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 // Support for intptr_t get_previous_sp()
714 //
715 // This routine is used to find the previous stack pointer for the
716 // caller.
717 address generate_get_previous_sp() {
718 StubCodeMark mark(this, "StubRoutines", "get_previous_sp");
719 address start = __ pc();
720
721 __ movptr(rax, rsp);
722 __ addptr(rax, 8); // return address is at the top of the stack.
723 __ ret(0);
724
725 return start;
726 }
727
728 //----------------------------------------------------------------------------------------------------
729 // Support for void verify_mxcsr()
730 //
731 // This routine is used with -Xcheck:jni to verify that native
732 // JNI code does not return to Java code without restoring the
733 // MXCSR register to our expected state.
734
735 address generate_verify_mxcsr() {
736 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
737 address start = __ pc();
738
739 const Address mxcsr_save(rsp, 0);
740
741 if (CheckJNICalls) {
742 Label ok_ret;
743 __ push(rax);
744 __ subptr(rsp, wordSize); // allocate a temp location
745 __ stmxcsr(mxcsr_save);
746 __ movl(rax, mxcsr_save);
747 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
748 __ cmpl(rax, *(int *)(StubRoutines::x86::mxcsr_std()));
749 __ jcc(Assembler::equal, ok_ret);
750
751 __ warn("MXCSR changed by native JNI code, use -XX:+RestoreMXCSROnJNICall");
752
753 __ ldmxcsr(ExternalAddress(StubRoutines::x86::mxcsr_std()));
754
755 __ bind(ok_ret);
756 __ addptr(rsp, wordSize);
757 __ pop(rax);
758 }
759
760 __ ret(0);
761
762 return start;
763 }
764
765 address generate_f2i_fixup() {
766 StubCodeMark mark(this, "StubRoutines", "f2i_fixup");
767 Address inout(rsp, 5 * wordSize); // return address + 4 saves
768
769 address start = __ pc();
770
771 Label L;
772
773 __ push(rax);
774 __ push(c_rarg3);
775 __ push(c_rarg2);
776 __ push(c_rarg1);
777
778 __ movl(rax, 0x7f800000);
779 __ xorl(c_rarg3, c_rarg3);
780 __ movl(c_rarg2, inout);
781 __ movl(c_rarg1, c_rarg2);
782 __ andl(c_rarg1, 0x7fffffff);
783 __ cmpl(rax, c_rarg1); // NaN? -> 0
784 __ jcc(Assembler::negative, L);
785 __ testl(c_rarg2, c_rarg2); // signed ? min_jint : max_jint
786 __ movl(c_rarg3, 0x80000000);
787 __ movl(rax, 0x7fffffff);
788 __ cmovl(Assembler::positive, c_rarg3, rax);
789
790 __ bind(L);
791 __ movptr(inout, c_rarg3);
792
793 __ pop(c_rarg1);
794 __ pop(c_rarg2);
795 __ pop(c_rarg3);
796 __ pop(rax);
797
798 __ ret(0);
799
800 return start;
801 }
802
803 address generate_f2l_fixup() {
804 StubCodeMark mark(this, "StubRoutines", "f2l_fixup");
805 Address inout(rsp, 5 * wordSize); // return address + 4 saves
806 address start = __ pc();
807
808 Label L;
809
810 __ push(rax);
811 __ push(c_rarg3);
812 __ push(c_rarg2);
813 __ push(c_rarg1);
814
815 __ movl(rax, 0x7f800000);
816 __ xorl(c_rarg3, c_rarg3);
817 __ movl(c_rarg2, inout);
818 __ movl(c_rarg1, c_rarg2);
819 __ andl(c_rarg1, 0x7fffffff);
820 __ cmpl(rax, c_rarg1); // NaN? -> 0
821 __ jcc(Assembler::negative, L);
822 __ testl(c_rarg2, c_rarg2); // signed ? min_jlong : max_jlong
823 __ mov64(c_rarg3, 0x8000000000000000);
824 __ mov64(rax, 0x7fffffffffffffff);
825 __ cmov(Assembler::positive, c_rarg3, rax);
826
827 __ bind(L);
828 __ movptr(inout, c_rarg3);
829
830 __ pop(c_rarg1);
831 __ pop(c_rarg2);
832 __ pop(c_rarg3);
833 __ pop(rax);
834
835 __ ret(0);
836
837 return start;
838 }
839
840 address generate_d2i_fixup() {
841 StubCodeMark mark(this, "StubRoutines", "d2i_fixup");
842 Address inout(rsp, 6 * wordSize); // return address + 5 saves
843
844 address start = __ pc();
845
846 Label L;
847
848 __ push(rax);
849 __ push(c_rarg3);
850 __ push(c_rarg2);
851 __ push(c_rarg1);
852 __ push(c_rarg0);
853
854 __ movl(rax, 0x7ff00000);
855 __ movq(c_rarg2, inout);
856 __ movl(c_rarg3, c_rarg2);
857 __ mov(c_rarg1, c_rarg2);
858 __ mov(c_rarg0, c_rarg2);
859 __ negl(c_rarg3);
860 __ shrptr(c_rarg1, 0x20);
861 __ orl(c_rarg3, c_rarg2);
862 __ andl(c_rarg1, 0x7fffffff);
863 __ xorl(c_rarg2, c_rarg2);
864 __ shrl(c_rarg3, 0x1f);
865 __ orl(c_rarg1, c_rarg3);
866 __ cmpl(rax, c_rarg1);
867 __ jcc(Assembler::negative, L); // NaN -> 0
868 __ testptr(c_rarg0, c_rarg0); // signed ? min_jint : max_jint
869 __ movl(c_rarg2, 0x80000000);
870 __ movl(rax, 0x7fffffff);
871 __ cmov(Assembler::positive, c_rarg2, rax);
872
873 __ bind(L);
874 __ movptr(inout, c_rarg2);
875
876 __ pop(c_rarg0);
877 __ pop(c_rarg1);
878 __ pop(c_rarg2);
879 __ pop(c_rarg3);
880 __ pop(rax);
881
882 __ ret(0);
883
884 return start;
885 }
886
887 address generate_d2l_fixup() {
888 StubCodeMark mark(this, "StubRoutines", "d2l_fixup");
889 Address inout(rsp, 6 * wordSize); // return address + 5 saves
890
891 address start = __ pc();
892
893 Label L;
894
895 __ push(rax);
896 __ push(c_rarg3);
897 __ push(c_rarg2);
898 __ push(c_rarg1);
899 __ push(c_rarg0);
900
901 __ movl(rax, 0x7ff00000);
902 __ movq(c_rarg2, inout);
903 __ movl(c_rarg3, c_rarg2);
904 __ mov(c_rarg1, c_rarg2);
905 __ mov(c_rarg0, c_rarg2);
906 __ negl(c_rarg3);
907 __ shrptr(c_rarg1, 0x20);
908 __ orl(c_rarg3, c_rarg2);
909 __ andl(c_rarg1, 0x7fffffff);
910 __ xorl(c_rarg2, c_rarg2);
911 __ shrl(c_rarg3, 0x1f);
912 __ orl(c_rarg1, c_rarg3);
913 __ cmpl(rax, c_rarg1);
914 __ jcc(Assembler::negative, L); // NaN -> 0
915 __ testq(c_rarg0, c_rarg0); // signed ? min_jlong : max_jlong
916 __ mov64(c_rarg2, 0x8000000000000000);
917 __ mov64(rax, 0x7fffffffffffffff);
918 __ cmovq(Assembler::positive, c_rarg2, rax);
919
920 __ bind(L);
921 __ movq(inout, c_rarg2);
922
923 __ pop(c_rarg0);
924 __ pop(c_rarg1);
925 __ pop(c_rarg2);
926 __ pop(c_rarg3);
927 __ pop(rax);
928
929 __ ret(0);
930
931 return start;
932 }
933
934 address generate_fp_mask(const char *stub_name, int64_t mask) {
935 __ align(CodeEntryAlignment);
936 StubCodeMark mark(this, "StubRoutines", stub_name);
937 address start = __ pc();
938
939 __ emit_data64( mask, relocInfo::none );
940 __ emit_data64( mask, relocInfo::none );
941
942 return start;
943 }
944
945 // The following routine generates a subroutine to throw an
946 // asynchronous UnknownError when an unsafe access gets a fault that
947 // could not be reasonably prevented by the programmer. (Example:
948 // SIGBUS/OBJERR.)
949 address generate_handler_for_unsafe_access() {
950 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
951 address start = __ pc();
952
953 __ push(0); // hole for return address-to-be
954 __ pusha(); // push registers
955 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
956
957 // FIXME: this probably needs alignment logic
958
959 __ subptr(rsp, frame::arg_reg_save_area_bytes);
960 BLOCK_COMMENT("call handle_unsafe_access");
961 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
962 __ addptr(rsp, frame::arg_reg_save_area_bytes);
963
964 __ movptr(next_pc, rax); // stuff next address
965 __ popa();
966 __ ret(0); // jump to next address
967
968 return start;
969 }
970
971 // Non-destructive plausibility checks for oops
972 //
973 // Arguments:
974 // all args on stack!
975 //
976 // Stack after saving c_rarg3:
977 // [tos + 0]: saved c_rarg3
978 // [tos + 1]: saved c_rarg2
979 // [tos + 2]: saved r12 (several TemplateTable methods use it)
980 // [tos + 3]: saved flags
981 // [tos + 4]: return address
982 // * [tos + 5]: error message (char*)
983 // * [tos + 6]: object to verify (oop)
984 // * [tos + 7]: saved rax - saved by caller and bashed
985 // * [tos + 8]: saved r10 (rscratch1) - saved by caller
986 // * = popped on exit
987 address generate_verify_oop() {
988 StubCodeMark mark(this, "StubRoutines", "verify_oop");
989 address start = __ pc();
990
991 Label exit, error;
992
993 __ pushf();
994 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
995
996 __ push(r12);
997
998 // save c_rarg2 and c_rarg3
999 __ push(c_rarg2);
1000 __ push(c_rarg3);
1001
1002 enum {
1003 // After previous pushes.
1004 oop_to_verify = 6 * wordSize,
1005 saved_rax = 7 * wordSize,
1006 saved_r10 = 8 * wordSize,
1007
1008 // Before the call to MacroAssembler::debug(), see below.
1009 return_addr = 16 * wordSize,
1010 error_msg = 17 * wordSize
1011 };
1012
1013 // get object
1014 __ movptr(rax, Address(rsp, oop_to_verify));
1015
1016 // make sure object is 'reasonable'
1017 __ testptr(rax, rax);
1018 __ jcc(Assembler::zero, exit); // if obj is NULL it is OK
1019 // Check if the oop is in the right area of memory
1020 __ movptr(c_rarg2, rax);
1021 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_mask());
1022 __ andptr(c_rarg2, c_rarg3);
1023 __ movptr(c_rarg3, (intptr_t) Universe::verify_oop_bits());
1024 __ cmpptr(c_rarg2, c_rarg3);
1025 __ jcc(Assembler::notZero, error);
1026
1027 // set r12 to heapbase for load_klass()
1028 __ reinit_heapbase();
1029
1030 // make sure klass is 'reasonable'
1031 __ load_klass(rax, rax); // get klass
1032 __ testptr(rax, rax);
1033 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
1034 // Check if the klass is in the right area of memory
1035 __ mov(c_rarg2, rax);
1036 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1037 __ andptr(c_rarg2, c_rarg3);
1038 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1039 __ cmpptr(c_rarg2, c_rarg3);
1040 __ jcc(Assembler::notZero, error);
1041
1042 // make sure klass' klass is 'reasonable'
1043 __ load_klass(rax, rax);
1044 __ testptr(rax, rax);
1045 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
1046 // Check if the klass' klass is in the right area of memory
1047 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_mask());
1048 __ andptr(rax, c_rarg3);
1049 __ movptr(c_rarg3, (intptr_t) Universe::verify_klass_bits());
1050 __ cmpptr(rax, c_rarg3);
1051 __ jcc(Assembler::notZero, error);
1052
1053 // return if everything seems ok
1054 __ bind(exit);
1055 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1056 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1057 __ pop(c_rarg3); // restore c_rarg3
1058 __ pop(c_rarg2); // restore c_rarg2
1059 __ pop(r12); // restore r12
1060 __ popf(); // restore flags
1061 __ ret(4 * wordSize); // pop caller saved stuff
1062
1063 // handle errors
1064 __ bind(error);
1065 __ movptr(rax, Address(rsp, saved_rax)); // get saved rax back
1066 __ movptr(rscratch1, Address(rsp, saved_r10)); // get saved r10 back
1067 __ pop(c_rarg3); // get saved c_rarg3 back
1068 __ pop(c_rarg2); // get saved c_rarg2 back
1069 __ pop(r12); // get saved r12 back
1070 __ popf(); // get saved flags off stack --
1071 // will be ignored
1072
1073 __ pusha(); // push registers
1074 // (rip is already
1075 // already pushed)
1076 // debug(char* msg, int64_t pc, int64_t regs[])
1077 // We've popped the registers we'd saved (c_rarg3, c_rarg2 and flags), and
1078 // pushed all the registers, so now the stack looks like:
1079 // [tos + 0] 16 saved registers
1080 // [tos + 16] return address
1081 // * [tos + 17] error message (char*)
1082 // * [tos + 18] object to verify (oop)
1083 // * [tos + 19] saved rax - saved by caller and bashed
1084 // * [tos + 20] saved r10 (rscratch1) - saved by caller
1085 // * = popped on exit
1086
1087 __ movptr(c_rarg0, Address(rsp, error_msg)); // pass address of error message
1088 __ movptr(c_rarg1, Address(rsp, return_addr)); // pass return address
1089 __ movq(c_rarg2, rsp); // pass address of regs on stack
1090 __ mov(r12, rsp); // remember rsp
1091 __ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
1092 __ andptr(rsp, -16); // align stack as required by ABI
1093 BLOCK_COMMENT("call MacroAssembler::debug");
1094 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64)));
1095 __ mov(rsp, r12); // restore rsp
1096 __ popa(); // pop registers (includes r12)
1097 __ ret(4 * wordSize); // pop caller saved stuff
1098
1099 return start;
1100 }
1101
1102 //
1103 // Verify that a register contains clean 32-bits positive value
1104 // (high 32-bits are 0) so it could be used in 64-bits shifts.
1105 //
1106 // Input:
1107 // Rint - 32-bits value
1108 // Rtmp - scratch
1109 //
1110 void assert_clean_int(Register Rint, Register Rtmp) {
1111 #ifdef ASSERT
1112 Label L;
1113 assert_different_registers(Rtmp, Rint);
1114 __ movslq(Rtmp, Rint);
1115 __ cmpq(Rtmp, Rint);
1116 __ jcc(Assembler::equal, L);
1117 __ stop("high 32-bits of int value are not 0");
1118 __ bind(L);
1119 #endif
1120 }
1121
1122 // Generate overlap test for array copy stubs
1123 //
1124 // Input:
1125 // c_rarg0 - from
1126 // c_rarg1 - to
1127 // c_rarg2 - element count
1128 //
1129 // Output:
1130 // rax - &from[element count - 1]
1131 //
1132 void array_overlap_test(address no_overlap_target, Address::ScaleFactor sf) {
1133 assert(no_overlap_target != NULL, "must be generated");
1134 array_overlap_test(no_overlap_target, NULL, sf);
1135 }
1136 void array_overlap_test(Label& L_no_overlap, Address::ScaleFactor sf) {
1137 array_overlap_test(NULL, &L_no_overlap, sf);
1138 }
1139 void array_overlap_test(address no_overlap_target, Label* NOLp, Address::ScaleFactor sf) {
1140 const Register from = c_rarg0;
1141 const Register to = c_rarg1;
1142 const Register count = c_rarg2;
1143 const Register end_from = rax;
1144
1145 __ cmpptr(to, from);
1146 __ lea(end_from, Address(from, count, sf, 0));
1147 if (NOLp == NULL) {
1148 ExternalAddress no_overlap(no_overlap_target);
1149 __ jump_cc(Assembler::belowEqual, no_overlap);
1150 __ cmpptr(to, end_from);
1151 __ jump_cc(Assembler::aboveEqual, no_overlap);
1152 } else {
1153 __ jcc(Assembler::belowEqual, (*NOLp));
1154 __ cmpptr(to, end_from);
1155 __ jcc(Assembler::aboveEqual, (*NOLp));
1156 }
1157 }
1158
1159 // Shuffle first three arg regs on Windows into Linux/Solaris locations.
1160 //
1161 // Outputs:
1162 // rdi - rcx
1163 // rsi - rdx
1164 // rdx - r8
1165 // rcx - r9
1166 //
1167 // Registers r9 and r10 are used to save rdi and rsi on Windows, which latter
1168 // are non-volatile. r9 and r10 should not be used by the caller.
1169 //
1170 void setup_arg_regs(int nargs = 3) {
1171 const Register saved_rdi = r9;
1172 const Register saved_rsi = r10;
1173 assert(nargs == 3 || nargs == 4, "else fix");
1174 #ifdef _WIN64
1175 assert(c_rarg0 == rcx && c_rarg1 == rdx && c_rarg2 == r8 && c_rarg3 == r9,
1176 "unexpected argument registers");
1177 if (nargs >= 4)
1178 __ mov(rax, r9); // r9 is also saved_rdi
1179 __ movptr(saved_rdi, rdi);
1180 __ movptr(saved_rsi, rsi);
1181 __ mov(rdi, rcx); // c_rarg0
1182 __ mov(rsi, rdx); // c_rarg1
1183 __ mov(rdx, r8); // c_rarg2
1184 if (nargs >= 4)
1185 __ mov(rcx, rax); // c_rarg3 (via rax)
1186 #else
1187 assert(c_rarg0 == rdi && c_rarg1 == rsi && c_rarg2 == rdx && c_rarg3 == rcx,
1188 "unexpected argument registers");
1189 #endif
1190 }
1191
1192 void restore_arg_regs() {
1193 const Register saved_rdi = r9;
1194 const Register saved_rsi = r10;
1195 #ifdef _WIN64
1196 __ movptr(rdi, saved_rdi);
1197 __ movptr(rsi, saved_rsi);
1198 #endif
1199 }
1200
1201 // Generate code for an array write pre barrier
1202 //
1203 // addr - starting address
1204 // count - element count
1205 // tmp - scratch register
1206 //
1207 // Destroy no registers!
1208 //
1209 void gen_write_ref_array_pre_barrier(Register addr, Register count, bool dest_uninitialized) {
1210 BarrierSet* bs = Universe::heap()->barrier_set();
1211 switch (bs->kind()) {
1212 case BarrierSet::G1SATBCT:
1213 case BarrierSet::G1SATBCTLogging:
1214 // With G1, don't generate the call if we statically know that the target in uninitialized
1215 if (!dest_uninitialized) {
1216 __ pusha(); // push registers
1217 if (count == c_rarg0) {
1218 if (addr == c_rarg1) {
1219 // exactly backwards!!
1220 __ xchgptr(c_rarg1, c_rarg0);
1221 } else {
1222 __ movptr(c_rarg1, count);
1223 __ movptr(c_rarg0, addr);
1224 }
1225 } else {
1226 __ movptr(c_rarg0, addr);
1227 __ movptr(c_rarg1, count);
1228 }
1229 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 2);
1230 __ popa();
1231 }
1232 break;
1233 case BarrierSet::CardTableModRef:
1234 case BarrierSet::CardTableExtension:
1235 case BarrierSet::ModRef:
1236 break;
1237 default:
1238 ShouldNotReachHere();
1239
1240 }
1241 }
1242
1243 //
1244 // Generate code for an array write post barrier
1245 //
1246 // Input:
1247 // start - register containing starting address of destination array
1248 // end - register containing ending address of destination array
1249 // scratch - scratch register
1250 //
1251 // The input registers are overwritten.
1252 // The ending address is inclusive.
1253 void gen_write_ref_array_post_barrier(Register start, Register end, Register scratch) {
1254 assert_different_registers(start, end, scratch);
1255 BarrierSet* bs = Universe::heap()->barrier_set();
1256 switch (bs->kind()) {
1257 case BarrierSet::G1SATBCT:
1258 case BarrierSet::G1SATBCTLogging:
1259
1260 {
1261 __ pusha(); // push registers (overkill)
1262 // must compute element count unless barrier set interface is changed (other platforms supply count)
1263 assert_different_registers(start, end, scratch);
1264 __ lea(scratch, Address(end, BytesPerHeapOop));
1265 __ subptr(scratch, start); // subtract start to get #bytes
1266 __ shrptr(scratch, LogBytesPerHeapOop); // convert to element count
1267 __ mov(c_rarg0, start);
1268 __ mov(c_rarg1, scratch);
1269 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 2);
1270 __ popa();
1271 }
1272 break;
1273 case BarrierSet::CardTableModRef:
1274 case BarrierSet::CardTableExtension:
1275 {
1276 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
1277 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
1278
1279 Label L_loop;
1280
1281 __ shrptr(start, CardTableModRefBS::card_shift);
1282 __ addptr(end, BytesPerHeapOop);
1283 __ shrptr(end, CardTableModRefBS::card_shift);
1284 __ subptr(end, start); // number of bytes to copy
1285
1286 intptr_t disp = (intptr_t) ct->byte_map_base;
1287 if (Assembler::is_simm32(disp)) {
1288 Address cardtable(noreg, noreg, Address::no_scale, disp);
1289 __ lea(scratch, cardtable);
1290 } else {
1291 ExternalAddress cardtable((address)disp);
1292 __ lea(scratch, cardtable);
1293 }
1294
1295 const Register count = end; // 'end' register contains bytes count now
1296 __ addptr(start, scratch);
1297 __ BIND(L_loop);
1298 __ movb(Address(start, count, Address::times_1), 0);
1299 __ decrement(count);
1300 __ jcc(Assembler::greaterEqual, L_loop);
1301 }
1302 break;
1303 default:
1304 ShouldNotReachHere();
1305
1306 }
1307 }
1308
1309
1310 // Copy big chunks forward
1311 //
1312 // Inputs:
1313 // end_from - source arrays end address
1314 // end_to - destination array end address
1315 // qword_count - 64-bits element count, negative
1316 // to - scratch
1317 // L_copy_32_bytes - entry label
1318 // L_copy_8_bytes - exit label
1319 //
1320 void copy_32_bytes_forward(Register end_from, Register end_to,
1321 Register qword_count, Register to,
1322 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1323 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1324 Label L_loop;
1325 __ align(OptoLoopAlignment);
1326 __ BIND(L_loop);
1327 if(UseUnalignedLoadStores) {
1328 __ movdqu(xmm0, Address(end_from, qword_count, Address::times_8, -24));
1329 __ movdqu(Address(end_to, qword_count, Address::times_8, -24), xmm0);
1330 __ movdqu(xmm1, Address(end_from, qword_count, Address::times_8, - 8));
1331 __ movdqu(Address(end_to, qword_count, Address::times_8, - 8), xmm1);
1332
1333 } else {
1334 __ movq(to, Address(end_from, qword_count, Address::times_8, -24));
1335 __ movq(Address(end_to, qword_count, Address::times_8, -24), to);
1336 __ movq(to, Address(end_from, qword_count, Address::times_8, -16));
1337 __ movq(Address(end_to, qword_count, Address::times_8, -16), to);
1338 __ movq(to, Address(end_from, qword_count, Address::times_8, - 8));
1339 __ movq(Address(end_to, qword_count, Address::times_8, - 8), to);
1340 __ movq(to, Address(end_from, qword_count, Address::times_8, - 0));
1341 __ movq(Address(end_to, qword_count, Address::times_8, - 0), to);
1342 }
1343 __ BIND(L_copy_32_bytes);
1344 __ addptr(qword_count, 4);
1345 __ jcc(Assembler::lessEqual, L_loop);
1346 __ subptr(qword_count, 4);
1347 __ jcc(Assembler::less, L_copy_8_bytes); // Copy trailing qwords
1348 }
1349
1350
1351 // Copy big chunks backward
1352 //
1353 // Inputs:
1354 // from - source arrays address
1355 // dest - destination array address
1356 // qword_count - 64-bits element count
1357 // to - scratch
1358 // L_copy_32_bytes - entry label
1359 // L_copy_8_bytes - exit label
1360 //
1361 void copy_32_bytes_backward(Register from, Register dest,
1362 Register qword_count, Register to,
1363 Label& L_copy_32_bytes, Label& L_copy_8_bytes) {
1364 DEBUG_ONLY(__ stop("enter at entry label, not here"));
1365 Label L_loop;
1366 __ align(OptoLoopAlignment);
1367 __ BIND(L_loop);
1368 if(UseUnalignedLoadStores) {
1369 __ movdqu(xmm0, Address(from, qword_count, Address::times_8, 16));
1370 __ movdqu(Address(dest, qword_count, Address::times_8, 16), xmm0);
1371 __ movdqu(xmm1, Address(from, qword_count, Address::times_8, 0));
1372 __ movdqu(Address(dest, qword_count, Address::times_8, 0), xmm1);
1373
1374 } else {
1375 __ movq(to, Address(from, qword_count, Address::times_8, 24));
1376 __ movq(Address(dest, qword_count, Address::times_8, 24), to);
1377 __ movq(to, Address(from, qword_count, Address::times_8, 16));
1378 __ movq(Address(dest, qword_count, Address::times_8, 16), to);
1379 __ movq(to, Address(from, qword_count, Address::times_8, 8));
1380 __ movq(Address(dest, qword_count, Address::times_8, 8), to);
1381 __ movq(to, Address(from, qword_count, Address::times_8, 0));
1382 __ movq(Address(dest, qword_count, Address::times_8, 0), to);
1383 }
1384 __ BIND(L_copy_32_bytes);
1385 __ subptr(qword_count, 4);
1386 __ jcc(Assembler::greaterEqual, L_loop);
1387 __ addptr(qword_count, 4);
1388 __ jcc(Assembler::greater, L_copy_8_bytes); // Copy trailing qwords
1389 }
1390
1391
1392 // Arguments:
1393 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1394 // ignored
1395 // name - stub name string
1396 //
1397 // Inputs:
1398 // c_rarg0 - source array address
1399 // c_rarg1 - destination array address
1400 // c_rarg2 - element count, treated as ssize_t, can be zero
1401 //
1402 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1403 // we let the hardware handle it. The one to eight bytes within words,
1404 // dwords or qwords that span cache line boundaries will still be loaded
1405 // and stored atomically.
1406 //
1407 // Side Effects:
1408 // disjoint_byte_copy_entry is set to the no-overlap entry point
1409 // used by generate_conjoint_byte_copy().
1410 //
1411 address generate_disjoint_byte_copy(bool aligned, address* entry, const char *name) {
1412 __ align(CodeEntryAlignment);
1413 StubCodeMark mark(this, "StubRoutines", name);
1414 address start = __ pc();
1415
1416 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1417 Label L_copy_byte, L_exit;
1418 const Register from = rdi; // source array address
1419 const Register to = rsi; // destination array address
1420 const Register count = rdx; // elements count
1421 const Register byte_count = rcx;
1422 const Register qword_count = count;
1423 const Register end_from = from; // source array end address
1424 const Register end_to = to; // destination array end address
1425 // End pointers are inclusive, and if count is not zero they point
1426 // to the last unit copied: end_to[0] := end_from[0]
1427
1428 __ enter(); // required for proper stackwalking of RuntimeStub frame
1429 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1430
1431 if (entry != NULL) {
1432 *entry = __ pc();
1433 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1434 BLOCK_COMMENT("Entry:");
1435 }
1436
1437 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1438 // r9 and r10 may be used to save non-volatile registers
1439
1440 // 'from', 'to' and 'count' are now valid
1441 __ movptr(byte_count, count);
1442 __ shrptr(count, 3); // count => qword_count
1443
1444 // Copy from low to high addresses. Use 'to' as scratch.
1445 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1446 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1447 __ negptr(qword_count); // make the count negative
1448 __ jmp(L_copy_32_bytes);
1449
1450 // Copy trailing qwords
1451 __ BIND(L_copy_8_bytes);
1452 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1453 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1454 __ increment(qword_count);
1455 __ jcc(Assembler::notZero, L_copy_8_bytes);
1456
1457 // Check for and copy trailing dword
1458 __ BIND(L_copy_4_bytes);
1459 __ testl(byte_count, 4);
1460 __ jccb(Assembler::zero, L_copy_2_bytes);
1461 __ movl(rax, Address(end_from, 8));
1462 __ movl(Address(end_to, 8), rax);
1463
1464 __ addptr(end_from, 4);
1465 __ addptr(end_to, 4);
1466
1467 // Check for and copy trailing word
1468 __ BIND(L_copy_2_bytes);
1469 __ testl(byte_count, 2);
1470 __ jccb(Assembler::zero, L_copy_byte);
1471 __ movw(rax, Address(end_from, 8));
1472 __ movw(Address(end_to, 8), rax);
1473
1474 __ addptr(end_from, 2);
1475 __ addptr(end_to, 2);
1476
1477 // Check for and copy trailing byte
1478 __ BIND(L_copy_byte);
1479 __ testl(byte_count, 1);
1480 __ jccb(Assembler::zero, L_exit);
1481 __ movb(rax, Address(end_from, 8));
1482 __ movb(Address(end_to, 8), rax);
1483
1484 __ BIND(L_exit);
1485 restore_arg_regs();
1486 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1487 __ xorptr(rax, rax); // return 0
1488 __ leave(); // required for proper stackwalking of RuntimeStub frame
1489 __ ret(0);
1490
1491 // Copy in 32-bytes chunks
1492 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1493 __ jmp(L_copy_4_bytes);
1494
1495 return start;
1496 }
1497
1498 // Arguments:
1499 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1500 // ignored
1501 // name - stub name string
1502 //
1503 // Inputs:
1504 // c_rarg0 - source array address
1505 // c_rarg1 - destination array address
1506 // c_rarg2 - element count, treated as ssize_t, can be zero
1507 //
1508 // If 'from' and/or 'to' are aligned on 4-, 2-, or 1-byte boundaries,
1509 // we let the hardware handle it. The one to eight bytes within words,
1510 // dwords or qwords that span cache line boundaries will still be loaded
1511 // and stored atomically.
1512 //
1513 address generate_conjoint_byte_copy(bool aligned, address nooverlap_target,
1514 address* entry, const char *name) {
1515 __ align(CodeEntryAlignment);
1516 StubCodeMark mark(this, "StubRoutines", name);
1517 address start = __ pc();
1518
1519 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_copy_2_bytes;
1520 const Register from = rdi; // source array address
1521 const Register to = rsi; // destination array address
1522 const Register count = rdx; // elements count
1523 const Register byte_count = rcx;
1524 const Register qword_count = count;
1525
1526 __ enter(); // required for proper stackwalking of RuntimeStub frame
1527 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1528
1529 if (entry != NULL) {
1530 *entry = __ pc();
1531 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1532 BLOCK_COMMENT("Entry:");
1533 }
1534
1535 array_overlap_test(nooverlap_target, Address::times_1);
1536 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1537 // r9 and r10 may be used to save non-volatile registers
1538
1539 // 'from', 'to' and 'count' are now valid
1540 __ movptr(byte_count, count);
1541 __ shrptr(count, 3); // count => qword_count
1542
1543 // Copy from high to low addresses.
1544
1545 // Check for and copy trailing byte
1546 __ testl(byte_count, 1);
1547 __ jcc(Assembler::zero, L_copy_2_bytes);
1548 __ movb(rax, Address(from, byte_count, Address::times_1, -1));
1549 __ movb(Address(to, byte_count, Address::times_1, -1), rax);
1550 __ decrement(byte_count); // Adjust for possible trailing word
1551
1552 // Check for and copy trailing word
1553 __ BIND(L_copy_2_bytes);
1554 __ testl(byte_count, 2);
1555 __ jcc(Assembler::zero, L_copy_4_bytes);
1556 __ movw(rax, Address(from, byte_count, Address::times_1, -2));
1557 __ movw(Address(to, byte_count, Address::times_1, -2), rax);
1558
1559 // Check for and copy trailing dword
1560 __ BIND(L_copy_4_bytes);
1561 __ testl(byte_count, 4);
1562 __ jcc(Assembler::zero, L_copy_32_bytes);
1563 __ movl(rax, Address(from, qword_count, Address::times_8));
1564 __ movl(Address(to, qword_count, Address::times_8), rax);
1565 __ jmp(L_copy_32_bytes);
1566
1567 // Copy trailing qwords
1568 __ BIND(L_copy_8_bytes);
1569 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1570 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1571 __ decrement(qword_count);
1572 __ jcc(Assembler::notZero, L_copy_8_bytes);
1573
1574 restore_arg_regs();
1575 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1576 __ xorptr(rax, rax); // return 0
1577 __ leave(); // required for proper stackwalking of RuntimeStub frame
1578 __ ret(0);
1579
1580 // Copy in 32-bytes chunks
1581 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1582
1583 restore_arg_regs();
1584 inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); // Update counter after rscratch1 is free
1585 __ xorptr(rax, rax); // return 0
1586 __ leave(); // required for proper stackwalking of RuntimeStub frame
1587 __ ret(0);
1588
1589 return start;
1590 }
1591
1592 // Arguments:
1593 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1594 // ignored
1595 // name - stub name string
1596 //
1597 // Inputs:
1598 // c_rarg0 - source array address
1599 // c_rarg1 - destination array address
1600 // c_rarg2 - element count, treated as ssize_t, can be zero
1601 //
1602 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1603 // let the hardware handle it. The two or four words within dwords
1604 // or qwords that span cache line boundaries will still be loaded
1605 // and stored atomically.
1606 //
1607 // Side Effects:
1608 // disjoint_short_copy_entry is set to the no-overlap entry point
1609 // used by generate_conjoint_short_copy().
1610 //
1611 address generate_disjoint_short_copy(bool aligned, address *entry, const char *name) {
1612 __ align(CodeEntryAlignment);
1613 StubCodeMark mark(this, "StubRoutines", name);
1614 address start = __ pc();
1615
1616 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes,L_copy_2_bytes,L_exit;
1617 const Register from = rdi; // source array address
1618 const Register to = rsi; // destination array address
1619 const Register count = rdx; // elements count
1620 const Register word_count = rcx;
1621 const Register qword_count = count;
1622 const Register end_from = from; // source array end address
1623 const Register end_to = to; // destination array end address
1624 // End pointers are inclusive, and if count is not zero they point
1625 // to the last unit copied: end_to[0] := end_from[0]
1626
1627 __ enter(); // required for proper stackwalking of RuntimeStub frame
1628 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1629
1630 if (entry != NULL) {
1631 *entry = __ pc();
1632 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1633 BLOCK_COMMENT("Entry:");
1634 }
1635
1636 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1637 // r9 and r10 may be used to save non-volatile registers
1638
1639 // 'from', 'to' and 'count' are now valid
1640 __ movptr(word_count, count);
1641 __ shrptr(count, 2); // count => qword_count
1642
1643 // Copy from low to high addresses. Use 'to' as scratch.
1644 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1645 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1646 __ negptr(qword_count);
1647 __ jmp(L_copy_32_bytes);
1648
1649 // Copy trailing qwords
1650 __ BIND(L_copy_8_bytes);
1651 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1652 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1653 __ increment(qword_count);
1654 __ jcc(Assembler::notZero, L_copy_8_bytes);
1655
1656 // Original 'dest' is trashed, so we can't use it as a
1657 // base register for a possible trailing word copy
1658
1659 // Check for and copy trailing dword
1660 __ BIND(L_copy_4_bytes);
1661 __ testl(word_count, 2);
1662 __ jccb(Assembler::zero, L_copy_2_bytes);
1663 __ movl(rax, Address(end_from, 8));
1664 __ movl(Address(end_to, 8), rax);
1665
1666 __ addptr(end_from, 4);
1667 __ addptr(end_to, 4);
1668
1669 // Check for and copy trailing word
1670 __ BIND(L_copy_2_bytes);
1671 __ testl(word_count, 1);
1672 __ jccb(Assembler::zero, L_exit);
1673 __ movw(rax, Address(end_from, 8));
1674 __ movw(Address(end_to, 8), rax);
1675
1676 __ BIND(L_exit);
1677 restore_arg_regs();
1678 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1679 __ xorptr(rax, rax); // return 0
1680 __ leave(); // required for proper stackwalking of RuntimeStub frame
1681 __ ret(0);
1682
1683 // Copy in 32-bytes chunks
1684 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1685 __ jmp(L_copy_4_bytes);
1686
1687 return start;
1688 }
1689
1690 address generate_fill(BasicType t, bool aligned, const char *name) {
1691 __ align(CodeEntryAlignment);
1692 StubCodeMark mark(this, "StubRoutines", name);
1693 address start = __ pc();
1694
1695 BLOCK_COMMENT("Entry:");
1696
1697 const Register to = c_rarg0; // source array address
1698 const Register value = c_rarg1; // value
1699 const Register count = c_rarg2; // elements count
1700
1701 __ enter(); // required for proper stackwalking of RuntimeStub frame
1702
1703 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1704
1705 __ leave(); // required for proper stackwalking of RuntimeStub frame
1706 __ ret(0);
1707 return start;
1708 }
1709
1710 // Arguments:
1711 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1712 // ignored
1713 // name - stub name string
1714 //
1715 // Inputs:
1716 // c_rarg0 - source array address
1717 // c_rarg1 - destination array address
1718 // c_rarg2 - element count, treated as ssize_t, can be zero
1719 //
1720 // If 'from' and/or 'to' are aligned on 4- or 2-byte boundaries, we
1721 // let the hardware handle it. The two or four words within dwords
1722 // or qwords that span cache line boundaries will still be loaded
1723 // and stored atomically.
1724 //
1725 address generate_conjoint_short_copy(bool aligned, address nooverlap_target,
1726 address *entry, const char *name) {
1727 __ align(CodeEntryAlignment);
1728 StubCodeMark mark(this, "StubRoutines", name);
1729 address start = __ pc();
1730
1731 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes;
1732 const Register from = rdi; // source array address
1733 const Register to = rsi; // destination array address
1734 const Register count = rdx; // elements count
1735 const Register word_count = rcx;
1736 const Register qword_count = count;
1737
1738 __ enter(); // required for proper stackwalking of RuntimeStub frame
1739 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1740
1741 if (entry != NULL) {
1742 *entry = __ pc();
1743 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1744 BLOCK_COMMENT("Entry:");
1745 }
1746
1747 array_overlap_test(nooverlap_target, Address::times_2);
1748 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1749 // r9 and r10 may be used to save non-volatile registers
1750
1751 // 'from', 'to' and 'count' are now valid
1752 __ movptr(word_count, count);
1753 __ shrptr(count, 2); // count => qword_count
1754
1755 // Copy from high to low addresses. Use 'to' as scratch.
1756
1757 // Check for and copy trailing word
1758 __ testl(word_count, 1);
1759 __ jccb(Assembler::zero, L_copy_4_bytes);
1760 __ movw(rax, Address(from, word_count, Address::times_2, -2));
1761 __ movw(Address(to, word_count, Address::times_2, -2), rax);
1762
1763 // Check for and copy trailing dword
1764 __ BIND(L_copy_4_bytes);
1765 __ testl(word_count, 2);
1766 __ jcc(Assembler::zero, L_copy_32_bytes);
1767 __ movl(rax, Address(from, qword_count, Address::times_8));
1768 __ movl(Address(to, qword_count, Address::times_8), rax);
1769 __ jmp(L_copy_32_bytes);
1770
1771 // Copy trailing qwords
1772 __ BIND(L_copy_8_bytes);
1773 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1774 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1775 __ decrement(qword_count);
1776 __ jcc(Assembler::notZero, L_copy_8_bytes);
1777
1778 restore_arg_regs();
1779 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1780 __ xorptr(rax, rax); // return 0
1781 __ leave(); // required for proper stackwalking of RuntimeStub frame
1782 __ ret(0);
1783
1784 // Copy in 32-bytes chunks
1785 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1786
1787 restore_arg_regs();
1788 inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); // Update counter after rscratch1 is free
1789 __ xorptr(rax, rax); // return 0
1790 __ leave(); // required for proper stackwalking of RuntimeStub frame
1791 __ ret(0);
1792
1793 return start;
1794 }
1795
1796 // Arguments:
1797 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1798 // ignored
1799 // is_oop - true => oop array, so generate store check code
1800 // name - stub name string
1801 //
1802 // Inputs:
1803 // c_rarg0 - source array address
1804 // c_rarg1 - destination array address
1805 // c_rarg2 - element count, treated as ssize_t, can be zero
1806 //
1807 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1808 // the hardware handle it. The two dwords within qwords that span
1809 // cache line boundaries will still be loaded and stored atomicly.
1810 //
1811 // Side Effects:
1812 // disjoint_int_copy_entry is set to the no-overlap entry point
1813 // used by generate_conjoint_int_oop_copy().
1814 //
1815 address generate_disjoint_int_oop_copy(bool aligned, bool is_oop, address* entry,
1816 const char *name, bool dest_uninitialized = false) {
1817 __ align(CodeEntryAlignment);
1818 StubCodeMark mark(this, "StubRoutines", name);
1819 address start = __ pc();
1820
1821 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_4_bytes, L_exit;
1822 const Register from = rdi; // source array address
1823 const Register to = rsi; // destination array address
1824 const Register count = rdx; // elements count
1825 const Register dword_count = rcx;
1826 const Register qword_count = count;
1827 const Register end_from = from; // source array end address
1828 const Register end_to = to; // destination array end address
1829 const Register saved_to = r11; // saved destination array address
1830 // End pointers are inclusive, and if count is not zero they point
1831 // to the last unit copied: end_to[0] := end_from[0]
1832
1833 __ enter(); // required for proper stackwalking of RuntimeStub frame
1834 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1835
1836 if (entry != NULL) {
1837 *entry = __ pc();
1838 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1839 BLOCK_COMMENT("Entry:");
1840 }
1841
1842 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1843 // r9 and r10 may be used to save non-volatile registers
1844 if (is_oop) {
1845 __ movq(saved_to, to);
1846 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1847 }
1848
1849 // 'from', 'to' and 'count' are now valid
1850 __ movptr(dword_count, count);
1851 __ shrptr(count, 1); // count => qword_count
1852
1853 // Copy from low to high addresses. Use 'to' as scratch.
1854 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
1855 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
1856 __ negptr(qword_count);
1857 __ jmp(L_copy_32_bytes);
1858
1859 // Copy trailing qwords
1860 __ BIND(L_copy_8_bytes);
1861 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
1862 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
1863 __ increment(qword_count);
1864 __ jcc(Assembler::notZero, L_copy_8_bytes);
1865
1866 // Check for and copy trailing dword
1867 __ BIND(L_copy_4_bytes);
1868 __ testl(dword_count, 1); // Only byte test since the value is 0 or 1
1869 __ jccb(Assembler::zero, L_exit);
1870 __ movl(rax, Address(end_from, 8));
1871 __ movl(Address(end_to, 8), rax);
1872
1873 __ BIND(L_exit);
1874 if (is_oop) {
1875 __ leaq(end_to, Address(saved_to, dword_count, Address::times_4, -4));
1876 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
1877 }
1878 restore_arg_regs();
1879 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1880 __ xorptr(rax, rax); // return 0
1881 __ leave(); // required for proper stackwalking of RuntimeStub frame
1882 __ ret(0);
1883
1884 // Copy 32-bytes chunks
1885 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1886 __ jmp(L_copy_4_bytes);
1887
1888 return start;
1889 }
1890
1891 // Arguments:
1892 // aligned - true => Input and output aligned on a HeapWord == 8-byte boundary
1893 // ignored
1894 // is_oop - true => oop array, so generate store check code
1895 // name - stub name string
1896 //
1897 // Inputs:
1898 // c_rarg0 - source array address
1899 // c_rarg1 - destination array address
1900 // c_rarg2 - element count, treated as ssize_t, can be zero
1901 //
1902 // If 'from' and/or 'to' are aligned on 4-byte boundaries, we let
1903 // the hardware handle it. The two dwords within qwords that span
1904 // cache line boundaries will still be loaded and stored atomicly.
1905 //
1906 address generate_conjoint_int_oop_copy(bool aligned, bool is_oop, address nooverlap_target,
1907 address *entry, const char *name,
1908 bool dest_uninitialized = false) {
1909 __ align(CodeEntryAlignment);
1910 StubCodeMark mark(this, "StubRoutines", name);
1911 address start = __ pc();
1912
1913 Label L_copy_32_bytes, L_copy_8_bytes, L_copy_2_bytes, L_exit;
1914 const Register from = rdi; // source array address
1915 const Register to = rsi; // destination array address
1916 const Register count = rdx; // elements count
1917 const Register dword_count = rcx;
1918 const Register qword_count = count;
1919
1920 __ enter(); // required for proper stackwalking of RuntimeStub frame
1921 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
1922
1923 if (entry != NULL) {
1924 *entry = __ pc();
1925 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
1926 BLOCK_COMMENT("Entry:");
1927 }
1928
1929 array_overlap_test(nooverlap_target, Address::times_4);
1930 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
1931 // r9 and r10 may be used to save non-volatile registers
1932
1933 if (is_oop) {
1934 // no registers are destroyed by this call
1935 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1936 }
1937
1938 assert_clean_int(count, rax); // Make sure 'count' is clean int.
1939 // 'from', 'to' and 'count' are now valid
1940 __ movptr(dword_count, count);
1941 __ shrptr(count, 1); // count => qword_count
1942
1943 // Copy from high to low addresses. Use 'to' as scratch.
1944
1945 // Check for and copy trailing dword
1946 __ testl(dword_count, 1);
1947 __ jcc(Assembler::zero, L_copy_32_bytes);
1948 __ movl(rax, Address(from, dword_count, Address::times_4, -4));
1949 __ movl(Address(to, dword_count, Address::times_4, -4), rax);
1950 __ jmp(L_copy_32_bytes);
1951
1952 // Copy trailing qwords
1953 __ BIND(L_copy_8_bytes);
1954 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
1955 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
1956 __ decrement(qword_count);
1957 __ jcc(Assembler::notZero, L_copy_8_bytes);
1958
1959 if (is_oop) {
1960 __ jmp(L_exit);
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 // Copy in 32-bytes chunks
1969 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
1970
1971 __ bind(L_exit);
1972 if (is_oop) {
1973 Register end_to = rdx;
1974 __ leaq(end_to, Address(to, dword_count, Address::times_4, -4));
1975 gen_write_ref_array_post_barrier(to, end_to, rax);
1976 }
1977 restore_arg_regs();
1978 inc_counter_np(SharedRuntime::_jint_array_copy_ctr); // Update counter after rscratch1 is free
1979 __ xorptr(rax, rax); // return 0
1980 __ leave(); // required for proper stackwalking of RuntimeStub frame
1981 __ ret(0);
1982
1983 return start;
1984 }
1985
1986 // Arguments:
1987 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
1988 // ignored
1989 // is_oop - true => oop array, so generate store check code
1990 // name - stub name string
1991 //
1992 // Inputs:
1993 // c_rarg0 - source array address
1994 // c_rarg1 - destination array address
1995 // c_rarg2 - element count, treated as ssize_t, can be zero
1996 //
1997 // Side Effects:
1998 // disjoint_oop_copy_entry or disjoint_long_copy_entry is set to the
1999 // no-overlap entry point used by generate_conjoint_long_oop_copy().
2000 //
2001 address generate_disjoint_long_oop_copy(bool aligned, bool is_oop, address *entry,
2002 const char *name, bool dest_uninitialized = false) {
2003 __ align(CodeEntryAlignment);
2004 StubCodeMark mark(this, "StubRoutines", name);
2005 address start = __ pc();
2006
2007 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2008 const Register from = rdi; // source array address
2009 const Register to = rsi; // destination array address
2010 const Register qword_count = rdx; // elements count
2011 const Register end_from = from; // source array end address
2012 const Register end_to = rcx; // destination array end address
2013 const Register saved_to = to;
2014 // End pointers are inclusive, and if count is not zero they point
2015 // to the last unit copied: end_to[0] := end_from[0]
2016
2017 __ enter(); // required for proper stackwalking of RuntimeStub frame
2018 // Save no-overlap entry point for generate_conjoint_long_oop_copy()
2019 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2020
2021 if (entry != NULL) {
2022 *entry = __ pc();
2023 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2024 BLOCK_COMMENT("Entry:");
2025 }
2026
2027 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2028 // r9 and r10 may be used to save non-volatile registers
2029 // 'from', 'to' and 'qword_count' are now valid
2030 if (is_oop) {
2031 // no registers are destroyed by this call
2032 gen_write_ref_array_pre_barrier(to, qword_count, dest_uninitialized);
2033 }
2034
2035 // Copy from low to high addresses. Use 'to' as scratch.
2036 __ lea(end_from, Address(from, qword_count, Address::times_8, -8));
2037 __ lea(end_to, Address(to, qword_count, Address::times_8, -8));
2038 __ negptr(qword_count);
2039 __ jmp(L_copy_32_bytes);
2040
2041 // Copy trailing qwords
2042 __ BIND(L_copy_8_bytes);
2043 __ movq(rax, Address(end_from, qword_count, Address::times_8, 8));
2044 __ movq(Address(end_to, qword_count, Address::times_8, 8), rax);
2045 __ increment(qword_count);
2046 __ jcc(Assembler::notZero, L_copy_8_bytes);
2047
2048 if (is_oop) {
2049 __ jmp(L_exit);
2050 } else {
2051 restore_arg_regs();
2052 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2053 __ xorptr(rax, rax); // return 0
2054 __ leave(); // required for proper stackwalking of RuntimeStub frame
2055 __ ret(0);
2056 }
2057
2058 // Copy 64-byte chunks
2059 copy_32_bytes_forward(end_from, end_to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2060
2061 if (is_oop) {
2062 __ BIND(L_exit);
2063 gen_write_ref_array_post_barrier(saved_to, end_to, rax);
2064 }
2065 restore_arg_regs();
2066 if (is_oop) {
2067 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2068 } else {
2069 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2070 }
2071 __ xorptr(rax, rax); // return 0
2072 __ leave(); // required for proper stackwalking of RuntimeStub frame
2073 __ ret(0);
2074
2075 return start;
2076 }
2077
2078 // Arguments:
2079 // aligned - true => Input and output aligned on a HeapWord boundary == 8 bytes
2080 // ignored
2081 // is_oop - true => oop array, so generate store check code
2082 // name - stub name string
2083 //
2084 // Inputs:
2085 // c_rarg0 - source array address
2086 // c_rarg1 - destination array address
2087 // c_rarg2 - element count, treated as ssize_t, can be zero
2088 //
2089 address generate_conjoint_long_oop_copy(bool aligned, bool is_oop,
2090 address nooverlap_target, address *entry,
2091 const char *name, bool dest_uninitialized = false) {
2092 __ align(CodeEntryAlignment);
2093 StubCodeMark mark(this, "StubRoutines", name);
2094 address start = __ pc();
2095
2096 Label L_copy_32_bytes, L_copy_8_bytes, L_exit;
2097 const Register from = rdi; // source array address
2098 const Register to = rsi; // destination array address
2099 const Register qword_count = rdx; // elements count
2100 const Register saved_count = rcx;
2101
2102 __ enter(); // required for proper stackwalking of RuntimeStub frame
2103 assert_clean_int(c_rarg2, rax); // Make sure 'count' is clean int.
2104
2105 if (entry != NULL) {
2106 *entry = __ pc();
2107 // caller can pass a 64-bit byte count here (from Unsafe.copyMemory)
2108 BLOCK_COMMENT("Entry:");
2109 }
2110
2111 array_overlap_test(nooverlap_target, Address::times_8);
2112 setup_arg_regs(); // from => rdi, to => rsi, count => rdx
2113 // r9 and r10 may be used to save non-volatile registers
2114 // 'from', 'to' and 'qword_count' are now valid
2115 if (is_oop) {
2116 // Save to and count for store barrier
2117 __ movptr(saved_count, qword_count);
2118 // No registers are destroyed by this call
2119 gen_write_ref_array_pre_barrier(to, saved_count, dest_uninitialized);
2120 }
2121
2122 __ jmp(L_copy_32_bytes);
2123
2124 // Copy trailing qwords
2125 __ BIND(L_copy_8_bytes);
2126 __ movq(rax, Address(from, qword_count, Address::times_8, -8));
2127 __ movq(Address(to, qword_count, Address::times_8, -8), rax);
2128 __ decrement(qword_count);
2129 __ jcc(Assembler::notZero, L_copy_8_bytes);
2130
2131 if (is_oop) {
2132 __ jmp(L_exit);
2133 } else {
2134 restore_arg_regs();
2135 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2136 __ xorptr(rax, rax); // return 0
2137 __ leave(); // required for proper stackwalking of RuntimeStub frame
2138 __ ret(0);
2139 }
2140
2141 // Copy in 32-bytes chunks
2142 copy_32_bytes_backward(from, to, qword_count, rax, L_copy_32_bytes, L_copy_8_bytes);
2143
2144 if (is_oop) {
2145 __ BIND(L_exit);
2146 __ lea(rcx, Address(to, saved_count, Address::times_8, -8));
2147 gen_write_ref_array_post_barrier(to, rcx, rax);
2148 }
2149 restore_arg_regs();
2150 if (is_oop) {
2151 inc_counter_np(SharedRuntime::_oop_array_copy_ctr); // Update counter after rscratch1 is free
2152 } else {
2153 inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); // Update counter after rscratch1 is free
2154 }
2155 __ xorptr(rax, rax); // return 0
2156 __ leave(); // required for proper stackwalking of RuntimeStub frame
2157 __ ret(0);
2158
2159 return start;
2160 }
2161
2162
2163 // Helper for generating a dynamic type check.
2164 // Smashes no registers.
2165 void generate_type_check(Register sub_klass,
2166 Register super_check_offset,
2167 Register super_klass,
2168 Label& L_success) {
2169 assert_different_registers(sub_klass, super_check_offset, super_klass);
2170
2171 BLOCK_COMMENT("type_check:");
2172
2173 Label L_miss;
2174
2175 __ check_klass_subtype_fast_path(sub_klass, super_klass, noreg, &L_success, &L_miss, NULL,
2176 super_check_offset);
2177 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, &L_success, NULL);
2178
2179 // Fall through on failure!
2180 __ BIND(L_miss);
2181 }
2182
2183 //
2184 // Generate checkcasting array copy stub
2185 //
2186 // Input:
2187 // c_rarg0 - source array address
2188 // c_rarg1 - destination array address
2189 // c_rarg2 - element count, treated as ssize_t, can be zero
2190 // c_rarg3 - size_t ckoff (super_check_offset)
2191 // not Win64
2192 // c_rarg4 - oop ckval (super_klass)
2193 // Win64
2194 // rsp+40 - oop ckval (super_klass)
2195 //
2196 // Output:
2197 // rax == 0 - success
2198 // rax == -1^K - failure, where K is partial transfer count
2199 //
2200 address generate_checkcast_copy(const char *name, address *entry,
2201 bool dest_uninitialized = false) {
2202
2203 Label L_load_element, L_store_element, L_do_card_marks, L_done;
2204
2205 // Input registers (after setup_arg_regs)
2206 const Register from = rdi; // source array address
2207 const Register to = rsi; // destination array address
2208 const Register length = rdx; // elements count
2209 const Register ckoff = rcx; // super_check_offset
2210 const Register ckval = r8; // super_klass
2211
2212 // Registers used as temps (r13, r14 are save-on-entry)
2213 const Register end_from = from; // source array end address
2214 const Register end_to = r13; // destination array end address
2215 const Register count = rdx; // -(count_remaining)
2216 const Register r14_length = r14; // saved copy of length
2217 // End pointers are inclusive, and if length is not zero they point
2218 // to the last unit copied: end_to[0] := end_from[0]
2219
2220 const Register rax_oop = rax; // actual oop copied
2221 const Register r11_klass = r11; // oop._klass
2222
2223 //---------------------------------------------------------------
2224 // Assembler stub will be used for this call to arraycopy
2225 // if the two arrays are subtypes of Object[] but the
2226 // destination array type is not equal to or a supertype
2227 // of the source type. Each element must be separately
2228 // checked.
2229
2230 __ align(CodeEntryAlignment);
2231 StubCodeMark mark(this, "StubRoutines", name);
2232 address start = __ pc();
2233
2234 __ enter(); // required for proper stackwalking of RuntimeStub frame
2235
2236 #ifdef ASSERT
2237 // caller guarantees that the arrays really are different
2238 // otherwise, we would have to make conjoint checks
2239 { Label L;
2240 array_overlap_test(L, TIMES_OOP);
2241 __ stop("checkcast_copy within a single array");
2242 __ bind(L);
2243 }
2244 #endif //ASSERT
2245
2246 setup_arg_regs(4); // from => rdi, to => rsi, length => rdx
2247 // ckoff => rcx, ckval => r8
2248 // r9 and r10 may be used to save non-volatile registers
2249 #ifdef _WIN64
2250 // last argument (#4) is on stack on Win64
2251 __ movptr(ckval, Address(rsp, 6 * wordSize));
2252 #endif
2253
2254 // Caller of this entry point must set up the argument registers.
2255 if (entry != NULL) {
2256 *entry = __ pc();
2257 BLOCK_COMMENT("Entry:");
2258 }
2259
2260 // allocate spill slots for r13, r14
2261 enum {
2262 saved_r13_offset,
2263 saved_r14_offset,
2264 saved_rbp_offset
2265 };
2266 __ subptr(rsp, saved_rbp_offset * wordSize);
2267 __ movptr(Address(rsp, saved_r13_offset * wordSize), r13);
2268 __ movptr(Address(rsp, saved_r14_offset * wordSize), r14);
2269
2270 // check that int operands are properly extended to size_t
2271 assert_clean_int(length, rax);
2272 assert_clean_int(ckoff, rax);
2273
2274 #ifdef ASSERT
2275 BLOCK_COMMENT("assert consistent ckoff/ckval");
2276 // The ckoff and ckval must be mutually consistent,
2277 // even though caller generates both.
2278 { Label L;
2279 int sco_offset = in_bytes(Klass::super_check_offset_offset());
2280 __ cmpl(ckoff, Address(ckval, sco_offset));
2281 __ jcc(Assembler::equal, L);
2282 __ stop("super_check_offset inconsistent");
2283 __ bind(L);
2284 }
2285 #endif //ASSERT
2286
2287 // Loop-invariant addresses. They are exclusive end pointers.
2288 Address end_from_addr(from, length, TIMES_OOP, 0);
2289 Address end_to_addr(to, length, TIMES_OOP, 0);
2290 // Loop-variant addresses. They assume post-incremented count < 0.
2291 Address from_element_addr(end_from, count, TIMES_OOP, 0);
2292 Address to_element_addr(end_to, count, TIMES_OOP, 0);
2293
2294 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
2295
2296 // Copy from low to high addresses, indexed from the end of each array.
2297 __ lea(end_from, end_from_addr);
2298 __ lea(end_to, end_to_addr);
2299 __ movptr(r14_length, length); // save a copy of the length
2300 assert(length == count, ""); // else fix next line:
2301 __ negptr(count); // negate and test the length
2302 __ jcc(Assembler::notZero, L_load_element);
2303
2304 // Empty array: Nothing to do.
2305 __ xorptr(rax, rax); // return 0 on (trivial) success
2306 __ jmp(L_done);
2307
2308 // ======== begin loop ========
2309 // (Loop is rotated; its entry is L_load_element.)
2310 // Loop control:
2311 // for (count = -count; count != 0; count++)
2312 // Base pointers src, dst are biased by 8*(count-1),to last element.
2313 __ align(OptoLoopAlignment);
2314
2315 __ BIND(L_store_element);
2316 __ store_heap_oop(to_element_addr, rax_oop); // store the oop
2317 __ increment(count); // increment the count toward zero
2318 __ jcc(Assembler::zero, L_do_card_marks);
2319
2320 // ======== loop entry is here ========
2321 __ BIND(L_load_element);
2322 __ load_heap_oop(rax_oop, from_element_addr); // load the oop
2323 __ testptr(rax_oop, rax_oop);
2324 __ jcc(Assembler::zero, L_store_element);
2325
2326 __ load_klass(r11_klass, rax_oop);// query the object klass
2327 generate_type_check(r11_klass, ckoff, ckval, L_store_element);
2328 // ======== end loop ========
2329
2330 // It was a real error; we must depend on the caller to finish the job.
2331 // Register rdx = -1 * number of *remaining* oops, r14 = *total* oops.
2332 // Emit GC store barriers for the oops we have copied (r14 + rdx),
2333 // and report their number to the caller.
2334 assert_different_registers(rax, r14_length, count, to, end_to, rcx);
2335 __ lea(end_to, to_element_addr);
2336 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2337 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2338 __ movptr(rax, r14_length); // original oops
2339 __ addptr(rax, count); // K = (original - remaining) oops
2340 __ notptr(rax); // report (-1^K) to caller
2341 __ jmp(L_done);
2342
2343 // Come here on success only.
2344 __ BIND(L_do_card_marks);
2345 __ addptr(end_to, -heapOopSize); // make an inclusive end pointer
2346 gen_write_ref_array_post_barrier(to, end_to, rscratch1);
2347 __ xorptr(rax, rax); // return 0 on success
2348
2349 // Common exit point (success or failure).
2350 __ BIND(L_done);
2351 __ movptr(r13, Address(rsp, saved_r13_offset * wordSize));
2352 __ movptr(r14, Address(rsp, saved_r14_offset * wordSize));
2353 restore_arg_regs();
2354 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); // Update counter after rscratch1 is free
2355 __ leave(); // required for proper stackwalking of RuntimeStub frame
2356 __ ret(0);
2357
2358 return start;
2359 }
2360
2361 //
2362 // Generate 'unsafe' array copy stub
2363 // Though just as safe as the other stubs, it takes an unscaled
2364 // size_t argument instead of an element count.
2365 //
2366 // Input:
2367 // c_rarg0 - source array address
2368 // c_rarg1 - destination array address
2369 // c_rarg2 - byte count, treated as ssize_t, can be zero
2370 //
2371 // Examines the alignment of the operands and dispatches
2372 // to a long, int, short, or byte copy loop.
2373 //
2374 address generate_unsafe_copy(const char *name,
2375 address byte_copy_entry, address short_copy_entry,
2376 address int_copy_entry, address long_copy_entry) {
2377
2378 Label L_long_aligned, L_int_aligned, L_short_aligned;
2379
2380 // Input registers (before setup_arg_regs)
2381 const Register from = c_rarg0; // source array address
2382 const Register to = c_rarg1; // destination array address
2383 const Register size = c_rarg2; // byte count (size_t)
2384
2385 // Register used as a temp
2386 const Register bits = rax; // test copy of low bits
2387
2388 __ align(CodeEntryAlignment);
2389 StubCodeMark mark(this, "StubRoutines", name);
2390 address start = __ pc();
2391
2392 __ enter(); // required for proper stackwalking of RuntimeStub frame
2393
2394 // bump this on entry, not on exit:
2395 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
2396
2397 __ mov(bits, from);
2398 __ orptr(bits, to);
2399 __ orptr(bits, size);
2400
2401 __ testb(bits, BytesPerLong-1);
2402 __ jccb(Assembler::zero, L_long_aligned);
2403
2404 __ testb(bits, BytesPerInt-1);
2405 __ jccb(Assembler::zero, L_int_aligned);
2406
2407 __ testb(bits, BytesPerShort-1);
2408 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
2409
2410 __ BIND(L_short_aligned);
2411 __ shrptr(size, LogBytesPerShort); // size => short_count
2412 __ jump(RuntimeAddress(short_copy_entry));
2413
2414 __ BIND(L_int_aligned);
2415 __ shrptr(size, LogBytesPerInt); // size => int_count
2416 __ jump(RuntimeAddress(int_copy_entry));
2417
2418 __ BIND(L_long_aligned);
2419 __ shrptr(size, LogBytesPerLong); // size => qword_count
2420 __ jump(RuntimeAddress(long_copy_entry));
2421
2422 return start;
2423 }
2424
2425 // Perform range checks on the proposed arraycopy.
2426 // Kills temp, but nothing else.
2427 // Also, clean the sign bits of src_pos and dst_pos.
2428 void arraycopy_range_checks(Register src, // source array oop (c_rarg0)
2429 Register src_pos, // source position (c_rarg1)
2430 Register dst, // destination array oo (c_rarg2)
2431 Register dst_pos, // destination position (c_rarg3)
2432 Register length,
2433 Register temp,
2434 Label& L_failed) {
2435 BLOCK_COMMENT("arraycopy_range_checks:");
2436
2437 // if (src_pos + length > arrayOop(src)->length()) FAIL;
2438 __ movl(temp, length);
2439 __ addl(temp, src_pos); // src_pos + length
2440 __ cmpl(temp, Address(src, arrayOopDesc::length_offset_in_bytes()));
2441 __ jcc(Assembler::above, L_failed);
2442
2443 // if (dst_pos + length > arrayOop(dst)->length()) FAIL;
2444 __ movl(temp, length);
2445 __ addl(temp, dst_pos); // dst_pos + length
2446 __ cmpl(temp, Address(dst, arrayOopDesc::length_offset_in_bytes()));
2447 __ jcc(Assembler::above, L_failed);
2448
2449 // Have to clean up high 32-bits of 'src_pos' and 'dst_pos'.
2450 // Move with sign extension can be used since they are positive.
2451 __ movslq(src_pos, src_pos);
2452 __ movslq(dst_pos, dst_pos);
2453
2454 BLOCK_COMMENT("arraycopy_range_checks done");
2455 }
2456
2457 //
2458 // Generate generic array copy stubs
2459 //
2460 // Input:
2461 // c_rarg0 - src oop
2462 // c_rarg1 - src_pos (32-bits)
2463 // c_rarg2 - dst oop
2464 // c_rarg3 - dst_pos (32-bits)
2465 // not Win64
2466 // c_rarg4 - element count (32-bits)
2467 // Win64
2468 // rsp+40 - element count (32-bits)
2469 //
2470 // Output:
2471 // rax == 0 - success
2472 // rax == -1^K - failure, where K is partial transfer count
2473 //
2474 address generate_generic_copy(const char *name,
2475 address byte_copy_entry, address short_copy_entry,
2476 address int_copy_entry, address oop_copy_entry,
2477 address long_copy_entry, address checkcast_copy_entry) {
2478
2479 Label L_failed, L_failed_0, L_objArray;
2480 Label L_copy_bytes, L_copy_shorts, L_copy_ints, L_copy_longs;
2481
2482 // Input registers
2483 const Register src = c_rarg0; // source array oop
2484 const Register src_pos = c_rarg1; // source position
2485 const Register dst = c_rarg2; // destination array oop
2486 const Register dst_pos = c_rarg3; // destination position
2487 #ifndef _WIN64
2488 const Register length = c_rarg4;
2489 #else
2490 const Address length(rsp, 6 * wordSize); // elements count is on stack on Win64
2491 #endif
2492
2493 { int modulus = CodeEntryAlignment;
2494 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
2495 int advance = target - (__ offset() % modulus);
2496 if (advance < 0) advance += modulus;
2497 if (advance > 0) __ nop(advance);
2498 }
2499 StubCodeMark mark(this, "StubRoutines", name);
2500
2501 // Short-hop target to L_failed. Makes for denser prologue code.
2502 __ BIND(L_failed_0);
2503 __ jmp(L_failed);
2504 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
2505
2506 __ align(CodeEntryAlignment);
2507 address start = __ pc();
2508
2509 __ enter(); // required for proper stackwalking of RuntimeStub frame
2510
2511 // bump this on entry, not on exit:
2512 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
2513
2514 //-----------------------------------------------------------------------
2515 // Assembler stub will be used for this call to arraycopy
2516 // if the following conditions are met:
2517 //
2518 // (1) src and dst must not be null.
2519 // (2) src_pos must not be negative.
2520 // (3) dst_pos must not be negative.
2521 // (4) length must not be negative.
2522 // (5) src klass and dst klass should be the same and not NULL.
2523 // (6) src and dst should be arrays.
2524 // (7) src_pos + length must not exceed length of src.
2525 // (8) dst_pos + length must not exceed length of dst.
2526 //
2527
2528 // if (src == NULL) return -1;
2529 __ testptr(src, src); // src oop
2530 size_t j1off = __ offset();
2531 __ jccb(Assembler::zero, L_failed_0);
2532
2533 // if (src_pos < 0) return -1;
2534 __ testl(src_pos, src_pos); // src_pos (32-bits)
2535 __ jccb(Assembler::negative, L_failed_0);
2536
2537 // if (dst == NULL) return -1;
2538 __ testptr(dst, dst); // dst oop
2539 __ jccb(Assembler::zero, L_failed_0);
2540
2541 // if (dst_pos < 0) return -1;
2542 __ testl(dst_pos, dst_pos); // dst_pos (32-bits)
2543 size_t j4off = __ offset();
2544 __ jccb(Assembler::negative, L_failed_0);
2545
2546 // The first four tests are very dense code,
2547 // but not quite dense enough to put four
2548 // jumps in a 16-byte instruction fetch buffer.
2549 // That's good, because some branch predicters
2550 // do not like jumps so close together.
2551 // Make sure of this.
2552 guarantee(((j1off ^ j4off) & ~15) != 0, "I$ line of 1st & 4th jumps");
2553
2554 // registers used as temp
2555 const Register r11_length = r11; // elements count to copy
2556 const Register r10_src_klass = r10; // array klass
2557
2558 // if (length < 0) return -1;
2559 __ movl(r11_length, length); // length (elements count, 32-bits value)
2560 __ testl(r11_length, r11_length);
2561 __ jccb(Assembler::negative, L_failed_0);
2562
2563 __ load_klass(r10_src_klass, src);
2564 #ifdef ASSERT
2565 // assert(src->klass() != NULL);
2566 {
2567 BLOCK_COMMENT("assert klasses not null {");
2568 Label L1, L2;
2569 __ testptr(r10_src_klass, r10_src_klass);
2570 __ jcc(Assembler::notZero, L2); // it is broken if klass is NULL
2571 __ bind(L1);
2572 __ stop("broken null klass");
2573 __ bind(L2);
2574 __ load_klass(rax, dst);
2575 __ cmpq(rax, 0);
2576 __ jcc(Assembler::equal, L1); // this would be broken also
2577 BLOCK_COMMENT("} assert klasses not null done");
2578 }
2579 #endif
2580
2581 // Load layout helper (32-bits)
2582 //
2583 // |array_tag| | header_size | element_type | |log2_element_size|
2584 // 32 30 24 16 8 2 0
2585 //
2586 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
2587 //
2588
2589 const int lh_offset = in_bytes(Klass::layout_helper_offset());
2590
2591 // Handle objArrays completely differently...
2592 const jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2593 __ cmpl(Address(r10_src_klass, lh_offset), objArray_lh);
2594 __ jcc(Assembler::equal, L_objArray);
2595
2596 // if (src->klass() != dst->klass()) return -1;
2597 __ load_klass(rax, dst);
2598 __ cmpq(r10_src_klass, rax);
2599 __ jcc(Assembler::notEqual, L_failed);
2600
2601 const Register rax_lh = rax; // layout helper
2602 __ movl(rax_lh, Address(r10_src_klass, lh_offset));
2603
2604 // if (!src->is_Array()) return -1;
2605 __ cmpl(rax_lh, Klass::_lh_neutral_value);
2606 __ jcc(Assembler::greaterEqual, L_failed);
2607
2608 // At this point, it is known to be a typeArray (array_tag 0x3).
2609 #ifdef ASSERT
2610 {
2611 BLOCK_COMMENT("assert primitive array {");
2612 Label L;
2613 __ cmpl(rax_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
2614 __ jcc(Assembler::greaterEqual, L);
2615 __ stop("must be a primitive array");
2616 __ bind(L);
2617 BLOCK_COMMENT("} assert primitive array done");
2618 }
2619 #endif
2620
2621 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2622 r10, L_failed);
2623
2624 // typeArrayKlass
2625 //
2626 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
2627 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
2628 //
2629
2630 const Register r10_offset = r10; // array offset
2631 const Register rax_elsize = rax_lh; // element size
2632
2633 __ movl(r10_offset, rax_lh);
2634 __ shrl(r10_offset, Klass::_lh_header_size_shift);
2635 __ andptr(r10_offset, Klass::_lh_header_size_mask); // array_offset
2636 __ addptr(src, r10_offset); // src array offset
2637 __ addptr(dst, r10_offset); // dst array offset
2638 BLOCK_COMMENT("choose copy loop based on element size");
2639 __ andl(rax_lh, Klass::_lh_log2_element_size_mask); // rax_lh -> rax_elsize
2640
2641 // next registers should be set before the jump to corresponding stub
2642 const Register from = c_rarg0; // source array address
2643 const Register to = c_rarg1; // destination array address
2644 const Register count = c_rarg2; // elements count
2645
2646 // 'from', 'to', 'count' registers should be set in such order
2647 // since they are the same as 'src', 'src_pos', 'dst'.
2648
2649 __ BIND(L_copy_bytes);
2650 __ cmpl(rax_elsize, 0);
2651 __ jccb(Assembler::notEqual, L_copy_shorts);
2652 __ lea(from, Address(src, src_pos, Address::times_1, 0));// src_addr
2653 __ lea(to, Address(dst, dst_pos, Address::times_1, 0));// dst_addr
2654 __ movl2ptr(count, r11_length); // length
2655 __ jump(RuntimeAddress(byte_copy_entry));
2656
2657 __ BIND(L_copy_shorts);
2658 __ cmpl(rax_elsize, LogBytesPerShort);
2659 __ jccb(Assembler::notEqual, L_copy_ints);
2660 __ lea(from, Address(src, src_pos, Address::times_2, 0));// src_addr
2661 __ lea(to, Address(dst, dst_pos, Address::times_2, 0));// dst_addr
2662 __ movl2ptr(count, r11_length); // length
2663 __ jump(RuntimeAddress(short_copy_entry));
2664
2665 __ BIND(L_copy_ints);
2666 __ cmpl(rax_elsize, LogBytesPerInt);
2667 __ jccb(Assembler::notEqual, L_copy_longs);
2668 __ lea(from, Address(src, src_pos, Address::times_4, 0));// src_addr
2669 __ lea(to, Address(dst, dst_pos, Address::times_4, 0));// dst_addr
2670 __ movl2ptr(count, r11_length); // length
2671 __ jump(RuntimeAddress(int_copy_entry));
2672
2673 __ BIND(L_copy_longs);
2674 #ifdef ASSERT
2675 {
2676 BLOCK_COMMENT("assert long copy {");
2677 Label L;
2678 __ cmpl(rax_elsize, LogBytesPerLong);
2679 __ jcc(Assembler::equal, L);
2680 __ stop("must be long copy, but elsize is wrong");
2681 __ bind(L);
2682 BLOCK_COMMENT("} assert long copy done");
2683 }
2684 #endif
2685 __ lea(from, Address(src, src_pos, Address::times_8, 0));// src_addr
2686 __ lea(to, Address(dst, dst_pos, Address::times_8, 0));// dst_addr
2687 __ movl2ptr(count, r11_length); // length
2688 __ jump(RuntimeAddress(long_copy_entry));
2689
2690 // objArrayKlass
2691 __ BIND(L_objArray);
2692 // live at this point: r10_src_klass, r11_length, src[_pos], dst[_pos]
2693
2694 Label L_plain_copy, L_checkcast_copy;
2695 // test array classes for subtyping
2696 __ load_klass(rax, dst);
2697 __ cmpq(r10_src_klass, rax); // usual case is exact equality
2698 __ jcc(Assembler::notEqual, L_checkcast_copy);
2699
2700 // Identically typed arrays can be copied without element-wise checks.
2701 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2702 r10, L_failed);
2703
2704 __ lea(from, Address(src, src_pos, TIMES_OOP,
2705 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
2706 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2707 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
2708 __ movl2ptr(count, r11_length); // length
2709 __ BIND(L_plain_copy);
2710 __ jump(RuntimeAddress(oop_copy_entry));
2711
2712 __ BIND(L_checkcast_copy);
2713 // live at this point: r10_src_klass, r11_length, rax (dst_klass)
2714 {
2715 // Before looking at dst.length, make sure dst is also an objArray.
2716 __ cmpl(Address(rax, lh_offset), objArray_lh);
2717 __ jcc(Assembler::notEqual, L_failed);
2718
2719 // It is safe to examine both src.length and dst.length.
2720 arraycopy_range_checks(src, src_pos, dst, dst_pos, r11_length,
2721 rax, L_failed);
2722
2723 const Register r11_dst_klass = r11;
2724 __ load_klass(r11_dst_klass, dst); // reload
2725
2726 // Marshal the base address arguments now, freeing registers.
2727 __ lea(from, Address(src, src_pos, TIMES_OOP,
2728 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2729 __ lea(to, Address(dst, dst_pos, TIMES_OOP,
2730 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
2731 __ movl(count, length); // length (reloaded)
2732 Register sco_temp = c_rarg3; // this register is free now
2733 assert_different_registers(from, to, count, sco_temp,
2734 r11_dst_klass, r10_src_klass);
2735 assert_clean_int(count, sco_temp);
2736
2737 // Generate the type check.
2738 const int sco_offset = in_bytes(Klass::super_check_offset_offset());
2739 __ movl(sco_temp, Address(r11_dst_klass, sco_offset));
2740 assert_clean_int(sco_temp, rax);
2741 generate_type_check(r10_src_klass, sco_temp, r11_dst_klass, L_plain_copy);
2742
2743 // Fetch destination element klass from the objArrayKlass header.
2744 int ek_offset = in_bytes(objArrayKlass::element_klass_offset());
2745 __ movptr(r11_dst_klass, Address(r11_dst_klass, ek_offset));
2746 __ movl( sco_temp, Address(r11_dst_klass, sco_offset));
2747 assert_clean_int(sco_temp, rax);
2748
2749 // the checkcast_copy loop needs two extra arguments:
2750 assert(c_rarg3 == sco_temp, "#3 already in place");
2751 // Set up arguments for checkcast_copy_entry.
2752 setup_arg_regs(4);
2753 __ movptr(r8, r11_dst_klass); // dst.klass.element_klass, r8 is c_rarg4 on Linux/Solaris
2754 __ jump(RuntimeAddress(checkcast_copy_entry));
2755 }
2756
2757 __ BIND(L_failed);
2758 __ xorptr(rax, rax);
2759 __ notptr(rax); // return -1
2760 __ leave(); // required for proper stackwalking of RuntimeStub frame
2761 __ ret(0);
2762
2763 return start;
2764 }
2765
2766 void generate_arraycopy_stubs() {
2767 address entry;
2768 address entry_jbyte_arraycopy;
2769 address entry_jshort_arraycopy;
2770 address entry_jint_arraycopy;
2771 address entry_oop_arraycopy;
2772 address entry_jlong_arraycopy;
2773 address entry_checkcast_arraycopy;
2774
2775 StubRoutines::_jbyte_disjoint_arraycopy = generate_disjoint_byte_copy(false, &entry,
2776 "jbyte_disjoint_arraycopy");
2777 StubRoutines::_jbyte_arraycopy = generate_conjoint_byte_copy(false, entry, &entry_jbyte_arraycopy,
2778 "jbyte_arraycopy");
2779
2780 StubRoutines::_jshort_disjoint_arraycopy = generate_disjoint_short_copy(false, &entry,
2781 "jshort_disjoint_arraycopy");
2782 StubRoutines::_jshort_arraycopy = generate_conjoint_short_copy(false, entry, &entry_jshort_arraycopy,
2783 "jshort_arraycopy");
2784
2785 StubRoutines::_jint_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, false, &entry,
2786 "jint_disjoint_arraycopy");
2787 StubRoutines::_jint_arraycopy = generate_conjoint_int_oop_copy(false, false, entry,
2788 &entry_jint_arraycopy, "jint_arraycopy");
2789
2790 StubRoutines::_jlong_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, false, &entry,
2791 "jlong_disjoint_arraycopy");
2792 StubRoutines::_jlong_arraycopy = generate_conjoint_long_oop_copy(false, false, entry,
2793 &entry_jlong_arraycopy, "jlong_arraycopy");
2794
2795
2796 if (UseCompressedOops) {
2797 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_int_oop_copy(false, true, &entry,
2798 "oop_disjoint_arraycopy");
2799 StubRoutines::_oop_arraycopy = generate_conjoint_int_oop_copy(false, true, entry,
2800 &entry_oop_arraycopy, "oop_arraycopy");
2801 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_int_oop_copy(false, true, &entry,
2802 "oop_disjoint_arraycopy_uninit",
2803 /*dest_uninitialized*/true);
2804 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_int_oop_copy(false, true, entry,
2805 NULL, "oop_arraycopy_uninit",
2806 /*dest_uninitialized*/true);
2807 } else {
2808 StubRoutines::_oop_disjoint_arraycopy = generate_disjoint_long_oop_copy(false, true, &entry,
2809 "oop_disjoint_arraycopy");
2810 StubRoutines::_oop_arraycopy = generate_conjoint_long_oop_copy(false, true, entry,
2811 &entry_oop_arraycopy, "oop_arraycopy");
2812 StubRoutines::_oop_disjoint_arraycopy_uninit = generate_disjoint_long_oop_copy(false, true, &entry,
2813 "oop_disjoint_arraycopy_uninit",
2814 /*dest_uninitialized*/true);
2815 StubRoutines::_oop_arraycopy_uninit = generate_conjoint_long_oop_copy(false, true, entry,
2816 NULL, "oop_arraycopy_uninit",
2817 /*dest_uninitialized*/true);
2818 }
2819
2820 StubRoutines::_checkcast_arraycopy = generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2821 StubRoutines::_checkcast_arraycopy_uninit = generate_checkcast_copy("checkcast_arraycopy_uninit", NULL,
2822 /*dest_uninitialized*/true);
2823
2824 StubRoutines::_unsafe_arraycopy = generate_unsafe_copy("unsafe_arraycopy",
2825 entry_jbyte_arraycopy,
2826 entry_jshort_arraycopy,
2827 entry_jint_arraycopy,
2828 entry_jlong_arraycopy);
2829 StubRoutines::_generic_arraycopy = generate_generic_copy("generic_arraycopy",
2830 entry_jbyte_arraycopy,
2831 entry_jshort_arraycopy,
2832 entry_jint_arraycopy,
2833 entry_oop_arraycopy,
2834 entry_jlong_arraycopy,
2835 entry_checkcast_arraycopy);
2836
2837 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2838 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2839 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2840 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2841 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2842 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2843
2844 // We don't generate specialized code for HeapWord-aligned source
2845 // arrays, so just use the code we've already generated
2846 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = StubRoutines::_jbyte_disjoint_arraycopy;
2847 StubRoutines::_arrayof_jbyte_arraycopy = StubRoutines::_jbyte_arraycopy;
2848
2849 StubRoutines::_arrayof_jshort_disjoint_arraycopy = StubRoutines::_jshort_disjoint_arraycopy;
2850 StubRoutines::_arrayof_jshort_arraycopy = StubRoutines::_jshort_arraycopy;
2851
2852 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2853 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2854
2855 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2856 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2857
2858 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2859 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2860
2861 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2862 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2863 }
2864
2865 void generate_math_stubs() {
2866 {
2867 StubCodeMark mark(this, "StubRoutines", "log");
2868 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2869
2870 __ subq(rsp, 8);
2871 __ movdbl(Address(rsp, 0), xmm0);
2872 __ fld_d(Address(rsp, 0));
2873 __ flog();
2874 __ fstp_d(Address(rsp, 0));
2875 __ movdbl(xmm0, Address(rsp, 0));
2876 __ addq(rsp, 8);
2877 __ ret(0);
2878 }
2879 {
2880 StubCodeMark mark(this, "StubRoutines", "log10");
2881 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2882
2883 __ subq(rsp, 8);
2884 __ movdbl(Address(rsp, 0), xmm0);
2885 __ fld_d(Address(rsp, 0));
2886 __ flog10();
2887 __ fstp_d(Address(rsp, 0));
2888 __ movdbl(xmm0, Address(rsp, 0));
2889 __ addq(rsp, 8);
2890 __ ret(0);
2891 }
2892 {
2893 StubCodeMark mark(this, "StubRoutines", "sin");
2894 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2895
2896 __ subq(rsp, 8);
2897 __ movdbl(Address(rsp, 0), xmm0);
2898 __ fld_d(Address(rsp, 0));
2899 __ trigfunc('s');
2900 __ fstp_d(Address(rsp, 0));
2901 __ movdbl(xmm0, Address(rsp, 0));
2902 __ addq(rsp, 8);
2903 __ ret(0);
2904 }
2905 {
2906 StubCodeMark mark(this, "StubRoutines", "cos");
2907 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2908
2909 __ subq(rsp, 8);
2910 __ movdbl(Address(rsp, 0), xmm0);
2911 __ fld_d(Address(rsp, 0));
2912 __ trigfunc('c');
2913 __ fstp_d(Address(rsp, 0));
2914 __ movdbl(xmm0, Address(rsp, 0));
2915 __ addq(rsp, 8);
2916 __ ret(0);
2917 }
2918 {
2919 StubCodeMark mark(this, "StubRoutines", "tan");
2920 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2921
2922 __ subq(rsp, 8);
2923 __ movdbl(Address(rsp, 0), xmm0);
2924 __ fld_d(Address(rsp, 0));
2925 __ trigfunc('t');
2926 __ fstp_d(Address(rsp, 0));
2927 __ movdbl(xmm0, Address(rsp, 0));
2928 __ addq(rsp, 8);
2929 __ ret(0);
2930 }
2931
2932 // The intrinsic version of these seem to return the same value as
2933 // the strict version.
2934 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2935 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2936 }
2937
2938 #undef __
2939 #define __ masm->
2940
2941 // Continuation point for throwing of implicit exceptions that are
2942 // not handled in the current activation. Fabricates an exception
2943 // oop and initiates normal exception dispatching in this
2944 // frame. Since we need to preserve callee-saved values (currently
2945 // only for C2, but done for C1 as well) we need a callee-saved oop
2946 // map and therefore have to make these stubs into RuntimeStubs
2947 // rather than BufferBlobs. If the compiler needs all registers to
2948 // be preserved between the fault point and the exception handler
2949 // then it must assume responsibility for that in
2950 // AbstractCompiler::continuation_for_implicit_null_exception or
2951 // continuation_for_implicit_division_by_zero_exception. All other
2952 // implicit exceptions (e.g., NullPointerException or
2953 // AbstractMethodError on entry) are either at call sites or
2954 // otherwise assume that stack unwinding will be initiated, so
2955 // caller saved registers were assumed volatile in the compiler.
2956 address generate_throw_exception(const char* name,
2957 address runtime_entry,
2958 Register arg1 = noreg,
2959 Register arg2 = noreg) {
2960 // Information about frame layout at time of blocking runtime call.
2961 // Note that we only have to preserve callee-saved registers since
2962 // the compilers are responsible for supplying a continuation point
2963 // if they expect all registers to be preserved.
2964 enum layout {
2965 rbp_off = frame::arg_reg_save_area_bytes/BytesPerInt,
2966 rbp_off2,
2967 return_off,
2968 return_off2,
2969 framesize // inclusive of return address
2970 };
2971
2972 int insts_size = 512;
2973 int locs_size = 64;
2974
2975 CodeBuffer code(name, insts_size, locs_size);
2976 OopMapSet* oop_maps = new OopMapSet();
2977 MacroAssembler* masm = new MacroAssembler(&code);
2978
2979 address start = __ pc();
2980
2981 // This is an inlined and slightly modified version of call_VM
2982 // which has the ability to fetch the return PC out of
2983 // thread-local storage and also sets up last_Java_sp slightly
2984 // differently than the real call_VM
2985
2986 __ enter(); // required for proper stackwalking of RuntimeStub frame
2987
2988 assert(is_even(framesize/2), "sp not 16-byte aligned");
2989
2990 // return address and rbp are already in place
2991 __ subptr(rsp, (framesize-4) << LogBytesPerInt); // prolog
2992
2993 int frame_complete = __ pc() - start;
2994
2995 // Set up last_Java_sp and last_Java_fp
2996 address the_pc = __ pc();
2997 __ set_last_Java_frame(rsp, rbp, the_pc);
2998 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
2999
3000 // Call runtime
3001 if (arg1 != noreg) {
3002 assert(arg2 != c_rarg1, "clobbered");
3003 __ movptr(c_rarg1, arg1);
3004 }
3005 if (arg2 != noreg) {
3006 __ movptr(c_rarg2, arg2);
3007 }
3008 __ movptr(c_rarg0, r15_thread);
3009 BLOCK_COMMENT("call runtime_entry");
3010 __ call(RuntimeAddress(runtime_entry));
3011
3012 // Generate oop map
3013 OopMap* map = new OopMap(framesize, 0);
3014
3015 oop_maps->add_gc_map(the_pc - start, map);
3016
3017 __ reset_last_Java_frame(true, true);
3018
3019 __ leave(); // required for proper stackwalking of RuntimeStub frame
3020
3021 // check for pending exceptions
3022 #ifdef ASSERT
3023 Label L;
3024 __ cmpptr(Address(r15_thread, Thread::pending_exception_offset()),
3025 (int32_t) NULL_WORD);
3026 __ jcc(Assembler::notEqual, L);
3027 __ should_not_reach_here();
3028 __ bind(L);
3029 #endif // ASSERT
3030 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
3031
3032
3033 // codeBlob framesize is in words (not VMRegImpl::slot_size)
3034 RuntimeStub* stub =
3035 RuntimeStub::new_runtime_stub(name,
3036 &code,
3037 frame_complete,
3038 (framesize >> (LogBytesPerWord - LogBytesPerInt)),
3039 oop_maps, false);
3040 return stub->entry_point();
3041 }
3042
3043 // Initialization
3044 void generate_initial() {
3045 // Generates all stubs and initializes the entry points
3046
3047 // This platform-specific stub is needed by generate_call_stub()
3048 StubRoutines::x86::_mxcsr_std = generate_fp_mask("mxcsr_std", 0x0000000000001F80);
3049
3050 // entry points that exist in all platforms Note: This is code
3051 // that could be shared among different platforms - however the
3052 // benefit seems to be smaller than the disadvantage of having a
3053 // much more complicated generator structure. See also comment in
3054 // stubRoutines.hpp.
3055
3056 StubRoutines::_forward_exception_entry = generate_forward_exception();
3057
3058 StubRoutines::_call_stub_entry =
3059 generate_call_stub(StubRoutines::_call_stub_return_address);
3060
3061 // is referenced by megamorphic call
3062 StubRoutines::_catch_exception_entry = generate_catch_exception();
3063
3064 // atomic calls
3065 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
3066 StubRoutines::_atomic_xchg_ptr_entry = generate_atomic_xchg_ptr();
3067 StubRoutines::_atomic_cmpxchg_entry = generate_atomic_cmpxchg();
3068 StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
3069 StubRoutines::_atomic_add_entry = generate_atomic_add();
3070 StubRoutines::_atomic_add_ptr_entry = generate_atomic_add_ptr();
3071 StubRoutines::_fence_entry = generate_orderaccess_fence();
3072
3073 StubRoutines::_handler_for_unsafe_access_entry =
3074 generate_handler_for_unsafe_access();
3075
3076 // platform dependent
3077 StubRoutines::x86::_get_previous_fp_entry = generate_get_previous_fp();
3078 StubRoutines::x86::_get_previous_sp_entry = generate_get_previous_sp();
3079
3080 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
3081
3082 // Build this early so it's available for the interpreter. Stub
3083 // expects the required and actual types as register arguments in
3084 // j_rarg0 and j_rarg1 respectively.
3085 StubRoutines::_throw_WrongMethodTypeException_entry =
3086 generate_throw_exception("WrongMethodTypeException throw_exception",
3087 CAST_FROM_FN_PTR(address, SharedRuntime::throw_WrongMethodTypeException),
3088 rax, rcx);
3089
3090 // Build this early so it's available for the interpreter.
3091 StubRoutines::_throw_StackOverflowError_entry =
3092 generate_throw_exception("StackOverflowError throw_exception",
3093 CAST_FROM_FN_PTR(address,
3094 SharedRuntime::
3095 throw_StackOverflowError));
3096 }
3097
3098 void generate_all() {
3099 // Generates all stubs and initializes the entry points
3100
3101 // These entry points require SharedInfo::stack0 to be set up in
3102 // non-core builds and need to be relocatable, so they each
3103 // fabricate a RuntimeStub internally.
3104 StubRoutines::_throw_AbstractMethodError_entry =
3105 generate_throw_exception("AbstractMethodError throw_exception",
3106 CAST_FROM_FN_PTR(address,
3107 SharedRuntime::
3108 throw_AbstractMethodError));
3109
3110 StubRoutines::_throw_IncompatibleClassChangeError_entry =
3111 generate_throw_exception("IncompatibleClassChangeError throw_exception",
3112 CAST_FROM_FN_PTR(address,
3113 SharedRuntime::
3114 throw_IncompatibleClassChangeError));
3115
3116 StubRoutines::_throw_NullPointerException_at_call_entry =
3117 generate_throw_exception("NullPointerException at call throw_exception",
3118 CAST_FROM_FN_PTR(address,
3119 SharedRuntime::
3120 throw_NullPointerException_at_call));
3121
3122 // entry points that are platform specific
3123 StubRoutines::x86::_f2i_fixup = generate_f2i_fixup();
3124 StubRoutines::x86::_f2l_fixup = generate_f2l_fixup();
3125 StubRoutines::x86::_d2i_fixup = generate_d2i_fixup();
3126 StubRoutines::x86::_d2l_fixup = generate_d2l_fixup();
3127
3128 StubRoutines::x86::_float_sign_mask = generate_fp_mask("float_sign_mask", 0x7FFFFFFF7FFFFFFF);
3129 StubRoutines::x86::_float_sign_flip = generate_fp_mask("float_sign_flip", 0x8000000080000000);
3130 StubRoutines::x86::_double_sign_mask = generate_fp_mask("double_sign_mask", 0x7FFFFFFFFFFFFFFF);
3131 StubRoutines::x86::_double_sign_flip = generate_fp_mask("double_sign_flip", 0x8000000000000000);
3132
3133 // support for verify_oop (must happen after universe_init)
3134 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3135
3136 // arraycopy stubs used by compilers
3137 generate_arraycopy_stubs();
3138
3139 generate_math_stubs();
3140 }
3141
3142 public:
3143 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3144 if (all) {
3145 generate_all();
3146 } else {
3147 generate_initial();
3148 }
3149 }
3150 }; // end class declaration
3151
3152 void StubGenerator_generate(CodeBuffer* code, bool all) {
3153 StubGenerator g(code, all);
3154 }