Thu, 24 May 2018 19:32:53 +0800
[Code Reorganization] update copyright 2018
1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2018, 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, notBool, 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 //ztos
705 __ bind(notByte);
706 __ daddi(T1, T3, (-1) * ztos);
707 __ bne(T1, R0, notBool);
708 __ delayed()->nop();
709 __ lb(V0, T0, 0);
710 __ b(xreturn_path);
711 __ delayed()->nop();
713 //stos
714 __ bind(notBool);
715 __ daddi(T1, T3, (-1) * stos);
716 __ bne(T1, R0, notShort);
717 __ delayed()->nop();
718 __ lh(V0, T0, 0);
719 __ b(xreturn_path);
720 __ delayed()->nop();
722 //ctos
723 __ bind(notShort);
724 __ daddi(T1, T3, (-1) * ctos);
725 __ bne(T1, R0, notChar);
726 __ delayed()->nop();
727 __ lhu(V0, T0, 0);
728 __ b(xreturn_path);
729 __ delayed()->nop();
731 //atos
732 __ bind(notChar);
733 __ daddi(T1, T3, (-1) * atos);
734 __ bne(T1, R0, notObj);
735 __ delayed()->nop();
736 //add for compressedoops
737 __ load_heap_oop(V0, Address(T0, 0));
738 __ b(xreturn_path);
739 __ delayed()->nop();
741 //itos
742 __ bind(notObj);
743 #ifdef ASSERT
744 Label okay;
745 __ daddi(T1, T3, (-1) * itos);
746 __ beq(T1, R0, okay);
747 __ delayed()->nop();
748 __ stop("what type is this?");
749 __ bind(okay);
750 #endif // ASSERT
751 __ lw(V0, T0, 0);
753 __ bind(xreturn_path);
755 // _ireturn/_areturn
756 //FIXME
757 __ move(SP, Rsender);//FIXME, set sender's fp to SP
758 __ jr(RA);
759 __ delayed()->nop();
761 // generate a vanilla interpreter entry as the slow path
762 __ bind(slow_path);
763 (void) generate_normal_entry(false);
764 } else {
765 (void) generate_normal_entry(false);
766 }
768 return entry_point;
769 }
771 // Method entry for java.lang.ref.Reference.get.
772 address InterpreterGenerator::generate_Reference_get_entry(void) {
773 #if INCLUDE_ALL_GCS
774 // Code: _aload_0, _getfield, _areturn
775 // parameter size = 1
776 //
777 // The code that gets generated by this routine is split into 2 parts:
778 // 1. The "intrinsified" code for G1 (or any SATB based GC),
779 // 2. The slow path - which is an expansion of the regular method entry.
780 //
781 // Notes:-
782 // * In the G1 code we do not check whether we need to block for
783 // a safepoint. If G1 is enabled then we must execute the specialized
784 // code for Reference.get (except when the Reference object is null)
785 // so that we can log the value in the referent field with an SATB
786 // update buffer.
787 // If the code for the getfield template is modified so that the
788 // G1 pre-barrier code is executed when the current method is
789 // Reference.get() then going through the normal method entry
790 // will be fine.
791 // * The G1 code can, however, check the receiver object (the instance
792 // of java.lang.Reference) and jump to the slow path if null. If the
793 // Reference object is null then we obviously cannot fetch the referent
794 // and so we don't need to call the G1 pre-barrier. Thus we can use the
795 // regular method entry code to generate the NPE.
796 //
797 // This code is based on generate_accessor_enty.
798 //
799 // rbx: Method* (Rmethod)
801 // r13: senderSP must preserve for slow path, set SP to it on fast path (Rsender)
803 // rax: V0
804 // rbx: Rmethod
805 // r13: Rsender
806 // rdi: T9
808 address entry = __ pc();
810 const int referent_offset = java_lang_ref_Reference::referent_offset;
811 guarantee(referent_offset > 0, "referent offset not initialized");
813 if (UseG1GC) {
814 Label slow_path;
816 // Check if local 0 != NULL
817 // If the receiver is null then it is OK to jump to the slow path.
818 __ ld(V0, SP, 0);
820 __ beq(V0, R0, slow_path);
821 __ nop();
823 // Generate the G1 pre-barrier code to log the value of
824 // the referent field in an SATB buffer.
826 // Load the value of the referent field.
827 const Address field_address(V0, referent_offset);
828 __ load_heap_oop(V0, field_address);
830 __ push(RA);
831 // Generate the G1 pre-barrier code to log the value of
832 // the referent field in an SATB buffer.
833 __ g1_write_barrier_pre(noreg /* obj */,
834 V0 /* pre_val */,
835 TREG /* thread */,
836 Rmethod /* tmp */,
837 true /* tosca_live */,
838 true /* expand_call */);
839 __ pop(RA);
841 __ jr(RA);
842 __ delayed()->daddu(SP, Rsender, R0); // set sp to sender sp
844 // generate a vanilla interpreter entry as the slow path
845 __ bind(slow_path);
846 (void) generate_normal_entry(false);
848 return entry;
849 }
850 #endif // INCLUDE_ALL_GCS
852 // If G1 is not enabled then attempt to go through the accessor entry point
853 // Reference.get is an accessor
854 return generate_accessor_entry();
855 }
857 // Interpreter stub for calling a native method. (asm interpreter)
858 // This sets up a somewhat different looking stack for calling the
859 // native method than the typical interpreter frame setup.
860 address InterpreterGenerator::generate_native_entry(bool synchronized) {
861 // determine code generation flags
862 bool inc_counter = UseCompiler || CountCompiledCalls;
863 // Rsender: sender's sp
864 // Rmethod: Method*
865 address entry_point = __ pc();
867 #ifndef CORE
868 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() + // Fu: 20130814
869 InvocationCounter::counter_offset()));
870 #endif
872 // get parameter size (always needed)
873 // the size in the java stack
874 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
875 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
877 // native calls don't need the stack size check since they have no expression stack
878 // and the arguments are already on the stack and we only add a handful of words
879 // to the stack
881 // Rmethod: Method*
882 // V0: size of parameters
883 // Layout of frame at this point
884 //
885 // [ argument word n-1 ] <--- sp
886 // ...
887 // [ argument word 0 ]
889 // for natives the size of locals is zero
891 // compute beginning of parameters (S7)
892 __ dsll(LVP, V0, Address::times_8);
893 __ daddiu(LVP, LVP, (-1) * wordSize);
894 __ dadd(LVP, LVP, SP);
897 // add 2 zero-initialized slots for native calls
898 __ daddi(SP, SP, (-2) * wordSize);
899 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
900 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
902 // Layout of frame at this point
903 // [ method holder mirror ] <--- sp
904 // [ result type info ]
905 // [ argument word n-1 ] <--- T0
906 // ...
907 // [ argument word 0 ] <--- LVP
910 #ifndef CORE
911 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
912 #endif
914 // initialize fixed part of activation frame
915 generate_fixed_frame(true);
916 // after this function, the layout of frame is as following
917 //
918 // [ monitor block top ] <--- sp ( the top monitor entry )
919 // [ byte code pointer (0) ] (if native, bcp = 0)
920 // [ constant pool cache ]
921 // [ Method* ]
922 // [ locals offset ]
923 // [ sender's sp ]
924 // [ sender's fp ]
925 // [ return address ] <--- fp
926 // [ method holder mirror ]
927 // [ result type info ]
928 // [ argumnet word n-1 ] <--- sender's sp
929 // ...
930 // [ argument word 0 ] <--- S7
933 // make sure method is native & not abstract
934 #ifdef ASSERT
935 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
936 {
937 Label L;
938 __ andi(AT, T0, JVM_ACC_NATIVE);
939 __ bne(AT, R0, L);
940 __ delayed()->nop();
941 __ stop("tried to execute native method as non-native");
942 __ bind(L);
943 }
944 {
945 Label L;
946 __ andi(AT, T0, JVM_ACC_ABSTRACT);
947 __ beq(AT, R0, L);
948 __ delayed()->nop();
949 __ stop("tried to execute abstract method in interpreter");
950 __ bind(L);
951 }
952 #endif
954 // Since at this point in the method invocation the exception handler
955 // would try to exit the monitor of synchronized methods which hasn't
956 // been entered yet, we set the thread local variable
957 // _do_not_unlock_if_synchronized to true. The remove_activation will
958 // check this flag.
959 Register thread = TREG;
960 #ifndef OPT_THREAD
961 __ get_thread(thread);
962 #endif
963 __ move(AT, (int)true);
964 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
966 #ifndef CORE
967 // increment invocation count & check for overflow
968 Label invocation_counter_overflow;
969 if (inc_counter) {
970 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
971 }
973 Label continue_after_compile;
974 __ bind(continue_after_compile);
975 #endif // CORE
977 bang_stack_shadow_pages(true);
979 // reset the _do_not_unlock_if_synchronized flag
980 #ifndef OPT_THREAD
981 __ get_thread(thread);
982 #endif
983 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
985 // check for synchronized methods
986 // Must happen AFTER invocation_counter check and stack overflow check,
987 // so method is not locked if overflows.
988 if (synchronized) {
989 lock_method();
990 } else {
991 // no synchronization necessary
992 #ifdef ASSERT
993 {
994 Label L;
995 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
996 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
997 __ beq(AT, R0, L);
998 __ delayed()->nop();
999 __ stop("method needs synchronization");
1000 __ bind(L);
1001 }
1002 #endif
1003 }
1005 // after method_lock, the layout of frame is as following
1006 //
1007 // [ monitor entry ] <--- sp
1008 // ...
1009 // [ monitor entry ]
1010 // [ monitor block top ] ( the top monitor entry )
1011 // [ byte code pointer (0) ] (if native, bcp = 0)
1012 // [ constant pool cache ]
1013 // [ Method* ]
1014 // [ locals offset ]
1015 // [ sender's sp ]
1016 // [ sender's fp ]
1017 // [ return address ] <--- fp
1018 // [ method holder mirror ]
1019 // [ result type info ]
1020 // [ argumnet word n-1 ] <--- ( sender's sp )
1021 // ...
1022 // [ argument word 0 ] <--- S7
1024 // start execution
1025 #ifdef ASSERT
1026 {
1027 Label L;
1028 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1029 __ beq(AT, SP, L);
1030 __ delayed()->nop();
1031 __ stop("broken stack frame setup in interpreter in asm");
1032 __ bind(L);
1033 }
1034 #endif
1036 // jvmti/jvmpi support
1037 __ notify_method_entry();
1039 // work registers
1040 const Register method = Rmethod;
1041 //const Register thread = T2;
1042 const Register t = RT4;
1044 __ get_method(method);
1045 __ verify_oop(method);
1046 {
1047 Label L, Lstatic;
1048 __ ld(t,method,in_bytes(Method::const_offset()));
1049 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset())); // Fu: 20130814
1050 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1051 //FIXME, aoqi: A1?
1052 // A0 and A1(if needed)
1053 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1054 __ andi(AT, AT, JVM_ACC_STATIC);
1055 __ beq(AT, R0, Lstatic);
1056 __ delayed()->nop();
1057 __ daddiu(t, t, 1);
1058 __ bind(Lstatic);
1059 __ daddiu(t, t, -7);
1060 __ blez(t, L);
1061 __ delayed()->nop();
1062 __ dsll(t, t, Address::times_8);
1063 __ dsub(SP, SP, t);
1064 __ bind(L);
1065 }
1066 __ move(AT, -(StackAlignmentInBytes));
1067 __ andr(SP, SP, AT);
1068 __ move(AT, SP);
1069 // [ ] <--- sp
1070 // ... (size of parameters - 8 )
1071 // [ monitor entry ]
1072 // ...
1073 // [ monitor entry ]
1074 // [ monitor block top ] ( the top monitor entry )
1075 // [ byte code pointer (0) ] (if native, bcp = 0)
1076 // [ constant pool cache ]
1077 // [ Method* ]
1078 // [ locals offset ]
1079 // [ sender's sp ]
1080 // [ sender's fp ]
1081 // [ return address ] <--- fp
1082 // [ method holder mirror ]
1083 // [ result type info ]
1084 // [ argumnet word n-1 ] <--- ( sender's sp )
1085 // ...
1086 // [ argument word 0 ] <--- LVP
1088 // get signature handler
1089 {
1090 Label L;
1091 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1092 __ bne(T9, R0, L);
1093 __ delayed()->nop();
1094 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1095 InterpreterRuntime::prepare_native_call), method);
1096 __ get_method(method);
1097 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1098 __ bind(L);
1099 }
1101 // call signature handler
1102 // FIXME: when change codes in InterpreterRuntime, note this point
1103 // from: begin of parameters
1104 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1105 // to: current sp
1106 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1107 // temp: T3
1108 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1110 __ jalr(T9);
1111 __ delayed()->nop();
1112 __ get_method(method); // slow path call blows EBX on DevStudio 5.0
1114 /*
1115 if native function is static, and its second parameter has type length of double word,
1116 and first parameter has type length of word, we have to reserve one word
1117 for the first parameter, according to mips o32 abi.
1118 if native function is not static, and its third parameter has type length of double word,
1119 and second parameter has type length of word, we have to reserve one word for the second
1120 parameter.
1121 */
1124 // result handler is in V0
1125 // set result handler
1126 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1128 #define FIRSTPARA_SHIFT_COUNT 5
1129 #define SECONDPARA_SHIFT_COUNT 9
1130 #define THIRDPARA_SHIFT_COUNT 13
1131 #define PARA_MASK 0xf
1133 // pass mirror handle if static call
1134 {
1135 Label L;
1136 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1137 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1138 __ andi(AT, t, JVM_ACC_STATIC);
1139 __ beq(AT, R0, L);
1140 __ delayed()->nop();
1142 // get mirror
1143 __ ld(t, method, in_bytes(Method:: const_offset()));
1144 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1145 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1146 __ ld(t, t, mirror_offset);
1147 // copy mirror into activation frame
1148 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1149 // pass handle to mirror
1150 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1151 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1152 __ move(A1, t);
1153 __ bind(L);
1154 }
1156 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1157 // [ ] |
1158 // ... size of parameters(or +1) |
1159 // [ monitor entry ] |
1160 // ... |
1161 // [ monitor entry ] |
1162 // [ monitor block top ] ( the top monitor entry ) |
1163 // [ byte code pointer (0) ] (if native, bcp = 0) |
1164 // [ constant pool cache ] |
1165 // [ Method* ] |
1166 // [ locals offset ] |
1167 // [ sender's sp ] |
1168 // [ sender's fp ] |
1169 // [ return address ] <--- fp |
1170 // [ method holder mirror ] <----------------------------|
1171 // [ result type info ]
1172 // [ argumnet word n-1 ] <--- ( sender's sp )
1173 // ...
1174 // [ argument word 0 ] <--- S7
1176 // get native function entry point
1177 { Label L;
1178 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1179 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1180 __ bne(V1, T9, L);
1181 __ delayed()->nop();
1182 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1183 __ get_method(method);
1184 __ verify_oop(method);
1185 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1186 __ bind(L);
1187 }
1188 /*
1189 __ pushad();
1190 __ move(A0, T9);
1191 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::func_debug),relocInfo::runtime_call_type);
1192 __ popad();
1193 */
1195 // pass JNIEnv
1196 // native function in T9
1197 #ifndef OPT_THREAD
1198 __ get_thread(thread);
1199 #endif
1200 __ daddi(t, thread, in_bytes(JavaThread::jni_environment_offset()));
1201 // stack,but I think it won't work when pass float,double etc @jerome,10/17,2006
1202 __ move(A0, t);
1203 // [ jni environment ] <--- sp
1204 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1205 // [ ] |
1206 // ... size of parameters |
1207 // [ monitor entry ] |
1208 // ... |
1209 // [ monitor entry ] |
1210 // [ monitor block top ] ( the top monitor entry ) |
1211 // [ byte code pointer (0) ] (if native, bcp = 0) |
1212 // [ constant pool cache ] |
1213 // [ Method* ] |
1214 // [ locals offset ] |
1215 // [ sender's sp ] |
1216 // [ sender's fp ] |
1217 // [ return address ] <--- fp |
1218 // [ method holder mirror ] <----------------------------|
1219 // [ result type info ]
1220 // [ argumnet word n-1 ] <--- ( sender's sp )
1221 // ...
1222 // [ argument word 0 ] <--- S7
1224 // set_last_Java_frame_before_call
1225 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1226 // Change state to native (we save the return address in the thread, since it might not
1227 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1228 // points into the right code segment. It does not have to be the correct return pc.
1229 __ li(t, __ pc());
1230 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1231 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1233 // change thread state
1234 #ifdef ASSERT
1235 {
1236 Label L;
1237 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1238 __ daddi(t, t, (-1) * _thread_in_Java);
1239 __ beq(t, R0, L);
1240 __ delayed()->nop();
1241 __ stop("Wrong thread state in native stub");
1242 __ bind(L);
1243 }
1244 #endif
1246 __ move(t, _thread_in_native);
1247 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1249 // call native method
1250 __ jalr(T9);
1251 __ delayed()->nop();
1252 // result potentially in V2:V1 or F0:F1
1255 // via _last_native_pc and not via _last_jave_sp
1256 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1257 // If the order changes or anything else is added to the stack the code in
1258 // interpreter_frame_result will have to be changed.
1259 //FIXME, should modify here
1260 // save return value to keep the value from being destroyed by other calls
1261 __ move(S1, V0);
1262 __ move(S3, V1);
1263 __ dmfc1(S4, F0);
1264 __ dmfc1(S2, F1);
1266 // change thread state
1267 __ get_thread(thread);
1268 __ move(t, _thread_in_native_trans);
1269 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1271 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1273 // check for safepoint operation in progress and/or pending suspend requests
1274 { Label Continue;
1276 // Don't use call_VM as it will see a possible pending exception and forward it
1277 // and never return here preventing us from clearing _last_native_pc down below.
1278 // Also can't use call_VM_leaf either as it will check to see if esi & edi are
1279 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1280 // by hand.
1281 //
1282 Label L;
1283 __ li(AT, SafepointSynchronize::address_of_state());
1284 __ lw(AT, AT, 0);
1285 __ bne(AT, R0, L);
1286 __ delayed()->nop();
1287 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1288 __ beq(AT, R0, Continue);
1289 __ delayed()->nop();
1290 __ bind(L);
1291 __ move(A0, thread);
1292 __ call(CAST_FROM_FN_PTR(address,
1293 JavaThread::check_special_condition_for_native_trans),
1294 relocInfo::runtime_call_type);
1295 __ delayed()->nop();
1297 #ifndef OPT_THREAD
1298 __ get_thread(thread);
1299 #endif
1300 //add for compressedoops
1301 __ reinit_heapbase();
1302 __ bind(Continue);
1303 }
1305 // change thread state
1306 __ move(t, _thread_in_Java);
1307 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1308 __ reset_last_Java_frame(thread, true, true);
1310 // reset handle block
1311 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1312 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1314 // If result was an oop then unbox and save it in the frame
1315 { Label L;
1316 Label no_oop, store_result;
1317 //FIXME, addi only support 16-bit imeditate
1318 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1319 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1320 __ bne(AT, T0, no_oop);
1321 __ delayed()->nop();
1322 __ move(V0, S1);
1323 __ beq(V0, R0, store_result);
1324 __ delayed()->nop();
1325 // unbox
1326 __ ld(V0, V0, 0);
1327 __ bind(store_result);
1328 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1329 // keep stack depth as expected by pushing oop which will eventually be discarded
1330 __ bind(no_oop);
1331 }
1332 {
1333 Label no_reguard;
1334 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1335 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1336 __ bne(t, AT, no_reguard);
1337 __ delayed()->nop();
1338 __ pushad();
1339 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::runtime_call_type);
1340 __ delayed()->nop();
1341 __ popad();
1342 //add for compressedoops
1343 __ reinit_heapbase();
1344 __ bind(no_reguard);
1345 }
1346 // restore esi to have legal interpreter frame,
1347 // i.e., bci == 0 <=> esi == code_base()
1348 // Can't call_VM until bcp is within reasonable.
1349 __ get_method(method); // method is junk from thread_in_native to now.
1350 __ verify_oop(method);
1351 __ ld(BCP, method, in_bytes(Method::const_offset()));
1352 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1353 // handle exceptions (exception handling will handle unlocking!)
1354 {
1355 Label L;
1356 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1357 __ beq(t, R0, L);
1358 __ delayed()->nop();
1359 // Note: At some point we may want to unify this with the code used in
1360 // call_VM_base();
1361 // i.e., we should use the StubRoutines::forward_exception code. For now this
1362 // doesn't work here because the esp is not correctly set at this point.
1363 __ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address,
1364 InterpreterRuntime::throw_pending_exception));
1365 __ should_not_reach_here();
1366 __ bind(L);
1367 }
1369 // do unlocking if necessary
1370 {
1371 Label L;
1372 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1373 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1374 __ beq(t, R0, L);
1375 // the code below should be shared with interpreter macro assembler implementation
1376 {
1377 Label unlock;
1378 // BasicObjectLock will be first in list,
1379 // since this is a synchronized method. However, need
1380 // to check that the object has not been unlocked by
1381 // an explicit monitorexit bytecode.
1382 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1383 // address of first monitor
1385 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1386 __ bne(t, R0, unlock);
1387 __ delayed()->nop();
1389 // Entry already unlocked, need to throw exception
1390 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1391 InterpreterRuntime::throw_illegal_monitor_state_exception));
1392 __ should_not_reach_here();
1394 __ bind(unlock);
1395 __ unlock_object(c_rarg0);
1396 }
1397 __ bind(L);
1398 }
1400 // jvmti/jvmpi support
1401 // Note: This must happen _after_ handling/throwing any exceptions since
1402 // the exception handler code notifies the runtime of method exits
1403 // too. If this happens before, method entry/exit notifications are
1404 // not properly paired (was bug - gri 11/22/99).
1405 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI );
1407 // restore potential result in V0:V1,
1408 // call result handler to restore potential result in ST0 & handle result
1409 //__ lw(V0, SP, 3 * wordSize);
1410 //__ lw(V1, SP, 2 * wordSize);
1411 //__ lwc1(F0, SP, 1 * wordSize);
1412 //__ lwc1(F1, SP, 0 * wordSize);
1413 //__ addi(SP, SP, 4 * wordSize);
1414 __ move(V0, S1);
1415 __ move(V1, S3);
1416 __ dmtc1(S4, F0);
1417 __ dmtc1(S2, F1);
1418 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1419 __ jalr(t);
1420 __ delayed()->nop();
1423 // remove activation
1424 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1425 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1426 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1427 __ jr(RA);
1428 __ delayed()->nop();
1430 #ifndef CORE
1431 if (inc_counter) {
1432 // Handle overflow of counter and compile method
1433 __ bind(invocation_counter_overflow);
1434 generate_counter_overflow(&continue_after_compile);
1435 // entry_point is the beginning of this
1436 // function and checks again for compiled code
1437 }
1438 #endif
1439 return entry_point;
1440 }
1442 //
1443 // Generic interpreted method entry to (asm) interpreter
1444 //
1445 // Layout of frame just at the entry
1446 //
1447 // [ argument word n-1 ] <--- sp
1448 // ...
1449 // [ argument word 0 ]
1450 // assume Method* in Rmethod before call this method.
1451 // prerequisites to the generated stub : the callee Method* in Rmethod
1452 // note you must save the caller bcp before call the generated stub
1453 //
1454 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1455 // determine code generation flags
1456 bool inc_counter = UseCompiler || CountCompiledCalls;
1458 // Rmethod: Method*
1459 // Rsender: sender 's sp
1460 address entry_point = __ pc();
1462 const Address invocation_counter(Rmethod,
1463 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1465 // get parameter size (always needed)
1466 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1467 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1469 // Rmethod: Method*
1470 // V0: size of parameters
1471 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1472 // get size of locals in words to T2
1473 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1474 // T2 = no. of additional locals, locals include parameters
1475 __ dsub(T2, T2, V0);
1477 // see if we've got enough room on the stack for locals plus overhead.
1478 // Layout of frame at this point
1479 //
1480 // [ argument word n-1 ] <--- sp
1481 // ...
1482 // [ argument word 0 ]
1483 generate_stack_overflow_check();
1484 // after this function, the layout of frame does not change
1486 // compute beginning of parameters (LVP)
1487 __ dsll(LVP, V0, LogBytesPerWord);
1488 __ daddiu(LVP, LVP, (-1) * wordSize);
1489 __ dadd(LVP, LVP, SP);
1491 // T2 - # of additional locals
1492 // allocate space for locals
1493 // explicitly initialize locals
1494 {
1495 Label exit, loop;
1496 __ beq(T2, R0, exit);
1497 __ delayed()->nop();
1499 __ bind(loop);
1500 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1501 __ daddiu(T2, T2, -1); // until everything initialized
1502 __ bne(T2, R0, loop);
1503 __ delayed();
1505 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1507 __ bind(exit);
1508 }
1510 //
1511 // [ local var m-1 ] <--- sp
1512 // ...
1513 // [ local var 0 ]
1514 // [ argument word n-1 ] <--- T0?
1515 // ...
1516 // [ argument word 0 ] <--- LVP
1518 // initialize fixed part of activation frame
1520 generate_fixed_frame(false);
1523 // after this function, the layout of frame is as following
1524 //
1525 // [ monitor block top ] <--- sp ( the top monitor entry )
1526 // [ byte code pointer ] (if native, bcp = 0)
1527 // [ constant pool cache ]
1528 // [ Method* ]
1529 // [ locals offset ]
1530 // [ sender's sp ]
1531 // [ sender's fp ] <--- fp
1532 // [ return address ]
1533 // [ local var m-1 ]
1534 // ...
1535 // [ local var 0 ]
1536 // [ argumnet word n-1 ] <--- ( sender's sp )
1537 // ...
1538 // [ argument word 0 ] <--- LVP
1541 // make sure method is not native & not abstract
1542 #ifdef ASSERT
1543 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1544 {
1545 Label L;
1546 __ andi(T2, AT, JVM_ACC_NATIVE);
1547 __ beq(T2, R0, L);
1548 __ delayed()->nop();
1549 __ stop("tried to execute native method as non-native");
1550 __ bind(L);
1551 }
1552 {
1553 Label L;
1554 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1555 __ beq(T2, R0, L);
1556 __ delayed()->nop();
1557 __ stop("tried to execute abstract method in interpreter");
1558 __ bind(L);
1559 }
1560 #endif
1562 // Since at this point in the method invocation the exception handler
1563 // would try to exit the monitor of synchronized methods which hasn't
1564 // been entered yet, we set the thread local variable
1565 // _do_not_unlock_if_synchronized to true. The remove_activation will
1566 // check this flag.
1568 #ifndef OPT_THREAD
1569 Register thread = T8;
1570 __ get_thread(thread);
1571 #else
1572 Register thread = TREG;
1573 #endif
1574 __ move(AT, (int)true);
1575 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1577 #ifndef CORE
1579 // 2014/11/24 Fu
1580 // mdp : T8
1581 // tmp1: T9
1582 // tmp2: T2
1583 __ profile_parameters_type(T8, T9, T2);
1585 // increment invocation count & check for overflow
1586 Label invocation_counter_overflow;
1587 Label profile_method;
1588 Label profile_method_continue;
1589 if (inc_counter) {
1590 generate_counter_incr(&invocation_counter_overflow,
1591 &profile_method,
1592 &profile_method_continue);
1593 if (ProfileInterpreter) {
1594 __ bind(profile_method_continue);
1595 }
1596 }
1598 Label continue_after_compile;
1599 __ bind(continue_after_compile);
1601 #endif // CORE
1603 bang_stack_shadow_pages(false);
1605 // reset the _do_not_unlock_if_synchronized flag
1606 #ifndef OPT_THREAD
1607 __ get_thread(thread);
1608 #endif
1609 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1611 // check for synchronized methods
1612 // Must happen AFTER invocation_counter check and stack overflow check,
1613 // so method is not locked if overflows.
1614 //
1615 if (synchronized) {
1616 // Allocate monitor and lock method
1617 lock_method();
1618 } else {
1619 // no synchronization necessary
1620 #ifdef ASSERT
1621 { Label L;
1622 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1623 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1624 __ beq(T2, R0, L);
1625 __ delayed()->nop();
1626 __ stop("method needs synchronization");
1627 __ bind(L);
1628 }
1629 #endif
1630 }
1632 // layout of frame after lock_method
1633 // [ monitor entry ] <--- sp
1634 // ...
1635 // [ monitor entry ]
1636 // [ monitor block top ] ( the top monitor entry )
1637 // [ byte code pointer ] (if native, bcp = 0)
1638 // [ constant pool cache ]
1639 // [ Method* ]
1640 // [ locals offset ]
1641 // [ sender's sp ]
1642 // [ sender's fp ]
1643 // [ return address ] <--- fp
1644 // [ local var m-1 ]
1645 // ...
1646 // [ local var 0 ]
1647 // [ argumnet word n-1 ] <--- ( sender's sp )
1648 // ...
1649 // [ argument word 0 ] <--- LVP
1652 // start execution
1653 #ifdef ASSERT
1654 {
1655 Label L;
1656 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1657 __ beq(AT, SP, L);
1658 __ delayed()->nop();
1659 __ stop("broken stack frame setup in interpreter in native");
1660 __ bind(L);
1661 }
1662 #endif
1664 // jvmti/jvmpi support
1665 __ notify_method_entry();
1667 __ dispatch_next(vtos);
1669 // invocation counter overflow
1670 if (inc_counter) {
1671 if (ProfileInterpreter) {
1672 // We have decided to profile this method in the interpreter
1673 __ bind(profile_method);
1674 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1675 InterpreterRuntime::profile_method));
1676 __ set_method_data_pointer_for_bcp();
1677 __ get_method(Rmethod);
1678 __ b(profile_method_continue);
1679 __ delayed()->nop();
1680 }
1681 // Handle overflow of counter and compile method
1682 __ bind(invocation_counter_overflow);
1683 generate_counter_overflow(&continue_after_compile);
1684 }
1686 return entry_point;
1687 }
1689 // Entry points
1690 //
1691 // Here we generate the various kind of entries into the interpreter.
1692 // The two main entry type are generic bytecode methods and native
1693 // call method. These both come in synchronized and non-synchronized
1694 // versions but the frame layout they create is very similar. The
1695 // other method entry types are really just special purpose entries
1696 // that are really entry and interpretation all in one. These are for
1697 // trivial methods like accessor, empty, or special math methods.
1698 //
1699 // When control flow reaches any of the entry types for the interpreter
1700 // the following holds ->
1701 //
1702 // Arguments:
1703 //
1704 // Rmethod: Method*
1705 // V0: receiver
1706 //
1707 //
1708 // Stack layout immediately at entry
1709 //
1710 // [ parameter n-1 ] <--- sp
1711 // ...
1712 // [ parameter 0 ]
1713 // [ expression stack ] (caller's java expression stack)
1715 // Assuming that we don't go to one of the trivial specialized entries
1716 // the stack will look like below when we are ready to execute the
1717 // first bytecode (or call the native routine). The register usage
1718 // will be as the template based interpreter expects (see
1719 // interpreter_amd64.hpp).
1720 //
1721 // local variables follow incoming parameters immediately; i.e.
1722 // the return address is moved to the end of the locals).
1723 //
1724 // [ monitor entry ] <--- sp
1725 // ...
1726 // [ monitor entry ]
1727 // [ monitor block top ] ( the top monitor entry )
1728 // [ byte code pointer ] (if native, bcp = 0)
1729 // [ constant pool cache ]
1730 // [ Method* ]
1731 // [ locals offset ]
1732 // [ sender's sp ]
1733 // [ sender's fp ]
1734 // [ return address ] <--- fp
1735 // [ local var m-1 ]
1736 // ...
1737 // [ local var 0 ]
1738 // [ argumnet word n-1 ] <--- ( sender's sp )
1739 // ...
1740 // [ argument word 0 ] <--- S7
1742 address AbstractInterpreterGenerator::generate_method_entry(
1743 AbstractInterpreter::MethodKind kind) {
1744 // determine code generation flags
1745 bool synchronized = false;
1746 address entry_point = NULL;
1747 switch (kind) {
1748 case Interpreter::zerolocals :
1749 break;
1750 case Interpreter::zerolocals_synchronized:
1751 synchronized = true;
1752 break;
1753 case Interpreter::native :
1754 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1755 break;
1756 case Interpreter::native_synchronized :
1757 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1758 break;
1759 case Interpreter::empty :
1760 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1761 break;
1762 case Interpreter::accessor :
1763 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1764 break;
1765 case Interpreter::abstract :
1766 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1767 break;
1769 case Interpreter::java_lang_math_sin : // fall thru
1770 case Interpreter::java_lang_math_cos : // fall thru
1771 case Interpreter::java_lang_math_tan : // fall thru
1772 case Interpreter::java_lang_math_log : // fall thru
1773 case Interpreter::java_lang_math_log10 : // fall thru
1774 case Interpreter::java_lang_math_pow : // fall thru
1775 case Interpreter::java_lang_math_exp : break;
1776 case Interpreter::java_lang_math_abs : // fall thru
1777 case Interpreter::java_lang_math_sqrt :
1778 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1779 case Interpreter::java_lang_ref_reference_get:
1780 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1781 default:
1782 fatal(err_msg("unexpected method kind: %d", kind));
1783 break;
1784 }
1785 if (entry_point) return entry_point;
1787 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1788 }
1790 // These should never be compiled since the interpreter will prefer
1791 // the compiled version to the intrinsic version.
1792 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1793 switch (method_kind(m)) {
1794 case Interpreter::java_lang_math_sin : // fall thru
1795 case Interpreter::java_lang_math_cos : // fall thru
1796 case Interpreter::java_lang_math_tan : // fall thru
1797 case Interpreter::java_lang_math_abs : // fall thru
1798 case Interpreter::java_lang_math_log : // fall thru
1799 case Interpreter::java_lang_math_log10 : // fall thru
1800 case Interpreter::java_lang_math_sqrt : // fall thru
1801 case Interpreter::java_lang_math_pow : // fall thru
1802 case Interpreter::java_lang_math_exp :
1803 return false;
1804 default:
1805 return true;
1806 }
1807 }
1809 // How much stack a method activation needs in words.
1810 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1812 const int entry_size = frame::interpreter_frame_monitor_size();
1814 // total overhead size: entry_size + (saved ebp thru expr stack bottom).
1815 // be sure to change this if you add/subtract anything to/from the overhead area
1816 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1818 const int stub_code = 6; // see generate_call_stub
1819 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1820 const int method_stack = (method->max_locals() + method->max_stack()) *
1821 Interpreter::stackElementWords;
1822 return overhead_size + method_stack + stub_code;
1823 }
1825 void AbstractInterpreter::layout_activation(Method* method,
1826 int tempcount,
1827 int popframe_extra_args,
1828 int moncount,
1829 int caller_actual_parameters,
1830 int callee_param_count,
1831 int callee_locals,
1832 frame* caller,
1833 frame* interpreter_frame,
1834 bool is_top_frame,
1835 bool is_bottom_frame) {
1836 // Note: This calculation must exactly parallel the frame setup
1837 // in AbstractInterpreterGenerator::generate_method_entry.
1838 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1839 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1840 // right size, as determined by a previous call to this method.
1841 // It is also guaranteed to be walkable even though it is in a skeletal state
1843 // fixed size of an interpreter frame:
1845 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1846 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1848 #ifdef ASSERT
1849 if (!EnableInvokeDynamic) {
1850 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1851 // Probably, since deoptimization doesn't work yet.
1852 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1853 }
1854 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1855 #endif
1857 interpreter_frame->interpreter_frame_set_method(method);
1858 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1859 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1860 // and sender_sp is fp+8
1861 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1863 #ifdef ASSERT
1864 if (caller->is_interpreted_frame()) {
1865 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1866 }
1867 #endif
1869 interpreter_frame->interpreter_frame_set_locals(locals);
1870 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1871 BasicObjectLock* monbot = montop - moncount;
1872 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1874 //set last sp;
1875 intptr_t* esp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1876 popframe_extra_args;
1877 interpreter_frame->interpreter_frame_set_last_sp(esp);
1878 // All frames but the initial interpreter frame we fill in have a
1879 // value for sender_sp that allows walking the stack but isn't
1880 // truly correct. Correct the value here.
1881 //
1882 if (extra_locals != 0 &&
1883 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1884 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1885 }
1886 *interpreter_frame->interpreter_frame_cache_addr() = method->constants()->cache();
1887 }
1889 //-----------------------------------------------------------------------------
1890 // Exceptions
1892 void TemplateInterpreterGenerator::generate_throw_exception() {
1893 // Entry point in previous activation (i.e., if the caller was
1894 // interpreted)
1895 Interpreter::_rethrow_exception_entry = __ pc();
1896 // Restore sp to interpreter_frame_last_sp even though we are going
1897 // to empty the expression stack for the exception processing.
1898 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
1900 // V0: exception
1901 // V1: return address/pc that threw exception
1902 __ restore_bcp(); // esi points to call/send
1903 __ restore_locals();
1905 //add for compressedoops
1906 __ reinit_heapbase();
1907 // Entry point for exceptions thrown within interpreter code
1908 Interpreter::_throw_exception_entry = __ pc();
1909 // expression stack is undefined here
1910 // V0: exception
1911 // BCP: exception bcp
1912 __ verify_oop(V0);
1914 // expression stack must be empty before entering the VM in case of an exception
1915 __ empty_expression_stack();
1916 // find exception handler address and preserve exception oop
1917 __ move(A1, V0);
1918 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
1919 // V0: exception handler entry point
1920 // V1: preserved exception oop
1921 // S0: bcp for exception handler
1922 __ daddi(SP, SP, (-1) * wordSize);
1923 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
1924 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
1925 __ delayed()->nop();
1927 // If the exception is not handled in the current frame the frame is removed and
1928 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1929 //
1930 // Note: At this point the bci is still the bxi for the instruction which caused
1931 // the exception and the expression stack is empty. Thus, for any VM calls
1932 // at this point, GC will find a legal oop map (with empty expression stack).
1934 // In current activation
1935 // V0: exception
1936 // BCP: exception bcp
1938 //
1939 // JVMTI PopFrame support
1940 //
1942 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1943 __ empty_expression_stack();
1944 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1945 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1946 // popframe handling cycles.
1947 #ifndef OPT_THREAD
1948 Register thread = T2;
1949 __ get_thread(T2);
1950 #else
1951 Register thread = TREG;
1952 #endif
1953 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1954 __ ori(T3, T3, JavaThread::popframe_processing_bit);
1955 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1957 #ifndef CORE
1958 {
1959 // Check to see whether we are returning to a deoptimized frame.
1960 // (The PopFrame call ensures that the caller of the popped frame is
1961 // either interpreted or compiled and deoptimizes it if compiled.)
1962 // In this case, we can't call dispatch_next() after the frame is
1963 // popped, but instead must save the incoming arguments and restore
1964 // them after deoptimization has occurred.
1965 //
1966 // Note that we don't compare the return PC against the
1967 // deoptimization blob's unpack entry because of the presence of
1968 // adapter frames in C2.
1969 Label caller_not_deoptimized;
1970 __ ld(A0, FP, frame::return_addr_offset * wordSize);
1971 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
1972 __ bne(V0, R0, caller_not_deoptimized);
1973 __ delayed()->nop();
1975 // Compute size of arguments for saving when returning to deoptimized caller
1976 __ get_method(A1);
1977 __ verify_oop(A1);
1978 __ ld(A1,A1,in_bytes(Method::const_offset()));
1979 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
1980 __ shl(A1, Interpreter::logStackElementSize);
1981 __ restore_locals();
1982 __ dsub(A2, LVP, T0);
1983 __ daddiu(A2, A2, wordSize);
1984 // Save these arguments
1985 #ifndef OPT_THREAD
1986 __ get_thread(A0);
1987 #else
1988 __ move(A0, TREG);
1989 #endif
1990 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
1992 __ remove_activation(vtos, T9, false, false, false);
1994 // Inform deoptimization that it is responsible for restoring these arguments
1995 #ifndef OPT_THREAD
1996 __ get_thread(thread);
1997 #endif
1998 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
1999 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2000 // Continue in deoptimization handler
2001 __ jr(T9);
2002 __ delayed()->nop();
2004 __ bind(caller_not_deoptimized);
2005 }
2006 #endif /* !CORE */
2008 __ remove_activation(vtos, T3,
2009 /* throw_monitor_exception */ false,
2010 /* install_monitor_exception */ false,
2011 /* notify_jvmdi */ false);
2013 // Clear the popframe condition flag
2014 // Finish with popframe handling
2015 // A previous I2C followed by a deoptimization might have moved the
2016 // outgoing arguments further up the stack. PopFrame expects the
2017 // mutations to those outgoing arguments to be preserved and other
2018 // constraints basically require this frame to look exactly as
2019 // though it had previously invoked an interpreted activation with
2020 // no space between the top of the expression stack (current
2021 // last_sp) and the top of stack. Rather than force deopt to
2022 // maintain this kind of invariant all the time we call a small
2023 // fixup routine to move the mutated arguments onto the top of our
2024 // expression stack if necessary.
2025 __ move(T8, SP);
2026 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2027 #ifndef OPT_THREAD
2028 __ get_thread(thread);
2029 #endif
2030 // PC must point into interpreter here
2031 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2032 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2033 __ reset_last_Java_frame(thread, true, true);
2034 // Restore the last_sp and null it out
2035 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2036 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2040 __ move(AT, JavaThread::popframe_inactive);
2041 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2043 // Finish with popframe handling
2044 __ restore_bcp();
2045 __ restore_locals();
2046 #ifndef CORE
2047 // The method data pointer was incremented already during
2048 // call profiling. We have to restore the mdp for the current bcp.
2049 if (ProfileInterpreter) {
2050 __ set_method_data_pointer_for_bcp();
2051 }
2052 #endif // !CORE
2053 // Clear the popframe condition flag
2054 #ifndef OPT_THREAD
2055 __ get_thread(thread);
2056 #endif
2057 __ move(AT, JavaThread::popframe_inactive);
2058 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2059 __ dispatch_next(vtos);
2060 // end of PopFrame support
2062 Interpreter::_remove_activation_entry = __ pc();
2064 // preserve exception over this code sequence
2065 __ ld(T0, SP, 0);
2066 __ daddi(SP, SP, wordSize);
2067 #ifndef OPT_THREAD
2068 __ get_thread(thread);
2069 #endif
2070 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2071 // remove the activation (without doing throws on illegalMonitorExceptions)
2072 __ remove_activation(vtos, T3, false, true, false);
2073 // restore exception
2074 __ get_vm_result(T0, thread);
2075 __ verify_oop(T0);
2077 // Inbetween activations - previous activation type unknown yet
2078 // compute continuation point - the continuation point expects
2079 // the following registers set up:
2080 //
2081 // T0: exception eax
2082 // T1: return address/pc that threw exception edx
2083 // SP: expression stack of caller esp
2084 // FP: ebp of caller ebp
2085 __ daddi(SP, SP, (-2) * wordSize);
2086 __ sd(T0, SP, wordSize); // save exception
2087 __ sd(T3, SP, 0); // save return address
2088 __ move(A1, T3);
2089 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2090 __ move(T9, V0); // save exception handler
2091 __ ld(V0, SP, wordSize); // restore exception
2092 __ ld(V1, SP, 0); // restore return address
2093 __ daddi(SP, SP, 2 * wordSize);
2095 // Note that an "issuing PC" is actually the next PC after the call
2096 __ jr(T9); // jump to exception handler of caller
2097 __ delayed()->nop();
2098 }
2101 //
2102 // JVMTI ForceEarlyReturn support
2103 //
2104 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2105 address entry = __ pc();
2106 __ restore_bcp();
2107 __ restore_locals();
2108 __ empty_expression_stack();
2109 __ empty_FPU_stack();
2110 __ load_earlyret_value(state);
2112 #ifndef OPT_THREAD
2113 __ get_thread(TREG);
2114 #endif
2115 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2116 //const Address cond_addr(ecx, JvmtiThreadState::earlyret_state_offset());
2117 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2118 // Clear the earlyret state
2119 __ move(AT,JvmtiThreadState::earlyret_inactive);
2120 __ sw(AT,cond_addr);
2121 __ sync();
2124 __ remove_activation(state, T0,
2125 false, /* throw_monitor_exception */
2126 false, /* install_monitor_exception */
2127 true); /* notify_jvmdi */
2128 __ sync();
2129 __ jr(T0);
2130 __ delayed()->nop();
2131 return entry;
2132 } // end of ForceEarlyReturn support
2135 //-----------------------------------------------------------------------------
2136 // Helper for vtos entry point generation
2138 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2139 address& bep,
2140 address& cep,
2141 address& sep,
2142 address& aep,
2143 address& iep,
2144 address& lep,
2145 address& fep,
2146 address& dep,
2147 address& vep) {
2148 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2149 Label L;
2150 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2151 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2152 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2153 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2154 bep = cep = sep =
2155 iep = __ pc(); __ push(itos);
2156 vep = __ pc();
2157 __ bind(L);
2158 generate_and_dispatch(t);
2159 }
2162 //-----------------------------------------------------------------------------
2163 // Generation of individual instructions
2165 // helpers for generate_and_dispatch
2168 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2169 : TemplateInterpreterGenerator(code) {
2170 generate_all(); // down here so it can be "virtual"
2171 }
2173 //-----------------------------------------------------------------------------
2175 // Non-product code
2176 #ifndef PRODUCT
2177 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2178 address entry = __ pc();
2180 // prepare expression stack
2181 __ push(state); // save tosca
2183 // tos & tos2, added by yjl 7/15/2005
2184 // trace_bytecode need actually 4 args, the last two is tos&tos2
2185 // this work fine for x86. but mips o32 call convention will store A2-A3
2186 // to the stack position it think is the tos&tos2
2187 // when the expression stack have no more than 2 data, error occur.
2188 __ ld(A2, SP, 0);
2189 __ ld(A3, SP, 1 * wordSize);
2191 // pass arguments & call tracer
2192 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2193 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2195 // restore expression stack
2196 __ pop(state); // restore tosca
2198 // return
2199 __ jr(RA);
2200 __ delayed()->nop();
2202 return entry;
2203 }
2205 void TemplateInterpreterGenerator::count_bytecode() {
2206 __ li(T8, (long)&BytecodeCounter::_counter_value);
2207 __ lw(AT, T8, 0);
2208 __ daddi(AT, AT, 1);
2209 __ sw(AT, T8, 0);
2210 }
2212 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2213 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2214 __ lw(AT, T8, 0);
2215 __ daddi(AT, AT, 1);
2216 __ sw(AT, T8, 0);
2217 }
2219 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2220 __ li(T8, (long)&BytecodePairHistogram::_index);
2221 __ lw(T9, T8, 0);
2222 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2223 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2224 __ orr(T9, T9, T8);
2225 __ li(T8, (long)&BytecodePairHistogram::_index);
2226 __ sw(T9, T8, 0);
2227 __ dsll(T9, T9, 2);
2228 __ li(T8, (long)BytecodePairHistogram::_counters);
2229 __ dadd(T8, T8, T9);
2230 __ lw(AT, T8, 0);
2231 __ daddi(AT, AT, 1);
2232 __ sw(AT, T8, 0);
2233 }
2236 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2237 // Call a little run-time stub to avoid blow-up for each bytecode.
2238 // The run-time runtime saves the right registers, depending on
2239 // the tosca in-state for the given template.
2241 address entry = Interpreter::trace_code(t->tos_in());
2242 assert(entry != NULL, "entry must have been generated");
2243 __ call(entry, relocInfo::none);
2244 __ delayed()->nop();
2245 //add for compressedoops
2246 __ reinit_heapbase();
2247 }
2250 void TemplateInterpreterGenerator::stop_interpreter_at() {
2251 Label L;
2252 __ li(T8, long(&BytecodeCounter::_counter_value));
2253 __ lw(T8, T8, 0);
2254 __ move(AT, StopInterpreterAt);
2255 __ bne(T8, AT, L);
2256 __ delayed()->nop();
2257 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2258 __ delayed()->nop();
2259 __ bind(L);
2260 }
2261 #endif // !PRODUCT
2262 #endif // ! CC_INTERP