1 /*
2 * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "oops/arrayOop.hpp"
30 #include "oops/markOop.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/method.hpp"
33 #include "prims/jvmtiExport.hpp"
34 #include "prims/jvmtiRedefineClassesTrace.hpp"
35 #include "prims/jvmtiThreadState.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/biasedLocking.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/thread.inline.hpp"
40
41
42 // Implementation of InterpreterMacroAssembler
43 #ifdef CC_INTERP
44 void InterpreterMacroAssembler::get_method(Register reg) {
45 movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize)));
46 movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
47 }
48 #endif // CC_INTERP
49
50 #ifndef CC_INTERP
51 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point,
52 int number_of_arguments) {
53 // interpreter specific
54 //
55 // Note: No need to save/restore bcp & locals (r13 & r14) pointer
56 // since these are callee saved registers and no blocking/
57 // GC can happen in leaf calls.
58 // Further Note: DO NOT save/restore bcp/locals. If a caller has
59 // already saved them so that it can use esi/edi as temporaries
60 // then a save/restore here will DESTROY the copy the caller
61 // saved! There used to be a save_bcp() that only happened in
62 // the ASSERT path (no restore_bcp). Which caused bizarre failures
63 // when jvm built with ASSERTs.
64 #ifdef ASSERT
65 {
66 Label L;
67 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
68 jcc(Assembler::equal, L);
69 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
70 " last_sp != NULL");
71 bind(L);
72 }
73 #endif
74 // super call
75 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
76 // interpreter specific
77 // Used to ASSERT that r13/r14 were equal to frame's bcp/locals
78 // but since they may not have been saved (and we don't want to
79 // save them here (see note above) the assert is invalid.
80 }
81
82 void InterpreterMacroAssembler::call_VM_base(Register oop_result,
83 Register java_thread,
84 Register last_java_sp,
85 address entry_point,
86 int number_of_arguments,
87 bool check_exceptions) {
88 // interpreter specific
89 //
90 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
91 // really make a difference for these runtime calls, since they are
92 // slow anyway. Btw., bcp must be saved/restored since it may change
93 // due to GC.
94 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");)
95 save_bcp();
96 #ifdef ASSERT
97 {
98 Label L;
99 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
100 jcc(Assembler::equal, L);
101 stop("InterpreterMacroAssembler::call_VM_leaf_base:"
102 " last_sp != NULL");
103 bind(L);
104 }
105 #endif /* ASSERT */
106 // super call
107 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp,
108 entry_point, number_of_arguments,
109 check_exceptions);
110 // interpreter specific
111 restore_bcp();
112 restore_locals();
113 }
114
115 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
116 if (JvmtiExport::can_pop_frame()) {
117 Label L;
118 // Initiate popframe handling only if it is not already being
119 // processed. If the flag has the popframe_processing bit set, it
120 // means that this code is called *during* popframe handling - we
121 // don't want to reenter.
122 // This method is only called just after the call into the vm in
123 // call_VM_base, so the arg registers are available.
124 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit
125 LP64_ONLY(c_rarg0);
126 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
127 testl(pop_cond, JavaThread::popframe_pending_bit);
128 jcc(Assembler::zero, L);
129 testl(pop_cond, JavaThread::popframe_processing_bit);
130 jcc(Assembler::notZero, L);
131 // Call Interpreter::remove_activation_preserving_args_entry() to get the
132 // address of the same-named entrypoint in the generated interpreter code.
133 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
134 jmp(rax);
135 bind(L);
136 NOT_LP64(get_thread(java_thread);)
137 }
138 }
139
140 void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
141 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
142 NOT_LP64(get_thread(thread);)
143 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset()));
144 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset());
145 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset());
146 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset());
147 #ifdef _LP64
148 switch (state) {
149 case atos: movptr(rax, oop_addr);
150 movptr(oop_addr, (int32_t)NULL_WORD);
151 verify_oop(rax, state); break;
152 case ltos: movptr(rax, val_addr); break;
153 case btos: // fall through
154 case ctos: // fall through
155 case stos: // fall through
156 case itos: movl(rax, val_addr); break;
157 case ftos: movflt(xmm0, val_addr); break;
158 case dtos: movdbl(xmm0, val_addr); break;
159 case vtos: /* nothing to do */ break;
160 default : ShouldNotReachHere();
161 }
162 // Clean up tos value in the thread object
163 movl(tos_addr, (int) ilgl);
164 movl(val_addr, (int32_t) NULL_WORD);
165 #else
166 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
167 + in_ByteSize(wordSize));
168 switch (state) {
169 case atos: movptr(rax, oop_addr);
170 movptr(oop_addr, NULL_WORD);
171 verify_oop(rax, state); break;
172 case ltos:
173 movl(rdx, val_addr1); // fall through
174 case btos: // fall through
175 case ctos: // fall through
176 case stos: // fall through
177 case itos: movl(rax, val_addr); break;
178 case ftos: fld_s(val_addr); break;
179 case dtos: fld_d(val_addr); break;
180 case vtos: /* nothing to do */ break;
181 default : ShouldNotReachHere();
182 }
183 #endif // _LP64
184 // Clean up tos value in the thread object
185 movl(tos_addr, (int32_t) ilgl);
186 movptr(val_addr, NULL_WORD);
187 NOT_LP64(movptr(val_addr1, NULL_WORD);)
188 }
189
190
191 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
192 if (JvmtiExport::can_force_early_return()) {
193 Label L;
194 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread);
195 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread);
196
197 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
198 testptr(tmp, tmp);
199 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
200
201 // Initiate earlyret handling only if it is not already being processed.
202 // If the flag has the earlyret_processing bit set, it means that this code
203 // is called *during* earlyret handling - we don't want to reenter.
204 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
205 cmpl(tmp, JvmtiThreadState::earlyret_pending);
206 jcc(Assembler::notEqual, L);
207
208 // Call Interpreter::remove_activation_early_entry() to get the address of the
209 // same-named entrypoint in the generated interpreter code.
210 NOT_LP64(get_thread(java_thread);)
211 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset()));
212 #ifdef _LP64
213 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
214 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp);
215 #else
216 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
217 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
218 #endif // _LP64
219 jmp(rax);
220 bind(L);
221 NOT_LP64(get_thread(java_thread);)
222 }
223 }
224
225 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
226 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
227 load_unsigned_short(reg, Address(_bcp_register, bcp_offset));
228 bswapl(reg);
229 shrl(reg, 16);
230 }
231
232 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index,
233 int bcp_offset,
234 size_t index_size) {
235 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
236 if (index_size == sizeof(u2)) {
237 load_unsigned_short(index, Address(_bcp_register, bcp_offset));
238 } else if (index_size == sizeof(u4)) {
239 movl(index, Address(_bcp_register, bcp_offset));
240 // Check if the secondary index definition is still ~x, otherwise
241 // we have to change the following assembler code to calculate the
242 // plain index.
243 assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
244 notl(index); // convert to plain index
245 } else if (index_size == sizeof(u1)) {
246 load_unsigned_byte(index, Address(_bcp_register, bcp_offset));
247 } else {
248 ShouldNotReachHere();
249 }
250 }
251
252 void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache,
253 Register index,
254 int bcp_offset,
255 size_t index_size) {
256 assert_different_registers(cache, index);
257 get_cache_index_at_bcp(index, bcp_offset, index_size);
258 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
259 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
260 // convert from field index to ConstantPoolCacheEntry index
261 assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
262 shll(index, 2);
263 }
264
265 void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
266 Register index,
267 Register bytecode,
268 int byte_no,
269 int bcp_offset,
270 size_t index_size) {
271 get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
272 // We use a 32-bit load here since the layout of 64-bit words on
273 // little-endian machines allow us that.
274 movl(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
275 const int shift_count = (1 + byte_no) * BitsPerByte;
276 assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
277 (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
278 "correct shift count");
279 shrl(bytecode, shift_count);
280 assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
281 andl(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
282 }
283
284 void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache,
285 Register tmp,
286 int bcp_offset,
287 size_t index_size) {
288 assert(cache != tmp, "must use different register");
289 get_cache_index_at_bcp(tmp, bcp_offset, index_size);
290 assert(sizeof(ConstantPoolCacheEntry) == 4 * wordSize, "adjust code below");
291 // convert from field index to ConstantPoolCacheEntry index
292 // and from word offset to byte offset
293 assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
294 shll(tmp, 2 + LogBytesPerWord);
295 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
296 // skip past the header
297 addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
298 addptr(cache, tmp); // construct pointer to cache entry
299 }
300
301 // Load object from cpool->resolved_references(index)
302 void InterpreterMacroAssembler::load_resolved_reference_at_index(
303 Register result, Register index) {
304 assert_different_registers(result, index);
305 // convert from field index to resolved_references() index and from
306 // word index to byte offset. Since this is a java object, it can be compressed
307 Register tmp = index; // reuse
308 shll(tmp, LogBytesPerHeapOop);
309
310 get_constant_pool(result);
311 // load pointer for resolved_references[] objArray
312 movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
313 // JNIHandles::resolve(obj);
314 movptr(result, Address(result, 0));
315 // Add in the index
316 addptr(result, tmp);
317 load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
318 }
319
320
321 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a
322 // subtype of super_klass.
323 //
324 // Args:
325 // rax: superklass
326 // Rsub_klass: subklass
327 //
328 // Kills:
329 // rcx, rdi
330 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
331 Label& ok_is_subtype) {
332 assert(Rsub_klass != rax, "rax holds superklass");
333 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");)
334 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");)
335 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length");
336 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr");
337
338 // Profile the not-null value's klass.
339 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
340
341 // Do the check.
342 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
343
344 // Profile the failure of the check.
345 profile_typecheck_failed(rcx); // blows rcx
346 }
347
348
349 #ifndef _LP64
350 void InterpreterMacroAssembler::f2ieee() {
351 if (IEEEPrecision) {
352 fstp_s(Address(rsp, 0));
353 fld_s(Address(rsp, 0));
354 }
355 }
356
357
358 void InterpreterMacroAssembler::d2ieee() {
359 if (IEEEPrecision) {
360 fstp_d(Address(rsp, 0));
361 fld_d(Address(rsp, 0));
362 }
363 }
364 #endif // _LP64
365
366 // Java Expression Stack
367
368 void InterpreterMacroAssembler::pop_ptr(Register r) {
369 pop(r);
370 }
371
372 void InterpreterMacroAssembler::push_ptr(Register r) {
373 push(r);
374 }
375
376 void InterpreterMacroAssembler::push_i(Register r) {
377 push(r);
378 }
379
380 #ifdef _LP64
381 void InterpreterMacroAssembler::pop_i(Register r) {
382 // XXX can't use pop currently, upper half non clean
383 movl(r, Address(rsp, 0));
384 addptr(rsp, wordSize);
385 }
386
387 void InterpreterMacroAssembler::pop_l(Register r) {
388 movq(r, Address(rsp, 0));
389 addptr(rsp, 2 * Interpreter::stackElementSize);
390 }
391
392 void InterpreterMacroAssembler::pop_f(XMMRegister r) {
393 movflt(r, Address(rsp, 0));
394 addptr(rsp, wordSize);
395 }
396
397 void InterpreterMacroAssembler::pop_d(XMMRegister r) {
398 movdbl(r, Address(rsp, 0));
399 addptr(rsp, 2 * Interpreter::stackElementSize);
400 }
401
402 void InterpreterMacroAssembler::push_l(Register r) {
403 subptr(rsp, 2 * wordSize);
404 movq(Address(rsp, 0), r);
405 }
406
407 void InterpreterMacroAssembler::push_f(XMMRegister r) {
408 subptr(rsp, wordSize);
409 movflt(Address(rsp, 0), r);
410 }
411
412 void InterpreterMacroAssembler::push_d(XMMRegister r) {
413 subptr(rsp, 2 * wordSize);
414 movdbl(Address(rsp, 0), r);
415 }
416
417 void InterpreterMacroAssembler::pop(TosState state) {
418 switch (state) {
419 case atos: pop_ptr(); break;
420 case btos:
421 case ctos:
422 case stos:
423 case itos: pop_i(); break;
424 case ltos: pop_l(); break;
425 case ftos: pop_f(); break;
426 case dtos: pop_d(); break;
427 case vtos: /* nothing to do */ break;
428 default: ShouldNotReachHere();
429 }
430 verify_oop(rax, state);
431 }
432
433 void InterpreterMacroAssembler::push(TosState state) {
434 verify_oop(rax, state);
435 switch (state) {
436 case atos: push_ptr(); break;
437 case btos:
438 case ctos:
439 case stos:
440 case itos: push_i(); break;
441 case ltos: push_l(); break;
442 case ftos: push_f(); break;
443 case dtos: push_d(); break;
444 case vtos: /* nothing to do */ break;
445 default : ShouldNotReachHere();
446 }
447 }
448 #else
449 void InterpreterMacroAssembler::pop_i(Register r) {
450 pop(r);
451 }
452
453 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
454 pop(lo);
455 pop(hi);
456 }
457
458 void InterpreterMacroAssembler::pop_f() {
459 fld_s(Address(rsp, 0));
460 addptr(rsp, 1 * wordSize);
461 }
462
463 void InterpreterMacroAssembler::pop_d() {
464 fld_d(Address(rsp, 0));
465 addptr(rsp, 2 * wordSize);
466 }
467
468
469 void InterpreterMacroAssembler::pop(TosState state) {
470 switch (state) {
471 case atos: pop_ptr(rax); break;
472 case btos: // fall through
473 case ctos: // fall through
474 case stos: // fall through
475 case itos: pop_i(rax); break;
476 case ltos: pop_l(rax, rdx); break;
477 case ftos: pop_f(); break;
478 case dtos: pop_d(); break;
479 case vtos: /* nothing to do */ break;
480 default : ShouldNotReachHere();
481 }
482 verify_oop(rax, state);
483 }
484
485
486 void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
487 push(hi);
488 push(lo);
489 }
490
491 void InterpreterMacroAssembler::push_f() {
492 // Do not schedule for no AGI! Never write beyond rsp!
493 subptr(rsp, 1 * wordSize);
494 fstp_s(Address(rsp, 0));
495 }
496
497 void InterpreterMacroAssembler::push_d(Register r) {
498 // Do not schedule for no AGI! Never write beyond rsp!
499 subptr(rsp, 2 * wordSize);
500 fstp_d(Address(rsp, 0));
501 }
502
503
504 void InterpreterMacroAssembler::push(TosState state) {
505 verify_oop(rax, state);
506 switch (state) {
507 case atos: push_ptr(rax); break;
508 case btos: // fall through
509 case ctos: // fall through
510 case stos: // fall through
511 case itos: push_i(rax); break;
512 case ltos: push_l(rax, rdx); break;
513 case ftos: push_f(); break;
514 case dtos: push_d(rax); break;
515 case vtos: /* nothing to do */ break;
516 default : ShouldNotReachHere();
517 }
518 }
519 #endif // _LP64
520
521
522 // Helpers for swap and dup
523 void InterpreterMacroAssembler::load_ptr(int n, Register val) {
524 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
525 }
526
527 void InterpreterMacroAssembler::store_ptr(int n, Register val) {
528 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
529 }
530
531
532 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
533 // set sender sp
534 lea(_bcp_register, Address(rsp, wordSize));
535 // record last_sp
536 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), _bcp_register);
537 }
538
539
540 // Jump to from_interpreted entry of a call unless single stepping is possible
541 // in this thread in which case we must call the i2i entry
542 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
543 prepare_to_jump_from_interpreted();
544
545 if (JvmtiExport::can_post_interpreter_events()) {
546 Label run_compiled_code;
547 // JVMTI events, such as single-stepping, are implemented partly by avoiding running
548 // compiled code in threads for which the event is enabled. Check here for
549 // interp_only_mode if these events CAN be enabled.
550 // interp_only is an int, on little endian it is sufficient to test the byte only
551 // Is a cmpl faster?
552 LP64_ONLY(temp = r15_thread;)
553 NOT_LP64(get_thread(temp);)
554 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
555 jccb(Assembler::zero, run_compiled_code);
556 jmp(Address(method, Method::interpreter_entry_offset()));
557 bind(run_compiled_code);
558 }
559
560 jmp(Address(method, Method::from_interpreted_offset()));
561 }
562
563 // The following two routines provide a hook so that an implementation
564 // can schedule the dispatch in two parts. x86 does not do this.
565 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
566 // Nothing x86 specific to be done here
567 }
568
569 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
570 dispatch_next(state, step);
571 }
572
573 void InterpreterMacroAssembler::dispatch_base(TosState state,
574 address* table,
575 bool verifyoop) {
576 verify_FPU(1, state);
577 if (VerifyActivationFrameSize) {
578 Label L;
579 mov(rcx, rbp);
580 subptr(rcx, rsp);
581 int32_t min_frame_size =
582 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) *
583 wordSize;
584 cmpptr(rcx, (int32_t)min_frame_size);
585 jcc(Assembler::greaterEqual, L);
586 stop("broken stack frame");
587 bind(L);
588 }
589 if (verifyoop) {
590 verify_oop(rax, state);
591 }
592 #ifdef _LP64
593 lea(rscratch1, ExternalAddress((address)table));
594 jmp(Address(rscratch1, rbx, Address::times_8));
595 #else
596 Address index(noreg, rbx, Address::times_ptr);
597 ExternalAddress tbl((address)table);
598 ArrayAddress dispatch(tbl, index);
599 jump(dispatch);
600 #endif // _LP64
601 }
602
603 void InterpreterMacroAssembler::dispatch_only(TosState state) {
604 dispatch_base(state, Interpreter::dispatch_table(state));
605 }
606
607 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
608 dispatch_base(state, Interpreter::normal_table(state));
609 }
610
611 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
612 dispatch_base(state, Interpreter::normal_table(state), false);
613 }
614
615
616 void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
617 // load next bytecode (load before advancing _bcp_register to prevent AGI)
618 load_unsigned_byte(rbx, Address(_bcp_register, step));
619 // advance _bcp_register
620 increment(_bcp_register, step);
621 dispatch_base(state, Interpreter::dispatch_table(state));
622 }
623
624 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
625 // load current bytecode
626 load_unsigned_byte(rbx, Address(_bcp_register, 0));
627 dispatch_base(state, table);
628 }
629
630 // remove activation
631 //
632 // Unlock the receiver if this is a synchronized method.
633 // Unlock any Java monitors from syncronized blocks.
634 // Remove the activation from the stack.
635 //
636 // If there are locked Java monitors
637 // If throw_monitor_exception
638 // throws IllegalMonitorStateException
639 // Else if install_monitor_exception
640 // installs IllegalMonitorStateException
641 // Else
642 // no error processing
643 void InterpreterMacroAssembler::remove_activation(
644 TosState state,
645 Register ret_addr,
646 bool throw_monitor_exception,
647 bool install_monitor_exception,
648 bool notify_jvmdi) {
649 // Note: Registers rdx xmm0 may be in use for the
650 // result check if synchronized method
651 Label unlocked, unlock, no_unlock;
652
653 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
654 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
655 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
656 // monitor pointers need different register
657 // because rdx may have the result in it
658 NOT_LP64(get_thread(rcx);)
659
660 // get the value of _do_not_unlock_if_synchronized into rdx
661 const Address do_not_unlock_if_synchronized(rthread,
662 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
663 movbool(rbx, do_not_unlock_if_synchronized);
664 movbool(do_not_unlock_if_synchronized, false); // reset the flag
665
666 // get method access flags
667 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
668 movl(rcx, Address(rcx, Method::access_flags_offset()));
669 testl(rcx, JVM_ACC_SYNCHRONIZED);
670 jcc(Assembler::zero, unlocked);
671
672 // Don't unlock anything if the _do_not_unlock_if_synchronized flag
673 // is set.
674 testbool(rbx);
675 jcc(Assembler::notZero, no_unlock);
676
677 // unlock monitor
678 push(state); // save result
679
680 // BasicObjectLock will be first in list, since this is a
681 // synchronized method. However, need to check that the object has
682 // not been unlocked by an explicit monitorexit bytecode.
683 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset *
684 wordSize - (int) sizeof(BasicObjectLock));
685 // We use c_rarg1/rdx so that if we go slow path it will be the correct
686 // register for unlock_object to pass to VM directly
687 lea(robj, monitor); // address of first monitor
688
689 movptr(rax, Address(robj, BasicObjectLock::obj_offset_in_bytes()));
690 testptr(rax, rax);
691 jcc(Assembler::notZero, unlock);
692
693 pop(state);
694 if (throw_monitor_exception) {
695 // Entry already unlocked, need to throw exception
696 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow
697 call_VM(noreg, CAST_FROM_FN_PTR(address,
698 InterpreterRuntime::throw_illegal_monitor_state_exception));
699 should_not_reach_here();
700 } else {
701 // Monitor already unlocked during a stack unroll. If requested,
702 // install an illegal_monitor_state_exception. Continue with
703 // stack unrolling.
704 if (install_monitor_exception) {
705 NOT_LP64(empty_FPU_stack();)
706 call_VM(noreg, CAST_FROM_FN_PTR(address,
707 InterpreterRuntime::new_illegal_monitor_state_exception));
708 }
709 jmp(unlocked);
710 }
711
712 bind(unlock);
713 unlock_object(robj);
714 pop(state);
715
716 // Check that for block-structured locking (i.e., that all locked
717 // objects has been unlocked)
718 bind(unlocked);
719
720 // rax, rdx: Might contain return value
721
722 // Check that all monitors are unlocked
723 {
724 Label loop, exception, entry, restart;
725 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
726 const Address monitor_block_top(
727 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
728 const Address monitor_block_bot(
729 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
730
731 bind(restart);
732 // We use c_rarg1 so that if we go slow path it will be the correct
733 // register for unlock_object to pass to VM directly
734 movptr(rmon, monitor_block_top); // points to current entry, starting
735 // with top-most entry
736 lea(rbx, monitor_block_bot); // points to word before bottom of
737 // monitor block
738 jmp(entry);
739
740 // Entry already locked, need to throw exception
741 bind(exception);
742
743 if (throw_monitor_exception) {
744 // Throw exception
745 NOT_LP64(empty_FPU_stack();)
746 MacroAssembler::call_VM(noreg,
747 CAST_FROM_FN_PTR(address, InterpreterRuntime::
748 throw_illegal_monitor_state_exception));
749 should_not_reach_here();
750 } else {
751 // Stack unrolling. Unlock object and install illegal_monitor_exception.
752 // Unlock does not block, so don't have to worry about the frame.
753 // We don't have to preserve c_rarg1 since we are going to throw an exception.
754
755 push(state);
756 mov(robj, rmon); // nop if robj and rmon are the same
757 unlock_object(robj);
758 pop(state);
759
760 if (install_monitor_exception) {
761 NOT_LP64(empty_FPU_stack();)
762 call_VM(noreg, CAST_FROM_FN_PTR(address,
763 InterpreterRuntime::
764 new_illegal_monitor_state_exception));
765 }
766
767 jmp(restart);
768 }
769
770 bind(loop);
771 // check if current entry is used
772 cmpptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL);
773 jcc(Assembler::notEqual, exception);
774
775 addptr(rmon, entry_size); // otherwise advance to next entry
776 bind(entry);
777 cmpptr(rmon, rbx); // check if bottom reached
778 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
779 }
780
781 bind(no_unlock);
782
783 // jvmti support
784 if (notify_jvmdi) {
785 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA
786 } else {
787 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
788 }
789
790 // remove activation
791 // get sender sp
792 movptr(rbx,
793 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize));
794 leave(); // remove frame anchor
795 pop(ret_addr); // get return address
796 mov(rsp, rbx); // set sp to sender sp
797 #ifndef _LP64
798 if (UseSSE) {
799 // float and double are returned in xmm register in SSE-mode
800 if (state == ftos && UseSSE >= 1) {
801 subptr(rsp, wordSize);
802 fstp_s(Address(rsp, 0));
803 movflt(xmm0, Address(rsp, 0));
804 addptr(rsp, wordSize);
805 } else if (state == dtos && UseSSE >= 2) {
806 subptr(rsp, 2*wordSize);
807 fstp_d(Address(rsp, 0));
808 movdbl(xmm0, Address(rsp, 0));
809 addptr(rsp, 2*wordSize);
810 }
811 }
812 #endif // _LP64
813 }
814 #endif /* !CC_INTERP */
815
816 void InterpreterMacroAssembler::get_method_counters(Register method,
817 Register mcs, Label& skip) {
818 Label has_counters;
819 movptr(mcs, Address(method, Method::method_counters_offset()));
820 testptr(mcs, mcs);
821 jcc(Assembler::notZero, has_counters);
822 call_VM(noreg, CAST_FROM_FN_PTR(address,
823 InterpreterRuntime::build_method_counters), method);
824 movptr(mcs, Address(method,Method::method_counters_offset()));
825 testptr(mcs, mcs);
826 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory
827 bind(has_counters);
828 }
829
830
831 // Lock object
832 //
833 // Args:
834 // rdx, c_rarg1: BasicObjectLock to be used for locking
835 //
836 // Kills:
837 // rax
838 // rscratch1 (scratch regs)
839 void InterpreterMacroAssembler::lock_object(Register lock_reg) {
840 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
841 "The argument is only for looks. It must be c_rarg1");
842
843 if (UseHeavyMonitors) {
844 call_VM(noreg,
845 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
846 lock_reg);
847 } else {
848 Label done;
849
850 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
851 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
852
853 const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
854 const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
855 const int mark_offset = lock_offset +
856 BasicLock::displaced_header_offset_in_bytes();
857
858 Label slow_case;
859
860 // Load object pointer into obj_reg
861 movptr(obj_reg, Address(lock_reg, obj_offset));
862
863 if (UseBiasedLocking) {
864 biased_locking_enter(lock_reg, obj_reg, swap_reg, rscratch1, false, done, &slow_case);
865 }
866
867 // Load immediate 1 into swap_reg %rax
868 movl(swap_reg, (int32_t)1);
869
870 // Load (object->mark() | 1) into swap_reg %rax
871 orptr(swap_reg, Address(obj_reg, 0));
872
873 // Save (object->mark() | 1) into BasicLock's displaced header
874 movptr(Address(lock_reg, mark_offset), swap_reg);
875
876 assert(lock_offset == 0,
877 "displached header must be first word in BasicObjectLock");
878
879 if (os::is_MP()) lock();
880 cmpxchgptr(lock_reg, Address(obj_reg, 0));
881 if (PrintBiasedLockingStatistics) {
882 cond_inc32(Assembler::zero,
883 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
884 }
885 jcc(Assembler::zero, done);
886
887 const int zero_bits = LP64_ONLY(7) NOT_LP64(3);
888
889 // Test if the oopMark is an obvious stack pointer, i.e.,
890 // 1) (mark & zero_bits) == 0, and
891 // 2) rsp <= mark < mark + os::pagesize()
892 //
893 // These 3 tests can be done by evaluating the following
894 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())),
895 // assuming both stack pointer and pagesize have their
896 // least significant bits clear.
897 // NOTE: the oopMark is in swap_reg %rax as the result of cmpxchg
898 subptr(swap_reg, rsp);
899 andptr(swap_reg, zero_bits - os::vm_page_size());
900
901 // Save the test result, for recursive case, the result is zero
902 movptr(Address(lock_reg, mark_offset), swap_reg);
903
904 if (PrintBiasedLockingStatistics) {
905 cond_inc32(Assembler::zero,
906 ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
907 }
908 jcc(Assembler::zero, done);
909
910 bind(slow_case);
911
912 // Call the runtime routine for slow case
913 call_VM(noreg,
914 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter),
915 lock_reg);
916
917 bind(done);
918 }
919 }
920
921
922 // Unlocks an object. Used in monitorexit bytecode and
923 // remove_activation. Throws an IllegalMonitorException if object is
924 // not locked by current thread.
925 //
926 // Args:
927 // rdx, c_rarg1: BasicObjectLock for lock
928 //
929 // Kills:
930 // rax
931 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs)
932 // rscratch1, rscratch2 (scratch regs)
933 // rax, rbx, rcx, rdx
934 void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
935 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx),
936 "The argument is only for looks. It must be c_rarg1");
937
938 if (UseHeavyMonitors) {
939 call_VM(noreg,
940 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
941 lock_reg);
942 } else {
943 Label done;
944
945 const Register swap_reg = rax; // Must use rax for cmpxchg instruction
946 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark
947 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop
948
949 save_bcp(); // Save in case of exception
950
951 // Convert from BasicObjectLock structure to object and BasicLock
952 // structure Store the BasicLock address into %rax
953 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
954
955 // Load oop into obj_reg(%c_rarg3)
956 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()));
957
958 // Free entry
959 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
960
961 if (UseBiasedLocking) {
962 biased_locking_exit(obj_reg, header_reg, done);
963 }
964
965 // Load the old header from BasicLock structure
966 movptr(header_reg, Address(swap_reg,
967 BasicLock::displaced_header_offset_in_bytes()));
968
969 // Test for recursion
970 testptr(header_reg, header_reg);
971
972 // zero for recursive case
973 jcc(Assembler::zero, done);
974
975 // Atomic swap back the old header
976 if (os::is_MP()) lock();
977 cmpxchgptr(header_reg, Address(obj_reg, 0));
978
979 // zero for recursive case
980 jcc(Assembler::zero, done);
981
982 // Call the runtime routine for slow case.
983 movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()),
984 obj_reg); // restore obj
985 call_VM(noreg,
986 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit),
987 lock_reg);
988
989 bind(done);
990
991 restore_bcp();
992 }
993 }
994 #ifndef CC_INTERP
995 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp,
996 Label& zero_continue) {
997 assert(ProfileInterpreter, "must be profiling interpreter");
998 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize));
999 testptr(mdp, mdp);
1000 jcc(Assembler::zero, zero_continue);
1001 }
1002
1003
1004 // Set the method data pointer for the current bcp.
1005 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
1006 assert(ProfileInterpreter, "must be profiling interpreter");
1007 Label set_mdp;
1008 push(rax);
1009 push(rbx);
1010
1011 get_method(rbx);
1012 // Test MDO to avoid the call if it is NULL.
1013 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
1014 testptr(rax, rax);
1015 jcc(Assembler::zero, set_mdp);
1016 // rbx: method
1017 // _bcp_register: bcp
1018 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register);
1019 // rax: mdi
1020 // mdo is guaranteed to be non-zero here, we checked for it before the call.
1021 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
1022 addptr(rbx, in_bytes(MethodData::data_offset()));
1023 addptr(rax, rbx);
1024 bind(set_mdp);
1025 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax);
1026 pop(rbx);
1027 pop(rax);
1028 }
1029
1030 void InterpreterMacroAssembler::verify_method_data_pointer() {
1031 assert(ProfileInterpreter, "must be profiling interpreter");
1032 #ifdef ASSERT
1033 Label verify_continue;
1034 push(rax);
1035 push(rbx);
1036 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
1037 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
1038 push(arg3_reg);
1039 push(arg2_reg);
1040 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue
1041 get_method(rbx);
1042
1043 // If the mdp is valid, it will point to a DataLayout header which is
1044 // consistent with the bcp. The converse is highly probable also.
1045 load_unsigned_short(arg2_reg,
1046 Address(arg3_reg, in_bytes(DataLayout::bci_offset())));
1047 addptr(arg2_reg, Address(rbx, Method::const_offset()));
1048 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset()));
1049 cmpptr(arg2_reg, _bcp_register);
1050 jcc(Assembler::equal, verify_continue);
1051 // rbx: method
1052 // _bcp_register: bcp
1053 // c_rarg3: mdp
1054 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp),
1055 rbx, _bcp_register, arg3_reg);
1056 bind(verify_continue);
1057 pop(arg2_reg);
1058 pop(arg3_reg);
1059 pop(rbx);
1060 pop(rax);
1061 #endif // ASSERT
1062 }
1063
1064
1065 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in,
1066 int constant,
1067 Register value) {
1068 assert(ProfileInterpreter, "must be profiling interpreter");
1069 Address data(mdp_in, constant);
1070 movptr(data, value);
1071 }
1072
1073
1074 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1075 int constant,
1076 bool decrement) {
1077 // Counter address
1078 Address data(mdp_in, constant);
1079
1080 increment_mdp_data_at(data, decrement);
1081 }
1082
1083 void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
1084 bool decrement) {
1085 assert(ProfileInterpreter, "must be profiling interpreter");
1086 // %%% this does 64bit counters at best it is wasting space
1087 // at worst it is a rare bug when counters overflow
1088
1089 if (decrement) {
1090 // Decrement the register. Set condition codes.
1091 addptr(data, (int32_t) -DataLayout::counter_increment);
1092 // If the decrement causes the counter to overflow, stay negative
1093 Label L;
1094 jcc(Assembler::negative, L);
1095 addptr(data, (int32_t) DataLayout::counter_increment);
1096 bind(L);
1097 } else {
1098 assert(DataLayout::counter_increment == 1,
1099 "flow-free idiom only works with 1");
1100 // Increment the register. Set carry flag.
1101 addptr(data, DataLayout::counter_increment);
1102 // If the increment causes the counter to overflow, pull back by 1.
1103 sbbptr(data, (int32_t)0);
1104 }
1105 }
1106
1107
1108 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
1109 Register reg,
1110 int constant,
1111 bool decrement) {
1112 Address data(mdp_in, reg, Address::times_1, constant);
1113
1114 increment_mdp_data_at(data, decrement);
1115 }
1116
1117 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in,
1118 int flag_byte_constant) {
1119 assert(ProfileInterpreter, "must be profiling interpreter");
1120 int header_offset = in_bytes(DataLayout::header_offset());
1121 int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
1122 // Set the flag
1123 orl(Address(mdp_in, header_offset), header_bits);
1124 }
1125
1126
1127
1128 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
1129 int offset,
1130 Register value,
1131 Register test_value_out,
1132 Label& not_equal_continue) {
1133 assert(ProfileInterpreter, "must be profiling interpreter");
1134 if (test_value_out == noreg) {
1135 cmpptr(value, Address(mdp_in, offset));
1136 } else {
1137 // Put the test value into a register, so caller can use it:
1138 movptr(test_value_out, Address(mdp_in, offset));
1139 cmpptr(test_value_out, value);
1140 }
1141 jcc(Assembler::notEqual, not_equal_continue);
1142 }
1143
1144
1145 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1146 int offset_of_disp) {
1147 assert(ProfileInterpreter, "must be profiling interpreter");
1148 Address disp_address(mdp_in, offset_of_disp);
1149 addptr(mdp_in, disp_address);
1150 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1151 }
1152
1153
1154 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in,
1155 Register reg,
1156 int offset_of_disp) {
1157 assert(ProfileInterpreter, "must be profiling interpreter");
1158 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
1159 addptr(mdp_in, disp_address);
1160 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1161 }
1162
1163
1164 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in,
1165 int constant) {
1166 assert(ProfileInterpreter, "must be profiling interpreter");
1167 addptr(mdp_in, constant);
1168 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in);
1169 }
1170
1171
1172 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
1173 assert(ProfileInterpreter, "must be profiling interpreter");
1174 push(return_bci); // save/restore across call_VM
1175 call_VM(noreg,
1176 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret),
1177 return_bci);
1178 pop(return_bci);
1179 }
1180
1181
1182 void InterpreterMacroAssembler::profile_taken_branch(Register mdp,
1183 Register bumped_count) {
1184 if (ProfileInterpreter) {
1185 Label profile_continue;
1186
1187 // If no method data exists, go to profile_continue.
1188 // Otherwise, assign to mdp
1189 test_method_data_pointer(mdp, profile_continue);
1190
1191 // We are taking a branch. Increment the taken count.
1192 // We inline increment_mdp_data_at to return bumped_count in a register
1193 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
1194 Address data(mdp, in_bytes(JumpData::taken_offset()));
1195 movptr(bumped_count, data);
1196 assert(DataLayout::counter_increment == 1,
1197 "flow-free idiom only works with 1");
1198 addptr(bumped_count, DataLayout::counter_increment);
1199 sbbptr(bumped_count, 0);
1200 movptr(data, bumped_count); // Store back out
1201
1202 // The method data pointer needs to be updated to reflect the new target.
1203 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
1204 bind(profile_continue);
1205 }
1206 }
1207
1208
1209 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
1210 if (ProfileInterpreter) {
1211 Label profile_continue;
1212
1213 // If no method data exists, go to profile_continue.
1214 test_method_data_pointer(mdp, profile_continue);
1215
1216 // We are taking a branch. Increment the not taken count.
1217 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
1218
1219 // The method data pointer needs to be updated to correspond to
1220 // the next bytecode
1221 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
1222 bind(profile_continue);
1223 }
1224 }
1225
1226 void InterpreterMacroAssembler::profile_call(Register mdp) {
1227 if (ProfileInterpreter) {
1228 Label profile_continue;
1229
1230 // If no method data exists, go to profile_continue.
1231 test_method_data_pointer(mdp, profile_continue);
1232
1233 // We are making a call. Increment the count.
1234 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1235
1236 // The method data pointer needs to be updated to reflect the new target.
1237 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
1238 bind(profile_continue);
1239 }
1240 }
1241
1242
1243 void InterpreterMacroAssembler::profile_final_call(Register mdp) {
1244 if (ProfileInterpreter) {
1245 Label profile_continue;
1246
1247 // If no method data exists, go to profile_continue.
1248 test_method_data_pointer(mdp, profile_continue);
1249
1250 // We are making a call. Increment the count.
1251 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1252
1253 // The method data pointer needs to be updated to reflect the new target.
1254 update_mdp_by_constant(mdp,
1255 in_bytes(VirtualCallData::
1256 virtual_call_data_size()));
1257 bind(profile_continue);
1258 }
1259 }
1260
1261
1262 void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
1263 Register mdp,
1264 Register reg2,
1265 bool receiver_can_be_null) {
1266 if (ProfileInterpreter) {
1267 Label profile_continue;
1268
1269 // If no method data exists, go to profile_continue.
1270 test_method_data_pointer(mdp, profile_continue);
1271
1272 Label skip_receiver_profile;
1273 if (receiver_can_be_null) {
1274 Label not_null;
1275 testptr(receiver, receiver);
1276 jccb(Assembler::notZero, not_null);
1277 // We are making a call. Increment the count for null receiver.
1278 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1279 jmp(skip_receiver_profile);
1280 bind(not_null);
1281 }
1282
1283 // Record the receiver type.
1284 record_klass_in_profile(receiver, mdp, reg2, true);
1285 bind(skip_receiver_profile);
1286
1287 // The method data pointer needs to be updated to reflect the new target.
1288 update_mdp_by_constant(mdp,
1289 in_bytes(VirtualCallData::
1290 virtual_call_data_size()));
1291 bind(profile_continue);
1292 }
1293 }
1294
1295 // This routine creates a state machine for updating the multi-row
1296 // type profile at a virtual call site (or other type-sensitive bytecode).
1297 // The machine visits each row (of receiver/count) until the receiver type
1298 // is found, or until it runs out of rows. At the same time, it remembers
1299 // the location of the first empty row. (An empty row records null for its
1300 // receiver, and can be allocated for a newly-observed receiver type.)
1301 // Because there are two degrees of freedom in the state, a simple linear
1302 // search will not work; it must be a decision tree. Hence this helper
1303 // function is recursive, to generate the required tree structured code.
1304 // It's the interpreter, so we are trading off code space for speed.
1305 // See below for example code.
1306 void InterpreterMacroAssembler::record_klass_in_profile_helper(
1307 Register receiver, Register mdp,
1308 Register reg2, int start_row,
1309 Label& done, bool is_virtual_call) {
1310 if (TypeProfileWidth == 0) {
1311 if (is_virtual_call) {
1312 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1313 }
1314 return;
1315 }
1316
1317 int last_row = VirtualCallData::row_limit() - 1;
1318 assert(start_row <= last_row, "must be work left to do");
1319 // Test this row for both the receiver and for null.
1320 // Take any of three different outcomes:
1321 // 1. found receiver => increment count and goto done
1322 // 2. found null => keep looking for case 1, maybe allocate this cell
1323 // 3. found something else => keep looking for cases 1 and 2
1324 // Case 3 is handled by a recursive call.
1325 for (int row = start_row; row <= last_row; row++) {
1326 Label next_test;
1327 bool test_for_null_also = (row == start_row);
1328
1329 // See if the receiver is receiver[n].
1330 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
1331 test_mdp_data_at(mdp, recvr_offset, receiver,
1332 (test_for_null_also ? reg2 : noreg),
1333 next_test);
1334 // (Reg2 now contains the receiver from the CallData.)
1335
1336 // The receiver is receiver[n]. Increment count[n].
1337 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
1338 increment_mdp_data_at(mdp, count_offset);
1339 jmp(done);
1340 bind(next_test);
1341
1342 if (test_for_null_also) {
1343 Label found_null;
1344 // Failed the equality check on receiver[n]... Test for null.
1345 testptr(reg2, reg2);
1346 if (start_row == last_row) {
1347 // The only thing left to do is handle the null case.
1348 if (is_virtual_call) {
1349 jccb(Assembler::zero, found_null);
1350 // Receiver did not match any saved receiver and there is no empty row for it.
1351 // Increment total counter to indicate polymorphic case.
1352 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1353 jmp(done);
1354 bind(found_null);
1355 } else {
1356 jcc(Assembler::notZero, done);
1357 }
1358 break;
1359 }
1360 // Since null is rare, make it be the branch-taken case.
1361 jcc(Assembler::zero, found_null);
1362
1363 // Put all the "Case 3" tests here.
1364 record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
1365
1366 // Found a null. Keep searching for a matching receiver,
1367 // but remember that this is an empty (unused) slot.
1368 bind(found_null);
1369 }
1370 }
1371
1372 // In the fall-through case, we found no matching receiver, but we
1373 // observed the receiver[start_row] is NULL.
1374
1375 // Fill in the receiver field and increment the count.
1376 int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
1377 set_mdp_data_at(mdp, recvr_offset, receiver);
1378 int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
1379 movl(reg2, DataLayout::counter_increment);
1380 set_mdp_data_at(mdp, count_offset, reg2);
1381 if (start_row > 0) {
1382 jmp(done);
1383 }
1384 }
1385
1386 // Example state machine code for three profile rows:
1387 // // main copy of decision tree, rooted at row[1]
1388 // if (row[0].rec == rec) { row[0].incr(); goto done; }
1389 // if (row[0].rec != NULL) {
1390 // // inner copy of decision tree, rooted at row[1]
1391 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1392 // if (row[1].rec != NULL) {
1393 // // degenerate decision tree, rooted at row[2]
1394 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1395 // if (row[2].rec != NULL) { count.incr(); goto done; } // overflow
1396 // row[2].init(rec); goto done;
1397 // } else {
1398 // // remember row[1] is empty
1399 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1400 // row[1].init(rec); goto done;
1401 // }
1402 // } else {
1403 // // remember row[0] is empty
1404 // if (row[1].rec == rec) { row[1].incr(); goto done; }
1405 // if (row[2].rec == rec) { row[2].incr(); goto done; }
1406 // row[0].init(rec); goto done;
1407 // }
1408 // done:
1409
1410 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
1411 Register mdp, Register reg2,
1412 bool is_virtual_call) {
1413 assert(ProfileInterpreter, "must be profiling");
1414 Label done;
1415
1416 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
1417
1418 bind (done);
1419 }
1420
1421 void InterpreterMacroAssembler::profile_ret(Register return_bci,
1422 Register mdp) {
1423 if (ProfileInterpreter) {
1424 Label profile_continue;
1425 uint row;
1426
1427 // If no method data exists, go to profile_continue.
1428 test_method_data_pointer(mdp, profile_continue);
1429
1430 // Update the total ret count.
1431 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
1432
1433 for (row = 0; row < RetData::row_limit(); row++) {
1434 Label next_test;
1435
1436 // See if return_bci is equal to bci[n]:
1437 test_mdp_data_at(mdp,
1438 in_bytes(RetData::bci_offset(row)),
1439 return_bci, noreg,
1440 next_test);
1441
1442 // return_bci is equal to bci[n]. Increment the count.
1443 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
1444
1445 // The method data pointer needs to be updated to reflect the new target.
1446 update_mdp_by_offset(mdp,
1447 in_bytes(RetData::bci_displacement_offset(row)));
1448 jmp(profile_continue);
1449 bind(next_test);
1450 }
1451
1452 update_mdp_for_ret(return_bci);
1453
1454 bind(profile_continue);
1455 }
1456 }
1457
1458
1459 void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
1460 if (ProfileInterpreter) {
1461 Label profile_continue;
1462
1463 // If no method data exists, go to profile_continue.
1464 test_method_data_pointer(mdp, profile_continue);
1465
1466 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
1467
1468 // The method data pointer needs to be updated.
1469 int mdp_delta = in_bytes(BitData::bit_data_size());
1470 if (TypeProfileCasts) {
1471 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1472 }
1473 update_mdp_by_constant(mdp, mdp_delta);
1474
1475 bind(profile_continue);
1476 }
1477 }
1478
1479
1480 void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
1481 if (ProfileInterpreter && TypeProfileCasts) {
1482 Label profile_continue;
1483
1484 // If no method data exists, go to profile_continue.
1485 test_method_data_pointer(mdp, profile_continue);
1486
1487 int count_offset = in_bytes(CounterData::count_offset());
1488 // Back up the address, since we have already bumped the mdp.
1489 count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
1490
1491 // *Decrement* the counter. We expect to see zero or small negatives.
1492 increment_mdp_data_at(mdp, count_offset, true);
1493
1494 bind (profile_continue);
1495 }
1496 }
1497
1498
1499 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
1500 if (ProfileInterpreter) {
1501 Label profile_continue;
1502
1503 // If no method data exists, go to profile_continue.
1504 test_method_data_pointer(mdp, profile_continue);
1505
1506 // The method data pointer needs to be updated.
1507 int mdp_delta = in_bytes(BitData::bit_data_size());
1508 if (TypeProfileCasts) {
1509 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
1510
1511 // Record the object type.
1512 record_klass_in_profile(klass, mdp, reg2, false);
1513 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");)
1514 NOT_LP64(restore_locals();) // Restore EDI
1515 }
1516 update_mdp_by_constant(mdp, mdp_delta);
1517
1518 bind(profile_continue);
1519 }
1520 }
1521
1522
1523 void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
1524 if (ProfileInterpreter) {
1525 Label profile_continue;
1526
1527 // If no method data exists, go to profile_continue.
1528 test_method_data_pointer(mdp, profile_continue);
1529
1530 // Update the default case count
1531 increment_mdp_data_at(mdp,
1532 in_bytes(MultiBranchData::default_count_offset()));
1533
1534 // The method data pointer needs to be updated.
1535 update_mdp_by_offset(mdp,
1536 in_bytes(MultiBranchData::
1537 default_displacement_offset()));
1538
1539 bind(profile_continue);
1540 }
1541 }
1542
1543
1544 void InterpreterMacroAssembler::profile_switch_case(Register index,
1545 Register mdp,
1546 Register reg2) {
1547 if (ProfileInterpreter) {
1548 Label profile_continue;
1549
1550 // If no method data exists, go to profile_continue.
1551 test_method_data_pointer(mdp, profile_continue);
1552
1553 // Build the base (index * per_case_size_in_bytes()) +
1554 // case_array_offset_in_bytes()
1555 movl(reg2, in_bytes(MultiBranchData::per_case_size()));
1556 imulptr(index, reg2); // XXX l ?
1557 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ?
1558
1559 // Update the case count
1560 increment_mdp_data_at(mdp,
1561 index,
1562 in_bytes(MultiBranchData::relative_count_offset()));
1563
1564 // The method data pointer needs to be updated.
1565 update_mdp_by_offset(mdp,
1566 index,
1567 in_bytes(MultiBranchData::
1568 relative_displacement_offset()));
1569
1570 bind(profile_continue);
1571 }
1572 }
1573
1574
1575
1576 void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
1577 if (state == atos) {
1578 MacroAssembler::verify_oop(reg);
1579 }
1580 }
1581
1582 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
1583 #ifndef _LP64
1584 if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
1585 #endif
1586 }
1587
1588 // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
1589 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
1590 int increment, Address mask,
1591 Register scratch, bool preloaded,
1592 Condition cond, Label* where) {
1593 if (!preloaded) {
1594 movl(scratch, counter_addr);
1595 }
1596 incrementl(scratch, increment);
1597 movl(counter_addr, scratch);
1598 andl(scratch, mask);
1599 jcc(cond, *where);
1600 }
1601 #endif /* CC_INTERP */
1602
1603 void InterpreterMacroAssembler::notify_method_entry() {
1604 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1605 // track stack depth. If it is possible to enter interp_only_mode we add
1606 // the code to check if the event should be sent.
1607 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1608 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1609 if (JvmtiExport::can_post_interpreter_events()) {
1610 Label L;
1611 NOT_LP64(get_thread(rthread);)
1612 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1613 testl(rdx, rdx);
1614 jcc(Assembler::zero, L);
1615 call_VM(noreg, CAST_FROM_FN_PTR(address,
1616 InterpreterRuntime::post_method_entry));
1617 bind(L);
1618 }
1619
1620 {
1621 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1622 NOT_LP64(get_thread(rthread);)
1623 get_method(rarg);
1624 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry),
1625 rthread, rarg);
1626 }
1627
1628 // RedefineClasses() tracing support for obsolete method entry
1629 if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
1630 NOT_LP64(get_thread(rthread);)
1631 get_method(rarg);
1632 call_VM_leaf(
1633 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
1634 rthread, rarg);
1635 }
1636 }
1637
1638
1639 void InterpreterMacroAssembler::notify_method_exit(
1640 TosState state, NotifyMethodExitMode mode) {
1641 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
1642 // track stack depth. If it is possible to enter interp_only_mode we add
1643 // the code to check if the event should be sent.
1644 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
1645 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
1646 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
1647 Label L;
1648 // Note: frame::interpreter_frame_result has a dependency on how the
1649 // method result is saved across the call to post_method_exit. If this
1650 // is changed then the interpreter_frame_result implementation will
1651 // need to be updated too.
1652
1653 // For c++ interpreter the result is always stored at a known location in the frame
1654 // template interpreter will leave it on the top of the stack.
1655 NOT_CC_INTERP(push(state);)
1656 get_thread(rthread);
1657 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset()));
1658 testl(rdx, rdx);
1659 jcc(Assembler::zero, L);
1660 call_VM(noreg,
1661 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
1662 bind(L);
1663 NOT_CC_INTERP(pop(state));
1664 }
1665
1666 {
1667 SkipIfEqual skip(this, &DTraceMethodProbes, false);
1668 NOT_CC_INTERP(push(state));
1669 get_thread(rthread);
1670 get_method(rarg);
1671 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
1672 rthread, rarg);
1673 NOT_CC_INTERP(pop(state));
1674 }
1675 }