Tue, 29 Jun 2010 10:34:00 -0700
6964774: Adjust optimization flags setting
Summary: Adjust performance flags settings.
Reviewed-by: never, phh
1 /*
2 * Copyright (c) 2003, 2010, 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 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_interpreter_x86_64.cpp.incl"
28 #define __ _masm->
30 #ifndef CC_INTERP
32 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
33 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
36 //-----------------------------------------------------------------------------
38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
39 address entry = __ pc();
41 #ifdef ASSERT
42 {
43 Label L;
44 __ lea(rax, Address(rbp,
45 frame::interpreter_frame_monitor_block_top_offset *
46 wordSize));
47 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
48 // grows negative)
49 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
50 __ stop ("interpreter frame not set up");
51 __ bind(L);
52 }
53 #endif // ASSERT
54 // Restore bcp under the assumption that the current frame is still
55 // interpreted
56 __ restore_bcp();
58 // expression stack must be empty before entering the VM if an
59 // exception happened
60 __ empty_expression_stack();
61 // throw exception
62 __ call_VM(noreg,
63 CAST_FROM_FN_PTR(address,
64 InterpreterRuntime::throw_StackOverflowError));
65 return entry;
66 }
68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
69 const char* name) {
70 address entry = __ pc();
71 // expression stack must be empty before entering the VM if an
72 // exception happened
73 __ empty_expression_stack();
74 // setup parameters
75 // ??? convention: expect aberrant index in register ebx
76 __ lea(c_rarg1, ExternalAddress((address)name));
77 __ call_VM(noreg,
78 CAST_FROM_FN_PTR(address,
79 InterpreterRuntime::
80 throw_ArrayIndexOutOfBoundsException),
81 c_rarg1, rbx);
82 return entry;
83 }
85 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
86 address entry = __ pc();
88 // object is at TOS
89 __ pop(c_rarg1);
91 // expression stack must be empty before entering the VM if an
92 // exception happened
93 __ empty_expression_stack();
95 __ call_VM(noreg,
96 CAST_FROM_FN_PTR(address,
97 InterpreterRuntime::
98 throw_ClassCastException),
99 c_rarg1);
100 return entry;
101 }
103 // Arguments are: required type at TOS+8, failing object (or NULL) at TOS+4.
104 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
105 address entry = __ pc();
107 __ pop(c_rarg2); // failing object is at TOS
108 __ pop(c_rarg1); // required type is at TOS+8
110 __ verify_oop(c_rarg1);
111 __ verify_oop(c_rarg2);
113 // Various method handle types use interpreter registers as temps.
114 __ restore_bcp();
115 __ restore_locals();
117 // Expression stack must be empty before entering the VM for an exception.
118 __ empty_expression_stack();
120 __ call_VM(noreg,
121 CAST_FROM_FN_PTR(address,
122 InterpreterRuntime::throw_WrongMethodTypeException),
123 // pass required type, failing object (or NULL)
124 c_rarg1, c_rarg2);
125 return entry;
126 }
128 address TemplateInterpreterGenerator::generate_exception_handler_common(
129 const char* name, const char* message, bool pass_oop) {
130 assert(!pass_oop || message == NULL, "either oop or message but not both");
131 address entry = __ pc();
132 if (pass_oop) {
133 // object is at TOS
134 __ pop(c_rarg2);
135 }
136 // expression stack must be empty before entering the VM if an
137 // exception happened
138 __ empty_expression_stack();
139 // setup parameters
140 __ lea(c_rarg1, ExternalAddress((address)name));
141 if (pass_oop) {
142 __ call_VM(rax, CAST_FROM_FN_PTR(address,
143 InterpreterRuntime::
144 create_klass_exception),
145 c_rarg1, c_rarg2);
146 } else {
147 // kind of lame ExternalAddress can't take NULL because
148 // external_word_Relocation will assert.
149 if (message != NULL) {
150 __ lea(c_rarg2, ExternalAddress((address)message));
151 } else {
152 __ movptr(c_rarg2, NULL_WORD);
153 }
154 __ call_VM(rax,
155 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
156 c_rarg1, c_rarg2);
157 }
158 // throw exception
159 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
160 return entry;
161 }
164 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
165 address entry = __ pc();
166 // NULL last_sp until next java call
167 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
168 __ dispatch_next(state);
169 return entry;
170 }
173 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state,
174 int step) {
176 // amd64 doesn't need to do anything special about compiled returns
177 // to the interpreter so the code that exists on x86 to place a sentinel
178 // here and the specialized cleanup code is not needed here.
180 address entry = __ pc();
182 // Restore stack bottom in case i2c adjusted stack
183 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
184 // and NULL it as marker that esp is now tos until next java call
185 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
187 __ restore_bcp();
188 __ restore_locals();
190 Label L_got_cache, L_giant_index;
191 if (EnableInvokeDynamic) {
192 __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
193 __ jcc(Assembler::equal, L_giant_index);
194 }
195 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
196 __ bind(L_got_cache);
197 __ movl(rbx, Address(rbx, rcx,
198 Address::times_ptr,
199 in_bytes(constantPoolCacheOopDesc::base_offset()) +
200 3 * wordSize));
201 __ andl(rbx, 0xFF);
202 __ lea(rsp, Address(rsp, rbx, Address::times_8));
203 __ dispatch_next(state, step);
205 // out of the main line of code...
206 if (EnableInvokeDynamic) {
207 __ bind(L_giant_index);
208 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
209 __ jmp(L_got_cache);
210 }
212 return entry;
213 }
216 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
217 int step) {
218 address entry = __ pc();
219 // NULL last_sp until next java call
220 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
221 __ restore_bcp();
222 __ restore_locals();
223 // handle exceptions
224 {
225 Label L;
226 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
227 __ jcc(Assembler::zero, L);
228 __ call_VM(noreg,
229 CAST_FROM_FN_PTR(address,
230 InterpreterRuntime::throw_pending_exception));
231 __ should_not_reach_here();
232 __ bind(L);
233 }
234 __ dispatch_next(state, step);
235 return entry;
236 }
238 int AbstractInterpreter::BasicType_as_index(BasicType type) {
239 int i = 0;
240 switch (type) {
241 case T_BOOLEAN: i = 0; break;
242 case T_CHAR : i = 1; break;
243 case T_BYTE : i = 2; break;
244 case T_SHORT : i = 3; break;
245 case T_INT : i = 4; break;
246 case T_LONG : i = 5; break;
247 case T_VOID : i = 6; break;
248 case T_FLOAT : i = 7; break;
249 case T_DOUBLE : i = 8; break;
250 case T_OBJECT : i = 9; break;
251 case T_ARRAY : i = 9; break;
252 default : ShouldNotReachHere();
253 }
254 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
255 "index out of bounds");
256 return i;
257 }
260 address TemplateInterpreterGenerator::generate_result_handler_for(
261 BasicType type) {
262 address entry = __ pc();
263 switch (type) {
264 case T_BOOLEAN: __ c2bool(rax); break;
265 case T_CHAR : __ movzwl(rax, rax); break;
266 case T_BYTE : __ sign_extend_byte(rax); break;
267 case T_SHORT : __ sign_extend_short(rax); break;
268 case T_INT : /* nothing to do */ break;
269 case T_LONG : /* nothing to do */ break;
270 case T_VOID : /* nothing to do */ break;
271 case T_FLOAT : /* nothing to do */ break;
272 case T_DOUBLE : /* nothing to do */ break;
273 case T_OBJECT :
274 // retrieve result from frame
275 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
276 // and verify it
277 __ verify_oop(rax);
278 break;
279 default : ShouldNotReachHere();
280 }
281 __ ret(0); // return from result handler
282 return entry;
283 }
285 address TemplateInterpreterGenerator::generate_safept_entry_for(
286 TosState state,
287 address runtime_entry) {
288 address entry = __ pc();
289 __ push(state);
290 __ call_VM(noreg, runtime_entry);
291 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
292 return entry;
293 }
297 // Helpers for commoning out cases in the various type of method entries.
298 //
301 // increment invocation count & check for overflow
302 //
303 // Note: checking for negative value instead of overflow
304 // so we have a 'sticky' overflow test
305 //
306 // rbx: method
307 // ecx: invocation counter
308 //
309 void InterpreterGenerator::generate_counter_incr(
310 Label* overflow,
311 Label* profile_method,
312 Label* profile_method_continue) {
314 const Address invocation_counter(rbx,
315 methodOopDesc::invocation_counter_offset() +
316 InvocationCounter::counter_offset());
317 const Address backedge_counter(rbx,
318 methodOopDesc::backedge_counter_offset() +
319 InvocationCounter::counter_offset());
321 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
322 __ incrementl(Address(rbx,
323 methodOopDesc::interpreter_invocation_counter_offset()));
324 }
325 // Update standard invocation counters
326 __ movl(rax, backedge_counter); // load backedge counter
328 __ incrementl(rcx, InvocationCounter::count_increment);
329 __ andl(rax, InvocationCounter::count_mask_value); // mask out the
330 // status bits
332 __ movl(invocation_counter, rcx); // save invocation count
333 __ addl(rcx, rax); // add both counters
335 // profile_method is non-null only for interpreted method so
336 // profile_method != NULL == !native_call
338 if (ProfileInterpreter && profile_method != NULL) {
339 // Test to see if we should create a method data oop
340 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
341 __ jcc(Assembler::less, *profile_method_continue);
343 // if no method data exists, go to profile_method
344 __ test_method_data_pointer(rax, *profile_method);
345 }
347 __ cmp32(rcx, ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
348 __ jcc(Assembler::aboveEqual, *overflow);
349 }
351 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
353 // Asm interpreter on entry
354 // r14 - locals
355 // r13 - bcp
356 // rbx - method
357 // edx - cpool --- DOES NOT APPEAR TO BE TRUE
358 // rbp - interpreter frame
360 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
361 // Everything as it was on entry
362 // rdx is not restored. Doesn't appear to really be set.
364 const Address size_of_parameters(rbx,
365 methodOopDesc::size_of_parameters_offset());
367 // InterpreterRuntime::frequency_counter_overflow takes two
368 // arguments, the first (thread) is passed by call_VM, the second
369 // indicates if the counter overflow occurs at a backwards branch
370 // (NULL bcp). We pass zero for it. The call returns the address
371 // of the verified entry point for the method or NULL if the
372 // compilation did not complete (either went background or bailed
373 // out).
374 __ movl(c_rarg1, 0);
375 __ call_VM(noreg,
376 CAST_FROM_FN_PTR(address,
377 InterpreterRuntime::frequency_counter_overflow),
378 c_rarg1);
380 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
381 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
382 // and jump to the interpreted entry.
383 __ jmp(*do_continue, relocInfo::none);
384 }
386 // See if we've got enough room on the stack for locals plus overhead.
387 // The expression stack grows down incrementally, so the normal guard
388 // page mechanism will work for that.
389 //
390 // NOTE: Since the additional locals are also always pushed (wasn't
391 // obvious in generate_method_entry) so the guard should work for them
392 // too.
393 //
394 // Args:
395 // rdx: number of additional locals this frame needs (what we must check)
396 // rbx: methodOop
397 //
398 // Kills:
399 // rax
400 void InterpreterGenerator::generate_stack_overflow_check(void) {
402 // monitor entry size: see picture of stack set
403 // (generate_method_entry) and frame_amd64.hpp
404 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
406 // total overhead size: entry_size + (saved rbp through expr stack
407 // bottom). be sure to change this if you add/subtract anything
408 // to/from the overhead area
409 const int overhead_size =
410 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
412 const int page_size = os::vm_page_size();
414 Label after_frame_check;
416 // see if the frame is greater than one page in size. If so,
417 // then we need to verify there is enough stack space remaining
418 // for the additional locals.
419 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
420 __ jcc(Assembler::belowEqual, after_frame_check);
422 // compute rsp as if this were going to be the last frame on
423 // the stack before the red zone
425 const Address stack_base(r15_thread, Thread::stack_base_offset());
426 const Address stack_size(r15_thread, Thread::stack_size_offset());
428 // locals + overhead, in bytes
429 __ mov(rax, rdx);
430 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
431 __ addptr(rax, overhead_size);
433 #ifdef ASSERT
434 Label stack_base_okay, stack_size_okay;
435 // verify that thread stack base is non-zero
436 __ cmpptr(stack_base, (int32_t)NULL_WORD);
437 __ jcc(Assembler::notEqual, stack_base_okay);
438 __ stop("stack base is zero");
439 __ bind(stack_base_okay);
440 // verify that thread stack size is non-zero
441 __ cmpptr(stack_size, 0);
442 __ jcc(Assembler::notEqual, stack_size_okay);
443 __ stop("stack size is zero");
444 __ bind(stack_size_okay);
445 #endif
447 // Add stack base to locals and subtract stack size
448 __ addptr(rax, stack_base);
449 __ subptr(rax, stack_size);
451 // Use the maximum number of pages we might bang.
452 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
453 (StackRedPages+StackYellowPages);
455 // add in the red and yellow zone sizes
456 __ addptr(rax, max_pages * page_size);
458 // check against the current stack bottom
459 __ cmpptr(rsp, rax);
460 __ jcc(Assembler::above, after_frame_check);
462 __ pop(rax); // get return address
463 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
465 // all done with frame size check
466 __ bind(after_frame_check);
467 }
469 // Allocate monitor and lock method (asm interpreter)
470 //
471 // Args:
472 // rbx: methodOop
473 // r14: locals
474 //
475 // Kills:
476 // rax
477 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
478 // rscratch1, rscratch2 (scratch regs)
479 void InterpreterGenerator::lock_method(void) {
480 // synchronize method
481 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
482 const Address monitor_block_top(
483 rbp,
484 frame::interpreter_frame_monitor_block_top_offset * wordSize);
485 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
487 #ifdef ASSERT
488 {
489 Label L;
490 __ movl(rax, access_flags);
491 __ testl(rax, JVM_ACC_SYNCHRONIZED);
492 __ jcc(Assembler::notZero, L);
493 __ stop("method doesn't need synchronization");
494 __ bind(L);
495 }
496 #endif // ASSERT
498 // get synchronization object
499 {
500 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
501 Klass::java_mirror_offset_in_bytes();
502 Label done;
503 __ movl(rax, access_flags);
504 __ testl(rax, JVM_ACC_STATIC);
505 // get receiver (assume this is frequent case)
506 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
507 __ jcc(Assembler::zero, done);
508 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
509 __ movptr(rax, Address(rax,
510 constantPoolOopDesc::pool_holder_offset_in_bytes()));
511 __ movptr(rax, Address(rax, mirror_offset));
513 #ifdef ASSERT
514 {
515 Label L;
516 __ testptr(rax, rax);
517 __ jcc(Assembler::notZero, L);
518 __ stop("synchronization object is NULL");
519 __ bind(L);
520 }
521 #endif // ASSERT
523 __ bind(done);
524 }
526 // add space for monitor & lock
527 __ subptr(rsp, entry_size); // add space for a monitor entry
528 __ movptr(monitor_block_top, rsp); // set new monitor block top
529 // store object
530 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
531 __ movptr(c_rarg1, rsp); // object address
532 __ lock_object(c_rarg1);
533 }
535 // Generate a fixed interpreter frame. This is identical setup for
536 // interpreted methods and for native methods hence the shared code.
537 //
538 // Args:
539 // rax: return address
540 // rbx: methodOop
541 // r14: pointer to locals
542 // r13: sender sp
543 // rdx: cp cache
544 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
545 // initialize fixed part of activation frame
546 __ push(rax); // save return address
547 __ enter(); // save old & set new rbp
548 __ push(r13); // set sender sp
549 __ push((int)NULL_WORD); // leave last_sp as null
550 __ movptr(r13, Address(rbx, methodOopDesc::const_offset())); // get constMethodOop
551 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
552 __ push(rbx); // save methodOop
553 if (ProfileInterpreter) {
554 Label method_data_continue;
555 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
556 __ testptr(rdx, rdx);
557 __ jcc(Assembler::zero, method_data_continue);
558 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
559 __ bind(method_data_continue);
560 __ push(rdx); // set the mdp (method data pointer)
561 } else {
562 __ push(0);
563 }
565 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
566 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
567 __ push(rdx); // set constant pool cache
568 __ push(r14); // set locals pointer
569 if (native_call) {
570 __ push(0); // no bcp
571 } else {
572 __ push(r13); // set bcp
573 }
574 __ push(0); // reserve word for pointer to expression stack bottom
575 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
576 }
578 // End of helpers
580 // Various method entries
581 //------------------------------------------------------------------------------------------------------------------------
582 //
583 //
585 // Call an accessor method (assuming it is resolved, otherwise drop
586 // into vanilla (slow path) entry
587 address InterpreterGenerator::generate_accessor_entry(void) {
588 // rbx: methodOop
590 // r13: senderSP must preserver for slow path, set SP to it on fast path
592 address entry_point = __ pc();
593 Label xreturn_path;
595 // do fastpath for resolved accessor methods
596 if (UseFastAccessorMethods) {
597 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites
598 // thereof; parameter size = 1
599 // Note: We can only use this code if the getfield has been resolved
600 // and if we don't have a null-pointer exception => check for
601 // these conditions first and use slow path if necessary.
602 Label slow_path;
603 // If we need a safepoint check, generate full interpreter entry.
604 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
605 SafepointSynchronize::_not_synchronized);
607 __ jcc(Assembler::notEqual, slow_path);
608 // rbx: method
609 __ movptr(rax, Address(rsp, wordSize));
611 // check if local 0 != NULL and read field
612 __ testptr(rax, rax);
613 __ jcc(Assembler::zero, slow_path);
615 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
616 // read first instruction word and extract bytecode @ 1 and index @ 2
617 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
618 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
619 // Shift codes right to get the index on the right.
620 // The bytecode fetched looks like <index><0xb4><0x2a>
621 __ shrl(rdx, 2 * BitsPerByte);
622 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
623 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
625 // rax: local 0
626 // rbx: method
627 // rdx: constant pool cache index
628 // rdi: constant pool cache
630 // check if getfield has been resolved and read constant pool cache entry
631 // check the validity of the cache entry by testing whether _indices field
632 // contains Bytecode::_getfield in b1 byte.
633 assert(in_words(ConstantPoolCacheEntry::size()) == 4,
634 "adjust shift below");
635 __ movl(rcx,
636 Address(rdi,
637 rdx,
638 Address::times_8,
639 constantPoolCacheOopDesc::base_offset() +
640 ConstantPoolCacheEntry::indices_offset()));
641 __ shrl(rcx, 2 * BitsPerByte);
642 __ andl(rcx, 0xFF);
643 __ cmpl(rcx, Bytecodes::_getfield);
644 __ jcc(Assembler::notEqual, slow_path);
646 // Note: constant pool entry is not valid before bytecode is resolved
647 __ movptr(rcx,
648 Address(rdi,
649 rdx,
650 Address::times_8,
651 constantPoolCacheOopDesc::base_offset() +
652 ConstantPoolCacheEntry::f2_offset()));
653 // edx: flags
654 __ movl(rdx,
655 Address(rdi,
656 rdx,
657 Address::times_8,
658 constantPoolCacheOopDesc::base_offset() +
659 ConstantPoolCacheEntry::flags_offset()));
661 Label notObj, notInt, notByte, notShort;
662 const Address field_address(rax, rcx, Address::times_1);
664 // Need to differentiate between igetfield, agetfield, bgetfield etc.
665 // because they are different sizes.
666 // Use the type from the constant pool cache
667 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
668 // Make sure we don't need to mask edx for tosBits after the above shift
669 ConstantPoolCacheEntry::verify_tosBits();
671 __ cmpl(rdx, atos);
672 __ jcc(Assembler::notEqual, notObj);
673 // atos
674 __ load_heap_oop(rax, field_address);
675 __ jmp(xreturn_path);
677 __ bind(notObj);
678 __ cmpl(rdx, itos);
679 __ jcc(Assembler::notEqual, notInt);
680 // itos
681 __ movl(rax, field_address);
682 __ jmp(xreturn_path);
684 __ bind(notInt);
685 __ cmpl(rdx, btos);
686 __ jcc(Assembler::notEqual, notByte);
687 // btos
688 __ load_signed_byte(rax, field_address);
689 __ jmp(xreturn_path);
691 __ bind(notByte);
692 __ cmpl(rdx, stos);
693 __ jcc(Assembler::notEqual, notShort);
694 // stos
695 __ load_signed_short(rax, field_address);
696 __ jmp(xreturn_path);
698 __ bind(notShort);
699 #ifdef ASSERT
700 Label okay;
701 __ cmpl(rdx, ctos);
702 __ jcc(Assembler::equal, okay);
703 __ stop("what type is this?");
704 __ bind(okay);
705 #endif
706 // ctos
707 __ load_unsigned_short(rax, field_address);
709 __ bind(xreturn_path);
711 // _ireturn/_areturn
712 __ pop(rdi);
713 __ mov(rsp, r13);
714 __ jmp(rdi);
715 __ ret(0);
717 // generate a vanilla interpreter entry as the slow path
718 __ bind(slow_path);
719 (void) generate_normal_entry(false);
720 } else {
721 (void) generate_normal_entry(false);
722 }
724 return entry_point;
725 }
727 // Interpreter stub for calling a native method. (asm interpreter)
728 // This sets up a somewhat different looking stack for calling the
729 // native method than the typical interpreter frame setup.
730 address InterpreterGenerator::generate_native_entry(bool synchronized) {
731 // determine code generation flags
732 bool inc_counter = UseCompiler || CountCompiledCalls;
734 // rbx: methodOop
735 // r13: sender sp
737 address entry_point = __ pc();
739 const Address size_of_parameters(rbx, methodOopDesc::
740 size_of_parameters_offset());
741 const Address invocation_counter(rbx, methodOopDesc::
742 invocation_counter_offset() +
743 InvocationCounter::counter_offset());
744 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
746 // get parameter size (always needed)
747 __ load_unsigned_short(rcx, size_of_parameters);
749 // native calls don't need the stack size check since they have no
750 // expression stack and the arguments are already on the stack and
751 // we only add a handful of words to the stack
753 // rbx: methodOop
754 // rcx: size of parameters
755 // r13: sender sp
756 __ pop(rax); // get return address
758 // for natives the size of locals is zero
760 // compute beginning of parameters (r14)
761 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
763 // add 2 zero-initialized slots for native calls
764 // initialize result_handler slot
765 __ push((int) NULL_WORD);
766 // slot for oop temp
767 // (static native method holder mirror/jni oop result)
768 __ push((int) NULL_WORD);
770 if (inc_counter) {
771 __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
772 }
774 // initialize fixed part of activation frame
775 generate_fixed_frame(true);
777 // make sure method is native & not abstract
778 #ifdef ASSERT
779 __ movl(rax, access_flags);
780 {
781 Label L;
782 __ testl(rax, JVM_ACC_NATIVE);
783 __ jcc(Assembler::notZero, L);
784 __ stop("tried to execute non-native method as native");
785 __ bind(L);
786 }
787 {
788 Label L;
789 __ testl(rax, JVM_ACC_ABSTRACT);
790 __ jcc(Assembler::zero, L);
791 __ stop("tried to execute abstract method in interpreter");
792 __ bind(L);
793 }
794 #endif
796 // Since at this point in the method invocation the exception handler
797 // would try to exit the monitor of synchronized methods which hasn't
798 // been entered yet, we set the thread local variable
799 // _do_not_unlock_if_synchronized to true. The remove_activation will
800 // check this flag.
802 const Address do_not_unlock_if_synchronized(r15_thread,
803 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
804 __ movbool(do_not_unlock_if_synchronized, true);
806 // increment invocation count & check for overflow
807 Label invocation_counter_overflow;
808 if (inc_counter) {
809 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
810 }
812 Label continue_after_compile;
813 __ bind(continue_after_compile);
815 bang_stack_shadow_pages(true);
817 // reset the _do_not_unlock_if_synchronized flag
818 __ movbool(do_not_unlock_if_synchronized, false);
820 // check for synchronized methods
821 // Must happen AFTER invocation_counter check and stack overflow check,
822 // so method is not locked if overflows.
823 if (synchronized) {
824 lock_method();
825 } else {
826 // no synchronization necessary
827 #ifdef ASSERT
828 {
829 Label L;
830 __ movl(rax, access_flags);
831 __ testl(rax, JVM_ACC_SYNCHRONIZED);
832 __ jcc(Assembler::zero, L);
833 __ stop("method needs synchronization");
834 __ bind(L);
835 }
836 #endif
837 }
839 // start execution
840 #ifdef ASSERT
841 {
842 Label L;
843 const Address monitor_block_top(rbp,
844 frame::interpreter_frame_monitor_block_top_offset * wordSize);
845 __ movptr(rax, monitor_block_top);
846 __ cmpptr(rax, rsp);
847 __ jcc(Assembler::equal, L);
848 __ stop("broken stack frame setup in interpreter");
849 __ bind(L);
850 }
851 #endif
853 // jvmti support
854 __ notify_method_entry();
856 // work registers
857 const Register method = rbx;
858 const Register t = r11;
860 // allocate space for parameters
861 __ get_method(method);
862 __ verify_oop(method);
863 __ load_unsigned_short(t,
864 Address(method,
865 methodOopDesc::size_of_parameters_offset()));
866 __ shll(t, Interpreter::logStackElementSize);
868 __ subptr(rsp, t);
869 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
870 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
872 // get signature handler
873 {
874 Label L;
875 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
876 __ testptr(t, t);
877 __ jcc(Assembler::notZero, L);
878 __ call_VM(noreg,
879 CAST_FROM_FN_PTR(address,
880 InterpreterRuntime::prepare_native_call),
881 method);
882 __ get_method(method);
883 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
884 __ bind(L);
885 }
887 // call signature handler
888 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
889 "adjust this code");
890 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
891 "adjust this code");
892 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
893 "adjust this code");
895 // The generated handlers do not touch RBX (the method oop).
896 // However, large signatures cannot be cached and are generated
897 // each time here. The slow-path generator can do a GC on return,
898 // so we must reload it after the call.
899 __ call(t);
900 __ get_method(method); // slow path can do a GC, reload RBX
903 // result handler is in rax
904 // set result handler
905 __ movptr(Address(rbp,
906 (frame::interpreter_frame_result_handler_offset) * wordSize),
907 rax);
909 // pass mirror handle if static call
910 {
911 Label L;
912 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() +
913 Klass::java_mirror_offset_in_bytes();
914 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
915 __ testl(t, JVM_ACC_STATIC);
916 __ jcc(Assembler::zero, L);
917 // get mirror
918 __ movptr(t, Address(method, methodOopDesc::constants_offset()));
919 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
920 __ movptr(t, Address(t, mirror_offset));
921 // copy mirror into activation frame
922 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
923 t);
924 // pass handle to mirror
925 __ lea(c_rarg1,
926 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
927 __ bind(L);
928 }
930 // get native function entry point
931 {
932 Label L;
933 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
934 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
935 __ movptr(rscratch2, unsatisfied.addr());
936 __ cmpptr(rax, rscratch2);
937 __ jcc(Assembler::notEqual, L);
938 __ call_VM(noreg,
939 CAST_FROM_FN_PTR(address,
940 InterpreterRuntime::prepare_native_call),
941 method);
942 __ get_method(method);
943 __ verify_oop(method);
944 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
945 __ bind(L);
946 }
948 // pass JNIEnv
949 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
951 // It is enough that the pc() points into the right code
952 // segment. It does not have to be the correct return pc.
953 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
955 // change thread state
956 #ifdef ASSERT
957 {
958 Label L;
959 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
960 __ cmpl(t, _thread_in_Java);
961 __ jcc(Assembler::equal, L);
962 __ stop("Wrong thread state in native stub");
963 __ bind(L);
964 }
965 #endif
967 // Change state to native
969 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
970 _thread_in_native);
972 // Call the native method.
973 __ call(rax);
974 // result potentially in rax or xmm0
976 // Depending on runtime options, either restore the MXCSR
977 // register after returning from the JNI Call or verify that
978 // it wasn't changed during -Xcheck:jni.
979 if (RestoreMXCSROnJNICalls) {
980 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
981 }
982 else if (CheckJNICalls) {
983 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::x86::verify_mxcsr_entry())));
984 }
986 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
987 // in order to extract the result of a method call. If the order of these
988 // pushes change or anything else is added to the stack then the code in
989 // interpreter_frame_result must also change.
991 __ push(dtos);
992 __ push(ltos);
994 // change thread state
995 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
996 _thread_in_native_trans);
998 if (os::is_MP()) {
999 if (UseMembar) {
1000 // Force this write out before the read below
1001 __ membar(Assembler::Membar_mask_bits(
1002 Assembler::LoadLoad | Assembler::LoadStore |
1003 Assembler::StoreLoad | Assembler::StoreStore));
1004 } else {
1005 // Write serialization page so VM thread can do a pseudo remote membar.
1006 // We use the current thread pointer to calculate a thread specific
1007 // offset to write to within the page. This minimizes bus traffic
1008 // due to cache line collision.
1009 __ serialize_memory(r15_thread, rscratch2);
1010 }
1011 }
1013 // check for safepoint operation in progress and/or pending suspend requests
1014 {
1015 Label Continue;
1016 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1017 SafepointSynchronize::_not_synchronized);
1019 Label L;
1020 __ jcc(Assembler::notEqual, L);
1021 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1022 __ jcc(Assembler::equal, Continue);
1023 __ bind(L);
1025 // Don't use call_VM as it will see a possible pending exception
1026 // and forward it and never return here preventing us from
1027 // clearing _last_native_pc down below. Also can't use
1028 // call_VM_leaf either as it will check to see if r13 & r14 are
1029 // preserved and correspond to the bcp/locals pointers. So we do a
1030 // runtime call by hand.
1031 //
1032 __ mov(c_rarg0, r15_thread);
1033 __ mov(r12, rsp); // remember sp
1034 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1035 __ andptr(rsp, -16); // align stack as required by ABI
1036 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1037 __ mov(rsp, r12); // restore sp
1038 __ reinit_heapbase();
1039 __ bind(Continue);
1040 }
1042 // change thread state
1043 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1045 // reset_last_Java_frame
1046 __ reset_last_Java_frame(true, true);
1048 // reset handle block
1049 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1050 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1052 // If result is an oop unbox and store it in frame where gc will see it
1053 // and result handler will pick it up
1055 {
1056 Label no_oop, store_result;
1057 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1058 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1059 __ jcc(Assembler::notEqual, no_oop);
1060 // retrieve result
1061 __ pop(ltos);
1062 __ testptr(rax, rax);
1063 __ jcc(Assembler::zero, store_result);
1064 __ movptr(rax, Address(rax, 0));
1065 __ bind(store_result);
1066 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1067 // keep stack depth as expected by pushing oop which will eventually be discarde
1068 __ push(ltos);
1069 __ bind(no_oop);
1070 }
1073 {
1074 Label no_reguard;
1075 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1076 JavaThread::stack_guard_yellow_disabled);
1077 __ jcc(Assembler::notEqual, no_reguard);
1079 __ pusha(); // XXX only save smashed registers
1080 __ mov(r12, rsp); // remember sp
1081 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1082 __ andptr(rsp, -16); // align stack as required by ABI
1083 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1084 __ mov(rsp, r12); // restore sp
1085 __ popa(); // XXX only restore smashed registers
1086 __ reinit_heapbase();
1088 __ bind(no_reguard);
1089 }
1092 // The method register is junk from after the thread_in_native transition
1093 // until here. Also can't call_VM until the bcp has been
1094 // restored. Need bcp for throwing exception below so get it now.
1095 __ get_method(method);
1096 __ verify_oop(method);
1098 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1099 // r13 == code_base()
1100 __ movptr(r13, Address(method, methodOopDesc::const_offset())); // get constMethodOop
1101 __ lea(r13, Address(r13, constMethodOopDesc::codes_offset())); // get codebase
1102 // handle exceptions (exception handling will handle unlocking!)
1103 {
1104 Label L;
1105 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1106 __ jcc(Assembler::zero, L);
1107 // Note: At some point we may want to unify this with the code
1108 // used in call_VM_base(); i.e., we should use the
1109 // StubRoutines::forward_exception code. For now this doesn't work
1110 // here because the rsp is not correctly set at this point.
1111 __ MacroAssembler::call_VM(noreg,
1112 CAST_FROM_FN_PTR(address,
1113 InterpreterRuntime::throw_pending_exception));
1114 __ should_not_reach_here();
1115 __ bind(L);
1116 }
1118 // do unlocking if necessary
1119 {
1120 Label L;
1121 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1122 __ testl(t, JVM_ACC_SYNCHRONIZED);
1123 __ jcc(Assembler::zero, L);
1124 // the code below should be shared with interpreter macro
1125 // assembler implementation
1126 {
1127 Label unlock;
1128 // BasicObjectLock will be first in list, since this is a
1129 // synchronized method. However, need to check that the object
1130 // has not been unlocked by an explicit monitorexit bytecode.
1131 const Address monitor(rbp,
1132 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1133 wordSize - sizeof(BasicObjectLock)));
1135 // monitor expect in c_rarg1 for slow unlock path
1136 __ lea(c_rarg1, monitor); // address of first monitor
1138 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1139 __ testptr(t, t);
1140 __ jcc(Assembler::notZero, unlock);
1142 // Entry already unlocked, need to throw exception
1143 __ MacroAssembler::call_VM(noreg,
1144 CAST_FROM_FN_PTR(address,
1145 InterpreterRuntime::throw_illegal_monitor_state_exception));
1146 __ should_not_reach_here();
1148 __ bind(unlock);
1149 __ unlock_object(c_rarg1);
1150 }
1151 __ bind(L);
1152 }
1154 // jvmti support
1155 // Note: This must happen _after_ handling/throwing any exceptions since
1156 // the exception handler code notifies the runtime of method exits
1157 // too. If this happens before, method entry/exit notifications are
1158 // not properly paired (was bug - gri 11/22/99).
1159 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1161 // restore potential result in edx:eax, call result handler to
1162 // restore potential result in ST0 & handle result
1164 __ pop(ltos);
1165 __ pop(dtos);
1167 __ movptr(t, Address(rbp,
1168 (frame::interpreter_frame_result_handler_offset) * wordSize));
1169 __ call(t);
1171 // remove activation
1172 __ movptr(t, Address(rbp,
1173 frame::interpreter_frame_sender_sp_offset *
1174 wordSize)); // get sender sp
1175 __ leave(); // remove frame anchor
1176 __ pop(rdi); // get return address
1177 __ mov(rsp, t); // set sp to sender sp
1178 __ jmp(rdi);
1180 if (inc_counter) {
1181 // Handle overflow of counter and compile method
1182 __ bind(invocation_counter_overflow);
1183 generate_counter_overflow(&continue_after_compile);
1184 }
1186 return entry_point;
1187 }
1189 //
1190 // Generic interpreted method entry to (asm) interpreter
1191 //
1192 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1193 // determine code generation flags
1194 bool inc_counter = UseCompiler || CountCompiledCalls;
1196 // ebx: methodOop
1197 // r13: sender sp
1198 address entry_point = __ pc();
1200 const Address size_of_parameters(rbx,
1201 methodOopDesc::size_of_parameters_offset());
1202 const Address size_of_locals(rbx, methodOopDesc::size_of_locals_offset());
1203 const Address invocation_counter(rbx,
1204 methodOopDesc::invocation_counter_offset() +
1205 InvocationCounter::counter_offset());
1206 const Address access_flags(rbx, methodOopDesc::access_flags_offset());
1208 // get parameter size (always needed)
1209 __ load_unsigned_short(rcx, size_of_parameters);
1211 // rbx: methodOop
1212 // rcx: size of parameters
1213 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1215 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1216 __ subl(rdx, rcx); // rdx = no. of additional locals
1218 // YYY
1219 // __ incrementl(rdx);
1220 // __ andl(rdx, -2);
1222 // see if we've got enough room on the stack for locals plus overhead.
1223 generate_stack_overflow_check();
1225 // get return address
1226 __ pop(rax);
1228 // compute beginning of parameters (r14)
1229 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1231 // rdx - # of additional locals
1232 // allocate space for locals
1233 // explicitly initialize locals
1234 {
1235 Label exit, loop;
1236 __ testl(rdx, rdx);
1237 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1238 __ bind(loop);
1239 __ push((int) NULL_WORD); // initialize local variables
1240 __ decrementl(rdx); // until everything initialized
1241 __ jcc(Assembler::greater, loop);
1242 __ bind(exit);
1243 }
1245 // (pre-)fetch invocation count
1246 if (inc_counter) {
1247 __ movl(rcx, invocation_counter);
1248 }
1249 // initialize fixed part of activation frame
1250 generate_fixed_frame(false);
1252 // make sure method is not native & not abstract
1253 #ifdef ASSERT
1254 __ movl(rax, access_flags);
1255 {
1256 Label L;
1257 __ testl(rax, JVM_ACC_NATIVE);
1258 __ jcc(Assembler::zero, L);
1259 __ stop("tried to execute native method as non-native");
1260 __ bind(L);
1261 }
1262 {
1263 Label L;
1264 __ testl(rax, JVM_ACC_ABSTRACT);
1265 __ jcc(Assembler::zero, L);
1266 __ stop("tried to execute abstract method in interpreter");
1267 __ bind(L);
1268 }
1269 #endif
1271 // Since at this point in the method invocation the exception
1272 // handler would try to exit the monitor of synchronized methods
1273 // which hasn't been entered yet, we set the thread local variable
1274 // _do_not_unlock_if_synchronized to true. The remove_activation
1275 // will check this flag.
1277 const Address do_not_unlock_if_synchronized(r15_thread,
1278 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1279 __ movbool(do_not_unlock_if_synchronized, true);
1281 // increment invocation count & check for overflow
1282 Label invocation_counter_overflow;
1283 Label profile_method;
1284 Label profile_method_continue;
1285 if (inc_counter) {
1286 generate_counter_incr(&invocation_counter_overflow,
1287 &profile_method,
1288 &profile_method_continue);
1289 if (ProfileInterpreter) {
1290 __ bind(profile_method_continue);
1291 }
1292 }
1294 Label continue_after_compile;
1295 __ bind(continue_after_compile);
1297 // check for synchronized interpreted methods
1298 bang_stack_shadow_pages(false);
1300 // reset the _do_not_unlock_if_synchronized flag
1301 __ movbool(do_not_unlock_if_synchronized, false);
1303 // check for synchronized methods
1304 // Must happen AFTER invocation_counter check and stack overflow check,
1305 // so method is not locked if overflows.
1306 if (synchronized) {
1307 // Allocate monitor and lock method
1308 lock_method();
1309 } else {
1310 // no synchronization necessary
1311 #ifdef ASSERT
1312 {
1313 Label L;
1314 __ movl(rax, access_flags);
1315 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1316 __ jcc(Assembler::zero, L);
1317 __ stop("method needs synchronization");
1318 __ bind(L);
1319 }
1320 #endif
1321 }
1323 // start execution
1324 #ifdef ASSERT
1325 {
1326 Label L;
1327 const Address monitor_block_top (rbp,
1328 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1329 __ movptr(rax, monitor_block_top);
1330 __ cmpptr(rax, rsp);
1331 __ jcc(Assembler::equal, L);
1332 __ stop("broken stack frame setup in interpreter");
1333 __ bind(L);
1334 }
1335 #endif
1337 // jvmti support
1338 __ notify_method_entry();
1340 __ dispatch_next(vtos);
1342 // invocation counter overflow
1343 if (inc_counter) {
1344 if (ProfileInterpreter) {
1345 // We have decided to profile this method in the interpreter
1346 __ bind(profile_method);
1348 __ call_VM(noreg,
1349 CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method),
1350 r13, true);
1352 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
1353 __ movptr(rax, Address(rbx,
1354 in_bytes(methodOopDesc::method_data_offset())));
1355 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1356 rax);
1357 __ test_method_data_pointer(rax, profile_method_continue);
1358 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1359 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1360 rax);
1361 __ jmp(profile_method_continue);
1362 }
1363 // Handle overflow of counter and compile method
1364 __ bind(invocation_counter_overflow);
1365 generate_counter_overflow(&continue_after_compile);
1366 }
1368 return entry_point;
1369 }
1371 // Entry points
1372 //
1373 // Here we generate the various kind of entries into the interpreter.
1374 // The two main entry type are generic bytecode methods and native
1375 // call method. These both come in synchronized and non-synchronized
1376 // versions but the frame layout they create is very similar. The
1377 // other method entry types are really just special purpose entries
1378 // that are really entry and interpretation all in one. These are for
1379 // trivial methods like accessor, empty, or special math methods.
1380 //
1381 // When control flow reaches any of the entry types for the interpreter
1382 // the following holds ->
1383 //
1384 // Arguments:
1385 //
1386 // rbx: methodOop
1387 //
1388 // Stack layout immediately at entry
1389 //
1390 // [ return address ] <--- rsp
1391 // [ parameter n ]
1392 // ...
1393 // [ parameter 1 ]
1394 // [ expression stack ] (caller's java expression stack)
1396 // Assuming that we don't go to one of the trivial specialized entries
1397 // the stack will look like below when we are ready to execute the
1398 // first bytecode (or call the native routine). The register usage
1399 // will be as the template based interpreter expects (see
1400 // interpreter_amd64.hpp).
1401 //
1402 // local variables follow incoming parameters immediately; i.e.
1403 // the return address is moved to the end of the locals).
1404 //
1405 // [ monitor entry ] <--- rsp
1406 // ...
1407 // [ monitor entry ]
1408 // [ expr. stack bottom ]
1409 // [ saved r13 ]
1410 // [ current r14 ]
1411 // [ methodOop ]
1412 // [ saved ebp ] <--- rbp
1413 // [ return address ]
1414 // [ local variable m ]
1415 // ...
1416 // [ local variable 1 ]
1417 // [ parameter n ]
1418 // ...
1419 // [ parameter 1 ] <--- r14
1421 address AbstractInterpreterGenerator::generate_method_entry(
1422 AbstractInterpreter::MethodKind kind) {
1423 // determine code generation flags
1424 bool synchronized = false;
1425 address entry_point = NULL;
1427 switch (kind) {
1428 case Interpreter::zerolocals : break;
1429 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1430 case Interpreter::native : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(false); break;
1431 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*) this)->generate_native_entry(true); break;
1432 case Interpreter::empty : entry_point = ((InterpreterGenerator*) this)->generate_empty_entry(); break;
1433 case Interpreter::accessor : entry_point = ((InterpreterGenerator*) this)->generate_accessor_entry(); break;
1434 case Interpreter::abstract : entry_point = ((InterpreterGenerator*) this)->generate_abstract_entry(); break;
1435 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*) this)->generate_method_handle_entry();break;
1437 case Interpreter::java_lang_math_sin : // fall thru
1438 case Interpreter::java_lang_math_cos : // fall thru
1439 case Interpreter::java_lang_math_tan : // fall thru
1440 case Interpreter::java_lang_math_abs : // fall thru
1441 case Interpreter::java_lang_math_log : // fall thru
1442 case Interpreter::java_lang_math_log10 : // fall thru
1443 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*) this)->generate_math_entry(kind); break;
1444 default : ShouldNotReachHere(); break;
1445 }
1447 if (entry_point) {
1448 return entry_point;
1449 }
1451 return ((InterpreterGenerator*) this)->
1452 generate_normal_entry(synchronized);
1453 }
1455 // These should never be compiled since the interpreter will prefer
1456 // the compiled version to the intrinsic version.
1457 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1458 switch (method_kind(m)) {
1459 case Interpreter::java_lang_math_sin : // fall thru
1460 case Interpreter::java_lang_math_cos : // fall thru
1461 case Interpreter::java_lang_math_tan : // fall thru
1462 case Interpreter::java_lang_math_abs : // fall thru
1463 case Interpreter::java_lang_math_log : // fall thru
1464 case Interpreter::java_lang_math_log10 : // fall thru
1465 case Interpreter::java_lang_math_sqrt :
1466 return false;
1467 default:
1468 return true;
1469 }
1470 }
1472 // How much stack a method activation needs in words.
1473 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1474 const int entry_size = frame::interpreter_frame_monitor_size();
1476 // total overhead size: entry_size + (saved rbp thru expr stack
1477 // bottom). be sure to change this if you add/subtract anything
1478 // to/from the overhead area
1479 const int overhead_size =
1480 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1482 const int stub_code = frame::entry_frame_after_call_words;
1483 const int extra_stack = methodOopDesc::extra_stack_entries();
1484 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1485 Interpreter::stackElementWords;
1486 return (overhead_size + method_stack + stub_code);
1487 }
1489 int AbstractInterpreter::layout_activation(methodOop method,
1490 int tempcount,
1491 int popframe_extra_args,
1492 int moncount,
1493 int callee_param_count,
1494 int callee_locals,
1495 frame* caller,
1496 frame* interpreter_frame,
1497 bool is_top_frame) {
1498 // Note: This calculation must exactly parallel the frame setup
1499 // in AbstractInterpreterGenerator::generate_method_entry.
1500 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1501 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1502 // right size, as determined by a previous call to this method.
1503 // It is also guaranteed to be walkable even though it is in a skeletal state
1505 // fixed size of an interpreter frame:
1506 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1507 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1508 Interpreter::stackElementWords;
1510 int overhead = frame::sender_sp_offset -
1511 frame::interpreter_frame_initial_sp_offset;
1512 // Our locals were accounted for by the caller (or last_frame_adjust
1513 // on the transistion) Since the callee parameters already account
1514 // for the callee's params we only need to account for the extra
1515 // locals.
1516 int size = overhead +
1517 (callee_locals - callee_param_count)*Interpreter::stackElementWords +
1518 moncount * frame::interpreter_frame_monitor_size() +
1519 tempcount* Interpreter::stackElementWords + popframe_extra_args;
1520 if (interpreter_frame != NULL) {
1521 #ifdef ASSERT
1522 if (!EnableMethodHandles)
1523 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1524 // Probably, since deoptimization doesn't work yet.
1525 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1526 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1527 #endif
1529 interpreter_frame->interpreter_frame_set_method(method);
1530 // NOTE the difference in using sender_sp and
1531 // interpreter_frame_sender_sp interpreter_frame_sender_sp is
1532 // the original sp of the caller (the unextended_sp) and
1533 // sender_sp is fp+16 XXX
1534 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1536 interpreter_frame->interpreter_frame_set_locals(locals);
1537 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1538 BasicObjectLock* monbot = montop - moncount;
1539 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1541 // Set last_sp
1542 intptr_t* esp = (intptr_t*) monbot -
1543 tempcount*Interpreter::stackElementWords -
1544 popframe_extra_args;
1545 interpreter_frame->interpreter_frame_set_last_sp(esp);
1547 // All frames but the initial (oldest) interpreter frame we fill in have
1548 // a value for sender_sp that allows walking the stack but isn't
1549 // truly correct. Correct the value here.
1550 if (extra_locals != 0 &&
1551 interpreter_frame->sender_sp() ==
1552 interpreter_frame->interpreter_frame_sender_sp()) {
1553 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() +
1554 extra_locals);
1555 }
1556 *interpreter_frame->interpreter_frame_cache_addr() =
1557 method->constants()->cache();
1558 }
1559 return size;
1560 }
1562 //-----------------------------------------------------------------------------
1563 // Exceptions
1565 void TemplateInterpreterGenerator::generate_throw_exception() {
1566 // Entry point in previous activation (i.e., if the caller was
1567 // interpreted)
1568 Interpreter::_rethrow_exception_entry = __ pc();
1569 // Restore sp to interpreter_frame_last_sp even though we are going
1570 // to empty the expression stack for the exception processing.
1571 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1572 // rax: exception
1573 // rdx: return address/pc that threw exception
1574 __ restore_bcp(); // r13 points to call/send
1575 __ restore_locals();
1576 __ reinit_heapbase(); // restore r12 as heapbase.
1577 // Entry point for exceptions thrown within interpreter code
1578 Interpreter::_throw_exception_entry = __ pc();
1579 // expression stack is undefined here
1580 // rax: exception
1581 // r13: exception bcp
1582 __ verify_oop(rax);
1583 __ mov(c_rarg1, rax);
1585 // expression stack must be empty before entering the VM in case of
1586 // an exception
1587 __ empty_expression_stack();
1588 // find exception handler address and preserve exception oop
1589 __ call_VM(rdx,
1590 CAST_FROM_FN_PTR(address,
1591 InterpreterRuntime::exception_handler_for_exception),
1592 c_rarg1);
1593 // rax: exception handler entry point
1594 // rdx: preserved exception oop
1595 // r13: bcp for exception handler
1596 __ push_ptr(rdx); // push exception which is now the only value on the stack
1597 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1599 // If the exception is not handled in the current frame the frame is
1600 // removed and the exception is rethrown (i.e. exception
1601 // continuation is _rethrow_exception).
1602 //
1603 // Note: At this point the bci is still the bxi for the instruction
1604 // which caused the exception and the expression stack is
1605 // empty. Thus, for any VM calls at this point, GC will find a legal
1606 // oop map (with empty expression stack).
1608 // In current activation
1609 // tos: exception
1610 // esi: exception bcp
1612 //
1613 // JVMTI PopFrame support
1614 //
1616 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1617 __ empty_expression_stack();
1618 // Set the popframe_processing bit in pending_popframe_condition
1619 // indicating that we are currently handling popframe, so that
1620 // call_VMs that may happen later do not trigger new popframe
1621 // handling cycles.
1622 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1623 __ orl(rdx, JavaThread::popframe_processing_bit);
1624 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1626 {
1627 // Check to see whether we are returning to a deoptimized frame.
1628 // (The PopFrame call ensures that the caller of the popped frame is
1629 // either interpreted or compiled and deoptimizes it if compiled.)
1630 // In this case, we can't call dispatch_next() after the frame is
1631 // popped, but instead must save the incoming arguments and restore
1632 // them after deoptimization has occurred.
1633 //
1634 // Note that we don't compare the return PC against the
1635 // deoptimization blob's unpack entry because of the presence of
1636 // adapter frames in C2.
1637 Label caller_not_deoptimized;
1638 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1639 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1640 InterpreterRuntime::interpreter_contains), c_rarg1);
1641 __ testl(rax, rax);
1642 __ jcc(Assembler::notZero, caller_not_deoptimized);
1644 // Compute size of arguments for saving when returning to
1645 // deoptimized caller
1646 __ get_method(rax);
1647 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::
1648 size_of_parameters_offset())));
1649 __ shll(rax, Interpreter::logStackElementSize);
1650 __ restore_locals(); // XXX do we need this?
1651 __ subptr(r14, rax);
1652 __ addptr(r14, wordSize);
1653 // Save these arguments
1654 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1655 Deoptimization::
1656 popframe_preserve_args),
1657 r15_thread, rax, r14);
1659 __ remove_activation(vtos, rdx,
1660 /* throw_monitor_exception */ false,
1661 /* install_monitor_exception */ false,
1662 /* notify_jvmdi */ false);
1664 // Inform deoptimization that it is responsible for restoring
1665 // these arguments
1666 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1667 JavaThread::popframe_force_deopt_reexecution_bit);
1669 // Continue in deoptimization handler
1670 __ jmp(rdx);
1672 __ bind(caller_not_deoptimized);
1673 }
1675 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1676 /* throw_monitor_exception */ false,
1677 /* install_monitor_exception */ false,
1678 /* notify_jvmdi */ false);
1680 // Finish with popframe handling
1681 // A previous I2C followed by a deoptimization might have moved the
1682 // outgoing arguments further up the stack. PopFrame expects the
1683 // mutations to those outgoing arguments to be preserved and other
1684 // constraints basically require this frame to look exactly as
1685 // though it had previously invoked an interpreted activation with
1686 // no space between the top of the expression stack (current
1687 // last_sp) and the top of stack. Rather than force deopt to
1688 // maintain this kind of invariant all the time we call a small
1689 // fixup routine to move the mutated arguments onto the top of our
1690 // expression stack if necessary.
1691 __ mov(c_rarg1, rsp);
1692 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1693 // PC must point into interpreter here
1694 __ set_last_Java_frame(noreg, rbp, __ pc());
1695 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1696 __ reset_last_Java_frame(true, true);
1697 // Restore the last_sp and null it out
1698 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1699 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1701 __ restore_bcp(); // XXX do we need this?
1702 __ restore_locals(); // XXX do we need this?
1703 // The method data pointer was incremented already during
1704 // call profiling. We have to restore the mdp for the current bcp.
1705 if (ProfileInterpreter) {
1706 __ set_method_data_pointer_for_bcp();
1707 }
1709 // Clear the popframe condition flag
1710 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1711 JavaThread::popframe_inactive);
1713 __ dispatch_next(vtos);
1714 // end of PopFrame support
1716 Interpreter::_remove_activation_entry = __ pc();
1718 // preserve exception over this code sequence
1719 __ pop_ptr(rax);
1720 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1721 // remove the activation (without doing throws on illegalMonitorExceptions)
1722 __ remove_activation(vtos, rdx, false, true, false);
1723 // restore exception
1724 __ movptr(rax, Address(r15_thread, JavaThread::vm_result_offset()));
1725 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), (int32_t)NULL_WORD);
1726 __ verify_oop(rax);
1728 // In between activations - previous activation type unknown yet
1729 // compute continuation point - the continuation point expects the
1730 // following registers set up:
1731 //
1732 // rax: exception
1733 // rdx: return address/pc that threw exception
1734 // rsp: expression stack of caller
1735 // rbp: ebp of caller
1736 __ push(rax); // save exception
1737 __ push(rdx); // save return address
1738 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1739 SharedRuntime::exception_handler_for_return_address),
1740 r15_thread, rdx);
1741 __ mov(rbx, rax); // save exception handler
1742 __ pop(rdx); // restore return address
1743 __ pop(rax); // restore exception
1744 // Note that an "issuing PC" is actually the next PC after the call
1745 __ jmp(rbx); // jump to exception
1746 // handler of caller
1747 }
1750 //
1751 // JVMTI ForceEarlyReturn support
1752 //
1753 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1754 address entry = __ pc();
1756 __ restore_bcp();
1757 __ restore_locals();
1758 __ empty_expression_stack();
1759 __ load_earlyret_value(state);
1761 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1762 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1764 // Clear the earlyret state
1765 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1767 __ remove_activation(state, rsi,
1768 false, /* throw_monitor_exception */
1769 false, /* install_monitor_exception */
1770 true); /* notify_jvmdi */
1771 __ jmp(rsi);
1773 return entry;
1774 } // end of ForceEarlyReturn support
1777 //-----------------------------------------------------------------------------
1778 // Helper for vtos entry point generation
1780 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1781 address& bep,
1782 address& cep,
1783 address& sep,
1784 address& aep,
1785 address& iep,
1786 address& lep,
1787 address& fep,
1788 address& dep,
1789 address& vep) {
1790 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1791 Label L;
1792 aep = __ pc(); __ push_ptr(); __ jmp(L);
1793 fep = __ pc(); __ push_f(); __ jmp(L);
1794 dep = __ pc(); __ push_d(); __ jmp(L);
1795 lep = __ pc(); __ push_l(); __ jmp(L);
1796 bep = cep = sep =
1797 iep = __ pc(); __ push_i();
1798 vep = __ pc();
1799 __ bind(L);
1800 generate_and_dispatch(t);
1801 }
1804 //-----------------------------------------------------------------------------
1805 // Generation of individual instructions
1807 // helpers for generate_and_dispatch
1810 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1811 : TemplateInterpreterGenerator(code) {
1812 generate_all(); // down here so it can be "virtual"
1813 }
1815 //-----------------------------------------------------------------------------
1817 // Non-product code
1818 #ifndef PRODUCT
1819 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1820 address entry = __ pc();
1822 __ push(state);
1823 __ push(c_rarg0);
1824 __ push(c_rarg1);
1825 __ push(c_rarg2);
1826 __ push(c_rarg3);
1827 __ mov(c_rarg2, rax); // Pass itos
1828 #ifdef _WIN64
1829 __ movflt(xmm3, xmm0); // Pass ftos
1830 #endif
1831 __ call_VM(noreg,
1832 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1833 c_rarg1, c_rarg2, c_rarg3);
1834 __ pop(c_rarg3);
1835 __ pop(c_rarg2);
1836 __ pop(c_rarg1);
1837 __ pop(c_rarg0);
1838 __ pop(state);
1839 __ ret(0); // return from result handler
1841 return entry;
1842 }
1844 void TemplateInterpreterGenerator::count_bytecode() {
1845 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1846 }
1848 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1849 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1850 }
1852 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1853 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1854 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1855 __ orl(rbx,
1856 ((int) t->bytecode()) <<
1857 BytecodePairHistogram::log2_number_of_codes);
1858 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1859 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1860 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1861 }
1864 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1865 // Call a little run-time stub to avoid blow-up for each bytecode.
1866 // The run-time runtime saves the right registers, depending on
1867 // the tosca in-state for the given template.
1869 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1870 "entry must have been generated");
1871 __ mov(r12, rsp); // remember sp
1872 __ andptr(rsp, -16); // align stack as required by ABI
1873 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1874 __ mov(rsp, r12); // restore sp
1875 __ reinit_heapbase();
1876 }
1879 void TemplateInterpreterGenerator::stop_interpreter_at() {
1880 Label L;
1881 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1882 StopInterpreterAt);
1883 __ jcc(Assembler::notEqual, L);
1884 __ int3();
1885 __ bind(L);
1886 }
1887 #endif // !PRODUCT
1888 #endif // ! CC_INTERP