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 // Arguments are: required type at TOS+4, failing object (or NULL) at TOS.
116 address TemplateInterpreterGenerator::generate_WrongMethodType_handler() {
117 address entry = __ pc();
118
119 __ pop(rbx); // actual failing object is at TOS
120 __ pop(rax); // required type is at TOS+4
121
122 __ verify_oop(rbx);
123 __ verify_oop(rax);
124
125 // Various method handle types use interpreter registers as temps.
126 __ restore_bcp();
127 __ restore_locals();
128
129 // Expression stack must be empty before entering the VM for an exception.
130 __ empty_expression_stack();
131 __ empty_FPU_stack();
132 __ call_VM(noreg,
133 CAST_FROM_FN_PTR(address,
134 InterpreterRuntime::throw_WrongMethodTypeException),
135 // pass required type, failing object (or NULL)
136 rax, rbx);
137 return entry;
138 }
139
140
141 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
142 assert(!pass_oop || message == NULL, "either oop or message but not both");
143 address entry = __ pc();
144 if (pass_oop) {
145 // object is at TOS
146 __ pop(rbx);
147 }
148 // expression stack must be empty before entering the VM if an exception happened
149 __ empty_expression_stack();
150 __ empty_FPU_stack();
151 // setup parameters
152 __ lea(rax, ExternalAddress((address)name));
153 if (pass_oop) {
154 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
155 } else {
156 if (message != NULL) {
157 __ lea(rbx, ExternalAddress((address)message));
158 } else {
159 __ movptr(rbx, NULL_WORD);
160 }
161 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
162 }
163 // throw exception
164 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
165 return entry;
166 }
167
168
169 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
170 address entry = __ pc();
171 // NULL last_sp until next java call
172 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
173 __ dispatch_next(state);
174 return entry;
175 }
176
177
178 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
179 TosState incoming_state = state;
180 address entry = __ pc();
181
182 #ifdef COMPILER2
183 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
184 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
185 for (int i = 1; i < 8; i++) {
186 __ ffree(i);
187 }
188 } else if (UseSSE < 2) {
189 __ empty_FPU_stack();
190 }
191 #endif
192 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
193 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
194 } else {
195 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
196 }
197
198 // In SSE mode, interpreter returns FP results in xmm0 but they need
199 // to end up back on the FPU so it can operate on them.
200 if (incoming_state == ftos && UseSSE >= 1) {
201 __ subptr(rsp, wordSize);
202 __ movflt(Address(rsp, 0), xmm0);
203 __ fld_s(Address(rsp, 0));
204 __ addptr(rsp, wordSize);
205 } else if (incoming_state == dtos && UseSSE >= 2) {
206 __ subptr(rsp, 2*wordSize);
207 __ movdbl(Address(rsp, 0), xmm0);
208 __ fld_d(Address(rsp, 0));
209 __ addptr(rsp, 2*wordSize);
210 }
211
212 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
213
214 // Restore stack bottom in case i2c adjusted stack
215 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
216 // and NULL it as marker that rsp is now tos until next java call
217 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
218
219 __ restore_bcp();
220 __ restore_locals();
221
222 Label L_got_cache, L_giant_index;
223 if (EnableInvokeDynamic) {
224 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
225 __ jcc(Assembler::equal, L_giant_index);
226 }
227 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
228 __ bind(L_got_cache);
229 __ movl(rbx, Address(rbx, rcx,
230 Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
231 ConstantPoolCacheEntry::flags_offset()));
232 __ andptr(rbx, 0xFF);
233 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
234 __ dispatch_next(state, step);
235
236 // out of the main line of code...
237 if (EnableInvokeDynamic) {
238 __ bind(L_giant_index);
239 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
240 __ jmp(L_got_cache);
241 }
242
243 return entry;
244 }
245
246
247 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
248 address entry = __ pc();
249
250 // In SSE mode, FP results are in xmm0
251 if (state == ftos && UseSSE > 0) {
252 __ subptr(rsp, wordSize);
253 __ movflt(Address(rsp, 0), xmm0);
254 __ fld_s(Address(rsp, 0));
255 __ addptr(rsp, wordSize);
256 } else if (state == dtos && UseSSE >= 2) {
257 __ subptr(rsp, 2*wordSize);
258 __ movdbl(Address(rsp, 0), xmm0);
259 __ fld_d(Address(rsp, 0));
260 __ addptr(rsp, 2*wordSize);
261 }
262
263 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
264
265 // The stack is not extended by deopt but we must NULL last_sp as this
266 // entry is like a "return".
267 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
268 __ restore_bcp();
269 __ restore_locals();
270 // handle exceptions
271 { Label L;
272 const Register thread = rcx;
273 __ get_thread(thread);
274 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
275 __ jcc(Assembler::zero, L);
276 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
277 __ should_not_reach_here();
278 __ bind(L);
279 }
280 __ dispatch_next(state, step);
281 return entry;
282 }
283
284
285 int AbstractInterpreter::BasicType_as_index(BasicType type) {
286 int i = 0;
287 switch (type) {
288 case T_BOOLEAN: i = 0; break;
289 case T_CHAR : i = 1; break;
290 case T_BYTE : i = 2; break;
291 case T_SHORT : i = 3; break;
292 case T_INT : // fall through
293 case T_LONG : // fall through
294 case T_VOID : i = 4; break;
295 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
296 case T_DOUBLE : i = 6; break;
297 case T_OBJECT : // fall through
298 case T_ARRAY : i = 7; break;
299 default : ShouldNotReachHere();
300 }
301 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
302 return i;
303 }
304
305
306 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
307 address entry = __ pc();
308 switch (type) {
309 case T_BOOLEAN: __ c2bool(rax); break;
310 case T_CHAR : __ andptr(rax, 0xFFFF); break;
311 case T_BYTE : __ sign_extend_byte (rax); break;
312 case T_SHORT : __ sign_extend_short(rax); break;
313 case T_INT : /* nothing to do */ break;
314 case T_DOUBLE :
315 case T_FLOAT :
316 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
317 __ pop(t); // remove return address first
318 // Must return a result for interpreter or compiler. In SSE
319 // mode, results are returned in xmm0 and the FPU stack must
320 // be empty.
321 if (type == T_FLOAT && UseSSE >= 1) {
322 // Load ST0
323 __ fld_d(Address(rsp, 0));
324 // Store as float and empty fpu stack
325 __ fstp_s(Address(rsp, 0));
326 // and reload
327 __ movflt(xmm0, Address(rsp, 0));
328 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
329 __ movdbl(xmm0, Address(rsp, 0));
330 } else {
331 // restore ST0
332 __ fld_d(Address(rsp, 0));
333 }
334 // and pop the temp
335 __ addptr(rsp, 2 * wordSize);
336 __ push(t); // restore return address
337 }
338 break;
339 case T_OBJECT :
340 // retrieve result from frame
341 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
342 // and verify it
343 __ verify_oop(rax);
344 break;
345 default : ShouldNotReachHere();
346 }
347 __ ret(0); // return from result handler
348 return entry;
349 }
350
351 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
352 address entry = __ pc();
353 __ push(state);
354 __ call_VM(noreg, runtime_entry);
355 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
356 return entry;
357 }
358
359
360 // Helpers for commoning out cases in the various type of method entries.
361 //
362
363 // increment invocation count & check for overflow
364 //
365 // Note: checking for negative value instead of overflow
366 // so we have a 'sticky' overflow test
367 //
368 // rbx,: method
369 // rcx: invocation counter
370 //
371 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
372 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
373 in_bytes(InvocationCounter::counter_offset()));
374 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
375 if (TieredCompilation) {
376 int increment = InvocationCounter::count_increment;
377 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
378 Label no_mdo, done;
379 if (ProfileInterpreter) {
380 // Are we profiling?
381 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
382 __ testptr(rax, rax);
383 __ jccb(Assembler::zero, no_mdo);
384 // Increment counter in the MDO
385 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
386 in_bytes(InvocationCounter::counter_offset()));
387 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
388 __ jmpb(done);
389 }
390 __ bind(no_mdo);
391 // Increment counter in methodOop (we don't need to load it, it's in rcx).
392 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
393 __ bind(done);
394 } else {
395 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() +
396 InvocationCounter::counter_offset());
397
398 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
399 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
400 }
401 // Update standard invocation counters
402 __ movl(rax, backedge_counter); // load backedge counter
403
404 __ incrementl(rcx, InvocationCounter::count_increment);
405 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
406
407 __ movl(invocation_counter, rcx); // save invocation count
408 __ addl(rcx, rax); // add both counters
409
410 // profile_method is non-null only for interpreted method so
411 // profile_method != NULL == !native_call
412 // BytecodeInterpreter only calls for native so code is elided.
413
414 if (ProfileInterpreter && profile_method != NULL) {
415 // Test to see if we should create a method data oop
416 __ cmp32(rcx,
417 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
418 __ jcc(Assembler::less, *profile_method_continue);
419
420 // if no method data exists, go to profile_method
421 __ test_method_data_pointer(rax, *profile_method);
422 }
423
424 __ cmp32(rcx,
425 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
426 __ jcc(Assembler::aboveEqual, *overflow);
427 }
428 }
429
430 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
431
432 // Asm interpreter on entry
433 // rdi - locals
434 // rsi - bcp
435 // rbx, - method
436 // rdx - cpool
437 // rbp, - interpreter frame
438
439 // C++ interpreter on entry
440 // rsi - new interpreter state pointer
441 // rbp - interpreter frame pointer
442 // rbx - method
443
444 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
445 // rbx, - method
446 // rcx - rcvr (assuming there is one)
447 // top of stack return address of interpreter caller
448 // rsp - sender_sp
449
450 // C++ interpreter only
451 // rsi - previous interpreter state pointer
452
453 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
454
455 // InterpreterRuntime::frequency_counter_overflow takes one argument
456 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
457 // The call returns the address of the verified entry point for the method or NULL
458 // if the compilation did not complete (either went background or bailed out).
459 __ movptr(rax, (intptr_t)false);
460 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
461
462 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
463
464 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
465 // and jump to the interpreted entry.
466 __ jmp(*do_continue, relocInfo::none);
467
468 }
469
470 void InterpreterGenerator::generate_stack_overflow_check(void) {
471 // see if we've got enough room on the stack for locals plus overhead.
472 // the expression stack grows down incrementally, so the normal guard
473 // page mechanism will work for that.
474 //
475 // Registers live on entry:
476 //
477 // Asm interpreter
478 // rdx: number of additional locals this frame needs (what we must check)
479 // rbx,: methodOop
480
481 // destroyed on exit
482 // rax,
483
484 // NOTE: since the additional locals are also always pushed (wasn't obvious in
485 // generate_method_entry) so the guard should work for them too.
486 //
487
488 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
489 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
490
491 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
492 // be sure to change this if you add/subtract anything to/from the overhead area
493 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
494
495 const int page_size = os::vm_page_size();
496
497 Label after_frame_check;
498
499 // see if the frame is greater than one page in size. If so,
500 // then we need to verify there is enough stack space remaining
501 // for the additional locals.
502 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
503 __ jcc(Assembler::belowEqual, after_frame_check);
504
505 // compute rsp as if this were going to be the last frame on
506 // the stack before the red zone
507
508 Label after_frame_check_pop;
509
510 __ push(rsi);
511
512 const Register thread = rsi;
513
514 __ get_thread(thread);
515
516 const Address stack_base(thread, Thread::stack_base_offset());
517 const Address stack_size(thread, Thread::stack_size_offset());
518
519 // locals + overhead, in bytes
520 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
521
522 #ifdef ASSERT
523 Label stack_base_okay, stack_size_okay;
524 // verify that thread stack base is non-zero
525 __ cmpptr(stack_base, (int32_t)NULL_WORD);
526 __ jcc(Assembler::notEqual, stack_base_okay);
527 __ stop("stack base is zero");
528 __ bind(stack_base_okay);
529 // verify that thread stack size is non-zero
530 __ cmpptr(stack_size, 0);
531 __ jcc(Assembler::notEqual, stack_size_okay);
532 __ stop("stack size is zero");
533 __ bind(stack_size_okay);
534 #endif
535
536 // Add stack base to locals and subtract stack size
537 __ addptr(rax, stack_base);
538 __ subptr(rax, stack_size);
539
540 // Use the maximum number of pages we might bang.
541 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
542 (StackRedPages+StackYellowPages);
543 __ addptr(rax, max_pages * page_size);
544
545 // check against the current stack bottom
546 __ cmpptr(rsp, rax);
547 __ jcc(Assembler::above, after_frame_check_pop);
548
549 __ pop(rsi); // get saved bcp / (c++ prev state ).
550
551 __ pop(rax); // get return address
552 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
553
554 // all done with frame size check
555 __ bind(after_frame_check_pop);
556 __ pop(rsi);
557
558 __ bind(after_frame_check);
559 }
560
561 // Allocate monitor and lock method (asm interpreter)
562 // rbx, - methodOop
563 //
564 void InterpreterGenerator::lock_method(void) {
565 // synchronize method
566 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
567 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
568 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
569
570 #ifdef ASSERT
571 { Label L;
572 __ movl(rax, access_flags);
573 __ testl(rax, JVM_ACC_SYNCHRONIZED);
574 __ jcc(Assembler::notZero, L);
575 __ stop("method doesn't need synchronization");
576 __ bind(L);
577 }
578 #endif // ASSERT
579 // get synchronization object
580 { Label done;
581 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
582 __ movl(rax, access_flags);
583 __ testl(rax, JVM_ACC_STATIC);
584 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
585 __ jcc(Assembler::zero, done);
586 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
587 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
588 __ movptr(rax, Address(rax, mirror_offset));
589 __ bind(done);
590 }
591 // add space for monitor & lock
592 __ subptr(rsp, entry_size); // add space for a monitor entry
593 __ movptr(monitor_block_top, rsp); // set new monitor block top
594 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
595 __ mov(rdx, rsp); // object address
596 __ lock_object(rdx);
597 }
598
599 //
600 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
601 // and for native methods hence the shared code.
602
603 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
604 // initialize fixed part of activation frame
605 __ push(rax); // save return address
606 __ enter(); // save old & set new rbp,
607
608
609 __ push(rsi); // set sender sp
610 __ push((int32_t)NULL_WORD); // leave last_sp as null
611 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
612 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
613 __ push(rbx); // save methodOop
614 if (ProfileInterpreter) {
615 Label method_data_continue;
616 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
617 __ testptr(rdx, rdx);
618 __ jcc(Assembler::zero, method_data_continue);
619 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
620 __ bind(method_data_continue);
621 __ push(rdx); // set the mdp (method data pointer)
622 } else {
623 __ push(0);
624 }
625
626 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
627 __ movptr(rdx, Address(rdx, constantPoolOopDesc::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,: methodOop
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 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
682 // read first instruction word and extract bytecode @ 1 and index @ 2
683 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
684 __ movl(rdx, Address(rdx, constMethodOopDesc::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, constantPoolOopDesc::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, constantPoolCacheOopDesc::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, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
717 __ movl(rdx,
718 Address(rdi,
719 rdx,
720 Address::times_ptr, constantPoolCacheOopDesc::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::tosBits);
729 // Make sure we don't need to mask rdx for tosBits after the above shift
730 ConstantPoolCacheEntry::verify_tosBits();
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 #ifndef SERIALGC
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,: methodOop
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 // SERIALGC
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,: methodOop
882 // rsi: sender sp
883 // rsi: previous interpreter state (C++ interpreter) must preserve
884 address entry_point = __ pc();
885
886
887 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
888 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
889 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
890
891 // get parameter size (always needed)
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,: methodOop
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 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
916 // initialize fixed part of activation frame
917
918 generate_fixed_frame(true);
919
920 // make sure method is native & not abstract
921 #ifdef ASSERT
922 __ movl(rax, access_flags);
923 {
924 Label L;
925 __ testl(rax, JVM_ACC_NATIVE);
926 __ jcc(Assembler::notZero, L);
927 __ stop("tried to execute non-native method as native");
928 __ bind(L);
929 }
930 { Label L;
931 __ testl(rax, JVM_ACC_ABSTRACT);
932 __ jcc(Assembler::zero, L);
933 __ stop("tried to execute abstract method in interpreter");
934 __ bind(L);
935 }
936 #endif
937
938 // Since at this point in the method invocation the exception handler
939 // would try to exit the monitor of synchronized methods which hasn't
940 // been entered yet, we set the thread local variable
941 // _do_not_unlock_if_synchronized to true. The remove_activation will
942 // check this flag.
943
944 __ get_thread(rax);
945 const Address do_not_unlock_if_synchronized(rax,
946 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
947 __ movbool(do_not_unlock_if_synchronized, true);
948
949 // increment invocation count & check for overflow
950 Label invocation_counter_overflow;
951 if (inc_counter) {
952 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
953 }
954
955 Label continue_after_compile;
956 __ bind(continue_after_compile);
957
958 bang_stack_shadow_pages(true);
959
960 // reset the _do_not_unlock_if_synchronized flag
961 __ get_thread(rax);
962 __ movbool(do_not_unlock_if_synchronized, false);
963
964 // check for synchronized methods
965 // Must happen AFTER invocation_counter check and stack overflow check,
966 // so method is not locked if overflows.
967 //
968 if (synchronized) {
969 lock_method();
970 } else {
971 // no synchronization necessary
972 #ifdef ASSERT
973 { Label L;
974 __ movl(rax, access_flags);
975 __ testl(rax, JVM_ACC_SYNCHRONIZED);
976 __ jcc(Assembler::zero, L);
977 __ stop("method needs synchronization");
978 __ bind(L);
979 }
980 #endif
981 }
982
983 // start execution
984 #ifdef ASSERT
985 { Label L;
986 const Address monitor_block_top (rbp,
987 frame::interpreter_frame_monitor_block_top_offset * wordSize);
988 __ movptr(rax, monitor_block_top);
989 __ cmpptr(rax, rsp);
990 __ jcc(Assembler::equal, L);
991 __ stop("broken stack frame setup in interpreter");
992 __ bind(L);
993 }
994 #endif
995
996 // jvmti/dtrace support
997 __ notify_method_entry();
998
999 // work registers
1000 const Register method = rbx;
1001 const Register thread = rdi;
1002 const Register t = rcx;
1003
1004 // allocate space for parameters
1005 __ get_method(method);
1006 __ verify_oop(method);
1007 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
1008 __ shlptr(t, Interpreter::logStackElementSize);
1009 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1010 __ subptr(rsp, t);
1011 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1012
1013 // get signature handler
1014 { Label L;
1015 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
1016 __ testptr(t, t);
1017 __ jcc(Assembler::notZero, L);
1018 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1019 __ get_method(method);
1020 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
1021 __ bind(L);
1022 }
1023
1024 // call signature handler
1025 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1026 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1027 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1028 // The generated handlers do not touch RBX (the method oop).
1029 // However, large signatures cannot be cached and are generated
1030 // each time here. The slow-path generator will blow RBX
1031 // sometime, so we must reload it after the call.
1032 __ call(t);
1033 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1034
1035 // result handler is in rax,
1036 // set result handler
1037 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1038
1039 // pass mirror handle if static call
1040 { Label L;
1041 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
1042 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1043 __ testl(t, JVM_ACC_STATIC);
1044 __ jcc(Assembler::zero, L);
1045 // get mirror
1046 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
1047 __ movptr(t, Address(t, constantPoolOopDesc::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, methodOopDesc::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 __ verify_oop(method);
1066 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
1067 __ bind(L);
1068 }
1069
1070 // pass JNIEnv
1071 __ get_thread(thread);
1072 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1073 __ movptr(Address(rsp, 0), t);
1074
1075 // set_last_Java_frame_before_call
1076 // It is enough that the pc()
1077 // points into the right code segment. It does not have to be the correct return pc.
1078 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1079
1080 // change thread state
1081 #ifdef ASSERT
1082 { Label L;
1083 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1084 __ cmpl(t, _thread_in_Java);
1085 __ jcc(Assembler::equal, L);
1086 __ stop("Wrong thread state in native stub");
1087 __ bind(L);
1088 }
1089 #endif
1090
1091 // Change state to native
1092 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1093 __ call(rax);
1094
1095 // result potentially in rdx:rax or ST0
1096
1097 // Either restore the MXCSR register after returning from the JNI Call
1098 // or verify that it wasn't changed.
1099 if (VM_Version::supports_sse()) {
1100 if (RestoreMXCSROnJNICalls) {
1101 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1102 }
1103 else if (CheckJNICalls ) {
1104 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1105 }
1106 }
1107
1108 // Either restore the x87 floating pointer control word after returning
1109 // from the JNI call or verify that it wasn't changed.
1110 if (CheckJNICalls) {
1111 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1112 }
1113
1114 // save potential result in ST(0) & rdx:rax
1115 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1116 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1117 // It is safe to do this push because state is _thread_in_native and return address will be found
1118 // via _last_native_pc and not via _last_jave_sp
1119
1120 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1121 // If the order changes or anything else is added to the stack the code in
1122 // interpreter_frame_result will have to be changed.
1123
1124 { Label L;
1125 Label push_double;
1126 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1127 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1128 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1129 float_handler.addr());
1130 __ jcc(Assembler::equal, push_double);
1131 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1132 double_handler.addr());
1133 __ jcc(Assembler::notEqual, L);
1134 __ bind(push_double);
1135 __ push(dtos);
1136 __ bind(L);
1137 }
1138 __ push(ltos);
1139
1140 // change thread state
1141 __ get_thread(thread);
1142 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1143 if(os::is_MP()) {
1144 if (UseMembar) {
1145 // Force this write out before the read below
1146 __ membar(Assembler::Membar_mask_bits(
1147 Assembler::LoadLoad | Assembler::LoadStore |
1148 Assembler::StoreLoad | Assembler::StoreStore));
1149 } else {
1150 // Write serialization page so VM thread can do a pseudo remote membar.
1151 // We use the current thread pointer to calculate a thread specific
1152 // offset to write to within the page. This minimizes bus traffic
1153 // due to cache line collision.
1154 __ serialize_memory(thread, rcx);
1155 }
1156 }
1157
1158 if (AlwaysRestoreFPU) {
1159 // Make sure the control word is correct.
1160 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1161 }
1162
1163 // check for safepoint operation in progress and/or pending suspend requests
1164 { Label Continue;
1165
1166 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1167 SafepointSynchronize::_not_synchronized);
1168
1169 Label L;
1170 __ jcc(Assembler::notEqual, L);
1171 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1172 __ jcc(Assembler::equal, Continue);
1173 __ bind(L);
1174
1175 // Don't use call_VM as it will see a possible pending exception and forward it
1176 // and never return here preventing us from clearing _last_native_pc down below.
1177 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1178 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1179 // by hand.
1180 //
1181 __ push(thread);
1182 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1183 JavaThread::check_special_condition_for_native_trans)));
1184 __ increment(rsp, wordSize);
1185 __ get_thread(thread);
1186
1187 __ bind(Continue);
1188 }
1189
1190 // change thread state
1191 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1192
1193 __ reset_last_Java_frame(thread, true, true);
1194
1195 // reset handle block
1196 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1197 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1198
1199 // If result was an oop then unbox and save it in the frame
1200 { Label L;
1201 Label no_oop, store_result;
1202 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1203 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1204 handler.addr());
1205 __ jcc(Assembler::notEqual, no_oop);
1206 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1207 __ pop(ltos);
1208 __ testptr(rax, rax);
1209 __ jcc(Assembler::zero, store_result);
1210 // unbox
1211 __ movptr(rax, Address(rax, 0));
1212 __ bind(store_result);
1213 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1214 // keep stack depth as expected by pushing oop which will eventually be discarded
1215 __ push(ltos);
1216 __ bind(no_oop);
1217 }
1218
1219 {
1220 Label no_reguard;
1221 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1222 __ jcc(Assembler::notEqual, no_reguard);
1223
1224 __ pusha();
1225 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1226 __ popa();
1227
1228 __ bind(no_reguard);
1229 }
1230
1231 // restore rsi to have legal interpreter frame,
1232 // i.e., bci == 0 <=> rsi == code_base()
1233 // Can't call_VM until bcp is within reasonable.
1234 __ get_method(method); // method is junk from thread_in_native to now.
1235 __ verify_oop(method);
1236 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
1237 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
1238
1239 // handle exceptions (exception handling will handle unlocking!)
1240 { Label L;
1241 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1242 __ jcc(Assembler::zero, L);
1243 // Note: At some point we may want to unify this with the code used in call_VM_base();
1244 // i.e., we should use the StubRoutines::forward_exception code. For now this
1245 // doesn't work here because the rsp is not correctly set at this point.
1246 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1247 __ should_not_reach_here();
1248 __ bind(L);
1249 }
1250
1251 // do unlocking if necessary
1252 { Label L;
1253 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1254 __ testl(t, JVM_ACC_SYNCHRONIZED);
1255 __ jcc(Assembler::zero, L);
1256 // the code below should be shared with interpreter macro assembler implementation
1257 { Label unlock;
1258 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1259 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1260 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1261
1262 __ lea(rdx, monitor); // address of first monitor
1263
1264 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1265 __ testptr(t, t);
1266 __ jcc(Assembler::notZero, unlock);
1267
1268 // Entry already unlocked, need to throw exception
1269 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1270 __ should_not_reach_here();
1271
1272 __ bind(unlock);
1273 __ unlock_object(rdx);
1274 }
1275 __ bind(L);
1276 }
1277
1278 // jvmti/dtrace support
1279 // Note: This must happen _after_ handling/throwing any exceptions since
1280 // the exception handler code notifies the runtime of method exits
1281 // too. If this happens before, method entry/exit notifications are
1282 // not properly paired (was bug - gri 11/22/99).
1283 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1284
1285 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1286 __ pop(ltos);
1287 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1288 __ call(t);
1289
1290 // remove activation
1291 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1292 __ leave(); // remove frame anchor
1293 __ pop(rdi); // get return address
1294 __ mov(rsp, t); // set sp to sender sp
1295 __ jmp(rdi);
1296
1297 if (inc_counter) {
1298 // Handle overflow of counter and compile method
1299 __ bind(invocation_counter_overflow);
1300 generate_counter_overflow(&continue_after_compile);
1301 }
1302
1303 return entry_point;
1304 }
1305
1306 //
1307 // Generic interpreted method entry to (asm) interpreter
1308 //
1309 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1310 // determine code generation flags
1311 bool inc_counter = UseCompiler || CountCompiledCalls;
1312
1313 // rbx,: methodOop
1314 // rsi: sender sp
1315 address entry_point = __ pc();
1316
1317
1318 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1319 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
1320 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1321 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
1322
1323 // get parameter size (always needed)
1324 __ load_unsigned_short(rcx, size_of_parameters);
1325
1326 // rbx,: methodOop
1327 // rcx: size of parameters
1328
1329 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1330
1331 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1332 __ subl(rdx, rcx); // rdx = no. of additional locals
1333
1334 // see if we've got enough room on the stack for locals plus overhead.
1335 generate_stack_overflow_check();
1336
1337 // get return address
1338 __ pop(rax);
1339
1340 // compute beginning of parameters (rdi)
1341 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1342
1343 // rdx - # of additional locals
1344 // allocate space for locals
1345 // explicitly initialize locals
1346 {
1347 Label exit, loop;
1348 __ testl(rdx, rdx);
1349 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1350 __ bind(loop);
1351 __ push((int32_t)NULL_WORD); // initialize local variables
1352 __ decrement(rdx); // until everything initialized
1353 __ jcc(Assembler::greater, loop);
1354 __ bind(exit);
1355 }
1356
1357 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1358 // initialize fixed part of activation frame
1359 generate_fixed_frame(false);
1360
1361 // make sure method is not native & not abstract
1362 #ifdef ASSERT
1363 __ movl(rax, access_flags);
1364 {
1365 Label L;
1366 __ testl(rax, JVM_ACC_NATIVE);
1367 __ jcc(Assembler::zero, L);
1368 __ stop("tried to execute native method as non-native");
1369 __ bind(L);
1370 }
1371 { Label L;
1372 __ testl(rax, JVM_ACC_ABSTRACT);
1373 __ jcc(Assembler::zero, L);
1374 __ stop("tried to execute abstract method in interpreter");
1375 __ bind(L);
1376 }
1377 #endif
1378
1379 // Since at this point in the method invocation the exception handler
1380 // would try to exit the monitor of synchronized methods which hasn't
1381 // been entered yet, we set the thread local variable
1382 // _do_not_unlock_if_synchronized to true. The remove_activation will
1383 // check this flag.
1384
1385 __ get_thread(rax);
1386 const Address do_not_unlock_if_synchronized(rax,
1387 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1388 __ movbool(do_not_unlock_if_synchronized, true);
1389
1390 // increment invocation count & check for overflow
1391 Label invocation_counter_overflow;
1392 Label profile_method;
1393 Label profile_method_continue;
1394 if (inc_counter) {
1395 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1396 if (ProfileInterpreter) {
1397 __ bind(profile_method_continue);
1398 }
1399 }
1400 Label continue_after_compile;
1401 __ bind(continue_after_compile);
1402
1403 bang_stack_shadow_pages(false);
1404
1405 // reset the _do_not_unlock_if_synchronized flag
1406 __ get_thread(rax);
1407 __ movbool(do_not_unlock_if_synchronized, false);
1408
1409 // check for synchronized methods
1410 // Must happen AFTER invocation_counter check and stack overflow check,
1411 // so method is not locked if overflows.
1412 //
1413 if (synchronized) {
1414 // Allocate monitor and lock method
1415 lock_method();
1416 } else {
1417 // no synchronization necessary
1418 #ifdef ASSERT
1419 { Label L;
1420 __ movl(rax, access_flags);
1421 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1422 __ jcc(Assembler::zero, L);
1423 __ stop("method needs synchronization");
1424 __ bind(L);
1425 }
1426 #endif
1427 }
1428
1429 // start execution
1430 #ifdef ASSERT
1431 { Label L;
1432 const Address monitor_block_top (rbp,
1433 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1434 __ movptr(rax, monitor_block_top);
1435 __ cmpptr(rax, rsp);
1436 __ jcc(Assembler::equal, L);
1437 __ stop("broken stack frame setup in interpreter");
1438 __ bind(L);
1439 }
1440 #endif
1441
1442 // jvmti support
1443 __ notify_method_entry();
1444
1445 __ dispatch_next(vtos);
1446
1447 // invocation counter overflow
1448 if (inc_counter) {
1449 if (ProfileInterpreter) {
1450 // We have decided to profile this method in the interpreter
1451 __ bind(profile_method);
1452 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1453 __ set_method_data_pointer_for_bcp();
1454 __ get_method(rbx);
1455 __ jmp(profile_method_continue);
1456 }
1457 // Handle overflow of counter and compile method
1458 __ bind(invocation_counter_overflow);
1459 generate_counter_overflow(&continue_after_compile);
1460 }
1461
1462 return entry_point;
1463 }
1464
1465 //------------------------------------------------------------------------------------------------------------------------
1466 // Entry points
1467 //
1468 // Here we generate the various kind of entries into the interpreter.
1469 // The two main entry type are generic bytecode methods and native call method.
1470 // These both come in synchronized and non-synchronized versions but the
1471 // frame layout they create is very similar. The other method entry
1472 // types are really just special purpose entries that are really entry
1473 // and interpretation all in one. These are for trivial methods like
1474 // accessor, empty, or special math methods.
1475 //
1476 // When control flow reaches any of the entry types for the interpreter
1477 // the following holds ->
1478 //
1479 // Arguments:
1480 //
1481 // rbx,: methodOop
1482 // rcx: receiver
1483 //
1484 //
1485 // Stack layout immediately at entry
1486 //
1487 // [ return address ] <--- rsp
1488 // [ parameter n ]
1489 // ...
1490 // [ parameter 1 ]
1491 // [ expression stack ] (caller's java expression stack)
1492
1493 // Assuming that we don't go to one of the trivial specialized
1494 // entries the stack will look like below when we are ready to execute
1495 // the first bytecode (or call the native routine). The register usage
1496 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1497 //
1498 // local variables follow incoming parameters immediately; i.e.
1499 // the return address is moved to the end of the locals).
1500 //
1501 // [ monitor entry ] <--- rsp
1502 // ...
1503 // [ monitor entry ]
1504 // [ expr. stack bottom ]
1505 // [ saved rsi ]
1506 // [ current rdi ]
1507 // [ methodOop ]
1508 // [ saved rbp, ] <--- rbp,
1509 // [ return address ]
1510 // [ local variable m ]
1511 // ...
1512 // [ local variable 1 ]
1513 // [ parameter n ]
1514 // ...
1515 // [ parameter 1 ] <--- rdi
1516
1517 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1518 // determine code generation flags
1519 bool synchronized = false;
1520 address entry_point = NULL;
1521
1522 switch (kind) {
1523 case Interpreter::zerolocals : break;
1524 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1525 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1526 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1527 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1528 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1529 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1530 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1531
1532 case Interpreter::java_lang_math_sin : // fall thru
1533 case Interpreter::java_lang_math_cos : // fall thru
1534 case Interpreter::java_lang_math_tan : // fall thru
1535 case Interpreter::java_lang_math_abs : // fall thru
1536 case Interpreter::java_lang_math_log : // fall thru
1537 case Interpreter::java_lang_math_log10 : // fall thru
1538 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1539 case Interpreter::java_lang_ref_reference_get
1540 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1541 default : ShouldNotReachHere(); break;
1542 }
1543
1544 if (entry_point) return entry_point;
1545
1546 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1547
1548 }
1549
1550 // These should never be compiled since the interpreter will prefer
1551 // the compiled version to the intrinsic version.
1552 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1553 switch (method_kind(m)) {
1554 case Interpreter::java_lang_math_sin : // fall thru
1555 case Interpreter::java_lang_math_cos : // fall thru
1556 case Interpreter::java_lang_math_tan : // fall thru
1557 case Interpreter::java_lang_math_abs : // fall thru
1558 case Interpreter::java_lang_math_log : // fall thru
1559 case Interpreter::java_lang_math_log10 : // fall thru
1560 case Interpreter::java_lang_math_sqrt :
1561 return false;
1562 default:
1563 return true;
1564 }
1565 }
1566
1567 // How much stack a method activation needs in words.
1568 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1569
1570 const int stub_code = 4; // see generate_call_stub
1571 // Save space for one monitor to get into the interpreted method in case
1572 // the method is synchronized
1573 int monitor_size = method->is_synchronized() ?
1574 1*frame::interpreter_frame_monitor_size() : 0;
1575
1576 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1577 // be sure to change this if you add/subtract anything to/from the overhead area
1578 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1579
1580 const int extra_stack = methodOopDesc::extra_stack_entries();
1581 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1582 Interpreter::stackElementWords;
1583 return overhead_size + method_stack + stub_code;
1584 }
1585
1586 // asm based interpreter deoptimization helpers
1587
1588 int AbstractInterpreter::layout_activation(methodOop method,
1589 int tempcount,
1590 int popframe_extra_args,
1591 int moncount,
1592 int caller_actual_parameters,
1593 int callee_param_count,
1594 int callee_locals,
1595 frame* caller,
1596 frame* interpreter_frame,
1597 bool is_top_frame) {
1598 // Note: This calculation must exactly parallel the frame setup
1599 // in AbstractInterpreterGenerator::generate_method_entry.
1600 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1601 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1602 // as determined by a previous call to this method.
1603 // It is also guaranteed to be walkable even though it is in a skeletal state
1604 // NOTE: return size is in words not bytes
1605
1606 // fixed size of an interpreter frame:
1607 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1608 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1609 Interpreter::stackElementWords;
1610
1611 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1612
1613 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1614 // Since the callee parameters already account for the callee's params we only need to account for
1615 // the extra locals.
1616
1617
1618 int size = overhead +
1619 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1620 (moncount*frame::interpreter_frame_monitor_size()) +
1621 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1622
1623 if (interpreter_frame != NULL) {
1624 #ifdef ASSERT
1625 if (!EnableInvokeDynamic)
1626 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1627 // Probably, since deoptimization doesn't work yet.
1628 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1629 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1630 #endif
1631
1632 interpreter_frame->interpreter_frame_set_method(method);
1633 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1634 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1635 // and sender_sp is fp+8
1636 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1637
1638 interpreter_frame->interpreter_frame_set_locals(locals);
1639 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1640 BasicObjectLock* monbot = montop - moncount;
1641 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1642
1643 // Set last_sp
1644 intptr_t* rsp = (intptr_t*) monbot -
1645 tempcount*Interpreter::stackElementWords -
1646 popframe_extra_args;
1647 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1648
1649 // All frames but the initial (oldest) interpreter frame we fill in have a
1650 // value for sender_sp that allows walking the stack but isn't
1651 // truly correct. Correct the value here.
1652
1653 if (extra_locals != 0 &&
1654 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1655 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1656 }
1657 *interpreter_frame->interpreter_frame_cache_addr() =
1658 method->constants()->cache();
1659 }
1660 return size;
1661 }
1662
1663
1664 //------------------------------------------------------------------------------------------------------------------------
1665 // Exceptions
1666
1667 void TemplateInterpreterGenerator::generate_throw_exception() {
1668 // Entry point in previous activation (i.e., if the caller was interpreted)
1669 Interpreter::_rethrow_exception_entry = __ pc();
1670 const Register thread = rcx;
1671
1672 // Restore sp to interpreter_frame_last_sp even though we are going
1673 // to empty the expression stack for the exception processing.
1674 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1675 // rax,: exception
1676 // rdx: return address/pc that threw exception
1677 __ restore_bcp(); // rsi points to call/send
1678 __ restore_locals();
1679
1680 // Entry point for exceptions thrown within interpreter code
1681 Interpreter::_throw_exception_entry = __ pc();
1682 // expression stack is undefined here
1683 // rax,: exception
1684 // rsi: exception bcp
1685 __ verify_oop(rax);
1686
1687 // expression stack must be empty before entering the VM in case of an exception
1688 __ empty_expression_stack();
1689 __ empty_FPU_stack();
1690 // find exception handler address and preserve exception oop
1691 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1692 // rax,: exception handler entry point
1693 // rdx: preserved exception oop
1694 // rsi: bcp for exception handler
1695 __ push_ptr(rdx); // push exception which is now the only value on the stack
1696 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1697
1698 // If the exception is not handled in the current frame the frame is removed and
1699 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1700 //
1701 // Note: At this point the bci is still the bxi for the instruction which caused
1702 // the exception and the expression stack is empty. Thus, for any VM calls
1703 // at this point, GC will find a legal oop map (with empty expression stack).
1704
1705 // In current activation
1706 // tos: exception
1707 // rsi: exception bcp
1708
1709 //
1710 // JVMTI PopFrame support
1711 //
1712
1713 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1714 __ empty_expression_stack();
1715 __ empty_FPU_stack();
1716 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1717 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1718 // popframe handling cycles.
1719 __ get_thread(thread);
1720 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1721 __ orl(rdx, JavaThread::popframe_processing_bit);
1722 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1723
1724 {
1725 // Check to see whether we are returning to a deoptimized frame.
1726 // (The PopFrame call ensures that the caller of the popped frame is
1727 // either interpreted or compiled and deoptimizes it if compiled.)
1728 // In this case, we can't call dispatch_next() after the frame is
1729 // popped, but instead must save the incoming arguments and restore
1730 // them after deoptimization has occurred.
1731 //
1732 // Note that we don't compare the return PC against the
1733 // deoptimization blob's unpack entry because of the presence of
1734 // adapter frames in C2.
1735 Label caller_not_deoptimized;
1736 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1737 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1738 __ testl(rax, rax);
1739 __ jcc(Assembler::notZero, caller_not_deoptimized);
1740
1741 // Compute size of arguments for saving when returning to deoptimized caller
1742 __ get_method(rax);
1743 __ verify_oop(rax);
1744 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1745 __ shlptr(rax, Interpreter::logStackElementSize);
1746 __ restore_locals();
1747 __ subptr(rdi, rax);
1748 __ addptr(rdi, wordSize);
1749 // Save these arguments
1750 __ get_thread(thread);
1751 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1752
1753 __ remove_activation(vtos, rdx,
1754 /* throw_monitor_exception */ false,
1755 /* install_monitor_exception */ false,
1756 /* notify_jvmdi */ false);
1757
1758 // Inform deoptimization that it is responsible for restoring these arguments
1759 __ get_thread(thread);
1760 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1761
1762 // Continue in deoptimization handler
1763 __ jmp(rdx);
1764
1765 __ bind(caller_not_deoptimized);
1766 }
1767
1768 __ remove_activation(vtos, rdx,
1769 /* throw_monitor_exception */ false,
1770 /* install_monitor_exception */ false,
1771 /* notify_jvmdi */ false);
1772
1773 // Finish with popframe handling
1774 // A previous I2C followed by a deoptimization might have moved the
1775 // outgoing arguments further up the stack. PopFrame expects the
1776 // mutations to those outgoing arguments to be preserved and other
1777 // constraints basically require this frame to look exactly as
1778 // though it had previously invoked an interpreted activation with
1779 // no space between the top of the expression stack (current
1780 // last_sp) and the top of stack. Rather than force deopt to
1781 // maintain this kind of invariant all the time we call a small
1782 // fixup routine to move the mutated arguments onto the top of our
1783 // expression stack if necessary.
1784 __ mov(rax, rsp);
1785 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1786 __ get_thread(thread);
1787 // PC must point into interpreter here
1788 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1789 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1790 __ get_thread(thread);
1791 __ reset_last_Java_frame(thread, true, true);
1792 // Restore the last_sp and null it out
1793 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1794 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1795
1796 __ restore_bcp();
1797 __ restore_locals();
1798 // The method data pointer was incremented already during
1799 // call profiling. We have to restore the mdp for the current bcp.
1800 if (ProfileInterpreter) {
1801 __ set_method_data_pointer_for_bcp();
1802 }
1803
1804 // Clear the popframe condition flag
1805 __ get_thread(thread);
1806 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1807
1808 __ dispatch_next(vtos);
1809 // end of PopFrame support
1810
1811 Interpreter::_remove_activation_entry = __ pc();
1812
1813 // preserve exception over this code sequence
1814 __ pop_ptr(rax);
1815 __ get_thread(thread);
1816 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1817 // remove the activation (without doing throws on illegalMonitorExceptions)
1818 __ remove_activation(vtos, rdx, false, true, false);
1819 // restore exception
1820 __ get_thread(thread);
1821 __ movptr(rax, Address(thread, JavaThread::vm_result_offset()));
1822 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
1823 __ verify_oop(rax);
1824
1825 // Inbetween activations - previous activation type unknown yet
1826 // compute continuation point - the continuation point expects
1827 // the following registers set up:
1828 //
1829 // rax: exception
1830 // rdx: return address/pc that threw exception
1831 // rsp: expression stack of caller
1832 // rbp: rbp, of caller
1833 __ push(rax); // save exception
1834 __ push(rdx); // save return address
1835 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1836 __ mov(rbx, rax); // save exception handler
1837 __ pop(rdx); // restore return address
1838 __ pop(rax); // restore exception
1839 // Note that an "issuing PC" is actually the next PC after the call
1840 __ jmp(rbx); // jump to exception handler of caller
1841 }
1842
1843
1844 //
1845 // JVMTI ForceEarlyReturn support
1846 //
1847 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1848 address entry = __ pc();
1849 const Register thread = rcx;
1850
1851 __ restore_bcp();
1852 __ restore_locals();
1853 __ empty_expression_stack();
1854 __ empty_FPU_stack();
1855 __ load_earlyret_value(state);
1856
1857 __ get_thread(thread);
1858 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1859 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1860
1861 // Clear the earlyret state
1862 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1863
1864 __ remove_activation(state, rsi,
1865 false, /* throw_monitor_exception */
1866 false, /* install_monitor_exception */
1867 true); /* notify_jvmdi */
1868 __ jmp(rsi);
1869 return entry;
1870 } // end of ForceEarlyReturn support
1871
1872
1873 //------------------------------------------------------------------------------------------------------------------------
1874 // Helper for vtos entry point generation
1875
1876 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) {
1877 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1878 Label L;
1879 fep = __ pc(); __ push(ftos); __ jmp(L);
1880 dep = __ pc(); __ push(dtos); __ jmp(L);
1881 lep = __ pc(); __ push(ltos); __ jmp(L);
1882 aep = __ pc(); __ push(atos); __ jmp(L);
1883 bep = cep = sep = // fall through
1884 iep = __ pc(); __ push(itos); // fall through
1885 vep = __ pc(); __ bind(L); // fall through
1886 generate_and_dispatch(t);
1887 }
1888
1889 //------------------------------------------------------------------------------------------------------------------------
1890 // Generation of individual instructions
1891
1892 // helpers for generate_and_dispatch
1893
1894
1895
1896 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1897 : TemplateInterpreterGenerator(code) {
1898 generate_all(); // down here so it can be "virtual"
1899 }
1900
1901 //------------------------------------------------------------------------------------------------------------------------
1902
1903 // Non-product code
1904 #ifndef PRODUCT
1905 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1906 address entry = __ pc();
1907
1908 // prepare expression stack
1909 __ pop(rcx); // pop return address so expression stack is 'pure'
1910 __ push(state); // save tosca
1911
1912 // pass tosca registers as arguments & call tracer
1913 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1914 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1915 __ pop(state); // restore tosca
1916
1917 // return
1918 __ jmp(rcx);
1919
1920 return entry;
1921 }
1922
1923
1924 void TemplateInterpreterGenerator::count_bytecode() {
1925 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1926 }
1927
1928
1929 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1930 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1931 }
1932
1933
1934 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1935 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1936 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1937 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1938 ExternalAddress table((address) BytecodePairHistogram::_counters);
1939 Address index(noreg, rbx, Address::times_4);
1940 __ incrementl(ArrayAddress(table, index));
1941 }
1942
1943
1944 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1945 // Call a little run-time stub to avoid blow-up for each bytecode.
1946 // The run-time runtime saves the right registers, depending on
1947 // the tosca in-state for the given template.
1948 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1949 "entry must have been generated");
1950 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1951 }
1952
1953
1954 void TemplateInterpreterGenerator::stop_interpreter_at() {
1955 Label L;
1956 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1957 StopInterpreterAt);
1958 __ jcc(Assembler::notEqual, L);
1959 __ int3();
1960 __ bind(L);
1961 }
1962 #endif // !PRODUCT
1963 #endif // CC_INTERP