1 /*
2 * Copyright (c) 1997, 2015, 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/interp_masm.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "oops/arrayOop.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/jvmtiExport.hpp"
38 #include "prims/jvmtiThreadState.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #include "utilities/macros.hpp"
49
50 #define __ _masm->
51
52
53 #ifndef CC_INTERP
54 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
55 const int bcp_offset = frame::interpreter_frame_bcp_offset * wordSize;
56 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
57
58 //------------------------------------------------------------------------------------------------------------------------
59
60 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
61 address entry = __ pc();
62
63 // Note: There should be a minimal interpreter frame set up when stack
64 // overflow occurs since we check explicitly for it now.
65 //
66 #ifdef ASSERT
67 { Label L;
68 __ lea(rax, Address(rbp,
69 frame::interpreter_frame_monitor_block_top_offset * wordSize));
70 __ cmpptr(rax, rsp); // rax, = maximal rsp for current rbp,
71 // (stack grows negative)
72 __ jcc(Assembler::aboveEqual, L); // check if frame is complete
73 __ stop ("interpreter frame not set up");
74 __ bind(L);
75 }
76 #endif // ASSERT
77 // Restore bcp under the assumption that the current frame is still
78 // interpreted
79 __ restore_bcp();
80
81 // expression stack must be empty before entering the VM if an exception
82 // happened
83 __ empty_expression_stack();
84 __ empty_FPU_stack();
85 // throw exception
86 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
87 return entry;
88 }
89
90 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) {
91 address entry = __ pc();
92 // expression stack must be empty before entering the VM if an exception happened
93 __ empty_expression_stack();
94 __ empty_FPU_stack();
95 // setup parameters
96 // ??? convention: expect aberrant index in register rbx,
97 __ lea(rax, ExternalAddress((address)name));
98 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), rax, rbx);
99 return entry;
100 }
101
102 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
103 address entry = __ pc();
104 // object is at TOS
105 __ pop(rax);
106 // expression stack must be empty before entering the VM if an exception
107 // happened
108 __ empty_expression_stack();
109 __ empty_FPU_stack();
110 __ call_VM(noreg,
111 CAST_FROM_FN_PTR(address,
112 InterpreterRuntime::throw_ClassCastException),
113 rax);
114 return entry;
115 }
116
117 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) {
118 assert(!pass_oop || message == NULL, "either oop or message but not both");
119 address entry = __ pc();
120 if (pass_oop) {
121 // object is at TOS
122 __ pop(rbx);
123 }
124 // expression stack must be empty before entering the VM if an exception happened
125 __ empty_expression_stack();
126 __ empty_FPU_stack();
127 // setup parameters
128 __ lea(rax, ExternalAddress((address)name));
129 if (pass_oop) {
130 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), rax, rbx);
131 } else {
132 if (message != NULL) {
133 __ lea(rbx, ExternalAddress((address)message));
134 } else {
135 __ movptr(rbx, NULL_WORD);
136 }
137 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), rax, rbx);
138 }
139 // throw exception
140 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
141 return entry;
142 }
143
144
145 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
146 address entry = __ pc();
147 // NULL last_sp until next java call
148 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
149 __ dispatch_next(state);
150 return entry;
151 }
152
153
154 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
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 ((state == ftos && UseSSE < 1) || (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 ((state == ftos && UseSSE < 1) || (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 if (state == ftos) {
174 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_return_entry_for in interpreter");
175 } else if (state == dtos) {
176 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_return_entry_for in interpreter");
177 }
178
179 // Restore stack bottom in case i2c adjusted stack
180 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
181 // and NULL it as marker that rsp is now tos until next java call
182 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
183
184 __ restore_bcp();
185 __ restore_locals();
186
187 if (state == atos) {
188 Register mdp = rbx;
189 Register tmp = rcx;
190 __ profile_return_type(mdp, rax, tmp);
191 }
192
193 const Register cache = rbx;
194 const Register index = rcx;
195 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
196
197 const Register flags = cache;
198 __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
199 __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
200 __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
201 __ dispatch_next(state, step);
202
203 return entry;
204 }
205
206
207 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) {
208 address entry = __ pc();
209
210 if (state == ftos) {
211 __ MacroAssembler::verify_FPU(UseSSE >= 1 ? 0 : 1, "generate_deopt_entry_for in interpreter");
212 } else if (state == dtos) {
213 __ MacroAssembler::verify_FPU(UseSSE >= 2 ? 0 : 1, "generate_deopt_entry_for in interpreter");
214 }
215
216 // The stack is not extended by deopt but we must NULL last_sp as this
217 // entry is like a "return".
218 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
219 __ restore_bcp();
220 __ restore_locals();
221 // handle exceptions
222 { Label L;
223 const Register thread = rcx;
224 __ get_thread(thread);
225 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
226 __ jcc(Assembler::zero, L);
227 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
228 __ should_not_reach_here();
229 __ bind(L);
230 }
231 __ dispatch_next(state, step);
232 return entry;
233 }
234
235
236 int AbstractInterpreter::BasicType_as_index(BasicType type) {
237 int i = 0;
238 switch (type) {
239 case T_BOOLEAN: i = 0; break;
240 case T_CHAR : i = 1; break;
241 case T_BYTE : i = 2; break;
242 case T_SHORT : i = 3; break;
243 case T_INT : // fall through
244 case T_LONG : // fall through
245 case T_VOID : i = 4; break;
246 case T_FLOAT : i = 5; break; // have to treat float and double separately for SSE
247 case T_DOUBLE : i = 6; break;
248 case T_OBJECT : // fall through
249 case T_ARRAY : i = 7; break;
250 default : ShouldNotReachHere();
251 }
252 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
253 return i;
254 }
255
256
257 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) {
258 address entry = __ pc();
259 switch (type) {
260 case T_BOOLEAN: __ c2bool(rax); break;
261 case T_CHAR : __ andptr(rax, 0xFFFF); break;
262 case T_BYTE : __ sign_extend_byte (rax); break;
263 case T_SHORT : __ sign_extend_short(rax); break;
264 case T_INT : /* nothing to do */ break;
265 case T_LONG : /* nothing to do */ break;
266 case T_VOID : /* nothing to do */ break;
267 case T_DOUBLE :
268 case T_FLOAT :
269 { const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
270 __ pop(t); // remove return address first
271 // Must return a result for interpreter or compiler. In SSE
272 // mode, results are returned in xmm0 and the FPU stack must
273 // be empty.
274 if (type == T_FLOAT && UseSSE >= 1) {
275 // Load ST0
276 __ fld_d(Address(rsp, 0));
277 // Store as float and empty fpu stack
278 __ fstp_s(Address(rsp, 0));
279 // and reload
280 __ movflt(xmm0, Address(rsp, 0));
281 } else if (type == T_DOUBLE && UseSSE >= 2 ) {
282 __ movdbl(xmm0, Address(rsp, 0));
283 } else {
284 // restore ST0
285 __ fld_d(Address(rsp, 0));
286 }
287 // and pop the temp
288 __ addptr(rsp, 2 * wordSize);
289 __ push(t); // restore return address
290 }
291 break;
292 case T_OBJECT :
293 // retrieve result from frame
294 __ movptr(rax, Address(rbp, frame::interpreter_frame_oop_temp_offset*wordSize));
295 // and verify it
296 __ verify_oop(rax);
297 break;
298 default : ShouldNotReachHere();
299 }
300 __ ret(0); // return from result handler
301 return entry;
302 }
303
304 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) {
305 address entry = __ pc();
306 __ push(state);
307 __ call_VM(noreg, runtime_entry);
308 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
309 return entry;
310 }
311
312
313 // Helpers for commoning out cases in the various type of method entries.
314 //
315
316 // increment invocation count & check for overflow
317 //
318 // Note: checking for negative value instead of overflow
319 // so we have a 'sticky' overflow test
320 //
321 // rbx,: method
322 // rcx: invocation counter
323 //
324 void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
325 Label done;
326 // Note: In tiered we increment either counters in MethodCounters* or in MDO
327 // depending if we're profiling or not.
328 if (TieredCompilation) {
329 int increment = InvocationCounter::count_increment;
330 Label no_mdo;
331 if (ProfileInterpreter) {
332 // Are we profiling?
333 __ movptr(rax, Address(rbx, Method::method_data_offset()));
334 __ testptr(rax, rax);
335 __ jccb(Assembler::zero, no_mdo);
336 // Increment counter in the MDO
337 const Address mdo_invocation_counter(rax, in_bytes(MethodData::invocation_counter_offset()) +
338 in_bytes(InvocationCounter::counter_offset()));
339 const Address mask(rax, in_bytes(MethodData::invoke_mask_offset()));
340 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, rcx, false, Assembler::zero, overflow);
341 __ jmp(done);
342 }
343 __ bind(no_mdo);
344 // Increment counter in MethodCounters
345 const Address invocation_counter(rax,
346 MethodCounters::invocation_counter_offset() +
347 InvocationCounter::counter_offset());
348
349 __ get_method_counters(rbx, rax, done);
350 const Address mask(rax, in_bytes(MethodCounters::invoke_mask_offset()));
351 __ increment_mask_and_jump(invocation_counter, increment, mask,
352 rcx, false, Assembler::zero, overflow);
353 __ bind(done);
354 } else { // not TieredCompilation
355 const Address backedge_counter(rax,
356 MethodCounters::backedge_counter_offset() +
357 InvocationCounter::counter_offset());
358 const Address invocation_counter(rax,
359 MethodCounters::invocation_counter_offset() +
360 InvocationCounter::counter_offset());
361
362 __ get_method_counters(rbx, rax, done);
363
364 if (ProfileInterpreter) {
365 __ incrementl(Address(rax,
366 MethodCounters::interpreter_invocation_counter_offset()));
367 }
368
369 // Update standard invocation counters
370 __ movl(rcx, invocation_counter);
371 __ incrementl(rcx, InvocationCounter::count_increment);
372 __ movl(invocation_counter, rcx); // save invocation count
373
374 __ movl(rax, backedge_counter); // load backedge counter
375 __ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
376
377 __ addl(rcx, rax); // add both counters
378
379 // profile_method is non-null only for interpreted method so
380 // profile_method != NULL == !native_call
381 // BytecodeInterpreter only calls for native so code is elided.
382
383 if (ProfileInterpreter && profile_method != NULL) {
384 // Test to see if we should create a method data oop
385 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
386 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
387 __ jcc(Assembler::less, *profile_method_continue);
388
389 // if no method data exists, go to profile_method
390 __ test_method_data_pointer(rax, *profile_method);
391 }
392
393 __ movptr(rax, Address(rbx, Method::method_counters_offset()));
394 __ cmp32(rcx, Address(rax, in_bytes(MethodCounters::interpreter_invocation_limit_offset())));
395 __ jcc(Assembler::aboveEqual, *overflow);
396 __ bind(done);
397 }
398 }
399
400 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
401
402 // Asm interpreter on entry
403 // rdi - locals
404 // rsi - bcp
405 // rbx, - method
406 // rdx - cpool
407 // rbp, - interpreter frame
408
409 // C++ interpreter on entry
410 // rsi - new interpreter state pointer
411 // rbp - interpreter frame pointer
412 // rbx - method
413
414 // On return (i.e. jump to entry_point) [ back to invocation of interpreter ]
415 // rbx, - method
416 // rcx - rcvr (assuming there is one)
417 // top of stack return address of interpreter caller
418 // rsp - sender_sp
419
420 // C++ interpreter only
421 // rsi - previous interpreter state pointer
422
423 // InterpreterRuntime::frequency_counter_overflow takes one argument
424 // indicating if the counter overflow occurs at a backwards branch (non-NULL bcp).
425 // The call returns the address of the verified entry point for the method or NULL
426 // if the compilation did not complete (either went background or bailed out).
427 __ movptr(rax, (intptr_t)false);
428 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
429
430 __ movptr(rbx, Address(rbp, method_offset)); // restore Method*
431
432 // Preserve invariant that rsi/rdi contain bcp/locals of sender frame
433 // and jump to the interpreted entry.
434 __ jmp(*do_continue, relocInfo::none);
435
436 }
437
438 void InterpreterGenerator::generate_stack_overflow_check(void) {
439 // see if we've got enough room on the stack for locals plus overhead.
440 // the expression stack grows down incrementally, so the normal guard
441 // page mechanism will work for that.
442 //
443 // Registers live on entry:
444 //
445 // Asm interpreter
446 // rdx: number of additional locals this frame needs (what we must check)
447 // rbx,: Method*
448
449 // destroyed on exit
450 // rax,
451
452 // NOTE: since the additional locals are also always pushed (wasn't obvious in
453 // generate_fixed_frame) so the guard should work for them too.
454 //
455
456 // monitor entry size: see picture of stack in frame_x86.hpp
457 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
458
459 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
460 // be sure to change this if you add/subtract anything to/from the overhead area
461 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize) + entry_size;
462
463 const int page_size = os::vm_page_size();
464
465 Label after_frame_check;
466
467 // see if the frame is greater than one page in size. If so,
468 // then we need to verify there is enough stack space remaining
469 // for the additional locals.
470 __ cmpl(rdx, (page_size - overhead_size)/Interpreter::stackElementSize);
471 __ jcc(Assembler::belowEqual, after_frame_check);
472
473 // compute rsp as if this were going to be the last frame on
474 // the stack before the red zone
475
476 Label after_frame_check_pop;
477
478 __ push(rsi);
479
480 const Register thread = rsi;
481
482 __ get_thread(thread);
483
484 const Address stack_base(thread, Thread::stack_base_offset());
485 const Address stack_size(thread, Thread::stack_size_offset());
486
487 // locals + overhead, in bytes
488 __ lea(rax, Address(noreg, rdx, Interpreter::stackElementScale(), overhead_size));
489
490 #ifdef ASSERT
491 Label stack_base_okay, stack_size_okay;
492 // verify that thread stack base is non-zero
493 __ cmpptr(stack_base, (int32_t)NULL_WORD);
494 __ jcc(Assembler::notEqual, stack_base_okay);
495 __ stop("stack base is zero");
496 __ bind(stack_base_okay);
497 // verify that thread stack size is non-zero
498 __ cmpptr(stack_size, 0);
499 __ jcc(Assembler::notEqual, stack_size_okay);
500 __ stop("stack size is zero");
501 __ bind(stack_size_okay);
502 #endif
503
504 // Add stack base to locals and subtract stack size
505 __ addptr(rax, stack_base);
506 __ subptr(rax, stack_size);
507
508 // Use the maximum number of pages we might bang.
509 const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
510 (StackRedPages+StackYellowPages);
511 __ addptr(rax, max_pages * page_size);
512
513 // check against the current stack bottom
514 __ cmpptr(rsp, rax);
515 __ jcc(Assembler::above, after_frame_check_pop);
516
517 __ pop(rsi); // get saved bcp / (c++ prev state ).
518
519 // Restore sender's sp as SP. This is necessary if the sender's
520 // frame is an extended compiled frame (see gen_c2i_adapter())
521 // and safer anyway in case of JSR292 adaptations.
522
523 __ pop(rax); // return address must be moved if SP is changed
524 __ mov(rsp, rsi);
525 __ push(rax);
526
527 // Note: the restored frame is not necessarily interpreted.
528 // Use the shared runtime version of the StackOverflowError.
529 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
530 __ jump(ExternalAddress(StubRoutines::throw_StackOverflowError_entry()));
531 // all done with frame size check
532 __ bind(after_frame_check_pop);
533 __ pop(rsi);
534
535 __ bind(after_frame_check);
536 }
537
538 // Allocate monitor and lock method (asm interpreter)
539 // rbx, - Method*
540 //
541 void InterpreterGenerator::lock_method(void) {
542 // synchronize method
543 const Address access_flags (rbx, Method::access_flags_offset());
544 const Address monitor_block_top (rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
545 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
546
547 #ifdef ASSERT
548 { Label L;
549 __ movl(rax, access_flags);
550 __ testl(rax, JVM_ACC_SYNCHRONIZED);
551 __ jcc(Assembler::notZero, L);
552 __ stop("method doesn't need synchronization");
553 __ bind(L);
554 }
555 #endif // ASSERT
556 // get synchronization object
557 { Label done;
558 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
559 __ movl(rax, access_flags);
560 __ testl(rax, JVM_ACC_STATIC);
561 __ movptr(rax, Address(rdi, Interpreter::local_offset_in_bytes(0))); // get receiver (assume this is frequent case)
562 __ jcc(Assembler::zero, done);
563 __ movptr(rax, Address(rbx, Method::const_offset()));
564 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
565 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
566 __ movptr(rax, Address(rax, mirror_offset));
567 __ bind(done);
568 }
569 // add space for monitor & lock
570 __ subptr(rsp, entry_size); // add space for a monitor entry
571 __ movptr(monitor_block_top, rsp); // set new monitor block top
572 __ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
573 __ mov(rdx, rsp); // object address
574 __ lock_object(rdx);
575 }
576
577 //
578 // Generate a fixed interpreter frame. This is identical setup for interpreted methods
579 // and for native methods hence the shared code.
580
581 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
582 // initialize fixed part of activation frame
583 __ push(rax); // save return address
584 __ enter(); // save old & set new rbp,
585
586
587 __ push(rsi); // set sender sp
588 __ push((int32_t)NULL_WORD); // leave last_sp as null
589 __ movptr(rsi, Address(rbx,Method::const_offset())); // get ConstMethod*
590 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
591 __ push(rbx); // save Method*
592 if (ProfileInterpreter) {
593 Label method_data_continue;
594 __ movptr(rdx, Address(rbx, in_bytes(Method::method_data_offset())));
595 __ testptr(rdx, rdx);
596 __ jcc(Assembler::zero, method_data_continue);
597 __ addptr(rdx, in_bytes(MethodData::data_offset()));
598 __ bind(method_data_continue);
599 __ push(rdx); // set the mdp (method data pointer)
600 } else {
601 __ push(0);
602 }
603
604 __ movptr(rdx, Address(rbx, Method::const_offset()));
605 __ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
606 __ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
607 __ push(rdx); // set constant pool cache
608 __ push(rdi); // set locals pointer
609 if (native_call) {
610 __ push(0); // no bcp
611 } else {
612 __ push(rsi); // set bcp
613 }
614 __ push(0); // reserve word for pointer to expression stack bottom
615 __ movptr(Address(rsp, 0), rsp); // set expression stack bottom
616 }
617
618
619 // Method entry for java.lang.ref.Reference.get.
620 address InterpreterGenerator::generate_Reference_get_entry(void) {
621 #if INCLUDE_ALL_GCS
622 // Code: _aload_0, _getfield, _areturn
623 // parameter size = 1
624 //
625 // The code that gets generated by this routine is split into 2 parts:
626 // 1. The "intrinsified" code for G1 (or any SATB based GC),
627 // 2. The slow path - which is an expansion of the regular method entry.
628 //
629 // Notes:-
630 // * In the G1 code we do not check whether we need to block for
631 // a safepoint. If G1 is enabled then we must execute the specialized
632 // code for Reference.get (except when the Reference object is null)
633 // so that we can log the value in the referent field with an SATB
634 // update buffer.
635 // If the code for the getfield template is modified so that the
636 // G1 pre-barrier code is executed when the current method is
637 // Reference.get() then going through the normal method entry
638 // will be fine.
639 // * The G1 code below can, however, check the receiver object (the instance
640 // of java.lang.Reference) and jump to the slow path if null. If the
641 // Reference object is null then we obviously cannot fetch the referent
642 // and so we don't need to call the G1 pre-barrier. Thus we can use the
643 // regular method entry code to generate the NPE.
644 //
645 // This code is based on generate_accessor_enty.
646
647 // rbx,: Method*
648 // rcx: receiver (preserve for slow entry into asm interpreter)
649
650 // rsi: senderSP must preserved for slow path, set SP to it on fast path
651
652 address entry = __ pc();
653
654 const int referent_offset = java_lang_ref_Reference::referent_offset;
655 guarantee(referent_offset > 0, "referent offset not initialized");
656
657 if (UseG1GC) {
658 Label slow_path;
659
660 // Check if local 0 != NULL
661 // If the receiver is null then it is OK to jump to the slow path.
662 __ movptr(rax, Address(rsp, wordSize));
663 __ testptr(rax, rax);
664 __ jcc(Assembler::zero, slow_path);
665
666 // rax: local 0 (must be preserved across the G1 barrier call)
667 //
668 // rbx: method (at this point it's scratch)
669 // rcx: receiver (at this point it's scratch)
670 // rdx: scratch
671 // rdi: scratch
672 //
673 // rsi: sender sp
674
675 // Preserve the sender sp in case the pre-barrier
676 // calls the runtime
677 __ push(rsi);
678
679 // Load the value of the referent field.
680 const Address field_address(rax, referent_offset);
681 __ movptr(rax, field_address);
682
683 // Generate the G1 pre-barrier code to log the value of
684 // the referent field in an SATB buffer.
685 __ get_thread(rcx);
686 __ g1_write_barrier_pre(noreg /* obj */,
687 rax /* pre_val */,
688 rcx /* thread */,
689 rbx /* tmp */,
690 true /* tosca_save */,
691 true /* expand_call */);
692
693 // _areturn
694 __ pop(rsi); // get sender sp
695 __ pop(rdi); // get return address
696 __ mov(rsp, rsi); // set sp to sender sp
697 __ jmp(rdi);
698
699 __ bind(slow_path);
700 (void) generate_normal_entry(false);
701
702 return entry;
703 }
704 #endif // INCLUDE_ALL_GCS
705
706 // If G1 is not enabled then attempt to go through the accessor entry point
707 // Reference.get is an accessor
708 return generate_jump_to_normal_entry();
709 }
710
711 /**
712 * Method entry for static native methods:
713 * int java.util.zip.CRC32.update(int crc, int b)
714 */
715 address InterpreterGenerator::generate_CRC32_update_entry() {
716 if (UseCRC32Intrinsics) {
717 address entry = __ pc();
718
719 // rbx: Method*
720 // rsi: senderSP must preserved for slow path, set SP to it on fast path
721 // rdx: scratch
722 // rdi: scratch
723
724 Label slow_path;
725 // If we need a safepoint check, generate full interpreter entry.
726 ExternalAddress state(SafepointSynchronize::address_of_state());
727 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
728 SafepointSynchronize::_not_synchronized);
729 __ jcc(Assembler::notEqual, slow_path);
730
731 // We don't generate local frame and don't align stack because
732 // we call stub code and there is no safepoint on this path.
733
734 // Load parameters
735 const Register crc = rax; // crc
736 const Register val = rdx; // source java byte value
737 const Register tbl = rdi; // scratch
738
739 // Arguments are reversed on java expression stack
740 __ movl(val, Address(rsp, wordSize)); // byte value
741 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
742
743 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
744 __ notl(crc); // ~crc
745 __ update_byte_crc32(crc, val, tbl);
746 __ notl(crc); // ~crc
747 // result in rax
748
749 // _areturn
750 __ pop(rdi); // get return address
751 __ mov(rsp, rsi); // set sp to sender sp
752 __ jmp(rdi);
753
754 // generate a vanilla native entry as the slow path
755 __ bind(slow_path);
756
757 (void) generate_native_entry(false);
758
759 return entry;
760 }
761 return generate_native_entry(false);
762 }
763
764 /**
765 * Method entry for static native methods:
766 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
767 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
768 */
769 address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
770 if (UseCRC32Intrinsics) {
771 address entry = __ pc();
772
773 // rbx,: Method*
774 // rsi: senderSP must preserved for slow path, set SP to it on fast path
775 // rdx: scratch
776 // rdi: scratch
777
778 Label slow_path;
779 // If we need a safepoint check, generate full interpreter entry.
780 ExternalAddress state(SafepointSynchronize::address_of_state());
781 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
782 SafepointSynchronize::_not_synchronized);
783 __ jcc(Assembler::notEqual, slow_path);
784
785 // We don't generate local frame and don't align stack because
786 // we call stub code and there is no safepoint on this path.
787
788 // Load parameters
789 const Register crc = rax; // crc
790 const Register buf = rdx; // source java byte array address
791 const Register len = rdi; // length
792
793 // value x86_32
794 // interp. arg ptr ESP + 4
795 // int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
796 // 3 2 1 0
797 // int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
798 // 4 2,3 1 0
799
800 // Arguments are reversed on java expression stack
801 __ movl(len, Address(rsp, 4 + 0)); // Length
802 // Calculate address of start element
803 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
804 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // long buf
805 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
806 __ movl(crc, Address(rsp, 4 + 4 * wordSize)); // Initial CRC
807 } else {
808 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // byte[] array
809 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
810 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
811 __ movl(crc, Address(rsp, 4 + 3 * wordSize)); // Initial CRC
812 }
813
814 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
815 // result in rax
816
817 // _areturn
818 __ pop(rdi); // get return address
819 __ mov(rsp, rsi); // set sp to sender sp
820 __ jmp(rdi);
821
822 // generate a vanilla native entry as the slow path
823 __ bind(slow_path);
824
825 (void) generate_native_entry(false);
826
827 return entry;
828 }
829 return generate_native_entry(false);
830 }
831
832 /**
833 * Method entry for static native methods:
834 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end)
835 * int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end)
836 */
837 address InterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
838 if (UseCRC32CIntrinsics) {
839 address entry = __ pc();
840 // Load parameters
841 const Register crc = rax; // crc
842 const Register buf = rcx; // source java byte array address
843 const Register len = rdx; // length
844 const Register end = len;
845
846 // value x86_32
847 // interp. arg ptr ESP + 4
848 // int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int end)
849 // 3 2 1 0
850 // int java.util.zip.CRC32.updateByteBuffer(int crc, long address, int off, int end)
851 // 4 2,3 1 0
852
853 // Arguments are reversed on java expression stack
854 __ movl(end, Address(rsp, 4 + 0)); // end
855 __ subl(len, Address(rsp, 4 + 1 * wordSize)); // end - offset == length
856 // Calculate address of start element
857 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
858 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // long address
859 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
860 __ movl(crc, Address(rsp, 4 + 4 * wordSize)); // Initial CRC
861 } else {
862 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // byte[] array
863 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
864 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
865 __ movl(crc, Address(rsp, 4 + 3 * wordSize)); // Initial CRC
866 }
867 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()), crc, buf, len);
868 // result in rax
869 // _areturn
870 __ pop(rdi); // get return address
871 __ mov(rsp, rsi); // set sp to sender sp
872 __ jmp(rdi);
873
874 return entry;
875 }
876 return generate_native_entry(false);
877 }
878
879 /**
880 * Method entry for static native method:
881 * java.lang.Float.intBitsToFloat(int bits)
882 */
883 address InterpreterGenerator::generate_Float_intBitsToFloat_entry() {
884 address entry;
885
886 if (UseSSE >= 1) {
887 entry = __ pc();
888
889 // rsi: the sender's SP
890
891 // Skip safepoint check (compiler intrinsic versions of this method
892 // do not perform safepoint checks either).
893
894 // Load 'bits' into xmm0 (interpreter returns results in xmm0)
895 __ movflt(xmm0, Address(rsp, wordSize));
896
897 // Return
898 __ pop(rdi); // get return address
899 __ mov(rsp, rsi); // set rsp to the sender's SP
900 __ jmp(rdi);
901 } else {
902 entry = generate_native_entry(false);
903 }
904
905 return entry;
906 }
907
908 /**
909 * Method entry for static native method:
910 * java.lang.Float.floatToRawIntBits(float value)
911 */
912 address InterpreterGenerator::generate_Float_floatToRawIntBits_entry() {
913 address entry;
914
915 if (UseSSE >= 1) {
916 entry = __ pc();
917
918 // rsi: the sender's SP
919
920 // Skip safepoint check (compiler intrinsic versions of this method
921 // do not perform safepoint checks either).
922
923 // Load the parameter (a floating-point value) into rax.
924 __ movl(rax, Address(rsp, wordSize));
925
926 // Return
927 __ pop(rdi); // get return address
928 __ mov(rsp, rsi); // set rsp to the sender's SP
929 __ jmp(rdi);
930 } else {
931 entry = generate_native_entry(false);
932 }
933
934 return entry;
935 }
936
937
938 /**
939 * Method entry for static native method:
940 * java.lang.Double.longBitsToDouble(long bits)
941 */
942 address InterpreterGenerator::generate_Double_longBitsToDouble_entry() {
943 address entry;
944
945 if (UseSSE >= 2) {
946 entry = __ pc();
947
948 // rsi: the sender's SP
949
950 // Skip safepoint check (compiler intrinsic versions of this method
951 // do not perform safepoint checks either).
952
953 // Load 'bits' into xmm0 (interpreter returns results in xmm0)
954 __ movdbl(xmm0, Address(rsp, wordSize));
955
956 // Return
957 __ pop(rdi); // get return address
958 __ mov(rsp, rsi); // set rsp to the sender's SP
959 __ jmp(rdi);
960 } else {
961 entry = generate_native_entry(false);
962 }
963
964 return entry;
965 }
966
967 /**
968 * Method entry for static native method:
969 * java.lang.Double.doubleToRawLongBits(double value)
970 */
971 address InterpreterGenerator::generate_Double_doubleToRawLongBits_entry() {
972 address entry;
973
974 if (UseSSE >= 2) {
975 entry = __ pc();
976
977 // rsi: the sender's SP
978
979 // Skip safepoint check (compiler intrinsic versions of this method
980 // do not perform safepoint checks either).
981
982 // Load the parameter (a floating-point value) into rax.
983 __ movl(rdx, Address(rsp, 2*wordSize));
984 __ movl(rax, Address(rsp, wordSize));
985
986 // Return
987 __ pop(rdi); // get return address
988 __ mov(rsp, rsi); // set rsp to the sender's SP
989 __ jmp(rdi);
990 } else {
991 entry = generate_native_entry(false);
992 }
993
994 return entry;
995 }
996
997 //
998 // Interpreter stub for calling a native method. (asm interpreter)
999 // This sets up a somewhat different looking stack for calling the native method
1000 // than the typical interpreter frame setup.
1001 //
1002
1003 address InterpreterGenerator::generate_native_entry(bool synchronized) {
1004 // determine code generation flags
1005 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1006
1007 // rbx,: Method*
1008 // rsi: sender sp
1009 // rsi: previous interpreter state (C++ interpreter) must preserve
1010 address entry_point = __ pc();
1011
1012 const Address constMethod (rbx, Method::const_offset());
1013 const Address access_flags (rbx, Method::access_flags_offset());
1014 const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
1015
1016 // get parameter size (always needed)
1017 __ movptr(rcx, constMethod);
1018 __ load_unsigned_short(rcx, size_of_parameters);
1019
1020 // native calls don't need the stack size check since they have no expression stack
1021 // and the arguments are already on the stack and we only add a handful of words
1022 // to the stack
1023
1024 // rbx,: Method*
1025 // rcx: size of parameters
1026 // rsi: sender sp
1027
1028 __ pop(rax); // get return address
1029 // for natives the size of locals is zero
1030
1031 // compute beginning of parameters (rdi)
1032 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1033
1034
1035 // add 2 zero-initialized slots for native calls
1036 // NULL result handler
1037 __ push((int32_t)NULL_WORD);
1038 // NULL oop temp (mirror or jni oop result)
1039 __ push((int32_t)NULL_WORD);
1040
1041 // initialize fixed part of activation frame
1042 generate_fixed_frame(true);
1043
1044 // make sure method is native & not abstract
1045 #ifdef ASSERT
1046 __ movl(rax, access_flags);
1047 {
1048 Label L;
1049 __ testl(rax, JVM_ACC_NATIVE);
1050 __ jcc(Assembler::notZero, L);
1051 __ stop("tried to execute non-native method as native");
1052 __ bind(L);
1053 }
1054 { Label L;
1055 __ testl(rax, JVM_ACC_ABSTRACT);
1056 __ jcc(Assembler::zero, L);
1057 __ stop("tried to execute abstract method in interpreter");
1058 __ bind(L);
1059 }
1060 #endif
1061
1062 // Since at this point in the method invocation the exception handler
1063 // would try to exit the monitor of synchronized methods which hasn't
1064 // been entered yet, we set the thread local variable
1065 // _do_not_unlock_if_synchronized to true. The remove_activation will
1066 // check this flag.
1067
1068 __ get_thread(rax);
1069 const Address do_not_unlock_if_synchronized(rax,
1070 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1071 __ movbool(do_not_unlock_if_synchronized, true);
1072
1073 // increment invocation count & check for overflow
1074 Label invocation_counter_overflow;
1075 if (inc_counter) {
1076 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1077 }
1078
1079 Label continue_after_compile;
1080 __ bind(continue_after_compile);
1081
1082 bang_stack_shadow_pages(true);
1083
1084 // reset the _do_not_unlock_if_synchronized flag
1085 __ get_thread(rax);
1086 __ movbool(do_not_unlock_if_synchronized, false);
1087
1088 // check for synchronized methods
1089 // Must happen AFTER invocation_counter check and stack overflow check,
1090 // so method is not locked if overflows.
1091 //
1092 if (synchronized) {
1093 lock_method();
1094 } else {
1095 // no synchronization necessary
1096 #ifdef ASSERT
1097 { Label L;
1098 __ movl(rax, access_flags);
1099 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1100 __ jcc(Assembler::zero, L);
1101 __ stop("method needs synchronization");
1102 __ bind(L);
1103 }
1104 #endif
1105 }
1106
1107 // start execution
1108 #ifdef ASSERT
1109 { Label L;
1110 const Address monitor_block_top (rbp,
1111 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1112 __ movptr(rax, monitor_block_top);
1113 __ cmpptr(rax, rsp);
1114 __ jcc(Assembler::equal, L);
1115 __ stop("broken stack frame setup in interpreter");
1116 __ bind(L);
1117 }
1118 #endif
1119
1120 // jvmti/dtrace support
1121 __ notify_method_entry();
1122
1123 // work registers
1124 const Register method = rbx;
1125 const Register thread = rdi;
1126 const Register t = rcx;
1127
1128 // allocate space for parameters
1129 __ get_method(method);
1130 __ movptr(t, Address(method, Method::const_offset()));
1131 __ load_unsigned_short(t, Address(t, ConstMethod::size_of_parameters_offset()));
1132
1133 __ shlptr(t, Interpreter::logStackElementSize);
1134 __ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
1135 __ subptr(rsp, t);
1136 __ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
1137
1138 // get signature handler
1139 { Label L;
1140 __ movptr(t, Address(method, Method::signature_handler_offset()));
1141 __ testptr(t, t);
1142 __ jcc(Assembler::notZero, L);
1143 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1144 __ get_method(method);
1145 __ movptr(t, Address(method, Method::signature_handler_offset()));
1146 __ bind(L);
1147 }
1148
1149 // call signature handler
1150 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == rdi, "adjust this code");
1151 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
1152 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1153 // The generated handlers do not touch RBX (the method oop).
1154 // However, large signatures cannot be cached and are generated
1155 // each time here. The slow-path generator will blow RBX
1156 // sometime, so we must reload it after the call.
1157 __ call(t);
1158 __ get_method(method); // slow path call blows RBX on DevStudio 5.0
1159
1160 // result handler is in rax,
1161 // set result handler
1162 __ movptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize), rax);
1163
1164 // pass mirror handle if static call
1165 { Label L;
1166 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1167 __ movl(t, Address(method, Method::access_flags_offset()));
1168 __ testl(t, JVM_ACC_STATIC);
1169 __ jcc(Assembler::zero, L);
1170 // get mirror
1171 __ movptr(t, Address(method, Method:: const_offset()));
1172 __ movptr(t, Address(t, ConstMethod::constants_offset()));
1173 __ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
1174 __ movptr(t, Address(t, mirror_offset));
1175 // copy mirror into activation frame
1176 __ movptr(Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize), t);
1177 // pass handle to mirror
1178 __ lea(t, Address(rbp, frame::interpreter_frame_oop_temp_offset * wordSize));
1179 __ movptr(Address(rsp, wordSize), t);
1180 __ bind(L);
1181 }
1182
1183 // get native function entry point
1184 { Label L;
1185 __ movptr(rax, Address(method, Method::native_function_offset()));
1186 ExternalAddress unsatisfied(SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1187 __ cmpptr(rax, unsatisfied.addr());
1188 __ jcc(Assembler::notEqual, L);
1189 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1190 __ get_method(method);
1191 __ movptr(rax, Address(method, Method::native_function_offset()));
1192 __ bind(L);
1193 }
1194
1195 // pass JNIEnv
1196 __ get_thread(thread);
1197 __ lea(t, Address(thread, JavaThread::jni_environment_offset()));
1198 __ movptr(Address(rsp, 0), t);
1199
1200 // set_last_Java_frame_before_call
1201 // It is enough that the pc()
1202 // points into the right code segment. It does not have to be the correct return pc.
1203 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1204
1205 // change thread state
1206 #ifdef ASSERT
1207 { Label L;
1208 __ movl(t, Address(thread, JavaThread::thread_state_offset()));
1209 __ cmpl(t, _thread_in_Java);
1210 __ jcc(Assembler::equal, L);
1211 __ stop("Wrong thread state in native stub");
1212 __ bind(L);
1213 }
1214 #endif
1215
1216 // Change state to native
1217 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
1218 __ call(rax);
1219
1220 // result potentially in rdx:rax or ST0
1221
1222 // Verify or restore cpu control state after JNI call
1223 __ restore_cpu_control_state_after_jni();
1224
1225 // save potential result in ST(0) & rdx:rax
1226 // (if result handler is the T_FLOAT or T_DOUBLE handler, result must be in ST0 -
1227 // the check is necessary to avoid potential Intel FPU overflow problems by saving/restoring 'empty' FPU registers)
1228 // It is safe to do this push because state is _thread_in_native and return address will be found
1229 // via _last_native_pc and not via _last_jave_sp
1230
1231 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1232 // If the order changes or anything else is added to the stack the code in
1233 // interpreter_frame_result will have to be changed.
1234
1235 { Label L;
1236 Label push_double;
1237 ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
1238 ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
1239 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1240 float_handler.addr());
1241 __ jcc(Assembler::equal, push_double);
1242 __ cmpptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset + 1)*wordSize),
1243 double_handler.addr());
1244 __ jcc(Assembler::notEqual, L);
1245 __ bind(push_double);
1246 __ push_d(); // FP values are returned using the FPU, so push FPU contents (even if UseSSE > 0).
1247 __ bind(L);
1248 }
1249 __ push(ltos);
1250
1251 // change thread state
1252 __ get_thread(thread);
1253 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
1254 if(os::is_MP()) {
1255 if (UseMembar) {
1256 // Force this write out before the read below
1257 __ membar(Assembler::Membar_mask_bits(
1258 Assembler::LoadLoad | Assembler::LoadStore |
1259 Assembler::StoreLoad | Assembler::StoreStore));
1260 } else {
1261 // Write serialization page so VM thread can do a pseudo remote membar.
1262 // We use the current thread pointer to calculate a thread specific
1263 // offset to write to within the page. This minimizes bus traffic
1264 // due to cache line collision.
1265 __ serialize_memory(thread, rcx);
1266 }
1267 }
1268
1269 if (AlwaysRestoreFPU) {
1270 // Make sure the control word is correct.
1271 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1272 }
1273
1274 // check for safepoint operation in progress and/or pending suspend requests
1275 { Label Continue;
1276
1277 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
1278 SafepointSynchronize::_not_synchronized);
1279
1280 Label L;
1281 __ jcc(Assembler::notEqual, L);
1282 __ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
1283 __ jcc(Assembler::equal, Continue);
1284 __ bind(L);
1285
1286 // Don't use call_VM as it will see a possible pending exception and forward it
1287 // and never return here preventing us from clearing _last_native_pc down below.
1288 // Also can't use call_VM_leaf either as it will check to see if rsi & rdi are
1289 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1290 // by hand.
1291 //
1292 __ push(thread);
1293 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address,
1294 JavaThread::check_special_condition_for_native_trans)));
1295 __ increment(rsp, wordSize);
1296 __ get_thread(thread);
1297
1298 __ bind(Continue);
1299 }
1300
1301 // change thread state
1302 __ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
1303
1304 __ reset_last_Java_frame(thread, true, true);
1305
1306 // reset handle block
1307 __ movptr(t, Address(thread, JavaThread::active_handles_offset()));
1308 __ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), NULL_WORD);
1309
1310 // If result was an oop then unbox and save it in the frame
1311 { Label L;
1312 Label no_oop, store_result;
1313 ExternalAddress handler(AbstractInterpreter::result_handler(T_OBJECT));
1314 __ cmpptr(Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize),
1315 handler.addr());
1316 __ jcc(Assembler::notEqual, no_oop);
1317 __ cmpptr(Address(rsp, 0), (int32_t)NULL_WORD);
1318 __ pop(ltos);
1319 __ testptr(rax, rax);
1320 __ jcc(Assembler::zero, store_result);
1321 // unbox
1322 __ movptr(rax, Address(rax, 0));
1323 __ bind(store_result);
1324 __ movptr(Address(rbp, (frame::interpreter_frame_oop_temp_offset)*wordSize), rax);
1325 // keep stack depth as expected by pushing oop which will eventually be discarded
1326 __ push(ltos);
1327 __ bind(no_oop);
1328 }
1329
1330 {
1331 Label no_reguard;
1332 __ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
1333 __ jcc(Assembler::notEqual, no_reguard);
1334
1335 __ pusha();
1336 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
1337 __ popa();
1338
1339 __ bind(no_reguard);
1340 }
1341
1342 // restore rsi to have legal interpreter frame,
1343 // i.e., bci == 0 <=> rsi == code_base()
1344 // Can't call_VM until bcp is within reasonable.
1345 __ get_method(method); // method is junk from thread_in_native to now.
1346 __ movptr(rsi, Address(method,Method::const_offset())); // get ConstMethod*
1347 __ lea(rsi, Address(rsi,ConstMethod::codes_offset())); // get codebase
1348
1349 // handle exceptions (exception handling will handle unlocking!)
1350 { Label L;
1351 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
1352 __ jcc(Assembler::zero, L);
1353 // Note: At some point we may want to unify this with the code used in call_VM_base();
1354 // i.e., we should use the StubRoutines::forward_exception code. For now this
1355 // doesn't work here because the rsp is not correctly set at this point.
1356 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
1357 __ should_not_reach_here();
1358 __ bind(L);
1359 }
1360
1361 // do unlocking if necessary
1362 { Label L;
1363 __ movl(t, Address(method, Method::access_flags_offset()));
1364 __ testl(t, JVM_ACC_SYNCHRONIZED);
1365 __ jcc(Assembler::zero, L);
1366 // the code below should be shared with interpreter macro assembler implementation
1367 { Label unlock;
1368 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1369 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1370 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1371
1372 __ lea(rdx, monitor); // address of first monitor
1373
1374 __ movptr(t, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
1375 __ testptr(t, t);
1376 __ jcc(Assembler::notZero, unlock);
1377
1378 // Entry already unlocked, need to throw exception
1379 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1380 __ should_not_reach_here();
1381
1382 __ bind(unlock);
1383 __ unlock_object(rdx);
1384 }
1385 __ bind(L);
1386 }
1387
1388 // jvmti/dtrace support
1389 // Note: This must happen _after_ handling/throwing any exceptions since
1390 // the exception handler code notifies the runtime of method exits
1391 // too. If this happens before, method entry/exit notifications are
1392 // not properly paired (was bug - gri 11/22/99).
1393 __ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
1394
1395 // restore potential result in rdx:rax, call result handler to restore potential result in ST0 & handle result
1396 __ pop(ltos);
1397 __ movptr(t, Address(rbp, frame::interpreter_frame_result_handler_offset*wordSize));
1398 __ call(t);
1399
1400 // remove activation
1401 __ movptr(t, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1402 __ leave(); // remove frame anchor
1403 __ pop(rdi); // get return address
1404 __ mov(rsp, t); // set sp to sender sp
1405 __ jmp(rdi);
1406
1407 if (inc_counter) {
1408 // Handle overflow of counter and compile method
1409 __ bind(invocation_counter_overflow);
1410 generate_counter_overflow(&continue_after_compile);
1411 }
1412
1413 return entry_point;
1414 }
1415
1416 //
1417 // Generic interpreted method entry to (asm) interpreter
1418 //
1419 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1420 // determine code generation flags
1421 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods;
1422
1423 // rbx,: Method*
1424 // rsi: sender sp
1425 address entry_point = __ pc();
1426
1427 const Address constMethod (rbx, Method::const_offset());
1428 const Address access_flags (rbx, Method::access_flags_offset());
1429 const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
1430 const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
1431
1432 // get parameter size (always needed)
1433 __ movptr(rdx, constMethod);
1434 __ load_unsigned_short(rcx, size_of_parameters);
1435
1436 // rbx,: Method*
1437 // rcx: size of parameters
1438
1439 // rsi: sender_sp (could differ from sp+wordSize if we were called via c2i )
1440
1441 __ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
1442 __ subl(rdx, rcx); // rdx = no. of additional locals
1443
1444 // see if we've got enough room on the stack for locals plus overhead.
1445 generate_stack_overflow_check();
1446
1447 // get return address
1448 __ pop(rax);
1449
1450 // compute beginning of parameters (rdi)
1451 __ lea(rdi, Address(rsp, rcx, Interpreter::stackElementScale(), -wordSize));
1452
1453 // rdx - # of additional locals
1454 // allocate space for locals
1455 // explicitly initialize locals
1456 {
1457 Label exit, loop;
1458 __ testl(rdx, rdx);
1459 __ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
1460 __ bind(loop);
1461 __ push((int32_t)NULL_WORD); // initialize local variables
1462 __ decrement(rdx); // until everything initialized
1463 __ jcc(Assembler::greater, loop);
1464 __ bind(exit);
1465 }
1466
1467 // initialize fixed part of activation frame
1468 generate_fixed_frame(false);
1469
1470 // make sure method is not native & not abstract
1471 #ifdef ASSERT
1472 __ movl(rax, access_flags);
1473 {
1474 Label L;
1475 __ testl(rax, JVM_ACC_NATIVE);
1476 __ jcc(Assembler::zero, L);
1477 __ stop("tried to execute native method as non-native");
1478 __ bind(L);
1479 }
1480 { Label L;
1481 __ testl(rax, JVM_ACC_ABSTRACT);
1482 __ jcc(Assembler::zero, L);
1483 __ stop("tried to execute abstract method in interpreter");
1484 __ bind(L);
1485 }
1486 #endif
1487
1488 // Since at this point in the method invocation the exception handler
1489 // would try to exit the monitor of synchronized methods which hasn't
1490 // been entered yet, we set the thread local variable
1491 // _do_not_unlock_if_synchronized to true. The remove_activation will
1492 // check this flag.
1493
1494 __ get_thread(rax);
1495 const Address do_not_unlock_if_synchronized(rax,
1496 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1497 __ movbool(do_not_unlock_if_synchronized, true);
1498
1499 __ profile_parameters_type(rax, rcx, rdx);
1500 // increment invocation count & check for overflow
1501 Label invocation_counter_overflow;
1502 Label profile_method;
1503 Label profile_method_continue;
1504 if (inc_counter) {
1505 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue);
1506 if (ProfileInterpreter) {
1507 __ bind(profile_method_continue);
1508 }
1509 }
1510 Label continue_after_compile;
1511 __ bind(continue_after_compile);
1512
1513 bang_stack_shadow_pages(false);
1514
1515 // reset the _do_not_unlock_if_synchronized flag
1516 __ get_thread(rax);
1517 __ movbool(do_not_unlock_if_synchronized, false);
1518
1519 // check for synchronized methods
1520 // Must happen AFTER invocation_counter check and stack overflow check,
1521 // so method is not locked if overflows.
1522 //
1523 if (synchronized) {
1524 // Allocate monitor and lock method
1525 lock_method();
1526 } else {
1527 // no synchronization necessary
1528 #ifdef ASSERT
1529 { Label L;
1530 __ movl(rax, access_flags);
1531 __ testl(rax, JVM_ACC_SYNCHRONIZED);
1532 __ jcc(Assembler::zero, L);
1533 __ stop("method needs synchronization");
1534 __ bind(L);
1535 }
1536 #endif
1537 }
1538
1539 // start execution
1540 #ifdef ASSERT
1541 { Label L;
1542 const Address monitor_block_top (rbp,
1543 frame::interpreter_frame_monitor_block_top_offset * wordSize);
1544 __ movptr(rax, monitor_block_top);
1545 __ cmpptr(rax, rsp);
1546 __ jcc(Assembler::equal, L);
1547 __ stop("broken stack frame setup in interpreter");
1548 __ bind(L);
1549 }
1550 #endif
1551
1552 // jvmti support
1553 __ notify_method_entry();
1554
1555 __ dispatch_next(vtos);
1556
1557 // invocation counter overflow
1558 if (inc_counter) {
1559 if (ProfileInterpreter) {
1560 // We have decided to profile this method in the interpreter
1561 __ bind(profile_method);
1562 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1563 __ set_method_data_pointer_for_bcp();
1564 __ get_method(rbx);
1565 __ jmp(profile_method_continue);
1566 }
1567 // Handle overflow of counter and compile method
1568 __ bind(invocation_counter_overflow);
1569 generate_counter_overflow(&continue_after_compile);
1570 }
1571
1572 return entry_point;
1573 }
1574
1575
1576 // These should never be compiled since the interpreter will prefer
1577 // the compiled version to the intrinsic version.
1578 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1579 switch (method_kind(m)) {
1580 case Interpreter::java_lang_math_sin : // fall thru
1581 case Interpreter::java_lang_math_cos : // fall thru
1582 case Interpreter::java_lang_math_tan : // fall thru
1583 case Interpreter::java_lang_math_abs : // fall thru
1584 case Interpreter::java_lang_math_log : // fall thru
1585 case Interpreter::java_lang_math_log10 : // fall thru
1586 case Interpreter::java_lang_math_sqrt : // fall thru
1587 case Interpreter::java_lang_math_pow : // fall thru
1588 case Interpreter::java_lang_math_exp :
1589 return false;
1590 default:
1591 return true;
1592 }
1593 }
1594
1595 // How much stack a method activation needs in words.
1596 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1597
1598 const int stub_code = 4; // see generate_call_stub
1599 // Save space for one monitor to get into the interpreted method in case
1600 // the method is synchronized
1601 int monitor_size = method->is_synchronized() ?
1602 1*frame::interpreter_frame_monitor_size() : 0;
1603
1604 // total overhead size: entry_size + (saved rbp, thru expr stack bottom).
1605 // be sure to change this if you add/subtract anything to/from the overhead area
1606 const int overhead_size = -frame::interpreter_frame_initial_sp_offset;
1607
1608 const int method_stack = (method->max_locals() + method->max_stack()) *
1609 Interpreter::stackElementWords;
1610 return overhead_size + method_stack + stub_code;
1611 }
1612
1613 //------------------------------------------------------------------------------------------------------------------------
1614 // Exceptions
1615
1616 void TemplateInterpreterGenerator::generate_throw_exception() {
1617 // Entry point in previous activation (i.e., if the caller was interpreted)
1618 Interpreter::_rethrow_exception_entry = __ pc();
1619 const Register thread = rcx;
1620
1621 // Restore sp to interpreter_frame_last_sp even though we are going
1622 // to empty the expression stack for the exception processing.
1623 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1624 // rax,: exception
1625 // rdx: return address/pc that threw exception
1626 __ restore_bcp(); // rsi points to call/send
1627 __ restore_locals();
1628
1629 // Entry point for exceptions thrown within interpreter code
1630 Interpreter::_throw_exception_entry = __ pc();
1631 // expression stack is undefined here
1632 // rax,: exception
1633 // rsi: exception bcp
1634 __ verify_oop(rax);
1635
1636 // expression stack must be empty before entering the VM in case of an exception
1637 __ empty_expression_stack();
1638 __ empty_FPU_stack();
1639 // find exception handler address and preserve exception oop
1640 __ call_VM(rdx, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), rax);
1641 // rax,: exception handler entry point
1642 // rdx: preserved exception oop
1643 // rsi: bcp for exception handler
1644 __ push_ptr(rdx); // push exception which is now the only value on the stack
1645 __ jmp(rax); // jump to exception handler (may be _remove_activation_entry!)
1646
1647 // If the exception is not handled in the current frame the frame is removed and
1648 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1649 //
1650 // Note: At this point the bci is still the bxi for the instruction which caused
1651 // the exception and the expression stack is empty. Thus, for any VM calls
1652 // at this point, GC will find a legal oop map (with empty expression stack).
1653
1654 // In current activation
1655 // tos: exception
1656 // rsi: exception bcp
1657
1658 //
1659 // JVMTI PopFrame support
1660 //
1661
1662 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1663 __ empty_expression_stack();
1664 __ empty_FPU_stack();
1665 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1666 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1667 // popframe handling cycles.
1668 __ get_thread(thread);
1669 __ movl(rdx, Address(thread, JavaThread::popframe_condition_offset()));
1670 __ orl(rdx, JavaThread::popframe_processing_bit);
1671 __ movl(Address(thread, JavaThread::popframe_condition_offset()), rdx);
1672
1673 {
1674 // Check to see whether we are returning to a deoptimized frame.
1675 // (The PopFrame call ensures that the caller of the popped frame is
1676 // either interpreted or compiled and deoptimizes it if compiled.)
1677 // In this case, we can't call dispatch_next() after the frame is
1678 // popped, but instead must save the incoming arguments and restore
1679 // them after deoptimization has occurred.
1680 //
1681 // Note that we don't compare the return PC against the
1682 // deoptimization blob's unpack entry because of the presence of
1683 // adapter frames in C2.
1684 Label caller_not_deoptimized;
1685 __ movptr(rdx, Address(rbp, frame::return_addr_offset * wordSize));
1686 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), rdx);
1687 __ testl(rax, rax);
1688 __ jcc(Assembler::notZero, caller_not_deoptimized);
1689
1690 // Compute size of arguments for saving when returning to deoptimized caller
1691 __ get_method(rax);
1692 __ movptr(rax, Address(rax, Method::const_offset()));
1693 __ load_unsigned_short(rax, Address(rax, ConstMethod::size_of_parameters_offset()));
1694 __ shlptr(rax, Interpreter::logStackElementSize);
1695 __ restore_locals();
1696 __ subptr(rdi, rax);
1697 __ addptr(rdi, wordSize);
1698 // Save these arguments
1699 __ get_thread(thread);
1700 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), thread, rax, rdi);
1701
1702 __ remove_activation(vtos, rdx,
1703 /* throw_monitor_exception */ false,
1704 /* install_monitor_exception */ false,
1705 /* notify_jvmdi */ false);
1706
1707 // Inform deoptimization that it is responsible for restoring these arguments
1708 __ get_thread(thread);
1709 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_force_deopt_reexecution_bit);
1710
1711 // Continue in deoptimization handler
1712 __ jmp(rdx);
1713
1714 __ bind(caller_not_deoptimized);
1715 }
1716
1717 __ remove_activation(vtos, rdx,
1718 /* throw_monitor_exception */ false,
1719 /* install_monitor_exception */ false,
1720 /* notify_jvmdi */ false);
1721
1722 // Finish with popframe handling
1723 // A previous I2C followed by a deoptimization might have moved the
1724 // outgoing arguments further up the stack. PopFrame expects the
1725 // mutations to those outgoing arguments to be preserved and other
1726 // constraints basically require this frame to look exactly as
1727 // though it had previously invoked an interpreted activation with
1728 // no space between the top of the expression stack (current
1729 // last_sp) and the top of stack. Rather than force deopt to
1730 // maintain this kind of invariant all the time we call a small
1731 // fixup routine to move the mutated arguments onto the top of our
1732 // expression stack if necessary.
1733 __ mov(rax, rsp);
1734 __ movptr(rbx, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1735 __ get_thread(thread);
1736 // PC must point into interpreter here
1737 __ set_last_Java_frame(thread, noreg, rbp, __ pc());
1738 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, rax, rbx);
1739 __ get_thread(thread);
1740 __ reset_last_Java_frame(thread, true, true);
1741 // Restore the last_sp and null it out
1742 __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
1743 __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD);
1744
1745 __ restore_bcp();
1746 __ restore_locals();
1747 // The method data pointer was incremented already during
1748 // call profiling. We have to restore the mdp for the current bcp.
1749 if (ProfileInterpreter) {
1750 __ set_method_data_pointer_for_bcp();
1751 }
1752
1753 // Clear the popframe condition flag
1754 __ get_thread(thread);
1755 __ movl(Address(thread, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
1756
1757 #if INCLUDE_JVMTI
1758 {
1759 Label L_done;
1760 const Register local0 = rdi;
1761
1762 __ cmpb(Address(rsi, 0), Bytecodes::_invokestatic);
1763 __ jcc(Assembler::notEqual, L_done);
1764
1765 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
1766 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
1767
1768 __ get_method(rdx);
1769 __ movptr(rax, Address(local0, 0));
1770 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), rax, rdx, rsi);
1771
1772 __ testptr(rax, rax);
1773 __ jcc(Assembler::zero, L_done);
1774
1775 __ movptr(Address(rbx, 0), rax);
1776 __ bind(L_done);
1777 }
1778 #endif // INCLUDE_JVMTI
1779
1780 __ dispatch_next(vtos);
1781 // end of PopFrame support
1782
1783 Interpreter::_remove_activation_entry = __ pc();
1784
1785 // preserve exception over this code sequence
1786 __ pop_ptr(rax);
1787 __ get_thread(thread);
1788 __ movptr(Address(thread, JavaThread::vm_result_offset()), rax);
1789 // remove the activation (without doing throws on illegalMonitorExceptions)
1790 __ remove_activation(vtos, rdx, false, true, false);
1791 // restore exception
1792 __ get_thread(thread);
1793 __ get_vm_result(rax, thread);
1794
1795 // Inbetween activations - previous activation type unknown yet
1796 // compute continuation point - the continuation point expects
1797 // the following registers set up:
1798 //
1799 // rax: exception
1800 // rdx: return address/pc that threw exception
1801 // rsp: expression stack of caller
1802 // rbp: rbp, of caller
1803 __ push(rax); // save exception
1804 __ push(rdx); // save return address
1805 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, rdx);
1806 __ mov(rbx, rax); // save exception handler
1807 __ pop(rdx); // restore return address
1808 __ pop(rax); // restore exception
1809 // Note that an "issuing PC" is actually the next PC after the call
1810 __ jmp(rbx); // jump to exception handler of caller
1811 }
1812
1813
1814 //
1815 // JVMTI ForceEarlyReturn support
1816 //
1817 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
1818 address entry = __ pc();
1819 const Register thread = rcx;
1820
1821 __ restore_bcp();
1822 __ restore_locals();
1823 __ empty_expression_stack();
1824 __ empty_FPU_stack();
1825 __ load_earlyret_value(state);
1826
1827 __ get_thread(thread);
1828 __ movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
1829 const Address cond_addr(rcx, JvmtiThreadState::earlyret_state_offset());
1830
1831 // Clear the earlyret state
1832 __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
1833
1834 __ remove_activation(state, rsi,
1835 false, /* throw_monitor_exception */
1836 false, /* install_monitor_exception */
1837 true); /* notify_jvmdi */
1838 __ jmp(rsi);
1839 return entry;
1840 } // end of ForceEarlyReturn support
1841
1842
1843 //------------------------------------------------------------------------------------------------------------------------
1844 // Helper for vtos entry point generation
1845
1846 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) {
1847 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
1848 Label L;
1849 fep = __ pc(); __ push(ftos); __ jmp(L);
1850 dep = __ pc(); __ push(dtos); __ jmp(L);
1851 lep = __ pc(); __ push(ltos); __ jmp(L);
1852 aep = __ pc(); __ push(atos); __ jmp(L);
1853 bep = cep = sep = // fall through
1854 iep = __ pc(); __ push(itos); // fall through
1855 vep = __ pc(); __ bind(L); // fall through
1856 generate_and_dispatch(t);
1857 }
1858
1859 //------------------------------------------------------------------------------------------------------------------------
1860 // Generation of individual instructions
1861
1862 // helpers for generate_and_dispatch
1863
1864
1865
1866 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
1867 : TemplateInterpreterGenerator(code) {
1868 generate_all(); // down here so it can be "virtual"
1869 }
1870
1871 //------------------------------------------------------------------------------------------------------------------------
1872
1873 // Non-product code
1874 #ifndef PRODUCT
1875 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
1876 address entry = __ pc();
1877
1878 // prepare expression stack
1879 __ pop(rcx); // pop return address so expression stack is 'pure'
1880 __ push(state); // save tosca
1881
1882 // pass tosca registers as arguments & call tracer
1883 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), rcx, rax, rdx);
1884 __ mov(rcx, rax); // make sure return address is not destroyed by pop(state)
1885 __ pop(state); // restore tosca
1886
1887 // return
1888 __ jmp(rcx);
1889
1890 return entry;
1891 }
1892
1893
1894 void TemplateInterpreterGenerator::count_bytecode() {
1895 __ incrementl(ExternalAddress((address) &BytecodeCounter::_counter_value));
1896 }
1897
1898
1899 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
1900 __ incrementl(ExternalAddress((address) &BytecodeHistogram::_counters[t->bytecode()]));
1901 }
1902
1903
1904 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
1905 __ mov32(ExternalAddress((address) &BytecodePairHistogram::_index), rbx);
1906 __ shrl(rbx, BytecodePairHistogram::log2_number_of_codes);
1907 __ orl(rbx, ((int)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
1908 ExternalAddress table((address) BytecodePairHistogram::_counters);
1909 Address index(noreg, rbx, Address::times_4);
1910 __ incrementl(ArrayAddress(table, index));
1911 }
1912
1913
1914 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
1915 // Call a little run-time stub to avoid blow-up for each bytecode.
1916 // The run-time runtime saves the right registers, depending on
1917 // the tosca in-state for the given template.
1918 assert(Interpreter::trace_code(t->tos_in()) != NULL,
1919 "entry must have been generated");
1920 __ call(RuntimeAddress(Interpreter::trace_code(t->tos_in())));
1921 }
1922
1923
1924 void TemplateInterpreterGenerator::stop_interpreter_at() {
1925 Label L;
1926 __ cmp32(ExternalAddress((address) &BytecodeCounter::_counter_value),
1927 StopInterpreterAt);
1928 __ jcc(Assembler::notEqual, L);
1929 __ int3();
1930 __ bind(L);
1931 }
1932 #endif // !PRODUCT
1933 #endif // CC_INTERP