Tue, 26 Sep 2017 17:05:14 +0800
[Interpreter] Use b*_far for InterpreterGenerator::generate_counter_incr.
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 // FIXME: why do not pass parameter thread ?
112 __ call_VM(NOREG, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
113 return entry;
114 }
116 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(
117 const char* name) {
118 address entry = __ pc();
119 // expression stack must be empty before entering the VM if an
120 // exception happened
121 __ empty_expression_stack();
122 __ li(A1, (long)name);
123 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
124 InterpreterRuntime::throw_ArrayIndexOutOfBoundsException), A1, A2);
125 return entry;
126 }
128 address TemplateInterpreterGenerator::generate_ClassCastException_handler() {
129 address entry = __ pc();
131 // object is at TOS
132 //FIXME, I am not sure if the object is at TOS as x86 do now @jerome, 04/20,2007
133 //__ pop(c_rarg1);
135 // expression stack must be empty before entering the VM if an
136 // exception happened
137 __ empty_expression_stack();
138 __ empty_FPU_stack();
139 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), FSR);
140 return entry;
141 }
143 address TemplateInterpreterGenerator::generate_exception_handler_common(
144 const char* name, const char* message, bool pass_oop) {
145 assert(!pass_oop || message == NULL, "either oop or message but not both");
146 address entry = __ pc();
148 // expression stack must be empty before entering the VM if an exception happened
149 __ empty_expression_stack();
150 // setup parameters
151 __ li(A1, (long)name);
152 if (pass_oop) {
153 __ call_VM(V0,
154 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), A1, FSR);
155 } else {
156 __ li(A2, (long)message);
157 __ call_VM(V0,
158 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), A1, A2);
159 }
160 // throw exception
161 __ jmp(Interpreter::throw_exception_entry(), relocInfo::none);
162 __ delayed()->nop();
163 return entry;
164 }
167 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
168 address entry = __ pc();
169 // NULL last_sp until next java call
170 __ sd(R0,Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
171 __ dispatch_next(state);
172 return entry;
173 }
176 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
178 address entry = __ pc();
180 // Restore stack bottom in case i2c adjusted stack
181 __ ld(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
182 // and NULL it as marker that esp is now tos until next java call
183 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
185 __ restore_bcp();
186 __ restore_locals();
188 // 2014/11/24 Fu
189 // mdp: T8
190 // ret: FSR
191 // tmp: T9
192 if (state == atos) {
193 Register mdp = T8;
194 Register tmp = T9;
195 __ profile_return_type(mdp, FSR, tmp);
196 }
199 const Register cache = T9;
200 const Register index = T3;
201 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
203 const Register flags = cache;
204 __ dsll(AT, index, Address::times_ptr);
205 __ daddu(AT, cache, AT);
206 __ lw(flags, AT, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
207 __ andi(flags, flags, ConstantPoolCacheEntry::parameter_size_mask);
208 __ dsll(AT, flags, Interpreter::stackElementScale());
209 __ daddu(SP, SP, AT);
211 __ dispatch_next(state, step);
213 return entry;
214 }
217 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
218 int step) {
219 address entry = __ pc();
220 // NULL last_sp until next java call
221 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
222 __ restore_bcp();
223 __ restore_locals();
224 // handle exceptions
225 {
226 Label L;
227 const Register thread = TREG;
228 #ifndef OPT_THREAD
229 __ get_thread(thread);
230 #endif
231 __ lw(AT, thread, in_bytes(Thread::pending_exception_offset()));
232 __ beq(AT, R0, L);
233 __ delayed()->nop();
234 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
235 __ should_not_reach_here();
236 __ bind(L);
237 }
238 __ dispatch_next(state, step);
239 return entry;
240 }
242 int AbstractInterpreter::BasicType_as_index(BasicType type) {
243 int i = 0;
244 switch (type) {
245 case T_BOOLEAN: i = 0; break;
246 case T_CHAR : i = 1; break;
247 case T_BYTE : i = 2; break;
248 case T_SHORT : i = 3; break;
249 case T_INT : // fall through
250 case T_LONG : // fall through
251 case T_VOID : i = 4; break;
252 case T_FLOAT : i = 5; break;
253 case T_DOUBLE : i = 6; break;
254 case T_OBJECT : // fall through
255 case T_ARRAY : i = 7; break;
256 default : ShouldNotReachHere();
257 }
258 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
259 "index out of bounds");
260 return i;
261 }
264 // why do not consider float and double , @jerome, 12/27,06, @jerome
265 //FIXME, aoqi
266 address TemplateInterpreterGenerator::generate_result_handler_for(
267 BasicType type) {
268 address entry = __ pc();
269 switch (type) {
270 case T_BOOLEAN: __ c2bool(V0); break;
271 case T_CHAR : __ andi(V0, V0, 0xFFFF); break;
272 case T_BYTE : __ sign_extend_byte (V0); break;
273 case T_SHORT : __ sign_extend_short(V0); break;
274 case T_INT : /* nothing to do */ break;
275 case T_FLOAT : /* nothing to do */ break;
276 case T_DOUBLE : /* nothing to do */ break;
277 case T_OBJECT :
278 {
279 __ ld(V0, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
280 __ verify_oop(V0); // and verify it
281 }
282 break;
283 default : ShouldNotReachHere();
284 }
285 __ jr(RA); // return from result handler
286 __ delayed()->nop();
287 return entry;
288 }
290 address TemplateInterpreterGenerator::generate_safept_entry_for(
291 TosState state,
292 address runtime_entry) {
293 address entry = __ pc();
294 __ push(state);
295 __ call_VM(noreg, runtime_entry);
296 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
297 return entry;
298 }
302 // Helpers for commoning out cases in the various type of method entries.
303 //
306 // increment invocation count & check for overflow
307 //
308 // Note: checking for negative value instead of overflow
309 // so we have a 'sticky' overflow test
310 //
311 // prerequisites : method in T0, invocation counter in T3
312 void InterpreterGenerator::generate_counter_incr(
313 Label* overflow,
314 Label* profile_method,
315 Label* profile_method_continue) {
316 Label done;
317 const Address invocation_counter(FSR, in_bytes(MethodCounters::invocation_counter_offset())
318 + in_bytes(InvocationCounter::counter_offset()));
319 const Address backedge_counter (FSR, in_bytes(MethodCounters::backedge_counter_offset())
320 + in_bytes(InvocationCounter::counter_offset()));
322 __ get_method_counters(Rmethod, FSR, done);
324 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
325 __ lw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
326 __ incrementl(T9, 1);
327 __ sw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
328 }
329 // Update standard invocation counters
330 __ lw(T3, invocation_counter);
331 __ increment(T3, InvocationCounter::count_increment);
332 __ sw(T3, invocation_counter); // save invocation count
334 __ lw(FSR, backedge_counter); // load backedge counter
335 __ li(AT, InvocationCounter::count_mask_value); // mask out the status bits
336 __ andr(FSR, FSR, AT);
338 __ dadd(T3, T3, FSR); // add both counters
340 if (ProfileInterpreter && profile_method != NULL) {
341 // Test to see if we should create a method data oop
342 if (Assembler::is_simm16(InvocationCounter::InterpreterProfileLimit)) {
343 __ slti(AT, T3, InvocationCounter::InterpreterProfileLimit);
344 } else {
345 __ li(AT, (long)&InvocationCounter::InterpreterProfileLimit);
346 __ lw(AT, AT, 0);
347 __ slt(AT, T3, AT);
348 }
350 __ bne_far(AT, R0, *profile_method_continue);
351 __ delayed()->nop();
353 // if no method data exists, go to profile_method
354 __ test_method_data_pointer(FSR, *profile_method);
355 }
357 if (Assembler::is_simm16(CompileThreshold)) {
358 __ srl(AT, T3, InvocationCounter::count_shift);
359 __ slti(AT, AT, CompileThreshold);
360 } else {
361 __ li(AT, (long)&InvocationCounter::InterpreterInvocationLimit);
362 __ lw(AT, AT, 0);
363 __ slt(AT, T3, AT);
364 }
366 __ beq_far(AT, R0, *overflow);
367 __ delayed()->nop();
368 __ bind(done);
369 }
371 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
373 // Asm interpreter on entry
374 // S7 - locals
375 // S0 - bcp
376 // Rmethod - method
377 // FP - interpreter frame
379 // On return (i.e. jump to entry_point)
380 // Rmethod - method
381 // RA - return address of interpreter caller
382 // tos - the last parameter to Java method
383 // SP - sender_sp
385 //const Address size_of_parameters(Rmethod,in_bytes( Method::size_of_parameters_offset()));
387 // the bcp is valid if and only if it's not null
388 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
389 InterpreterRuntime::frequency_counter_overflow), R0);
390 __ ld(Rmethod, FP, method_offset);
391 // Preserve invariant that esi/edi contain bcp/locals of sender frame
392 __ b_far(*do_continue);
393 __ delayed()->nop();
394 }
396 // See if we've got enough room on the stack for locals plus overhead.
397 // The expression stack grows down incrementally, so the normal guard
398 // page mechanism will work for that.
399 //
400 // NOTE: Since the additional locals are also always pushed (wasn't
401 // obvious in generate_method_entry) so the guard should work for them
402 // too.
403 //
404 // Args:
405 // rdx: number of additional locals this frame needs (what we must check)
406 // rbx: Method*
407 //
408 // Kills:
409 // rax
410 void InterpreterGenerator::generate_stack_overflow_check(void) {
411 // see if we've got enough room on the stack for locals plus overhead.
412 // the expression stack grows down incrementally, so the normal guard
413 // page mechanism will work for that.
414 //
415 // Registers live on entry:
416 //
417 // T0: Method*
418 // T2: number of additional locals this frame needs (what we must check)
420 // NOTE: since the additional locals are also always pushed (wasn't obvious in
421 // generate_method_entry) so the guard should work for them too.
422 //
424 // monitor entry size: see picture of stack set (generate_method_entry) and frame_i486.hpp
425 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
427 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
428 // be sure to change this if you add/subtract anything to/from the overhead area
429 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize)
430 + entry_size;
432 const int page_size = os::vm_page_size();
434 Label after_frame_check;
436 // see if the frame is greater than one page in size. If so,
437 // then we need to verify there is enough stack space remaining
438 // for the additional locals.
439 __ move(AT, (page_size - overhead_size) / Interpreter::stackElementSize);
440 __ slt(AT, AT, T2);
441 __ beq(AT, R0, after_frame_check);
442 __ delayed()->nop();
444 // compute sp as if this were going to be the last frame on
445 // the stack before the red zone
446 #ifndef OPT_THREAD
447 Register thread = T1;
448 __ get_thread(thread);
449 #else
450 Register thread = TREG;
451 #endif
453 // locals + overhead, in bytes
454 //FIXME aoqi
455 __ dsll(T3, T2, Interpreter::stackElementScale());
456 __ daddiu(T3, T3, overhead_size); // locals * 4 + overhead_size --> T3
458 #ifdef ASSERT
459 Label stack_base_okay, stack_size_okay;
460 // verify that thread stack base is non-zero
461 __ ld(AT, thread, in_bytes(Thread::stack_base_offset()));
462 __ bne(AT, R0, stack_base_okay);
463 __ delayed()->nop();
464 __ stop("stack base is zero");
465 __ bind(stack_base_okay);
466 // verify that thread stack size is non-zero
467 __ ld(AT, thread, in_bytes(Thread::stack_size_offset()));
468 __ bne(AT, R0, stack_size_okay);
469 __ delayed()->nop();
470 __ stop("stack size is zero");
471 __ bind(stack_size_okay);
472 #endif
474 // Add stack base to locals and subtract stack size
475 __ ld(AT, thread, in_bytes(Thread::stack_base_offset())); // stack_base --> AT
476 __ dadd(T3, T3, AT); // locals * 4 + overhead_size + stack_base--> T3
477 __ ld(AT, thread, in_bytes(Thread::stack_size_offset())); // stack_size --> AT
478 __ dsub(T3, T3, AT); // locals * 4 + overhead_size + stack_base - stack_size --> T3
481 // add in the redzone and yellow size
482 __ move(AT, (StackRedPages+StackYellowPages) * page_size);
483 __ add(T3, T3, AT);
485 // check against the current stack bottom
486 __ slt(AT, T3, SP);
487 __ bne(AT, R0, after_frame_check);
488 __ delayed()->nop();
490 // Note: the restored frame is not necessarily interpreted.
491 // Use the shared runtime version of the StackOverflowError.
492 __ move(SP, Rsender);
493 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
494 __ jmp(StubRoutines::throw_StackOverflowError_entry(), relocInfo::runtime_call_type);
495 __ delayed()->nop();
497 // all done with frame size check
498 __ bind(after_frame_check);
499 }
501 // Allocate monitor and lock method (asm interpreter)
502 // Rmethod - Method*
503 void InterpreterGenerator::lock_method(void) {
504 // synchronize method
505 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
507 #ifdef ASSERT
508 { Label L;
509 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
510 __ andi(T0, T0, JVM_ACC_SYNCHRONIZED);
511 __ bne(T0, R0, L);
512 __ delayed()->nop();
513 __ stop("method doesn't need synchronization");
514 __ bind(L);
515 }
516 #endif // ASSERT
517 // get synchronization object
518 {
519 Label done;
520 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
521 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
522 __ andi(T2, T0, JVM_ACC_STATIC);
523 __ ld(T0, LVP, Interpreter::local_offset_in_bytes(0));
524 __ beq(T2, R0, done);
525 __ delayed()->nop();
526 __ ld(T0, Rmethod, in_bytes(Method::const_offset()));
527 __ ld(T0, T0, in_bytes(ConstMethod::constants_offset()));
528 __ ld(T0, T0, ConstantPool::pool_holder_offset_in_bytes());
529 __ ld(T0, T0, mirror_offset);
530 __ bind(done);
531 }
532 // add space for monitor & lock
533 __ daddi(SP, SP, (-1) * entry_size); // add space for a monitor entry
534 __ sd(SP, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
535 // set new monitor block top
536 __ sd(T0, SP, BasicObjectLock::obj_offset_in_bytes()); // store object
537 // FIXME: I do not know what lock_object will do and what it will need
538 __ move(c_rarg0, SP); // object address
539 __ lock_object(c_rarg0);
540 }
542 // Generate a fixed interpreter frame. This is identical setup for
543 // interpreted methods and for native methods hence the shared code.
544 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
546 // [ local var m-1 ] <--- sp
547 // ...
548 // [ local var 0 ]
549 // [ argumnet word n-1 ] <--- T0(sender's sp)
550 // ...
551 // [ argument word 0 ] <--- S7
553 // initialize fixed part of activation frame
554 // sender's sp in Rsender
555 int i = 0;
556 __ sd(RA, SP, (-1) * wordSize); // save return address
557 __ sd(FP, SP, (-2) * wordSize); // save sender's fp
558 __ daddiu(FP, SP, (-2) * wordSize);
559 __ sd(Rsender, FP, (-++i) * wordSize); // save sender's sp
560 __ sd(R0, FP,(-++i)*wordSize); //save last_sp as null, FIXME aoqi
561 __ sd(LVP, FP, (-++i) * wordSize); // save locals offset
562 __ ld(BCP, Rmethod, in_bytes(Method::const_offset())); // get constMethodOop
563 __ daddiu(BCP, BCP, in_bytes(ConstMethod::codes_offset())); // get codebase
564 __ sd(Rmethod, FP, (-++i) * wordSize); // save Method*
565 #ifndef CORE
566 if (ProfileInterpreter) {
567 Label method_data_continue;
568 __ ld(AT, Rmethod, in_bytes(Method::method_data_offset()));
569 __ beq(AT, R0, method_data_continue);
570 __ delayed()->nop();
571 __ daddi(AT, AT, in_bytes(MethodData::data_offset()));
572 __ bind(method_data_continue);
573 __ sd(AT, FP, (-++i) * wordSize);
574 } else {
575 __ sd(R0, FP, (-++i) * wordSize);
576 }
577 #endif // !CORE
579 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
580 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
581 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
582 __ sd(T2, FP, (-++i) * wordSize); // set constant pool cache
583 if (native_call) {
584 __ sd(R0, FP, (-++i) * wordSize); // no bcp
585 } else {
586 __ sd(BCP, FP, (-++i) * wordSize); // set bcp
587 }
588 __ daddiu(SP, FP, (-++i) * wordSize);
589 __ sd(SP, FP, (-i) * wordSize); // reserve word for pointer to expression stack bottom
590 }
592 // End of helpers
594 // Various method entries
595 //------------------------------------------------------------------------------------------------------------------------
596 //
597 //
599 // Call an accessor method (assuming it is resolved, otherwise drop
600 // into vanilla (slow path) entry
601 address InterpreterGenerator::generate_accessor_entry(void) {
603 // Rmethod: Method*
604 // V0: receiver (preserve for slow entry into asm interpreter)
605 // Rsender: senderSP must preserved for slow path, set SP to it on fast path
607 address entry_point = __ pc();
608 Label xreturn_path;
609 // do fastpath for resolved accessor methods
610 if (UseFastAccessorMethods) {
611 Label slow_path;
612 __ li(T2, SafepointSynchronize::address_of_state());
613 __ lw(AT, T2, 0);
614 __ daddi(AT, AT, -(SafepointSynchronize::_not_synchronized));
615 __ bne(AT, R0, slow_path);
616 __ delayed()->nop();
617 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof;
618 // parameter size = 1
619 // Note: We can only use this code if the getfield has been resolved
620 // and if we don't have a null-pointer exception => check for
621 // these conditions first and use slow path if necessary.
622 // Rmethod: method
623 // V0: receiver
625 // [ receiver ] <-- sp
626 __ ld(T0, SP, 0);
628 // check if local 0 != NULL and read field
629 __ beq(T0, R0, slow_path);
630 __ delayed()->nop();
631 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
632 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
633 // read first instruction word and extract bytecode @ 1 and index @ 2
634 __ ld(T3, Rmethod, in_bytes(Method::const_offset()));
635 __ lw(T3, T3, in_bytes(ConstMethod::codes_offset()));
636 // Shift codes right to get the index on the right.
637 // The bytecode fetched looks like <index><0xb4><0x2a>
638 __ dsrl(T3, T3, 2 * BitsPerByte);
639 // FIXME: maybe it's wrong
640 __ dsll(T3, T3, exact_log2(in_words(ConstantPoolCacheEntry::size())));
641 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
643 // T0: local 0 eax
644 // Rmethod: method ebx
645 // V0: receiver - do not destroy since it is needed for slow path! ecx
646 // ecx: scratch use which register instead ?
647 // T1: scratch use which register instead ?
648 // T3: constant pool cache index edx
649 // T2: constant pool cache edi
650 // esi: send's sp
651 // Rsender: send's sp
652 // check if getfield has been resolved and read constant pool cache entry
653 // check the validity of the cache entry by testing whether _indices field
654 // contains Bytecode::_getfield in b1 byte.
655 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
656 // __ movl(esi,
657 // Address(edi,
658 // edx,
659 // Address::times_4, ConstantPoolCache::base_offset()
660 // + ConstantPoolCacheEntry::indices_offset()));
663 __ dsll(T8, T3, Address::times_8);
664 __ move(T1, in_bytes(ConstantPoolCache::base_offset()
665 + ConstantPoolCacheEntry::indices_offset()));
666 __ dadd(T1, T8, T1);
667 __ dadd(T1, T1, T2);
668 __ lw(T1, T1, 0);
669 __ dsrl(T1, T1, 2 * BitsPerByte);
670 __ andi(T1, T1, 0xFF);
671 __ daddi(T1, T1, (-1) * Bytecodes::_getfield);
672 __ bne(T1, R0, slow_path);
673 __ delayed()->nop();
675 // Note: constant pool entry is not valid before bytecode is resolved
677 __ move(T1, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
678 __ dadd(T1, T1, T8);
679 __ dadd(T1, T1, T2);
680 __ lw(AT, T1, 0);
682 __ move(T1, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
683 __ dadd(T1, T1, T8);
684 __ dadd(T1, T1, T2);
685 __ lw(T3, T1, 0);
687 Label notByte, notShort, notChar, notObj;
688 // const Address field_address (eax, esi, Address::times_1);
690 // Need to differentiate between igetfield, agetfield, bgetfield etc.
691 // because they are different sizes.
692 // Use the type from the constant pool cache
693 __ dsrl(T3, T3, ConstantPoolCacheEntry::tos_state_shift);
694 // Make sure we don't need to mask edx for tosBits after the above shift
695 ConstantPoolCacheEntry::verify_tos_state_shift();
696 // btos = 0
697 __ bne(T3, R0, notByte);
698 __ delayed()->dadd(T0, T0, AT);
700 __ lb(V0, T0, 0);
701 __ b(xreturn_path);
702 __ delayed()->nop();
704 //stos
705 __ bind(notByte);
706 __ daddi(T1, T3, (-1) * stos);
707 __ bne(T1, R0, notShort);
708 __ delayed()->nop();
709 __ lh(V0, T0, 0);
710 __ b(xreturn_path);
711 __ delayed()->nop();
713 //ctos
714 __ bind(notShort);
715 __ daddi(T1, T3, (-1) * ctos);
716 __ bne(T1, R0, notChar);
717 __ delayed()->nop();
718 __ lhu(V0, T0, 0);
719 __ b(xreturn_path);
720 __ delayed()->nop();
722 //atos
723 __ bind(notChar);
724 __ daddi(T1, T3, (-1) * atos);
725 __ bne(T1, R0, notObj);
726 __ delayed()->nop();
727 //add for compressedoops
728 __ load_heap_oop(V0, Address(T0, 0));
729 __ b(xreturn_path);
730 __ delayed()->nop();
732 //itos
733 __ bind(notObj);
734 #ifdef ASSERT
735 Label okay;
736 __ daddi(T1, T3, (-1) * itos);
737 __ beq(T1, R0, okay);
738 __ delayed()->nop();
739 __ stop("what type is this?");
740 __ bind(okay);
741 #endif // ASSERT
742 __ lw(V0, T0, 0);
744 __ bind(xreturn_path);
746 // _ireturn/_areturn
747 //FIXME
748 __ move(SP, Rsender);//FIXME, set sender's fp to SP
749 __ jr(RA);
750 __ delayed()->nop();
752 // generate a vanilla interpreter entry as the slow path
753 __ bind(slow_path);
754 (void) generate_normal_entry(false);
755 } else {
756 (void) generate_normal_entry(false);
757 }
759 return entry_point;
760 }
762 // Method entry for java.lang.ref.Reference.get.
763 address InterpreterGenerator::generate_Reference_get_entry(void) {
764 #if INCLUDE_ALL_GCS
765 // Code: _aload_0, _getfield, _areturn
766 // parameter size = 1
767 //
768 // The code that gets generated by this routine is split into 2 parts:
769 // 1. The "intrinsified" code for G1 (or any SATB based GC),
770 // 2. The slow path - which is an expansion of the regular method entry.
771 //
772 // Notes:-
773 // * In the G1 code we do not check whether we need to block for
774 // a safepoint. If G1 is enabled then we must execute the specialized
775 // code for Reference.get (except when the Reference object is null)
776 // so that we can log the value in the referent field with an SATB
777 // update buffer.
778 // If the code for the getfield template is modified so that the
779 // G1 pre-barrier code is executed when the current method is
780 // Reference.get() then going through the normal method entry
781 // will be fine.
782 // * The G1 code can, however, check the receiver object (the instance
783 // of java.lang.Reference) and jump to the slow path if null. If the
784 // Reference object is null then we obviously cannot fetch the referent
785 // and so we don't need to call the G1 pre-barrier. Thus we can use the
786 // regular method entry code to generate the NPE.
787 //
788 // This code is based on generate_accessor_enty.
789 //
790 // rbx: Method*
792 // r13: senderSP must preserve for slow path, set SP to it on fast path
794 address entry = __ pc();
796 const int referent_offset = java_lang_ref_Reference::referent_offset;
797 guarantee(referent_offset > 0, "referent offset not initialized");
799 if (UseG1GC) {
800 Unimplemented();
801 Label slow_path;
802 // rbx: method
804 // Check if local 0 != NULL
805 // If the receiver is null then it is OK to jump to the slow path.
807 // rax: local 0
808 // rbx: method (but can be used as scratch now)
809 // rdx: scratch
810 // rdi: scratch
812 // Generate the G1 pre-barrier code to log the value of
813 // the referent field in an SATB buffer.
815 // Load the value of the referent field.
817 return entry;
818 }
819 #endif // INCLUDE_ALL_GCS
821 // If G1 is not enabled then attempt to go through the accessor entry point
822 // Reference.get is an accessor
823 return generate_accessor_entry();
824 }
825 // Interpreter stub for calling a native method. (asm interpreter)
826 // This sets up a somewhat different looking stack for calling the
827 // native method than the typical interpreter frame setup.
828 address InterpreterGenerator::generate_native_entry(bool synchronized) {
829 // determine code generation flags
830 bool inc_counter = UseCompiler || CountCompiledCalls;
831 // Rsender: sender's sp
832 // Rmethod: Method*
833 address entry_point = __ pc();
835 #ifndef CORE
836 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() + // Fu: 20130814
837 InvocationCounter::counter_offset()));
838 #endif
840 // get parameter size (always needed)
841 // the size in the java stack
842 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
843 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
845 // native calls don't need the stack size check since they have no expression stack
846 // and the arguments are already on the stack and we only add a handful of words
847 // to the stack
849 // Rmethod: Method*
850 // V0: size of parameters
851 // Layout of frame at this point
852 //
853 // [ argument word n-1 ] <--- sp
854 // ...
855 // [ argument word 0 ]
857 // for natives the size of locals is zero
859 // compute beginning of parameters (S7)
860 __ dsll(LVP, V0, Address::times_8);
861 __ daddiu(LVP, LVP, (-1) * wordSize);
862 __ dadd(LVP, LVP, SP);
865 // add 2 zero-initialized slots for native calls
866 __ daddi(SP, SP, (-2) * wordSize);
867 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
868 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
870 // Layout of frame at this point
871 // [ method holder mirror ] <--- sp
872 // [ result type info ]
873 // [ argument word n-1 ] <--- T0
874 // ...
875 // [ argument word 0 ] <--- LVP
878 #ifndef CORE
879 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
880 #endif
882 // initialize fixed part of activation frame
883 generate_fixed_frame(true);
884 // after this function, the layout of frame is as following
885 //
886 // [ monitor block top ] <--- sp ( the top monitor entry )
887 // [ byte code pointer (0) ] (if native, bcp = 0)
888 // [ constant pool cache ]
889 // [ Method* ]
890 // [ locals offset ]
891 // [ sender's sp ]
892 // [ sender's fp ]
893 // [ return address ] <--- fp
894 // [ method holder mirror ]
895 // [ result type info ]
896 // [ argumnet word n-1 ] <--- sender's sp
897 // ...
898 // [ argument word 0 ] <--- S7
901 // make sure method is native & not abstract
902 #ifdef ASSERT
903 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
904 {
905 Label L;
906 __ andi(AT, T0, JVM_ACC_NATIVE);
907 __ bne(AT, R0, L);
908 __ delayed()->nop();
909 __ stop("tried to execute native method as non-native");
910 __ bind(L);
911 }
912 {
913 Label L;
914 __ andi(AT, T0, JVM_ACC_ABSTRACT);
915 __ beq(AT, R0, L);
916 __ delayed()->nop();
917 __ stop("tried to execute abstract method in interpreter");
918 __ bind(L);
919 }
920 #endif
922 // Since at this point in the method invocation the exception handler
923 // would try to exit the monitor of synchronized methods which hasn't
924 // been entered yet, we set the thread local variable
925 // _do_not_unlock_if_synchronized to true. The remove_activation will
926 // check this flag.
927 Register thread = TREG;
928 #ifndef OPT_THREAD
929 __ get_thread(thread);
930 #endif
931 __ move(AT, (int)true);
932 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
934 #ifndef CORE
935 // increment invocation count & check for overflow
936 Label invocation_counter_overflow;
937 if (inc_counter) {
938 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
939 }
941 Label continue_after_compile;
942 __ bind(continue_after_compile);
943 #endif // CORE
945 bang_stack_shadow_pages(true);
947 // reset the _do_not_unlock_if_synchronized flag
948 #ifndef OPT_THREAD
949 __ get_thread(thread);
950 #endif
951 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
953 // check for synchronized methods
954 // Must happen AFTER invocation_counter check and stack overflow check,
955 // so method is not locked if overflows.
956 if (synchronized) {
957 lock_method();
958 } else {
959 // no synchronization necessary
960 #ifdef ASSERT
961 {
962 Label L;
963 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
964 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
965 __ beq(AT, R0, L);
966 __ delayed()->nop();
967 __ stop("method needs synchronization");
968 __ bind(L);
969 }
970 #endif
971 }
973 // after method_lock, the layout of frame is as following
974 //
975 // [ monitor entry ] <--- sp
976 // ...
977 // [ monitor entry ]
978 // [ monitor block top ] ( the top monitor entry )
979 // [ byte code pointer (0) ] (if native, bcp = 0)
980 // [ constant pool cache ]
981 // [ Method* ]
982 // [ locals offset ]
983 // [ sender's sp ]
984 // [ sender's fp ]
985 // [ return address ] <--- fp
986 // [ method holder mirror ]
987 // [ result type info ]
988 // [ argumnet word n-1 ] <--- ( sender's sp )
989 // ...
990 // [ argument word 0 ] <--- S7
992 // start execution
993 #ifdef ASSERT
994 {
995 Label L;
996 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
997 __ beq(AT, SP, L);
998 __ delayed()->nop();
999 __ stop("broken stack frame setup in interpreter in asm");
1000 __ bind(L);
1001 }
1002 #endif
1004 // jvmti/jvmpi support
1005 __ notify_method_entry();
1007 // work registers
1008 const Register method = Rmethod;
1009 //const Register thread = T2;
1010 const Register t = RT4;
1012 __ get_method(method);
1013 __ verify_oop(method);
1014 {
1015 Label L, Lstatic;
1016 __ ld(t,method,in_bytes(Method::const_offset()));
1017 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
1018 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1019 //FIXME, aoqi: A1?
1020 // A0 and A1(if needed)
1021 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1022 __ andi(AT, AT, JVM_ACC_STATIC);
1023 __ beq(AT, R0, Lstatic);
1024 __ delayed()->nop();
1025 __ daddiu(t, t, 1);
1026 __ bind(Lstatic);
1027 __ daddiu(t, t, -7);
1028 __ blez(t, L);
1029 __ delayed()->nop();
1030 __ dsll(t, t, Address::times_8);
1031 __ dsub(SP, SP, t);
1032 __ bind(L);
1033 }
1034 __ move(AT, -(StackAlignmentInBytes));
1035 __ andr(SP, SP, AT);
1036 __ move(AT, SP);
1037 // [ ] <--- sp
1038 // ... (size of parameters - 8 )
1039 // [ monitor entry ]
1040 // ...
1041 // [ monitor entry ]
1042 // [ monitor block top ] ( the top monitor entry )
1043 // [ byte code pointer (0) ] (if native, bcp = 0)
1044 // [ constant pool cache ]
1045 // [ Method* ]
1046 // [ locals offset ]
1047 // [ sender's sp ]
1048 // [ sender's fp ]
1049 // [ return address ] <--- fp
1050 // [ method holder mirror ]
1051 // [ result type info ]
1052 // [ argumnet word n-1 ] <--- ( sender's sp )
1053 // ...
1054 // [ argument word 0 ] <--- LVP
1056 // get signature handler
1057 {
1058 Label L;
1059 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1060 __ bne(T9, R0, L);
1061 __ delayed()->nop();
1062 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1063 InterpreterRuntime::prepare_native_call), method);
1064 __ get_method(method);
1065 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1066 __ bind(L);
1067 }
1069 // call signature handler
1070 // FIXME: when change codes in InterpreterRuntime, note this point
1071 // from: begin of parameters
1072 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1073 // to: current sp
1074 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1075 // temp: T3
1076 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1078 __ jalr(T9);
1079 __ delayed()->nop();
1080 __ get_method(method); // slow path call blows EBX on DevStudio 5.0
1082 /*
1083 if native function is static, and its second parameter has type length of double word,
1084 and first parameter has type length of word, we have to reserve one word
1085 for the first parameter, according to mips o32 abi.
1086 if native function is not static, and its third parameter has type length of double word,
1087 and second parameter has type length of word, we have to reserve one word for the second
1088 parameter.
1089 */
1092 // result handler is in V0
1093 // set result handler
1094 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1096 #define FIRSTPARA_SHIFT_COUNT 5
1097 #define SECONDPARA_SHIFT_COUNT 9
1098 #define THIRDPARA_SHIFT_COUNT 13
1099 #define PARA_MASK 0xf
1101 // pass mirror handle if static call
1102 {
1103 Label L;
1104 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1105 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1106 __ andi(AT, t, JVM_ACC_STATIC);
1107 __ beq(AT, R0, L);
1108 __ delayed()->nop();
1110 // get mirror
1111 __ ld(t, method, in_bytes(Method:: const_offset()));
1112 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1113 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1114 __ ld(t, t, mirror_offset);
1115 // copy mirror into activation frame
1116 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1117 // pass handle to mirror
1118 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1119 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1120 __ move(A1, t);
1121 __ bind(L);
1122 }
1124 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1125 // [ ] |
1126 // ... size of parameters(or +1) |
1127 // [ monitor entry ] |
1128 // ... |
1129 // [ monitor entry ] |
1130 // [ monitor block top ] ( the top monitor entry ) |
1131 // [ byte code pointer (0) ] (if native, bcp = 0) |
1132 // [ constant pool cache ] |
1133 // [ Method* ] |
1134 // [ locals offset ] |
1135 // [ sender's sp ] |
1136 // [ sender's fp ] |
1137 // [ return address ] <--- fp |
1138 // [ method holder mirror ] <----------------------------|
1139 // [ result type info ]
1140 // [ argumnet word n-1 ] <--- ( sender's sp )
1141 // ...
1142 // [ argument word 0 ] <--- S7
1144 // get native function entry point
1145 { Label L;
1146 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1147 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1148 __ bne(V1, T9, L);
1149 __ delayed()->nop();
1150 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1151 __ get_method(method);
1152 __ verify_oop(method);
1153 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1154 __ bind(L);
1155 }
1156 /*
1157 __ pushad();
1158 __ move(A0, T9);
1159 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
1160 __ popad();
1161 */
1163 // pass JNIEnv
1164 // native function in T9
1165 #ifndef OPT_THREAD
1166 __ get_thread(thread);
1167 #endif
1168 __ daddi(t, thread, in_bytes(JavaThread::jni_environment_offset()));
1169 // stack,but I think it won't work when pass float,double etc @jerome,10/17,2006
1170 __ move(A0, t);
1171 // [ jni environment ] <--- sp
1172 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1173 // [ ] |
1174 // ... size of parameters |
1175 // [ monitor entry ] |
1176 // ... |
1177 // [ monitor entry ] |
1178 // [ monitor block top ] ( the top monitor entry ) |
1179 // [ byte code pointer (0) ] (if native, bcp = 0) |
1180 // [ constant pool cache ] |
1181 // [ Method* ] |
1182 // [ locals offset ] |
1183 // [ sender's sp ] |
1184 // [ sender's fp ] |
1185 // [ return address ] <--- fp |
1186 // [ method holder mirror ] <----------------------------|
1187 // [ result type info ]
1188 // [ argumnet word n-1 ] <--- ( sender's sp )
1189 // ...
1190 // [ argument word 0 ] <--- S7
1192 // set_last_Java_frame_before_call
1193 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1194 // Change state to native (we save the return address in the thread, since it might not
1195 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1196 // points into the right code segment. It does not have to be the correct return pc.
1197 __ li(t, __ pc());
1198 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1199 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1201 // change thread state
1202 #ifdef ASSERT
1203 {
1204 Label L;
1205 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1206 __ daddi(t, t, (-1) * _thread_in_Java);
1207 __ beq(t, R0, L);
1208 __ delayed()->nop();
1209 __ stop("Wrong thread state in native stub");
1210 __ bind(L);
1211 }
1212 #endif
1214 __ move(t, _thread_in_native);
1215 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1217 // call native method
1218 __ jalr(T9);
1219 __ delayed()->nop();
1220 // result potentially in V2:V1 or F0:F1
1223 // via _last_native_pc and not via _last_jave_sp
1224 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1225 // If the order changes or anything else is added to the stack the code in
1226 // interpreter_frame_result will have to be changed.
1227 //FIXME, should modify here
1228 // save return value to keep the value from being destroyed by other calls
1229 __ move(S1, V0);
1230 __ move(S3, V1);
1231 __ dmfc1(S4, F0);
1232 __ dmfc1(S2, F1);
1234 // change thread state
1235 __ get_thread(thread);
1236 __ move(t, _thread_in_native_trans);
1237 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1239 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1241 // check for safepoint operation in progress and/or pending suspend requests
1242 { Label Continue;
1244 // Don't use call_VM as it will see a possible pending exception and forward it
1245 // and never return here preventing us from clearing _last_native_pc down below.
1246 // Also can't use call_VM_leaf either as it will check to see if esi & edi are
1247 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1248 // by hand.
1249 //
1250 Label L;
1251 __ li(AT, SafepointSynchronize::address_of_state());
1252 __ lw(AT, AT, 0);
1253 __ bne(AT, R0, L);
1254 __ delayed()->nop();
1255 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1256 __ beq(AT, R0, Continue);
1257 __ delayed()->nop();
1258 __ bind(L);
1259 __ move(A0, thread);
1260 __ call(CAST_FROM_FN_PTR(address,
1261 JavaThread::check_special_condition_for_native_trans),
1262 relocInfo::runtime_call_type);
1263 __ delayed()->nop();
1265 #ifndef OPT_THREAD
1266 __ get_thread(thread);
1267 #endif
1268 //add for compressedoops
1269 __ reinit_heapbase();
1270 __ bind(Continue);
1271 }
1273 // change thread state
1274 __ move(t, _thread_in_Java);
1275 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1276 __ reset_last_Java_frame(thread, true, true);
1278 // reset handle block
1279 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1280 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1282 // If result was an oop then unbox and save it in the frame
1283 { Label L;
1284 Label no_oop, store_result;
1285 //FIXME, addi only support 16-bit imeditate
1286 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1287 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1288 __ bne(AT, T0, no_oop);
1289 __ delayed()->nop();
1290 __ move(V0, S1);
1291 __ beq(V0, R0, store_result);
1292 __ delayed()->nop();
1293 // unbox
1294 __ ld(V0, V0, 0);
1295 __ bind(store_result);
1296 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1297 // keep stack depth as expected by pushing oop which will eventually be discarded
1298 __ bind(no_oop);
1299 }
1300 {
1301 Label no_reguard;
1302 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1303 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1304 __ bne(t, AT, no_reguard);
1305 __ delayed()->nop();
1306 __ pushad();
1307 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::runtime_call_type);
1308 __ delayed()->nop();
1309 __ popad();
1310 //add for compressedoops
1311 __ reinit_heapbase();
1312 __ bind(no_reguard);
1313 }
1314 // restore esi to have legal interpreter frame,
1315 // i.e., bci == 0 <=> esi == code_base()
1316 // Can't call_VM until bcp is within reasonable.
1317 __ get_method(method); // method is junk from thread_in_native to now.
1318 __ verify_oop(method);
1319 __ ld(BCP, method, in_bytes(Method::const_offset()));
1320 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1321 // handle exceptions (exception handling will handle unlocking!)
1322 {
1323 Label L;
1324 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1325 __ beq(t, R0, L);
1326 __ delayed()->nop();
1327 // Note: At some point we may want to unify this with the code used in
1328 // call_VM_base();
1329 // i.e., we should use the StubRoutines::forward_exception code. For now this
1330 // doesn't work here because the esp is not correctly set at this point.
1331 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address,
1332 InterpreterRuntime::throw_pending_exception));
1333 __ should_not_reach_here();
1334 __ bind(L);
1335 }
1337 // do unlocking if necessary
1338 {
1339 Label L;
1340 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1341 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1342 __ beq(t, R0, L);
1343 // the code below should be shared with interpreter macro assembler implementation
1344 {
1345 Label unlock;
1346 // BasicObjectLock will be first in list,
1347 // since this is a synchronized method. However, need
1348 // to check that the object has not been unlocked by
1349 // an explicit monitorexit bytecode.
1350 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1351 // address of first monitor
1353 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1354 __ bne(t, R0, unlock);
1355 __ delayed()->nop();
1357 // Entry already unlocked, need to throw exception
1358 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1359 InterpreterRuntime::throw_illegal_monitor_state_exception));
1360 __ should_not_reach_here();
1362 __ bind(unlock);
1363 __ unlock_object(c_rarg0);
1364 }
1365 __ bind(L);
1366 }
1368 // jvmti/jvmpi support
1369 // Note: This must happen _after_ handling/throwing any exceptions since
1370 // the exception handler code notifies the runtime of method exits
1371 // too. If this happens before, method entry/exit notifications are
1372 // not properly paired (was bug - gri 11/22/99).
1373 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI );
1375 // restore potential result in V0:V1,
1376 // call result handler to restore potential result in ST0 & handle result
1377 //__ lw(V0, SP, 3 * wordSize);
1378 //__ lw(V1, SP, 2 * wordSize);
1379 //__ lwc1(F0, SP, 1 * wordSize);
1380 //__ lwc1(F1, SP, 0 * wordSize);
1381 //__ addi(SP, SP, 4 * wordSize);
1382 __ move(V0, S1);
1383 __ move(V1, S3);
1384 __ dmtc1(S4, F0);
1385 __ dmtc1(S2, F1);
1386 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1387 __ jalr(t);
1388 __ delayed()->nop();
1391 // remove activation
1392 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1393 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1394 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1395 __ jr(RA);
1396 __ delayed()->nop();
1398 #ifndef CORE
1399 if (inc_counter) {
1400 // Handle overflow of counter and compile method
1401 __ bind(invocation_counter_overflow);
1402 generate_counter_overflow(&continue_after_compile);
1403 // entry_point is the beginning of this
1404 // function and checks again for compiled code
1405 }
1406 #endif
1407 return entry_point;
1408 }
1410 //
1411 // Generic interpreted method entry to (asm) interpreter
1412 //
1413 // Layout of frame just at the entry
1414 //
1415 // [ argument word n-1 ] <--- sp
1416 // ...
1417 // [ argument word 0 ]
1418 // assume Method* in Rmethod before call this method.
1419 // prerequisites to the generated stub : the callee Method* in Rmethod
1420 // note you must save the caller bcp before call the generated stub
1421 //
1422 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1423 // determine code generation flags
1424 bool inc_counter = UseCompiler || CountCompiledCalls;
1426 // Rmethod: Method*
1427 // Rsender: sender 's sp
1428 address entry_point = __ pc();
1430 const Address invocation_counter(Rmethod,
1431 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1433 // get parameter size (always needed)
1434 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1435 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1437 // Rmethod: Method*
1438 // V0: size of parameters
1439 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1440 // get size of locals in words to T2
1441 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1442 // T2 = no. of additional locals, locals include parameters
1443 __ dsub(T2, T2, V0);
1445 // see if we've got enough room on the stack for locals plus overhead.
1446 // Layout of frame at this point
1447 //
1448 // [ argument word n-1 ] <--- sp
1449 // ...
1450 // [ argument word 0 ]
1451 generate_stack_overflow_check();
1452 // after this function, the layout of frame does not change
1454 // compute beginning of parameters (LVP)
1455 __ dsll(LVP, V0, LogBytesPerWord);
1456 __ daddiu(LVP, LVP, (-1) * wordSize);
1457 __ dadd(LVP, LVP, SP);
1459 // T2 - # of additional locals
1460 // allocate space for locals
1461 // explicitly initialize locals
1462 {
1463 Label exit, loop;
1464 __ beq(T2, R0, exit);
1465 __ delayed()->nop();
1467 __ bind(loop);
1468 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1469 __ daddiu(T2, T2, -1); // until everything initialized
1470 __ bne(T2, R0, loop);
1471 __ delayed();
1473 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1475 __ bind(exit);
1476 }
1478 //
1479 // [ local var m-1 ] <--- sp
1480 // ...
1481 // [ local var 0 ]
1482 // [ argument word n-1 ] <--- T0?
1483 // ...
1484 // [ argument word 0 ] <--- LVP
1486 // initialize fixed part of activation frame
1488 generate_fixed_frame(false);
1491 // after this function, the layout of frame is as following
1492 //
1493 // [ monitor block top ] <--- sp ( the top monitor entry )
1494 // [ byte code pointer ] (if native, bcp = 0)
1495 // [ constant pool cache ]
1496 // [ Method* ]
1497 // [ locals offset ]
1498 // [ sender's sp ]
1499 // [ sender's fp ] <--- fp
1500 // [ return address ]
1501 // [ local var m-1 ]
1502 // ...
1503 // [ local var 0 ]
1504 // [ argumnet word n-1 ] <--- ( sender's sp )
1505 // ...
1506 // [ argument word 0 ] <--- LVP
1509 // make sure method is not native & not abstract
1510 #ifdef ASSERT
1511 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1512 {
1513 Label L;
1514 __ andi(T2, AT, JVM_ACC_NATIVE);
1515 __ beq(T2, R0, L);
1516 __ delayed()->nop();
1517 __ stop("tried to execute native method as non-native");
1518 __ bind(L);
1519 }
1520 {
1521 Label L;
1522 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1523 __ beq(T2, R0, L);
1524 __ delayed()->nop();
1525 __ stop("tried to execute abstract method in interpreter");
1526 __ bind(L);
1527 }
1528 #endif
1530 // Since at this point in the method invocation the exception handler
1531 // would try to exit the monitor of synchronized methods which hasn't
1532 // been entered yet, we set the thread local variable
1533 // _do_not_unlock_if_synchronized to true. The remove_activation will
1534 // check this flag.
1536 #ifndef OPT_THREAD
1537 Register thread = T8;
1538 __ get_thread(thread);
1539 #else
1540 Register thread = TREG;
1541 #endif
1542 __ move(AT, (int)true);
1543 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1545 #ifndef CORE
1547 // 2014/11/24 Fu
1548 // mdp : T8
1549 // tmp1: T9
1550 // tmp2: T2
1551 __ profile_parameters_type(T8, T9, T2);
1553 // increment invocation count & check for overflow
1554 Label invocation_counter_overflow;
1555 Label profile_method;
1556 Label profile_method_continue;
1557 if (inc_counter) {
1558 generate_counter_incr(&invocation_counter_overflow,
1559 &profile_method,
1560 &profile_method_continue);
1561 if (ProfileInterpreter) {
1562 __ bind(profile_method_continue);
1563 }
1564 }
1566 Label continue_after_compile;
1567 __ bind(continue_after_compile);
1569 #endif // CORE
1571 bang_stack_shadow_pages(false);
1573 // reset the _do_not_unlock_if_synchronized flag
1574 #ifndef OPT_THREAD
1575 __ get_thread(thread);
1576 #endif
1577 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1579 // check for synchronized methods
1580 // Must happen AFTER invocation_counter check and stack overflow check,
1581 // so method is not locked if overflows.
1582 //
1583 if (synchronized) {
1584 // Allocate monitor and lock method
1585 lock_method();
1586 } else {
1587 // no synchronization necessary
1588 #ifdef ASSERT
1589 { Label L;
1590 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1591 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1592 __ beq(T2, R0, L);
1593 __ delayed()->nop();
1594 __ stop("method needs synchronization");
1595 __ bind(L);
1596 }
1597 #endif
1598 }
1600 // layout of frame after lock_method
1601 // [ monitor entry ] <--- sp
1602 // ...
1603 // [ monitor entry ]
1604 // [ monitor block top ] ( the top monitor entry )
1605 // [ byte code pointer ] (if native, bcp = 0)
1606 // [ constant pool cache ]
1607 // [ Method* ]
1608 // [ locals offset ]
1609 // [ sender's sp ]
1610 // [ sender's fp ]
1611 // [ return address ] <--- fp
1612 // [ local var m-1 ]
1613 // ...
1614 // [ local var 0 ]
1615 // [ argumnet word n-1 ] <--- ( sender's sp )
1616 // ...
1617 // [ argument word 0 ] <--- LVP
1620 // start execution
1621 #ifdef ASSERT
1622 {
1623 Label L;
1624 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1625 __ beq(AT, SP, L);
1626 __ delayed()->nop();
1627 __ stop("broken stack frame setup in interpreter in native");
1628 __ bind(L);
1629 }
1630 #endif
1632 // jvmti/jvmpi support
1633 __ notify_method_entry();
1635 __ dispatch_next(vtos);
1637 // invocation counter overflow
1638 if (inc_counter) {
1639 if (ProfileInterpreter) {
1640 // We have decided to profile this method in the interpreter
1641 __ bind(profile_method);
1642 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1643 InterpreterRuntime::profile_method));
1644 __ set_method_data_pointer_for_bcp();
1645 __ get_method(Rmethod);
1646 __ b(profile_method_continue);
1647 __ delayed()->nop();
1648 }
1649 // Handle overflow of counter and compile method
1650 __ bind(invocation_counter_overflow);
1651 generate_counter_overflow(&continue_after_compile);
1652 }
1654 return entry_point;
1655 }
1657 // Entry points
1658 //
1659 // Here we generate the various kind of entries into the interpreter.
1660 // The two main entry type are generic bytecode methods and native
1661 // call method. These both come in synchronized and non-synchronized
1662 // versions but the frame layout they create is very similar. The
1663 // other method entry types are really just special purpose entries
1664 // that are really entry and interpretation all in one. These are for
1665 // trivial methods like accessor, empty, or special math methods.
1666 //
1667 // When control flow reaches any of the entry types for the interpreter
1668 // the following holds ->
1669 //
1670 // Arguments:
1671 //
1672 // Rmethod: Method*
1673 // V0: receiver
1674 //
1675 //
1676 // Stack layout immediately at entry
1677 //
1678 // [ parameter n-1 ] <--- sp
1679 // ...
1680 // [ parameter 0 ]
1681 // [ expression stack ] (caller's java expression stack)
1683 // Assuming that we don't go to one of the trivial specialized entries
1684 // the stack will look like below when we are ready to execute the
1685 // first bytecode (or call the native routine). The register usage
1686 // will be as the template based interpreter expects (see
1687 // interpreter_amd64.hpp).
1688 //
1689 // local variables follow incoming parameters immediately; i.e.
1690 // the return address is moved to the end of the locals).
1691 //
1692 // [ monitor entry ] <--- sp
1693 // ...
1694 // [ monitor entry ]
1695 // [ monitor block top ] ( the top monitor entry )
1696 // [ byte code pointer ] (if native, bcp = 0)
1697 // [ constant pool cache ]
1698 // [ Method* ]
1699 // [ locals offset ]
1700 // [ sender's sp ]
1701 // [ sender's fp ]
1702 // [ return address ] <--- fp
1703 // [ local var m-1 ]
1704 // ...
1705 // [ local var 0 ]
1706 // [ argumnet word n-1 ] <--- ( sender's sp )
1707 // ...
1708 // [ argument word 0 ] <--- S7
1710 address AbstractInterpreterGenerator::generate_method_entry(
1711 AbstractInterpreter::MethodKind kind) {
1712 // determine code generation flags
1713 bool synchronized = false;
1714 address entry_point = NULL;
1715 switch (kind) {
1716 case Interpreter::zerolocals :
1717 break;
1718 case Interpreter::zerolocals_synchronized:
1719 synchronized = true;
1720 break;
1721 case Interpreter::native :
1722 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1723 break;
1724 case Interpreter::native_synchronized :
1725 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1726 break;
1727 case Interpreter::empty :
1728 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1729 break;
1730 case Interpreter::accessor :
1731 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1732 break;
1733 case Interpreter::abstract :
1734 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1735 break;
1737 case Interpreter::java_lang_math_sin : // fall thru
1738 case Interpreter::java_lang_math_cos : // fall thru
1739 case Interpreter::java_lang_math_tan : // fall thru
1740 case Interpreter::java_lang_math_log : // fall thru
1741 case Interpreter::java_lang_math_log10 : // fall thru
1742 case Interpreter::java_lang_math_pow : // fall thru
1743 case Interpreter::java_lang_math_exp : break;
1744 case Interpreter::java_lang_math_abs : // fall thru
1745 case Interpreter::java_lang_math_sqrt :
1746 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1747 case Interpreter::java_lang_ref_reference_get:
1748 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1749 default:
1750 fatal(err_msg("unexpected method kind: %d", kind));
1751 break;
1752 }
1753 if (entry_point) return entry_point;
1755 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1756 }
1758 // These should never be compiled since the interpreter will prefer
1759 // the compiled version to the intrinsic version.
1760 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1761 switch (method_kind(m)) {
1762 case Interpreter::java_lang_math_sin : // fall thru
1763 case Interpreter::java_lang_math_cos : // fall thru
1764 case Interpreter::java_lang_math_tan : // fall thru
1765 case Interpreter::java_lang_math_abs : // fall thru
1766 case Interpreter::java_lang_math_log : // fall thru
1767 case Interpreter::java_lang_math_log10 : // fall thru
1768 case Interpreter::java_lang_math_sqrt : // fall thru
1769 case Interpreter::java_lang_math_pow : // fall thru
1770 case Interpreter::java_lang_math_exp :
1771 return false;
1772 default:
1773 return true;
1774 }
1775 }
1777 // How much stack a method activation needs in words.
1778 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1780 const int entry_size = frame::interpreter_frame_monitor_size();
1782 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
1783 // be sure to change this if you add/subtract anything to/from the overhead area
1784 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1786 const int stub_code = 6; // see generate_call_stub
1787 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1788 const int method_stack = (method->max_locals() + method->max_stack()) *
1789 Interpreter::stackElementWords;
1790 return overhead_size + method_stack + stub_code;
1791 }
1793 void AbstractInterpreter::layout_activation(Method* method,
1794 int tempcount,
1795 int popframe_extra_args,
1796 int moncount,
1797 int caller_actual_parameters,
1798 int callee_param_count,
1799 int callee_locals,
1800 frame* caller,
1801 frame* interpreter_frame,
1802 bool is_top_frame,
1803 bool is_bottom_frame) {
1804 // Note: This calculation must exactly parallel the frame setup
1805 // in AbstractInterpreterGenerator::generate_method_entry.
1806 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1807 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1808 // right size, as determined by a previous call to this method.
1809 // It is also guaranteed to be walkable even though it is in a skeletal state
1811 // fixed size of an interpreter frame:
1813 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1814 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1816 #ifdef ASSERT
1817 if (!EnableInvokeDynamic) {
1818 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1819 // Probably, since deoptimization doesn't work yet.
1820 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1821 }
1822 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1823 #endif
1825 interpreter_frame->interpreter_frame_set_method(method);
1826 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1827 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1828 // and sender_sp is fp+8
1829 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1831 #ifdef ASSERT
1832 if (caller->is_interpreted_frame()) {
1833 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1834 }
1835 #endif
1837 interpreter_frame->interpreter_frame_set_locals(locals);
1838 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1839 BasicObjectLock* monbot = montop - moncount;
1840 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1842 //set last sp;
1843 intptr_t* esp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1844 popframe_extra_args;
1845 interpreter_frame->interpreter_frame_set_last_sp(esp);
1846 // All frames but the initial interpreter frame we fill in have a
1847 // value for sender_sp that allows walking the stack but isn't
1848 // truly correct. Correct the value here.
1849 //
1850 if (extra_locals != 0 &&
1851 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1852 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1853 }
1854 *interpreter_frame->interpreter_frame_cache_addr() = method->constants()->cache();
1855 }
1857 //-----------------------------------------------------------------------------
1858 // Exceptions
1860 void TemplateInterpreterGenerator::generate_throw_exception() {
1861 // Entry point in previous activation (i.e., if the caller was
1862 // interpreted)
1863 Interpreter::_rethrow_exception_entry = __ pc();
1864 // Restore sp to interpreter_frame_last_sp even though we are going
1865 // to empty the expression stack for the exception processing.
1866 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
1868 // V0: exception
1869 // V1: return address/pc that threw exception
1870 __ restore_bcp(); // esi points to call/send
1871 __ restore_locals();
1873 //add for compressedoops
1874 __ reinit_heapbase();
1875 // Entry point for exceptions thrown within interpreter code
1876 Interpreter::_throw_exception_entry = __ pc();
1877 // expression stack is undefined here
1878 // V0: exception
1879 // BCP: exception bcp
1880 __ verify_oop(V0);
1882 // expression stack must be empty before entering the VM in case of an exception
1883 __ empty_expression_stack();
1884 // find exception handler address and preserve exception oop
1885 __ move(A1, V0);
1886 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
1887 // V0: exception handler entry point
1888 // V1: preserved exception oop
1889 // S0: bcp for exception handler
1890 __ daddi(SP, SP, (-1) * wordSize);
1891 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
1892 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
1893 __ delayed()->nop();
1895 // If the exception is not handled in the current frame the frame is removed and
1896 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1897 //
1898 // Note: At this point the bci is still the bxi for the instruction which caused
1899 // the exception and the expression stack is empty. Thus, for any VM calls
1900 // at this point, GC will find a legal oop map (with empty expression stack).
1902 // In current activation
1903 // V0: exception
1904 // BCP: exception bcp
1906 //
1907 // JVMTI PopFrame support
1908 //
1910 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1911 __ empty_expression_stack();
1912 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1913 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1914 // popframe handling cycles.
1915 #ifndef OPT_THREAD
1916 Register thread = T2;
1917 __ get_thread(T2);
1918 #else
1919 Register thread = TREG;
1920 #endif
1921 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1922 __ ori(T3, T3, JavaThread::popframe_processing_bit);
1923 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1925 #ifndef CORE
1926 {
1927 // Check to see whether we are returning to a deoptimized frame.
1928 // (The PopFrame call ensures that the caller of the popped frame is
1929 // either interpreted or compiled and deoptimizes it if compiled.)
1930 // In this case, we can't call dispatch_next() after the frame is
1931 // popped, but instead must save the incoming arguments and restore
1932 // them after deoptimization has occurred.
1933 //
1934 // Note that we don't compare the return PC against the
1935 // deoptimization blob's unpack entry because of the presence of
1936 // adapter frames in C2.
1937 Label caller_not_deoptimized;
1938 __ ld(A0, FP, frame::return_addr_offset * wordSize);
1939 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
1940 __ bne(V0, R0, caller_not_deoptimized);
1941 __ delayed()->nop();
1943 // Compute size of arguments for saving when returning to deoptimized caller
1944 __ get_method(A1);
1945 __ verify_oop(A1);
1946 __ ld(A1,A1,in_bytes(Method::const_offset()));
1947 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
1948 __ shl(A1, Interpreter::logStackElementSize);
1949 __ restore_locals();
1950 __ dsub(A2, LVP, T0);
1951 __ daddiu(A2, A2, wordSize);
1952 // Save these arguments
1953 #ifndef OPT_THREAD
1954 __ get_thread(A0);
1955 #else
1956 __ move(A0, TREG);
1957 #endif
1958 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
1960 __ remove_activation(vtos, T9, false, false, false);
1962 // Inform deoptimization that it is responsible for restoring these arguments
1963 #ifndef OPT_THREAD
1964 __ get_thread(thread);
1965 #endif
1966 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
1967 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
1968 // Continue in deoptimization handler
1969 __ jr(T9);
1970 __ delayed()->nop();
1972 __ bind(caller_not_deoptimized);
1973 }
1974 #endif /* !CORE */
1976 __ remove_activation(vtos, T3,
1977 /* throw_monitor_exception */ false,
1978 /* install_monitor_exception */ false,
1979 /* notify_jvmdi */ false);
1981 // Clear the popframe condition flag
1982 // Finish with popframe handling
1983 // A previous I2C followed by a deoptimization might have moved the
1984 // outgoing arguments further up the stack. PopFrame expects the
1985 // mutations to those outgoing arguments to be preserved and other
1986 // constraints basically require this frame to look exactly as
1987 // though it had previously invoked an interpreted activation with
1988 // no space between the top of the expression stack (current
1989 // last_sp) and the top of stack. Rather than force deopt to
1990 // maintain this kind of invariant all the time we call a small
1991 // fixup routine to move the mutated arguments onto the top of our
1992 // expression stack if necessary.
1993 __ move(T8, SP);
1994 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
1995 #ifndef OPT_THREAD
1996 __ get_thread(thread);
1997 #endif
1998 // PC must point into interpreter here
1999 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2000 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2001 __ reset_last_Java_frame(thread, true, true);
2002 // Restore the last_sp and null it out
2003 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2004 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2008 __ move(AT, JavaThread::popframe_inactive);
2009 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2011 // Finish with popframe handling
2012 __ restore_bcp();
2013 __ restore_locals();
2014 #ifndef CORE
2015 // The method data pointer was incremented already during
2016 // call profiling. We have to restore the mdp for the current bcp.
2017 if (ProfileInterpreter) {
2018 __ set_method_data_pointer_for_bcp();
2019 }
2020 #endif // !CORE
2021 // Clear the popframe condition flag
2022 #ifndef OPT_THREAD
2023 __ get_thread(thread);
2024 #endif
2025 __ move(AT, JavaThread::popframe_inactive);
2026 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2027 __ dispatch_next(vtos);
2028 // end of PopFrame support
2030 Interpreter::_remove_activation_entry = __ pc();
2032 // preserve exception over this code sequence
2033 __ ld(T0, SP, 0);
2034 __ daddi(SP, SP, wordSize);
2035 #ifndef OPT_THREAD
2036 __ get_thread(thread);
2037 #endif
2038 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2039 // remove the activation (without doing throws on illegalMonitorExceptions)
2040 __ remove_activation(vtos, T3, false, true, false);
2041 // restore exception
2042 __ get_vm_result(T0, thread);
2043 __ verify_oop(T0);
2045 // Inbetween activations - previous activation type unknown yet
2046 // compute continuation point - the continuation point expects
2047 // the following registers set up:
2048 //
2049 // T0: exception eax
2050 // T1: return address/pc that threw exception edx
2051 // SP: expression stack of caller esp
2052 // FP: ebp of caller ebp
2053 __ daddi(SP, SP, (-2) * wordSize);
2054 __ sd(T0, SP, wordSize); // save exception
2055 __ sd(T3, SP, 0); // save return address
2056 __ move(A1, T3);
2057 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2058 __ move(T9, V0); // save exception handler
2059 __ ld(V0, SP, wordSize); // restore exception
2060 __ ld(V1, SP, 0); // restore return address
2061 __ daddi(SP, SP, 2 * wordSize);
2063 // Note that an "issuing PC" is actually the next PC after the call
2064 __ jr(T9); // jump to exception handler of caller
2065 __ delayed()->nop();
2066 }
2069 //
2070 // JVMTI ForceEarlyReturn support
2071 //
2072 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2073 address entry = __ pc();
2074 __ restore_bcp();
2075 __ restore_locals();
2076 __ empty_expression_stack();
2077 __ empty_FPU_stack();
2078 __ load_earlyret_value(state);
2080 #ifndef OPT_THREAD
2081 __ get_thread(TREG);
2082 #endif
2083 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2084 //const Address cond_addr(ecx, JvmtiThreadState::earlyret_state_offset());
2085 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2086 // Clear the earlyret state
2087 __ move(AT,JvmtiThreadState::earlyret_inactive);
2088 __ sw(AT,cond_addr);
2089 __ sync();
2092 __ remove_activation(state, T0,
2093 false, /* throw_monitor_exception */
2094 false, /* install_monitor_exception */
2095 true); /* notify_jvmdi */
2096 __ sync();
2097 __ jr(T0);
2098 __ delayed()->nop();
2099 return entry;
2100 } // end of ForceEarlyReturn support
2103 //-----------------------------------------------------------------------------
2104 // Helper for vtos entry point generation
2106 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2107 address& bep,
2108 address& cep,
2109 address& sep,
2110 address& aep,
2111 address& iep,
2112 address& lep,
2113 address& fep,
2114 address& dep,
2115 address& vep) {
2116 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2117 Label L;
2118 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2119 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2120 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2121 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2122 bep = cep = sep =
2123 iep = __ pc(); __ push(itos);
2124 vep = __ pc();
2125 __ bind(L);
2126 generate_and_dispatch(t);
2127 }
2130 //-----------------------------------------------------------------------------
2131 // Generation of individual instructions
2133 // helpers for generate_and_dispatch
2136 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2137 : TemplateInterpreterGenerator(code) {
2138 generate_all(); // down here so it can be "virtual"
2139 }
2141 //-----------------------------------------------------------------------------
2143 // Non-product code
2144 #ifndef PRODUCT
2145 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2146 address entry = __ pc();
2148 // prepare expression stack
2149 __ push(state); // save tosca
2151 // tos & tos2, added by yjl 7/15/2005
2152 // trace_bytecode need actually 4 args, the last two is tos&tos2
2153 // this work fine for x86. but mips o32 call convention will store A2-A3
2154 // to the stack position it think is the tos&tos2
2155 // when the expression stack have no more than 2 data, error occur.
2156 __ ld(A2, SP, 0);
2157 __ ld(A3, SP, 1 * wordSize);
2159 // pass arguments & call tracer
2160 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2161 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2163 // restore expression stack
2164 __ pop(state); // restore tosca
2166 // return
2167 __ jr(RA);
2168 __ delayed()->nop();
2170 return entry;
2171 }
2173 void TemplateInterpreterGenerator::count_bytecode() {
2174 __ li(T8, (long)&BytecodeCounter::_counter_value);
2175 __ lw(AT, T8, 0);
2176 __ daddi(AT, AT, 1);
2177 __ sw(AT, T8, 0);
2178 }
2180 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2181 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2182 __ lw(AT, T8, 0);
2183 __ daddi(AT, AT, 1);
2184 __ sw(AT, T8, 0);
2185 }
2187 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2188 __ li(T8, (long)&BytecodePairHistogram::_index);
2189 __ lw(T9, T8, 0);
2190 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2191 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2192 __ orr(T9, T9, T8);
2193 __ li(T8, (long)&BytecodePairHistogram::_index);
2194 __ sw(T9, T8, 0);
2195 __ dsll(T9, T9, 2);
2196 __ li(T8, (long)BytecodePairHistogram::_counters);
2197 __ dadd(T8, T8, T9);
2198 __ lw(AT, T8, 0);
2199 __ daddi(AT, AT, 1);
2200 __ sw(AT, T8, 0);
2201 }
2204 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2205 // Call a little run-time stub to avoid blow-up for each bytecode.
2206 // The run-time runtime saves the right registers, depending on
2207 // the tosca in-state for the given template.
2209 address entry = Interpreter::trace_code(t->tos_in());
2210 assert(entry != NULL, "entry must have been generated");
2211 __ call(entry, relocInfo::none);
2212 __ delayed()->nop();
2213 //add for compressedoops
2214 __ reinit_heapbase();
2215 }
2218 void TemplateInterpreterGenerator::stop_interpreter_at() {
2219 Label L;
2220 __ li(T8, long(&BytecodeCounter::_counter_value));
2221 __ lw(T8, T8, 0);
2222 __ move(AT, StopInterpreterAt);
2223 __ bne(T8, AT, L);
2224 __ delayed()->nop();
2225 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2226 __ delayed()->nop();
2227 __ bind(L);
2228 }
2229 #endif // !PRODUCT
2230 #endif // ! CC_INTERP