1 /*
2 * Copyright 2003-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25 #include "incls/_precompiled.incl"
26 #include "incls/_interp_masm_x86_64.cpp.incl"
27
28
29 // Implementation of InterpreterMacroAssembler
30
31 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
32 int number_of_arguments) {
33 // interpreter specific
34 //
35 // Note: No need to save/restore bcp & locals (r13 & r14) pointer
36 // since these are callee saved registers and no blocking/
37 // GC can happen in leaf calls.
38 #ifdef ASSERT
39 save_bcp();
40 {
41 Label L;
42 cmpq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD);
43 jcc(Assembler::equal, L);
44 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
45 " last_sp != NULL");
46 bind(L);
47 }
48 #endif
49 // super call
50 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
51 // interpreter specific
52 #ifdef ASSERT
53 {
54 Label L;
55 cmpq(r13, Address(rbp, frame::interpreter_frame_bcx_offset * wordSize));
56 jcc(Assembler::equal, L);
57 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
58 " r13 not callee saved?");
59 bind(L);
60 }
61 {
62 Label L;
63 cmpq(r14, Address(rbp, frame::interpreter_frame_locals_offset * wordSize));
64 jcc(Assembler::equal, L);
65 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
66 " r14 not callee saved?");
67 bind(L);
68 }
69 #endif
70 }
71
72 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
73 Register java_thread,
74 Register last_java_sp,
75 address entry_point,
76 int number_of_arguments,
77 bool check_exceptions) {
78 // interpreter specific
79 //
80 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
81 // really make a difference for these runtime calls, since they are
82 // slow anyway. Btw., bcp must be saved/restored since it may change
83 // due to GC.
84 // assert(java_thread == noreg , "not expecting a precomputed java thread");
85 save_bcp();
86 #ifdef ASSERT
87 {
88 Label L;
89 cmpq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int)NULL_WORD);
90 jcc(Assembler::equal, L);
91 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
92 " last_sp != NULL");
93 bind(L);
94 }
95 #endif /* ASSERT */
96 // super call
97 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
98 entry_point, number_of_arguments,
99 check_exceptions);
100 // interpreter specific
101 restore_bcp();
102 restore_locals();
103 }
104
105
106 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
107 if (JvmtiExport::can_pop_frame()) {
108 Label L;
109 // Initiate popframe handling only if it is not already being
110 // processed. If the flag has the popframe_processing bit set, it
111 // means that this code is called *during* popframe handling - we
112 // don't want to reenter.
113 // This method is only called just after the call into the vm in
114 // call_VM_base, so the arg registers are available.
115 movl(c_rarg0, Address(r15_thread, JavaThread::popframe_condition_offset()));
116 testl(c_rarg0, JavaThread::popframe_pending_bit);
117 jcc(Assembler::zero, L);
118 testl(c_rarg0, JavaThread::popframe_processing_bit);
119 jcc(Assembler::notZero, L);
120 // Call Interpreter::remove_activation_preserving_args_entry() to get the
121 // address of the same-named entrypoint in the generated interpreter code.
122 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
123 jmp(rax);
124 bind(L);
125 }
126 }
127
128
129 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
130 movq(rcx, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
131 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
132 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
133 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
134 switch (state) {
135 case atos: movq(rax, oop_addr);
136 movptr(oop_addr, NULL_WORD);
137 verify_oop(rax, state); break;
138 case ltos: movq(rax, val_addr); break;
139 case btos: // fall through
140 case ctos: // fall through
141 case stos: // fall through
142 case itos: movl(rax, val_addr); break;
143 case ftos: movflt(xmm0, val_addr); break;
144 case dtos: movdbl(xmm0, val_addr); break;
145 case vtos: /* nothing to do */ break;
146 default : ShouldNotReachHere();
147 }
148 // Clean up tos value in the thread object
149 movl(tos_addr, (int) ilgl);
150 movl(val_addr, (int) NULL_WORD);
151 }
152
153
154 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
155 if (JvmtiExport::can_force_early_return()) {
156 Label L;
157 movq(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
158 testq(c_rarg0, c_rarg0);
159 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
160
161 // Initiate earlyret handling only if it is not already being processed.
162 // If the flag has the earlyret_processing bit set, it means that this code
163 // is called *during* earlyret handling - we don't want to reenter.
164 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_state_offset()));
165 cmpl(c_rarg0, JvmtiThreadState::earlyret_pending);
166 jcc(Assembler::notEqual, L);
167
168 // Call Interpreter::remove_activation_early_entry() to get the address of the
169 // same-named entrypoint in the generated interpreter code.
170 movq(c_rarg0, Address(r15_thread, JavaThread::jvmti_thread_state_offset()));
171 movl(c_rarg0, Address(c_rarg0, JvmtiThreadState::earlyret_tos_offset()));
172 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), c_rarg0);
173 jmp(rax);
174 bind(L);
175 }
176 }
177
178
179 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(
180 Register reg,
181 int bcp_offset) {
182 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
183 movl(reg, Address(r13, bcp_offset));
184 bswapl(reg);
185 shrl(reg, 16);
186 }
187
188
189 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
190 Register index,
191 int bcp_offset) {
192 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
193 assert(cache != index, "must use different registers");
194 load_unsigned_word(index, Address(r13, bcp_offset));
195 movq(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
196 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
197 // convert from field index to ConstantPoolCacheEntry index
198 shll(index, 2);
199 }
200
201
202 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
203 Register tmp,
204 int bcp_offset) {
205 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
206 assert(cache != tmp, "must use different register");
207 load_unsigned_word(tmp, Address(r13, bcp_offset));
208 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
209 // convert from field index to ConstantPoolCacheEntry index
210 // and from word offset to byte offset
211 shll(tmp, 2 + LogBytesPerWord);
212 movq(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
213 // skip past the header
214 addq(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
215 addq(cache, tmp); // construct pointer to cache entry
216 }
217
218
219 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
220 // subtype of super_klass.
221 //
222 // Args:
223 // rax: superklass
224 // Rsub_klass: subklass
225 //
226 // Kills:
227 // rcx, rdi
228 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
229 Label& ok_is_subtype) {
230 assert(Rsub_klass != rax, "rax holds superklass");
231 assert(Rsub_klass != r14, "r14 holds locals");
232 assert(Rsub_klass != r13, "r13 holds bcp");
233 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
234 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
235
236 Label not_subtype, loop;
237
238 // Profile the not-null value's klass.
239 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, rdi
240
241 // Load the super-klass's check offset into rcx
242 movl(rcx, Address(rax, sizeof(oopDesc) +
243 Klass::super_check_offset_offset_in_bytes()));
244 // Load from the sub-klass's super-class display list, or a 1-word
245 // cache of the secondary superclass list, or a failing value with a
246 // sentinel offset if the super-klass is an interface or
247 // exceptionally deep in the Java hierarchy and we have to scan the
248 // secondary superclass list the hard way. See if we get an
249 // immediate positive hit
250 cmpq(rax, Address(Rsub_klass, rcx, Address::times_1));
251 jcc(Assembler::equal,ok_is_subtype);
252
253 // Check for immediate negative hit
254 cmpl(rcx, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
255 jcc( Assembler::notEqual, not_subtype );
256 // Check for self
257 cmpq(Rsub_klass, rax);
258 jcc(Assembler::equal, ok_is_subtype);
259
260 // Now do a linear scan of the secondary super-klass chain.
261 movq(rdi, Address(Rsub_klass, sizeof(oopDesc) +
262 Klass::secondary_supers_offset_in_bytes()));
263 // rdi holds the objArrayOop of secondary supers.
264 // Load the array length
265 movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
266 // Skip to start of data; also clear Z flag incase rcx is zero
267 addq(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
268 // Scan rcx words at [rdi] for occurance of rax
269 // Set NZ/Z based on last compare
270
271 // this part is kind tricky, as values in supers array could be 32 or 64 bit wide
272 // and we store values in objArrays always encoded, thus we need to encode value
273 // before repne
274 if (UseCompressedOops) {
275 encode_heap_oop(rax);
276 repne_scanl();
277 // Not equal?
278 jcc(Assembler::notEqual, not_subtype);
279 // decode heap oop here for movq
280 decode_heap_oop(rax);
281 } else {
282 repne_scanq();
283 jcc(Assembler::notEqual, not_subtype);
284 }
285 // Must be equal but missed in cache. Update cache.
286 movq(Address(Rsub_klass, sizeof(oopDesc) +
287 Klass::secondary_super_cache_offset_in_bytes()), rax);
288 jmp(ok_is_subtype);
289
290 bind(not_subtype);
291 // decode heap oop here for miss
292 if (UseCompressedOops) decode_heap_oop(rax);
293 profile_typecheck_failed(rcx); // blows rcx
294 }
295
296
297 // Java Expression Stack
298
299 #ifdef ASSERT
300 // Verifies that the stack tag matches. Must be called before the stack
301 // value is popped off the stack.
302 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t) {
303 if (TaggedStackInterpreter) {
304 frame::Tag tag = t;
305 if (t == frame::TagCategory2) {
306 tag = frame::TagValue;
307 Label hokay;
308 cmpq(Address(rsp, 3*wordSize), (int)tag);
309 jcc(Assembler::equal, hokay);
310 stop("Java Expression stack tag high value is bad");
311 bind(hokay);
312 }
313 Label okay;
314 cmpq(Address(rsp, wordSize), (int)tag);
315 jcc(Assembler::equal, okay);
316 // Also compare if the stack value is zero, then the tag might
317 // not have been set coming from deopt.
318 cmpq(Address(rsp, 0), 0);
319 jcc(Assembler::equal, okay);
320 stop("Java Expression stack tag value is bad");
321 bind(okay);
322 }
323 }
324 #endif // ASSERT
325
326 void InterpreterMacroAssembler::pop_ptr(Register r) {
327 debug_only(verify_stack_tag(frame::TagReference));
328 popq(r);
329 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
330 }
331
332 void InterpreterMacroAssembler::pop_ptr(Register r, Register tag) {
333 popq(r);
334 if (TaggedStackInterpreter) popq(tag);
335 }
336
337 void InterpreterMacroAssembler::pop_i(Register r) {
338 // XXX can't use popq currently, upper half non clean
339 debug_only(verify_stack_tag(frame::TagValue));
340 movl(r, Address(rsp, 0));
341 addq(rsp, wordSize);
342 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
343 }
344
345 void InterpreterMacroAssembler::pop_l(Register r) {
346 debug_only(verify_stack_tag(frame::TagCategory2));
347 movq(r, Address(rsp, 0));
348 addq(rsp, 2 * Interpreter::stackElementSize());
349 }
350
351 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
352 debug_only(verify_stack_tag(frame::TagValue));
353 movflt(r, Address(rsp, 0));
354 addq(rsp, wordSize);
355 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
356 }
357
358 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
359 debug_only(verify_stack_tag(frame::TagCategory2));
360 movdbl(r, Address(rsp, 0));
361 addq(rsp, 2 * Interpreter::stackElementSize());
362 }
363
364 void InterpreterMacroAssembler::push_ptr(Register r) {
365 if (TaggedStackInterpreter) pushq(frame::TagReference);
366 pushq(r);
367 }
368
369 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) {
370 if (TaggedStackInterpreter) pushq(tag);
371 pushq(r);
372 }
373
374 void InterpreterMacroAssembler::push_i(Register r) {
375 if (TaggedStackInterpreter) pushq(frame::TagValue);
376 pushq(r);
377 }
378
379 void InterpreterMacroAssembler::push_l(Register r) {
380 if (TaggedStackInterpreter) {
381 pushq(frame::TagValue);
382 subq(rsp, 1 * wordSize);
383 pushq(frame::TagValue);
384 subq(rsp, 1 * wordSize);
385 } else {
386 subq(rsp, 2 * wordSize);
387 }
388 movq(Address(rsp, 0), r);
389 }
390
391 void InterpreterMacroAssembler::push_f(XMMRegister r) {
392 if (TaggedStackInterpreter) pushq(frame::TagValue);
393 subq(rsp, wordSize);
394 movflt(Address(rsp, 0), r);
395 }
396
397 void InterpreterMacroAssembler::push_d(XMMRegister r) {
398 if (TaggedStackInterpreter) {
399 pushq(frame::TagValue);
400 subq(rsp, 1 * wordSize);
401 pushq(frame::TagValue);
402 subq(rsp, 1 * wordSize);
403 } else {
404 subq(rsp, 2 * wordSize);
405 }
406 movdbl(Address(rsp, 0), r);
407 }
408
409 void InterpreterMacroAssembler::pop(TosState state) {
410 switch (state) {
411 case atos: pop_ptr(); break;
412 case btos:
413 case ctos:
414 case stos:
415 case itos: pop_i(); break;
416 case ltos: pop_l(); break;
417 case ftos: pop_f(); break;
418 case dtos: pop_d(); break;
419 case vtos: /* nothing to do */ break;
420 default: ShouldNotReachHere();
421 }
422 verify_oop(rax, state);
423 }
424
425 void InterpreterMacroAssembler::push(TosState state) {
426 verify_oop(rax, state);
427 switch (state) {
428 case atos: push_ptr(); break;
429 case btos:
430 case ctos:
431 case stos:
432 case itos: push_i(); break;
433 case ltos: push_l(); break;
434 case ftos: push_f(); break;
435 case dtos: push_d(); break;
436 case vtos: /* nothing to do */ break;
437 default : ShouldNotReachHere();
438 }
439 }
440
441
442 // Tagged stack helpers for swap and dup
443 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val,
444 Register tag) {
445 movq(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
446 if (TaggedStackInterpreter) {
447 movq(tag, Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)));
448 }
449 }
450
451 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val,
452 Register tag) {
453 movq(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
454 if (TaggedStackInterpreter) {
455 movq(Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)), tag);
456 }
457 }
458
459
460 // Tagged local support
461 void InterpreterMacroAssembler::tag_local(frame::Tag tag, int n) {
462 if (TaggedStackInterpreter) {
463 if (tag == frame::TagCategory2) {
464 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)),
465 (intptr_t)frame::TagValue);
466 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)),
467 (intptr_t)frame::TagValue);
468 } else {
469 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (intptr_t)tag);
470 }
471 }
472 }
473
474 void InterpreterMacroAssembler::tag_local(frame::Tag tag, Register idx) {
475 if (TaggedStackInterpreter) {
476 if (tag == frame::TagCategory2) {
477 mov64(Address(r14, idx, Address::times_8,
478 Interpreter::local_tag_offset_in_bytes(1)), (intptr_t)frame::TagValue);
479 mov64(Address(r14, idx, Address::times_8,
480 Interpreter::local_tag_offset_in_bytes(0)), (intptr_t)frame::TagValue);
481 } else {
482 mov64(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)),
483 (intptr_t)tag);
484 }
485 }
486 }
487
488 void InterpreterMacroAssembler::tag_local(Register tag, Register idx) {
489 if (TaggedStackInterpreter) {
490 // can only be TagValue or TagReference
491 movq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), tag);
492 }
493 }
494
495
496 void InterpreterMacroAssembler::tag_local(Register tag, int n) {
497 if (TaggedStackInterpreter) {
498 // can only be TagValue or TagReference
499 movq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), tag);
500 }
501 }
502
503 #ifdef ASSERT
504 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, int n) {
505 if (TaggedStackInterpreter) {
506 frame::Tag t = tag;
507 if (tag == frame::TagCategory2) {
508 Label nbl;
509 t = frame::TagValue; // change to what is stored in locals
510 cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)), (int)t);
511 jcc(Assembler::equal, nbl);
512 stop("Local tag is bad for long/double");
513 bind(nbl);
514 }
515 Label notBad;
516 cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (int)t);
517 jcc(Assembler::equal, notBad);
518 // Also compare if the local value is zero, then the tag might
519 // not have been set coming from deopt.
520 cmpq(Address(r14, Interpreter::local_offset_in_bytes(n)), 0);
521 jcc(Assembler::equal, notBad);
522 stop("Local tag is bad");
523 bind(notBad);
524 }
525 }
526
527 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, Register idx) {
528 if (TaggedStackInterpreter) {
529 frame::Tag t = tag;
530 if (tag == frame::TagCategory2) {
531 Label nbl;
532 t = frame::TagValue; // change to what is stored in locals
533 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(1)), (int)t);
534 jcc(Assembler::equal, nbl);
535 stop("Local tag is bad for long/double");
536 bind(nbl);
537 }
538 Label notBad;
539 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), (int)t);
540 jcc(Assembler::equal, notBad);
541 // Also compare if the local value is zero, then the tag might
542 // not have been set coming from deopt.
543 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_offset_in_bytes(0)), 0);
544 jcc(Assembler::equal, notBad);
545 stop("Local tag is bad");
546 bind(notBad);
547 }
548 }
549 #endif // ASSERT
550
551
552 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
553 MacroAssembler::call_VM_leaf_base(entry_point, 0);
554 }
555
556
557 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
558 Register arg_1) {
559 if (c_rarg0 != arg_1) {
560 movq(c_rarg0, arg_1);
561 }
562 MacroAssembler::call_VM_leaf_base(entry_point, 1);
563 }
564
565
566 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
567 Register arg_1,
568 Register arg_2) {
569 assert(c_rarg0 != arg_2, "smashed argument");
570 assert(c_rarg1 != arg_1, "smashed argument");
571 if (c_rarg0 != arg_1) {
572 movq(c_rarg0, arg_1);
573 }
574 if (c_rarg1 != arg_2) {
575 movq(c_rarg1, arg_2);
576 }
577 MacroAssembler::call_VM_leaf_base(entry_point, 2);
578 }
579
580 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
581 Register arg_1,
582 Register arg_2,
583 Register arg_3) {
584 assert(c_rarg0 != arg_2, "smashed argument");
585 assert(c_rarg0 != arg_3, "smashed argument");
586 assert(c_rarg1 != arg_1, "smashed argument");
587 assert(c_rarg1 != arg_3, "smashed argument");
588 assert(c_rarg2 != arg_1, "smashed argument");
589 assert(c_rarg2 != arg_2, "smashed argument");
590 if (c_rarg0 != arg_1) {
591 movq(c_rarg0, arg_1);
592 }
593 if (c_rarg1 != arg_2) {
594 movq(c_rarg1, arg_2);
595 }
596 if (c_rarg2 != arg_3) {
597 movq(c_rarg2, arg_3);
598 }
599 MacroAssembler::call_VM_leaf_base(entry_point, 3);
600 }
601
602 // Jump to from_interpreted entry of a call unless single stepping is possible
603 // in this thread in which case we must call the i2i entry
604 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
605 // set sender sp
606 leaq(r13, Address(rsp, wordSize));
607 // record last_sp
608 movq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
609
610 if (JvmtiExport::can_post_interpreter_events()) {
611 Label run_compiled_code;
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 get_thread(temp);
616 // interp_only is an int, on little endian it is sufficient to test the byte only
617 // Is a cmpl faster (ce
618 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
619 jcc(Assembler::zero, run_compiled_code);
620 jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
621 bind(run_compiled_code);
622 }
623
624 jmp(Address(method, methodOopDesc::from_interpreted_offset()));
625
626 }
627
628
629 // The following two routines provide a hook so that an implementation
630 // can schedule the dispatch in two parts. amd64 does not do this.
631 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
632 // Nothing amd64 specific to be done here
633 }
634
635 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
636 dispatch_next(state, step);
637 }
638
639 void InterpreterMacroAssembler::dispatch_base(TosState state,
640 address* table,
641 bool verifyoop) {
642 verify_FPU(1, state);
643 if (VerifyActivationFrameSize) {
644 Label L;
645 movq(rcx, rbp);
646 subq(rcx, rsp);
647 int min_frame_size =
648 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
649 wordSize;
650 cmpq(rcx, min_frame_size);
651 jcc(Assembler::greaterEqual, L);
652 stop("broken stack frame");
653 bind(L);
654 }
655 if (verifyoop) {
656 verify_oop(rax, state);
657 }
658 lea(rscratch1, ExternalAddress((address)table));
659 jmp(Address(rscratch1, rbx, Address::times_8));
660 }
661
662 void InterpreterMacroAssembler::dispatch_only(TosState state) {
663 dispatch_base(state, Interpreter::dispatch_table(state));
664 }
665
666 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
667 dispatch_base(state, Interpreter::normal_table(state));
668 }
669
670 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
671 dispatch_base(state, Interpreter::normal_table(state), false);
672 }
673
674
675 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
676 // load next bytecode (load before advancing r13 to prevent AGI)
677 load_unsigned_byte(rbx, Address(r13, step));
678 // advance r13
679 incrementq(r13, step);
680 dispatch_base(state, Interpreter::dispatch_table(state));
681 }
682
683 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
684 // load current bytecode
685 load_unsigned_byte(rbx, Address(r13, 0));
686 dispatch_base(state, table);
687 }
688
689 // remove activation
690 //
691 // Unlock the receiver if this is a synchronized method.
692 // Unlock any Java monitors from syncronized blocks.
693 // Remove the activation from the stack.
694 //
695 // If there are locked Java monitors
696 // If throw_monitor_exception
697 // throws IllegalMonitorStateException
698 // Else if install_monitor_exception
699 // installs IllegalMonitorStateException
700 // Else
701 // no error processing
702 void InterpreterMacroAssembler::remove_activation(
703 TosState state,
704 Register ret_addr,
705 bool throw_monitor_exception,
706 bool install_monitor_exception,
707 bool notify_jvmdi) {
708 // Note: Registers rdx xmm0 may be in use for the
709 // result check if synchronized method
710 Label unlocked, unlock, no_unlock;
711
712 // get the value of _do_not_unlock_if_synchronized into rdx
713 const Address do_not_unlock_if_synchronized(r15_thread,
714 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
715 movbool(rdx, do_not_unlock_if_synchronized);
716 movbool(do_not_unlock_if_synchronized, false); // reset the flag
717
718 // get method access flags
719 movq(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
720 movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
721 testl(rcx, JVM_ACC_SYNCHRONIZED);
722 jcc(Assembler::zero, unlocked);
723
724 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
725 // is set.
726 testbool(rdx);
727 jcc(Assembler::notZero, no_unlock);
728
729 // unlock monitor
730 push(state); // save result
731
732 // BasicObjectLock will be first in list, since this is a
733 // synchronized method. However, need to check that the object has
734 // not been unlocked by an explicit monitorexit bytecode.
735 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
736 wordSize - (int) sizeof(BasicObjectLock));
737 // We use c_rarg1 so that if we go slow path it will be the correct
738 // register for unlock_object to pass to VM directly
739 leaq(c_rarg1, monitor); // address of first monitor
740
741 movq(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
742 testq(rax, rax);
743 jcc(Assembler::notZero, unlock);
744
745 pop(state);
746 if (throw_monitor_exception) {
747 // Entry already unlocked, need to throw exception
748 call_VM(noreg, CAST_FROM_FN_PTR(address,
749 InterpreterRuntime::throw_illegal_monitor_state_exception));
750 should_not_reach_here();
751 } else {
752 // Monitor already unlocked during a stack unroll. If requested,
753 // install an illegal_monitor_state_exception. Continue with
754 // stack unrolling.
755 if (install_monitor_exception) {
756 call_VM(noreg, CAST_FROM_FN_PTR(address,
757 InterpreterRuntime::new_illegal_monitor_state_exception));
758 }
759 jmp(unlocked);
760 }
761
762 bind(unlock);
763 unlock_object(c_rarg1);
764 pop(state);
765
766 // Check that for block-structured locking (i.e., that all locked
767 // objects has been unlocked)
768 bind(unlocked);
769
770 // rax: Might contain return value
771
772 // Check that all monitors are unlocked
773 {
774 Label loop, exception, entry, restart;
775 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
776 const Address monitor_block_top(
777 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
778 const Address monitor_block_bot(
779 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
780
781 bind(restart);
782 // We use c_rarg1 so that if we go slow path it will be the correct
783 // register for unlock_object to pass to VM directly
784 movq(c_rarg1, monitor_block_top); // points to current entry, starting
785 // with top-most entry
786 leaq(rbx, monitor_block_bot); // points to word before bottom of
787 // monitor block
788 jmp(entry);
789
790 // Entry already locked, need to throw exception
791 bind(exception);
792
793 if (throw_monitor_exception) {
794 // Throw exception
795 MacroAssembler::call_VM(noreg,
796 CAST_FROM_FN_PTR(address, InterpreterRuntime::
797 throw_illegal_monitor_state_exception));
798 should_not_reach_here();
799 } else {
800 // Stack unrolling. Unlock object and install illegal_monitor_exception.
801 // Unlock does not block, so don't have to worry about the frame.
802 // We don't have to preserve c_rarg1 since we are going to throw an exception.
803
804 push(state);
805 unlock_object(c_rarg1);
806 pop(state);
807
808 if (install_monitor_exception) {
809 call_VM(noreg, CAST_FROM_FN_PTR(address,
810 InterpreterRuntime::
811 new_illegal_monitor_state_exception));
812 }
813
814 jmp(restart);
815 }
816
817 bind(loop);
818 // check if current entry is used
819 cmpq(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int) NULL);
820 jcc(Assembler::notEqual, exception);
821
822 addq(c_rarg1, entry_size); // otherwise advance to next entry
823 bind(entry);
824 cmpq(c_rarg1, rbx); // check if bottom reached
825 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
826 }
827
828 bind(no_unlock);
829
830 // jvmti support
831 if (notify_jvmdi) {
832 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
833 } else {
834 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
835 }
836
837 // remove activation
838 // get sender sp
839 movq(rbx,
840 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
841 leave(); // remove frame anchor
842 popq(ret_addr); // get return address
843 movq(rsp, rbx); // set sp to sender sp
844 }
845
846 // Lock object
847 //
848 // Args:
849 // c_rarg1: BasicObjectLock to be used for locking
850 //
851 // Kills:
852 // rax
853 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
854 // rscratch1, rscratch2 (scratch regs)
855 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
856 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
857
858 if (UseHeavyMonitors) {
859 call_VM(noreg,
860 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
861 lock_reg);
862 } else {
863 Label done;
864
865 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
866 const Register obj_reg = c_rarg3; // Will contain the oop
867
868 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
869 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
870 const int mark_offset = lock_offset +
871 BasicLock::displaced_header_offset_in_bytes();
872
873 Label slow_case;
874
875 // Load object pointer into obj_reg %c_rarg3
876 movq(obj_reg, Address(lock_reg, obj_offset));
877
878 if (UseBiasedLocking) {
879 biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
880 }
881
882 // Load immediate 1 into swap_reg %rax
883 movl(swap_reg, 1);
884
885 // Load (object->mark() | 1) into swap_reg %rax
886 orq(swap_reg, Address(obj_reg, 0));
887
888 // Save (object->mark() | 1) into BasicLock's displaced header
889 movq(Address(lock_reg, mark_offset), swap_reg);
890
891 assert(lock_offset == 0,
892 "displached header must be first word in BasicObjectLock");
893
894 if (os::is_MP()) lock();
895 cmpxchgq(lock_reg, Address(obj_reg, 0));
896 if (PrintBiasedLockingStatistics) {
897 cond_inc32(Assembler::zero,
898 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
899 }
900 jcc(Assembler::zero, done);
901
902 // Test if the oopMark is an obvious stack pointer, i.e.,
903 // 1) (mark & 7) == 0, and
904 // 2) rsp <= mark < mark + os::pagesize()
905 //
906 // These 3 tests can be done by evaluating the following
907 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
908 // assuming both stack pointer and pagesize have their
909 // least significant 3 bits clear.
910 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
911 subq(swap_reg, rsp);
912 andq(swap_reg, 7 - os::vm_page_size());
913
914 // Save the test result, for recursive case, the result is zero
915 movq(Address(lock_reg, mark_offset), swap_reg);
916
917 if (PrintBiasedLockingStatistics) {
918 cond_inc32(Assembler::zero,
919 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
920 }
921 jcc(Assembler::zero, done);
922
923 bind(slow_case);
924
925 // Call the runtime routine for slow case
926 call_VM(noreg,
927 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
928 lock_reg);
929
930 bind(done);
931 }
932 }
933
934
935 // Unlocks an object. Used in monitorexit bytecode and
936 // remove_activation. Throws an IllegalMonitorException if object is
937 // not locked by current thread.
938 //
939 // Args:
940 // c_rarg1: BasicObjectLock for lock
941 //
942 // Kills:
943 // rax
944 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
945 // rscratch1, rscratch2 (scratch regs)
946 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
947 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
948
949 if (UseHeavyMonitors) {
950 call_VM(noreg,
951 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
952 lock_reg);
953 } else {
954 Label done;
955
956 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
957 const Register header_reg = c_rarg2; // Will contain the old oopMark
958 const Register obj_reg = c_rarg3; // Will contain the oop
959
960 save_bcp(); // Save in case of exception
961
962 // Convert from BasicObjectLock structure to object and BasicLock
963 // structure Store the BasicLock address into %rax
964 leaq(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
965
966 // Load oop into obj_reg(%c_rarg3)
967 movq(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
968
969 // Free entry
970 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
971
972 if (UseBiasedLocking) {
973 biased_locking_exit(obj_reg, header_reg, done);
974 }
975
976 // Load the old header from BasicLock structure
977 movq(header_reg, Address(swap_reg,
978 BasicLock::displaced_header_offset_in_bytes()));
979
980 // Test for recursion
981 testq(header_reg, header_reg);
982
983 // zero for recursive case
984 jcc(Assembler::zero, done);
985
986 // Atomic swap back the old header
987 if (os::is_MP()) lock();
988 cmpxchgq(header_reg, Address(obj_reg, 0));
989
990 // zero for recursive case
991 jcc(Assembler::zero, done);
992
993 // Call the runtime routine for slow case.
994 movq(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
995 obj_reg); // restore obj
996 call_VM(noreg,
997 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
998 lock_reg);
999
1000 bind(done);
1001
1002 restore_bcp();
1003 }
1004 }
1005
1006
1007 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1008 Label& zero_continue) {
1009 assert(ProfileInterpreter, "must be profiling interpreter");
1010 movq(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
1011 testq(mdp, mdp);
1012 jcc(Assembler::zero, zero_continue);
1013 }
1014
1015
1016 // Set the method data pointer for the current bcp.
1017 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1018 assert(ProfileInterpreter, "must be profiling interpreter");
1019 Label zero_continue;
1020 pushq(rax);
1021 pushq(rbx);
1022
1023 get_method(rbx);
1024 // Test MDO to avoid the call if it is NULL.
1025 movq(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1026 testq(rax, rax);
1027 jcc(Assembler::zero, zero_continue);
1028
1029 // rbx: method
1030 // r13: bcp
1031 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
1032 // rax: mdi
1033
1034 movq(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1035 testq(rbx, rbx);
1036 jcc(Assembler::zero, zero_continue);
1037 addq(rbx, in_bytes(methodDataOopDesc::data_offset()));
1038 addq(rbx, rax);
1039 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
1040
1041 bind(zero_continue);
1042 popq(rbx);
1043 popq(rax);
1044 }
1045
1046 void InterpreterMacroAssembler::verify_method_data_pointer() {
1047 assert(ProfileInterpreter, "must be profiling interpreter");
1048 #ifdef ASSERT
1049 Label verify_continue;
1050 pushq(rax);
1051 pushq(rbx);
1052 pushq(c_rarg3);
1053 pushq(c_rarg2);
1054 test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
1055 get_method(rbx);
1056
1057 // If the mdp is valid, it will point to a DataLayout header which is
1058 // consistent with the bcp. The converse is highly probable also.
1059 load_unsigned_word(c_rarg2,
1060 Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
1061 addq(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
1062 leaq(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
1063 cmpq(c_rarg2, r13);
1064 jcc(Assembler::equal, verify_continue);
1065 // rbx: method
1066 // r13: bcp
1067 // c_rarg3: mdp
1068 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1069 rbx, r13, c_rarg3);
1070 bind(verify_continue);
1071 popq(c_rarg2);
1072 popq(c_rarg3);
1073 popq(rbx);
1074 popq(rax);
1075 #endif // ASSERT
1076 }
1077
1078
1079 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1080 int constant,
1081 Register value) {
1082 assert(ProfileInterpreter, "must be profiling interpreter");
1083 Address data(mdp_in, constant);
1084 movq(data, value);
1085 }
1086
1087
1088 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1089 int constant,
1090 bool decrement) {
1091 // Counter address
1092 Address data(mdp_in, constant);
1093
1094 increment_mdp_data_at(data, decrement);
1095 }
1096
1097 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1098 bool decrement) {
1099 assert(ProfileInterpreter, "must be profiling interpreter");
1100
1101 if (decrement) {
1102 // Decrement the register. Set condition codes.
1103 addq(data, -DataLayout::counter_increment);
1104 // If the decrement causes the counter to overflow, stay negative
1105 Label L;
1106 jcc(Assembler::negative, L);
1107 addq(data, DataLayout::counter_increment);
1108 bind(L);
1109 } else {
1110 assert(DataLayout::counter_increment == 1,
1111 "flow-free idiom only works with 1");
1112 // Increment the register. Set carry flag.
1113 addq(data, DataLayout::counter_increment);
1114 // If the increment causes the counter to overflow, pull back by 1.
1115 sbbq(data, 0);
1116 }
1117 }
1118
1119
1120 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1121 Register reg,
1122 int constant,
1123 bool decrement) {
1124 Address data(mdp_in, reg, Address::times_1, constant);
1125
1126 increment_mdp_data_at(data, decrement);
1127 }
1128
1129 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1130 int flag_byte_constant) {
1131 assert(ProfileInterpreter, "must be profiling interpreter");
1132 int header_offset = in_bytes(DataLayout::header_offset());
1133 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1134 // Set the flag
1135 orl(Address(mdp_in, header_offset), header_bits);
1136 }
1137
1138
1139
1140 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1141 int offset,
1142 Register value,
1143 Register test_value_out,
1144 Label& not_equal_continue) {
1145 assert(ProfileInterpreter, "must be profiling interpreter");
1146 if (test_value_out == noreg) {
1147 cmpq(value, Address(mdp_in, offset));
1148 } else {
1149 // Put the test value into a register, so caller can use it:
1150 movq(test_value_out, Address(mdp_in, offset));
1151 cmpq(test_value_out, value);
1152 }
1153 jcc(Assembler::notEqual, not_equal_continue);
1154 }
1155
1156
1157 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1158 int offset_of_disp) {
1159 assert(ProfileInterpreter, "must be profiling interpreter");
1160 Address disp_address(mdp_in, offset_of_disp);
1161 addq(mdp_in, disp_address);
1162 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1163 }
1164
1165
1166 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1167 Register reg,
1168 int offset_of_disp) {
1169 assert(ProfileInterpreter, "must be profiling interpreter");
1170 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1171 addq(mdp_in, disp_address);
1172 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1173 }
1174
1175
1176 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1177 int constant) {
1178 assert(ProfileInterpreter, "must be profiling interpreter");
1179 addq(mdp_in, constant);
1180 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1181 }
1182
1183
1184 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1185 assert(ProfileInterpreter, "must be profiling interpreter");
1186 pushq(return_bci); // save/restore across call_VM
1187 call_VM(noreg,
1188 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1189 return_bci);
1190 popq(return_bci);
1191 }
1192
1193
1194 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1195 Register bumped_count) {
1196 if (ProfileInterpreter) {
1197 Label profile_continue;
1198
1199 // If no method data exists, go to profile_continue.
1200 // Otherwise, assign to mdp
1201 test_method_data_pointer(mdp, profile_continue);
1202
1203 // We are taking a branch. Increment the taken count.
1204 // We inline increment_mdp_data_at to return bumped_count in a register
1205 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1206 Address data(mdp, in_bytes(JumpData::taken_offset()));
1207 movq(bumped_count, data);
1208 assert(DataLayout::counter_increment == 1,
1209 "flow-free idiom only works with 1");
1210 addq(bumped_count, DataLayout::counter_increment);
1211 sbbq(bumped_count, 0);
1212 movq(data, bumped_count); // Store back out
1213
1214 // The method data pointer needs to be updated to reflect the new target.
1215 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1216 bind(profile_continue);
1217 }
1218 }
1219
1220
1221 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1222 if (ProfileInterpreter) {
1223 Label profile_continue;
1224
1225 // If no method data exists, go to profile_continue.
1226 test_method_data_pointer(mdp, profile_continue);
1227
1228 // We are taking a branch. Increment the not taken count.
1229 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1230
1231 // The method data pointer needs to be updated to correspond to
1232 // the next bytecode
1233 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1234 bind(profile_continue);
1235 }
1236 }
1237
1238
1239 void InterpreterMacroAssembler::profile_call(Register mdp) {
1240 if (ProfileInterpreter) {
1241 Label profile_continue;
1242
1243 // If no method data exists, go to profile_continue.
1244 test_method_data_pointer(mdp, profile_continue);
1245
1246 // We are making a call. Increment the count.
1247 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1248
1249 // The method data pointer needs to be updated to reflect the new target.
1250 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1251 bind(profile_continue);
1252 }
1253 }
1254
1255
1256 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1257 if (ProfileInterpreter) {
1258 Label profile_continue;
1259
1260 // If no method data exists, go to profile_continue.
1261 test_method_data_pointer(mdp, profile_continue);
1262
1263 // We are making a call. Increment the count.
1264 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1265
1266 // The method data pointer needs to be updated to reflect the new target.
1267 update_mdp_by_constant(mdp,
1268 in_bytes(VirtualCallData::
1269 virtual_call_data_size()));
1270 bind(profile_continue);
1271 }
1272 }
1273
1274
1275 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1276 Register mdp,
1277 Register reg2) {
1278 if (ProfileInterpreter) {
1279 Label profile_continue;
1280
1281 // If no method data exists, go to profile_continue.
1282 test_method_data_pointer(mdp, profile_continue);
1283
1284 // We are making a call. Increment the count.
1285 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1286
1287 // Record the receiver type.
1288 record_klass_in_profile(receiver, mdp, reg2);
1289
1290 // The method data pointer needs to be updated to reflect the new target.
1291 update_mdp_by_constant(mdp,
1292 in_bytes(VirtualCallData::
1293 virtual_call_data_size()));
1294 bind(profile_continue);
1295 }
1296 }
1297
1298 // This routine creates a state machine for updating the multi-row
1299 // type profile at a virtual call site (or other type-sensitive bytecode).
1300 // The machine visits each row (of receiver/count) until the receiver type
1301 // is found, or until it runs out of rows. At the same time, it remembers
1302 // the location of the first empty row. (An empty row records null for its
1303 // receiver, and can be allocated for a newly-observed receiver type.)
1304 // Because there are two degrees of freedom in the state, a simple linear
1305 // search will not work; it must be a decision tree. Hence this helper
1306 // function is recursive, to generate the required tree structured code.
1307 // It's the interpreter, so we are trading off code space for speed.
1308 // See below for example code.
1309 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1310 Register receiver, Register mdp,
1311 Register reg2,
1312 int start_row, Label& done) {
1313 int last_row = VirtualCallData::row_limit() - 1;
1314 assert(start_row <= last_row, "must be work left to do");
1315 // Test this row for both the receiver and for null.
1316 // Take any of three different outcomes:
1317 // 1. found receiver => increment count and goto done
1318 // 2. found null => keep looking for case 1, maybe allocate this cell
1319 // 3. found something else => keep looking for cases 1 and 2
1320 // Case 3 is handled by a recursive call.
1321 for (int row = start_row; row <= last_row; row++) {
1322 Label next_test;
1323 bool test_for_null_also = (row == start_row);
1324
1325 // See if the receiver is receiver[n].
1326 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1327 test_mdp_data_at(mdp, recvr_offset, receiver,
1328 (test_for_null_also ? reg2 : noreg),
1329 next_test);
1330 // (Reg2 now contains the receiver from the CallData.)
1331
1332 // The receiver is receiver[n]. Increment count[n].
1333 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1334 increment_mdp_data_at(mdp, count_offset);
1335 jmp(done);
1336 bind(next_test);
1337
1338 if (test_for_null_also) {
1339 // Failed the equality check on receiver[n]... Test for null.
1340 testq(reg2, reg2);
1341 if (start_row == last_row) {
1342 // The only thing left to do is handle the null case.
1343 jcc(Assembler::notZero, done);
1344 break;
1345 }
1346 // Since null is rare, make it be the branch-taken case.
1347 Label found_null;
1348 jcc(Assembler::zero, found_null);
1349
1350 // Put all the "Case 3" tests here.
1351 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done);
1352
1353 // Found a null. Keep searching for a matching receiver,
1354 // but remember that this is an empty (unused) slot.
1355 bind(found_null);
1356 }
1357 }
1358
1359 // In the fall-through case, we found no matching receiver, but we
1360 // observed the receiver[start_row] is NULL.
1361
1362 // Fill in the receiver field and increment the count.
1363 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1364 set_mdp_data_at(mdp, recvr_offset, receiver);
1365 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1366 movl(reg2, DataLayout::counter_increment);
1367 set_mdp_data_at(mdp, count_offset, reg2);
1368 jmp(done);
1369 }
1370
1371 // Example state machine code for three profile rows:
1372 // // main copy of decision tree, rooted at row[1]
1373 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1374 // if (row[0].rec != NULL) {
1375 // // inner copy of decision tree, rooted at row[1]
1376 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1377 // if (row[1].rec != NULL) {
1378 // // degenerate decision tree, rooted at row[2]
1379 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1380 // if (row[2].rec != NULL) { goto done; } // overflow
1381 // row[2].init(rec); goto done;
1382 // } else {
1383 // // remember row[1] is empty
1384 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1385 // row[1].init(rec); goto done;
1386 // }
1387 // } else {
1388 // // remember row[0] is empty
1389 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1390 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1391 // row[0].init(rec); goto done;
1392 // }
1393
1394 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1395 Register mdp,
1396 Register reg2) {
1397 assert(ProfileInterpreter, "must be profiling");
1398 Label done;
1399
1400 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1401
1402 bind (done);
1403 }
1404
1405 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1406 Register mdp) {
1407 if (ProfileInterpreter) {
1408 Label profile_continue;
1409 uint row;
1410
1411 // If no method data exists, go to profile_continue.
1412 test_method_data_pointer(mdp, profile_continue);
1413
1414 // Update the total ret count.
1415 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1416
1417 for (row = 0; row < RetData::row_limit(); row++) {
1418 Label next_test;
1419
1420 // See if return_bci is equal to bci[n]:
1421 test_mdp_data_at(mdp,
1422 in_bytes(RetData::bci_offset(row)),
1423 return_bci, noreg,
1424 next_test);
1425
1426 // return_bci is equal to bci[n]. Increment the count.
1427 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1428
1429 // The method data pointer needs to be updated to reflect the new target.
1430 update_mdp_by_offset(mdp,
1431 in_bytes(RetData::bci_displacement_offset(row)));
1432 jmp(profile_continue);
1433 bind(next_test);
1434 }
1435
1436 update_mdp_for_ret(return_bci);
1437
1438 bind(profile_continue);
1439 }
1440 }
1441
1442
1443 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1444 if (ProfileInterpreter) {
1445 Label profile_continue;
1446
1447 // If no method data exists, go to profile_continue.
1448 test_method_data_pointer(mdp, profile_continue);
1449
1450 // The method data pointer needs to be updated.
1451 int mdp_delta = in_bytes(BitData::bit_data_size());
1452 if (TypeProfileCasts) {
1453 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1454 }
1455 update_mdp_by_constant(mdp, mdp_delta);
1456
1457 bind(profile_continue);
1458 }
1459 }
1460
1461
1462 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1463 if (ProfileInterpreter && TypeProfileCasts) {
1464 Label profile_continue;
1465
1466 // If no method data exists, go to profile_continue.
1467 test_method_data_pointer(mdp, profile_continue);
1468
1469 int count_offset = in_bytes(CounterData::count_offset());
1470 // Back up the address, since we have already bumped the mdp.
1471 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1472
1473 // *Decrement* the counter. We expect to see zero or small negatives.
1474 increment_mdp_data_at(mdp, count_offset, true);
1475
1476 bind (profile_continue);
1477 }
1478 }
1479
1480
1481 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1482 if (ProfileInterpreter) {
1483 Label profile_continue;
1484
1485 // If no method data exists, go to profile_continue.
1486 test_method_data_pointer(mdp, profile_continue);
1487
1488 // The method data pointer needs to be updated.
1489 int mdp_delta = in_bytes(BitData::bit_data_size());
1490 if (TypeProfileCasts) {
1491 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1492
1493 // Record the object type.
1494 record_klass_in_profile(klass, mdp, reg2);
1495 }
1496 update_mdp_by_constant(mdp, mdp_delta);
1497
1498 bind(profile_continue);
1499 }
1500 }
1501
1502
1503 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1504 if (ProfileInterpreter) {
1505 Label profile_continue;
1506
1507 // If no method data exists, go to profile_continue.
1508 test_method_data_pointer(mdp, profile_continue);
1509
1510 // Update the default case count
1511 increment_mdp_data_at(mdp,
1512 in_bytes(MultiBranchData::default_count_offset()));
1513
1514 // The method data pointer needs to be updated.
1515 update_mdp_by_offset(mdp,
1516 in_bytes(MultiBranchData::
1517 default_displacement_offset()));
1518
1519 bind(profile_continue);
1520 }
1521 }
1522
1523
1524 void InterpreterMacroAssembler::profile_switch_case(Register index,
1525 Register mdp,
1526 Register reg2) {
1527 if (ProfileInterpreter) {
1528 Label profile_continue;
1529
1530 // If no method data exists, go to profile_continue.
1531 test_method_data_pointer(mdp, profile_continue);
1532
1533 // Build the base (index * per_case_size_in_bytes()) +
1534 // case_array_offset_in_bytes()
1535 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1536 imulq(index, reg2); // XXX l ?
1537 addq(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1538
1539 // Update the case count
1540 increment_mdp_data_at(mdp,
1541 index,
1542 in_bytes(MultiBranchData::relative_count_offset()));
1543
1544 // The method data pointer needs to be updated.
1545 update_mdp_by_offset(mdp,
1546 index,
1547 in_bytes(MultiBranchData::
1548 relative_displacement_offset()));
1549
1550 bind(profile_continue);
1551 }
1552 }
1553
1554
1555 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1556 if (state == atos) {
1557 MacroAssembler::verify_oop(reg);
1558 }
1559 }
1560
1561 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1562 }
1563
1564
1565 void InterpreterMacroAssembler::notify_method_entry() {
1566 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1567 // track stack depth. If it is possible to enter interp_only_mode we add
1568 // the code to check if the event should be sent.
1569 if (JvmtiExport::can_post_interpreter_events()) {
1570 Label L;
1571 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1572 testl(rdx, rdx);
1573 jcc(Assembler::zero, L);
1574 call_VM(noreg, CAST_FROM_FN_PTR(address,
1575 InterpreterRuntime::post_method_entry));
1576 bind(L);
1577 }
1578
1579 {
1580 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1581 get_method(c_rarg1);
1582 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1583 r15_thread, c_rarg1);
1584 }
1585 }
1586
1587
1588 void InterpreterMacroAssembler::notify_method_exit(
1589 TosState state, NotifyMethodExitMode mode) {
1590 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1591 // track stack depth. If it is possible to enter interp_only_mode we add
1592 // the code to check if the event should be sent.
1593 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1594 Label L;
1595 // Note: frame::interpreter_frame_result has a dependency on how the
1596 // method result is saved across the call to post_method_exit. If this
1597 // is changed then the interpreter_frame_result implementation will
1598 // need to be updated too.
1599 push(state);
1600 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1601 testl(rdx, rdx);
1602 jcc(Assembler::zero, L);
1603 call_VM(noreg,
1604 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1605 bind(L);
1606 pop(state);
1607 }
1608
1609 {
1610 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1611 push(state);
1612 get_method(c_rarg1);
1613 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1614 r15_thread, c_rarg1);
1615 pop(state);
1616 }
1617 }