos);
721 __ jcc(Assembler::notEqual, notShort);
722 // stos
723 __ load_signed_short(rax, field_address);
724 __ jmp(xreturn_path);
725
726 __ bind(notShort);
727 #ifdef ASSERT
728 Label okay;
729 __ cmpl(rdx, ctos);
730 __ jcc(Assembler::equal, okay);
731 __ stop("what type is this?");
732 __ bind(okay);
733 #endif
734 // ctos
735 __ load_unsigned_short(rax, field_address);
736
737 __ bind(xreturn_path);
738
739 // _ireturn/_areturn
740 __ pop(rdi);
741 __ mov(rsp, r13);
742 __ jmp(rdi);
743 __ ret(0);
744
745 // generate a vanilla interpreter entry as the slow path
746 __ bind(slow_path);
747 (void) generate_normal_entry(false);
748 } else {
749 (void) generate_normal_entry(false);
750 }
751
752 return entry_point;
753 }
754
755 // Method entry for java.lang.ref.Reference.get.
756 address InterpreterGenerator::generate_Reference_get_entry(void) {
757 #if INCLUDE_ALL_GCS
758 // Code: _aload_0, _getfield, _areturn
759 // parameter size = 1
760 //
761 // The code that gets generated by this routine is split into 2 parts:
762 // 1. The "intrinsified" code for G1 (or any SATB based GC),
763 // 2. The slow path - which is an expansion of the regular method entry.
764 //
765 // Notes:-
766 // * In the G1 code we do not check whether we need to block for
767 // a safepoint. If G1 is enabled then we must execute the specialized
768 // code for Reference.get (except when the Reference object is null)
769 // so that we can log the value in the referent field with an SATB
770 // update buffer.
771 // If the code for the getfield template is modified so that the
772 // G1 pre-barrier code is executed when the current method is
773 // Reference.get() then going through the normal method entry
774 // will be fine.
775 // * The G1 code can, however, check the receiver object (the instance
776 // of java.lang.Reference) and jump to the slow path if null. If the
777 // Reference object is null then we obviously cannot fetch the referent
778 // and so we don't need to call the G1 pre-barrier. Thus we can use the
779 // regular method entry code to generate the NPE.
780 //
781 // This code is based on generate_accessor_enty.
782 //
783 // rbx: Method*
784
785 // r13: senderSP must preserve for slow path, set SP to it on fast path
786
787 address entry = __ pc();
788
789 const int referent_offset = java_lang_ref_Reference::referent_offset;
790 guarantee(referent_offset > 0, "referent offset not initialized");
791
792 if (UseG1GC) {
793 Label slow_path;
794 // rbx: method
795
796 // Check if local 0 != NULL
797 // If the receiver is null then it is OK to jump to the slow path.
798 __ movptr(rax, Address(rsp, wordSize));
799
800 __ testptr(rax, rax);
801 __ jcc(Assembler::zero, slow_path);
802
803 // rax: local 0
804 // rbx: method (but can be used as scratch now)
805 // rdx: scratch
806 // rdi: scratch
807
808 // Generate the G1 pre-barrier code to log the value of
809 // the referent field in an SATB buffer.
810
811 // Load the value of the referent field.
812 const Address field_address(rax, referent_offset);
813 __ load_heap_oop(rax, field_address);
814
815 // Generate the G1 pre-barrier code to log the value of
816 // the referent field in an SATB buffer.
817 __ g1_write_barrier_pre(noreg /* obj */,
818 rax /* pre_val */,
819 r15_thread /* thread */,
820 rbx /* tmp */,
821 true /* tosca_live */,
822 true /* expand_call */);
823
824 // _areturn
825 __ pop(rdi); // get return address
826 __ mov(rsp, r13); // set sp to sender sp
827 __ jmp(rdi);
828 __ ret(0);
829
830 // generate a vanilla interpreter entry as the slow path
831 __ bind(slow_path);
832 (void) generate_normal_entry(false);
833
834 return entry;
835 }
836 #endif // INCLUDE_ALL_GCS
837
838 // If G1 is not enabled then attempt to go through the accessor entry point
839 // Reference.get is an accessor
840 return generate_accessor_entry();
841 }
842
843 /**
844 * Method entry for static native methods:
845 * int java.util.zip.CRC32.update(int crc, int b)
846 */
847 address InterpreterGenerator::generate_CRC32_update_entry() {
848 if (UseCRC32Intrinsics) {
849 address entry = __ pc();
850
851 // rbx,: Method*
852 // r13: senderSP must preserved for slow path, set SP to it on fast path
853 // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
854 // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
855
856 Label slow_path;
857 // If we need a safepoint check, generate full interpreter entry.
858 ExternalAddress state(SafepointSynchronize::address_of_state());
859 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
860 SafepointSynchronize::_not_synchronized);
861 __ jcc(Assembler::notEqual, slow_path);
862
863 // We don't generate local frame and don't align stack because
864 // we call stub code and there is no safepoint on this path.
865
866 // Load parameters
867 const Register crc = rax; // crc
868 const Register val = c_rarg0; // source java byte value
869 const Register tbl = c_rarg1; // scratch
870
871 // Arguments are reversed on java expression stack
872 __ movl(val, Address(rsp, wordSize)); // byte value
873 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
874
875 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
876 __ notl(crc); // ~crc
877 __ update_byte_crc32(crc, val, tbl);
878 __ notl(crc); // ~crc
879 // result in rax
880
881 // _areturn
882 __ pop(rdi); // get return address
883 __ mov(rsp, r13); // set sp to sender sp
884 __ jmp(rdi);
885
886 // generate a vanilla native entry as the slow path
887 __ bind(slow_path);
888
889 (void) generate_native_entry(false);
890
891 return entry;
892 }
893 return generate_native_entry(false);
894 }
895
896 /**
897 * Method entry for static native methods:
898 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
899 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
900 */
901 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
902 if (UseCRC32Intrinsics) {
903 address entry = __ pc();
904
905 // rbx,: Method*
906 // r13: senderSP must preserved for slow path, set SP to it on fast path
907
908 Label slow_path;
909 // If we need a safepoint check, generate full interpreter entry.
910 ExternalAddress state(SafepointSynchronize::address_of_state());
911 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
912 SafepointSynchronize::_not_synchronized);
913 __ jcc(Assembler::notEqual, slow_path);
914
915 // We don't generate local frame and don't align stack because
916 // we call stub code and there is no safepoint on this path.
917
918 // Load parameters
919 const Register crc = c_rarg0; // crc
920 const Register buf = c_rarg1; // source java byte array address
921 const Register len = c_rarg2; // length
922 const Register off = len; // offset (never overlaps with 'len')
923
924 // Arguments are reversed on java expression stack
925 // Calculate address of start element
926 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
927 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
928 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
929 __ addq(buf, off); // + offset
930 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
931 } else {
932 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
933 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
934 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
935 __ addq(buf, off); // + offset
936 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
937 }
938 // Can now load 'len' since we're finished with 'off'
939 __ movl(len, Address(rsp, wordSize)); // Length
940
941 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
942 // result in rax
943
944 // _areturn
945 __ pop(rdi); // get return address
946 __ mov(rsp, r13); // set sp to sender sp
947 __ jmp(rdi);
948
949 // generate a vanilla native entry as the slow path
950 __ bind(slow_path);
951
952 (void) generate_native_entry(false);
953
954 return entry;
955 }
956 return generate_native_entry(false);
957 }
958
959 // Interpreter stub for calling a native method. (asm interpreter)
960 // This sets up a somewhat different looking stack for calling the
961 // native method than the typical interpreter frame setup.
962 address InterpreterGenerator::generate_native_entry(bool synchronized) {
963 // determine code generation flags
964 bool inc_counter = UseCompiler || CountCompiledCalls;
965
966 // rbx: Method*
967 // r13: sender sp
968
969 address entry_point = __ pc();
970
971 const Address constMethod (rbx, Method::const_offset());
972 const Address access_flags (rbx, Method::access_flags_offset());
973 const Address size_of_parameters(rcx, ConstMethod::
974 size_of_parameters_offset());
975
976
977 // get parameter size (always needed)
978 __ movptr(rcx, constMethod);
979 __ load_unsigned_short(rcx, size_of_parameters);
980
981 // native calls don't need the stack size check since they have no
982 // expression stack and the arguments are already on the stack and
983 // we only add a handful of words to the stack
984
985 // rbx: Method*
986 // rcx: size of parameters
987 // r13: sender sp
988 __ pop(rax); // get return address
989
990 // for natives the size of locals is zero
991
992 // compute beginning of parameters (r14)
993 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
994
995 // add 2 zero-initialized slots for native calls
996 // initialize result_handler slot
997 __ push((int) NULL_WORD);
998 // slot for oop temp
999 // (static native method holder mirror/jni oop result)
1000 __ push((int) NULL_WORD);
1001
1002 // initialize fixed part of activation frame
1003 generate_fixed_frame(true);
1004
1005 // make sure method is native & not abstract
1006 #ifdef ASSERT
1007 __ movl(rax, access_flags);
1008 {
1009 Label L;
1010 __ testl(rax, JVM_ACC_NATIVE);
1011 __ jcc(Assembler::notZero, L);
1012 __ stop("tried to execute non-native method as native");
1013 __ bind(L);
1014 }
1015 {
1016 Label L;
1017 __ testl(rax, JVM_ACC_ABSTRACT);
1018 __ jcc(Assembler::zero, L);
1019 __ stop("tried to execute abstract method in interpreter");
1020 __ bind(L);
1021 }
1022 #endif
1023
1024 // Since at this point in the method invocation the exception handler
1025 // would try to exit the monitor of synchronized methods which hasn't
1026 // been entered yet, we set the thread local variable
1027 // _do_not_unlock_if_synchronized to true. The remove_activation will
1028 // check this flag.
1029
1030 const Address do_not_unlock_if_synchronized(r15_thread,
1031 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1032 __ movbool(do_not_unlock_if_synchronized, true);
1033
1034 // increment invocation count & check for overflow
1035 Label invocation_counter_overflow;
1036 if (inc_counter) {
1037 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1038 }
1039
1040 Label continue_after_compile;
1041 __ bind(continue_after_compile);
1042
1043 bang_stack_shadow_pages(true);
1044
1045 // reset the _do_not_unlock_if_synchronized flag
1046 __ movbool(do_not_unlock_if_synchronized, false);
1047
1048 // check for synchronized methods
1049 // Must happen AFTER invocation_counter check and stack overflow check,
1050 // so method is not locked if overflows.
1051 if (synchronized) {
1052 lock_method();
1053 } else {
1054 // no synchronization necessary
1055 #ifdef ASSERT
1056 {
1057 Label L;
1058 __ movl(rax, access_flags);
1059 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1060 __ jcc(Assembler::zero, L);
1061 __ stop("method needs synchronization");
1062 __ bind(L);
1063 }
1064 #endif
1065 }
1066
1067 // start execution
1068 #ifdef ASSERT
1069 {
1070 Label L;
1071 const Address monitor_block_top(rbp,
1072 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1073 __ movptr(rax, monitor_block_top);
1074 __ cmpptr(rax, rsp);
1075 __ jcc(Assembler::equal, L);
1076 __ stop("broken stack frame setup in interpreter");
1077 __ bind(L);
1078 }
1079 #endif
1080
1081 // jvmti support
1082 __ notify_method_entry();
1083
1084 // work registers
1085 const Register method = rbx;
1086 const Register t = r11;
1087
1088 // allocate space for parameters
1089 __ get_method(method);
1090 __ movptr(t, Address(method, Method::const_offset()));
1091 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1092 __ shll(t, Interpreter::logStackElementSize);
1093
1094 __ subptr(rsp, t);
1095 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1096 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
1097
1098 // get signature handler
1099 {
1100 Label L;
1101 __ movptr(t, Address(method, Method::signature_handler_offset()));
1102 __ testptr(t, t);
1103 __ jcc(Assembler::notZero, L);
1104 __ call_VM(noreg,
1105 CAST_FROM_FN_PTR(address,
1106 InterpreterRuntime::prepare_native_call),
1107 method);
1108 __ get_method(method);
1109 __ movptr(t, Address(method, Method::signature_handler_offset()));
1110 __ bind(L);
1111 }
1112
1113 // call signature handler
1114 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
1115 "adjust this code");
1116 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1117 "adjust this code");
1118 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1119 "adjust this code");
1120
1121 // The generated handlers do not touch RBX (the method oop).
1122 // However, large signatures cannot be cached and are generated
1123 // each time here. The slow-path generator can do a GC on return,
1124 // so we must reload it after the call.
1125 __ call(t);
1126 __ get_method(method); // slow path can do a GC, reload RBX
1127
1128
1129 // result handler is in rax
1130 // set result handler
1131 __ movptr(Address(rbp,
1132 (frame::interpreter_frame_result_handler_offset) * wordSize),
1133 rax);
1134
1135 // pass mirror handle if static call
1136 {
1137 Label L;
1138 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1139 __ movl(t, Address(method, Method::access_flags_offset()));
1140 __ testl(t, JVM_ACC_STATIC);
1141 __ jcc(Assembler::zero, L);
1142 // get mirror
1143 __ movptr(t, Address(method, Method::const_offset()));
1144 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1145 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1146 __ movptr(t, Address(t, mirror_offset));
1147 // copy mirror into activation frame
1148 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1149 t);
1150 // pass handle to mirror
1151 __ lea(c_rarg1,
1152 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1153 __ bind(L);
1154 }
1155
1156 // get native function entry point
1157 {
1158 Label L;
1159 __ movptr(rax, Address(method, Method::native_function_offset()));
1160 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1161 __ movptr(rscratch2, unsatisfied.addr());
1162 __ cmpptr(rax, rscratch2);
1163 __ jcc(Assembler::notEqual, L);
1164 __ call_VM(noreg,
1165 CAST_FROM_FN_PTR(address,
1166 InterpreterRuntime::prepare_native_call),
1167 method);
1168 __ get_method(method);
1169 __ movptr(rax, Address(method, Method::native_function_offset()));
1170 __ bind(L);
1171 }
1172
1173 // pass JNIEnv
1174 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1175
1176 // It is enough that the pc() points into the right code
1177 // segment. It does not have to be the correct return pc.
1178 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1179
1180 // change thread state
1181 #ifdef ASSERT
1182 {
1183 Label L;
1184 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1185 __ cmpl(t, _thread_in_Java);
1186 __ jcc(Assembler::equal, L);
1187 __ stop("Wrong thread state in native stub");
1188 __ bind(L);
1189 }
1190 #endif
1191
1192 // Change state to native
1193
1194 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1195 _thread_in_native);
1196
1197 // Call the native method.
1198 __ call(rax);
1199 // result potentially in rax or xmm0
1200
1201 // Verify or restore cpu control state after JNI call
1202 __ restore_cpu_control_state_after_jni();
1203
1204 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1205 // in order to extract the result of a method call. If the order of these
1206 // pushes change or anything else is added to the stack then the code in
1207 // interpreter_frame_result must also change.
1208
1209 __ push(dtos);
1210 __ push(ltos);
1211
1212 // change thread state
1213 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1214 _thread_in_native_trans);
1215
1216 if (os::is_MP()) {
1217 if (UseMembar) {
1218 // Force this write out before the read below
1219 __ membar(Assembler::Membar_mask_bits(
1220 Assembler::LoadLoad | Assembler::LoadStore |
1221 Assembler::StoreLoad | Assembler::StoreStore));
1222 } else {
1223 // Write serialization page so VM thread can do a pseudo remote membar.
1224 // We use the current thread pointer to calculate a thread specific
1225 // offset to write to within the page. This minimizes bus traffic
1226 // due to cache line collision.
1227 __ serialize_memory(r15_thread, rscratch2);
1228 }
1229 }
1230
1231 // check for safepoint operation in progress and/or pending suspend requests
1232 {
1233 Label Continue;
1234 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1235 SafepointSynchronize::_not_synchronized);
1236
1237 Label L;
1238 __ jcc(Assembler::notEqual, L);
1239 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1240 __ jcc(Assembler::equal, Continue);
1241 __ bind(L);
1242
1243 // Don't use call_VM as it will see a possible pending exception
1244 // and forward it and never return here preventing us from
1245 // clearing _last_native_pc down below. Also can't use
1246 // call_VM_leaf either as it will check to see if r13 & r14 are
1247 // preserved and correspond to the bcp/locals pointers. So we do a
1248 // runtime call by hand.
1249 //
1250 __ mov(c_rarg0, r15_thread);
1251 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1252 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1253 __ andptr(rsp, -16); // align stack as required by ABI
1254 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1255 __ mov(rsp, r12); // restore sp
1256 __ reinit_heapbase();
1257 __ bind(Continue);
1258 }
1259
1260 // change thread state
1261 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1262
1263 // reset_last_Java_frame
1264 __ reset_last_Java_frame(true, true);
1265
1266 // reset handle block
1267 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1268 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1269
1270 // If result is an oop unbox and store it in frame where gc will see it
1271 // and result handler will pick it up
1272
1273 {
1274 Label no_oop, store_result;
1275 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1276 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1277 __ jcc(Assembler::notEqual, no_oop);
1278 // retrieve result
1279 __ pop(ltos);
1280 __ testptr(rax, rax);
1281 __ jcc(Assembler::zero, store_result);
1282 __ movptr(rax, Address(rax, 0));
1283 __ bind(store_result);
1284 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1285 // keep stack depth as expected by pushing oop which will eventually be discarde
1286 __ push(ltos);
1287 __ bind(no_oop);
1288 }
1289
1290
1291 {
1292 Label no_reguard;
1293 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1294 JavaThread::stack_guard_yellow_disabled);
1295 __ jcc(Assembler::notEqual, no_reguard);
1296
1297 __ pusha(); // XXX only save smashed registers
1298 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1299 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1300 __ andptr(rsp, -16); // align stack as required by ABI
1301 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1302 __ mov(rsp, r12); // restore sp
1303 __ popa(); // XXX only restore smashed registers
1304 __ reinit_heapbase();
1305
1306 __ bind(no_reguard);
1307 }
1308
1309
1310 // The method register is junk from after the thread_in_native transition
1311 // until here. Also can't call_VM until the bcp has been
1312 // restored. Need bcp for throwing exception below so get it now.
1313 __ get_method(method);
1314
1315 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1316 // r13 == code_base()
1317 __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod*
1318 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
1319 // handle exceptions (exception handling will handle unlocking!)
1320 {
1321 Label L;
1322 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1323 __ jcc(Assembler::zero, L);
1324 // Note: At some point we may want to unify this with the code
1325 // used in call_VM_base(); i.e., we should use the
1326 // StubRoutines::forward_exception code. For now this doesn't work
1327 // here because the rsp is not correctly set at this point.
1328 __ MacroAssembler::call_VM(noreg,
1329 CAST_FROM_FN_PTR(address,
1330 InterpreterRuntime::throw_pending_exception));
1331 __ should_not_reach_here();
1332 __ bind(L);
1333 }
1334
1335 // do unlocking if necessary
1336 {
1337 Label L;
1338 __ movl(t, Address(method, Method::access_flags_offset()));
1339 __ testl(t, JVM_ACC_SYNCHRONIZED);
1340 __ jcc(Assembler::zero, L);
1341 // the code below should be shared with interpreter macro
1342 // assembler implementation
1343 {
1344 Label unlock;
1345 // BasicObjectLock will be first in list, since this is a
1346 // synchronized method. However, need to check that the object
1347 // has not been unlocked by an explicit monitorexit bytecode.
1348 const Address monitor(rbp,
1349 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1350 wordSize - sizeof(BasicObjectLock)));
1351
1352 // monitor expect in c_rarg1 for slow unlock path
1353 __ lea(c_rarg1, monitor); // address of first monitor
1354
1355 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1356 __ testptr(t, t);
1357 __ jcc(Assembler::notZero, unlock);
1358
1359 // Entry already unlocked, need to throw exception
1360 __ MacroAssembler::call_VM(noreg,
1361 CAST_FROM_FN_PTR(address,
1362 InterpreterRuntime::throw_illegal_monitor_state_exception));
1363 __ should_not_reach_here();
1364
1365 __ bind(unlock);
1366 __ unlock_object(c_rarg1);
1367 }
1368 __ bind(L);
1369 }
1370
1371 // jvmti support
1372 // Note: This must happen _after_ handling/throwing any exceptions since
1373 // the exception handler code notifies the runtime of method exits
1374 // too. If this happens before, method entry/exit notifications are
1375 // not properly paired (was bug - gri 11/22/99).
1376 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1377
1378 // restore potential result in edx:eax, call result handler to
1379 // restore potential result in ST0 & handle result
1380
1381 __ pop(ltos);
1382 __ pop(dtos);
1383
1384 __ movptr(t, Address(rbp,
1385 (frame::interpreter_frame_result_handler_offset) * wordSize));
1386 __ call(t);
1387
1388 // remove activation
1389 __ movptr(t, Address(rbp,
1390 frame::interpreter_frame_sender_sp_offset *
1391 wordSize)); // get sender sp
1392 __ leave(); // remove frame anchor
1393 __ pop(rdi); // get return address
1394 __ mov(rsp, t); // set sp to sender sp
1395 __ jmp(rdi);
1396
1397 if (inc_counter) {
1398 // Handle overflow of counter and compile method
1399 __ bind(invocation_counter_overflow);
1400 generate_counter_overflow(&continue_after_compile);
1401 }
1402
1403 return entry_point;
1404 }
1405
1406 //
1407 // Generic interpreted method entry to (asm) interpreter
1408 //
1409 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1410 // determine code generation flags
1411 bool inc_counter = UseCompiler || CountCompiledCalls;
1412
1413 // ebx: Method*
1414 // r13: sender sp
1415 address entry_point = __ pc();
1416
1417 const Address constMethod(rbx, Method::const_offset());
1418 const Address access_flags(rbx, Method::access_flags_offset());
1419 const Address size_of_parameters(rdx,
1420 ConstMethod::size_of_parameters_offset());
1421 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1422
1423
1424 // get parameter size (always needed)
1425 __ movptr(rdx, constMethod);
1426 __ load_unsigned_short(rcx, size_of_parameters);
1427
1428 // rbx: Method*
1429 // rcx: size of parameters
1430 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1431
1432 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1433 __ subl(rdx, rcx); // rdx = no. of additional locals
1434
1435 // YYY
1436 // __ incrementl(rdx);
1437 // __ andl(rdx, -2);
1438
1439 // see if we've got enough room on the stack for locals plus overhead.
1440 generate_stack_overflow_check();
1441
1442 // get return address
1443 __ pop(rax);
1444
1445 // compute beginning of parameters (r14)
1446 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1447
1448 // rdx - # of additional locals
1449 // allocate space for locals
1450 // explicitly initialize locals
1451 {
1452 Label exit, loop;
1453 __ testl(rdx, rdx);
1454 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1455 __ bind(loop);
1456 __ push((int) NULL_WORD); // initialize local variables
1457 __ decrementl(rdx); // until everything initialized
1458 __ jcc(Assembler::greater, loop);
1459 __ bind(exit);
1460 }
1461
1462 // initialize fixed part of activation frame
1463 generate_fixed_frame(false);
1464
1465 // make sure method is not native & not abstract
1466 #ifdef ASSERT
1467 __ movl(rax, access_flags);
1468 {
1469 Label L;
1470 __ testl(rax, JVM_ACC_NATIVE);
1471 __ jcc(Assembler::zero, L);
1472 __ stop("tried to execute native method as non-native");
1473 __ bind(L);
1474 }
1475 {
1476 Label L;
1477 __ testl(rax, JVM_ACC_ABSTRACT);
1478 __ jcc(Assembler::zero, L);
1479 __ stop("tried to execute abstract method in interpreter");
1480 __ bind(L);
1481 }
1482 #endif
1483
1484 // Since at this point in the method invocation the exception
1485 // handler would try to exit the monitor of synchronized methods
1486 // which hasn't been entered yet, we set the thread local variable
1487 // _do_not_unlock_if_synchronized to true. The remove_activation
1488 // will check this flag.
1489
1490 const Address do_not_unlock_if_synchronized(r15_thread,
1491 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1492 __ movbool(do_not_unlock_if_synchronized, true);
1493
1494 // increment invocation count & check for overflow
1495 Label invocation_counter_overflow;
1496 Label profile_method;
1497 Label profile_method_continue;
1498 if (inc_counter) {
1499 generate_counter_incr(&invocation_counter_overflow,
1500 &profile_method,
1501 &profile_method_continue);
1502 if (ProfileInterpreter) {
1503 __ bind(profile_method_continue);
1504 }
1505 }
1506
1507 Label continue_after_compile;
1508 __ bind(continue_after_compile);
1509
1510 // check for synchronized interpreted methods
1511 bang_stack_shadow_pages(false);
1512
1513 // reset the _do_not_unlock_if_synchronized flag
1514 __ movbool(do_not_unlock_if_synchronized, false);
1515
1516 // check for synchronized methods
1517 // Must happen AFTER invocation_counter check and stack overflow check,
1518 // so method is not locked if overflows.
1519 if (synchronized) {
1520 // Allocate monitor and lock method
1521 lock_method();
1522 } else {
1523 // no synchronization necessary
1524 #ifdef ASSERT
1525 {
1526 Label L;
1527 __ movl(rax, access_flags);
1528 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1529 __ jcc(Assembler::zero, L);
1530 __ stop("method needs synchronization");
1531 __ bind(L);
1532 }
1533 #endif
1534 }
1535
1536 // start execution
1537 #ifdef ASSERT
1538 {
1539 Label L;
1540 const Address monitor_block_top (rbp,
1541 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1542 __ movptr(rax, monitor_block_top);
1543 __ cmpptr(rax, rsp);
1544 __ jcc(Assembler::equal, L);
1545 __ stop("broken stack frame setup in interpreter");
1546 __ bind(L);
1547 }
1548 #endif
1549
1550 // jvmti support
1551 __ notify_method_entry();
1552
1553 __ dispatch_next(vtos);
1554
1555 // invocation counter overflow
1556 if (inc_counter) {
1557 if (ProfileInterpreter) {
1558 // We have decided to profile this method in the interpreter
1559 __ bind(profile_method);
1560 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1561 __ set_method_data_pointer_for_bcp();
1562 __ get_method(rbx);
1563 __ jmp(profile_method_continue);
1564 }
1565 // Handle overflow of counter and compile method
1566 __ bind(invocation_counter_overflow);
1567 generate_counter_overflow(&continue_after_compile);
1568 }
1569
1570 return entry_point;
1571 }
1572
1573 // Entry points
1574 //
1575 // Here we generate the various kind of entries into the interpreter.
1576 // The two main entry type are generic bytecode methods and native
1577 // call method. These both come in synchronized and non-synchronized
1578 // versions but the frame layout they create is very similar. The
1579 // other method entry types are really just special purpose entries
1580 // that are really entry and interpretation all in one. These are for
1581 // trivial methods like accessor, empty, or special math methods.
1582 //
1583 // When control flow reaches any of the entry types for the interpreter
1584 // the following holds ->
1585 //
1586 // Arguments:
1587 //
1588 // rbx: Method*
1589 //
1590 // Stack layout immediately at entry
1591 //
1592 // [ return address ] <--- rsp
1593 // [ parameter n ]
1594 // ...
1595 // [ parameter 1 ]
1596 // [ expression stack ] (caller's java expression stack)
1597
1598 // Assuming that we don't go to one of the trivial specialized entries
1599 // the stack will look like below when we are ready to execute the
1600 // first bytecode (or call the native routine). The register usage
1601 // will be as the template based interpreter expects (see
1602 // interpreter_amd64.hpp).
1603 //
1604 // local variables follow incoming parameters immediately; i.e.
1605 // the return address is moved to the end of the locals).
1606 //
1607 // [ monitor entry ] <--- rsp
1608 // ...
1609 // [ monitor entry ]
1610 // [ expr. stack bottom ]
1611 // [ saved r13 ]
1612 // [ current r14 ]
1613 // [ Method* ]
1614 // [ saved ebp ] <--- rbp
1615 // [ return address ]
1616 // [ local variable m ]
1617 // ...
1618 // [ local variable 1 ]
1619 // [ parameter n ]
1620 // ...
1621 // [ parameter 1 ] <--- r14
1622
1623 address AbstractInterpreterGenerator::generate_method_entry(
1624 AbstractInterpreter::MethodKind kind) {
1625 // determine code generation flags
1626 bool synchronized = false;
1627 address entry_point = NULL;
1628 InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1629
1630 switch (kind) {
1631 case Interpreter::zerolocals : break;
1632 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1633 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
1634 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
1635 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
1636 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
1637 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
1638
1639 case Interpreter::java_lang_math_sin : // fall thru
1640 case Interpreter::java_lang_math_cos : // fall thru
1641 case Interpreter::java_lang_math_tan : // fall thru
1642 case Interpreter::java_lang_math_abs : // fall thru
1643 case Interpreter::java_lang_math_log : // fall thru
1644 case Interpreter::java_lang_math_log10 : // fall thru
1645 case Interpreter::java_lang_math_sqrt : // fall thru
1646 case Interpreter::java_lang_math_pow : // fall thru
1647 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
1648 case Interpreter::java_lang_ref_reference_get
1649 : entry_point = ig_this->generate_Reference_get_entry(); break;
1650 case Interpreter::java_util_zip_CRC32_update
1651 : entry_point = ig_this->generate_CRC32_update_entry(); break;
1652 case Interpreter::java_util_zip_CRC32_updateBytes
1653 : // fall thru
1654 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1655 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1656 default:
1657 fatal(err_msg("unexpected method kind: %d", kind));
1658 break;
1659 }
1660
1661 if (entry_point) {
1662 return entry_point;
1663 }
1664
1665 return ig_this->generate_normal_entry(synchronized);
1666 }
1667
1668 // These should never be compiled since the interpreter will prefer
1669 // the compiled version to the intrinsic version.
1670 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1671 switch (method_kind(m)) {
1672 case Interpreter::java_lang_math_sin : // fall thru
1673 case Interpreter::java_lang_math_cos : // fall thru
1674 case Interpreter::java_lang_math_tan : // fall thru
1675 case Interpreter::java_lang_math_abs : // fall thru
1676 case Interpreter::java_lang_math_log : // fall thru
1677 case Interpreter::java_lang_math_log10 : // fall thru
1678 case Interpreter::java_lang_math_sqrt : // fall thru
1679 case Interpreter::java_lang_math_pow : // fall thru
1680 case Interpreter::java_lang_math_exp :
1681 return false;
1682 default:
1683 return true;
1684 }
1685 }
1686
1687 // How much stack a method activation needs in words.
1688 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1689 const int entry_size = frame::interpreter_frame_monitor_size();
1690
1691 // total overhead size: entry_size + (saved rbp thru expr stack
1692 // bottom). be sure to change this if you add/subtract anything
1693 // to/from the overhead area
1694 const int overhead_size =
1695 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1696
1697 const int stub_code = frame::entry_frame_after_call_words;
1698 const int method_stack = (method->max_locals() + method->max_stack()) *
1699 Interpreter::stackElementWords;
1700 return (overhead_size + method_stack + stub_code);
1701 }
1702
1703 int AbstractInterpreter::layout_activation(Method* method,
1704 int tempcount,
1705 int popframe_extra_args,
1706 int moncount,
1707 int caller_actual_parameters,
1708 int callee_param_count,
1709 int callee_locals,
1710 frame* caller,
1711 frame* interpreter_frame,
1712 bool is_top_frame,
1713 bool is_bottom_frame) {
1714 // Note: This calculation must exactly parallel the frame setup
1715 // in AbstractInterpreterGenerator::generate_method_entry.
1716 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1717 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1718 // right size, as determined by a previous call to this method.
1719 // It is also guaranteed to be walkable even though it is in a skeletal state
1720
1721 // fixed size of an interpreter frame:
1722 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1723 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1724 Interpreter::stackElementWords;
1725
1726 int overhead = frame::sender_sp_offset -
1727 frame::interpreter_frame_initial_sp_offset;
1728 // Our locals were accounted for by the caller (or last_frame_adjust
1729 // on the transistion) Since the callee parameters already account
1730 // for the callee's params we only need to account for the extra
1731 // locals.
1732 int size = overhead +
1733 (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1734 moncount * frame::interpreter_frame_monitor_size() +
1735 tempcount* Interpreter::stackElementWords + popframe_extra_args;
1736 if (interpreter_frame != NULL) {
1737 #ifdef ASSERT
1738 if (!EnableInvokeDynamic)
1739 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1740 // Probably, since deoptimization doesn't work yet.
1741 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1742 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1743 #endif
1744
1745 interpreter_frame->interpreter_frame_set_method(method);
1746 // NOTE the difference in using sender_sp and
1747 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1748 // the original sp of the caller (the unextended_sp) and
1749 // sender_sp is fp+16 XXX
1750 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1751
1752 #ifdef ASSERT
1753 if (caller->is_interpreted_frame()) {
1754 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1755 }
1756 #endif
1757
1758 interpreter_frame->interpreter_frame_set_locals(locals);
1759 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1760 BasicObjectLock* monbot = montop - moncount;
1761 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1762
1763 // Set last_sp
1764 intptr_t* esp = (intptr_t*) monbot -
1765 tempcount*Interpreter::stackElementWords -
1766 popframe_extra_args;
1767 interpreter_frame->interpreter_frame_set_last_sp(esp);
1768
1769 // All frames but the initial (oldest) interpreter frame we fill in have
1770 // a value for sender_sp that allows walking the stack but isn't
1771 // truly correct. Correct the value here.
1772 if (extra_locals != 0 &&
1773 interpreter_frame->sender_sp() ==
1774 interpreter_frame->interpreter_frame_sender_sp()) {
1775 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1776 extra_locals);
1777 }
1778 *interpreter_frame->interpreter_frame_cache_addr() =
1779 method->constants()->cache();
1780 }
1781 return size;
1782 }
1783
1784 //-----------------------------------------------------------------------------
1785 // Exceptions
1786
1787 void TemplateInterpreterGenerator::generate_throw_exception() {
1788 // Entry point in previous activation (i.e., if the caller was
1789 // interpreted)
1790 Interpreter::_rethrow_exception_entry = __ pc();
1791 // Restore sp to interpreter_frame_last_sp even though we are going
1792 // to empty the expression stack for the exception processing.
1793 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1794 // rax: exception
1795 // rdx: return address/pc that threw exception
1796 __ restore_bcp(); // r13 points to call/send
1797 __ restore_locals();
1798 __ reinit_heapbase(); // restore r12 as heapbase.
1799 // Entry point for exceptions thrown within interpreter code
1800 Interpreter::_throw_exception_entry = __ pc();
1801 // expression stack is undefined here
1802 // rax: exception
1803 // r13: exception bcp
1804 __ verify_oop(rax);
1805 __ mov(c_rarg1, rax);
1806
1807 // expression stack must be empty before entering the VM in case of
1808 // an exception
1809 __ empty_expression_stack();
1810 // find exception handler address and preserve exception oop
1811 __ call_VM(rdx,
1812 CAST_FROM_FN_PTR(address,
1813 InterpreterRuntime::exception_handler_for_exception),
1814 c_rarg1);
1815 // rax: exception handler entry point
1816 // rdx: preserved exception oop
1817 // r13: bcp for exception handler
1818 __ push_ptr(rdx); // push exception which is now the only value on the stack
1819 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1820
1821 // If the exception is not handled in the current frame the frame is
1822 // removed and the exception is rethrown (i.e. exception
1823 // continuation is _rethrow_exception).
1824 //
1825 // Note: At this point the bci is still the bxi for the instruction
1826 // which caused the exception and the expression stack is
1827 // empty. Thus, for any VM calls at this point, GC will find a legal
1828 // oop map (with empty expression stack).
1829
1830 // In current activation
1831 // tos: exception
1832 // esi: exception bcp
1833
1834 //
1835 // JVMTI PopFrame support
1836 //
1837
1838 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1839 __ empty_expression_stack();
1840 // Set the popframe_processing bit in pending_popframe_condition
1841 // indicating that we are currently handling popframe, so that
1842 // call_VMs that may happen later do not trigger new popframe
1843 // handling cycles.
1844 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1845 __ orl(rdx, JavaThread::popframe_processing_bit);
1846 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1847
1848 {
1849 // Check to see whether we are returning to a deoptimized frame.
1850 // (The PopFrame call ensures that the caller of the popped frame is
1851 // either interpreted or compiled and deoptimizes it if compiled.)
1852 // In this case, we can't call dispatch_next() after the frame is
1853 // popped, but instead must save the incoming arguments and restore
1854 // them after deoptimization has occurred.
1855 //
1856 // Note that we don't compare the return PC against the
1857 // deoptimization blob's unpack entry because of the presence of
1858 // adapter frames in C2.
1859 Label caller_not_deoptimized;
1860 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1861 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1862 InterpreterRuntime::interpreter_contains), c_rarg1);
1863 __ testl(rax, rax);
1864 __ jcc(Assembler::notZero, caller_not_deoptimized);
1865
1866 // Compute size of arguments for saving when returning to
1867 // deoptimized caller
1868 __ get_method(rax);
1869 __ movptr(rax, Address(rax, Method::const_offset()));
1870 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1871 size_of_parameters_offset())));
1872 __ shll(rax, Interpreter::logStackElementSize);
1873 __ restore_locals(); // XXX do we need this?
1874 __ subptr(r14, rax);
1875 __ addptr(r14, wordSize);
1876 // Save these arguments
1877 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1878 Deoptimization::
1879 popframe_preserve_args),
1880 r15_thread, rax, r14);
1881
1882 __ remove_activation(vtos, rdx,
1883 /* throw_monitor_exception */ false,
1884 /* install_monitor_exception */ false,
1885 /* notify_jvmdi */ false);
1886
1887 // Inform deoptimization that it is responsible for restoring
1888 // these arguments
1889 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1890 JavaThread::popframe_force_deopt_reexecution_bit);
1891
1892 // Continue in deoptimization handler
1893 __ jmp(rdx);
1894
1895 __ bind(caller_not_deoptimized);
1896 }
1897
1898 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1899 /* throw_monitor_exception */ false,
1900 /* install_monitor_exception */ false,
1901 /* notify_jvmdi */ false);
1902
1903 // Finish with popframe handling
1904 // A previous I2C followed by a deoptimization might have moved the
1905 // outgoing arguments further up the stack. PopFrame expects the
1906 // mutations to those outgoing arguments to be preserved and other
1907 // constraints basically require this frame to look exactly as
1908 // though it had previously invoked an interpreted activation with
1909 // no space between the top of the expression stack (current
1910 // last_sp) and the top of stack. Rather than force deopt to
1911 // maintain this kind of invariant all the time we call a small
1912 // fixup routine to move the mutated arguments onto the top of our
1913 // expression stack if necessary.
1914 __ mov(c_rarg1, rsp);
1915 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1916 // PC must point into interpreter here
1917 __ set_last_Java_frame(noreg, rbp, __ pc());
1918 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1919 __ reset_last_Java_frame(true, true);
1920 // Restore the last_sp and null it out
1921 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1922 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1923
1924 __ restore_bcp(); // XXX do we need this?
1925 __ restore_locals(); // XXX do we need this?
1926 // The method data pointer was incremented already during
1927 // call profiling. We have to restore the mdp for the current bcp.
1928 if (ProfileInterpreter) {
1929 __ set_method_data_pointer_for_bcp();
1930 }
1931
1932 // Clear the popframe condition flag
1933 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1934 JavaThread::popframe_inactive);
1935
1936 #if INCLUDE_JVMTI
1937 if (EnableInvokeDynamic) {
1938 Label L_done;
1939 const Register local0 = r14;
1940
1941 __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1942 __ jcc(Assembler::notEqual, L_done);
1943
1944 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1945 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1946
1947 __ get_method(rdx);
1948 __ movptr(rax, Address(local0, 0));
1949 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1950
1951 __ testptr(rax, rax);
1952 __ jcc(Assembler::zero, L_done);
1953
1954 __ movptr(Address(rbx, 0), rax);
1955 __ bind(L_done);
1956 }
1957 #endif // INCLUDE_JVMTI
1958
1959 __ dispatch_next(vtos);
1960 // end of PopFrame support
1961
1962 Interpreter::_remove_activation_entry = __ pc();
1963
1964 // preserve exception over this code sequence
1965 __ pop_ptr(rax);
1966 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1967 // remove the activation (without doing throws on illegalMonitorExceptions)
1968 __ remove_activation(vtos, rdx, false, true, false);
1969 // restore exception
1970 __ get_vm_result(rax, r15_thread);
1971
1972 // In between activations - previous activation type unknown yet
1973 // compute continuation point - the continuation point expects the
1974 // following registers set up:
1975 //
1976 // rax: exception
1977 // rdx: return address/pc that threw exception
1978 // rsp: expression stack of caller
1979 // rbp: ebp of caller
1980 __ push(rax); // save exception
1981 __ push(rdx); // save return address
1982 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1983 SharedRuntime::exception_handler_for_return_address),
1984 r15_thread, rdx);
1985 __ mov(rbx, rax); // save exception handler
1986 __ pop(rdx); // restore return address
1987 __ pop(rax); // restore exception
1988 // Note that an "issuing PC" is actually the next PC after the call
1989 __ jmp(rbx); // jump to exception
1990 // handler of caller
1991 }
1992
1993
1994 //
1995 // JVMTI ForceEarlyReturn support
1996 //
1997 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1998 address entry = __ pc();
1999
2000 __ restore_bcp();
2001 __ restore_locals();
2002 __ empty_expression_stack();
2003 __ load_earlyret_value(state);
2004
2005 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
2006 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
2007
2008 // Clear the earlyret state
2009 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
2010
2011 __ remove_activation(state, rsi,
2012 false, /* throw_monitor_exception */
2013 false, /* install_monitor_exception */
2014 true); /* notify_jvmdi */
2015 __ jmp(rsi);
2016
2017 return entry;
2018 } // end of ForceEarlyReturn support
2019
2020
2021 //-----------------------------------------------------------------------------
2022 // Helper for vtos entry point generation
2023
2024 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2025 address& bep,
2026 address& cep,
2027 address& sep,
2028 address& aep,
2029 address& iep,
2030 address& lep,
2031 address& fep,
2032 address& dep,
2033 address& vep) {
2034 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2035 Label L;
2036 aep = __ pc(); __ push_ptr(); __ jmp(L);
2037 fep = __ pc(); __ push_f(); __ jmp(L);
2038 dep = __ pc(); __ push_d(); __ jmp(L);
2039 lep = __ pc(); __ push_l(); __ jmp(L);
2040 bep = cep = sep =
2041 iep = __ pc(); __ push_i();
2042 vep = __ pc();
2043 __ bind(L);
2044 generate_and_dispatch(t);
2045 }
2046
2047
2048 //-----------------------------------------------------------------------------
2049 // Generation of individual instructions
2050
2051 // helpers for generate_and_dispatch
2052
2053
2054 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2055 : TemplateInterpreterGenerator(code) {
2056 generate_all(); // down here so it can be "virtual"
2057 }
2058
2059 //-----------------------------------------------------------------------------
2060
2061 // Non-product code
2062 #ifndef PRODUCT
2063 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2064 address entry = __ pc();
2065
2066 __ push(state);
2067 __ push(c_rarg0);
2068 __ push(c_rarg1);
2069 __ push(c_rarg2);
2070 __ push(c_rarg3);
2071 __ mov(c_rarg2, rax); // Pass itos
2072 #ifdef _WIN64
2073 __ movflt(xmm3, xmm0); // Pass ftos
2074 #endif
2075 __ call_VM(noreg,
2076 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
2077 c_rarg1, c_rarg2, c_rarg3);
2078 __ pop(c_rarg3);
2079 __ pop(c_rarg2);
2080 __ pop(c_rarg1);
2081 __ pop(c_rarg0);
2082 __ pop(state);
2083 __ ret(0); // return from result handler
2084
2085 return entry;
2086 }
2087
2088 void TemplateInterpreterGenerator::count_bytecode() {
2089 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
2090 }
2091
2092 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2093 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
2094 }
2095
2096 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2097 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
2098 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
2099 __ orl(rbx,
2100 ((int) t->bytecode()) <<
2101 BytecodePairHistogram::log2_number_of_codes);
2102 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
2103 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
2104 __ incrementl(Address(rscratch1, rbx, Address::times_4));
2105 }
2106
2107
2108 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2109 // Call a little run-time stub to avoid blow-up for each bytecode.
2110 // The run-time runtime saves the right registers, depending on
2111 // the tosca in-state for the given template.
2112
2113 assert(Interpreter::trace_code(t->tos_in()) != NULL,
2114 "entry must have been generated");
2115 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
2116 __ andptr(rsp, -16); // align stack as required by ABI
2117 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
2118 __ mov(rsp, r12); // restore sp
2119 __ reinit_heapbase();
2120 }
2121
2122
2123 void TemplateInterpreterGenerator::stop_interpreter_at() {
2124 Label L;
2125 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2126 StopInterpreterAt);
2127 __ jcc(Assembler::notEqual, L);
2128 __ int3();
2129 __ bind(L);
2130 }
2131 #endif // !PRODUCT
2132 #endif // ! CC_INTERP