Tue, 16 Jun 2020 18:24:46 +0800
#13346 Backport of #10019 SP must be decremented first.
Reviewed-by: aoqi
1 /*
2 * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2015, 2020, 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 // expression stack must be empty before entering the VM if an
132 // exception happened
133 __ empty_expression_stack();
134 __ empty_FPU_stack();
135 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), FSR);
136 return entry;
137 }
139 address TemplateInterpreterGenerator::generate_exception_handler_common(
140 const char* name, const char* message, bool pass_oop) {
141 assert(!pass_oop || message == NULL, "either oop or message but not both");
142 address entry = __ pc();
144 // expression stack must be empty before entering the VM if an exception happened
145 __ empty_expression_stack();
146 // setup parameters
147 __ li(A1, (long)name);
148 if (pass_oop) {
149 __ call_VM(V0,
150 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), A1, FSR);
151 } else {
152 __ li(A2, (long)message);
153 __ call_VM(V0,
154 CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), A1, A2);
155 }
156 // throw exception
157 __ jmp(Interpreter::throw_exception_entry(), relocInfo::none);
158 __ delayed()->nop();
159 return entry;
160 }
163 address TemplateInterpreterGenerator::generate_continuation_for(TosState state) {
164 address entry = __ pc();
165 // NULL last_sp until next java call
166 __ sd(R0,Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
167 __ dispatch_next(state);
168 return entry;
169 }
172 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) {
174 address entry = __ pc();
176 // Restore stack bottom in case i2c adjusted stack
177 __ ld(SP, Address(FP, frame::interpreter_frame_last_sp_offset * wordSize));
178 // and NULL it as marker that sp is now tos until next java call
179 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
181 __ restore_bcp();
182 __ restore_locals();
184 // mdp: T8
185 // ret: FSR
186 // tmp: T9
187 if (state == atos) {
188 Register mdp = T8;
189 Register tmp = T9;
190 __ profile_return_type(mdp, FSR, tmp);
191 }
194 const Register cache = T9;
195 const Register index = T3;
196 __ get_cache_and_index_at_bcp(cache, index, 1, index_size);
198 const Register flags = cache;
199 __ dsll(AT, index, Address::times_ptr);
200 __ daddu(AT, cache, AT);
201 __ lw(flags, AT, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
202 __ andi(flags, flags, ConstantPoolCacheEntry::parameter_size_mask);
203 __ dsll(AT, flags, Interpreter::stackElementScale());
204 __ daddu(SP, SP, AT);
206 __ dispatch_next(state, step);
208 return entry;
209 }
212 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state,
213 int step) {
214 address entry = __ pc();
215 // NULL last_sp until next java call
216 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
217 __ restore_bcp();
218 __ restore_locals();
219 // handle exceptions
220 {
221 Label L;
222 const Register thread = TREG;
223 #ifndef OPT_THREAD
224 __ get_thread(thread);
225 #endif
226 __ lw(AT, thread, in_bytes(Thread::pending_exception_offset()));
227 __ beq(AT, R0, L);
228 __ delayed()->nop();
229 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_pending_exception));
230 __ should_not_reach_here();
231 __ bind(L);
232 }
233 __ dispatch_next(state, step);
234 return entry;
235 }
237 int AbstractInterpreter::BasicType_as_index(BasicType type) {
238 int i = 0;
239 switch (type) {
240 case T_BOOLEAN: i = 0; break;
241 case T_CHAR : i = 1; break;
242 case T_BYTE : i = 2; break;
243 case T_SHORT : i = 3; break;
244 case T_INT : // fall through
245 case T_LONG : // fall through
246 case T_VOID : i = 4; break;
247 case T_FLOAT : i = 5; break;
248 case T_DOUBLE : i = 6; break;
249 case T_OBJECT : // fall through
250 case T_ARRAY : i = 7; break;
251 default : ShouldNotReachHere();
252 }
253 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers,
254 "index out of bounds");
255 return i;
256 }
259 address TemplateInterpreterGenerator::generate_result_handler_for(
260 BasicType type) {
261 address entry = __ pc();
262 switch (type) {
263 case T_BOOLEAN: __ c2bool(V0); break;
264 case T_CHAR : __ andi(V0, V0, 0xFFFF); break;
265 case T_BYTE : __ sign_extend_byte (V0); break;
266 case T_SHORT : __ sign_extend_short(V0); break;
267 case T_INT : /* nothing to do */ break;
268 case T_FLOAT : /* nothing to do */ break;
269 case T_DOUBLE : /* nothing to do */ break;
270 case T_OBJECT :
271 {
272 __ ld(V0, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
273 __ verify_oop(V0); // and verify it
274 }
275 break;
276 default : ShouldNotReachHere();
277 }
278 __ jr(RA); // return from result handler
279 __ delayed()->nop();
280 return entry;
281 }
283 address TemplateInterpreterGenerator::generate_safept_entry_for(
284 TosState state,
285 address runtime_entry) {
286 address entry = __ pc();
287 __ push(state);
288 __ call_VM(noreg, runtime_entry);
289 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos));
290 return entry;
291 }
295 // Helpers for commoning out cases in the various type of method entries.
296 //
299 // increment invocation count & check for overflow
300 //
301 // Note: checking for negative value instead of overflow
302 // so we have a 'sticky' overflow test
303 //
304 // Rmethod: method
305 // T3 : invocation counter
306 //
307 void InterpreterGenerator::generate_counter_incr(
308 Label* overflow,
309 Label* profile_method,
310 Label* profile_method_continue) {
311 Label done;
312 if (TieredCompilation) {
313 int increment = InvocationCounter::count_increment;
314 int mask = ((1 << Tier0InvokeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
315 Label no_mdo;
316 if (ProfileInterpreter) {
317 // Are we profiling?
318 __ ld(FSR, Address(Rmethod, Method::method_data_offset()));
319 __ beq(FSR, R0, no_mdo);
320 __ delayed()->nop();
321 // Increment counter in the MDO
322 const Address mdo_invocation_counter(FSR, in_bytes(MethodData::invocation_counter_offset()) +
323 in_bytes(InvocationCounter::counter_offset()));
324 __ increment_mask_and_jump(mdo_invocation_counter, increment, mask, T3, false, Assembler::zero, overflow);
325 __ beq(R0, R0, done);
326 __ delayed()->nop();
327 }
328 __ bind(no_mdo);
329 // Increment counter in MethodCounters
330 const Address invocation_counter(FSR,
331 MethodCounters::invocation_counter_offset() +
332 InvocationCounter::counter_offset());
333 __ get_method_counters(Rmethod, FSR, done);
334 __ increment_mask_and_jump(invocation_counter, increment, mask, T3, false, Assembler::zero, overflow);
335 __ bind(done);
336 } else {
337 const Address invocation_counter(FSR, in_bytes(MethodCounters::invocation_counter_offset())
338 + in_bytes(InvocationCounter::counter_offset()));
339 const Address backedge_counter (FSR, in_bytes(MethodCounters::backedge_counter_offset())
340 + in_bytes(InvocationCounter::counter_offset()));
342 __ get_method_counters(Rmethod, FSR, done);
344 if (ProfileInterpreter) { // %%% Merge this into methodDataOop
345 __ lw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
346 __ incrementl(T9, 1);
347 __ sw(T9, FSR, in_bytes(MethodCounters::interpreter_invocation_counter_offset()));
348 }
349 // Update standard invocation counters
350 __ lw(T3, invocation_counter);
351 __ increment(T3, InvocationCounter::count_increment);
352 __ sw(T3, invocation_counter); // save invocation count
354 __ lw(FSR, backedge_counter); // load backedge counter
355 __ li(AT, InvocationCounter::count_mask_value); // mask out the status bits
356 __ andr(FSR, FSR, AT);
358 __ dadd(T3, T3, FSR); // add both counters
360 if (ProfileInterpreter && profile_method != NULL) {
361 // Test to see if we should create a method data oop
362 if (Assembler::is_simm16(InvocationCounter::InterpreterProfileLimit)) {
363 __ slti(AT, T3, InvocationCounter::InterpreterProfileLimit);
364 } else {
365 __ li(AT, (long)&InvocationCounter::InterpreterProfileLimit);
366 __ lw(AT, AT, 0);
367 __ slt(AT, T3, AT);
368 }
370 __ bne_far(AT, R0, *profile_method_continue);
371 __ delayed()->nop();
373 // if no method data exists, go to profile_method
374 __ test_method_data_pointer(FSR, *profile_method);
375 }
377 if (Assembler::is_simm16(CompileThreshold)) {
378 __ srl(AT, T3, InvocationCounter::count_shift);
379 __ slti(AT, AT, CompileThreshold);
380 } else {
381 __ li(AT, (long)&InvocationCounter::InterpreterInvocationLimit);
382 __ lw(AT, AT, 0);
383 __ slt(AT, T3, AT);
384 }
386 __ beq_far(AT, R0, *overflow);
387 __ delayed()->nop();
388 __ bind(done);
389 }
390 }
392 void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
394 // Asm interpreter on entry
395 // S7 - locals
396 // S0 - bcp
397 // Rmethod - method
398 // FP - interpreter frame
400 // On return (i.e. jump to entry_point)
401 // Rmethod - method
402 // RA - return address of interpreter caller
403 // tos - the last parameter to Java method
404 // SP - sender_sp
407 // the bcp is valid if and only if it's not null
408 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
409 InterpreterRuntime::frequency_counter_overflow), R0);
410 __ ld(Rmethod, FP, method_offset);
411 // Preserve invariant that S0/S7 contain bcp/locals of sender frame
412 __ b_far(*do_continue);
413 __ delayed()->nop();
414 }
416 // See if we've got enough room on the stack for locals plus overhead.
417 // The expression stack grows down incrementally, so the normal guard
418 // page mechanism will work for that.
419 //
420 // NOTE: Since the additional locals are also always pushed (wasn't
421 // obvious in generate_method_entry) so the guard should work for them
422 // too.
423 //
424 // Args:
425 // T2: number of additional locals this frame needs (what we must check)
426 // T0: Method*
427 //
428 void InterpreterGenerator::generate_stack_overflow_check(void) {
429 // see if we've got enough room on the stack for locals plus overhead.
430 // the expression stack grows down incrementally, so the normal guard
431 // page mechanism will work for that.
432 //
433 // Registers live on entry:
434 //
435 // T0: Method*
436 // T2: number of additional locals this frame needs (what we must check)
438 // NOTE: since the additional locals are also always pushed (wasn't obvious in
439 // generate_method_entry) so the guard should work for them too.
440 //
442 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
444 // total overhead size: entry_size + (saved fp thru expr stack bottom).
445 // be sure to change this if you add/subtract anything to/from the overhead area
446 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset*wordSize)
447 + entry_size;
449 const int page_size = os::vm_page_size();
451 Label after_frame_check;
453 // see if the frame is greater than one page in size. If so,
454 // then we need to verify there is enough stack space remaining
455 // for the additional locals.
456 __ move(AT, (page_size - overhead_size) / Interpreter::stackElementSize);
457 __ slt(AT, AT, T2);
458 __ beq(AT, R0, after_frame_check);
459 __ delayed()->nop();
461 // compute sp as if this were going to be the last frame on
462 // the stack before the red zone
463 #ifndef OPT_THREAD
464 Register thread = T1;
465 __ get_thread(thread);
466 #else
467 Register thread = TREG;
468 #endif
470 // locals + overhead, in bytes
471 __ dsll(T3, T2, Interpreter::stackElementScale());
472 __ daddiu(T3, T3, overhead_size); // locals * 4 + overhead_size --> T3
474 #ifdef ASSERT
475 Label stack_base_okay, stack_size_okay;
476 // verify that thread stack base is non-zero
477 __ ld(AT, thread, in_bytes(Thread::stack_base_offset()));
478 __ bne(AT, R0, stack_base_okay);
479 __ delayed()->nop();
480 __ stop("stack base is zero");
481 __ bind(stack_base_okay);
482 // verify that thread stack size is non-zero
483 __ ld(AT, thread, in_bytes(Thread::stack_size_offset()));
484 __ bne(AT, R0, stack_size_okay);
485 __ delayed()->nop();
486 __ stop("stack size is zero");
487 __ bind(stack_size_okay);
488 #endif
490 // Add stack base to locals and subtract stack size
491 __ ld(AT, thread, in_bytes(Thread::stack_base_offset())); // stack_base --> AT
492 __ dadd(T3, T3, AT); // locals * 4 + overhead_size + stack_base--> T3
493 __ ld(AT, thread, in_bytes(Thread::stack_size_offset())); // stack_size --> AT
494 __ dsub(T3, T3, AT); // locals * 4 + overhead_size + stack_base - stack_size --> T3
497 // add in the redzone and yellow size
498 __ move(AT, (StackRedPages+StackYellowPages) * page_size);
499 __ add(T3, T3, AT);
501 // check against the current stack bottom
502 __ slt(AT, T3, SP);
503 __ bne(AT, R0, after_frame_check);
504 __ delayed()->nop();
506 // Note: the restored frame is not necessarily interpreted.
507 // Use the shared runtime version of the StackOverflowError.
508 __ move(SP, Rsender);
509 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "stub not yet generated");
510 __ jmp(StubRoutines::throw_StackOverflowError_entry(), relocInfo::runtime_call_type);
511 __ delayed()->nop();
513 // all done with frame size check
514 __ bind(after_frame_check);
515 }
517 // Allocate monitor and lock method (asm interpreter)
518 // Rmethod - Method*
519 void InterpreterGenerator::lock_method(void) {
520 // synchronize method
521 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
523 #ifdef ASSERT
524 { Label L;
525 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
526 __ andi(T0, T0, JVM_ACC_SYNCHRONIZED);
527 __ bne(T0, R0, L);
528 __ delayed()->nop();
529 __ stop("method doesn't need synchronization");
530 __ bind(L);
531 }
532 #endif // ASSERT
533 // get synchronization object
534 {
535 Label done;
536 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
537 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
538 __ andi(T2, T0, JVM_ACC_STATIC);
539 __ ld(T0, LVP, Interpreter::local_offset_in_bytes(0));
540 __ beq(T2, R0, done);
541 __ delayed()->nop();
542 __ ld(T0, Rmethod, in_bytes(Method::const_offset()));
543 __ ld(T0, T0, in_bytes(ConstMethod::constants_offset()));
544 __ ld(T0, T0, ConstantPool::pool_holder_offset_in_bytes());
545 __ ld(T0, T0, mirror_offset);
546 __ bind(done);
547 }
548 // add space for monitor & lock
549 __ daddi(SP, SP, (-1) * entry_size); // add space for a monitor entry
550 __ sd(SP, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
551 // set new monitor block top
552 __ sd(T0, SP, BasicObjectLock::obj_offset_in_bytes()); // store object
553 // FIXME: I do not know what lock_object will do and what it will need
554 __ move(c_rarg0, SP); // object address
555 __ lock_object(c_rarg0);
556 }
558 // Generate a fixed interpreter frame. This is identical setup for
559 // interpreted methods and for native methods hence the shared code.
560 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call) {
562 // [ local var m-1 ] <--- sp
563 // ...
564 // [ local var 0 ]
565 // [ argumnet word n-1 ] <--- T0(sender's sp)
566 // ...
567 // [ argument word 0 ] <--- S7
569 // initialize fixed part of activation frame
570 // sender's sp in Rsender
571 int i = 0;
572 int frame_size = 9;
573 #ifndef CORE
574 ++frame_size;
575 #endif
576 __ daddiu(SP, SP, (-frame_size) * wordSize);
577 __ sd(RA, SP, (frame_size - 1) * wordSize); // save return address
578 __ sd(FP, SP, (frame_size - 2) * wordSize); // save sender's fp
579 __ daddiu(FP, SP, (frame_size - 2) * wordSize);
580 __ sd(Rsender, FP, (-++i) * wordSize); // save sender's sp
581 __ sd(R0, FP,(-++i) * wordSize); //save last_sp as null
582 __ sd(LVP, FP, (-++i) * wordSize); // save locals offset
583 __ ld(BCP, Rmethod, in_bytes(Method::const_offset())); // get constMethodOop
584 __ daddiu(BCP, BCP, in_bytes(ConstMethod::codes_offset())); // get codebase
585 __ sd(Rmethod, FP, (-++i) * wordSize); // save Method*
586 #ifndef CORE
587 if (ProfileInterpreter) {
588 Label method_data_continue;
589 __ ld(AT, Rmethod, in_bytes(Method::method_data_offset()));
590 __ beq(AT, R0, method_data_continue);
591 __ delayed()->nop();
592 __ daddi(AT, AT, in_bytes(MethodData::data_offset()));
593 __ bind(method_data_continue);
594 __ sd(AT, FP, (-++i) * wordSize);
595 } else {
596 __ sd(R0, FP, (-++i) * wordSize);
597 }
598 #endif // !CORE
600 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
601 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
602 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
603 __ sd(T2, FP, (-++i) * wordSize); // set constant pool cache
604 if (native_call) {
605 __ sd(R0, FP, (-++i) * wordSize); // no bcp
606 } else {
607 __ sd(BCP, FP, (-++i) * wordSize); // set bcp
608 }
609 __ sd(SP, FP, (-++i) * wordSize); // reserve word for pointer to expression stack bottom
610 assert(i + 2 == frame_size, "i + 2 should be equal to frame_size");
611 }
613 // End of helpers
615 // Various method entries
616 //------------------------------------------------------------------------------------------------------------------------
617 //
618 //
620 // Call an accessor method (assuming it is resolved, otherwise drop
621 // into vanilla (slow path) entry
622 address InterpreterGenerator::generate_accessor_entry(void) {
624 // Rmethod: Method*
625 // V0: receiver (preserve for slow entry into asm interpreter)
626 // Rsender: senderSP must preserved for slow path, set SP to it on fast path
628 address entry_point = __ pc();
629 Label xreturn_path;
630 // do fastpath for resolved accessor methods
631 if (UseFastAccessorMethods) {
632 Label slow_path;
633 __ li(T2, SafepointSynchronize::address_of_state());
634 __ lw(AT, T2, 0);
635 __ daddi(AT, AT, -(SafepointSynchronize::_not_synchronized));
636 __ bne(AT, R0, slow_path);
637 __ delayed()->nop();
638 // Code: _aload_0, _(i|a)getfield, _(i|a)return or any rewrites thereof;
639 // parameter size = 1
640 // Note: We can only use this code if the getfield has been resolved
641 // and if we don't have a null-pointer exception => check for
642 // these conditions first and use slow path if necessary.
643 // Rmethod: method
644 // V0: receiver
646 // [ receiver ] <-- sp
647 __ ld(T0, SP, 0);
649 // check if local 0 != NULL and read field
650 __ beq(T0, R0, slow_path);
651 __ delayed()->nop();
652 __ ld(T2, Rmethod, in_bytes(Method::const_offset()));
653 __ ld(T2, T2, in_bytes(ConstMethod::constants_offset()));
654 // read first instruction word and extract bytecode @ 1 and index @ 2
655 __ ld(T3, Rmethod, in_bytes(Method::const_offset()));
656 __ lw(T3, T3, in_bytes(ConstMethod::codes_offset()));
657 // Shift codes right to get the index on the right.
658 // The bytecode fetched looks like <index><0xb4><0x2a>
659 __ dsrl(T3, T3, 2 * BitsPerByte);
660 // FIXME: maybe it's wrong
661 __ dsll(T3, T3, exact_log2(in_words(ConstantPoolCacheEntry::size())));
662 __ ld(T2, T2, ConstantPool::cache_offset_in_bytes());
664 // T0: local 0
665 // Rmethod: method
666 // V0: receiver - do not destroy since it is needed for slow path!
667 // T1: scratch use which register instead ?
668 // T3: constant pool cache index
669 // T2: constant pool cache
670 // Rsender: send's sp
671 // check if getfield has been resolved and read constant pool cache entry
672 // check the validity of the cache entry by testing whether _indices field
673 // contains Bytecode::_getfield in b1 byte.
674 assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
676 __ dsll(T8, T3, Address::times_8);
677 __ move(T1, in_bytes(ConstantPoolCache::base_offset()
678 + ConstantPoolCacheEntry::indices_offset()));
679 __ dadd(T1, T8, T1);
680 __ dadd(T1, T1, T2);
681 __ lw(T1, T1, 0);
682 __ dsrl(T1, T1, 2 * BitsPerByte);
683 __ andi(T1, T1, 0xFF);
684 __ daddi(T1, T1, (-1) * Bytecodes::_getfield);
685 __ bne(T1, R0, slow_path);
686 __ delayed()->nop();
688 // Note: constant pool entry is not valid before bytecode is resolved
690 __ move(T1, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
691 __ dadd(T1, T1, T8);
692 __ dadd(T1, T1, T2);
693 __ lw(AT, T1, 0);
695 __ move(T1, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
696 __ dadd(T1, T1, T8);
697 __ dadd(T1, T1, T2);
698 __ lw(T3, T1, 0);
700 Label notByte, notBool, notShort, notChar, notObj;
702 // Need to differentiate between igetfield, agetfield, bgetfield etc.
703 // because they are different sizes.
704 // Use the type from the constant pool cache
705 __ srl(T3, T3, ConstantPoolCacheEntry::tos_state_shift);
706 // Make sure we don't need to mask T3 for tosBits after the above shift
707 ConstantPoolCacheEntry::verify_tos_state_shift();
708 // btos = 0
709 __ bne(T3, R0, notByte);
710 __ delayed()->dadd(T0, T0, AT);
712 __ lb(V0, T0, 0);
713 __ b(xreturn_path);
714 __ delayed()->nop();
716 //ztos
717 __ bind(notByte);
718 __ daddi(T1, T3, (-1) * ztos);
719 __ bne(T1, R0, notBool);
720 __ delayed()->nop();
721 __ lb(V0, T0, 0);
722 __ b(xreturn_path);
723 __ delayed()->nop();
725 //stos
726 __ bind(notBool);
727 __ daddi(T1, T3, (-1) * stos);
728 __ bne(T1, R0, notShort);
729 __ delayed()->nop();
730 __ lh(V0, T0, 0);
731 __ b(xreturn_path);
732 __ delayed()->nop();
734 //ctos
735 __ bind(notShort);
736 __ daddi(T1, T3, (-1) * ctos);
737 __ bne(T1, R0, notChar);
738 __ delayed()->nop();
739 __ lhu(V0, T0, 0);
740 __ b(xreturn_path);
741 __ delayed()->nop();
743 //atos
744 __ bind(notChar);
745 __ daddi(T1, T3, (-1) * atos);
746 __ bne(T1, R0, notObj);
747 __ delayed()->nop();
748 //add for compressedoops
749 __ load_heap_oop(V0, Address(T0, 0));
750 __ b(xreturn_path);
751 __ delayed()->nop();
753 //itos
754 __ bind(notObj);
755 #ifdef ASSERT
756 Label okay;
757 __ daddi(T1, T3, (-1) * itos);
758 __ beq(T1, R0, okay);
759 __ delayed()->nop();
760 __ stop("what type is this?");
761 __ bind(okay);
762 #endif // ASSERT
763 __ lw(V0, T0, 0);
765 __ bind(xreturn_path);
767 // _ireturn/_areturn
768 //FIXME
769 __ move(SP, Rsender);//FIXME, set sender's fp to SP
770 __ jr(RA);
771 __ delayed()->nop();
773 // generate a vanilla interpreter entry as the slow path
774 __ bind(slow_path);
775 (void) generate_normal_entry(false);
776 } else {
777 (void) generate_normal_entry(false);
778 }
780 return entry_point;
781 }
783 // Method entry for java.lang.ref.Reference.get.
784 address InterpreterGenerator::generate_Reference_get_entry(void) {
785 #if INCLUDE_ALL_GCS
786 // Code: _aload_0, _getfield, _areturn
787 // parameter size = 1
788 //
789 // The code that gets generated by this routine is split into 2 parts:
790 // 1. The "intrinsified" code for G1 (or any SATB based GC),
791 // 2. The slow path - which is an expansion of the regular method entry.
792 //
793 // Notes:-
794 // * In the G1 code we do not check whether we need to block for
795 // a safepoint. If G1 is enabled then we must execute the specialized
796 // code for Reference.get (except when the Reference object is null)
797 // so that we can log the value in the referent field with an SATB
798 // update buffer.
799 // If the code for the getfield template is modified so that the
800 // G1 pre-barrier code is executed when the current method is
801 // Reference.get() then going through the normal method entry
802 // will be fine.
803 // * The G1 code can, however, check the receiver object (the instance
804 // of java.lang.Reference) and jump to the slow path if null. If the
805 // Reference object is null then we obviously cannot fetch the referent
806 // and so we don't need to call the G1 pre-barrier. Thus we can use the
807 // regular method entry code to generate the NPE.
808 //
809 // This code is based on generate_accessor_enty.
810 //
811 // Rmethod: Method*
813 // Rsender: senderSP must preserve for slow path, set SP to it on fast path (Rsender)
815 address entry = __ pc();
817 const int referent_offset = java_lang_ref_Reference::referent_offset;
818 guarantee(referent_offset > 0, "referent offset not initialized");
820 if (UseG1GC) {
821 Label slow_path;
823 // Check if local 0 != NULL
824 // If the receiver is null then it is OK to jump to the slow path.
825 __ ld(V0, SP, 0);
827 __ beq(V0, R0, slow_path);
828 __ delayed()->nop();
830 // Generate the G1 pre-barrier code to log the value of
831 // the referent field in an SATB buffer.
833 // Load the value of the referent field.
834 const Address field_address(V0, referent_offset);
835 __ load_heap_oop(V0, field_address);
837 __ push(RA);
838 // Generate the G1 pre-barrier code to log the value of
839 // the referent field in an SATB buffer.
840 __ g1_write_barrier_pre(noreg /* obj */,
841 V0 /* pre_val */,
842 TREG /* thread */,
843 Rmethod /* tmp */,
844 true /* tosca_live */,
845 true /* expand_call */);
846 __ pop(RA);
848 __ jr(RA);
849 __ delayed()->daddu(SP, Rsender, R0); // set sp to sender sp
851 // generate a vanilla interpreter entry as the slow path
852 __ bind(slow_path);
853 (void) generate_normal_entry(false);
855 return entry;
856 }
857 #endif // INCLUDE_ALL_GCS
859 // If G1 is not enabled then attempt to go through the accessor entry point
860 // Reference.get is an accessor
861 return generate_accessor_entry();
862 }
864 // Interpreter stub for calling a native method. (asm interpreter)
865 // This sets up a somewhat different looking stack for calling the
866 // native method than the typical interpreter frame setup.
867 address InterpreterGenerator::generate_native_entry(bool synchronized) {
868 // determine code generation flags
869 bool inc_counter = UseCompiler || CountCompiledCalls;
870 // Rsender: sender's sp
871 // Rmethod: Method*
872 address entry_point = __ pc();
874 #ifndef CORE
875 const Address invocation_counter(Rmethod,in_bytes(MethodCounters::invocation_counter_offset() +
876 InvocationCounter::counter_offset()));
877 #endif
879 // get parameter size (always needed)
880 // the size in the java stack
881 __ ld(V0, Rmethod, in_bytes(Method::const_offset()));
882 __ lhu(V0, V0, in_bytes(ConstMethod::size_of_parameters_offset()));
884 // native calls don't need the stack size check since they have no expression stack
885 // and the arguments are already on the stack and we only add a handful of words
886 // to the stack
888 // Rmethod: Method*
889 // V0: size of parameters
890 // Layout of frame at this point
891 //
892 // [ argument word n-1 ] <--- sp
893 // ...
894 // [ argument word 0 ]
896 // for natives the size of locals is zero
898 // compute beginning of parameters (S7)
899 __ dsll(LVP, V0, Address::times_8);
900 __ daddiu(LVP, LVP, (-1) * wordSize);
901 __ dadd(LVP, LVP, SP);
904 // add 2 zero-initialized slots for native calls
905 __ daddi(SP, SP, (-2) * wordSize);
906 __ sd(R0, SP, 1 * wordSize); // slot for native oop temp offset (setup via runtime)
907 __ sd(R0, SP, 0 * wordSize); // slot for static native result handler3 (setup via runtime)
909 // Layout of frame at this point
910 // [ method holder mirror ] <--- sp
911 // [ result type info ]
912 // [ argument word n-1 ] <--- T0
913 // ...
914 // [ argument word 0 ] <--- LVP
917 #ifndef CORE
918 if (inc_counter) __ lw(T3, invocation_counter); // (pre-)fetch invocation count
919 #endif
921 // initialize fixed part of activation frame
922 generate_fixed_frame(true);
923 // after this function, the layout of frame is as following
924 //
925 // [ monitor block top ] <--- sp ( the top monitor entry )
926 // [ byte code pointer (0) ] (if native, bcp = 0)
927 // [ constant pool cache ]
928 // [ Method* ]
929 // [ locals offset ]
930 // [ sender's sp ]
931 // [ sender's fp ]
932 // [ return address ] <--- fp
933 // [ method holder mirror ]
934 // [ result type info ]
935 // [ argumnet word n-1 ] <--- sender's sp
936 // ...
937 // [ argument word 0 ] <--- S7
940 // make sure method is native & not abstract
941 #ifdef ASSERT
942 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
943 {
944 Label L;
945 __ andi(AT, T0, JVM_ACC_NATIVE);
946 __ bne(AT, R0, L);
947 __ delayed()->nop();
948 __ stop("tried to execute native method as non-native");
949 __ bind(L);
950 }
951 {
952 Label L;
953 __ andi(AT, T0, JVM_ACC_ABSTRACT);
954 __ beq(AT, R0, L);
955 __ delayed()->nop();
956 __ stop("tried to execute abstract method in interpreter");
957 __ bind(L);
958 }
959 #endif
961 // Since at this point in the method invocation the exception handler
962 // would try to exit the monitor of synchronized methods which hasn't
963 // been entered yet, we set the thread local variable
964 // _do_not_unlock_if_synchronized to true. The remove_activation will
965 // check this flag.
966 Register thread = TREG;
967 #ifndef OPT_THREAD
968 __ get_thread(thread);
969 #endif
970 __ move(AT, (int)true);
971 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
973 #ifndef CORE
974 // increment invocation count & check for overflow
975 Label invocation_counter_overflow;
976 if (inc_counter) {
977 generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
978 }
980 Label continue_after_compile;
981 __ bind(continue_after_compile);
982 #endif // CORE
984 bang_stack_shadow_pages(true);
986 // reset the _do_not_unlock_if_synchronized flag
987 #ifndef OPT_THREAD
988 __ get_thread(thread);
989 #endif
990 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
992 // check for synchronized methods
993 // Must happen AFTER invocation_counter check and stack overflow check,
994 // so method is not locked if overflows.
995 if (synchronized) {
996 lock_method();
997 } else {
998 // no synchronization necessary
999 #ifdef ASSERT
1000 {
1001 Label L;
1002 __ lw(T0, Rmethod, in_bytes(Method::access_flags_offset()));
1003 __ andi(AT, T0, JVM_ACC_SYNCHRONIZED);
1004 __ beq(AT, R0, L);
1005 __ delayed()->nop();
1006 __ stop("method needs synchronization");
1007 __ bind(L);
1008 }
1009 #endif
1010 }
1012 // after method_lock, the layout of frame is as following
1013 //
1014 // [ monitor entry ] <--- sp
1015 // ...
1016 // [ monitor entry ]
1017 // [ monitor block top ] ( the top monitor entry )
1018 // [ byte code pointer (0) ] (if native, bcp = 0)
1019 // [ constant pool cache ]
1020 // [ Method* ]
1021 // [ locals offset ]
1022 // [ sender's sp ]
1023 // [ sender's fp ]
1024 // [ return address ] <--- fp
1025 // [ method holder mirror ]
1026 // [ result type info ]
1027 // [ argumnet word n-1 ] <--- ( sender's sp )
1028 // ...
1029 // [ argument word 0 ] <--- S7
1031 // start execution
1032 #ifdef ASSERT
1033 {
1034 Label L;
1035 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1036 __ beq(AT, SP, L);
1037 __ delayed()->nop();
1038 __ stop("broken stack frame setup in interpreter in asm");
1039 __ bind(L);
1040 }
1041 #endif
1043 // jvmti/jvmpi support
1044 __ notify_method_entry();
1046 // work registers
1047 const Register method = Rmethod;
1048 //const Register thread = T2;
1049 const Register t = RT4;
1051 __ get_method(method);
1052 __ verify_oop(method);
1053 {
1054 Label L, Lstatic;
1055 __ ld(t,method,in_bytes(Method::const_offset()));
1056 __ lhu(t, t, in_bytes(ConstMethod::size_of_parameters_offset()));
1057 // MIPS n64 ABI: caller does not reserve space for the register auguments.
1058 // A0 and A1(if needed)
1059 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1060 __ andi(AT, AT, JVM_ACC_STATIC);
1061 __ beq(AT, R0, Lstatic);
1062 __ delayed()->nop();
1063 __ daddiu(t, t, 1);
1064 __ bind(Lstatic);
1065 __ daddiu(t, t, -7);
1066 __ blez(t, L);
1067 __ delayed()->nop();
1068 __ dsll(t, t, Address::times_8);
1069 __ dsub(SP, SP, t);
1070 __ bind(L);
1071 }
1072 __ move(AT, -(StackAlignmentInBytes));
1073 __ andr(SP, SP, AT);
1074 __ move(AT, SP);
1075 // [ ] <--- sp
1076 // ... (size of parameters - 8 )
1077 // [ monitor entry ]
1078 // ...
1079 // [ monitor entry ]
1080 // [ monitor block top ] ( the top monitor entry )
1081 // [ byte code pointer (0) ] (if native, bcp = 0)
1082 // [ constant pool cache ]
1083 // [ Method* ]
1084 // [ locals offset ]
1085 // [ sender's sp ]
1086 // [ sender's fp ]
1087 // [ return address ] <--- fp
1088 // [ method holder mirror ]
1089 // [ result type info ]
1090 // [ argumnet word n-1 ] <--- ( sender's sp )
1091 // ...
1092 // [ argument word 0 ] <--- LVP
1094 // get signature handler
1095 {
1096 Label L;
1097 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1098 __ bne(T9, R0, L);
1099 __ delayed()->nop();
1100 __ call_VM(NOREG, CAST_FROM_FN_PTR(address,
1101 InterpreterRuntime::prepare_native_call), method);
1102 __ get_method(method);
1103 __ ld(T9, method, in_bytes(Method::signature_handler_offset()));
1104 __ bind(L);
1105 }
1107 // call signature handler
1108 // FIXME: when change codes in InterpreterRuntime, note this point
1109 // from: begin of parameters
1110 assert(InterpreterRuntime::SignatureHandlerGenerator::from() == LVP, "adjust this code");
1111 // to: current sp
1112 assert(InterpreterRuntime::SignatureHandlerGenerator::to () == SP, "adjust this code");
1113 // temp: T3
1114 assert(InterpreterRuntime::SignatureHandlerGenerator::temp() == t , "adjust this code");
1116 __ jalr(T9);
1117 __ delayed()->nop();
1118 __ get_method(method);
1120 //
1121 // if native function is static, and its second parameter has type length of double word,
1122 // and first parameter has type length of word, we have to reserve one word
1123 // for the first parameter, according to mips o32 abi.
1124 // if native function is not static, and its third parameter has type length of double word,
1125 // and second parameter has type length of word, we have to reserve one word for the second
1126 // parameter.
1127 //
1130 // result handler is in V0
1131 // set result handler
1132 __ sd(V0, FP, (frame::interpreter_frame_result_handler_offset)*wordSize);
1134 #define FIRSTPARA_SHIFT_COUNT 5
1135 #define SECONDPARA_SHIFT_COUNT 9
1136 #define THIRDPARA_SHIFT_COUNT 13
1137 #define PARA_MASK 0xf
1139 // pass mirror handle if static call
1140 {
1141 Label L;
1142 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
1143 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1144 __ andi(AT, t, JVM_ACC_STATIC);
1145 __ beq(AT, R0, L);
1146 __ delayed()->nop();
1148 // get mirror
1149 __ ld(t, method, in_bytes(Method:: const_offset()));
1150 __ ld(t, t, in_bytes(ConstMethod::constants_offset())); //??
1151 __ ld(t, t, ConstantPool::pool_holder_offset_in_bytes());
1152 __ ld(t, t, mirror_offset);
1153 // copy mirror into activation frame
1154 //__ sw(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1155 // pass handle to mirror
1156 __ sd(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1157 __ daddi(t, FP, frame::interpreter_frame_oop_temp_offset * wordSize);
1158 __ move(A1, t);
1159 __ bind(L);
1160 }
1162 // [ mthd holder mirror ptr ] <--- sp --------------------| (only for static method)
1163 // [ ] |
1164 // ... size of parameters(or +1) |
1165 // [ monitor entry ] |
1166 // ... |
1167 // [ monitor entry ] |
1168 // [ monitor block top ] ( the top monitor entry ) |
1169 // [ byte code pointer (0) ] (if native, bcp = 0) |
1170 // [ constant pool cache ] |
1171 // [ Method* ] |
1172 // [ locals offset ] |
1173 // [ sender's sp ] |
1174 // [ sender's fp ] |
1175 // [ return address ] <--- fp |
1176 // [ method holder mirror ] <----------------------------|
1177 // [ result type info ]
1178 // [ argumnet word n-1 ] <--- ( sender's sp )
1179 // ...
1180 // [ argument word 0 ] <--- S7
1182 // get native function entry point
1183 { Label L;
1184 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1185 __ li(V1, SharedRuntime::native_method_throw_unsatisfied_link_error_entry());
1186 __ bne(V1, T9, L);
1187 __ delayed()->nop();
1188 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
1189 __ get_method(method);
1190 __ verify_oop(method);
1191 __ ld(T9, method, in_bytes(Method::native_function_offset()));
1192 __ bind(L);
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 __ move(A0, t);
1202 // [ jni environment ] <--- sp
1203 // [ mthd holder mirror ptr ] ---------------------------->| (only for static method)
1204 // [ ] |
1205 // ... size of parameters |
1206 // [ monitor entry ] |
1207 // ... |
1208 // [ monitor entry ] |
1209 // [ monitor block top ] ( the top monitor entry ) |
1210 // [ byte code pointer (0) ] (if native, bcp = 0) |
1211 // [ constant pool cache ] |
1212 // [ Method* ] |
1213 // [ locals offset ] |
1214 // [ sender's sp ] |
1215 // [ sender's fp ] |
1216 // [ return address ] <--- fp |
1217 // [ method holder mirror ] <----------------------------|
1218 // [ result type info ]
1219 // [ argumnet word n-1 ] <--- ( sender's sp )
1220 // ...
1221 // [ argument word 0 ] <--- S7
1223 // set_last_Java_frame_before_call
1224 __ sd(FP, thread, in_bytes(JavaThread::last_Java_fp_offset()));
1225 // Change state to native (we save the return address in the thread, since it might not
1226 // be pushed on the stack when we do a a stack traversal). It is enough that the pc()
1227 // points into the right code segment. It does not have to be the correct return pc.
1228 __ li(t, __ pc());
1229 __ sd(t, thread, in_bytes(JavaThread::last_Java_pc_offset()));
1230 __ sd(SP, thread, in_bytes(JavaThread::last_Java_sp_offset()));
1232 // change thread state
1233 #ifdef ASSERT
1234 {
1235 Label L;
1236 __ lw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1237 __ daddi(t, t, (-1) * _thread_in_Java);
1238 __ beq(t, R0, L);
1239 __ delayed()->nop();
1240 __ stop("Wrong thread state in native stub");
1241 __ bind(L);
1242 }
1243 #endif
1245 __ move(t, _thread_in_native);
1246 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1248 // call native method
1249 __ jalr(T9);
1250 __ delayed()->nop();
1251 // result potentially in V0 or F0
1254 // via _last_native_pc and not via _last_jave_sp
1255 // NOTE: the order of theses push(es) is known to frame::interpreter_frame_result.
1256 // If the order changes or anything else is added to the stack the code in
1257 // interpreter_frame_result will have to be changed.
1258 //FIXME, should modify here
1259 // save return value to keep the value from being destroyed by other calls
1260 __ push(dtos);
1261 __ push(ltos);
1263 // change thread state
1264 __ get_thread(thread);
1265 __ move(t, _thread_in_native_trans);
1266 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1268 if( os::is_MP() ) __ sync(); // Force this write out before the read below
1270 // check for safepoint operation in progress and/or pending suspend requests
1271 { Label Continue;
1273 // Don't use call_VM as it will see a possible pending exception and forward it
1274 // and never return here preventing us from clearing _last_native_pc down below.
1275 // Also can't use call_VM_leaf either as it will check to see if BCP & LVP are
1276 // preserved and correspond to the bcp/locals pointers. So we do a runtime call
1277 // by hand.
1278 //
1279 Label L;
1280 __ li(AT, SafepointSynchronize::address_of_state());
1281 __ lw(AT, AT, 0);
1282 __ bne(AT, R0, L);
1283 __ delayed()->nop();
1284 __ lw(AT, thread, in_bytes(JavaThread::suspend_flags_offset()));
1285 __ beq(AT, R0, Continue);
1286 __ delayed()->nop();
1287 __ bind(L);
1288 __ move(A0, thread);
1289 __ call(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
1290 relocInfo::runtime_call_type);
1291 __ delayed()->nop();
1293 #ifndef OPT_THREAD
1294 __ get_thread(thread);
1295 #endif
1296 //add for compressedoops
1297 __ reinit_heapbase();
1298 __ bind(Continue);
1299 }
1301 // change thread state
1302 __ move(t, _thread_in_Java);
1303 __ sw(t, thread, in_bytes(JavaThread::thread_state_offset()));
1304 __ reset_last_Java_frame(thread, true);
1306 // reset handle block
1307 __ ld(t, thread, in_bytes(JavaThread::active_handles_offset()));
1308 __ sw(R0, t, JNIHandleBlock::top_offset_in_bytes());
1310 // If result was an oop then unbox and save it in the frame
1311 {
1312 Label no_oop;
1313 //FIXME, addi only support 16-bit imeditate
1314 __ ld(AT, FP, frame::interpreter_frame_result_handler_offset*wordSize);
1315 __ li(T0, AbstractInterpreter::result_handler(T_OBJECT));
1316 __ bne(AT, T0, no_oop);
1317 __ delayed()->nop();
1318 __ pop(ltos);
1319 // Unbox oop result, e.g. JNIHandles::resolve value.
1320 __ resolve_jobject(V0, thread, T9);
1321 __ sd(V0, FP, (frame::interpreter_frame_oop_temp_offset)*wordSize);
1322 // keep stack depth as expected by pushing oop which will eventually be discarded
1323 __ push(ltos);
1324 __ bind(no_oop);
1325 }
1326 {
1327 Label no_reguard;
1328 __ lw(t, thread, in_bytes(JavaThread::stack_guard_state_offset()));
1329 __ move(AT,(int) JavaThread::stack_guard_yellow_disabled);
1330 __ bne(t, AT, no_reguard);
1331 __ delayed()->nop();
1332 __ pushad();
1333 __ move(S5_heapbase, SP);
1334 __ move(AT, -StackAlignmentInBytes);
1335 __ andr(SP, SP, AT);
1336 __ call(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages), relocInfo::runtime_call_type);
1337 __ delayed()->nop();
1338 __ move(SP, S5_heapbase);
1339 __ popad();
1340 //add for compressedoops
1341 __ reinit_heapbase();
1342 __ bind(no_reguard);
1343 }
1344 // restore BCP to have legal interpreter frame,
1345 // i.e., bci == 0 <=> BCP == code_base()
1346 // Can't call_VM until bcp is within reasonable.
1347 __ get_method(method); // method is junk from thread_in_native to now.
1348 __ verify_oop(method);
1349 __ ld(BCP, method, in_bytes(Method::const_offset()));
1350 __ lea(BCP, Address(BCP, in_bytes(ConstMethod::codes_offset())));
1351 // handle exceptions (exception handling will handle unlocking!)
1352 {
1353 Label L;
1354 __ lw(t, thread, in_bytes(Thread::pending_exception_offset()));
1355 __ beq(t, R0, L);
1356 __ delayed()->nop();
1357 // Note: At some point we may want to unify this with the code used in
1358 // call_VM_base();
1359 // i.e., we should use the StubRoutines::forward_exception code. For now this
1360 // doesn't work here because the sp is not correctly set at this point.
1361 __ MacroAssembler::call_VM(noreg,
1362 CAST_FROM_FN_PTR(address,
1363 InterpreterRuntime::throw_pending_exception));
1364 __ should_not_reach_here();
1365 __ bind(L);
1366 }
1368 // do unlocking if necessary
1369 {
1370 Label L;
1371 __ lw(t, method, in_bytes(Method::access_flags_offset()));
1372 __ andi(t, t, JVM_ACC_SYNCHRONIZED);
1373 __ beq(t, R0, L);
1374 // the code below should be shared with interpreter macro assembler implementation
1375 {
1376 Label unlock;
1377 // BasicObjectLock will be first in list,
1378 // since this is a synchronized method. However, need
1379 // to check that the object has not been unlocked by
1380 // an explicit monitorexit bytecode.
1381 __ delayed()->daddi(c_rarg0, FP, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
1382 // address of first monitor
1384 __ ld(t, c_rarg0, BasicObjectLock::obj_offset_in_bytes());
1385 __ bne(t, R0, unlock);
1386 __ delayed()->nop();
1388 // Entry already unlocked, need to throw exception
1389 __ MacroAssembler::call_VM(NOREG, CAST_FROM_FN_PTR(address,
1390 InterpreterRuntime::throw_illegal_monitor_state_exception));
1391 __ should_not_reach_here();
1393 __ bind(unlock);
1394 __ unlock_object(c_rarg0);
1395 }
1396 __ bind(L);
1397 }
1399 // jvmti/jvmpi support
1400 // Note: This must happen _after_ handling/throwing any exceptions since
1401 // the exception handler code notifies the runtime of method exits
1402 // too. If this happens before, method entry/exit notifications are
1403 // not properly paired (was bug - gri 11/22/99).
1404 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::NotifyJVMTI);
1406 // restore potential result in V0,
1407 // call result handler to restore potential result in ST0 & handle result
1409 __ pop(ltos);
1410 __ pop(dtos);
1412 __ ld(t, FP, (frame::interpreter_frame_result_handler_offset) * wordSize);
1413 __ jalr(t);
1414 __ delayed()->nop();
1417 // remove activation
1418 __ ld(SP, FP, frame::interpreter_frame_sender_sp_offset * wordSize); // get sender sp
1419 __ ld(RA, FP, frame::interpreter_frame_return_addr_offset * wordSize); // get return address
1420 __ ld(FP, FP, frame::interpreter_frame_sender_fp_offset * wordSize); // restore sender's fp
1421 __ jr(RA);
1422 __ delayed()->nop();
1424 #ifndef CORE
1425 if (inc_counter) {
1426 // Handle overflow of counter and compile method
1427 __ bind(invocation_counter_overflow);
1428 generate_counter_overflow(&continue_after_compile);
1429 // entry_point is the beginning of this
1430 // function and checks again for compiled code
1431 }
1432 #endif
1433 return entry_point;
1434 }
1436 //
1437 // Generic interpreted method entry to (asm) interpreter
1438 //
1439 // Layout of frame just at the entry
1440 //
1441 // [ argument word n-1 ] <--- sp
1442 // ...
1443 // [ argument word 0 ]
1444 // assume Method* in Rmethod before call this method.
1445 // prerequisites to the generated stub : the callee Method* in Rmethod
1446 // note you must save the caller bcp before call the generated stub
1447 //
1448 address InterpreterGenerator::generate_normal_entry(bool synchronized) {
1449 // determine code generation flags
1450 bool inc_counter = UseCompiler || CountCompiledCalls;
1452 // Rmethod: Method*
1453 // Rsender: sender 's sp
1454 address entry_point = __ pc();
1456 const Address invocation_counter(Rmethod,
1457 in_bytes(MethodCounters::invocation_counter_offset() + InvocationCounter::counter_offset()));
1459 // get parameter size (always needed)
1460 __ ld(T3, Rmethod, in_bytes(Method::const_offset())); //T3 --> Rmethod._constMethod
1461 __ lhu(V0, T3, in_bytes(ConstMethod::size_of_parameters_offset()));
1463 // Rmethod: Method*
1464 // V0: size of parameters
1465 // Rsender: sender 's sp ,could be different frome sp+ wordSize if we call via c2i
1466 // get size of locals in words to T2
1467 __ lhu(T2, T3, in_bytes(ConstMethod::size_of_locals_offset()));
1468 // T2 = no. of additional locals, locals include parameters
1469 __ dsub(T2, T2, V0);
1471 // see if we've got enough room on the stack for locals plus overhead.
1472 // Layout of frame at this point
1473 //
1474 // [ argument word n-1 ] <--- sp
1475 // ...
1476 // [ argument word 0 ]
1477 generate_stack_overflow_check();
1478 // after this function, the layout of frame does not change
1480 // compute beginning of parameters (LVP)
1481 __ dsll(LVP, V0, LogBytesPerWord);
1482 __ daddiu(LVP, LVP, (-1) * wordSize);
1483 __ dadd(LVP, LVP, SP);
1485 // T2 - # of additional locals
1486 // allocate space for locals
1487 // explicitly initialize locals
1488 {
1489 Label exit, loop;
1490 __ beq(T2, R0, exit);
1491 __ delayed()->nop();
1493 __ bind(loop);
1494 __ sd(R0, SP, -1 * wordSize); // initialize local variables
1495 __ daddiu(T2, T2, -1); // until everything initialized
1496 __ bne(T2, R0, loop);
1497 __ delayed();
1499 __ daddiu(SP, SP, (-1) * wordSize); //fill delay slot
1501 __ bind(exit);
1502 }
1504 //
1505 // [ local var m-1 ] <--- sp
1506 // ...
1507 // [ local var 0 ]
1508 // [ argument word n-1 ] <--- T0?
1509 // ...
1510 // [ argument word 0 ] <--- LVP
1512 // initialize fixed part of activation frame
1514 generate_fixed_frame(false);
1517 // after this function, the layout of frame is as following
1518 //
1519 // [ monitor block top ] <--- sp ( the top monitor entry )
1520 // [ byte code pointer ] (if native, bcp = 0)
1521 // [ constant pool cache ]
1522 // [ Method* ]
1523 // [ locals offset ]
1524 // [ sender's sp ]
1525 // [ sender's fp ] <--- fp
1526 // [ return address ]
1527 // [ local var m-1 ]
1528 // ...
1529 // [ local var 0 ]
1530 // [ argumnet word n-1 ] <--- ( sender's sp )
1531 // ...
1532 // [ argument word 0 ] <--- LVP
1535 // make sure method is not native & not abstract
1536 #ifdef ASSERT
1537 __ ld(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1538 {
1539 Label L;
1540 __ andi(T2, AT, JVM_ACC_NATIVE);
1541 __ beq(T2, R0, L);
1542 __ delayed()->nop();
1543 __ stop("tried to execute native method as non-native");
1544 __ bind(L);
1545 }
1546 {
1547 Label L;
1548 __ andi(T2, AT, JVM_ACC_ABSTRACT);
1549 __ beq(T2, R0, L);
1550 __ delayed()->nop();
1551 __ stop("tried to execute abstract method in interpreter");
1552 __ bind(L);
1553 }
1554 #endif
1556 // Since at this point in the method invocation the exception handler
1557 // would try to exit the monitor of synchronized methods which hasn't
1558 // been entered yet, we set the thread local variable
1559 // _do_not_unlock_if_synchronized to true. The remove_activation will
1560 // check this flag.
1562 #ifndef OPT_THREAD
1563 Register thread = T8;
1564 __ get_thread(thread);
1565 #else
1566 Register thread = TREG;
1567 #endif
1568 __ move(AT, (int)true);
1569 __ sb(AT, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1571 #ifndef CORE
1573 // mdp : T8
1574 // tmp1: T9
1575 // tmp2: T2
1576 __ profile_parameters_type(T8, T9, T2);
1578 // increment invocation count & check for overflow
1579 Label invocation_counter_overflow;
1580 Label profile_method;
1581 Label profile_method_continue;
1582 if (inc_counter) {
1583 generate_counter_incr(&invocation_counter_overflow,
1584 &profile_method,
1585 &profile_method_continue);
1586 if (ProfileInterpreter) {
1587 __ bind(profile_method_continue);
1588 }
1589 }
1591 Label continue_after_compile;
1592 __ bind(continue_after_compile);
1594 #endif // CORE
1596 bang_stack_shadow_pages(false);
1598 // reset the _do_not_unlock_if_synchronized flag
1599 #ifndef OPT_THREAD
1600 __ get_thread(thread);
1601 #endif
1602 __ sb(R0, thread, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
1604 // check for synchronized methods
1605 // Must happen AFTER invocation_counter check and stack overflow check,
1606 // so method is not locked if overflows.
1607 //
1608 if (synchronized) {
1609 // Allocate monitor and lock method
1610 lock_method();
1611 } else {
1612 // no synchronization necessary
1613 #ifdef ASSERT
1614 { Label L;
1615 __ lw(AT, Rmethod, in_bytes(Method::access_flags_offset()));
1616 __ andi(T2, AT, JVM_ACC_SYNCHRONIZED);
1617 __ beq(T2, R0, L);
1618 __ delayed()->nop();
1619 __ stop("method needs synchronization");
1620 __ bind(L);
1621 }
1622 #endif
1623 }
1625 // layout of frame after lock_method
1626 // [ monitor entry ] <--- sp
1627 // ...
1628 // [ monitor entry ]
1629 // [ monitor block top ] ( the top monitor entry )
1630 // [ byte code pointer ] (if native, bcp = 0)
1631 // [ constant pool cache ]
1632 // [ Method* ]
1633 // [ locals offset ]
1634 // [ sender's sp ]
1635 // [ sender's fp ]
1636 // [ return address ] <--- fp
1637 // [ local var m-1 ]
1638 // ...
1639 // [ local var 0 ]
1640 // [ argumnet word n-1 ] <--- ( sender's sp )
1641 // ...
1642 // [ argument word 0 ] <--- LVP
1645 // start execution
1646 #ifdef ASSERT
1647 {
1648 Label L;
1649 __ ld(AT, FP, frame::interpreter_frame_monitor_block_top_offset * wordSize);
1650 __ beq(AT, SP, L);
1651 __ delayed()->nop();
1652 __ stop("broken stack frame setup in interpreter in native");
1653 __ bind(L);
1654 }
1655 #endif
1657 // jvmti/jvmpi support
1658 __ notify_method_entry();
1660 __ dispatch_next(vtos);
1662 // invocation counter overflow
1663 if (inc_counter) {
1664 if (ProfileInterpreter) {
1665 // We have decided to profile this method in the interpreter
1666 __ bind(profile_method);
1667 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1668 InterpreterRuntime::profile_method));
1669 __ set_method_data_pointer_for_bcp();
1670 __ get_method(Rmethod);
1671 __ b(profile_method_continue);
1672 __ delayed()->nop();
1673 }
1674 // Handle overflow of counter and compile method
1675 __ bind(invocation_counter_overflow);
1676 generate_counter_overflow(&continue_after_compile);
1677 }
1679 return entry_point;
1680 }
1682 // Entry points
1683 //
1684 // Here we generate the various kind of entries into the interpreter.
1685 // The two main entry type are generic bytecode methods and native
1686 // call method. These both come in synchronized and non-synchronized
1687 // versions but the frame layout they create is very similar. The
1688 // other method entry types are really just special purpose entries
1689 // that are really entry and interpretation all in one. These are for
1690 // trivial methods like accessor, empty, or special math methods.
1691 //
1692 // When control flow reaches any of the entry types for the interpreter
1693 // the following holds ->
1694 //
1695 // Arguments:
1696 //
1697 // Rmethod: Method*
1698 // V0: receiver
1699 //
1700 //
1701 // Stack layout immediately at entry
1702 //
1703 // [ parameter n-1 ] <--- sp
1704 // ...
1705 // [ parameter 0 ]
1706 // [ expression stack ] (caller's java expression stack)
1708 // Assuming that we don't go to one of the trivial specialized entries
1709 // the stack will look like below when we are ready to execute the
1710 // first bytecode (or call the native routine). The register usage
1711 // will be as the template based interpreter expects (see
1712 // interpreter_mips_64.hpp).
1713 //
1714 // local variables follow incoming parameters immediately; i.e.
1715 // the return address is moved to the end of the locals).
1716 //
1717 // [ monitor entry ] <--- sp
1718 // ...
1719 // [ monitor entry ]
1720 // [ monitor block top ] ( the top monitor entry )
1721 // [ byte code pointer ] (if native, bcp = 0)
1722 // [ constant pool cache ]
1723 // [ Method* ]
1724 // [ locals offset ]
1725 // [ sender's sp ]
1726 // [ sender's fp ]
1727 // [ return address ] <--- fp
1728 // [ local var m-1 ]
1729 // ...
1730 // [ local var 0 ]
1731 // [ argumnet word n-1 ] <--- ( sender's sp )
1732 // ...
1733 // [ argument word 0 ] <--- S7
1735 address AbstractInterpreterGenerator::generate_method_entry(
1736 AbstractInterpreter::MethodKind kind) {
1737 // determine code generation flags
1738 bool synchronized = false;
1739 address entry_point = NULL;
1740 switch (kind) {
1741 case Interpreter::zerolocals :
1742 break;
1743 case Interpreter::zerolocals_synchronized:
1744 synchronized = true;
1745 break;
1746 case Interpreter::native :
1747 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false);
1748 break;
1749 case Interpreter::native_synchronized :
1750 entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true);
1751 break;
1752 case Interpreter::empty :
1753 entry_point = ((InterpreterGenerator*)this)->generate_empty_entry();
1754 break;
1755 case Interpreter::accessor :
1756 entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry();
1757 break;
1758 case Interpreter::abstract :
1759 entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry();
1760 break;
1762 case Interpreter::java_lang_math_sin : // fall thru
1763 case Interpreter::java_lang_math_cos : // fall thru
1764 case Interpreter::java_lang_math_tan : // fall thru
1765 case Interpreter::java_lang_math_log : // fall thru
1766 case Interpreter::java_lang_math_log10 : // fall thru
1767 case Interpreter::java_lang_math_pow : // fall thru
1768 case Interpreter::java_lang_math_exp : break;
1769 case Interpreter::java_lang_math_abs : // fall thru
1770 case Interpreter::java_lang_math_sqrt :
1771 entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
1772 case Interpreter::java_lang_ref_reference_get:
1773 entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
1774 default:
1775 fatal(err_msg("unexpected method kind: %d", kind));
1776 break;
1777 }
1778 if (entry_point) return entry_point;
1780 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
1781 }
1783 // These should never be compiled since the interpreter will prefer
1784 // the compiled version to the intrinsic version.
1785 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
1786 switch (method_kind(m)) {
1787 case Interpreter::java_lang_math_sin : // fall thru
1788 case Interpreter::java_lang_math_cos : // fall thru
1789 case Interpreter::java_lang_math_tan : // fall thru
1790 case Interpreter::java_lang_math_abs : // fall thru
1791 case Interpreter::java_lang_math_log : // fall thru
1792 case Interpreter::java_lang_math_log10 : // fall thru
1793 case Interpreter::java_lang_math_sqrt : // fall thru
1794 case Interpreter::java_lang_math_pow : // fall thru
1795 case Interpreter::java_lang_math_exp :
1796 return false;
1797 default:
1798 return true;
1799 }
1800 }
1802 // How much stack a method activation needs in words.
1803 int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
1805 const int entry_size = frame::interpreter_frame_monitor_size();
1807 // total overhead size: entry_size + (saved fp thru expr stack bottom).
1808 // be sure to change this if you add/subtract anything to/from the overhead area
1809 const int overhead_size = -(frame::interpreter_frame_initial_sp_offset) + entry_size;
1811 const int stub_code = 6; // see generate_call_stub
1812 // return overhead_size + method->max_locals() + method->max_stack() + stub_code;
1813 const int method_stack = (method->max_locals() + method->max_stack()) *
1814 Interpreter::stackElementWords;
1815 return overhead_size + method_stack + stub_code;
1816 }
1818 void AbstractInterpreter::layout_activation(Method* method,
1819 int tempcount,
1820 int popframe_extra_args,
1821 int moncount,
1822 int caller_actual_parameters,
1823 int callee_param_count,
1824 int callee_locals,
1825 frame* caller,
1826 frame* interpreter_frame,
1827 bool is_top_frame,
1828 bool is_bottom_frame) {
1829 // Note: This calculation must exactly parallel the frame setup
1830 // in AbstractInterpreterGenerator::generate_method_entry.
1831 // If interpreter_frame!=NULL, set up the method, locals, and monitors.
1832 // The frame interpreter_frame, if not NULL, is guaranteed to be the
1833 // right size, as determined by a previous call to this method.
1834 // It is also guaranteed to be walkable even though it is in a skeletal state
1836 // fixed size of an interpreter frame:
1838 int max_locals = method->max_locals() * Interpreter::stackElementWords;
1839 int extra_locals = (method->max_locals() - method->size_of_parameters()) * Interpreter::stackElementWords;
1841 #ifdef ASSERT
1842 if (!EnableInvokeDynamic) {
1843 // @@@ FIXME: Should we correct interpreter_frame_sender_sp in the calling sequences?
1844 // Probably, since deoptimization doesn't work yet.
1845 assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
1846 }
1847 assert(caller->sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable(2)");
1848 #endif
1850 interpreter_frame->interpreter_frame_set_method(method);
1851 // NOTE the difference in using sender_sp and interpreter_frame_sender_sp
1852 // interpreter_frame_sender_sp is the original sp of the caller (the unextended_sp)
1853 // and sender_sp is fp+8
1854 intptr_t* locals = interpreter_frame->sender_sp() + max_locals - 1;
1856 #ifdef ASSERT
1857 if (caller->is_interpreted_frame()) {
1858 assert(locals < caller->fp() + frame::interpreter_frame_initial_sp_offset, "bad placement");
1859 }
1860 #endif
1862 interpreter_frame->interpreter_frame_set_locals(locals);
1863 BasicObjectLock* montop = interpreter_frame->interpreter_frame_monitor_begin();
1864 BasicObjectLock* monbot = montop - moncount;
1865 interpreter_frame->interpreter_frame_set_monitor_end(montop - moncount);
1867 //set last sp;
1868 intptr_t* sp = (intptr_t*) monbot - tempcount*Interpreter::stackElementWords -
1869 popframe_extra_args;
1870 interpreter_frame->interpreter_frame_set_last_sp(sp);
1871 // All frames but the initial interpreter frame we fill in have a
1872 // value for sender_sp that allows walking the stack but isn't
1873 // truly correct. Correct the value here.
1874 //
1875 if (extra_locals != 0 &&
1876 interpreter_frame->sender_sp() == interpreter_frame->interpreter_frame_sender_sp() ) {
1877 interpreter_frame->set_interpreter_frame_sender_sp(caller->sp() + extra_locals);
1878 }
1879 *interpreter_frame->interpreter_frame_cache_addr() = method->constants()->cache();
1880 }
1882 //-----------------------------------------------------------------------------
1883 // Exceptions
1885 void TemplateInterpreterGenerator::generate_throw_exception() {
1886 // Entry point in previous activation (i.e., if the caller was
1887 // interpreted)
1888 Interpreter::_rethrow_exception_entry = __ pc();
1889 // Restore sp to interpreter_frame_last_sp even though we are going
1890 // to empty the expression stack for the exception processing.
1891 __ sd(R0,FP, frame::interpreter_frame_last_sp_offset * wordSize);
1893 // V0: exception
1894 // V1: return address/pc that threw exception
1895 __ restore_bcp(); // BCP points to call/send
1896 __ restore_locals();
1898 //add for compressedoops
1899 __ reinit_heapbase();
1900 // Entry point for exceptions thrown within interpreter code
1901 Interpreter::_throw_exception_entry = __ pc();
1902 // expression stack is undefined here
1903 // V0: exception
1904 // BCP: exception bcp
1905 __ verify_oop(V0);
1907 // expression stack must be empty before entering the VM in case of an exception
1908 __ empty_expression_stack();
1909 // find exception handler address and preserve exception oop
1910 __ move(A1, V0);
1911 __ call_VM(V1, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), A1);
1912 // V0: exception handler entry point
1913 // V1: preserved exception oop
1914 // S0: bcp for exception handler
1915 __ daddi(SP, SP, (-1) * wordSize);
1916 __ sd(V1, SP, 0); // push exception which is now the only value on the stack
1917 __ jr(V0); // jump to exception handler (may be _remove_activation_entry!)
1918 __ delayed()->nop();
1920 // If the exception is not handled in the current frame the frame is removed and
1921 // the exception is rethrown (i.e. exception continuation is _rethrow_exception).
1922 //
1923 // Note: At this point the bci is still the bxi for the instruction which caused
1924 // the exception and the expression stack is empty. Thus, for any VM calls
1925 // at this point, GC will find a legal oop map (with empty expression stack).
1927 // In current activation
1928 // V0: exception
1929 // BCP: exception bcp
1931 //
1932 // JVMTI PopFrame support
1933 //
1935 Interpreter::_remove_activation_preserving_args_entry = __ pc();
1936 __ empty_expression_stack();
1937 // Set the popframe_processing bit in pending_popframe_condition indicating that we are
1938 // currently handling popframe, so that call_VMs that may happen later do not trigger new
1939 // popframe handling cycles.
1940 #ifndef OPT_THREAD
1941 Register thread = T2;
1942 __ get_thread(T2);
1943 #else
1944 Register thread = TREG;
1945 #endif
1946 __ lw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1947 __ ori(T3, T3, JavaThread::popframe_processing_bit);
1948 __ sw(T3, thread, in_bytes(JavaThread::popframe_condition_offset()));
1950 #ifndef CORE
1951 {
1952 // Check to see whether we are returning to a deoptimized frame.
1953 // (The PopFrame call ensures that the caller of the popped frame is
1954 // either interpreted or compiled and deoptimizes it if compiled.)
1955 // In this case, we can't call dispatch_next() after the frame is
1956 // popped, but instead must save the incoming arguments and restore
1957 // them after deoptimization has occurred.
1958 //
1959 // Note that we don't compare the return PC against the
1960 // deoptimization blob's unpack entry because of the presence of
1961 // adapter frames in C2.
1962 Label caller_not_deoptimized;
1963 __ ld(A0, FP, frame::return_addr_offset * wordSize);
1964 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), A0);
1965 __ bne(V0, R0, caller_not_deoptimized);
1966 __ delayed()->nop();
1968 // Compute size of arguments for saving when returning to deoptimized caller
1969 __ get_method(A1);
1970 __ verify_oop(A1);
1971 __ ld(A1, A1, in_bytes(Method::const_offset()));
1972 __ lhu(A1, A1, in_bytes(ConstMethod::size_of_parameters_offset()));
1973 __ shl(A1, Interpreter::logStackElementSize);
1974 __ restore_locals();
1975 __ dsub(A2, LVP, A1);
1976 __ daddiu(A2, A2, wordSize);
1977 // Save these arguments
1978 #ifndef OPT_THREAD
1979 __ get_thread(A0);
1980 #else
1981 __ move(A0, TREG);
1982 #endif
1983 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), A0, A1, A2);
1985 __ remove_activation(vtos, T9, false, false, false);
1987 // Inform deoptimization that it is responsible for restoring these arguments
1988 #ifndef OPT_THREAD
1989 __ get_thread(thread);
1990 #endif
1991 __ move(AT, JavaThread::popframe_force_deopt_reexecution_bit);
1992 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
1993 // Continue in deoptimization handler
1994 __ jr(T9);
1995 __ delayed()->nop();
1997 __ bind(caller_not_deoptimized);
1998 }
1999 #endif /* !CORE */
2001 __ remove_activation(vtos, T3,
2002 /* throw_monitor_exception */ false,
2003 /* install_monitor_exception */ false,
2004 /* notify_jvmdi */ false);
2006 // Clear the popframe condition flag
2007 // Finish with popframe handling
2008 // A previous I2C followed by a deoptimization might have moved the
2009 // outgoing arguments further up the stack. PopFrame expects the
2010 // mutations to those outgoing arguments to be preserved and other
2011 // constraints basically require this frame to look exactly as
2012 // though it had previously invoked an interpreted activation with
2013 // no space between the top of the expression stack (current
2014 // last_sp) and the top of stack. Rather than force deopt to
2015 // maintain this kind of invariant all the time we call a small
2016 // fixup routine to move the mutated arguments onto the top of our
2017 // expression stack if necessary.
2018 __ move(T8, SP);
2019 __ ld(A2, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2020 #ifndef OPT_THREAD
2021 __ get_thread(thread);
2022 #endif
2023 // PC must point into interpreter here
2024 __ set_last_Java_frame(thread, noreg, FP, __ pc());
2025 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::popframe_move_outgoing_args), thread, T8, A2);
2026 __ reset_last_Java_frame(thread, true);
2027 // Restore the last_sp and null it out
2028 __ ld(SP, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2029 __ sd(R0, FP, frame::interpreter_frame_last_sp_offset * wordSize);
2033 __ move(AT, JavaThread::popframe_inactive);
2034 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2036 // Finish with popframe handling
2037 __ restore_bcp();
2038 __ restore_locals();
2039 #ifndef CORE
2040 // The method data pointer was incremented already during
2041 // call profiling. We have to restore the mdp for the current bcp.
2042 if (ProfileInterpreter) {
2043 __ set_method_data_pointer_for_bcp();
2044 }
2045 #endif // !CORE
2046 // Clear the popframe condition flag
2047 #ifndef OPT_THREAD
2048 __ get_thread(thread);
2049 #endif
2050 __ move(AT, JavaThread::popframe_inactive);
2051 __ sw(AT, thread, in_bytes(JavaThread::popframe_condition_offset()));
2053 #if INCLUDE_JVMTI
2054 {
2055 Label L_done;
2057 __ lbu(AT, BCP, 0);
2058 __ daddiu(AT, AT, -1 * Bytecodes::_invokestatic);
2059 __ bne(AT, R0, L_done);
2060 __ delayed()->nop();
2062 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call.
2063 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL.
2065 __ get_method(T9);
2066 __ ld(T8, LVP, 0);
2067 __ call_VM(T8, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), T8, T9, BCP);
2069 __ beq(T8, R0, L_done);
2070 __ delayed()->nop();
2072 __ sd(T8, SP, 0);
2073 __ bind(L_done);
2074 }
2075 #endif // INCLUDE_JVMTI
2077 __ dispatch_next(vtos);
2078 // end of PopFrame support
2080 Interpreter::_remove_activation_entry = __ pc();
2082 // preserve exception over this code sequence
2083 __ ld(T0, SP, 0);
2084 __ daddi(SP, SP, wordSize);
2085 #ifndef OPT_THREAD
2086 __ get_thread(thread);
2087 #endif
2088 __ sd(T0, thread, in_bytes(JavaThread::vm_result_offset()));
2089 // remove the activation (without doing throws on illegalMonitorExceptions)
2090 __ remove_activation(vtos, T3, false, true, false);
2091 // restore exception
2092 __ get_vm_result(T0, thread);
2093 __ verify_oop(T0);
2095 // In between activations - previous activation type unknown yet
2096 // compute continuation point - the continuation point expects
2097 // the following registers set up:
2098 //
2099 // T0: exception
2100 // T1: return address/pc that threw exception
2101 // SP: expression stack of caller
2102 // FP: fp of caller
2103 __ daddi(SP, SP, (-2) * wordSize);
2104 __ sd(T0, SP, wordSize); // save exception
2105 __ sd(T3, SP, 0); // save return address
2106 __ move(A1, T3);
2107 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, A1);
2108 __ move(T9, V0); // save exception handler
2109 __ ld(V0, SP, wordSize); // restore exception
2110 __ ld(V1, SP, 0); // restore return address
2111 __ daddi(SP, SP, 2 * wordSize);
2113 // Note that an "issuing PC" is actually the next PC after the call
2114 __ jr(T9); // jump to exception handler of caller
2115 __ delayed()->nop();
2116 }
2119 //
2120 // JVMTI ForceEarlyReturn support
2121 //
2122 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) {
2123 address entry = __ pc();
2124 __ restore_bcp();
2125 __ restore_locals();
2126 __ empty_expression_stack();
2127 __ empty_FPU_stack();
2128 __ load_earlyret_value(state);
2130 #ifndef OPT_THREAD
2131 __ get_thread(TREG);
2132 #endif
2133 __ ld_ptr(T9, TREG, in_bytes(JavaThread::jvmti_thread_state_offset()));
2134 const Address cond_addr(T9, in_bytes(JvmtiThreadState::earlyret_state_offset()));
2135 // Clear the earlyret state
2136 __ move(AT, JvmtiThreadState::earlyret_inactive);
2137 __ sw(AT, cond_addr);
2138 __ sync();
2141 __ remove_activation(state, T0,
2142 false, /* throw_monitor_exception */
2143 false, /* install_monitor_exception */
2144 true); /* notify_jvmdi */
2145 __ sync();
2146 __ jr(T0);
2147 __ delayed()->nop();
2148 return entry;
2149 } // end of ForceEarlyReturn support
2152 //-----------------------------------------------------------------------------
2153 // Helper for vtos entry point generation
2155 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t,
2156 address& bep,
2157 address& cep,
2158 address& sep,
2159 address& aep,
2160 address& iep,
2161 address& lep,
2162 address& fep,
2163 address& dep,
2164 address& vep) {
2165 assert(t->is_valid() && t->tos_in() == vtos, "illegal template");
2166 Label L;
2167 fep = __ pc(); __ push(ftos); __ b(L); __ delayed()->nop();
2168 dep = __ pc(); __ push(dtos); __ b(L); __ delayed()->nop();
2169 lep = __ pc(); __ push(ltos); __ b(L); __ delayed()->nop();
2170 aep =__ pc(); __ push(atos); __ b(L); __ delayed()->nop();
2171 bep = cep = sep =
2172 iep = __ pc(); __ push(itos);
2173 vep = __ pc();
2174 __ bind(L);
2175 generate_and_dispatch(t);
2176 }
2179 //-----------------------------------------------------------------------------
2180 // Generation of individual instructions
2182 // helpers for generate_and_dispatch
2185 InterpreterGenerator::InterpreterGenerator(StubQueue* code)
2186 : TemplateInterpreterGenerator(code) {
2187 generate_all(); // down here so it can be "virtual"
2188 }
2190 //-----------------------------------------------------------------------------
2192 // Non-product code
2193 #ifndef PRODUCT
2194 address TemplateInterpreterGenerator::generate_trace_code(TosState state) {
2195 address entry = __ pc();
2197 // prepare expression stack
2198 __ push(state); // save tosca
2200 // tos & tos2
2201 // trace_bytecode need actually 4 args, the last two is tos&tos2
2202 // this work fine for x86. but mips o32 call convention will store A2-A3
2203 // to the stack position it think is the tos&tos2
2204 // when the expression stack have no more than 2 data, error occur.
2205 __ ld(A2, SP, 0);
2206 __ ld(A3, SP, 1 * wordSize);
2208 // pass arguments & call tracer
2209 __ call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::trace_bytecode), RA, A2, A3);
2210 __ move(RA, V0); // make sure return address is not destroyed by pop(state)
2212 // restore expression stack
2213 __ pop(state); // restore tosca
2215 // return
2216 __ jr(RA);
2217 __ delayed()->nop();
2219 return entry;
2220 }
2222 void TemplateInterpreterGenerator::count_bytecode() {
2223 __ li(T8, (long)&BytecodeCounter::_counter_value);
2224 __ lw(AT, T8, 0);
2225 __ daddi(AT, AT, 1);
2226 __ sw(AT, T8, 0);
2227 }
2229 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) {
2230 __ li(T8, (long)&BytecodeHistogram::_counters[t->bytecode()]);
2231 __ lw(AT, T8, 0);
2232 __ daddi(AT, AT, 1);
2233 __ sw(AT, T8, 0);
2234 }
2236 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) {
2237 __ li(T8, (long)&BytecodePairHistogram::_index);
2238 __ lw(T9, T8, 0);
2239 __ dsrl(T9, T9, BytecodePairHistogram::log2_number_of_codes);
2240 __ li(T8, ((long)t->bytecode()) << BytecodePairHistogram::log2_number_of_codes);
2241 __ orr(T9, T9, T8);
2242 __ li(T8, (long)&BytecodePairHistogram::_index);
2243 __ sw(T9, T8, 0);
2244 __ dsll(T9, T9, 2);
2245 __ li(T8, (long)BytecodePairHistogram::_counters);
2246 __ dadd(T8, T8, T9);
2247 __ lw(AT, T8, 0);
2248 __ daddi(AT, AT, 1);
2249 __ sw(AT, T8, 0);
2250 }
2253 void TemplateInterpreterGenerator::trace_bytecode(Template* t) {
2254 // Call a little run-time stub to avoid blow-up for each bytecode.
2255 // The run-time runtime saves the right registers, depending on
2256 // the tosca in-state for the given template.
2258 address entry = Interpreter::trace_code(t->tos_in());
2259 assert(entry != NULL, "entry must have been generated");
2260 __ call(entry, relocInfo::none);
2261 __ delayed()->nop();
2262 //add for compressedoops
2263 __ reinit_heapbase();
2264 }
2267 void TemplateInterpreterGenerator::stop_interpreter_at() {
2268 Label L;
2269 __ li(T8, long(&BytecodeCounter::_counter_value));
2270 __ lw(T8, T8, 0);
2271 __ move(AT, StopInterpreterAt);
2272 __ bne(T8, AT, L);
2273 __ delayed()->nop();
2274 __ call(CAST_FROM_FN_PTR(address, os::breakpoint), relocInfo::runtime_call_type);
2275 __ delayed()->nop();
2276 __ bind(L);
2277 }
2278 #endif // !PRODUCT
2279 #endif // ! CC_INTERP