1 /*
2 * Copyright (c) 1997, 2011, 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/assembler.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/methodDataOop.hpp"
34 #include "oops/methodOop.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
48 #define __ _masm->
49
50
51 #ifndef CC_INTERP
52 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
53 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
54 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
55
56 //------------------------------------------------------------------------------------------------------------------------
57
58 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
59 address entry = __ pc();
60
61 // Note: There should be a minimal interpreter frame set up when stack
62 // overflow occurs since we check explicitly for it now.
63 //
64 #ifdef ASSERT
65 { Label L;
66 __ lea(rax, Address(rbp,
67 frame::interpreter_frame_monitor_block_top_offset * wordSize));
68 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
69 // (stack grows negative)
70 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
71 __ stop ("interpreter frame not set up");
72 __ bind(L);
73 }
74 #endif // ASSERT
75 // Restore bcp under the assumption that the current frame is still
76 // interpreted
77 __ restore_bcp();
78
79 // expression stack must be empty before entering the VM if an exception
80 // happened
81 __ empty_expression_stack();
82 __ empty_FPU_stack();
83 // throw exception
84 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
85 return entry;
86 }
87
88 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
89 address entry = __ pc();
90 // expression stack must be empty before entering the VM if an exception happened
91 __ empty_expression_stack();
92 __ empty_FPU_stack();
93 // setup parameters
94 // ??? convention: expect aberrant index in register rbx,
95 __ lea(rax, ExternalAddress((address)name));
96 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
97 return entry;
98 }
99
100 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
101 address entry = __ pc();
102 // object is at TOS
103 __ pop(rax);
104 // expression stack must be empty before entering the VM if an exception
105 // happened
106 __ empty_expression_stack();
107 __ empty_FPU_stack();
108 __ call_VM(noreg,
109 CAST_FROM_FN_PTR(address,
110 InterpreterRuntime::throw_ClassCastException),
111 rax);
112 return entry;
113 }
114
115 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
116 assert(!pass_oop || message == NULL, "either oop or message but not both");
117 address entry = __ pc();
118 if (pass_oop) {
119 // object is at TOS
120 __ pop(rbx);
121 }
122 // expression stack must be empty before entering the VM if an exception happened
123 __ empty_expression_stack();
124 __ empty_FPU_stack();
125 // setup parameters
126 __ lea(rax, ExternalAddress((address)name));
127 if (pass_oop) {
128 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
129 } else {
130 if (message != NULL) {
131 __ lea(rbx, ExternalAddress((address)message));
132 } else {
133 __ movptr(rbx, NULL_WORD);
134 }
135 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
136 }
137 // throw exception
138 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
139 return entry;
140 }
141
142
143 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
144 address entry = __ pc();
145 // NULL last_sp until next java call
146 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
147 __ dispatch_next(state);
148 return entry;
149 }
150
151
152 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
153 TosState incoming_state = state;
154 address entry = __ pc();
155
156 #ifdef COMPILER2
157 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
158 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
159 for (int i = 1; i < 8; i++) {
160 __ ffree(i);
161 }
162 } else if (UseSSE < 2) {
163 __ empty_FPU_stack();
164 }
165 #endif
166 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
167 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
168 } else {
169 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
170 }
171
172 // In SSE mode, interpreter returns FP results in xmm0 but they need
173 // to end up back on the FPU so it can operate on them.
174 if (incoming_state == ftos && UseSSE >= 1) {
175 __ subptr(rsp, wordSize);
176 __ movflt(Address(rsp, 0), xmm0);
177 __ fld_s(Address(rsp, 0));
178 __ addptr(rsp, wordSize);
179 } else if (incoming_state == dtos && UseSSE >= 2) {
180 __ subptr(rsp, 2*wordSize);
181 __ movdbl(Address(rsp, 0), xmm0);
182 __ fld_d(Address(rsp, 0));
183 __ addptr(rsp, 2*wordSize);
184 }
185
186 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
187
188 // Restore stack bottom in case i2c adjusted stack
189 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
190 // and NULL it as marker that rsp is now tos until next java call
191 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
192
193 __ restore_bcp();
194 __ restore_locals();
195
196 Label L_got_cache, L_giant_index;
197 if (EnableInvokeDynamic) {
198 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
199 __ jcc(Assembler::equal, L_giant_index);
200 }
201 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
202 __ bind(L_got_cache);
203 __ movl(rbx, Address(rbx, rcx,
204 Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
205 ConstantPoolCacheEntry::flags_offset()));
206 __ andptr(rbx, 0xFF);
207 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
208 __ dispatch_next(state, step);
209
210 // out of the main line of code...
211 if (EnableInvokeDynamic) {
212 __ bind(L_giant_index);
213 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
214 __ jmp(L_got_cache);
215 }
216
217 return entry;
218 }
219
220
221 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
222 address entry = __ pc();
223
224 // In SSE mode, FP results are in xmm0
225 if (state == ftos && UseSSE > 0) {
226 __ subptr(rsp, wordSize);
227 __ movflt(Address(rsp, 0), xmm0);
228 __ fld_s(Address(rsp, 0));
229 __ addptr(rsp, wordSize);
230 } else if (state == dtos && UseSSE >= 2) {
231 __ subptr(rsp, 2*wordSize);
232 __ movdbl(Address(rsp, 0), xmm0);
233 __ fld_d(Address(rsp, 0));
234 __ addptr(rsp, 2*wordSize);
235 }
236
237 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
238
239 // The stack is not extended by deopt but we must NULL last_sp as this
240 // entry is like a "return".
241 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
242 __ restore_bcp();
243 __ restore_locals();
244 // handle exceptions
245 { Label L;
246 const Register thread = rcx;
247 __ get_thread(thread);
248 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
249 __ jcc(Assembler::zero, L);
250 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
251 __ should_not_reach_here();
252 __ bind(L);
253 }
254 __ dispatch_next(state, step);
255 return entry;
256 }
257
258
259 int AbstractInterpreter::BasicType_as_index(BasicType type) {
260 int i = 0;
261 switch (type) {
262 case T_BOOLEAN: i = 0; break;
263 case T_CHAR : i = 1; break;
264 case T_BYTE : i = 2; break;
265 case T_SHORT : i = 3; break;
266 case T_INT : // fall through
267 case T_LONG : // fall through
268 case T_VOID : i = 4; break;
269 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
270 case T_DOUBLE : i = 6; break;
271 case T_OBJECT : // fall through
272 case T_ARRAY : i = 7; break;
273 default : ShouldNotReachHere();
274 }
275 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
276 return i;
277 }
278
279
280 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
281 address entry = __ pc();
282 switch (type) {
283 case T_BOOLEAN: __ c2bool(rax); break;
284 case T_CHAR : __ andptr(rax, 0xFFFF); break;
285 case T_BYTE : __ sign_extend_byte (rax); break;
286 case T_SHORT : __ sign_extend_short(rax); break;
287 case T_INT : /* nothing to do */ break;
288 case T_DOUBLE :
289 case T_FLOAT :
290 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
291 __ pop(t); // remove return address first
292 // Must return a result for interpreter or compiler. In SSE
293 // mode, results are returned in xmm0 and the FPU stack must
294 // be empty.
295 if (type == T_FLOAT && UseSSE >= 1) {
296 // Load ST0
297 __ fld_d(Address(rsp, 0));
298 // Store as float and empty fpu stack
299 __ fstp_s(Address(rsp, 0));
300 // and reload
301 __ movflt(xmm0, Address(rsp, 0));
302 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
303 __ movdbl(xmm0, Address(rsp, 0));
304 } else {
305 // restore ST0
306 __ fld_d(Address(rsp, 0));
307 }
308 // and pop the temp
309 __ addptr(rsp, 2 * wordSize);
310 __ push(t); // restore return address
311 }
312 break;
313 case T_OBJECT :
314 // retrieve result from frame
315 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
316 // and verify it
317 __ verify_oop(rax);
318 break;
319 default : ShouldNotReachHere();
320 }
321 __ ret(0); // return from result handler
322 return entry;
323 }
324
325 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
326 address entry = __ pc();
327 __ push(state);
328 __ call_VM(noreg, runtime_entry);
329 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
330 return entry;
331 }
332
333
334 // Helpers for commoning out cases in the various type of method entries.
335 //
336
337 // increment invocation count & check for overflow
338 //
339 // Note: checking for negative value instead of overflow
340 // so we have a 'sticky' overflow test
341 //
342 // rbx,: method
343 // rcx: invocation counter
344 //
345 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
346 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
347 in_bytes(InvocationCounter::counter_offset()));
348 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
349 if (TieredCompilation) {
350 int increment = InvocationCounter::count_increment;
351 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
352 Label no_mdo, done;
353 if (ProfileInterpreter) {
354 // Are we profiling?
355 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
356 __ testptr(rax, rax);
357 __ jccb(Assembler::zero, no_mdo);
358 // Increment counter in the MDO
359 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
360 in_bytes(InvocationCounter::counter_offset()));
361 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
362 __ jmpb(done);
363 }
364 __ bind(no_mdo);
365 // Increment counter in methodOop (we don't need to load it, it's in rcx).
366 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
367 __ bind(done);
368 } else {
369 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() +
370 InvocationCounter::counter_offset());
371
372 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
373 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
374 }
375 // Update standard invocation counters
376 __ movl(rax, backedge_counter); // load backedge counter
377
378 __ incrementl(rcx, InvocationCounter::count_increment);
379 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
380
381 __ movl(invocation_counter, rcx); // save invocation count
382 __ addl(rcx, rax); // add both counters
383
384 // profile_method is non-null only for interpreted method so
385 // profile_method != NULL == !native_call
386 // BytecodeInterpreter only calls for native so code is elided.
387
388 if (ProfileInterpreter && profile_method != NULL) {
389 // Test to see if we should create a method data oop
390 __ cmp32(rcx,
391 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
392 __ jcc(Assembler::less, *profile_method_continue);
393
394 // if no method data exists, go to profile_method
395 __ test_method_data_pointer(rax, *profile_method);
396 }
397
398 __ cmp32(rcx,
399 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
400 __ jcc(Assembler::aboveEqual, *overflow);
401 }
402 }
403
404 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
405
406 // Asm interpreter on entry
407 // rdi - locals
408 // rsi - bcp
409 // rbx, - method
410 // rdx - cpool
411 // rbp, - interpreter frame
412
413 // C++ interpreter on entry
414 // rsi - new interpreter state pointer
415 // rbp - interpreter frame pointer
416 // rbx - method
417
418 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
419 // rbx, - method
420 // rcx - rcvr (assuming there is one)
421 // top of stack return address of interpreter caller
422 // rsp - sender_sp
423
424 // C++ interpreter only
425 // rsi - previous interpreter state pointer
426
427 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
428
429 // InterpreterRuntime::frequency_counter_overflow takes one argument
430 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
431 // The call returns the address of the verified entry point for the method or NULL
432 // if the compilation did not complete (either went background or bailed out).
433 __ movptr(rax, (intptr_t)false);
434 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
435
436 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
437
438 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
439 // and jump to the interpreted entry.
440 __ jmp(*do_continue, relocInfo::none);
441
442 }
443
444 void InterpreterGenerator::generate_stack_overflow_check(void) {
445 // see if we've got enough room on the stack for locals plus overhead.
446 // the expression stack grows down incrementally, so the normal guard
447 // page mechanism will work for that.
448 //
449 // Registers live on entry:
450 //
451 // Asm interpreter
452 // rdx: number of additional locals this frame needs (what we must check)
453 // rbx,: methodOop
454
455 // destroyed on exit
456 // rax,
457
458 // NOTE: since the additional locals are also always pushed (wasn't obvious in
459 // generate_method_entry) so the guard should work for them too.
460 //
461
462 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
463 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
464
465 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
466 // be sure to change this if you add/subtract anything to/from the overhead area
467 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
468
469 const int page_size = os::vm_page_size();
470
471 Label after_frame_check;
472
473 // see if the frame is greater than one page in size. If so,
474 // then we need to verify there is enough stack space remaining
475 // for the additional locals.
476 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
477 __ jcc(Assembler::belowEqual, after_frame_check);
478
479 // compute rsp as if this were going to be the last frame on
480 // the stack before the red zone
481
482 Label after_frame_check_pop;
483
484 __ push(rsi);
485
486 const Register thread = rsi;
487
488 __ get_thread(thread);
489
490 const Address stack_base(thread, Thread::stack_base_offset());
491 const Address stack_size(thread, Thread::stack_size_offset());
492
493 // locals + overhead, in bytes
494 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
495
496 #ifdef ASSERT
497 Label stack_base_okay, stack_size_okay;
498 // verify that thread stack base is non-zero
499 __ cmpptr(stack_base, (int32_t)NULL_WORD);
500 __ jcc(Assembler::notEqual, stack_base_okay);
501 __ stop("stack base is zero");
502 __ bind(stack_base_okay);
503 // verify that thread stack size is non-zero
504 __ cmpptr(stack_size, 0);
505 __ jcc(Assembler::notEqual, stack_size_okay);
506 __ stop("stack size is zero");
507 __ bind(stack_size_okay);
508 #endif
509
510 // Add stack base to locals and subtract stack size
511 __ addptr(rax, stack_base);
512 __ subptr(rax, stack_size);
513
514 // Use the maximum number of pages we might bang.
515 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
516 (StackRedPages+StackYellowPages);
517 __ addptr(rax, max_pages * page_size);
518
519 // check against the current stack bottom
520 __ cmpptr(rsp, rax);
521 __ jcc(Assembler::above, after_frame_check_pop);
522
523 __ pop(rsi); // get saved bcp / (c++ prev state ).
524
525 __ pop(rax); // get return address
526 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
527
528 // all done with frame size check
529 __ bind(after_frame_check_pop);
530 __ pop(rsi);
531
532 __ bind(after_frame_check);
533 }
534
535 // Allocate monitor and lock method (asm interpreter)
536 // rbx, - methodOop
537 //
538 void InterpreterGenerator::lock_method(void) {
539 // synchronize method
540 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
541 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
542 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
543
544 #ifdef ASSERT
545 { Label L;
546 __ movl(rax, access_flags);
547 __ testl(rax, JVM_ACC_SYNCHRONIZED);
548 __ jcc(Assembler::notZero, L);
549 __ stop("method doesn't need synchronization");
550 __ bind(L);
551 }
552 #endif // ASSERT
553 // get synchronization object
554 { Label done;
555 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
556 __ movl(rax, access_flags);
557 __ testl(rax, JVM_ACC_STATIC);
558 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
559 __ jcc(Assembler::zero, done);
560 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
561 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
562 __ movptr(rax, Address(rax, mirror_offset));
563 __ bind(done);
564 }
565 // add space for monitor & lock
566 __ subptr(rsp, entry_size); // add space for a monitor entry
567 __ movptr(monitor_block_top, rsp); // set new monitor block top
568 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
569 __ mov(rdx, rsp); // object address
570 __ lock_object(rdx);
571 }
572
573 //
574 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
575 // and for native methods hence the shared code.
576
577 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
578 // initialize fixed part of activation frame
579 __ push(rax); // save return address
580 __ enter(); // save old & set new rbp,
581
582
583 __ push(rsi); // set sender sp
584 __ push((int32_t)NULL_WORD); // leave last_sp as null
585 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
586 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
587 __ push(rbx); // save methodOop
588 if (ProfileInterpreter) {
589 Label method_data_continue;
590 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
591 __ testptr(rdx, rdx);
592 __ jcc(Assembler::zero, method_data_continue);
593 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
594 __ bind(method_data_continue);
595 __ push(rdx); // set the mdp (method data pointer)
596 } else {
597 __ push(0);
598 }
599
600 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
601 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
602 __ push(rdx); // set constant pool cache
603 __ push(rdi); // set locals pointer
604 if (native_call) {
605 __ push(0); // no bcp
606 } else {
607 __ push(rsi); // set bcp
608 }
609 __ push(0); // reserve word for pointer to expression stack bottom
610 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
611 }
612
613 // End of helpers
614
615 //
616 // Various method entries
617 //------------------------------------------------------------------------------------------------------------------------
618 //
619 //
620
621 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
622
623 address InterpreterGenerator::generate_accessor_entry(void) {
624
625 // rbx,: methodOop
626 // rcx: receiver (preserve for slow entry into asm interpreter)
627
628 // rsi: senderSP must preserved for slow path, set SP to it on fast path
629
630 address entry_point = __ pc();
631 Label xreturn_path;
632
633 // do fastpath for resolved accessor methods
634 if (UseFastAccessorMethods) {
635 Label slow_path;
636 // If we need a safepoint check, generate full interpreter entry.
637 ExternalAddress state(SafepointSynchronize::address_of_state());
638 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
639 SafepointSynchronize::_not_synchronized);
640
641 __ jcc(Assembler::notEqual, slow_path);
642 // ASM/C++ Interpreter
643 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
644 // Note: We can only use this code if the getfield has been resolved
645 // and if we don't have a null-pointer exception => check for
646 // these conditions first and use slow path if necessary.
647 // rbx,: method
648 // rcx: receiver
649 __ movptr(rax, Address(rsp, wordSize));
650
651 // check if local 0 != NULL and read field
652 __ testptr(rax, rax);
653 __ jcc(Assembler::zero, slow_path);
654
655 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
656 // read first instruction word and extract bytecode @ 1 and index @ 2
657 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
658 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
659 // Shift codes right to get the index on the right.
660 // The bytecode fetched looks like <index><0xb4><0x2a>
661 __ shrl(rdx, 2*BitsPerByte);
662 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
663 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
664
665 // rax,: local 0
666 // rbx,: method
667 // rcx: receiver - do not destroy since it is needed for slow path!
668 // rcx: scratch
669 // rdx: constant pool cache index
670 // rdi: constant pool cache
671 // rsi: sender sp
672
673 // check if getfield has been resolved and read constant pool cache entry
674 // check the validity of the cache entry by testing whether _indices field
675 // contains Bytecode::_getfield in b1 byte.
676 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
677 __ movl(rcx,
678 Address(rdi,
679 rdx,
680 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
681 __ shrl(rcx, 2*BitsPerByte);
682 __ andl(rcx, 0xFF);
683 __ cmpl(rcx, Bytecodes::_getfield);
684 __ jcc(Assembler::notEqual, slow_path);
685
686 // Note: constant pool entry is not valid before bytecode is resolved
687 __ movptr(rcx,
688 Address(rdi,
689 rdx,
690 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
691 __ movl(rdx,
692 Address(rdi,
693 rdx,
694 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
695
696 Label notByte, notShort, notChar;
697 const Address field_address (rax, rcx, Address::times_1);
698
699 // Need to differentiate between igetfield, agetfield, bgetfield etc.
700 // because they are different sizes.
701 // Use the type from the constant pool cache
702 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
703 // Make sure we don't need to mask rdx for tosBits after the above shift
704 ConstantPoolCacheEntry::verify_tosBits();
705 __ cmpl(rdx, btos);
706 __ jcc(Assembler::notEqual, notByte);
707 __ load_signed_byte(rax, field_address);
708 __ jmp(xreturn_path);
709
710 __ bind(notByte);
711 __ cmpl(rdx, stos);
712 __ jcc(Assembler::notEqual, notShort);
713 __ load_signed_short(rax, field_address);
714 __ jmp(xreturn_path);
715
716 __ bind(notShort);
717 __ cmpl(rdx, ctos);
718 __ jcc(Assembler::notEqual, notChar);
719 __ load_unsigned_short(rax, field_address);
720 __ jmp(xreturn_path);
721
722 __ bind(notChar);
723 #ifdef ASSERT
724 Label okay;
725 __ cmpl(rdx, atos);
726 __ jcc(Assembler::equal, okay);
727 __ cmpl(rdx, itos);
728 __ jcc(Assembler::equal, okay);
729 __ stop("what type is this?");
730 __ bind(okay);
731 #endif // ASSERT
732 // All the rest are a 32 bit wordsize
733 // This is ok for now. Since fast accessors should be going away
734 __ movptr(rax, field_address);
735
736 __ bind(xreturn_path);
737
738 // _ireturn/_areturn
739 __ pop(rdi); // get return address
740 __ mov(rsp, rsi); // set sp to sender sp
741 __ jmp(rdi);
742
743 // generate a vanilla interpreter entry as the slow path
744 __ bind(slow_path);
745
746 (void) generate_normal_entry(false);
747 return entry_point;
748 }
749 return NULL;
750
751 }
752
753 // Method entry for java.lang.ref.Reference.get.
754 address InterpreterGenerator::generate_Reference_get_entry(void) {
755 #ifndef SERIALGC
756 // Code: _aload_0, _getfield, _areturn
757 // parameter size = 1
758 //
759 // The code that gets generated by this routine is split into 2 parts:
760 // 1. The "intrinsified" code for G1 (or any SATB based GC),
761 // 2. The slow path - which is an expansion of the regular method entry.
762 //
763 // Notes:-
764 // * In the G1 code we do not check whether we need to block for
765 // a safepoint. If G1 is enabled then we must execute the specialized
766 // code for Reference.get (except when the Reference object is null)
767 // so that we can log the value in the referent field with an SATB
768 // update buffer.
769 // If the code for the getfield template is modified so that the
770 // G1 pre-barrier code is executed when the current method is
771 // Reference.get() then going through the normal method entry
772 // will be fine.
773 // * The G1 code below can, however, check the receiver object (the instance
774 // of java.lang.Reference) and jump to the slow path if null. If the
775 // Reference object is null then we obviously cannot fetch the referent
776 // and so we don't need to call the G1 pre-barrier. Thus we can use the
777 // regular method entry code to generate the NPE.
778 //
779 // This code is based on generate_accessor_enty.
780
781 // rbx,: methodOop
782 // rcx: receiver (preserve for slow entry into asm interpreter)
783
784 // rsi: senderSP must preserved for slow path, set SP to it on fast path
785
786 address entry = __ pc();
787
788 const int referent_offset = java_lang_ref_Reference::referent_offset;
789 guarantee(referent_offset > 0, "referent offset not initialized");
790
791 if (UseG1GC) {
792 Label slow_path;
793
794 // Check if local 0 != NULL
795 // If the receiver is null then it is OK to jump to the slow path.
796 __ movptr(rax, Address(rsp, wordSize));
797 __ testptr(rax, rax);
798 __ jcc(Assembler::zero, slow_path);
799
800 // rax: local 0 (must be preserved across the G1 barrier call)
801 //
802 // rbx: method (at this point it's scratch)
803 // rcx: receiver (at this point it's scratch)
804 // rdx: scratch
805 // rdi: scratch
806 //
807 // rsi: sender sp
808
809 // Preserve the sender sp in case the pre-barrier
810 // calls the runtime
811 __ push(rsi);
812
813 // Load the value of the referent field.
814 const Address field_address(rax, referent_offset);
815 __ movptr(rax, field_address);
816
817 // Generate the G1 pre-barrier code to log the value of
818 // the referent field in an SATB buffer.
819 __ get_thread(rcx);
820 __ g1_write_barrier_pre(noreg /* obj */,
821 rax /* pre_val */,
822 rcx /* thread */,
823 rbx /* tmp */,
824 true /* tosca_save */,
825 true /* expand_call */);
826
827 // _areturn
828 __ pop(rsi); // get sender sp
829 __ pop(rdi); // get return address
830 __ mov(rsp, rsi); // set sp to sender sp
831 __ jmp(rdi);
832
833 __ bind(slow_path);
834 (void) generate_normal_entry(false);
835
836 return entry;
837 }
838 #endif // SERIALGC
839
840 // If G1 is not enabled then attempt to go through the accessor entry point
841 // Reference.get is an accessor
842 return generate_accessor_entry();
843 }
844
845 //
846 // Interpreter stub for calling a native method. (asm interpreter)
847 // This sets up a somewhat different looking stack for calling the native method
848 // than the typical interpreter frame setup.
849 //
850
851 address InterpreterGenerator::generate_native_entry(bool synchronized) {
852 // determine code generation flags
853 bool inc_counter = UseCompiler || CountCompiledCalls;
854
855 // rbx,: methodOop
856 // rsi: sender sp
857 // rsi: previous interpreter state (C++ interpreter) must preserve
858 address entry_point = __ pc();
859
860
861 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
862 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
863 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
864
865 // get parameter size (always needed)
866 __ load_unsigned_short(rcx, size_of_parameters);
867
868 // native calls don't need the stack size check since they have no expression stack
869 // and the arguments are already on the stack and we only add a handful of words
870 // to the stack
871
872 // rbx,: methodOop
873 // rcx: size of parameters
874 // rsi: sender sp
875
876 __ pop(rax); // get return address
877 // for natives the size of locals is zero
878
879 // compute beginning of parameters (rdi)
880 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
881
882
883 // add 2 zero-initialized slots for native calls
884 // NULL result handler
885 __ push((int32_t)NULL_WORD);
886 // NULL oop temp (mirror or jni oop result)
887 __ push((int32_t)NULL_WORD);
888
889 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
890 // initialize fixed part of activation frame
891
892 generate_fixed_frame(true);
893
894 // make sure method is native & not abstract
895 #ifdef ASSERT
896 __ movl(rax, access_flags);
897 {
898 Label L;
899 __ testl(rax, JVM_ACC_NATIVE);
900 __ jcc(Assembler::notZero, L);
901 __ stop("tried to execute non-native method as native");
902 __ bind(L);
903 }
904 { Label L;
905 __ testl(rax, JVM_ACC_ABSTRACT);
906 __ jcc(Assembler::zero, L);
907 __ stop("tried to execute abstract method in interpreter");
908 __ bind(L);
909 }
910 #endif
911
912 // Since at this point in the method invocation the exception handler
913 // would try to exit the monitor of synchronized methods which hasn't
914 // been entered yet, we set the thread local variable
915 // _do_not_unlock_if_synchronized to true. The remove_activation will
916 // check this flag.
917
918 __ get_thread(rax);
919 const Address do_not_unlock_if_synchronized(rax,
920 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
921 __ movbool(do_not_unlock_if_synchronized, true);
922
923 // increment invocation count & check for overflow
924 Label invocation_counter_overflow;
925 if (inc_counter) {
926 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
927 }
928
929 Label continue_after_compile;
930 __ bind(continue_after_compile);
931
932 bang_stack_shadow_pages(true);
933
934 // reset the _do_not_unlock_if_synchronized flag
935 __ get_thread(rax);
936 __ movbool(do_not_unlock_if_synchronized, false);
937
938 // check for synchronized methods
939 // Must happen AFTER invocation_counter check and stack overflow check,
940 // so method is not locked if overflows.
941 //
942 if (synchronized) {
943 lock_method();
944 } else {
945 // no synchronization necessary
946 #ifdef ASSERT
947 { Label L;
948 __ movl(rax, access_flags);
949 __ testl(rax, JVM_ACC_SYNCHRONIZED);
950 __ jcc(Assembler::zero, L);
951 __ stop("method needs synchronization");
952 __ bind(L);
953 }
954 #endif
955 }
956
957 // start execution
958 #ifdef ASSERT
959 { Label L;
960 const Address monitor_block_top (rbp,
961 frame::interpreter_frame_monitor_block_top_offset * wordSize);
962 __ movptr(rax, monitor_block_top);
963 __ cmpptr(rax, rsp);
964 __ jcc(Assembler::equal, L);
965 __ stop("broken stack frame setup in interpreter");
966 __ bind(L);
967 }
968 #endif
969
970 // jvmti/dtrace support
971 __ notify_method_entry();
972
973 // work registers
974 const Register method = rbx;
975 const Register thread = rdi;
976 const Register t = rcx;
977
978 // allocate space for parameters
979 __ get_method(method);
980 __ verify_oop(method);
981 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
982 __ shlptr(t, Interpreter::logStackElementSize);
983 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
984 __ subptr(rsp, t);
985 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
986
987 // get signature handler
988 { Label L;
989 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
990 __ testptr(t, t);
991 __ jcc(Assembler::notZero, L);
992 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
993 __ get_method(method);
994 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
995 __ bind(L);
996 }
997
998 // call signature handler
999 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1000 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1001 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1002 // The generated handlers do not touch RBX (the method oop).
1003 // However, large signatures cannot be cached and are generated
1004 // each time here. The slow-path generator will blow RBX
1005 // sometime, so we must reload it after the call.
1006 __ call(t);
1007 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1008
1009 // result handler is in rax,
1010 // set result handler
1011 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1012
1013 // pass mirror handle if static call
1014 { Label L;
1015 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1016 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1017 __ testl(t, JVM_ACC_STATIC);
1018 __ jcc(Assembler::zero, L);
1019 // get mirror
1020 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
1021 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
1022 __ movptr(t, Address(t, mirror_offset));
1023 // copy mirror into activation frame
1024 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1025 // pass handle to mirror
1026 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1027 __ movptr(Address(rsp, wordSize), t);
1028 __ bind(L);
1029 }
1030
1031 // get native function entry point
1032 { Label L;
1033 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1034 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1035 __ cmpptr(rax, unsatisfied.addr());
1036 __ jcc(Assembler::notEqual, L);
1037 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1038 __ get_method(method);
1039 __ verify_oop(method);
1040 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1041 __ bind(L);
1042 }
1043
1044 // pass JNIEnv
1045 __ get_thread(thread);
1046 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1047 __ movptr(Address(rsp, 0), t);
1048
1049 // set_last_Java_frame_before_call
1050 // It is enough that the pc()
1051 // points into the right code segment. It does not have to be the correct return pc.
1052 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1053
1054 // change thread state
1055 #ifdef ASSERT
1056 { Label L;
1057 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1058 __ cmpl(t, _thread_in_Java);
1059 __ jcc(Assembler::equal, L);
1060 __ stop("Wrong thread state in native stub");
1061 __ bind(L);
1062 }
1063 #endif
1064
1065 // Change state to native
1066 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1067 __ call(rax);
1068
1069 // result potentially in rdx:rax or ST0
1070
1071 // Either restore the MXCSR register after returning from the JNI Call
1072 // or verify that it wasn't changed.
1073 if (VM_Version::supports_sse()) {
1074 if (RestoreMXCSROnJNICalls) {
1075 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1076 }
1077 else if (CheckJNICalls ) {
1078 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1079 }
1080 }
1081
1082 // Either restore the x87 floating pointer control word after returning
1083 // from the JNI call or verify that it wasn't changed.
1084 if (CheckJNICalls) {
1085 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1086 }
1087
1088 // save potential result in ST(0) & rdx:rax
1089 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1090 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1091 // It is safe to do this push because state is _thread_in_native and return address will be found
1092 // via _last_native_pc and not via _last_jave_sp
1093
1094 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1095 // If the order changes or anything else is added to the stack the code in
1096 // interpreter_frame_result will have to be changed.
1097
1098 { Label L;
1099 Label push_double;
1100 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1101 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1102 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1103 float_handler.addr());
1104 __ jcc(Assembler::equal, push_double);
1105 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1106 double_handler.addr());
1107 __ jcc(Assembler::notEqual, L);
1108 __ bind(push_double);
1109 __ push(dtos);
1110 __ bind(L);
1111 }
1112 __ push(ltos);
1113
1114 // change thread state
1115 __ get_thread(thread);
1116 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1117 if(os::is_MP()) {
1118 if (UseMembar) {
1119 // Force this write out before the read below
1120 __ membar(Assembler::Membar_mask_bits(
1121 Assembler::LoadLoad | Assembler::LoadStore |
1122 Assembler::StoreLoad | Assembler::StoreStore));
1123 } else {
1124 // Write serialization page so VM thread can do a pseudo remote membar.
1125 // We use the current thread pointer to calculate a thread specific
1126 // offset to write to within the page. This minimizes bus traffic
1127 // due to cache line collision.
1128 __ serialize_memory(thread, rcx);
1129 }
1130 }
1131
1132 if (AlwaysRestoreFPU) {
1133 // Make sure the control word is correct.
1134 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1135 }
1136
1137 // check for safepoint operation in progress and/or pending suspend requests
1138 { Label Continue;
1139
1140 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1141 SafepointSynchronize::_not_synchronized);
1142
1143 Label L;
1144 __ jcc(Assembler::notEqual, L);
1145 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1146 __ jcc(Assembler::equal, Continue);
1147 __ bind(L);
1148
1149 // Don't use call_VM as it will see a possible pending exception and forward it
1150 // and never return here preventing us from clearing _last_native_pc down below.
1151 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1152 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1153 // by hand.
1154 //
1155 __ push(thread);
1156 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1157 JavaThread::check_special_condition_for_native_trans)));
1158 __ increment(rsp, wordSize);
1159 __ get_thread(thread);
1160
1161 __ bind(Continue);
1162 }
1163
1164 // change thread state
1165 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1166
1167 __ reset_last_Java_frame(thread, true, true);
1168
1169 // reset handle block
1170 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1171 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1172
1173 // If result was an oop then unbox and save it in the frame
1174 { Label L;
1175 Label no_oop, store_result;
1176 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1177 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1178 handler.addr());
1179 __ jcc(Assembler::notEqual, no_oop);
1180 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1181 __ pop(ltos);
1182 __ testptr(rax, rax);
1183 __ jcc(Assembler::zero, store_result);
1184 // unbox
1185 __ movptr(rax, Address(rax, 0));
1186 __ bind(store_result);
1187 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1188 // keep stack depth as expected by pushing oop which will eventually be discarded
1189 __ push(ltos);
1190 __ bind(no_oop);
1191 }
1192
1193 {
1194 Label no_reguard;
1195 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1196 __ jcc(Assembler::notEqual, no_reguard);
1197
1198 __ pusha();
1199 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1200 __ popa();
1201
1202 __ bind(no_reguard);
1203 }
1204
1205 // restore rsi to have legal interpreter frame,
1206 // i.e., bci == 0 <=> rsi == code_base()
1207 // Can't call_VM until bcp is within reasonable.
1208 __ get_method(method); // method is junk from thread_in_native to now.
1209 __ verify_oop(method);
1210 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
1211 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
1212
1213 // handle exceptions (exception handling will handle unlocking!)
1214 { Label L;
1215 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1216 __ jcc(Assembler::zero, L);
1217 // Note: At some point we may want to unify this with the code used in call_VM_base();
1218 // i.e., we should use the StubRoutines::forward_exception code. For now this
1219 // doesn't work here because the rsp is not correctly set at this point.
1220 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1221 __ should_not_reach_here();
1222 __ bind(L);
1223 }
1224
1225 // do unlocking if necessary
1226 { Label L;
1227 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1228 __ testl(t, JVM_ACC_SYNCHRONIZED);
1229 __ jcc(Assembler::zero, L);
1230 // the code below should be shared with interpreter macro assembler implementation
1231 { Label unlock;
1232 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1233 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1234 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1235
1236 __ lea(rdx, monitor); // address of first monitor
1237
1238 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1239 __ testptr(t, t);
1240 __ jcc(Assembler::notZero, unlock);
1241
1242 // Entry already unlocked, need to throw exception
1243 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1244 __ should_not_reach_here();
1245
1246 __ bind(unlock);
1247 __ unlock_object(rdx);
1248 }
1249 __ bind(L);
1250 }
1251
1252 // jvmti/dtrace support
1253 // Note: This must happen _after_ handling/throwing any exceptions since
1254 // the exception handler code notifies the runtime of method exits
1255 // too. If this happens before, method entry/exit notifications are
1256 // not properly paired (was bug - gri 11/22/99).
1257 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1258
1259 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1260 __ pop(ltos);
1261 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1262 __ call(t);
1263
1264 // remove activation
1265 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1266 __ leave(); // remove frame anchor
1267 __ pop(rdi); // get return address
1268 __ mov(rsp, t); // set sp to sender sp
1269 __ jmp(rdi);
1270
1271 if (inc_counter) {
1272 // Handle overflow of counter and compile method
1273 __ bind(invocation_counter_overflow);
1274 generate_counter_overflow(&continue_after_compile);
1275 }
1276
1277 return entry_point;
1278 }
1279
1280 //
1281 // Generic interpreted method entry to (asm) interpreter
1282 //
1283 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1284 // determine code generation flags
1285 bool inc_counter = UseCompiler || CountCompiledCalls;
1286
1287 // rbx,: methodOop
1288 // rsi: sender sp
1289 address entry_point = __ pc();
1290
1291
1292 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1293 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
1294 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1295 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
1296
1297 // get parameter size (always needed)
1298 __ load_unsigned_short(rcx, size_of_parameters);
1299
1300 // rbx,: methodOop
1301 // rcx: size of parameters
1302
1303 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1304
1305 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1306 __ subl(rdx, rcx); // rdx = no. of additional locals
1307
1308 // see if we've got enough room on the stack for locals plus overhead.
1309 generate_stack_overflow_check();
1310
1311 // get return address
1312 __ pop(rax);
1313
1314 // compute beginning of parameters (rdi)
1315 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1316
1317 // rdx - # of additional locals
1318 // allocate space for locals
1319 // explicitly initialize locals
1320 {
1321 Label exit, loop;
1322 __ testl(rdx, rdx);
1323 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1324 __ bind(loop);
1325 __ push((int32_t)NULL_WORD); // initialize local variables
1326 __ decrement(rdx); // until everything initialized
1327 __ jcc(Assembler::greater, loop);
1328 __ bind(exit);
1329 }
1330
1331 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1332 // initialize fixed part of activation frame
1333 generate_fixed_frame(false);
1334
1335 // make sure method is not native & not abstract
1336 #ifdef ASSERT
1337 __ movl(rax, access_flags);
1338 {
1339 Label L;
1340 __ testl(rax, JVM_ACC_NATIVE);
1341 __ jcc(Assembler::zero, L);
1342 __ stop("tried to execute native method as non-native");
1343 __ bind(L);
1344 }
1345 { Label L;
1346 __ testl(rax, JVM_ACC_ABSTRACT);
1347 __ jcc(Assembler::zero, L);
1348 __ stop("tried to execute abstract method in interpreter");
1349 __ bind(L);
1350 }
1351 #endif
1352
1353 // Since at this point in the method invocation the exception handler
1354 // would try to exit the monitor of synchronized methods which hasn't
1355 // been entered yet, we set the thread local variable
1356 // _do_not_unlock_if_synchronized to true. The remove_activation will
1357 // check this flag.
1358
1359 __ get_thread(rax);
1360 const Address do_not_unlock_if_synchronized(rax,
1361 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1362 __ movbool(do_not_unlock_if_synchronized, true);
1363
1364 // increment invocation count & check for overflow
1365 Label invocation_counter_overflow;
1366 Label profile_method;
1367 Label profile_method_continue;
1368 if (inc_counter) {
1369 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1370 if (ProfileInterpreter) {
1371 __ bind(profile_method_continue);
1372 }
1373 }
1374 Label continue_after_compile;
1375 __ bind(continue_after_compile);
1376
1377 bang_stack_shadow_pages(false);
1378
1379 // reset the _do_not_unlock_if_synchronized flag
1380 __ get_thread(rax);
1381 __ movbool(do_not_unlock_if_synchronized, false);
1382
1383 // check for synchronized methods
1384 // Must happen AFTER invocation_counter check and stack overflow check,
1385 // so method is not locked if overflows.
1386 //
1387 if (synchronized) {
1388 // Allocate monitor and lock method
1389 lock_method();
1390 } else {
1391 // no synchronization necessary
1392 #ifdef ASSERT
1393 { Label L;
1394 __ movl(rax, access_flags);
1395 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1396 __ jcc(Assembler::zero, L);
1397 __ stop("method needs synchronization");
1398 __ bind(L);
1399 }
1400 #endif
1401 }
1402
1403 // start execution
1404 #ifdef ASSERT
1405 { Label L;
1406 const Address monitor_block_top (rbp,
1407 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1408 __ movptr(rax, monitor_block_top);
1409 __ cmpptr(rax, rsp);
1410 __ jcc(Assembler::equal, L);
1411 __ stop("broken stack frame setup in interpreter");
1412 __ bind(L);
1413 }
1414 #endif
1415
1416 // jvmti support
1417 __ notify_method_entry();
1418
1419 __ dispatch_next(vtos);
1420
1421 // invocation counter overflow
1422 if (inc_counter) {
1423 if (ProfileInterpreter) {
1424 // We have decided to profile this method in the interpreter
1425 __ bind(profile_method);
1426 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1427 __ set_method_data_pointer_for_bcp();
1428 __ get_method(rbx);
1429 __ jmp(profile_method_continue);
1430 }
1431 // Handle overflow of counter and compile method
1432 __ bind(invocation_counter_overflow);
1433 generate_counter_overflow(&continue_after_compile);
1434 }
1435
1436 return entry_point;
1437 }
1438
1439 //------------------------------------------------------------------------------------------------------------------------
1440 // Entry points
1441 //
1442 // Here we generate the various kind of entries into the interpreter.
1443 // The two main entry type are generic bytecode methods and native call method.
1444 // These both come in synchronized and non-synchronized versions but the
1445 // frame layout they create is very similar. The other method entry
1446 // types are really just special purpose entries that are really entry
1447 // and interpretation all in one. These are for trivial methods like
1448 // accessor, empty, or special math methods.
1449 //
1450 // When control flow reaches any of the entry types for the interpreter
1451 // the following holds ->
1452 //
1453 // Arguments:
1454 //
1455 // rbx,: methodOop
1456 // rcx: receiver
1457 //
1458 //
1459 // Stack layout immediately at entry
1460 //
1461 // [ return address ] <--- rsp
1462 // [ parameter n ]
1463 // ...
1464 // [ parameter 1 ]
1465 // [ expression stack ] (caller's java expression stack)
1466
1467 // Assuming that we don't go to one of the trivial specialized
1468 // entries the stack will look like below when we are ready to execute
1469 // the first bytecode (or call the native routine). The register usage
1470 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1471 //
1472 // local variables follow incoming parameters immediately; i.e.
1473 // the return address is moved to the end of the locals).
1474 //
1475 // [ monitor entry ] <--- rsp
1476 // ...
1477 // [ monitor entry ]
1478 // [ expr. stack bottom ]
1479 // [ saved rsi ]
1480 // [ current rdi ]
1481 // [ methodOop ]
1482 // [ saved rbp, ] <--- rbp,
1483 // [ return address ]
1484 // [ local variable m ]
1485 // ...
1486 // [ local variable 1 ]
1487 // [ parameter n ]
1488 // ...
1489 // [ parameter 1 ] <--- rdi
1490
1491 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1492 // determine code generation flags
1493 bool synchronized = false;
1494 address entry_point = NULL;
1495
1496 switch (kind) {
1497 case Interpreter::zerolocals : break;
1498 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1499 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1500 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1501 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1502 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1503 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1504 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1505
1506 case Interpreter::java_lang_math_sin : // fall thru
1507 case Interpreter::java_lang_math_cos : // fall thru
1508 case Interpreter::java_lang_math_tan : // fall thru
1509 case Interpreter::java_lang_math_abs : // fall thru
1510 case Interpreter::java_lang_math_log : // fall thru
1511 case Interpreter::java_lang_math_log10 : // fall thru
1512 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1513 case Interpreter::java_lang_ref_reference_get
1514 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1515 default : ShouldNotReachHere(); break;
1516 }
1517
1518 if (entry_point) return entry_point;
1519
1520 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1521
1522 }
1523
1524 // These should never be compiled since the interpreter will prefer
1525 // the compiled version to the intrinsic version.
1526 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1527 switch (method_kind(m)) {
1528 case Interpreter::java_lang_math_sin : // fall thru
1529 case Interpreter::java_lang_math_cos : // fall thru
1530 case Interpreter::java_lang_math_tan : // fall thru
1531 case Interpreter::java_lang_math_abs : // fall thru
1532 case Interpreter::java_lang_math_log : // fall thru
1533 case Interpreter::java_lang_math_log10 : // fall thru
1534 case Interpreter::java_lang_math_sqrt :
1535 return false;
1536 default:
1537 return true;
1538 }
1539 }
1540
1541 // How much stack a method activation needs in words.
1542 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1543
1544 const int stub_code = 4; // see generate_call_stub
1545 // Save space for one monitor to get into the interpreted method in case
1546 // the method is synchronized
1547 int monitor_size = method->is_synchronized() ?
1548 1*frame::interpreter_frame_monitor_size() : 0;
1549
1550 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1551 // be sure to change this if you add/subtract anything to/from the overhead area
1552 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1553
1554 const int extra_stack = methodOopDesc::extra_stack_entries();
1555 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1556 Interpreter::stackElementWords;
1557 return overhead_size + method_stack + stub_code;
1558 }
1559
1560 // asm based interpreter deoptimization helpers
1561
1562 int AbstractInterpreter::layout_activation(methodOop method,
1563 int tempcount,
1564 int popframe_extra_args,
1565 int moncount,
1566 int caller_actual_parameters,
1567 int callee_param_count,
1568 int callee_locals,
1569 frame* caller,
1570 frame* interpreter_frame,
1571 bool is_top_frame) {
1572 // Note: This calculation must exactly parallel the frame setup
1573 // in AbstractInterpreterGenerator::generate_method_entry.
1574 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1575 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1576 // as determined by a previous call to this method.
1577 // It is also guaranteed to be walkable even though it is in a skeletal state
1578 // NOTE: return size is in words not bytes
1579
1580 // fixed size of an interpreter frame:
1581 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1582 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1583 Interpreter::stackElementWords;
1584
1585 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1586
1587 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1588 // Since the callee parameters already account for the callee's params we only need to account for
1589 // the extra locals.
1590
1591
1592 int size = overhead +
1593 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1594 (moncount*frame::interpreter_frame_monitor_size()) +
1595 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1596
1597 if (interpreter_frame != NULL) {
1598 #ifdef ASSERT
1599 if (!EnableInvokeDynamic)
1600 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1601 // Probably, since deoptimization doesn't work yet.
1602 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1603 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1604 #endif
1605
1606 interpreter_frame->interpreter_frame_set_method(method);
1607 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1608 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1609 // and sender_sp is fp+8
1610 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1611
1612 #ifdef ASSERT
1613 if (caller->is_interpreted_frame()) {
1614 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1615 }
1616 #endif
1617
1618 interpreter_frame->interpreter_frame_set_locals(locals);
1619 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1620 BasicObjectLock* monbot = montop - moncount;
1621 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1622
1623 // Set last_sp
1624 intptr_t* rsp = (intptr_t*) monbot -
1625 tempcount*Interpreter::stackElementWords -
1626 popframe_extra_args;
1627 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1628
1629 // All frames but the initial (oldest) interpreter frame we fill in have a
1630 // value for sender_sp that allows walking the stack but isn't
1631 // truly correct. Correct the value here.
1632
1633 if (extra_locals != 0 &&
1634 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1635 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1636 }
1637 *interpreter_frame->interpreter_frame_cache_addr() =
1638 method->constants()->cache();
1639 }
1640 return size;
1641 }
1642
1643
1644 //------------------------------------------------------------------------------------------------------------------------
1645 // Exceptions
1646
1647 void TemplateInterpreterGenerator::generate_throw_exception() {
1648 // Entry point in previous activation (i.e., if the caller was interpreted)
1649 Interpreter::_rethrow_exception_entry = __ pc();
1650 const Register thread = rcx;
1651
1652 // Restore sp to interpreter_frame_last_sp even though we are going
1653 // to empty the expression stack for the exception processing.
1654 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1655 // rax,: exception
1656 // rdx: return address/pc that threw exception
1657 __ restore_bcp(); // rsi points to call/send
1658 __ restore_locals();
1659
1660 // Entry point for exceptions thrown within interpreter code
1661 Interpreter::_throw_exception_entry = __ pc();
1662 // expression stack is undefined here
1663 // rax,: exception
1664 // rsi: exception bcp
1665 __ verify_oop(rax);
1666
1667 // expression stack must be empty before entering the VM in case of an exception
1668 __ empty_expression_stack();
1669 __ empty_FPU_stack();
1670 // find exception handler address and preserve exception oop
1671 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1672 // rax,: exception handler entry point
1673 // rdx: preserved exception oop
1674 // rsi: bcp for exception handler
1675 __ push_ptr(rdx); // push exception which is now the only value on the stack
1676 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1677
1678 // If the exception is not handled in the current frame the frame is removed and
1679 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1680 //
1681 // Note: At this point the bci is still the bxi for the instruction which caused
1682 // the exception and the expression stack is empty. Thus, for any VM calls
1683 // at this point, GC will find a legal oop map (with empty expression stack).
1684
1685 // In current activation
1686 // tos: exception
1687 // rsi: exception bcp
1688
1689 //
1690 // JVMTI PopFrame support
1691 //
1692
1693 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1694 __ empty_expression_stack();
1695 __ empty_FPU_stack();
1696 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1697 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1698 // popframe handling cycles.
1699 __ get_thread(thread);
1700 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1701 __ orl(rdx, JavaThread::popframe_processing_bit);
1702 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1703
1704 {
1705 // Check to see whether we are returning to a deoptimized frame.
1706 // (The PopFrame call ensures that the caller of the popped frame is
1707 // either interpreted or compiled and deoptimizes it if compiled.)
1708 // In this case, we can't call dispatch_next() after the frame is
1709 // popped, but instead must save the incoming arguments and restore
1710 // them after deoptimization has occurred.
1711 //
1712 // Note that we don't compare the return PC against the
1713 // deoptimization blob's unpack entry because of the presence of
1714 // adapter frames in C2.
1715 Label caller_not_deoptimized;
1716 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1717 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1718 __ testl(rax, rax);
1719 __ jcc(Assembler::notZero, caller_not_deoptimized);
1720
1721 // Compute size of arguments for saving when returning to deoptimized caller
1722 __ get_method(rax);
1723 __ verify_oop(rax);
1724 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1725 __ shlptr(rax, Interpreter::logStackElementSize);
1726 __ restore_locals();
1727 __ subptr(rdi, rax);
1728 __ addptr(rdi, wordSize);
1729 // Save these arguments
1730 __ get_thread(thread);
1731 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1732
1733 __ remove_activation(vtos, rdx,
1734 /* throw_monitor_exception */ false,
1735 /* install_monitor_exception */ false,
1736 /* notify_jvmdi */ false);
1737
1738 // Inform deoptimization that it is responsible for restoring these arguments
1739 __ get_thread(thread);
1740 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1741
1742 // Continue in deoptimization handler
1743 __ jmp(rdx);
1744
1745 __ bind(caller_not_deoptimized);
1746 }
1747
1748 __ remove_activation(vtos, rdx,
1749 /* throw_monitor_exception */ false,
1750 /* install_monitor_exception */ false,
1751 /* notify_jvmdi */ false);
1752
1753 // Finish with popframe handling
1754 // A previous I2C followed by a deoptimization might have moved the
1755 // outgoing arguments further up the stack. PopFrame expects the
1756 // mutations to those outgoing arguments to be preserved and other
1757 // constraints basically require this frame to look exactly as
1758 // though it had previously invoked an interpreted activation with
1759 // no space between the top of the expression stack (current
1760 // last_sp) and the top of stack. Rather than force deopt to
1761 // maintain this kind of invariant all the time we call a small
1762 // fixup routine to move the mutated arguments onto the top of our
1763 // expression stack if necessary.
1764 __ mov(rax, rsp);
1765 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1766 __ get_thread(thread);
1767 // PC must point into interpreter here
1768 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1769 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1770 __ get_thread(thread);
1771 __ reset_last_Java_frame(thread, true, true);
1772 // Restore the last_sp and null it out
1773 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1774 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1775
1776 __ restore_bcp();
1777 __ restore_locals();
1778 // The method data pointer was incremented already during
1779 // call profiling. We have to restore the mdp for the current bcp.
1780 if (ProfileInterpreter) {
1781 __ set_method_data_pointer_for_bcp();
1782 }
1783
1784 // Clear the popframe condition flag
1785 __ get_thread(thread);
1786 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1787
1788 __ dispatch_next(vtos);
1789 // end of PopFrame support
1790
1791 Interpreter::_remove_activation_entry = __ pc();
1792
1793 // preserve exception over this code sequence
1794 __ pop_ptr(rax);
1795 __ get_thread(thread);
1796 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1797 // remove the activation (without doing throws on illegalMonitorExceptions)
1798 __ remove_activation(vtos, rdx, false, true, false);
1799 // restore exception
1800 __ get_thread(thread);
1801 __ movptr(rax, Address(thread, JavaThread::vm_result_offset()));
1802 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
1803 __ verify_oop(rax);
1804
1805 // Inbetween activations - previous activation type unknown yet
1806 // compute continuation point - the continuation point expects
1807 // the following registers set up:
1808 //
1809 // rax: exception
1810 // rdx: return address/pc that threw exception
1811 // rsp: expression stack of caller
1812 // rbp: rbp, of caller
1813 __ push(rax); // save exception
1814 __ push(rdx); // save return address
1815 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1816 __ mov(rbx, rax); // save exception handler
1817 __ pop(rdx); // restore return address
1818 __ pop(rax); // restore exception
1819 // Note that an "issuing PC" is actually the next PC after the call
1820 __ jmp(rbx); // jump to exception handler of caller
1821 }
1822
1823
1824 //
1825 // JVMTI ForceEarlyReturn support
1826 //
1827 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1828 address entry = __ pc();
1829 const Register thread = rcx;
1830
1831 __ restore_bcp();
1832 __ restore_locals();
1833 __ empty_expression_stack();
1834 __ empty_FPU_stack();
1835 __ load_earlyret_value(state);
1836
1837 __ get_thread(thread);
1838 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1839 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1840
1841 // Clear the earlyret state
1842 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1843
1844 __ remove_activation(state, rsi,
1845 false, /* throw_monitor_exception */
1846 false, /* install_monitor_exception */
1847 true); /* notify_jvmdi */
1848 __ jmp(rsi);
1849 return entry;
1850 } // end of ForceEarlyReturn support
1851
1852
1853 //------------------------------------------------------------------------------------------------------------------------
1854 // Helper for vtos entry point generation
1855
1856 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) {
1857 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1858 Label L;
1859 fep = __ pc(); __ push(ftos); __ jmp(L);
1860 dep = __ pc(); __ push(dtos); __ jmp(L);
1861 lep = __ pc(); __ push(ltos); __ jmp(L);
1862 aep = __ pc(); __ push(atos); __ jmp(L);
1863 bep = cep = sep = // fall through
1864 iep = __ pc(); __ push(itos); // fall through
1865 vep = __ pc(); __ bind(L); // fall through
1866 generate_and_dispatch(t);
1867 }
1868
1869 //------------------------------------------------------------------------------------------------------------------------
1870 // Generation of individual instructions
1871
1872 // helpers for generate_and_dispatch
1873
1874
1875
1876 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1877 : TemplateInterpreterGenerator(code) {
1878 generate_all(); // down here so it can be "virtual"
1879 }
1880
1881 //------------------------------------------------------------------------------------------------------------------------
1882
1883 // Non-product code
1884 #ifndef PRODUCT
1885 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1886 address entry = __ pc();
1887
1888 // prepare expression stack
1889 __ pop(rcx); // pop return address so expression stack is 'pure'
1890 __ push(state); // save tosca
1891
1892 // pass tosca registers as arguments & call tracer
1893 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1894 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1895 __ pop(state); // restore tosca
1896
1897 // return
1898 __ jmp(rcx);
1899
1900 return entry;
1901 }
1902
1903
1904 void TemplateInterpreterGenerator::count_bytecode() {
1905 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1906 }
1907
1908
1909 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1910 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1911 }
1912
1913
1914 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1915 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1916 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1917 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1918 ExternalAddress table((address) BytecodePairHistogram::_counters);
1919 Address index(noreg, rbx, Address::times_4);
1920 __ incrementl(ArrayAddress(table, index));
1921 }
1922
1923
1924 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1925 // Call a little run-time stub to avoid blow-up for each bytecode.
1926 // The run-time runtime saves the right registers, depending on
1927 // the tosca in-state for the given template.
1928 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1929 "entry must have been generated");
1930 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1931 }
1932
1933
1934 void TemplateInterpreterGenerator::stop_interpreter_at() {
1935 Label L;
1936 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1937 StopInterpreterAt);
1938 __ jcc(Assembler::notEqual, L);
1939 __ int3();
1940 __ bind(L);
1941 }
1942 #endif // !PRODUCT
1943 #endif // CC_INTERP