1 /*
2 * Copyright (c) 2003, 2015, 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/interp_masm.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #define __ _masm->
51
52 #ifndef CC_INTERP
53
54 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
55 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
56 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
57
58 //-----------------------------------------------------------------------------
59
60 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
61 address entry = __ pc();
62
63 #ifdef ASSERT
64 {
65 Label L;
66 __ lea(rax, Address(rbp,
67 frame::interpreter_frame_monitor_block_top_offset *
68 wordSize));
69 __ cmpptr(rax, rsp); // rax = maximal rsp for current rbp (stack
70 // grows negative)
71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
72 __ stop ("interpreter frame not set up");
73 __ bind(L);
74 }
75 #endif // ASSERT
76 // Restore bcp under the assumption that the current frame is still
77 // interpreted
78 __ restore_bcp();
79
80 // expression stack must be empty before entering the VM if an
81 // exception happened
82 __ empty_expression_stack();
83 // throw exception
84 __ call_VM(noreg,
85 CAST_FROM_FN_PTR(address,
86 InterpreterRuntime::throw_StackOverflowError));
87 return entry;
88 }
89
90 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
91 const char* name) {
92 address entry = __ pc();
93 // expression stack must be empty before entering the VM if an
94 // exception happened
95 __ empty_expression_stack();
96 // setup parameters
97 // ??? convention: expect aberrant index in register ebx
98 __ lea(c_rarg1, ExternalAddress((address)name));
99 __ call_VM(noreg,
100 CAST_FROM_FN_PTR(address,
101 InterpreterRuntime::
102 throw_ArrayIndexOutOfBoundsException),
103 c_rarg1, rbx);
104 return entry;
105 }
106
107 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
108 address entry = __ pc();
109
110 // object is at TOS
111 __ pop(c_rarg1);
112
113 // expression stack must be empty before entering the VM if an
114 // exception happened
115 __ empty_expression_stack();
116
117 __ call_VM(noreg,
118 CAST_FROM_FN_PTR(address,
119 InterpreterRuntime::
120 throw_ClassCastException),
121 c_rarg1);
122 return entry;
123 }
124
125 address TemplateInterpreterGenerator::generate_exception_handler_common(
126 const char* name, const char* message, bool pass_oop) {
127 assert(!pass_oop || message == NULL, "either oop or message but not both");
128 address entry = __ pc();
129 if (pass_oop) {
130 // object is at TOS
131 __ pop(c_rarg2);
132 }
133 // expression stack must be empty before entering the VM if an
134 // exception happened
135 __ empty_expression_stack();
136 // setup parameters
137 __ lea(c_rarg1, ExternalAddress((address)name));
138 if (pass_oop) {
139 __ call_VM(rax, CAST_FROM_FN_PTR(address,
140 InterpreterRuntime::
141 create_klass_exception),
142 c_rarg1, c_rarg2);
143 } else {
144 // kind of lame ExternalAddress can't take NULL because
145 // external_word_Relocation will assert.
146 if (message != NULL) {
147 __ lea(c_rarg2, ExternalAddress((address)message));
148 } else {
149 __ movptr(c_rarg2, NULL_WORD);
150 }
151 __ call_VM(rax,
152 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception),
153 c_rarg1, c_rarg2);
154 }
155 // throw exception
156 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
157 return entry;
158 }
159
160
161 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
162 address entry = __ pc();
163 // NULL last_sp until next java call
164 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
165 __ dispatch_next(state);
166 return entry;
167 }
168
169
170 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
171 address entry = __ pc();
172
173 // Restore stack bottom in case i2c adjusted stack
174 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
175 // and NULL it as marker that esp is now tos until next java call
176 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
177
178 __ restore_bcp();
179 __ restore_locals();
180
181 if (state == atos) {
182 Register mdp = rbx;
183 Register tmp = rcx;
184 __ profile_return_type(mdp, rax, tmp);
185 }
186
187 const Register cache = rbx;
188 const Register index = rcx;
189 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
190
191 const Register flags = cache;
192 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
193 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
194 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
195 __ dispatch_next(state, step);
196
197 return entry;
198 }
199
200
201 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
202 int step) {
203 address entry = __ pc();
204 // NULL last_sp until next java call
205 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
206 __ restore_bcp();
207 __ restore_locals();
208 // handle exceptions
209 {
210 Label L;
211 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
212 __ jcc(Assembler::zero, L);
213 __ call_VM(noreg,
214 CAST_FROM_FN_PTR(address,
215 InterpreterRuntime::throw_pending_exception));
216 __ should_not_reach_here();
217 __ bind(L);
218 }
219 __ dispatch_next(state, step);
220 return entry;
221 }
222
223 int AbstractInterpreter::BasicType_as_index(BasicType type) {
224 int i = 0;
225 switch (type) {
226 case T_BOOLEAN: i = 0; break;
227 case T_CHAR : i = 1; break;
228 case T_BYTE : i = 2; break;
229 case T_SHORT : i = 3; break;
230 case T_INT : i = 4; break;
231 case T_LONG : i = 5; break;
232 case T_VOID : i = 6; break;
233 case T_FLOAT : i = 7; break;
234 case T_DOUBLE : i = 8; break;
235 case T_OBJECT : i = 9; break;
236 case T_ARRAY : i = 9; break;
237 default : ShouldNotReachHere();
238 }
239 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
240 "index out of bounds");
241 return i;
242 }
243
244
245 address TemplateInterpreterGenerator::generate_result_handler_for(
246 BasicType type) {
247 address entry = __ pc();
248 switch (type) {
249 case T_BOOLEAN: __ c2bool(rax); break;
250 case T_CHAR : __ movzwl(rax, rax); break;
251 case T_BYTE : __ sign_extend_byte(rax); break;
252 case T_SHORT : __ sign_extend_short(rax); break;
253 case T_INT : /* nothing to do */ break;
254 case T_LONG : /* nothing to do */ break;
255 case T_VOID : /* nothing to do */ break;
256 case T_FLOAT : /* nothing to do */ break;
257 case T_DOUBLE : /* nothing to do */ break;
258 case T_OBJECT :
259 // retrieve result from frame
260 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
261 // and verify it
262 __ verify_oop(rax);
263 break;
264 default : ShouldNotReachHere();
265 }
266 __ ret(0); // return from result handler
267 return entry;
268 }
269
270 address TemplateInterpreterGenerator::generate_safept_entry_for(
271 TosState state,
272 address runtime_entry) {
273 address entry = __ pc();
274 __ push(state);
275 __ call_VM(noreg, runtime_entry);
276 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
277 return entry;
278 }
279
280
281
282 // Helpers for commoning out cases in the various type of method entries.
283 //
284
285
286 // increment invocation count & check for overflow
287 //
288 // Note: checking for negative value instead of overflow
289 // so we have a 'sticky' overflow test
290 //
291 // rbx: method
292 // ecx: invocation counter
293 //
294 void InterpreterGenerator::generate_counter_incr(
295 Label* overflow,
296 Label* profile_method,
297 Label* profile_method_continue) {
298 Label done;
299 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
300 if (TieredCompilation) {
301 int increment = InvocationCounter::count_increment;
302 Label no_mdo;
303 if (ProfileInterpreter) {
304 // Are we profiling?
305 __ movptr(rax, Address(rbx, Method::method_data_offset()));
306 __ testptr(rax, rax);
307 __ jccb(Assembler::zero, no_mdo);
308 // Increment counter in the MDO
309 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
310 in_bytes(InvocationCounter::counter_offset()));
311 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
312 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
313 __ jmp(done);
314 }
315 __ bind(no_mdo);
316 // Increment counter in MethodCounters
317 const Address invocation_counter(rax,
318 MethodCounters::invocation_counter_offset() +
319 InvocationCounter::counter_offset());
320 __ get_method_counters(rbx, rax, done);
321 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
322 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx,
323 false, Assembler::zero, overflow);
324 __ bind(done);
325 } else { // not TieredCompilation
326 const Address backedge_counter(rax,
327 MethodCounters::backedge_counter_offset() +
328 InvocationCounter::counter_offset());
329 const Address invocation_counter(rax,
330 MethodCounters::invocation_counter_offset() +
331 InvocationCounter::counter_offset());
332
333 __ get_method_counters(rbx, rax, done);
334
335 if (ProfileInterpreter) {
336 __ incrementl(Address(rax,
337 MethodCounters::interpreter_invocation_counter_offset()));
338 }
339 // Update standard invocation counters
340 __ movl(rcx, invocation_counter);
341 __ incrementl(rcx, InvocationCounter::count_increment);
342 __ movl(invocation_counter, rcx); // save invocation count
343
344 __ movl(rax, backedge_counter); // load backedge counter
345 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
346
347 __ addl(rcx, rax); // add both counters
348
349 // profile_method is non-null only for interpreted method so
350 // profile_method != NULL == !native_call
351
352 if (ProfileInterpreter && profile_method != NULL) {
353 // Test to see if we should create a method data oop
354 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
355 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
356 __ jcc(Assembler::less, *profile_method_continue);
357
358 // if no method data exists, go to profile_method
359 __ test_method_data_pointer(rax, *profile_method);
360 }
361
362 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
363 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
364 __ jcc(Assembler::aboveEqual, *overflow);
365 __ bind(done);
366 }
367 }
368
369 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
370
371 // Asm interpreter on entry
372 // r14 - locals
373 // r13 - bcp
374 // rbx - method
375 // edx - cpool --- DOES NOT APPEAR TO BE TRUE
376 // rbp - interpreter frame
377
378 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
379 // Everything as it was on entry
380 // rdx is not restored. Doesn't appear to really be set.
381
382 // InterpreterRuntime::frequency_counter_overflow takes two
383 // arguments, the first (thread) is passed by call_VM, the second
384 // indicates if the counter overflow occurs at a backwards branch
385 // (NULL bcp). We pass zero for it. The call returns the address
386 // of the verified entry point for the method or NULL if the
387 // compilation did not complete (either went background or bailed
388 // out).
389 __ movl(c_rarg1, 0);
390 __ call_VM(noreg,
391 CAST_FROM_FN_PTR(address,
392 InterpreterRuntime::frequency_counter_overflow),
393 c_rarg1);
394
395 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
396 // Preserve invariant that r13/r14 contain bcp/locals of sender frame
397 // and jump to the interpreted entry.
398 __ jmp(*do_continue, relocInfo::none);
399 }
400
401 // See if we've got enough room on the stack for locals plus overhead.
402 // The expression stack grows down incrementally, so the normal guard
403 // page mechanism will work for that.
404 //
405 // NOTE: Since the additional locals are also always pushed (wasn't
406 // obvious in generate_fixed_frame) so the guard should work for them
407 // too.
408 //
409 // Args:
410 // rdx: number of additional locals this frame needs (what we must check)
411 // rbx: Method*
412 //
413 // Kills:
414 // rax
415 void InterpreterGenerator::generate_stack_overflow_check(void) {
416
417 // monitor entry size: see picture of stack in frame_x86.hpp
418 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
419
420 // total overhead size: entry_size + (saved rbp through expr stack
421 // bottom). be sure to change this if you add/subtract anything
422 // to/from the overhead area
423 const int overhead_size =
424 -(frame::interpreter_frame_initial_sp_offset * wordSize) + entry_size;
425
426 const int page_size = os::vm_page_size();
427
428 Label after_frame_check;
429
430 // see if the frame is greater than one page in size. If so,
431 // then we need to verify there is enough stack space remaining
432 // for the additional locals.
433 __ cmpl(rdx, (page_size - overhead_size) / Interpreter::stackElementSize);
434 __ jcc(Assembler::belowEqual, after_frame_check);
435
436 // compute rsp as if this were going to be the last frame on
437 // the stack before the red zone
438
439 const Address stack_base(r15_thread, Thread::stack_base_offset());
440 const Address stack_size(r15_thread, Thread::stack_size_offset());
441
442 // locals + overhead, in bytes
443 __ mov(rax, rdx);
444 __ shlptr(rax, Interpreter::logStackElementSize); // 2 slots per parameter.
445 __ addptr(rax, overhead_size);
446
447 #ifdef ASSERT
448 Label stack_base_okay, stack_size_okay;
449 // verify that thread stack base is non-zero
450 __ cmpptr(stack_base, (int32_t)NULL_WORD);
451 __ jcc(Assembler::notEqual, stack_base_okay);
452 __ stop("stack base is zero");
453 __ bind(stack_base_okay);
454 // verify that thread stack size is non-zero
455 __ cmpptr(stack_size, 0);
456 __ jcc(Assembler::notEqual, stack_size_okay);
457 __ stop("stack size is zero");
458 __ bind(stack_size_okay);
459 #endif
460
461 // Add stack base to locals and subtract stack size
462 __ addptr(rax, stack_base);
463 __ subptr(rax, stack_size);
464
465 // Use the maximum number of pages we might bang.
466 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
467 (StackRedPages+StackYellowPages);
468
469 // add in the red and yellow zone sizes
470 __ addptr(rax, max_pages * page_size);
471
472 // check against the current stack bottom
473 __ cmpptr(rsp, rax);
474 __ jcc(Assembler::above, after_frame_check);
475
476 // Restore sender's sp as SP. This is necessary if the sender's
477 // frame is an extended compiled frame (see gen_c2i_adapter())
478 // and safer anyway in case of JSR292 adaptations.
479
480 __ pop(rax); // return address must be moved if SP is changed
481 __ mov(rsp, r13);
482 __ push(rax);
483
484 // Note: the restored frame is not necessarily interpreted.
485 // Use the shared runtime version of the StackOverflowError.
486 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
487 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
488
489 // all done with frame size check
490 __ bind(after_frame_check);
491 }
492
493 // Allocate monitor and lock method (asm interpreter)
494 //
495 // Args:
496 // rbx: Method*
497 // r14: locals
498 //
499 // Kills:
500 // rax
501 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ...(param regs)
502 // rscratch1, rscratch2 (scratch regs)
503 void InterpreterGenerator::lock_method(void) {
504 // synchronize method
505 const Address access_flags(rbx, Method::access_flags_offset());
506 const Address monitor_block_top(
507 rbp,
508 frame::interpreter_frame_monitor_block_top_offset * wordSize);
509 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
510
511 #ifdef ASSERT
512 {
513 Label L;
514 __ movl(rax, access_flags);
515 __ testl(rax, JVM_ACC_SYNCHRONIZED);
516 __ jcc(Assembler::notZero, L);
517 __ stop("method doesn't need synchronization");
518 __ bind(L);
519 }
520 #endif // ASSERT
521
522 // get synchronization object
523 {
524 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
525 Label done;
526 __ movl(rax, access_flags);
527 __ testl(rax, JVM_ACC_STATIC);
528 // get receiver (assume this is frequent case)
529 __ movptr(rax, Address(r14, Interpreter::local_offset_in_bytes(0)));
530 __ jcc(Assembler::zero, done);
531 __ movptr(rax, Address(rbx, Method::const_offset()));
532 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
533 __ movptr(rax, Address(rax,
534 ConstantPool::pool_holder_offset_in_bytes()));
535 __ movptr(rax, Address(rax, mirror_offset));
536
537 #ifdef ASSERT
538 {
539 Label L;
540 __ testptr(rax, rax);
541 __ jcc(Assembler::notZero, L);
542 __ stop("synchronization object is NULL");
543 __ bind(L);
544 }
545 #endif // ASSERT
546
547 __ bind(done);
548 }
549
550 // add space for monitor & lock
551 __ subptr(rsp, entry_size); // add space for a monitor entry
552 __ movptr(monitor_block_top, rsp); // set new monitor block top
553 // store object
554 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
555 __ movptr(c_rarg1, rsp); // object address
556 __ lock_object(c_rarg1);
557 }
558
559 // Generate a fixed interpreter frame. This is identical setup for
560 // interpreted methods and for native methods hence the shared code.
561 //
562 // Args:
563 // rax: return address
564 // rbx: Method*
565 // r14: pointer to locals
566 // r13: sender sp
567 // rdx: cp cache
568 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
569 // initialize fixed part of activation frame
570 __ push(rax); // save return address
571 __ enter(); // save old & set new rbp
572 __ push(r13); // set sender sp
573 __ push((int)NULL_WORD); // leave last_sp as null
574 __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod*
575 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
576 __ push(rbx); // save Method*
577 if (ProfileInterpreter) {
578 Label method_data_continue;
579 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
580 __ testptr(rdx, rdx);
581 __ jcc(Assembler::zero, method_data_continue);
582 __ addptr(rdx, in_bytes(MethodData::data_offset()));
583 __ bind(method_data_continue);
584 __ push(rdx); // set the mdp (method data pointer)
585 } else {
586 __ push(0);
587 }
588
589 __ movptr(rdx, Address(rbx, Method::const_offset()));
590 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
591 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
592 __ push(rdx); // set constant pool cache
593 __ push(r14); // set locals pointer
594 if (native_call) {
595 __ push(0); // no bcp
596 } else {
597 __ push(r13); // set bcp
598 }
599 __ push(0); // reserve word for pointer to expression stack bottom
600 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
601 }
602
603 // End of helpers
604
605 // Method entry for java.lang.ref.Reference.get.
606 address InterpreterGenerator::generate_Reference_get_entry(void) {
607 #if INCLUDE_ALL_GCS
608 // Code: _aload_0, _getfield, _areturn
609 // parameter size = 1
610 //
611 // The code that gets generated by this routine is split into 2 parts:
612 // 1. The "intrinsified" code for G1 (or any SATB based GC),
613 // 2. The slow path - which is an expansion of the regular method entry.
614 //
615 // Notes:-
616 // * In the G1 code we do not check whether we need to block for
617 // a safepoint. If G1 is enabled then we must execute the specialized
618 // code for Reference.get (except when the Reference object is null)
619 // so that we can log the value in the referent field with an SATB
620 // update buffer.
621 // If the code for the getfield template is modified so that the
622 // G1 pre-barrier code is executed when the current method is
623 // Reference.get() then going through the normal method entry
624 // will be fine.
625 // * The G1 code can, however, check the receiver object (the instance
626 // of java.lang.Reference) and jump to the slow path if null. If the
627 // Reference object is null then we obviously cannot fetch the referent
628 // and so we don't need to call the G1 pre-barrier. Thus we can use the
629 // regular method entry code to generate the NPE.
630 //
631 // rbx: Method*
632
633 // r13: senderSP must preserve for slow path, set SP to it on fast path
634
635 address entry = __ pc();
636
637 const int referent_offset = java_lang_ref_Reference::referent_offset;
638 guarantee(referent_offset > 0, "referent offset not initialized");
639
640 if (UseG1GC) {
641 Label slow_path;
642 // rbx: method
643
644 // Check if local 0 != NULL
645 // If the receiver is null then it is OK to jump to the slow path.
646 __ movptr(rax, Address(rsp, wordSize));
647
648 __ testptr(rax, rax);
649 __ jcc(Assembler::zero, slow_path);
650
651 // rax: local 0
652 // rbx: method (but can be used as scratch now)
653 // rdx: scratch
654 // rdi: scratch
655
656 // Generate the G1 pre-barrier code to log the value of
657 // the referent field in an SATB buffer.
658
659 // Load the value of the referent field.
660 const Address field_address(rax, referent_offset);
661 __ load_heap_oop(rax, field_address);
662
663 // Generate the G1 pre-barrier code to log the value of
664 // the referent field in an SATB buffer.
665 __ g1_write_barrier_pre(noreg /* obj */,
666 rax /* pre_val */,
667 r15_thread /* thread */,
668 rbx /* tmp */,
669 true /* tosca_live */,
670 true /* expand_call */);
671
672 // _areturn
673 __ pop(rdi); // get return address
674 __ mov(rsp, r13); // set sp to sender sp
675 __ jmp(rdi);
676 __ ret(0);
677
678 // generate a vanilla interpreter entry as the slow path
679 __ bind(slow_path);
680 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals));
681 return entry;
682 }
683 #endif // INCLUDE_ALL_GCS
684
685 // If G1 is not enabled then attempt to go through the accessor entry point
686 // Reference.get is an accessor
687 return NULL;
688 }
689
690 /**
691 * Method entry for static native methods:
692 * int java.util.zip.CRC32.update(int crc, int b)
693 */
694 address InterpreterGenerator::generate_CRC32_update_entry() {
695 if (UseCRC32Intrinsics) {
696 address entry = __ pc();
697
698 // rbx,: Method*
699 // r13: senderSP must preserved for slow path, set SP to it on fast path
700 // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
701 // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
702
703 Label slow_path;
704 // If we need a safepoint check, generate full interpreter entry.
705 ExternalAddress state(SafepointSynchronize::address_of_state());
706 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
707 SafepointSynchronize::_not_synchronized);
708 __ jcc(Assembler::notEqual, slow_path);
709
710 // We don't generate local frame and don't align stack because
711 // we call stub code and there is no safepoint on this path.
712
713 // Load parameters
714 const Register crc = rax; // crc
715 const Register val = c_rarg0; // source java byte value
716 const Register tbl = c_rarg1; // scratch
717
718 // Arguments are reversed on java expression stack
719 __ movl(val, Address(rsp, wordSize)); // byte value
720 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
721
722 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
723 __ notl(crc); // ~crc
724 __ update_byte_crc32(crc, val, tbl);
725 __ notl(crc); // ~crc
726 // result in rax
727
728 // _areturn
729 __ pop(rdi); // get return address
730 __ mov(rsp, r13); // set sp to sender sp
731 __ jmp(rdi);
732
733 // generate a vanilla native entry as the slow path
734 __ bind(slow_path);
735 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
736 return entry;
737 }
738 return NULL;
739 }
740
741 /**
742 * Method entry for static native methods:
743 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
744 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
745 */
746 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
747 if (UseCRC32Intrinsics) {
748 address entry = __ pc();
749
750 // rbx,: Method*
751 // r13: senderSP must preserved for slow path, set SP to it on fast path
752
753 Label slow_path;
754 // If we need a safepoint check, generate full interpreter entry.
755 ExternalAddress state(SafepointSynchronize::address_of_state());
756 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
757 SafepointSynchronize::_not_synchronized);
758 __ jcc(Assembler::notEqual, slow_path);
759
760 // We don't generate local frame and don't align stack because
761 // we call stub code and there is no safepoint on this path.
762
763 // Load parameters
764 const Register crc = c_rarg0; // crc
765 const Register buf = c_rarg1; // source java byte array address
766 const Register len = c_rarg2; // length
767 const Register off = len; // offset (never overlaps with 'len')
768
769 // Arguments are reversed on java expression stack
770 // Calculate address of start element
771 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
772 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
773 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
774 __ addq(buf, off); // + offset
775 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
776 } else {
777 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
778 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
779 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
780 __ addq(buf, off); // + offset
781 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
782 }
783 // Can now load 'len' since we're finished with 'off'
784 __ movl(len, Address(rsp, wordSize)); // Length
785
786 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
787 // result in rax
788
789 // _areturn
790 __ pop(rdi); // get return address
791 __ mov(rsp, r13); // set sp to sender sp
792 __ jmp(rdi);
793
794 // generate a vanilla native entry as the slow path
795 __ bind(slow_path);
796 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
797 return entry;
798 }
799 return NULL;
800 }
801
802 /**
803 * Method entry for static native methods:
804 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
805 * int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end)
806 */
807 address InterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
808 if (UseCRC32CIntrinsics) {
809 address entry = __ pc();
810 // Load parameters
811 const Register crc = c_rarg0; // crc
812 const Register buf = c_rarg1; // source java byte array address
813 const Register len = c_rarg2;
814 const Register off = c_rarg3; // offset
815 const Register end = len;
816
817 // Arguments are reversed on java expression stack
818 // Calculate address of start element
819 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) {
820 __ movptr(buf, Address(rsp, 3 * wordSize)); // long buf
821 __ movl2ptr(off, Address(rsp, 2 * wordSize)); // offset
822 __ addq(buf, off); // + offset
823 __ movl(crc, Address(rsp, 5 * wordSize)); // Initial CRC
824 // Note on 5 * wordSize vs. 4 * wordSize:
825 // * int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end)
826 // 4 2,3 1 0
827 // end starts at SP + 8
828 // The Java(R) Virtual Machine Specification Java SE 7 Edition
829 // 4.10.2.3. Values of Types long and double
830 // "When calculating operand stack length, values of type long and double have length two."
831 } else {
832 __ movptr(buf, Address(rsp, 3 * wordSize)); // byte[] array
833 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
834 __ movl2ptr(off, Address(rsp, 2 * wordSize)); // offset
835 __ addq(buf, off); // + offset
836 __ movl(crc, Address(rsp, 4 * wordSize)); // Initial CRC
837 }
838 __ movl(end, Address(rsp, wordSize)); // end
839 __ subl(end, off); // end - off
840 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()), crc, buf, len);
841 // result in rax
842 // _areturn
843 __ pop(rdi); // get return address
844 __ mov(rsp, r13); // set sp to sender sp
845 __ jmp(rdi);
846
847 return entry;
848 }
849
850 return NULL;
851 }
852
853 // Interpreter stub for calling a native method. (asm interpreter)
854 // This sets up a somewhat different looking stack for calling the
855 // native method than the typical interpreter frame setup.
856 address InterpreterGenerator::generate_native_entry(bool synchronized) {
857 // determine code generation flags
858 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
859
860 // rbx: Method*
861 // r13: sender sp
862
863 address entry_point = __ pc();
864
865 const Address constMethod (rbx, Method::const_offset());
866 const Address access_flags (rbx, Method::access_flags_offset());
867 const Address size_of_parameters(rcx, ConstMethod::
868 size_of_parameters_offset());
869
870
871 // get parameter size (always needed)
872 __ movptr(rcx, constMethod);
873 __ load_unsigned_short(rcx, size_of_parameters);
874
875 // native calls don't need the stack size check since they have no
876 // expression stack and the arguments are already on the stack and
877 // we only add a handful of words to the stack
878
879 // rbx: Method*
880 // rcx: size of parameters
881 // r13: sender sp
882 __ pop(rax); // get return address
883
884 // for natives the size of locals is zero
885
886 // compute beginning of parameters (r14)
887 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
888
889 // add 2 zero-initialized slots for native calls
890 // initialize result_handler slot
891 __ push((int) NULL_WORD);
892 // slot for oop temp
893 // (static native method holder mirror/jni oop result)
894 __ push((int) NULL_WORD);
895
896 // initialize fixed part of activation frame
897 generate_fixed_frame(true);
898
899 // make sure method is native & not abstract
900 #ifdef ASSERT
901 __ movl(rax, access_flags);
902 {
903 Label L;
904 __ testl(rax, JVM_ACC_NATIVE);
905 __ jcc(Assembler::notZero, L);
906 __ stop("tried to execute non-native method as native");
907 __ bind(L);
908 }
909 {
910 Label L;
911 __ testl(rax, JVM_ACC_ABSTRACT);
912 __ jcc(Assembler::zero, L);
913 __ stop("tried to execute abstract method in interpreter");
914 __ bind(L);
915 }
916 #endif
917
918 // Since at this point in the method invocation the exception handler
919 // would try to exit the monitor of synchronized methods which hasn't
920 // been entered yet, we set the thread local variable
921 // _do_not_unlock_if_synchronized to true. The remove_activation will
922 // check this flag.
923
924 const Address do_not_unlock_if_synchronized(r15_thread,
925 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
926 __ movbool(do_not_unlock_if_synchronized, true);
927
928 // increment invocation count & check for overflow
929 Label invocation_counter_overflow;
930 if (inc_counter) {
931 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
932 }
933
934 Label continue_after_compile;
935 __ bind(continue_after_compile);
936
937 bang_stack_shadow_pages(true);
938
939 // reset the _do_not_unlock_if_synchronized flag
940 __ movbool(do_not_unlock_if_synchronized, false);
941
942 // check for synchronized methods
943 // Must happen AFTER invocation_counter check and stack overflow check,
944 // so method is not locked if overflows.
945 if (synchronized) {
946 lock_method();
947 } else {
948 // no synchronization necessary
949 #ifdef ASSERT
950 {
951 Label L;
952 __ movl(rax, access_flags);
953 __ testl(rax, JVM_ACC_SYNCHRONIZED);
954 __ jcc(Assembler::zero, L);
955 __ stop("method needs synchronization");
956 __ bind(L);
957 }
958 #endif
959 }
960
961 // start execution
962 #ifdef ASSERT
963 {
964 Label L;
965 const Address monitor_block_top(rbp,
966 frame::interpreter_frame_monitor_block_top_offset * wordSize);
967 __ movptr(rax, monitor_block_top);
968 __ cmpptr(rax, rsp);
969 __ jcc(Assembler::equal, L);
970 __ stop("broken stack frame setup in interpreter");
971 __ bind(L);
972 }
973 #endif
974
975 // jvmti support
976 __ notify_method_entry();
977
978 // work registers
979 const Register method = rbx;
980 const Register t = r11;
981
982 // allocate space for parameters
983 __ get_method(method);
984 __ movptr(t, Address(method, Method::const_offset()));
985 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
986 __ shll(t, Interpreter::logStackElementSize);
987
988 __ subptr(rsp, t);
989 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
990 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
991
992 // get signature handler
993 {
994 Label L;
995 __ movptr(t, Address(method, Method::signature_handler_offset()));
996 __ testptr(t, t);
997 __ jcc(Assembler::notZero, L);
998 __ call_VM(noreg,
999 CAST_FROM_FN_PTR(address,
1000 InterpreterRuntime::prepare_native_call),
1001 method);
1002 __ get_method(method);
1003 __ movptr(t, Address(method, Method::signature_handler_offset()));
1004 __ bind(L);
1005 }
1006
1007 // call signature handler
1008 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
1009 "adjust this code");
1010 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
1011 "adjust this code");
1012 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
1013 "adjust this code");
1014
1015 // The generated handlers do not touch RBX (the method oop).
1016 // However, large signatures cannot be cached and are generated
1017 // each time here. The slow-path generator can do a GC on return,
1018 // so we must reload it after the call.
1019 __ call(t);
1020 __ get_method(method); // slow path can do a GC, reload RBX
1021
1022
1023 // result handler is in rax
1024 // set result handler
1025 __ movptr(Address(rbp,
1026 (frame::interpreter_frame_result_handler_offset) * wordSize),
1027 rax);
1028
1029 // pass mirror handle if static call
1030 {
1031 Label L;
1032 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1033 __ movl(t, Address(method, Method::access_flags_offset()));
1034 __ testl(t, JVM_ACC_STATIC);
1035 __ jcc(Assembler::zero, L);
1036 // get mirror
1037 __ movptr(t, Address(method, Method::const_offset()));
1038 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1039 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1040 __ movptr(t, Address(t, mirror_offset));
1041 // copy mirror into activation frame
1042 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1043 t);
1044 // pass handle to mirror
1045 __ lea(c_rarg1,
1046 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1047 __ bind(L);
1048 }
1049
1050 // get native function entry point
1051 {
1052 Label L;
1053 __ movptr(rax, Address(method, Method::native_function_offset()));
1054 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1055 __ movptr(rscratch2, unsatisfied.addr());
1056 __ cmpptr(rax, rscratch2);
1057 __ jcc(Assembler::notEqual, L);
1058 __ call_VM(noreg,
1059 CAST_FROM_FN_PTR(address,
1060 InterpreterRuntime::prepare_native_call),
1061 method);
1062 __ get_method(method);
1063 __ movptr(rax, Address(method, Method::native_function_offset()));
1064 __ bind(L);
1065 }
1066
1067 // pass JNIEnv
1068 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1069
1070 // It is enough that the pc() points into the right code
1071 // segment. It does not have to be the correct return pc.
1072 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1073
1074 // change thread state
1075 #ifdef ASSERT
1076 {
1077 Label L;
1078 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1079 __ cmpl(t, _thread_in_Java);
1080 __ jcc(Assembler::equal, L);
1081 __ stop("Wrong thread state in native stub");
1082 __ bind(L);
1083 }
1084 #endif
1085
1086 // Change state to native
1087
1088 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1089 _thread_in_native);
1090
1091 // Call the native method.
1092 __ call(rax);
1093 // result potentially in rax or xmm0
1094
1095 // Verify or restore cpu control state after JNI call
1096 __ restore_cpu_control_state_after_jni();
1097
1098 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1099 // in order to extract the result of a method call. If the order of these
1100 // pushes change or anything else is added to the stack then the code in
1101 // interpreter_frame_result must also change.
1102
1103 __ push(dtos);
1104 __ push(ltos);
1105
1106 // change thread state
1107 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1108 _thread_in_native_trans);
1109
1110 if (os::is_MP()) {
1111 if (UseMembar) {
1112 // Force this write out before the read below
1113 __ membar(Assembler::Membar_mask_bits(
1114 Assembler::LoadLoad | Assembler::LoadStore |
1115 Assembler::StoreLoad | Assembler::StoreStore));
1116 } else {
1117 // Write serialization page so VM thread can do a pseudo remote membar.
1118 // We use the current thread pointer to calculate a thread specific
1119 // offset to write to within the page. This minimizes bus traffic
1120 // due to cache line collision.
1121 __ serialize_memory(r15_thread, rscratch2);
1122 }
1123 }
1124
1125 // check for safepoint operation in progress and/or pending suspend requests
1126 {
1127 Label Continue;
1128 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1129 SafepointSynchronize::_not_synchronized);
1130
1131 Label L;
1132 __ jcc(Assembler::notEqual, L);
1133 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1134 __ jcc(Assembler::equal, Continue);
1135 __ bind(L);
1136
1137 // Don't use call_VM as it will see a possible pending exception
1138 // and forward it and never return here preventing us from
1139 // clearing _last_native_pc down below. Also can't use
1140 // call_VM_leaf either as it will check to see if r13 & r14 are
1141 // preserved and correspond to the bcp/locals pointers. So we do a
1142 // runtime call by hand.
1143 //
1144 __ mov(c_rarg0, r15_thread);
1145 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1146 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1147 __ andptr(rsp, -16); // align stack as required by ABI
1148 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1149 __ mov(rsp, r12); // restore sp
1150 __ reinit_heapbase();
1151 __ bind(Continue);
1152 }
1153
1154 // change thread state
1155 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1156
1157 // reset_last_Java_frame
1158 __ reset_last_Java_frame(true, true);
1159
1160 // reset handle block
1161 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1162 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1163
1164 // If result is an oop unbox and store it in frame where gc will see it
1165 // and result handler will pick it up
1166
1167 {
1168 Label no_oop, store_result;
1169 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1170 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1171 __ jcc(Assembler::notEqual, no_oop);
1172 // retrieve result
1173 __ pop(ltos);
1174 __ testptr(rax, rax);
1175 __ jcc(Assembler::zero, store_result);
1176 __ movptr(rax, Address(rax, 0));
1177 __ bind(store_result);
1178 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1179 // keep stack depth as expected by pushing oop which will eventually be discarde
1180 __ push(ltos);
1181 __ bind(no_oop);
1182 }
1183
1184
1185 {
1186 Label no_reguard;
1187 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1188 JavaThread::stack_guard_yellow_disabled);
1189 __ jcc(Assembler::notEqual, no_reguard);
1190
1191 __ pusha(); // XXX only save smashed registers
1192 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1193 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1194 __ andptr(rsp, -16); // align stack as required by ABI
1195 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1196 __ mov(rsp, r12); // restore sp
1197 __ popa(); // XXX only restore smashed registers
1198 __ reinit_heapbase();
1199
1200 __ bind(no_reguard);
1201 }
1202
1203
1204 // The method register is junk from after the thread_in_native transition
1205 // until here. Also can't call_VM until the bcp has been
1206 // restored. Need bcp for throwing exception below so get it now.
1207 __ get_method(method);
1208
1209 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1210 // r13 == code_base()
1211 __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod*
1212 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
1213 // handle exceptions (exception handling will handle unlocking!)
1214 {
1215 Label L;
1216 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1217 __ jcc(Assembler::zero, L);
1218 // Note: At some point we may want to unify this with the code
1219 // used in call_VM_base(); i.e., we should use the
1220 // StubRoutines::forward_exception code. For now this doesn't work
1221 // here because the rsp is not correctly set at this point.
1222 __ MacroAssembler::call_VM(noreg,
1223 CAST_FROM_FN_PTR(address,
1224 InterpreterRuntime::throw_pending_exception));
1225 __ should_not_reach_here();
1226 __ bind(L);
1227 }
1228
1229 // do unlocking if necessary
1230 {
1231 Label L;
1232 __ movl(t, Address(method, Method::access_flags_offset()));
1233 __ testl(t, JVM_ACC_SYNCHRONIZED);
1234 __ jcc(Assembler::zero, L);
1235 // the code below should be shared with interpreter macro
1236 // assembler implementation
1237 {
1238 Label unlock;
1239 // BasicObjectLock will be first in list, since this is a
1240 // synchronized method. However, need to check that the object
1241 // has not been unlocked by an explicit monitorexit bytecode.
1242 const Address monitor(rbp,
1243 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1244 wordSize - sizeof(BasicObjectLock)));
1245
1246 // monitor expect in c_rarg1 for slow unlock path
1247 __ lea(c_rarg1, monitor); // address of first monitor
1248
1249 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1250 __ testptr(t, t);
1251 __ jcc(Assembler::notZero, unlock);
1252
1253 // Entry already unlocked, need to throw exception
1254 __ MacroAssembler::call_VM(noreg,
1255 CAST_FROM_FN_PTR(address,
1256 InterpreterRuntime::throw_illegal_monitor_state_exception));
1257 __ should_not_reach_here();
1258
1259 __ bind(unlock);
1260 __ unlock_object(c_rarg1);
1261 }
1262 __ bind(L);
1263 }
1264
1265 // jvmti support
1266 // Note: This must happen _after_ handling/throwing any exceptions since
1267 // the exception handler code notifies the runtime of method exits
1268 // too. If this happens before, method entry/exit notifications are
1269 // not properly paired (was bug - gri 11/22/99).
1270 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1271
1272 // restore potential result in edx:eax, call result handler to
1273 // restore potential result in ST0 & handle result
1274
1275 __ pop(ltos);
1276 __ pop(dtos);
1277
1278 __ movptr(t, Address(rbp,
1279 (frame::interpreter_frame_result_handler_offset) * wordSize));
1280 __ call(t);
1281
1282 // remove activation
1283 __ movptr(t, Address(rbp,
1284 frame::interpreter_frame_sender_sp_offset *
1285 wordSize)); // get sender sp
1286 __ leave(); // remove frame anchor
1287 __ pop(rdi); // get return address
1288 __ mov(rsp, t); // set sp to sender sp
1289 __ jmp(rdi);
1290
1291 if (inc_counter) {
1292 // Handle overflow of counter and compile method
1293 __ bind(invocation_counter_overflow);
1294 generate_counter_overflow(&continue_after_compile);
1295 }
1296
1297 return entry_point;
1298 }
1299
1300 //
1301 // Generic interpreted method entry to (asm) interpreter
1302 //
1303 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1304 // determine code generation flags
1305 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1306
1307 // ebx: Method*
1308 // r13: sender sp
1309 address entry_point = __ pc();
1310
1311 const Address constMethod(rbx, Method::const_offset());
1312 const Address access_flags(rbx, Method::access_flags_offset());
1313 const Address size_of_parameters(rdx,
1314 ConstMethod::size_of_parameters_offset());
1315 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1316
1317
1318 // get parameter size (always needed)
1319 __ movptr(rdx, constMethod);
1320 __ load_unsigned_short(rcx, size_of_parameters);
1321
1322 // rbx: Method*
1323 // rcx: size of parameters
1324 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1325
1326 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1327 __ subl(rdx, rcx); // rdx = no. of additional locals
1328
1329 // YYY
1330 // __ incrementl(rdx);
1331 // __ andl(rdx, -2);
1332
1333 // see if we've got enough room on the stack for locals plus overhead.
1334 generate_stack_overflow_check();
1335
1336 // get return address
1337 __ pop(rax);
1338
1339 // compute beginning of parameters (r14)
1340 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1341
1342 // rdx - # of additional locals
1343 // allocate space for locals
1344 // explicitly initialize locals
1345 {
1346 Label exit, loop;
1347 __ testl(rdx, rdx);
1348 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1349 __ bind(loop);
1350 __ push((int) NULL_WORD); // initialize local variables
1351 __ decrementl(rdx); // until everything initialized
1352 __ jcc(Assembler::greater, loop);
1353 __ bind(exit);
1354 }
1355
1356 // initialize fixed part of activation frame
1357 generate_fixed_frame(false);
1358
1359 // make sure method is not native & not abstract
1360 #ifdef ASSERT
1361 __ movl(rax, access_flags);
1362 {
1363 Label L;
1364 __ testl(rax, JVM_ACC_NATIVE);
1365 __ jcc(Assembler::zero, L);
1366 __ stop("tried to execute native method as non-native");
1367 __ bind(L);
1368 }
1369 {
1370 Label L;
1371 __ testl(rax, JVM_ACC_ABSTRACT);
1372 __ jcc(Assembler::zero, L);
1373 __ stop("tried to execute abstract method in interpreter");
1374 __ bind(L);
1375 }
1376 #endif
1377
1378 // Since at this point in the method invocation the exception
1379 // handler would try to exit the monitor of synchronized methods
1380 // which hasn't been entered yet, we set the thread local variable
1381 // _do_not_unlock_if_synchronized to true. The remove_activation
1382 // will check this flag.
1383
1384 const Address do_not_unlock_if_synchronized(r15_thread,
1385 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1386 __ movbool(do_not_unlock_if_synchronized, true);
1387
1388 __ profile_parameters_type(rax, rcx, rdx);
1389 // increment invocation count & check for overflow
1390 Label invocation_counter_overflow;
1391 Label profile_method;
1392 Label profile_method_continue;
1393 if (inc_counter) {
1394 generate_counter_incr(&invocation_counter_overflow,
1395 &profile_method,
1396 &profile_method_continue);
1397 if (ProfileInterpreter) {
1398 __ bind(profile_method_continue);
1399 }
1400 }
1401
1402 Label continue_after_compile;
1403 __ bind(continue_after_compile);
1404
1405 // check for synchronized interpreted methods
1406 bang_stack_shadow_pages(false);
1407
1408 // reset the _do_not_unlock_if_synchronized flag
1409 __ movbool(do_not_unlock_if_synchronized, false);
1410
1411 // check for synchronized methods
1412 // Must happen AFTER invocation_counter check and stack overflow check,
1413 // so method is not locked if overflows.
1414 if (synchronized) {
1415 // Allocate monitor and lock method
1416 lock_method();
1417 } else {
1418 // no synchronization necessary
1419 #ifdef ASSERT
1420 {
1421 Label L;
1422 __ movl(rax, access_flags);
1423 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1424 __ jcc(Assembler::zero, L);
1425 __ stop("method needs synchronization");
1426 __ bind(L);
1427 }
1428 #endif
1429 }
1430
1431 // start execution
1432 #ifdef ASSERT
1433 {
1434 Label L;
1435 const Address monitor_block_top (rbp,
1436 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1437 __ movptr(rax, monitor_block_top);
1438 __ cmpptr(rax, rsp);
1439 __ jcc(Assembler::equal, L);
1440 __ stop("broken stack frame setup in interpreter");
1441 __ bind(L);
1442 }
1443 #endif
1444
1445 // jvmti support
1446 __ notify_method_entry();
1447
1448 __ dispatch_next(vtos);
1449
1450 // invocation counter overflow
1451 if (inc_counter) {
1452 if (ProfileInterpreter) {
1453 // We have decided to profile this method in the interpreter
1454 __ bind(profile_method);
1455 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1456 __ set_method_data_pointer_for_bcp();
1457 __ get_method(rbx);
1458 __ jmp(profile_method_continue);
1459 }
1460 // Handle overflow of counter and compile method
1461 __ bind(invocation_counter_overflow);
1462 generate_counter_overflow(&continue_after_compile);
1463 }
1464
1465 return entry_point;
1466 }
1467
1468
1469 // These should never be compiled since the interpreter will prefer
1470 // the compiled version to the intrinsic version.
1471 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1472 switch (method_kind(m)) {
1473 case Interpreter::java_lang_math_sin : // fall thru
1474 case Interpreter::java_lang_math_cos : // fall thru
1475 case Interpreter::java_lang_math_tan : // fall thru
1476 case Interpreter::java_lang_math_abs : // fall thru
1477 case Interpreter::java_lang_math_log : // fall thru
1478 case Interpreter::java_lang_math_log10 : // fall thru
1479 case Interpreter::java_lang_math_sqrt : // fall thru
1480 case Interpreter::java_lang_math_pow : // fall thru
1481 case Interpreter::java_lang_math_exp :
1482 return false;
1483 default:
1484 return true;
1485 }
1486 }
1487
1488 // How much stack a method activation needs in words.
1489 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1490 const int entry_size = frame::interpreter_frame_monitor_size();
1491
1492 // total overhead size: entry_size + (saved rbp thru expr stack
1493 // bottom). be sure to change this if you add/subtract anything
1494 // to/from the overhead area
1495 const int overhead_size =
1496 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1497
1498 const int stub_code = frame::entry_frame_after_call_words;
1499 const int method_stack = (method->max_locals() + method->max_stack()) *
1500 Interpreter::stackElementWords;
1501 return (overhead_size + method_stack + stub_code);
1502 }
1503
1504 //-----------------------------------------------------------------------------
1505 // Exceptions
1506
1507 void TemplateInterpreterGenerator::generate_throw_exception() {
1508 // Entry point in previous activation (i.e., if the caller was
1509 // interpreted)
1510 Interpreter::_rethrow_exception_entry = __ pc();
1511 // Restore sp to interpreter_frame_last_sp even though we are going
1512 // to empty the expression stack for the exception processing.
1513 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1514 // rax: exception
1515 // rdx: return address/pc that threw exception
1516 __ restore_bcp(); // r13 points to call/send
1517 __ restore_locals();
1518 __ reinit_heapbase(); // restore r12 as heapbase.
1519 // Entry point for exceptions thrown within interpreter code
1520 Interpreter::_throw_exception_entry = __ pc();
1521 // expression stack is undefined here
1522 // rax: exception
1523 // r13: exception bcp
1524 __ verify_oop(rax);
1525 __ mov(c_rarg1, rax);
1526
1527 // expression stack must be empty before entering the VM in case of
1528 // an exception
1529 __ empty_expression_stack();
1530 // find exception handler address and preserve exception oop
1531 __ call_VM(rdx,
1532 CAST_FROM_FN_PTR(address,
1533 InterpreterRuntime::exception_handler_for_exception),
1534 c_rarg1);
1535 // rax: exception handler entry point
1536 // rdx: preserved exception oop
1537 // r13: bcp for exception handler
1538 __ push_ptr(rdx); // push exception which is now the only value on the stack
1539 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1540
1541 // If the exception is not handled in the current frame the frame is
1542 // removed and the exception is rethrown (i.e. exception
1543 // continuation is _rethrow_exception).
1544 //
1545 // Note: At this point the bci is still the bxi for the instruction
1546 // which caused the exception and the expression stack is
1547 // empty. Thus, for any VM calls at this point, GC will find a legal
1548 // oop map (with empty expression stack).
1549
1550 // In current activation
1551 // tos: exception
1552 // esi: exception bcp
1553
1554 //
1555 // JVMTI PopFrame support
1556 //
1557
1558 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1559 __ empty_expression_stack();
1560 // Set the popframe_processing bit in pending_popframe_condition
1561 // indicating that we are currently handling popframe, so that
1562 // call_VMs that may happen later do not trigger new popframe
1563 // handling cycles.
1564 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1565 __ orl(rdx, JavaThread::popframe_processing_bit);
1566 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1567
1568 {
1569 // Check to see whether we are returning to a deoptimized frame.
1570 // (The PopFrame call ensures that the caller of the popped frame is
1571 // either interpreted or compiled and deoptimizes it if compiled.)
1572 // In this case, we can't call dispatch_next() after the frame is
1573 // popped, but instead must save the incoming arguments and restore
1574 // them after deoptimization has occurred.
1575 //
1576 // Note that we don't compare the return PC against the
1577 // deoptimization blob's unpack entry because of the presence of
1578 // adapter frames in C2.
1579 Label caller_not_deoptimized;
1580 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1581 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1582 InterpreterRuntime::interpreter_contains), c_rarg1);
1583 __ testl(rax, rax);
1584 __ jcc(Assembler::notZero, caller_not_deoptimized);
1585
1586 // Compute size of arguments for saving when returning to
1587 // deoptimized caller
1588 __ get_method(rax);
1589 __ movptr(rax, Address(rax, Method::const_offset()));
1590 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1591 size_of_parameters_offset())));
1592 __ shll(rax, Interpreter::logStackElementSize);
1593 __ restore_locals(); // XXX do we need this?
1594 __ subptr(r14, rax);
1595 __ addptr(r14, wordSize);
1596 // Save these arguments
1597 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1598 Deoptimization::
1599 popframe_preserve_args),
1600 r15_thread, rax, r14);
1601
1602 __ remove_activation(vtos, rdx,
1603 /* throw_monitor_exception */ false,
1604 /* install_monitor_exception */ false,
1605 /* notify_jvmdi */ false);
1606
1607 // Inform deoptimization that it is responsible for restoring
1608 // these arguments
1609 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1610 JavaThread::popframe_force_deopt_reexecution_bit);
1611
1612 // Continue in deoptimization handler
1613 __ jmp(rdx);
1614
1615 __ bind(caller_not_deoptimized);
1616 }
1617
1618 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1619 /* throw_monitor_exception */ false,
1620 /* install_monitor_exception */ false,
1621 /* notify_jvmdi */ false);
1622
1623 // Finish with popframe handling
1624 // A previous I2C followed by a deoptimization might have moved the
1625 // outgoing arguments further up the stack. PopFrame expects the
1626 // mutations to those outgoing arguments to be preserved and other
1627 // constraints basically require this frame to look exactly as
1628 // though it had previously invoked an interpreted activation with
1629 // no space between the top of the expression stack (current
1630 // last_sp) and the top of stack. Rather than force deopt to
1631 // maintain this kind of invariant all the time we call a small
1632 // fixup routine to move the mutated arguments onto the top of our
1633 // expression stack if necessary.
1634 __ mov(c_rarg1, rsp);
1635 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1636 // PC must point into interpreter here
1637 __ set_last_Java_frame(noreg, rbp, __ pc());
1638 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1639 __ reset_last_Java_frame(true, true);
1640 // Restore the last_sp and null it out
1641 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1642 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1643
1644 __ restore_bcp(); // XXX do we need this?
1645 __ restore_locals(); // XXX do we need this?
1646 // The method data pointer was incremented already during
1647 // call profiling. We have to restore the mdp for the current bcp.
1648 if (ProfileInterpreter) {
1649 __ set_method_data_pointer_for_bcp();
1650 }
1651
1652 // Clear the popframe condition flag
1653 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1654 JavaThread::popframe_inactive);
1655
1656 #if INCLUDE_JVMTI
1657 {
1658 Label L_done;
1659 const Register local0 = r14;
1660
1661 __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1662 __ jcc(Assembler::notEqual, L_done);
1663
1664 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1665 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1666
1667 __ get_method(rdx);
1668 __ movptr(rax, Address(local0, 0));
1669 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1670
1671 __ testptr(rax, rax);
1672 __ jcc(Assembler::zero, L_done);
1673
1674 __ movptr(Address(rbx, 0), rax);
1675 __ bind(L_done);
1676 }
1677 #endif // INCLUDE_JVMTI
1678
1679 __ dispatch_next(vtos);
1680 // end of PopFrame support
1681
1682 Interpreter::_remove_activation_entry = __ pc();
1683
1684 // preserve exception over this code sequence
1685 __ pop_ptr(rax);
1686 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1687 // remove the activation (without doing throws on illegalMonitorExceptions)
1688 __ remove_activation(vtos, rdx, false, true, false);
1689 // restore exception
1690 __ get_vm_result(rax, r15_thread);
1691
1692 // In between activations - previous activation type unknown yet
1693 // compute continuation point - the continuation point expects the
1694 // following registers set up:
1695 //
1696 // rax: exception
1697 // rdx: return address/pc that threw exception
1698 // rsp: expression stack of caller
1699 // rbp: ebp of caller
1700 __ push(rax); // save exception
1701 __ push(rdx); // save return address
1702 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1703 SharedRuntime::exception_handler_for_return_address),
1704 r15_thread, rdx);
1705 __ mov(rbx, rax); // save exception handler
1706 __ pop(rdx); // restore return address
1707 __ pop(rax); // restore exception
1708 // Note that an "issuing PC" is actually the next PC after the call
1709 __ jmp(rbx); // jump to exception
1710 // handler of caller
1711 }
1712
1713
1714 //
1715 // JVMTI ForceEarlyReturn support
1716 //
1717 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1718 address entry = __ pc();
1719
1720 __ restore_bcp();
1721 __ restore_locals();
1722 __ empty_expression_stack();
1723 __ load_earlyret_value(state);
1724
1725 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1726 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1727
1728 // Clear the earlyret state
1729 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1730
1731 __ remove_activation(state, rsi,
1732 false, /* throw_monitor_exception */
1733 false, /* install_monitor_exception */
1734 true); /* notify_jvmdi */
1735 __ jmp(rsi);
1736
1737 return entry;
1738 } // end of ForceEarlyReturn support
1739
1740
1741 //-----------------------------------------------------------------------------
1742 // Helper for vtos entry point generation
1743
1744 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1745 address& bep,
1746 address& cep,
1747 address& sep,
1748 address& aep,
1749 address& iep,
1750 address& lep,
1751 address& fep,
1752 address& dep,
1753 address& vep) {
1754 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1755 Label L;
1756 aep = __ pc(); __ push_ptr(); __ jmp(L);
1757 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1758 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1759 lep = __ pc(); __ push_l(); __ jmp(L);
1760 bep = cep = sep =
1761 iep = __ pc(); __ push_i();
1762 vep = __ pc();
1763 __ bind(L);
1764 generate_and_dispatch(t);
1765 }
1766
1767
1768 //-----------------------------------------------------------------------------
1769 // Generation of individual instructions
1770
1771 // helpers for generate_and_dispatch
1772
1773
1774 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1775 : TemplateInterpreterGenerator(code) {
1776 generate_all(); // down here so it can be "virtual"
1777 }
1778
1779 //-----------------------------------------------------------------------------
1780
1781 // Non-product code
1782 #ifndef PRODUCT
1783 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1784 address entry = __ pc();
1785
1786 __ push(state);
1787 __ push(c_rarg0);
1788 __ push(c_rarg1);
1789 __ push(c_rarg2);
1790 __ push(c_rarg3);
1791 __ mov(c_rarg2, rax); // Pass itos
1792 #ifdef _WIN64
1793 __ movflt(xmm3, xmm0); // Pass ftos
1794 #endif
1795 __ call_VM(noreg,
1796 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1797 c_rarg1, c_rarg2, c_rarg3);
1798 __ pop(c_rarg3);
1799 __ pop(c_rarg2);
1800 __ pop(c_rarg1);
1801 __ pop(c_rarg0);
1802 __ pop(state);
1803 __ ret(0); // return from result handler
1804
1805 return entry;
1806 }
1807
1808 void TemplateInterpreterGenerator::count_bytecode() {
1809 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1810 }
1811
1812 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1813 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1814 }
1815
1816 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1817 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1818 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1819 __ orl(rbx,
1820 ((int) t->bytecode()) <<
1821 BytecodePairHistogram::log2_number_of_codes);
1822 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1823 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1824 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1825 }
1826
1827
1828 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1829 // Call a little run-time stub to avoid blow-up for each bytecode.
1830 // The run-time runtime saves the right registers, depending on
1831 // the tosca in-state for the given template.
1832
1833 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1834 "entry must have been generated");
1835 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1836 __ andptr(rsp, -16); // align stack as required by ABI
1837 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1838 __ mov(rsp, r12); // restore sp
1839 __ reinit_heapbase();
1840 }
1841
1842
1843 void TemplateInterpreterGenerator::stop_interpreter_at() {
1844 Label L;
1845 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1846 StopInterpreterAt);
1847 __ jcc(Assembler::notEqual, L);
1848 __ int3();
1849 __ bind(L);
1850 }
1851 #endif // !PRODUCT
1852 #endif // ! CC_INTERP