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 oopDesc::bs()->interpreter_read_barrier(_masm, rax);
549 }
550
551 // add space for monitor & lock
552 __ subptr(rsp, entry_size); // add space for a monitor entry
553 __ movptr(monitor_block_top, rsp); // set new monitor block top
554 // store object
555 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax);
556 __ movptr(c_rarg1, rsp); // object address
557 __ lock_object(c_rarg1);
558 }
559
560 // Generate a fixed interpreter frame. This is identical setup for
561 // interpreted methods and for native methods hence the shared code.
562 //
563 // Args:
564 // rax: return address
565 // rbx: Method*
566 // r14: pointer to locals
567 // r13: sender sp
568 // rdx: cp cache
569 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
570 // initialize fixed part of activation frame
571 __ push(rax); // save return address
572 __ enter(); // save old & set new rbp
573 __ push(r13); // set sender sp
574 __ push((int)NULL_WORD); // leave last_sp as null
575 __ movptr(r13, Address(rbx, Method::const_offset())); // get ConstMethod*
576 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
577 __ push(rbx); // save Method*
578 if (ProfileInterpreter) {
579 Label method_data_continue;
580 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
581 __ testptr(rdx, rdx);
582 __ jcc(Assembler::zero, method_data_continue);
583 __ addptr(rdx, in_bytes(MethodData::data_offset()));
584 __ bind(method_data_continue);
585 __ push(rdx); // set the mdp (method data pointer)
586 } else {
587 __ push(0);
588 }
589
590 __ movptr(rdx, Address(rbx, Method::const_offset()));
591 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
592 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
593 __ push(rdx); // set constant pool cache
594 __ push(r14); // set locals pointer
595 if (native_call) {
596 __ push(0); // no bcp
597 } else {
598 __ push(r13); // set bcp
599 }
600 __ push(0); // reserve word for pointer to expression stack bottom
601 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
602 }
603
604 // End of helpers
605
606 // Method entry for java.lang.ref.Reference.get.
607 address InterpreterGenerator::generate_Reference_get_entry(void) {
608 #if INCLUDE_ALL_GCS
609 // Code: _aload_0, _getfield, _areturn
610 // parameter size = 1
611 //
612 // The code that gets generated by this routine is split into 2 parts:
613 // 1. The "intrinsified" code for G1 (or any SATB based GC),
614 // 2. The slow path - which is an expansion of the regular method entry.
615 //
616 // Notes:-
617 // * In the G1 code we do not check whether we need to block for
618 // a safepoint. If G1 is enabled then we must execute the specialized
619 // code for Reference.get (except when the Reference object is null)
620 // so that we can log the value in the referent field with an SATB
621 // update buffer.
622 // If the code for the getfield template is modified so that the
623 // G1 pre-barrier code is executed when the current method is
624 // Reference.get() then going through the normal method entry
625 // will be fine.
626 // * The G1 code can, however, check the receiver object (the instance
627 // of java.lang.Reference) and jump to the slow path if null. If the
628 // Reference object is null then we obviously cannot fetch the referent
629 // and so we don't need to call the G1 pre-barrier. Thus we can use the
630 // regular method entry code to generate the NPE.
631 //
632 // rbx: Method*
633
634 // r13: senderSP must preserve for slow path, set SP to it on fast path
635
636 address entry = __ pc();
637
638 const int referent_offset = java_lang_ref_Reference::referent_offset;
639 guarantee(referent_offset > 0, "referent offset not initialized");
640
641 if (UseG1GC || UseShenandoahGC) {
642 Label slow_path;
643 // rbx: method
644
645 // Check if local 0 != NULL
646 // If the receiver is null then it is OK to jump to the slow path.
647 __ movptr(rax, Address(rsp, wordSize));
648
649 __ testptr(rax, rax);
650 __ jcc(Assembler::zero, slow_path);
651
652 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
653
654 // rax: local 0
655 // rbx: method (but can be used as scratch now)
656 // rdx: scratch
657 // rdi: scratch
658
659 // Generate the G1 pre-barrier code to log the value of
660 // the referent field in an SATB buffer.
661
662 // Load the value of the referent field.
663 const Address field_address(rax, referent_offset);
664 __ load_heap_oop(rax, field_address);
665
666 // Generate the G1 pre-barrier code to log the value of
667 // the referent field in an SATB buffer.
668 __ g1_write_barrier_pre(noreg /* obj */,
669 rax /* pre_val */,
670 r15_thread /* thread */,
671 rbx /* tmp */,
672 true /* tosca_live */,
673 true /* expand_call */);
674
675 // _areturn
676 __ pop(rdi); // get return address
677 __ mov(rsp, r13); // set sp to sender sp
678 __ jmp(rdi);
679 __ ret(0);
680
681 // generate a vanilla interpreter entry as the slow path
682 __ bind(slow_path);
683 (void) generate_normal_entry(false);
684
685 return entry;
686 }
687 #endif // INCLUDE_ALL_GCS
688
689 // If G1 is not enabled then attempt to go through the accessor entry point
690 // Reference.get is an accessor
691 return generate_jump_to_normal_entry();
692 }
693
694 /**
695 * Method entry for static native methods:
696 * int java.util.zip.CRC32.update(int crc, int b)
697 */
698 address InterpreterGenerator::generate_CRC32_update_entry() {
699 if (UseCRC32Intrinsics) {
700 address entry = __ pc();
701
702 // rbx,: Method*
703 // r13: senderSP must preserved for slow path, set SP to it on fast path
704 // c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
705 // c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
706
707 Label slow_path;
708 // If we need a safepoint check, generate full interpreter entry.
709 ExternalAddress state(SafepointSynchronize::address_of_state());
710 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
711 SafepointSynchronize::_not_synchronized);
712 __ jcc(Assembler::notEqual, slow_path);
713
714 // We don't generate local frame and don't align stack because
715 // we call stub code and there is no safepoint on this path.
716
717 // Load parameters
718 const Register crc = rax; // crc
719 const Register val = c_rarg0; // source java byte value
720 const Register tbl = c_rarg1; // scratch
721
722 // Arguments are reversed on java expression stack
723 __ movl(val, Address(rsp, wordSize)); // byte value
724 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
725
726 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
727 __ notl(crc); // ~crc
728 __ update_byte_crc32(crc, val, tbl);
729 __ notl(crc); // ~crc
730 // result in rax
731
732 // _areturn
733 __ pop(rdi); // get return address
734 __ mov(rsp, r13); // set sp to sender sp
735 __ jmp(rdi);
736
737 // generate a vanilla native entry as the slow path
738 __ bind(slow_path);
739
740 (void) generate_native_entry(false);
741
742 return entry;
743 }
744 return generate_native_entry(false);
745 }
746
747 /**
748 * Method entry for static native methods:
749 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
750 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
751 */
752 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
753 if (UseCRC32Intrinsics) {
754 address entry = __ pc();
755
756 // rbx,: Method*
757 // r13: senderSP must preserved for slow path, set SP to it on fast path
758
759 Label slow_path;
760 // If we need a safepoint check, generate full interpreter entry.
761 ExternalAddress state(SafepointSynchronize::address_of_state());
762 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
763 SafepointSynchronize::_not_synchronized);
764 __ jcc(Assembler::notEqual, slow_path);
765
766 // We don't generate local frame and don't align stack because
767 // we call stub code and there is no safepoint on this path.
768
769 // Load parameters
770 const Register crc = c_rarg0; // crc
771 const Register buf = c_rarg1; // source java byte array address
772 const Register len = c_rarg2; // length
773 const Register off = len; // offset (never overlaps with 'len')
774
775 // Arguments are reversed on java expression stack
776 // Calculate address of start element
777 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
778 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
779 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
780 __ addq(buf, off); // + offset
781 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
782 } else {
783 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
784 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, buf);
785 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
786 __ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
787 __ addq(buf, off); // + offset
788 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
789 }
790 // Can now load 'len' since we're finished with 'off'
791 __ movl(len, Address(rsp, wordSize)); // Length
792
793 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
794 // result in rax
795
796 // _areturn
797 __ pop(rdi); // get return address
798 __ mov(rsp, r13); // set sp to sender sp
799 __ jmp(rdi);
800
801 // generate a vanilla native entry as the slow path
802 __ bind(slow_path);
803
804 (void) generate_native_entry(false);
805
806 return entry;
807 }
808 return generate_native_entry(false);
809 }
810
811 // Interpreter stub for calling a native method. (asm interpreter)
812 // This sets up a somewhat different looking stack for calling the
813 // native method than the typical interpreter frame setup.
814 address InterpreterGenerator::generate_native_entry(bool synchronized) {
815 // determine code generation flags
816 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
817
818 // rbx: Method*
819 // r13: sender sp
820
821 address entry_point = __ pc();
822
823 const Address constMethod (rbx, Method::const_offset());
824 const Address access_flags (rbx, Method::access_flags_offset());
825 const Address size_of_parameters(rcx, ConstMethod::
826 size_of_parameters_offset());
827
828
829 // get parameter size (always needed)
830 __ movptr(rcx, constMethod);
831 __ load_unsigned_short(rcx, size_of_parameters);
832
833 // native calls don't need the stack size check since they have no
834 // expression stack and the arguments are already on the stack and
835 // we only add a handful of words to the stack
836
837 // rbx: Method*
838 // rcx: size of parameters
839 // r13: sender sp
840 __ pop(rax); // get return address
841
842 // for natives the size of locals is zero
843
844 // compute beginning of parameters (r14)
845 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
846
847 // add 2 zero-initialized slots for native calls
848 // initialize result_handler slot
849 __ push((int) NULL_WORD);
850 // slot for oop temp
851 // (static native method holder mirror/jni oop result)
852 __ push((int) NULL_WORD);
853
854 // initialize fixed part of activation frame
855 generate_fixed_frame(true);
856
857 // make sure method is native & not abstract
858 #ifdef ASSERT
859 __ movl(rax, access_flags);
860 {
861 Label L;
862 __ testl(rax, JVM_ACC_NATIVE);
863 __ jcc(Assembler::notZero, L);
864 __ stop("tried to execute non-native method as native");
865 __ bind(L);
866 }
867 {
868 Label L;
869 __ testl(rax, JVM_ACC_ABSTRACT);
870 __ jcc(Assembler::zero, L);
871 __ stop("tried to execute abstract method in interpreter");
872 __ bind(L);
873 }
874 #endif
875
876 // Since at this point in the method invocation the exception handler
877 // would try to exit the monitor of synchronized methods which hasn't
878 // been entered yet, we set the thread local variable
879 // _do_not_unlock_if_synchronized to true. The remove_activation will
880 // check this flag.
881
882 const Address do_not_unlock_if_synchronized(r15_thread,
883 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
884 __ movbool(do_not_unlock_if_synchronized, true);
885
886 // increment invocation count & check for overflow
887 Label invocation_counter_overflow;
888 if (inc_counter) {
889 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
890 }
891
892 Label continue_after_compile;
893 __ bind(continue_after_compile);
894
895 bang_stack_shadow_pages(true);
896
897 // reset the _do_not_unlock_if_synchronized flag
898 __ movbool(do_not_unlock_if_synchronized, false);
899
900 // check for synchronized methods
901 // Must happen AFTER invocation_counter check and stack overflow check,
902 // so method is not locked if overflows.
903 if (synchronized) {
904 lock_method();
905 } else {
906 // no synchronization necessary
907 #ifdef ASSERT
908 {
909 Label L;
910 __ movl(rax, access_flags);
911 __ testl(rax, JVM_ACC_SYNCHRONIZED);
912 __ jcc(Assembler::zero, L);
913 __ stop("method needs synchronization");
914 __ bind(L);
915 }
916 #endif
917 }
918
919 // start execution
920 #ifdef ASSERT
921 {
922 Label L;
923 const Address monitor_block_top(rbp,
924 frame::interpreter_frame_monitor_block_top_offset * wordSize);
925 __ movptr(rax, monitor_block_top);
926 __ cmpptr(rax, rsp);
927 __ jcc(Assembler::equal, L);
928 __ stop("broken stack frame setup in interpreter");
929 __ bind(L);
930 }
931 #endif
932
933 // jvmti support
934 __ notify_method_entry();
935
936 // work registers
937 const Register method = rbx;
938 const Register t = r11;
939
940 // allocate space for parameters
941 __ get_method(method);
942 __ movptr(t, Address(method, Method::const_offset()));
943 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
944 __ shll(t, Interpreter::logStackElementSize);
945
946 __ subptr(rsp, t);
947 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
948 __ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
949
950 // get signature handler
951 {
952 Label L;
953 __ movptr(t, Address(method, Method::signature_handler_offset()));
954 __ testptr(t, t);
955 __ jcc(Assembler::notZero, L);
956 __ call_VM(noreg,
957 CAST_FROM_FN_PTR(address,
958 InterpreterRuntime::prepare_native_call),
959 method);
960 __ get_method(method);
961 __ movptr(t, Address(method, Method::signature_handler_offset()));
962 __ bind(L);
963 }
964
965 // call signature handler
966 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == r14,
967 "adjust this code");
968 assert(InterpreterRuntime::SignatureHandlerGenerator::to() == rsp,
969 "adjust this code");
970 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == rscratch1,
971 "adjust this code");
972
973 // The generated handlers do not touch RBX (the method oop).
974 // However, large signatures cannot be cached and are generated
975 // each time here. The slow-path generator can do a GC on return,
976 // so we must reload it after the call.
977 __ call(t);
978 __ get_method(method); // slow path can do a GC, reload RBX
979
980
981 // result handler is in rax
982 // set result handler
983 __ movptr(Address(rbp,
984 (frame::interpreter_frame_result_handler_offset) * wordSize),
985 rax);
986
987 // pass mirror handle if static call
988 {
989 Label L;
990 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
991 __ movl(t, Address(method, Method::access_flags_offset()));
992 __ testl(t, JVM_ACC_STATIC);
993 __ jcc(Assembler::zero, L);
994 // get mirror
995 __ movptr(t, Address(method, Method::const_offset()));
996 __ movptr(t, Address(t, ConstMethod::constants_offset()));
997 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
998 __ movptr(t, Address(t, mirror_offset));
999 // copy mirror into activation frame
1000 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize),
1001 t);
1002 // pass handle to mirror
1003 __ lea(c_rarg1,
1004 Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1005 __ bind(L);
1006 }
1007
1008 // get native function entry point
1009 {
1010 Label L;
1011 __ movptr(rax, Address(method, Method::native_function_offset()));
1012 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1013 __ movptr(rscratch2, unsatisfied.addr());
1014 __ cmpptr(rax, rscratch2);
1015 __ jcc(Assembler::notEqual, L);
1016 __ call_VM(noreg,
1017 CAST_FROM_FN_PTR(address,
1018 InterpreterRuntime::prepare_native_call),
1019 method);
1020 __ get_method(method);
1021 __ movptr(rax, Address(method, Method::native_function_offset()));
1022 __ bind(L);
1023 }
1024
1025 // pass JNIEnv
1026 __ lea(c_rarg0, Address(r15_thread, JavaThread::jni_environment_offset()));
1027
1028 // It is enough that the pc() points into the right code
1029 // segment. It does not have to be the correct return pc.
1030 __ set_last_Java_frame(rsp, rbp, (address) __ pc());
1031
1032 // change thread state
1033 #ifdef ASSERT
1034 {
1035 Label L;
1036 __ movl(t, Address(r15_thread, JavaThread::thread_state_offset()));
1037 __ cmpl(t, _thread_in_Java);
1038 __ jcc(Assembler::equal, L);
1039 __ stop("Wrong thread state in native stub");
1040 __ bind(L);
1041 }
1042 #endif
1043
1044 // Change state to native
1045
1046 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1047 _thread_in_native);
1048
1049 // Call the native method.
1050 __ call(rax);
1051 // result potentially in rax or xmm0
1052
1053 // Verify or restore cpu control state after JNI call
1054 __ restore_cpu_control_state_after_jni();
1055
1056 // NOTE: The order of these pushes is known to frame::interpreter_frame_result
1057 // in order to extract the result of a method call. If the order of these
1058 // pushes change or anything else is added to the stack then the code in
1059 // interpreter_frame_result must also change.
1060
1061 __ push(dtos);
1062 __ push(ltos);
1063
1064 // change thread state
1065 __ movl(Address(r15_thread, JavaThread::thread_state_offset()),
1066 _thread_in_native_trans);
1067
1068 if (os::is_MP()) {
1069 if (UseMembar) {
1070 // Force this write out before the read below
1071 __ membar(Assembler::Membar_mask_bits(
1072 Assembler::LoadLoad | Assembler::LoadStore |
1073 Assembler::StoreLoad | Assembler::StoreStore));
1074 } else {
1075 // Write serialization page so VM thread can do a pseudo remote membar.
1076 // We use the current thread pointer to calculate a thread specific
1077 // offset to write to within the page. This minimizes bus traffic
1078 // due to cache line collision.
1079 __ serialize_memory(r15_thread, rscratch2);
1080 }
1081 }
1082
1083 // check for safepoint operation in progress and/or pending suspend requests
1084 {
1085 Label Continue;
1086 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1087 SafepointSynchronize::_not_synchronized);
1088
1089 Label L;
1090 __ jcc(Assembler::notEqual, L);
1091 __ cmpl(Address(r15_thread, JavaThread::suspend_flags_offset()), 0);
1092 __ jcc(Assembler::equal, Continue);
1093 __ bind(L);
1094
1095 // Don't use call_VM as it will see a possible pending exception
1096 // and forward it and never return here preventing us from
1097 // clearing _last_native_pc down below. Also can't use
1098 // call_VM_leaf either as it will check to see if r13 & r14 are
1099 // preserved and correspond to the bcp/locals pointers. So we do a
1100 // runtime call by hand.
1101 //
1102 __ mov(c_rarg0, r15_thread);
1103 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1104 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1105 __ andptr(rsp, -16); // align stack as required by ABI
1106 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans)));
1107 __ mov(rsp, r12); // restore sp
1108 __ reinit_heapbase();
1109 __ bind(Continue);
1110 }
1111
1112 // change thread state
1113 __ movl(Address(r15_thread, JavaThread::thread_state_offset()), _thread_in_Java);
1114
1115 // reset_last_Java_frame
1116 __ reset_last_Java_frame(true, true);
1117
1118 // reset handle block
1119 __ movptr(t, Address(r15_thread, JavaThread::active_handles_offset()));
1120 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
1121
1122 // If result is an oop unbox and store it in frame where gc will see it
1123 // and result handler will pick it up
1124
1125 {
1126 Label no_oop, store_result;
1127 __ lea(t, ExternalAddress(AbstractInterpreter::result_handler(T_OBJECT)));
1128 __ cmpptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1129 __ jcc(Assembler::notEqual, no_oop);
1130 // retrieve result
1131 __ pop(ltos);
1132 __ testptr(rax, rax);
1133 __ jcc(Assembler::zero, store_result);
1134 __ movptr(rax, Address(rax, 0));
1135 __ bind(store_result);
1136 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize), rax);
1137 // keep stack depth as expected by pushing oop which will eventually be discarde
1138 __ push(ltos);
1139 __ bind(no_oop);
1140 }
1141
1142
1143 {
1144 Label no_reguard;
1145 __ cmpl(Address(r15_thread, JavaThread::stack_guard_state_offset()),
1146 JavaThread::stack_guard_yellow_disabled);
1147 __ jcc(Assembler::notEqual, no_reguard);
1148
1149 __ pusha(); // XXX only save smashed registers
1150 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1151 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1152 __ andptr(rsp, -16); // align stack as required by ABI
1153 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1154 __ mov(rsp, r12); // restore sp
1155 __ popa(); // XXX only restore smashed registers
1156 __ reinit_heapbase();
1157
1158 __ bind(no_reguard);
1159 }
1160
1161
1162 // The method register is junk from after the thread_in_native transition
1163 // until here. Also can't call_VM until the bcp has been
1164 // restored. Need bcp for throwing exception below so get it now.
1165 __ get_method(method);
1166
1167 // restore r13 to have legal interpreter frame, i.e., bci == 0 <=>
1168 // r13 == code_base()
1169 __ movptr(r13, Address(method, Method::const_offset())); // get ConstMethod*
1170 __ lea(r13, Address(r13, ConstMethod::codes_offset())); // get codebase
1171 // handle exceptions (exception handling will handle unlocking!)
1172 {
1173 Label L;
1174 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD);
1175 __ jcc(Assembler::zero, L);
1176 // Note: At some point we may want to unify this with the code
1177 // used in call_VM_base(); i.e., we should use the
1178 // StubRoutines::forward_exception code. For now this doesn't work
1179 // here because the rsp is not correctly set at this point.
1180 __ MacroAssembler::call_VM(noreg,
1181 CAST_FROM_FN_PTR(address,
1182 InterpreterRuntime::throw_pending_exception));
1183 __ should_not_reach_here();
1184 __ bind(L);
1185 }
1186
1187 // do unlocking if necessary
1188 {
1189 Label L;
1190 __ movl(t, Address(method, Method::access_flags_offset()));
1191 __ testl(t, JVM_ACC_SYNCHRONIZED);
1192 __ jcc(Assembler::zero, L);
1193 // the code below should be shared with interpreter macro
1194 // assembler implementation
1195 {
1196 Label unlock;
1197 // BasicObjectLock will be first in list, since this is a
1198 // synchronized method. However, need to check that the object
1199 // has not been unlocked by an explicit monitorexit bytecode.
1200 const Address monitor(rbp,
1201 (intptr_t)(frame::interpreter_frame_initial_sp_offset *
1202 wordSize - sizeof(BasicObjectLock)));
1203
1204 // monitor expect in c_rarg1 for slow unlock path
1205 __ lea(c_rarg1, monitor); // address of first monitor
1206
1207 __ movptr(t, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
1208 __ testptr(t, t);
1209 __ jcc(Assembler::notZero, unlock);
1210
1211 // Entry already unlocked, need to throw exception
1212 __ MacroAssembler::call_VM(noreg,
1213 CAST_FROM_FN_PTR(address,
1214 InterpreterRuntime::throw_illegal_monitor_state_exception));
1215 __ should_not_reach_here();
1216
1217 __ bind(unlock);
1218 __ unlock_object(c_rarg1);
1219 }
1220 __ bind(L);
1221 }
1222
1223 // jvmti support
1224 // Note: This must happen _after_ handling/throwing any exceptions since
1225 // the exception handler code notifies the runtime of method exits
1226 // too. If this happens before, method entry/exit notifications are
1227 // not properly paired (was bug - gri 11/22/99).
1228 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1229
1230 // restore potential result in edx:eax, call result handler to
1231 // restore potential result in ST0 & handle result
1232
1233 __ pop(ltos);
1234 __ pop(dtos);
1235
1236 __ movptr(t, Address(rbp,
1237 (frame::interpreter_frame_result_handler_offset) * wordSize));
1238 __ call(t);
1239
1240 // remove activation
1241 __ movptr(t, Address(rbp,
1242 frame::interpreter_frame_sender_sp_offset *
1243 wordSize)); // get sender sp
1244 __ leave(); // remove frame anchor
1245 __ pop(rdi); // get return address
1246 __ mov(rsp, t); // set sp to sender sp
1247 __ jmp(rdi);
1248
1249 if (inc_counter) {
1250 // Handle overflow of counter and compile method
1251 __ bind(invocation_counter_overflow);
1252 generate_counter_overflow(&continue_after_compile);
1253 }
1254
1255 return entry_point;
1256 }
1257
1258 //
1259 // Generic interpreted method entry to (asm) interpreter
1260 //
1261 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1262 // determine code generation flags
1263 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1264
1265 // ebx: Method*
1266 // r13: sender sp
1267 address entry_point = __ pc();
1268
1269 const Address constMethod(rbx, Method::const_offset());
1270 const Address access_flags(rbx, Method::access_flags_offset());
1271 const Address size_of_parameters(rdx,
1272 ConstMethod::size_of_parameters_offset());
1273 const Address size_of_locals(rdx, ConstMethod::size_of_locals_offset());
1274
1275
1276 // get parameter size (always needed)
1277 __ movptr(rdx, constMethod);
1278 __ load_unsigned_short(rcx, size_of_parameters);
1279
1280 // rbx: Method*
1281 // rcx: size of parameters
1282 // r13: sender_sp (could differ from sp+wordSize if we were called via c2i )
1283
1284 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1285 __ subl(rdx, rcx); // rdx = no. of additional locals
1286
1287 // YYY
1288 // __ incrementl(rdx);
1289 // __ andl(rdx, -2);
1290
1291 // see if we've got enough room on the stack for locals plus overhead.
1292 generate_stack_overflow_check();
1293
1294 // get return address
1295 __ pop(rax);
1296
1297 // compute beginning of parameters (r14)
1298 __ lea(r14, Address(rsp, rcx, Address::times_8, -wordSize));
1299
1300 // rdx - # of additional locals
1301 // allocate space for locals
1302 // explicitly initialize locals
1303 {
1304 Label exit, loop;
1305 __ testl(rdx, rdx);
1306 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1307 __ bind(loop);
1308 __ push((int) NULL_WORD); // initialize local variables
1309 __ decrementl(rdx); // until everything initialized
1310 __ jcc(Assembler::greater, loop);
1311 __ bind(exit);
1312 }
1313
1314 // initialize fixed part of activation frame
1315 generate_fixed_frame(false);
1316
1317 // make sure method is not native & not abstract
1318 #ifdef ASSERT
1319 __ movl(rax, access_flags);
1320 {
1321 Label L;
1322 __ testl(rax, JVM_ACC_NATIVE);
1323 __ jcc(Assembler::zero, L);
1324 __ stop("tried to execute native method as non-native");
1325 __ bind(L);
1326 }
1327 {
1328 Label L;
1329 __ testl(rax, JVM_ACC_ABSTRACT);
1330 __ jcc(Assembler::zero, L);
1331 __ stop("tried to execute abstract method in interpreter");
1332 __ bind(L);
1333 }
1334 #endif
1335
1336 // Since at this point in the method invocation the exception
1337 // handler would try to exit the monitor of synchronized methods
1338 // which hasn't been entered yet, we set the thread local variable
1339 // _do_not_unlock_if_synchronized to true. The remove_activation
1340 // will check this flag.
1341
1342 const Address do_not_unlock_if_synchronized(r15_thread,
1343 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1344 __ movbool(do_not_unlock_if_synchronized, true);
1345
1346 __ profile_parameters_type(rax, rcx, rdx);
1347 // increment invocation count & check for overflow
1348 Label invocation_counter_overflow;
1349 Label profile_method;
1350 Label profile_method_continue;
1351 if (inc_counter) {
1352 generate_counter_incr(&invocation_counter_overflow,
1353 &profile_method,
1354 &profile_method_continue);
1355 if (ProfileInterpreter) {
1356 __ bind(profile_method_continue);
1357 }
1358 }
1359
1360 Label continue_after_compile;
1361 __ bind(continue_after_compile);
1362
1363 // check for synchronized interpreted methods
1364 bang_stack_shadow_pages(false);
1365
1366 // reset the _do_not_unlock_if_synchronized flag
1367 __ movbool(do_not_unlock_if_synchronized, false);
1368
1369 // check for synchronized methods
1370 // Must happen AFTER invocation_counter check and stack overflow check,
1371 // so method is not locked if overflows.
1372 if (synchronized) {
1373 // Allocate monitor and lock method
1374 lock_method();
1375 } else {
1376 // no synchronization necessary
1377 #ifdef ASSERT
1378 {
1379 Label L;
1380 __ movl(rax, access_flags);
1381 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1382 __ jcc(Assembler::zero, L);
1383 __ stop("method needs synchronization");
1384 __ bind(L);
1385 }
1386 #endif
1387 }
1388
1389 // start execution
1390 #ifdef ASSERT
1391 {
1392 Label L;
1393 const Address monitor_block_top (rbp,
1394 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1395 __ movptr(rax, monitor_block_top);
1396 __ cmpptr(rax, rsp);
1397 __ jcc(Assembler::equal, L);
1398 __ stop("broken stack frame setup in interpreter");
1399 __ bind(L);
1400 }
1401 #endif
1402
1403 // jvmti support
1404 __ notify_method_entry();
1405
1406 __ dispatch_next(vtos);
1407
1408 // invocation counter overflow
1409 if (inc_counter) {
1410 if (ProfileInterpreter) {
1411 // We have decided to profile this method in the interpreter
1412 __ bind(profile_method);
1413 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1414 __ set_method_data_pointer_for_bcp();
1415 __ get_method(rbx);
1416 __ jmp(profile_method_continue);
1417 }
1418 // Handle overflow of counter and compile method
1419 __ bind(invocation_counter_overflow);
1420 generate_counter_overflow(&continue_after_compile);
1421 }
1422
1423 return entry_point;
1424 }
1425
1426
1427 // These should never be compiled since the interpreter will prefer
1428 // the compiled version to the intrinsic version.
1429 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1430 switch (method_kind(m)) {
1431 case Interpreter::java_lang_math_sin : // fall thru
1432 case Interpreter::java_lang_math_cos : // fall thru
1433 case Interpreter::java_lang_math_tan : // fall thru
1434 case Interpreter::java_lang_math_abs : // fall thru
1435 case Interpreter::java_lang_math_log : // fall thru
1436 case Interpreter::java_lang_math_log10 : // fall thru
1437 case Interpreter::java_lang_math_sqrt : // fall thru
1438 case Interpreter::java_lang_math_pow : // fall thru
1439 case Interpreter::java_lang_math_exp :
1440 return false;
1441 default:
1442 return true;
1443 }
1444 }
1445
1446 // How much stack a method activation needs in words.
1447 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1448 const int entry_size = frame::interpreter_frame_monitor_size();
1449
1450 // total overhead size: entry_size + (saved rbp thru expr stack
1451 // bottom). be sure to change this if you add/subtract anything
1452 // to/from the overhead area
1453 const int overhead_size =
1454 -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1455
1456 const int stub_code = frame::entry_frame_after_call_words;
1457 const int method_stack = (method->max_locals() + method->max_stack()) *
1458 Interpreter::stackElementWords;
1459 return (overhead_size + method_stack + stub_code);
1460 }
1461
1462 //-----------------------------------------------------------------------------
1463 // Exceptions
1464
1465 void TemplateInterpreterGenerator::generate_throw_exception() {
1466 // Entry point in previous activation (i.e., if the caller was
1467 // interpreted)
1468 Interpreter::_rethrow_exception_entry = __ pc();
1469 // Restore sp to interpreter_frame_last_sp even though we are going
1470 // to empty the expression stack for the exception processing.
1471 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1472 // rax: exception
1473 // rdx: return address/pc that threw exception
1474 __ restore_bcp(); // r13 points to call/send
1475 __ restore_locals();
1476 __ reinit_heapbase(); // restore r12 as heapbase.
1477 // Entry point for exceptions thrown within interpreter code
1478 Interpreter::_throw_exception_entry = __ pc();
1479 // expression stack is undefined here
1480 // rax: exception
1481 // r13: exception bcp
1482 __ verify_oop(rax);
1483 __ mov(c_rarg1, rax);
1484
1485 // expression stack must be empty before entering the VM in case of
1486 // an exception
1487 __ empty_expression_stack();
1488 // find exception handler address and preserve exception oop
1489 __ call_VM(rdx,
1490 CAST_FROM_FN_PTR(address,
1491 InterpreterRuntime::exception_handler_for_exception),
1492 c_rarg1);
1493 // rax: exception handler entry point
1494 // rdx: preserved exception oop
1495 // r13: bcp for exception handler
1496 __ push_ptr(rdx); // push exception which is now the only value on the stack
1497 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1498
1499 // If the exception is not handled in the current frame the frame is
1500 // removed and the exception is rethrown (i.e. exception
1501 // continuation is _rethrow_exception).
1502 //
1503 // Note: At this point the bci is still the bxi for the instruction
1504 // which caused the exception and the expression stack is
1505 // empty. Thus, for any VM calls at this point, GC will find a legal
1506 // oop map (with empty expression stack).
1507
1508 // In current activation
1509 // tos: exception
1510 // esi: exception bcp
1511
1512 //
1513 // JVMTI PopFrame support
1514 //
1515
1516 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1517 __ empty_expression_stack();
1518 // Set the popframe_processing bit in pending_popframe_condition
1519 // indicating that we are currently handling popframe, so that
1520 // call_VMs that may happen later do not trigger new popframe
1521 // handling cycles.
1522 __ movl(rdx, Address(r15_thread, JavaThread::popframe_condition_offset()));
1523 __ orl(rdx, JavaThread::popframe_processing_bit);
1524 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()), rdx);
1525
1526 {
1527 // Check to see whether we are returning to a deoptimized frame.
1528 // (The PopFrame call ensures that the caller of the popped frame is
1529 // either interpreted or compiled and deoptimizes it if compiled.)
1530 // In this case, we can't call dispatch_next() after the frame is
1531 // popped, but instead must save the incoming arguments and restore
1532 // them after deoptimization has occurred.
1533 //
1534 // Note that we don't compare the return PC against the
1535 // deoptimization blob's unpack entry because of the presence of
1536 // adapter frames in C2.
1537 Label caller_not_deoptimized;
1538 __ movptr(c_rarg1, Address(rbp, frame::return_addr_offset * wordSize));
1539 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1540 InterpreterRuntime::interpreter_contains), c_rarg1);
1541 __ testl(rax, rax);
1542 __ jcc(Assembler::notZero, caller_not_deoptimized);
1543
1544 // Compute size of arguments for saving when returning to
1545 // deoptimized caller
1546 __ get_method(rax);
1547 __ movptr(rax, Address(rax, Method::const_offset()));
1548 __ load_unsigned_short(rax, Address(rax, in_bytes(ConstMethod::
1549 size_of_parameters_offset())));
1550 __ shll(rax, Interpreter::logStackElementSize);
1551 __ restore_locals(); // XXX do we need this?
1552 __ subptr(r14, rax);
1553 __ addptr(r14, wordSize);
1554 // Save these arguments
1555 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1556 Deoptimization::
1557 popframe_preserve_args),
1558 r15_thread, rax, r14);
1559
1560 __ remove_activation(vtos, rdx,
1561 /* throw_monitor_exception */ false,
1562 /* install_monitor_exception */ false,
1563 /* notify_jvmdi */ false);
1564
1565 // Inform deoptimization that it is responsible for restoring
1566 // these arguments
1567 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1568 JavaThread::popframe_force_deopt_reexecution_bit);
1569
1570 // Continue in deoptimization handler
1571 __ jmp(rdx);
1572
1573 __ bind(caller_not_deoptimized);
1574 }
1575
1576 __ remove_activation(vtos, rdx, /* rdx result (retaddr) is not used */
1577 /* throw_monitor_exception */ false,
1578 /* install_monitor_exception */ false,
1579 /* notify_jvmdi */ false);
1580
1581 // Finish with popframe handling
1582 // A previous I2C followed by a deoptimization might have moved the
1583 // outgoing arguments further up the stack. PopFrame expects the
1584 // mutations to those outgoing arguments to be preserved and other
1585 // constraints basically require this frame to look exactly as
1586 // though it had previously invoked an interpreted activation with
1587 // no space between the top of the expression stack (current
1588 // last_sp) and the top of stack. Rather than force deopt to
1589 // maintain this kind of invariant all the time we call a small
1590 // fixup routine to move the mutated arguments onto the top of our
1591 // expression stack if necessary.
1592 __ mov(c_rarg1, rsp);
1593 __ movptr(c_rarg2, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1594 // PC must point into interpreter here
1595 __ set_last_Java_frame(noreg, rbp, __ pc());
1596 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), r15_thread, c_rarg1, c_rarg2);
1597 __ reset_last_Java_frame(true, true);
1598 // Restore the last_sp and null it out
1599 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1600 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
1601
1602 __ restore_bcp(); // XXX do we need this?
1603 __ restore_locals(); // XXX do we need this?
1604 // The method data pointer was incremented already during
1605 // call profiling. We have to restore the mdp for the current bcp.
1606 if (ProfileInterpreter) {
1607 __ set_method_data_pointer_for_bcp();
1608 }
1609
1610 // Clear the popframe condition flag
1611 __ movl(Address(r15_thread, JavaThread::popframe_condition_offset()),
1612 JavaThread::popframe_inactive);
1613
1614 #if INCLUDE_JVMTI
1615 {
1616 Label L_done;
1617 const Register local0 = r14;
1618
1619 __ cmpb(Address(r13, 0), Bytecodes::_invokestatic);
1620 __ jcc(Assembler::notEqual, L_done);
1621
1622 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1623 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1624
1625 __ get_method(rdx);
1626 __ movptr(rax, Address(local0, 0));
1627 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, r13);
1628
1629 __ testptr(rax, rax);
1630 __ jcc(Assembler::zero, L_done);
1631
1632 __ movptr(Address(rbx, 0), rax);
1633 __ bind(L_done);
1634 }
1635 #endif // INCLUDE_JVMTI
1636
1637 __ dispatch_next(vtos);
1638 // end of PopFrame support
1639
1640 Interpreter::_remove_activation_entry = __ pc();
1641
1642 // preserve exception over this code sequence
1643 __ pop_ptr(rax);
1644 __ movptr(Address(r15_thread, JavaThread::vm_result_offset()), rax);
1645 // remove the activation (without doing throws on illegalMonitorExceptions)
1646 __ remove_activation(vtos, rdx, false, true, false);
1647 // restore exception
1648 __ get_vm_result(rax, r15_thread);
1649
1650 // In between activations - previous activation type unknown yet
1651 // compute continuation point - the continuation point expects the
1652 // following registers set up:
1653 //
1654 // rax: exception
1655 // rdx: return address/pc that threw exception
1656 // rsp: expression stack of caller
1657 // rbp: ebp of caller
1658 __ push(rax); // save exception
1659 __ push(rdx); // save return address
1660 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address,
1661 SharedRuntime::exception_handler_for_return_address),
1662 r15_thread, rdx);
1663 __ mov(rbx, rax); // save exception handler
1664 __ pop(rdx); // restore return address
1665 __ pop(rax); // restore exception
1666 // Note that an "issuing PC" is actually the next PC after the call
1667 __ jmp(rbx); // jump to exception
1668 // handler of caller
1669 }
1670
1671
1672 //
1673 // JVMTI ForceEarlyReturn support
1674 //
1675 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1676 address entry = __ pc();
1677
1678 __ restore_bcp();
1679 __ restore_locals();
1680 __ empty_expression_stack();
1681 __ load_earlyret_value(state);
1682
1683 __ movptr(rdx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
1684 Address cond_addr(rdx, JvmtiThreadState::earlyret_state_offset());
1685
1686 // Clear the earlyret state
1687 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1688
1689 __ remove_activation(state, rsi,
1690 false, /* throw_monitor_exception */
1691 false, /* install_monitor_exception */
1692 true); /* notify_jvmdi */
1693 __ jmp(rsi);
1694
1695 return entry;
1696 } // end of ForceEarlyReturn support
1697
1698
1699 //-----------------------------------------------------------------------------
1700 // Helper for vtos entry point generation
1701
1702 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
1703 address& bep,
1704 address& cep,
1705 address& sep,
1706 address& aep,
1707 address& iep,
1708 address& lep,
1709 address& fep,
1710 address& dep,
1711 address& vep) {
1712 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1713 Label L;
1714 aep = __ pc(); __ push_ptr(); __ jmp(L);
1715 fep = __ pc(); __ push_f(xmm0); __ jmp(L);
1716 dep = __ pc(); __ push_d(xmm0); __ jmp(L);
1717 lep = __ pc(); __ push_l(); __ jmp(L);
1718 bep = cep = sep =
1719 iep = __ pc(); __ push_i();
1720 vep = __ pc();
1721 __ bind(L);
1722 generate_and_dispatch(t);
1723 }
1724
1725
1726 //-----------------------------------------------------------------------------
1727 // Generation of individual instructions
1728
1729 // helpers for generate_and_dispatch
1730
1731
1732 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1733 : TemplateInterpreterGenerator(code) {
1734 generate_all(); // down here so it can be "virtual"
1735 }
1736
1737 //-----------------------------------------------------------------------------
1738
1739 // Non-product code
1740 #ifndef PRODUCT
1741 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1742 address entry = __ pc();
1743
1744 __ push(state);
1745 __ push(c_rarg0);
1746 __ push(c_rarg1);
1747 __ push(c_rarg2);
1748 __ push(c_rarg3);
1749 __ mov(c_rarg2, rax); // Pass itos
1750 #ifdef _WIN64
1751 __ movflt(xmm3, xmm0); // Pass ftos
1752 #endif
1753 __ call_VM(noreg,
1754 CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode),
1755 c_rarg1, c_rarg2, c_rarg3);
1756 __ pop(c_rarg3);
1757 __ pop(c_rarg2);
1758 __ pop(c_rarg1);
1759 __ pop(c_rarg0);
1760 __ pop(state);
1761 __ ret(0); // return from result handler
1762
1763 return entry;
1764 }
1765
1766 void TemplateInterpreterGenerator::count_bytecode() {
1767 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1768 }
1769
1770 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1771 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1772 }
1773
1774 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1775 __ mov32(rbx, ExternalAddress((address) &BytecodePairHistogram::_index));
1776 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1777 __ orl(rbx,
1778 ((int) t->bytecode()) <<
1779 BytecodePairHistogram::log2_number_of_codes);
1780 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1781 __ lea(rscratch1, ExternalAddress((address) BytecodePairHistogram::_counters));
1782 __ incrementl(Address(rscratch1, rbx, Address::times_4));
1783 }
1784
1785
1786 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1787 // Call a little run-time stub to avoid blow-up for each bytecode.
1788 // The run-time runtime saves the right registers, depending on
1789 // the tosca in-state for the given template.
1790
1791 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1792 "entry must have been generated");
1793 __ mov(r12, rsp); // remember sp (can only use r12 if not using call_VM)
1794 __ andptr(rsp, -16); // align stack as required by ABI
1795 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1796 __ mov(rsp, r12); // restore sp
1797 __ reinit_heapbase();
1798 }
1799
1800
1801 void TemplateInterpreterGenerator::stop_interpreter_at() {
1802 Label L;
1803 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1804 StopInterpreterAt);
1805 __ jcc(Assembler::notEqual, L);
1806 __ int3();
1807 __ bind(L);
1808 }
1809 #endif // !PRODUCT
1810 #endif // ! CC_INTERP