1 /*
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "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
181 Label interpreter_entry;
182 address compiled_entry = __ pc();
183
184 #ifdef COMPILER2
185 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
186 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
187 for (int i = 1; i < 8; i++) {
188 __ ffree(i);
189 }
190 } else if (UseSSE < 2) {
191 __ empty_FPU_stack();
192 }
193 #endif
194 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
195 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
196 } else {
197 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
198 }
199
200 __ jmp(interpreter_entry, relocInfo::none);
201 // emit a sentinel we can test for when converting an interpreter
202 // entry point to a compiled entry point.
203 __ a_long(Interpreter::return_sentinel);
204 __ a_long((int)compiled_entry);
205 address entry = __ pc();
206 __ bind(interpreter_entry);
207
208 // In SSE mode, interpreter returns FP results in xmm0 but they need
209 // to end up back on the FPU so it can operate on them.
210 if (incoming_state == ftos && UseSSE >= 1) {
211 __ subptr(rsp, wordSize);
212 __ movflt(Address(rsp, 0), xmm0);
213 __ fld_s(Address(rsp, 0));
214 __ addptr(rsp, wordSize);
215 } else if (incoming_state == dtos && UseSSE >= 2) {
216 __ subptr(rsp, 2*wordSize);
217 __ movdbl(Address(rsp, 0), xmm0);
218 __ fld_d(Address(rsp, 0));
219 __ addptr(rsp, 2*wordSize);
220 }
221
222 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
223
224 // Restore stack bottom in case i2c adjusted stack
225 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
226 // and NULL it as marker that rsp is now tos until next java call
227 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
228
229 __ restore_bcp();
230 __ restore_locals();
231
232 Label L_got_cache, L_giant_index;
233 if (EnableInvokeDynamic) {
234 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
235 __ jcc(Assembler::equal, L_giant_index);
236 }
237 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
238 __ bind(L_got_cache);
239 __ movl(rbx, Address(rbx, rcx,
240 Address::times_ptr, constantPoolCacheOopDesc::base_offset() +
241 ConstantPoolCacheEntry::flags_offset()));
242 __ andptr(rbx, 0xFF);
243 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
244 __ dispatch_next(state, step);
245
246 // out of the main line of code...
247 if (EnableInvokeDynamic) {
248 __ bind(L_giant_index);
249 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
250 __ jmp(L_got_cache);
251 }
252
253 return entry;
254 }
255
256
257 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
258 address entry = __ pc();
259
260 // In SSE mode, FP results are in xmm0
261 if (state == ftos && UseSSE > 0) {
262 __ subptr(rsp, wordSize);
263 __ movflt(Address(rsp, 0), xmm0);
264 __ fld_s(Address(rsp, 0));
265 __ addptr(rsp, wordSize);
266 } else if (state == dtos && UseSSE >= 2) {
267 __ subptr(rsp, 2*wordSize);
268 __ movdbl(Address(rsp, 0), xmm0);
269 __ fld_d(Address(rsp, 0));
270 __ addptr(rsp, 2*wordSize);
271 }
272
273 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
274
275 // The stack is not extended by deopt but we must NULL last_sp as this
276 // entry is like a "return".
277 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
278 __ restore_bcp();
279 __ restore_locals();
280 // handle exceptions
281 { Label L;
282 const Register thread = rcx;
283 __ get_thread(thread);
284 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
285 __ jcc(Assembler::zero, L);
286 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
287 __ should_not_reach_here();
288 __ bind(L);
289 }
290 __ dispatch_next(state, step);
291 return entry;
292 }
293
294
295 int AbstractInterpreter::BasicType_as_index(BasicType type) {
296 int i = 0;
297 switch (type) {
298 case T_BOOLEAN: i = 0; break;
299 case T_CHAR : i = 1; break;
300 case T_BYTE : i = 2; break;
301 case T_SHORT : i = 3; break;
302 case T_INT : // fall through
303 case T_LONG : // fall through
304 case T_VOID : i = 4; break;
305 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
306 case T_DOUBLE : i = 6; break;
307 case T_OBJECT : // fall through
308 case T_ARRAY : i = 7; break;
309 default : ShouldNotReachHere();
310 }
311 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
312 return i;
313 }
314
315
316 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
317 address entry = __ pc();
318 switch (type) {
319 case T_BOOLEAN: __ c2bool(rax); break;
320 case T_CHAR : __ andptr(rax, 0xFFFF); break;
321 case T_BYTE : __ sign_extend_byte (rax); break;
322 case T_SHORT : __ sign_extend_short(rax); break;
323 case T_INT : /* nothing to do */ break;
324 case T_DOUBLE :
325 case T_FLOAT :
326 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
327 __ pop(t); // remove return address first
328 // Must return a result for interpreter or compiler. In SSE
329 // mode, results are returned in xmm0 and the FPU stack must
330 // be empty.
331 if (type == T_FLOAT && UseSSE >= 1) {
332 // Load ST0
333 __ fld_d(Address(rsp, 0));
334 // Store as float and empty fpu stack
335 __ fstp_s(Address(rsp, 0));
336 // and reload
337 __ movflt(xmm0, Address(rsp, 0));
338 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
339 __ movdbl(xmm0, Address(rsp, 0));
340 } else {
341 // restore ST0
342 __ fld_d(Address(rsp, 0));
343 }
344 // and pop the temp
345 __ addptr(rsp, 2 * wordSize);
346 __ push(t); // restore return address
347 }
348 break;
349 case T_OBJECT :
350 // retrieve result from frame
351 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
352 // and verify it
353 __ verify_oop(rax);
354 break;
355 default : ShouldNotReachHere();
356 }
357 __ ret(0); // return from result handler
358 return entry;
359 }
360
361 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
362 address entry = __ pc();
363 __ push(state);
364 __ call_VM(noreg, runtime_entry);
365 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
366 return entry;
367 }
368
369
370 // Helpers for commoning out cases in the various type of method entries.
371 //
372
373 // increment invocation count & check for overflow
374 //
375 // Note: checking for negative value instead of overflow
376 // so we have a 'sticky' overflow test
377 //
378 // rbx,: method
379 // rcx: invocation counter
380 //
381 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
382 const Address invocation_counter(rbx, in_bytes(methodOopDesc::invocation_counter_offset()) +
383 in_bytes(InvocationCounter::counter_offset()));
384 // Note: In tiered we increment either counters in methodOop or in MDO depending if we're profiling or not.
385 if (TieredCompilation) {
386 int increment = InvocationCounter::count_increment;
387 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
388 Label no_mdo, done;
389 if (ProfileInterpreter) {
390 // Are we profiling?
391 __ movptr(rax, Address(rbx, methodOopDesc::method_data_offset()));
392 __ testptr(rax, rax);
393 __ jccb(Assembler::zero, no_mdo);
394 // Increment counter in the MDO
395 const Address mdo_invocation_counter(rax, in_bytes(methodDataOopDesc::invocation_counter_offset()) +
396 in_bytes(InvocationCounter::counter_offset()));
397 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
398 __ jmpb(done);
399 }
400 __ bind(no_mdo);
401 // Increment counter in methodOop (we don't need to load it, it's in rcx).
402 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
403 __ bind(done);
404 } else {
405 const Address backedge_counter (rbx, methodOopDesc::backedge_counter_offset() +
406 InvocationCounter::counter_offset());
407
408 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
409 __ incrementl(Address(rbx,methodOopDesc::interpreter_invocation_counter_offset()));
410 }
411 // Update standard invocation counters
412 __ movl(rax, backedge_counter); // load backedge counter
413
414 __ incrementl(rcx, InvocationCounter::count_increment);
415 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
416
417 __ movl(invocation_counter, rcx); // save invocation count
418 __ addl(rcx, rax); // add both counters
419
420 // profile_method is non-null only for interpreted method so
421 // profile_method != NULL == !native_call
422 // BytecodeInterpreter only calls for native so code is elided.
423
424 if (ProfileInterpreter && profile_method != NULL) {
425 // Test to see if we should create a method data oop
426 __ cmp32(rcx,
427 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
428 __ jcc(Assembler::less, *profile_method_continue);
429
430 // if no method data exists, go to profile_method
431 __ test_method_data_pointer(rax, *profile_method);
432 }
433
434 __ cmp32(rcx,
435 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
436 __ jcc(Assembler::aboveEqual, *overflow);
437 }
438 }
439
440 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
441
442 // Asm interpreter on entry
443 // rdi - locals
444 // rsi - bcp
445 // rbx, - method
446 // rdx - cpool
447 // rbp, - interpreter frame
448
449 // C++ interpreter on entry
450 // rsi - new interpreter state pointer
451 // rbp - interpreter frame pointer
452 // rbx - method
453
454 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
455 // rbx, - method
456 // rcx - rcvr (assuming there is one)
457 // top of stack return address of interpreter caller
458 // rsp - sender_sp
459
460 // C++ interpreter only
461 // rsi - previous interpreter state pointer
462
463 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
464
465 // InterpreterRuntime::frequency_counter_overflow takes one argument
466 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
467 // The call returns the address of the verified entry point for the method or NULL
468 // if the compilation did not complete (either went background or bailed out).
469 __ movptr(rax, (intptr_t)false);
470 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
471
472 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
473
474 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
475 // and jump to the interpreted entry.
476 __ jmp(*do_continue, relocInfo::none);
477
478 }
479
480 void InterpreterGenerator::generate_stack_overflow_check(void) {
481 // see if we've got enough room on the stack for locals plus overhead.
482 // the expression stack grows down incrementally, so the normal guard
483 // page mechanism will work for that.
484 //
485 // Registers live on entry:
486 //
487 // Asm interpreter
488 // rdx: number of additional locals this frame needs (what we must check)
489 // rbx,: methodOop
490
491 // destroyed on exit
492 // rax,
493
494 // NOTE: since the additional locals are also always pushed (wasn't obvious in
495 // generate_method_entry) so the guard should work for them too.
496 //
497
498 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
499 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
500
501 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
502 // be sure to change this if you add/subtract anything to/from the overhead area
503 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
504
505 const int page_size = os::vm_page_size();
506
507 Label after_frame_check;
508
509 // see if the frame is greater than one page in size. If so,
510 // then we need to verify there is enough stack space remaining
511 // for the additional locals.
512 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
513 __ jcc(Assembler::belowEqual, after_frame_check);
514
515 // compute rsp as if this were going to be the last frame on
516 // the stack before the red zone
517
518 Label after_frame_check_pop;
519
520 __ push(rsi);
521
522 const Register thread = rsi;
523
524 __ get_thread(thread);
525
526 const Address stack_base(thread, Thread::stack_base_offset());
527 const Address stack_size(thread, Thread::stack_size_offset());
528
529 // locals + overhead, in bytes
530 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
531
532 #ifdef ASSERT
533 Label stack_base_okay, stack_size_okay;
534 // verify that thread stack base is non-zero
535 __ cmpptr(stack_base, (int32_t)NULL_WORD);
536 __ jcc(Assembler::notEqual, stack_base_okay);
537 __ stop("stack base is zero");
538 __ bind(stack_base_okay);
539 // verify that thread stack size is non-zero
540 __ cmpptr(stack_size, 0);
541 __ jcc(Assembler::notEqual, stack_size_okay);
542 __ stop("stack size is zero");
543 __ bind(stack_size_okay);
544 #endif
545
546 // Add stack base to locals and subtract stack size
547 __ addptr(rax, stack_base);
548 __ subptr(rax, stack_size);
549
550 // Use the maximum number of pages we might bang.
551 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
552 (StackRedPages+StackYellowPages);
553 __ addptr(rax, max_pages * page_size);
554
555 // check against the current stack bottom
556 __ cmpptr(rsp, rax);
557 __ jcc(Assembler::above, after_frame_check_pop);
558
559 __ pop(rsi); // get saved bcp / (c++ prev state ).
560
561 __ pop(rax); // get return address
562 __ jump(ExternalAddress(Interpreter::throw_StackOverflowError_entry()));
563
564 // all done with frame size check
565 __ bind(after_frame_check_pop);
566 __ pop(rsi);
567
568 __ bind(after_frame_check);
569 }
570
571 // Allocate monitor and lock method (asm interpreter)
572 // rbx, - methodOop
573 //
574 void InterpreterGenerator::lock_method(void) {
575 // synchronize method
576 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
577 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
578 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
579
580 #ifdef ASSERT
581 { Label L;
582 __ movl(rax, access_flags);
583 __ testl(rax, JVM_ACC_SYNCHRONIZED);
584 __ jcc(Assembler::notZero, L);
585 __ stop("method doesn't need synchronization");
586 __ bind(L);
587 }
588 #endif // ASSERT
589 // get synchronization object
590 { Label done;
591 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
592 __ movl(rax, access_flags);
593 __ testl(rax, JVM_ACC_STATIC);
594 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
595 __ jcc(Assembler::zero, done);
596 __ movptr(rax, Address(rbx, methodOopDesc::constants_offset()));
597 __ movptr(rax, Address(rax, constantPoolOopDesc::pool_holder_offset_in_bytes()));
598 __ movptr(rax, Address(rax, mirror_offset));
599 __ bind(done);
600 }
601 // add space for monitor & lock
602 __ subptr(rsp, entry_size); // add space for a monitor entry
603 __ movptr(monitor_block_top, rsp); // set new monitor block top
604 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
605 __ mov(rdx, rsp); // object address
606 __ lock_object(rdx);
607 }
608
609 //
610 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
611 // and for native methods hence the shared code.
612
613 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
614 // initialize fixed part of activation frame
615 __ push(rax); // save return address
616 __ enter(); // save old & set new rbp,
617
618
619 __ push(rsi); // set sender sp
620 __ push((int32_t)NULL_WORD); // leave last_sp as null
621 __ movptr(rsi, Address(rbx,methodOopDesc::const_offset())); // get constMethodOop
622 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
623 __ push(rbx); // save methodOop
624 if (ProfileInterpreter) {
625 Label method_data_continue;
626 __ movptr(rdx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
627 __ testptr(rdx, rdx);
628 __ jcc(Assembler::zero, method_data_continue);
629 __ addptr(rdx, in_bytes(methodDataOopDesc::data_offset()));
630 __ bind(method_data_continue);
631 __ push(rdx); // set the mdp (method data pointer)
632 } else {
633 __ push(0);
634 }
635
636 __ movptr(rdx, Address(rbx, methodOopDesc::constants_offset()));
637 __ movptr(rdx, Address(rdx, constantPoolOopDesc::cache_offset_in_bytes()));
638 __ push(rdx); // set constant pool cache
639 __ push(rdi); // set locals pointer
640 if (native_call) {
641 __ push(0); // no bcp
642 } else {
643 __ push(rsi); // set bcp
644 }
645 __ push(0); // reserve word for pointer to expression stack bottom
646 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
647 }
648
649 // End of helpers
650
651 //
652 // Various method entries
653 //------------------------------------------------------------------------------------------------------------------------
654 //
655 //
656
657 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
658
659 address InterpreterGenerator::generate_accessor_entry(void) {
660
661 // rbx,: methodOop
662 // rcx: receiver (preserve for slow entry into asm interpreter)
663
664 // rsi: senderSP must preserved for slow path, set SP to it on fast path
665
666 address entry_point = __ pc();
667 Label xreturn_path;
668
669 // do fastpath for resolved accessor methods
670 if (UseFastAccessorMethods) {
671 Label slow_path;
672 // If we need a safepoint check, generate full interpreter entry.
673 ExternalAddress state(SafepointSynchronize::address_of_state());
674 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
675 SafepointSynchronize::_not_synchronized);
676
677 __ jcc(Assembler::notEqual, slow_path);
678 // ASM/C++ Interpreter
679 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
680 // Note: We can only use this code if the getfield has been resolved
681 // and if we don't have a null-pointer exception => check for
682 // these conditions first and use slow path if necessary.
683 // rbx,: method
684 // rcx: receiver
685 __ movptr(rax, Address(rsp, wordSize));
686
687 // check if local 0 != NULL and read field
688 __ testptr(rax, rax);
689 __ jcc(Assembler::zero, slow_path);
690
691 __ movptr(rdi, Address(rbx, methodOopDesc::constants_offset()));
692 // read first instruction word and extract bytecode @ 1 and index @ 2
693 __ movptr(rdx, Address(rbx, methodOopDesc::const_offset()));
694 __ movl(rdx, Address(rdx, constMethodOopDesc::codes_offset()));
695 // Shift codes right to get the index on the right.
696 // The bytecode fetched looks like <index><0xb4><0x2a>
697 __ shrl(rdx, 2*BitsPerByte);
698 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
699 __ movptr(rdi, Address(rdi, constantPoolOopDesc::cache_offset_in_bytes()));
700
701 // rax,: local 0
702 // rbx,: method
703 // rcx: receiver - do not destroy since it is needed for slow path!
704 // rcx: scratch
705 // rdx: constant pool cache index
706 // rdi: constant pool cache
707 // rsi: sender sp
708
709 // check if getfield has been resolved and read constant pool cache entry
710 // check the validity of the cache entry by testing whether _indices field
711 // contains Bytecode::_getfield in b1 byte.
712 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
713 __ movl(rcx,
714 Address(rdi,
715 rdx,
716 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
717 __ shrl(rcx, 2*BitsPerByte);
718 __ andl(rcx, 0xFF);
719 __ cmpl(rcx, Bytecodes::_getfield);
720 __ jcc(Assembler::notEqual, slow_path);
721
722 // Note: constant pool entry is not valid before bytecode is resolved
723 __ movptr(rcx,
724 Address(rdi,
725 rdx,
726 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset()));
727 __ movl(rdx,
728 Address(rdi,
729 rdx,
730 Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::flags_offset()));
731
732 Label notByte, notShort, notChar;
733 const Address field_address (rax, rcx, Address::times_1);
734
735 // Need to differentiate between igetfield, agetfield, bgetfield etc.
736 // because they are different sizes.
737 // Use the type from the constant pool cache
738 __ shrl(rdx, ConstantPoolCacheEntry::tosBits);
739 // Make sure we don't need to mask rdx for tosBits after the above shift
740 ConstantPoolCacheEntry::verify_tosBits();
741 __ cmpl(rdx, btos);
742 __ jcc(Assembler::notEqual, notByte);
743 __ load_signed_byte(rax, field_address);
744 __ jmp(xreturn_path);
745
746 __ bind(notByte);
747 __ cmpl(rdx, stos);
748 __ jcc(Assembler::notEqual, notShort);
749 __ load_signed_short(rax, field_address);
750 __ jmp(xreturn_path);
751
752 __ bind(notShort);
753 __ cmpl(rdx, ctos);
754 __ jcc(Assembler::notEqual, notChar);
755 __ load_unsigned_short(rax, field_address);
756 __ jmp(xreturn_path);
757
758 __ bind(notChar);
759 #ifdef ASSERT
760 Label okay;
761 __ cmpl(rdx, atos);
762 __ jcc(Assembler::equal, okay);
763 __ cmpl(rdx, itos);
764 __ jcc(Assembler::equal, okay);
765 __ stop("what type is this?");
766 __ bind(okay);
767 #endif // ASSERT
768 // All the rest are a 32 bit wordsize
769 // This is ok for now. Since fast accessors should be going away
770 __ movptr(rax, field_address);
771
772 __ bind(xreturn_path);
773
774 // _ireturn/_areturn
775 __ pop(rdi); // get return address
776 __ mov(rsp, rsi); // set sp to sender sp
777 __ jmp(rdi);
778
779 // generate a vanilla interpreter entry as the slow path
780 __ bind(slow_path);
781
782 (void) generate_normal_entry(false);
783 return entry_point;
784 }
785 return NULL;
786
787 }
788
789 //
790 // Interpreter stub for calling a native method. (asm interpreter)
791 // This sets up a somewhat different looking stack for calling the native method
792 // than the typical interpreter frame setup.
793 //
794
795 address InterpreterGenerator::generate_native_entry(bool synchronized) {
796 // determine code generation flags
797 bool inc_counter = UseCompiler || CountCompiledCalls;
798
799 // rbx,: methodOop
800 // rsi: sender sp
801 // rsi: previous interpreter state (C++ interpreter) must preserve
802 address entry_point = __ pc();
803
804
805 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
806 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
807 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
808
809 // get parameter size (always needed)
810 __ load_unsigned_short(rcx, size_of_parameters);
811
812 // native calls don't need the stack size check since they have no expression stack
813 // and the arguments are already on the stack and we only add a handful of words
814 // to the stack
815
816 // rbx,: methodOop
817 // rcx: size of parameters
818 // rsi: sender sp
819
820 __ pop(rax); // get return address
821 // for natives the size of locals is zero
822
823 // compute beginning of parameters (rdi)
824 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
825
826
827 // add 2 zero-initialized slots for native calls
828 // NULL result handler
829 __ push((int32_t)NULL_WORD);
830 // NULL oop temp (mirror or jni oop result)
831 __ push((int32_t)NULL_WORD);
832
833 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
834 // initialize fixed part of activation frame
835
836 generate_fixed_frame(true);
837
838 // make sure method is native & not abstract
839 #ifdef ASSERT
840 __ movl(rax, access_flags);
841 {
842 Label L;
843 __ testl(rax, JVM_ACC_NATIVE);
844 __ jcc(Assembler::notZero, L);
845 __ stop("tried to execute non-native method as native");
846 __ bind(L);
847 }
848 { Label L;
849 __ testl(rax, JVM_ACC_ABSTRACT);
850 __ jcc(Assembler::zero, L);
851 __ stop("tried to execute abstract method in interpreter");
852 __ bind(L);
853 }
854 #endif
855
856 // Since at this point in the method invocation the exception handler
857 // would try to exit the monitor of synchronized methods which hasn't
858 // been entered yet, we set the thread local variable
859 // _do_not_unlock_if_synchronized to true. The remove_activation will
860 // check this flag.
861
862 __ get_thread(rax);
863 const Address do_not_unlock_if_synchronized(rax,
864 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
865 __ movbool(do_not_unlock_if_synchronized, true);
866
867 // increment invocation count & check for overflow
868 Label invocation_counter_overflow;
869 if (inc_counter) {
870 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
871 }
872
873 Label continue_after_compile;
874 __ bind(continue_after_compile);
875
876 bang_stack_shadow_pages(true);
877
878 // reset the _do_not_unlock_if_synchronized flag
879 __ get_thread(rax);
880 __ movbool(do_not_unlock_if_synchronized, false);
881
882 // check for synchronized methods
883 // Must happen AFTER invocation_counter check and stack overflow check,
884 // so method is not locked if overflows.
885 //
886 if (synchronized) {
887 lock_method();
888 } else {
889 // no synchronization necessary
890 #ifdef ASSERT
891 { Label L;
892 __ movl(rax, access_flags);
893 __ testl(rax, JVM_ACC_SYNCHRONIZED);
894 __ jcc(Assembler::zero, L);
895 __ stop("method needs synchronization");
896 __ bind(L);
897 }
898 #endif
899 }
900
901 // start execution
902 #ifdef ASSERT
903 { Label L;
904 const Address monitor_block_top (rbp,
905 frame::interpreter_frame_monitor_block_top_offset * wordSize);
906 __ movptr(rax, monitor_block_top);
907 __ cmpptr(rax, rsp);
908 __ jcc(Assembler::equal, L);
909 __ stop("broken stack frame setup in interpreter");
910 __ bind(L);
911 }
912 #endif
913
914 // jvmti/dtrace support
915 __ notify_method_entry();
916
917 // work registers
918 const Register method = rbx;
919 const Register thread = rdi;
920 const Register t = rcx;
921
922 // allocate space for parameters
923 __ get_method(method);
924 __ verify_oop(method);
925 __ load_unsigned_short(t, Address(method, methodOopDesc::size_of_parameters_offset()));
926 __ shlptr(t, Interpreter::logStackElementSize);
927 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
928 __ subptr(rsp, t);
929 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
930
931 // get signature handler
932 { Label L;
933 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
934 __ testptr(t, t);
935 __ jcc(Assembler::notZero, L);
936 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
937 __ get_method(method);
938 __ movptr(t, Address(method, methodOopDesc::signature_handler_offset()));
939 __ bind(L);
940 }
941
942 // call signature handler
943 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
944 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
945 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
946 // The generated handlers do not touch RBX (the method oop).
947 // However, large signatures cannot be cached and are generated
948 // each time here. The slow-path generator will blow RBX
949 // sometime, so we must reload it after the call.
950 __ call(t);
951 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
952
953 // result handler is in rax,
954 // set result handler
955 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
956
957 // pass mirror handle if static call
958 { Label L;
959 const int mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
960 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
961 __ testl(t, JVM_ACC_STATIC);
962 __ jcc(Assembler::zero, L);
963 // get mirror
964 __ movptr(t, Address(method, methodOopDesc:: constants_offset()));
965 __ movptr(t, Address(t, constantPoolOopDesc::pool_holder_offset_in_bytes()));
966 __ movptr(t, Address(t, mirror_offset));
967 // copy mirror into activation frame
968 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
969 // pass handle to mirror
970 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
971 __ movptr(Address(rsp, wordSize), t);
972 __ bind(L);
973 }
974
975 // get native function entry point
976 { Label L;
977 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
978 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
979 __ cmpptr(rax, unsatisfied.addr());
980 __ jcc(Assembler::notEqual, L);
981 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
982 __ get_method(method);
983 __ verify_oop(method);
984 __ movptr(rax, Address(method, methodOopDesc::native_function_offset()));
985 __ bind(L);
986 }
987
988 // pass JNIEnv
989 __ get_thread(thread);
990 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
991 __ movptr(Address(rsp, 0), t);
992
993 // set_last_Java_frame_before_call
994 // It is enough that the pc()
995 // points into the right code segment. It does not have to be the correct return pc.
996 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
997
998 // change thread state
999 #ifdef ASSERT
1000 { Label L;
1001 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1002 __ cmpl(t, _thread_in_Java);
1003 __ jcc(Assembler::equal, L);
1004 __ stop("Wrong thread state in native stub");
1005 __ bind(L);
1006 }
1007 #endif
1008
1009 // Change state to native
1010 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1011 __ call(rax);
1012
1013 // result potentially in rdx:rax or ST0
1014
1015 // Either restore the MXCSR register after returning from the JNI Call
1016 // or verify that it wasn't changed.
1017 if (VM_Version::supports_sse()) {
1018 if (RestoreMXCSROnJNICalls) {
1019 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1020 }
1021 else if (CheckJNICalls ) {
1022 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1023 }
1024 }
1025
1026 // Either restore the x87 floating pointer control word after returning
1027 // from the JNI call or verify that it wasn't changed.
1028 if (CheckJNICalls) {
1029 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1030 }
1031
1032 // save potential result in ST(0) & rdx:rax
1033 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1034 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1035 // It is safe to do this push because state is _thread_in_native and return address will be found
1036 // via _last_native_pc and not via _last_jave_sp
1037
1038 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1039 // If the order changes or anything else is added to the stack the code in
1040 // interpreter_frame_result will have to be changed.
1041
1042 { Label L;
1043 Label push_double;
1044 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1045 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1046 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1047 float_handler.addr());
1048 __ jcc(Assembler::equal, push_double);
1049 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1050 double_handler.addr());
1051 __ jcc(Assembler::notEqual, L);
1052 __ bind(push_double);
1053 __ push(dtos);
1054 __ bind(L);
1055 }
1056 __ push(ltos);
1057
1058 // change thread state
1059 __ get_thread(thread);
1060 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1061 if(os::is_MP()) {
1062 if (UseMembar) {
1063 // Force this write out before the read below
1064 __ membar(Assembler::Membar_mask_bits(
1065 Assembler::LoadLoad | Assembler::LoadStore |
1066 Assembler::StoreLoad | Assembler::StoreStore));
1067 } else {
1068 // Write serialization page so VM thread can do a pseudo remote membar.
1069 // We use the current thread pointer to calculate a thread specific
1070 // offset to write to within the page. This minimizes bus traffic
1071 // due to cache line collision.
1072 __ serialize_memory(thread, rcx);
1073 }
1074 }
1075
1076 if (AlwaysRestoreFPU) {
1077 // Make sure the control word is correct.
1078 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1079 }
1080
1081 // check for safepoint operation in progress and/or pending suspend requests
1082 { Label Continue;
1083
1084 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1085 SafepointSynchronize::_not_synchronized);
1086
1087 Label L;
1088 __ jcc(Assembler::notEqual, L);
1089 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1090 __ jcc(Assembler::equal, Continue);
1091 __ bind(L);
1092
1093 // Don't use call_VM as it will see a possible pending exception and forward it
1094 // and never return here preventing us from clearing _last_native_pc down below.
1095 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1096 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1097 // by hand.
1098 //
1099 __ push(thread);
1100 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1101 JavaThread::check_special_condition_for_native_trans)));
1102 __ increment(rsp, wordSize);
1103 __ get_thread(thread);
1104
1105 __ bind(Continue);
1106 }
1107
1108 // change thread state
1109 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1110
1111 __ reset_last_Java_frame(thread, true, true);
1112
1113 // reset handle block
1114 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1115 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1116
1117 // If result was an oop then unbox and save it in the frame
1118 { Label L;
1119 Label no_oop, store_result;
1120 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1121 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1122 handler.addr());
1123 __ jcc(Assembler::notEqual, no_oop);
1124 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1125 __ pop(ltos);
1126 __ testptr(rax, rax);
1127 __ jcc(Assembler::zero, store_result);
1128 // unbox
1129 __ movptr(rax, Address(rax, 0));
1130 __ bind(store_result);
1131 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1132 // keep stack depth as expected by pushing oop which will eventually be discarded
1133 __ push(ltos);
1134 __ bind(no_oop);
1135 }
1136
1137 {
1138 Label no_reguard;
1139 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1140 __ jcc(Assembler::notEqual, no_reguard);
1141
1142 __ pusha();
1143 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1144 __ popa();
1145
1146 __ bind(no_reguard);
1147 }
1148
1149 // restore rsi to have legal interpreter frame,
1150 // i.e., bci == 0 <=> rsi == code_base()
1151 // Can't call_VM until bcp is within reasonable.
1152 __ get_method(method); // method is junk from thread_in_native to now.
1153 __ verify_oop(method);
1154 __ movptr(rsi, Address(method,methodOopDesc::const_offset())); // get constMethodOop
1155 __ lea(rsi, Address(rsi,constMethodOopDesc::codes_offset())); // get codebase
1156
1157 // handle exceptions (exception handling will handle unlocking!)
1158 { Label L;
1159 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1160 __ jcc(Assembler::zero, L);
1161 // Note: At some point we may want to unify this with the code used in call_VM_base();
1162 // i.e., we should use the StubRoutines::forward_exception code. For now this
1163 // doesn't work here because the rsp is not correctly set at this point.
1164 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1165 __ should_not_reach_here();
1166 __ bind(L);
1167 }
1168
1169 // do unlocking if necessary
1170 { Label L;
1171 __ movl(t, Address(method, methodOopDesc::access_flags_offset()));
1172 __ testl(t, JVM_ACC_SYNCHRONIZED);
1173 __ jcc(Assembler::zero, L);
1174 // the code below should be shared with interpreter macro assembler implementation
1175 { Label unlock;
1176 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1177 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1178 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1179
1180 __ lea(rdx, monitor); // address of first monitor
1181
1182 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1183 __ testptr(t, t);
1184 __ jcc(Assembler::notZero, unlock);
1185
1186 // Entry already unlocked, need to throw exception
1187 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1188 __ should_not_reach_here();
1189
1190 __ bind(unlock);
1191 __ unlock_object(rdx);
1192 }
1193 __ bind(L);
1194 }
1195
1196 // jvmti/dtrace support
1197 // Note: This must happen _after_ handling/throwing any exceptions since
1198 // the exception handler code notifies the runtime of method exits
1199 // too. If this happens before, method entry/exit notifications are
1200 // not properly paired (was bug - gri 11/22/99).
1201 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1202
1203 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1204 __ pop(ltos);
1205 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1206 __ call(t);
1207
1208 // remove activation
1209 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1210 __ leave(); // remove frame anchor
1211 __ pop(rdi); // get return address
1212 __ mov(rsp, t); // set sp to sender sp
1213 __ jmp(rdi);
1214
1215 if (inc_counter) {
1216 // Handle overflow of counter and compile method
1217 __ bind(invocation_counter_overflow);
1218 generate_counter_overflow(&continue_after_compile);
1219 }
1220
1221 return entry_point;
1222 }
1223
1224 //
1225 // Generic interpreted method entry to (asm) interpreter
1226 //
1227 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1228 // determine code generation flags
1229 bool inc_counter = UseCompiler || CountCompiledCalls;
1230
1231 // rbx,: methodOop
1232 // rsi: sender sp
1233 address entry_point = __ pc();
1234
1235
1236 const Address size_of_parameters(rbx, methodOopDesc::size_of_parameters_offset());
1237 const Address size_of_locals (rbx, methodOopDesc::size_of_locals_offset());
1238 const Address invocation_counter(rbx, methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset());
1239 const Address access_flags (rbx, methodOopDesc::access_flags_offset());
1240
1241 // get parameter size (always needed)
1242 __ load_unsigned_short(rcx, size_of_parameters);
1243
1244 // rbx,: methodOop
1245 // rcx: size of parameters
1246
1247 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1248
1249 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1250 __ subl(rdx, rcx); // rdx = no. of additional locals
1251
1252 // see if we've got enough room on the stack for locals plus overhead.
1253 generate_stack_overflow_check();
1254
1255 // get return address
1256 __ pop(rax);
1257
1258 // compute beginning of parameters (rdi)
1259 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1260
1261 // rdx - # of additional locals
1262 // allocate space for locals
1263 // explicitly initialize locals
1264 {
1265 Label exit, loop;
1266 __ testl(rdx, rdx);
1267 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1268 __ bind(loop);
1269 __ push((int32_t)NULL_WORD); // initialize local variables
1270 __ decrement(rdx); // until everything initialized
1271 __ jcc(Assembler::greater, loop);
1272 __ bind(exit);
1273 }
1274
1275 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1276 // initialize fixed part of activation frame
1277 generate_fixed_frame(false);
1278
1279 // make sure method is not native & not abstract
1280 #ifdef ASSERT
1281 __ movl(rax, access_flags);
1282 {
1283 Label L;
1284 __ testl(rax, JVM_ACC_NATIVE);
1285 __ jcc(Assembler::zero, L);
1286 __ stop("tried to execute native method as non-native");
1287 __ bind(L);
1288 }
1289 { Label L;
1290 __ testl(rax, JVM_ACC_ABSTRACT);
1291 __ jcc(Assembler::zero, L);
1292 __ stop("tried to execute abstract method in interpreter");
1293 __ bind(L);
1294 }
1295 #endif
1296
1297 // Since at this point in the method invocation the exception handler
1298 // would try to exit the monitor of synchronized methods which hasn't
1299 // been entered yet, we set the thread local variable
1300 // _do_not_unlock_if_synchronized to true. The remove_activation will
1301 // check this flag.
1302
1303 __ get_thread(rax);
1304 const Address do_not_unlock_if_synchronized(rax,
1305 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1306 __ movbool(do_not_unlock_if_synchronized, true);
1307
1308 // increment invocation count & check for overflow
1309 Label invocation_counter_overflow;
1310 Label profile_method;
1311 Label profile_method_continue;
1312 if (inc_counter) {
1313 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1314 if (ProfileInterpreter) {
1315 __ bind(profile_method_continue);
1316 }
1317 }
1318 Label continue_after_compile;
1319 __ bind(continue_after_compile);
1320
1321 bang_stack_shadow_pages(false);
1322
1323 // reset the _do_not_unlock_if_synchronized flag
1324 __ get_thread(rax);
1325 __ movbool(do_not_unlock_if_synchronized, false);
1326
1327 // check for synchronized methods
1328 // Must happen AFTER invocation_counter check and stack overflow check,
1329 // so method is not locked if overflows.
1330 //
1331 if (synchronized) {
1332 // Allocate monitor and lock method
1333 lock_method();
1334 } else {
1335 // no synchronization necessary
1336 #ifdef ASSERT
1337 { Label L;
1338 __ movl(rax, access_flags);
1339 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1340 __ jcc(Assembler::zero, L);
1341 __ stop("method needs synchronization");
1342 __ bind(L);
1343 }
1344 #endif
1345 }
1346
1347 // start execution
1348 #ifdef ASSERT
1349 { Label L;
1350 const Address monitor_block_top (rbp,
1351 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1352 __ movptr(rax, monitor_block_top);
1353 __ cmpptr(rax, rsp);
1354 __ jcc(Assembler::equal, L);
1355 __ stop("broken stack frame setup in interpreter");
1356 __ bind(L);
1357 }
1358 #endif
1359
1360 // jvmti support
1361 __ notify_method_entry();
1362
1363 __ dispatch_next(vtos);
1364
1365 // invocation counter overflow
1366 if (inc_counter) {
1367 if (ProfileInterpreter) {
1368 // We have decided to profile this method in the interpreter
1369 __ bind(profile_method);
1370
1371 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi, true);
1372
1373 __ movptr(rbx, Address(rbp, method_offset)); // restore methodOop
1374 __ movptr(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1375 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1376 __ test_method_data_pointer(rax, profile_method_continue);
1377 __ addptr(rax, in_bytes(methodDataOopDesc::data_offset()));
1378 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
1379 __ jmp(profile_method_continue);
1380 }
1381 // Handle overflow of counter and compile method
1382 __ bind(invocation_counter_overflow);
1383 generate_counter_overflow(&continue_after_compile);
1384 }
1385
1386 return entry_point;
1387 }
1388
1389 //------------------------------------------------------------------------------------------------------------------------
1390 // Entry points
1391 //
1392 // Here we generate the various kind of entries into the interpreter.
1393 // The two main entry type are generic bytecode methods and native call method.
1394 // These both come in synchronized and non-synchronized versions but the
1395 // frame layout they create is very similar. The other method entry
1396 // types are really just special purpose entries that are really entry
1397 // and interpretation all in one. These are for trivial methods like
1398 // accessor, empty, or special math methods.
1399 //
1400 // When control flow reaches any of the entry types for the interpreter
1401 // the following holds ->
1402 //
1403 // Arguments:
1404 //
1405 // rbx,: methodOop
1406 // rcx: receiver
1407 //
1408 //
1409 // Stack layout immediately at entry
1410 //
1411 // [ return address ] <--- rsp
1412 // [ parameter n ]
1413 // ...
1414 // [ parameter 1 ]
1415 // [ expression stack ] (caller's java expression stack)
1416
1417 // Assuming that we don't go to one of the trivial specialized
1418 // entries the stack will look like below when we are ready to execute
1419 // the first bytecode (or call the native routine). The register usage
1420 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1421 //
1422 // local variables follow incoming parameters immediately; i.e.
1423 // the return address is moved to the end of the locals).
1424 //
1425 // [ monitor entry ] <--- rsp
1426 // ...
1427 // [ monitor entry ]
1428 // [ expr. stack bottom ]
1429 // [ saved rsi ]
1430 // [ current rdi ]
1431 // [ methodOop ]
1432 // [ saved rbp, ] <--- rbp,
1433 // [ return address ]
1434 // [ local variable m ]
1435 // ...
1436 // [ local variable 1 ]
1437 // [ parameter n ]
1438 // ...
1439 // [ parameter 1 ] <--- rdi
1440
1441 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1442 // determine code generation flags
1443 bool synchronized = false;
1444 address entry_point = NULL;
1445
1446 switch (kind) {
1447 case Interpreter::zerolocals : break;
1448 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1449 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1450 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1451 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1452 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1453 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1454 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
1455
1456 case Interpreter::java_lang_math_sin : // fall thru
1457 case Interpreter::java_lang_math_cos : // fall thru
1458 case Interpreter::java_lang_math_tan : // fall thru
1459 case Interpreter::java_lang_math_abs : // fall thru
1460 case Interpreter::java_lang_math_log : // fall thru
1461 case Interpreter::java_lang_math_log10 : // fall thru
1462 case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1463 default : ShouldNotReachHere(); break;
1464 }
1465
1466 if (entry_point) return entry_point;
1467
1468 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1469
1470 }
1471
1472 // These should never be compiled since the interpreter will prefer
1473 // the compiled version to the intrinsic version.
1474 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1475 switch (method_kind(m)) {
1476 case Interpreter::java_lang_math_sin : // fall thru
1477 case Interpreter::java_lang_math_cos : // fall thru
1478 case Interpreter::java_lang_math_tan : // fall thru
1479 case Interpreter::java_lang_math_abs : // fall thru
1480 case Interpreter::java_lang_math_log : // fall thru
1481 case Interpreter::java_lang_math_log10 : // fall thru
1482 case Interpreter::java_lang_math_sqrt :
1483 return false;
1484 default:
1485 return true;
1486 }
1487 }
1488
1489 // How much stack a method activation needs in words.
1490 int AbstractInterpreter::size_top_interpreter_activation(methodOop method) {
1491
1492 const int stub_code = 4; // see generate_call_stub
1493 // Save space for one monitor to get into the interpreted method in case
1494 // the method is synchronized
1495 int monitor_size = method->is_synchronized() ?
1496 1*frame::interpreter_frame_monitor_size() : 0;
1497
1498 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1499 // be sure to change this if you add/subtract anything to/from the overhead area
1500 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1501
1502 const int extra_stack = methodOopDesc::extra_stack_entries();
1503 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1504 Interpreter::stackElementWords;
1505 return overhead_size + method_stack + stub_code;
1506 }
1507
1508 // asm based interpreter deoptimization helpers
1509
1510 int AbstractInterpreter::layout_activation(methodOop method,
1511 int tempcount,
1512 int popframe_extra_args,
1513 int moncount,
1514 int callee_param_count,
1515 int callee_locals,
1516 frame* caller,
1517 frame* interpreter_frame,
1518 bool is_top_frame) {
1519 // Note: This calculation must exactly parallel the frame setup
1520 // in AbstractInterpreterGenerator::generate_method_entry.
1521 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1522 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1523 // as determined by a previous call to this method.
1524 // It is also guaranteed to be walkable even though it is in a skeletal state
1525 // NOTE: return size is in words not bytes
1526
1527 // fixed size of an interpreter frame:
1528 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1529 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1530 Interpreter::stackElementWords;
1531
1532 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1533
1534 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1535 // Since the callee parameters already account for the callee's params we only need to account for
1536 // the extra locals.
1537
1538
1539 int size = overhead +
1540 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1541 (moncount*frame::interpreter_frame_monitor_size()) +
1542 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1543
1544 if (interpreter_frame != NULL) {
1545 #ifdef ASSERT
1546 if (!EnableMethodHandles)
1547 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1548 // Probably, since deoptimization doesn't work yet.
1549 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1550 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1551 #endif
1552
1553 interpreter_frame->interpreter_frame_set_method(method);
1554 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1555 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1556 // and sender_sp is fp+8
1557 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1558
1559 interpreter_frame->interpreter_frame_set_locals(locals);
1560 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1561 BasicObjectLock* monbot = montop - moncount;
1562 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1563
1564 // Set last_sp
1565 intptr_t* rsp = (intptr_t*) monbot -
1566 tempcount*Interpreter::stackElementWords -
1567 popframe_extra_args;
1568 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1569
1570 // All frames but the initial (oldest) interpreter frame we fill in have a
1571 // value for sender_sp that allows walking the stack but isn't
1572 // truly correct. Correct the value here.
1573
1574 if (extra_locals != 0 &&
1575 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1576 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1577 }
1578 *interpreter_frame->interpreter_frame_cache_addr() =
1579 method->constants()->cache();
1580 }
1581 return size;
1582 }
1583
1584
1585 //------------------------------------------------------------------------------------------------------------------------
1586 // Exceptions
1587
1588 void TemplateInterpreterGenerator::generate_throw_exception() {
1589 // Entry point in previous activation (i.e., if the caller was interpreted)
1590 Interpreter::_rethrow_exception_entry = __ pc();
1591 const Register thread = rcx;
1592
1593 // Restore sp to interpreter_frame_last_sp even though we are going
1594 // to empty the expression stack for the exception processing.
1595 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1596 // rax,: exception
1597 // rdx: return address/pc that threw exception
1598 __ restore_bcp(); // rsi points to call/send
1599 __ restore_locals();
1600
1601 // Entry point for exceptions thrown within interpreter code
1602 Interpreter::_throw_exception_entry = __ pc();
1603 // expression stack is undefined here
1604 // rax,: exception
1605 // rsi: exception bcp
1606 __ verify_oop(rax);
1607
1608 // expression stack must be empty before entering the VM in case of an exception
1609 __ empty_expression_stack();
1610 __ empty_FPU_stack();
1611 // find exception handler address and preserve exception oop
1612 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1613 // rax,: exception handler entry point
1614 // rdx: preserved exception oop
1615 // rsi: bcp for exception handler
1616 __ push_ptr(rdx); // push exception which is now the only value on the stack
1617 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1618
1619 // If the exception is not handled in the current frame the frame is removed and
1620 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1621 //
1622 // Note: At this point the bci is still the bxi for the instruction which caused
1623 // the exception and the expression stack is empty. Thus, for any VM calls
1624 // at this point, GC will find a legal oop map (with empty expression stack).
1625
1626 // In current activation
1627 // tos: exception
1628 // rsi: exception bcp
1629
1630 //
1631 // JVMTI PopFrame support
1632 //
1633
1634 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1635 __ empty_expression_stack();
1636 __ empty_FPU_stack();
1637 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1638 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1639 // popframe handling cycles.
1640 __ get_thread(thread);
1641 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1642 __ orl(rdx, JavaThread::popframe_processing_bit);
1643 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1644
1645 {
1646 // Check to see whether we are returning to a deoptimized frame.
1647 // (The PopFrame call ensures that the caller of the popped frame is
1648 // either interpreted or compiled and deoptimizes it if compiled.)
1649 // In this case, we can't call dispatch_next() after the frame is
1650 // popped, but instead must save the incoming arguments and restore
1651 // them after deoptimization has occurred.
1652 //
1653 // Note that we don't compare the return PC against the
1654 // deoptimization blob's unpack entry because of the presence of
1655 // adapter frames in C2.
1656 Label caller_not_deoptimized;
1657 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1658 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1659 __ testl(rax, rax);
1660 __ jcc(Assembler::notZero, caller_not_deoptimized);
1661
1662 // Compute size of arguments for saving when returning to deoptimized caller
1663 __ get_method(rax);
1664 __ verify_oop(rax);
1665 __ load_unsigned_short(rax, Address(rax, in_bytes(methodOopDesc::size_of_parameters_offset())));
1666 __ shlptr(rax, Interpreter::logStackElementSize);
1667 __ restore_locals();
1668 __ subptr(rdi, rax);
1669 __ addptr(rdi, wordSize);
1670 // Save these arguments
1671 __ get_thread(thread);
1672 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1673
1674 __ remove_activation(vtos, rdx,
1675 /* throw_monitor_exception */ false,
1676 /* install_monitor_exception */ false,
1677 /* notify_jvmdi */ false);
1678
1679 // Inform deoptimization that it is responsible for restoring these arguments
1680 __ get_thread(thread);
1681 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1682
1683 // Continue in deoptimization handler
1684 __ jmp(rdx);
1685
1686 __ bind(caller_not_deoptimized);
1687 }
1688
1689 __ remove_activation(vtos, rdx,
1690 /* throw_monitor_exception */ false,
1691 /* install_monitor_exception */ false,
1692 /* notify_jvmdi */ false);
1693
1694 // Finish with popframe handling
1695 // A previous I2C followed by a deoptimization might have moved the
1696 // outgoing arguments further up the stack. PopFrame expects the
1697 // mutations to those outgoing arguments to be preserved and other
1698 // constraints basically require this frame to look exactly as
1699 // though it had previously invoked an interpreted activation with
1700 // no space between the top of the expression stack (current
1701 // last_sp) and the top of stack. Rather than force deopt to
1702 // maintain this kind of invariant all the time we call a small
1703 // fixup routine to move the mutated arguments onto the top of our
1704 // expression stack if necessary.
1705 __ mov(rax, rsp);
1706 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1707 __ get_thread(thread);
1708 // PC must point into interpreter here
1709 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1710 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1711 __ get_thread(thread);
1712 __ reset_last_Java_frame(thread, true, true);
1713 // Restore the last_sp and null it out
1714 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1715 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1716
1717 __ restore_bcp();
1718 __ restore_locals();
1719 // The method data pointer was incremented already during
1720 // call profiling. We have to restore the mdp for the current bcp.
1721 if (ProfileInterpreter) {
1722 __ set_method_data_pointer_for_bcp();
1723 }
1724
1725 // Clear the popframe condition flag
1726 __ get_thread(thread);
1727 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1728
1729 __ dispatch_next(vtos);
1730 // end of PopFrame support
1731
1732 Interpreter::_remove_activation_entry = __ pc();
1733
1734 // preserve exception over this code sequence
1735 __ pop_ptr(rax);
1736 __ get_thread(thread);
1737 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1738 // remove the activation (without doing throws on illegalMonitorExceptions)
1739 __ remove_activation(vtos, rdx, false, true, false);
1740 // restore exception
1741 __ get_thread(thread);
1742 __ movptr(rax, Address(thread, JavaThread::vm_result_offset()));
1743 __ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
1744 __ verify_oop(rax);
1745
1746 // Inbetween activations - previous activation type unknown yet
1747 // compute continuation point - the continuation point expects
1748 // the following registers set up:
1749 //
1750 // rax: exception
1751 // rdx: return address/pc that threw exception
1752 // rsp: expression stack of caller
1753 // rbp: rbp, of caller
1754 __ push(rax); // save exception
1755 __ push(rdx); // save return address
1756 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1757 __ mov(rbx, rax); // save exception handler
1758 __ pop(rdx); // restore return address
1759 __ pop(rax); // restore exception
1760 // Note that an "issuing PC" is actually the next PC after the call
1761 __ jmp(rbx); // jump to exception handler of caller
1762 }
1763
1764
1765 //
1766 // JVMTI ForceEarlyReturn support
1767 //
1768 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1769 address entry = __ pc();
1770 const Register thread = rcx;
1771
1772 __ restore_bcp();
1773 __ restore_locals();
1774 __ empty_expression_stack();
1775 __ empty_FPU_stack();
1776 __ load_earlyret_value(state);
1777
1778 __ get_thread(thread);
1779 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1780 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1781
1782 // Clear the earlyret state
1783 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1784
1785 __ remove_activation(state, rsi,
1786 false, /* throw_monitor_exception */
1787 false, /* install_monitor_exception */
1788 true); /* notify_jvmdi */
1789 __ jmp(rsi);
1790 return entry;
1791 } // end of ForceEarlyReturn support
1792
1793
1794 //------------------------------------------------------------------------------------------------------------------------
1795 // Helper for vtos entry point generation
1796
1797 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) {
1798 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1799 Label L;
1800 fep = __ pc(); __ push(ftos); __ jmp(L);
1801 dep = __ pc(); __ push(dtos); __ jmp(L);
1802 lep = __ pc(); __ push(ltos); __ jmp(L);
1803 aep = __ pc(); __ push(atos); __ jmp(L);
1804 bep = cep = sep = // fall through
1805 iep = __ pc(); __ push(itos); // fall through
1806 vep = __ pc(); __ bind(L); // fall through
1807 generate_and_dispatch(t);
1808 }
1809
1810 //------------------------------------------------------------------------------------------------------------------------
1811 // Generation of individual instructions
1812
1813 // helpers for generate_and_dispatch
1814
1815
1816
1817 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1818 : TemplateInterpreterGenerator(code) {
1819 generate_all(); // down here so it can be "virtual"
1820 }
1821
1822 //------------------------------------------------------------------------------------------------------------------------
1823
1824 // Non-product code
1825 #ifndef PRODUCT
1826 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1827 address entry = __ pc();
1828
1829 // prepare expression stack
1830 __ pop(rcx); // pop return address so expression stack is 'pure'
1831 __ push(state); // save tosca
1832
1833 // pass tosca registers as arguments & call tracer
1834 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1835 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1836 __ pop(state); // restore tosca
1837
1838 // return
1839 __ jmp(rcx);
1840
1841 return entry;
1842 }
1843
1844
1845 void TemplateInterpreterGenerator::count_bytecode() {
1846 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1847 }
1848
1849
1850 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1851 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1852 }
1853
1854
1855 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1856 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1857 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1858 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1859 ExternalAddress table((address) BytecodePairHistogram::_counters);
1860 Address index(noreg, rbx, Address::times_4);
1861 __ incrementl(ArrayAddress(table, index));
1862 }
1863
1864
1865 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1866 // Call a little run-time stub to avoid blow-up for each bytecode.
1867 // The run-time runtime saves the right registers, depending on
1868 // the tosca in-state for the given template.
1869 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1870 "entry must have been generated");
1871 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1872 }
1873
1874
1875 void TemplateInterpreterGenerator::stop_interpreter_at() {
1876 Label L;
1877 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1878 StopInterpreterAt);
1879 __ jcc(Assembler::notEqual, L);
1880 __ int3();
1881 __ bind(L);
1882 }
1883 #endif // !PRODUCT
1884 #endif // CC_INTERP