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) {
154 TosState incoming_state = state;
155 address entry = __ pc();
156
157 #ifdef COMPILER2
158 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
159 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
160 for (int i = 1; i < 8; i++) {
161 __ ffree(i);
162 }
163 } else if (UseSSE < 2) {
164 __ empty_FPU_stack();
165 }
166 #endif
167 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
168 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
169 } else {
170 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
171 }
172
173 // In SSE mode, interpreter returns FP results in xmm0 but they need
174 // to end up back on the FPU so it can operate on them.
175 if (incoming_state == ftos && UseSSE >= 1) {
176 __ subptr(rsp, wordSize);
177 __ movflt(Address(rsp, 0), xmm0);
178 __ fld_s(Address(rsp, 0));
179 __ addptr(rsp, wordSize);
180 } else if (incoming_state == dtos && UseSSE >= 2) {
181 __ subptr(rsp, 2*wordSize);
182 __ movdbl(Address(rsp, 0), xmm0);
183 __ fld_d(Address(rsp, 0));
184 __ addptr(rsp, 2*wordSize);
185 }
186
187 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
188
189 // Restore stack bottom in case i2c adjusted stack
190 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
191 // and NULL it as marker that rsp is now tos until next java call
192 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
193
194 __ restore_bcp();
195 __ restore_locals();
196
197 Label L_got_cache, L_giant_index;
198 if (EnableInvokeDynamic) {
199 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
200 __ jcc(Assembler::equal, L_giant_index);
201 }
202 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
203 __ bind(L_got_cache);
204 __ movl(rbx, Address(rbx, rcx,
205 Address::times_ptr, ConstantPoolCache::base_offset() +
206 ConstantPoolCacheEntry::flags_offset()));
207 __ andptr(rbx, 0xFF);
208 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
209 __ dispatch_next(state, step);
210
211 // out of the main line of code...
212 if (EnableInvokeDynamic) {
213 __ bind(L_giant_index);
214 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
215 __ jmp(L_got_cache);
216 }
217
218 return entry;
219 }
220
221
222 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
223 address entry = __ pc();
224
225 // In SSE mode, FP results are in xmm0
226 if (state == ftos && UseSSE > 0) {
227 __ subptr(rsp, wordSize);
228 __ movflt(Address(rsp, 0), xmm0);
229 __ fld_s(Address(rsp, 0));
230 __ addptr(rsp, wordSize);
231 } else if (state == dtos && UseSSE >= 2) {
232 __ subptr(rsp, 2*wordSize);
233 __ movdbl(Address(rsp, 0), xmm0);
234 __ fld_d(Address(rsp, 0));
235 __ addptr(rsp, 2*wordSize);
236 }
237
238 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
239
240 // The stack is not extended by deopt but we must NULL last_sp as this
241 // entry is like a "return".
242 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
243 __ restore_bcp();
244 __ restore_locals();
245 // handle exceptions
246 { Label L;
247 const Register thread = rcx;
248 __ get_thread(thread);
249 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
250 __ jcc(Assembler::zero, L);
251 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
252 __ should_not_reach_here();
253 __ bind(L);
254 }
255 __ dispatch_next(state, step);
256 return entry;
257 }
258
259
260 int AbstractInterpreter::BasicType_as_index(BasicType type) {
261 int i = 0;
262 switch (type) {
263 case T_BOOLEAN: i = 0; break;
264 case T_CHAR : i = 1; break;
265 case T_BYTE : i = 2; break;
266 case T_SHORT : i = 3; break;
267 case T_INT : // fall through
268 case T_LONG : // fall through
269 case T_VOID : i = 4; break;
270 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
271 case T_DOUBLE : i = 6; break;
272 case T_OBJECT : // fall through
273 case T_ARRAY : i = 7; break;
274 default : ShouldNotReachHere();
275 }
276 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
277 return i;
278 }
279
280
281 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
282 address entry = __ pc();
283 switch (type) {
284 case T_BOOLEAN: __ c2bool(rax); break;
285 case T_CHAR : __ andptr(rax, 0xFFFF); break;
286 case T_BYTE : __ sign_extend_byte (rax); break;
287 case T_SHORT : __ sign_extend_short(rax); break;
288 case T_INT : /* nothing to do */ break;
289 case T_DOUBLE :
290 case T_FLOAT :
291 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
292 __ pop(t); // remove return address first
293 // Must return a result for interpreter or compiler. In SSE
294 // mode, results are returned in xmm0 and the FPU stack must
295 // be empty.
296 if (type == T_FLOAT && UseSSE >= 1) {
297 // Load ST0
298 __ fld_d(Address(rsp, 0));
299 // Store as float and empty fpu stack
300 __ fstp_s(Address(rsp, 0));
301 // and reload
302 __ movflt(xmm0, Address(rsp, 0));
303 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
304 __ movdbl(xmm0, Address(rsp, 0));
305 } else {
306 // restore ST0
307 __ fld_d(Address(rsp, 0));
308 }
309 // and pop the temp
310 __ addptr(rsp, 2 * wordSize);
311 __ push(t); // restore return address
312 }
313 break;
314 case T_OBJECT :
315 // retrieve result from frame
316 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
317 // and verify it
318 __ verify_oop(rax);
319 break;
320 default : ShouldNotReachHere();
321 }
322 __ ret(0); // return from result handler
323 return entry;
324 }
325
326 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
327 address entry = __ pc();
328 __ push(state);
329 __ call_VM(noreg, runtime_entry);
330 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
331 return entry;
332 }
333
334
335 // Helpers for commoning out cases in the various type of method entries.
336 //
337
338 // increment invocation count & check for overflow
339 //
340 // Note: checking for negative value instead of overflow
341 // so we have a 'sticky' overflow test
342 //
343 // rbx,: method
344 // rcx: invocation counter
345 //
346 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
347 Label done;
348 // Note: In tiered we increment either counters in MethodCounters* or in MDO
349 // depending if we're profiling or not.
350 if (TieredCompilation) {
351 int increment = InvocationCounter::count_increment;
352 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
353 Label no_mdo;
354 if (ProfileInterpreter) {
355 // Are we profiling?
356 __ movptr(rax, Address(rbx, Method::method_data_offset()));
357 __ testptr(rax, rax);
358 __ jccb(Assembler::zero, no_mdo);
359 // Increment counter in the MDO
360 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
361 in_bytes(InvocationCounter::counter_offset()));
362 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
363 __ jmp(done);
364 }
365 __ bind(no_mdo);
366 // Increment counter in MethodCounters
367 const Address invocation_counter(rax,
368 MethodCounters::invocation_counter_offset() +
369 InvocationCounter::counter_offset());
370
371 __ get_method_counters(rbx, rax, done);
372 __ increment_mask_and_jump(invocation_counter, increment, mask,
373 rcx, false, Assembler::zero, overflow);
374 __ bind(done);
375 } else {
376 const Address backedge_counter (rax,
377 MethodCounters::backedge_counter_offset() +
378 InvocationCounter::counter_offset());
379 const Address invocation_counter(rax,
380 MethodCounters::invocation_counter_offset() +
381 InvocationCounter::counter_offset());
382
383 __ get_method_counters(rbx, rax, done);
384
385 if (ProfileInterpreter) {
386 __ incrementl(Address(rax,
387 MethodCounters::interpreter_invocation_counter_offset()));
388 }
389
390 // Update standard invocation counters
391 __ movl(rcx, invocation_counter);
392 __ incrementl(rcx, InvocationCounter::count_increment);
393 __ movl(invocation_counter, rcx); // save invocation count
394
395 __ movl(rax, backedge_counter); // load backedge counter
396 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
397
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 __ bind(done);
418 }
419 }
420
421 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
422
423 // Asm interpreter on entry
424 // rdi - locals
425 // rsi - bcp
426 // rbx, - method
427 // rdx - cpool
428 // rbp, - interpreter frame
429
430 // C++ interpreter on entry
431 // rsi - new interpreter state pointer
432 // rbp - interpreter frame pointer
433 // rbx - method
434
435 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
436 // rbx, - method
437 // rcx - rcvr (assuming there is one)
438 // top of stack return address of interpreter caller
439 // rsp - sender_sp
440
441 // C++ interpreter only
442 // rsi - previous interpreter state pointer
443
444 // InterpreterRuntime::frequency_counter_overflow takes one argument
445 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
446 // The call returns the address of the verified entry point for the method or NULL
447 // if the compilation did not complete (either went background or bailed out).
448 __ movptr(rax, (intptr_t)false);
449 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
450
451 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
452
453 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
454 // and jump to the interpreted entry.
455 __ jmp(*do_continue, relocInfo::none);
456
457 }
458
459 void InterpreterGenerator::generate_stack_overflow_check(void) {
460 // see if we've got enough room on the stack for locals plus overhead.
461 // the expression stack grows down incrementally, so the normal guard
462 // page mechanism will work for that.
463 //
464 // Registers live on entry:
465 //
466 // Asm interpreter
467 // rdx: number of additional locals this frame needs (what we must check)
468 // rbx,: Method*
469
470 // destroyed on exit
471 // rax,
472
473 // NOTE: since the additional locals are also always pushed (wasn't obvious in
474 // generate_method_entry) so the guard should work for them too.
475 //
476
477 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
478 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
479
480 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
481 // be sure to change this if you add/subtract anything to/from the overhead area
482 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
483
484 const int page_size = os::vm_page_size();
485
486 Label after_frame_check;
487
488 // see if the frame is greater than one page in size. If so,
489 // then we need to verify there is enough stack space remaining
490 // for the additional locals.
491 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
492 __ jcc(Assembler::belowEqual, after_frame_check);
493
494 // compute rsp as if this were going to be the last frame on
495 // the stack before the red zone
496
497 Label after_frame_check_pop;
498
499 __ push(rsi);
500
501 const Register thread = rsi;
502
503 __ get_thread(thread);
504
505 const Address stack_base(thread, Thread::stack_base_offset());
506 const Address stack_size(thread, Thread::stack_size_offset());
507
508 // locals + overhead, in bytes
509 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
510
511 #ifdef ASSERT
512 Label stack_base_okay, stack_size_okay;
513 // verify that thread stack base is non-zero
514 __ cmpptr(stack_base, (int32_t)NULL_WORD);
515 __ jcc(Assembler::notEqual, stack_base_okay);
516 __ stop("stack base is zero");
517 __ bind(stack_base_okay);
518 // verify that thread stack size is non-zero
519 __ cmpptr(stack_size, 0);
520 __ jcc(Assembler::notEqual, stack_size_okay);
521 __ stop("stack size is zero");
522 __ bind(stack_size_okay);
523 #endif
524
525 // Add stack base to locals and subtract stack size
526 __ addptr(rax, stack_base);
527 __ subptr(rax, stack_size);
528
529 // Use the maximum number of pages we might bang.
530 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
531 (StackRedPages+StackYellowPages);
532 __ addptr(rax, max_pages * page_size);
533
534 // check against the current stack bottom
535 __ cmpptr(rsp, rax);
536 __ jcc(Assembler::above, after_frame_check_pop);
537
538 __ pop(rsi); // get saved bcp / (c++ prev state ).
539
540 // Restore sender's sp as SP. This is necessary if the sender's
541 // frame is an extended compiled frame (see gen_c2i_adapter())
542 // and safer anyway in case of JSR292 adaptations.
543
544 __ pop(rax); // return address must be moved if SP is changed
545 __ mov(rsp, rsi);
546 __ push(rax);
547
548 // Note: the restored frame is not necessarily interpreted.
549 // Use the shared runtime version of the StackOverflowError.
550 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
551 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
552 // all done with frame size check
553 __ bind(after_frame_check_pop);
554 __ pop(rsi);
555
556 __ bind(after_frame_check);
557 }
558
559 // Allocate monitor and lock method (asm interpreter)
560 // rbx, - Method*
561 //
562 void InterpreterGenerator::lock_method(void) {
563 // synchronize method
564 const Address access_flags (rbx, Method::access_flags_offset());
565 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
566 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
567
568 #ifdef ASSERT
569 { Label L;
570 __ movl(rax, access_flags);
571 __ testl(rax, JVM_ACC_SYNCHRONIZED);
572 __ jcc(Assembler::notZero, L);
573 __ stop("method doesn't need synchronization");
574 __ bind(L);
575 }
576 #endif // ASSERT
577 // get synchronization object
578 { Label done;
579 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
580 __ movl(rax, access_flags);
581 __ testl(rax, JVM_ACC_STATIC);
582 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
583 __ jcc(Assembler::zero, done);
584 __ movptr(rax, Address(rbx, Method::const_offset()));
585 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
586 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
587 __ movptr(rax, Address(rax, mirror_offset));
588 __ bind(done);
589 }
590 // add space for monitor & lock
591 __ subptr(rsp, entry_size); // add space for a monitor entry
592 __ movptr(monitor_block_top, rsp); // set new monitor block top
593 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
594 __ mov(rdx, rsp); // object address
595 __ lock_object(rdx);
596 }
597
598 //
599 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
600 // and for native methods hence the shared code.
601
602 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
603 // initialize fixed part of activation frame
604 __ push(rax); // save return address
605 __ enter(); // save old & set new rbp,
606
607
608 __ push(rsi); // set sender sp
609 __ push((int32_t)NULL_WORD); // leave last_sp as null
610 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
611 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
612 __ push(rbx); // save Method*
613 if (ProfileInterpreter) {
614 Label method_data_continue;
615 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
616 __ testptr(rdx, rdx);
617 __ jcc(Assembler::zero, method_data_continue);
618 __ addptr(rdx, in_bytes(MethodData::data_offset()));
619 __ bind(method_data_continue);
620 __ push(rdx); // set the mdp (method data pointer)
621 } else {
622 __ push(0);
623 }
624
625 __ movptr(rdx, Address(rbx, Method::const_offset()));
626 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
627 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
628 __ push(rdx); // set constant pool cache
629 __ push(rdi); // set locals pointer
630 if (native_call) {
631 __ push(0); // no bcp
632 } else {
633 __ push(rsi); // set bcp
634 }
635 __ push(0); // reserve word for pointer to expression stack bottom
636 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
637 }
638
639 // End of helpers
640
641 //
642 // Various method entries
643 //------------------------------------------------------------------------------------------------------------------------
644 //
645 //
646
647 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
648
649 address InterpreterGenerator::generate_accessor_entry(void) {
650
651 // rbx,: Method*
652 // rcx: receiver (preserve for slow entry into asm interpreter)
653
654 // rsi: senderSP must preserved for slow path, set SP to it on fast path
655
656 address entry_point = __ pc();
657 Label xreturn_path;
658
659 // do fastpath for resolved accessor methods
660 if (UseFastAccessorMethods) {
661 Label slow_path;
662 // If we need a safepoint check, generate full interpreter entry.
663 ExternalAddress state(SafepointSynchronize::address_of_state());
664 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
665 SafepointSynchronize::_not_synchronized);
666
667 __ jcc(Assembler::notEqual, slow_path);
668 // ASM/C++ Interpreter
669 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
670 // Note: We can only use this code if the getfield has been resolved
671 // and if we don't have a null-pointer exception => check for
672 // these conditions first and use slow path if necessary.
673 // rbx,: method
674 // rcx: receiver
675 __ movptr(rax, Address(rsp, wordSize));
676
677 // check if local 0 != NULL and read field
678 __ testptr(rax, rax);
679 __ jcc(Assembler::zero, slow_path);
680
681 // read first instruction word and extract bytecode @ 1 and index @ 2
682 __ movptr(rdx, Address(rbx, Method::const_offset()));
683 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
684 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
685 // Shift codes right to get the index on the right.
686 // The bytecode fetched looks like <index><0xb4><0x2a>
687 __ shrl(rdx, 2*BitsPerByte);
688 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
689 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
690
691 // rax,: local 0
692 // rbx,: method
693 // rcx: receiver - do not destroy since it is needed for slow path!
694 // rcx: scratch
695 // rdx: constant pool cache index
696 // rdi: constant pool cache
697 // rsi: sender sp
698
699 // check if getfield has been resolved and read constant pool cache entry
700 // check the validity of the cache entry by testing whether _indices field
701 // contains Bytecode::_getfield in b1 byte.
702 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
703 __ movl(rcx,
704 Address(rdi,
705 rdx,
706 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
707 __ shrl(rcx, 2*BitsPerByte);
708 __ andl(rcx, 0xFF);
709 __ cmpl(rcx, Bytecodes::_getfield);
710 __ jcc(Assembler::notEqual, slow_path);
711
712 // Note: constant pool entry is not valid before bytecode is resolved
713 __ movptr(rcx,
714 Address(rdi,
715 rdx,
716 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
717 __ movl(rdx,
718 Address(rdi,
719 rdx,
720 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
721
722 Label notByte, notShort, notChar;
723 const Address field_address (rax, rcx, Address::times_1);
724
725 // Need to differentiate between igetfield, agetfield, bgetfield etc.
726 // because they are different sizes.
727 // Use the type from the constant pool cache
728 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
729 // Make sure we don't need to mask rdx after the above shift
730 ConstantPoolCacheEntry::verify_tos_state_shift();
731 __ cmpl(rdx, btos);
732 __ jcc(Assembler::notEqual, notByte);
733 __ load_signed_byte(rax, field_address);
734 __ jmp(xreturn_path);
735
736 __ bind(notByte);
737 __ cmpl(rdx, stos);
738 __ jcc(Assembler::notEqual, notShort);
739 __ load_signed_short(rax, field_address);
740 __ jmp(xreturn_path);
741
742 __ bind(notShort);
743 __ cmpl(rdx, ctos);
744 __ jcc(Assembler::notEqual, notChar);
745 __ load_unsigned_short(rax, field_address);
746 __ jmp(xreturn_path);
747
748 __ bind(notChar);
749 #ifdef ASSERT
750 Label okay;
751 __ cmpl(rdx, atos);
752 __ jcc(Assembler::equal, okay);
753 __ cmpl(rdx, itos);
754 __ jcc(Assembler::equal, okay);
755 __ stop("what type is this?");
756 __ bind(okay);
757 #endif // ASSERT
758 // All the rest are a 32 bit wordsize
759 // This is ok for now. Since fast accessors should be going away
760 __ movptr(rax, field_address);
761
762 __ bind(xreturn_path);
763
764 // _ireturn/_areturn
765 __ pop(rdi); // get return address
766 __ mov(rsp, rsi); // set sp to sender sp
767 __ jmp(rdi);
768
769 // generate a vanilla interpreter entry as the slow path
770 __ bind(slow_path);
771
772 (void) generate_normal_entry(false);
773 return entry_point;
774 }
775 return NULL;
776
777 }
778
779 // Method entry for java.lang.ref.Reference.get.
780 address InterpreterGenerator::generate_Reference_get_entry(void) {
781 #if INCLUDE_ALL_GCS
782 // Code: _aload_0, _getfield, _areturn
783 // parameter size = 1
784 //
785 // The code that gets generated by this routine is split into 2 parts:
786 // 1. The "intrinsified" code for G1 (or any SATB based GC),
787 // 2. The slow path - which is an expansion of the regular method entry.
788 //
789 // Notes:-
790 // * In the G1 code we do not check whether we need to block for
791 // a safepoint. If G1 is enabled then we must execute the specialized
792 // code for Reference.get (except when the Reference object is null)
793 // so that we can log the value in the referent field with an SATB
794 // update buffer.
795 // If the code for the getfield template is modified so that the
796 // G1 pre-barrier code is executed when the current method is
797 // Reference.get() then going through the normal method entry
798 // will be fine.
799 // * The G1 code below can, however, check the receiver object (the instance
800 // of java.lang.Reference) and jump to the slow path if null. If the
801 // Reference object is null then we obviously cannot fetch the referent
802 // and so we don't need to call the G1 pre-barrier. Thus we can use the
803 // regular method entry code to generate the NPE.
804 //
805 // This code is based on generate_accessor_enty.
806
807 // rbx,: Method*
808 // rcx: receiver (preserve for slow entry into asm interpreter)
809
810 // rsi: senderSP must preserved for slow path, set SP to it on fast path
811
812 address entry = __ pc();
813
814 const int referent_offset = java_lang_ref_Reference::referent_offset;
815 guarantee(referent_offset > 0, "referent offset not initialized");
816
817 if (UseG1GC) {
818 Label slow_path;
819
820 // Check if local 0 != NULL
821 // If the receiver is null then it is OK to jump to the slow path.
822 __ movptr(rax, Address(rsp, wordSize));
823 __ testptr(rax, rax);
824 __ jcc(Assembler::zero, slow_path);
825
826 // rax: local 0 (must be preserved across the G1 barrier call)
827 //
828 // rbx: method (at this point it's scratch)
829 // rcx: receiver (at this point it's scratch)
830 // rdx: scratch
831 // rdi: scratch
832 //
833 // rsi: sender sp
834
835 // Preserve the sender sp in case the pre-barrier
836 // calls the runtime
837 __ push(rsi);
838
839 // Load the value of the referent field.
840 const Address field_address(rax, referent_offset);
841 __ movptr(rax, field_address);
842
843 // Generate the G1 pre-barrier code to log the value of
844 // the referent field in an SATB buffer.
845 __ get_thread(rcx);
846 __ g1_write_barrier_pre(noreg /* obj */,
847 rax /* pre_val */,
848 rcx /* thread */,
849 rbx /* tmp */,
850 true /* tosca_save */,
851 true /* expand_call */);
852
853 // _areturn
854 __ pop(rsi); // get sender sp
855 __ pop(rdi); // get return address
856 __ mov(rsp, rsi); // set sp to sender sp
857 __ jmp(rdi);
858
859 __ bind(slow_path);
860 (void) generate_normal_entry(false);
861
862 return entry;
863 }
864 #endif // INCLUDE_ALL_GCS
865
866 // If G1 is not enabled then attempt to go through the accessor entry point
867 // Reference.get is an accessor
868 return generate_accessor_entry();
869 }
870
871 //
872 // Interpreter stub for calling a native method. (asm interpreter)
873 // This sets up a somewhat different looking stack for calling the native method
874 // than the typical interpreter frame setup.
875 //
876
877 address InterpreterGenerator::generate_native_entry(bool synchronized) {
878 // determine code generation flags
879 bool inc_counter = UseCompiler || CountCompiledCalls;
880
881 // rbx,: Method*
882 // rsi: sender sp
883 // rsi: previous interpreter state (C++ interpreter) must preserve
884 address entry_point = __ pc();
885
886 const Address constMethod (rbx, Method::const_offset());
887 const Address access_flags (rbx, Method::access_flags_offset());
888 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
889
890 // get parameter size (always needed)
891 __ movptr(rcx, constMethod);
892 __ load_unsigned_short(rcx, size_of_parameters);
893
894 // native calls don't need the stack size check since they have no expression stack
895 // and the arguments are already on the stack and we only add a handful of words
896 // to the stack
897
898 // rbx,: Method*
899 // rcx: size of parameters
900 // rsi: sender sp
901
902 __ pop(rax); // get return address
903 // for natives the size of locals is zero
904
905 // compute beginning of parameters (rdi)
906 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
907
908
909 // add 2 zero-initialized slots for native calls
910 // NULL result handler
911 __ push((int32_t)NULL_WORD);
912 // NULL oop temp (mirror or jni oop result)
913 __ push((int32_t)NULL_WORD);
914
915 // initialize fixed part of activation frame
916 generate_fixed_frame(true);
917
918 // make sure method is native & not abstract
919 #ifdef ASSERT
920 __ movl(rax, access_flags);
921 {
922 Label L;
923 __ testl(rax, JVM_ACC_NATIVE);
924 __ jcc(Assembler::notZero, L);
925 __ stop("tried to execute non-native method as native");
926 __ bind(L);
927 }
928 { Label L;
929 __ testl(rax, JVM_ACC_ABSTRACT);
930 __ jcc(Assembler::zero, L);
931 __ stop("tried to execute abstract method in interpreter");
932 __ bind(L);
933 }
934 #endif
935
936 // Since at this point in the method invocation the exception handler
937 // would try to exit the monitor of synchronized methods which hasn't
938 // been entered yet, we set the thread local variable
939 // _do_not_unlock_if_synchronized to true. The remove_activation will
940 // check this flag.
941
942 __ get_thread(rax);
943 const Address do_not_unlock_if_synchronized(rax,
944 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
945 __ movbool(do_not_unlock_if_synchronized, true);
946
947 // increment invocation count & check for overflow
948 Label invocation_counter_overflow;
949 if (inc_counter) {
950 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
951 }
952
953 Label continue_after_compile;
954 __ bind(continue_after_compile);
955
956 bang_stack_shadow_pages(true);
957
958 // reset the _do_not_unlock_if_synchronized flag
959 __ get_thread(rax);
960 __ movbool(do_not_unlock_if_synchronized, false);
961
962 // check for synchronized methods
963 // Must happen AFTER invocation_counter check and stack overflow check,
964 // so method is not locked if overflows.
965 //
966 if (synchronized) {
967 lock_method();
968 } else {
969 // no synchronization necessary
970 #ifdef ASSERT
971 { Label L;
972 __ movl(rax, access_flags);
973 __ testl(rax, JVM_ACC_SYNCHRONIZED);
974 __ jcc(Assembler::zero, L);
975 __ stop("method needs synchronization");
976 __ bind(L);
977 }
978 #endif
979 }
980
981 // start execution
982 #ifdef ASSERT
983 { Label L;
984 const Address monitor_block_top (rbp,
985 frame::interpreter_frame_monitor_block_top_offset * wordSize);
986 __ movptr(rax, monitor_block_top);
987 __ cmpptr(rax, rsp);
988 __ jcc(Assembler::equal, L);
989 __ stop("broken stack frame setup in interpreter");
990 __ bind(L);
991 }
992 #endif
993
994 // jvmti/dtrace support
995 __ notify_method_entry();
996
997 // work registers
998 const Register method = rbx;
999 const Register thread = rdi;
1000 const Register t = rcx;
1001
1002 // allocate space for parameters
1003 __ get_method(method);
1004 __ movptr(t, Address(method, Method::const_offset()));
1005 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1006
1007 __ shlptr(t, Interpreter::logStackElementSize);
1008 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1009 __ subptr(rsp, t);
1010 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1011
1012 // get signature handler
1013 { Label L;
1014 __ movptr(t, Address(method, Method::signature_handler_offset()));
1015 __ testptr(t, t);
1016 __ jcc(Assembler::notZero, L);
1017 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1018 __ get_method(method);
1019 __ movptr(t, Address(method, Method::signature_handler_offset()));
1020 __ bind(L);
1021 }
1022
1023 // call signature handler
1024 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1025 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1026 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1027 // The generated handlers do not touch RBX (the method oop).
1028 // However, large signatures cannot be cached and are generated
1029 // each time here. The slow-path generator will blow RBX
1030 // sometime, so we must reload it after the call.
1031 __ call(t);
1032 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1033
1034 // result handler is in rax,
1035 // set result handler
1036 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1037
1038 // pass mirror handle if static call
1039 { Label L;
1040 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1041 __ movl(t, Address(method, Method::access_flags_offset()));
1042 __ testl(t, JVM_ACC_STATIC);
1043 __ jcc(Assembler::zero, L);
1044 // get mirror
1045 __ movptr(t, Address(method, Method:: const_offset()));
1046 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1047 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1048 __ movptr(t, Address(t, mirror_offset));
1049 // copy mirror into activation frame
1050 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1051 // pass handle to mirror
1052 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1053 __ movptr(Address(rsp, wordSize), t);
1054 __ bind(L);
1055 }
1056
1057 // get native function entry point
1058 { Label L;
1059 __ movptr(rax, Address(method, Method::native_function_offset()));
1060 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1061 __ cmpptr(rax, unsatisfied.addr());
1062 __ jcc(Assembler::notEqual, L);
1063 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1064 __ get_method(method);
1065 __ movptr(rax, Address(method, Method::native_function_offset()));
1066 __ bind(L);
1067 }
1068
1069 // pass JNIEnv
1070 __ get_thread(thread);
1071 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1072 __ movptr(Address(rsp, 0), t);
1073
1074 // set_last_Java_frame_before_call
1075 // It is enough that the pc()
1076 // points into the right code segment. It does not have to be the correct return pc.
1077 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1078
1079 // change thread state
1080 #ifdef ASSERT
1081 { Label L;
1082 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1083 __ cmpl(t, _thread_in_Java);
1084 __ jcc(Assembler::equal, L);
1085 __ stop("Wrong thread state in native stub");
1086 __ bind(L);
1087 }
1088 #endif
1089
1090 // Change state to native
1091 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1092 __ call(rax);
1093
1094 // result potentially in rdx:rax or ST0
1095
1096 // Verify or restore cpu control state after JNI call
1097 __ restore_cpu_control_state_after_jni();
1098
1099 // save potential result in ST(0) & rdx:rax
1100 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1101 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1102 // It is safe to do this push because state is _thread_in_native and return address will be found
1103 // via _last_native_pc and not via _last_jave_sp
1104
1105 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1106 // If the order changes or anything else is added to the stack the code in
1107 // interpreter_frame_result will have to be changed.
1108
1109 { Label L;
1110 Label push_double;
1111 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1112 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1113 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1114 float_handler.addr());
1115 __ jcc(Assembler::equal, push_double);
1116 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1117 double_handler.addr());
1118 __ jcc(Assembler::notEqual, L);
1119 __ bind(push_double);
1120 __ push(dtos);
1121 __ bind(L);
1122 }
1123 __ push(ltos);
1124
1125 // change thread state
1126 __ get_thread(thread);
1127 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1128 if(os::is_MP()) {
1129 if (UseMembar) {
1130 // Force this write out before the read below
1131 __ membar(Assembler::Membar_mask_bits(
1132 Assembler::LoadLoad | Assembler::LoadStore |
1133 Assembler::StoreLoad | Assembler::StoreStore));
1134 } else {
1135 // Write serialization page so VM thread can do a pseudo remote membar.
1136 // We use the current thread pointer to calculate a thread specific
1137 // offset to write to within the page. This minimizes bus traffic
1138 // due to cache line collision.
1139 __ serialize_memory(thread, rcx);
1140 }
1141 }
1142
1143 if (AlwaysRestoreFPU) {
1144 // Make sure the control word is correct.
1145 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1146 }
1147
1148 // check for safepoint operation in progress and/or pending suspend requests
1149 { Label Continue;
1150
1151 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1152 SafepointSynchronize::_not_synchronized);
1153
1154 Label L;
1155 __ jcc(Assembler::notEqual, L);
1156 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1157 __ jcc(Assembler::equal, Continue);
1158 __ bind(L);
1159
1160 // Don't use call_VM as it will see a possible pending exception and forward it
1161 // and never return here preventing us from clearing _last_native_pc down below.
1162 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1163 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1164 // by hand.
1165 //
1166 __ push(thread);
1167 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1168 JavaThread::check_special_condition_for_native_trans)));
1169 __ increment(rsp, wordSize);
1170 __ get_thread(thread);
1171
1172 __ bind(Continue);
1173 }
1174
1175 // change thread state
1176 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1177
1178 __ reset_last_Java_frame(thread, true, true);
1179
1180 // reset handle block
1181 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1182 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1183
1184 // If result was an oop then unbox and save it in the frame
1185 { Label L;
1186 Label no_oop, store_result;
1187 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1188 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1189 handler.addr());
1190 __ jcc(Assembler::notEqual, no_oop);
1191 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1192 __ pop(ltos);
1193 __ testptr(rax, rax);
1194 __ jcc(Assembler::zero, store_result);
1195 // unbox
1196 __ movptr(rax, Address(rax, 0));
1197 __ bind(store_result);
1198 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1199 // keep stack depth as expected by pushing oop which will eventually be discarded
1200 __ push(ltos);
1201 __ bind(no_oop);
1202 }
1203
1204 {
1205 Label no_reguard;
1206 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1207 __ jcc(Assembler::notEqual, no_reguard);
1208
1209 __ pusha();
1210 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1211 __ popa();
1212
1213 __ bind(no_reguard);
1214 }
1215
1216 // restore rsi to have legal interpreter frame,
1217 // i.e., bci == 0 <=> rsi == code_base()
1218 // Can't call_VM until bcp is within reasonable.
1219 __ get_method(method); // method is junk from thread_in_native to now.
1220 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1221 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1222
1223 // handle exceptions (exception handling will handle unlocking!)
1224 { Label L;
1225 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1226 __ jcc(Assembler::zero, L);
1227 // Note: At some point we may want to unify this with the code used in call_VM_base();
1228 // i.e., we should use the StubRoutines::forward_exception code. For now this
1229 // doesn't work here because the rsp is not correctly set at this point.
1230 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1231 __ should_not_reach_here();
1232 __ bind(L);
1233 }
1234
1235 // do unlocking if necessary
1236 { Label L;
1237 __ movl(t, Address(method, Method::access_flags_offset()));
1238 __ testl(t, JVM_ACC_SYNCHRONIZED);
1239 __ jcc(Assembler::zero, L);
1240 // the code below should be shared with interpreter macro assembler implementation
1241 { Label unlock;
1242 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1243 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1244 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1245
1246 __ lea(rdx, monitor); // address of first monitor
1247
1248 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1249 __ testptr(t, t);
1250 __ jcc(Assembler::notZero, unlock);
1251
1252 // Entry already unlocked, need to throw exception
1253 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1254 __ should_not_reach_here();
1255
1256 __ bind(unlock);
1257 __ unlock_object(rdx);
1258 }
1259 __ bind(L);
1260 }
1261
1262 // jvmti/dtrace support
1263 // Note: This must happen _after_ handling/throwing any exceptions since
1264 // the exception handler code notifies the runtime of method exits
1265 // too. If this happens before, method entry/exit notifications are
1266 // not properly paired (was bug - gri 11/22/99).
1267 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1268
1269 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1270 __ pop(ltos);
1271 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1272 __ call(t);
1273
1274 // remove activation
1275 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1276 __ leave(); // remove frame anchor
1277 __ pop(rdi); // get return address
1278 __ mov(rsp, t); // set sp to sender sp
1279 __ jmp(rdi);
1280
1281 if (inc_counter) {
1282 // Handle overflow of counter and compile method
1283 __ bind(invocation_counter_overflow);
1284 generate_counter_overflow(&continue_after_compile);
1285 }
1286
1287 return entry_point;
1288 }
1289
1290 //
1291 // Generic interpreted method entry to (asm) interpreter
1292 //
1293 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1294 // determine code generation flags
1295 bool inc_counter = UseCompiler || CountCompiledCalls;
1296
1297 // rbx,: Method*
1298 // rsi: sender sp
1299 address entry_point = __ pc();
1300
1301 const Address constMethod (rbx, Method::const_offset());
1302 const Address access_flags (rbx, Method::access_flags_offset());
1303 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1304 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1305
1306 // get parameter size (always needed)
1307 __ movptr(rdx, constMethod);
1308 __ load_unsigned_short(rcx, size_of_parameters);
1309
1310 // rbx,: Method*
1311 // rcx: size of parameters
1312
1313 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1314
1315 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1316 __ subl(rdx, rcx); // rdx = no. of additional locals
1317
1318 // see if we've got enough room on the stack for locals plus overhead.
1319 generate_stack_overflow_check();
1320
1321 // get return address
1322 __ pop(rax);
1323
1324 // compute beginning of parameters (rdi)
1325 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1326
1327 // rdx - # of additional locals
1328 // allocate space for locals
1329 // explicitly initialize locals
1330 {
1331 Label exit, loop;
1332 __ testl(rdx, rdx);
1333 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1334 __ bind(loop);
1335 __ push((int32_t)NULL_WORD); // initialize local variables
1336 __ decrement(rdx); // until everything initialized
1337 __ jcc(Assembler::greater, loop);
1338 __ bind(exit);
1339 }
1340
1341 // initialize fixed part of activation frame
1342 generate_fixed_frame(false);
1343
1344 // make sure method is not native & not abstract
1345 #ifdef ASSERT
1346 __ movl(rax, access_flags);
1347 {
1348 Label L;
1349 __ testl(rax, JVM_ACC_NATIVE);
1350 __ jcc(Assembler::zero, L);
1351 __ stop("tried to execute native method as non-native");
1352 __ bind(L);
1353 }
1354 { Label L;
1355 __ testl(rax, JVM_ACC_ABSTRACT);
1356 __ jcc(Assembler::zero, L);
1357 __ stop("tried to execute abstract method in interpreter");
1358 __ bind(L);
1359 }
1360 #endif
1361
1362 // Since at this point in the method invocation the exception handler
1363 // would try to exit the monitor of synchronized methods which hasn't
1364 // been entered yet, we set the thread local variable
1365 // _do_not_unlock_if_synchronized to true. The remove_activation will
1366 // check this flag.
1367
1368 __ get_thread(rax);
1369 const Address do_not_unlock_if_synchronized(rax,
1370 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1371 __ movbool(do_not_unlock_if_synchronized, true);
1372
1373 // increment invocation count & check for overflow
1374 Label invocation_counter_overflow;
1375 Label profile_method;
1376 Label profile_method_continue;
1377 if (inc_counter) {
1378 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1379 if (ProfileInterpreter) {
1380 __ bind(profile_method_continue);
1381 }
1382 }
1383 Label continue_after_compile;
1384 __ bind(continue_after_compile);
1385
1386 bang_stack_shadow_pages(false);
1387
1388 // reset the _do_not_unlock_if_synchronized flag
1389 __ get_thread(rax);
1390 __ movbool(do_not_unlock_if_synchronized, false);
1391
1392 // check for synchronized methods
1393 // Must happen AFTER invocation_counter check and stack overflow check,
1394 // so method is not locked if overflows.
1395 //
1396 if (synchronized) {
1397 // Allocate monitor and lock method
1398 lock_method();
1399 } else {
1400 // no synchronization necessary
1401 #ifdef ASSERT
1402 { Label L;
1403 __ movl(rax, access_flags);
1404 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1405 __ jcc(Assembler::zero, L);
1406 __ stop("method needs synchronization");
1407 __ bind(L);
1408 }
1409 #endif
1410 }
1411
1412 // start execution
1413 #ifdef ASSERT
1414 { Label L;
1415 const Address monitor_block_top (rbp,
1416 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1417 __ movptr(rax, monitor_block_top);
1418 __ cmpptr(rax, rsp);
1419 __ jcc(Assembler::equal, L);
1420 __ stop("broken stack frame setup in interpreter");
1421 __ bind(L);
1422 }
1423 #endif
1424
1425 // jvmti support
1426 __ notify_method_entry();
1427
1428 __ dispatch_next(vtos);
1429
1430 // invocation counter overflow
1431 if (inc_counter) {
1432 if (ProfileInterpreter) {
1433 // We have decided to profile this method in the interpreter
1434 __ bind(profile_method);
1435 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1436 __ set_method_data_pointer_for_bcp();
1437 __ get_method(rbx);
1438 __ jmp(profile_method_continue);
1439 }
1440 // Handle overflow of counter and compile method
1441 __ bind(invocation_counter_overflow);
1442 generate_counter_overflow(&continue_after_compile);
1443 }
1444
1445 return entry_point;
1446 }
1447
1448 //------------------------------------------------------------------------------------------------------------------------
1449 // Entry points
1450 //
1451 // Here we generate the various kind of entries into the interpreter.
1452 // The two main entry type are generic bytecode methods and native call method.
1453 // These both come in synchronized and non-synchronized versions but the
1454 // frame layout they create is very similar. The other method entry
1455 // types are really just special purpose entries that are really entry
1456 // and interpretation all in one. These are for trivial methods like
1457 // accessor, empty, or special math methods.
1458 //
1459 // When control flow reaches any of the entry types for the interpreter
1460 // the following holds ->
1461 //
1462 // Arguments:
1463 //
1464 // rbx,: Method*
1465 // rcx: receiver
1466 //
1467 //
1468 // Stack layout immediately at entry
1469 //
1470 // [ return address ] <--- rsp
1471 // [ parameter n ]
1472 // ...
1473 // [ parameter 1 ]
1474 // [ expression stack ] (caller's java expression stack)
1475
1476 // Assuming that we don't go to one of the trivial specialized
1477 // entries the stack will look like below when we are ready to execute
1478 // the first bytecode (or call the native routine). The register usage
1479 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1480 //
1481 // local variables follow incoming parameters immediately; i.e.
1482 // the return address is moved to the end of the locals).
1483 //
1484 // [ monitor entry ] <--- rsp
1485 // ...
1486 // [ monitor entry ]
1487 // [ expr. stack bottom ]
1488 // [ saved rsi ]
1489 // [ current rdi ]
1490 // [ Method* ]
1491 // [ saved rbp, ] <--- rbp,
1492 // [ return address ]
1493 // [ local variable m ]
1494 // ...
1495 // [ local variable 1 ]
1496 // [ parameter n ]
1497 // ...
1498 // [ parameter 1 ] <--- rdi
1499
1500 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1501 // determine code generation flags
1502 bool synchronized = false;
1503 address entry_point = NULL;
1504
1505 switch (kind) {
1506 case Interpreter::zerolocals : break;
1507 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1508 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1509 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1510 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1511 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1512 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1513
1514 case Interpreter::java_lang_math_sin : // fall thru
1515 case Interpreter::java_lang_math_cos : // fall thru
1516 case Interpreter::java_lang_math_tan : // fall thru
1517 case Interpreter::java_lang_math_abs : // fall thru
1518 case Interpreter::java_lang_math_log : // fall thru
1519 case Interpreter::java_lang_math_log10 : // fall thru
1520 case Interpreter::java_lang_math_sqrt : // fall thru
1521 case Interpreter::java_lang_math_pow : // fall thru
1522 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1523 case Interpreter::java_lang_ref_reference_get
1524 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1525 default:
1526 fatal(err_msg("unexpected method kind: %d", kind));
1527 break;
1528 }
1529
1530 if (entry_point) return entry_point;
1531
1532 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1533
1534 }
1535
1536 // These should never be compiled since the interpreter will prefer
1537 // the compiled version to the intrinsic version.
1538 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1539 switch (method_kind(m)) {
1540 case Interpreter::java_lang_math_sin : // fall thru
1541 case Interpreter::java_lang_math_cos : // fall thru
1542 case Interpreter::java_lang_math_tan : // fall thru
1543 case Interpreter::java_lang_math_abs : // fall thru
1544 case Interpreter::java_lang_math_log : // fall thru
1545 case Interpreter::java_lang_math_log10 : // fall thru
1546 case Interpreter::java_lang_math_sqrt : // fall thru
1547 case Interpreter::java_lang_math_pow : // fall thru
1548 case Interpreter::java_lang_math_exp :
1549 return false;
1550 default:
1551 return true;
1552 }
1553 }
1554
1555 // How much stack a method activation needs in words.
1556 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1557
1558 const int stub_code = 4; // see generate_call_stub
1559 // Save space for one monitor to get into the interpreted method in case
1560 // the method is synchronized
1561 int monitor_size = method->is_synchronized() ?
1562 1*frame::interpreter_frame_monitor_size() : 0;
1563
1564 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1565 // be sure to change this if you add/subtract anything to/from the overhead area
1566 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1567
1568 const int method_stack = (method->max_locals() + method->max_stack()) *
1569 Interpreter::stackElementWords;
1570 return overhead_size + method_stack + stub_code;
1571 }
1572
1573 // asm based interpreter deoptimization helpers
1574
1575 int AbstractInterpreter::layout_activation(Method* method,
1576 int tempcount,
1577 int popframe_extra_args,
1578 int moncount,
1579 int caller_actual_parameters,
1580 int callee_param_count,
1581 int callee_locals,
1582 frame* caller,
1583 frame* interpreter_frame,
1584 bool is_top_frame,
1585 bool is_bottom_frame) {
1586 // Note: This calculation must exactly parallel the frame setup
1587 // in AbstractInterpreterGenerator::generate_method_entry.
1588 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1589 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1590 // as determined by a previous call to this method.
1591 // It is also guaranteed to be walkable even though it is in a skeletal state
1592 // NOTE: return size is in words not bytes
1593
1594 // fixed size of an interpreter frame:
1595 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1596 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1597 Interpreter::stackElementWords;
1598
1599 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1600
1601 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1602 // Since the callee parameters already account for the callee's params we only need to account for
1603 // the extra locals.
1604
1605
1606 int size = overhead +
1607 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1608 (moncount*frame::interpreter_frame_monitor_size()) +
1609 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1610
1611 if (interpreter_frame != NULL) {
1612 #ifdef ASSERT
1613 if (!EnableInvokeDynamic)
1614 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1615 // Probably, since deoptimization doesn't work yet.
1616 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1617 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1618 #endif
1619
1620 interpreter_frame->interpreter_frame_set_method(method);
1621 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1622 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1623 // and sender_sp is fp+8
1624 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1625
1626 #ifdef ASSERT
1627 if (caller->is_interpreted_frame()) {
1628 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1629 }
1630 #endif
1631
1632 interpreter_frame->interpreter_frame_set_locals(locals);
1633 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1634 BasicObjectLock* monbot = montop - moncount;
1635 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1636
1637 // Set last_sp
1638 intptr_t* rsp = (intptr_t*) monbot -
1639 tempcount*Interpreter::stackElementWords -
1640 popframe_extra_args;
1641 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1642
1643 // All frames but the initial (oldest) interpreter frame we fill in have a
1644 // value for sender_sp that allows walking the stack but isn't
1645 // truly correct. Correct the value here.
1646
1647 if (extra_locals != 0 &&
1648 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1649 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1650 }
1651 *interpreter_frame->interpreter_frame_cache_addr() =
1652 method->constants()->cache();
1653 }
1654 return size;
1655 }
1656
1657
1658 //------------------------------------------------------------------------------------------------------------------------
1659 // Exceptions
1660
1661 void TemplateInterpreterGenerator::generate_throw_exception() {
1662 // Entry point in previous activation (i.e., if the caller was interpreted)
1663 Interpreter::_rethrow_exception_entry = __ pc();
1664 const Register thread = rcx;
1665
1666 // Restore sp to interpreter_frame_last_sp even though we are going
1667 // to empty the expression stack for the exception processing.
1668 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1669 // rax,: exception
1670 // rdx: return address/pc that threw exception
1671 __ restore_bcp(); // rsi points to call/send
1672 __ restore_locals();
1673
1674 // Entry point for exceptions thrown within interpreter code
1675 Interpreter::_throw_exception_entry = __ pc();
1676 // expression stack is undefined here
1677 // rax,: exception
1678 // rsi: exception bcp
1679 __ verify_oop(rax);
1680
1681 // expression stack must be empty before entering the VM in case of an exception
1682 __ empty_expression_stack();
1683 __ empty_FPU_stack();
1684 // find exception handler address and preserve exception oop
1685 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1686 // rax,: exception handler entry point
1687 // rdx: preserved exception oop
1688 // rsi: bcp for exception handler
1689 __ push_ptr(rdx); // push exception which is now the only value on the stack
1690 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1691
1692 // If the exception is not handled in the current frame the frame is removed and
1693 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1694 //
1695 // Note: At this point the bci is still the bxi for the instruction which caused
1696 // the exception and the expression stack is empty. Thus, for any VM calls
1697 // at this point, GC will find a legal oop map (with empty expression stack).
1698
1699 // In current activation
1700 // tos: exception
1701 // rsi: exception bcp
1702
1703 //
1704 // JVMTI PopFrame support
1705 //
1706
1707 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1708 __ empty_expression_stack();
1709 __ empty_FPU_stack();
1710 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1711 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1712 // popframe handling cycles.
1713 __ get_thread(thread);
1714 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1715 __ orl(rdx, JavaThread::popframe_processing_bit);
1716 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1717
1718 {
1719 // Check to see whether we are returning to a deoptimized frame.
1720 // (The PopFrame call ensures that the caller of the popped frame is
1721 // either interpreted or compiled and deoptimizes it if compiled.)
1722 // In this case, we can't call dispatch_next() after the frame is
1723 // popped, but instead must save the incoming arguments and restore
1724 // them after deoptimization has occurred.
1725 //
1726 // Note that we don't compare the return PC against the
1727 // deoptimization blob's unpack entry because of the presence of
1728 // adapter frames in C2.
1729 Label caller_not_deoptimized;
1730 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1731 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1732 __ testl(rax, rax);
1733 __ jcc(Assembler::notZero, caller_not_deoptimized);
1734
1735 // Compute size of arguments for saving when returning to deoptimized caller
1736 __ get_method(rax);
1737 __ movptr(rax, Address(rax, Method::const_offset()));
1738 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1739 __ shlptr(rax, Interpreter::logStackElementSize);
1740 __ restore_locals();
1741 __ subptr(rdi, rax);
1742 __ addptr(rdi, wordSize);
1743 // Save these arguments
1744 __ get_thread(thread);
1745 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1746
1747 __ remove_activation(vtos, rdx,
1748 /* throw_monitor_exception */ false,
1749 /* install_monitor_exception */ false,
1750 /* notify_jvmdi */ false);
1751
1752 // Inform deoptimization that it is responsible for restoring these arguments
1753 __ get_thread(thread);
1754 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1755
1756 // Continue in deoptimization handler
1757 __ jmp(rdx);
1758
1759 __ bind(caller_not_deoptimized);
1760 }
1761
1762 __ remove_activation(vtos, rdx,
1763 /* throw_monitor_exception */ false,
1764 /* install_monitor_exception */ false,
1765 /* notify_jvmdi */ false);
1766
1767 // Finish with popframe handling
1768 // A previous I2C followed by a deoptimization might have moved the
1769 // outgoing arguments further up the stack. PopFrame expects the
1770 // mutations to those outgoing arguments to be preserved and other
1771 // constraints basically require this frame to look exactly as
1772 // though it had previously invoked an interpreted activation with
1773 // no space between the top of the expression stack (current
1774 // last_sp) and the top of stack. Rather than force deopt to
1775 // maintain this kind of invariant all the time we call a small
1776 // fixup routine to move the mutated arguments onto the top of our
1777 // expression stack if necessary.
1778 __ mov(rax, rsp);
1779 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1780 __ get_thread(thread);
1781 // PC must point into interpreter here
1782 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1783 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1784 __ get_thread(thread);
1785 __ reset_last_Java_frame(thread, true, true);
1786 // Restore the last_sp and null it out
1787 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1788 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1789
1790 __ restore_bcp();
1791 __ restore_locals();
1792 // The method data pointer was incremented already during
1793 // call profiling. We have to restore the mdp for the current bcp.
1794 if (ProfileInterpreter) {
1795 __ set_method_data_pointer_for_bcp();
1796 }
1797
1798 // Clear the popframe condition flag
1799 __ get_thread(thread);
1800 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1801
1802 __ dispatch_next(vtos);
1803 // end of PopFrame support
1804
1805 Interpreter::_remove_activation_entry = __ pc();
1806
1807 // preserve exception over this code sequence
1808 __ pop_ptr(rax);
1809 __ get_thread(thread);
1810 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1811 // remove the activation (without doing throws on illegalMonitorExceptions)
1812 __ remove_activation(vtos, rdx, false, true, false);
1813 // restore exception
1814 __ get_thread(thread);
1815 __ get_vm_result(rax, thread);
1816
1817 // Inbetween activations - previous activation type unknown yet
1818 // compute continuation point - the continuation point expects
1819 // the following registers set up:
1820 //
1821 // rax: exception
1822 // rdx: return address/pc that threw exception
1823 // rsp: expression stack of caller
1824 // rbp: rbp, of caller
1825 __ push(rax); // save exception
1826 __ push(rdx); // save return address
1827 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1828 __ mov(rbx, rax); // save exception handler
1829 __ pop(rdx); // restore return address
1830 __ pop(rax); // restore exception
1831 // Note that an "issuing PC" is actually the next PC after the call
1832 __ jmp(rbx); // jump to exception handler of caller
1833 }
1834
1835
1836 //
1837 // JVMTI ForceEarlyReturn support
1838 //
1839 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1840 address entry = __ pc();
1841 const Register thread = rcx;
1842
1843 __ restore_bcp();
1844 __ restore_locals();
1845 __ empty_expression_stack();
1846 __ empty_FPU_stack();
1847 __ load_earlyret_value(state);
1848
1849 __ get_thread(thread);
1850 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1851 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1852
1853 // Clear the earlyret state
1854 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1855
1856 __ remove_activation(state, rsi,
1857 false, /* throw_monitor_exception */
1858 false, /* install_monitor_exception */
1859 true); /* notify_jvmdi */
1860 __ jmp(rsi);
1861 return entry;
1862 } // end of ForceEarlyReturn support
1863
1864
1865 //------------------------------------------------------------------------------------------------------------------------
1866 // Helper for vtos entry point generation
1867
1868 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) {
1869 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1870 Label L;
1871 fep = __ pc(); __ push(ftos); __ jmp(L);
1872 dep = __ pc(); __ push(dtos); __ jmp(L);
1873 lep = __ pc(); __ push(ltos); __ jmp(L);
1874 aep = __ pc(); __ push(atos); __ jmp(L);
1875 bep = cep = sep = // fall through
1876 iep = __ pc(); __ push(itos); // fall through
1877 vep = __ pc(); __ bind(L); // fall through
1878 generate_and_dispatch(t);
1879 }
1880
1881 //------------------------------------------------------------------------------------------------------------------------
1882 // Generation of individual instructions
1883
1884 // helpers for generate_and_dispatch
1885
1886
1887
1888 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1889 : TemplateInterpreterGenerator(code) {
1890 generate_all(); // down here so it can be "virtual"
1891 }
1892
1893 //------------------------------------------------------------------------------------------------------------------------
1894
1895 // Non-product code
1896 #ifndef PRODUCT
1897 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1898 address entry = __ pc();
1899
1900 // prepare expression stack
1901 __ pop(rcx); // pop return address so expression stack is 'pure'
1902 __ push(state); // save tosca
1903
1904 // pass tosca registers as arguments & call tracer
1905 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1906 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1907 __ pop(state); // restore tosca
1908
1909 // return
1910 __ jmp(rcx);
1911
1912 return entry;
1913 }
1914
1915
1916 void TemplateInterpreterGenerator::count_bytecode() {
1917 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1918 }
1919
1920
1921 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1922 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1923 }
1924
1925
1926 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1927 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1928 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1929 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1930 ExternalAddress table((address) BytecodePairHistogram::_counters);
1931 Address index(noreg, rbx, Address::times_4);
1932 __ incrementl(ArrayAddress(table, index));
1933 }
1934
1935
1936 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1937 // Call a little run-time stub to avoid blow-up for each bytecode.
1938 // The run-time runtime saves the right registers, depending on
1939 // the tosca in-state for the given template.
1940 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1941 "entry must have been generated");
1942 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1943 }
1944
1945
1946 void TemplateInterpreterGenerator::stop_interpreter_at() {
1947 Label L;
1948 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1949 StopInterpreterAt);
1950 __ jcc(Assembler::notEqual, L);
1951 __ int3();
1952 __ bind(L);
1953 }
1954 #endif // !PRODUCT
1955 #endif // CC_INTERP