1 /*
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "incls/_precompiled.incl"
26 #include "incls/_templateInterpreter_x86_32.cpp.incl"
27
28 #define __ _masm->
29
30
31 #ifndef CC_INTERP
32 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
33 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
34 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
35
36 //------------------------------------------------------------------------------------------------------------------------
37
38 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
39 address entry = __ pc();
40
41 // Note: There should be a minimal interpreter frame set up when stack
42 // overflow occurs since we check explicitly for it now.
43 //
44 #ifdef ASSERT
45 { Label L;
46 __ lea(rax, Address(rbp,
47 frame::interpreter_frame_monitor_block_top_offset * wordSize));
48 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
49 // (stack grows negative)
50 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
51 __ stop ("interpreter frame not set up");
52 __ bind(L);
53 }
54 #endif // ASSERT
55 // Restore bcp under the assumption that the current frame is still
56 // interpreted
57 __ restore_bcp();
58
59 // expression stack must be empty before entering the VM if an exception
60 // happened
61 __ empty_expression_stack();
62 __ empty_FPU_stack();
63 // throw exception
64 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
65 return entry;
66 }
67
68 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
69 address entry = __ pc();
70 // expression stack must be empty before entering the VM if an exception happened
71 __ empty_expression_stack();
72 __ empty_FPU_stack();
73 // setup parameters
74 // ??? convention: expect aberrant index in register rbx,
75 __ lea(rax, ExternalAddress((address)name));
76 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
77 return entry;
78 }
79
80 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
81 address entry = __ pc();
82 // object is at TOS
83 __ pop(rax);
84 // expression stack must be empty before entering the VM if an exception
85 // happened
86 __ empty_expression_stack();
87 __ empty_FPU_stack();
88 __ call_VM(noreg,
89 CAST_FROM_FN_PTR(address,
90 InterpreterRuntime::throw_ClassCastException),
91 rax);
92 return entry;
93 }
94
95 // Arguments are: required type at TOS+4, failing object (or NULL) at TOS.
96 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
97 address entry = __ pc();
98
99 __ pop(rbx); // actual failing object is at TOS
100 __ pop(rax); // required type is at TOS+4
101
102 __ verify_oop(rbx);
103 __ verify_oop(rax);
104
105 // Various method handle types use interpreter registers as temps.
106 __ restore_bcp();
107 __ restore_locals();
108
109 // Expression stack must be empty before entering the VM for an exception.
110 __ empty_expression_stack();
111 __ empty_FPU_stack();
112 __ call_VM(noreg,
113 CAST_FROM_FN_PTR(address,
114 InterpreterRuntime::throw_WrongMethodTypeException),
115 // pass required type, failing object (or NULL)
116 rax, rbx);
117 return entry;
118 }
119
120
121 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
122 assert(!pass_oop || message == NULL, "either oop or message but not both");
123 address entry = __ pc();
124 if (pass_oop) {
125 // object is at TOS
126 __ pop(rbx);
127 }
128 // expression stack must be empty before entering the VM if an exception happened
129 __ empty_expression_stack();
130 __ empty_FPU_stack();
131 // setup parameters
132 __ lea(rax, ExternalAddress((address)name));
133 if (pass_oop) {
134 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
135 } else {
136 if (message != NULL) {
137 __ lea(rbx, ExternalAddress((address)message));
138 } else {
139 __ movptr(rbx, NULL_WORD);
140 }
141 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
142 }
143 // throw exception
144 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
145 return entry;
146 }
147
148
149 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
150 address entry = __ pc();
151 // NULL last_sp until next java call
152 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
153 __ dispatch_next(state);
154 return entry;
155 }
156
157
158 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
159 TosState incoming_state = state;
160
161 Label interpreter_entry;
162 address compiled_entry = __ pc();
163
164 #ifdef COMPILER2
165 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
166 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
167 for (int i = 1; i < 8; i++) {
168 __ ffree(i);
169 }
170 } else if (UseSSE < 2) {
171 __ empty_FPU_stack();
172 }
173 #endif
174 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
175 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
176 } else {
177 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
178 }
179
180 __ jmp(interpreter_entry, relocInfo::none);
181 // emit a sentinel we can test for when converting an interpreter
182 // entry point to a compiled entry point.
183 __ a_long(Interpreter::return_sentinel);
184 __ a_long((int)compiled_entry);
185 address entry = __ pc();
186 __ bind(interpreter_entry);
187
188 // In SSE mode, interpreter returns FP results in xmm0 but they need
189 // to end up back on the FPU so it can operate on them.
190 if (incoming_state == ftos && UseSSE >= 1) {
191 __ subptr(rsp, wordSize);
192 __ movflt(Address(rsp, 0), xmm0);
193 __ fld_s(Address(rsp, 0));
194 __ addptr(rsp, wordSize);
195 } else if (incoming_state == dtos && UseSSE >= 2) {
196 __ subptr(rsp, 2*wordSize);
197 __ movdbl(Address(rsp, 0), xmm0);
198 __ fld_d(Address(rsp, 0));
199 __ addptr(rsp, 2*wordSize);
200 }
201
202 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
203
204 // Restore stack bottom in case i2c adjusted stack
205 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
206 // and NULL it as marker that rsp is now tos until next java call
207 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
208
209 __ restore_bcp();
210 __ restore_locals();
211
212 Label L_got_cache, L_giant_index;
213 if (EnableInvokeDynamic) {
214 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
215 __ jcc(Assembler::equal, L_giant_index);
216 }
217 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
218 __ bind(L_got_cache);
219 __ movl(rbx, Address(rbx, rcx,
220 Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
221 ConstantPoolCacheEntry::flags_offset()));
222 __ andptr(rbx, 0xFF);
223 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
224 __ dispatch_next(state, step);
225
226 // out of the main line of code...
227 if (EnableInvokeDynamic) {
228 __ bind(L_giant_index);
229 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
230 __ jmp(L_got_cache);
231 }
232
233 return entry;
234 }
235
236
237 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
238 address entry = __ pc();
239
240 // In SSE mode, FP results are in xmm0
241 if (state == ftos && UseSSE > 0) {
242 __ subptr(rsp, wordSize);
243 __ movflt(Address(rsp, 0), xmm0);
244 __ fld_s(Address(rsp, 0));
245 __ addptr(rsp, wordSize);
246 } else if (state == dtos && UseSSE >= 2) {
247 __ subptr(rsp, 2*wordSize);
248 __ movdbl(Address(rsp, 0), xmm0);
249 __ fld_d(Address(rsp, 0));
250 __ addptr(rsp, 2*wordSize);
251 }
252
253 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
254
255 // The stack is not extended by deopt but we must NULL last_sp as this
256 // entry is like a "return".
257 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
258 __ restore_bcp();
259 __ restore_locals();
260 // handle exceptions
261 { Label L;
262 const Register thread = rcx;
263 __ get_thread(thread);
264 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
265 __ jcc(Assembler::zero, L);
266 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
267 __ should_not_reach_here();
268 __ bind(L);
269 }
270 __ dispatch_next(state, step);
271 return entry;
272 }
273
274
275 int AbstractInterpreter::BasicType_as_index(BasicType type) {
276 int i = 0;
277 switch (type) {
278 case T_BOOLEAN: i = 0; break;
279 case T_CHAR : i = 1; break;
280 case T_BYTE : i = 2; break;
281 case T_SHORT : i = 3; break;
282 case T_INT : // fall through
283 case T_LONG : // fall through
284 case T_VOID : i = 4; break;
285 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
286 case T_DOUBLE : i = 6; break;
287 case T_OBJECT : // fall through
288 case T_ARRAY : i = 7; break;
289 default : ShouldNotReachHere();
290 }
291 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
292 return i;
293 }
294
295
296 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
297 address entry = __ pc();
298 switch (type) {
299 case T_BOOLEAN: __ c2bool(rax); break;
300 case T_CHAR : __ andptr(rax, 0xFFFF); break;
301 case T_BYTE : __ sign_extend_byte (rax); break;
302 case T_SHORT : __ sign_extend_short(rax); break;
303 case T_INT : /* nothing to do */ break;
304 case T_DOUBLE :
305 case T_FLOAT :
306 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
307 __ pop(t); // remove return address first
308 // Must return a result for interpreter or compiler. In SSE
309 // mode, results are returned in xmm0 and the FPU stack must
310 // be empty.
311 if (type == T_FLOAT && UseSSE >= 1) {
312 // Load ST0
313 __ fld_d(Address(rsp, 0));
314 // Store as float and empty fpu stack
315 __ fstp_s(Address(rsp, 0));
316 // and reload
317 __ movflt(xmm0, Address(rsp, 0));
318 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
319 __ movdbl(xmm0, Address(rsp, 0));
320 } else {
321 // restore ST0
322 __ fld_d(Address(rsp, 0));
323 }
324 // and pop the temp
325 __ addptr(rsp, 2 * wordSize);
326 __ push(t); // restore return address
327 }
328 break;
329 case T_OBJECT :
330 // retrieve result from frame
331 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
332 // and verify it
333 __ verify_oop(rax);
334 break;
335 default : ShouldNotReachHere();
336 }
337 __ ret(0); // return from result handler
338 return entry;
339 }
340
341 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
342 address entry = __ pc();
343 __ push(state);
344 __ call_VM(noreg, runtime_entry);
345 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
346 return entry;
347 }
348
349
350 // Helpers for commoning out cases in the various type of method entries.
351 //
352
353 // increment invocation count & check for overflow
354 //
355 // Note: checking for negative value instead of overflow
356 // so we have a 'sticky' overflow test
357 //
358 // rbx,: method
359 // rcx: invocation counter
360 //
361 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
362 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
363 in_bytes(InvocationCounter::counter_offset()));
364 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
365 if (TieredCompilation) {
366 int increment = InvocationCounter::count_increment;
367 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
368 Label no_mdo, done;
369 if (ProfileInterpreter) {
370 // Are we profiling?
371 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
372 __ testptr(rax, rax);
373 __ jccb(Assembler::zero, no_mdo);
374 // Increment counter in the MDO
375 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
376 in_bytes(InvocationCounter::counter_offset()));
377 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
378 __ jmpb(done);
379 }
380 __ bind(no_mdo);
381 // Increment counter in methodOop (we don't need to load it, it's in rcx).
382 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
383 __ bind(done);
384 } else {
385 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() +
386 InvocationCounter::counter_offset());
387
388 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
389 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
390 }
391 // Update standard invocation counters
392 __ movl(rax, backedge_counter); // load backedge counter
393
394 __ incrementl(rcx, InvocationCounter::count_increment);
395 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
396
397 __ movl(invocation_counter, rcx); // save invocation count
398 __ addl(rcx, rax); // add both counters
399
400 // profile_method is non-null only for interpreted method so
401 // profile_method != NULL == !native_call
402 // BytecodeInterpreter only calls for native so code is elided.
403
404 if (ProfileInterpreter && profile_method != NULL) {
405 // Test to see if we should create a method data oop
406 __ cmp32(rcx,
407 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
408 __ jcc(Assembler::less, *profile_method_continue);
409
410 // if no method data exists, go to profile_method
411 __ test_method_data_pointer(rax, *profile_method);
412 }
413
414 __ cmp32(rcx,
415 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
416 __ jcc(Assembler::aboveEqual, *overflow);
417 }
418 }
419
420 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
421
422 // Asm interpreter on entry
423 // rdi - locals
424 // rsi - bcp
425 // rbx, - method
426 // rdx - cpool
427 // rbp, - interpreter frame
428
429 // C++ interpreter on entry
430 // rsi - new interpreter state pointer
431 // rbp - interpreter frame pointer
432 // rbx - method
433
434 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
435 // rbx, - method
436 // rcx - rcvr (assuming there is one)
437 // top of stack return address of interpreter caller
438 // rsp - sender_sp
439
440 // C++ interpreter only
441 // rsi - previous interpreter state pointer
442
443 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
444
445 // InterpreterRuntime::frequency_counter_overflow takes one argument
446 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
447 // The call returns the address of the verified entry point for the method or NULL
448 // if the compilation did not complete (either went background or bailed out).
449 __ movptr(rax, (intptr_t)false);
450 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
451
452 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
453
454 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
455 // and jump to the interpreted entry.
456 __ jmp(*do_continue, relocInfo::none);
457
458 }
459
460 void InterpreterGenerator::generate_stack_overflow_check(void) {
461 // see if we've got enough room on the stack for locals plus overhead.
462 // the expression stack grows down incrementally, so the normal guard
463 // page mechanism will work for that.
464 //
465 // Registers live on entry:
466 //
467 // Asm interpreter
468 // rdx: number of additional locals this frame needs (what we must check)
469 // rbx,: methodOop
470
471 // destroyed on exit
472 // rax,
473
474 // NOTE: since the additional locals are also always pushed (wasn't obvious in
475 // generate_method_entry) so the guard should work for them too.
476 //
477
478 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
479 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
480
481 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
482 // be sure to change this if you add/subtract anything to/from the overhead area
483 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
484
485 const int page_size = os::vm_page_size();
486
487 Label after_frame_check;
488
489 // see if the frame is greater than one page in size. If so,
490 // then we need to verify there is enough stack space remaining
491 // for the additional locals.
492 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
493 __ jcc(Assembler::belowEqual, after_frame_check);
494
495 // compute rsp as if this were going to be the last frame on
496 // the stack before the red zone
497
498 Label after_frame_check_pop;
499
500 __ push(rsi);
501
502 const Register thread = rsi;
503
504 __ get_thread(thread);
505
506 const Address stack_base(thread, Thread::stack_base_offset());
507 const Address stack_size(thread, Thread::stack_size_offset());
508
509 // locals + overhead, in bytes
510 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
511
512 #ifdef ASSERT
513 Label stack_base_okay, stack_size_okay;
514 // verify that thread stack base is non-zero
515 __ cmpptr(stack_base, (int32_t)NULL_WORD);
516 __ jcc(Assembler::notEqual, stack_base_okay);
517 __ stop("stack base is zero");
518 __ bind(stack_base_okay);
519 // verify that thread stack size is non-zero
520 __ cmpptr(stack_size, 0);
521 __ jcc(Assembler::notEqual, stack_size_okay);
522 __ stop("stack size is zero");
523 __ bind(stack_size_okay);
524 #endif
525
526 // Add stack base to locals and subtract stack size
527 __ addptr(rax, stack_base);
528 __ subptr(rax, stack_size);
529
530 // Use the maximum number of pages we might bang.
531 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
532 (StackRedPages+StackYellowPages);
533 __ addptr(rax, max_pages * page_size);
534
535 // check against the current stack bottom
536 __ cmpptr(rsp, rax);
537 __ jcc(Assembler::above, after_frame_check_pop);
538
539 __ pop(rsi); // get saved bcp / (c++ prev state ).
540
541 __ pop(rax); // get return address
542 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
543
544 // all done with frame size check
545 __ bind(after_frame_check_pop);
546 __ pop(rsi);
547
548 __ bind(after_frame_check);
549 }
550
551 // Allocate monitor and lock method (asm interpreter)
552 // rbx, - methodOop
553 //
554 void InterpreterGenerator::lock_method(void) {
555 // synchronize method
556 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
557 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
558 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
559
560 #ifdef ASSERT
561 { Label L;
562 __ movl(rax, access_flags);
563 __ testl(rax, JVM_ACC_SYNCHRONIZED);
564 __ jcc(Assembler::notZero, L);
565 __ stop("method doesn't need synchronization");
566 __ bind(L);
567 }
568 #endif // ASSERT
569 // get synchronization object
570 { Label done;
571 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
572 __ movl(rax, access_flags);
573 __ testl(rax, JVM_ACC_STATIC);
574 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
575 __ jcc(Assembler::zero, done);
576 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
577 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
578 __ movptr(rax, Address(rax, mirror_offset));
579 __ bind(done);
580 }
581 // add space for monitor & lock
582 __ subptr(rsp, entry_size); // add space for a monitor entry
583 __ movptr(monitor_block_top, rsp); // set new monitor block top
584 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
585 __ mov(rdx, rsp); // object address
586 __ lock_object(rdx);
587 }
588
589 //
590 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
591 // and for native methods hence the shared code.
592
593 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
594 // initialize fixed part of activation frame
595 __ push(rax); // save return address
596 __ enter(); // save old & set new rbp,
597
598
599 __ push(rsi); // set sender sp
600 __ push((int32_t)NULL_WORD); // leave last_sp as null
601 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
602 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
603 __ push(rbx); // save methodOop
604 if (ProfileInterpreter) {
605 Label method_data_continue;
606 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
607 __ testptr(rdx, rdx);
608 __ jcc(Assembler::zero, method_data_continue);
609 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
610 __ bind(method_data_continue);
611 __ push(rdx); // set the mdp (method data pointer)
612 } else {
613 __ push(0);
614 }
615
616 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
617 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
618 __ push(rdx); // set constant pool cache
619 __ push(rdi); // set locals pointer
620 if (native_call) {
621 __ push(0); // no bcp
622 } else {
623 __ push(rsi); // set bcp
624 }
625 __ push(0); // reserve word for pointer to expression stack bottom
626 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
627 }
628
629 // End of helpers
630
631 //
632 // Various method entries
633 //------------------------------------------------------------------------------------------------------------------------
634 //
635 //
636
637 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
638
639 address InterpreterGenerator::generate_accessor_entry(void) {
640
641 // rbx,: methodOop
642 // rcx: receiver (preserve for slow entry into asm interpreter)
643
644 // rsi: senderSP must preserved for slow path, set SP to it on fast path
645
646 address entry_point = __ pc();
647 Label xreturn_path;
648
649 // do fastpath for resolved accessor methods
650 if (UseFastAccessorMethods) {
651 Label slow_path;
652 // If we need a safepoint check, generate full interpreter entry.
653 ExternalAddress state(SafepointSynchronize::address_of_state());
654 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
655 SafepointSynchronize::_not_synchronized);
656
657 __ jcc(Assembler::notEqual, slow_path);
658 // ASM/C++ Interpreter
659 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
660 // Note: We can only use this code if the getfield has been resolved
661 // and if we don't have a null-pointer exception => check for
662 // these conditions first and use slow path if necessary.
663 // rbx,: method
664 // rcx: receiver
665 __ movptr(rax, Address(rsp, wordSize));
666
667 // check if local 0 != NULL and read field
668 __ testptr(rax, rax);
669 __ jcc(Assembler::zero, slow_path);
670
671 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
672 // read first instruction word and extract bytecode @ 1 and index @ 2
673 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
674 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
675 // Shift codes right to get the index on the right.
676 // The bytecode fetched looks like <index><0xb4><0x2a>
677 __ shrl(rdx, 2*BitsPerByte);
678 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
679 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
680
681 // rax,: local 0
682 // rbx,: method
683 // rcx: receiver - do not destroy since it is needed for slow path!
684 // rcx: scratch
685 // rdx: constant pool cache index
686 // rdi: constant pool cache
687 // rsi: sender sp
688
689 // check if getfield has been resolved and read constant pool cache entry
690 // check the validity of the cache entry by testing whether _indices field
691 // contains Bytecode::_getfield in b1 byte.
692 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
693 __ movl(rcx,
694 Address(rdi,
695 rdx,
696 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
697 __ shrl(rcx, 2*BitsPerByte);
698 __ andl(rcx, 0xFF);
699 __ cmpl(rcx, Bytecodes::_getfield);
700 __ jcc(Assembler::notEqual, slow_path);
701
702 // Note: constant pool entry is not valid before bytecode is resolved
703 __ movptr(rcx,
704 Address(rdi,
705 rdx,
706 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
707 __ movl(rdx,
708 Address(rdi,
709 rdx,
710 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
711
712 Label notByte, notShort, notChar;
713 const Address field_address (rax, rcx, Address::times_1);
714
715 // Need to differentiate between igetfield, agetfield, bgetfield etc.
716 // because they are different sizes.
717 // Use the type from the constant pool cache
718 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
719 // Make sure we don't need to mask rdx for tosBits after the above shift
720 ConstantPoolCacheEntry::verify_tosBits();
721 __ cmpl(rdx, btos);
722 __ jcc(Assembler::notEqual, notByte);
723 __ load_signed_byte(rax, field_address);
724 __ jmp(xreturn_path);
725
726 __ bind(notByte);
727 __ cmpl(rdx, stos);
728 __ jcc(Assembler::notEqual, notShort);
729 __ load_signed_short(rax, field_address);
730 __ jmp(xreturn_path);
731
732 __ bind(notShort);
733 __ cmpl(rdx, ctos);
734 __ jcc(Assembler::notEqual, notChar);
735 __ load_unsigned_short(rax, field_address);
736 __ jmp(xreturn_path);
737
738 __ bind(notChar);
739 #ifdef ASSERT
740 Label okay;
741 __ cmpl(rdx, atos);
742 __ jcc(Assembler::equal, okay);
743 __ cmpl(rdx, itos);
744 __ jcc(Assembler::equal, okay);
745 __ stop("what type is this?");
746 __ bind(okay);
747 #endif // ASSERT
748 // All the rest are a 32 bit wordsize
749 // This is ok for now. Since fast accessors should be going away
750 __ movptr(rax, field_address);
751
752 __ bind(xreturn_path);
753
754 // _ireturn/_areturn
755 __ pop(rdi); // get return address
756 __ mov(rsp, rsi); // set sp to sender sp
757 __ jmp(rdi);
758
759 // generate a vanilla interpreter entry as the slow path
760 __ bind(slow_path);
761
762 (void) generate_normal_entry(false);
763 return entry_point;
764 }
765 return NULL;
766
767 }
768
769 //
770 // Interpreter stub for calling a native method. (asm interpreter)
771 // This sets up a somewhat different looking stack for calling the native method
772 // than the typical interpreter frame setup.
773 //
774
775 address InterpreterGenerator::generate_native_entry(bool synchronized) {
776 // determine code generation flags
777 bool inc_counter = UseCompiler || CountCompiledCalls;
778
779 // rbx,: methodOop
780 // rsi: sender sp
781 // rsi: previous interpreter state (C++ interpreter) must preserve
782 address entry_point = __ pc();
783
784
785 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
786 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
787 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
788
789 // get parameter size (always needed)
790 __ load_unsigned_short(rcx, size_of_parameters);
791
792 // native calls don't need the stack size check since they have no expression stack
793 // and the arguments are already on the stack and we only add a handful of words
794 // to the stack
795
796 // rbx,: methodOop
797 // rcx: size of parameters
798 // rsi: sender sp
799
800 __ pop(rax); // get return address
801 // for natives the size of locals is zero
802
803 // compute beginning of parameters (rdi)
804 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
805
806
807 // add 2 zero-initialized slots for native calls
808 // NULL result handler
809 __ push((int32_t)NULL_WORD);
810 // NULL oop temp (mirror or jni oop result)
811 __ push((int32_t)NULL_WORD);
812
813 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
814 // initialize fixed part of activation frame
815
816 generate_fixed_frame(true);
817
818 // make sure method is native & not abstract
819 #ifdef ASSERT
820 __ movl(rax, access_flags);
821 {
822 Label L;
823 __ testl(rax, JVM_ACC_NATIVE);
824 __ jcc(Assembler::notZero, L);
825 __ stop("tried to execute non-native method as native");
826 __ bind(L);
827 }
828 { Label L;
829 __ testl(rax, JVM_ACC_ABSTRACT);
830 __ jcc(Assembler::zero, L);
831 __ stop("tried to execute abstract method in interpreter");
832 __ bind(L);
833 }
834 #endif
835
836 // Since at this point in the method invocation the exception handler
837 // would try to exit the monitor of synchronized methods which hasn't
838 // been entered yet, we set the thread local variable
839 // _do_not_unlock_if_synchronized to true. The remove_activation will
840 // check this flag.
841
842 __ get_thread(rax);
843 const Address do_not_unlock_if_synchronized(rax,
844 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
845 __ movbool(do_not_unlock_if_synchronized, true);
846
847 // increment invocation count & check for overflow
848 Label invocation_counter_overflow;
849 if (inc_counter) {
850 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
851 }
852
853 Label continue_after_compile;
854 __ bind(continue_after_compile);
855
856 bang_stack_shadow_pages(true);
857
858 // reset the _do_not_unlock_if_synchronized flag
859 __ get_thread(rax);
860 __ movbool(do_not_unlock_if_synchronized, false);
861
862 // check for synchronized methods
863 // Must happen AFTER invocation_counter check and stack overflow check,
864 // so method is not locked if overflows.
865 //
866 if (synchronized) {
867 lock_method();
868 } else {
869 // no synchronization necessary
870 #ifdef ASSERT
871 { Label L;
872 __ movl(rax, access_flags);
873 __ testl(rax, JVM_ACC_SYNCHRONIZED);
874 __ jcc(Assembler::zero, L);
875 __ stop("method needs synchronization");
876 __ bind(L);
877 }
878 #endif
879 }
880
881 // start execution
882 #ifdef ASSERT
883 { Label L;
884 const Address monitor_block_top (rbp,
885 frame::interpreter_frame_monitor_block_top_offset * wordSize);
886 __ movptr(rax, monitor_block_top);
887 __ cmpptr(rax, rsp);
888 __ jcc(Assembler::equal, L);
889 __ stop("broken stack frame setup in interpreter");
890 __ bind(L);
891 }
892 #endif
893
894 // jvmti/dtrace support
895 __ notify_method_entry();
896
897 // work registers
898 const Register method = rbx;
899 const Register thread = rdi;
900 const Register t = rcx;
901
902 // allocate space for parameters
903 __ get_method(method);
904 __ verify_oop(method);
905 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
906 __ shlptr(t, Interpreter::logStackElementSize);
907 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
908 __ subptr(rsp, t);
909 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
910
911 // get signature handler
912 { Label L;
913 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
914 __ testptr(t, t);
915 __ jcc(Assembler::notZero, L);
916 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
917 __ get_method(method);
918 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
919 __ bind(L);
920 }
921
922 // call signature handler
923 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
924 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
925 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
926 // The generated handlers do not touch RBX (the method oop).
927 // However, large signatures cannot be cached and are generated
928 // each time here. The slow-path generator will blow RBX
929 // sometime, so we must reload it after the call.
930 __ call(t);
931 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
932
933 // result handler is in rax,
934 // set result handler
935 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
936
937 // pass mirror handle if static call
938 { Label L;
939 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
940 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
941 __ testl(t, JVM_ACC_STATIC);
942 __ jcc(Assembler::zero, L);
943 // get mirror
944 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
945 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
946 __ movptr(t, Address(t, mirror_offset));
947 // copy mirror into activation frame
948 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
949 // pass handle to mirror
950 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
951 __ movptr(Address(rsp, wordSize), t);
952 __ bind(L);
953 }
954
955 // get native function entry point
956 { Label L;
957 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
958 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
959 __ cmpptr(rax, unsatisfied.addr());
960 __ jcc(Assembler::notEqual, L);
961 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
962 __ get_method(method);
963 __ verify_oop(method);
964 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
965 __ bind(L);
966 }
967
968 // pass JNIEnv
969 __ get_thread(thread);
970 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
971 __ movptr(Address(rsp, 0), t);
972
973 // set_last_Java_frame_before_call
974 // It is enough that the pc()
975 // points into the right code segment. It does not have to be the correct return pc.
976 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
977
978 // change thread state
979 #ifdef ASSERT
980 { Label L;
981 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
982 __ cmpl(t, _thread_in_Java);
983 __ jcc(Assembler::equal, L);
984 __ stop("Wrong thread state in native stub");
985 __ bind(L);
986 }
987 #endif
988
989 // Change state to native
990 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
991 __ call(rax);
992
993 // result potentially in rdx:rax or ST0
994
995 // Either restore the MXCSR register after returning from the JNI Call
996 // or verify that it wasn't changed.
997 if (VM_Version::supports_sse()) {
998 if (RestoreMXCSROnJNICalls) {
999 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1000 }
1001 else if (CheckJNICalls ) {
1002 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1003 }
1004 }
1005
1006 // Either restore the x87 floating pointer control word after returning
1007 // from the JNI call or verify that it wasn't changed.
1008 if (CheckJNICalls) {
1009 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1010 }
1011
1012 // save potential result in ST(0) & rdx:rax
1013 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1014 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1015 // It is safe to do this push because state is _thread_in_native and return address will be found
1016 // via _last_native_pc and not via _last_jave_sp
1017
1018 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1019 // If the order changes or anything else is added to the stack the code in
1020 // interpreter_frame_result will have to be changed.
1021
1022 { Label L;
1023 Label push_double;
1024 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1025 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1026 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1027 float_handler.addr());
1028 __ jcc(Assembler::equal, push_double);
1029 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1030 double_handler.addr());
1031 __ jcc(Assembler::notEqual, L);
1032 __ bind(push_double);
1033 __ push(dtos);
1034 __ bind(L);
1035 }
1036 __ push(ltos);
1037
1038 // change thread state
1039 __ get_thread(thread);
1040 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1041 if(os::is_MP()) {
1042 if (UseMembar) {
1043 // Force this write out before the read below
1044 __ membar(Assembler::Membar_mask_bits(
1045 Assembler::LoadLoad | Assembler::LoadStore |
1046 Assembler::StoreLoad | Assembler::StoreStore));
1047 } else {
1048 // Write serialization page so VM thread can do a pseudo remote membar.
1049 // We use the current thread pointer to calculate a thread specific
1050 // offset to write to within the page. This minimizes bus traffic
1051 // due to cache line collision.
1052 __ serialize_memory(thread, rcx);
1053 }
1054 }
1055
1056 if (AlwaysRestoreFPU) {
1057 // Make sure the control word is correct.
1058 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1059 }
1060
1061 // check for safepoint operation in progress and/or pending suspend requests
1062 { Label Continue;
1063
1064 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1065 SafepointSynchronize::_not_synchronized);
1066
1067 Label L;
1068 __ jcc(Assembler::notEqual, L);
1069 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1070 __ jcc(Assembler::equal, Continue);
1071 __ bind(L);
1072
1073 // Don't use call_VM as it will see a possible pending exception and forward it
1074 // and never return here preventing us from clearing _last_native_pc down below.
1075 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1076 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1077 // by hand.
1078 //
1079 __ push(thread);
1080 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1081 JavaThread::check_special_condition_for_native_trans)));
1082 __ increment(rsp, wordSize);
1083 __ get_thread(thread);
1084
1085 __ bind(Continue);
1086 }
1087
1088 // change thread state
1089 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1090
1091 __ reset_last_Java_frame(thread, true, true);
1092
1093 // reset handle block
1094 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1095 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1096
1097 // If result was an oop then unbox and save it in the frame
1098 { Label L;
1099 Label no_oop, store_result;
1100 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1101 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1102 handler.addr());
1103 __ jcc(Assembler::notEqual, no_oop);
1104 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1105 __ pop(ltos);
1106 __ testptr(rax, rax);
1107 __ jcc(Assembler::zero, store_result);
1108 // unbox
1109 __ movptr(rax, Address(rax, 0));
1110 __ bind(store_result);
1111 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1112 // keep stack depth as expected by pushing oop which will eventually be discarded
1113 __ push(ltos);
1114 __ bind(no_oop);
1115 }
1116
1117 {
1118 Label no_reguard;
1119 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1120 __ jcc(Assembler::notEqual, no_reguard);
1121
1122 __ pusha();
1123 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1124 __ popa();
1125
1126 __ bind(no_reguard);
1127 }
1128
1129 // restore rsi to have legal interpreter frame,
1130 // i.e., bci == 0 <=> rsi == code_base()
1131 // Can't call_VM until bcp is within reasonable.
1132 __ get_method(method); // method is junk from thread_in_native to now.
1133 __ verify_oop(method);
1134 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
1135 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
1136
1137 // handle exceptions (exception handling will handle unlocking!)
1138 { Label L;
1139 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1140 __ jcc(Assembler::zero, L);
1141 // Note: At some point we may want to unify this with the code used in call_VM_base();
1142 // i.e., we should use the StubRoutines::forward_exception code. For now this
1143 // doesn't work here because the rsp is not correctly set at this point.
1144 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1145 __ should_not_reach_here();
1146 __ bind(L);
1147 }
1148
1149 // do unlocking if necessary
1150 { Label L;
1151 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1152 __ testl(t, JVM_ACC_SYNCHRONIZED);
1153 __ jcc(Assembler::zero, L);
1154 // the code below should be shared with interpreter macro assembler implementation
1155 { Label unlock;
1156 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1157 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1158 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1159
1160 __ lea(rdx, monitor); // address of first monitor
1161
1162 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1163 __ testptr(t, t);
1164 __ jcc(Assembler::notZero, unlock);
1165
1166 // Entry already unlocked, need to throw exception
1167 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1168 __ should_not_reach_here();
1169
1170 __ bind(unlock);
1171 __ unlock_object(rdx);
1172 }
1173 __ bind(L);
1174 }
1175
1176 // jvmti/dtrace support
1177 // Note: This must happen _after_ handling/throwing any exceptions since
1178 // the exception handler code notifies the runtime of method exits
1179 // too. If this happens before, method entry/exit notifications are
1180 // not properly paired (was bug - gri 11/22/99).
1181 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1182
1183 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1184 __ pop(ltos);
1185 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1186 __ call(t);
1187
1188 // remove activation
1189 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1190 __ leave(); // remove frame anchor
1191 __ pop(rdi); // get return address
1192 __ mov(rsp, t); // set sp to sender sp
1193 __ jmp(rdi);
1194
1195 if (inc_counter) {
1196 // Handle overflow of counter and compile method
1197 __ bind(invocation_counter_overflow);
1198 generate_counter_overflow(&continue_after_compile);
1199 }
1200
1201 return entry_point;
1202 }
1203
1204 //
1205 // Generic interpreted method entry to (asm) interpreter
1206 //
1207 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1208 // determine code generation flags
1209 bool inc_counter = UseCompiler || CountCompiledCalls;
1210
1211 // rbx,: methodOop
1212 // rsi: sender sp
1213 address entry_point = __ pc();
1214
1215
1216 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1217 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
1218 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1219 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
1220
1221 // get parameter size (always needed)
1222 __ load_unsigned_short(rcx, size_of_parameters);
1223
1224 // rbx,: methodOop
1225 // rcx: size of parameters
1226
1227 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1228
1229 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1230 __ subl(rdx, rcx); // rdx = no. of additional locals
1231
1232 // see if we've got enough room on the stack for locals plus overhead.
1233 generate_stack_overflow_check();
1234
1235 // get return address
1236 __ pop(rax);
1237
1238 // compute beginning of parameters (rdi)
1239 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1240
1241 // rdx - # of additional locals
1242 // allocate space for locals
1243 // explicitly initialize locals
1244 {
1245 Label exit, loop;
1246 __ testl(rdx, rdx);
1247 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1248 __ bind(loop);
1249 __ push((int32_t)NULL_WORD); // initialize local variables
1250 __ decrement(rdx); // until everything initialized
1251 __ jcc(Assembler::greater, loop);
1252 __ bind(exit);
1253 }
1254
1255 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1256 // initialize fixed part of activation frame
1257 generate_fixed_frame(false);
1258
1259 // make sure method is not native & not abstract
1260 #ifdef ASSERT
1261 __ movl(rax, access_flags);
1262 {
1263 Label L;
1264 __ testl(rax, JVM_ACC_NATIVE);
1265 __ jcc(Assembler::zero, L);
1266 __ stop("tried to execute native method as non-native");
1267 __ bind(L);
1268 }
1269 { Label L;
1270 __ testl(rax, JVM_ACC_ABSTRACT);
1271 __ jcc(Assembler::zero, L);
1272 __ stop("tried to execute abstract method in interpreter");
1273 __ bind(L);
1274 }
1275 #endif
1276
1277 // Since at this point in the method invocation the exception handler
1278 // would try to exit the monitor of synchronized methods which hasn't
1279 // been entered yet, we set the thread local variable
1280 // _do_not_unlock_if_synchronized to true. The remove_activation will
1281 // check this flag.
1282
1283 __ get_thread(rax);
1284 const Address do_not_unlock_if_synchronized(rax,
1285 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1286 __ movbool(do_not_unlock_if_synchronized, true);
1287
1288 // increment invocation count & check for overflow
1289 Label invocation_counter_overflow;
1290 Label profile_method;
1291 Label profile_method_continue;
1292 if (inc_counter) {
1293 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1294 if (ProfileInterpreter) {
1295 __ bind(profile_method_continue);
1296 }
1297 }
1298 Label continue_after_compile;
1299 __ bind(continue_after_compile);
1300
1301 bang_stack_shadow_pages(false);
1302
1303 // reset the _do_not_unlock_if_synchronized flag
1304 __ get_thread(rax);
1305 __ movbool(do_not_unlock_if_synchronized, false);
1306
1307 // check for synchronized methods
1308 // Must happen AFTER invocation_counter check and stack overflow check,
1309 // so method is not locked if overflows.
1310 //
1311 if (synchronized) {
1312 // Allocate monitor and lock method
1313 lock_method();
1314 } else {
1315 // no synchronization necessary
1316 #ifdef ASSERT
1317 { Label L;
1318 __ movl(rax, access_flags);
1319 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1320 __ jcc(Assembler::zero, L);
1321 __ stop("method needs synchronization");
1322 __ bind(L);
1323 }
1324 #endif
1325 }
1326
1327 // start execution
1328 #ifdef ASSERT
1329 { Label L;
1330 const Address monitor_block_top (rbp,
1331 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1332 __ movptr(rax, monitor_block_top);
1333 __ cmpptr(rax, rsp);
1334 __ jcc(Assembler::equal, L);
1335 __ stop("broken stack frame setup in interpreter");
1336 __ bind(L);
1337 }
1338 #endif
1339
1340 // jvmti support
1341 __ notify_method_entry();
1342
1343 __ dispatch_next(vtos);
1344
1345 // invocation counter overflow
1346 if (inc_counter) {
1347 if (ProfileInterpreter) {
1348 // We have decided to profile this method in the interpreter
1349 __ bind(profile_method);
1350
1351 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
1352
1353 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
1354 __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1355 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1356 __ test_method_data_pointer(rax, profile_method_continue);
1357 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1358 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1359 __ jmp(profile_method_continue);
1360 }
1361 // Handle overflow of counter and compile method
1362 __ bind(invocation_counter_overflow);
1363 generate_counter_overflow(&continue_after_compile);
1364 }
1365
1366 return entry_point;
1367 }
1368
1369 //------------------------------------------------------------------------------------------------------------------------
1370 // Entry points
1371 //
1372 // Here we generate the various kind of entries into the interpreter.
1373 // The two main entry type are generic bytecode methods and native call method.
1374 // These both come in synchronized and non-synchronized versions but the
1375 // frame layout they create is very similar. The other method entry
1376 // types are really just special purpose entries that are really entry
1377 // and interpretation all in one. These are for trivial methods like
1378 // accessor, empty, or special math methods.
1379 //
1380 // When control flow reaches any of the entry types for the interpreter
1381 // the following holds ->
1382 //
1383 // Arguments:
1384 //
1385 // rbx,: methodOop
1386 // rcx: receiver
1387 //
1388 //
1389 // Stack layout immediately at entry
1390 //
1391 // [ return address ] <--- rsp
1392 // [ parameter n ]
1393 // ...
1394 // [ parameter 1 ]
1395 // [ expression stack ] (caller's java expression stack)
1396
1397 // Assuming that we don't go to one of the trivial specialized
1398 // entries the stack will look like below when we are ready to execute
1399 // the first bytecode (or call the native routine). The register usage
1400 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1401 //
1402 // local variables follow incoming parameters immediately; i.e.
1403 // the return address is moved to the end of the locals).
1404 //
1405 // [ monitor entry ] <--- rsp
1406 // ...
1407 // [ monitor entry ]
1408 // [ expr. stack bottom ]
1409 // [ saved rsi ]
1410 // [ current rdi ]
1411 // [ methodOop ]
1412 // [ saved rbp, ] <--- rbp,
1413 // [ return address ]
1414 // [ local variable m ]
1415 // ...
1416 // [ local variable 1 ]
1417 // [ parameter n ]
1418 // ...
1419 // [ parameter 1 ] <--- rdi
1420
1421 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1422 // determine code generation flags
1423 bool synchronized = false;
1424 address entry_point = NULL;
1425
1426 switch (kind) {
1427 case Interpreter::zerolocals : break;
1428 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1429 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1430 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1431 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1432 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1433 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1434 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1435
1436 case Interpreter::java_lang_math_sin : // fall thru
1437 case Interpreter::java_lang_math_cos : // fall thru
1438 case Interpreter::java_lang_math_tan : // fall thru
1439 case Interpreter::java_lang_math_abs : // fall thru
1440 case Interpreter::java_lang_math_log : // fall thru
1441 case Interpreter::java_lang_math_log10 : // fall thru
1442 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1443 default : ShouldNotReachHere(); break;
1444 }
1445
1446 if (entry_point) return entry_point;
1447
1448 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1449
1450 }
1451
1452 // These should never be compiled since the interpreter will prefer
1453 // the compiled version to the intrinsic version.
1454 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1455 switch (method_kind(m)) {
1456 case Interpreter::java_lang_math_sin : // fall thru
1457 case Interpreter::java_lang_math_cos : // fall thru
1458 case Interpreter::java_lang_math_tan : // fall thru
1459 case Interpreter::java_lang_math_abs : // fall thru
1460 case Interpreter::java_lang_math_log : // fall thru
1461 case Interpreter::java_lang_math_log10 : // fall thru
1462 case Interpreter::java_lang_math_sqrt :
1463 return false;
1464 default:
1465 return true;
1466 }
1467 }
1468
1469 // How much stack a method activation needs in words.
1470 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1471
1472 const int stub_code = 4; // see generate_call_stub
1473 // Save space for one monitor to get into the interpreted method in case
1474 // the method is synchronized
1475 int monitor_size = method->is_synchronized() ?
1476 1*frame::interpreter_frame_monitor_size() : 0;
1477
1478 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1479 // be sure to change this if you add/subtract anything to/from the overhead area
1480 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1481
1482 const int extra_stack = methodOopDesc::extra_stack_entries();
1483 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1484 Interpreter::stackElementWords;
1485 return overhead_size + method_stack + stub_code;
1486 }
1487
1488 // asm based interpreter deoptimization helpers
1489
1490 int AbstractInterpreter::layout_activation(methodOop method,
1491 int tempcount,
1492 int popframe_extra_args,
1493 int moncount,
1494 int callee_param_count,
1495 int callee_locals,
1496 frame* caller,
1497 frame* interpreter_frame,
1498 bool is_top_frame) {
1499 // Note: This calculation must exactly parallel the frame setup
1500 // in AbstractInterpreterGenerator::generate_method_entry.
1501 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1502 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1503 // as determined by a previous call to this method.
1504 // It is also guaranteed to be walkable even though it is in a skeletal state
1505 // NOTE: return size is in words not bytes
1506
1507 // fixed size of an interpreter frame:
1508 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1509 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1510 Interpreter::stackElementWords;
1511
1512 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1513
1514 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1515 // Since the callee parameters already account for the callee's params we only need to account for
1516 // the extra locals.
1517
1518
1519 int size = overhead +
1520 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1521 (moncount*frame::interpreter_frame_monitor_size()) +
1522 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1523
1524 if (interpreter_frame != NULL) {
1525 #ifdef ASSERT
1526 if (!EnableMethodHandles)
1527 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1528 // Probably, since deoptimization doesn't work yet.
1529 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1530 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1531 #endif
1532
1533 interpreter_frame->interpreter_frame_set_method(method);
1534 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1535 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1536 // and sender_sp is fp+8
1537 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1538
1539 interpreter_frame->interpreter_frame_set_locals(locals);
1540 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1541 BasicObjectLock* monbot = montop - moncount;
1542 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1543
1544 // Set last_sp
1545 intptr_t* rsp = (intptr_t*) monbot -
1546 tempcount*Interpreter::stackElementWords -
1547 popframe_extra_args;
1548 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1549
1550 // All frames but the initial (oldest) interpreter frame we fill in have a
1551 // value for sender_sp that allows walking the stack but isn't
1552 // truly correct. Correct the value here.
1553
1554 if (extra_locals != 0 &&
1555 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1556 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1557 }
1558 *interpreter_frame->interpreter_frame_cache_addr() =
1559 method->constants()->cache();
1560 }
1561 return size;
1562 }
1563
1564
1565 //------------------------------------------------------------------------------------------------------------------------
1566 // Exceptions
1567
1568 void TemplateInterpreterGenerator::generate_throw_exception() {
1569 // Entry point in previous activation (i.e., if the caller was interpreted)
1570 Interpreter::_rethrow_exception_entry = __ pc();
1571 const Register thread = rcx;
1572
1573 // Restore sp to interpreter_frame_last_sp even though we are going
1574 // to empty the expression stack for the exception processing.
1575 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1576 // rax,: exception
1577 // rdx: return address/pc that threw exception
1578 __ restore_bcp(); // rsi points to call/send
1579 __ restore_locals();
1580
1581 // Entry point for exceptions thrown within interpreter code
1582 Interpreter::_throw_exception_entry = __ pc();
1583 // expression stack is undefined here
1584 // rax,: exception
1585 // rsi: exception bcp
1586 __ verify_oop(rax);
1587
1588 // expression stack must be empty before entering the VM in case of an exception
1589 __ empty_expression_stack();
1590 __ empty_FPU_stack();
1591 // find exception handler address and preserve exception oop
1592 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1593 // rax,: exception handler entry point
1594 // rdx: preserved exception oop
1595 // rsi: bcp for exception handler
1596 __ push_ptr(rdx); // push exception which is now the only value on the stack
1597 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1598
1599 // If the exception is not handled in the current frame the frame is removed and
1600 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1601 //
1602 // Note: At this point the bci is still the bxi for the instruction which caused
1603 // the exception and the expression stack is empty. Thus, for any VM calls
1604 // at this point, GC will find a legal oop map (with empty expression stack).
1605
1606 // In current activation
1607 // tos: exception
1608 // rsi: exception bcp
1609
1610 //
1611 // JVMTI PopFrame support
1612 //
1613
1614 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1615 __ empty_expression_stack();
1616 __ empty_FPU_stack();
1617 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1618 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1619 // popframe handling cycles.
1620 __ get_thread(thread);
1621 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1622 __ orl(rdx, JavaThread::popframe_processing_bit);
1623 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1624
1625 {
1626 // Check to see whether we are returning to a deoptimized frame.
1627 // (The PopFrame call ensures that the caller of the popped frame is
1628 // either interpreted or compiled and deoptimizes it if compiled.)
1629 // In this case, we can't call dispatch_next() after the frame is
1630 // popped, but instead must save the incoming arguments and restore
1631 // them after deoptimization has occurred.
1632 //
1633 // Note that we don't compare the return PC against the
1634 // deoptimization blob's unpack entry because of the presence of
1635 // adapter frames in C2.
1636 Label caller_not_deoptimized;
1637 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1638 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1639 __ testl(rax, rax);
1640 __ jcc(Assembler::notZero, caller_not_deoptimized);
1641
1642 // Compute size of arguments for saving when returning to deoptimized caller
1643 __ get_method(rax);
1644 __ verify_oop(rax);
1645 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1646 __ shlptr(rax, Interpreter::logStackElementSize);
1647 __ restore_locals();
1648 __ subptr(rdi, rax);
1649 __ addptr(rdi, wordSize);
1650 // Save these arguments
1651 __ get_thread(thread);
1652 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1653
1654 __ remove_activation(vtos, rdx,
1655 /* throw_monitor_exception */ false,
1656 /* install_monitor_exception */ false,
1657 /* notify_jvmdi */ false);
1658
1659 // Inform deoptimization that it is responsible for restoring these arguments
1660 __ get_thread(thread);
1661 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1662
1663 // Continue in deoptimization handler
1664 __ jmp(rdx);
1665
1666 __ bind(caller_not_deoptimized);
1667 }
1668
1669 __ remove_activation(vtos, rdx,
1670 /* throw_monitor_exception */ false,
1671 /* install_monitor_exception */ false,
1672 /* notify_jvmdi */ false);
1673
1674 // Finish with popframe handling
1675 // A previous I2C followed by a deoptimization might have moved the
1676 // outgoing arguments further up the stack. PopFrame expects the
1677 // mutations to those outgoing arguments to be preserved and other
1678 // constraints basically require this frame to look exactly as
1679 // though it had previously invoked an interpreted activation with
1680 // no space between the top of the expression stack (current
1681 // last_sp) and the top of stack. Rather than force deopt to
1682 // maintain this kind of invariant all the time we call a small
1683 // fixup routine to move the mutated arguments onto the top of our
1684 // expression stack if necessary.
1685 __ mov(rax, rsp);
1686 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1687 __ get_thread(thread);
1688 // PC must point into interpreter here
1689 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1690 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1691 __ get_thread(thread);
1692 __ reset_last_Java_frame(thread, true, true);
1693 // Restore the last_sp and null it out
1694 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1695 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1696
1697 __ restore_bcp();
1698 __ restore_locals();
1699 // The method data pointer was incremented already during
1700 // call profiling. We have to restore the mdp for the current bcp.
1701 if (ProfileInterpreter) {
1702 __ set_method_data_pointer_for_bcp();
1703 }
1704
1705 // Clear the popframe condition flag
1706 __ get_thread(thread);
1707 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1708
1709 __ dispatch_next(vtos);
1710 // end of PopFrame support
1711
1712 Interpreter::_remove_activation_entry = __ pc();
1713
1714 // preserve exception over this code sequence
1715 __ pop_ptr(rax);
1716 __ get_thread(thread);
1717 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1718 // remove the activation (without doing throws on illegalMonitorExceptions)
1719 __ remove_activation(vtos, rdx, false, true, false);
1720 // restore exception
1721 __ get_thread(thread);
1722 __ movptr(rax, Address(thread, JavaThread::vm_result_offset()));
1723 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
1724 __ verify_oop(rax);
1725
1726 // Inbetween activations - previous activation type unknown yet
1727 // compute continuation point - the continuation point expects
1728 // the following registers set up:
1729 //
1730 // rax: exception
1731 // rdx: return address/pc that threw exception
1732 // rsp: expression stack of caller
1733 // rbp: rbp, of caller
1734 __ push(rax); // save exception
1735 __ push(rdx); // save return address
1736 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1737 __ mov(rbx, rax); // save exception handler
1738 __ pop(rdx); // restore return address
1739 __ pop(rax); // restore exception
1740 // Note that an "issuing PC" is actually the next PC after the call
1741 __ jmp(rbx); // jump to exception handler of caller
1742 }
1743
1744
1745 //
1746 // JVMTI ForceEarlyReturn support
1747 //
1748 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1749 address entry = __ pc();
1750 const Register thread = rcx;
1751
1752 __ restore_bcp();
1753 __ restore_locals();
1754 __ empty_expression_stack();
1755 __ empty_FPU_stack();
1756 __ load_earlyret_value(state);
1757
1758 __ get_thread(thread);
1759 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1760 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1761
1762 // Clear the earlyret state
1763 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1764
1765 __ remove_activation(state, rsi,
1766 false, /* throw_monitor_exception */
1767 false, /* install_monitor_exception */
1768 true); /* notify_jvmdi */
1769 __ jmp(rsi);
1770 return entry;
1771 } // end of ForceEarlyReturn support
1772
1773
1774 //------------------------------------------------------------------------------------------------------------------------
1775 // Helper for vtos entry point generation
1776
1777 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) {
1778 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1779 Label L;
1780 fep = __ pc(); __ push(ftos); __ jmp(L);
1781 dep = __ pc(); __ push(dtos); __ jmp(L);
1782 lep = __ pc(); __ push(ltos); __ jmp(L);
1783 aep = __ pc(); __ push(atos); __ jmp(L);
1784 bep = cep = sep = // fall through
1785 iep = __ pc(); __ push(itos); // fall through
1786 vep = __ pc(); __ bind(L); // fall through
1787 generate_and_dispatch(t);
1788 }
1789
1790 //------------------------------------------------------------------------------------------------------------------------
1791 // Generation of individual instructions
1792
1793 // helpers for generate_and_dispatch
1794
1795
1796
1797 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1798 : TemplateInterpreterGenerator(code) {
1799 generate_all(); // down here so it can be "virtual"
1800 }
1801
1802 //------------------------------------------------------------------------------------------------------------------------
1803
1804 // Non-product code
1805 #ifndef PRODUCT
1806 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1807 address entry = __ pc();
1808
1809 // prepare expression stack
1810 __ pop(rcx); // pop return address so expression stack is 'pure'
1811 __ push(state); // save tosca
1812
1813 // pass tosca registers as arguments & call tracer
1814 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1815 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1816 __ pop(state); // restore tosca
1817
1818 // return
1819 __ jmp(rcx);
1820
1821 return entry;
1822 }
1823
1824
1825 void TemplateInterpreterGenerator::count_bytecode() {
1826 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1827 }
1828
1829
1830 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1831 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1832 }
1833
1834
1835 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1836 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1837 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1838 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1839 ExternalAddress table((address) BytecodePairHistogram::_counters);
1840 Address index(noreg, rbx, Address::times_4);
1841 __ incrementl(ArrayAddress(table, index));
1842 }
1843
1844
1845 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1846 // Call a little run-time stub to avoid blow-up for each bytecode.
1847 // The run-time runtime saves the right registers, depending on
1848 // the tosca in-state for the given template.
1849 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1850 "entry must have been generated");
1851 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1852 }
1853
1854
1855 void TemplateInterpreterGenerator::stop_interpreter_at() {
1856 Label L;
1857 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1858 StopInterpreterAt);
1859 __ jcc(Assembler::notEqual, L);
1860 __ int3();
1861 __ bind(L);
1862 }
1863 #endif // !PRODUCT
1864 #endif // CC_INTERP