1 /*
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "runtime/arguments.hpp"
39 #include "runtime/deoptimization.hpp"
40 #include "runtime/frame.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "runtime/synchronizer.hpp"
44 #include "runtime/timer.hpp"
45 #include "runtime/vframeArray.hpp"
46 #include "utilities/debug.hpp"
47 #include "utilities/macros.hpp"
48
49 #define __ _masm->
50
51
52 #ifndef CC_INTERP
53 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
54 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
55 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
56
57 //------------------------------------------------------------------------------------------------------------------------
58
59 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
60 address entry = __ pc();
61
62 // Note: There should be a minimal interpreter frame set up when stack
63 // overflow occurs since we check explicitly for it now.
64 //
65 #ifdef ASSERT
66 { Label L;
67 __ lea(rax, Address(rbp,
68 frame::interpreter_frame_monitor_block_top_offset * wordSize));
69 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
70 // (stack grows negative)
71 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
72 __ stop ("interpreter frame not set up");
73 __ bind(L);
74 }
75 #endif // ASSERT
76 // Restore bcp under the assumption that the current frame is still
77 // interpreted
78 __ restore_bcp();
79
80 // expression stack must be empty before entering the VM if an exception
81 // happened
82 __ empty_expression_stack();
83 __ empty_FPU_stack();
84 // throw exception
85 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
86 return entry;
87 }
88
89 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
90 address entry = __ pc();
91 // expression stack must be empty before entering the VM if an exception happened
92 __ empty_expression_stack();
93 __ empty_FPU_stack();
94 // setup parameters
95 // ??? convention: expect aberrant index in register rbx,
96 __ lea(rax, ExternalAddress((address)name));
97 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
98 return entry;
99 }
100
101 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
102 address entry = __ pc();
103 // object is at TOS
104 __ pop(rax);
105 // expression stack must be empty before entering the VM if an exception
106 // happened
107 __ empty_expression_stack();
108 __ empty_FPU_stack();
109 __ call_VM(noreg,
110 CAST_FROM_FN_PTR(address,
111 InterpreterRuntime::throw_ClassCastException),
112 rax);
113 return entry;
114 }
115
116 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
117 assert(!pass_oop || message == NULL, "either oop or message but not both");
118 address entry = __ pc();
119 if (pass_oop) {
120 // object is at TOS
121 __ pop(rbx);
122 }
123 // expression stack must be empty before entering the VM if an exception happened
124 __ empty_expression_stack();
125 __ empty_FPU_stack();
126 // setup parameters
127 __ lea(rax, ExternalAddress((address)name));
128 if (pass_oop) {
129 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
130 } else {
131 if (message != NULL) {
132 __ lea(rbx, ExternalAddress((address)message));
133 } else {
134 __ movptr(rbx, NULL_WORD);
135 }
136 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
137 }
138 // throw exception
139 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
140 return entry;
141 }
142
143
144 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
145 address entry = __ pc();
146 // NULL last_sp until next java call
147 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
148 __ dispatch_next(state);
149 return entry;
150 }
151
152
153 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step) {
154 TosState incoming_state = state;
155 address entry = __ pc();
156
157 #ifdef COMPILER2
158 // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
159 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
160 for (int i = 1; i < 8; i++) {
161 __ ffree(i);
162 }
163 } else if (UseSSE < 2) {
164 __ empty_FPU_stack();
165 }
166 #endif
167 if ((incoming_state == ftos && UseSSE < 1) || (incoming_state == dtos && UseSSE < 2)) {
168 __ MacroAssembler::verify_FPU(1, "generate_return_entry_for compiled");
169 } else {
170 __ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
171 }
172
173 // In SSE mode, interpreter returns FP results in xmm0 but they need
174 // to end up back on the FPU so it can operate on them.
175 if (incoming_state == ftos && UseSSE >= 1) {
176 __ subptr(rsp, wordSize);
177 __ movflt(Address(rsp, 0), xmm0);
178 __ fld_s(Address(rsp, 0));
179 __ addptr(rsp, wordSize);
180 } else if (incoming_state == dtos && UseSSE >= 2) {
181 __ subptr(rsp, 2*wordSize);
182 __ movdbl(Address(rsp, 0), xmm0);
183 __ fld_d(Address(rsp, 0));
184 __ addptr(rsp, 2*wordSize);
185 }
186
187 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_return_entry_for in interpreter");
188
189 // Restore stack bottom in case i2c adjusted stack
190 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
191 // and NULL it as marker that rsp is now tos until next java call
192 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
193
194 __ restore_bcp();
195 __ restore_locals();
196
197 Label L_got_cache, L_giant_index;
198 if (EnableInvokeDynamic) {
199 __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
200 __ jcc(Assembler::equal, L_giant_index);
201 }
202 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u2));
203 __ bind(L_got_cache);
204 __ movl(rbx, Address(rbx, rcx,
205 Address::times_ptr, ConstantPoolCache::base_offset() +
206 ConstantPoolCacheEntry::flags_offset()));
207 __ andptr(rbx, 0xFF);
208 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale()));
209 __ dispatch_next(state, step);
210
211 // out of the main line of code...
212 if (EnableInvokeDynamic) {
213 __ bind(L_giant_index);
214 __ get_cache_and_index_at_bcp(rbx, rcx, 1, sizeof(u4));
215 __ jmp(L_got_cache);
216 }
217
218 return entry;
219 }
220
221
222 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
223 address entry = __ pc();
224
225 // In SSE mode, FP results are in xmm0
226 if (state == ftos && UseSSE > 0) {
227 __ subptr(rsp, wordSize);
228 __ movflt(Address(rsp, 0), xmm0);
229 __ fld_s(Address(rsp, 0));
230 __ addptr(rsp, wordSize);
231 } else if (state == dtos && UseSSE >= 2) {
232 __ subptr(rsp, 2*wordSize);
233 __ movdbl(Address(rsp, 0), xmm0);
234 __ fld_d(Address(rsp, 0));
235 __ addptr(rsp, 2*wordSize);
236 }
237
238 __ MacroAssembler::verify_FPU(state == ftos || state == dtos ? 1 : 0, "generate_deopt_entry_for in interpreter");
239
240 // The stack is not extended by deopt but we must NULL last_sp as this
241 // entry is like a "return".
242 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
243 __ restore_bcp();
244 __ restore_locals();
245 // handle exceptions
246 { Label L;
247 const Register thread = rcx;
248 __ get_thread(thread);
249 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
250 __ jcc(Assembler::zero, L);
251 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
252 __ should_not_reach_here();
253 __ bind(L);
254 }
255 __ dispatch_next(state, step);
256 return entry;
257 }
258
259
260 int AbstractInterpreter::BasicType_as_index(BasicType type) {
261 int i = 0;
262 switch (type) {
263 case T_BOOLEAN: i = 0; break;
264 case T_CHAR : i = 1; break;
265 case T_BYTE : i = 2; break;
266 case T_SHORT : i = 3; break;
267 case T_INT : // fall through
268 case T_LONG : // fall through
269 case T_VOID : i = 4; break;
270 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
271 case T_DOUBLE : i = 6; break;
272 case T_OBJECT : // fall through
273 case T_ARRAY : i = 7; break;
274 default : ShouldNotReachHere();
275 }
276 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
277 return i;
278 }
279
280
281 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
282 address entry = __ pc();
283 switch (type) {
284 case T_BOOLEAN: __ c2bool(rax); break;
285 case T_CHAR : __ andptr(rax, 0xFFFF); break;
286 case T_BYTE : __ sign_extend_byte (rax); break;
287 case T_SHORT : __ sign_extend_short(rax); break;
288 case T_INT : /* nothing to do */ break;
289 case T_DOUBLE :
290 case T_FLOAT :
291 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
292 __ pop(t); // remove return address first
293 // Must return a result for interpreter or compiler. In SSE
294 // mode, results are returned in xmm0 and the FPU stack must
295 // be empty.
296 if (type == T_FLOAT && UseSSE >= 1) {
297 // Load ST0
298 __ fld_d(Address(rsp, 0));
299 // Store as float and empty fpu stack
300 __ fstp_s(Address(rsp, 0));
301 // and reload
302 __ movflt(xmm0, Address(rsp, 0));
303 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
304 __ movdbl(xmm0, Address(rsp, 0));
305 } else {
306 // restore ST0
307 __ fld_d(Address(rsp, 0));
308 }
309 // and pop the temp
310 __ addptr(rsp, 2 * wordSize);
311 __ push(t); // restore return address
312 }
313 break;
314 case T_OBJECT :
315 // retrieve result from frame
316 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
317 // and verify it
318 __ verify_oop(rax);
319 break;
320 default : ShouldNotReachHere();
321 }
322 __ ret(0); // return from result handler
323 return entry;
324 }
325
326 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
327 address entry = __ pc();
328 __ push(state);
329 __ call_VM(noreg, runtime_entry);
330 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
331 return entry;
332 }
333
334
335 // Helpers for commoning out cases in the various type of method entries.
336 //
337
338 // increment invocation count & check for overflow
339 //
340 // Note: checking for negative value instead of overflow
341 // so we have a 'sticky' overflow test
342 //
343 // rbx,: method
344 // rcx: invocation counter
345 //
346 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
347 const Address invocation_counter(rbx, in_bytes(Method::invocation_counter_offset()) +
348 in_bytes(InvocationCounter::counter_offset()));
349 // Note: In tiered we increment either counters in Method* or in MDO depending if we're profiling or not.
350 if (TieredCompilation) {
351 int increment = InvocationCounter::count_increment;
352 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
353 Label no_mdo, done;
354 if (ProfileInterpreter) {
355 // Are we profiling?
356 __ movptr(rax, Address(rbx, Method::method_data_offset()));
357 __ testptr(rax, rax);
358 __ jccb(Assembler::zero, no_mdo);
359 // Increment counter in the MDO
360 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
361 in_bytes(InvocationCounter::counter_offset()));
362 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
363 __ jmpb(done);
364 }
365 __ bind(no_mdo);
366 // Increment counter in Method* (we don't need to load it, it's in rcx).
367 __ increment_mask_and_jump(invocation_counter, increment, mask, rcx, true, Assembler::zero, overflow);
368 __ bind(done);
369 } else {
370 const Address backedge_counter (rbx, Method::backedge_counter_offset() +
371 InvocationCounter::counter_offset());
372
373 if (ProfileInterpreter) { // %%% Merge this into MethodData*
374 __ incrementl(Address(rbx,Method::interpreter_invocation_counter_offset()));
375 }
376 // Update standard invocation counters
377 __ movl(rax, backedge_counter); // load backedge counter
378
379 __ incrementl(rcx, InvocationCounter::count_increment);
380 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
381
382 __ movl(invocation_counter, rcx); // save invocation count
383 __ addl(rcx, rax); // add both counters
384
385 // profile_method is non-null only for interpreted method so
386 // profile_method != NULL == !native_call
387 // BytecodeInterpreter only calls for native so code is elided.
388
389 if (ProfileInterpreter && profile_method != NULL) {
390 // Test to see if we should create a method data oop
391 __ cmp32(rcx,
392 ExternalAddress((address)&InvocationCounter::InterpreterProfileLimit));
393 __ jcc(Assembler::less, *profile_method_continue);
394
395 // if no method data exists, go to profile_method
396 __ test_method_data_pointer(rax, *profile_method);
397 }
398
399 __ cmp32(rcx,
400 ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
401 __ jcc(Assembler::aboveEqual, *overflow);
402 }
403 }
404
405 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
406
407 // Asm interpreter on entry
408 // rdi - locals
409 // rsi - bcp
410 // rbx, - method
411 // rdx - cpool
412 // rbp, - interpreter frame
413
414 // C++ interpreter on entry
415 // rsi - new interpreter state pointer
416 // rbp - interpreter frame pointer
417 // rbx - method
418
419 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
420 // rbx, - method
421 // rcx - rcvr (assuming there is one)
422 // top of stack return address of interpreter caller
423 // rsp - sender_sp
424
425 // C++ interpreter only
426 // rsi - previous interpreter state pointer
427
428 // InterpreterRuntime::frequency_counter_overflow takes one argument
429 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
430 // The call returns the address of the verified entry point for the method or NULL
431 // if the compilation did not complete (either went background or bailed out).
432 __ movptr(rax, (intptr_t)false);
433 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
434
435 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
436
437 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
438 // and jump to the interpreted entry.
439 __ jmp(*do_continue, relocInfo::none);
440
441 }
442
443 void InterpreterGenerator::generate_stack_overflow_check(void) {
444 // see if we've got enough room on the stack for locals plus overhead.
445 // the expression stack grows down incrementally, so the normal guard
446 // page mechanism will work for that.
447 //
448 // Registers live on entry:
449 //
450 // Asm interpreter
451 // rdx: number of additional locals this frame needs (what we must check)
452 // rbx,: Method*
453
454 // destroyed on exit
455 // rax,
456
457 // NOTE: since the additional locals are also always pushed (wasn't obvious in
458 // generate_method_entry) so the guard should work for them too.
459 //
460
461 // monitor entry size: see picture of stack set (generate_method_entry) and frame_x86.hpp
462 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
463
464 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
465 // be sure to change this if you add/subtract anything to/from the overhead area
466 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
467
468 const int page_size = os::vm_page_size();
469
470 Label after_frame_check;
471
472 // see if the frame is greater than one page in size. If so,
473 // then we need to verify there is enough stack space remaining
474 // for the additional locals.
475 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
476 __ jcc(Assembler::belowEqual, after_frame_check);
477
478 // compute rsp as if this were going to be the last frame on
479 // the stack before the red zone
480
481 Label after_frame_check_pop;
482
483 __ push(rsi);
484
485 const Register thread = rsi;
486
487 __ get_thread(thread);
488
489 const Address stack_base(thread, Thread::stack_base_offset());
490 const Address stack_size(thread, Thread::stack_size_offset());
491
492 // locals + overhead, in bytes
493 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
494
495 #ifdef ASSERT
496 Label stack_base_okay, stack_size_okay;
497 // verify that thread stack base is non-zero
498 __ cmpptr(stack_base, (int32_t)NULL_WORD);
499 __ jcc(Assembler::notEqual, stack_base_okay);
500 __ stop("stack base is zero");
501 __ bind(stack_base_okay);
502 // verify that thread stack size is non-zero
503 __ cmpptr(stack_size, 0);
504 __ jcc(Assembler::notEqual, stack_size_okay);
505 __ stop("stack size is zero");
506 __ bind(stack_size_okay);
507 #endif
508
509 // Add stack base to locals and subtract stack size
510 __ addptr(rax, stack_base);
511 __ subptr(rax, stack_size);
512
513 // Use the maximum number of pages we might bang.
514 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
515 (StackRedPages+StackYellowPages);
516 __ addptr(rax, max_pages * page_size);
517
518 // check against the current stack bottom
519 __ cmpptr(rsp, rax);
520 __ jcc(Assembler::above, after_frame_check_pop);
521
522 __ pop(rsi); // get saved bcp / (c++ prev state ).
523
524 // Restore sender's sp as SP. This is necessary if the sender's
525 // frame is an extended compiled frame (see gen_c2i_adapter())
526 // and safer anyway in case of JSR292 adaptations.
527
528 __ pop(rax); // return address must be moved if SP is changed
529 __ mov(rsp, rsi);
530 __ push(rax);
531
532 // Note: the restored frame is not necessarily interpreted.
533 // Use the shared runtime version of the StackOverflowError.
534 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
535 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
536 // all done with frame size check
537 __ bind(after_frame_check_pop);
538 __ pop(rsi);
539
540 __ bind(after_frame_check);
541 }
542
543 // Allocate monitor and lock method (asm interpreter)
544 // rbx, - Method*
545 //
546 void InterpreterGenerator::lock_method(void) {
547 // synchronize method
548 const Address access_flags (rbx, Method::access_flags_offset());
549 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
550 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
551
552 #ifdef ASSERT
553 { Label L;
554 __ movl(rax, access_flags);
555 __ testl(rax, JVM_ACC_SYNCHRONIZED);
556 __ jcc(Assembler::notZero, L);
557 __ stop("method doesn't need synchronization");
558 __ bind(L);
559 }
560 #endif // ASSERT
561 // get synchronization object
562 { Label done;
563 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
564 __ movl(rax, access_flags);
565 __ testl(rax, JVM_ACC_STATIC);
566 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
567 __ jcc(Assembler::zero, done);
568 __ movptr(rax, Address(rbx, Method::const_offset()));
569 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
570 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
571 __ movptr(rax, Address(rax, mirror_offset));
572 __ bind(done);
573 }
574 // add space for monitor & lock
575 __ subptr(rsp, entry_size); // add space for a monitor entry
576 __ movptr(monitor_block_top, rsp); // set new monitor block top
577 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
578 __ mov(rdx, rsp); // object address
579 __ lock_object(rdx);
580 }
581
582 //
583 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
584 // and for native methods hence the shared code.
585
586 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
587 // initialize fixed part of activation frame
588 __ push(rax); // save return address
589 __ enter(); // save old & set new rbp,
590
591
592 __ push(rsi); // set sender sp
593 __ push((int32_t)NULL_WORD); // leave last_sp as null
594 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
595 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
596 __ push(rbx); // save Method*
597 if (ProfileInterpreter) {
598 Label method_data_continue;
599 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
600 __ testptr(rdx, rdx);
601 __ jcc(Assembler::zero, method_data_continue);
602 __ addptr(rdx, in_bytes(MethodData::data_offset()));
603 __ bind(method_data_continue);
604 __ push(rdx); // set the mdp (method data pointer)
605 } else {
606 __ push(0);
607 }
608
609 __ movptr(rdx, Address(rbx, Method::const_offset()));
610 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
611 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
612 __ push(rdx); // set constant pool cache
613 __ push(rdi); // set locals pointer
614 if (native_call) {
615 __ push(0); // no bcp
616 } else {
617 __ push(rsi); // set bcp
618 }
619 __ push(0); // reserve word for pointer to expression stack bottom
620 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
621 }
622
623 // End of helpers
624
625 //
626 // Various method entries
627 //------------------------------------------------------------------------------------------------------------------------
628 //
629 //
630
631 // Call an accessor method (assuming it is resolved, otherwise drop into vanilla (slow path) entry
632
633 address InterpreterGenerator::generate_accessor_entry(void) {
634
635 // rbx,: Method*
636 // rcx: receiver (preserve for slow entry into asm interpreter)
637
638 // rsi: senderSP must preserved for slow path, set SP to it on fast path
639
640 address entry_point = __ pc();
641 Label xreturn_path;
642
643 // do fastpath for resolved accessor methods
644 if (UseFastAccessorMethods) {
645 Label slow_path;
646 // If we need a safepoint check, generate full interpreter entry.
647 ExternalAddress state(SafepointSynchronize::address_of_state());
648 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
649 SafepointSynchronize::_not_synchronized);
650
651 __ jcc(Assembler::notEqual, slow_path);
652 // ASM/C++ Interpreter
653 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof; parameter size = 1
654 // Note: We can only use this code if the getfield has been resolved
655 // and if we don't have a null-pointer exception => check for
656 // these conditions first and use slow path if necessary.
657 // rbx,: method
658 // rcx: receiver
659 __ movptr(rax, Address(rsp, wordSize));
660
661 // check if local 0 != NULL and read field
662 __ testptr(rax, rax);
663 __ jcc(Assembler::zero, slow_path);
664
665 // read first instruction word and extract bytecode @ 1 and index @ 2
666 __ movptr(rdx, Address(rbx, Method::const_offset()));
667 __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
668 __ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
669 // Shift codes right to get the index on the right.
670 // The bytecode fetched looks like <index><0xb4><0x2a>
671 __ shrl(rdx, 2*BitsPerByte);
672 __ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
673 __ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
674
675 // rax,: local 0
676 // rbx,: method
677 // rcx: receiver - do not destroy since it is needed for slow path!
678 // rcx: scratch
679 // rdx: constant pool cache index
680 // rdi: constant pool cache
681 // rsi: sender sp
682
683 // check if getfield has been resolved and read constant pool cache entry
684 // check the validity of the cache entry by testing whether _indices field
685 // contains Bytecode::_getfield in b1 byte.
686 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
687 __ movl(rcx,
688 Address(rdi,
689 rdx,
690 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
691 __ shrl(rcx, 2*BitsPerByte);
692 __ andl(rcx, 0xFF);
693 __ cmpl(rcx, Bytecodes::_getfield);
694 __ jcc(Assembler::notEqual, slow_path);
695
696 // Note: constant pool entry is not valid before bytecode is resolved
697 __ movptr(rcx,
698 Address(rdi,
699 rdx,
700 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
701 __ movl(rdx,
702 Address(rdi,
703 rdx,
704 Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
705
706 Label notByte, notShort, notChar;
707 const Address field_address (rax, rcx, Address::times_1);
708
709 // Need to differentiate between igetfield, agetfield, bgetfield etc.
710 // because they are different sizes.
711 // Use the type from the constant pool cache
712 __ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
713 // Make sure we don't need to mask rdx after the above shift
714 ConstantPoolCacheEntry::verify_tos_state_shift();
715 __ cmpl(rdx, btos);
716 __ jcc(Assembler::notEqual, notByte);
717 __ load_signed_byte(rax, field_address);
718 __ jmp(xreturn_path);
719
720 __ bind(notByte);
721 __ cmpl(rdx, stos);
722 __ jcc(Assembler::notEqual, notShort);
723 __ load_signed_short(rax, field_address);
724 __ jmp(xreturn_path);
725
726 __ bind(notShort);
727 __ cmpl(rdx, ctos);
728 __ jcc(Assembler::notEqual, notChar);
729 __ load_unsigned_short(rax, field_address);
730 __ jmp(xreturn_path);
731
732 __ bind(notChar);
733 #ifdef ASSERT
734 Label okay;
735 __ cmpl(rdx, atos);
736 __ jcc(Assembler::equal, okay);
737 __ cmpl(rdx, itos);
738 __ jcc(Assembler::equal, okay);
739 __ stop("what type is this?");
740 __ bind(okay);
741 #endif // ASSERT
742 // All the rest are a 32 bit wordsize
743 // This is ok for now. Since fast accessors should be going away
744 __ movptr(rax, field_address);
745
746 __ bind(xreturn_path);
747
748 // _ireturn/_areturn
749 __ pop(rdi); // get return address
750 __ mov(rsp, rsi); // set sp to sender sp
751 __ jmp(rdi);
752
753 // generate a vanilla interpreter entry as the slow path
754 __ bind(slow_path);
755
756 (void) generate_normal_entry(false);
757 return entry_point;
758 }
759 return NULL;
760
761 }
762
763 // Method entry for java.lang.ref.Reference.get.
764 address InterpreterGenerator::generate_Reference_get_entry(void) {
765 #if INCLUDE_ALL_GCS
766 // Code: _aload_0, _getfield, _areturn
767 // parameter size = 1
768 //
769 // The code that gets generated by this routine is split into 2 parts:
770 // 1. The "intrinsified" code for G1 (or any SATB based GC),
771 // 2. The slow path - which is an expansion of the regular method entry.
772 //
773 // Notes:-
774 // * In the G1 code we do not check whether we need to block for
775 // a safepoint. If G1 is enabled then we must execute the specialized
776 // code for Reference.get (except when the Reference object is null)
777 // so that we can log the value in the referent field with an SATB
778 // update buffer.
779 // If the code for the getfield template is modified so that the
780 // G1 pre-barrier code is executed when the current method is
781 // Reference.get() then going through the normal method entry
782 // will be fine.
783 // * The G1 code below can, however, check the receiver object (the instance
784 // of java.lang.Reference) and jump to the slow path if null. If the
785 // Reference object is null then we obviously cannot fetch the referent
786 // and so we don't need to call the G1 pre-barrier. Thus we can use the
787 // regular method entry code to generate the NPE.
788 //
789 // This code is based on generate_accessor_enty.
790
791 // rbx,: Method*
792 // rcx: receiver (preserve for slow entry into asm interpreter)
793
794 // rsi: senderSP must preserved for slow path, set SP to it on fast path
795
796 address entry = __ pc();
797
798 const int referent_offset = java_lang_ref_Reference::referent_offset;
799 guarantee(referent_offset > 0, "referent offset not initialized");
800
801 if (UseG1GC) {
802 Label slow_path;
803
804 // Check if local 0 != NULL
805 // If the receiver is null then it is OK to jump to the slow path.
806 __ movptr(rax, Address(rsp, wordSize));
807 __ testptr(rax, rax);
808 __ jcc(Assembler::zero, slow_path);
809
810 // rax: local 0 (must be preserved across the G1 barrier call)
811 //
812 // rbx: method (at this point it's scratch)
813 // rcx: receiver (at this point it's scratch)
814 // rdx: scratch
815 // rdi: scratch
816 //
817 // rsi: sender sp
818
819 // Preserve the sender sp in case the pre-barrier
820 // calls the runtime
821 __ push(rsi);
822
823 // Load the value of the referent field.
824 const Address field_address(rax, referent_offset);
825 __ movptr(rax, field_address);
826
827 // Generate the G1 pre-barrier code to log the value of
828 // the referent field in an SATB buffer.
829 __ get_thread(rcx);
830 __ g1_write_barrier_pre(noreg /* obj */,
831 rax /* pre_val */,
832 rcx /* thread */,
833 rbx /* tmp */,
834 true /* tosca_save */,
835 true /* expand_call */);
836
837 // _areturn
838 __ pop(rsi); // get sender sp
839 __ pop(rdi); // get return address
840 __ mov(rsp, rsi); // set sp to sender sp
841 __ jmp(rdi);
842
843 __ bind(slow_path);
844 (void) generate_normal_entry(false);
845
846 return entry;
847 }
848 #endif // INCLUDE_ALL_GCS
849
850 // If G1 is not enabled then attempt to go through the accessor entry point
851 // Reference.get is an accessor
852 return generate_accessor_entry();
853 }
854
855 //
856 // Interpreter stub for calling a native method. (asm interpreter)
857 // This sets up a somewhat different looking stack for calling the native method
858 // than the typical interpreter frame setup.
859 //
860
861 address InterpreterGenerator::generate_native_entry(bool synchronized) {
862 // determine code generation flags
863 bool inc_counter = UseCompiler || CountCompiledCalls;
864
865 // rbx,: Method*
866 // rsi: sender sp
867 // rsi: previous interpreter state (C++ interpreter) must preserve
868 address entry_point = __ pc();
869
870 const Address constMethod (rbx, Method::const_offset());
871 const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
872 const Address access_flags (rbx, Method::access_flags_offset());
873 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
874
875 // get parameter size (always needed)
876 __ movptr(rcx, constMethod);
877 __ load_unsigned_short(rcx, size_of_parameters);
878
879 // native calls don't need the stack size check since they have no expression stack
880 // and the arguments are already on the stack and we only add a handful of words
881 // to the stack
882
883 // rbx,: Method*
884 // rcx: size of parameters
885 // rsi: sender sp
886
887 __ pop(rax); // get return address
888 // for natives the size of locals is zero
889
890 // compute beginning of parameters (rdi)
891 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
892
893
894 // add 2 zero-initialized slots for native calls
895 // NULL result handler
896 __ push((int32_t)NULL_WORD);
897 // NULL oop temp (mirror or jni oop result)
898 __ push((int32_t)NULL_WORD);
899
900 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
901 // initialize fixed part of activation frame
902
903 generate_fixed_frame(true);
904
905 // make sure method is native & not abstract
906 #ifdef ASSERT
907 __ movl(rax, access_flags);
908 {
909 Label L;
910 __ testl(rax, JVM_ACC_NATIVE);
911 __ jcc(Assembler::notZero, L);
912 __ stop("tried to execute non-native method as native");
913 __ bind(L);
914 }
915 { Label L;
916 __ testl(rax, JVM_ACC_ABSTRACT);
917 __ jcc(Assembler::zero, L);
918 __ stop("tried to execute abstract method in interpreter");
919 __ bind(L);
920 }
921 #endif
922
923 // Since at this point in the method invocation the exception handler
924 // would try to exit the monitor of synchronized methods which hasn't
925 // been entered yet, we set the thread local variable
926 // _do_not_unlock_if_synchronized to true. The remove_activation will
927 // check this flag.
928
929 __ get_thread(rax);
930 const Address do_not_unlock_if_synchronized(rax,
931 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
932 __ movbool(do_not_unlock_if_synchronized, true);
933
934 // increment invocation count & check for overflow
935 Label invocation_counter_overflow;
936 if (inc_counter) {
937 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
938 }
939
940 Label continue_after_compile;
941 __ bind(continue_after_compile);
942
943 bang_stack_shadow_pages(true);
944
945 // reset the _do_not_unlock_if_synchronized flag
946 __ get_thread(rax);
947 __ movbool(do_not_unlock_if_synchronized, false);
948
949 // check for synchronized methods
950 // Must happen AFTER invocation_counter check and stack overflow check,
951 // so method is not locked if overflows.
952 //
953 if (synchronized) {
954 lock_method();
955 } else {
956 // no synchronization necessary
957 #ifdef ASSERT
958 { Label L;
959 __ movl(rax, access_flags);
960 __ testl(rax, JVM_ACC_SYNCHRONIZED);
961 __ jcc(Assembler::zero, L);
962 __ stop("method needs synchronization");
963 __ bind(L);
964 }
965 #endif
966 }
967
968 // start execution
969 #ifdef ASSERT
970 { Label L;
971 const Address monitor_block_top (rbp,
972 frame::interpreter_frame_monitor_block_top_offset * wordSize);
973 __ movptr(rax, monitor_block_top);
974 __ cmpptr(rax, rsp);
975 __ jcc(Assembler::equal, L);
976 __ stop("broken stack frame setup in interpreter");
977 __ bind(L);
978 }
979 #endif
980
981 // jvmti/dtrace support
982 __ notify_method_entry();
983
984 // work registers
985 const Register method = rbx;
986 const Register thread = rdi;
987 const Register t = rcx;
988
989 // allocate space for parameters
990 __ get_method(method);
991 __ movptr(t, Address(method, Method::const_offset()));
992 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
993
994 __ shlptr(t, Interpreter::logStackElementSize);
995 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
996 __ subptr(rsp, t);
997 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
998
999 // get signature handler
1000 { Label L;
1001 __ movptr(t, Address(method, Method::signature_handler_offset()));
1002 __ testptr(t, t);
1003 __ jcc(Assembler::notZero, L);
1004 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1005 __ get_method(method);
1006 __ movptr(t, Address(method, Method::signature_handler_offset()));
1007 __ bind(L);
1008 }
1009
1010 // call signature handler
1011 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1012 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1013 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1014 // The generated handlers do not touch RBX (the method oop).
1015 // However, large signatures cannot be cached and are generated
1016 // each time here. The slow-path generator will blow RBX
1017 // sometime, so we must reload it after the call.
1018 __ call(t);
1019 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1020
1021 // result handler is in rax,
1022 // set result handler
1023 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1024
1025 // pass mirror handle if static call
1026 { Label L;
1027 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1028 __ movl(t, Address(method, Method::access_flags_offset()));
1029 __ testl(t, JVM_ACC_STATIC);
1030 __ jcc(Assembler::zero, L);
1031 // get mirror
1032 __ movptr(t, Address(method, Method:: const_offset()));
1033 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1034 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1035 __ movptr(t, Address(t, mirror_offset));
1036 // copy mirror into activation frame
1037 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1038 // pass handle to mirror
1039 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1040 __ movptr(Address(rsp, wordSize), t);
1041 __ bind(L);
1042 }
1043
1044 // get native function entry point
1045 { Label L;
1046 __ movptr(rax, Address(method, Method::native_function_offset()));
1047 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1048 __ cmpptr(rax, unsatisfied.addr());
1049 __ jcc(Assembler::notEqual, L);
1050 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1051 __ get_method(method);
1052 __ movptr(rax, Address(method, Method::native_function_offset()));
1053 __ bind(L);
1054 }
1055
1056 // pass JNIEnv
1057 __ get_thread(thread);
1058 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1059 __ movptr(Address(rsp, 0), t);
1060
1061 // set_last_Java_frame_before_call
1062 // It is enough that the pc()
1063 // points into the right code segment. It does not have to be the correct return pc.
1064 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1065
1066 // change thread state
1067 #ifdef ASSERT
1068 { Label L;
1069 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1070 __ cmpl(t, _thread_in_Java);
1071 __ jcc(Assembler::equal, L);
1072 __ stop("Wrong thread state in native stub");
1073 __ bind(L);
1074 }
1075 #endif
1076
1077 // Change state to native
1078 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1079 __ call(rax);
1080
1081 // result potentially in rdx:rax or ST0
1082
1083 // Either restore the MXCSR register after returning from the JNI Call
1084 // or verify that it wasn't changed.
1085 if (VM_Version::supports_sse()) {
1086 if (RestoreMXCSROnJNICalls) {
1087 __ ldmxcsr(ExternalAddress(StubRoutines::addr_mxcsr_std()));
1088 }
1089 else if (CheckJNICalls ) {
1090 __ call(RuntimeAddress(StubRoutines::x86::verify_mxcsr_entry()));
1091 }
1092 }
1093
1094 // Either restore the x87 floating pointer control word after returning
1095 // from the JNI call or verify that it wasn't changed.
1096 if (CheckJNICalls) {
1097 __ call(RuntimeAddress(StubRoutines::x86::verify_fpu_cntrl_wrd_entry()));
1098 }
1099
1100 // save potential result in ST(0) & rdx:rax
1101 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1102 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1103 // It is safe to do this push because state is _thread_in_native and return address will be found
1104 // via _last_native_pc and not via _last_jave_sp
1105
1106 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1107 // If the order changes or anything else is added to the stack the code in
1108 // interpreter_frame_result will have to be changed.
1109
1110 { Label L;
1111 Label push_double;
1112 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1113 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1114 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1115 float_handler.addr());
1116 __ jcc(Assembler::equal, push_double);
1117 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1118 double_handler.addr());
1119 __ jcc(Assembler::notEqual, L);
1120 __ bind(push_double);
1121 __ push(dtos);
1122 __ bind(L);
1123 }
1124 __ push(ltos);
1125
1126 // change thread state
1127 __ get_thread(thread);
1128 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1129 if(os::is_MP()) {
1130 if (UseMembar) {
1131 // Force this write out before the read below
1132 __ membar(Assembler::Membar_mask_bits(
1133 Assembler::LoadLoad | Assembler::LoadStore |
1134 Assembler::StoreLoad | Assembler::StoreStore));
1135 } else {
1136 // Write serialization page so VM thread can do a pseudo remote membar.
1137 // We use the current thread pointer to calculate a thread specific
1138 // offset to write to within the page. This minimizes bus traffic
1139 // due to cache line collision.
1140 __ serialize_memory(thread, rcx);
1141 }
1142 }
1143
1144 if (AlwaysRestoreFPU) {
1145 // Make sure the control word is correct.
1146 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1147 }
1148
1149 // check for safepoint operation in progress and/or pending suspend requests
1150 { Label Continue;
1151
1152 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1153 SafepointSynchronize::_not_synchronized);
1154
1155 Label L;
1156 __ jcc(Assembler::notEqual, L);
1157 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1158 __ jcc(Assembler::equal, Continue);
1159 __ bind(L);
1160
1161 // Don't use call_VM as it will see a possible pending exception and forward it
1162 // and never return here preventing us from clearing _last_native_pc down below.
1163 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1164 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1165 // by hand.
1166 //
1167 __ push(thread);
1168 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1169 JavaThread::check_special_condition_for_native_trans)));
1170 __ increment(rsp, wordSize);
1171 __ get_thread(thread);
1172
1173 __ bind(Continue);
1174 }
1175
1176 // change thread state
1177 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1178
1179 __ reset_last_Java_frame(thread, true, true);
1180
1181 // reset handle block
1182 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1183 __ movptr(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1184
1185 // If result was an oop then unbox and save it in the frame
1186 { Label L;
1187 Label no_oop, store_result;
1188 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1189 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1190 handler.addr());
1191 __ jcc(Assembler::notEqual, no_oop);
1192 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1193 __ pop(ltos);
1194 __ testptr(rax, rax);
1195 __ jcc(Assembler::zero, store_result);
1196 // unbox
1197 __ movptr(rax, Address(rax, 0));
1198 __ bind(store_result);
1199 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1200 // keep stack depth as expected by pushing oop which will eventually be discarded
1201 __ push(ltos);
1202 __ bind(no_oop);
1203 }
1204
1205 {
1206 Label no_reguard;
1207 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1208 __ jcc(Assembler::notEqual, no_reguard);
1209
1210 __ pusha();
1211 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1212 __ popa();
1213
1214 __ bind(no_reguard);
1215 }
1216
1217 // restore rsi to have legal interpreter frame,
1218 // i.e., bci == 0 <=> rsi == code_base()
1219 // Can't call_VM until bcp is within reasonable.
1220 __ get_method(method); // method is junk from thread_in_native to now.
1221 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1222 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1223
1224 // handle exceptions (exception handling will handle unlocking!)
1225 { Label L;
1226 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1227 __ jcc(Assembler::zero, L);
1228 // Note: At some point we may want to unify this with the code used in call_VM_base();
1229 // i.e., we should use the StubRoutines::forward_exception code. For now this
1230 // doesn't work here because the rsp is not correctly set at this point.
1231 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1232 __ should_not_reach_here();
1233 __ bind(L);
1234 }
1235
1236 // do unlocking if necessary
1237 { Label L;
1238 __ movl(t, Address(method, Method::access_flags_offset()));
1239 __ testl(t, JVM_ACC_SYNCHRONIZED);
1240 __ jcc(Assembler::zero, L);
1241 // the code below should be shared with interpreter macro assembler implementation
1242 { Label unlock;
1243 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1244 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1245 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1246
1247 __ lea(rdx, monitor); // address of first monitor
1248
1249 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1250 __ testptr(t, t);
1251 __ jcc(Assembler::notZero, unlock);
1252
1253 // Entry already unlocked, need to throw exception
1254 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1255 __ should_not_reach_here();
1256
1257 __ bind(unlock);
1258 __ unlock_object(rdx);
1259 }
1260 __ bind(L);
1261 }
1262
1263 // jvmti/dtrace support
1264 // Note: This must happen _after_ handling/throwing any exceptions since
1265 // the exception handler code notifies the runtime of method exits
1266 // too. If this happens before, method entry/exit notifications are
1267 // not properly paired (was bug - gri 11/22/99).
1268 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1269
1270 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1271 __ pop(ltos);
1272 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1273 __ call(t);
1274
1275 // remove activation
1276 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1277 __ leave(); // remove frame anchor
1278 __ pop(rdi); // get return address
1279 __ mov(rsp, t); // set sp to sender sp
1280 __ jmp(rdi);
1281
1282 if (inc_counter) {
1283 // Handle overflow of counter and compile method
1284 __ bind(invocation_counter_overflow);
1285 generate_counter_overflow(&continue_after_compile);
1286 }
1287
1288 return entry_point;
1289 }
1290
1291 //
1292 // Generic interpreted method entry to (asm) interpreter
1293 //
1294 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1295 // determine code generation flags
1296 bool inc_counter = UseCompiler || CountCompiledCalls;
1297
1298 // rbx,: Method*
1299 // rsi: sender sp
1300 address entry_point = __ pc();
1301
1302 const Address constMethod (rbx, Method::const_offset());
1303 const Address invocation_counter(rbx, Method::invocation_counter_offset() + InvocationCounter::counter_offset());
1304 const Address access_flags (rbx, Method::access_flags_offset());
1305 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1306 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1307
1308 // get parameter size (always needed)
1309 __ movptr(rdx, constMethod);
1310 __ load_unsigned_short(rcx, size_of_parameters);
1311
1312 // rbx,: Method*
1313 // rcx: size of parameters
1314
1315 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1316
1317 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1318 __ subl(rdx, rcx); // rdx = no. of additional locals
1319
1320 // see if we've got enough room on the stack for locals plus overhead.
1321 generate_stack_overflow_check();
1322
1323 // get return address
1324 __ pop(rax);
1325
1326 // compute beginning of parameters (rdi)
1327 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1328
1329 // rdx - # of additional locals
1330 // allocate space for locals
1331 // explicitly initialize locals
1332 {
1333 Label exit, loop;
1334 __ testl(rdx, rdx);
1335 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1336 __ bind(loop);
1337 __ push((int32_t)NULL_WORD); // initialize local variables
1338 __ decrement(rdx); // until everything initialized
1339 __ jcc(Assembler::greater, loop);
1340 __ bind(exit);
1341 }
1342
1343 if (inc_counter) __ movl(rcx, invocation_counter); // (pre-)fetch invocation count
1344 // initialize fixed part of activation frame
1345 generate_fixed_frame(false);
1346
1347 // make sure method is not native & not abstract
1348 #ifdef ASSERT
1349 __ movl(rax, access_flags);
1350 {
1351 Label L;
1352 __ testl(rax, JVM_ACC_NATIVE);
1353 __ jcc(Assembler::zero, L);
1354 __ stop("tried to execute native method as non-native");
1355 __ bind(L);
1356 }
1357 { Label L;
1358 __ testl(rax, JVM_ACC_ABSTRACT);
1359 __ jcc(Assembler::zero, L);
1360 __ stop("tried to execute abstract method in interpreter");
1361 __ bind(L);
1362 }
1363 #endif
1364
1365 // Since at this point in the method invocation the exception handler
1366 // would try to exit the monitor of synchronized methods which hasn't
1367 // been entered yet, we set the thread local variable
1368 // _do_not_unlock_if_synchronized to true. The remove_activation will
1369 // check this flag.
1370
1371 __ get_thread(rax);
1372 const Address do_not_unlock_if_synchronized(rax,
1373 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1374 __ movbool(do_not_unlock_if_synchronized, true);
1375
1376 // increment invocation count & check for overflow
1377 Label invocation_counter_overflow;
1378 Label profile_method;
1379 Label profile_method_continue;
1380 if (inc_counter) {
1381 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1382 if (ProfileInterpreter) {
1383 __ bind(profile_method_continue);
1384 }
1385 }
1386 Label continue_after_compile;
1387 __ bind(continue_after_compile);
1388
1389 bang_stack_shadow_pages(false);
1390
1391 // reset the _do_not_unlock_if_synchronized flag
1392 __ get_thread(rax);
1393 __ movbool(do_not_unlock_if_synchronized, false);
1394
1395 // check for synchronized methods
1396 // Must happen AFTER invocation_counter check and stack overflow check,
1397 // so method is not locked if overflows.
1398 //
1399 if (synchronized) {
1400 // Allocate monitor and lock method
1401 lock_method();
1402 } else {
1403 // no synchronization necessary
1404 #ifdef ASSERT
1405 { Label L;
1406 __ movl(rax, access_flags);
1407 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1408 __ jcc(Assembler::zero, L);
1409 __ stop("method needs synchronization");
1410 __ bind(L);
1411 }
1412 #endif
1413 }
1414
1415 // start execution
1416 #ifdef ASSERT
1417 { Label L;
1418 const Address monitor_block_top (rbp,
1419 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1420 __ movptr(rax, monitor_block_top);
1421 __ cmpptr(rax, rsp);
1422 __ jcc(Assembler::equal, L);
1423 __ stop("broken stack frame setup in interpreter");
1424 __ bind(L);
1425 }
1426 #endif
1427
1428 // jvmti support
1429 __ notify_method_entry();
1430
1431 __ dispatch_next(vtos);
1432
1433 // invocation counter overflow
1434 if (inc_counter) {
1435 if (ProfileInterpreter) {
1436 // We have decided to profile this method in the interpreter
1437 __ bind(profile_method);
1438 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1439 __ set_method_data_pointer_for_bcp();
1440 __ get_method(rbx);
1441 __ jmp(profile_method_continue);
1442 }
1443 // Handle overflow of counter and compile method
1444 __ bind(invocation_counter_overflow);
1445 generate_counter_overflow(&continue_after_compile);
1446 }
1447
1448 return entry_point;
1449 }
1450
1451 //------------------------------------------------------------------------------------------------------------------------
1452 // Entry points
1453 //
1454 // Here we generate the various kind of entries into the interpreter.
1455 // The two main entry type are generic bytecode methods and native call method.
1456 // These both come in synchronized and non-synchronized versions but the
1457 // frame layout they create is very similar. The other method entry
1458 // types are really just special purpose entries that are really entry
1459 // and interpretation all in one. These are for trivial methods like
1460 // accessor, empty, or special math methods.
1461 //
1462 // When control flow reaches any of the entry types for the interpreter
1463 // the following holds ->
1464 //
1465 // Arguments:
1466 //
1467 // rbx,: Method*
1468 // rcx: receiver
1469 //
1470 //
1471 // Stack layout immediately at entry
1472 //
1473 // [ return address ] <--- rsp
1474 // [ parameter n ]
1475 // ...
1476 // [ parameter 1 ]
1477 // [ expression stack ] (caller's java expression stack)
1478
1479 // Assuming that we don't go to one of the trivial specialized
1480 // entries the stack will look like below when we are ready to execute
1481 // the first bytecode (or call the native routine). The register usage
1482 // will be as the template based interpreter expects (see interpreter_x86.hpp).
1483 //
1484 // local variables follow incoming parameters immediately; i.e.
1485 // the return address is moved to the end of the locals).
1486 //
1487 // [ monitor entry ] <--- rsp
1488 // ...
1489 // [ monitor entry ]
1490 // [ expr. stack bottom ]
1491 // [ saved rsi ]
1492 // [ current rdi ]
1493 // [ Method* ]
1494 // [ saved rbp, ] <--- rbp,
1495 // [ return address ]
1496 // [ local variable m ]
1497 // ...
1498 // [ local variable 1 ]
1499 // [ parameter n ]
1500 // ...
1501 // [ parameter 1 ] <--- rdi
1502
1503 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
1504 // determine code generation flags
1505 bool synchronized = false;
1506 address entry_point = NULL;
1507
1508 switch (kind) {
1509 case Interpreter::zerolocals : break;
1510 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1511 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
1512 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
1513 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
1514 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
1515 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
1516
1517 case Interpreter::java_lang_math_sin : // fall thru
1518 case Interpreter::java_lang_math_cos : // fall thru
1519 case Interpreter::java_lang_math_tan : // fall thru
1520 case Interpreter::java_lang_math_abs : // fall thru
1521 case Interpreter::java_lang_math_log : // fall thru
1522 case Interpreter::java_lang_math_log10 : // fall thru
1523 case Interpreter::java_lang_math_sqrt : // fall thru
1524 case Interpreter::java_lang_math_pow : // fall thru
1525 case Interpreter::java_lang_math_exp : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1526 case Interpreter::java_lang_ref_reference_get
1527 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1528 default:
1529 fatal(err_msg("unexpected method kind: %d", kind));
1530 break;
1531 }
1532
1533 if (entry_point) return entry_point;
1534
1535 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1536
1537 }
1538
1539 // These should never be compiled since the interpreter will prefer
1540 // the compiled version to the intrinsic version.
1541 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1542 switch (method_kind(m)) {
1543 case Interpreter::java_lang_math_sin : // fall thru
1544 case Interpreter::java_lang_math_cos : // fall thru
1545 case Interpreter::java_lang_math_tan : // fall thru
1546 case Interpreter::java_lang_math_abs : // fall thru
1547 case Interpreter::java_lang_math_log : // fall thru
1548 case Interpreter::java_lang_math_log10 : // fall thru
1549 case Interpreter::java_lang_math_sqrt : // fall thru
1550 case Interpreter::java_lang_math_pow : // fall thru
1551 case Interpreter::java_lang_math_exp :
1552 return false;
1553 default:
1554 return true;
1555 }
1556 }
1557
1558 // How much stack a method activation needs in words.
1559 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1560
1561 const int stub_code = 4; // see generate_call_stub
1562 // Save space for one monitor to get into the interpreted method in case
1563 // the method is synchronized
1564 int monitor_size = method->is_synchronized() ?
1565 1*frame::interpreter_frame_monitor_size() : 0;
1566
1567 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1568 // be sure to change this if you add/subtract anything to/from the overhead area
1569 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1570
1571 const int extra_stack = Method::extra_stack_entries();
1572 const int method_stack = (method->max_locals() + method->max_stack() + extra_stack) *
1573 Interpreter::stackElementWords;
1574 return overhead_size + method_stack + stub_code;
1575 }
1576
1577 // asm based interpreter deoptimization helpers
1578
1579 int AbstractInterpreter::layout_activation(Method* method,
1580 int tempcount,
1581 int popframe_extra_args,
1582 int moncount,
1583 int caller_actual_parameters,
1584 int callee_param_count,
1585 int callee_locals,
1586 frame* caller,
1587 frame* interpreter_frame,
1588 bool is_top_frame,
1589 bool is_bottom_frame) {
1590 // Note: This calculation must exactly parallel the frame setup
1591 // in AbstractInterpreterGenerator::generate_method_entry.
1592 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1593 // The frame interpreter_frame, if not NULL, is guaranteed to be the right size,
1594 // as determined by a previous call to this method.
1595 // It is also guaranteed to be walkable even though it is in a skeletal state
1596 // NOTE: return size is in words not bytes
1597
1598 // fixed size of an interpreter frame:
1599 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1600 int extra_locals = (method->max_locals() - method->size_of_parameters()) *
1601 Interpreter::stackElementWords;
1602
1603 int overhead = frame::sender_sp_offset - frame::interpreter_frame_initial_sp_offset;
1604
1605 // Our locals were accounted for by the caller (or last_frame_adjust on the transistion)
1606 // Since the callee parameters already account for the callee's params we only need to account for
1607 // the extra locals.
1608
1609
1610 int size = overhead +
1611 ((callee_locals - callee_param_count)*Interpreter::stackElementWords) +
1612 (moncount*frame::interpreter_frame_monitor_size()) +
1613 tempcount*Interpreter::stackElementWords + popframe_extra_args;
1614
1615 if (interpreter_frame != NULL) {
1616 #ifdef ASSERT
1617 if (!EnableInvokeDynamic)
1618 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1619 // Probably, since deoptimization doesn't work yet.
1620 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1621 assert(caller->sp() == interpreter_frame->sender_sp(), "Frame not properly walkable(2)");
1622 #endif
1623
1624 interpreter_frame->interpreter_frame_set_method(method);
1625 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1626 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1627 // and sender_sp is fp+8
1628 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1629
1630 #ifdef ASSERT
1631 if (caller->is_interpreted_frame()) {
1632 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1633 }
1634 #endif
1635
1636 interpreter_frame->interpreter_frame_set_locals(locals);
1637 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1638 BasicObjectLock* monbot = montop - moncount;
1639 interpreter_frame->interpreter_frame_set_monitor_end(monbot);
1640
1641 // Set last_sp
1642 intptr_t* rsp = (intptr_t*) monbot -
1643 tempcount*Interpreter::stackElementWords -
1644 popframe_extra_args;
1645 interpreter_frame->interpreter_frame_set_last_sp(rsp);
1646
1647 // All frames but the initial (oldest) interpreter frame we fill in have a
1648 // value for sender_sp that allows walking the stack but isn't
1649 // truly correct. Correct the value here.
1650
1651 if (extra_locals != 0 &&
1652 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1653 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1654 }
1655 *interpreter_frame->interpreter_frame_cache_addr() =
1656 method->constants()->cache();
1657 }
1658 return size;
1659 }
1660
1661
1662 //------------------------------------------------------------------------------------------------------------------------
1663 // Exceptions
1664
1665 void TemplateInterpreterGenerator::generate_throw_exception() {
1666 // Entry point in previous activation (i.e., if the caller was interpreted)
1667 Interpreter::_rethrow_exception_entry = __ pc();
1668 const Register thread = rcx;
1669
1670 // Restore sp to interpreter_frame_last_sp even though we are going
1671 // to empty the expression stack for the exception processing.
1672 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1673 // rax,: exception
1674 // rdx: return address/pc that threw exception
1675 __ restore_bcp(); // rsi points to call/send
1676 __ restore_locals();
1677
1678 // Entry point for exceptions thrown within interpreter code
1679 Interpreter::_throw_exception_entry = __ pc();
1680 // expression stack is undefined here
1681 // rax,: exception
1682 // rsi: exception bcp
1683 __ verify_oop(rax);
1684
1685 // expression stack must be empty before entering the VM in case of an exception
1686 __ empty_expression_stack();
1687 __ empty_FPU_stack();
1688 // find exception handler address and preserve exception oop
1689 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1690 // rax,: exception handler entry point
1691 // rdx: preserved exception oop
1692 // rsi: bcp for exception handler
1693 __ push_ptr(rdx); // push exception which is now the only value on the stack
1694 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1695
1696 // If the exception is not handled in the current frame the frame is removed and
1697 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1698 //
1699 // Note: At this point the bci is still the bxi for the instruction which caused
1700 // the exception and the expression stack is empty. Thus, for any VM calls
1701 // at this point, GC will find a legal oop map (with empty expression stack).
1702
1703 // In current activation
1704 // tos: exception
1705 // rsi: exception bcp
1706
1707 //
1708 // JVMTI PopFrame support
1709 //
1710
1711 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1712 __ empty_expression_stack();
1713 __ empty_FPU_stack();
1714 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1715 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1716 // popframe handling cycles.
1717 __ get_thread(thread);
1718 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1719 __ orl(rdx, JavaThread::popframe_processing_bit);
1720 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1721
1722 {
1723 // Check to see whether we are returning to a deoptimized frame.
1724 // (The PopFrame call ensures that the caller of the popped frame is
1725 // either interpreted or compiled and deoptimizes it if compiled.)
1726 // In this case, we can't call dispatch_next() after the frame is
1727 // popped, but instead must save the incoming arguments and restore
1728 // them after deoptimization has occurred.
1729 //
1730 // Note that we don't compare the return PC against the
1731 // deoptimization blob's unpack entry because of the presence of
1732 // adapter frames in C2.
1733 Label caller_not_deoptimized;
1734 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1735 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1736 __ testl(rax, rax);
1737 __ jcc(Assembler::notZero, caller_not_deoptimized);
1738
1739 // Compute size of arguments for saving when returning to deoptimized caller
1740 __ get_method(rax);
1741 __ movptr(rax, Address(rax, Method::const_offset()));
1742 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1743 __ shlptr(rax, Interpreter::logStackElementSize);
1744 __ restore_locals();
1745 __ subptr(rdi, rax);
1746 __ addptr(rdi, wordSize);
1747 // Save these arguments
1748 __ get_thread(thread);
1749 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1750
1751 __ remove_activation(vtos, rdx,
1752 /* throw_monitor_exception */ false,
1753 /* install_monitor_exception */ false,
1754 /* notify_jvmdi */ false);
1755
1756 // Inform deoptimization that it is responsible for restoring these arguments
1757 __ get_thread(thread);
1758 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1759
1760 // Continue in deoptimization handler
1761 __ jmp(rdx);
1762
1763 __ bind(caller_not_deoptimized);
1764 }
1765
1766 __ remove_activation(vtos, rdx,
1767 /* throw_monitor_exception */ false,
1768 /* install_monitor_exception */ false,
1769 /* notify_jvmdi */ false);
1770
1771 // Finish with popframe handling
1772 // A previous I2C followed by a deoptimization might have moved the
1773 // outgoing arguments further up the stack. PopFrame expects the
1774 // mutations to those outgoing arguments to be preserved and other
1775 // constraints basically require this frame to look exactly as
1776 // though it had previously invoked an interpreted activation with
1777 // no space between the top of the expression stack (current
1778 // last_sp) and the top of stack. Rather than force deopt to
1779 // maintain this kind of invariant all the time we call a small
1780 // fixup routine to move the mutated arguments onto the top of our
1781 // expression stack if necessary.
1782 __ mov(rax, rsp);
1783 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1784 __ get_thread(thread);
1785 // PC must point into interpreter here
1786 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1787 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1788 __ get_thread(thread);
1789 __ reset_last_Java_frame(thread, true, true);
1790 // Restore the last_sp and null it out
1791 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1792 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1793
1794 __ restore_bcp();
1795 __ restore_locals();
1796 // The method data pointer was incremented already during
1797 // call profiling. We have to restore the mdp for the current bcp.
1798 if (ProfileInterpreter) {
1799 __ set_method_data_pointer_for_bcp();
1800 }
1801
1802 // Clear the popframe condition flag
1803 __ get_thread(thread);
1804 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1805
1806 __ dispatch_next(vtos);
1807 // end of PopFrame support
1808
1809 Interpreter::_remove_activation_entry = __ pc();
1810
1811 // preserve exception over this code sequence
1812 __ pop_ptr(rax);
1813 __ get_thread(thread);
1814 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1815 // remove the activation (without doing throws on illegalMonitorExceptions)
1816 __ remove_activation(vtos, rdx, false, true, false);
1817 // restore exception
1818 __ get_thread(thread);
1819 __ get_vm_result(rax, thread);
1820
1821 // Inbetween activations - previous activation type unknown yet
1822 // compute continuation point - the continuation point expects
1823 // the following registers set up:
1824 //
1825 // rax: exception
1826 // rdx: return address/pc that threw exception
1827 // rsp: expression stack of caller
1828 // rbp: rbp, of caller
1829 __ push(rax); // save exception
1830 __ push(rdx); // save return address
1831 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1832 __ mov(rbx, rax); // save exception handler
1833 __ pop(rdx); // restore return address
1834 __ pop(rax); // restore exception
1835 // Note that an "issuing PC" is actually the next PC after the call
1836 __ jmp(rbx); // jump to exception handler of caller
1837 }
1838
1839
1840 //
1841 // JVMTI ForceEarlyReturn support
1842 //
1843 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1844 address entry = __ pc();
1845 const Register thread = rcx;
1846
1847 __ restore_bcp();
1848 __ restore_locals();
1849 __ empty_expression_stack();
1850 __ empty_FPU_stack();
1851 __ load_earlyret_value(state);
1852
1853 __ get_thread(thread);
1854 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1855 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1856
1857 // Clear the earlyret state
1858 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1859
1860 __ remove_activation(state, rsi,
1861 false, /* throw_monitor_exception */
1862 false, /* install_monitor_exception */
1863 true); /* notify_jvmdi */
1864 __ jmp(rsi);
1865 return entry;
1866 } // end of ForceEarlyReturn support
1867
1868
1869 //------------------------------------------------------------------------------------------------------------------------
1870 // Helper for vtos entry point generation
1871
1872 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) {
1873 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1874 Label L;
1875 fep = __ pc(); __ push(ftos); __ jmp(L);
1876 dep = __ pc(); __ push(dtos); __ jmp(L);
1877 lep = __ pc(); __ push(ltos); __ jmp(L);
1878 aep = __ pc(); __ push(atos); __ jmp(L);
1879 bep = cep = sep = // fall through
1880 iep = __ pc(); __ push(itos); // fall through
1881 vep = __ pc(); __ bind(L); // fall through
1882 generate_and_dispatch(t);
1883 }
1884
1885 //------------------------------------------------------------------------------------------------------------------------
1886 // Generation of individual instructions
1887
1888 // helpers for generate_and_dispatch
1889
1890
1891
1892 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1893 : TemplateInterpreterGenerator(code) {
1894 generate_all(); // down here so it can be "virtual"
1895 }
1896
1897 //------------------------------------------------------------------------------------------------------------------------
1898
1899 // Non-product code
1900 #ifndef PRODUCT
1901 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1902 address entry = __ pc();
1903
1904 // prepare expression stack
1905 __ pop(rcx); // pop return address so expression stack is 'pure'
1906 __ push(state); // save tosca
1907
1908 // pass tosca registers as arguments & call tracer
1909 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1910 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1911 __ pop(state); // restore tosca
1912
1913 // return
1914 __ jmp(rcx);
1915
1916 return entry;
1917 }
1918
1919
1920 void TemplateInterpreterGenerator::count_bytecode() {
1921 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1922 }
1923
1924
1925 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1926 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1927 }
1928
1929
1930 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1931 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1932 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1933 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1934 ExternalAddress table((address) BytecodePairHistogram::_counters);
1935 Address index(noreg, rbx, Address::times_4);
1936 __ incrementl(ArrayAddress(table, index));
1937 }
1938
1939
1940 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1941 // Call a little run-time stub to avoid blow-up for each bytecode.
1942 // The run-time runtime saves the right registers, depending on
1943 // the tosca in-state for the given template.
1944 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1945 "entry must have been generated");
1946 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1947 }
1948
1949
1950 void TemplateInterpreterGenerator::stop_interpreter_at() {
1951 Label L;
1952 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1953 StopInterpreterAt);
1954 __ jcc(Assembler::notEqual, L);
1955 __ int3();
1956 __ bind(L);
1957 }
1958 #endif // !PRODUCT
1959 #endif // CC_INTERP