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, not_subtype_pop, 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 pushq(rax);
276 encode_heap_oop(rax);
277 repne_scanl();
278 // Not equal?
279 jcc(Assembler::notEqual, not_subtype_pop);
280 // restore heap oop here for movq
281 popq(rax);
282 } else {
283 repne_scanq();
284 jcc(Assembler::notEqual, not_subtype);
285 }
286 // Must be equal but missed in cache. Update cache.
287 movq(Address(Rsub_klass, sizeof(oopDesc) +
288 Klass::secondary_super_cache_offset_in_bytes()), rax);
289 jmp(ok_is_subtype);
290
291 bind(not_subtype_pop);
292 // restore heap oop here for miss
293 if (UseCompressedOops) popq(rax);
294 bind(not_subtype);
295 profile_typecheck_failed(rcx); // blows rcx
296 }
297
298
299 // Java Expression Stack
300
301 #ifdef ASSERT
302 // Verifies that the stack tag matches. Must be called before the stack
303 // value is popped off the stack.
304 void InterpreterMacroAssembler::verify_stack_tag(frame::Tag t) {
305 if (TaggedStackInterpreter) {
306 frame::Tag tag = t;
307 if (t == frame::TagCategory2) {
308 tag = frame::TagValue;
309 Label hokay;
310 cmpq(Address(rsp, 3*wordSize), (int)tag);
311 jcc(Assembler::equal, hokay);
312 stop("Java Expression stack tag high value is bad");
313 bind(hokay);
314 }
315 Label okay;
316 cmpq(Address(rsp, wordSize), (int)tag);
317 jcc(Assembler::equal, okay);
318 // Also compare if the stack value is zero, then the tag might
319 // not have been set coming from deopt.
320 cmpq(Address(rsp, 0), 0);
321 jcc(Assembler::equal, okay);
322 stop("Java Expression stack tag value is bad");
323 bind(okay);
324 }
325 }
326 #endif // ASSERT
327
328 void InterpreterMacroAssembler::pop_ptr(Register r) {
329 debug_only(verify_stack_tag(frame::TagReference));
330 popq(r);
331 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
332 }
333
334 void InterpreterMacroAssembler::pop_ptr(Register r, Register tag) {
335 popq(r);
336 if (TaggedStackInterpreter) popq(tag);
337 }
338
339 void InterpreterMacroAssembler::pop_i(Register r) {
340 // XXX can't use popq currently, upper half non clean
341 debug_only(verify_stack_tag(frame::TagValue));
342 movl(r, Address(rsp, 0));
343 addq(rsp, wordSize);
344 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
345 }
346
347 void InterpreterMacroAssembler::pop_l(Register r) {
348 debug_only(verify_stack_tag(frame::TagCategory2));
349 movq(r, Address(rsp, 0));
350 addq(rsp, 2 * Interpreter::stackElementSize());
351 }
352
353 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
354 debug_only(verify_stack_tag(frame::TagValue));
355 movflt(r, Address(rsp, 0));
356 addq(rsp, wordSize);
357 if (TaggedStackInterpreter) addq(rsp, 1 * wordSize);
358 }
359
360 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
361 debug_only(verify_stack_tag(frame::TagCategory2));
362 movdbl(r, Address(rsp, 0));
363 addq(rsp, 2 * Interpreter::stackElementSize());
364 }
365
366 void InterpreterMacroAssembler::push_ptr(Register r) {
367 if (TaggedStackInterpreter) pushq(frame::TagReference);
368 pushq(r);
369 }
370
371 void InterpreterMacroAssembler::push_ptr(Register r, Register tag) {
372 if (TaggedStackInterpreter) pushq(tag);
373 pushq(r);
374 }
375
376 void InterpreterMacroAssembler::push_i(Register r) {
377 if (TaggedStackInterpreter) pushq(frame::TagValue);
378 pushq(r);
379 }
380
381 void InterpreterMacroAssembler::push_l(Register r) {
382 if (TaggedStackInterpreter) {
383 pushq(frame::TagValue);
384 subq(rsp, 1 * wordSize);
385 pushq(frame::TagValue);
386 subq(rsp, 1 * wordSize);
387 } else {
388 subq(rsp, 2 * wordSize);
389 }
390 movq(Address(rsp, 0), r);
391 }
392
393 void InterpreterMacroAssembler::push_f(XMMRegister r) {
394 if (TaggedStackInterpreter) pushq(frame::TagValue);
395 subq(rsp, wordSize);
396 movflt(Address(rsp, 0), r);
397 }
398
399 void InterpreterMacroAssembler::push_d(XMMRegister r) {
400 if (TaggedStackInterpreter) {
401 pushq(frame::TagValue);
402 subq(rsp, 1 * wordSize);
403 pushq(frame::TagValue);
404 subq(rsp, 1 * wordSize);
405 } else {
406 subq(rsp, 2 * wordSize);
407 }
408 movdbl(Address(rsp, 0), r);
409 }
410
411 void InterpreterMacroAssembler::pop(TosState state) {
412 switch (state) {
413 case atos: pop_ptr(); break;
414 case btos:
415 case ctos:
416 case stos:
417 case itos: pop_i(); break;
418 case ltos: pop_l(); break;
419 case ftos: pop_f(); break;
420 case dtos: pop_d(); break;
421 case vtos: /* nothing to do */ break;
422 default: ShouldNotReachHere();
423 }
424 verify_oop(rax, state);
425 }
426
427 void InterpreterMacroAssembler::push(TosState state) {
428 verify_oop(rax, state);
429 switch (state) {
430 case atos: push_ptr(); break;
431 case btos:
432 case ctos:
433 case stos:
434 case itos: push_i(); break;
435 case ltos: push_l(); break;
436 case ftos: push_f(); break;
437 case dtos: push_d(); break;
438 case vtos: /* nothing to do */ break;
439 default : ShouldNotReachHere();
440 }
441 }
442
443
444 // Tagged stack helpers for swap and dup
445 void InterpreterMacroAssembler::load_ptr_and_tag(int n, Register val,
446 Register tag) {
447 movq(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
448 if (TaggedStackInterpreter) {
449 movq(tag, Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)));
450 }
451 }
452
453 void InterpreterMacroAssembler::store_ptr_and_tag(int n, Register val,
454 Register tag) {
455 movq(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
456 if (TaggedStackInterpreter) {
457 movq(Address(rsp, Interpreter::expr_tag_offset_in_bytes(n)), tag);
458 }
459 }
460
461
462 // Tagged local support
463 void InterpreterMacroAssembler::tag_local(frame::Tag tag, int n) {
464 if (TaggedStackInterpreter) {
465 if (tag == frame::TagCategory2) {
466 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)),
467 (intptr_t)frame::TagValue);
468 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)),
469 (intptr_t)frame::TagValue);
470 } else {
471 mov64(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (intptr_t)tag);
472 }
473 }
474 }
475
476 void InterpreterMacroAssembler::tag_local(frame::Tag tag, Register idx) {
477 if (TaggedStackInterpreter) {
478 if (tag == frame::TagCategory2) {
479 mov64(Address(r14, idx, Address::times_8,
480 Interpreter::local_tag_offset_in_bytes(1)), (intptr_t)frame::TagValue);
481 mov64(Address(r14, idx, Address::times_8,
482 Interpreter::local_tag_offset_in_bytes(0)), (intptr_t)frame::TagValue);
483 } else {
484 mov64(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)),
485 (intptr_t)tag);
486 }
487 }
488 }
489
490 void InterpreterMacroAssembler::tag_local(Register tag, Register idx) {
491 if (TaggedStackInterpreter) {
492 // can only be TagValue or TagReference
493 movq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), tag);
494 }
495 }
496
497
498 void InterpreterMacroAssembler::tag_local(Register tag, int n) {
499 if (TaggedStackInterpreter) {
500 // can only be TagValue or TagReference
501 movq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), tag);
502 }
503 }
504
505 #ifdef ASSERT
506 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, int n) {
507 if (TaggedStackInterpreter) {
508 frame::Tag t = tag;
509 if (tag == frame::TagCategory2) {
510 Label nbl;
511 t = frame::TagValue; // change to what is stored in locals
512 cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n+1)), (int)t);
513 jcc(Assembler::equal, nbl);
514 stop("Local tag is bad for long/double");
515 bind(nbl);
516 }
517 Label notBad;
518 cmpq(Address(r14, Interpreter::local_tag_offset_in_bytes(n)), (int)t);
519 jcc(Assembler::equal, notBad);
520 // Also compare if the local value is zero, then the tag might
521 // not have been set coming from deopt.
522 cmpq(Address(r14, Interpreter::local_offset_in_bytes(n)), 0);
523 jcc(Assembler::equal, notBad);
524 stop("Local tag is bad");
525 bind(notBad);
526 }
527 }
528
529 void InterpreterMacroAssembler::verify_local_tag(frame::Tag tag, Register idx) {
530 if (TaggedStackInterpreter) {
531 frame::Tag t = tag;
532 if (tag == frame::TagCategory2) {
533 Label nbl;
534 t = frame::TagValue; // change to what is stored in locals
535 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(1)), (int)t);
536 jcc(Assembler::equal, nbl);
537 stop("Local tag is bad for long/double");
538 bind(nbl);
539 }
540 Label notBad;
541 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_tag_offset_in_bytes(0)), (int)t);
542 jcc(Assembler::equal, notBad);
543 // Also compare if the local value is zero, then the tag might
544 // not have been set coming from deopt.
545 cmpq(Address(r14, idx, Address::times_8, Interpreter::local_offset_in_bytes(0)), 0);
546 jcc(Assembler::equal, notBad);
547 stop("Local tag is bad");
548 bind(notBad);
549 }
550 }
551 #endif // ASSERT
552
553
554 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point) {
555 MacroAssembler::call_VM_leaf_base(entry_point, 0);
556 }
557
558
559 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
560 Register arg_1) {
561 if (c_rarg0 != arg_1) {
562 movq(c_rarg0, arg_1);
563 }
564 MacroAssembler::call_VM_leaf_base(entry_point, 1);
565 }
566
567
568 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
569 Register arg_1,
570 Register arg_2) {
571 assert(c_rarg0 != arg_2, "smashed argument");
572 assert(c_rarg1 != arg_1, "smashed argument");
573 if (c_rarg0 != arg_1) {
574 movq(c_rarg0, arg_1);
575 }
576 if (c_rarg1 != arg_2) {
577 movq(c_rarg1, arg_2);
578 }
579 MacroAssembler::call_VM_leaf_base(entry_point, 2);
580 }
581
582 void InterpreterMacroAssembler::super_call_VM_leaf(address entry_point,
583 Register arg_1,
584 Register arg_2,
585 Register arg_3) {
586 assert(c_rarg0 != arg_2, "smashed argument");
587 assert(c_rarg0 != arg_3, "smashed argument");
588 assert(c_rarg1 != arg_1, "smashed argument");
589 assert(c_rarg1 != arg_3, "smashed argument");
590 assert(c_rarg2 != arg_1, "smashed argument");
591 assert(c_rarg2 != arg_2, "smashed argument");
592 if (c_rarg0 != arg_1) {
593 movq(c_rarg0, arg_1);
594 }
595 if (c_rarg1 != arg_2) {
596 movq(c_rarg1, arg_2);
597 }
598 if (c_rarg2 != arg_3) {
599 movq(c_rarg2, arg_3);
600 }
601 MacroAssembler::call_VM_leaf_base(entry_point, 3);
602 }
603
604 // Jump to from_interpreted entry of a call unless single stepping is possible
605 // in this thread in which case we must call the i2i entry
606 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
607 // set sender sp
608 leaq(r13, Address(rsp, wordSize));
609 // record last_sp
610 movq(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), r13);
611
612 if (JvmtiExport::can_post_interpreter_events()) {
613 Label run_compiled_code;
614 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
615 // compiled code in threads for which the event is enabled. Check here for
616 // interp_only_mode if these events CAN be enabled.
617 get_thread(temp);
618 // interp_only is an int, on little endian it is sufficient to test the byte only
619 // Is a cmpl faster (ce
620 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
621 jcc(Assembler::zero, run_compiled_code);
622 jmp(Address(method, methodOopDesc::interpreter_entry_offset()));
623 bind(run_compiled_code);
624 }
625
626 jmp(Address(method, methodOopDesc::from_interpreted_offset()));
627
628 }
629
630
631 // The following two routines provide a hook so that an implementation
632 // can schedule the dispatch in two parts. amd64 does not do this.
633 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
634 // Nothing amd64 specific to be done here
635 }
636
637 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
638 dispatch_next(state, step);
639 }
640
641 void InterpreterMacroAssembler::dispatch_base(TosState state,
642 address* table,
643 bool verifyoop) {
644 verify_FPU(1, state);
645 if (VerifyActivationFrameSize) {
646 Label L;
647 movq(rcx, rbp);
648 subq(rcx, rsp);
649 int min_frame_size =
650 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
651 wordSize;
652 cmpq(rcx, min_frame_size);
653 jcc(Assembler::greaterEqual, L);
654 stop("broken stack frame");
655 bind(L);
656 }
657 if (verifyoop) {
658 verify_oop(rax, state);
659 }
660 lea(rscratch1, ExternalAddress((address)table));
661 jmp(Address(rscratch1, rbx, Address::times_8));
662 }
663
664 void InterpreterMacroAssembler::dispatch_only(TosState state) {
665 dispatch_base(state, Interpreter::dispatch_table(state));
666 }
667
668 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
669 dispatch_base(state, Interpreter::normal_table(state));
670 }
671
672 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
673 dispatch_base(state, Interpreter::normal_table(state), false);
674 }
675
676
677 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
678 // load next bytecode (load before advancing r13 to prevent AGI)
679 load_unsigned_byte(rbx, Address(r13, step));
680 // advance r13
681 incrementq(r13, step);
682 dispatch_base(state, Interpreter::dispatch_table(state));
683 }
684
685 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
686 // load current bytecode
687 load_unsigned_byte(rbx, Address(r13, 0));
688 dispatch_base(state, table);
689 }
690
691 // remove activation
692 //
693 // Unlock the receiver if this is a synchronized method.
694 // Unlock any Java monitors from syncronized blocks.
695 // Remove the activation from the stack.
696 //
697 // If there are locked Java monitors
698 // If throw_monitor_exception
699 // throws IllegalMonitorStateException
700 // Else if install_monitor_exception
701 // installs IllegalMonitorStateException
702 // Else
703 // no error processing
704 void InterpreterMacroAssembler::remove_activation(
705 TosState state,
706 Register ret_addr,
707 bool throw_monitor_exception,
708 bool install_monitor_exception,
709 bool notify_jvmdi) {
710 // Note: Registers rdx xmm0 may be in use for the
711 // result check if synchronized method
712 Label unlocked, unlock, no_unlock;
713
714 // get the value of _do_not_unlock_if_synchronized into rdx
715 const Address do_not_unlock_if_synchronized(r15_thread,
716 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
717 movbool(rdx, do_not_unlock_if_synchronized);
718 movbool(do_not_unlock_if_synchronized, false); // reset the flag
719
720 // get method access flags
721 movq(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
722 movl(rcx, Address(rbx, methodOopDesc::access_flags_offset()));
723 testl(rcx, JVM_ACC_SYNCHRONIZED);
724 jcc(Assembler::zero, unlocked);
725
726 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
727 // is set.
728 testbool(rdx);
729 jcc(Assembler::notZero, no_unlock);
730
731 // unlock monitor
732 push(state); // save result
733
734 // BasicObjectLock will be first in list, since this is a
735 // synchronized method. However, need to check that the object has
736 // not been unlocked by an explicit monitorexit bytecode.
737 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
738 wordSize - (int) sizeof(BasicObjectLock));
739 // We use c_rarg1 so that if we go slow path it will be the correct
740 // register for unlock_object to pass to VM directly
741 leaq(c_rarg1, monitor); // address of first monitor
742
743 movq(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
744 testq(rax, rax);
745 jcc(Assembler::notZero, unlock);
746
747 pop(state);
748 if (throw_monitor_exception) {
749 // Entry already unlocked, need to throw exception
750 call_VM(noreg, CAST_FROM_FN_PTR(address,
751 InterpreterRuntime::throw_illegal_monitor_state_exception));
752 should_not_reach_here();
753 } else {
754 // Monitor already unlocked during a stack unroll. If requested,
755 // install an illegal_monitor_state_exception. Continue with
756 // stack unrolling.
757 if (install_monitor_exception) {
758 call_VM(noreg, CAST_FROM_FN_PTR(address,
759 InterpreterRuntime::new_illegal_monitor_state_exception));
760 }
761 jmp(unlocked);
762 }
763
764 bind(unlock);
765 unlock_object(c_rarg1);
766 pop(state);
767
768 // Check that for block-structured locking (i.e., that all locked
769 // objects has been unlocked)
770 bind(unlocked);
771
772 // rax: Might contain return value
773
774 // Check that all monitors are unlocked
775 {
776 Label loop, exception, entry, restart;
777 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
778 const Address monitor_block_top(
779 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
780 const Address monitor_block_bot(
781 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
782
783 bind(restart);
784 // We use c_rarg1 so that if we go slow path it will be the correct
785 // register for unlock_object to pass to VM directly
786 movq(c_rarg1, monitor_block_top); // points to current entry, starting
787 // with top-most entry
788 leaq(rbx, monitor_block_bot); // points to word before bottom of
789 // monitor block
790 jmp(entry);
791
792 // Entry already locked, need to throw exception
793 bind(exception);
794
795 if (throw_monitor_exception) {
796 // Throw exception
797 MacroAssembler::call_VM(noreg,
798 CAST_FROM_FN_PTR(address, InterpreterRuntime::
799 throw_illegal_monitor_state_exception));
800 should_not_reach_here();
801 } else {
802 // Stack unrolling. Unlock object and install illegal_monitor_exception.
803 // Unlock does not block, so don't have to worry about the frame.
804 // We don't have to preserve c_rarg1 since we are going to throw an exception.
805
806 push(state);
807 unlock_object(c_rarg1);
808 pop(state);
809
810 if (install_monitor_exception) {
811 call_VM(noreg, CAST_FROM_FN_PTR(address,
812 InterpreterRuntime::
813 new_illegal_monitor_state_exception));
814 }
815
816 jmp(restart);
817 }
818
819 bind(loop);
820 // check if current entry is used
821 cmpq(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), (int) NULL);
822 jcc(Assembler::notEqual, exception);
823
824 addq(c_rarg1, entry_size); // otherwise advance to next entry
825 bind(entry);
826 cmpq(c_rarg1, rbx); // check if bottom reached
827 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
828 }
829
830 bind(no_unlock);
831
832 // jvmti support
833 if (notify_jvmdi) {
834 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
835 } else {
836 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
837 }
838
839 // remove activation
840 // get sender sp
841 movq(rbx,
842 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
843 leave(); // remove frame anchor
844 popq(ret_addr); // get return address
845 movq(rsp, rbx); // set sp to sender sp
846 }
847
848 // Lock object
849 //
850 // Args:
851 // c_rarg1: BasicObjectLock to be used for locking
852 //
853 // Kills:
854 // rax
855 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, .. (param regs)
856 // rscratch1, rscratch2 (scratch regs)
857 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
858 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be c_rarg1");
859
860 if (UseHeavyMonitors) {
861 call_VM(noreg,
862 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
863 lock_reg);
864 } else {
865 Label done;
866
867 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
868 const Register obj_reg = c_rarg3; // Will contain the oop
869
870 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
871 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
872 const int mark_offset = lock_offset +
873 BasicLock::displaced_header_offset_in_bytes();
874
875 Label slow_case;
876
877 // Load object pointer into obj_reg %c_rarg3
878 movq(obj_reg, Address(lock_reg, obj_offset));
879
880 if (UseBiasedLocking) {
881 biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
882 }
883
884 // Load immediate 1 into swap_reg %rax
885 movl(swap_reg, 1);
886
887 // Load (object->mark() | 1) into swap_reg %rax
888 orq(swap_reg, Address(obj_reg, 0));
889
890 // Save (object->mark() | 1) into BasicLock's displaced header
891 movq(Address(lock_reg, mark_offset), swap_reg);
892
893 assert(lock_offset == 0,
894 "displached header must be first word in BasicObjectLock");
895
896 if (os::is_MP()) lock();
897 cmpxchgq(lock_reg, Address(obj_reg, 0));
898 if (PrintBiasedLockingStatistics) {
899 cond_inc32(Assembler::zero,
900 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
901 }
902 jcc(Assembler::zero, done);
903
904 // Test if the oopMark is an obvious stack pointer, i.e.,
905 // 1) (mark & 7) == 0, and
906 // 2) rsp <= mark < mark + os::pagesize()
907 //
908 // These 3 tests can be done by evaluating the following
909 // expression: ((mark - rsp) & (7 - os::vm_page_size())),
910 // assuming both stack pointer and pagesize have their
911 // least significant 3 bits clear.
912 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
913 subq(swap_reg, rsp);
914 andq(swap_reg, 7 - os::vm_page_size());
915
916 // Save the test result, for recursive case, the result is zero
917 movq(Address(lock_reg, mark_offset), swap_reg);
918
919 if (PrintBiasedLockingStatistics) {
920 cond_inc32(Assembler::zero,
921 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
922 }
923 jcc(Assembler::zero, done);
924
925 bind(slow_case);
926
927 // Call the runtime routine for slow case
928 call_VM(noreg,
929 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
930 lock_reg);
931
932 bind(done);
933 }
934 }
935
936
937 // Unlocks an object. Used in monitorexit bytecode and
938 // remove_activation. Throws an IllegalMonitorException if object is
939 // not locked by current thread.
940 //
941 // Args:
942 // c_rarg1: BasicObjectLock for lock
943 //
944 // Kills:
945 // rax
946 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
947 // rscratch1, rscratch2 (scratch regs)
948 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
949 assert(lock_reg == c_rarg1, "The argument is only for looks. It must be rarg1");
950
951 if (UseHeavyMonitors) {
952 call_VM(noreg,
953 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
954 lock_reg);
955 } else {
956 Label done;
957
958 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
959 const Register header_reg = c_rarg2; // Will contain the old oopMark
960 const Register obj_reg = c_rarg3; // Will contain the oop
961
962 save_bcp(); // Save in case of exception
963
964 // Convert from BasicObjectLock structure to object and BasicLock
965 // structure Store the BasicLock address into %rax
966 leaq(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
967
968 // Load oop into obj_reg(%c_rarg3)
969 movq(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
970
971 // Free entry
972 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
973
974 if (UseBiasedLocking) {
975 biased_locking_exit(obj_reg, header_reg, done);
976 }
977
978 // Load the old header from BasicLock structure
979 movq(header_reg, Address(swap_reg,
980 BasicLock::displaced_header_offset_in_bytes()));
981
982 // Test for recursion
983 testq(header_reg, header_reg);
984
985 // zero for recursive case
986 jcc(Assembler::zero, done);
987
988 // Atomic swap back the old header
989 if (os::is_MP()) lock();
990 cmpxchgq(header_reg, Address(obj_reg, 0));
991
992 // zero for recursive case
993 jcc(Assembler::zero, done);
994
995 // Call the runtime routine for slow case.
996 movq(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
997 obj_reg); // restore obj
998 call_VM(noreg,
999 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
1000 lock_reg);
1001
1002 bind(done);
1003
1004 restore_bcp();
1005 }
1006 }
1007
1008
1009 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
1010 Label& zero_continue) {
1011 assert(ProfileInterpreter, "must be profiling interpreter");
1012 movq(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
1013 testq(mdp, mdp);
1014 jcc(Assembler::zero, zero_continue);
1015 }
1016
1017
1018 // Set the method data pointer for the current bcp.
1019 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1020 assert(ProfileInterpreter, "must be profiling interpreter");
1021 Label zero_continue;
1022 pushq(rax);
1023 pushq(rbx);
1024
1025 get_method(rbx);
1026 // Test MDO to avoid the call if it is NULL.
1027 movq(rax, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1028 testq(rax, rax);
1029 jcc(Assembler::zero, zero_continue);
1030
1031 // rbx: method
1032 // r13: bcp
1033 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, r13);
1034 // rax: mdi
1035
1036 movq(rbx, Address(rbx, in_bytes(methodOopDesc::method_data_offset())));
1037 testq(rbx, rbx);
1038 jcc(Assembler::zero, zero_continue);
1039 addq(rbx, in_bytes(methodDataOopDesc::data_offset()));
1040 addq(rbx, rax);
1041 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rbx);
1042
1043 bind(zero_continue);
1044 popq(rbx);
1045 popq(rax);
1046 }
1047
1048 void InterpreterMacroAssembler::verify_method_data_pointer() {
1049 assert(ProfileInterpreter, "must be profiling interpreter");
1050 #ifdef ASSERT
1051 Label verify_continue;
1052 pushq(rax);
1053 pushq(rbx);
1054 pushq(c_rarg3);
1055 pushq(c_rarg2);
1056 test_method_data_pointer(c_rarg3, verify_continue); // If mdp is zero, continue
1057 get_method(rbx);
1058
1059 // If the mdp is valid, it will point to a DataLayout header which is
1060 // consistent with the bcp. The converse is highly probable also.
1061 load_unsigned_word(c_rarg2,
1062 Address(c_rarg3, in_bytes(DataLayout::bci_offset())));
1063 addq(c_rarg2, Address(rbx, methodOopDesc::const_offset()));
1064 leaq(c_rarg2, Address(c_rarg2, constMethodOopDesc::codes_offset()));
1065 cmpq(c_rarg2, r13);
1066 jcc(Assembler::equal, verify_continue);
1067 // rbx: method
1068 // r13: bcp
1069 // c_rarg3: mdp
1070 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1071 rbx, r13, c_rarg3);
1072 bind(verify_continue);
1073 popq(c_rarg2);
1074 popq(c_rarg3);
1075 popq(rbx);
1076 popq(rax);
1077 #endif // ASSERT
1078 }
1079
1080
1081 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1082 int constant,
1083 Register value) {
1084 assert(ProfileInterpreter, "must be profiling interpreter");
1085 Address data(mdp_in, constant);
1086 movq(data, value);
1087 }
1088
1089
1090 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1091 int constant,
1092 bool decrement) {
1093 // Counter address
1094 Address data(mdp_in, constant);
1095
1096 increment_mdp_data_at(data, decrement);
1097 }
1098
1099 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1100 bool decrement) {
1101 assert(ProfileInterpreter, "must be profiling interpreter");
1102
1103 if (decrement) {
1104 // Decrement the register. Set condition codes.
1105 addq(data, -DataLayout::counter_increment);
1106 // If the decrement causes the counter to overflow, stay negative
1107 Label L;
1108 jcc(Assembler::negative, L);
1109 addq(data, DataLayout::counter_increment);
1110 bind(L);
1111 } else {
1112 assert(DataLayout::counter_increment == 1,
1113 "flow-free idiom only works with 1");
1114 // Increment the register. Set carry flag.
1115 addq(data, DataLayout::counter_increment);
1116 // If the increment causes the counter to overflow, pull back by 1.
1117 sbbq(data, 0);
1118 }
1119 }
1120
1121
1122 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1123 Register reg,
1124 int constant,
1125 bool decrement) {
1126 Address data(mdp_in, reg, Address::times_1, constant);
1127
1128 increment_mdp_data_at(data, decrement);
1129 }
1130
1131 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1132 int flag_byte_constant) {
1133 assert(ProfileInterpreter, "must be profiling interpreter");
1134 int header_offset = in_bytes(DataLayout::header_offset());
1135 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1136 // Set the flag
1137 orl(Address(mdp_in, header_offset), header_bits);
1138 }
1139
1140
1141
1142 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1143 int offset,
1144 Register value,
1145 Register test_value_out,
1146 Label& not_equal_continue) {
1147 assert(ProfileInterpreter, "must be profiling interpreter");
1148 if (test_value_out == noreg) {
1149 cmpq(value, Address(mdp_in, offset));
1150 } else {
1151 // Put the test value into a register, so caller can use it:
1152 movq(test_value_out, Address(mdp_in, offset));
1153 cmpq(test_value_out, value);
1154 }
1155 jcc(Assembler::notEqual, not_equal_continue);
1156 }
1157
1158
1159 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1160 int offset_of_disp) {
1161 assert(ProfileInterpreter, "must be profiling interpreter");
1162 Address disp_address(mdp_in, offset_of_disp);
1163 addq(mdp_in, disp_address);
1164 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1165 }
1166
1167
1168 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1169 Register reg,
1170 int offset_of_disp) {
1171 assert(ProfileInterpreter, "must be profiling interpreter");
1172 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1173 addq(mdp_in, disp_address);
1174 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1175 }
1176
1177
1178 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1179 int constant) {
1180 assert(ProfileInterpreter, "must be profiling interpreter");
1181 addq(mdp_in, constant);
1182 movq(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
1183 }
1184
1185
1186 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1187 assert(ProfileInterpreter, "must be profiling interpreter");
1188 pushq(return_bci); // save/restore across call_VM
1189 call_VM(noreg,
1190 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1191 return_bci);
1192 popq(return_bci);
1193 }
1194
1195
1196 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1197 Register bumped_count) {
1198 if (ProfileInterpreter) {
1199 Label profile_continue;
1200
1201 // If no method data exists, go to profile_continue.
1202 // Otherwise, assign to mdp
1203 test_method_data_pointer(mdp, profile_continue);
1204
1205 // We are taking a branch. Increment the taken count.
1206 // We inline increment_mdp_data_at to return bumped_count in a register
1207 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1208 Address data(mdp, in_bytes(JumpData::taken_offset()));
1209 movq(bumped_count, data);
1210 assert(DataLayout::counter_increment == 1,
1211 "flow-free idiom only works with 1");
1212 addq(bumped_count, DataLayout::counter_increment);
1213 sbbq(bumped_count, 0);
1214 movq(data, bumped_count); // Store back out
1215
1216 // The method data pointer needs to be updated to reflect the new target.
1217 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1218 bind(profile_continue);
1219 }
1220 }
1221
1222
1223 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1224 if (ProfileInterpreter) {
1225 Label profile_continue;
1226
1227 // If no method data exists, go to profile_continue.
1228 test_method_data_pointer(mdp, profile_continue);
1229
1230 // We are taking a branch. Increment the not taken count.
1231 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1232
1233 // The method data pointer needs to be updated to correspond to
1234 // the next bytecode
1235 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1236 bind(profile_continue);
1237 }
1238 }
1239
1240
1241 void InterpreterMacroAssembler::profile_call(Register mdp) {
1242 if (ProfileInterpreter) {
1243 Label profile_continue;
1244
1245 // If no method data exists, go to profile_continue.
1246 test_method_data_pointer(mdp, profile_continue);
1247
1248 // We are making a call. Increment the count.
1249 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1250
1251 // The method data pointer needs to be updated to reflect the new target.
1252 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1253 bind(profile_continue);
1254 }
1255 }
1256
1257
1258 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1259 if (ProfileInterpreter) {
1260 Label profile_continue;
1261
1262 // If no method data exists, go to profile_continue.
1263 test_method_data_pointer(mdp, profile_continue);
1264
1265 // We are making a call. Increment the count.
1266 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1267
1268 // The method data pointer needs to be updated to reflect the new target.
1269 update_mdp_by_constant(mdp,
1270 in_bytes(VirtualCallData::
1271 virtual_call_data_size()));
1272 bind(profile_continue);
1273 }
1274 }
1275
1276
1277 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1278 Register mdp,
1279 Register reg2) {
1280 if (ProfileInterpreter) {
1281 Label profile_continue;
1282
1283 // If no method data exists, go to profile_continue.
1284 test_method_data_pointer(mdp, profile_continue);
1285
1286 // We are making a call. Increment the count.
1287 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1288
1289 // Record the receiver type.
1290 record_klass_in_profile(receiver, mdp, reg2);
1291
1292 // The method data pointer needs to be updated to reflect the new target.
1293 update_mdp_by_constant(mdp,
1294 in_bytes(VirtualCallData::
1295 virtual_call_data_size()));
1296 bind(profile_continue);
1297 }
1298 }
1299
1300 // This routine creates a state machine for updating the multi-row
1301 // type profile at a virtual call site (or other type-sensitive bytecode).
1302 // The machine visits each row (of receiver/count) until the receiver type
1303 // is found, or until it runs out of rows. At the same time, it remembers
1304 // the location of the first empty row. (An empty row records null for its
1305 // receiver, and can be allocated for a newly-observed receiver type.)
1306 // Because there are two degrees of freedom in the state, a simple linear
1307 // search will not work; it must be a decision tree. Hence this helper
1308 // function is recursive, to generate the required tree structured code.
1309 // It's the interpreter, so we are trading off code space for speed.
1310 // See below for example code.
1311 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1312 Register receiver, Register mdp,
1313 Register reg2,
1314 int start_row, Label& done) {
1315 int last_row = VirtualCallData::row_limit() - 1;
1316 assert(start_row <= last_row, "must be work left to do");
1317 // Test this row for both the receiver and for null.
1318 // Take any of three different outcomes:
1319 // 1. found receiver => increment count and goto done
1320 // 2. found null => keep looking for case 1, maybe allocate this cell
1321 // 3. found something else => keep looking for cases 1 and 2
1322 // Case 3 is handled by a recursive call.
1323 for (int row = start_row; row <= last_row; row++) {
1324 Label next_test;
1325 bool test_for_null_also = (row == start_row);
1326
1327 // See if the receiver is receiver[n].
1328 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1329 test_mdp_data_at(mdp, recvr_offset, receiver,
1330 (test_for_null_also ? reg2 : noreg),
1331 next_test);
1332 // (Reg2 now contains the receiver from the CallData.)
1333
1334 // The receiver is receiver[n]. Increment count[n].
1335 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1336 increment_mdp_data_at(mdp, count_offset);
1337 jmp(done);
1338 bind(next_test);
1339
1340 if (test_for_null_also) {
1341 // Failed the equality check on receiver[n]... Test for null.
1342 testq(reg2, reg2);
1343 if (start_row == last_row) {
1344 // The only thing left to do is handle the null case.
1345 jcc(Assembler::notZero, done);
1346 break;
1347 }
1348 // Since null is rare, make it be the branch-taken case.
1349 Label found_null;
1350 jcc(Assembler::zero, found_null);
1351
1352 // Put all the "Case 3" tests here.
1353 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done);
1354
1355 // Found a null. Keep searching for a matching receiver,
1356 // but remember that this is an empty (unused) slot.
1357 bind(found_null);
1358 }
1359 }
1360
1361 // In the fall-through case, we found no matching receiver, but we
1362 // observed the receiver[start_row] is NULL.
1363
1364 // Fill in the receiver field and increment the count.
1365 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1366 set_mdp_data_at(mdp, recvr_offset, receiver);
1367 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1368 movl(reg2, DataLayout::counter_increment);
1369 set_mdp_data_at(mdp, count_offset, reg2);
1370 jmp(done);
1371 }
1372
1373 // Example state machine code for three profile rows:
1374 // // main copy of decision tree, rooted at row[1]
1375 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1376 // if (row[0].rec != NULL) {
1377 // // inner copy of decision tree, rooted at row[1]
1378 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1379 // if (row[1].rec != NULL) {
1380 // // degenerate decision tree, rooted at row[2]
1381 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1382 // if (row[2].rec != NULL) { goto done; } // overflow
1383 // row[2].init(rec); goto done;
1384 // } else {
1385 // // remember row[1] is empty
1386 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1387 // row[1].init(rec); goto done;
1388 // }
1389 // } else {
1390 // // remember row[0] is empty
1391 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1392 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1393 // row[0].init(rec); goto done;
1394 // }
1395
1396 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1397 Register mdp,
1398 Register reg2) {
1399 assert(ProfileInterpreter, "must be profiling");
1400 Label done;
1401
1402 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
1403
1404 bind (done);
1405 }
1406
1407 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1408 Register mdp) {
1409 if (ProfileInterpreter) {
1410 Label profile_continue;
1411 uint row;
1412
1413 // If no method data exists, go to profile_continue.
1414 test_method_data_pointer(mdp, profile_continue);
1415
1416 // Update the total ret count.
1417 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1418
1419 for (row = 0; row < RetData::row_limit(); row++) {
1420 Label next_test;
1421
1422 // See if return_bci is equal to bci[n]:
1423 test_mdp_data_at(mdp,
1424 in_bytes(RetData::bci_offset(row)),
1425 return_bci, noreg,
1426 next_test);
1427
1428 // return_bci is equal to bci[n]. Increment the count.
1429 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1430
1431 // The method data pointer needs to be updated to reflect the new target.
1432 update_mdp_by_offset(mdp,
1433 in_bytes(RetData::bci_displacement_offset(row)));
1434 jmp(profile_continue);
1435 bind(next_test);
1436 }
1437
1438 update_mdp_for_ret(return_bci);
1439
1440 bind(profile_continue);
1441 }
1442 }
1443
1444
1445 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1446 if (ProfileInterpreter) {
1447 Label profile_continue;
1448
1449 // If no method data exists, go to profile_continue.
1450 test_method_data_pointer(mdp, profile_continue);
1451
1452 // The method data pointer needs to be updated.
1453 int mdp_delta = in_bytes(BitData::bit_data_size());
1454 if (TypeProfileCasts) {
1455 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1456 }
1457 update_mdp_by_constant(mdp, mdp_delta);
1458
1459 bind(profile_continue);
1460 }
1461 }
1462
1463
1464 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1465 if (ProfileInterpreter && TypeProfileCasts) {
1466 Label profile_continue;
1467
1468 // If no method data exists, go to profile_continue.
1469 test_method_data_pointer(mdp, profile_continue);
1470
1471 int count_offset = in_bytes(CounterData::count_offset());
1472 // Back up the address, since we have already bumped the mdp.
1473 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1474
1475 // *Decrement* the counter. We expect to see zero or small negatives.
1476 increment_mdp_data_at(mdp, count_offset, true);
1477
1478 bind (profile_continue);
1479 }
1480 }
1481
1482
1483 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1484 if (ProfileInterpreter) {
1485 Label profile_continue;
1486
1487 // If no method data exists, go to profile_continue.
1488 test_method_data_pointer(mdp, profile_continue);
1489
1490 // The method data pointer needs to be updated.
1491 int mdp_delta = in_bytes(BitData::bit_data_size());
1492 if (TypeProfileCasts) {
1493 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1494
1495 // Record the object type.
1496 record_klass_in_profile(klass, mdp, reg2);
1497 }
1498 update_mdp_by_constant(mdp, mdp_delta);
1499
1500 bind(profile_continue);
1501 }
1502 }
1503
1504
1505 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1506 if (ProfileInterpreter) {
1507 Label profile_continue;
1508
1509 // If no method data exists, go to profile_continue.
1510 test_method_data_pointer(mdp, profile_continue);
1511
1512 // Update the default case count
1513 increment_mdp_data_at(mdp,
1514 in_bytes(MultiBranchData::default_count_offset()));
1515
1516 // The method data pointer needs to be updated.
1517 update_mdp_by_offset(mdp,
1518 in_bytes(MultiBranchData::
1519 default_displacement_offset()));
1520
1521 bind(profile_continue);
1522 }
1523 }
1524
1525
1526 void InterpreterMacroAssembler::profile_switch_case(Register index,
1527 Register mdp,
1528 Register reg2) {
1529 if (ProfileInterpreter) {
1530 Label profile_continue;
1531
1532 // If no method data exists, go to profile_continue.
1533 test_method_data_pointer(mdp, profile_continue);
1534
1535 // Build the base (index * per_case_size_in_bytes()) +
1536 // case_array_offset_in_bytes()
1537 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1538 imulq(index, reg2); // XXX l ?
1539 addq(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1540
1541 // Update the case count
1542 increment_mdp_data_at(mdp,
1543 index,
1544 in_bytes(MultiBranchData::relative_count_offset()));
1545
1546 // The method data pointer needs to be updated.
1547 update_mdp_by_offset(mdp,
1548 index,
1549 in_bytes(MultiBranchData::
1550 relative_displacement_offset()));
1551
1552 bind(profile_continue);
1553 }
1554 }
1555
1556
1557 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1558 if (state == atos) {
1559 MacroAssembler::verify_oop(reg);
1560 }
1561 }
1562
1563 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1564 }
1565
1566
1567 void InterpreterMacroAssembler::notify_method_entry() {
1568 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1569 // track stack depth. If it is possible to enter interp_only_mode we add
1570 // the code to check if the event should be sent.
1571 if (JvmtiExport::can_post_interpreter_events()) {
1572 Label L;
1573 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1574 testl(rdx, rdx);
1575 jcc(Assembler::zero, L);
1576 call_VM(noreg, CAST_FROM_FN_PTR(address,
1577 InterpreterRuntime::post_method_entry));
1578 bind(L);
1579 }
1580
1581 {
1582 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1583 get_method(c_rarg1);
1584 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1585 r15_thread, c_rarg1);
1586 }
1587 }
1588
1589
1590 void InterpreterMacroAssembler::notify_method_exit(
1591 TosState state, NotifyMethodExitMode mode) {
1592 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1593 // track stack depth. If it is possible to enter interp_only_mode we add
1594 // the code to check if the event should be sent.
1595 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1596 Label L;
1597 // Note: frame::interpreter_frame_result has a dependency on how the
1598 // method result is saved across the call to post_method_exit. If this
1599 // is changed then the interpreter_frame_result implementation will
1600 // need to be updated too.
1601 push(state);
1602 movl(rdx, Address(r15_thread, JavaThread::interp_only_mode_offset()));
1603 testl(rdx, rdx);
1604 jcc(Assembler::zero, L);
1605 call_VM(noreg,
1606 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1607 bind(L);
1608 pop(state);
1609 }
1610
1611 {
1612 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1613 push(state);
1614 get_method(c_rarg1);
1615 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1616 r15_thread, c_rarg1);
1617 pop(state);
1618 }
1619 }