Wed, 08 Nov 2017 09:28:23 +0800
[GC] 17 out of 18 jtreg tests for g1 have passed (the same as x86 with jdk8u60-b32).
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* (Rmethod)
792 // r13: senderSP must preserve for slow path, set SP to it on fast path (Rsender)
794 // rax: V0
795 // rbx: Rmethod
796 // r13: Rsender
797 // rdi: T9
799 address entry = __ pc();
801 const int referent_offset = java_lang_ref_Reference::referent_offset;
802 guarantee(referent_offset > 0, "referent offset not initialized");
804 if (UseG1GC) {
805 Label slow_path;
807 // Check if local 0 != NULL
808 // If the receiver is null then it is OK to jump to the slow path.
809 __ ld(V0, SP, 0);
811 __ beq(V0, R0, slow_path);
812 __ nop();
814 // Generate the G1 pre-barrier code to log the value of
815 // the referent field in an SATB buffer.
817 // Load the value of the referent field.
818 const Address field_address(V0, referent_offset);
819 __ load_heap_oop(V0, field_address);
821 // Generate the G1 pre-barrier code to log the value of
822 // the referent field in an SATB buffer.
823 __ g1_write_barrier_pre(noreg /* obj */,
824 V0 /* pre_val */,
825 TREG /* thread */,
826 Rmethod /* tmp */,
827 true /* tosca_live */,
828 true /* expand_call */);
830 __ move(SP, Rsender); // set sp to sender sp
832 __ jr(RA);
833 __ nop();
835 // generate a vanilla interpreter entry as the slow path
836 __ bind(slow_path);
837 (void) generate_normal_entry(false);
839 return entry;
840 }
841 #endif // INCLUDE_ALL_GCS
843 // If G1 is not enabled then attempt to go through the accessor entry point
844 // Reference.get is an accessor
845 return generate_accessor_entry();
846 }
848 // Interpreter stub for calling a native method. (asm interpreter)
849 // This sets up a somewhat different looking stack for calling the
850 // native method than the typical interpreter frame setup.
851 address InterpreterGenerator::generate_native_entry(bool synchronized) {
852 // determine code generation flags
853 bool inc_counter = UseCompiler || CountCompiledCalls;
854 // Rsender: sender's sp
855 // Rmethod: Method*
856 address entry_point = __ pc();
858 #ifndef CORE
859 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() + // Fu: 20130814
860 InvocationCounter::counter_offset()));
861 #endif
863 // get parameter size (always needed)
864 // the size in the java stack
865 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
866 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
868 // native calls don't need the stack size check since they have no expression stack
869 // and the arguments are already on the stack and we only add a handful of words
870 // to the stack
872 // Rmethod: Method*
873 // V0: size of parameters
874 // Layout of frame at this point
875 //
876 // [ argument word n-1 ] <--- sp
877 // ...
878 // [ argument word 0 ]
880 // for natives the size of locals is zero
882 // compute beginning of parameters (S7)
883 __ dsll(LVP, V0, Address::times_8);
884 __ daddiu(LVP, LVP, (-1) * wordSize);
885 __ dadd(LVP, LVP, SP);
888 // add 2 zero-initialized slots for native calls
889 __ daddi(SP, SP, (-2) * wordSize);
890 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
891 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
893 // Layout of frame at this point
894 // [ method holder mirror ] <--- sp
895 // [ result type info ]
896 // [ argument word n-1 ] <--- T0
897 // ...
898 // [ argument word 0 ] <--- LVP
901 #ifndef CORE
902 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
903 #endif
905 // initialize fixed part of activation frame
906 generate_fixed_frame(true);
907 // after this function, the layout of frame is as following
908 //
909 // [ monitor block top ] <--- sp ( the top monitor entry )
910 // [ byte code pointer (0) ] (if native, bcp = 0)
911 // [ constant pool cache ]
912 // [ Method* ]
913 // [ locals offset ]
914 // [ sender's sp ]
915 // [ sender's fp ]
916 // [ return address ] <--- fp
917 // [ method holder mirror ]
918 // [ result type info ]
919 // [ argumnet word n-1 ] <--- sender's sp
920 // ...
921 // [ argument word 0 ] <--- S7
924 // make sure method is native & not abstract
925 #ifdef ASSERT
926 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
927 {
928 Label L;
929 __ andi(AT, T0, JVM_ACC_NATIVE);
930 __ bne(AT, R0, L);
931 __ delayed()->nop();
932 __ stop("tried to execute native method as non-native");
933 __ bind(L);
934 }
935 {
936 Label L;
937 __ andi(AT, T0, JVM_ACC_ABSTRACT);
938 __ beq(AT, R0, L);
939 __ delayed()->nop();
940 __ stop("tried to execute abstract method in interpreter");
941 __ bind(L);
942 }
943 #endif
945 // Since at this point in the method invocation the exception handler
946 // would try to exit the monitor of synchronized methods which hasn't
947 // been entered yet, we set the thread local variable
948 // _do_not_unlock_if_synchronized to true. The remove_activation will
949 // check this flag.
950 Register thread = TREG;
951 #ifndef OPT_THREAD
952 __ get_thread(thread);
953 #endif
954 __ move(AT, (int)true);
955 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
957 #ifndef CORE
958 // increment invocation count & check for overflow
959 Label invocation_counter_overflow;
960 if (inc_counter) {
961 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
962 }
964 Label continue_after_compile;
965 __ bind(continue_after_compile);
966 #endif // CORE
968 bang_stack_shadow_pages(true);
970 // reset the _do_not_unlock_if_synchronized flag
971 #ifndef OPT_THREAD
972 __ get_thread(thread);
973 #endif
974 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
976 // check for synchronized methods
977 // Must happen AFTER invocation_counter check and stack overflow check,
978 // so method is not locked if overflows.
979 if (synchronized) {
980 lock_method();
981 } else {
982 // no synchronization necessary
983 #ifdef ASSERT
984 {
985 Label L;
986 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
987 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
988 __ beq(AT, R0, L);
989 __ delayed()->nop();
990 __ stop("method needs synchronization");
991 __ bind(L);
992 }
993 #endif
994 }
996 // after method_lock, the layout of frame is as following
997 //
998 // [ monitor entry ] <--- sp
999 // ...
1000 // [ monitor entry ]
1001 // [ monitor block top ] ( the top monitor entry )
1002 // [ byte code pointer (0) ] (if native, bcp = 0)
1003 // [ constant pool cache ]
1004 // [ Method* ]
1005 // [ locals offset ]
1006 // [ sender's sp ]
1007 // [ sender's fp ]
1008 // [ return address ] <--- fp
1009 // [ method holder mirror ]
1010 // [ result type info ]
1011 // [ argumnet word n-1 ] <--- ( sender's sp )
1012 // ...
1013 // [ argument word 0 ] <--- S7
1015 // start execution
1016 #ifdef ASSERT
1017 {
1018 Label L;
1019 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1020 __ beq(AT, SP, L);
1021 __ delayed()->nop();
1022 __ stop("broken stack frame setup in interpreter in asm");
1023 __ bind(L);
1024 }
1025 #endif
1027 // jvmti/jvmpi support
1028 __ notify_method_entry();
1030 // work registers
1031 const Register method = Rmethod;
1032 //const Register thread = T2;
1033 const Register t = RT4;
1035 __ get_method(method);
1036 __ verify_oop(method);
1037 {
1038 Label L, Lstatic;
1039 __ ld(t,method,in_bytes(Method::const_offset()));
1040 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
1041 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1042 //FIXME, aoqi: A1?
1043 // A0 and A1(if needed)
1044 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1045 __ andi(AT, AT, JVM_ACC_STATIC);
1046 __ beq(AT, R0, Lstatic);
1047 __ delayed()->nop();
1048 __ daddiu(t, t, 1);
1049 __ bind(Lstatic);
1050 __ daddiu(t, t, -7);
1051 __ blez(t, L);
1052 __ delayed()->nop();
1053 __ dsll(t, t, Address::times_8);
1054 __ dsub(SP, SP, t);
1055 __ bind(L);
1056 }
1057 __ move(AT, -(StackAlignmentInBytes));
1058 __ andr(SP, SP, AT);
1059 __ move(AT, SP);
1060 // [ ] <--- sp
1061 // ... (size of parameters - 8 )
1062 // [ monitor entry ]
1063 // ...
1064 // [ monitor entry ]
1065 // [ monitor block top ] ( the top monitor entry )
1066 // [ byte code pointer (0) ] (if native, bcp = 0)
1067 // [ constant pool cache ]
1068 // [ Method* ]
1069 // [ locals offset ]
1070 // [ sender's sp ]
1071 // [ sender's fp ]
1072 // [ return address ] <--- fp
1073 // [ method holder mirror ]
1074 // [ result type info ]
1075 // [ argumnet word n-1 ] <--- ( sender's sp )
1076 // ...
1077 // [ argument word 0 ] <--- LVP
1079 // get signature handler
1080 {
1081 Label L;
1082 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1083 __ bne(T9, R0, L);
1084 __ delayed()->nop();
1085 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1086 InterpreterRuntime::prepare_native_call), method);
1087 __ get_method(method);
1088 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1089 __ bind(L);
1090 }
1092 // call signature handler
1093 // FIXME: when change codes in InterpreterRuntime, note this point
1094 // from: begin of parameters
1095 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1096 // to: current sp
1097 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1098 // temp: T3
1099 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1101 __ jalr(T9);
1102 __ delayed()->nop();
1103 __ get_method(method); // slow path call blows EBX on DevStudio 5.0
1105 /*
1106 if native function is static, and its second parameter has type length of double word,
1107 and first parameter has type length of word, we have to reserve one word
1108 for the first parameter, according to mips o32 abi.
1109 if native function is not static, and its third parameter has type length of double word,
1110 and second parameter has type length of word, we have to reserve one word for the second
1111 parameter.
1112 */
1115 // result handler is in V0
1116 // set result handler
1117 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1119 #define FIRSTPARA_SHIFT_COUNT 5
1120 #define SECONDPARA_SHIFT_COUNT 9
1121 #define THIRDPARA_SHIFT_COUNT 13
1122 #define PARA_MASK 0xf
1124 // pass mirror handle if static call
1125 {
1126 Label L;
1127 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1128 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1129 __ andi(AT, t, JVM_ACC_STATIC);
1130 __ beq(AT, R0, L);
1131 __ delayed()->nop();
1133 // get mirror
1134 __ ld(t, method, in_bytes(Method:: const_offset()));
1135 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1136 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1137 __ ld(t, t, mirror_offset);
1138 // copy mirror into activation frame
1139 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1140 // pass handle to mirror
1141 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1142 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1143 __ move(A1, t);
1144 __ bind(L);
1145 }
1147 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1148 // [ ] |
1149 // ... size of parameters(or +1) |
1150 // [ monitor entry ] |
1151 // ... |
1152 // [ monitor entry ] |
1153 // [ monitor block top ] ( the top monitor entry ) |
1154 // [ byte code pointer (0) ] (if native, bcp = 0) |
1155 // [ constant pool cache ] |
1156 // [ Method* ] |
1157 // [ locals offset ] |
1158 // [ sender's sp ] |
1159 // [ sender's fp ] |
1160 // [ return address ] <--- fp |
1161 // [ method holder mirror ] <----------------------------|
1162 // [ result type info ]
1163 // [ argumnet word n-1 ] <--- ( sender's sp )
1164 // ...
1165 // [ argument word 0 ] <--- S7
1167 // get native function entry point
1168 { Label L;
1169 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1170 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1171 __ bne(V1, T9, L);
1172 __ delayed()->nop();
1173 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1174 __ get_method(method);
1175 __ verify_oop(method);
1176 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1177 __ bind(L);
1178 }
1179 /*
1180 __ pushad();
1181 __ move(A0, T9);
1182 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
1183 __ popad();
1184 */
1186 // pass JNIEnv
1187 // native function in T9
1188 #ifndef OPT_THREAD
1189 __ get_thread(thread);
1190 #endif
1191 __ daddi(t, thread, in_bytes(JavaThread::jni_environment_offset()));
1192 // stack,but I think it won't work when pass float,double etc @jerome,10/17,2006
1193 __ move(A0, t);
1194 // [ jni environment ] <--- sp
1195 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1196 // [ ] |
1197 // ... size of parameters |
1198 // [ monitor entry ] |
1199 // ... |
1200 // [ monitor entry ] |
1201 // [ monitor block top ] ( the top monitor entry ) |
1202 // [ byte code pointer (0) ] (if native, bcp = 0) |
1203 // [ constant pool cache ] |
1204 // [ Method* ] |
1205 // [ locals offset ] |
1206 // [ sender's sp ] |
1207 // [ sender's fp ] |
1208 // [ return address ] <--- fp |
1209 // [ method holder mirror ] <----------------------------|
1210 // [ result type info ]
1211 // [ argumnet word n-1 ] <--- ( sender's sp )
1212 // ...
1213 // [ argument word 0 ] <--- S7
1215 // set_last_Java_frame_before_call
1216 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1217 // Change state to native (we save the return address in the thread, since it might not
1218 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1219 // points into the right code segment. It does not have to be the correct return pc.
1220 __ li(t, __ pc());
1221 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1222 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1224 // change thread state
1225 #ifdef ASSERT
1226 {
1227 Label L;
1228 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1229 __ daddi(t, t, (-1) * _thread_in_Java);
1230 __ beq(t, R0, L);
1231 __ delayed()->nop();
1232 __ stop("Wrong thread state in native stub");
1233 __ bind(L);
1234 }
1235 #endif
1237 __ move(t, _thread_in_native);
1238 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1240 // call native method
1241 __ jalr(T9);
1242 __ delayed()->nop();
1243 // result potentially in V2:V1 or F0:F1
1246 // via _last_native_pc and not via _last_jave_sp
1247 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1248 // If the order changes or anything else is added to the stack the code in
1249 // interpreter_frame_result will have to be changed.
1250 //FIXME, should modify here
1251 // save return value to keep the value from being destroyed by other calls
1252 __ move(S1, V0);
1253 __ move(S3, V1);
1254 __ dmfc1(S4, F0);
1255 __ dmfc1(S2, F1);
1257 // change thread state
1258 __ get_thread(thread);
1259 __ move(t, _thread_in_native_trans);
1260 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1262 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1264 // check for safepoint operation in progress and/or pending suspend requests
1265 { Label Continue;
1267 // Don't use call_VM as it will see a possible pending exception and forward it
1268 // and never return here preventing us from clearing _last_native_pc down below.
1269 // Also can't use call_VM_leaf either as it will check to see if esi & edi are
1270 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1271 // by hand.
1272 //
1273 Label L;
1274 __ li(AT, SafepointSynchronize::address_of_state());
1275 __ lw(AT, AT, 0);
1276 __ bne(AT, R0, L);
1277 __ delayed()->nop();
1278 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1279 __ beq(AT, R0, Continue);
1280 __ delayed()->nop();
1281 __ bind(L);
1282 __ move(A0, thread);
1283 __ call(CAST_FROM_FN_PTR(address,
1284 JavaThread::check_special_condition_for_native_trans),
1285 relocInfo::runtime_call_type);
1286 __ delayed()->nop();
1288 #ifndef OPT_THREAD
1289 __ get_thread(thread);
1290 #endif
1291 //add for compressedoops
1292 __ reinit_heapbase();
1293 __ bind(Continue);
1294 }
1296 // change thread state
1297 __ move(t, _thread_in_Java);
1298 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1299 __ reset_last_Java_frame(thread, true, true);
1301 // reset handle block
1302 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1303 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1305 // If result was an oop then unbox and save it in the frame
1306 { Label L;
1307 Label no_oop, store_result;
1308 //FIXME, addi only support 16-bit imeditate
1309 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1310 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1311 __ bne(AT, T0, no_oop);
1312 __ delayed()->nop();
1313 __ move(V0, S1);
1314 __ beq(V0, R0, store_result);
1315 __ delayed()->nop();
1316 // unbox
1317 __ ld(V0, V0, 0);
1318 __ bind(store_result);
1319 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1320 // keep stack depth as expected by pushing oop which will eventually be discarded
1321 __ bind(no_oop);
1322 }
1323 {
1324 Label no_reguard;
1325 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1326 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1327 __ bne(t, AT, no_reguard);
1328 __ delayed()->nop();
1329 __ pushad();
1330 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::runtime_call_type);
1331 __ delayed()->nop();
1332 __ popad();
1333 //add for compressedoops
1334 __ reinit_heapbase();
1335 __ bind(no_reguard);
1336 }
1337 // restore esi to have legal interpreter frame,
1338 // i.e., bci == 0 <=> esi == code_base()
1339 // Can't call_VM until bcp is within reasonable.
1340 __ get_method(method); // method is junk from thread_in_native to now.
1341 __ verify_oop(method);
1342 __ ld(BCP, method, in_bytes(Method::const_offset()));
1343 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1344 // handle exceptions (exception handling will handle unlocking!)
1345 {
1346 Label L;
1347 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1348 __ beq(t, R0, L);
1349 __ delayed()->nop();
1350 // Note: At some point we may want to unify this with the code used in
1351 // call_VM_base();
1352 // i.e., we should use the StubRoutines::forward_exception code. For now this
1353 // doesn't work here because the esp is not correctly set at this point.
1354 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address,
1355 InterpreterRuntime::throw_pending_exception));
1356 __ should_not_reach_here();
1357 __ bind(L);
1358 }
1360 // do unlocking if necessary
1361 {
1362 Label L;
1363 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1364 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1365 __ beq(t, R0, L);
1366 // the code below should be shared with interpreter macro assembler implementation
1367 {
1368 Label unlock;
1369 // BasicObjectLock will be first in list,
1370 // since this is a synchronized method. However, need
1371 // to check that the object has not been unlocked by
1372 // an explicit monitorexit bytecode.
1373 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1374 // address of first monitor
1376 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1377 __ bne(t, R0, unlock);
1378 __ delayed()->nop();
1380 // Entry already unlocked, need to throw exception
1381 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1382 InterpreterRuntime::throw_illegal_monitor_state_exception));
1383 __ should_not_reach_here();
1385 __ bind(unlock);
1386 __ unlock_object(c_rarg0);
1387 }
1388 __ bind(L);
1389 }
1391 // jvmti/jvmpi support
1392 // Note: This must happen _after_ handling/throwing any exceptions since
1393 // the exception handler code notifies the runtime of method exits
1394 // too. If this happens before, method entry/exit notifications are
1395 // not properly paired (was bug - gri 11/22/99).
1396 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI );
1398 // restore potential result in V0:V1,
1399 // call result handler to restore potential result in ST0 & handle result
1400 //__ lw(V0, SP, 3 * wordSize);
1401 //__ lw(V1, SP, 2 * wordSize);
1402 //__ lwc1(F0, SP, 1 * wordSize);
1403 //__ lwc1(F1, SP, 0 * wordSize);
1404 //__ addi(SP, SP, 4 * wordSize);
1405 __ move(V0, S1);
1406 __ move(V1, S3);
1407 __ dmtc1(S4, F0);
1408 __ dmtc1(S2, F1);
1409 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1410 __ jalr(t);
1411 __ delayed()->nop();
1414 // remove activation
1415 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1416 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1417 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1418 __ jr(RA);
1419 __ delayed()->nop();
1421 #ifndef CORE
1422 if (inc_counter) {
1423 // Handle overflow of counter and compile method
1424 __ bind(invocation_counter_overflow);
1425 generate_counter_overflow(&continue_after_compile);
1426 // entry_point is the beginning of this
1427 // function and checks again for compiled code
1428 }
1429 #endif
1430 return entry_point;
1431 }
1433 //
1434 // Generic interpreted method entry to (asm) interpreter
1435 //
1436 // Layout of frame just at the entry
1437 //
1438 // [ argument word n-1 ] <--- sp
1439 // ...
1440 // [ argument word 0 ]
1441 // assume Method* in Rmethod before call this method.
1442 // prerequisites to the generated stub : the callee Method* in Rmethod
1443 // note you must save the caller bcp before call the generated stub
1444 //
1445 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1446 // determine code generation flags
1447 bool inc_counter = UseCompiler || CountCompiledCalls;
1449 // Rmethod: Method*
1450 // Rsender: sender 's sp
1451 address entry_point = __ pc();
1453 const Address invocation_counter(Rmethod,
1454 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1456 // get parameter size (always needed)
1457 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1458 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1460 // Rmethod: Method*
1461 // V0: size of parameters
1462 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1463 // get size of locals in words to T2
1464 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1465 // T2 = no. of additional locals, locals include parameters
1466 __ dsub(T2, T2, V0);
1468 // see if we've got enough room on the stack for locals plus overhead.
1469 // Layout of frame at this point
1470 //
1471 // [ argument word n-1 ] <--- sp
1472 // ...
1473 // [ argument word 0 ]
1474 generate_stack_overflow_check();
1475 // after this function, the layout of frame does not change
1477 // compute beginning of parameters (LVP)
1478 __ dsll(LVP, V0, LogBytesPerWord);
1479 __ daddiu(LVP, LVP, (-1) * wordSize);
1480 __ dadd(LVP, LVP, SP);
1482 // T2 - # of additional locals
1483 // allocate space for locals
1484 // explicitly initialize locals
1485 {
1486 Label exit, loop;
1487 __ beq(T2, R0, exit);
1488 __ delayed()->nop();
1490 __ bind(loop);
1491 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1492 __ daddiu(T2, T2, -1); // until everything initialized
1493 __ bne(T2, R0, loop);
1494 __ delayed();
1496 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1498 __ bind(exit);
1499 }
1501 //
1502 // [ local var m-1 ] <--- sp
1503 // ...
1504 // [ local var 0 ]
1505 // [ argument word n-1 ] <--- T0?
1506 // ...
1507 // [ argument word 0 ] <--- LVP
1509 // initialize fixed part of activation frame
1511 generate_fixed_frame(false);
1514 // after this function, the layout of frame is as following
1515 //
1516 // [ monitor block top ] <--- sp ( the top monitor entry )
1517 // [ byte code pointer ] (if native, bcp = 0)
1518 // [ constant pool cache ]
1519 // [ Method* ]
1520 // [ locals offset ]
1521 // [ sender's sp ]
1522 // [ sender's fp ] <--- fp
1523 // [ return address ]
1524 // [ local var m-1 ]
1525 // ...
1526 // [ local var 0 ]
1527 // [ argumnet word n-1 ] <--- ( sender's sp )
1528 // ...
1529 // [ argument word 0 ] <--- LVP
1532 // make sure method is not native & not abstract
1533 #ifdef ASSERT
1534 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1535 {
1536 Label L;
1537 __ andi(T2, AT, JVM_ACC_NATIVE);
1538 __ beq(T2, R0, L);
1539 __ delayed()->nop();
1540 __ stop("tried to execute native method as non-native");
1541 __ bind(L);
1542 }
1543 {
1544 Label L;
1545 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1546 __ beq(T2, R0, L);
1547 __ delayed()->nop();
1548 __ stop("tried to execute abstract method in interpreter");
1549 __ bind(L);
1550 }
1551 #endif
1553 // Since at this point in the method invocation the exception handler
1554 // would try to exit the monitor of synchronized methods which hasn't
1555 // been entered yet, we set the thread local variable
1556 // _do_not_unlock_if_synchronized to true. The remove_activation will
1557 // check this flag.
1559 #ifndef OPT_THREAD
1560 Register thread = T8;
1561 __ get_thread(thread);
1562 #else
1563 Register thread = TREG;
1564 #endif
1565 __ move(AT, (int)true);
1566 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1568 #ifndef CORE
1570 // 2014/11/24 Fu
1571 // mdp : T8
1572 // tmp1: T9
1573 // tmp2: T2
1574 __ profile_parameters_type(T8, T9, T2);
1576 // increment invocation count & check for overflow
1577 Label invocation_counter_overflow;
1578 Label profile_method;
1579 Label profile_method_continue;
1580 if (inc_counter) {
1581 generate_counter_incr(&invocation_counter_overflow,
1582 &profile_method,
1583 &profile_method_continue);
1584 if (ProfileInterpreter) {
1585 __ bind(profile_method_continue);
1586 }
1587 }
1589 Label continue_after_compile;
1590 __ bind(continue_after_compile);
1592 #endif // CORE
1594 bang_stack_shadow_pages(false);
1596 // reset the _do_not_unlock_if_synchronized flag
1597 #ifndef OPT_THREAD
1598 __ get_thread(thread);
1599 #endif
1600 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1602 // check for synchronized methods
1603 // Must happen AFTER invocation_counter check and stack overflow check,
1604 // so method is not locked if overflows.
1605 //
1606 if (synchronized) {
1607 // Allocate monitor and lock method
1608 lock_method();
1609 } else {
1610 // no synchronization necessary
1611 #ifdef ASSERT
1612 { Label L;
1613 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1614 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1615 __ beq(T2, R0, L);
1616 __ delayed()->nop();
1617 __ stop("method needs synchronization");
1618 __ bind(L);
1619 }
1620 #endif
1621 }
1623 // layout of frame after lock_method
1624 // [ monitor entry ] <--- sp
1625 // ...
1626 // [ monitor entry ]
1627 // [ monitor block top ] ( the top monitor entry )
1628 // [ byte code pointer ] (if native, bcp = 0)
1629 // [ constant pool cache ]
1630 // [ Method* ]
1631 // [ locals offset ]
1632 // [ sender's sp ]
1633 // [ sender's fp ]
1634 // [ return address ] <--- fp
1635 // [ local var m-1 ]
1636 // ...
1637 // [ local var 0 ]
1638 // [ argumnet word n-1 ] <--- ( sender's sp )
1639 // ...
1640 // [ argument word 0 ] <--- LVP
1643 // start execution
1644 #ifdef ASSERT
1645 {
1646 Label L;
1647 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1648 __ beq(AT, SP, L);
1649 __ delayed()->nop();
1650 __ stop("broken stack frame setup in interpreter in native");
1651 __ bind(L);
1652 }
1653 #endif
1655 // jvmti/jvmpi support
1656 __ notify_method_entry();
1658 __ dispatch_next(vtos);
1660 // invocation counter overflow
1661 if (inc_counter) {
1662 if (ProfileInterpreter) {
1663 // We have decided to profile this method in the interpreter
1664 __ bind(profile_method);
1665 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1666 InterpreterRuntime::profile_method));
1667 __ set_method_data_pointer_for_bcp();
1668 __ get_method(Rmethod);
1669 __ b(profile_method_continue);
1670 __ delayed()->nop();
1671 }
1672 // Handle overflow of counter and compile method
1673 __ bind(invocation_counter_overflow);
1674 generate_counter_overflow(&continue_after_compile);
1675 }
1677 return entry_point;
1678 }
1680 // Entry points
1681 //
1682 // Here we generate the various kind of entries into the interpreter.
1683 // The two main entry type are generic bytecode methods and native
1684 // call method. These both come in synchronized and non-synchronized
1685 // versions but the frame layout they create is very similar. The
1686 // other method entry types are really just special purpose entries
1687 // that are really entry and interpretation all in one. These are for
1688 // trivial methods like accessor, empty, or special math methods.
1689 //
1690 // When control flow reaches any of the entry types for the interpreter
1691 // the following holds ->
1692 //
1693 // Arguments:
1694 //
1695 // Rmethod: Method*
1696 // V0: receiver
1697 //
1698 //
1699 // Stack layout immediately at entry
1700 //
1701 // [ parameter n-1 ] <--- sp
1702 // ...
1703 // [ parameter 0 ]
1704 // [ expression stack ] (caller's java expression stack)
1706 // Assuming that we don't go to one of the trivial specialized entries
1707 // the stack will look like below when we are ready to execute the
1708 // first bytecode (or call the native routine). The register usage
1709 // will be as the template based interpreter expects (see
1710 // interpreter_amd64.hpp).
1711 //
1712 // local variables follow incoming parameters immediately; i.e.
1713 // the return address is moved to the end of the locals).
1714 //
1715 // [ monitor entry ] <--- sp
1716 // ...
1717 // [ monitor entry ]
1718 // [ monitor block top ] ( the top monitor entry )
1719 // [ byte code pointer ] (if native, bcp = 0)
1720 // [ constant pool cache ]
1721 // [ Method* ]
1722 // [ locals offset ]
1723 // [ sender's sp ]
1724 // [ sender's fp ]
1725 // [ return address ] <--- fp
1726 // [ local var m-1 ]
1727 // ...
1728 // [ local var 0 ]
1729 // [ argumnet word n-1 ] <--- ( sender's sp )
1730 // ...
1731 // [ argument word 0 ] <--- S7
1733 address AbstractInterpreterGenerator::generate_method_entry(
1734 AbstractInterpreter::MethodKind kind) {
1735 // determine code generation flags
1736 bool synchronized = false;
1737 address entry_point = NULL;
1738 switch (kind) {
1739 case Interpreter::zerolocals :
1740 break;
1741 case Interpreter::zerolocals_synchronized:
1742 synchronized = true;
1743 break;
1744 case Interpreter::native :
1745 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1746 break;
1747 case Interpreter::native_synchronized :
1748 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1749 break;
1750 case Interpreter::empty :
1751 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1752 break;
1753 case Interpreter::accessor :
1754 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1755 break;
1756 case Interpreter::abstract :
1757 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1758 break;
1760 case Interpreter::java_lang_math_sin : // fall thru
1761 case Interpreter::java_lang_math_cos : // fall thru
1762 case Interpreter::java_lang_math_tan : // fall thru
1763 case Interpreter::java_lang_math_log : // fall thru
1764 case Interpreter::java_lang_math_log10 : // fall thru
1765 case Interpreter::java_lang_math_pow : // fall thru
1766 case Interpreter::java_lang_math_exp : break;
1767 case Interpreter::java_lang_math_abs : // fall thru
1768 case Interpreter::java_lang_math_sqrt :
1769 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1770 case Interpreter::java_lang_ref_reference_get:
1771 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1772 default:
1773 fatal(err_msg("unexpected method kind: %d", kind));
1774 break;
1775 }
1776 if (entry_point) return entry_point;
1778 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1779 }
1781 // These should never be compiled since the interpreter will prefer
1782 // the compiled version to the intrinsic version.
1783 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1784 switch (method_kind(m)) {
1785 case Interpreter::java_lang_math_sin : // fall thru
1786 case Interpreter::java_lang_math_cos : // fall thru
1787 case Interpreter::java_lang_math_tan : // fall thru
1788 case Interpreter::java_lang_math_abs : // fall thru
1789 case Interpreter::java_lang_math_log : // fall thru
1790 case Interpreter::java_lang_math_log10 : // fall thru
1791 case Interpreter::java_lang_math_sqrt : // fall thru
1792 case Interpreter::java_lang_math_pow : // fall thru
1793 case Interpreter::java_lang_math_exp :
1794 return false;
1795 default:
1796 return true;
1797 }
1798 }
1800 // How much stack a method activation needs in words.
1801 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1803 const int entry_size = frame::interpreter_frame_monitor_size();
1805 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
1806 // be sure to change this if you add/subtract anything to/from the overhead area
1807 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1809 const int stub_code = 6; // see generate_call_stub
1810 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1811 const int method_stack = (method->max_locals() + method->max_stack()) *
1812 Interpreter::stackElementWords;
1813 return overhead_size + method_stack + stub_code;
1814 }
1816 void AbstractInterpreter::layout_activation(Method* method,
1817 int tempcount,
1818 int popframe_extra_args,
1819 int moncount,
1820 int caller_actual_parameters,
1821 int callee_param_count,
1822 int callee_locals,
1823 frame* caller,
1824 frame* interpreter_frame,
1825 bool is_top_frame,
1826 bool is_bottom_frame) {
1827 // Note: This calculation must exactly parallel the frame setup
1828 // in AbstractInterpreterGenerator::generate_method_entry.
1829 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1830 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1831 // right size, as determined by a previous call to this method.
1832 // It is also guaranteed to be walkable even though it is in a skeletal state
1834 // fixed size of an interpreter frame:
1836 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1837 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1839 #ifdef ASSERT
1840 if (!EnableInvokeDynamic) {
1841 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1842 // Probably, since deoptimization doesn't work yet.
1843 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1844 }
1845 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1846 #endif
1848 interpreter_frame->interpreter_frame_set_method(method);
1849 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1850 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1851 // and sender_sp is fp+8
1852 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1854 #ifdef ASSERT
1855 if (caller->is_interpreted_frame()) {
1856 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1857 }
1858 #endif
1860 interpreter_frame->interpreter_frame_set_locals(locals);
1861 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1862 BasicObjectLock* monbot = montop - moncount;
1863 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1865 //set last sp;
1866 intptr_t* esp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1867 popframe_extra_args;
1868 interpreter_frame->interpreter_frame_set_last_sp(esp);
1869 // All frames but the initial interpreter frame we fill in have a
1870 // value for sender_sp that allows walking the stack but isn't
1871 // truly correct. Correct the value here.
1872 //
1873 if (extra_locals != 0 &&
1874 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1875 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1876 }
1877 *interpreter_frame->interpreter_frame_cache_addr() = method->constants()->cache();
1878 }
1880 //-----------------------------------------------------------------------------
1881 // Exceptions
1883 void TemplateInterpreterGenerator::generate_throw_exception() {
1884 // Entry point in previous activation (i.e., if the caller was
1885 // interpreted)
1886 Interpreter::_rethrow_exception_entry = __ pc();
1887 // Restore sp to interpreter_frame_last_sp even though we are going
1888 // to empty the expression stack for the exception processing.
1889 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
1891 // V0: exception
1892 // V1: return address/pc that threw exception
1893 __ restore_bcp(); // esi points to call/send
1894 __ restore_locals();
1896 //add for compressedoops
1897 __ reinit_heapbase();
1898 // Entry point for exceptions thrown within interpreter code
1899 Interpreter::_throw_exception_entry = __ pc();
1900 // expression stack is undefined here
1901 // V0: exception
1902 // BCP: exception bcp
1903 __ verify_oop(V0);
1905 // expression stack must be empty before entering the VM in case of an exception
1906 __ empty_expression_stack();
1907 // find exception handler address and preserve exception oop
1908 __ move(A1, V0);
1909 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
1910 // V0: exception handler entry point
1911 // V1: preserved exception oop
1912 // S0: bcp for exception handler
1913 __ daddi(SP, SP, (-1) * wordSize);
1914 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
1915 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
1916 __ delayed()->nop();
1918 // If the exception is not handled in the current frame the frame is removed and
1919 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1920 //
1921 // Note: At this point the bci is still the bxi for the instruction which caused
1922 // the exception and the expression stack is empty. Thus, for any VM calls
1923 // at this point, GC will find a legal oop map (with empty expression stack).
1925 // In current activation
1926 // V0: exception
1927 // BCP: exception bcp
1929 //
1930 // JVMTI PopFrame support
1931 //
1933 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1934 __ empty_expression_stack();
1935 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1936 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1937 // popframe handling cycles.
1938 #ifndef OPT_THREAD
1939 Register thread = T2;
1940 __ get_thread(T2);
1941 #else
1942 Register thread = TREG;
1943 #endif
1944 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1945 __ ori(T3, T3, JavaThread::popframe_processing_bit);
1946 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1948 #ifndef CORE
1949 {
1950 // Check to see whether we are returning to a deoptimized frame.
1951 // (The PopFrame call ensures that the caller of the popped frame is
1952 // either interpreted or compiled and deoptimizes it if compiled.)
1953 // In this case, we can't call dispatch_next() after the frame is
1954 // popped, but instead must save the incoming arguments and restore
1955 // them after deoptimization has occurred.
1956 //
1957 // Note that we don't compare the return PC against the
1958 // deoptimization blob's unpack entry because of the presence of
1959 // adapter frames in C2.
1960 Label caller_not_deoptimized;
1961 __ ld(A0, FP, frame::return_addr_offset * wordSize);
1962 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
1963 __ bne(V0, R0, caller_not_deoptimized);
1964 __ delayed()->nop();
1966 // Compute size of arguments for saving when returning to deoptimized caller
1967 __ get_method(A1);
1968 __ verify_oop(A1);
1969 __ ld(A1,A1,in_bytes(Method::const_offset()));
1970 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
1971 __ shl(A1, Interpreter::logStackElementSize);
1972 __ restore_locals();
1973 __ dsub(A2, LVP, T0);
1974 __ daddiu(A2, A2, wordSize);
1975 // Save these arguments
1976 #ifndef OPT_THREAD
1977 __ get_thread(A0);
1978 #else
1979 __ move(A0, TREG);
1980 #endif
1981 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
1983 __ remove_activation(vtos, T9, false, false, false);
1985 // Inform deoptimization that it is responsible for restoring these arguments
1986 #ifndef OPT_THREAD
1987 __ get_thread(thread);
1988 #endif
1989 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
1990 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
1991 // Continue in deoptimization handler
1992 __ jr(T9);
1993 __ delayed()->nop();
1995 __ bind(caller_not_deoptimized);
1996 }
1997 #endif /* !CORE */
1999 __ remove_activation(vtos, T3,
2000 /* throw_monitor_exception */ false,
2001 /* install_monitor_exception */ false,
2002 /* notify_jvmdi */ false);
2004 // Clear the popframe condition flag
2005 // Finish with popframe handling
2006 // A previous I2C followed by a deoptimization might have moved the
2007 // outgoing arguments further up the stack. PopFrame expects the
2008 // mutations to those outgoing arguments to be preserved and other
2009 // constraints basically require this frame to look exactly as
2010 // though it had previously invoked an interpreted activation with
2011 // no space between the top of the expression stack (current
2012 // last_sp) and the top of stack. Rather than force deopt to
2013 // maintain this kind of invariant all the time we call a small
2014 // fixup routine to move the mutated arguments onto the top of our
2015 // expression stack if necessary.
2016 __ move(T8, SP);
2017 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2018 #ifndef OPT_THREAD
2019 __ get_thread(thread);
2020 #endif
2021 // PC must point into interpreter here
2022 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2023 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2024 __ reset_last_Java_frame(thread, true, true);
2025 // Restore the last_sp and null it out
2026 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2027 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2031 __ move(AT, JavaThread::popframe_inactive);
2032 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2034 // Finish with popframe handling
2035 __ restore_bcp();
2036 __ restore_locals();
2037 #ifndef CORE
2038 // The method data pointer was incremented already during
2039 // call profiling. We have to restore the mdp for the current bcp.
2040 if (ProfileInterpreter) {
2041 __ set_method_data_pointer_for_bcp();
2042 }
2043 #endif // !CORE
2044 // Clear the popframe condition flag
2045 #ifndef OPT_THREAD
2046 __ get_thread(thread);
2047 #endif
2048 __ move(AT, JavaThread::popframe_inactive);
2049 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2050 __ dispatch_next(vtos);
2051 // end of PopFrame support
2053 Interpreter::_remove_activation_entry = __ pc();
2055 // preserve exception over this code sequence
2056 __ ld(T0, SP, 0);
2057 __ daddi(SP, SP, wordSize);
2058 #ifndef OPT_THREAD
2059 __ get_thread(thread);
2060 #endif
2061 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2062 // remove the activation (without doing throws on illegalMonitorExceptions)
2063 __ remove_activation(vtos, T3, false, true, false);
2064 // restore exception
2065 __ get_vm_result(T0, thread);
2066 __ verify_oop(T0);
2068 // Inbetween activations - previous activation type unknown yet
2069 // compute continuation point - the continuation point expects
2070 // the following registers set up:
2071 //
2072 // T0: exception eax
2073 // T1: return address/pc that threw exception edx
2074 // SP: expression stack of caller esp
2075 // FP: ebp of caller ebp
2076 __ daddi(SP, SP, (-2) * wordSize);
2077 __ sd(T0, SP, wordSize); // save exception
2078 __ sd(T3, SP, 0); // save return address
2079 __ move(A1, T3);
2080 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2081 __ move(T9, V0); // save exception handler
2082 __ ld(V0, SP, wordSize); // restore exception
2083 __ ld(V1, SP, 0); // restore return address
2084 __ daddi(SP, SP, 2 * wordSize);
2086 // Note that an "issuing PC" is actually the next PC after the call
2087 __ jr(T9); // jump to exception handler of caller
2088 __ delayed()->nop();
2089 }
2092 //
2093 // JVMTI ForceEarlyReturn support
2094 //
2095 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2096 address entry = __ pc();
2097 __ restore_bcp();
2098 __ restore_locals();
2099 __ empty_expression_stack();
2100 __ empty_FPU_stack();
2101 __ load_earlyret_value(state);
2103 #ifndef OPT_THREAD
2104 __ get_thread(TREG);
2105 #endif
2106 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2107 //const Address cond_addr(ecx, JvmtiThreadState::earlyret_state_offset());
2108 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2109 // Clear the earlyret state
2110 __ move(AT,JvmtiThreadState::earlyret_inactive);
2111 __ sw(AT,cond_addr);
2112 __ sync();
2115 __ remove_activation(state, T0,
2116 false, /* throw_monitor_exception */
2117 false, /* install_monitor_exception */
2118 true); /* notify_jvmdi */
2119 __ sync();
2120 __ jr(T0);
2121 __ delayed()->nop();
2122 return entry;
2123 } // end of ForceEarlyReturn support
2126 //-----------------------------------------------------------------------------
2127 // Helper for vtos entry point generation
2129 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2130 address& bep,
2131 address& cep,
2132 address& sep,
2133 address& aep,
2134 address& iep,
2135 address& lep,
2136 address& fep,
2137 address& dep,
2138 address& vep) {
2139 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2140 Label L;
2141 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2142 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2143 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2144 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2145 bep = cep = sep =
2146 iep = __ pc(); __ push(itos);
2147 vep = __ pc();
2148 __ bind(L);
2149 generate_and_dispatch(t);
2150 }
2153 //-----------------------------------------------------------------------------
2154 // Generation of individual instructions
2156 // helpers for generate_and_dispatch
2159 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2160 : TemplateInterpreterGenerator(code) {
2161 generate_all(); // down here so it can be "virtual"
2162 }
2164 //-----------------------------------------------------------------------------
2166 // Non-product code
2167 #ifndef PRODUCT
2168 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2169 address entry = __ pc();
2171 // prepare expression stack
2172 __ push(state); // save tosca
2174 // tos & tos2, added by yjl 7/15/2005
2175 // trace_bytecode need actually 4 args, the last two is tos&tos2
2176 // this work fine for x86. but mips o32 call convention will store A2-A3
2177 // to the stack position it think is the tos&tos2
2178 // when the expression stack have no more than 2 data, error occur.
2179 __ ld(A2, SP, 0);
2180 __ ld(A3, SP, 1 * wordSize);
2182 // pass arguments & call tracer
2183 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2184 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2186 // restore expression stack
2187 __ pop(state); // restore tosca
2189 // return
2190 __ jr(RA);
2191 __ delayed()->nop();
2193 return entry;
2194 }
2196 void TemplateInterpreterGenerator::count_bytecode() {
2197 __ li(T8, (long)&BytecodeCounter::_counter_value);
2198 __ lw(AT, T8, 0);
2199 __ daddi(AT, AT, 1);
2200 __ sw(AT, T8, 0);
2201 }
2203 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2204 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2205 __ lw(AT, T8, 0);
2206 __ daddi(AT, AT, 1);
2207 __ sw(AT, T8, 0);
2208 }
2210 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2211 __ li(T8, (long)&BytecodePairHistogram::_index);
2212 __ lw(T9, T8, 0);
2213 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2214 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2215 __ orr(T9, T9, T8);
2216 __ li(T8, (long)&BytecodePairHistogram::_index);
2217 __ sw(T9, T8, 0);
2218 __ dsll(T9, T9, 2);
2219 __ li(T8, (long)BytecodePairHistogram::_counters);
2220 __ dadd(T8, T8, T9);
2221 __ lw(AT, T8, 0);
2222 __ daddi(AT, AT, 1);
2223 __ sw(AT, T8, 0);
2224 }
2227 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2228 // Call a little run-time stub to avoid blow-up for each bytecode.
2229 // The run-time runtime saves the right registers, depending on
2230 // the tosca in-state for the given template.
2232 address entry = Interpreter::trace_code(t->tos_in());
2233 assert(entry != NULL, "entry must have been generated");
2234 __ call(entry, relocInfo::none);
2235 __ delayed()->nop();
2236 //add for compressedoops
2237 __ reinit_heapbase();
2238 }
2241 void TemplateInterpreterGenerator::stop_interpreter_at() {
2242 Label L;
2243 __ li(T8, long(&BytecodeCounter::_counter_value));
2244 __ lw(T8, T8, 0);
2245 __ move(AT, StopInterpreterAt);
2246 __ bne(T8, AT, L);
2247 __ delayed()->nop();
2248 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2249 __ delayed()->nop();
2250 __ bind(L);
2251 }
2252 #endif // !PRODUCT
2253 #endif // ! CC_INTERP