Thu, 10 Nov 2016 10:14:33 +0800
[Interpreter] Remove redundant sync operations.
1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2016, Loongson Technology. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.hpp"
28 #include "interpreter/bytecodeHistogram.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterGenerator.hpp"
31 #include "interpreter/interpreterRuntime.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"
49 #define __ _masm->
51 #ifndef CC_INTERP
53 // asm based interpreter deoptimization helpers
54 int AbstractInterpreter::size_activation(int max_stack,
55 int temps,
56 int extra_args,
57 int monitors,
58 int callee_params,
59 int callee_locals,
60 bool is_top_frame) {
61 // Note: This calculation must exactly parallel the frame setup
62 // in AbstractInterpreterGenerator::generate_method_entry.
64 // fixed size of an interpreter frame:
65 int overhead = frame::sender_sp_offset -
66 frame::interpreter_frame_initial_sp_offset;
67 // Our locals were accounted for by the caller (or last_frame_adjust
68 // on the transistion) Since the callee parameters already account
69 // for the callee's params we only need to account for the extra
70 // locals.
71 int size = overhead +
72 (callee_locals - callee_params)*Interpreter::stackElementWords +
73 monitors * frame::interpreter_frame_monitor_size() +
74 temps* Interpreter::stackElementWords + extra_args;
76 return size;
77 }
80 const int Interpreter::return_sentinel = 0xfeedbeed;
81 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
82 const int bci_offset = frame::interpreter_frame_bcx_offset * wordSize;
83 const int locals_offset = frame::interpreter_frame_locals_offset * wordSize;
85 //-----------------------------------------------------------------------------
87 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() {
88 address entry = __ pc();
90 #ifdef ASSERT
91 {
92 Label L;
93 __ addi(T1, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
94 __ sub(T1, T1, SP); // T1 = maximal sp for current fp
95 __ bgez(T1, L); // check if frame is complete
96 __ delayed()->nop();
97 __ stop("interpreter frame not set up");
98 __ bind(L);
99 }
100 #endif // ASSERT
101 // Restore bcp under the assumption that the current frame is still
102 // interpreted
103 // FIXME: please change the func restore_bcp
104 // S0 is the conventional register for bcp
105 __ restore_bcp();
107 // expression stack must be empty before entering the VM if an
108 // exception happened
109 __ empty_expression_stack();
110 // throw exception
111 //__ call_VM(noreg,
112 // CAST_FROM_FN_PTR(address,
113 // InterpreterRuntime::throw_StackOverflowError));
114 // FIXME: why do not pass parameter thread ?
115 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
116 return entry;
117 }
119 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
120 const char* name) {
121 address entry = __ pc();
122 // expression stack must be empty before entering the VM if an
123 // exception happened
124 __ empty_expression_stack();
125 __ li(A1, (long)name);
126 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
127 InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), A1, A2);
128 return entry;
129 }
131 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
132 address entry = __ pc();
134 // object is at TOS
135 //FIXME, I am not sure if the object is at TOS as x86 do now @jerome, 04/20,2007
136 //__ pop(c_rarg1);
138 // expression stack must be empty before entering the VM if an
139 // exception happened
140 __ empty_expression_stack();
141 __ empty_FPU_stack();
142 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), FSR);
143 return entry;
144 }
146 address TemplateInterpreterGenerator::generate_exception_handler_common(
147 const char* name, const char* message, bool pass_oop) {
148 assert(!pass_oop || message == NULL, "either oop or message but not both");
149 address entry = __ pc();
151 // expression stack must be empty before entering the VM if an exception happened
152 __ empty_expression_stack();
153 // setup parameters
154 __ li(A1, (long)name);
155 if (pass_oop) {
156 __ call_VM(V0,
157 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), A1, FSR);
158 } else {
159 __ li(A2, (long)message);
160 __ call_VM(V0,
161 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), A1, A2);
162 }
163 // throw exception
164 __ jmp(Interpreter::throw_exception_entry(), relocInfo::none);
165 __ delayed()->nop();
166 return entry;
167 }
170 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
171 address entry = __ pc();
172 // NULL last_sp until next java call
173 __ sd(R0,Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
174 __ dispatch_next(state);
175 return entry;
176 }
179 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
181 address entry = __ pc();
183 // Restore stack bottom in case i2c adjusted stack
184 // __ movptr(rsp, Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize));
185 __ ld(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
186 // and NULL it as marker that esp is now tos until next java call
187 // __ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
188 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
190 __ restore_bcp();
191 __ restore_locals();
193 // 2014/11/24 Fu
194 // mdp: T8
195 // ret: FSR
196 // tmp: T9
197 if (state == atos) {
198 Register mdp = T8;
199 Register tmp = T9;
200 __ profile_return_type(mdp, FSR, tmp);
201 }
204 const Register cache = T9;
205 const Register index = T3;
206 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
208 const Register flags = cache;
209 // __ movl(flags, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
210 __ dsll(AT, index, Address::times_ptr);
211 __ daddu(AT, cache, AT);
212 __ lw(flags, AT, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
213 // __ andl(flags, ConstantPoolCacheEntry::parameter_size_mask);
214 __ andi(flags, flags, ConstantPoolCacheEntry::parameter_size_mask);
215 // __ lea(rsp, Address(rsp, flags, Interpreter::stackElementScale()));
216 __ dsll(AT, flags, Interpreter::stackElementScale());
217 __ daddu(SP, SP, AT);
219 __ dispatch_next(state, step);
221 return entry;
222 }
225 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
226 int step) {
227 address entry = __ pc();
228 // NULL last_sp until next java call
229 //__ movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
230 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
231 __ restore_bcp();
232 __ restore_locals();
233 // handle exceptions
234 {
235 Label L;
236 const Register thread = TREG;
237 #ifndef OPT_THREAD
238 __ get_thread(thread);
239 #endif
240 __ lw(AT, thread, in_bytes(Thread::pending_exception_offset()));
241 __ beq(AT, R0, L);
242 __ delayed()->nop();
243 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
244 __ should_not_reach_here();
245 __ bind(L);
246 }
247 __ dispatch_next(state, step);
248 return entry;
249 }
251 int AbstractInterpreter::BasicType_as_index(BasicType type) {
252 int i = 0;
253 /*
254 switch (type) {
255 case T_BOOLEAN: i = 0; break;
256 case T_CHAR : i = 1; break;
257 case T_BYTE : i = 2; break;
258 case T_SHORT : i = 3; break;
259 case T_INT : i = 4; break;
260 case T_LONG : i = 5; break;
261 case T_VOID : i = 6; break;
262 case T_FLOAT : i = 7; break;
263 case T_DOUBLE : i = 8; break;
264 case T_OBJECT : i = 9; break;
265 case T_ARRAY : i = 9; break;
266 default : ShouldNotReachHere();
267 }
268 */
269 switch (type) {
270 case T_BOOLEAN: i = 0; break;
271 case T_CHAR : i = 1; break;
272 case T_BYTE : i = 2; break;
273 case T_SHORT : i = 3; break;
274 case T_INT : // fall through
275 case T_LONG : // fall through
276 case T_VOID : i = 4; break;
277 case T_FLOAT : i = 5; break;
278 case T_DOUBLE : i = 6; break;
279 case T_OBJECT : // fall through
280 case T_ARRAY : i = 7; break;
281 default : ShouldNotReachHere();
282 }
283 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
284 "index out of bounds");
285 return i;
286 }
289 // why do not consider float and double , @jerome, 12/27,06, @jerome
290 //FIXME, aoqi
291 address TemplateInterpreterGenerator::generate_result_handler_for(
292 BasicType type) {
293 address entry = __ pc();
294 switch (type) {
295 case T_BOOLEAN: __ c2bool(V0); break;
296 case T_CHAR : __ andi(V0, V0, 0xFFFF); break;
297 case T_BYTE : __ sign_extend_byte (V0); break;
298 case T_SHORT : __ sign_extend_short(V0); break;
299 case T_INT : /* nothing to do */ break;
300 case T_FLOAT : /* nothing to do */ break;
301 case T_DOUBLE : /* nothing to do */ break;
302 case T_OBJECT :
303 {
304 // __ beq(V0, R0, L); // test if NULL handle
305 // __ delayed()->nop(); // if not then
306 // __ lw(V0, V0, 0); // unbox result
307 __ ld(V0, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
308 __ verify_oop(V0); // and verify it
309 }
310 break;
311 default : ShouldNotReachHere();
312 }
313 __ jr(RA); // return from result handler
314 __ delayed()->nop();
315 return entry;
316 }
318 address TemplateInterpreterGenerator::generate_safept_entry_for(
319 TosState state,
320 address runtime_entry) {
321 address entry = __ pc();
322 __ push(state);
323 __ call_VM(noreg, runtime_entry);
324 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
325 return entry;
326 }
330 // Helpers for commoning out cases in the various type of method entries.
331 //
334 // increment invocation count & check for overflow
335 //
336 // Note: checking for negative value instead of overflow
337 // so we have a 'sticky' overflow test
338 //
339 // prerequisites : method in T0, invocation counter in T3
340 void InterpreterGenerator::generate_counter_incr(
341 Label* overflow,
342 Label* profile_method,
343 Label* profile_method_continue) {
344 Label done;
345 const Address invocation_counter(FSR, in_bytes(MethodCounters::invocation_counter_offset()) //Fu:20130814 Wang:Rmethod --> FSR
346 + in_bytes(InvocationCounter::counter_offset()));
347 const Address backedge_counter (FSR, in_bytes(MethodCounters::backedge_counter_offset())
348 + in_bytes(InvocationCounter::counter_offset()));
350 __ get_method_counters(Rmethod, FSR, done);
352 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
353 __ lw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
354 __ incrementl(T9, 1);
355 __ sw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
356 }
357 // Update standard invocation counters
358 __ lw(T3, invocation_counter);
359 __ increment(T3, InvocationCounter::count_increment);
360 __ sw(T3, invocation_counter); // save invocation count
362 __ lw(FSR, backedge_counter); // load backedge counter
363 __ li(AT, InvocationCounter::count_mask_value); // mask out the status bits
364 __ andr(FSR, FSR, AT);
366 __ dadd(T3, T3, FSR); // add both counters
368 // profile_method is non-null only for interpreted method so
369 // profile_method != NULL == !native_call
371 if (ProfileInterpreter && profile_method != NULL) {
372 // Test to see if we should create a method data oop
373 /*
374 __ lui(AT, Assembler::split_high(
375 int(&InvocationCounter::InterpreterProfileLimit)));
376 __ ld(AT, AT, Assembler::split_low(
377 int(&InvocationCounter::InterpreterProfileLimit)));
378 */
379 __ li(AT, (long)&InvocationCounter::InterpreterProfileLimit);
380 __ lw(AT, AT, 0);
381 __ slt(AT, T3, AT);
382 __ bne_far(AT, R0, *profile_method_continue);
383 __ delayed()->nop();
385 // if no method data exists, go to profile_method
386 __ test_method_data_pointer(FSR, *profile_method);
387 }
389 //__ lui(AT, Assembler::split_high(intptr_t(&InvocationCounter::InterpreterInvocationLimit)));
390 //__ ld(AT, AT, Assembler::split_low(intptr_t(&InvocationCounter::InterpreterInvocationLimit)));
391 __ li(AT, (long)&InvocationCounter::InterpreterInvocationLimit);
392 __ lw(AT, AT, 0);
393 __ slt(AT, T3, AT);
394 __ beq_far(AT, R0, *overflow);
395 __ delayed()->nop();
396 __ bind(done);
397 }
399 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
401 // Asm interpreter on entry
402 // S7 - locals
403 // S0 - bcp
404 // Rmethod - method
405 // FP - interpreter frame
407 // On return (i.e. jump to entry_point)
408 // Rmethod - method
409 // RA - return address of interpreter caller
410 // tos - the last parameter to Java method
411 // SP - sender_sp
413 //const Address size_of_parameters(Rmethod,in_bytes( Method::size_of_parameters_offset()));
415 // the bcp is valid if and only if it's not null
416 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
417 InterpreterRuntime::frequency_counter_overflow), R0);
418 __ ld(Rmethod, FP, method_offset);
419 /*
420 // method has been compiled - remove activation frame
421 // (leave return address on stack) and continue at
422 // verified entry point (eax). (eax in some past life maybe, seems to use methodoop these days)
423 //
424 // Note: continuation at verified entry point works if the method that has been
425 // compiled is the right one (in case of virtual calls); i.e., the inline
426 // cache check must have happened before the invocation counter overflow
427 // check.
428 __ lhu(V0, size_of_parameters);
429 __ move(SP, FP);
430 __ lw(FP, SP, frame::interpreter_frame_sender_fp_offset * wordSize);
431 __ lw(RA, SP, frame::interpreter_frame_return_addr_offset * wordSize);
432 __ sll(V0, V0, 2);
433 __ addi(V0, V0, - 1 * wordSize);
434 __ sub(SP, LVP, V0);
435 // __ lw(T0, LVP, 0);
436 */
437 // Preserve invariant that esi/edi contain bcp/locals of sender frame
438 __ b_far(*do_continue);
439 __ delayed()->nop();
440 }
442 // See if we've got enough room on the stack for locals plus overhead.
443 // The expression stack grows down incrementally, so the normal guard
444 // page mechanism will work for that.
445 //
446 // NOTE: Since the additional locals are also always pushed (wasn't
447 // obvious in generate_method_entry) so the guard should work for them
448 // too.
449 //
450 // Args:
451 // rdx: number of additional locals this frame needs (what we must check)
452 // rbx: Method*
453 //
454 // Kills:
455 // rax
456 void InterpreterGenerator::generate_stack_overflow_check(void) {
457 // see if we've got enough room on the stack for locals plus overhead.
458 // the expression stack grows down incrementally, so the normal guard
459 // page mechanism will work for that.
460 //
461 // Registers live on entry:
462 //
463 // T0: Method*
464 // T2: number of additional locals this frame needs (what we must check)
466 // NOTE: since the additional locals are also always pushed (wasn't obvious in
467 // generate_method_entry) so the guard should work for them too.
468 //
470 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp
471 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
473 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
474 // be sure to change this if you add/subtract anything to/from the overhead area
475 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize)
476 + entry_size;
478 const int page_size = os::vm_page_size();
480 Label after_frame_check;
482 // see if the frame is greater than one page in size. If so,
483 // then we need to verify there is enough stack space remaining
484 // for the additional locals.
485 __ move(AT, (page_size - overhead_size) / Interpreter::stackElementSize);
486 __ slt(AT, AT, T2);
487 __ beq(AT, R0, after_frame_check);
488 __ delayed()->nop();
490 // compute sp as if this were going to be the last frame on
491 // the stack before the red zone
492 #ifndef OPT_THREAD
493 Register thread = T1;
494 __ get_thread(thread);
495 #else
496 Register thread = TREG;
497 #endif
499 // locals + overhead, in bytes
500 //FIXME aoqi
501 __ dsll(T3, T2, Interpreter::stackElementScale());
502 //__ dsll(T9, T9, Address::times_8);
503 __ daddiu(T3, T3, overhead_size); // locals * 4 + overhead_size --> T3
505 #ifdef ASSERT
506 Label stack_base_okay, stack_size_okay;
507 // verify that thread stack base is non-zero
508 __ ld(AT, thread, in_bytes(Thread::stack_base_offset()));
509 __ bne(AT, R0, stack_base_okay);
510 __ delayed()->nop();
511 __ stop("stack base is zero");
512 __ bind(stack_base_okay);
513 // verify that thread stack size is non-zero
514 __ ld(AT, thread, in_bytes(Thread::stack_size_offset()));
515 __ bne(AT, R0, stack_size_okay);
516 __ delayed()->nop();
517 __ stop("stack size is zero");
518 __ bind(stack_size_okay);
519 #endif
521 // Add stack base to locals and subtract stack size
522 __ ld(AT, thread, in_bytes(Thread::stack_base_offset())); // stack_base --> AT
523 __ dadd(T3, T3, AT); // locals * 4 + overhead_size + stack_base--> T3
524 __ ld(AT, thread, in_bytes(Thread::stack_size_offset())); // stack_size --> AT
525 __ dsub(T3, T3, AT); // locals * 4 + overhead_size + stack_base - stack_size --> T3
528 // add in the redzone and yellow size
529 __ move(AT, (StackRedPages+StackYellowPages) * page_size);
530 __ add(T3, T3, AT);
532 // check against the current stack bottom
533 __ slt(AT, T3, SP);
534 __ bne(AT, R0, after_frame_check);
535 __ delayed()->nop();
536 // x86 version pop saved bcp and return address here, FIXME
537 __ jmp(Interpreter::throw_StackOverflowError_entry(), relocInfo::runtime_call_type);
538 __ delayed()->nop();
540 // all done with frame size check
541 __ bind(after_frame_check);
542 }
544 // Allocate monitor and lock method (asm interpreter)
545 // Rmethod - Method*
546 void InterpreterGenerator::lock_method(void) {
547 // synchronize method
548 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
550 #ifdef ASSERT
551 { Label L;
552 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
553 __ andi(T0, T0, JVM_ACC_SYNCHRONIZED);
554 __ bne(T0, R0, L);
555 __ delayed()->nop();
556 __ stop("method doesn't need synchronization");
557 __ bind(L);
558 }
559 #endif // ASSERT
560 // get synchronization object
561 { Label done;
562 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
563 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
564 __ andi(T2, T0, JVM_ACC_STATIC);
565 __ ld(T0, LVP, Interpreter::local_offset_in_bytes(0));
566 __ beq(T2, R0, done);
567 __ delayed()->nop();
568 __ ld(T0, Rmethod, in_bytes(Method::const_offset()));
569 __ ld(T0, T0, in_bytes(ConstMethod::constants_offset()));
570 __ ld(T0, T0, ConstantPool::pool_holder_offset_in_bytes());
571 __ ld(T0, T0, mirror_offset);
572 __ bind(done);
573 }
574 // add space for monitor & lock
575 __ daddi(SP, SP, (-1) * entry_size); // add space for a monitor entry
576 __ sd(SP, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
577 // set new monitor block top
578 __ sd(T0, SP, BasicObjectLock::obj_offset_in_bytes()); // store object
579 // FIXME: I do not know what lock_object will do and what it will need
580 __ move(c_rarg0, SP); // object address
581 __ lock_object(c_rarg0);
582 }
584 // Generate a fixed interpreter frame. This is identical setup for
585 // interpreted methods and for native methods hence the shared code.
586 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
588 // [ local var m-1 ] <--- sp
589 // ...
590 // [ local var 0 ]
591 // [ argumnet word n-1 ] <--- T0(sender's sp)
592 // ...
593 // [ argument word 0 ] <--- S7
595 // initialize fixed part of activation frame
596 // sender's sp in Rsender
597 int i = 0;
598 __ sd(RA, SP, (-1) * wordSize); // save return address
599 __ sd(FP, SP, (-2) * wordSize); // save sender's fp
600 __ daddiu(FP, SP, (-2) * wordSize);
601 __ sd(Rsender, FP, (-++i) * wordSize); // save sender's sp
602 __ sd(R0, FP,(-++i)*wordSize); //save last_sp as null, FIXME aoqi
603 __ sd(LVP, FP, (-++i) * wordSize); // save locals offset
604 __ ld(BCP, Rmethod, in_bytes(Method::const_offset())); // get constMethodOop
605 __ daddiu(BCP, BCP, in_bytes(ConstMethod::codes_offset())); // get codebase
606 __ sd(Rmethod, FP, (-++i) * wordSize); // save Method*
607 #ifndef CORE
608 if (ProfileInterpreter) {
609 Label method_data_continue;
610 __ ld(AT, Rmethod, in_bytes(Method::method_data_offset()));
611 __ beq(AT, R0, method_data_continue);
612 __ delayed()->nop();
613 __ daddi(AT, AT, in_bytes(MethodData::data_offset()));
614 __ bind(method_data_continue);
615 __ sd(AT, FP, (-++i) * wordSize);
616 } else {
617 __ sd(R0, FP, (-++i) * wordSize);
618 }
619 #endif // !CORE
621 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
622 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
623 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
624 __ sd(T2, FP, (-++i) * wordSize); // set constant pool cache
625 if (native_call) {
626 __ sd(R0, FP, (-++i) * wordSize); // no bcp
627 } else {
628 __ sd(BCP, FP, (-++i) * wordSize); // set bcp
629 }
630 __ daddiu(SP, FP, (-++i) * wordSize);
631 __ sd(SP, FP, (-i) * wordSize); // reserve word for pointer to expression stack bottom
632 }
634 // End of helpers
636 // Various method entries
637 //------------------------------------------------------------------------------------------------------------------------
638 //
639 //
641 // Call an accessor method (assuming it is resolved, otherwise drop
642 // into vanilla (slow path) entry
643 address InterpreterGenerator::generate_accessor_entry(void) {
645 // Rmethod: Method*
646 // V0: receiver (preserve for slow entry into asm interpreter)
647 // Rsender: senderSP must preserved for slow path, set SP to it on fast path
649 address entry_point = __ pc();
650 Label xreturn_path;
651 // do fastpath for resolved accessor methods
652 if (UseFastAccessorMethods) {
653 Label slow_path;
654 __ li(T2, SafepointSynchronize::address_of_state());
655 __ lw(AT, T2, 0);
656 __ daddi(AT, AT, -(SafepointSynchronize::_not_synchronized));
657 __ bne(AT, R0, slow_path);
658 __ delayed()->nop();
659 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof;
660 // parameter size = 1
661 // Note: We can only use this code if the getfield has been resolved
662 // and if we don't have a null-pointer exception => check for
663 // these conditions first and use slow path if necessary.
664 // Rmethod: method
665 // V0: receiver
667 // [ receiver ] <-- sp
668 __ ld(T0, SP, 0);
670 // check if local 0 != NULL and read field
671 __ beq(T0, R0, slow_path);
672 __ delayed()->nop();
673 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
674 // __ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
675 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
676 // read first instruction word and extract bytecode @ 1 and index @ 2
677 __ ld(T3, Rmethod, in_bytes(Method::const_offset()));
678 __ lw(T3, T3, in_bytes(ConstMethod::codes_offset()));
679 // Shift codes right to get the index on the right.
680 // The bytecode fetched looks like <index><0xb4><0x2a>
681 __ dsrl(T3, T3, 2 * BitsPerByte);
682 // FIXME: maybe it's wrong
683 __ dsll(T3, T3, exact_log2(in_words(ConstantPoolCacheEntry::size())));
684 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
686 // T0: local 0 eax
687 // Rmethod: method ebx
688 // V0: receiver - do not destroy since it is needed for slow path! ecx
689 // ecx: scratch use which register instead ?
690 // T1: scratch use which register instead ?
691 // T3: constant pool cache index edx
692 // T2: constant pool cache edi
693 // esi: send's sp
694 // Rsender: send's sp
695 // check if getfield has been resolved and read constant pool cache entry
696 // check the validity of the cache entry by testing whether _indices field
697 // contains Bytecode::_getfield in b1 byte.
698 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
699 // __ movl(esi,
700 // Address(edi,
701 // edx,
702 // Address::times_4, ConstantPoolCache::base_offset()
703 // + ConstantPoolCacheEntry::indices_offset()));
706 __ dsll(T8, T3, Address::times_8);
707 __ move(T1, in_bytes(ConstantPoolCache::base_offset()
708 + ConstantPoolCacheEntry::indices_offset()));
709 __ dadd(T1, T8, T1);
710 __ dadd(T1, T1, T2);
711 __ lw(T1, T1, 0);
712 __ dsrl(T1, T1, 2 * BitsPerByte);
713 __ andi(T1, T1, 0xFF);
714 __ daddi(T1, T1, (-1) * Bytecodes::_getfield);
715 __ bne(T1, R0, slow_path);
716 __ delayed()->nop();
718 // __ shrl(esi, 2*BitsPerByte);
719 // __ andl(esi, 0xFF);
720 // __ cmpl(esi, Bytecodes::_getfield);
721 // __ jcc(Assembler::notEqual, slow_path);
723 // Note: constant pool entry is not valid before bytecode is resolved
725 // __ movl(esi,
726 // Address(edi,
727 // edx,
728 // Address::times_4, ConstantPoolCache::base_offset()
729 // + ConstantPoolCacheEntry::f2_offset()));
730 __ move(T1, in_bytes(ConstantPoolCache::base_offset()
731 + ConstantPoolCacheEntry::f2_offset()));
732 __ dadd(T1, T1, T8);
733 __ dadd(T1, T1, T2);
734 __ lw(AT, T1, 0);
735 // __ movl(edx,
736 // Address(edi,
737 // edx,
738 // Address::times_4, ConstantPoolCache::base_offset()
739 // + ConstantPoolCacheEntry::flags_offset()));
742 __ move(T1, in_bytes(ConstantPoolCache::base_offset()
743 + ConstantPoolCacheEntry::flags_offset()));
744 __ dadd(T1, T1, T8);
745 __ dadd(T1, T1, T2);
746 __ lw(T3, T1, 0);
748 Label notByte, notShort, notChar, notObj;
749 // const Address field_address (eax, esi, Address::times_1);
751 // Need to differentiate between igetfield, agetfield, bgetfield etc.
752 // because they are different sizes.
753 // Use the type from the constant pool cache
754 __ dsrl(T3, T3, ConstantPoolCacheEntry::tos_state_shift);
755 // Make sure we don't need to mask edx for tosBits after the above shift
756 ConstantPoolCacheEntry::verify_tos_state_shift();
757 // btos = 0
758 __ bne(T3, R0, notByte);
759 __ delayed()->dadd(T0, T0, AT);
761 __ lb(V0, T0, 0);
762 __ b(xreturn_path);
763 __ delayed()->nop();
765 //stos
766 __ bind(notByte);
767 __ daddi(T1, T3, (-1) * stos);
768 __ bne(T1, R0, notShort);
769 __ delayed()->nop();
770 __ lh(V0, T0, 0);
771 __ b(xreturn_path);
772 __ delayed()->nop();
774 //ctos
775 __ bind(notShort);
776 __ daddi(T1, T3, (-1) * ctos);
777 __ bne(T1, R0, notChar);
778 __ delayed()->nop();
779 __ lhu(V0, T0, 0);
780 __ b(xreturn_path);
781 __ delayed()->nop();
783 //atos
784 __ bind(notChar);
785 __ daddi(T1, T3, (-1) * atos);
786 __ bne(T1, R0, notObj);
787 __ delayed()->nop();
788 //add for compressedoops
789 __ load_heap_oop(V0, Address(T0, 0));
790 __ b(xreturn_path);
791 __ delayed()->nop();
793 //itos
794 __ bind(notObj);
795 #ifdef ASSERT
796 Label okay;
797 __ daddi(T1, T3, (-1) * itos);
798 __ beq(T1, R0, okay);
799 __ delayed()->nop();
800 __ stop("what type is this?");
801 __ bind(okay);
802 #endif // ASSERT
803 __ lw(V0, T0, 0);
805 __ bind(xreturn_path);
807 // _ireturn/_areturn
808 //FIXME
809 __ move(SP, Rsender);//FIXME, set sender's fp to SP
810 __ jr(RA);
811 __ delayed()->nop();
813 // generate a vanilla interpreter entry as the slow path
814 __ bind(slow_path);
815 (void) generate_normal_entry(false);
816 } else {
817 (void) generate_normal_entry(false);
818 }
820 return entry_point;
821 }
823 // Method entry for java.lang.ref.Reference.get.
824 address InterpreterGenerator::generate_Reference_get_entry(void) {
825 #ifndef SERIALGC
826 // Code: _aload_0, _getfield, _areturn
827 // parameter size = 1
828 //
829 // The code that gets generated by this routine is split into 2 parts:
830 // 1. The "intrinsified" code for G1 (or any SATB based GC),
831 // 2. The slow path - which is an expansion of the regular method entry.
832 //
833 // Notes:-
834 // * In the G1 code we do not check whether we need to block for
835 // a safepoint. If G1 is enabled then we must execute the specialized
836 // code for Reference.get (except when the Reference object is null)
837 // so that we can log the value in the referent field with an SATB
838 // update buffer.
839 // If the code for the getfield template is modified so that the
840 // G1 pre-barrier code is executed when the current method is
841 // Reference.get() then going through the normal method entry
842 // will be fine.
843 // * The G1 code can, however, check the receiver object (the instance
844 // of java.lang.Reference) and jump to the slow path if null. If the
845 // Reference object is null then we obviously cannot fetch the referent
846 // and so we don't need to call the G1 pre-barrier. Thus we can use the
847 // regular method entry code to generate the NPE.
848 //
849 // This code is based on generate_accessor_enty.
850 //
851 // rbx: Method*
853 // r13: senderSP must preserve for slow path, set SP to it on fast path
855 address entry = __ pc();
857 const int referent_offset = java_lang_ref_Reference::referent_offset;
858 guarantee(referent_offset > 0, "referent offset not initialized");
860 if (UseG1GC) {
861 Label slow_path;
862 // rbx: method
864 // Check if local 0 != NULL
865 // If the receiver is null then it is OK to jump to the slow path.
866 /* __ movptr(rax, Address(rsp, wordSize));
868 __ testptr(rax, rax);
869 __ jcc(Assembler::zero, slow_path);*/
871 // rax: local 0
872 // rbx: method (but can be used as scratch now)
873 // rdx: scratch
874 // rdi: scratch
876 // Generate the G1 pre-barrier code to log the value of
877 // the referent field in an SATB buffer.
879 // Load the value of the referent field.
880 //const Address field_address(rax, referent_offset);
881 // __ load_heap_oop(rax, field_address);
883 // Generate the G1 pre-barrier code to log the value of
884 // the referent field in an SATB buffer.
885 // __ g1_write_barrier_pre(noreg /* obj */,
886 // rax /* pre_val */,
887 // r15_thread /* thread */,
888 // rbx /* tmp */,
889 // true /* tosca_live */,
890 // true /* expand_call */);
892 // _areturn
893 /* __ pop(rdi); // get return address
894 __ mov(rsp, r13); // set sp to sender sp
895 __ jmp(rdi);
896 __ ret(0);
898 // generate a vanilla interpreter entry as the slow path
899 __ bind(slow_path);
900 (void) generate_normal_entry(false);
901 */
902 return entry;
903 }
904 #endif // SERIALGC
906 // If G1 is not enabled then attempt to go through the accessor entry point
907 // Reference.get is an accessor
908 return generate_accessor_entry();
909 }
910 // Interpreter stub for calling a native method. (asm interpreter)
911 // This sets up a somewhat different looking stack for calling the
912 // native method than the typical interpreter frame setup.
913 address InterpreterGenerator::generate_native_entry(bool synchronized) {
914 // determine code generation flags
915 bool inc_counter = UseCompiler || CountCompiledCalls;
916 // Rsender: sender's sp
917 // Rmethod: Method*
918 address entry_point = __ pc();
920 #ifndef CORE
921 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() + // Fu: 20130814
922 InvocationCounter::counter_offset()));
923 #endif
925 // get parameter size (always needed)
926 // the size in the java stack
927 //__ lhu(V0, Rmethod, in_bytes(Method::size_of_parameters_offset()));
928 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
929 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
931 // native calls don't need the stack size check since they have no expression stack
932 // and the arguments are already on the stack and we only add a handful of words
933 // to the stack
935 // Rmethod: Method*
936 // V0: size of parameters
937 // Layout of frame at this point
938 //
939 // [ argument word n-1 ] <--- sp
940 // ...
941 // [ argument word 0 ]
943 // for natives the size of locals is zero
945 // compute beginning of parameters (S7)
946 __ dsll(LVP, V0, Address::times_8);
947 __ daddiu(LVP, LVP, (-1) * wordSize);
948 __ dadd(LVP, LVP, SP);
950 //__ move(T0, SP); // remember sender sp for generate_fixed_frame
953 // add 2 zero-initialized slots for native calls
954 __ daddi(SP, SP, (-2) * wordSize);
955 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
956 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
958 // Layout of frame at this point
959 // [ method holder mirror ] <--- sp
960 // [ result type info ]
961 // [ argument word n-1 ] <--- T0
962 // ...
963 // [ argument word 0 ] <--- LVP
966 #ifndef CORE
967 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
968 #endif
970 // initialize fixed part of activation frame
971 generate_fixed_frame(true);
972 // after this function, the layout of frame is as following
973 //
974 // [ monitor block top ] <--- sp ( the top monitor entry )
975 // [ byte code pointer (0) ] (if native, bcp = 0)
976 // [ constant pool cache ]
977 // [ Method* ]
978 // [ locals offset ]
979 // [ sender's sp ]
980 // [ sender's fp ]
981 // [ return address ] <--- fp
982 // [ method holder mirror ]
983 // [ result type info ]
984 // [ argumnet word n-1 ] <--- sender's sp
985 // ...
986 // [ argument word 0 ] <--- S7
989 // make sure method is native & not abstract
990 #ifdef ASSERT
991 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
992 {
993 Label L;
994 __ andi(AT, T0, JVM_ACC_NATIVE);
995 __ bne(AT, R0, L);
996 __ delayed()->nop();
997 __ stop("tried to execute native method as non-native");
998 __ bind(L);
999 }
1000 { Label L;
1001 __ andi(AT, T0, JVM_ACC_ABSTRACT);
1002 __ beq(AT, R0, L);
1003 __ delayed()->nop();
1004 __ stop("tried to execute abstract method in interpreter");
1005 __ bind(L);
1006 }
1007 #endif
1009 // Since at this point in the method invocation the exception handler
1010 // would try to exit the monitor of synchronized methods which hasn't
1011 // been entered yet, we set the thread local variable
1012 // _do_not_unlock_if_synchronized to true. The remove_activation will
1013 // check this flag.
1014 Register thread = TREG;
1015 #ifndef OPT_THREAD
1016 __ get_thread(thread);
1017 #endif
1018 __ move(AT, (int)true);
1019 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1021 #ifndef CORE
1022 // increment invocation count & check for overflow
1023 Label invocation_counter_overflow;
1024 if (inc_counter) {
1025 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
1026 }
1027 Label continue_after_compile;
1028 __ bind(continue_after_compile);
1029 #endif // CORE
1031 bang_stack_shadow_pages(true);
1033 // reset the _do_not_unlock_if_synchronized flag
1034 #ifndef OPT_THREAD
1035 __ get_thread(thread);
1036 #endif
1037 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1039 // check for synchronized methods
1040 // Must happen AFTER invocation_counter check and stack overflow check,
1041 // so method is not locked if overflows.
1042 if (synchronized) {
1043 lock_method();
1044 } else {
1045 // no synchronization necessary
1046 #ifdef ASSERT
1047 {
1048 Label L;
1049 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
1050 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
1051 __ beq(AT, R0, L);
1052 __ delayed()->nop();
1053 __ stop("method needs synchronization");
1054 __ bind(L);
1055 }
1056 #endif
1057 }
1059 // after method_lock, the layout of frame is as following
1060 //
1061 // [ monitor entry ] <--- sp
1062 // ...
1063 // [ monitor entry ]
1064 // [ monitor block top ] ( the top monitor entry )
1065 // [ byte code pointer (0) ] (if native, bcp = 0)
1066 // [ constant pool cache ]
1067 // [ Method* ]
1068 // [ locals offset ]
1069 // [ sender's sp ]
1070 // [ sender's fp ]
1071 // [ return address ] <--- fp
1072 // [ method holder mirror ]
1073 // [ result type info ]
1074 // [ argumnet word n-1 ] <--- ( sender's sp )
1075 // ...
1076 // [ argument word 0 ] <--- S7
1078 // start execution
1079 #ifdef ASSERT
1080 { Label L;
1081 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1082 __ beq(AT, SP, L);
1083 __ delayed()->nop();
1084 __ stop("broken stack frame setup in interpreter in asm");
1085 __ bind(L);
1086 }
1087 #endif
1089 // jvmti/jvmpi support
1090 __ notify_method_entry();
1092 // work registers
1093 const Register method = Rmethod;
1094 //const Register thread = T2;
1095 const Register t = RT4;
1097 __ get_method(method);
1098 __ verify_oop(method);
1099 { Label L, Lstatic;
1100 __ ld(t,method,in_bytes(Method::const_offset()));
1101 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
1102 {//aoqi_test
1103 Label L;
1104 __ daddi(AT, t, -8);
1105 __ blez(AT, L);
1106 __ delayed()->nop();
1107 __ bind(L);
1108 }
1109 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1110 //FIXME, aoqi: A1?
1111 // A0 and A1(if needed)
1112 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1113 __ andi(AT, AT, JVM_ACC_STATIC);
1114 __ beq(AT, R0, Lstatic);
1115 __ delayed()->nop();
1116 __ daddiu(t, t, 1);
1117 __ bind(Lstatic);
1118 __ daddiu(t, t, -7);
1119 __ blez(t, L);
1120 __ delayed()->nop();
1121 __ dsll(t, t, Address::times_8);
1122 __ dsub(SP, SP, t);
1123 __ bind(L);
1124 }
1125 __ move(AT, -(StackAlignmentInBytes));
1126 __ andr(SP, SP, AT);
1127 __ move(AT, SP);
1128 // [ ] <--- sp
1129 // ... (size of parameters - 8 )
1130 // [ monitor entry ]
1131 // ...
1132 // [ monitor entry ]
1133 // [ monitor block top ] ( the top monitor entry )
1134 // [ byte code pointer (0) ] (if native, bcp = 0)
1135 // [ constant pool cache ]
1136 // [ Method* ]
1137 // [ locals offset ]
1138 // [ sender's sp ]
1139 // [ sender's fp ]
1140 // [ return address ] <--- fp
1141 // [ method holder mirror ]
1142 // [ result type info ]
1143 // [ argumnet word n-1 ] <--- ( sender's sp )
1144 // ...
1145 // [ argument word 0 ] <--- LVP
1147 // get signature handler
1148 {
1149 Label L;
1150 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1151 __ bne(T9, R0, L);
1152 __ delayed()->nop();
1153 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1154 InterpreterRuntime::prepare_native_call), method);
1155 __ get_method(method);
1156 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1157 __ bind(L);
1158 }
1160 // call signature handler
1161 // FIXME: when change codes in InterpreterRuntime, note this point
1162 // from: begin of parameters
1163 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1164 // to: current sp
1165 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1166 // temp: T3
1167 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1169 __ jalr(T9);
1170 __ delayed()->nop();
1171 __ get_method(method); // slow path call blows EBX on DevStudio 5.0
1173 /*
1174 if native function is static, and its second parameter has type length of double word,
1175 and first parameter has type length of word, we have to reserve one word
1176 for the first parameter, according to mips o32 abi.
1177 if native function is not static, and its third parameter has type length of double word,
1178 and second parameter has type length of word, we have to reserve one word for the second
1179 parameter.
1180 */
1183 // result handler is in V0
1184 // set result handler
1185 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1187 #define FIRSTPARA_SHIFT_COUNT 5
1188 #define SECONDPARA_SHIFT_COUNT 9
1189 #define THIRDPARA_SHIFT_COUNT 13
1190 #define PARA_MASK 0xf
1192 // pass mirror handle if static call
1193 {
1194 Label L;
1195 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1196 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1197 __ andi(AT, t, JVM_ACC_STATIC);
1198 __ beq(AT, R0, L);
1199 __ delayed()->nop();
1201 // get mirror
1202 __ ld(t, method, in_bytes(Method:: const_offset()));
1203 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1204 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1205 __ ld(t, t, mirror_offset);
1206 // copy mirror into activation frame
1207 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1208 // pass handle to mirror
1209 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1210 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1211 // __ ld_ptr(t,Address(SP ,wordSize));
1212 //FIXME, aoqi
1213 //__ st_ptr(t, Address(SP, wordSize));
1214 __ move(A1, t);
1215 __ bind(L);
1216 }
1218 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1219 // [ ] |
1220 // ... size of parameters(or +1) |
1221 // [ monitor entry ] |
1222 // ... |
1223 // [ monitor entry ] |
1224 // [ monitor block top ] ( the top monitor entry ) |
1225 // [ byte code pointer (0) ] (if native, bcp = 0) |
1226 // [ constant pool cache ] |
1227 // [ Method* ] |
1228 // [ locals offset ] |
1229 // [ sender's sp ] |
1230 // [ sender's fp ] |
1231 // [ return address ] <--- fp |
1232 // [ method holder mirror ] <----------------------------|
1233 // [ result type info ]
1234 // [ argumnet word n-1 ] <--- ( sender's sp )
1235 // ...
1236 // [ argument word 0 ] <--- S7
1238 // get native function entry point
1239 { Label L;
1240 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1241 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1242 __ bne(V1, T9, L);
1243 __ delayed()->nop();
1244 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1245 __ get_method(method);
1246 __ verify_oop(method);
1247 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1248 __ bind(L);
1249 }
1250 /*
1251 __ pushad();
1252 __ move(A0, T9);
1253 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
1254 __ popad();
1255 */
1257 // pass JNIEnv
1258 // native function in T9
1259 #ifndef OPT_THREAD
1260 __ get_thread(thread);
1261 #endif
1262 __ daddi(t, thread, in_bytes(JavaThread::jni_environment_offset()));
1263 //__ addi(SP, SP, (-1) * wordSize);
1264 //__ sw(t, SP, 0);
1265 // stack,but I think it won't work when pass float,double etc @jerome,10/17,2006
1266 __ move(A0, t);
1267 // [ jni environment ] <--- sp
1268 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1269 // [ ] |
1270 // ... size of parameters |
1271 // [ monitor entry ] |
1272 // ... |
1273 // [ monitor entry ] |
1274 // [ monitor block top ] ( the top monitor entry ) |
1275 // [ byte code pointer (0) ] (if native, bcp = 0) |
1276 // [ constant pool cache ] |
1277 // [ Method* ] |
1278 // [ locals offset ] |
1279 // [ sender's sp ] |
1280 // [ sender's fp ] |
1281 // [ return address ] <--- fp |
1282 // [ method holder mirror ] <----------------------------|
1283 // [ result type info ]
1284 // [ argumnet word n-1 ] <--- ( sender's sp )
1285 // ...
1286 // [ argument word 0 ] <--- S7
1288 /*
1289 // reset handle block
1290 __ lw(t, thread, in_bytes(JavaThread::active_handles_offset()));
1291 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1293 */
1294 // set_last_Java_frame_before_call
1295 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1296 // Change state to native (we save the return address in the thread, since it might not
1297 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1298 // points into the right code segment. It does not have to be the correct return pc.
1299 __ li(t, __ pc());
1300 // __ sw(t, thread, in_bytes(JavaThread::frame_anchor_offset()
1301 // + JavaFrameAnchor::last_Java_pc_offset()));
1302 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1303 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1305 // change thread state
1306 #ifdef ASSERT
1307 { Label L;
1308 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1309 __ daddi(t, t, (-1) * _thread_in_Java);
1310 __ beq(t, R0, L);
1311 __ delayed()->nop();
1312 __ stop("Wrong thread state in native stub");
1313 __ bind(L);
1314 }
1315 #endif
1317 __ move(t, _thread_in_native);
1318 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1320 // call native method
1321 __ jalr(T9);
1322 __ delayed()->nop();
1323 // result potentially in V2:V1 or F0:F1
1326 if (CheckJNICalls) {
1327 //FIXME
1328 // __ call(StubRoutines::gs2::verify_fpu_cntrl_wrd_entry(),
1329 // relocInfo::runtime_call_type);
1330 }
1332 // restore S0 to have legal interpreter frame, i.e., bci == 0 <=> S0 == code_base()
1333 //__ lw(BCP, method, in_bytes(Method::const_offset())); // get constMethodOop
1334 //__ addi(BCP, BCP, in_bytes(ConstMethod::codes_offset())); // get codebase
1336 // via _last_native_pc and not via _last_jave_sp
1337 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1338 // If the order changes or anything else is added to the stack the code in
1339 // interpreter_frame_result will have to be changed.
1340 //FIXME, should modify here
1341 // save return value to keep the value from being destroyed by other calls
1342 //__ addi(SP, SP, (-4) * wordSize);
1343 //__ sw(V0, SP, 3 * wordSize);
1344 //__ sw(V1, SP, 2 * wordSize);
1345 //__ swc1(F0, SP, 1 * wordSize);
1346 //__ swc1(F1, SP, 0 * wordSize);
1347 __ move(S1, V0);
1348 __ move(S3, V1);
1349 __ dmfc1(S4, F0);
1350 __ dmfc1(S2, F1);
1352 // change thread state
1353 __ get_thread(thread);
1354 __ move(t, _thread_in_native_trans);
1355 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1357 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1359 // check for safepoint operation in progress and/or pending suspend requests
1360 { Label Continue;
1362 // Don't use call_VM as it will see a possible pending exception and forward it
1363 // and never return here preventing us from clearing _last_native_pc down below.
1364 // Also can't use call_VM_leaf either as it will check to see if esi & edi are
1365 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1366 // by hand.
1367 //
1368 Label L;
1369 __ li(AT, SafepointSynchronize::address_of_state());
1370 __ lw(AT, AT, 0);
1371 __ bne(AT, R0, L);
1372 __ delayed()->nop();
1373 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1374 __ beq(AT, R0, Continue);
1375 __ delayed()->nop();
1376 __ bind(L);
1377 // __ addi(SP, SP, (-1) * wordSize);
1378 __ move(A0, thread);
1379 __ call(CAST_FROM_FN_PTR(address,
1380 JavaThread::check_special_condition_for_native_trans),
1381 relocInfo::runtime_call_type);
1382 __ delayed()->nop();
1383 // __ addi(SP, SP, wordSize);
1385 // __ get_method(method);
1386 #ifndef OPT_THREAD
1387 __ get_thread(thread);
1388 #endif
1389 //add for compressedoops
1390 __ reinit_heapbase();
1391 __ bind(Continue);
1392 }
1394 // change thread state
1395 __ move(t, _thread_in_Java);
1396 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1397 __ reset_last_Java_frame(thread, true, true);
1399 // reset handle block
1400 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1401 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1403 // If result was an oop then unbox and save it in the frame
1404 { Label L;
1405 Label no_oop, store_result;
1406 //FIXME, addi only support 16-bit imeditate
1407 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1408 // __ addi(AT,AT,-(int)AbstractInterpreter::result_handler(T_OBJECT));
1409 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1410 __ bne(AT, T0, no_oop);
1411 __ delayed()->nop();
1412 //__ cmpl(Address(esp), NULL_WORD);
1413 //FIXME, do we need pop here ? @jerome
1414 //__ pop(ltos);
1415 //__ testl(eax, eax);
1416 //__ jcc(Assembler::zero, store_result);
1417 __ move(V0, S1);
1418 __ beq(V0, R0, store_result);
1419 __ delayed()->nop();
1420 // unbox
1421 __ ld(V0, V0, 0);
1422 __ bind(store_result);
1423 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1424 // keep stack depth as expected by pushing oop which will eventually be discarded
1425 __ bind(no_oop);
1426 }
1427 {
1428 Label no_reguard;
1429 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1430 //__ bne(t, JavaThread::stack_guard_yellow_disabled, no_reguard);
1431 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1432 __ bne(t, AT, no_reguard);
1433 __ delayed()->nop();
1434 __ pushad();
1435 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages),
1436 relocInfo::runtime_call_type);
1437 __ delayed()->nop();
1438 __ popad();
1439 //add for compressedoops
1440 __ reinit_heapbase();
1441 __ bind(no_reguard);
1442 }
1443 // restore esi to have legal interpreter frame,
1444 // i.e., bci == 0 <=> esi == code_base()
1445 // Can't call_VM until bcp is within reasonable.
1446 __ get_method(method); // method is junk from thread_in_native to now.
1447 __ verify_oop(method);
1448 // __ movl(esi, Address(method,Method::const_offset())); // get constMethodOop
1449 __ ld(BCP, method, in_bytes(Method::const_offset()));
1450 // __ leal(esi, Address(esi,ConstMethod::codes_offset())); // get codebase
1451 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1452 // handle exceptions (exception handling will handle unlocking!)
1453 {
1454 Label L;
1455 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1456 __ beq(t, R0, L);
1457 __ delayed()->nop();
1458 // Note: At some point we may want to unify this with the code used in
1459 // call_VM_base();
1460 // i.e., we should use the StubRoutines::forward_exception code. For now this
1461 // doesn't work here because the esp is not correctly set at this point.
1462 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address,
1463 InterpreterRuntime::throw_pending_exception));
1464 __ should_not_reach_here();
1465 __ bind(L);
1466 }
1468 // do unlocking if necessary
1469 { Label L;
1470 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1471 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1472 __ beq(t, R0, L);
1473 // the code below should be shared with interpreter macro assembler implementation
1474 { Label unlock;
1475 // BasicObjectLock will be first in list,
1476 // since this is a synchronized method. However, need
1477 // to check that the object has not been unlocked by
1478 // an explicit monitorexit bytecode.
1479 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset
1480 * wordSize - (int)sizeof(BasicObjectLock));
1481 // address of first monitor
1483 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1484 __ bne(t, R0, unlock);
1485 __ delayed()->nop();
1487 // Entry already unlocked, need to throw exception
1488 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1489 InterpreterRuntime::throw_illegal_monitor_state_exception));
1490 __ should_not_reach_here();
1492 __ bind(unlock);
1493 __ unlock_object(c_rarg0);
1494 }
1495 __ bind(L);
1496 }
1498 // jvmti/jvmpi support
1499 // Note: This must happen _after_ handling/throwing any exceptions since
1500 // the exception handler code notifies the runtime of method exits
1501 // too. If this happens before, method entry/exit notifications are
1502 // not properly paired (was bug - gri 11/22/99).
1503 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI );
1505 // restore potential result in V0:V1,
1506 // call result handler to restore potential result in ST0 & handle result
1507 //__ lw(V0, SP, 3 * wordSize);
1508 //__ lw(V1, SP, 2 * wordSize);
1509 //__ lwc1(F0, SP, 1 * wordSize);
1510 //__ lwc1(F1, SP, 0 * wordSize);
1511 //__ addi(SP, SP, 4 * wordSize);
1512 __ move(V0, S1);
1513 __ move(V1, S3);
1514 __ dmtc1(S4, F0);
1515 __ dmtc1(S2, F1);
1516 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1517 __ jalr(t);
1518 __ delayed()->nop();
1519 //jerome_for_debug
1520 //__ move(AT, (int)(&jerome4));
1521 //__ sw(FP, AT, 0);
1524 // remove activation
1525 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1526 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1527 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1528 __ jr(RA);
1529 __ delayed()->nop();
1531 #ifndef CORE
1532 if (inc_counter) {
1533 // Handle overflow of counter and compile method
1534 __ bind(invocation_counter_overflow);
1535 generate_counter_overflow(&continue_after_compile);
1536 // entry_point is the beginning of this
1537 // function and checks again for compiled code
1538 }
1539 #endif
1540 return entry_point;
1541 }
1543 //
1544 // Generic interpreted method entry to (asm) interpreter
1545 //
1546 // Layout of frame just at the entry
1547 //
1548 // [ argument word n-1 ] <--- sp
1549 // ...
1550 // [ argument word 0 ]
1551 // assume Method* in Rmethod before call this method.
1552 // prerequisites to the generated stub : the callee Method* in Rmethod
1553 // note you must save the caller bcp before call the generated stub
1554 //
1555 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1556 // determine code generation flags
1557 bool inc_counter = UseCompiler || CountCompiledCalls;
1559 // Rmethod: Method*
1560 // Rsender: sender 's sp
1561 address entry_point = __ pc();
1562 /*
1563 #ifndef CORE
1564 // check if compiled code exists
1565 Label run_compiled_code;
1566 if (!CompileTheWorld) {
1567 check_for_compiled_code(run_compiled_code);
1568 }
1569 #endif
1570 */
1571 #ifndef CORE
1572 const Address invocation_counter(Rmethod,
1573 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1574 #endif
1576 // get parameter size (always needed)
1577 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1578 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1580 // Rmethod: Method*
1581 // V0: size of parameters
1582 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1583 // get size of locals in words to T2
1584 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1585 // T2 = no. of additional locals, locals include parameters
1586 __ dsub(T2, T2, V0);
1588 // see if we've got enough room on the stack for locals plus overhead.
1589 // Layout of frame at this point
1590 //
1591 // [ argument word n-1 ] <--- sp
1592 // ...
1593 // [ argument word 0 ]
1594 generate_stack_overflow_check();
1595 // after this function, the layout of frame does not change
1597 // compute beginning of parameters (LVP)
1598 __ dsll(LVP, V0, LogBytesPerWord);
1599 __ daddiu(LVP, LVP, (-1) * wordSize);
1600 __ dadd(LVP, LVP, SP);
1602 // T2 - # of additional locals
1603 // allocate space for locals
1604 // explicitly initialize locals
1605 {
1606 Label exit, loop;
1607 // for test
1608 // __ slt(AT, R0, T2);
1609 // __ beq(AT, R0, exit);
1610 __ beq(T2, R0, exit);
1611 __ delayed()->nop();
1612 __ bind(loop);
1613 // if(TaggedStackInterpreter)__ daddi(SP, SP, -1 * wordSize);
1614 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1615 __ daddiu(T2, T2, -1); // until everything initialized
1616 __ bne(T2, R0, loop);
1617 // __ slt(AT, R0, T2);
1618 // __ bne(AT, R0, loop);
1619 __ delayed();
1620 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1621 __ bind(exit);
1622 }
1624 #ifndef CORE
1625 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
1626 #endif
1627 //
1628 // [ local var m-1 ] <--- sp
1629 // ...
1630 // [ local var 0 ]
1631 // [ argument word n-1 ] <--- T0?
1632 // ...
1633 // [ argument word 0 ] <--- LVP
1635 // initialize fixed part of activation frame
1637 generate_fixed_frame(false);
1640 // after this function, the layout of frame is as following
1641 //
1642 // [ monitor block top ] <--- sp ( the top monitor entry )
1643 // [ byte code pointer ] (if native, bcp = 0)
1644 // [ constant pool cache ]
1645 // [ Method* ]
1646 // [ locals offset ]
1647 // [ sender's sp ]
1648 // [ sender's fp ] <--- fp
1649 // [ return address ]
1650 // [ local var m-1 ]
1651 // ...
1652 // [ local var 0 ]
1653 // [ argumnet word n-1 ] <--- ( sender's sp )
1654 // ...
1655 // [ argument word 0 ] <--- LVP
1658 // make sure method is not native & not abstract
1659 #ifdef ASSERT
1660 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1661 {
1662 Label L;
1663 __ andi(T2, AT, JVM_ACC_NATIVE);
1664 __ beq(T2, R0, L);
1665 __ delayed()->nop();
1666 __ stop("tried to execute native method as non-native");
1667 __ bind(L);
1668 }
1669 { Label L;
1670 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1671 __ beq(T2, R0, L);
1672 __ delayed()->nop();
1673 __ stop("tried to execute abstract method in interpreter");
1674 __ bind(L);
1675 }
1676 #endif
1678 // Since at this point in the method invocation the exception handler
1679 // would try to exit the monitor of synchronized methods which hasn't
1680 // been entered yet, we set the thread local variable
1681 // _do_not_unlock_if_synchronized to true. The remove_activation will
1682 // check this flag.
1684 #ifndef OPT_THREAD
1685 Register thread = T8;
1686 __ get_thread(thread);
1687 #else
1688 Register thread = TREG;
1689 #endif
1690 __ move(AT, (int)true);
1691 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1693 #ifndef CORE
1695 // 2014/11/24 Fu
1696 // mdp : T8
1697 // tmp1: T9
1698 // tmp2: T2
1699 __ profile_parameters_type(T8, T9, T2);
1701 // increment invocation count & check for overflow
1702 Label invocation_counter_overflow;
1703 Label profile_method;
1704 Label profile_method_continue;
1705 if (inc_counter) {
1706 generate_counter_incr(&invocation_counter_overflow, &profile_method,
1707 &profile_method_continue);
1708 if (ProfileInterpreter) {
1709 __ bind(profile_method_continue);
1710 }
1711 }
1713 Label continue_after_compile;
1714 __ bind(continue_after_compile);
1716 #endif // CORE
1718 bang_stack_shadow_pages(false);
1720 // reset the _do_not_unlock_if_synchronized flag
1721 #ifndef OPT_THREAD
1722 __ get_thread(thread);
1723 #endif
1724 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1726 // check for synchronized methods
1727 // Must happen AFTER invocation_counter check and stack overflow check,
1728 // so method is not locked if overflows.
1729 //
1730 if (synchronized) {
1731 // Allocate monitor and lock method
1732 lock_method();
1733 } else {
1734 // no synchronization necessary
1735 #ifdef ASSERT
1736 { Label L;
1737 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1738 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1739 __ beq(T2, R0, L);
1740 __ delayed()->nop();
1741 __ stop("method needs synchronization");
1742 __ bind(L);
1743 }
1744 #endif
1745 }
1747 // layout of frame after lock_method
1748 // [ monitor entry ] <--- sp
1749 // ...
1750 // [ monitor entry ]
1751 // [ monitor block top ] ( the top monitor entry )
1752 // [ byte code pointer ] (if native, bcp = 0)
1753 // [ constant pool cache ]
1754 // [ Method* ]
1755 // [ locals offset ]
1756 // [ sender's sp ]
1757 // [ sender's fp ]
1758 // [ return address ] <--- fp
1759 // [ local var m-1 ]
1760 // ...
1761 // [ local var 0 ]
1762 // [ argumnet word n-1 ] <--- ( sender's sp )
1763 // ...
1764 // [ argument word 0 ] <--- LVP
1767 // start execution
1768 #ifdef ASSERT
1769 { Label L;
1770 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1771 __ beq(AT, SP, L);
1772 __ delayed()->nop();
1773 __ stop("broken stack frame setup in interpreter in native");
1774 __ bind(L);
1775 }
1776 #endif
1778 // jvmti/jvmpi support
1779 __ notify_method_entry();
1781 __ dispatch_next(vtos);
1783 #ifndef CORE
1784 // invocation counter overflow
1785 if (inc_counter) {
1786 if (ProfileInterpreter) {
1787 // We have decided to profile this method in the interpreter
1788 __ bind(profile_method);
1789 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1790 InterpreterRuntime::profile_method));
1791 __ set_method_data_pointer_for_bcp();
1792 __ get_method(Rmethod);
1793 __ b(profile_method_continue);
1794 __ delayed()->nop();
1795 }
1796 // Handle overflow of counter and compile method
1797 __ bind(invocation_counter_overflow);
1798 generate_counter_overflow(&continue_after_compile);
1799 }
1801 #endif
1802 return entry_point;
1803 }
1805 // Entry points
1806 //
1807 // Here we generate the various kind of entries into the interpreter.
1808 // The two main entry type are generic bytecode methods and native
1809 // call method. These both come in synchronized and non-synchronized
1810 // versions but the frame layout they create is very similar. The
1811 // other method entry types are really just special purpose entries
1812 // that are really entry and interpretation all in one. These are for
1813 // trivial methods like accessor, empty, or special math methods.
1814 //
1815 // When control flow reaches any of the entry types for the interpreter
1816 // the following holds ->
1817 //
1818 // Arguments:
1819 //
1820 // Rmethod: Method*
1821 // V0: receiver
1822 //
1823 //
1824 // Stack layout immediately at entry
1825 //
1826 // [ parameter n-1 ] <--- sp
1827 // ...
1828 // [ parameter 0 ]
1829 // [ expression stack ] (caller's java expression stack)
1831 // Assuming that we don't go to one of the trivial specialized entries
1832 // the stack will look like below when we are ready to execute the
1833 // first bytecode (or call the native routine). The register usage
1834 // will be as the template based interpreter expects (see
1835 // interpreter_amd64.hpp).
1836 //
1837 // local variables follow incoming parameters immediately; i.e.
1838 // the return address is moved to the end of the locals).
1839 //
1840 // [ monitor entry ] <--- sp
1841 // ...
1842 // [ monitor entry ]
1843 // [ monitor block top ] ( the top monitor entry )
1844 // [ byte code pointer ] (if native, bcp = 0)
1845 // [ constant pool cache ]
1846 // [ Method* ]
1847 // [ locals offset ]
1848 // [ sender's sp ]
1849 // [ sender's fp ]
1850 // [ return address ] <--- fp
1851 // [ local var m-1 ]
1852 // ...
1853 // [ local var 0 ]
1854 // [ argumnet word n-1 ] <--- ( sender's sp )
1855 // ...
1856 // [ argument word 0 ] <--- S7
1858 address AbstractInterpreterGenerator::generate_method_entry(
1859 AbstractInterpreter::MethodKind kind) {
1860 // determine code generation flags
1861 bool synchronized = false;
1862 address entry_point = NULL;
1863 switch (kind) {
1864 case Interpreter::zerolocals : break;
1865 case Interpreter::zerolocals_synchronized: synchronized = true; break;
1866 case Interpreter::native :
1867 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1868 break;
1869 case Interpreter::native_synchronized :
1870 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1871 break;
1872 case Interpreter::empty :
1873 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1874 break;
1875 case Interpreter::accessor :
1876 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1877 break;
1878 case Interpreter::abstract :
1879 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1880 break;
1882 case Interpreter::java_lang_math_sin : // fall thru
1883 case Interpreter::java_lang_math_cos : // fall thru
1884 case Interpreter::java_lang_math_tan : // fall thru
1885 case Interpreter::java_lang_math_log : // fall thru
1886 case Interpreter::java_lang_math_log10 : // fall thru
1887 case Interpreter::java_lang_math_pow : // fall thru
1888 case Interpreter::java_lang_math_exp : break;
1889 case Interpreter::java_lang_math_abs : // fall thru
1890 case Interpreter::java_lang_math_sqrt :
1891 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1892 case Interpreter::java_lang_ref_reference_get:
1893 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1894 default:
1895 fatal(err_msg("unexpected method kind: %d", kind));
1896 break;
1897 }
1898 if (entry_point) return entry_point;
1900 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1901 }
1903 // These should never be compiled since the interpreter will prefer
1904 // // the compiled version to the intrinsic version.
1905 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1906 switch (method_kind(m)) {
1907 case Interpreter::java_lang_math_sin : // fall thru
1908 case Interpreter::java_lang_math_cos : // fall thru
1909 case Interpreter::java_lang_math_tan : // fall thru
1910 case Interpreter::java_lang_math_abs : // fall thru
1911 case Interpreter::java_lang_math_log : // fall thru
1912 case Interpreter::java_lang_math_log10 : // fall thru
1913 case Interpreter::java_lang_math_sqrt : // fall thru
1914 case Interpreter::java_lang_math_pow : // fall thru
1915 case Interpreter::java_lang_math_exp :
1916 return false;
1917 default:
1918 return true;
1919 }
1920 }
1922 // How much stack a method activation needs in words.
1923 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1925 const int entry_size = frame::interpreter_frame_monitor_size();
1927 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
1928 // be sure to change this if you add/subtract anything to/from the overhead area
1929 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1931 const int stub_code = 6; // see generate_call_stub
1932 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1933 const int method_stack = (method->max_locals() + method->max_stack()) *
1934 Interpreter::stackElementWords;
1935 return overhead_size + method_stack + stub_code;
1936 }
1938 void AbstractInterpreter::layout_activation(Method* method,
1939 int tempcount,
1940 int popframe_extra_args,
1941 int moncount,
1942 int caller_actual_parameters,
1943 int callee_param_count,
1944 int callee_locals,
1945 frame* caller,
1946 frame* interpreter_frame,
1947 bool is_top_frame,
1948 bool is_bottom_frame) {
1949 // Note: This calculation must exactly parallel the frame setup
1950 // in AbstractInterpreterGenerator::generate_method_entry.
1951 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1952 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1953 // right size, as determined by a previous call to this method.
1954 // It is also guaranteed to be walkable even though it is in a skeletal state
1956 // fixed size of an interpreter frame:
1957 // int max_locals = method->max_locals();
1959 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1960 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1962 #ifdef ASSERT
1963 if (!EnableInvokeDynamic) {
1964 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1965 // Probably, since deoptimization doesn't work yet.
1966 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1967 }
1968 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1969 #endif
1971 interpreter_frame->interpreter_frame_set_method(method);
1972 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1973 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1974 // and sender_sp is fp+8
1975 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1977 #ifdef ASSERT
1978 if (caller->is_interpreted_frame()) {
1979 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1980 }
1981 #endif
1983 interpreter_frame->interpreter_frame_set_locals(locals);
1984 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1985 BasicObjectLock* monbot = montop - moncount;
1986 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1988 //set last sp;
1989 intptr_t* esp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1990 popframe_extra_args;
1991 interpreter_frame->interpreter_frame_set_last_sp(esp);
1992 // All frames but the initial interpreter frame we fill in have a
1993 // value for sender_sp that allows walking the stack but isn't
1994 // truly correct. Correct the value here.
1995 //
1996 // int extra_locals = method->max_locals() - method->size_of_parameters();
1997 if (extra_locals != 0 &&
1998 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1999 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
2000 }
2001 *interpreter_frame->interpreter_frame_cache_addr() =
2002 method->constants()->cache();
2003 }
2005 //-----------------------------------------------------------------------------
2006 // Exceptions
2008 void TemplateInterpreterGenerator::generate_throw_exception() {
2009 // Entry point in previous activation (i.e., if the caller was
2010 // interpreted)
2011 Interpreter::_rethrow_exception_entry = __ pc();
2013 // Restore sp to interpreter_frame_last_sp even though we are going
2014 // to empty the expression stack for the exception processing.
2015 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
2017 // V0: exception
2018 // V1: return address/pc that threw exception
2019 __ restore_bcp(); // esi points to call/send
2020 __ restore_locals();
2022 //add for compressedoops
2023 __ reinit_heapbase();
2024 // Entry point for exceptions thrown within interpreter code
2025 Interpreter::_throw_exception_entry = __ pc();
2026 // expression stack is undefined here
2027 // V0: exception
2028 // BCP: exception bcp
2029 __ verify_oop(V0);
2031 // expression stack must be empty before entering the VM in case of an exception
2032 __ empty_expression_stack();
2033 // find exception handler address and preserve exception oop
2034 __ move(A1, V0);
2035 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
2036 // V0: exception handler entry point
2037 // V1: preserved exception oop
2038 // S0: bcp for exception handler
2039 __ daddi(SP, SP, (-1) * wordSize);
2040 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
2041 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
2042 __ delayed()->nop();
2044 // If the exception is not handled in the current frame the frame is removed and
2045 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
2046 //
2047 // Note: At this point the bci is still the bxi for the instruction which caused
2048 // the exception and the expression stack is empty. Thus, for any VM calls
2049 // at this point, GC will find a legal oop map (with empty expression stack).
2051 // In current activation
2052 // V0: exception
2053 // BCP: exception bcp
2055 //
2056 // JVMTI PopFrame support
2057 //
2059 Interpreter::_remove_activation_preserving_args_entry = __ pc();
2060 __ empty_expression_stack();
2061 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
2062 // currently handling popframe, so that call_VMs that may happen later do not trigger new
2063 // popframe handling cycles.
2064 #ifndef OPT_THREAD
2065 Register thread = T2;
2066 __ get_thread(T2);
2067 #else
2068 Register thread = TREG;
2069 #endif
2070 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
2071 __ ori(T3, T3, JavaThread::popframe_processing_bit);
2072 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
2074 #ifndef CORE
2075 {
2076 // Check to see whether we are returning to a deoptimized frame.
2077 // (The PopFrame call ensures that the caller of the popped frame is
2078 // either interpreted or compiled and deoptimizes it if compiled.)
2079 // In this case, we can't call dispatch_next() after the frame is
2080 // popped, but instead must save the incoming arguments and restore
2081 // them after deoptimization has occurred.
2082 //
2083 // Note that we don't compare the return PC against the
2084 // deoptimization blob's unpack entry because of the presence of
2085 // adapter frames in C2.
2086 Label caller_not_deoptimized;
2087 __ ld(A0, FP, frame::return_addr_offset * wordSize);
2088 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
2089 __ bne(V0, R0, caller_not_deoptimized);
2090 __ delayed()->nop();
2092 // Compute size of arguments for saving when returning to deoptimized caller
2093 __ get_method(A1);
2094 __ verify_oop(A1);
2095 __ ld(A1,A1,in_bytes(Method::const_offset()));
2096 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
2097 __ shl(A1, Interpreter::logStackElementSize);
2098 __ restore_locals();
2099 __ dsub(A2, LVP, T0);
2100 __ daddiu(A2, A2, wordSize);
2101 // Save these arguments
2102 #ifndef OPT_THREAD
2103 __ get_thread(A0);
2104 #else
2105 __ move(A0, TREG);
2106 #endif
2107 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
2111 __ remove_activation(vtos, T9, false, false, false);
2113 // Inform deoptimization that it is responsible for restoring these arguments
2114 #ifndef OPT_THREAD
2115 __ get_thread(thread);
2116 #endif
2117 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
2118 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2119 // Continue in deoptimization handler
2120 ///__ jmp(edx);
2121 __ jr(T9);
2122 __ delayed()->nop();
2124 __ bind(caller_not_deoptimized);
2125 }
2126 #endif /* !CORE */
2129 __ remove_activation(vtos, T3,
2130 /* throw_monitor_exception */ false,
2131 /* install_monitor_exception */ false,
2132 /* notify_jvmdi */ false);
2134 // Clear the popframe condition flag
2135 // Finish with popframe handling
2136 // A previous I2C followed by a deoptimization might have moved the
2137 // outgoing arguments further up the stack. PopFrame expects the
2138 // mutations to those outgoing arguments to be preserved and other
2139 // constraints basically require this frame to look exactly as
2140 // though it had previously invoked an interpreted activation with
2141 // no space between the top of the expression stack (current
2142 // last_sp) and the top of stack. Rather than force deopt to
2143 // maintain this kind of invariant all the time we call a small
2144 // fixup routine to move the mutated arguments onto the top of our
2145 // expression stack if necessary.
2146 //why x86 write this , i think it is no use ,@jerome
2147 //__ movl(eax, esp);
2148 //__ movl(ebx, Address(ebp, frame::interpreter_frame_last_sp_offset * wordSize));
2149 __ move(T8, SP);
2150 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2151 #ifndef OPT_THREAD
2152 __ get_thread(thread);
2153 #endif
2154 // PC must point into interpreter here
2155 //__ set_last_Java_frame(ecx, noreg, ebp, __ pc());
2156 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2157 // __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), ecx, eax, ebx);
2158 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2159 __ reset_last_Java_frame(thread, true, true);
2160 // Restore the last_sp and null it out
2161 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2162 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2166 __ move(AT, JavaThread::popframe_inactive);
2167 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2169 // Finish with popframe handling
2170 __ restore_bcp();
2171 __ restore_locals();
2172 #ifndef CORE
2173 // The method data pointer was incremented already during
2174 // call profiling. We have to restore the mdp for the current bcp.
2175 if (ProfileInterpreter) {
2176 __ set_method_data_pointer_for_bcp();
2177 }
2178 #endif // !CORE
2179 // Clear the popframe condition flag
2180 // __ get_thread(ecx);
2181 // __ movl(Address(ecx, JavaThread::popframe_condition_offset()), JavaThread::popframe_inactive);
2182 #ifndef OPT_THREAD
2183 __ get_thread(thread);
2184 #endif
2185 __ move(AT, JavaThread::popframe_inactive);
2186 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2187 __ dispatch_next(vtos);
2188 // end of PopFrame support
2190 Interpreter::_remove_activation_entry = __ pc();
2192 // preserve exception over this code sequence
2193 __ ld(T0, SP, 0);
2194 __ daddi(SP, SP, wordSize);
2195 #ifndef OPT_THREAD
2196 __ get_thread(thread);
2197 #endif
2198 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2199 // remove the activation (without doing throws on illegalMonitorExceptions)
2200 __ remove_activation(vtos, T3, false, true, false);
2201 // restore exception
2202 __ get_vm_result(T0, thread);
2203 __ verify_oop(T0);
2205 // Inbetween activations - previous activation type unknown yet
2206 // compute continuation point - the continuation point expects
2207 // the following registers set up:
2208 //
2209 // T0: exception eax
2210 // T1: return address/pc that threw exception edx
2211 // SP: expression stack of caller esp
2212 // FP: ebp of caller ebp
2213 __ daddi(SP, SP, (-2) * wordSize);
2214 __ sd(T0, SP, wordSize); // save exception
2215 __ sd(T3, SP, 0); // save return address
2216 __ move(A1, T3);
2217 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2218 __ move(T9, V0); // save exception handler
2219 __ ld(V0, SP, wordSize); // restore exception
2220 __ ld(V1, SP, 0); // restore return address
2221 __ daddi(SP, SP, 2 * wordSize);
2223 // Note that an "issuing PC" is actually the next PC after the call
2224 __ jr(T9); // jump to exception handler of caller
2225 __ delayed()->nop();
2226 }
2229 //
2230 // JVMTI ForceEarlyReturn support
2231 //
2232 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2233 address entry = __ pc();
2234 __ restore_bcp();
2235 __ restore_locals();
2236 __ empty_expression_stack();
2237 __ empty_FPU_stack();
2238 __ load_earlyret_value(state);
2240 //__ get_thread(ecx);
2241 #ifndef OPT_THREAD
2242 __ get_thread(TREG);
2243 #endif
2244 // __ movl(TREG, Address(TREG, JavaThread::jvmti_thread_state_offset()));
2245 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2246 //const Address cond_addr(ecx, JvmtiThreadState::earlyret_state_offset());
2247 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2248 // Clear the earlyret state
2249 // __ movl(cond_addr, JvmtiThreadState::earlyret_inactive);
2250 __ move(AT,JvmtiThreadState::earlyret_inactive);
2251 __ sw(AT,cond_addr);
2252 __ sync();
2253 //__ remove_activation(state, esi,
2257 __ remove_activation(state, T0,
2258 false, /* throw_monitor_exception */
2259 false, /* install_monitor_exception */
2260 true); /* notify_jvmdi */
2261 __ sync();
2262 // __ jmp(esi);
2263 //__ jmp(T0);
2264 __ jr(T0);
2265 __ delayed()->nop();
2266 return entry;
2267 } // end of ForceEarlyReturn support
2270 //-----------------------------------------------------------------------------
2271 // Helper for vtos entry point generation
2273 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2274 address& bep,
2275 address& cep,
2276 address& sep,
2277 address& aep,
2278 address& iep,
2279 address& lep,
2280 address& fep,
2281 address& dep,
2282 address& vep) {
2283 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2284 Label L;
2285 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2286 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2287 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2288 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2289 bep = cep = sep = iep = __ pc(); __ push(itos);
2290 vep = __ pc(); __ bind(L); // fall through
2291 generate_and_dispatch(t);
2292 }
2295 //-----------------------------------------------------------------------------
2296 // Generation of individual instructions
2298 // helpers for generate_and_dispatch
2301 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2302 : TemplateInterpreterGenerator(code) {
2303 generate_all(); // down here so it can be "virtual"
2304 }
2306 //-----------------------------------------------------------------------------
2308 // Non-product code
2309 #ifndef PRODUCT
2310 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2311 address entry = __ pc();
2313 // prepare expression stack
2314 __ push(state); // save tosca
2316 // tos & tos2, added by yjl 7/15/2005
2317 // trace_bytecode need actually 4 args, the last two is tos&tos2
2318 // this work fine for x86. but mips o32 call convention will store A2-A3
2319 // to the stack position it think is the tos&tos2
2320 // when the expression stack have no more than 2 data, error occur.
2321 __ ld(A2, SP, 0);
2322 __ ld(A3, SP, 1 * wordSize);
2323 {
2324 //aoqi_test
2325 /*
2326 __ pushad();
2327 __ li(A0, (long)0);
2328 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
2329 __ popad();
2330 */
2331 }
2333 // pass arguments & call tracer
2334 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2335 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2337 // restore expression stack
2338 __ pop(state); // restore tosca
2340 // return
2341 __ jr(RA);
2342 __ delayed()->nop();
2344 return entry;
2345 }
2347 void TemplateInterpreterGenerator::count_bytecode() {
2348 __ li(T8, (long)&BytecodeCounter::_counter_value);
2349 __ lw(AT, T8, 0);
2350 __ daddi(AT, AT, 1);
2351 __ sw(AT, T8, 0);
2352 }
2354 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2355 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2356 __ lw(AT, T8, 0);
2357 __ daddi(AT, AT, 1);
2358 __ sw(AT, T8, 0);
2359 }
2361 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2362 __ li(T8, (long)&BytecodePairHistogram::_index);
2363 __ lw(T9, T8, 0);
2364 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2365 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2366 __ orr(T9, T9, T8);
2367 __ li(T8, (long)&BytecodePairHistogram::_index);
2368 __ sw(T9, T8, 0);
2369 __ dsll(T9, T9, 2);
2370 __ li(T8, (long)BytecodePairHistogram::_counters);
2371 __ dadd(T8, T8, T9);
2372 __ lw(AT, T8, 0);
2373 __ daddi(AT, AT, 1);
2374 __ sw(AT, T8, 0);
2375 }
2378 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2379 // Call a little run-time stub to avoid blow-up for each bytecode.
2380 // The run-time runtime saves the right registers, depending on
2381 // the tosca in-state for the given template.
2383 address entry = Interpreter::trace_code(t->tos_in());
2384 assert(entry != NULL, "entry must have been generated");
2385 __ call(entry, relocInfo::none);
2386 __ delayed()->nop();
2387 //add for compressedoops
2388 __ reinit_heapbase();
2389 }
2392 void TemplateInterpreterGenerator::stop_interpreter_at() {
2393 Label L;
2394 __ li(T8, long(&BytecodeCounter::_counter_value));
2395 __ lw(T8, T8, 0);
2396 __ move(AT, StopInterpreterAt);
2397 __ bne(T8, AT, L);
2398 __ delayed()->nop();
2399 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2400 __ delayed()->nop();
2401 __ bind(L);
2402 }
2403 #endif // !PRODUCT
2404 #endif // ! CC_INTERP