1 /*
2 * Copyright (c) 2003, 2011, 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 "asm/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/methodOop.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/frame.inline.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "utilities/top.hpp"
41 #ifdef TARGET_OS_FAMILY_linux
42 # include "thread_linux.inline.hpp"
43 #endif
44 #ifdef TARGET_OS_FAMILY_solaris
45 # include "thread_solaris.inline.hpp"
46 #endif
47 #ifdef TARGET_OS_FAMILY_windows
48 # include "thread_windows.inline.hpp"
49 #endif
50 #ifdef TARGET_OS_FAMILY_bsd
51 # include "thread_bsd.inline.hpp"
52 #endif
53 #ifdef COMPILER2
54 #include "opto/runtime.hpp"
55 #endif
56
57 // Declaration and definition of StubGenerator (no .hpp file).
58 // For a more detailed description of the stub routine structure
59 // see the comment in stubRoutines.hpp
60
61 #define __ _masm->
62 #define TIMES_OOP (UseCompressedOops ? Address::times_4 : Address::times_8)
63 #define a__ ((Assembler*)_masm)->
64
65 #ifdef PRODUCT
66 #define BLOCK_COMMENT(str) /* nothing */
67 #else
68 #define BLOCK_COMMENT(str) __ block_comment(str)
69 #endif
70
71 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
72 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
73
74 // Stub Code definitions
75
76 static address handle_unsafe_access() {
77 JavaThread* thread = JavaThread::current();
78 address pc = thread->saved_exception_pc();
79 // pc is the instruction which we must emulate
80 // doing a no-op is fine: return garbage from the load
81 // therefore, compute npc
82 address npc = Assembler::locate_next_instruction(pc);
83
84 // request an async exception
85 thread->set_pending_unsafe_access_error();
86
87 // return address of next instruction to execute
88 return npc;
89 }
90
91 class StubGenerator: public StubCodeGenerator {
92 private:
93
94 #ifdef PRODUCT
95 #define inc_counter_np(counter) (0)
96 #else
97 void inc_counter_np_(int& counter) {
98 // This can destroy rscratch1 if counter is far from the code cache
99 __ incrementl(ExternalAddress((address)&counter));
100 }
101 #define inc_counter_np(counter) \
102 BLOCK_COMMENT("inc_counter " #counter); \
103 inc_counter_np_(counter);
104 #endif
105
106 // Call stubs are used to call Java from C
107 //
108 // Linux Arguments:
109 // c_rarg0: call wrapper address address
110 // c_rarg1: result address
111 // c_rarg2: result type BasicType
112 // c_rarg3: method methodOop
113 // c_rarg4: (interpreter) entry point address
114 // c_rarg5: parameters intptr_t*
115 // 16(rbp): parameter size (in words) int
116 // 24(rbp): thread Thread*
117 //
118 // [ return_from_Java ] <--- rsp
119 // [ argument word n ]
120 // ...
121 // -12 [ argument word 1 ]
122 // -11 [ saved r15 ] <--- rsp_after_call
123 // -10 [ saved r14 ]
124 // -9 [ saved r13 ]
125 // -8 [ saved r12 ]
126 // -7 [ saved rbx ]
127 // -6 [ call wrapper ]
128 // -5 [ result ]
129 // -4 [ result type ]
130 // -3 [ method ]
131 // -2 [ entry point ]
132 // -1 [ parameters ]
133 // 0 [ saved rbp ] <--- rbp
134 // 1 [ return address ]
135 // 2 [ parameter size ]
136 // 3 [ thread ]
137 //
138 // Windows Arguments:
139 // c_rarg0: call wrapper address address
140 // c_rarg1: result address
141 // c_rarg2: result type BasicType
142 // c_rarg3: method methodOop
143 // 48(rbp): (interpreter) entry point address
144 // 56(rbp): parameters intptr_t*
145 // 64(rbp): parameter size (in words) int
146 // 72(rbp): thread Thread*
147 //
148 // [ return_from_Java ] <--- rsp
149 // [ argument word n ]
150 // ...
151 // -28 [ argument word 1 ]
152 // -27 [ saved xmm15 ] <--- rsp_after_call
153 // [ saved xmm7-xmm14 ]
154 // -9 [ saved xmm6 ] (each xmm register takes 2 slots)
155 // -7 [ saved r15 ]
156 // -6 [ saved r14 ]
157 // -5 [ saved r13 ]
158 // -4 [ saved r12 ]
159 // -3 [ saved rdi ]
160 // -2 [ saved rsi ]
161 // -1 [ saved rbx ]
162 // 0 [ saved rbp ] <--- rbp
163 // 1 [ return address ]
164 // 2 [ call wrapper ]
165 // 3 [ result ]
166 // 4 [ result type ]
167 // 5 [ method ]
168 // 6 [ entry point ]
169 // 7 [ parameters ]
170 // 8 [ parameter size ]
171 // 9 [ thread ]
172 //
173 // Windows reserves the callers stack space for arguments 1-4.
174 // We spill c_rarg0-c_rarg3 to this space.
175
176 // Call stub stack layout word offsets from rbp
177 enum call_stub_layout {
178 #ifdef _WIN64
179 xmm_save_first = 6, // save from xmm6
180 xmm_save_last = 15, // to xmm15
181 xmm_save_base = -9,
182 rsp_after_call_off = xmm_save_base - 2 * (xmm_save_last - xmm_save_first), // -27
183 r15_off = -7,
184 r14_off = -6,
185 r13_off = -5,
186 r12_off = -4,
187 rdi_off = -3,
188 rsi_off = -2,
189 rbx_off = -1,
190 rbp_off = 0,
191 retaddr_off = 1,
192 call_wrapper_off = 2,
193 result_off = 3,
194 result_type_off = 4,
195 method_off = 5,
196 entry_point_off = 6,
197 parameters_off = 7,
198 parameter_size_off = 8,
199 thread_off = 9
200 #else
201 rsp_after_call_off = -12,
202 mxcsr_off = rsp_after_call_off,
203 r15_off = -11,
204 r14_off = -10,
205 r13_off = -9,
206 r12_off = -8,
207 rbx_off = -7,
208 call_wrapper_off = -6,
209 result_off = -5,
210 result_type_off = -4,
211 method_off = -3,
212 entry_point_off = -2,
213 parameters_off = -1,
214 rbp_off = 0,
215 retaddr_off = 1,
216 parameter_size_off = 2,
217 thread_off = 3
218 #endif
219 };
220
221 #ifdef _WIN64
222 Address xmm_save(int reg) {
223 assert(reg >= xmm_save_first && reg <= xmm_save_last, "XMM register number out of range");
224 return Address(rbp, (xmm_save_base - (reg - xmm_save_first) * 2) * wordSize);
225 }
226 #endif
227
228 address generate_call_stub(address& return_address) {
229 assert((int)frame::entry_frame_after_call_words == -(int)rsp_after_call_off + 1 &&
230 (int)frame::entry_frame_call_wrapper_offset == (int)call_wrapper_off,
231 "adjust this code");
232 StubCodeMark mark(this, "StubRoutines", "call_stub");
233 address start = __ pc();
234
235 // same as in generate_catch_exception()!
236 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
237
238 const Address call_wrapper (rbp, call_wrapper_off * wordSize);
239 const Address result (rbp, result_off * wordSize);
240 const Address result_type (rbp, result_type_off * wordSize);
241 const Address method (rbp, method_off * wordSize);
242 const Address entry_point (rbp, entry_point_off * wordSize);
243 const Address parameters (rbp, parameters_off * wordSize);
244 const Address parameter_size(rbp, parameter_size_off * wordSize);
245
246 // same as in generate_catch_exception()!
247 const Address thread (rbp, thread_off * wordSize);
248
249 const Address r15_save(rbp, r15_off * wordSize);
250 const Address r14_save(rbp, r14_off * wordSize);
251 const Address r13_save(rbp, r13_off * wordSize);
252 const Address r12_save(rbp, r12_off * wordSize);
253 const Address rbx_save(rbp, rbx_off * wordSize);
254
255 // stub code
256 __ enter();
257 __ subptr(rsp, -rsp_after_call_off * wordSize);
258
259 // save register parameters
260 #ifndef _WIN64
261 __ movptr(parameters, c_rarg5); // parameters
262 __ movptr(entry_point, c_rarg4); // entry_point
263 #endif
264
265 __ movptr(method, c_rarg3); // method
266 __ movl(result_type, c_rarg2); // result type
267 __ movptr(result, c_rarg1); // result
268 __ movptr(call_wrapper, c_rarg0); // call wrapper
269
270 // save regs belonging to calling function
271 __ movptr(rbx_save, rbx);
272 __ movptr(r12_save, r12);
273 __ movptr(r13_save, r13);
274 __ movptr(r14_save, r14);
275 __ movptr(r15_save, r15);
276 #ifdef _WIN64
277 for (int i = 6; i <= 15; i++) {
278 __ movdqu(xmm_save(i), as_XMMRegister(i));
279 }
280
281 const Address rdi_save(rbp, rdi_off * wordSize);
282 const Address rsi_save(rbp, rsi_off * wordSize);
283
284 __ movptr(rsi_save, rsi);
285 __ movptr(rdi_save, rdi);
286 #else
287 const Address mxcsr_save(rbp, mxcsr_off * wordSize);
288 {
289 Label skip_ldmx;
290 __ stmxcsr(mxcsr_save);
291 __ movl(rax, mxcsr_save);
292 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
293 ExternalAddress mxcsr_std(StubRoutines::x86::mxcsr_std());
294 __ cmp32(rax, mxcsr_std);
295 __ jcc(Assembler::equal, skip_ldmx);
296 __ ldmxcsr(mxcsr_std);
297 __ bind(skip_ldmx);
298 }
299 #endif
300
301 // Load up thread register
302 __ movptr(r15_thread, thread);
303 __ reinit_heapbase();
304
305 #ifdef ASSERT
306 // make sure we have no pending exceptions
307 {
308 Label L;
309 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
310 __ jcc(Assembler::equal, L);
311 __ stop("StubRoutines::call_stub: entered with pending exception");
312 __ bind(L);
313 }
314 #endif
315
316 // pass parameters if any
317 BLOCK_COMMENT("pass parameters if any");
318 Label parameters_done;
319 __ movl(c_rarg3, parameter_size);
320 __ testl(c_rarg3, c_rarg3);
321 __ jcc(Assembler::zero, parameters_done);
322
323 Label loop;
324 __ movptr(c_rarg2, parameters); // parameter pointer
325 __ movl(c_rarg1, c_rarg3); // parameter counter is in c_rarg1
326 __ BIND(loop);
327 __ movptr(rax, Address(c_rarg2, 0));// get parameter
328 __ addptr(c_rarg2, wordSize); // advance to next parameter
329 __ decrementl(c_rarg1); // decrement counter
330 __ push(rax); // pass parameter
331 __ jcc(Assembler::notZero, loop);
332
333 // call Java function
334 __ BIND(parameters_done);
335 __ movptr(rbx, method); // get methodOop
336 __ movptr(c_rarg1, entry_point); // get entry_point
337 __ mov(r13, rsp); // set sender sp
338 BLOCK_COMMENT("call Java function");
339 __ call(c_rarg1);
340
341 BLOCK_COMMENT("call_stub_return_address:");
342 return_address = __ pc();
343
344 // store result depending on type (everything that is not
345 // T_OBJECT, T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
346 __ movptr(c_rarg0, result);
347 Label is_long, is_float, is_double, exit;
348 __ movl(c_rarg1, result_type);
349 __ cmpl(c_rarg1, T_OBJECT);
350 __ jcc(Assembler::equal, is_long);
351 __ cmpl(c_rarg1, T_LONG);
352 __ jcc(Assembler::equal, is_long);
353 __ cmpl(c_rarg1, T_FLOAT);
354 __ jcc(Assembler::equal, is_float);
355 __ cmpl(c_rarg1, T_DOUBLE);
356 __ jcc(Assembler::equal, is_double);
357
358 // handle T_INT case
359 __ movl(Address(c_rarg0, 0), rax);
360
361 __ BIND(exit);
362
363 // pop parameters
364 __ lea(rsp, rsp_after_call);
365
366 #ifdef ASSERT
367 // verify that threads correspond
368 {
369 Label L, S;
370 __ cmpptr(r15_thread, thread);
371 __ jcc(Assembler::notEqual, S);
372 __ get_thread(rbx);
373 __ cmpptr(r15_thread, rbx);
374 __ jcc(Assembler::equal, L);
375 __ bind(S);
376 __ jcc(Assembler::equal, L);
377 __ stop("StubRoutines::call_stub: threads must correspond");
378 __ bind(L);
379 }
380 #endif
381
382 // restore regs belonging to calling function
383 #ifdef _WIN64
384 for (int i = 15; i >= 6; i--) {
385 __ movdqu(as_XMMRegister(i), xmm_save(i));
386 }
387 #endif
388 __ movptr(r15, r15_save);
389 __ movptr(r14, r14_save);
390 __ movptr(r13, r13_save);
391 __ movptr(r12, r12_save);
392 __ movptr(rbx, rbx_save);
393
394 #ifdef _WIN64
395 __ movptr(rdi, rdi_save);
396 __ movptr(rsi, rsi_save);
397 #else
398 __ ldmxcsr(mxcsr_save);
399 #endif
400
401 // restore rsp
402 __ addptr(rsp, -rsp_after_call_off * wordSize);
403
404 // return
405 __ pop(rbp);
406 __ ret(0);
407
408 // handle return types different from T_INT
409 __ BIND(is_long);
410 __ movq(Address(c_rarg0, 0), rax);
411 __ jmp(exit);
412
413 __ BIND(is_float);
414 __ movflt(Address(c_rarg0, 0), xmm0);
415 __ jmp(exit);
416
417 __ BIND(is_double);
418 __ movdbl(Address(c_rarg0, 0), xmm0);
419 __ jmp(exit);
420
421 return start;
422 }
423
424 // Return point for a Java call if there's an exception thrown in
425 // Java code. The exception is caught and transformed into a
426 // pending exception stored in JavaThread that can be tested from
427 // within the VM.
428 //
429 // Note: Usually the parameters are removed by the callee. In case
430 // of an exception crossing an activation frame boundary, that is
431 // not the case if the callee is compiled code => need to setup the
432 // rsp.
433 //
434 // rax: exception oop
435
436 address generate_catch_exception() {
437 StubCodeMark mark(this, "StubRoutines", "catch_exception");
438 address start = __ pc();
439
440 // same as in generate_call_stub():
441 const Address rsp_after_call(rbp, rsp_after_call_off * wordSize);
442 const Address thread (rbp, thread_off * wordSize);
443
444 #ifdef ASSERT
445 // verify that threads correspond
446 {
447 Label L, S;
448 __ cmpptr(r15_thread, thread);
449 __ jcc(Assembler::notEqual, S);
450 __ get_thread(rbx);
451 __ cmpptr(r15_thread, rbx);
452 __ jcc(Assembler::equal, L);
453 __ bind(S);
454 __ stop("StubRoutines::catch_exception: threads must correspond");
455 __ bind(L);
456 }
457 #endif
458
459 // set pending exception
460 __ verify_oop(rax);
461
462 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
463 __ lea(rscratch1, ExternalAddress((address)__FILE__));
464 __ movptr(Address(r15_thread, Thread::exception_file_offset()), rscratch1);
465 __ movl(Address(r15_thread, Thread::exception_line_offset()), (int) __LINE__);
466
467 // complete return to VM
468 assert(StubRoutines::_call_stub_return_address != NULL,
469 "_call_stub_return_address must have been generated before");
470 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
471
472 return start;
473 }
474
475 // Continuation point for runtime calls returning with a pending
476 // exception. The pending exception check happened in the runtime
477 // or native call stub. The pending exception in Thread is
478 // converted into a Java-level exception.
479 //
480 // Contract with Java-level exception handlers:
481 // rax: exception
482 // rdx: throwing pc
483 //
484 // NOTE: At entry of this stub, exception-pc must be on stack !!
485
486 address generate_forward_exception() {
487 StubCodeMark mark(this, "StubRoutines", "forward exception");
488 address start = __ pc();
489
490 // Upon entry, the sp points to the return address returning into
491 // Java (interpreted or compiled) code; i.e., the return address
492 // becomes the throwing pc.
493 //
494 // Arguments pushed before the runtime call are still on the stack
495 // but the exception handler will reset the stack pointer ->
496 // ignore them. A potential result in registers can be ignored as
497 // well.
498
499 #ifdef ASSERT
500 // make sure this code is only executed if there is a pending exception
501 {
502 Label L;
503 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t) NULL);
504 __ jcc(Assembler::notEqual, L);
505 __ stop("StubRoutines::forward exception: no pending exception (1)");
506 __ bind(L);
507 }
508 #endif
509
510 // compute exception handler into rbx
511 __ movptr(c_rarg0, Address(rsp, 0));
512 BLOCK_COMMENT("call exception_handler_for_return_address");
513 __ call_VM_leaf(CAST_FROM_FN_PTR(address,
514 SharedRuntime::exception_handler_for_return_address),
515 r15_thread, c_rarg0);
516 __ mov(rbx, rax);
517
518 // setup rax & rdx, remove return address & clear pending exception
519 __ pop(rdx);
520 __ movptr(rax, Address(r15_thread, Thread::pending_exception_offset()));
521 __ movptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
522
523 #ifdef ASSERT
524 // make sure exception is set
525 {
526 Label L;
527 __ testptr(rax, rax);
528 __ jcc(Assembler::notEqual, L);
529 __ stop("StubRoutines::forward exception: no pending exception (2)");
530 __ bind(L);
531 }
532 #endif
533
534 // continue at exception handler (return address removed)
535 // rax: exception
536 // rbx: exception handler
537 // rdx: throwing pc
538 __ verify_oop(rax);
539 __ jmp(rbx);
540
541 return start;
542 }
543
544 // Support for jint atomic::xchg(jint exchange_value, volatile jint* dest)
545 //
546 // Arguments :
547 // c_rarg0: exchange_value
548 // c_rarg0: dest
549 //
550 // Result:
551 // *dest <- ex, return (orig *dest)
552 address generate_atomic_xchg() {
553 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
554 address start = __ pc();
555
556 __ movl(rax, c_rarg0); // Copy to eax we need a return value anyhow
557 __ xchgl(rax, Address(c_rarg1, 0)); // automatic LOCK
558 __ ret(0);
559
560 return start;
561 }
562
563 // Support for intptr_t atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest)
564 //
565 // Arguments :
566 // c_rarg0: exchange_value
567 // c_rarg1: dest
568 //
569 // Result:
570 // *dest <- ex, return (orig *dest)
571 address generate_atomic_xchg_ptr() {
572 StubCodeMark mark(this, "StubRoutines", "atomic_xchg_ptr");
573 address start = __ pc();
574
575 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
576 __ xchgptr(rax, Address(c_rarg1, 0)); // automatic LOCK
577 __ ret(0);
578
579 return start;
580 }
581
582 // Support for jint atomic::atomic_cmpxchg(jint exchange_value, volatile jint* dest,
583 // jint compare_value)
584 //
585 // Arguments :
586 // c_rarg0: exchange_value
587 // c_rarg1: dest
588 // c_rarg2: compare_value
589 //
590 // Result:
591 // if ( compare_value == *dest ) {
592 // *dest = exchange_value
593 // return compare_value;
594 // else
595 // return *dest;
596 address generate_atomic_cmpxchg() {
597 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg");
598 address start = __ pc();
599
600 __ movl(rax, c_rarg2);
601 if ( os::is_MP() ) __ lock();
602 __ cmpxchgl(c_rarg0, Address(c_rarg1, 0));
603 __ ret(0);
604
605 return start;
606 }
607
608 // Support for jint atomic::atomic_cmpxchg_long(jlong exchange_value,
609 // volatile jlong* dest,
610 // jlong compare_value)
611 // Arguments :
612 // c_rarg0: exchange_value
613 // c_rarg1: dest
614 // c_rarg2: compare_value
615 //
616 // Result:
617 // if ( compare_value == *dest ) {
618 // *dest = exchange_value
619 // return compare_value;
620 // else
621 // return *dest;
622 address generate_atomic_cmpxchg_long() {
623 StubCodeMark mark(this, "StubRoutines", "atomic_cmpxchg_long");
624 address start = __ pc();
625
626 __ movq(rax, c_rarg2);
627 if ( os::is_MP() ) __ lock();
628 __ cmpxchgq(c_rarg0, Address(c_rarg1, 0));
629 __ ret(0);
630
631 return start;
632 }
633
634 // Support for jint atomic::add(jint add_value, volatile jint* dest)
635 //
636 // Arguments :
637 // c_rarg0: add_value
638 // c_rarg1: dest
639 //
640 // Result:
641 // *dest += add_value
642 // return *dest;
643 address generate_atomic_add() {
644 StubCodeMark mark(this, "StubRoutines", "atomic_add");
645 address start = __ pc();
646
647 __ movl(rax, c_rarg0);
648 if ( os::is_MP() ) __ lock();
649 __ xaddl(Address(c_rarg1, 0), c_rarg0);
650 __ addl(rax, c_rarg0);
651 __ ret(0);
652
653 return start;
654 }
655
656 // Support for intptr_t atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest)
657 //
658 // Arguments :
659 // c_rarg0: add_value
660 // c_rarg1: dest
661 //
662 // Result:
663 // *dest += add_value
664 // return *dest;
665 address generate_atomic_add_ptr() {
666 StubCodeMark mark(this, "StubRoutines", "atomic_add_ptr");
667 address start = __ pc();
668
669 __ movptr(rax, c_rarg0); // Copy to eax we need a return value anyhow
670 if ( os::is_MP() ) __ lock();
671 __ xaddptr(Address(c_rarg1, 0), c_rarg0);
672 __ addptr(rax, c_rarg0);
673 __ ret(0);
674
675 return start;
676 }
677
678 // Support for intptr_t OrderAccess::fence()
679 //
680 // Arguments :
681 //
682 // Result:
683 address generate_orderaccess_fence() {
684 StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
685 address start = __ pc();
686 __ membar(Assembler::StoreLoad);
687 __ ret(0);
688
689 return start;
690 }
691
692 // Support for intptr_t get_previous_fp()
693 //
694 // This routine is used to find the previous frame pointer for the
695 // caller (current_frame_guess). This is used as part of debugging
696 // ps() is seemingly lost trying to find frames.
697 // This code assumes that caller current_frame_guess) has a frame.
698 address generate_get_previous_fp() {
699 StubCodeMark mark(this, "StubRoutines", "get_previous_fp");
700 const Address old_fp(rbp, 0);
701 const Address older_fp(rax, 0);
702 address start = __ pc();
703
704 __ enter();
705 __ movptr(rax, old_fp); // callers fp
706 __ movptr(rax, older_fp); // the frame for ps()
707 __ pop(rbp);
708 __ ret(0);
709
710 return start;
711 }
712
713 //----------------------------------------------------------------------------------------------------
714 // Support for void verify_mxcsr()
715 //
716 // This routine is used with -Xcheck:jni to verify that native
717 // JNI code does not return to Java code without restoring the
718 // MXCSR register to our expected state.
719
720 address generate_verify_mxcsr() {
721 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
722 address start = __ pc();
723
724 const Address mxcsr_save(rsp, 0);
725
726 if (CheckJNICalls) {
727 Label ok_ret;
728 __ push(rax);
729 __ subptr(rsp, wordSize); // allocate a temp location
730 __ stmxcsr(mxcsr_save);
731 __ movl(rax, mxcsr_save);
732 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
733 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
734 __ jcc(Assembler::equal, ok_ret);
735
736 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
737
738 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
739
740 __ bind(ok_ret);
741 __ addptr(rsp, wordSize);
742 __ pop(rax);
743 }
744
745 __ ret(0);
746
747 return start;
748 }
749
750 address generate_f2i_fixup() {
751 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
752 Address inout(rsp, 5 * wordSize); // return address + 4 saves
753
754 address start = __ pc();
755
756 Label L;
757
758 __ push(rax);
759 __ push(c_rarg3);
760 __ push(c_rarg2);
761 __ push(c_rarg1);
762
763 __ movl(rax, 0x7f800000);
764 __ xorl(c_rarg3, c_rarg3);
765 __ movl(c_rarg2, inout);
766 __ movl(c_rarg1, c_rarg2);
767 __ andl(c_rarg1, 0x7fffffff);
768 __ cmpl(rax, c_rarg1); // NaN? -> 0
769 __ jcc(Assembler::negative, L);
770 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
771 __ movl(c_rarg3, 0x80000000);
772 __ movl(rax, 0x7fffffff);
773 __ cmovl(Assembler::positive, c_rarg3, rax);
774
775 __ bind(L);
776 __ movptr(inout, c_rarg3);
777
778 __ pop(c_rarg1);
779 __ pop(c_rarg2);
780 __ pop(c_rarg3);
781 __ pop(rax);
782
783 __ ret(0);
784
785 return start;
786 }
787
788 address generate_f2l_fixup() {
789 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
790 Address inout(rsp, 5 * wordSize); // return address + 4 saves
791 address start = __ pc();
792
793 Label L;
794
795 __ push(rax);
796 __ push(c_rarg3);
797 __ push(c_rarg2);
798 __ push(c_rarg1);
799
800 __ movl(rax, 0x7f800000);
801 __ xorl(c_rarg3, c_rarg3);
802 __ movl(c_rarg2, inout);
803 __ movl(c_rarg1, c_rarg2);
804 __ andl(c_rarg1, 0x7fffffff);
805 __ cmpl(rax, c_rarg1); // NaN? -> 0
806 __ jcc(Assembler::negative, L);
807 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
808 __ mov64(c_rarg3, 0x8000000000000000);
809 __ mov64(rax, 0x7fffffffffffffff);
810 __ cmov(Assembler::positive, c_rarg3, rax);
811
812 __ bind(L);
813 __ movptr(inout, c_rarg3);
814
815 __ pop(c_rarg1);
816 __ pop(c_rarg2);
817 __ pop(c_rarg3);
818 __ pop(rax);
819
820 __ ret(0);
821
822 return start;
823 }
824
825 address generate_d2i_fixup() {
826 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
827 Address inout(rsp, 6 * wordSize); // return address + 5 saves
828
829 address start = __ pc();
830
831 Label L;
832
833 __ push(rax);
834 __ push(c_rarg3);
835 __ push(c_rarg2);
836 __ push(c_rarg1);
837 __ push(c_rarg0);
838
839 __ movl(rax, 0x7ff00000);
840 __ movq(c_rarg2, inout);
841 __ movl(c_rarg3, c_rarg2);
842 __ mov(c_rarg1, c_rarg2);
843 __ mov(c_rarg0, c_rarg2);
844 __ negl(c_rarg3);
845 __ shrptr(c_rarg1, 0x20);
846 __ orl(c_rarg3, c_rarg2);
847 __ andl(c_rarg1, 0x7fffffff);
848 __ xorl(c_rarg2, c_rarg2);
849 __ shrl(c_rarg3, 0x1f);
850 __ orl(c_rarg1, c_rarg3);
851 __ cmpl(rax, c_rarg1);
852 __ jcc(Assembler::negative, L); // NaN -> 0
853 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
854 __ movl(c_rarg2, 0x80000000);
855 __ movl(rax, 0x7fffffff);
856 __ cmov(Assembler::positive, c_rarg2, rax);
857
858 __ bind(L);
859 __ movptr(inout, c_rarg2);
860
861 __ pop(c_rarg0);
862 __ pop(c_rarg1);
863 __ pop(c_rarg2);
864 __ pop(c_rarg3);
865 __ pop(rax);
866
867 __ ret(0);
868
869 return start;
870 }
871
872 address generate_d2l_fixup() {
873 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
874 Address inout(rsp, 6 * wordSize); // return address + 5 saves
875
876 address start = __ pc();
877
878 Label L;
879
880 __ push(rax);
881 __ push(c_rarg3);
882 __ push(c_rarg2);
883 __ push(c_rarg1);
884 __ push(c_rarg0);
885
886 __ movl(rax, 0x7ff00000);
887 __ movq(c_rarg2, inout);
888 __ movl(c_rarg3, c_rarg2);
889 __ mov(c_rarg1, c_rarg2);
890 __ mov(c_rarg0, c_rarg2);
891 __ negl(c_rarg3);
892 __ shrptr(c_rarg1, 0x20);
893 __ orl(c_rarg3, c_rarg2);
894 __ andl(c_rarg1, 0x7fffffff);
895 __ xorl(c_rarg2, c_rarg2);
896 __ shrl(c_rarg3, 0x1f);
897 __ orl(c_rarg1, c_rarg3);
898 __ cmpl(rax, c_rarg1);
899 __ jcc(Assembler::negative, L); // NaN -> 0
900 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
901 __ mov64(c_rarg2, 0x8000000000000000);
902 __ mov64(rax, 0x7fffffffffffffff);
903 __ cmovq(Assembler::positive, c_rarg2, rax);
904
905 __ bind(L);
906 __ movq(inout, c_rarg2);
907
908 __ pop(c_rarg0);
909 __ pop(c_rarg1);
910 __ pop(c_rarg2);
911 __ pop(c_rarg3);
912 __ pop(rax);
913
914 __ ret(0);
915
916 return start;
917 }
918
919 address generate_fp_mask(const char *stub_name, int64_t mask) {
920 __ align(CodeEntryAlignment);
921 StubCodeMark mark(this, "StubRoutines", stub_name);
922 address start = __ pc();
923
924 __ emit_data64( mask, relocInfo::none );
925 __ emit_data64( mask, relocInfo::none );
926
927 return start;
928 }
929
930 // The following routine generates a subroutine to throw an
931 // asynchronous UnknownError when an unsafe access gets a fault that
932 // could not be reasonably prevented by the programmer. (Example:
933 // SIGBUS/OBJERR.)
934 address generate_handler_for_unsafe_access() {
935 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
936 address start = __ pc();
937
938 __ push(0); // hole for return address-to-be
939 __ pusha(); // push registers
940 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
941
942 // FIXME: this probably needs alignment logic
943
944 __ subptr(rsp, frame::arg_reg_save_area_bytes);
945 BLOCK_COMMENT("call handle_unsafe_access");
946 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
947 __ addptr(rsp, frame::arg_reg_save_area_bytes);
948
949 __ movptr(next_pc, rax); // stuff next address
950 __ popa();
951 __ ret(0); // jump to next address
952
953 return start;
954 }
955
956 // Non-destructive plausibility checks for oops
957 //
958 // Arguments:
959 // all args on stack!
960 //
961 // Stack after saving c_rarg3:
962 // [tos + 0]: saved c_rarg3
963 // [tos + 1]: saved c_rarg2
964 // [tos + 2]: saved r12 (several TemplateTable methods use it)
965 // [tos + 3]: saved flags
966 // [tos + 4]: return address
967 // * [tos + 5]: error message (char*)
968 // * [tos + 6]: object to verify (oop)
969 // * [tos + 7]: saved rax - saved by caller and bashed
970 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
971 // * = popped on exit
972 address generate_verify_oop() {
973 StubCodeMark mark(this, "StubRoutines", "verify_oop");
974 address start = __ pc();
975
976 Label exit, error;
977
978 __ pushf();
979 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
980
981 __ push(r12);
982
983 // save c_rarg2 and c_rarg3
984 __ push(c_rarg2);
985 __ push(c_rarg3);
986
987 enum {
988 // After previous pushes.
989 oop_to_verify = 6 * wordSize,
990 saved_rax = 7 * wordSize,
991 saved_r10 = 8 * wordSize,
992
993 // Before the call to MacroAssembler::debug(), see below.
994 return_addr = 16 * wordSize,
995 error_msg = 17 * wordSize
996 };
997
998 // get object
999 __ movptr(rax, Address(rsp, oop_to_verify));
1000
1001 // make sure object is 'reasonable'
1002 __ testptr(rax, rax);
1003 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1004 // Check if the oop is in the right area of memory
1005 __ movptr(c_rarg2, rax);
1006 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1007 __ andptr(c_rarg2, c_rarg3);
1008 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1009 __ cmpptr(c_rarg2, c_rarg3);
1010 __ jcc(Assembler::notZero, error);
1011
1012 // set r12 to heapbase for load_klass()
1013 __ reinit_heapbase();
1014
1015 // make sure klass is 'reasonable'
1016 __ load_klass(rax, rax); // get klass
1017 __ testptr(rax, rax);
1018 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1019 // Check if the klass is in the right area of memory
1020 __ mov(c_rarg2, rax);
1021 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1022 __ andptr(c_rarg2, c_rarg3);
1023 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1024 __ cmpptr(c_rarg2, c_rarg3);
1025 __ jcc(Assembler::notZero, error);
1026
1027 // make sure klass' klass is 'reasonable'
1028 __ load_klass(rax, rax);
1029 __ testptr(rax, rax);
1030 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
1031 // Check if the klass' klass is in the right area of memory
1032 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1033 __ andptr(rax, c_rarg3);
1034 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1035 __ cmpptr(rax, c_rarg3);
1036 __ jcc(Assembler::notZero, error);
1037
1038 // return if everything seems ok
1039 __ bind(exit);
1040 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1041 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1042 __ pop(c_rarg3); // restore c_rarg3
1043 __ pop(c_rarg2); // restore c_rarg2
1044 __ pop(r12); // restore r12
1045 __ popf(); // restore flags
1046 __ ret(4 * wordSize); // pop caller saved stuff
1047
1048 // handle errors
1049 __ bind(error);
1050 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1051 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1052 __ pop(c_rarg3); // get saved c_rarg3 back
1053 __ pop(c_rarg2); // get saved c_rarg2 back
1054 __ pop(r12); // get saved r12 back
1055 __ popf(); // get saved flags off stack --
1056 // will be ignored
1057
1058 __ pusha(); // push registers
1059 // (rip is already
1060 // already pushed)
1061 // debug(char* msg, int64_t pc, int64_t regs[])
1062 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1063 // pushed all the registers, so now the stack looks like:
1064 // [tos + 0] 16 saved registers
1065 // [tos + 16] return address
1066 // * [tos + 17] error message (char*)
1067 // * [tos + 18] object to verify (oop)
1068 // * [tos + 19] saved rax - saved by caller and bashed
1069 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1070 // * = popped on exit
1071
1072 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1073 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1074 __ movq(c_rarg2, rsp); // pass address of regs on stack
1075 __ mov(r12, rsp); // remember rsp
1076 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1077 __ andptr(rsp, -16); // align stack as required by ABI
1078 BLOCK_COMMENT("call MacroAssembler::debug");
1079 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1080 __ mov(rsp, r12); // restore rsp
1081 __ popa(); // pop registers (includes r12)
1082 __ ret(4 * wordSize); // pop caller saved stuff
1083
1084 return start;
1085 }
1086
1087 //
1088 // Verify that a register contains clean 32-bits positive value
1089 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1090 //
1091 // Input:
1092 // Rint - 32-bits value
1093 // Rtmp - scratch
1094 //
1095 void assert_clean_int(Register Rint, Register Rtmp) {
1096 #ifdef ASSERT
1097 Label L;
1098 assert_different_registers(Rtmp, Rint);
1099 __ movslq(Rtmp, Rint);
1100 __ cmpq(Rtmp, Rint);
1101 __ jcc(Assembler::equal, L);
1102 __ stop("high 32-bits of int value are not 0");
1103 __ bind(L);
1104 #endif
1105 }
1106
1107 // Generate overlap test for array copy stubs
1108 //
1109 // Input:
1110 // c_rarg0 - from
1111 // c_rarg1 - to
1112 // c_rarg2 - element count
1113 //
1114 // Output:
1115 // rax - &from[element count - 1]
1116 //
1117 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1118 assert(no_overlap_target != NULL, "must be generated");
1119 array_overlap_test(no_overlap_target, NULL, sf);
1120 }
1121 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1122 array_overlap_test(NULL, &L_no_overlap, sf);
1123 }
1124 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1125 const Register from = c_rarg0;
1126 const Register to = c_rarg1;
1127 const Register count = c_rarg2;
1128 const Register end_from = rax;
1129
1130 __ cmpptr(to, from);
1131 __ lea(end_from, Address(from, count, sf, 0));
1132 if (NOLp == NULL) {
1133 ExternalAddress no_overlap(no_overlap_target);
1134 __ jump_cc(Assembler::belowEqual, no_overlap);
1135 __ cmpptr(to, end_from);
1136 __ jump_cc(Assembler::aboveEqual, no_overlap);
1137 } else {
1138 __ jcc(Assembler::belowEqual, (*NOLp));
1139 __ cmpptr(to, end_from);
1140 __ jcc(Assembler::aboveEqual, (*NOLp));
1141 }
1142 }
1143
1144 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1145 //
1146 // Outputs:
1147 // rdi - rcx
1148 // rsi - rdx
1149 // rdx - r8
1150 // rcx - r9
1151 //
1152 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1153 // are non-volatile. r9 and r10 should not be used by the caller.
1154 //
1155 void setup_arg_regs(int nargs = 3) {
1156 const Register saved_rdi = r9;
1157 const Register saved_rsi = r10;
1158 assert(nargs == 3 || nargs == 4, "else fix");
1159 #ifdef _WIN64
1160 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1161 "unexpected argument registers");
1162 if (nargs >= 4)
1163 __ mov(rax, r9); // r9 is also saved_rdi
1164 __ movptr(saved_rdi, rdi);
1165 __ movptr(saved_rsi, rsi);
1166 __ mov(rdi, rcx); // c_rarg0
1167 __ mov(rsi, rdx); // c_rarg1
1168 __ mov(rdx, r8); // c_rarg2
1169 if (nargs >= 4)
1170 __ mov(rcx, rax); // c_rarg3 (via rax)
1171 #else
1172 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1173 "unexpected argument registers");
1174 #endif
1175 }
1176
1177 void restore_arg_regs() {
1178 const Register saved_rdi = r9;
1179 const Register saved_rsi = r10;
1180 #ifdef _WIN64
1181 __ movptr(rdi, saved_rdi);
1182 __ movptr(rsi, saved_rsi);
1183 #endif
1184 }
1185
1186 // Generate code for an array write pre barrier
1187 //
1188 // addr - starting address
1189 // count - element count
1190 // tmp - scratch register
1191 //
1192 // Destroy no registers!
1193 //
1194 void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1195 BarrierSet* bs = Universe::heap()->barrier_set();
1196 switch (bs->kind()) {
1197 case BarrierSet::G1SATBCT:
1198 case BarrierSet::G1SATBCTLogging:
1199 // With G1, don't generate the call if we statically know that the target in uninitialized
1200 if (!dest_uninitialized) {
1201 __ pusha(); // push registers
1202 if (count == c_rarg0) {
1203 if (addr == c_rarg1) {
1204 // exactly backwards!!
1205 __ xchgptr(c_rarg1, c_rarg0);
1206 } else {
1207 __ movptr(c_rarg1, count);
1208 __ movptr(c_rarg0, addr);
1209 }
1210 } else {
1211 __ movptr(c_rarg0, addr);
1212 __ movptr(c_rarg1, count);
1213 }
1214 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1215 __ popa();
1216 }
1217 break;
1218 case BarrierSet::CardTableModRef:
1219 case BarrierSet::CardTableExtension:
1220 case BarrierSet::ModRef:
1221 break;
1222 default:
1223 ShouldNotReachHere();
1224
1225 }
1226 }
1227
1228 //
1229 // Generate code for an array write post barrier
1230 //
1231 // Input:
1232 // start - register containing starting address of destination array
1233 // end - register containing ending address of destination array
1234 // scratch - scratch register
1235 //
1236 // The input registers are overwritten.
1237 // The ending address is inclusive.
1238 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1239 assert_different_registers(start, end, scratch);
1240 BarrierSet* bs = Universe::heap()->barrier_set();
1241 switch (bs->kind()) {
1242 case BarrierSet::G1SATBCT:
1243 case BarrierSet::G1SATBCTLogging:
1244
1245 {
1246 __ pusha(); // push registers (overkill)
1247 // must compute element count unless barrier set interface is changed (other platforms supply count)
1248 assert_different_registers(start, end, scratch);
1249 __ lea(scratch, Address(end, BytesPerHeapOop));
1250 __ subptr(scratch, start); // subtract start to get #bytes
1251 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1252 __ mov(c_rarg0, start);
1253 __ mov(c_rarg1, scratch);
1254 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1255 __ popa();
1256 }
1257 break;
1258 case BarrierSet::CardTableModRef:
1259 case BarrierSet::CardTableExtension:
1260 {
1261 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1262 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1263
1264 Label L_loop;
1265
1266 __ shrptr(start, CardTableModRefBS::card_shift);
1267 __ addptr(end, BytesPerHeapOop);
1268 __ shrptr(end, CardTableModRefBS::card_shift);
1269 __ subptr(end, start); // number of bytes to copy
1270
1271 intptr_t disp = (intptr_t) ct->byte_map_base;
1272 if (Assembler::is_simm32(disp)) {
1273 Address cardtable(noreg, noreg, Address::no_scale, disp);
1274 __ lea(scratch, cardtable);
1275 } else {
1276 ExternalAddress cardtable((address)disp);
1277 __ lea(scratch, cardtable);
1278 }
1279
1280 const Register count = end; // 'end' register contains bytes count now
1281 __ addptr(start, scratch);
1282 __ BIND(L_loop);
1283 __ movb(Address(start, count, Address::times_1), 0);
1284 __ decrement(count);
1285 __ jcc(Assembler::greaterEqual, L_loop);
1286 }
1287 break;
1288 default:
1289 ShouldNotReachHere();
1290
1291 }
1292 }
1293
1294
1295 // Copy big chunks forward
1296 //
1297 // Inputs:
1298 // end_from - source arrays end address
1299 // end_to - destination array end address
1300 // qword_count - 64-bits element count, negative
1301 // to - scratch
1302 // L_copy_32_bytes - entry label
1303 // L_copy_8_bytes - exit label
1304 //
1305 void copy_32_bytes_forward(Register end_from, Register end_to,
1306 Register qword_count, Register to,
1307 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1308 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1309 Label L_loop;
1310 __ align(OptoLoopAlignment);
1311 __ BIND(L_loop);
1312 if(UseUnalignedLoadStores) {
1313 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1314 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1315 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1316 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1317
1318 } else {
1319 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1320 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1321 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1322 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1323 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1324 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1325 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1326 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1327 }
1328 __ BIND(L_copy_32_bytes);
1329 __ addptr(qword_count, 4);
1330 __ jcc(Assembler::lessEqual, L_loop);
1331 __ subptr(qword_count, 4);
1332 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1333 }
1334
1335
1336 // Copy big chunks backward
1337 //
1338 // Inputs:
1339 // from - source arrays address
1340 // dest - destination array address
1341 // qword_count - 64-bits element count
1342 // to - scratch
1343 // L_copy_32_bytes - entry label
1344 // L_copy_8_bytes - exit label
1345 //
1346 void copy_32_bytes_backward(Register from, Register dest,
1347 Register qword_count, Register to,
1348 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1349 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1350 Label L_loop;
1351 __ align(OptoLoopAlignment);
1352 __ BIND(L_loop);
1353 if(UseUnalignedLoadStores) {
1354 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1355 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1356 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1357 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1358
1359 } else {
1360 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1361 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1362 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1363 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1364 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1365 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1366 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1367 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1368 }
1369 __ BIND(L_copy_32_bytes);
1370 __ subptr(qword_count, 4);
1371 __ jcc(Assembler::greaterEqual, L_loop);
1372 __ addptr(qword_count, 4);
1373 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1374 }
1375
1376
1377 // Arguments:
1378 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1379 // ignored
1380 // name - stub name string
1381 //
1382 // Inputs:
1383 // c_rarg0 - source array address
1384 // c_rarg1 - destination array address
1385 // c_rarg2 - element count, treated as ssize_t, can be zero
1386 //
1387 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1388 // we let the hardware handle it. The one to eight bytes within words,
1389 // dwords or qwords that span cache line boundaries will still be loaded
1390 // and stored atomically.
1391 //
1392 // Side Effects:
1393 // disjoint_byte_copy_entry is set to the no-overlap entry point
1394 // used by generate_conjoint_byte_copy().
1395 //
1396 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1397 __ align(CodeEntryAlignment);
1398 StubCodeMark mark(this, "StubRoutines", name);
1399 address start = __ pc();
1400
1401 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1402 Label L_copy_byte, L_exit;
1403 const Register from = rdi; // source array address
1404 const Register to = rsi; // destination array address
1405 const Register count = rdx; // elements count
1406 const Register byte_count = rcx;
1407 const Register qword_count = count;
1408 const Register end_from = from; // source array end address
1409 const Register end_to = to; // destination array end address
1410 // End pointers are inclusive, and if count is not zero they point
1411 // to the last unit copied: end_to[0] := end_from[0]
1412
1413 __ enter(); // required for proper stackwalking of RuntimeStub frame
1414 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1415
1416 if (entry != NULL) {
1417 *entry = __ pc();
1418 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1419 BLOCK_COMMENT("Entry:");
1420 }
1421
1422 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1423 // r9 and r10 may be used to save non-volatile registers
1424
1425 // 'from', 'to' and 'count' are now valid
1426 __ movptr(byte_count, count);
1427 __ shrptr(count, 3); // count => qword_count
1428
1429 // Copy from low to high addresses. Use 'to' as scratch.
1430 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1431 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1432 __ negptr(qword_count); // make the count negative
1433 __ jmp(L_copy_32_bytes);
1434
1435 // Copy trailing qwords
1436 __ BIND(L_copy_8_bytes);
1437 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1438 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1439 __ increment(qword_count);
1440 __ jcc(Assembler::notZero, L_copy_8_bytes);
1441
1442 // Check for and copy trailing dword
1443 __ BIND(L_copy_4_bytes);
1444 __ testl(byte_count, 4);
1445 __ jccb(Assembler::zero, L_copy_2_bytes);
1446 __ movl(rax, Address(end_from, 8));
1447 __ movl(Address(end_to, 8), rax);
1448
1449 __ addptr(end_from, 4);
1450 __ addptr(end_to, 4);
1451
1452 // Check for and copy trailing word
1453 __ BIND(L_copy_2_bytes);
1454 __ testl(byte_count, 2);
1455 __ jccb(Assembler::zero, L_copy_byte);
1456 __ movw(rax, Address(end_from, 8));
1457 __ movw(Address(end_to, 8), rax);
1458
1459 __ addptr(end_from, 2);
1460 __ addptr(end_to, 2);
1461
1462 // Check for and copy trailing byte
1463 __ BIND(L_copy_byte);
1464 __ testl(byte_count, 1);
1465 __ jccb(Assembler::zero, L_exit);
1466 __ movb(rax, Address(end_from, 8));
1467 __ movb(Address(end_to, 8), rax);
1468
1469 __ BIND(L_exit);
1470 restore_arg_regs();
1471 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1472 __ xorptr(rax, rax); // return 0
1473 __ leave(); // required for proper stackwalking of RuntimeStub frame
1474 __ ret(0);
1475
1476 // Copy in 32-bytes chunks
1477 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1478 __ jmp(L_copy_4_bytes);
1479
1480 return start;
1481 }
1482
1483 // Arguments:
1484 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1485 // ignored
1486 // name - stub name string
1487 //
1488 // Inputs:
1489 // c_rarg0 - source array address
1490 // c_rarg1 - destination array address
1491 // c_rarg2 - element count, treated as ssize_t, can be zero
1492 //
1493 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1494 // we let the hardware handle it. The one to eight bytes within words,
1495 // dwords or qwords that span cache line boundaries will still be loaded
1496 // and stored atomically.
1497 //
1498 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1499 address* entry, const char *name) {
1500 __ align(CodeEntryAlignment);
1501 StubCodeMark mark(this, "StubRoutines", name);
1502 address start = __ pc();
1503
1504 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1505 const Register from = rdi; // source array address
1506 const Register to = rsi; // destination array address
1507 const Register count = rdx; // elements count
1508 const Register byte_count = rcx;
1509 const Register qword_count = count;
1510
1511 __ enter(); // required for proper stackwalking of RuntimeStub frame
1512 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1513
1514 if (entry != NULL) {
1515 *entry = __ pc();
1516 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1517 BLOCK_COMMENT("Entry:");
1518 }
1519
1520 array_overlap_test(nooverlap_target, Address::times_1);
1521 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1522 // r9 and r10 may be used to save non-volatile registers
1523
1524 // 'from', 'to' and 'count' are now valid
1525 __ movptr(byte_count, count);
1526 __ shrptr(count, 3); // count => qword_count
1527
1528 // Copy from high to low addresses.
1529
1530 // Check for and copy trailing byte
1531 __ testl(byte_count, 1);
1532 __ jcc(Assembler::zero, L_copy_2_bytes);
1533 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1534 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1535 __ decrement(byte_count); // Adjust for possible trailing word
1536
1537 // Check for and copy trailing word
1538 __ BIND(L_copy_2_bytes);
1539 __ testl(byte_count, 2);
1540 __ jcc(Assembler::zero, L_copy_4_bytes);
1541 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1542 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1543
1544 // Check for and copy trailing dword
1545 __ BIND(L_copy_4_bytes);
1546 __ testl(byte_count, 4);
1547 __ jcc(Assembler::zero, L_copy_32_bytes);
1548 __ movl(rax, Address(from, qword_count, Address::times_8));
1549 __ movl(Address(to, qword_count, Address::times_8), rax);
1550 __ jmp(L_copy_32_bytes);
1551
1552 // Copy trailing qwords
1553 __ BIND(L_copy_8_bytes);
1554 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1555 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1556 __ decrement(qword_count);
1557 __ jcc(Assembler::notZero, L_copy_8_bytes);
1558
1559 restore_arg_regs();
1560 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1561 __ xorptr(rax, rax); // return 0
1562 __ leave(); // required for proper stackwalking of RuntimeStub frame
1563 __ ret(0);
1564
1565 // Copy in 32-bytes chunks
1566 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1567
1568 restore_arg_regs();
1569 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1570 __ xorptr(rax, rax); // return 0
1571 __ leave(); // required for proper stackwalking of RuntimeStub frame
1572 __ ret(0);
1573
1574 return start;
1575 }
1576
1577 // Arguments:
1578 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1579 // ignored
1580 // name - stub name string
1581 //
1582 // Inputs:
1583 // c_rarg0 - source array address
1584 // c_rarg1 - destination array address
1585 // c_rarg2 - element count, treated as ssize_t, can be zero
1586 //
1587 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1588 // let the hardware handle it. The two or four words within dwords
1589 // or qwords that span cache line boundaries will still be loaded
1590 // and stored atomically.
1591 //
1592 // Side Effects:
1593 // disjoint_short_copy_entry is set to the no-overlap entry point
1594 // used by generate_conjoint_short_copy().
1595 //
1596 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1597 __ align(CodeEntryAlignment);
1598 StubCodeMark mark(this, "StubRoutines", name);
1599 address start = __ pc();
1600
1601 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1602 const Register from = rdi; // source array address
1603 const Register to = rsi; // destination array address
1604 const Register count = rdx; // elements count
1605 const Register word_count = rcx;
1606 const Register qword_count = count;
1607 const Register end_from = from; // source array end address
1608 const Register end_to = to; // destination array end address
1609 // End pointers are inclusive, and if count is not zero they point
1610 // to the last unit copied: end_to[0] := end_from[0]
1611
1612 __ enter(); // required for proper stackwalking of RuntimeStub frame
1613 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1614
1615 if (entry != NULL) {
1616 *entry = __ pc();
1617 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1618 BLOCK_COMMENT("Entry:");
1619 }
1620
1621 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1622 // r9 and r10 may be used to save non-volatile registers
1623
1624 // 'from', 'to' and 'count' are now valid
1625 __ movptr(word_count, count);
1626 __ shrptr(count, 2); // count => qword_count
1627
1628 // Copy from low to high addresses. Use 'to' as scratch.
1629 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1630 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1631 __ negptr(qword_count);
1632 __ jmp(L_copy_32_bytes);
1633
1634 // Copy trailing qwords
1635 __ BIND(L_copy_8_bytes);
1636 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1637 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1638 __ increment(qword_count);
1639 __ jcc(Assembler::notZero, L_copy_8_bytes);
1640
1641 // Original 'dest' is trashed, so we can't use it as a
1642 // base register for a possible trailing word copy
1643
1644 // Check for and copy trailing dword
1645 __ BIND(L_copy_4_bytes);
1646 __ testl(word_count, 2);
1647 __ jccb(Assembler::zero, L_copy_2_bytes);
1648 __ movl(rax, Address(end_from, 8));
1649 __ movl(Address(end_to, 8), rax);
1650
1651 __ addptr(end_from, 4);
1652 __ addptr(end_to, 4);
1653
1654 // Check for and copy trailing word
1655 __ BIND(L_copy_2_bytes);
1656 __ testl(word_count, 1);
1657 __ jccb(Assembler::zero, L_exit);
1658 __ movw(rax, Address(end_from, 8));
1659 __ movw(Address(end_to, 8), rax);
1660
1661 __ BIND(L_exit);
1662 restore_arg_regs();
1663 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1664 __ xorptr(rax, rax); // return 0
1665 __ leave(); // required for proper stackwalking of RuntimeStub frame
1666 __ ret(0);
1667
1668 // Copy in 32-bytes chunks
1669 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1670 __ jmp(L_copy_4_bytes);
1671
1672 return start;
1673 }
1674
1675 address generate_fill(BasicType t, bool aligned, const char *name) {
1676 __ align(CodeEntryAlignment);
1677 StubCodeMark mark(this, "StubRoutines", name);
1678 address start = __ pc();
1679
1680 BLOCK_COMMENT("Entry:");
1681
1682 const Register to = c_rarg0; // source array address
1683 const Register value = c_rarg1; // value
1684 const Register count = c_rarg2; // elements count
1685
1686 __ enter(); // required for proper stackwalking of RuntimeStub frame
1687
1688 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1689
1690 __ leave(); // required for proper stackwalking of RuntimeStub frame
1691 __ ret(0);
1692 return start;
1693 }
1694
1695 // Arguments:
1696 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1697 // ignored
1698 // name - stub name string
1699 //
1700 // Inputs:
1701 // c_rarg0 - source array address
1702 // c_rarg1 - destination array address
1703 // c_rarg2 - element count, treated as ssize_t, can be zero
1704 //
1705 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1706 // let the hardware handle it. The two or four words within dwords
1707 // or qwords that span cache line boundaries will still be loaded
1708 // and stored atomically.
1709 //
1710 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1711 address *entry, const char *name) {
1712 __ align(CodeEntryAlignment);
1713 StubCodeMark mark(this, "StubRoutines", name);
1714 address start = __ pc();
1715
1716 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1717 const Register from = rdi; // source array address
1718 const Register to = rsi; // destination array address
1719 const Register count = rdx; // elements count
1720 const Register word_count = rcx;
1721 const Register qword_count = count;
1722
1723 __ enter(); // required for proper stackwalking of RuntimeStub frame
1724 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1725
1726 if (entry != NULL) {
1727 *entry = __ pc();
1728 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1729 BLOCK_COMMENT("Entry:");
1730 }
1731
1732 array_overlap_test(nooverlap_target, Address::times_2);
1733 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1734 // r9 and r10 may be used to save non-volatile registers
1735
1736 // 'from', 'to' and 'count' are now valid
1737 __ movptr(word_count, count);
1738 __ shrptr(count, 2); // count => qword_count
1739
1740 // Copy from high to low addresses. Use 'to' as scratch.
1741
1742 // Check for and copy trailing word
1743 __ testl(word_count, 1);
1744 __ jccb(Assembler::zero, L_copy_4_bytes);
1745 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1746 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1747
1748 // Check for and copy trailing dword
1749 __ BIND(L_copy_4_bytes);
1750 __ testl(word_count, 2);
1751 __ jcc(Assembler::zero, L_copy_32_bytes);
1752 __ movl(rax, Address(from, qword_count, Address::times_8));
1753 __ movl(Address(to, qword_count, Address::times_8), rax);
1754 __ jmp(L_copy_32_bytes);
1755
1756 // Copy trailing qwords
1757 __ BIND(L_copy_8_bytes);
1758 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1759 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1760 __ decrement(qword_count);
1761 __ jcc(Assembler::notZero, L_copy_8_bytes);
1762
1763 restore_arg_regs();
1764 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1765 __ xorptr(rax, rax); // return 0
1766 __ leave(); // required for proper stackwalking of RuntimeStub frame
1767 __ ret(0);
1768
1769 // Copy in 32-bytes chunks
1770 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1771
1772 restore_arg_regs();
1773 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1774 __ xorptr(rax, rax); // return 0
1775 __ leave(); // required for proper stackwalking of RuntimeStub frame
1776 __ ret(0);
1777
1778 return start;
1779 }
1780
1781 // Arguments:
1782 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1783 // ignored
1784 // is_oop - true => oop array, so generate store check code
1785 // name - stub name string
1786 //
1787 // Inputs:
1788 // c_rarg0 - source array address
1789 // c_rarg1 - destination array address
1790 // c_rarg2 - element count, treated as ssize_t, can be zero
1791 //
1792 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1793 // the hardware handle it. The two dwords within qwords that span
1794 // cache line boundaries will still be loaded and stored atomicly.
1795 //
1796 // Side Effects:
1797 // disjoint_int_copy_entry is set to the no-overlap entry point
1798 // used by generate_conjoint_int_oop_copy().
1799 //
1800 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1801 const char *name, bool dest_uninitialized = false) {
1802 __ align(CodeEntryAlignment);
1803 StubCodeMark mark(this, "StubRoutines", name);
1804 address start = __ pc();
1805
1806 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1807 const Register from = rdi; // source array address
1808 const Register to = rsi; // destination array address
1809 const Register count = rdx; // elements count
1810 const Register dword_count = rcx;
1811 const Register qword_count = count;
1812 const Register end_from = from; // source array end address
1813 const Register end_to = to; // destination array end address
1814 const Register saved_to = r11; // saved destination array address
1815 // End pointers are inclusive, and if count is not zero they point
1816 // to the last unit copied: end_to[0] := end_from[0]
1817
1818 __ enter(); // required for proper stackwalking of RuntimeStub frame
1819 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1820
1821 if (entry != NULL) {
1822 *entry = __ pc();
1823 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1824 BLOCK_COMMENT("Entry:");
1825 }
1826
1827 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1828 // r9 and r10 may be used to save non-volatile registers
1829 if (is_oop) {
1830 __ movq(saved_to, to);
1831 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1832 }
1833
1834 // 'from', 'to' and 'count' are now valid
1835 __ movptr(dword_count, count);
1836 __ shrptr(count, 1); // count => qword_count
1837
1838 // Copy from low to high addresses. Use 'to' as scratch.
1839 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1840 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1841 __ negptr(qword_count);
1842 __ jmp(L_copy_32_bytes);
1843
1844 // Copy trailing qwords
1845 __ BIND(L_copy_8_bytes);
1846 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1847 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1848 __ increment(qword_count);
1849 __ jcc(Assembler::notZero, L_copy_8_bytes);
1850
1851 // Check for and copy trailing dword
1852 __ BIND(L_copy_4_bytes);
1853 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1854 __ jccb(Assembler::zero, L_exit);
1855 __ movl(rax, Address(end_from, 8));
1856 __ movl(Address(end_to, 8), rax);
1857
1858 __ BIND(L_exit);
1859 if (is_oop) {
1860 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1861 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1862 }
1863 restore_arg_regs();
1864 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1865 __ xorptr(rax, rax); // return 0
1866 __ leave(); // required for proper stackwalking of RuntimeStub frame
1867 __ ret(0);
1868
1869 // Copy 32-bytes chunks
1870 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1871 __ jmp(L_copy_4_bytes);
1872
1873 return start;
1874 }
1875
1876 // Arguments:
1877 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1878 // ignored
1879 // is_oop - true => oop array, so generate store check code
1880 // name - stub name string
1881 //
1882 // Inputs:
1883 // c_rarg0 - source array address
1884 // c_rarg1 - destination array address
1885 // c_rarg2 - element count, treated as ssize_t, can be zero
1886 //
1887 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1888 // the hardware handle it. The two dwords within qwords that span
1889 // cache line boundaries will still be loaded and stored atomicly.
1890 //
1891 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1892 address *entry, const char *name,
1893 bool dest_uninitialized = false) {
1894 __ align(CodeEntryAlignment);
1895 StubCodeMark mark(this, "StubRoutines", name);
1896 address start = __ pc();
1897
1898 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1899 const Register from = rdi; // source array address
1900 const Register to = rsi; // destination array address
1901 const Register count = rdx; // elements count
1902 const Register dword_count = rcx;
1903 const Register qword_count = count;
1904
1905 __ enter(); // required for proper stackwalking of RuntimeStub frame
1906 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1907
1908 if (entry != NULL) {
1909 *entry = __ pc();
1910 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1911 BLOCK_COMMENT("Entry:");
1912 }
1913
1914 array_overlap_test(nooverlap_target, Address::times_4);
1915 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1916 // r9 and r10 may be used to save non-volatile registers
1917
1918 if (is_oop) {
1919 // no registers are destroyed by this call
1920 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1921 }
1922
1923 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1924 // 'from', 'to' and 'count' are now valid
1925 __ movptr(dword_count, count);
1926 __ shrptr(count, 1); // count => qword_count
1927
1928 // Copy from high to low addresses. Use 'to' as scratch.
1929
1930 // Check for and copy trailing dword
1931 __ testl(dword_count, 1);
1932 __ jcc(Assembler::zero, L_copy_32_bytes);
1933 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1934 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1935 __ jmp(L_copy_32_bytes);
1936
1937 // Copy trailing qwords
1938 __ BIND(L_copy_8_bytes);
1939 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1940 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1941 __ decrement(qword_count);
1942 __ jcc(Assembler::notZero, L_copy_8_bytes);
1943
1944 if (is_oop) {
1945 __ jmp(L_exit);
1946 }
1947 restore_arg_regs();
1948 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1949 __ xorptr(rax, rax); // return 0
1950 __ leave(); // required for proper stackwalking of RuntimeStub frame
1951 __ ret(0);
1952
1953 // Copy in 32-bytes chunks
1954 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1955
1956 __ bind(L_exit);
1957 if (is_oop) {
1958 Register end_to = rdx;
1959 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1960 gen_write_ref_array_post_barrier(to, end_to, rax);
1961 }
1962 restore_arg_regs();
1963 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1964 __ xorptr(rax, rax); // return 0
1965 __ leave(); // required for proper stackwalking of RuntimeStub frame
1966 __ ret(0);
1967
1968 return start;
1969 }
1970
1971 // Arguments:
1972 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1973 // ignored
1974 // is_oop - true => oop array, so generate store check code
1975 // name - stub name string
1976 //
1977 // Inputs:
1978 // c_rarg0 - source array address
1979 // c_rarg1 - destination array address
1980 // c_rarg2 - element count, treated as ssize_t, can be zero
1981 //
1982 // Side Effects:
1983 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1984 // no-overlap entry point used by generate_conjoint_long_oop_copy().
1985 //
1986 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
1987 const char *name, bool dest_uninitialized = false) {
1988 __ align(CodeEntryAlignment);
1989 StubCodeMark mark(this, "StubRoutines", name);
1990 address start = __ pc();
1991
1992 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
1993 const Register from = rdi; // source array address
1994 const Register to = rsi; // destination array address
1995 const Register qword_count = rdx; // elements count
1996 const Register end_from = from; // source array end address
1997 const Register end_to = rcx; // destination array end address
1998 const Register saved_to = to;
1999 // End pointers are inclusive, and if count is not zero they point
2000 // to the last unit copied: end_to[0] := end_from[0]
2001
2002 __ enter(); // required for proper stackwalking of RuntimeStub frame
2003 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2004 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2005
2006 if (entry != NULL) {
2007 *entry = __ pc();
2008 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2009 BLOCK_COMMENT("Entry:");
2010 }
2011
2012 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2013 // r9 and r10 may be used to save non-volatile registers
2014 // 'from', 'to' and 'qword_count' are now valid
2015 if (is_oop) {
2016 // no registers are destroyed by this call
2017 gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2018 }
2019
2020 // Copy from low to high addresses. Use 'to' as scratch.
2021 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2022 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2023 __ negptr(qword_count);
2024 __ jmp(L_copy_32_bytes);
2025
2026 // Copy trailing qwords
2027 __ BIND(L_copy_8_bytes);
2028 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2029 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2030 __ increment(qword_count);
2031 __ jcc(Assembler::notZero, L_copy_8_bytes);
2032
2033 if (is_oop) {
2034 __ jmp(L_exit);
2035 } else {
2036 restore_arg_regs();
2037 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2038 __ xorptr(rax, rax); // return 0
2039 __ leave(); // required for proper stackwalking of RuntimeStub frame
2040 __ ret(0);
2041 }
2042
2043 // Copy 64-byte chunks
2044 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2045
2046 if (is_oop) {
2047 __ BIND(L_exit);
2048 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
2049 }
2050 restore_arg_regs();
2051 if (is_oop) {
2052 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2053 } else {
2054 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2055 }
2056 __ xorptr(rax, rax); // return 0
2057 __ leave(); // required for proper stackwalking of RuntimeStub frame
2058 __ ret(0);
2059
2060 return start;
2061 }
2062
2063 // Arguments:
2064 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2065 // ignored
2066 // is_oop - true => oop array, so generate store check code
2067 // name - stub name string
2068 //
2069 // Inputs:
2070 // c_rarg0 - source array address
2071 // c_rarg1 - destination array address
2072 // c_rarg2 - element count, treated as ssize_t, can be zero
2073 //
2074 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2075 address nooverlap_target, address *entry,
2076 const char *name, bool dest_uninitialized = false) {
2077 __ align(CodeEntryAlignment);
2078 StubCodeMark mark(this, "StubRoutines", name);
2079 address start = __ pc();
2080
2081 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2082 const Register from = rdi; // source array address
2083 const Register to = rsi; // destination array address
2084 const Register qword_count = rdx; // elements count
2085 const Register saved_count = rcx;
2086
2087 __ enter(); // required for proper stackwalking of RuntimeStub frame
2088 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2089
2090 if (entry != NULL) {
2091 *entry = __ pc();
2092 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2093 BLOCK_COMMENT("Entry:");
2094 }
2095
2096 array_overlap_test(nooverlap_target, Address::times_8);
2097 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2098 // r9 and r10 may be used to save non-volatile registers
2099 // 'from', 'to' and 'qword_count' are now valid
2100 if (is_oop) {
2101 // Save to and count for store barrier
2102 __ movptr(saved_count, qword_count);
2103 // No registers are destroyed by this call
2104 gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2105 }
2106
2107 __ jmp(L_copy_32_bytes);
2108
2109 // Copy trailing qwords
2110 __ BIND(L_copy_8_bytes);
2111 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2112 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2113 __ decrement(qword_count);
2114 __ jcc(Assembler::notZero, L_copy_8_bytes);
2115
2116 if (is_oop) {
2117 __ jmp(L_exit);
2118 } else {
2119 restore_arg_regs();
2120 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2121 __ xorptr(rax, rax); // return 0
2122 __ leave(); // required for proper stackwalking of RuntimeStub frame
2123 __ ret(0);
2124 }
2125
2126 // Copy in 32-bytes chunks
2127 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2128
2129 if (is_oop) {
2130 __ BIND(L_exit);
2131 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2132 gen_write_ref_array_post_barrier(to, rcx, rax);
2133 }
2134 restore_arg_regs();
2135 if (is_oop) {
2136 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2137 } else {
2138 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2139 }
2140 __ xorptr(rax, rax); // return 0
2141 __ leave(); // required for proper stackwalking of RuntimeStub frame
2142 __ ret(0);
2143
2144 return start;
2145 }
2146
2147
2148 // Helper for generating a dynamic type check.
2149 // Smashes no registers.
2150 void generate_type_check(Register sub_klass,
2151 Register super_check_offset,
2152 Register super_klass,
2153 Label& L_success) {
2154 assert_different_registers(sub_klass, super_check_offset, super_klass);
2155
2156 BLOCK_COMMENT("type_check:");
2157
2158 Label L_miss;
2159
2160 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2161 super_check_offset);
2162 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2163
2164 // Fall through on failure!
2165 __ BIND(L_miss);
2166 }
2167
2168 //
2169 // Generate checkcasting array copy stub
2170 //
2171 // Input:
2172 // c_rarg0 - source array address
2173 // c_rarg1 - destination array address
2174 // c_rarg2 - element count, treated as ssize_t, can be zero
2175 // c_rarg3 - size_t ckoff (super_check_offset)
2176 // not Win64
2177 // c_rarg4 - oop ckval (super_klass)
2178 // Win64
2179 // rsp+40 - oop ckval (super_klass)
2180 //
2181 // Output:
2182 // rax == 0 - success
2183 // rax == -1^K - failure, where K is partial transfer count
2184 //
2185 address generate_checkcast_copy(const char *name, address *entry,
2186 bool dest_uninitialized = false) {
2187
2188 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2189
2190 // Input registers (after setup_arg_regs)
2191 const Register from = rdi; // source array address
2192 const Register to = rsi; // destination array address
2193 const Register length = rdx; // elements count
2194 const Register ckoff = rcx; // super_check_offset
2195 const Register ckval = r8; // super_klass
2196
2197 // Registers used as temps (r13, r14 are save-on-entry)
2198 const Register end_from = from; // source array end address
2199 const Register end_to = r13; // destination array end address
2200 const Register count = rdx; // -(count_remaining)
2201 const Register r14_length = r14; // saved copy of length
2202 // End pointers are inclusive, and if length is not zero they point
2203 // to the last unit copied: end_to[0] := end_from[0]
2204
2205 const Register rax_oop = rax; // actual oop copied
2206 const Register r11_klass = r11; // oop._klass
2207
2208 //---------------------------------------------------------------
2209 // Assembler stub will be used for this call to arraycopy
2210 // if the two arrays are subtypes of Object[] but the
2211 // destination array type is not equal to or a supertype
2212 // of the source type. Each element must be separately
2213 // checked.
2214
2215 __ align(CodeEntryAlignment);
2216 StubCodeMark mark(this, "StubRoutines", name);
2217 address start = __ pc();
2218
2219 __ enter(); // required for proper stackwalking of RuntimeStub frame
2220
2221 #ifdef ASSERT
2222 // caller guarantees that the arrays really are different
2223 // otherwise, we would have to make conjoint checks
2224 { Label L;
2225 array_overlap_test(L, TIMES_OOP);
2226 __ stop("checkcast_copy within a single array");
2227 __ bind(L);
2228 }
2229 #endif //ASSERT
2230
2231 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2232 // ckoff => rcx, ckval => r8
2233 // r9 and r10 may be used to save non-volatile registers
2234 #ifdef _WIN64
2235 // last argument (#4) is on stack on Win64
2236 __ movptr(ckval, Address(rsp, 6 * wordSize));
2237 #endif
2238
2239 // Caller of this entry point must set up the argument registers.
2240 if (entry != NULL) {
2241 *entry = __ pc();
2242 BLOCK_COMMENT("Entry:");
2243 }
2244
2245 // allocate spill slots for r13, r14
2246 enum {
2247 saved_r13_offset,
2248 saved_r14_offset,
2249 saved_rbp_offset
2250 };
2251 __ subptr(rsp, saved_rbp_offset * wordSize);
2252 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2253 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2254
2255 // check that int operands are properly extended to size_t
2256 assert_clean_int(length, rax);
2257 assert_clean_int(ckoff, rax);
2258
2259 #ifdef ASSERT
2260 BLOCK_COMMENT("assert consistent ckoff/ckval");
2261 // The ckoff and ckval must be mutually consistent,
2262 // even though caller generates both.
2263 { Label L;
2264 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2265 Klass::super_check_offset_offset_in_bytes());
2266 __ cmpl(ckoff, Address(ckval, sco_offset));
2267 __ jcc(Assembler::equal, L);
2268 __ stop("super_check_offset inconsistent");
2269 __ bind(L);
2270 }
2271 #endif //ASSERT
2272
2273 // Loop-invariant addresses. They are exclusive end pointers.
2274 Address end_from_addr(from, length, TIMES_OOP, 0);
2275 Address end_to_addr(to, length, TIMES_OOP, 0);
2276 // Loop-variant addresses. They assume post-incremented count < 0.
2277 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2278 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2279
2280 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2281
2282 // Copy from low to high addresses, indexed from the end of each array.
2283 __ lea(end_from, end_from_addr);
2284 __ lea(end_to, end_to_addr);
2285 __ movptr(r14_length, length); // save a copy of the length
2286 assert(length == count, ""); // else fix next line:
2287 __ negptr(count); // negate and test the length
2288 __ jcc(Assembler::notZero, L_load_element);
2289
2290 // Empty array: Nothing to do.
2291 __ xorptr(rax, rax); // return 0 on (trivial) success
2292 __ jmp(L_done);
2293
2294 // ======== begin loop ========
2295 // (Loop is rotated; its entry is L_load_element.)
2296 // Loop control:
2297 // for (count = -count; count != 0; count++)
2298 // Base pointers src, dst are biased by 8*(count-1),to last element.
2299 __ align(OptoLoopAlignment);
2300
2301 __ BIND(L_store_element);
2302 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2303 __ increment(count); // increment the count toward zero
2304 __ jcc(Assembler::zero, L_do_card_marks);
2305
2306 // ======== loop entry is here ========
2307 __ BIND(L_load_element);
2308 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2309 __ testptr(rax_oop, rax_oop);
2310 __ jcc(Assembler::zero, L_store_element);
2311
2312 __ load_klass(r11_klass, rax_oop);// query the object klass
2313 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2314 // ======== end loop ========
2315
2316 // It was a real error; we must depend on the caller to finish the job.
2317 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2318 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2319 // and report their number to the caller.
2320 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2321 __ lea(end_to, to_element_addr);
2322 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2323 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2324 __ movptr(rax, r14_length); // original oops
2325 __ addptr(rax, count); // K = (original - remaining) oops
2326 __ notptr(rax); // report (-1^K) to caller
2327 __ jmp(L_done);
2328
2329 // Come here on success only.
2330 __ BIND(L_do_card_marks);
2331 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2332 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2333 __ xorptr(rax, rax); // return 0 on success
2334
2335 // Common exit point (success or failure).
2336 __ BIND(L_done);
2337 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2338 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2339 restore_arg_regs();
2340 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2341 __ leave(); // required for proper stackwalking of RuntimeStub frame
2342 __ ret(0);
2343
2344 return start;
2345 }
2346
2347 //
2348 // Generate 'unsafe' array copy stub
2349 // Though just as safe as the other stubs, it takes an unscaled
2350 // size_t argument instead of an element count.
2351 //
2352 // Input:
2353 // c_rarg0 - source array address
2354 // c_rarg1 - destination array address
2355 // c_rarg2 - byte count, treated as ssize_t, can be zero
2356 //
2357 // Examines the alignment of the operands and dispatches
2358 // to a long, int, short, or byte copy loop.
2359 //
2360 address generate_unsafe_copy(const char *name,
2361 address byte_copy_entry, address short_copy_entry,
2362 address int_copy_entry, address long_copy_entry) {
2363
2364 Label L_long_aligned, L_int_aligned, L_short_aligned;
2365
2366 // Input registers (before setup_arg_regs)
2367 const Register from = c_rarg0; // source array address
2368 const Register to = c_rarg1; // destination array address
2369 const Register size = c_rarg2; // byte count (size_t)
2370
2371 // Register used as a temp
2372 const Register bits = rax; // test copy of low bits
2373
2374 __ align(CodeEntryAlignment);
2375 StubCodeMark mark(this, "StubRoutines", name);
2376 address start = __ pc();
2377
2378 __ enter(); // required for proper stackwalking of RuntimeStub frame
2379
2380 // bump this on entry, not on exit:
2381 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2382
2383 __ mov(bits, from);
2384 __ orptr(bits, to);
2385 __ orptr(bits, size);
2386
2387 __ testb(bits, BytesPerLong-1);
2388 __ jccb(Assembler::zero, L_long_aligned);
2389
2390 __ testb(bits, BytesPerInt-1);
2391 __ jccb(Assembler::zero, L_int_aligned);
2392
2393 __ testb(bits, BytesPerShort-1);
2394 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2395
2396 __ BIND(L_short_aligned);
2397 __ shrptr(size, LogBytesPerShort); // size => short_count
2398 __ jump(RuntimeAddress(short_copy_entry));
2399
2400 __ BIND(L_int_aligned);
2401 __ shrptr(size, LogBytesPerInt); // size => int_count
2402 __ jump(RuntimeAddress(int_copy_entry));
2403
2404 __ BIND(L_long_aligned);
2405 __ shrptr(size, LogBytesPerLong); // size => qword_count
2406 __ jump(RuntimeAddress(long_copy_entry));
2407
2408 return start;
2409 }
2410
2411 // Perform range checks on the proposed arraycopy.
2412 // Kills temp, but nothing else.
2413 // Also, clean the sign bits of src_pos and dst_pos.
2414 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2415 Register src_pos, // source position (c_rarg1)
2416 Register dst, // destination array oo (c_rarg2)
2417 Register dst_pos, // destination position (c_rarg3)
2418 Register length,
2419 Register temp,
2420 Label& L_failed) {
2421 BLOCK_COMMENT("arraycopy_range_checks:");
2422
2423 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2424 __ movl(temp, length);
2425 __ addl(temp, src_pos); // src_pos + length
2426 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2427 __ jcc(Assembler::above, L_failed);
2428
2429 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2430 __ movl(temp, length);
2431 __ addl(temp, dst_pos); // dst_pos + length
2432 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2433 __ jcc(Assembler::above, L_failed);
2434
2435 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2436 // Move with sign extension can be used since they are positive.
2437 __ movslq(src_pos, src_pos);
2438 __ movslq(dst_pos, dst_pos);
2439
2440 BLOCK_COMMENT("arraycopy_range_checks done");
2441 }
2442
2443 //
2444 // Generate generic array copy stubs
2445 //
2446 // Input:
2447 // c_rarg0 - src oop
2448 // c_rarg1 - src_pos (32-bits)
2449 // c_rarg2 - dst oop
2450 // c_rarg3 - dst_pos (32-bits)
2451 // not Win64
2452 // c_rarg4 - element count (32-bits)
2453 // Win64
2454 // rsp+40 - element count (32-bits)
2455 //
2456 // Output:
2457 // rax == 0 - success
2458 // rax == -1^K - failure, where K is partial transfer count
2459 //
2460 address generate_generic_copy(const char *name,
2461 address byte_copy_entry, address short_copy_entry,
2462 address int_copy_entry, address oop_copy_entry,
2463 address long_copy_entry, address checkcast_copy_entry) {
2464
2465 Label L_failed, L_failed_0, L_objArray;
2466 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2467
2468 // Input registers
2469 const Register src = c_rarg0; // source array oop
2470 const Register src_pos = c_rarg1; // source position
2471 const Register dst = c_rarg2; // destination array oop
2472 const Register dst_pos = c_rarg3; // destination position
2473 #ifndef _WIN64
2474 const Register length = c_rarg4;
2475 #else
2476 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2477 #endif
2478
2479 { int modulus = CodeEntryAlignment;
2480 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2481 int advance = target - (__ offset() % modulus);
2482 if (advance < 0) advance += modulus;
2483 if (advance > 0) __ nop(advance);
2484 }
2485 StubCodeMark mark(this, "StubRoutines", name);
2486
2487 // Short-hop target to L_failed. Makes for denser prologue code.
2488 __ BIND(L_failed_0);
2489 __ jmp(L_failed);
2490 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2491
2492 __ align(CodeEntryAlignment);
2493 address start = __ pc();
2494
2495 __ enter(); // required for proper stackwalking of RuntimeStub frame
2496
2497 // bump this on entry, not on exit:
2498 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2499
2500 //-----------------------------------------------------------------------
2501 // Assembler stub will be used for this call to arraycopy
2502 // if the following conditions are met:
2503 //
2504 // (1) src and dst must not be null.
2505 // (2) src_pos must not be negative.
2506 // (3) dst_pos must not be negative.
2507 // (4) length must not be negative.
2508 // (5) src klass and dst klass should be the same and not NULL.
2509 // (6) src and dst should be arrays.
2510 // (7) src_pos + length must not exceed length of src.
2511 // (8) dst_pos + length must not exceed length of dst.
2512 //
2513
2514 // if (src == NULL) return -1;
2515 __ testptr(src, src); // src oop
2516 size_t j1off = __ offset();
2517 __ jccb(Assembler::zero, L_failed_0);
2518
2519 // if (src_pos < 0) return -1;
2520 __ testl(src_pos, src_pos); // src_pos (32-bits)
2521 __ jccb(Assembler::negative, L_failed_0);
2522
2523 // if (dst == NULL) return -1;
2524 __ testptr(dst, dst); // dst oop
2525 __ jccb(Assembler::zero, L_failed_0);
2526
2527 // if (dst_pos < 0) return -1;
2528 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2529 size_t j4off = __ offset();
2530 __ jccb(Assembler::negative, L_failed_0);
2531
2532 // The first four tests are very dense code,
2533 // but not quite dense enough to put four
2534 // jumps in a 16-byte instruction fetch buffer.
2535 // That's good, because some branch predicters
2536 // do not like jumps so close together.
2537 // Make sure of this.
2538 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2539
2540 // registers used as temp
2541 const Register r11_length = r11; // elements count to copy
2542 const Register r10_src_klass = r10; // array klass
2543
2544 // if (length < 0) return -1;
2545 __ movl(r11_length, length); // length (elements count, 32-bits value)
2546 __ testl(r11_length, r11_length);
2547 __ jccb(Assembler::negative, L_failed_0);
2548
2549 __ load_klass(r10_src_klass, src);
2550 #ifdef ASSERT
2551 // assert(src->klass() != NULL);
2552 {
2553 BLOCK_COMMENT("assert klasses not null {");
2554 Label L1, L2;
2555 __ testptr(r10_src_klass, r10_src_klass);
2556 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2557 __ bind(L1);
2558 __ stop("broken null klass");
2559 __ bind(L2);
2560 __ load_klass(rax, dst);
2561 __ cmpq(rax, 0);
2562 __ jcc(Assembler::equal, L1); // this would be broken also
2563 BLOCK_COMMENT("} assert klasses not null done");
2564 }
2565 #endif
2566
2567 // Load layout helper (32-bits)
2568 //
2569 // |array_tag| | header_size | element_type | |log2_element_size|
2570 // 32 30 24 16 8 2 0
2571 //
2572 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2573 //
2574
2575 const int lh_offset = klassOopDesc::header_size() * HeapWordSize +
2576 Klass::layout_helper_offset_in_bytes();
2577
2578 // Handle objArrays completely differently...
2579 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2580 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2581 __ jcc(Assembler::equal, L_objArray);
2582
2583 // if (src->klass() != dst->klass()) return -1;
2584 __ load_klass(rax, dst);
2585 __ cmpq(r10_src_klass, rax);
2586 __ jcc(Assembler::notEqual, L_failed);
2587
2588 const Register rax_lh = rax; // layout helper
2589 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2590
2591 // if (!src->is_Array()) return -1;
2592 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2593 __ jcc(Assembler::greaterEqual, L_failed);
2594
2595 // At this point, it is known to be a typeArray (array_tag 0x3).
2596 #ifdef ASSERT
2597 {
2598 BLOCK_COMMENT("assert primitive array {");
2599 Label L;
2600 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2601 __ jcc(Assembler::greaterEqual, L);
2602 __ stop("must be a primitive array");
2603 __ bind(L);
2604 BLOCK_COMMENT("} assert primitive array done");
2605 }
2606 #endif
2607
2608 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2609 r10, L_failed);
2610
2611 // typeArrayKlass
2612 //
2613 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2614 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2615 //
2616
2617 const Register r10_offset = r10; // array offset
2618 const Register rax_elsize = rax_lh; // element size
2619
2620 __ movl(r10_offset, rax_lh);
2621 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2622 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2623 __ addptr(src, r10_offset); // src array offset
2624 __ addptr(dst, r10_offset); // dst array offset
2625 BLOCK_COMMENT("choose copy loop based on element size");
2626 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2627
2628 // next registers should be set before the jump to corresponding stub
2629 const Register from = c_rarg0; // source array address
2630 const Register to = c_rarg1; // destination array address
2631 const Register count = c_rarg2; // elements count
2632
2633 // 'from', 'to', 'count' registers should be set in such order
2634 // since they are the same as 'src', 'src_pos', 'dst'.
2635
2636 __ BIND(L_copy_bytes);
2637 __ cmpl(rax_elsize, 0);
2638 __ jccb(Assembler::notEqual, L_copy_shorts);
2639 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2640 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2641 __ movl2ptr(count, r11_length); // length
2642 __ jump(RuntimeAddress(byte_copy_entry));
2643
2644 __ BIND(L_copy_shorts);
2645 __ cmpl(rax_elsize, LogBytesPerShort);
2646 __ jccb(Assembler::notEqual, L_copy_ints);
2647 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2648 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2649 __ movl2ptr(count, r11_length); // length
2650 __ jump(RuntimeAddress(short_copy_entry));
2651
2652 __ BIND(L_copy_ints);
2653 __ cmpl(rax_elsize, LogBytesPerInt);
2654 __ jccb(Assembler::notEqual, L_copy_longs);
2655 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2656 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2657 __ movl2ptr(count, r11_length); // length
2658 __ jump(RuntimeAddress(int_copy_entry));
2659
2660 __ BIND(L_copy_longs);
2661 #ifdef ASSERT
2662 {
2663 BLOCK_COMMENT("assert long copy {");
2664 Label L;
2665 __ cmpl(rax_elsize, LogBytesPerLong);
2666 __ jcc(Assembler::equal, L);
2667 __ stop("must be long copy, but elsize is wrong");
2668 __ bind(L);
2669 BLOCK_COMMENT("} assert long copy done");
2670 }
2671 #endif
2672 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2673 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2674 __ movl2ptr(count, r11_length); // length
2675 __ jump(RuntimeAddress(long_copy_entry));
2676
2677 // objArrayKlass
2678 __ BIND(L_objArray);
2679 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2680
2681 Label L_plain_copy, L_checkcast_copy;
2682 // test array classes for subtyping
2683 __ load_klass(rax, dst);
2684 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2685 __ jcc(Assembler::notEqual, L_checkcast_copy);
2686
2687 // Identically typed arrays can be copied without element-wise checks.
2688 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2689 r10, L_failed);
2690
2691 __ lea(from, Address(src, src_pos, TIMES_OOP,
2692 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2693 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2694 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2695 __ movl2ptr(count, r11_length); // length
2696 __ BIND(L_plain_copy);
2697 __ jump(RuntimeAddress(oop_copy_entry));
2698
2699 __ BIND(L_checkcast_copy);
2700 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2701 {
2702 // Before looking at dst.length, make sure dst is also an objArray.
2703 __ cmpl(Address(rax, lh_offset), objArray_lh);
2704 __ jcc(Assembler::notEqual, L_failed);
2705
2706 // It is safe to examine both src.length and dst.length.
2707 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2708 rax, L_failed);
2709
2710 const Register r11_dst_klass = r11;
2711 __ load_klass(r11_dst_klass, dst); // reload
2712
2713 // Marshal the base address arguments now, freeing registers.
2714 __ lea(from, Address(src, src_pos, TIMES_OOP,
2715 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2716 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2717 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2718 __ movl(count, length); // length (reloaded)
2719 Register sco_temp = c_rarg3; // this register is free now
2720 assert_different_registers(from, to, count, sco_temp,
2721 r11_dst_klass, r10_src_klass);
2722 assert_clean_int(count, sco_temp);
2723
2724 // Generate the type check.
2725 const int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
2726 Klass::super_check_offset_offset_in_bytes());
2727 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2728 assert_clean_int(sco_temp, rax);
2729 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2730
2731 // Fetch destination element klass from the objArrayKlass header.
2732 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
2733 objArrayKlass::element_klass_offset_in_bytes());
2734 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2735 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2736 assert_clean_int(sco_temp, rax);
2737
2738 // the checkcast_copy loop needs two extra arguments:
2739 assert(c_rarg3 == sco_temp, "#3 already in place");
2740 // Set up arguments for checkcast_copy_entry.
2741 setup_arg_regs(4);
2742 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2743 __ jump(RuntimeAddress(checkcast_copy_entry));
2744 }
2745
2746 __ BIND(L_failed);
2747 __ xorptr(rax, rax);
2748 __ notptr(rax); // return -1
2749 __ leave(); // required for proper stackwalking of RuntimeStub frame
2750 __ ret(0);
2751
2752 return start;
2753 }
2754
2755 void generate_arraycopy_stubs() {
2756 address entry;
2757 address entry_jbyte_arraycopy;
2758 address entry_jshort_arraycopy;
2759 address entry_jint_arraycopy;
2760 address entry_oop_arraycopy;
2761 address entry_jlong_arraycopy;
2762 address entry_checkcast_arraycopy;
2763
2764 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2765 "jbyte_disjoint_arraycopy");
2766 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2767 "jbyte_arraycopy");
2768
2769 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2770 "jshort_disjoint_arraycopy");
2771 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2772 "jshort_arraycopy");
2773
2774 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2775 "jint_disjoint_arraycopy");
2776 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2777 &entry_jint_arraycopy, "jint_arraycopy");
2778
2779 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2780 "jlong_disjoint_arraycopy");
2781 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2782 &entry_jlong_arraycopy, "jlong_arraycopy");
2783
2784
2785 if (UseCompressedOops) {
2786 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2787 "oop_disjoint_arraycopy");
2788 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2789 &entry_oop_arraycopy, "oop_arraycopy");
2790 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2791 "oop_disjoint_arraycopy_uninit",
2792 /*dest_uninitialized*/true);
2793 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2794 NULL, "oop_arraycopy_uninit",
2795 /*dest_uninitialized*/true);
2796 } else {
2797 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2798 "oop_disjoint_arraycopy");
2799 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2800 &entry_oop_arraycopy, "oop_arraycopy");
2801 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2802 "oop_disjoint_arraycopy_uninit",
2803 /*dest_uninitialized*/true);
2804 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2805 NULL, "oop_arraycopy_uninit",
2806 /*dest_uninitialized*/true);
2807 }
2808
2809 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2810 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2811 /*dest_uninitialized*/true);
2812
2813 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2814 entry_jbyte_arraycopy,
2815 entry_jshort_arraycopy,
2816 entry_jint_arraycopy,
2817 entry_jlong_arraycopy);
2818 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2819 entry_jbyte_arraycopy,
2820 entry_jshort_arraycopy,
2821 entry_jint_arraycopy,
2822 entry_oop_arraycopy,
2823 entry_jlong_arraycopy,
2824 entry_checkcast_arraycopy);
2825
2826 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2827 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2828 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2829 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2830 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2831 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2832
2833 // We don't generate specialized code for HeapWord-aligned source
2834 // arrays, so just use the code we've already generated
2835 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2836 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2837
2838 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2839 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2840
2841 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2842 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2843
2844 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2845 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2846
2847 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2848 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2849
2850 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2851 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2852 }
2853
2854 void generate_math_stubs() {
2855 {
2856 StubCodeMark mark(this, "StubRoutines", "log");
2857 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2858
2859 __ subq(rsp, 8);
2860 __ movdbl(Address(rsp, 0), xmm0);
2861 __ fld_d(Address(rsp, 0));
2862 __ flog();
2863 __ fstp_d(Address(rsp, 0));
2864 __ movdbl(xmm0, Address(rsp, 0));
2865 __ addq(rsp, 8);
2866 __ ret(0);
2867 }
2868 {
2869 StubCodeMark mark(this, "StubRoutines", "log10");
2870 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2871
2872 __ subq(rsp, 8);
2873 __ movdbl(Address(rsp, 0), xmm0);
2874 __ fld_d(Address(rsp, 0));
2875 __ flog10();
2876 __ fstp_d(Address(rsp, 0));
2877 __ movdbl(xmm0, Address(rsp, 0));
2878 __ addq(rsp, 8);
2879 __ ret(0);
2880 }
2881 {
2882 StubCodeMark mark(this, "StubRoutines", "sin");
2883 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2884
2885 __ subq(rsp, 8);
2886 __ movdbl(Address(rsp, 0), xmm0);
2887 __ fld_d(Address(rsp, 0));
2888 __ trigfunc('s');
2889 __ fstp_d(Address(rsp, 0));
2890 __ movdbl(xmm0, Address(rsp, 0));
2891 __ addq(rsp, 8);
2892 __ ret(0);
2893 }
2894 {
2895 StubCodeMark mark(this, "StubRoutines", "cos");
2896 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2897
2898 __ subq(rsp, 8);
2899 __ movdbl(Address(rsp, 0), xmm0);
2900 __ fld_d(Address(rsp, 0));
2901 __ trigfunc('c');
2902 __ fstp_d(Address(rsp, 0));
2903 __ movdbl(xmm0, Address(rsp, 0));
2904 __ addq(rsp, 8);
2905 __ ret(0);
2906 }
2907 {
2908 StubCodeMark mark(this, "StubRoutines", "tan");
2909 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2910
2911 __ subq(rsp, 8);
2912 __ movdbl(Address(rsp, 0), xmm0);
2913 __ fld_d(Address(rsp, 0));
2914 __ trigfunc('t');
2915 __ fstp_d(Address(rsp, 0));
2916 __ movdbl(xmm0, Address(rsp, 0));
2917 __ addq(rsp, 8);
2918 __ ret(0);
2919 }
2920
2921 // The intrinsic version of these seem to return the same value as
2922 // the strict version.
2923 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2924 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2925 }
2926
2927 #undef __
2928 #define __ masm->
2929
2930 // Continuation point for throwing of implicit exceptions that are
2931 // not handled in the current activation. Fabricates an exception
2932 // oop and initiates normal exception dispatching in this
2933 // frame. Since we need to preserve callee-saved values (currently
2934 // only for C2, but done for C1 as well) we need a callee-saved oop
2935 // map and therefore have to make these stubs into RuntimeStubs
2936 // rather than BufferBlobs. If the compiler needs all registers to
2937 // be preserved between the fault point and the exception handler
2938 // then it must assume responsibility for that in
2939 // AbstractCompiler::continuation_for_implicit_null_exception or
2940 // continuation_for_implicit_division_by_zero_exception. All other
2941 // implicit exceptions (e.g., NullPointerException or
2942 // AbstractMethodError on entry) are either at call sites or
2943 // otherwise assume that stack unwinding will be initiated, so
2944 // caller saved registers were assumed volatile in the compiler.
2945 address generate_throw_exception(const char* name,
2946 address runtime_entry,
2947 Register arg1 = noreg,
2948 Register arg2 = noreg) {
2949 // Information about frame layout at time of blocking runtime call.
2950 // Note that we only have to preserve callee-saved registers since
2951 // the compilers are responsible for supplying a continuation point
2952 // if they expect all registers to be preserved.
2953 enum layout {
2954 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2955 rbp_off2,
2956 return_off,
2957 return_off2,
2958 framesize // inclusive of return address
2959 };
2960
2961 int insts_size = 512;
2962 int locs_size = 64;
2963
2964 CodeBuffer code(name, insts_size, locs_size);
2965 OopMapSet* oop_maps = new OopMapSet();
2966 MacroAssembler* masm = new MacroAssembler(&code);
2967
2968 address start = __ pc();
2969
2970 // This is an inlined and slightly modified version of call_VM
2971 // which has the ability to fetch the return PC out of
2972 // thread-local storage and also sets up last_Java_sp slightly
2973 // differently than the real call_VM
2974
2975 __ enter(); // required for proper stackwalking of RuntimeStub frame
2976
2977 assert(is_even(framesize/2), "sp not 16-byte aligned");
2978
2979 // return address and rbp are already in place
2980 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2981
2982 int frame_complete = __ pc() - start;
2983
2984 // Set up last_Java_sp and last_Java_fp
2985 __ set_last_Java_frame(rsp, rbp, NULL);
2986
2987 // Call runtime
2988 if (arg1 != noreg) {
2989 assert(arg2 != c_rarg1, "clobbered");
2990 __ movptr(c_rarg1, arg1);
2991 }
2992 if (arg2 != noreg) {
2993 __ movptr(c_rarg2, arg2);
2994 }
2995 __ movptr(c_rarg0, r15_thread);
2996 BLOCK_COMMENT("call runtime_entry");
2997 __ call(RuntimeAddress(runtime_entry));
2998
2999 // Generate oop map
3000 OopMap* map = new OopMap(framesize, 0);
3001
3002 oop_maps->add_gc_map(__ pc() - start, map);
3003
3004 __ reset_last_Java_frame(true, false);
3005
3006 __ leave(); // required for proper stackwalking of RuntimeStub frame
3007
3008 // check for pending exceptions
3009 #ifdef ASSERT
3010 Label L;
3011 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3012 (int32_t) NULL_WORD);
3013 __ jcc(Assembler::notEqual, L);
3014 __ should_not_reach_here();
3015 __ bind(L);
3016 #endif // ASSERT
3017 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3018
3019
3020 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3021 RuntimeStub* stub =
3022 RuntimeStub::new_runtime_stub(name,
3023 &code,
3024 frame_complete,
3025 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3026 oop_maps, false);
3027 return stub->entry_point();
3028 }
3029
3030 // Initialization
3031 void generate_initial() {
3032 // Generates all stubs and initializes the entry points
3033
3034 // This platform-specific stub is needed by generate_call_stub()
3035 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
3036
3037 // entry points that exist in all platforms Note: This is code
3038 // that could be shared among different platforms - however the
3039 // benefit seems to be smaller than the disadvantage of having a
3040 // much more complicated generator structure. See also comment in
3041 // stubRoutines.hpp.
3042
3043 StubRoutines::_forward_exception_entry = generate_forward_exception();
3044
3045 StubRoutines::_call_stub_entry =
3046 generate_call_stub(StubRoutines::_call_stub_return_address);
3047
3048 // is referenced by megamorphic call
3049 StubRoutines::_catch_exception_entry = generate_catch_exception();
3050
3051 // atomic calls
3052 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3053 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
3054 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
3055 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3056 StubRoutines::_atomic_add_entry = generate_atomic_add();
3057 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
3058 StubRoutines::_fence_entry = generate_orderaccess_fence();
3059
3060 StubRoutines::_handler_for_unsafe_access_entry =
3061 generate_handler_for_unsafe_access();
3062
3063 // platform dependent
3064 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3065
3066 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3067
3068 // Build this early so it's available for the interpreter. Stub
3069 // expects the required and actual types as register arguments in
3070 // j_rarg0 and j_rarg1 respectively.
3071 StubRoutines::_throw_WrongMethodTypeException_entry =
3072 generate_throw_exception("WrongMethodTypeException throw_exception",
3073 CAST_FROM_FN_PTR(address, SharedRuntime::throw_WrongMethodTypeException),
3074 rax, rcx);
3075 }
3076
3077 void generate_all() {
3078 // Generates all stubs and initializes the entry points
3079
3080 // These entry points require SharedInfo::stack0 to be set up in
3081 // non-core builds and need to be relocatable, so they each
3082 // fabricate a RuntimeStub internally.
3083 StubRoutines::_throw_AbstractMethodError_entry =
3084 generate_throw_exception("AbstractMethodError throw_exception",
3085 CAST_FROM_FN_PTR(address,
3086 SharedRuntime::
3087 throw_AbstractMethodError));
3088
3089 StubRoutines::_throw_IncompatibleClassChangeError_entry =
3090 generate_throw_exception("IncompatibleClassChangeError throw_exception",
3091 CAST_FROM_FN_PTR(address,
3092 SharedRuntime::
3093 throw_IncompatibleClassChangeError));
3094
3095 StubRoutines::_throw_NullPointerException_at_call_entry =
3096 generate_throw_exception("NullPointerException at call throw_exception",
3097 CAST_FROM_FN_PTR(address,
3098 SharedRuntime::
3099 throw_NullPointerException_at_call));
3100
3101 StubRoutines::_throw_StackOverflowError_entry =
3102 generate_throw_exception("StackOverflowError throw_exception",
3103 CAST_FROM_FN_PTR(address,
3104 SharedRuntime::
3105 throw_StackOverflowError));
3106
3107 // entry points that are platform specific
3108 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3109 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3110 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3111 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3112
3113 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3114 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3115 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3116 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3117
3118 // support for verify_oop (must happen after universe_init)
3119 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3120
3121 // arraycopy stubs used by compilers
3122 generate_arraycopy_stubs();
3123
3124 generate_math_stubs();
3125 }
3126
3127 public:
3128 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3129 if (all) {
3130 generate_all();
3131 } else {
3132 generate_initial();
3133 }
3134 }
3135 }; // end class declaration
3136
3137 void StubGenerator_generate(CodeBuffer* code, bool all) {
3138 StubGenerator g(code, all);
3139 }