Mon, 23 Oct 2017 17:07:19 +0800
[G1] Initial porting of MacroAssembler::g1_write_barrier_{pre/post}
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 warning("UseG1GC unimplemented yet!");
801 __ should_not_reach_here();
802 Label slow_path;
803 // rbx: method
805 // Check if local 0 != NULL
806 // If the receiver is null then it is OK to jump to the slow path.
808 // rax: local 0
809 // rbx: method (but can be used as scratch now)
810 // rdx: scratch
811 // rdi: scratch
813 // Generate the G1 pre-barrier code to log the value of
814 // the referent field in an SATB buffer.
816 // Load the value of the referent field.
818 return entry;
819 }
820 #endif // INCLUDE_ALL_GCS
822 // If G1 is not enabled then attempt to go through the accessor entry point
823 // Reference.get is an accessor
824 return generate_accessor_entry();
825 }
826 // Interpreter stub for calling a native method. (asm interpreter)
827 // This sets up a somewhat different looking stack for calling the
828 // native method than the typical interpreter frame setup.
829 address InterpreterGenerator::generate_native_entry(bool synchronized) {
830 // determine code generation flags
831 bool inc_counter = UseCompiler || CountCompiledCalls;
832 // Rsender: sender's sp
833 // Rmethod: Method*
834 address entry_point = __ pc();
836 #ifndef CORE
837 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() + // Fu: 20130814
838 InvocationCounter::counter_offset()));
839 #endif
841 // get parameter size (always needed)
842 // the size in the java stack
843 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
844 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
846 // native calls don't need the stack size check since they have no expression stack
847 // and the arguments are already on the stack and we only add a handful of words
848 // to the stack
850 // Rmethod: Method*
851 // V0: size of parameters
852 // Layout of frame at this point
853 //
854 // [ argument word n-1 ] <--- sp
855 // ...
856 // [ argument word 0 ]
858 // for natives the size of locals is zero
860 // compute beginning of parameters (S7)
861 __ dsll(LVP, V0, Address::times_8);
862 __ daddiu(LVP, LVP, (-1) * wordSize);
863 __ dadd(LVP, LVP, SP);
866 // add 2 zero-initialized slots for native calls
867 __ daddi(SP, SP, (-2) * wordSize);
868 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
869 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
871 // Layout of frame at this point
872 // [ method holder mirror ] <--- sp
873 // [ result type info ]
874 // [ argument word n-1 ] <--- T0
875 // ...
876 // [ argument word 0 ] <--- LVP
879 #ifndef CORE
880 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
881 #endif
883 // initialize fixed part of activation frame
884 generate_fixed_frame(true);
885 // after this function, the layout of frame is as following
886 //
887 // [ monitor block top ] <--- sp ( the top monitor entry )
888 // [ byte code pointer (0) ] (if native, bcp = 0)
889 // [ constant pool cache ]
890 // [ Method* ]
891 // [ locals offset ]
892 // [ sender's sp ]
893 // [ sender's fp ]
894 // [ return address ] <--- fp
895 // [ method holder mirror ]
896 // [ result type info ]
897 // [ argumnet word n-1 ] <--- sender's sp
898 // ...
899 // [ argument word 0 ] <--- S7
902 // make sure method is native & not abstract
903 #ifdef ASSERT
904 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
905 {
906 Label L;
907 __ andi(AT, T0, JVM_ACC_NATIVE);
908 __ bne(AT, R0, L);
909 __ delayed()->nop();
910 __ stop("tried to execute native method as non-native");
911 __ bind(L);
912 }
913 {
914 Label L;
915 __ andi(AT, T0, JVM_ACC_ABSTRACT);
916 __ beq(AT, R0, L);
917 __ delayed()->nop();
918 __ stop("tried to execute abstract method in interpreter");
919 __ bind(L);
920 }
921 #endif
923 // Since at this point in the method invocation the exception handler
924 // would try to exit the monitor of synchronized methods which hasn't
925 // been entered yet, we set the thread local variable
926 // _do_not_unlock_if_synchronized to true. The remove_activation will
927 // check this flag.
928 Register thread = TREG;
929 #ifndef OPT_THREAD
930 __ get_thread(thread);
931 #endif
932 __ move(AT, (int)true);
933 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
935 #ifndef CORE
936 // increment invocation count & check for overflow
937 Label invocation_counter_overflow;
938 if (inc_counter) {
939 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
940 }
942 Label continue_after_compile;
943 __ bind(continue_after_compile);
944 #endif // CORE
946 bang_stack_shadow_pages(true);
948 // reset the _do_not_unlock_if_synchronized flag
949 #ifndef OPT_THREAD
950 __ get_thread(thread);
951 #endif
952 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
954 // check for synchronized methods
955 // Must happen AFTER invocation_counter check and stack overflow check,
956 // so method is not locked if overflows.
957 if (synchronized) {
958 lock_method();
959 } else {
960 // no synchronization necessary
961 #ifdef ASSERT
962 {
963 Label L;
964 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
965 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
966 __ beq(AT, R0, L);
967 __ delayed()->nop();
968 __ stop("method needs synchronization");
969 __ bind(L);
970 }
971 #endif
972 }
974 // after method_lock, the layout of frame is as following
975 //
976 // [ monitor entry ] <--- sp
977 // ...
978 // [ monitor entry ]
979 // [ monitor block top ] ( the top monitor entry )
980 // [ byte code pointer (0) ] (if native, bcp = 0)
981 // [ constant pool cache ]
982 // [ Method* ]
983 // [ locals offset ]
984 // [ sender's sp ]
985 // [ sender's fp ]
986 // [ return address ] <--- fp
987 // [ method holder mirror ]
988 // [ result type info ]
989 // [ argumnet word n-1 ] <--- ( sender's sp )
990 // ...
991 // [ argument word 0 ] <--- S7
993 // start execution
994 #ifdef ASSERT
995 {
996 Label L;
997 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
998 __ beq(AT, SP, L);
999 __ delayed()->nop();
1000 __ stop("broken stack frame setup in interpreter in asm");
1001 __ bind(L);
1002 }
1003 #endif
1005 // jvmti/jvmpi support
1006 __ notify_method_entry();
1008 // work registers
1009 const Register method = Rmethod;
1010 //const Register thread = T2;
1011 const Register t = RT4;
1013 __ get_method(method);
1014 __ verify_oop(method);
1015 {
1016 Label L, Lstatic;
1017 __ ld(t,method,in_bytes(Method::const_offset()));
1018 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
1019 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1020 //FIXME, aoqi: A1?
1021 // A0 and A1(if needed)
1022 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1023 __ andi(AT, AT, JVM_ACC_STATIC);
1024 __ beq(AT, R0, Lstatic);
1025 __ delayed()->nop();
1026 __ daddiu(t, t, 1);
1027 __ bind(Lstatic);
1028 __ daddiu(t, t, -7);
1029 __ blez(t, L);
1030 __ delayed()->nop();
1031 __ dsll(t, t, Address::times_8);
1032 __ dsub(SP, SP, t);
1033 __ bind(L);
1034 }
1035 __ move(AT, -(StackAlignmentInBytes));
1036 __ andr(SP, SP, AT);
1037 __ move(AT, SP);
1038 // [ ] <--- sp
1039 // ... (size of parameters - 8 )
1040 // [ monitor entry ]
1041 // ...
1042 // [ monitor entry ]
1043 // [ monitor block top ] ( the top monitor entry )
1044 // [ byte code pointer (0) ] (if native, bcp = 0)
1045 // [ constant pool cache ]
1046 // [ Method* ]
1047 // [ locals offset ]
1048 // [ sender's sp ]
1049 // [ sender's fp ]
1050 // [ return address ] <--- fp
1051 // [ method holder mirror ]
1052 // [ result type info ]
1053 // [ argumnet word n-1 ] <--- ( sender's sp )
1054 // ...
1055 // [ argument word 0 ] <--- LVP
1057 // get signature handler
1058 {
1059 Label L;
1060 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1061 __ bne(T9, R0, L);
1062 __ delayed()->nop();
1063 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1064 InterpreterRuntime::prepare_native_call), method);
1065 __ get_method(method);
1066 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1067 __ bind(L);
1068 }
1070 // call signature handler
1071 // FIXME: when change codes in InterpreterRuntime, note this point
1072 // from: begin of parameters
1073 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1074 // to: current sp
1075 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1076 // temp: T3
1077 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1079 __ jalr(T9);
1080 __ delayed()->nop();
1081 __ get_method(method); // slow path call blows EBX on DevStudio 5.0
1083 /*
1084 if native function is static, and its second parameter has type length of double word,
1085 and first parameter has type length of word, we have to reserve one word
1086 for the first parameter, according to mips o32 abi.
1087 if native function is not static, and its third parameter has type length of double word,
1088 and second parameter has type length of word, we have to reserve one word for the second
1089 parameter.
1090 */
1093 // result handler is in V0
1094 // set result handler
1095 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1097 #define FIRSTPARA_SHIFT_COUNT 5
1098 #define SECONDPARA_SHIFT_COUNT 9
1099 #define THIRDPARA_SHIFT_COUNT 13
1100 #define PARA_MASK 0xf
1102 // pass mirror handle if static call
1103 {
1104 Label L;
1105 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1106 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1107 __ andi(AT, t, JVM_ACC_STATIC);
1108 __ beq(AT, R0, L);
1109 __ delayed()->nop();
1111 // get mirror
1112 __ ld(t, method, in_bytes(Method:: const_offset()));
1113 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1114 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1115 __ ld(t, t, mirror_offset);
1116 // copy mirror into activation frame
1117 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1118 // pass handle to mirror
1119 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1120 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1121 __ move(A1, t);
1122 __ bind(L);
1123 }
1125 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1126 // [ ] |
1127 // ... size of parameters(or +1) |
1128 // [ monitor entry ] |
1129 // ... |
1130 // [ monitor entry ] |
1131 // [ monitor block top ] ( the top monitor entry ) |
1132 // [ byte code pointer (0) ] (if native, bcp = 0) |
1133 // [ constant pool cache ] |
1134 // [ Method* ] |
1135 // [ locals offset ] |
1136 // [ sender's sp ] |
1137 // [ sender's fp ] |
1138 // [ return address ] <--- fp |
1139 // [ method holder mirror ] <----------------------------|
1140 // [ result type info ]
1141 // [ argumnet word n-1 ] <--- ( sender's sp )
1142 // ...
1143 // [ argument word 0 ] <--- S7
1145 // get native function entry point
1146 { Label L;
1147 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1148 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1149 __ bne(V1, T9, L);
1150 __ delayed()->nop();
1151 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1152 __ get_method(method);
1153 __ verify_oop(method);
1154 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1155 __ bind(L);
1156 }
1157 /*
1158 __ pushad();
1159 __ move(A0, T9);
1160 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
1161 __ popad();
1162 */
1164 // pass JNIEnv
1165 // native function in T9
1166 #ifndef OPT_THREAD
1167 __ get_thread(thread);
1168 #endif
1169 __ daddi(t, thread, in_bytes(JavaThread::jni_environment_offset()));
1170 // stack,but I think it won't work when pass float,double etc @jerome,10/17,2006
1171 __ move(A0, t);
1172 // [ jni environment ] <--- sp
1173 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1174 // [ ] |
1175 // ... size of parameters |
1176 // [ monitor entry ] |
1177 // ... |
1178 // [ monitor entry ] |
1179 // [ monitor block top ] ( the top monitor entry ) |
1180 // [ byte code pointer (0) ] (if native, bcp = 0) |
1181 // [ constant pool cache ] |
1182 // [ Method* ] |
1183 // [ locals offset ] |
1184 // [ sender's sp ] |
1185 // [ sender's fp ] |
1186 // [ return address ] <--- fp |
1187 // [ method holder mirror ] <----------------------------|
1188 // [ result type info ]
1189 // [ argumnet word n-1 ] <--- ( sender's sp )
1190 // ...
1191 // [ argument word 0 ] <--- S7
1193 // set_last_Java_frame_before_call
1194 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1195 // Change state to native (we save the return address in the thread, since it might not
1196 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1197 // points into the right code segment. It does not have to be the correct return pc.
1198 __ li(t, __ pc());
1199 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1200 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1202 // change thread state
1203 #ifdef ASSERT
1204 {
1205 Label L;
1206 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1207 __ daddi(t, t, (-1) * _thread_in_Java);
1208 __ beq(t, R0, L);
1209 __ delayed()->nop();
1210 __ stop("Wrong thread state in native stub");
1211 __ bind(L);
1212 }
1213 #endif
1215 __ move(t, _thread_in_native);
1216 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1218 // call native method
1219 __ jalr(T9);
1220 __ delayed()->nop();
1221 // result potentially in V2:V1 or F0:F1
1224 // via _last_native_pc and not via _last_jave_sp
1225 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1226 // If the order changes or anything else is added to the stack the code in
1227 // interpreter_frame_result will have to be changed.
1228 //FIXME, should modify here
1229 // save return value to keep the value from being destroyed by other calls
1230 __ move(S1, V0);
1231 __ move(S3, V1);
1232 __ dmfc1(S4, F0);
1233 __ dmfc1(S2, F1);
1235 // change thread state
1236 __ get_thread(thread);
1237 __ move(t, _thread_in_native_trans);
1238 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1240 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1242 // check for safepoint operation in progress and/or pending suspend requests
1243 { Label Continue;
1245 // Don't use call_VM as it will see a possible pending exception and forward it
1246 // and never return here preventing us from clearing _last_native_pc down below.
1247 // Also can't use call_VM_leaf either as it will check to see if esi & edi are
1248 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1249 // by hand.
1250 //
1251 Label L;
1252 __ li(AT, SafepointSynchronize::address_of_state());
1253 __ lw(AT, AT, 0);
1254 __ bne(AT, R0, L);
1255 __ delayed()->nop();
1256 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1257 __ beq(AT, R0, Continue);
1258 __ delayed()->nop();
1259 __ bind(L);
1260 __ move(A0, thread);
1261 __ call(CAST_FROM_FN_PTR(address,
1262 JavaThread::check_special_condition_for_native_trans),
1263 relocInfo::runtime_call_type);
1264 __ delayed()->nop();
1266 #ifndef OPT_THREAD
1267 __ get_thread(thread);
1268 #endif
1269 //add for compressedoops
1270 __ reinit_heapbase();
1271 __ bind(Continue);
1272 }
1274 // change thread state
1275 __ move(t, _thread_in_Java);
1276 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1277 __ reset_last_Java_frame(thread, true, true);
1279 // reset handle block
1280 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1281 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1283 // If result was an oop then unbox and save it in the frame
1284 { Label L;
1285 Label no_oop, store_result;
1286 //FIXME, addi only support 16-bit imeditate
1287 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1288 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1289 __ bne(AT, T0, no_oop);
1290 __ delayed()->nop();
1291 __ move(V0, S1);
1292 __ beq(V0, R0, store_result);
1293 __ delayed()->nop();
1294 // unbox
1295 __ ld(V0, V0, 0);
1296 __ bind(store_result);
1297 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1298 // keep stack depth as expected by pushing oop which will eventually be discarded
1299 __ bind(no_oop);
1300 }
1301 {
1302 Label no_reguard;
1303 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1304 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1305 __ bne(t, AT, no_reguard);
1306 __ delayed()->nop();
1307 __ pushad();
1308 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::runtime_call_type);
1309 __ delayed()->nop();
1310 __ popad();
1311 //add for compressedoops
1312 __ reinit_heapbase();
1313 __ bind(no_reguard);
1314 }
1315 // restore esi to have legal interpreter frame,
1316 // i.e., bci == 0 <=> esi == code_base()
1317 // Can't call_VM until bcp is within reasonable.
1318 __ get_method(method); // method is junk from thread_in_native to now.
1319 __ verify_oop(method);
1320 __ ld(BCP, method, in_bytes(Method::const_offset()));
1321 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1322 // handle exceptions (exception handling will handle unlocking!)
1323 {
1324 Label L;
1325 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1326 __ beq(t, R0, L);
1327 __ delayed()->nop();
1328 // Note: At some point we may want to unify this with the code used in
1329 // call_VM_base();
1330 // i.e., we should use the StubRoutines::forward_exception code. For now this
1331 // doesn't work here because the esp is not correctly set at this point.
1332 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address,
1333 InterpreterRuntime::throw_pending_exception));
1334 __ should_not_reach_here();
1335 __ bind(L);
1336 }
1338 // do unlocking if necessary
1339 {
1340 Label L;
1341 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1342 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1343 __ beq(t, R0, L);
1344 // the code below should be shared with interpreter macro assembler implementation
1345 {
1346 Label unlock;
1347 // BasicObjectLock will be first in list,
1348 // since this is a synchronized method. However, need
1349 // to check that the object has not been unlocked by
1350 // an explicit monitorexit bytecode.
1351 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1352 // address of first monitor
1354 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1355 __ bne(t, R0, unlock);
1356 __ delayed()->nop();
1358 // Entry already unlocked, need to throw exception
1359 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1360 InterpreterRuntime::throw_illegal_monitor_state_exception));
1361 __ should_not_reach_here();
1363 __ bind(unlock);
1364 __ unlock_object(c_rarg0);
1365 }
1366 __ bind(L);
1367 }
1369 // jvmti/jvmpi support
1370 // Note: This must happen _after_ handling/throwing any exceptions since
1371 // the exception handler code notifies the runtime of method exits
1372 // too. If this happens before, method entry/exit notifications are
1373 // not properly paired (was bug - gri 11/22/99).
1374 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI );
1376 // restore potential result in V0:V1,
1377 // call result handler to restore potential result in ST0 & handle result
1378 //__ lw(V0, SP, 3 * wordSize);
1379 //__ lw(V1, SP, 2 * wordSize);
1380 //__ lwc1(F0, SP, 1 * wordSize);
1381 //__ lwc1(F1, SP, 0 * wordSize);
1382 //__ addi(SP, SP, 4 * wordSize);
1383 __ move(V0, S1);
1384 __ move(V1, S3);
1385 __ dmtc1(S4, F0);
1386 __ dmtc1(S2, F1);
1387 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1388 __ jalr(t);
1389 __ delayed()->nop();
1392 // remove activation
1393 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1394 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1395 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1396 __ jr(RA);
1397 __ delayed()->nop();
1399 #ifndef CORE
1400 if (inc_counter) {
1401 // Handle overflow of counter and compile method
1402 __ bind(invocation_counter_overflow);
1403 generate_counter_overflow(&continue_after_compile);
1404 // entry_point is the beginning of this
1405 // function and checks again for compiled code
1406 }
1407 #endif
1408 return entry_point;
1409 }
1411 //
1412 // Generic interpreted method entry to (asm) interpreter
1413 //
1414 // Layout of frame just at the entry
1415 //
1416 // [ argument word n-1 ] <--- sp
1417 // ...
1418 // [ argument word 0 ]
1419 // assume Method* in Rmethod before call this method.
1420 // prerequisites to the generated stub : the callee Method* in Rmethod
1421 // note you must save the caller bcp before call the generated stub
1422 //
1423 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1424 // determine code generation flags
1425 bool inc_counter = UseCompiler || CountCompiledCalls;
1427 // Rmethod: Method*
1428 // Rsender: sender 's sp
1429 address entry_point = __ pc();
1431 const Address invocation_counter(Rmethod,
1432 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1434 // get parameter size (always needed)
1435 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1436 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1438 // Rmethod: Method*
1439 // V0: size of parameters
1440 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1441 // get size of locals in words to T2
1442 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1443 // T2 = no. of additional locals, locals include parameters
1444 __ dsub(T2, T2, V0);
1446 // see if we've got enough room on the stack for locals plus overhead.
1447 // Layout of frame at this point
1448 //
1449 // [ argument word n-1 ] <--- sp
1450 // ...
1451 // [ argument word 0 ]
1452 generate_stack_overflow_check();
1453 // after this function, the layout of frame does not change
1455 // compute beginning of parameters (LVP)
1456 __ dsll(LVP, V0, LogBytesPerWord);
1457 __ daddiu(LVP, LVP, (-1) * wordSize);
1458 __ dadd(LVP, LVP, SP);
1460 // T2 - # of additional locals
1461 // allocate space for locals
1462 // explicitly initialize locals
1463 {
1464 Label exit, loop;
1465 __ beq(T2, R0, exit);
1466 __ delayed()->nop();
1468 __ bind(loop);
1469 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1470 __ daddiu(T2, T2, -1); // until everything initialized
1471 __ bne(T2, R0, loop);
1472 __ delayed();
1474 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1476 __ bind(exit);
1477 }
1479 //
1480 // [ local var m-1 ] <--- sp
1481 // ...
1482 // [ local var 0 ]
1483 // [ argument word n-1 ] <--- T0?
1484 // ...
1485 // [ argument word 0 ] <--- LVP
1487 // initialize fixed part of activation frame
1489 generate_fixed_frame(false);
1492 // after this function, the layout of frame is as following
1493 //
1494 // [ monitor block top ] <--- sp ( the top monitor entry )
1495 // [ byte code pointer ] (if native, bcp = 0)
1496 // [ constant pool cache ]
1497 // [ Method* ]
1498 // [ locals offset ]
1499 // [ sender's sp ]
1500 // [ sender's fp ] <--- fp
1501 // [ return address ]
1502 // [ local var m-1 ]
1503 // ...
1504 // [ local var 0 ]
1505 // [ argumnet word n-1 ] <--- ( sender's sp )
1506 // ...
1507 // [ argument word 0 ] <--- LVP
1510 // make sure method is not native & not abstract
1511 #ifdef ASSERT
1512 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1513 {
1514 Label L;
1515 __ andi(T2, AT, JVM_ACC_NATIVE);
1516 __ beq(T2, R0, L);
1517 __ delayed()->nop();
1518 __ stop("tried to execute native method as non-native");
1519 __ bind(L);
1520 }
1521 {
1522 Label L;
1523 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1524 __ beq(T2, R0, L);
1525 __ delayed()->nop();
1526 __ stop("tried to execute abstract method in interpreter");
1527 __ bind(L);
1528 }
1529 #endif
1531 // Since at this point in the method invocation the exception handler
1532 // would try to exit the monitor of synchronized methods which hasn't
1533 // been entered yet, we set the thread local variable
1534 // _do_not_unlock_if_synchronized to true. The remove_activation will
1535 // check this flag.
1537 #ifndef OPT_THREAD
1538 Register thread = T8;
1539 __ get_thread(thread);
1540 #else
1541 Register thread = TREG;
1542 #endif
1543 __ move(AT, (int)true);
1544 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1546 #ifndef CORE
1548 // 2014/11/24 Fu
1549 // mdp : T8
1550 // tmp1: T9
1551 // tmp2: T2
1552 __ profile_parameters_type(T8, T9, T2);
1554 // increment invocation count & check for overflow
1555 Label invocation_counter_overflow;
1556 Label profile_method;
1557 Label profile_method_continue;
1558 if (inc_counter) {
1559 generate_counter_incr(&invocation_counter_overflow,
1560 &profile_method,
1561 &profile_method_continue);
1562 if (ProfileInterpreter) {
1563 __ bind(profile_method_continue);
1564 }
1565 }
1567 Label continue_after_compile;
1568 __ bind(continue_after_compile);
1570 #endif // CORE
1572 bang_stack_shadow_pages(false);
1574 // reset the _do_not_unlock_if_synchronized flag
1575 #ifndef OPT_THREAD
1576 __ get_thread(thread);
1577 #endif
1578 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1580 // check for synchronized methods
1581 // Must happen AFTER invocation_counter check and stack overflow check,
1582 // so method is not locked if overflows.
1583 //
1584 if (synchronized) {
1585 // Allocate monitor and lock method
1586 lock_method();
1587 } else {
1588 // no synchronization necessary
1589 #ifdef ASSERT
1590 { Label L;
1591 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1592 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1593 __ beq(T2, R0, L);
1594 __ delayed()->nop();
1595 __ stop("method needs synchronization");
1596 __ bind(L);
1597 }
1598 #endif
1599 }
1601 // layout of frame after lock_method
1602 // [ monitor entry ] <--- sp
1603 // ...
1604 // [ monitor entry ]
1605 // [ monitor block top ] ( the top monitor entry )
1606 // [ byte code pointer ] (if native, bcp = 0)
1607 // [ constant pool cache ]
1608 // [ Method* ]
1609 // [ locals offset ]
1610 // [ sender's sp ]
1611 // [ sender's fp ]
1612 // [ return address ] <--- fp
1613 // [ local var m-1 ]
1614 // ...
1615 // [ local var 0 ]
1616 // [ argumnet word n-1 ] <--- ( sender's sp )
1617 // ...
1618 // [ argument word 0 ] <--- LVP
1621 // start execution
1622 #ifdef ASSERT
1623 {
1624 Label L;
1625 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1626 __ beq(AT, SP, L);
1627 __ delayed()->nop();
1628 __ stop("broken stack frame setup in interpreter in native");
1629 __ bind(L);
1630 }
1631 #endif
1633 // jvmti/jvmpi support
1634 __ notify_method_entry();
1636 __ dispatch_next(vtos);
1638 // invocation counter overflow
1639 if (inc_counter) {
1640 if (ProfileInterpreter) {
1641 // We have decided to profile this method in the interpreter
1642 __ bind(profile_method);
1643 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1644 InterpreterRuntime::profile_method));
1645 __ set_method_data_pointer_for_bcp();
1646 __ get_method(Rmethod);
1647 __ b(profile_method_continue);
1648 __ delayed()->nop();
1649 }
1650 // Handle overflow of counter and compile method
1651 __ bind(invocation_counter_overflow);
1652 generate_counter_overflow(&continue_after_compile);
1653 }
1655 return entry_point;
1656 }
1658 // Entry points
1659 //
1660 // Here we generate the various kind of entries into the interpreter.
1661 // The two main entry type are generic bytecode methods and native
1662 // call method. These both come in synchronized and non-synchronized
1663 // versions but the frame layout they create is very similar. The
1664 // other method entry types are really just special purpose entries
1665 // that are really entry and interpretation all in one. These are for
1666 // trivial methods like accessor, empty, or special math methods.
1667 //
1668 // When control flow reaches any of the entry types for the interpreter
1669 // the following holds ->
1670 //
1671 // Arguments:
1672 //
1673 // Rmethod: Method*
1674 // V0: receiver
1675 //
1676 //
1677 // Stack layout immediately at entry
1678 //
1679 // [ parameter n-1 ] <--- sp
1680 // ...
1681 // [ parameter 0 ]
1682 // [ expression stack ] (caller's java expression stack)
1684 // Assuming that we don't go to one of the trivial specialized entries
1685 // the stack will look like below when we are ready to execute the
1686 // first bytecode (or call the native routine). The register usage
1687 // will be as the template based interpreter expects (see
1688 // interpreter_amd64.hpp).
1689 //
1690 // local variables follow incoming parameters immediately; i.e.
1691 // the return address is moved to the end of the locals).
1692 //
1693 // [ monitor entry ] <--- sp
1694 // ...
1695 // [ monitor entry ]
1696 // [ monitor block top ] ( the top monitor entry )
1697 // [ byte code pointer ] (if native, bcp = 0)
1698 // [ constant pool cache ]
1699 // [ Method* ]
1700 // [ locals offset ]
1701 // [ sender's sp ]
1702 // [ sender's fp ]
1703 // [ return address ] <--- fp
1704 // [ local var m-1 ]
1705 // ...
1706 // [ local var 0 ]
1707 // [ argumnet word n-1 ] <--- ( sender's sp )
1708 // ...
1709 // [ argument word 0 ] <--- S7
1711 address AbstractInterpreterGenerator::generate_method_entry(
1712 AbstractInterpreter::MethodKind kind) {
1713 // determine code generation flags
1714 bool synchronized = false;
1715 address entry_point = NULL;
1716 switch (kind) {
1717 case Interpreter::zerolocals :
1718 break;
1719 case Interpreter::zerolocals_synchronized:
1720 synchronized = true;
1721 break;
1722 case Interpreter::native :
1723 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1724 break;
1725 case Interpreter::native_synchronized :
1726 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1727 break;
1728 case Interpreter::empty :
1729 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1730 break;
1731 case Interpreter::accessor :
1732 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1733 break;
1734 case Interpreter::abstract :
1735 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1736 break;
1738 case Interpreter::java_lang_math_sin : // fall thru
1739 case Interpreter::java_lang_math_cos : // fall thru
1740 case Interpreter::java_lang_math_tan : // fall thru
1741 case Interpreter::java_lang_math_log : // fall thru
1742 case Interpreter::java_lang_math_log10 : // fall thru
1743 case Interpreter::java_lang_math_pow : // fall thru
1744 case Interpreter::java_lang_math_exp : break;
1745 case Interpreter::java_lang_math_abs : // fall thru
1746 case Interpreter::java_lang_math_sqrt :
1747 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1748 case Interpreter::java_lang_ref_reference_get:
1749 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1750 default:
1751 fatal(err_msg("unexpected method kind: %d", kind));
1752 break;
1753 }
1754 if (entry_point) return entry_point;
1756 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1757 }
1759 // These should never be compiled since the interpreter will prefer
1760 // the compiled version to the intrinsic version.
1761 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1762 switch (method_kind(m)) {
1763 case Interpreter::java_lang_math_sin : // fall thru
1764 case Interpreter::java_lang_math_cos : // fall thru
1765 case Interpreter::java_lang_math_tan : // fall thru
1766 case Interpreter::java_lang_math_abs : // fall thru
1767 case Interpreter::java_lang_math_log : // fall thru
1768 case Interpreter::java_lang_math_log10 : // fall thru
1769 case Interpreter::java_lang_math_sqrt : // fall thru
1770 case Interpreter::java_lang_math_pow : // fall thru
1771 case Interpreter::java_lang_math_exp :
1772 return false;
1773 default:
1774 return true;
1775 }
1776 }
1778 // How much stack a method activation needs in words.
1779 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1781 const int entry_size = frame::interpreter_frame_monitor_size();
1783 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
1784 // be sure to change this if you add/subtract anything to/from the overhead area
1785 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1787 const int stub_code = 6; // see generate_call_stub
1788 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1789 const int method_stack = (method->max_locals() + method->max_stack()) *
1790 Interpreter::stackElementWords;
1791 return overhead_size + method_stack + stub_code;
1792 }
1794 void AbstractInterpreter::layout_activation(Method* method,
1795 int tempcount,
1796 int popframe_extra_args,
1797 int moncount,
1798 int caller_actual_parameters,
1799 int callee_param_count,
1800 int callee_locals,
1801 frame* caller,
1802 frame* interpreter_frame,
1803 bool is_top_frame,
1804 bool is_bottom_frame) {
1805 // Note: This calculation must exactly parallel the frame setup
1806 // in AbstractInterpreterGenerator::generate_method_entry.
1807 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1808 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1809 // right size, as determined by a previous call to this method.
1810 // It is also guaranteed to be walkable even though it is in a skeletal state
1812 // fixed size of an interpreter frame:
1814 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1815 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1817 #ifdef ASSERT
1818 if (!EnableInvokeDynamic) {
1819 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1820 // Probably, since deoptimization doesn't work yet.
1821 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1822 }
1823 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1824 #endif
1826 interpreter_frame->interpreter_frame_set_method(method);
1827 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1828 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1829 // and sender_sp is fp+8
1830 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1832 #ifdef ASSERT
1833 if (caller->is_interpreted_frame()) {
1834 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1835 }
1836 #endif
1838 interpreter_frame->interpreter_frame_set_locals(locals);
1839 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1840 BasicObjectLock* monbot = montop - moncount;
1841 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1843 //set last sp;
1844 intptr_t* esp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1845 popframe_extra_args;
1846 interpreter_frame->interpreter_frame_set_last_sp(esp);
1847 // All frames but the initial interpreter frame we fill in have a
1848 // value for sender_sp that allows walking the stack but isn't
1849 // truly correct. Correct the value here.
1850 //
1851 if (extra_locals != 0 &&
1852 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1853 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1854 }
1855 *interpreter_frame->interpreter_frame_cache_addr() = method->constants()->cache();
1856 }
1858 //-----------------------------------------------------------------------------
1859 // Exceptions
1861 void TemplateInterpreterGenerator::generate_throw_exception() {
1862 // Entry point in previous activation (i.e., if the caller was
1863 // interpreted)
1864 Interpreter::_rethrow_exception_entry = __ pc();
1865 // Restore sp to interpreter_frame_last_sp even though we are going
1866 // to empty the expression stack for the exception processing.
1867 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
1869 // V0: exception
1870 // V1: return address/pc that threw exception
1871 __ restore_bcp(); // esi points to call/send
1872 __ restore_locals();
1874 //add for compressedoops
1875 __ reinit_heapbase();
1876 // Entry point for exceptions thrown within interpreter code
1877 Interpreter::_throw_exception_entry = __ pc();
1878 // expression stack is undefined here
1879 // V0: exception
1880 // BCP: exception bcp
1881 __ verify_oop(V0);
1883 // expression stack must be empty before entering the VM in case of an exception
1884 __ empty_expression_stack();
1885 // find exception handler address and preserve exception oop
1886 __ move(A1, V0);
1887 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
1888 // V0: exception handler entry point
1889 // V1: preserved exception oop
1890 // S0: bcp for exception handler
1891 __ daddi(SP, SP, (-1) * wordSize);
1892 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
1893 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
1894 __ delayed()->nop();
1896 // If the exception is not handled in the current frame the frame is removed and
1897 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1898 //
1899 // Note: At this point the bci is still the bxi for the instruction which caused
1900 // the exception and the expression stack is empty. Thus, for any VM calls
1901 // at this point, GC will find a legal oop map (with empty expression stack).
1903 // In current activation
1904 // V0: exception
1905 // BCP: exception bcp
1907 //
1908 // JVMTI PopFrame support
1909 //
1911 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1912 __ empty_expression_stack();
1913 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1914 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1915 // popframe handling cycles.
1916 #ifndef OPT_THREAD
1917 Register thread = T2;
1918 __ get_thread(T2);
1919 #else
1920 Register thread = TREG;
1921 #endif
1922 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1923 __ ori(T3, T3, JavaThread::popframe_processing_bit);
1924 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1926 #ifndef CORE
1927 {
1928 // Check to see whether we are returning to a deoptimized frame.
1929 // (The PopFrame call ensures that the caller of the popped frame is
1930 // either interpreted or compiled and deoptimizes it if compiled.)
1931 // In this case, we can't call dispatch_next() after the frame is
1932 // popped, but instead must save the incoming arguments and restore
1933 // them after deoptimization has occurred.
1934 //
1935 // Note that we don't compare the return PC against the
1936 // deoptimization blob's unpack entry because of the presence of
1937 // adapter frames in C2.
1938 Label caller_not_deoptimized;
1939 __ ld(A0, FP, frame::return_addr_offset * wordSize);
1940 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
1941 __ bne(V0, R0, caller_not_deoptimized);
1942 __ delayed()->nop();
1944 // Compute size of arguments for saving when returning to deoptimized caller
1945 __ get_method(A1);
1946 __ verify_oop(A1);
1947 __ ld(A1,A1,in_bytes(Method::const_offset()));
1948 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
1949 __ shl(A1, Interpreter::logStackElementSize);
1950 __ restore_locals();
1951 __ dsub(A2, LVP, T0);
1952 __ daddiu(A2, A2, wordSize);
1953 // Save these arguments
1954 #ifndef OPT_THREAD
1955 __ get_thread(A0);
1956 #else
1957 __ move(A0, TREG);
1958 #endif
1959 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
1961 __ remove_activation(vtos, T9, false, false, false);
1963 // Inform deoptimization that it is responsible for restoring these arguments
1964 #ifndef OPT_THREAD
1965 __ get_thread(thread);
1966 #endif
1967 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
1968 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
1969 // Continue in deoptimization handler
1970 __ jr(T9);
1971 __ delayed()->nop();
1973 __ bind(caller_not_deoptimized);
1974 }
1975 #endif /* !CORE */
1977 __ remove_activation(vtos, T3,
1978 /* throw_monitor_exception */ false,
1979 /* install_monitor_exception */ false,
1980 /* notify_jvmdi */ false);
1982 // Clear the popframe condition flag
1983 // Finish with popframe handling
1984 // A previous I2C followed by a deoptimization might have moved the
1985 // outgoing arguments further up the stack. PopFrame expects the
1986 // mutations to those outgoing arguments to be preserved and other
1987 // constraints basically require this frame to look exactly as
1988 // though it had previously invoked an interpreted activation with
1989 // no space between the top of the expression stack (current
1990 // last_sp) and the top of stack. Rather than force deopt to
1991 // maintain this kind of invariant all the time we call a small
1992 // fixup routine to move the mutated arguments onto the top of our
1993 // expression stack if necessary.
1994 __ move(T8, SP);
1995 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
1996 #ifndef OPT_THREAD
1997 __ get_thread(thread);
1998 #endif
1999 // PC must point into interpreter here
2000 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2001 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2002 __ reset_last_Java_frame(thread, true, true);
2003 // Restore the last_sp and null it out
2004 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2005 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2009 __ move(AT, JavaThread::popframe_inactive);
2010 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2012 // Finish with popframe handling
2013 __ restore_bcp();
2014 __ restore_locals();
2015 #ifndef CORE
2016 // The method data pointer was incremented already during
2017 // call profiling. We have to restore the mdp for the current bcp.
2018 if (ProfileInterpreter) {
2019 __ set_method_data_pointer_for_bcp();
2020 }
2021 #endif // !CORE
2022 // Clear the popframe condition flag
2023 #ifndef OPT_THREAD
2024 __ get_thread(thread);
2025 #endif
2026 __ move(AT, JavaThread::popframe_inactive);
2027 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2028 __ dispatch_next(vtos);
2029 // end of PopFrame support
2031 Interpreter::_remove_activation_entry = __ pc();
2033 // preserve exception over this code sequence
2034 __ ld(T0, SP, 0);
2035 __ daddi(SP, SP, wordSize);
2036 #ifndef OPT_THREAD
2037 __ get_thread(thread);
2038 #endif
2039 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2040 // remove the activation (without doing throws on illegalMonitorExceptions)
2041 __ remove_activation(vtos, T3, false, true, false);
2042 // restore exception
2043 __ get_vm_result(T0, thread);
2044 __ verify_oop(T0);
2046 // Inbetween activations - previous activation type unknown yet
2047 // compute continuation point - the continuation point expects
2048 // the following registers set up:
2049 //
2050 // T0: exception eax
2051 // T1: return address/pc that threw exception edx
2052 // SP: expression stack of caller esp
2053 // FP: ebp of caller ebp
2054 __ daddi(SP, SP, (-2) * wordSize);
2055 __ sd(T0, SP, wordSize); // save exception
2056 __ sd(T3, SP, 0); // save return address
2057 __ move(A1, T3);
2058 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2059 __ move(T9, V0); // save exception handler
2060 __ ld(V0, SP, wordSize); // restore exception
2061 __ ld(V1, SP, 0); // restore return address
2062 __ daddi(SP, SP, 2 * wordSize);
2064 // Note that an "issuing PC" is actually the next PC after the call
2065 __ jr(T9); // jump to exception handler of caller
2066 __ delayed()->nop();
2067 }
2070 //
2071 // JVMTI ForceEarlyReturn support
2072 //
2073 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2074 address entry = __ pc();
2075 __ restore_bcp();
2076 __ restore_locals();
2077 __ empty_expression_stack();
2078 __ empty_FPU_stack();
2079 __ load_earlyret_value(state);
2081 #ifndef OPT_THREAD
2082 __ get_thread(TREG);
2083 #endif
2084 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2085 //const Address cond_addr(ecx, JvmtiThreadState::earlyret_state_offset());
2086 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2087 // Clear the earlyret state
2088 __ move(AT,JvmtiThreadState::earlyret_inactive);
2089 __ sw(AT,cond_addr);
2090 __ sync();
2093 __ remove_activation(state, T0,
2094 false, /* throw_monitor_exception */
2095 false, /* install_monitor_exception */
2096 true); /* notify_jvmdi */
2097 __ sync();
2098 __ jr(T0);
2099 __ delayed()->nop();
2100 return entry;
2101 } // end of ForceEarlyReturn support
2104 //-----------------------------------------------------------------------------
2105 // Helper for vtos entry point generation
2107 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2108 address& bep,
2109 address& cep,
2110 address& sep,
2111 address& aep,
2112 address& iep,
2113 address& lep,
2114 address& fep,
2115 address& dep,
2116 address& vep) {
2117 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2118 Label L;
2119 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2120 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2121 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2122 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2123 bep = cep = sep =
2124 iep = __ pc(); __ push(itos);
2125 vep = __ pc();
2126 __ bind(L);
2127 generate_and_dispatch(t);
2128 }
2131 //-----------------------------------------------------------------------------
2132 // Generation of individual instructions
2134 // helpers for generate_and_dispatch
2137 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2138 : TemplateInterpreterGenerator(code) {
2139 generate_all(); // down here so it can be "virtual"
2140 }
2142 //-----------------------------------------------------------------------------
2144 // Non-product code
2145 #ifndef PRODUCT
2146 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2147 address entry = __ pc();
2149 // prepare expression stack
2150 __ push(state); // save tosca
2152 // tos & tos2, added by yjl 7/15/2005
2153 // trace_bytecode need actually 4 args, the last two is tos&tos2
2154 // this work fine for x86. but mips o32 call convention will store A2-A3
2155 // to the stack position it think is the tos&tos2
2156 // when the expression stack have no more than 2 data, error occur.
2157 __ ld(A2, SP, 0);
2158 __ ld(A3, SP, 1 * wordSize);
2160 // pass arguments & call tracer
2161 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2162 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2164 // restore expression stack
2165 __ pop(state); // restore tosca
2167 // return
2168 __ jr(RA);
2169 __ delayed()->nop();
2171 return entry;
2172 }
2174 void TemplateInterpreterGenerator::count_bytecode() {
2175 __ li(T8, (long)&BytecodeCounter::_counter_value);
2176 __ lw(AT, T8, 0);
2177 __ daddi(AT, AT, 1);
2178 __ sw(AT, T8, 0);
2179 }
2181 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2182 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2183 __ lw(AT, T8, 0);
2184 __ daddi(AT, AT, 1);
2185 __ sw(AT, T8, 0);
2186 }
2188 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2189 __ li(T8, (long)&BytecodePairHistogram::_index);
2190 __ lw(T9, T8, 0);
2191 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2192 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2193 __ orr(T9, T9, T8);
2194 __ li(T8, (long)&BytecodePairHistogram::_index);
2195 __ sw(T9, T8, 0);
2196 __ dsll(T9, T9, 2);
2197 __ li(T8, (long)BytecodePairHistogram::_counters);
2198 __ dadd(T8, T8, T9);
2199 __ lw(AT, T8, 0);
2200 __ daddi(AT, AT, 1);
2201 __ sw(AT, T8, 0);
2202 }
2205 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2206 // Call a little run-time stub to avoid blow-up for each bytecode.
2207 // The run-time runtime saves the right registers, depending on
2208 // the tosca in-state for the given template.
2210 address entry = Interpreter::trace_code(t->tos_in());
2211 assert(entry != NULL, "entry must have been generated");
2212 __ call(entry, relocInfo::none);
2213 __ delayed()->nop();
2214 //add for compressedoops
2215 __ reinit_heapbase();
2216 }
2219 void TemplateInterpreterGenerator::stop_interpreter_at() {
2220 Label L;
2221 __ li(T8, long(&BytecodeCounter::_counter_value));
2222 __ lw(T8, T8, 0);
2223 __ move(AT, StopInterpreterAt);
2224 __ bne(T8, AT, L);
2225 __ delayed()->nop();
2226 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2227 __ delayed()->nop();
2228 __ bind(L);
2229 }
2230 #endif // !PRODUCT
2231 #endif // ! CC_INTERP