1 /*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_sparc.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/method.hpp"
33 #include "oops/methodCounters.hpp"
34 #include "prims/jvmtiExport.hpp"
35 #include "prims/jvmtiRedefineClassesTrace.hpp"
36 #include "prims/jvmtiThreadState.hpp"
37 #include "runtime/basicLock.hpp"
38 #include "runtime/biasedLocking.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/thread.inline.hpp"
41
42 #ifndef CC_INTERP
43 #ifndef FAST_DISPATCH
44 #define FAST_DISPATCH 1
45 #endif
46 #undef FAST_DISPATCH
47
48 // Implementation of InterpreterMacroAssembler
49
50 // This file specializes the assember with interpreter-specific macros
51
52 const Address InterpreterMacroAssembler::l_tmp(FP, (frame::interpreter_frame_l_scratch_fp_offset * wordSize) + STACK_BIAS);
53 const Address InterpreterMacroAssembler::d_tmp(FP, (frame::interpreter_frame_d_scratch_fp_offset * wordSize) + STACK_BIAS);
54
55 #else // CC_INTERP
56 #ifndef STATE
57 #define STATE(field_name) Lstate, in_bytes(byte_offset_of(BytecodeInterpreter, field_name))
58 #endif // STATE
59
60 #endif // CC_INTERP
61
62 void InterpreterMacroAssembler::compute_extra_locals_size_in_bytes(Register args_size, Register locals_size, Register delta) {
63 // Note: this algorithm is also used by C1's OSR entry sequence.
64 // Any changes should also be applied to CodeEmitter::emit_osr_entry().
65 assert_different_registers(args_size, locals_size);
66 // max_locals*2 for TAGS. Assumes that args_size has already been adjusted.
67 subcc(locals_size, args_size, delta);// extra space for non-arguments locals in words
68 // Use br/mov combination because it works on both V8 and V9 and is
69 // faster.
70 Label skip_move;
71 br(Assembler::negative, true, Assembler::pt, skip_move);
72 delayed()->mov(G0, delta);
73 bind(skip_move);
74 round_to(delta, WordsPerLong); // make multiple of 2 (SP must be 2-word aligned)
75 sll(delta, LogBytesPerWord, delta); // extra space for locals in bytes
76 }
77
78 #ifndef CC_INTERP
79
80 // Dispatch code executed in the prolog of a bytecode which does not do it's
81 // own dispatch. The dispatch address is computed and placed in IdispatchAddress
82 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) {
83 assert_not_delayed();
84 #ifdef FAST_DISPATCH
85 // FAST_DISPATCH and ProfileInterpreter are mutually exclusive since
86 // they both use I2.
87 assert(!ProfileInterpreter, "FAST_DISPATCH and +ProfileInterpreter are mutually exclusive");
88 ldub(Lbcp, bcp_incr, Lbyte_code); // load next bytecode
89 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
90 // add offset to correct dispatch table
91 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
92 ld_ptr(IdispatchTables, Lbyte_code, IdispatchAddress);// get entry addr
93 #else
94 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
95 // dispatch table to use
96 AddressLiteral tbl(Interpreter::dispatch_table(state));
97 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
98 set(tbl, G3_scratch); // compute addr of table
99 ld_ptr(G3_scratch, Lbyte_code, IdispatchAddress); // get entry addr
100 #endif
101 }
102
103
104 // Dispatch code executed in the epilog of a bytecode which does not do it's
105 // own dispatch. The dispatch address in IdispatchAddress is used for the
106 // dispatch.
107 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) {
108 assert_not_delayed();
109 verify_FPU(1, state);
110 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
111 jmp( IdispatchAddress, 0 );
112 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
113 else delayed()->nop();
114 }
115
116
117 void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) {
118 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
119 assert_not_delayed();
120 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
121 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr);
122 }
123
124
125 void InterpreterMacroAssembler::dispatch_next_noverify_oop(TosState state, int bcp_incr) {
126 // %%%% consider branching to a single shared dispatch stub (for each bcp_incr)
127 assert_not_delayed();
128 ldub( Lbcp, bcp_incr, Lbyte_code); // load next bytecode
129 dispatch_Lbyte_code(state, Interpreter::dispatch_table(state), bcp_incr, false);
130 }
131
132
133 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
134 // load current bytecode
135 assert_not_delayed();
136 ldub( Lbcp, 0, Lbyte_code); // load next bytecode
137 dispatch_base(state, table);
138 }
139
140
141 void InterpreterMacroAssembler::call_VM_leaf_base(
142 Register java_thread,
143 address entry_point,
144 int number_of_arguments
145 ) {
146 if (!java_thread->is_valid())
147 java_thread = L7_thread_cache;
148 // super call
149 MacroAssembler::call_VM_leaf_base(java_thread, entry_point, number_of_arguments);
150 }
151
152
153 void InterpreterMacroAssembler::call_VM_base(
154 Register oop_result,
155 Register java_thread,
156 Register last_java_sp,
157 address entry_point,
158 int number_of_arguments,
159 bool check_exception
160 ) {
161 if (!java_thread->is_valid())
162 java_thread = L7_thread_cache;
163 // See class ThreadInVMfromInterpreter, which assumes that the interpreter
164 // takes responsibility for setting its own thread-state on call-out.
165 // However, ThreadInVMfromInterpreter resets the state to "in_Java".
166
167 //save_bcp(); // save bcp
168 MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exception);
169 //restore_bcp(); // restore bcp
170 //restore_locals(); // restore locals pointer
171 }
172
173
174 void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) {
175 if (JvmtiExport::can_pop_frame()) {
176 Label L;
177
178 // Check the "pending popframe condition" flag in the current thread
179 ld(G2_thread, JavaThread::popframe_condition_offset(), scratch_reg);
180
181 // Initiate popframe handling only if it is not already being processed. If the flag
182 // has the popframe_processing bit set, it means that this code is called *during* popframe
183 // handling - we don't want to reenter.
184 btst(JavaThread::popframe_pending_bit, scratch_reg);
185 br(zero, false, pt, L);
186 delayed()->nop();
187 btst(JavaThread::popframe_processing_bit, scratch_reg);
188 br(notZero, false, pt, L);
189 delayed()->nop();
190
191 // Call Interpreter::remove_activation_preserving_args_entry() to get the
192 // address of the same-named entrypoint in the generated interpreter code.
193 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
194
195 // Jump to Interpreter::_remove_activation_preserving_args_entry
196 jmpl(O0, G0, G0);
197 delayed()->nop();
198 bind(L);
199 }
200 }
201
202
203 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
204 Register thr_state = G4_scratch;
205 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
206 const Address tos_addr(thr_state, JvmtiThreadState::earlyret_tos_offset());
207 const Address oop_addr(thr_state, JvmtiThreadState::earlyret_oop_offset());
208 const Address val_addr(thr_state, JvmtiThreadState::earlyret_value_offset());
209 switch (state) {
210 case ltos: ld_long(val_addr, Otos_l); break;
211 case atos: ld_ptr(oop_addr, Otos_l);
212 st_ptr(G0, oop_addr); break;
213 case btos: // fall through
214 case ctos: // fall through
215 case stos: // fall through
216 case itos: ld(val_addr, Otos_l1); break;
217 case ftos: ldf(FloatRegisterImpl::S, val_addr, Ftos_f); break;
218 case dtos: ldf(FloatRegisterImpl::D, val_addr, Ftos_d); break;
219 case vtos: /* nothing to do */ break;
220 default : ShouldNotReachHere();
221 }
222 // Clean up tos value in the jvmti thread state
223 or3(G0, ilgl, G3_scratch);
224 stw(G3_scratch, tos_addr);
225 st_long(G0, val_addr);
226 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
227 }
228
229
230 void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) {
231 if (JvmtiExport::can_force_early_return()) {
232 Label L;
233 Register thr_state = G3_scratch;
234 ld_ptr(G2_thread, JavaThread::jvmti_thread_state_offset(), thr_state);
235 br_null_short(thr_state, pt, L); // if (thread->jvmti_thread_state() == NULL) exit;
236
237 // Initiate earlyret handling only if it is not already being processed.
238 // If the flag has the earlyret_processing bit set, it means that this code
239 // is called *during* earlyret handling - we don't want to reenter.
240 ld(thr_state, JvmtiThreadState::earlyret_state_offset(), G4_scratch);
241 cmp_and_br_short(G4_scratch, JvmtiThreadState::earlyret_pending, Assembler::notEqual, pt, L);
242
243 // Call Interpreter::remove_activation_early_entry() to get the address of the
244 // same-named entrypoint in the generated interpreter code
245 ld(thr_state, JvmtiThreadState::earlyret_tos_offset(), Otos_l1);
246 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), Otos_l1);
247
248 // Jump to Interpreter::_remove_activation_early_entry
249 jmpl(O0, G0, G0);
250 delayed()->nop();
251 bind(L);
252 }
253 }
254
255
256 void InterpreterMacroAssembler::super_call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) {
257 mov(arg_1, O0);
258 mov(arg_2, O1);
259 MacroAssembler::call_VM_leaf_base(thread_cache, entry_point, 2);
260 }
261 #endif /* CC_INTERP */
262
263
264 #ifndef CC_INTERP
265
266 void InterpreterMacroAssembler::dispatch_base(TosState state, address* table) {
267 assert_not_delayed();
268 dispatch_Lbyte_code(state, table);
269 }
270
271
272 void InterpreterMacroAssembler::dispatch_normal(TosState state) {
273 dispatch_base(state, Interpreter::normal_table(state));
274 }
275
276
277 void InterpreterMacroAssembler::dispatch_only(TosState state) {
278 dispatch_base(state, Interpreter::dispatch_table(state));
279 }
280
281
282 // common code to dispatch and dispatch_only
283 // dispatch value in Lbyte_code and increment Lbcp
284
285 void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, address* table, int bcp_incr, bool verify) {
286 verify_FPU(1, state);
287 // %%%%% maybe implement +VerifyActivationFrameSize here
288 //verify_thread(); //too slow; we will just verify on method entry & exit
289 if (verify) interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
290 #ifdef FAST_DISPATCH
291 if (table == Interpreter::dispatch_table(state)) {
292 // use IdispatchTables
293 add(Lbyte_code, Interpreter::distance_from_dispatch_table(state), Lbyte_code);
294 // add offset to correct dispatch table
295 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
296 ld_ptr(IdispatchTables, Lbyte_code, G3_scratch); // get entry addr
297 } else {
298 #endif
299 // dispatch table to use
300 AddressLiteral tbl(table);
301 sll(Lbyte_code, LogBytesPerWord, Lbyte_code); // multiply by wordSize
302 set(tbl, G3_scratch); // compute addr of table
303 ld_ptr(G3_scratch, Lbyte_code, G3_scratch); // get entry addr
304 #ifdef FAST_DISPATCH
305 }
306 #endif
307 jmp( G3_scratch, 0 );
308 if (bcp_incr != 0) delayed()->inc(Lbcp, bcp_incr);
309 else delayed()->nop();
310 }
311
312
313 // Helpers for expression stack
314
315 // Longs and doubles are Category 2 computational types in the
316 // JVM specification (section 3.11.1) and take 2 expression stack or
317 // local slots.
318 // Aligning them on 32 bit with tagged stacks is hard because the code generated
319 // for the dup* bytecodes depends on what types are already on the stack.
320 // If the types are split into the two stack/local slots, that is much easier
321 // (and we can use 0 for non-reference tags).
322
323 // Known good alignment in _LP64 but unknown otherwise
324 void InterpreterMacroAssembler::load_unaligned_double(Register r1, int offset, FloatRegister d) {
325 assert_not_delayed();
326
327 #ifdef _LP64
328 ldf(FloatRegisterImpl::D, r1, offset, d);
329 #else
330 ldf(FloatRegisterImpl::S, r1, offset, d);
331 ldf(FloatRegisterImpl::S, r1, offset + Interpreter::stackElementSize, d->successor());
332 #endif
333 }
334
335 // Known good alignment in _LP64 but unknown otherwise
336 void InterpreterMacroAssembler::store_unaligned_double(FloatRegister d, Register r1, int offset) {
337 assert_not_delayed();
338
339 #ifdef _LP64
340 stf(FloatRegisterImpl::D, d, r1, offset);
341 // store something more useful here
342 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
343 #else
344 stf(FloatRegisterImpl::S, d, r1, offset);
345 stf(FloatRegisterImpl::S, d->successor(), r1, offset + Interpreter::stackElementSize);
346 #endif
347 }
348
349
350 // Known good alignment in _LP64 but unknown otherwise
351 void InterpreterMacroAssembler::load_unaligned_long(Register r1, int offset, Register rd) {
352 assert_not_delayed();
353 #ifdef _LP64
354 ldx(r1, offset, rd);
355 #else
356 ld(r1, offset, rd);
357 ld(r1, offset + Interpreter::stackElementSize, rd->successor());
358 #endif
359 }
360
361 // Known good alignment in _LP64 but unknown otherwise
362 void InterpreterMacroAssembler::store_unaligned_long(Register l, Register r1, int offset) {
363 assert_not_delayed();
364
365 #ifdef _LP64
366 stx(l, r1, offset);
367 // store something more useful here
368 debug_only(stx(G0, r1, offset+Interpreter::stackElementSize);)
369 #else
370 st(l, r1, offset);
371 st(l->successor(), r1, offset + Interpreter::stackElementSize);
372 #endif
373 }
374
375 void InterpreterMacroAssembler::pop_i(Register r) {
376 assert_not_delayed();
377 ld(Lesp, Interpreter::expr_offset_in_bytes(0), r);
378 inc(Lesp, Interpreter::stackElementSize);
379 debug_only(verify_esp(Lesp));
380 }
381
382 void InterpreterMacroAssembler::pop_ptr(Register r, Register scratch) {
383 assert_not_delayed();
384 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(0), r);
385 inc(Lesp, Interpreter::stackElementSize);
386 debug_only(verify_esp(Lesp));
387 }
388
389 void InterpreterMacroAssembler::pop_l(Register r) {
390 assert_not_delayed();
391 load_unaligned_long(Lesp, Interpreter::expr_offset_in_bytes(0), r);
392 inc(Lesp, 2*Interpreter::stackElementSize);
393 debug_only(verify_esp(Lesp));
394 }
395
396
397 void InterpreterMacroAssembler::pop_f(FloatRegister f, Register scratch) {
398 assert_not_delayed();
399 ldf(FloatRegisterImpl::S, Lesp, Interpreter::expr_offset_in_bytes(0), f);
400 inc(Lesp, Interpreter::stackElementSize);
401 debug_only(verify_esp(Lesp));
402 }
403
404
405 void InterpreterMacroAssembler::pop_d(FloatRegister f, Register scratch) {
406 assert_not_delayed();
407 load_unaligned_double(Lesp, Interpreter::expr_offset_in_bytes(0), f);
408 inc(Lesp, 2*Interpreter::stackElementSize);
409 debug_only(verify_esp(Lesp));
410 }
411
412
413 void InterpreterMacroAssembler::push_i(Register r) {
414 assert_not_delayed();
415 debug_only(verify_esp(Lesp));
416 st(r, Lesp, 0);
417 dec(Lesp, Interpreter::stackElementSize);
418 }
419
420 void InterpreterMacroAssembler::push_ptr(Register r) {
421 assert_not_delayed();
422 st_ptr(r, Lesp, 0);
423 dec(Lesp, Interpreter::stackElementSize);
424 }
425
426 // remember: our convention for longs in SPARC is:
427 // O0 (Otos_l1) has high-order part in first word,
428 // O1 (Otos_l2) has low-order part in second word
429
430 void InterpreterMacroAssembler::push_l(Register r) {
431 assert_not_delayed();
432 debug_only(verify_esp(Lesp));
433 // Longs are stored in memory-correct order, even if unaligned.
434 int offset = -Interpreter::stackElementSize;
435 store_unaligned_long(r, Lesp, offset);
436 dec(Lesp, 2 * Interpreter::stackElementSize);
437 }
438
439
440 void InterpreterMacroAssembler::push_f(FloatRegister f) {
441 assert_not_delayed();
442 debug_only(verify_esp(Lesp));
443 stf(FloatRegisterImpl::S, f, Lesp, 0);
444 dec(Lesp, Interpreter::stackElementSize);
445 }
446
447
448 void InterpreterMacroAssembler::push_d(FloatRegister d) {
449 assert_not_delayed();
450 debug_only(verify_esp(Lesp));
451 // Longs are stored in memory-correct order, even if unaligned.
452 int offset = -Interpreter::stackElementSize;
453 store_unaligned_double(d, Lesp, offset);
454 dec(Lesp, 2 * Interpreter::stackElementSize);
455 }
456
457
458 void InterpreterMacroAssembler::push(TosState state) {
459 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
460 switch (state) {
461 case atos: push_ptr(); break;
462 case btos: push_i(); break;
463 case ctos:
464 case stos: push_i(); break;
465 case itos: push_i(); break;
466 case ltos: push_l(); break;
467 case ftos: push_f(); break;
468 case dtos: push_d(); break;
469 case vtos: /* nothing to do */ break;
470 default : ShouldNotReachHere();
471 }
472 }
473
474
475 void InterpreterMacroAssembler::pop(TosState state) {
476 switch (state) {
477 case atos: pop_ptr(); break;
478 case btos: pop_i(); break;
479 case ctos:
480 case stos: pop_i(); break;
481 case itos: pop_i(); break;
482 case ltos: pop_l(); break;
483 case ftos: pop_f(); break;
484 case dtos: pop_d(); break;
485 case vtos: /* nothing to do */ break;
486 default : ShouldNotReachHere();
487 }
488 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
489 }
490
491
492 // Helpers for swap and dup
493 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
494 ld_ptr(Lesp, Interpreter::expr_offset_in_bytes(n), val);
495 }
496 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
497 st_ptr(val, Lesp, Interpreter::expr_offset_in_bytes(n));
498 }
499
500
501 void InterpreterMacroAssembler::load_receiver(Register param_count,
502 Register recv) {
503 sll(param_count, Interpreter::logStackElementSize, param_count);
504 ld_ptr(Lesp, param_count, recv); // gets receiver oop
505 }
506
507 void InterpreterMacroAssembler::empty_expression_stack() {
508 // Reset Lesp.
509 sub( Lmonitors, wordSize, Lesp );
510
511 // Reset SP by subtracting more space from Lesp.
512 Label done;
513 assert(G4_scratch != Gframe_size, "Only you can prevent register aliasing!");
514
515 // A native does not need to do this, since its callee does not change SP.
516 ld(Lmethod, Method::access_flags_offset(), Gframe_size); // Load access flags.
517 btst(JVM_ACC_NATIVE, Gframe_size);
518 br(Assembler::notZero, false, Assembler::pt, done);
519 delayed()->nop();
520
521 // Compute max expression stack+register save area
522 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Gframe_size);
523 lduh(Gframe_size, in_bytes(ConstMethod::max_stack_offset()), Gframe_size); // Load max stack.
524 add(Gframe_size, frame::memory_parameter_word_sp_offset+Method::extra_stack_entries(), Gframe_size );
525
526 //
527 // now set up a stack frame with the size computed above
528 //
529 //round_to( Gframe_size, WordsPerLong ); // -- moved down to the "and" below
530 sll( Gframe_size, LogBytesPerWord, Gframe_size );
531 sub( Lesp, Gframe_size, Gframe_size );
532 and3( Gframe_size, -(2 * wordSize), Gframe_size ); // align SP (downwards) to an 8/16-byte boundary
533 debug_only(verify_sp(Gframe_size, G4_scratch));
534 #ifdef _LP64
535 sub(Gframe_size, STACK_BIAS, Gframe_size );
536 #endif
537 mov(Gframe_size, SP);
538
539 bind(done);
540 }
541
542
543 #ifdef ASSERT
544 void InterpreterMacroAssembler::verify_sp(Register Rsp, Register Rtemp) {
545 Label Bad, OK;
546
547 // Saved SP must be aligned.
548 #ifdef _LP64
549 btst(2*BytesPerWord-1, Rsp);
550 #else
551 btst(LongAlignmentMask, Rsp);
552 #endif
553 br(Assembler::notZero, false, Assembler::pn, Bad);
554 delayed()->nop();
555
556 // Saved SP, plus register window size, must not be above FP.
557 add(Rsp, frame::register_save_words * wordSize, Rtemp);
558 #ifdef _LP64
559 sub(Rtemp, STACK_BIAS, Rtemp); // Bias Rtemp before cmp to FP
560 #endif
561 cmp_and_brx_short(Rtemp, FP, Assembler::greaterUnsigned, Assembler::pn, Bad);
562
563 // Saved SP must not be ridiculously below current SP.
564 size_t maxstack = MAX2(JavaThread::stack_size_at_create(), (size_t) 4*K*K);
565 set(maxstack, Rtemp);
566 sub(SP, Rtemp, Rtemp);
567 #ifdef _LP64
568 add(Rtemp, STACK_BIAS, Rtemp); // Unbias Rtemp before cmp to Rsp
569 #endif
570 cmp_and_brx_short(Rsp, Rtemp, Assembler::lessUnsigned, Assembler::pn, Bad);
571
572 ba_short(OK);
573
574 bind(Bad);
575 stop("on return to interpreted call, restored SP is corrupted");
576
577 bind(OK);
578 }
579
580
581 void InterpreterMacroAssembler::verify_esp(Register Resp) {
582 // about to read or write Resp[0]
583 // make sure it is not in the monitors or the register save area
584 Label OK1, OK2;
585
586 cmp(Resp, Lmonitors);
587 brx(Assembler::lessUnsigned, true, Assembler::pt, OK1);
588 delayed()->sub(Resp, frame::memory_parameter_word_sp_offset * wordSize, Resp);
589 stop("too many pops: Lesp points into monitor area");
590 bind(OK1);
591 #ifdef _LP64
592 sub(Resp, STACK_BIAS, Resp);
593 #endif
594 cmp(Resp, SP);
595 brx(Assembler::greaterEqualUnsigned, false, Assembler::pt, OK2);
596 delayed()->add(Resp, STACK_BIAS + frame::memory_parameter_word_sp_offset * wordSize, Resp);
597 stop("too many pushes: Lesp points into register window");
598 bind(OK2);
599 }
600 #endif // ASSERT
601
602 // Load compiled (i2c) or interpreter entry when calling from interpreted and
603 // do the call. Centralized so that all interpreter calls will do the same actions.
604 // If jvmti single stepping is on for a thread we must not call compiled code.
605 void InterpreterMacroAssembler::call_from_interpreter(Register target, Register scratch, Register Rret) {
606
607 // Assume we want to go compiled if available
608
609 ld_ptr(G5_method, in_bytes(Method::from_interpreted_offset()), target);
610
611 if (JvmtiExport::can_post_interpreter_events()) {
612 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
613 // compiled code in threads for which the event is enabled. Check here for
614 // interp_only_mode if these events CAN be enabled.
615 verify_thread();
616 Label skip_compiled_code;
617
618 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
619 ld(interp_only, scratch);
620 cmp_zero_and_br(Assembler::notZero, scratch, skip_compiled_code, true, Assembler::pn);
621 delayed()->ld_ptr(G5_method, in_bytes(Method::interpreter_entry_offset()), target);
622 bind(skip_compiled_code);
623 }
624
625 // the i2c_adapters need Method* in G5_method (right? %%%)
626 // do the call
627 #ifdef ASSERT
628 {
629 Label ok;
630 br_notnull_short(target, Assembler::pt, ok);
631 stop("null entry point");
632 bind(ok);
633 }
634 #endif // ASSERT
635
636 // Adjust Rret first so Llast_SP can be same as Rret
637 add(Rret, -frame::pc_return_offset, O7);
638 add(Lesp, BytesPerWord, Gargs); // setup parameter pointer
639 // Record SP so we can remove any stack space allocated by adapter transition
640 jmp(target, 0);
641 delayed()->mov(SP, Llast_SP);
642 }
643
644 void InterpreterMacroAssembler::if_cmp(Condition cc, bool ptr_compare) {
645 assert_not_delayed();
646
647 Label not_taken;
648 if (ptr_compare) brx(cc, false, Assembler::pn, not_taken);
649 else br (cc, false, Assembler::pn, not_taken);
650 delayed()->nop();
651
652 TemplateTable::branch(false,false);
653
654 bind(not_taken);
655
656 profile_not_taken_branch(G3_scratch);
657 }
658
659
660 void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(
661 int bcp_offset,
662 Register Rtmp,
663 Register Rdst,
664 signedOrNot is_signed,
665 setCCOrNot should_set_CC ) {
666 assert(Rtmp != Rdst, "need separate temp register");
667 assert_not_delayed();
668 switch (is_signed) {
669 default: ShouldNotReachHere();
670
671 case Signed: ldsb( Lbcp, bcp_offset, Rdst ); break; // high byte
672 case Unsigned: ldub( Lbcp, bcp_offset, Rdst ); break; // high byte
673 }
674 ldub( Lbcp, bcp_offset + 1, Rtmp ); // low byte
675 sll( Rdst, BitsPerByte, Rdst);
676 switch (should_set_CC ) {
677 default: ShouldNotReachHere();
678
679 case set_CC: orcc( Rdst, Rtmp, Rdst ); break;
680 case dont_set_CC: or3( Rdst, Rtmp, Rdst ); break;
681 }
682 }
683
684
685 void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(
686 int bcp_offset,
687 Register Rtmp,
688 Register Rdst,
689 setCCOrNot should_set_CC ) {
690 assert(Rtmp != Rdst, "need separate temp register");
691 assert_not_delayed();
692 add( Lbcp, bcp_offset, Rtmp);
693 andcc( Rtmp, 3, G0);
694 Label aligned;
695 switch (should_set_CC ) {
696 default: ShouldNotReachHere();
697
698 case set_CC: break;
699 case dont_set_CC: break;
700 }
701
702 br(Assembler::zero, true, Assembler::pn, aligned);
703 #ifdef _LP64
704 delayed()->ldsw(Rtmp, 0, Rdst);
705 #else
706 delayed()->ld(Rtmp, 0, Rdst);
707 #endif
708
709 ldub(Lbcp, bcp_offset + 3, Rdst);
710 ldub(Lbcp, bcp_offset + 2, Rtmp); sll(Rtmp, 8, Rtmp); or3(Rtmp, Rdst, Rdst);
711 ldub(Lbcp, bcp_offset + 1, Rtmp); sll(Rtmp, 16, Rtmp); or3(Rtmp, Rdst, Rdst);
712 #ifdef _LP64
713 ldsb(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
714 #else
715 // Unsigned load is faster than signed on some implementations
716 ldub(Lbcp, bcp_offset + 0, Rtmp); sll(Rtmp, 24, Rtmp);
717 #endif
718 or3(Rtmp, Rdst, Rdst );
719
720 bind(aligned);
721 if (should_set_CC == set_CC) tst(Rdst);
722 }
723
724 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register temp, Register index,
725 int bcp_offset, size_t index_size) {
726 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
727 if (index_size == sizeof(u2)) {
728 get_2_byte_integer_at_bcp(bcp_offset, temp, index, Unsigned);
729 } else if (index_size == sizeof(u4)) {
730 assert(EnableInvokeDynamic, "giant index used only for JSR 292");
731 get_4_byte_integer_at_bcp(bcp_offset, temp, index);
732 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
733 xor3(index, -1, index); // convert to plain index
734 } else if (index_size == sizeof(u1)) {
735 ldub(Lbcp, bcp_offset, index);
736 } else {
737 ShouldNotReachHere();
738 }
739 }
740
741
742 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register tmp,
743 int bcp_offset, size_t index_size) {
744 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
745 assert_different_registers(cache, tmp);
746 assert_not_delayed();
747 get_cache_index_at_bcp(cache, tmp, bcp_offset, index_size);
748 // convert from field index to ConstantPoolCacheEntry index and from
749 // word index to byte offset
750 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
751 add(LcpoolCache, tmp, cache);
752 }
753
754
755 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
756 Register temp,
757 Register bytecode,
758 int byte_no,
759 int bcp_offset,
760 size_t index_size) {
761 get_cache_and_index_at_bcp(cache, temp, bcp_offset, index_size);
762 ld_ptr(cache, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset(), bytecode);
763 const int shift_count = (1 + byte_no) * BitsPerByte;
764 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
765 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
766 "correct shift count");
767 srl(bytecode, shift_count, bytecode);
768 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
769 and3(bytecode, ConstantPoolCacheEntry::bytecode_1_mask, bytecode);
770 }
771
772
773 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
774 int bcp_offset, size_t index_size) {
775 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
776 assert_different_registers(cache, tmp);
777 assert_not_delayed();
778 if (index_size == sizeof(u2)) {
779 get_2_byte_integer_at_bcp(bcp_offset, cache, tmp, Unsigned);
780 } else {
781 ShouldNotReachHere(); // other sizes not supported here
782 }
783 // convert from field index to ConstantPoolCacheEntry index
784 // and from word index to byte offset
785 sll(tmp, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord), tmp);
786 // skip past the header
787 add(tmp, in_bytes(ConstantPoolCache::base_offset()), tmp);
788 // construct pointer to cache entry
789 add(LcpoolCache, tmp, cache);
790 }
791
792
793 // Load object from cpool->resolved_references(index)
794 void InterpreterMacroAssembler::load_resolved_reference_at_index(
795 Register result, Register index) {
796 assert_different_registers(result, index);
797 assert_not_delayed();
798 // convert from field index to resolved_references() index and from
799 // word index to byte offset. Since this is a java object, it can be compressed
800 Register tmp = index; // reuse
801 sll(index, LogBytesPerHeapOop, tmp);
802 get_constant_pool(result);
803 // load pointer for resolved_references[] objArray
804 ld_ptr(result, ConstantPool::resolved_references_offset_in_bytes(), result);
805 // JNIHandles::resolve(result)
806 ld_ptr(result, 0, result);
807 // Add in the index
808 add(result, tmp, result);
809 load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result);
810 }
811
812
813 // Generate a subtype check: branch to ok_is_subtype if sub_klass is
814 // a subtype of super_klass. Blows registers Rsuper_klass, Rsub_klass, tmp1, tmp2.
815 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
816 Register Rsuper_klass,
817 Register Rtmp1,
818 Register Rtmp2,
819 Register Rtmp3,
820 Label &ok_is_subtype ) {
821 Label not_subtype;
822
823 // Profile the not-null value's klass.
824 profile_typecheck(Rsub_klass, Rtmp1);
825
826 check_klass_subtype_fast_path(Rsub_klass, Rsuper_klass,
827 Rtmp1, Rtmp2,
828 &ok_is_subtype, ¬_subtype, NULL);
829
830 check_klass_subtype_slow_path(Rsub_klass, Rsuper_klass,
831 Rtmp1, Rtmp2, Rtmp3, /*hack:*/ noreg,
832 &ok_is_subtype, NULL);
833
834 bind(not_subtype);
835 profile_typecheck_failed(Rtmp1);
836 }
837
838 // Separate these two to allow for delay slot in middle
839 // These are used to do a test and full jump to exception-throwing code.
840
841 // %%%%% Could possibly reoptimize this by testing to see if could use
842 // a single conditional branch (i.e. if span is small enough.
843 // If you go that route, than get rid of the split and give up
844 // on the delay-slot hack.
845
846 void InterpreterMacroAssembler::throw_if_not_1_icc( Condition ok_condition,
847 Label& ok ) {
848 assert_not_delayed();
849 br(ok_condition, true, pt, ok);
850 // DELAY SLOT
851 }
852
853 void InterpreterMacroAssembler::throw_if_not_1_xcc( Condition ok_condition,
854 Label& ok ) {
855 assert_not_delayed();
856 bp( ok_condition, true, Assembler::xcc, pt, ok);
857 // DELAY SLOT
858 }
859
860 void InterpreterMacroAssembler::throw_if_not_1_x( Condition ok_condition,
861 Label& ok ) {
862 assert_not_delayed();
863 brx(ok_condition, true, pt, ok);
864 // DELAY SLOT
865 }
866
867 void InterpreterMacroAssembler::throw_if_not_2( address throw_entry_point,
868 Register Rscratch,
869 Label& ok ) {
870 assert(throw_entry_point != NULL, "entry point must be generated by now");
871 AddressLiteral dest(throw_entry_point);
872 jump_to(dest, Rscratch);
873 delayed()->nop();
874 bind(ok);
875 }
876
877
878 // And if you cannot use the delay slot, here is a shorthand:
879
880 void InterpreterMacroAssembler::throw_if_not_icc( Condition ok_condition,
881 address throw_entry_point,
882 Register Rscratch ) {
883 Label ok;
884 if (ok_condition != never) {
885 throw_if_not_1_icc( ok_condition, ok);
886 delayed()->nop();
887 }
888 throw_if_not_2( throw_entry_point, Rscratch, ok);
889 }
890 void InterpreterMacroAssembler::throw_if_not_xcc( Condition ok_condition,
891 address throw_entry_point,
892 Register Rscratch ) {
893 Label ok;
894 if (ok_condition != never) {
895 throw_if_not_1_xcc( ok_condition, ok);
896 delayed()->nop();
897 }
898 throw_if_not_2( throw_entry_point, Rscratch, ok);
899 }
900 void InterpreterMacroAssembler::throw_if_not_x( Condition ok_condition,
901 address throw_entry_point,
902 Register Rscratch ) {
903 Label ok;
904 if (ok_condition != never) {
905 throw_if_not_1_x( ok_condition, ok);
906 delayed()->nop();
907 }
908 throw_if_not_2( throw_entry_point, Rscratch, ok);
909 }
910
911 // Check that index is in range for array, then shift index by index_shift, and put arrayOop + shifted_index into res
912 // Note: res is still shy of address by array offset into object.
913
914 void InterpreterMacroAssembler::index_check_without_pop(Register array, Register index, int index_shift, Register tmp, Register res) {
915 assert_not_delayed();
916
917 verify_oop(array);
918 #ifdef _LP64
919 // sign extend since tos (index) can be a 32bit value
920 sra(index, G0, index);
921 #endif // _LP64
922
923 // check array
924 Label ptr_ok;
925 tst(array);
926 throw_if_not_1_x( notZero, ptr_ok );
927 delayed()->ld( array, arrayOopDesc::length_offset_in_bytes(), tmp ); // check index
928 throw_if_not_2( Interpreter::_throw_NullPointerException_entry, G3_scratch, ptr_ok);
929
930 Label index_ok;
931 cmp(index, tmp);
932 throw_if_not_1_icc( lessUnsigned, index_ok );
933 if (index_shift > 0) delayed()->sll(index, index_shift, index);
934 else delayed()->add(array, index, res); // addr - const offset in index
935 // convention: move aberrant index into G3_scratch for exception message
936 mov(index, G3_scratch);
937 throw_if_not_2( Interpreter::_throw_ArrayIndexOutOfBoundsException_entry, G4_scratch, index_ok);
938
939 // add offset if didn't do it in delay slot
940 if (index_shift > 0) add(array, index, res); // addr - const offset in index
941 }
942
943
944 void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) {
945 assert_not_delayed();
946
947 // pop array
948 pop_ptr(array);
949
950 // check array
951 index_check_without_pop(array, index, index_shift, tmp, res);
952 }
953
954
955 void InterpreterMacroAssembler::get_const(Register Rdst) {
956 ld_ptr(Lmethod, in_bytes(Method::const_offset()), Rdst);
957 }
958
959
960 void InterpreterMacroAssembler::get_constant_pool(Register Rdst) {
961 get_const(Rdst);
962 ld_ptr(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst);
963 }
964
965
966 void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) {
967 get_constant_pool(Rdst);
968 ld_ptr(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst);
969 }
970
971
972 void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) {
973 get_constant_pool(Rcpool);
974 ld_ptr(Rcpool, ConstantPool::tags_offset_in_bytes(), Rtags);
975 }
976
977
978 // unlock if synchronized method
979 //
980 // Unlock the receiver if this is a synchronized method.
981 // Unlock any Java monitors from syncronized blocks.
982 //
983 // If there are locked Java monitors
984 // If throw_monitor_exception
985 // throws IllegalMonitorStateException
986 // Else if install_monitor_exception
987 // installs IllegalMonitorStateException
988 // Else
989 // no error processing
990 void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state,
991 bool throw_monitor_exception,
992 bool install_monitor_exception) {
993 Label unlocked, unlock, no_unlock;
994
995 // get the value of _do_not_unlock_if_synchronized into G1_scratch
996 const Address do_not_unlock_if_synchronized(G2_thread,
997 JavaThread::do_not_unlock_if_synchronized_offset());
998 ldbool(do_not_unlock_if_synchronized, G1_scratch);
999 stbool(G0, do_not_unlock_if_synchronized); // reset the flag
1000
1001 // check if synchronized method
1002 const Address access_flags(Lmethod, Method::access_flags_offset());
1003 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1004 push(state); // save tos
1005 ld(access_flags, G3_scratch); // Load access flags.
1006 btst(JVM_ACC_SYNCHRONIZED, G3_scratch);
1007 br(zero, false, pt, unlocked);
1008 delayed()->nop();
1009
1010 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
1011 // is set.
1012 cmp_zero_and_br(Assembler::notZero, G1_scratch, no_unlock);
1013 delayed()->nop();
1014
1015 // BasicObjectLock will be first in list, since this is a synchronized method. However, need
1016 // to check that the object has not been unlocked by an explicit monitorexit bytecode.
1017
1018 //Intel: if (throw_monitor_exception) ... else ...
1019 // Entry already unlocked, need to throw exception
1020 //...
1021
1022 // pass top-most monitor elem
1023 add( top_most_monitor(), O1 );
1024
1025 ld_ptr(O1, BasicObjectLock::obj_offset_in_bytes(), G3_scratch);
1026 br_notnull_short(G3_scratch, pt, unlock);
1027
1028 if (throw_monitor_exception) {
1029 // Entry already unlocked need to throw an exception
1030 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1031 should_not_reach_here();
1032 } else {
1033 // Monitor already unlocked during a stack unroll.
1034 // If requested, install an illegal_monitor_state_exception.
1035 // Continue with stack unrolling.
1036 if (install_monitor_exception) {
1037 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1038 }
1039 ba_short(unlocked);
1040 }
1041
1042 bind(unlock);
1043
1044 unlock_object(O1);
1045
1046 bind(unlocked);
1047
1048 // I0, I1: Might contain return value
1049
1050 // Check that all monitors are unlocked
1051 { Label loop, exception, entry, restart;
1052
1053 Register Rmptr = O0;
1054 Register Rtemp = O1;
1055 Register Rlimit = Lmonitors;
1056 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
1057 assert( (delta & LongAlignmentMask) == 0,
1058 "sizeof BasicObjectLock must be even number of doublewords");
1059
1060 #ifdef ASSERT
1061 add(top_most_monitor(), Rmptr, delta);
1062 { Label L;
1063 // ensure that Rmptr starts out above (or at) Rlimit
1064 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1065 stop("monitor stack has negative size");
1066 bind(L);
1067 }
1068 #endif
1069 bind(restart);
1070 ba(entry);
1071 delayed()->
1072 add(top_most_monitor(), Rmptr, delta); // points to current entry, starting with bottom-most entry
1073
1074 // Entry is still locked, need to throw exception
1075 bind(exception);
1076 if (throw_monitor_exception) {
1077 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
1078 should_not_reach_here();
1079 } else {
1080 // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception.
1081 // Unlock does not block, so don't have to worry about the frame
1082 unlock_object(Rmptr);
1083 if (install_monitor_exception) {
1084 MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
1085 }
1086 ba_short(restart);
1087 }
1088
1089 bind(loop);
1090 cmp(Rtemp, G0); // check if current entry is used
1091 brx(Assembler::notEqual, false, pn, exception);
1092 delayed()->
1093 dec(Rmptr, delta); // otherwise advance to next entry
1094 #ifdef ASSERT
1095 { Label L;
1096 // ensure that Rmptr has not somehow stepped below Rlimit
1097 cmp_and_brx_short(Rmptr, Rlimit, Assembler::greaterEqualUnsigned, pn, L);
1098 stop("ran off the end of the monitor stack");
1099 bind(L);
1100 }
1101 #endif
1102 bind(entry);
1103 cmp(Rmptr, Rlimit); // check if bottom reached
1104 brx(Assembler::notEqual, true, pn, loop); // if not at bottom then check this entry
1105 delayed()->
1106 ld_ptr(Rmptr, BasicObjectLock::obj_offset_in_bytes() - delta, Rtemp);
1107 }
1108
1109 bind(no_unlock);
1110 pop(state);
1111 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1112 }
1113
1114
1115 // remove activation
1116 //
1117 // Unlock the receiver if this is a synchronized method.
1118 // Unlock any Java monitors from syncronized blocks.
1119 // Remove the activation from the stack.
1120 //
1121 // If there are locked Java monitors
1122 // If throw_monitor_exception
1123 // throws IllegalMonitorStateException
1124 // Else if install_monitor_exception
1125 // installs IllegalMonitorStateException
1126 // Else
1127 // no error processing
1128 void InterpreterMacroAssembler::remove_activation(TosState state,
1129 bool throw_monitor_exception,
1130 bool install_monitor_exception) {
1131
1132 unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
1133
1134 // save result (push state before jvmti call and pop it afterwards) and notify jvmti
1135 notify_method_exit(false, state, NotifyJVMTI);
1136
1137 interp_verify_oop(Otos_i, state, __FILE__, __LINE__);
1138 verify_thread();
1139
1140 // return tos
1141 assert(Otos_l1 == Otos_i, "adjust code below");
1142 switch (state) {
1143 #ifdef _LP64
1144 case ltos: mov(Otos_l, Otos_l->after_save()); break; // O0 -> I0
1145 #else
1146 case ltos: mov(Otos_l2, Otos_l2->after_save()); // fall through // O1 -> I1
1147 #endif
1148 case btos: // fall through
1149 case ctos:
1150 case stos: // fall through
1151 case atos: // fall through
1152 case itos: mov(Otos_l1, Otos_l1->after_save()); break; // O0 -> I0
1153 case ftos: // fall through
1154 case dtos: // fall through
1155 case vtos: /* nothing to do */ break;
1156 default : ShouldNotReachHere();
1157 }
1158
1159 #if defined(COMPILER2) && !defined(_LP64)
1160 if (state == ltos) {
1161 // C2 expects long results in G1 we can't tell if we're returning to interpreted
1162 // or compiled so just be safe use G1 and O0/O1
1163
1164 // Shift bits into high (msb) of G1
1165 sllx(Otos_l1->after_save(), 32, G1);
1166 // Zero extend low bits
1167 srl (Otos_l2->after_save(), 0, Otos_l2->after_save());
1168 or3 (Otos_l2->after_save(), G1, G1);
1169 }
1170 #endif /* COMPILER2 */
1171
1172 }
1173 #endif /* CC_INTERP */
1174
1175
1176 // Lock object
1177 //
1178 // Argument - lock_reg points to the BasicObjectLock to be used for locking,
1179 // it must be initialized with the object to lock
1180 void InterpreterMacroAssembler::lock_object(Register lock_reg, Register Object) {
1181 if (UseHeavyMonitors) {
1182 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1183 }
1184 else {
1185 Register obj_reg = Object;
1186 Register mark_reg = G4_scratch;
1187 Register temp_reg = G1_scratch;
1188 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes());
1189 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1190 Label done;
1191
1192 Label slow_case;
1193
1194 assert_different_registers(lock_reg, obj_reg, mark_reg, temp_reg);
1195
1196 // load markOop from object into mark_reg
1197 ld_ptr(mark_addr, mark_reg);
1198
1199 if (UseBiasedLocking) {
1200 biased_locking_enter(obj_reg, mark_reg, temp_reg, done, &slow_case);
1201 }
1202
1203 // get the address of basicLock on stack that will be stored in the object
1204 // we need a temporary register here as we do not want to clobber lock_reg
1205 // (cas clobbers the destination register)
1206 mov(lock_reg, temp_reg);
1207 // set mark reg to be (markOop of object | UNLOCK_VALUE)
1208 or3(mark_reg, markOopDesc::unlocked_value, mark_reg);
1209 // initialize the box (Must happen before we update the object mark!)
1210 st_ptr(mark_reg, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1211 // compare and exchange object_addr, markOop | 1, stack address of basicLock
1212 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1213 cas_ptr(mark_addr.base(), mark_reg, temp_reg);
1214
1215 // if the compare and exchange succeeded we are done (we saw an unlocked object)
1216 cmp_and_brx_short(mark_reg, temp_reg, Assembler::equal, Assembler::pt, done);
1217
1218 // We did not see an unlocked object so try the fast recursive case
1219
1220 // Check if owner is self by comparing the value in the markOop of object
1221 // with the stack pointer
1222 sub(temp_reg, SP, temp_reg);
1223 #ifdef _LP64
1224 sub(temp_reg, STACK_BIAS, temp_reg);
1225 #endif
1226 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
1227
1228 // Composite "andcc" test:
1229 // (a) %sp -vs- markword proximity check, and,
1230 // (b) verify mark word LSBs == 0 (Stack-locked).
1231 //
1232 // FFFFF003/FFFFFFFFFFFF003 is (markOopDesc::lock_mask_in_place | -os::vm_page_size())
1233 // Note that the page size used for %sp proximity testing is arbitrary and is
1234 // unrelated to the actual MMU page size. We use a 'logical' page size of
1235 // 4096 bytes. F..FFF003 is designed to fit conveniently in the SIMM13 immediate
1236 // field of the andcc instruction.
1237 andcc (temp_reg, 0xFFFFF003, G0) ;
1238
1239 // if condition is true we are done and hence we can store 0 in the displaced
1240 // header indicating it is a recursive lock and be done
1241 brx(Assembler::zero, true, Assembler::pt, done);
1242 delayed()->st_ptr(G0, lock_addr, BasicLock::displaced_header_offset_in_bytes());
1243
1244 // none of the above fast optimizations worked so we have to get into the
1245 // slow case of monitor enter
1246 bind(slow_case);
1247 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
1248
1249 bind(done);
1250 }
1251 }
1252
1253 // Unlocks an object. Used in monitorexit bytecode and remove_activation.
1254 //
1255 // Argument - lock_reg points to the BasicObjectLock for lock
1256 // Throw IllegalMonitorException if object is not locked by current thread
1257 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
1258 if (UseHeavyMonitors) {
1259 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1260 } else {
1261 Register obj_reg = G3_scratch;
1262 Register mark_reg = G4_scratch;
1263 Register displaced_header_reg = G1_scratch;
1264 Address lockobj_addr(lock_reg, BasicObjectLock::obj_offset_in_bytes());
1265 Address mark_addr(obj_reg, oopDesc::mark_offset_in_bytes());
1266 Label done;
1267
1268 if (UseBiasedLocking) {
1269 // load the object out of the BasicObjectLock
1270 ld_ptr(lockobj_addr, obj_reg);
1271 biased_locking_exit(mark_addr, mark_reg, done, true);
1272 st_ptr(G0, lockobj_addr); // free entry
1273 }
1274
1275 // Test first if we are in the fast recursive case
1276 Address lock_addr(lock_reg, BasicObjectLock::lock_offset_in_bytes() + BasicLock::displaced_header_offset_in_bytes());
1277 ld_ptr(lock_addr, displaced_header_reg);
1278 br_null(displaced_header_reg, true, Assembler::pn, done);
1279 delayed()->st_ptr(G0, lockobj_addr); // free entry
1280
1281 // See if it is still a light weight lock, if so we just unlock
1282 // the object and we are done
1283
1284 if (!UseBiasedLocking) {
1285 // load the object out of the BasicObjectLock
1286 ld_ptr(lockobj_addr, obj_reg);
1287 }
1288
1289 // we have the displaced header in displaced_header_reg
1290 // we expect to see the stack address of the basicLock in case the
1291 // lock is still a light weight lock (lock_reg)
1292 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
1293 cas_ptr(mark_addr.base(), lock_reg, displaced_header_reg);
1294 cmp(lock_reg, displaced_header_reg);
1295 brx(Assembler::equal, true, Assembler::pn, done);
1296 delayed()->st_ptr(G0, lockobj_addr); // free entry
1297
1298 // The lock has been converted into a heavy lock and hence
1299 // we need to get into the slow case
1300
1301 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
1302
1303 bind(done);
1304 }
1305 }
1306
1307 #ifndef CC_INTERP
1308
1309 // Get the method data pointer from the Method* and set the
1310 // specified register to its value.
1311
1312 void InterpreterMacroAssembler::set_method_data_pointer() {
1313 assert(ProfileInterpreter, "must be profiling interpreter");
1314 Label get_continue;
1315
1316 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1317 test_method_data_pointer(get_continue);
1318 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1319 bind(get_continue);
1320 }
1321
1322 // Set the method data pointer for the current bcp.
1323
1324 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1325 assert(ProfileInterpreter, "must be profiling interpreter");
1326 Label zero_continue;
1327
1328 // Test MDO to avoid the call if it is NULL.
1329 ld_ptr(Lmethod, in_bytes(Method::method_data_offset()), ImethodDataPtr);
1330 test_method_data_pointer(zero_continue);
1331 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), Lmethod, Lbcp);
1332 add(ImethodDataPtr, in_bytes(MethodData::data_offset()), ImethodDataPtr);
1333 add(ImethodDataPtr, O0, ImethodDataPtr);
1334 bind(zero_continue);
1335 }
1336
1337 // Test ImethodDataPtr. If it is null, continue at the specified label
1338
1339 void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) {
1340 assert(ProfileInterpreter, "must be profiling interpreter");
1341 br_null_short(ImethodDataPtr, Assembler::pn, zero_continue);
1342 }
1343
1344 void InterpreterMacroAssembler::verify_method_data_pointer() {
1345 assert(ProfileInterpreter, "must be profiling interpreter");
1346 #ifdef ASSERT
1347 Label verify_continue;
1348 test_method_data_pointer(verify_continue);
1349
1350 // If the mdp is valid, it will point to a DataLayout header which is
1351 // consistent with the bcp. The converse is highly probable also.
1352 lduh(ImethodDataPtr, in_bytes(DataLayout::bci_offset()), G3_scratch);
1353 ld_ptr(Lmethod, Method::const_offset(), O5);
1354 add(G3_scratch, in_bytes(ConstMethod::codes_offset()), G3_scratch);
1355 add(G3_scratch, O5, G3_scratch);
1356 cmp(Lbcp, G3_scratch);
1357 brx(Assembler::equal, false, Assembler::pt, verify_continue);
1358
1359 Register temp_reg = O5;
1360 delayed()->mov(ImethodDataPtr, temp_reg);
1361 // %%% should use call_VM_leaf here?
1362 //call_VM_leaf(noreg, ..., Lmethod, Lbcp, ImethodDataPtr);
1363 save_frame_and_mov(sizeof(jdouble) / wordSize, Lmethod, O0, Lbcp, O1);
1364 Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
1365 stf(FloatRegisterImpl::D, Ftos_d, d_save);
1366 mov(temp_reg->after_save(), O2);
1367 save_thread(L7_thread_cache);
1368 call(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), relocInfo::none);
1369 delayed()->nop();
1370 restore_thread(L7_thread_cache);
1371 ldf(FloatRegisterImpl::D, d_save, Ftos_d);
1372 restore();
1373 bind(verify_continue);
1374 #endif // ASSERT
1375 }
1376
1377 void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count,
1378 Register Rtmp,
1379 Label &profile_continue) {
1380 assert(ProfileInterpreter, "must be profiling interpreter");
1381 // Control will flow to "profile_continue" if the counter is less than the
1382 // limit or if we call profile_method()
1383
1384 Label done;
1385
1386 // if no method data exists, and the counter is high enough, make one
1387 br_notnull_short(ImethodDataPtr, Assembler::pn, done);
1388
1389 // Test to see if we should create a method data oop
1390 AddressLiteral profile_limit((address) &InvocationCounter::InterpreterProfileLimit);
1391 sethi(profile_limit, Rtmp);
1392 ld(Rtmp, profile_limit.low10(), Rtmp);
1393 cmp(invocation_count, Rtmp);
1394 // Use long branches because call_VM() code and following code generated by
1395 // test_backedge_count_for_osr() is large in debug VM.
1396 br(Assembler::lessUnsigned, false, Assembler::pn, profile_continue);
1397 delayed()->nop();
1398
1399 // Build it now.
1400 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1401 set_method_data_pointer_for_bcp();
1402 ba(profile_continue);
1403 delayed()->nop();
1404 bind(done);
1405 }
1406
1407 // Store a value at some constant offset from the method data pointer.
1408
1409 void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) {
1410 assert(ProfileInterpreter, "must be profiling interpreter");
1411 st_ptr(value, ImethodDataPtr, constant);
1412 }
1413
1414 void InterpreterMacroAssembler::increment_mdp_data_at(Address counter,
1415 Register bumped_count,
1416 bool decrement) {
1417 assert(ProfileInterpreter, "must be profiling interpreter");
1418
1419 // Load the counter.
1420 ld_ptr(counter, bumped_count);
1421
1422 if (decrement) {
1423 // Decrement the register. Set condition codes.
1424 subcc(bumped_count, DataLayout::counter_increment, bumped_count);
1425
1426 // If the decrement causes the counter to overflow, stay negative
1427 Label L;
1428 brx(Assembler::negative, true, Assembler::pn, L);
1429
1430 // Store the decremented counter, if it is still negative.
1431 delayed()->st_ptr(bumped_count, counter);
1432 bind(L);
1433 } else {
1434 // Increment the register. Set carry flag.
1435 addcc(bumped_count, DataLayout::counter_increment, bumped_count);
1436
1437 // If the increment causes the counter to overflow, pull back by 1.
1438 assert(DataLayout::counter_increment == 1, "subc works");
1439 subc(bumped_count, G0, bumped_count);
1440
1441 // Store the incremented counter.
1442 st_ptr(bumped_count, counter);
1443 }
1444 }
1445
1446 // Increment the value at some constant offset from the method data pointer.
1447
1448 void InterpreterMacroAssembler::increment_mdp_data_at(int constant,
1449 Register bumped_count,
1450 bool decrement) {
1451 // Locate the counter at a fixed offset from the mdp:
1452 Address counter(ImethodDataPtr, constant);
1453 increment_mdp_data_at(counter, bumped_count, decrement);
1454 }
1455
1456 // Increment the value at some non-fixed (reg + constant) offset from
1457 // the method data pointer.
1458
1459 void InterpreterMacroAssembler::increment_mdp_data_at(Register reg,
1460 int constant,
1461 Register bumped_count,
1462 Register scratch2,
1463 bool decrement) {
1464 // Add the constant to reg to get the offset.
1465 add(ImethodDataPtr, reg, scratch2);
1466 Address counter(scratch2, constant);
1467 increment_mdp_data_at(counter, bumped_count, decrement);
1468 }
1469
1470 // Set a flag value at the current method data pointer position.
1471 // Updates a single byte of the header, to avoid races with other header bits.
1472
1473 void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant,
1474 Register scratch) {
1475 assert(ProfileInterpreter, "must be profiling interpreter");
1476 // Load the data header
1477 ldub(ImethodDataPtr, in_bytes(DataLayout::flags_offset()), scratch);
1478
1479 // Set the flag
1480 or3(scratch, flag_constant, scratch);
1481
1482 // Store the modified header.
1483 stb(scratch, ImethodDataPtr, in_bytes(DataLayout::flags_offset()));
1484 }
1485
1486 // Test the location at some offset from the method data pointer.
1487 // If it is not equal to value, branch to the not_equal_continue Label.
1488 // Set condition codes to match the nullness of the loaded value.
1489
1490 void InterpreterMacroAssembler::test_mdp_data_at(int offset,
1491 Register value,
1492 Label& not_equal_continue,
1493 Register scratch) {
1494 assert(ProfileInterpreter, "must be profiling interpreter");
1495 ld_ptr(ImethodDataPtr, offset, scratch);
1496 cmp(value, scratch);
1497 brx(Assembler::notEqual, false, Assembler::pn, not_equal_continue);
1498 delayed()->tst(scratch);
1499 }
1500
1501 // Update the method data pointer by the displacement located at some fixed
1502 // offset from the method data pointer.
1503
1504 void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp,
1505 Register scratch) {
1506 assert(ProfileInterpreter, "must be profiling interpreter");
1507 ld_ptr(ImethodDataPtr, offset_of_disp, scratch);
1508 add(ImethodDataPtr, scratch, ImethodDataPtr);
1509 }
1510
1511 // Update the method data pointer by the displacement located at the
1512 // offset (reg + offset_of_disp).
1513
1514 void InterpreterMacroAssembler::update_mdp_by_offset(Register reg,
1515 int offset_of_disp,
1516 Register scratch) {
1517 assert(ProfileInterpreter, "must be profiling interpreter");
1518 add(reg, offset_of_disp, scratch);
1519 ld_ptr(ImethodDataPtr, scratch, scratch);
1520 add(ImethodDataPtr, scratch, ImethodDataPtr);
1521 }
1522
1523 // Update the method data pointer by a simple constant displacement.
1524
1525 void InterpreterMacroAssembler::update_mdp_by_constant(int constant) {
1526 assert(ProfileInterpreter, "must be profiling interpreter");
1527 add(ImethodDataPtr, constant, ImethodDataPtr);
1528 }
1529
1530 // Update the method data pointer for a _ret bytecode whose target
1531 // was not among our cached targets.
1532
1533 void InterpreterMacroAssembler::update_mdp_for_ret(TosState state,
1534 Register return_bci) {
1535 assert(ProfileInterpreter, "must be profiling interpreter");
1536 push(state);
1537 st_ptr(return_bci, l_tmp); // protect return_bci, in case it is volatile
1538 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
1539 ld_ptr(l_tmp, return_bci);
1540 pop(state);
1541 }
1542
1543 // Count a taken branch in the bytecodes.
1544
1545 void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) {
1546 if (ProfileInterpreter) {
1547 Label profile_continue;
1548
1549 // If no method data exists, go to profile_continue.
1550 test_method_data_pointer(profile_continue);
1551
1552 // We are taking a branch. Increment the taken count.
1553 increment_mdp_data_at(in_bytes(JumpData::taken_offset()), bumped_count);
1554
1555 // The method data pointer needs to be updated to reflect the new target.
1556 update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch);
1557 bind (profile_continue);
1558 }
1559 }
1560
1561
1562 // Count a not-taken branch in the bytecodes.
1563
1564 void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch) {
1565 if (ProfileInterpreter) {
1566 Label profile_continue;
1567
1568 // If no method data exists, go to profile_continue.
1569 test_method_data_pointer(profile_continue);
1570
1571 // We are taking a branch. Increment the not taken count.
1572 increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch);
1573
1574 // The method data pointer needs to be updated to correspond to the
1575 // next bytecode.
1576 update_mdp_by_constant(in_bytes(BranchData::branch_data_size()));
1577 bind (profile_continue);
1578 }
1579 }
1580
1581
1582 // Count a non-virtual call in the bytecodes.
1583
1584 void InterpreterMacroAssembler::profile_call(Register scratch) {
1585 if (ProfileInterpreter) {
1586 Label profile_continue;
1587
1588 // If no method data exists, go to profile_continue.
1589 test_method_data_pointer(profile_continue);
1590
1591 // We are making a call. Increment the count.
1592 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1593
1594 // The method data pointer needs to be updated to reflect the new target.
1595 update_mdp_by_constant(in_bytes(CounterData::counter_data_size()));
1596 bind (profile_continue);
1597 }
1598 }
1599
1600
1601 // Count a final call in the bytecodes.
1602
1603 void InterpreterMacroAssembler::profile_final_call(Register scratch) {
1604 if (ProfileInterpreter) {
1605 Label profile_continue;
1606
1607 // If no method data exists, go to profile_continue.
1608 test_method_data_pointer(profile_continue);
1609
1610 // We are making a call. Increment the count.
1611 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1612
1613 // The method data pointer needs to be updated to reflect the new target.
1614 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1615 bind (profile_continue);
1616 }
1617 }
1618
1619
1620 // Count a virtual call in the bytecodes.
1621
1622 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1623 Register scratch,
1624 bool receiver_can_be_null) {
1625 if (ProfileInterpreter) {
1626 Label profile_continue;
1627
1628 // If no method data exists, go to profile_continue.
1629 test_method_data_pointer(profile_continue);
1630
1631
1632 Label skip_receiver_profile;
1633 if (receiver_can_be_null) {
1634 Label not_null;
1635 br_notnull_short(receiver, Assembler::pt, not_null);
1636 // We are making a call. Increment the count for null receiver.
1637 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1638 ba_short(skip_receiver_profile);
1639 bind(not_null);
1640 }
1641
1642 // Record the receiver type.
1643 record_klass_in_profile(receiver, scratch, true);
1644 bind(skip_receiver_profile);
1645
1646 // The method data pointer needs to be updated to reflect the new target.
1647 update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size()));
1648 bind (profile_continue);
1649 }
1650 }
1651
1652 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1653 Register receiver, Register scratch,
1654 int start_row, Label& done, bool is_virtual_call) {
1655 if (TypeProfileWidth == 0) {
1656 if (is_virtual_call) {
1657 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1658 }
1659 return;
1660 }
1661
1662 int last_row = VirtualCallData::row_limit() - 1;
1663 assert(start_row <= last_row, "must be work left to do");
1664 // Test this row for both the receiver and for null.
1665 // Take any of three different outcomes:
1666 // 1. found receiver => increment count and goto done
1667 // 2. found null => keep looking for case 1, maybe allocate this cell
1668 // 3. found something else => keep looking for cases 1 and 2
1669 // Case 3 is handled by a recursive call.
1670 for (int row = start_row; row <= last_row; row++) {
1671 Label next_test;
1672 bool test_for_null_also = (row == start_row);
1673
1674 // See if the receiver is receiver[n].
1675 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1676 test_mdp_data_at(recvr_offset, receiver, next_test, scratch);
1677 // delayed()->tst(scratch);
1678
1679 // The receiver is receiver[n]. Increment count[n].
1680 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1681 increment_mdp_data_at(count_offset, scratch);
1682 ba_short(done);
1683 bind(next_test);
1684
1685 if (test_for_null_also) {
1686 Label found_null;
1687 // Failed the equality check on receiver[n]... Test for null.
1688 if (start_row == last_row) {
1689 // The only thing left to do is handle the null case.
1690 if (is_virtual_call) {
1691 brx(Assembler::zero, false, Assembler::pn, found_null);
1692 delayed()->nop();
1693 // Receiver did not match any saved receiver and there is no empty row for it.
1694 // Increment total counter to indicate polymorphic case.
1695 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1696 ba_short(done);
1697 bind(found_null);
1698 } else {
1699 brx(Assembler::notZero, false, Assembler::pt, done);
1700 delayed()->nop();
1701 }
1702 break;
1703 }
1704 // Since null is rare, make it be the branch-taken case.
1705 brx(Assembler::zero, false, Assembler::pn, found_null);
1706 delayed()->nop();
1707
1708 // Put all the "Case 3" tests here.
1709 record_klass_in_profile_helper(receiver, scratch, start_row + 1, done, is_virtual_call);
1710
1711 // Found a null. Keep searching for a matching receiver,
1712 // but remember that this is an empty (unused) slot.
1713 bind(found_null);
1714 }
1715 }
1716
1717 // In the fall-through case, we found no matching receiver, but we
1718 // observed the receiver[start_row] is NULL.
1719
1720 // Fill in the receiver field and increment the count.
1721 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1722 set_mdp_data_at(recvr_offset, receiver);
1723 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1724 mov(DataLayout::counter_increment, scratch);
1725 set_mdp_data_at(count_offset, scratch);
1726 if (start_row > 0) {
1727 ba_short(done);
1728 }
1729 }
1730
1731 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1732 Register scratch, bool is_virtual_call) {
1733 assert(ProfileInterpreter, "must be profiling");
1734 Label done;
1735
1736 record_klass_in_profile_helper(receiver, scratch, 0, done, is_virtual_call);
1737
1738 bind (done);
1739 }
1740
1741
1742 // Count a ret in the bytecodes.
1743
1744 void InterpreterMacroAssembler::profile_ret(TosState state,
1745 Register return_bci,
1746 Register scratch) {
1747 if (ProfileInterpreter) {
1748 Label profile_continue;
1749 uint row;
1750
1751 // If no method data exists, go to profile_continue.
1752 test_method_data_pointer(profile_continue);
1753
1754 // Update the total ret count.
1755 increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch);
1756
1757 for (row = 0; row < RetData::row_limit(); row++) {
1758 Label next_test;
1759
1760 // See if return_bci is equal to bci[n]:
1761 test_mdp_data_at(in_bytes(RetData::bci_offset(row)),
1762 return_bci, next_test, scratch);
1763
1764 // return_bci is equal to bci[n]. Increment the count.
1765 increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch);
1766
1767 // The method data pointer needs to be updated to reflect the new target.
1768 update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch);
1769 ba_short(profile_continue);
1770 bind(next_test);
1771 }
1772
1773 update_mdp_for_ret(state, return_bci);
1774
1775 bind (profile_continue);
1776 }
1777 }
1778
1779 // Profile an unexpected null in the bytecodes.
1780 void InterpreterMacroAssembler::profile_null_seen(Register scratch) {
1781 if (ProfileInterpreter) {
1782 Label profile_continue;
1783
1784 // If no method data exists, go to profile_continue.
1785 test_method_data_pointer(profile_continue);
1786
1787 set_mdp_flag_at(BitData::null_seen_byte_constant(), scratch);
1788
1789 // The method data pointer needs to be updated.
1790 int mdp_delta = in_bytes(BitData::bit_data_size());
1791 if (TypeProfileCasts) {
1792 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1793 }
1794 update_mdp_by_constant(mdp_delta);
1795
1796 bind (profile_continue);
1797 }
1798 }
1799
1800 void InterpreterMacroAssembler::profile_typecheck(Register klass,
1801 Register scratch) {
1802 if (ProfileInterpreter) {
1803 Label profile_continue;
1804
1805 // If no method data exists, go to profile_continue.
1806 test_method_data_pointer(profile_continue);
1807
1808 int mdp_delta = in_bytes(BitData::bit_data_size());
1809 if (TypeProfileCasts) {
1810 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1811
1812 // Record the object type.
1813 record_klass_in_profile(klass, scratch, false);
1814 }
1815
1816 // The method data pointer needs to be updated.
1817 update_mdp_by_constant(mdp_delta);
1818
1819 bind (profile_continue);
1820 }
1821 }
1822
1823 void InterpreterMacroAssembler::profile_typecheck_failed(Register scratch) {
1824 if (ProfileInterpreter && TypeProfileCasts) {
1825 Label profile_continue;
1826
1827 // If no method data exists, go to profile_continue.
1828 test_method_data_pointer(profile_continue);
1829
1830 int count_offset = in_bytes(CounterData::count_offset());
1831 // Back up the address, since we have already bumped the mdp.
1832 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1833
1834 // *Decrement* the counter. We expect to see zero or small negatives.
1835 increment_mdp_data_at(count_offset, scratch, true);
1836
1837 bind (profile_continue);
1838 }
1839 }
1840
1841 // Count the default case of a switch construct.
1842
1843 void InterpreterMacroAssembler::profile_switch_default(Register scratch) {
1844 if (ProfileInterpreter) {
1845 Label profile_continue;
1846
1847 // If no method data exists, go to profile_continue.
1848 test_method_data_pointer(profile_continue);
1849
1850 // Update the default case count
1851 increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()),
1852 scratch);
1853
1854 // The method data pointer needs to be updated.
1855 update_mdp_by_offset(
1856 in_bytes(MultiBranchData::default_displacement_offset()),
1857 scratch);
1858
1859 bind (profile_continue);
1860 }
1861 }
1862
1863 // Count the index'th case of a switch construct.
1864
1865 void InterpreterMacroAssembler::profile_switch_case(Register index,
1866 Register scratch,
1867 Register scratch2,
1868 Register scratch3) {
1869 if (ProfileInterpreter) {
1870 Label profile_continue;
1871
1872 // If no method data exists, go to profile_continue.
1873 test_method_data_pointer(profile_continue);
1874
1875 // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
1876 set(in_bytes(MultiBranchData::per_case_size()), scratch);
1877 smul(index, scratch, scratch);
1878 add(scratch, in_bytes(MultiBranchData::case_array_offset()), scratch);
1879
1880 // Update the case count
1881 increment_mdp_data_at(scratch,
1882 in_bytes(MultiBranchData::relative_count_offset()),
1883 scratch2,
1884 scratch3);
1885
1886 // The method data pointer needs to be updated.
1887 update_mdp_by_offset(scratch,
1888 in_bytes(MultiBranchData::relative_displacement_offset()),
1889 scratch2);
1890
1891 bind (profile_continue);
1892 }
1893 }
1894
1895 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr, Register tmp) {
1896 Label not_null, do_nothing, do_update;
1897
1898 assert_different_registers(obj, mdo_addr.base(), tmp);
1899
1900 verify_oop(obj);
1901
1902 ld_ptr(mdo_addr, tmp);
1903
1904 br_notnull_short(obj, pt, not_null);
1905 or3(tmp, TypeEntries::null_seen, tmp);
1906 ba_short(do_update);
1907
1908 bind(not_null);
1909 load_klass(obj, obj);
1910
1911 xor3(obj, tmp, obj);
1912 btst(TypeEntries::type_klass_mask, obj);
1913 // klass seen before, nothing to do. The unknown bit may have been
1914 // set already but no need to check.
1915 brx(zero, false, pt, do_nothing);
1916 delayed()->
1917
1918 btst(TypeEntries::type_unknown, obj);
1919 // already unknown. Nothing to do anymore.
1920 brx(notZero, false, pt, do_nothing);
1921 delayed()->
1922
1923 btst(TypeEntries::type_mask, tmp);
1924 brx(zero, true, pt, do_update);
1925 // first time here. Set profile type.
1926 delayed()->or3(tmp, obj, tmp);
1927
1928 // different than before. Cannot keep accurate profile.
1929 or3(tmp, TypeEntries::type_unknown, tmp);
1930
1931 bind(do_update);
1932 // update profile
1933 st_ptr(tmp, mdo_addr);
1934
1935 bind(do_nothing);
1936 }
1937
1938 void InterpreterMacroAssembler::profile_arguments_type(Register callee, Register tmp1, Register tmp2, bool is_virtual) {
1939 if (!ProfileInterpreter) {
1940 return;
1941 }
1942
1943 assert_different_registers(callee, tmp1, tmp2, ImethodDataPtr);
1944
1945 if (MethodData::profile_arguments() || MethodData::profile_return()) {
1946 Label profile_continue;
1947
1948 test_method_data_pointer(profile_continue);
1949
1950 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
1951
1952 ldub(ImethodDataPtr, in_bytes(DataLayout::tag_offset()) - off_to_start, tmp1);
1953 cmp_and_br_short(tmp1, is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag, notEqual, pn, profile_continue);
1954
1955 if (MethodData::profile_arguments()) {
1956 Label done;
1957 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
1958 add(ImethodDataPtr, off_to_args, ImethodDataPtr);
1959
1960 for (int i = 0; i < TypeProfileArgsLimit; i++) {
1961 if (i > 0 || MethodData::profile_return()) {
1962 // If return value type is profiled we may have no argument to profile
1963 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
1964 sub(tmp1, i*TypeStackSlotEntries::per_arg_count(), tmp1);
1965 cmp_and_br_short(tmp1, TypeStackSlotEntries::per_arg_count(), less, pn, done);
1966 }
1967 ld_ptr(Address(callee, Method::const_offset()), tmp1);
1968 lduh(Address(tmp1, ConstMethod::size_of_parameters_offset()), tmp1);
1969 // stack offset o (zero based) from the start of the argument
1970 // list, for n arguments translates into offset n - o - 1 from
1971 // the end of the argument list. But there's an extra slot at
1972 // the stop of the stack. So the offset is n - o from Lesp.
1973 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args, tmp2);
1974 sub(tmp1, tmp2, tmp1);
1975
1976 // Can't use MacroAssembler::argument_address() which needs Gargs to be set up
1977 sll(tmp1, Interpreter::logStackElementSize, tmp1);
1978 ld_ptr(Lesp, tmp1, tmp1);
1979
1980 Address mdo_arg_addr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
1981 profile_obj_type(tmp1, mdo_arg_addr, tmp2);
1982
1983 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
1984 add(ImethodDataPtr, to_add, ImethodDataPtr);
1985 off_to_args += to_add;
1986 }
1987
1988 if (MethodData::profile_return()) {
1989 ld_ptr(ImethodDataPtr, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args, tmp1);
1990 sub(tmp1, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count(), tmp1);
1991 }
1992
1993 bind(done);
1994
1995 if (MethodData::profile_return()) {
1996 // We're right after the type profile for the last
1997 // argument. tmp1 is the number of cells left in the
1998 // CallTypeData/VirtualCallTypeData to reach its end. Non null
1999 // if there's a return to profile.
2000 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
2001 sll(tmp1, exact_log2(DataLayout::cell_size), tmp1);
2002 add(ImethodDataPtr, tmp1, ImethodDataPtr);
2003 }
2004 } else {
2005 assert(MethodData::profile_return(), "either profile call args or call ret");
2006 update_mdp_by_constant(in_bytes(ReturnTypeEntry::size()));
2007 }
2008
2009 // mdp points right after the end of the
2010 // CallTypeData/VirtualCallTypeData, right after the cells for the
2011 // return value type if there's one.
2012
2013 bind(profile_continue);
2014 }
2015 }
2016
2017 void InterpreterMacroAssembler::profile_return_type(Register ret, Register tmp1, Register tmp2) {
2018 assert_different_registers(ret, tmp1, tmp2);
2019 if (ProfileInterpreter && MethodData::profile_return()) {
2020 Label profile_continue, done;
2021
2022 test_method_data_pointer(profile_continue);
2023
2024 if (MethodData::profile_return_jsr292_only()) {
2025 // If we don't profile all invoke bytecodes we must make sure
2026 // it's a bytecode we indeed profile. We can't go back to the
2027 // begining of the ProfileData we intend to update to check its
2028 // type because we're right after it and we don't known its
2029 // length.
2030 Label do_profile;
2031 ldub(Lbcp, 0, tmp1);
2032 cmp_and_br_short(tmp1, Bytecodes::_invokedynamic, equal, pn, do_profile);
2033 cmp(tmp1, Bytecodes::_invokehandle);
2034 br(equal, false, pn, do_profile);
2035 delayed()->ldub(Lmethod, Method::intrinsic_id_offset_in_bytes(), tmp1);
2036 cmp_and_br_short(tmp1, vmIntrinsics::_compiledLambdaForm, notEqual, pt, profile_continue);
2037
2038 bind(do_profile);
2039 }
2040
2041 Address mdo_ret_addr(ImethodDataPtr, -in_bytes(ReturnTypeEntry::size()));
2042 mov(ret, tmp1);
2043 profile_obj_type(tmp1, mdo_ret_addr, tmp2);
2044
2045 bind(profile_continue);
2046 }
2047 }
2048
2049 void InterpreterMacroAssembler::profile_parameters_type(Register tmp1, Register tmp2, Register tmp3, Register tmp4) {
2050 if (ProfileInterpreter && MethodData::profile_parameters()) {
2051 Label profile_continue, done;
2052
2053 test_method_data_pointer(profile_continue);
2054
2055 // Load the offset of the area within the MDO used for
2056 // parameters. If it's negative we're not profiling any parameters.
2057 lduw(ImethodDataPtr, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()), tmp1);
2058 cmp_and_br_short(tmp1, 0, less, pn, profile_continue);
2059
2060 // Compute a pointer to the area for parameters from the offset
2061 // and move the pointer to the slot for the last
2062 // parameters. Collect profiling from last parameter down.
2063 // mdo start + parameters offset + array length - 1
2064
2065 // Pointer to the parameter area in the MDO
2066 Register mdp = tmp1;
2067 add(ImethodDataPtr, tmp1, mdp);
2068
2069 // offset of the current profile entry to update
2070 Register entry_offset = tmp2;
2071 // entry_offset = array len in number of cells
2072 ld_ptr(mdp, ArrayData::array_len_offset(), entry_offset);
2073
2074 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
2075 assert(off_base % DataLayout::cell_size == 0, "should be a number of cells");
2076
2077 // entry_offset (number of cells) = array len - size of 1 entry + offset of the stack slot field
2078 sub(entry_offset, TypeStackSlotEntries::per_arg_count() - (off_base / DataLayout::cell_size), entry_offset);
2079 // entry_offset in bytes
2080 sll(entry_offset, exact_log2(DataLayout::cell_size), entry_offset);
2081
2082 Label loop;
2083 bind(loop);
2084
2085 // load offset on the stack from the slot for this parameter
2086 ld_ptr(mdp, entry_offset, tmp3);
2087 sll(tmp3,Interpreter::logStackElementSize, tmp3);
2088 neg(tmp3);
2089 // read the parameter from the local area
2090 ld_ptr(Llocals, tmp3, tmp3);
2091
2092 // make entry_offset now point to the type field for this parameter
2093 int type_base = in_bytes(ParametersTypeData::type_offset(0));
2094 assert(type_base > off_base, "unexpected");
2095 add(entry_offset, type_base - off_base, entry_offset);
2096
2097 // profile the parameter
2098 Address arg_type(mdp, entry_offset);
2099 profile_obj_type(tmp3, arg_type, tmp4);
2100
2101 // go to next parameter
2102 sub(entry_offset, TypeStackSlotEntries::per_arg_count() * DataLayout::cell_size + (type_base - off_base), entry_offset);
2103 cmp_and_br_short(entry_offset, off_base, greaterEqual, pt, loop);
2104
2105 bind(profile_continue);
2106 }
2107 }
2108
2109 // add a InterpMonitorElem to stack (see frame_sparc.hpp)
2110
2111 void InterpreterMacroAssembler::add_monitor_to_stack( bool stack_is_empty,
2112 Register Rtemp,
2113 Register Rtemp2 ) {
2114
2115 Register Rlimit = Lmonitors;
2116 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2117 assert( (delta & LongAlignmentMask) == 0,
2118 "sizeof BasicObjectLock must be even number of doublewords");
2119
2120 sub( SP, delta, SP);
2121 sub( Lesp, delta, Lesp);
2122 sub( Lmonitors, delta, Lmonitors);
2123
2124 if (!stack_is_empty) {
2125
2126 // must copy stack contents down
2127
2128 Label start_copying, next;
2129
2130 // untested("monitor stack expansion");
2131 compute_stack_base(Rtemp);
2132 ba(start_copying);
2133 delayed()->cmp(Rtemp, Rlimit); // done? duplicated below
2134
2135 // note: must copy from low memory upwards
2136 // On entry to loop,
2137 // Rtemp points to new base of stack, Lesp points to new end of stack (1 past TOS)
2138 // Loop mutates Rtemp
2139
2140 bind( next);
2141
2142 st_ptr(Rtemp2, Rtemp, 0);
2143 inc(Rtemp, wordSize);
2144 cmp(Rtemp, Rlimit); // are we done? (duplicated above)
2145
2146 bind( start_copying );
2147
2148 brx( notEqual, true, pn, next );
2149 delayed()->ld_ptr( Rtemp, delta, Rtemp2 );
2150
2151 // done copying stack
2152 }
2153 }
2154
2155 // Locals
2156 void InterpreterMacroAssembler::access_local_ptr( Register index, Register dst ) {
2157 assert_not_delayed();
2158 sll(index, Interpreter::logStackElementSize, index);
2159 sub(Llocals, index, index);
2160 ld_ptr(index, 0, dst);
2161 // Note: index must hold the effective address--the iinc template uses it
2162 }
2163
2164 // Just like access_local_ptr but the tag is a returnAddress
2165 void InterpreterMacroAssembler::access_local_returnAddress(Register index,
2166 Register dst ) {
2167 assert_not_delayed();
2168 sll(index, Interpreter::logStackElementSize, index);
2169 sub(Llocals, index, index);
2170 ld_ptr(index, 0, dst);
2171 }
2172
2173 void InterpreterMacroAssembler::access_local_int( Register index, Register dst ) {
2174 assert_not_delayed();
2175 sll(index, Interpreter::logStackElementSize, index);
2176 sub(Llocals, index, index);
2177 ld(index, 0, dst);
2178 // Note: index must hold the effective address--the iinc template uses it
2179 }
2180
2181
2182 void InterpreterMacroAssembler::access_local_long( Register index, Register dst ) {
2183 assert_not_delayed();
2184 sll(index, Interpreter::logStackElementSize, index);
2185 sub(Llocals, index, index);
2186 // First half stored at index n+1 (which grows down from Llocals[n])
2187 load_unaligned_long(index, Interpreter::local_offset_in_bytes(1), dst);
2188 }
2189
2190
2191 void InterpreterMacroAssembler::access_local_float( Register index, FloatRegister dst ) {
2192 assert_not_delayed();
2193 sll(index, Interpreter::logStackElementSize, index);
2194 sub(Llocals, index, index);
2195 ldf(FloatRegisterImpl::S, index, 0, dst);
2196 }
2197
2198
2199 void InterpreterMacroAssembler::access_local_double( Register index, FloatRegister dst ) {
2200 assert_not_delayed();
2201 sll(index, Interpreter::logStackElementSize, index);
2202 sub(Llocals, index, index);
2203 load_unaligned_double(index, Interpreter::local_offset_in_bytes(1), dst);
2204 }
2205
2206
2207 #ifdef ASSERT
2208 void InterpreterMacroAssembler::check_for_regarea_stomp(Register Rindex, int offset, Register Rlimit, Register Rscratch, Register Rscratch1) {
2209 Label L;
2210
2211 assert(Rindex != Rscratch, "Registers cannot be same");
2212 assert(Rindex != Rscratch1, "Registers cannot be same");
2213 assert(Rlimit != Rscratch, "Registers cannot be same");
2214 assert(Rlimit != Rscratch1, "Registers cannot be same");
2215 assert(Rscratch1 != Rscratch, "Registers cannot be same");
2216
2217 // untested("reg area corruption");
2218 add(Rindex, offset, Rscratch);
2219 add(Rlimit, 64 + STACK_BIAS, Rscratch1);
2220 cmp_and_brx_short(Rscratch, Rscratch1, Assembler::greaterEqualUnsigned, pn, L);
2221 stop("regsave area is being clobbered");
2222 bind(L);
2223 }
2224 #endif // ASSERT
2225
2226
2227 void InterpreterMacroAssembler::store_local_int( Register index, Register src ) {
2228 assert_not_delayed();
2229 sll(index, Interpreter::logStackElementSize, index);
2230 sub(Llocals, index, index);
2231 debug_only(check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);)
2232 st(src, index, 0);
2233 }
2234
2235 void InterpreterMacroAssembler::store_local_ptr( Register index, Register src ) {
2236 assert_not_delayed();
2237 sll(index, Interpreter::logStackElementSize, index);
2238 sub(Llocals, index, index);
2239 #ifdef ASSERT
2240 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2241 #endif
2242 st_ptr(src, index, 0);
2243 }
2244
2245
2246
2247 void InterpreterMacroAssembler::store_local_ptr( int n, Register src ) {
2248 st_ptr(src, Llocals, Interpreter::local_offset_in_bytes(n));
2249 }
2250
2251 void InterpreterMacroAssembler::store_local_long( Register index, Register src ) {
2252 assert_not_delayed();
2253 sll(index, Interpreter::logStackElementSize, index);
2254 sub(Llocals, index, index);
2255 #ifdef ASSERT
2256 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2257 #endif
2258 store_unaligned_long(src, index, Interpreter::local_offset_in_bytes(1)); // which is n+1
2259 }
2260
2261
2262 void InterpreterMacroAssembler::store_local_float( Register index, FloatRegister src ) {
2263 assert_not_delayed();
2264 sll(index, Interpreter::logStackElementSize, index);
2265 sub(Llocals, index, index);
2266 #ifdef ASSERT
2267 check_for_regarea_stomp(index, 0, FP, G1_scratch, G4_scratch);
2268 #endif
2269 stf(FloatRegisterImpl::S, src, index, 0);
2270 }
2271
2272
2273 void InterpreterMacroAssembler::store_local_double( Register index, FloatRegister src ) {
2274 assert_not_delayed();
2275 sll(index, Interpreter::logStackElementSize, index);
2276 sub(Llocals, index, index);
2277 #ifdef ASSERT
2278 check_for_regarea_stomp(index, Interpreter::local_offset_in_bytes(1), FP, G1_scratch, G4_scratch);
2279 #endif
2280 store_unaligned_double(src, index, Interpreter::local_offset_in_bytes(1));
2281 }
2282
2283
2284 int InterpreterMacroAssembler::top_most_monitor_byte_offset() {
2285 const jint delta = frame::interpreter_frame_monitor_size() * wordSize;
2286 int rounded_vm_local_words = ::round_to(frame::interpreter_frame_vm_local_words, WordsPerLong);
2287 return ((-rounded_vm_local_words * wordSize) - delta ) + STACK_BIAS;
2288 }
2289
2290
2291 Address InterpreterMacroAssembler::top_most_monitor() {
2292 return Address(FP, top_most_monitor_byte_offset());
2293 }
2294
2295
2296 void InterpreterMacroAssembler::compute_stack_base( Register Rdest ) {
2297 add( Lesp, wordSize, Rdest );
2298 }
2299
2300 #endif /* CC_INTERP */
2301
2302 void InterpreterMacroAssembler::get_method_counters(Register method,
2303 Register Rcounters,
2304 Label& skip) {
2305 Label has_counters;
2306 Address method_counters(method, in_bytes(Method::method_counters_offset()));
2307 ld_ptr(method_counters, Rcounters);
2308 br_notnull_short(Rcounters, Assembler::pt, has_counters);
2309 call_VM(noreg, CAST_FROM_FN_PTR(address,
2310 InterpreterRuntime::build_method_counters), method);
2311 ld_ptr(method_counters, Rcounters);
2312 br_null(Rcounters, false, Assembler::pn, skip); // No MethodCounters, OutOfMemory
2313 delayed()->nop();
2314 bind(has_counters);
2315 }
2316
2317 void InterpreterMacroAssembler::increment_invocation_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2318 assert(UseCompiler, "incrementing must be useful");
2319 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2320
2321 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2322 InvocationCounter::counter_offset());
2323 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2324 InvocationCounter::counter_offset());
2325 int delta = InvocationCounter::count_increment;
2326
2327 // Load each counter in a register
2328 ld( inv_counter, Rtmp );
2329 ld( be_counter, Rtmp2 );
2330
2331 assert( is_simm13( delta ), " delta too large.");
2332
2333 // Add the delta to the invocation counter and store the result
2334 add( Rtmp, delta, Rtmp );
2335
2336 // Mask the backedge counter
2337 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2338
2339 // Store value
2340 st( Rtmp, inv_counter);
2341
2342 // Add invocation counter + backedge counter
2343 add( Rtmp, Rtmp2, Rtmp);
2344
2345 // Note that this macro must leave the backedge_count + invocation_count in Rtmp!
2346 }
2347
2348 void InterpreterMacroAssembler::increment_backedge_counter( Register Rcounters, Register Rtmp, Register Rtmp2 ) {
2349 assert(UseCompiler, "incrementing must be useful");
2350 assert_different_registers(Rcounters, Rtmp, Rtmp2);
2351
2352 Address be_counter (Rcounters, MethodCounters::backedge_counter_offset() +
2353 InvocationCounter::counter_offset());
2354 Address inv_counter(Rcounters, MethodCounters::invocation_counter_offset() +
2355 InvocationCounter::counter_offset());
2356
2357 int delta = InvocationCounter::count_increment;
2358 // Load each counter in a register
2359 ld( be_counter, Rtmp );
2360 ld( inv_counter, Rtmp2 );
2361
2362 // Add the delta to the backedge counter
2363 add( Rtmp, delta, Rtmp );
2364
2365 // Mask the invocation counter, add to backedge counter
2366 and3( Rtmp2, InvocationCounter::count_mask_value, Rtmp2 );
2367
2368 // and store the result to memory
2369 st( Rtmp, be_counter );
2370
2371 // Add backedge + invocation counter
2372 add( Rtmp, Rtmp2, Rtmp );
2373
2374 // Note that this macro must leave backedge_count + invocation_count in Rtmp!
2375 }
2376
2377 #ifndef CC_INTERP
2378 void InterpreterMacroAssembler::test_backedge_count_for_osr( Register backedge_count,
2379 Register branch_bcp,
2380 Register Rtmp ) {
2381 Label did_not_overflow;
2382 Label overflow_with_error;
2383 assert_different_registers(backedge_count, Rtmp, branch_bcp);
2384 assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr");
2385
2386 AddressLiteral limit(&InvocationCounter::InterpreterBackwardBranchLimit);
2387 load_contents(limit, Rtmp);
2388 cmp_and_br_short(backedge_count, Rtmp, Assembler::lessUnsigned, Assembler::pt, did_not_overflow);
2389
2390 // When ProfileInterpreter is on, the backedge_count comes from the
2391 // MethodData*, which value does not get reset on the call to
2392 // frequency_counter_overflow(). To avoid excessive calls to the overflow
2393 // routine while the method is being compiled, add a second test to make sure
2394 // the overflow function is called only once every overflow_frequency.
2395 if (ProfileInterpreter) {
2396 const int overflow_frequency = 1024;
2397 andcc(backedge_count, overflow_frequency-1, Rtmp);
2398 brx(Assembler::notZero, false, Assembler::pt, did_not_overflow);
2399 delayed()->nop();
2400 }
2401
2402 // overflow in loop, pass branch bytecode
2403 set(6,Rtmp);
2404 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, Rtmp);
2405
2406 // Was an OSR adapter generated?
2407 // O0 = osr nmethod
2408 br_null_short(O0, Assembler::pn, overflow_with_error);
2409
2410 // Has the nmethod been invalidated already?
2411 ld(O0, nmethod::entry_bci_offset(), O2);
2412 cmp_and_br_short(O2, InvalidOSREntryBci, Assembler::equal, Assembler::pn, overflow_with_error);
2413
2414 // migrate the interpreter frame off of the stack
2415
2416 mov(G2_thread, L7);
2417 // save nmethod
2418 mov(O0, L6);
2419 set_last_Java_frame(SP, noreg);
2420 call_VM_leaf(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), L7);
2421 reset_last_Java_frame();
2422 mov(L7, G2_thread);
2423
2424 // move OSR nmethod to I1
2425 mov(L6, I1);
2426
2427 // OSR buffer to I0
2428 mov(O0, I0);
2429
2430 // remove the interpreter frame
2431 restore(I5_savedSP, 0, SP);
2432
2433 // Jump to the osr code.
2434 ld_ptr(O1, nmethod::osr_entry_point_offset(), O2);
2435 jmp(O2, G0);
2436 delayed()->nop();
2437
2438 bind(overflow_with_error);
2439
2440 bind(did_not_overflow);
2441 }
2442
2443
2444
2445 void InterpreterMacroAssembler::interp_verify_oop(Register reg, TosState state, const char * file, int line) {
2446 if (state == atos) { MacroAssembler::_verify_oop(reg, "broken oop ", file, line); }
2447 }
2448
2449
2450 // local helper function for the verify_oop_or_return_address macro
2451 static bool verify_return_address(Method* m, int bci) {
2452 #ifndef PRODUCT
2453 address pc = (address)(m->constMethod())
2454 + in_bytes(ConstMethod::codes_offset()) + bci;
2455 // assume it is a valid return address if it is inside m and is preceded by a jsr
2456 if (!m->contains(pc)) return false;
2457 address jsr_pc;
2458 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr);
2459 if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true;
2460 jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w);
2461 if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true;
2462 #endif // PRODUCT
2463 return false;
2464 }
2465
2466
2467 void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) {
2468 if (!VerifyOops) return;
2469 // the VM documentation for the astore[_wide] bytecode allows
2470 // the TOS to be not only an oop but also a return address
2471 Label test;
2472 Label skip;
2473 // See if it is an address (in the current method):
2474
2475 mov(reg, Rtmp);
2476 const int log2_bytecode_size_limit = 16;
2477 srl(Rtmp, log2_bytecode_size_limit, Rtmp);
2478 br_notnull_short( Rtmp, pt, test );
2479
2480 // %%% should use call_VM_leaf here?
2481 save_frame_and_mov(0, Lmethod, O0, reg, O1);
2482 save_thread(L7_thread_cache);
2483 call(CAST_FROM_FN_PTR(address,verify_return_address), relocInfo::none);
2484 delayed()->nop();
2485 restore_thread(L7_thread_cache);
2486 br_notnull( O0, false, pt, skip );
2487 delayed()->restore();
2488
2489 // Perform a more elaborate out-of-line call
2490 // Not an address; verify it:
2491 bind(test);
2492 verify_oop(reg);
2493 bind(skip);
2494 }
2495
2496
2497 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
2498 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
2499 }
2500
2501
2502 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
2503 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
2504 int increment, int mask,
2505 Register scratch1, Register scratch2,
2506 Condition cond, Label *where) {
2507 ld(counter_addr, scratch1);
2508 add(scratch1, increment, scratch1);
2509 if (is_simm13(mask)) {
2510 andcc(scratch1, mask, G0);
2511 } else {
2512 set(mask, scratch2);
2513 andcc(scratch1, scratch2, G0);
2514 }
2515 br(cond, false, Assembler::pn, *where);
2516 delayed()->st(scratch1, counter_addr);
2517 }
2518 #endif /* CC_INTERP */
2519
2520 // Inline assembly for:
2521 //
2522 // if (thread is in interp_only_mode) {
2523 // InterpreterRuntime::post_method_entry();
2524 // }
2525 // if (DTraceMethodProbes) {
2526 // SharedRuntime::dtrace_method_entry(method, receiver);
2527 // }
2528 // if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2529 // SharedRuntime::rc_trace_method_entry(method, receiver);
2530 // }
2531
2532 void InterpreterMacroAssembler::notify_method_entry() {
2533
2534 // C++ interpreter only uses this for native methods.
2535
2536 // Whenever JVMTI puts a thread in interp_only_mode, method
2537 // entry/exit events are sent for that thread to track stack
2538 // depth. If it is possible to enter interp_only_mode we add
2539 // the code to check if the event should be sent.
2540 if (JvmtiExport::can_post_interpreter_events()) {
2541 Label L;
2542 Register temp_reg = O5;
2543 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2544 ld(interp_only, temp_reg);
2545 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2546 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
2547 bind(L);
2548 }
2549
2550 {
2551 Register temp_reg = O5;
2552 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2553 call_VM_leaf(noreg,
2554 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
2555 G2_thread, Lmethod);
2556 }
2557
2558 // RedefineClasses() tracing support for obsolete method entry
2559 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
2560 call_VM_leaf(noreg,
2561 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
2562 G2_thread, Lmethod);
2563 }
2564 }
2565
2566
2567 // Inline assembly for:
2568 //
2569 // if (thread is in interp_only_mode) {
2570 // // save result
2571 // InterpreterRuntime::post_method_exit();
2572 // // restore result
2573 // }
2574 // if (DTraceMethodProbes) {
2575 // SharedRuntime::dtrace_method_exit(thread, method);
2576 // }
2577 //
2578 // Native methods have their result stored in d_tmp and l_tmp
2579 // Java methods have their result stored in the expression stack
2580
2581 void InterpreterMacroAssembler::notify_method_exit(bool is_native_method,
2582 TosState state,
2583 NotifyMethodExitMode mode) {
2584 // C++ interpreter only uses this for native methods.
2585
2586 // Whenever JVMTI puts a thread in interp_only_mode, method
2587 // entry/exit events are sent for that thread to track stack
2588 // depth. If it is possible to enter interp_only_mode we add
2589 // the code to check if the event should be sent.
2590 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
2591 Label L;
2592 Register temp_reg = O5;
2593 const Address interp_only(G2_thread, JavaThread::interp_only_mode_offset());
2594 ld(interp_only, temp_reg);
2595 cmp_and_br_short(temp_reg, 0, equal, pt, L);
2596
2597 // Note: frame::interpreter_frame_result has a dependency on how the
2598 // method result is saved across the call to post_method_exit. For
2599 // native methods it assumes the result registers are saved to
2600 // l_scratch and d_scratch. If this changes then the interpreter_frame_result
2601 // implementation will need to be updated too.
2602
2603 save_return_value(state, is_native_method);
2604 call_VM(noreg,
2605 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
2606 restore_return_value(state, is_native_method);
2607 bind(L);
2608 }
2609
2610 {
2611 Register temp_reg = O5;
2612 // Dtrace notification
2613 SkipIfEqual skip_if(this, temp_reg, &DTraceMethodProbes, zero);
2614 save_return_value(state, is_native_method);
2615 call_VM_leaf(
2616 noreg,
2617 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
2618 G2_thread, Lmethod);
2619 restore_return_value(state, is_native_method);
2620 }
2621 }
2622
2623 void InterpreterMacroAssembler::save_return_value(TosState state, bool is_native_call) {
2624 #ifdef CC_INTERP
2625 // result potentially in O0/O1: save it across calls
2626 stf(FloatRegisterImpl::D, F0, STATE(_native_fresult));
2627 #ifdef _LP64
2628 stx(O0, STATE(_native_lresult));
2629 #else
2630 std(O0, STATE(_native_lresult));
2631 #endif
2632 #else // CC_INTERP
2633 if (is_native_call) {
2634 stf(FloatRegisterImpl::D, F0, d_tmp);
2635 #ifdef _LP64
2636 stx(O0, l_tmp);
2637 #else
2638 std(O0, l_tmp);
2639 #endif
2640 } else {
2641 push(state);
2642 }
2643 #endif // CC_INTERP
2644 }
2645
2646 void InterpreterMacroAssembler::restore_return_value( TosState state, bool is_native_call) {
2647 #ifdef CC_INTERP
2648 ldf(FloatRegisterImpl::D, STATE(_native_fresult), F0);
2649 #ifdef _LP64
2650 ldx(STATE(_native_lresult), O0);
2651 #else
2652 ldd(STATE(_native_lresult), O0);
2653 #endif
2654 #else // CC_INTERP
2655 if (is_native_call) {
2656 ldf(FloatRegisterImpl::D, d_tmp, F0);
2657 #ifdef _LP64
2658 ldx(l_tmp, O0);
2659 #else
2660 ldd(l_tmp, O0);
2661 #endif
2662 } else {
2663 pop(state);
2664 }
2665 #endif // CC_INTERP
2666 }