1 /*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/arguments.hpp"
39 #include "runtime/deoptimization.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/synchronizer.hpp"
44 #include "runtime/timer.hpp"
45 #include "runtime/vframeArray.hpp"
46 #include "utilities/debug.hpp"
47 #include "utilities/macros.hpp"
48
49 #define __ _masm->
50
51
52 #ifndef CC_INTERP
53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
56
57 //------------------------------------------------------------------------------------------------------------------------
58
59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
60 address entry = __ pc();
61
62 // Note: There should be a minimal interpreter frame set up when stack
63 // overflow occurs since we check explicitly for it now.
64 //
65 #ifdef ASSERT
66 { Label L;
67 __ lea(rax, Address(rbp,
68 frame::interpreter_frame_monitor_block_top_offset * wordSize));
69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
70 // (stack 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 exception
81 // happened
82 __ empty_expression_stack();
83 __ empty_FPU_stack();
84 // throw exception
85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
86 return entry;
87 }
88
89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
90 address entry = __ pc();
91 // expression stack must be empty before entering the VM if an exception happened
92 __ empty_expression_stack();
93 __ empty_FPU_stack();
94 // setup parameters
95 // ??? convention: expect aberrant index in register rbx,
96 __ lea(rax, ExternalAddress((address)name));
97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
98 return entry;
99 }
100
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // object is at TOS
104 __ pop(rax);
105 // expression stack must be empty before entering the VM if an exception
106 // happened
107 __ empty_expression_stack();
108 __ empty_FPU_stack();
109 __ call_VM(noreg,
110 CAST_FROM_FN_PTR(address,
111 InterpreterRuntime::throw_ClassCastException),
112 rax);
113 return entry;
114 }
115
116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
117 assert(!pass_oop || message == NULL, "either oop or message but not both");
118 address entry = __ pc();
119 if (pass_oop) {
120 // object is at TOS
121 __ pop(rbx);
122 }
123 // expression stack must be empty before entering the VM if an exception happened
124 __ empty_expression_stack();
125 __ empty_FPU_stack();
126 // setup parameters
127 __ lea(rax, ExternalAddress((address)name));
128 if (pass_oop) {
129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
130 } else {
131 if (message != NULL) {
132 __ lea(rbx, ExternalAddress((address)message));
133 } else {
134 __ movptr(rbx, NULL_WORD);
135 }
136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
137 }
138 // throw exception
139 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
140 return entry;
141 }
142
143
144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
145 address entry = __ pc();
146 // NULL last_sp until next java call
147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
148 __ dispatch_next(state);
149 return entry;
150 }
151
152
153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
154 address entry = __ pc();
155
156 #ifdef COMPILER2
157 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
158 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
159 for (int i = 1; i < 8; i++) {
160 __ ffree(i);
161 }
162 } else if (UseSSE < 2) {
163 __ empty_FPU_stack();
164 }
165 #endif
166 if ((state == ftos && UseSSE < 1) || (state == dtos && UseSSE < 2)) {
167 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
168 } else {
169 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
170 }
171
172 // In SSE mode, interpreter returns FP results in xmm0 but they need
173 // to end up back on the FPU so it can operate on them.
174 if (state == ftos && UseSSE >= 1) {
175 __ subptr(rsp, wordSize);
176 __ movflt(Address(rsp, 0), xmm0);
177 __ fld_s(Address(rsp, 0));
178 __ addptr(rsp, wordSize);
179 } else if (state == dtos && UseSSE >= 2) {
180 __ subptr(rsp, 2*wordSize);
181 __ movdbl(Address(rsp, 0), xmm0);
182 __ fld_d(Address(rsp, 0));
183 __ addptr(rsp, 2*wordSize);
184 }
185
186 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
187
188 // Restore stack bottom in case i2c adjusted stack
189 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
190 // and NULL it as marker that rsp is now tos until next java call
191 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
192
193 __ restore_bcp();
194 __ restore_locals();
195
196 if (state == atos) {
197 Register mdp = rbx;
198 Register tmp = rcx;
199 __ profile_return_type(mdp, rax, tmp);
200 }
201
202 const Register cache = rbx;
203 const Register index = rcx;
204 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
205
206 const Register flags = cache;
207 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
208 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
209 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
210 __ dispatch_next(state, step);
211
212 return entry;
213 }
214
215
216 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
217 address entry = __ pc();
218
219 // In SSE mode, FP results are in xmm0
220 if (state == ftos && UseSSE > 0) {
221 __ subptr(rsp, wordSize);
222 __ movflt(Address(rsp, 0), xmm0);
223 __ fld_s(Address(rsp, 0));
224 __ addptr(rsp, wordSize);
225 } else if (state == dtos && UseSSE >= 2) {
226 __ subptr(rsp, 2*wordSize);
227 __ movdbl(Address(rsp, 0), xmm0);
228 __ fld_d(Address(rsp, 0));
229 __ addptr(rsp, 2*wordSize);
230 }
231
232 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
233
234 // The stack is not extended by deopt but we must NULL last_sp as this
235 // entry is like a "return".
236 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
237 __ restore_bcp();
238 __ restore_locals();
239 // handle exceptions
240 { Label L;
241 const Register thread = rcx;
242 __ get_thread(thread);
243 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
244 __ jcc(Assembler::zero, L);
245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
246 __ should_not_reach_here();
247 __ bind(L);
248 }
249 __ dispatch_next(state, step);
250 return entry;
251 }
252
253
254 int AbstractInterpreter::BasicType_as_index(BasicType type) {
255 int i = 0;
256 switch (type) {
257 case T_BOOLEAN: i = 0; break;
258 case T_CHAR : i = 1; break;
259 case T_BYTE : i = 2; break;
260 case T_SHORT : i = 3; break;
261 case T_INT : // fall through
262 case T_LONG : // fall through
263 case T_VOID : i = 4; break;
264 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
265 case T_DOUBLE : i = 6; break;
266 case T_OBJECT : // fall through
267 case T_ARRAY : i = 7; break;
268 default : ShouldNotReachHere();
269 }
270 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
271 return i;
272 }
273
274
275 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
276 address entry = __ pc();
277 switch (type) {
278 case T_BOOLEAN: __ c2bool(rax); break;
279 case T_CHAR : __ andptr(rax, 0xFFFF); break;
280 case T_BYTE : __ sign_extend_byte (rax); break;
281 case T_SHORT : __ sign_extend_short(rax); break;
282 case T_INT : /* nothing to do */ break;
283 case T_LONG : /* nothing to do */ break;
284 case T_VOID : /* nothing to do */ break;
285 case T_DOUBLE :
286 case T_FLOAT :
287 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
288 __ pop(t); // remove return address first
289 // Must return a result for interpreter or compiler. In SSE
290 // mode, results are returned in xmm0 and the FPU stack must
291 // be empty.
292 if (type == T_FLOAT && UseSSE >= 1) {
293 // Load ST0
294 __ fld_d(Address(rsp, 0));
295 // Store as float and empty fpu stack
296 __ fstp_s(Address(rsp, 0));
297 // and reload
298 __ movflt(xmm0, Address(rsp, 0));
299 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
300 __ movdbl(xmm0, Address(rsp, 0));
301 } else {
302 // restore ST0
303 __ fld_d(Address(rsp, 0));
304 }
305 // and pop the temp
306 __ addptr(rsp, 2 * wordSize);
307 __ push(t); // restore return address
308 }
309 break;
310 case T_OBJECT :
311 // retrieve result from frame
312 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
313 // and verify it
314 __ verify_oop(rax);
315 break;
316 default : ShouldNotReachHere();
317 }
318 __ ret(0); // return from result handler
319 return entry;
320 }
321
322 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
323 address entry = __ pc();
324 __ push(state);
325 __ call_VM(noreg, runtime_entry);
326 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
327 return entry;
328 }
329
330
331 // Helpers for commoning out cases in the various type of method entries.
332 //
333
334 // increment invocation count & check for overflow
335 //
336 // Note: checking for negative value instead of overflow
337 // so we have a 'sticky' overflow test
338 //
339 // rbx,: method
340 // rcx: invocation counter
341 //
342 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
343 Label done;
344 // Note: In tiered we increment either counters in MethodCounters* or in MDO
345 // depending if we're profiling or not.
346 if (TieredCompilation) {
347 int increment = InvocationCounter::count_increment;
348 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
349 Label no_mdo;
350 if (ProfileInterpreter) {
351 // Are we profiling?
352 __ movptr(rax, Address(rbx, Method::method_data_offset()));
353 __ testptr(rax, rax);
354 __ jccb(Assembler::zero, no_mdo);
355 // Increment counter in the MDO
356 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
357 in_bytes(InvocationCounter::counter_offset()));
358 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
359 __ jmp(done);
360 }
361 __ bind(no_mdo);
362 // Increment counter in MethodCounters
363 const Address invocation_counter(rax,
364 MethodCounters::invocation_counter_offset() +
365 InvocationCounter::counter_offset());
366
367 __ get_method_counters(rbx, rax, done);
368 __ increment_mask_and_jump(invocation_counter, increment, mask,
369 rcx, false, Assembler::zero, overflow);
370 __ bind(done);
371 } else {
372 const Address backedge_counter (rax,
373 MethodCounters::backedge_counter_offset() +
374 InvocationCounter::counter_offset());
375 const Address invocation_counter(rax,
376 MethodCounters::invocation_counter_offset() +
377 InvocationCounter::counter_offset());
378
379 __ get_method_counters(rbx, rax, done);
380
381 if (ProfileInterpreter) {
382 __ incrementl(Address(rax,
383 MethodCounters::interpreter_invocation_counter_offset()));
384 }
385
386 // Update standard invocation counters
387 __ movl(rcx, invocation_counter);
388 __ incrementl(rcx, InvocationCounter::count_increment);
389 __ movl(invocation_counter, rcx); // save invocation count
390
391 __ movl(rax, backedge_counter); // load backedge counter
392 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
393
394 __ addl(rcx, rax); // add both counters
395
396 // profile_method is non-null only for interpreted method so
397 // profile_method != NULL == !native_call
398 // BytecodeInterpreter only calls for native so code is elided.
399
400 if (ProfileInterpreter && profile_method != NULL) {
401 // Test to see if we should create a method data oop
402 __ cmp32(rcx,
403 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
404 __ jcc(Assembler::less, *profile_method_continue);
405
406 // if no method data exists, go to profile_method
407 __ test_method_data_pointer(rax, *profile_method);
408 }
409
410 __ cmp32(rcx,
411 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
412 __ jcc(Assembler::aboveEqual, *overflow);
413 __ bind(done);
414 }
415 }
416
417 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
418
419 // Asm interpreter on entry
420 // rdi - locals
421 // rsi - bcp
422 // rbx, - method
423 // rdx - cpool
424 // rbp, - interpreter frame
425
426 // C++ interpreter on entry
427 // rsi - new interpreter state pointer
428 // rbp - interpreter frame pointer
429 // rbx - method
430
431 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
432 // rbx, - method
433 // rcx - rcvr (assuming there is one)
434 // top of stack return address of interpreter caller
435 // rsp - sender_sp
436
437 // C++ interpreter only
438 // rsi - previous interpreter state pointer
439
440 // InterpreterRuntime::frequency_counter_overflow takes one argument
441 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
442 // The call returns the address of the verified entry point for the method or NULL
443 // if the compilation did not complete (either went background or bailed out).
444 __ movptr(rax, (intptr_t)false);
445 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
446
447 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
448
449 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
450 // and jump to the interpreted entry.
451 __ jmp(*do_continue, relocInfo::none);
452
453 }
454
455 void InterpreterGenerator::generate_stack_overflow_check(void) {
456 // see if we've got enough room on the stack for locals plus overhead.
457 // the expression stack grows down incrementally, so the normal guard
458 // page mechanism will work for that.
459 //
460 // Registers live on entry:
461 //
462 // Asm interpreter
463 // rdx: number of additional locals this frame needs (what we must check)
464 // rbx,: Method*
465
466 // destroyed on exit
467 // rax,
468
469 // NOTE: since the additional locals are also always pushed (wasn't obvious in
470 // generate_method_entry) so the guard should work for them too.
471 //
472
473 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
474 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
475
476 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
477 // be sure to change this if you add/subtract anything to/from the overhead area
478 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
479
480 const int page_size = os::vm_page_size();
481
482 Label after_frame_check;
483
484 // see if the frame is greater than one page in size. If so,
485 // then we need to verify there is enough stack space remaining
486 // for the additional locals.
487 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
488 __ jcc(Assembler::belowEqual, after_frame_check);
489
490 // compute rsp as if this were going to be the last frame on
491 // the stack before the red zone
492
493 Label after_frame_check_pop;
494
495 __ push(rsi);
496
497 const Register thread = rsi;
498
499 __ get_thread(thread);
500
501 const Address stack_base(thread, Thread::stack_base_offset());
502 const Address stack_size(thread, Thread::stack_size_offset());
503
504 // locals + overhead, in bytes
505 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
506
507 #ifdef ASSERT
508 Label stack_base_okay, stack_size_okay;
509 // verify that thread stack base is non-zero
510 __ cmpptr(stack_base, (int32_t)NULL_WORD);
511 __ jcc(Assembler::notEqual, stack_base_okay);
512 __ stop("stack base is zero");
513 __ bind(stack_base_okay);
514 // verify that thread stack size is non-zero
515 __ cmpptr(stack_size, 0);
516 __ jcc(Assembler::notEqual, stack_size_okay);
517 __ stop("stack size is zero");
518 __ bind(stack_size_okay);
519 #endif
520
521 // Add stack base to locals and subtract stack size
522 __ addptr(rax, stack_base);
523 __ subptr(rax, stack_size);
524
525 // Use the maximum number of pages we might bang.
526 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
527 (StackRedPages+StackYellowPages);
528 __ addptr(rax, max_pages * page_size);
529
530 // check against the current stack bottom
531 __ cmpptr(rsp, rax);
532 __ jcc(Assembler::above, after_frame_check_pop);
533
534 __ pop(rsi); // get saved bcp / (c++ prev state ).
535
536 // Restore sender's sp as SP. This is necessary if the sender's
537 // frame is an extended compiled frame (see gen_c2i_adapter())
538 // and safer anyway in case of JSR292 adaptations.
539
540 __ pop(rax); // return address must be moved if SP is changed
541 __ mov(rsp, rsi);
542 __ push(rax);
543
544 // Note: the restored frame is not necessarily interpreted.
545 // Use the shared runtime version of the StackOverflowError.
546 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
547 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
548 // all done with frame size check
549 __ bind(after_frame_check_pop);
550 __ pop(rsi);
551
552 __ bind(after_frame_check);
553 }
554
555 // Allocate monitor and lock method (asm interpreter)
556 // rbx, - Method*
557 //
558 void InterpreterGenerator::lock_method(void) {
559 // synchronize method
560 const Address access_flags (rbx, Method::access_flags_offset());
561 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
562 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
563
564 #ifdef ASSERT
565 { Label L;
566 __ movl(rax, access_flags);
567 __ testl(rax, JVM_ACC_SYNCHRONIZED);
568 __ jcc(Assembler::notZero, L);
569 __ stop("method doesn't need synchronization");
570 __ bind(L);
571 }
572 #endif // ASSERT
573 // get synchronization object
574 { Label done;
575 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
576 __ movl(rax, access_flags);
577 __ testl(rax, JVM_ACC_STATIC);
578 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
579 __ jcc(Assembler::zero, done);
580 __ movptr(rax, Address(rbx, Method::const_offset()));
581 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
582 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
583 __ movptr(rax, Address(rax, mirror_offset));
584 __ bind(done);
585 }
586 // add space for monitor & lock
587 __ subptr(rsp, entry_size); // add space for a monitor entry
588 __ movptr(monitor_block_top, rsp); // set new monitor block top
589 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
590 __ mov(rdx, rsp); // object address
591 __ lock_object(rdx);
592 }
593
594 //
595 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
596 // and for native methods hence the shared code.
597
598 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
599 // initialize fixed part of activation frame
600 __ push(rax); // save return address
601 __ enter(); // save old & set new rbp,
602
603
604 __ push(rsi); // set sender sp
605 __ push((int32_t)NULL_WORD); // leave last_sp as null
606 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
607 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
608 __ push(rbx); // save Method*
609 if (ProfileInterpreter) {
610 Label method_data_continue;
611 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
612 __ testptr(rdx, rdx);
613 __ jcc(Assembler::zero, method_data_continue);
614 __ addptr(rdx, in_bytes(MethodData::data_offset()));
615 __ bind(method_data_continue);
616 __ push(rdx); // set the mdp (method data pointer)
617 } else {
618 __ push(0);
619 }
620
621 __ movptr(rdx, Address(rbx, Method::const_offset()));
622 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
623 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
624 __ push(rdx); // set constant pool cache
625 __ push(rdi); // set locals pointer
626 if (native_call) {
627 __ push(0); // no bcp
628 } else {
629 __ push(rsi); // set bcp
630 }
631 __ push(0); // reserve word for pointer to expression stack bottom
632 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
633 }
634
635 // End of helpers
636
637 //
638 // Various method entries
639 //------------------------------------------------------------------------------------------------------------------------
640 //
641 //
642
643 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
644
645 address InterpreterGenerator::generate_accessor_entry(void) {
646
647 // rbx,: Method*
648 // rcx: receiver (preserve for slow entry into asm interpreter)
649
650 // rsi: senderSP must preserved for slow path, set SP to it on fast path
651
652 address entry_point = __ pc();
653 Label xreturn_path;
654
655 // do fastpath for resolved accessor methods
656 if (UseFastAccessorMethods) {
657 Label slow_path;
658 // If we need a safepoint check, generate full interpreter entry.
659 ExternalAddress state(SafepointSynchronize::address_of_state());
660 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
661 SafepointSynchronize::_not_synchronized);
662
663 __ jcc(Assembler::notEqual, slow_path);
664 // ASM/C++ Interpreter
665 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
666 // Note: We can only use this code if the getfield has been resolved
667 // and if we don't have a null-pointer exception => check for
668 // these conditions first and use slow path if necessary.
669 // rbx,: method
670 // rcx: receiver
671 __ movptr(rax, Address(rsp, wordSize));
672
673 // check if local 0 != NULL and read field
674 __ testptr(rax, rax);
675 __ jcc(Assembler::zero, slow_path);
676
677 // read first instruction word and extract bytecode @ 1 and index @ 2
678 __ movptr(rdx, Address(rbx, Method::const_offset()));
679 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
680 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
681 // Shift codes right to get the index on the right.
682 // The bytecode fetched looks like <index><0xb4><0x2a>
683 __ shrl(rdx, 2*BitsPerByte);
684 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
685 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
686
687 // rax,: local 0
688 // rbx,: method
689 // rcx: receiver - do not destroy since it is needed for slow path!
690 // rcx: scratch
691 // rdx: constant pool cache index
692 // rdi: constant pool cache
693 // rsi: sender sp
694
695 // check if getfield has been resolved and read constant pool cache entry
696 // check the validity of the cache entry by testing whether _indices field
697 // contains Bytecode::_getfield in b1 byte.
698 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
699 __ movl(rcx,
700 Address(rdi,
701 rdx,
702 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
703 __ shrl(rcx, 2*BitsPerByte);
704 __ andl(rcx, 0xFF);
705 __ cmpl(rcx, Bytecodes::_getfield);
706 __ jcc(Assembler::notEqual, slow_path);
707
708 // Note: constant pool entry is not valid before bytecode is resolved
709 __ movptr(rcx,
710 Address(rdi,
711 rdx,
712 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
713 __ movl(rdx,
714 Address(rdi,
715 rdx,
716 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
717
718 Label notByte, notShort, notChar;
719 const Address field_address (rax, rcx, Address::times_1);
720
721 // Need to differentiate between igetfield, agetfield, bgetfield etc.
722 // because they are different sizes.
723 // Use the type from the constant pool cache
724 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
725 // Make sure we don't need to mask rdx after the above shift
726 ConstantPoolCacheEntry::verify_tos_state_shift();
727 __ cmpl(rdx, btos);
728 __ jcc(Assembler::notEqual, notByte);
729 __ load_signed_byte(rax, field_address);
730 __ jmp(xreturn_path);
731
732 __ bind(notByte);
733 __ cmpl(rdx, stos);
734 __ jcc(Assembler::notEqual, notShort);
735 __ load_signed_short(rax, field_address);
736 __ jmp(xreturn_path);
737
738 __ bind(notShort);
739 __ cmpl(rdx, ctos);
740 __ jcc(Assembler::notEqual, notChar);
741 __ load_unsigned_short(rax, field_address);
742 __ jmp(xreturn_path);
743
744 __ bind(notChar);
745 #ifdef ASSERT
746 Label okay;
747 __ cmpl(rdx, atos);
748 __ jcc(Assembler::equal, okay);
749 __ cmpl(rdx, itos);
750 __ jcc(Assembler::equal, okay);
751 __ stop("what type is this?");
752 __ bind(okay);
753 #endif // ASSERT
754 // All the rest are a 32 bit wordsize
755 // This is ok for now. Since fast accessors should be going away
756 __ movptr(rax, field_address);
757
758 __ bind(xreturn_path);
759
760 // _ireturn/_areturn
761 __ pop(rdi); // get return address
762 __ mov(rsp, rsi); // set sp to sender sp
763 __ jmp(rdi);
764
765 // generate a vanilla interpreter entry as the slow path
766 __ bind(slow_path);
767
768 (void) generate_normal_entry(false);
769 return entry_point;
770 }
771 return NULL;
772
773 }
774
775 // Method entry for java.lang.ref.Reference.get.
776 address InterpreterGenerator::generate_Reference_get_entry(void) {
777 #if INCLUDE_ALL_GCS
778 // Code: _aload_0, _getfield, _areturn
779 // parameter size = 1
780 //
781 // The code that gets generated by this routine is split into 2 parts:
782 // 1. The "intrinsified" code for G1 (or any SATB based GC),
783 // 2. The slow path - which is an expansion of the regular method entry.
784 //
785 // Notes:-
786 // * In the G1 code we do not check whether we need to block for
787 // a safepoint. If G1 is enabled then we must execute the specialized
788 // code for Reference.get (except when the Reference object is null)
789 // so that we can log the value in the referent field with an SATB
790 // update buffer.
791 // If the code for the getfield template is modified so that the
792 // G1 pre-barrier code is executed when the current method is
793 // Reference.get() then going through the normal method entry
794 // will be fine.
795 // * The G1 code below can, however, check the receiver object (the instance
796 // of java.lang.Reference) and jump to the slow path if null. If the
797 // Reference object is null then we obviously cannot fetch the referent
798 // and so we don't need to call the G1 pre-barrier. Thus we can use the
799 // regular method entry code to generate the NPE.
800 //
801 // This code is based on generate_accessor_enty.
802
803 // rbx,: Method*
804 // rcx: receiver (preserve for slow entry into asm interpreter)
805
806 // rsi: senderSP must preserved for slow path, set SP to it on fast path
807
808 address entry = __ pc();
809
810 const int referent_offset = java_lang_ref_Reference::referent_offset;
811 guarantee(referent_offset > 0, "referent offset not initialized");
812
813 if (UseG1GC) {
814 Label slow_path;
815
816 // Check if local 0 != NULL
817 // If the receiver is null then it is OK to jump to the slow path.
818 __ movptr(rax, Address(rsp, wordSize));
819 __ testptr(rax, rax);
820 __ jcc(Assembler::zero, slow_path);
821
822 // rax: local 0 (must be preserved across the G1 barrier call)
823 //
824 // rbx: method (at this point it's scratch)
825 // rcx: receiver (at this point it's scratch)
826 // rdx: scratch
827 // rdi: scratch
828 //
829 // rsi: sender sp
830
831 // Preserve the sender sp in case the pre-barrier
832 // calls the runtime
833 __ push(rsi);
834
835 // Load the value of the referent field.
836 const Address field_address(rax, referent_offset);
837 __ movptr(rax, field_address);
838
839 // Generate the G1 pre-barrier code to log the value of
840 // the referent field in an SATB buffer.
841 __ get_thread(rcx);
842 __ g1_write_barrier_pre(noreg /* obj */,
843 rax /* pre_val */,
844 rcx /* thread */,
845 rbx /* tmp */,
846 true /* tosca_save */,
847 true /* expand_call */);
848
849 // _areturn
850 __ pop(rsi); // get sender sp
851 __ pop(rdi); // get return address
852 __ mov(rsp, rsi); // set sp to sender sp
853 __ jmp(rdi);
854
855 __ bind(slow_path);
856 (void) generate_normal_entry(false);
857
858 return entry;
859 }
860 #endif // INCLUDE_ALL_GCS
861
862 // If G1 is not enabled then attempt to go through the accessor entry point
863 // Reference.get is an accessor
864 return generate_accessor_entry();
865 }
866
867 /**
868 * Method entry for static native methods:
869 * int java.util.zip.CRC32.update(int crc, int b)
870 */
871 address InterpreterGenerator::generate_CRC32_update_entry() {
872 if (UseCRC32Intrinsics) {
873 address entry = __ pc();
874
875 // rbx,: Method*
876 // rsi: senderSP must preserved for slow path, set SP to it on fast path
877 // rdx: scratch
878 // rdi: scratch
879
880 Label slow_path;
881 // If we need a safepoint check, generate full interpreter entry.
882 ExternalAddress state(SafepointSynchronize::address_of_state());
883 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
884 SafepointSynchronize::_not_synchronized);
885 __ jcc(Assembler::notEqual, slow_path);
886
887 // We don't generate local frame and don't align stack because
888 // we call stub code and there is no safepoint on this path.
889
890 // Load parameters
891 const Register crc = rax; // crc
892 const Register val = rdx; // source java byte value
893 const Register tbl = rdi; // scratch
894
895 // Arguments are reversed on java expression stack
896 __ movl(val, Address(rsp, wordSize)); // byte value
897 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
898
899 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
900 __ notl(crc); // ~crc
901 __ update_byte_crc32(crc, val, tbl);
902 __ notl(crc); // ~crc
903 // result in rax
904
905 // _areturn
906 __ pop(rdi); // get return address
907 __ mov(rsp, rsi); // set sp to sender sp
908 __ jmp(rdi);
909
910 // generate a vanilla native entry as the slow path
911 __ bind(slow_path);
912
913 (void) generate_native_entry(false);
914
915 return entry;
916 }
917 return generate_native_entry(false);
918 }
919
920 /**
921 * Method entry for static native methods:
922 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
923 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
924 */
925 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
926 if (UseCRC32Intrinsics) {
927 address entry = __ pc();
928
929 // rbx,: Method*
930 // rsi: senderSP must preserved for slow path, set SP to it on fast path
931 // rdx: scratch
932 // rdi: scratch
933
934 Label slow_path;
935 // If we need a safepoint check, generate full interpreter entry.
936 ExternalAddress state(SafepointSynchronize::address_of_state());
937 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
938 SafepointSynchronize::_not_synchronized);
939 __ jcc(Assembler::notEqual, slow_path);
940
941 // We don't generate local frame and don't align stack because
942 // we call stub code and there is no safepoint on this path.
943
944 // Load parameters
945 const Register crc = rax; // crc
946 const Register buf = rdx; // source java byte array address
947 const Register len = rdi; // length
948
949 // Arguments are reversed on java expression stack
950 __ movl(len, Address(rsp, wordSize)); // Length
951 // Calculate address of start element
952 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
953 __ movptr(buf, Address(rsp, 3*wordSize)); // long buf
954 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
955 __ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
956 } else {
957 __ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
958 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
959 __ addptr(buf, Address(rsp, 2*wordSize)); // + offset
960 __ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
961 }
962
963 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
964 // result in rax
965
966 // _areturn
967 __ pop(rdi); // get return address
968 __ mov(rsp, rsi); // set sp to sender sp
969 __ jmp(rdi);
970
971 // generate a vanilla native entry as the slow path
972 __ bind(slow_path);
973
974 (void) generate_native_entry(false);
975
976 return entry;
977 }
978 return generate_native_entry(false);
979 }
980
981 //
982 // Interpreter stub for calling a native method. (asm interpreter)
983 // This sets up a somewhat different looking stack for calling the native method
984 // than the typical interpreter frame setup.
985 //
986
987 address InterpreterGenerator::generate_native_entry(bool synchronized) {
988 // determine code generation flags
989 bool inc_counter = UseCompiler || CountCompiledCalls;
990
991 // rbx,: Method*
992 // rsi: sender sp
993 // rsi: previous interpreter state (C++ interpreter) must preserve
994 address entry_point = __ pc();
995
996 const Address constMethod (rbx, Method::const_offset());
997 const Address access_flags (rbx, Method::access_flags_offset());
998 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
999
1000 // get parameter size (always needed)
1001 __ movptr(rcx, constMethod);
1002 __ load_unsigned_short(rcx, size_of_parameters);
1003
1004 // native calls don't need the stack size check since they have no expression stack
1005 // and the arguments are already on the stack and we only add a handful of words
1006 // to the stack
1007
1008 // rbx,: Method*
1009 // rcx: size of parameters
1010 // rsi: sender sp
1011
1012 __ pop(rax); // get return address
1013 // for natives the size of locals is zero
1014
1015 // compute beginning of parameters (rdi)
1016 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1017
1018
1019 // add 2 zero-initialized slots for native calls
1020 // NULL result handler
1021 __ push((int32_t)NULL_WORD);
1022 // NULL oop temp (mirror or jni oop result)
1023 __ push((int32_t)NULL_WORD);
1024
1025 // initialize fixed part of activation frame
1026 generate_fixed_frame(true);
1027
1028 // make sure method is native & not abstract
1029 #ifdef ASSERT
1030 __ movl(rax, access_flags);
1031 {
1032 Label L;
1033 __ testl(rax, JVM_ACC_NATIVE);
1034 __ jcc(Assembler::notZero, L);
1035 __ stop("tried to execute non-native method as native");
1036 __ bind(L);
1037 }
1038 { Label L;
1039 __ testl(rax, JVM_ACC_ABSTRACT);
1040 __ jcc(Assembler::zero, L);
1041 __ stop("tried to execute abstract method in interpreter");
1042 __ bind(L);
1043 }
1044 #endif
1045
1046 // Since at this point in the method invocation the exception handler
1047 // would try to exit the monitor of synchronized methods which hasn't
1048 // been entered yet, we set the thread local variable
1049 // _do_not_unlock_if_synchronized to true. The remove_activation will
1050 // check this flag.
1051
1052 __ get_thread(rax);
1053 const Address do_not_unlock_if_synchronized(rax,
1054 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1055 __ movbool(do_not_unlock_if_synchronized, true);
1056
1057 // increment invocation count & check for overflow
1058 Label invocation_counter_overflow;
1059 if (inc_counter) {
1060 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1061 }
1062
1063 Label continue_after_compile;
1064 __ bind(continue_after_compile);
1065
1066 bang_stack_shadow_pages(true);
1067
1068 // reset the _do_not_unlock_if_synchronized flag
1069 __ get_thread(rax);
1070 __ movbool(do_not_unlock_if_synchronized, false);
1071
1072 // check for synchronized methods
1073 // Must happen AFTER invocation_counter check and stack overflow check,
1074 // so method is not locked if overflows.
1075 //
1076 if (synchronized) {
1077 lock_method();
1078 } else {
1079 // no synchronization necessary
1080 #ifdef ASSERT
1081 { Label L;
1082 __ movl(rax, access_flags);
1083 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1084 __ jcc(Assembler::zero, L);
1085 __ stop("method needs synchronization");
1086 __ bind(L);
1087 }
1088 #endif
1089 }
1090
1091 // start execution
1092 #ifdef ASSERT
1093 { Label L;
1094 const Address monitor_block_top (rbp,
1095 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1096 __ movptr(rax, monitor_block_top);
1097 __ cmpptr(rax, rsp);
1098 __ jcc(Assembler::equal, L);
1099 __ stop("broken stack frame setup in interpreter");
1100 __ bind(L);
1101 }
1102 #endif
1103
1104 // jvmti/dtrace support
1105 __ notify_method_entry();
1106
1107 // work registers
1108 const Register method = rbx;
1109 const Register thread = rdi;
1110 const Register t = rcx;
1111
1112 // allocate space for parameters
1113 __ get_method(method);
1114 __ movptr(t, Address(method, Method::const_offset()));
1115 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1116
1117 __ shlptr(t, Interpreter::logStackElementSize);
1118 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1119 __ subptr(rsp, t);
1120 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1121
1122 // get signature handler
1123 { Label L;
1124 __ movptr(t, Address(method, Method::signature_handler_offset()));
1125 __ testptr(t, t);
1126 __ jcc(Assembler::notZero, L);
1127 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1128 __ get_method(method);
1129 __ movptr(t, Address(method, Method::signature_handler_offset()));
1130 __ bind(L);
1131 }
1132
1133 // call signature handler
1134 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1135 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1136 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1137 // The generated handlers do not touch RBX (the method oop).
1138 // However, large signatures cannot be cached and are generated
1139 // each time here. The slow-path generator will blow RBX
1140 // sometime, so we must reload it after the call.
1141 __ call(t);
1142 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1143
1144 // result handler is in rax,
1145 // set result handler
1146 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1147
1148 // pass mirror handle if static call
1149 { Label L;
1150 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1151 __ movl(t, Address(method, Method::access_flags_offset()));
1152 __ testl(t, JVM_ACC_STATIC);
1153 __ jcc(Assembler::zero, L);
1154 // get mirror
1155 __ movptr(t, Address(method, Method:: const_offset()));
1156 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1157 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1158 __ movptr(t, Address(t, mirror_offset));
1159 // copy mirror into activation frame
1160 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1161 // pass handle to mirror
1162 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1163 __ movptr(Address(rsp, wordSize), t);
1164 __ bind(L);
1165 }
1166
1167 // get native function entry point
1168 { Label L;
1169 __ movptr(rax, Address(method, Method::native_function_offset()));
1170 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1171 __ cmpptr(rax, unsatisfied.addr());
1172 __ jcc(Assembler::notEqual, L);
1173 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1174 __ get_method(method);
1175 __ movptr(rax, Address(method, Method::native_function_offset()));
1176 __ bind(L);
1177 }
1178
1179 // pass JNIEnv
1180 __ get_thread(thread);
1181 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1182 __ movptr(Address(rsp, 0), t);
1183
1184 // set_last_Java_frame_before_call
1185 // It is enough that the pc()
1186 // points into the right code segment. It does not have to be the correct return pc.
1187 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1188
1189 // change thread state
1190 #ifdef ASSERT
1191 { Label L;
1192 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1193 __ cmpl(t, _thread_in_Java);
1194 __ jcc(Assembler::equal, L);
1195 __ stop("Wrong thread state in native stub");
1196 __ bind(L);
1197 }
1198 #endif
1199
1200 // Change state to native
1201 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1202 __ call(rax);
1203
1204 // result potentially in rdx:rax or ST0
1205
1206 // Verify or restore cpu control state after JNI call
1207 __ restore_cpu_control_state_after_jni();
1208
1209 // save potential result in ST(0) & rdx:rax
1210 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1211 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1212 // It is safe to do this push because state is _thread_in_native and return address will be found
1213 // via _last_native_pc and not via _last_jave_sp
1214
1215 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1216 // If the order changes or anything else is added to the stack the code in
1217 // interpreter_frame_result will have to be changed.
1218
1219 { Label L;
1220 Label push_double;
1221 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1222 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1223 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1224 float_handler.addr());
1225 __ jcc(Assembler::equal, push_double);
1226 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1227 double_handler.addr());
1228 __ jcc(Assembler::notEqual, L);
1229 __ bind(push_double);
1230 __ push(dtos);
1231 __ bind(L);
1232 }
1233 __ push(ltos);
1234
1235 // change thread state
1236 __ get_thread(thread);
1237 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1238 if(os::is_MP()) {
1239 if (UseMembar) {
1240 // Force this write out before the read below
1241 __ membar(Assembler::Membar_mask_bits(
1242 Assembler::LoadLoad | Assembler::LoadStore |
1243 Assembler::StoreLoad | Assembler::StoreStore));
1244 } else {
1245 // Write serialization page so VM thread can do a pseudo remote membar.
1246 // We use the current thread pointer to calculate a thread specific
1247 // offset to write to within the page. This minimizes bus traffic
1248 // due to cache line collision.
1249 __ serialize_memory(thread, rcx);
1250 }
1251 }
1252
1253 if (AlwaysRestoreFPU) {
1254 // Make sure the control word is correct.
1255 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1256 }
1257
1258 // check for safepoint operation in progress and/or pending suspend requests
1259 { Label Continue;
1260
1261 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1262 SafepointSynchronize::_not_synchronized);
1263
1264 Label L;
1265 __ jcc(Assembler::notEqual, L);
1266 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1267 __ jcc(Assembler::equal, Continue);
1268 __ bind(L);
1269
1270 // Don't use call_VM as it will see a possible pending exception and forward it
1271 // and never return here preventing us from clearing _last_native_pc down below.
1272 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1273 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1274 // by hand.
1275 //
1276 __ push(thread);
1277 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1278 JavaThread::check_special_condition_for_native_trans)));
1279 __ increment(rsp, wordSize);
1280 __ get_thread(thread);
1281
1282 __ bind(Continue);
1283 }
1284
1285 // change thread state
1286 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1287
1288 __ reset_last_Java_frame(thread, true, true);
1289
1290 // reset handle block
1291 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1292 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1293
1294 // If result was an oop then unbox and save it in the frame
1295 { Label L;
1296 Label no_oop, store_result;
1297 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1298 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1299 handler.addr());
1300 __ jcc(Assembler::notEqual, no_oop);
1301 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1302 __ pop(ltos);
1303 __ testptr(rax, rax);
1304 __ jcc(Assembler::zero, store_result);
1305 // unbox
1306 __ movptr(rax, Address(rax, 0));
1307 __ bind(store_result);
1308 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1309 // keep stack depth as expected by pushing oop which will eventually be discarded
1310 __ push(ltos);
1311 __ bind(no_oop);
1312 }
1313
1314 {
1315 Label no_reguard;
1316 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1317 __ jcc(Assembler::notEqual, no_reguard);
1318
1319 __ pusha();
1320 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1321 __ popa();
1322
1323 __ bind(no_reguard);
1324 }
1325
1326 // restore rsi to have legal interpreter frame,
1327 // i.e., bci == 0 <=> rsi == code_base()
1328 // Can't call_VM until bcp is within reasonable.
1329 __ get_method(method); // method is junk from thread_in_native to now.
1330 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1331 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1332
1333 // handle exceptions (exception handling will handle unlocking!)
1334 { Label L;
1335 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1336 __ jcc(Assembler::zero, L);
1337 // Note: At some point we may want to unify this with the code used in call_VM_base();
1338 // i.e., we should use the StubRoutines::forward_exception code. For now this
1339 // doesn't work here because the rsp is not correctly set at this point.
1340 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1341 __ should_not_reach_here();
1342 __ bind(L);
1343 }
1344
1345 // do unlocking if necessary
1346 { Label L;
1347 __ movl(t, Address(method, Method::access_flags_offset()));
1348 __ testl(t, JVM_ACC_SYNCHRONIZED);
1349 __ jcc(Assembler::zero, L);
1350 // the code below should be shared with interpreter macro assembler implementation
1351 { Label unlock;
1352 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1353 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1354 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1355
1356 __ lea(rdx, monitor); // address of first monitor
1357
1358 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1359 __ testptr(t, t);
1360 __ jcc(Assembler::notZero, unlock);
1361
1362 // Entry already unlocked, need to throw exception
1363 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1364 __ should_not_reach_here();
1365
1366 __ bind(unlock);
1367 __ unlock_object(rdx);
1368 }
1369 __ bind(L);
1370 }
1371
1372 // jvmti/dtrace support
1373 // Note: This must happen _after_ handling/throwing any exceptions since
1374 // the exception handler code notifies the runtime of method exits
1375 // too. If this happens before, method entry/exit notifications are
1376 // not properly paired (was bug - gri 11/22/99).
1377 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1378
1379 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1380 __ pop(ltos);
1381 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1382 __ call(t);
1383
1384 // remove activation
1385 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1386 __ leave(); // remove frame anchor
1387 __ pop(rdi); // get return address
1388 __ mov(rsp, t); // set sp to sender sp
1389 __ jmp(rdi);
1390
1391 if (inc_counter) {
1392 // Handle overflow of counter and compile method
1393 __ bind(invocation_counter_overflow);
1394 generate_counter_overflow(&continue_after_compile);
1395 }
1396
1397 return entry_point;
1398 }
1399
1400 //
1401 // Generic interpreted method entry to (asm) interpreter
1402 //
1403 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1404 // determine code generation flags
1405 bool inc_counter = UseCompiler || CountCompiledCalls;
1406
1407 // rbx,: Method*
1408 // rsi: sender sp
1409 address entry_point = __ pc();
1410
1411 const Address constMethod (rbx, Method::const_offset());
1412 const Address access_flags (rbx, Method::access_flags_offset());
1413 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1414 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1415
1416 // get parameter size (always needed)
1417 __ movptr(rdx, constMethod);
1418 __ load_unsigned_short(rcx, size_of_parameters);
1419
1420 // rbx,: Method*
1421 // rcx: size of parameters
1422
1423 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1424
1425 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1426 __ subl(rdx, rcx); // rdx = no. of additional locals
1427
1428 // see if we've got enough room on the stack for locals plus overhead.
1429 generate_stack_overflow_check();
1430
1431 // get return address
1432 __ pop(rax);
1433
1434 // compute beginning of parameters (rdi)
1435 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1436
1437 // rdx - # of additional locals
1438 // allocate space for locals
1439 // explicitly initialize locals
1440 {
1441 Label exit, loop;
1442 __ testl(rdx, rdx);
1443 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1444 __ bind(loop);
1445 __ push((int32_t)NULL_WORD); // initialize local variables
1446 __ decrement(rdx); // until everything initialized
1447 __ jcc(Assembler::greater, loop);
1448 __ bind(exit);
1449 }
1450
1451 // initialize fixed part of activation frame
1452 generate_fixed_frame(false);
1453
1454 // make sure method is not native & not abstract
1455 #ifdef ASSERT
1456 __ movl(rax, access_flags);
1457 {
1458 Label L;
1459 __ testl(rax, JVM_ACC_NATIVE);
1460 __ jcc(Assembler::zero, L);
1461 __ stop("tried to execute native method as non-native");
1462 __ bind(L);
1463 }
1464 { Label L;
1465 __ testl(rax, JVM_ACC_ABSTRACT);
1466 __ jcc(Assembler::zero, L);
1467 __ stop("tried to execute abstract method in interpreter");
1468 __ bind(L);
1469 }
1470 #endif
1471
1472 // Since at this point in the method invocation the exception handler
1473 // would try to exit the monitor of synchronized methods which hasn't
1474 // been entered yet, we set the thread local variable
1475 // _do_not_unlock_if_synchronized to true. The remove_activation will
1476 // check this flag.
1477
1478 __ get_thread(rax);
1479 const Address do_not_unlock_if_synchronized(rax,
1480 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1481 __ movbool(do_not_unlock_if_synchronized, true);
1482
1483 __ profile_parameters_type(rax, rcx, rdx);
1484 // increment invocation count & check for overflow
1485 Label invocation_counter_overflow;
1486 Label profile_method;
1487 Label profile_method_continue;
1488 if (inc_counter) {
1489 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1490 if (ProfileInterpreter) {
1491 __ bind(profile_method_continue);
1492 }
1493 }
1494 Label continue_after_compile;
1495 __ bind(continue_after_compile);
1496
1497 bang_stack_shadow_pages(false);
1498
1499 // reset the _do_not_unlock_if_synchronized flag
1500 __ get_thread(rax);
1501 __ movbool(do_not_unlock_if_synchronized, false);
1502
1503 // check for synchronized methods
1504 // Must happen AFTER invocation_counter check and stack overflow check,
1505 // so method is not locked if overflows.
1506 //
1507 if (synchronized) {
1508 // Allocate monitor and lock method
1509 lock_method();
1510 } else {
1511 // no synchronization necessary
1512 #ifdef ASSERT
1513 { Label L;
1514 __ movl(rax, access_flags);
1515 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1516 __ jcc(Assembler::zero, L);
1517 __ stop("method needs synchronization");
1518 __ bind(L);
1519 }
1520 #endif
1521 }
1522
1523 // start execution
1524 #ifdef ASSERT
1525 { Label L;
1526 const Address monitor_block_top (rbp,
1527 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1528 __ movptr(rax, monitor_block_top);
1529 __ cmpptr(rax, rsp);
1530 __ jcc(Assembler::equal, L);
1531 __ stop("broken stack frame setup in interpreter");
1532 __ bind(L);
1533 }
1534 #endif
1535
1536 // jvmti support
1537 __ notify_method_entry();
1538
1539 __ dispatch_next(vtos);
1540
1541 // invocation counter overflow
1542 if (inc_counter) {
1543 if (ProfileInterpreter) {
1544 // We have decided to profile this method in the interpreter
1545 __ bind(profile_method);
1546 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1547 __ set_method_data_pointer_for_bcp();
1548 __ get_method(rbx);
1549 __ jmp(profile_method_continue);
1550 }
1551 // Handle overflow of counter and compile method
1552 __ bind(invocation_counter_overflow);
1553 generate_counter_overflow(&continue_after_compile);
1554 }
1555
1556 return entry_point;
1557 }
1558
1559 //------------------------------------------------------------------------------------------------------------------------
1560 // Entry points
1561 //
1562 // Here we generate the various kind of entries into the interpreter.
1563 // The two main entry type are generic bytecode methods and native call method.
1564 // These both come in synchronized and non-synchronized versions but the
1565 // frame layout they create is very similar. The other method entry
1566 // types are really just special purpose entries that are really entry
1567 // and interpretation all in one. These are for trivial methods like
1568 // accessor, empty, or special math methods.
1569 //
1570 // When control flow reaches any of the entry types for the interpreter
1571 // the following holds ->
1572 //
1573 // Arguments:
1574 //
1575 // rbx,: Method*
1576 // rcx: receiver
1577 //
1578 //
1579 // Stack layout immediately at entry
1580 //
1581 // [ return address ] <--- rsp
1582 // [ parameter n ]
1583 // ...
1584 // [ parameter 1 ]
1585 // [ expression stack ] (caller's java expression stack)
1586
1587 // Assuming that we don't go to one of the trivial specialized
1588 // entries the stack will look like below when we are ready to execute
1589 // the first bytecode (or call the native routine). The register usage
1590 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1591 //
1592 // local variables follow incoming parameters immediately; i.e.
1593 // the return address is moved to the end of the locals).
1594 //
1595 // [ monitor entry ] <--- rsp
1596 // ...
1597 // [ monitor entry ]
1598 // [ expr. stack bottom ]
1599 // [ saved rsi ]
1600 // [ current rdi ]
1601 // [ Method* ]
1602 // [ saved rbp, ] <--- rbp,
1603 // [ return address ]
1604 // [ local variable m ]
1605 // ...
1606 // [ local variable 1 ]
1607 // [ parameter n ]
1608 // ...
1609 // [ parameter 1 ] <--- rdi
1610
1611 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1612 // determine code generation flags
1613 bool synchronized = false;
1614 address entry_point = NULL;
1615 InterpreterGenerator* ig_this = (InterpreterGenerator*)this;
1616
1617 switch (kind) {
1618 case Interpreter::zerolocals : break;
1619 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1620 case Interpreter::native : entry_point = ig_this->generate_native_entry(false); break;
1621 case Interpreter::native_synchronized : entry_point = ig_this->generate_native_entry(true); break;
1622 case Interpreter::empty : entry_point = ig_this->generate_empty_entry(); break;
1623 case Interpreter::accessor : entry_point = ig_this->generate_accessor_entry(); break;
1624 case Interpreter::abstract : entry_point = ig_this->generate_abstract_entry(); break;
1625
1626 case Interpreter::java_lang_math_sin : // fall thru
1627 case Interpreter::java_lang_math_cos : // fall thru
1628 case Interpreter::java_lang_math_tan : // fall thru
1629 case Interpreter::java_lang_math_abs : // fall thru
1630 case Interpreter::java_lang_math_log : // fall thru
1631 case Interpreter::java_lang_math_log10 : // fall thru
1632 case Interpreter::java_lang_math_sqrt : // fall thru
1633 case Interpreter::java_lang_math_pow : // fall thru
1634 case Interpreter::java_lang_math_exp : entry_point = ig_this->generate_math_entry(kind); break;
1635 case Interpreter::java_lang_ref_reference_get
1636 : entry_point = ig_this->generate_Reference_get_entry(); break;
1637 case Interpreter::java_util_zip_CRC32_update
1638 : entry_point = ig_this->generate_CRC32_update_entry(); break;
1639 case Interpreter::java_util_zip_CRC32_updateBytes
1640 : // fall thru
1641 case Interpreter::java_util_zip_CRC32_updateByteBuffer
1642 : entry_point = ig_this->generate_CRC32_updateBytes_entry(kind); break;
1643 default:
1644 fatal(err_msg("unexpected method kind: %d", kind));
1645 break;
1646 }
1647
1648 if (entry_point) return entry_point;
1649
1650 return ig_this->generate_normal_entry(synchronized);
1651
1652 }
1653
1654 // These should never be compiled since the interpreter will prefer
1655 // the compiled version to the intrinsic version.
1656 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1657 switch (method_kind(m)) {
1658 case Interpreter::java_lang_math_sin : // fall thru
1659 case Interpreter::java_lang_math_cos : // fall thru
1660 case Interpreter::java_lang_math_tan : // fall thru
1661 case Interpreter::java_lang_math_abs : // fall thru
1662 case Interpreter::java_lang_math_log : // fall thru
1663 case Interpreter::java_lang_math_log10 : // fall thru
1664 case Interpreter::java_lang_math_sqrt : // fall thru
1665 case Interpreter::java_lang_math_pow : // fall thru
1666 case Interpreter::java_lang_math_exp :
1667 return false;
1668 default:
1669 return true;
1670 }
1671 }
1672
1673 // How much stack a method activation needs in words.
1674 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1675
1676 const int stub_code = 4; // see generate_call_stub
1677 // Save space for one monitor to get into the interpreted method in case
1678 // the method is synchronized
1679 int monitor_size = method->is_synchronized() ?
1680 1*frame::interpreter_frame_monitor_size() : 0;
1681
1682 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1683 // be sure to change this if you add/subtract anything to/from the overhead area
1684 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1685
1686 const int method_stack = (method->max_locals() + method->max_stack()) *
1687 Interpreter::stackElementWords;
1688 return overhead_size + method_stack + stub_code;
1689 }
1690
1691 //------------------------------------------------------------------------------------------------------------------------
1692 // Exceptions
1693
1694 void TemplateInterpreterGenerator::generate_throw_exception() {
1695 // Entry point in previous activation (i.e., if the caller was interpreted)
1696 Interpreter::_rethrow_exception_entry = __ pc();
1697 const Register thread = rcx;
1698
1699 // Restore sp to interpreter_frame_last_sp even though we are going
1700 // to empty the expression stack for the exception processing.
1701 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1702 // rax,: exception
1703 // rdx: return address/pc that threw exception
1704 __ restore_bcp(); // rsi points to call/send
1705 __ restore_locals();
1706
1707 // Entry point for exceptions thrown within interpreter code
1708 Interpreter::_throw_exception_entry = __ pc();
1709 // expression stack is undefined here
1710 // rax,: exception
1711 // rsi: exception bcp
1712 __ verify_oop(rax);
1713
1714 // expression stack must be empty before entering the VM in case of an exception
1715 __ empty_expression_stack();
1716 __ empty_FPU_stack();
1717 // find exception handler address and preserve exception oop
1718 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1719 // rax,: exception handler entry point
1720 // rdx: preserved exception oop
1721 // rsi: bcp for exception handler
1722 __ push_ptr(rdx); // push exception which is now the only value on the stack
1723 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1724
1725 // If the exception is not handled in the current frame the frame is removed and
1726 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1727 //
1728 // Note: At this point the bci is still the bxi for the instruction which caused
1729 // the exception and the expression stack is empty. Thus, for any VM calls
1730 // at this point, GC will find a legal oop map (with empty expression stack).
1731
1732 // In current activation
1733 // tos: exception
1734 // rsi: exception bcp
1735
1736 //
1737 // JVMTI PopFrame support
1738 //
1739
1740 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1741 __ empty_expression_stack();
1742 __ empty_FPU_stack();
1743 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1744 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1745 // popframe handling cycles.
1746 __ get_thread(thread);
1747 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1748 __ orl(rdx, JavaThread::popframe_processing_bit);
1749 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1750
1751 {
1752 // Check to see whether we are returning to a deoptimized frame.
1753 // (The PopFrame call ensures that the caller of the popped frame is
1754 // either interpreted or compiled and deoptimizes it if compiled.)
1755 // In this case, we can't call dispatch_next() after the frame is
1756 // popped, but instead must save the incoming arguments and restore
1757 // them after deoptimization has occurred.
1758 //
1759 // Note that we don't compare the return PC against the
1760 // deoptimization blob's unpack entry because of the presence of
1761 // adapter frames in C2.
1762 Label caller_not_deoptimized;
1763 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1764 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1765 __ testl(rax, rax);
1766 __ jcc(Assembler::notZero, caller_not_deoptimized);
1767
1768 // Compute size of arguments for saving when returning to deoptimized caller
1769 __ get_method(rax);
1770 __ movptr(rax, Address(rax, Method::const_offset()));
1771 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1772 __ shlptr(rax, Interpreter::logStackElementSize);
1773 __ restore_locals();
1774 __ subptr(rdi, rax);
1775 __ addptr(rdi, wordSize);
1776 // Save these arguments
1777 __ get_thread(thread);
1778 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1779
1780 __ remove_activation(vtos, rdx,
1781 /* throw_monitor_exception */ false,
1782 /* install_monitor_exception */ false,
1783 /* notify_jvmdi */ false);
1784
1785 // Inform deoptimization that it is responsible for restoring these arguments
1786 __ get_thread(thread);
1787 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1788
1789 // Continue in deoptimization handler
1790 __ jmp(rdx);
1791
1792 __ bind(caller_not_deoptimized);
1793 }
1794
1795 __ remove_activation(vtos, rdx,
1796 /* throw_monitor_exception */ false,
1797 /* install_monitor_exception */ false,
1798 /* notify_jvmdi */ false);
1799
1800 // Finish with popframe handling
1801 // A previous I2C followed by a deoptimization might have moved the
1802 // outgoing arguments further up the stack. PopFrame expects the
1803 // mutations to those outgoing arguments to be preserved and other
1804 // constraints basically require this frame to look exactly as
1805 // though it had previously invoked an interpreted activation with
1806 // no space between the top of the expression stack (current
1807 // last_sp) and the top of stack. Rather than force deopt to
1808 // maintain this kind of invariant all the time we call a small
1809 // fixup routine to move the mutated arguments onto the top of our
1810 // expression stack if necessary.
1811 __ mov(rax, rsp);
1812 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1813 __ get_thread(thread);
1814 // PC must point into interpreter here
1815 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1816 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1817 __ get_thread(thread);
1818 __ reset_last_Java_frame(thread, true, true);
1819 // Restore the last_sp and null it out
1820 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1821 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1822
1823 __ restore_bcp();
1824 __ restore_locals();
1825 // The method data pointer was incremented already during
1826 // call profiling. We have to restore the mdp for the current bcp.
1827 if (ProfileInterpreter) {
1828 __ set_method_data_pointer_for_bcp();
1829 }
1830
1831 // Clear the popframe condition flag
1832 __ get_thread(thread);
1833 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1834
1835 #if INCLUDE_JVMTI
1836 {
1837 Label L_done;
1838 const Register local0 = rdi;
1839
1840 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1841 __ jcc(Assembler::notEqual, L_done);
1842
1843 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1844 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1845
1846 __ get_method(rdx);
1847 __ movptr(rax, Address(local0, 0));
1848 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1849
1850 __ testptr(rax, rax);
1851 __ jcc(Assembler::zero, L_done);
1852
1853 __ movptr(Address(rbx, 0), rax);
1854 __ bind(L_done);
1855 }
1856 #endif // INCLUDE_JVMTI
1857
1858 __ dispatch_next(vtos);
1859 // end of PopFrame support
1860
1861 Interpreter::_remove_activation_entry = __ pc();
1862
1863 // preserve exception over this code sequence
1864 __ pop_ptr(rax);
1865 __ get_thread(thread);
1866 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1867 // remove the activation (without doing throws on illegalMonitorExceptions)
1868 __ remove_activation(vtos, rdx, false, true, false);
1869 // restore exception
1870 __ get_thread(thread);
1871 __ get_vm_result(rax, thread);
1872
1873 // Inbetween activations - previous activation type unknown yet
1874 // compute continuation point - the continuation point expects
1875 // the following registers set up:
1876 //
1877 // rax: exception
1878 // rdx: return address/pc that threw exception
1879 // rsp: expression stack of caller
1880 // rbp: rbp, of caller
1881 __ push(rax); // save exception
1882 __ push(rdx); // save return address
1883 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1884 __ mov(rbx, rax); // save exception handler
1885 __ pop(rdx); // restore return address
1886 __ pop(rax); // restore exception
1887 // Note that an "issuing PC" is actually the next PC after the call
1888 __ jmp(rbx); // jump to exception handler of caller
1889 }
1890
1891
1892 //
1893 // JVMTI ForceEarlyReturn support
1894 //
1895 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1896 address entry = __ pc();
1897 const Register thread = rcx;
1898
1899 __ restore_bcp();
1900 __ restore_locals();
1901 __ empty_expression_stack();
1902 __ empty_FPU_stack();
1903 __ load_earlyret_value(state);
1904
1905 __ get_thread(thread);
1906 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1907 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1908
1909 // Clear the earlyret state
1910 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1911
1912 __ remove_activation(state, rsi,
1913 false, /* throw_monitor_exception */
1914 false, /* install_monitor_exception */
1915 true); /* notify_jvmdi */
1916 __ jmp(rsi);
1917 return entry;
1918 } // end of ForceEarlyReturn support
1919
1920
1921 //------------------------------------------------------------------------------------------------------------------------
1922 // Helper for vtos entry point generation
1923
1924 void TemplateInterpreterGenerator::set_vtos_entry_points (Template* t, address& bep, address& cep, address& sep, address& aep, address& iep, address& lep, address& fep, address& dep, address& vep) {
1925 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1926 Label L;
1927 fep = __ pc(); __ push(ftos); __ jmp(L);
1928 dep = __ pc(); __ push(dtos); __ jmp(L);
1929 lep = __ pc(); __ push(ltos); __ jmp(L);
1930 aep = __ pc(); __ push(atos); __ jmp(L);
1931 bep = cep = sep = // fall through
1932 iep = __ pc(); __ push(itos); // fall through
1933 vep = __ pc(); __ bind(L); // fall through
1934 generate_and_dispatch(t);
1935 }
1936
1937 //------------------------------------------------------------------------------------------------------------------------
1938 // Generation of individual instructions
1939
1940 // helpers for generate_and_dispatch
1941
1942
1943
1944 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1945 : TemplateInterpreterGenerator(code) {
1946 generate_all(); // down here so it can be "virtual"
1947 }
1948
1949 //------------------------------------------------------------------------------------------------------------------------
1950
1951 // Non-product code
1952 #ifndef PRODUCT
1953 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1954 address entry = __ pc();
1955
1956 // prepare expression stack
1957 __ pop(rcx); // pop return address so expression stack is 'pure'
1958 __ push(state); // save tosca
1959
1960 // pass tosca registers as arguments & call tracer
1961 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1962 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1963 __ pop(state); // restore tosca
1964
1965 // return
1966 __ jmp(rcx);
1967
1968 return entry;
1969 }
1970
1971
1972 void TemplateInterpreterGenerator::count_bytecode() {
1973 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1974 }
1975
1976
1977 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1978 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1979 }
1980
1981
1982 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1983 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1984 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1985 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1986 ExternalAddress table((address) BytecodePairHistogram::_counters);
1987 Address index(noreg, rbx, Address::times_4);
1988 __ incrementl(ArrayAddress(table, index));
1989 }
1990
1991
1992 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1993 // Call a little run-time stub to avoid blow-up for each bytecode.
1994 // The run-time runtime saves the right registers, depending on
1995 // the tosca in-state for the given template.
1996 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1997 "entry must have been generated");
1998 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1999 }
2000
2001
2002 void TemplateInterpreterGenerator::stop_interpreter_at() {
2003 Label L;
2004 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
2005 StopInterpreterAt);
2006 __ jcc(Assembler::notEqual, L);
2007 __ int3();
2008 __ bind(L);
2009 }
2010 #endif // !PRODUCT
2011 #endif // CC_INTERP