1 /*
2 * Copyright (c) 1999, 2013, 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/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/method.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 "runtime/thread.inline.hpp"
41 #include "utilities/top.hpp"
42 #ifdef COMPILER2
43 #include "opto/runtime.hpp"
44 #endif
45
46 // Declaration and definition of StubGenerator (no .hpp file).
47 // For a more detailed description of the stub routine structure
48 // see the comment in stubRoutines.hpp
49
50 #define __ _masm->
51 #define a__ ((Assembler*)_masm)->
52
53 #ifdef PRODUCT
54 #define BLOCK_COMMENT(str) /* nothing */
55 #else
56 #define BLOCK_COMMENT(str) __ block_comment(str)
57 #endif
58
59 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
60
61 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
62 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
63
64 // -------------------------------------------------------------------------------------------------------------------------
65 // Stub Code definitions
66
67 static address handle_unsafe_access() {
68 JavaThread* thread = JavaThread::current();
69 address pc = thread->saved_exception_pc();
70 // pc is the instruction which we must emulate
71 // doing a no-op is fine: return garbage from the load
72 // therefore, compute npc
73 address npc = Assembler::locate_next_instruction(pc);
74
75 // request an async exception
76 thread->set_pending_unsafe_access_error();
77
78 // return address of next instruction to execute
79 return npc;
80 }
81
82 class StubGenerator: public StubCodeGenerator {
83 private:
84
85 #ifdef PRODUCT
86 #define inc_counter_np(counter) ((void)0)
87 #else
88 void inc_counter_np_(int& counter) {
89 __ incrementl(ExternalAddress((address)&counter));
90 }
91 #define inc_counter_np(counter) \
92 BLOCK_COMMENT("inc_counter " #counter); \
93 inc_counter_np_(counter);
94 #endif //PRODUCT
95
96 void inc_copy_counter_np(BasicType t) {
97 #ifndef PRODUCT
98 switch (t) {
99 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
100 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
101 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
102 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
103 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
104 }
105 ShouldNotReachHere();
106 #endif //PRODUCT
107 }
108
109 //------------------------------------------------------------------------------------------------------------------------
110 // Call stubs are used to call Java from C
111 //
112 // [ return_from_Java ] <--- rsp
113 // [ argument word n ]
114 // ...
115 // -N [ argument word 1 ]
116 // -7 [ Possible padding for stack alignment ]
117 // -6 [ Possible padding for stack alignment ]
118 // -5 [ Possible padding for stack alignment ]
119 // -4 [ mxcsr save ] <--- rsp_after_call
120 // -3 [ saved rbx, ]
121 // -2 [ saved rsi ]
122 // -1 [ saved rdi ]
123 // 0 [ saved rbp, ] <--- rbp,
124 // 1 [ return address ]
125 // 2 [ ptr. to call wrapper ]
126 // 3 [ result ]
127 // 4 [ result_type ]
128 // 5 [ method ]
129 // 6 [ entry_point ]
130 // 7 [ parameters ]
131 // 8 [ parameter_size ]
132 // 9 [ thread ]
133
134
135 address generate_call_stub(address& return_address) {
136 StubCodeMark mark(this, "StubRoutines", "call_stub");
137 address start = __ pc();
138
139 // stub code parameters / addresses
140 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
141 bool sse_save = false;
142 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
143 const int locals_count_in_bytes (4*wordSize);
144 const Address mxcsr_save (rbp, -4 * wordSize);
145 const Address saved_rbx (rbp, -3 * wordSize);
146 const Address saved_rsi (rbp, -2 * wordSize);
147 const Address saved_rdi (rbp, -1 * wordSize);
148 const Address result (rbp, 3 * wordSize);
149 const Address result_type (rbp, 4 * wordSize);
150 const Address method (rbp, 5 * wordSize);
151 const Address entry_point (rbp, 6 * wordSize);
152 const Address parameters (rbp, 7 * wordSize);
153 const Address parameter_size(rbp, 8 * wordSize);
154 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
155 sse_save = UseSSE > 0;
156
157 // stub code
158 __ enter();
159 __ movptr(rcx, parameter_size); // parameter counter
160 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
161 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
162 __ subptr(rsp, rcx);
163 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
164
165 // save rdi, rsi, & rbx, according to C calling conventions
166 __ movptr(saved_rdi, rdi);
167 __ movptr(saved_rsi, rsi);
168 __ movptr(saved_rbx, rbx);
169 // save and initialize %mxcsr
170 if (sse_save) {
171 Label skip_ldmx;
172 __ stmxcsr(mxcsr_save);
173 __ movl(rax, mxcsr_save);
174 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
175 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
176 __ cmp32(rax, mxcsr_std);
177 __ jcc(Assembler::equal, skip_ldmx);
178 __ ldmxcsr(mxcsr_std);
179 __ bind(skip_ldmx);
180 }
181
182 // make sure the control word is correct.
183 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
184
185 #ifdef ASSERT
186 // make sure we have no pending exceptions
187 { Label L;
188 __ movptr(rcx, thread);
189 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
190 __ jcc(Assembler::equal, L);
191 __ stop("StubRoutines::call_stub: entered with pending exception");
192 __ bind(L);
193 }
194 #endif
195
196 // pass parameters if any
197 BLOCK_COMMENT("pass parameters if any");
198 Label parameters_done;
199 __ movl(rcx, parameter_size); // parameter counter
200 __ testl(rcx, rcx);
201 __ jcc(Assembler::zero, parameters_done);
202
203 // parameter passing loop
204
205 Label loop;
206 // Copy Java parameters in reverse order (receiver last)
207 // Note that the argument order is inverted in the process
208 // source is rdx[rcx: N-1..0]
209 // dest is rsp[rbx: 0..N-1]
210
211 __ movptr(rdx, parameters); // parameter pointer
212 __ xorptr(rbx, rbx);
213
214 __ BIND(loop);
215
216 // get parameter
217 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
218 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
219 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
220 __ increment(rbx);
221 __ decrement(rcx);
222 __ jcc(Assembler::notZero, loop);
223
224 // call Java function
225 __ BIND(parameters_done);
226 __ movptr(rbx, method); // get Method*
227 __ movptr(rax, entry_point); // get entry_point
228 __ mov(rsi, rsp); // set sender sp
229 BLOCK_COMMENT("call Java function");
230 __ call(rax);
231
232 BLOCK_COMMENT("call_stub_return_address:");
233 return_address = __ pc();
234
235 #ifdef COMPILER2
236 {
237 Label L_skip;
238 if (UseSSE >= 2) {
239 __ verify_FPU(0, "call_stub_return");
240 } else {
241 for (int i = 1; i < 8; i++) {
242 __ ffree(i);
243 }
244
245 // UseSSE <= 1 so double result should be left on TOS
246 __ movl(rsi, result_type);
247 __ cmpl(rsi, T_DOUBLE);
248 __ jcc(Assembler::equal, L_skip);
249 if (UseSSE == 0) {
250 // UseSSE == 0 so float result should be left on TOS
251 __ cmpl(rsi, T_FLOAT);
252 __ jcc(Assembler::equal, L_skip);
253 }
254 __ ffree(0);
255 }
256 __ BIND(L_skip);
257 }
258 #endif // COMPILER2
259
260 // store result depending on type
261 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
262 __ movptr(rdi, result);
263 Label is_long, is_float, is_double, exit;
264 __ movl(rsi, result_type);
265 __ cmpl(rsi, T_LONG);
266 __ jcc(Assembler::equal, is_long);
267 __ cmpl(rsi, T_FLOAT);
268 __ jcc(Assembler::equal, is_float);
269 __ cmpl(rsi, T_DOUBLE);
270 __ jcc(Assembler::equal, is_double);
271
272 // handle T_INT case
273 __ movl(Address(rdi, 0), rax);
274 __ BIND(exit);
275
276 // check that FPU stack is empty
277 __ verify_FPU(0, "generate_call_stub");
278
279 // pop parameters
280 __ lea(rsp, rsp_after_call);
281
282 // restore %mxcsr
283 if (sse_save) {
284 __ ldmxcsr(mxcsr_save);
285 }
286
287 // restore rdi, rsi and rbx,
288 __ movptr(rbx, saved_rbx);
289 __ movptr(rsi, saved_rsi);
290 __ movptr(rdi, saved_rdi);
291 __ addptr(rsp, 4*wordSize);
292
293 // return
294 __ pop(rbp);
295 __ ret(0);
296
297 // handle return types different from T_INT
298 __ BIND(is_long);
299 __ movl(Address(rdi, 0 * wordSize), rax);
300 __ movl(Address(rdi, 1 * wordSize), rdx);
301 __ jmp(exit);
302
303 __ BIND(is_float);
304 // interpreter uses xmm0 for return values
305 if (UseSSE >= 1) {
306 __ movflt(Address(rdi, 0), xmm0);
307 } else {
308 __ fstp_s(Address(rdi, 0));
309 }
310 __ jmp(exit);
311
312 __ BIND(is_double);
313 // interpreter uses xmm0 for return values
314 if (UseSSE >= 2) {
315 __ movdbl(Address(rdi, 0), xmm0);
316 } else {
317 __ fstp_d(Address(rdi, 0));
318 }
319 __ jmp(exit);
320
321 return start;
322 }
323
324
325 //------------------------------------------------------------------------------------------------------------------------
326 // Return point for a Java call if there's an exception thrown in Java code.
327 // The exception is caught and transformed into a pending exception stored in
328 // JavaThread that can be tested from within the VM.
329 //
330 // Note: Usually the parameters are removed by the callee. In case of an exception
331 // crossing an activation frame boundary, that is not the case if the callee
332 // is compiled code => need to setup the rsp.
333 //
334 // rax,: exception oop
335
336 address generate_catch_exception() {
337 StubCodeMark mark(this, "StubRoutines", "catch_exception");
338 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
339 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
340 address start = __ pc();
341
342 // get thread directly
343 __ movptr(rcx, thread);
344 #ifdef ASSERT
345 // verify that threads correspond
346 { Label L;
347 __ get_thread(rbx);
348 __ cmpptr(rbx, rcx);
349 __ jcc(Assembler::equal, L);
350 __ stop("StubRoutines::catch_exception: threads must correspond");
351 __ bind(L);
352 }
353 #endif
354 // set pending exception
355 __ verify_oop(rax);
356 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
357 __ lea(Address(rcx, Thread::exception_file_offset ()),
358 ExternalAddress((address)__FILE__));
359 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
360 // complete return to VM
361 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
362 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
363
364 return start;
365 }
366
367
368 //------------------------------------------------------------------------------------------------------------------------
369 // Continuation point for runtime calls returning with a pending exception.
370 // The pending exception check happened in the runtime or native call stub.
371 // The pending exception in Thread is converted into a Java-level exception.
372 //
373 // Contract with Java-level exception handlers:
374 // rax: exception
375 // rdx: throwing pc
376 //
377 // NOTE: At entry of this stub, exception-pc must be on stack !!
378
379 address generate_forward_exception() {
380 StubCodeMark mark(this, "StubRoutines", "forward exception");
381 address start = __ pc();
382 const Register thread = rcx;
383
384 // other registers used in this stub
385 const Register exception_oop = rax;
386 const Register handler_addr = rbx;
387 const Register exception_pc = rdx;
388
389 // Upon entry, the sp points to the return address returning into Java
390 // (interpreted or compiled) code; i.e., the return address becomes the
391 // throwing pc.
392 //
393 // Arguments pushed before the runtime call are still on the stack but
394 // the exception handler will reset the stack pointer -> ignore them.
395 // A potential result in registers can be ignored as well.
396
397 #ifdef ASSERT
398 // make sure this code is only executed if there is a pending exception
399 { Label L;
400 __ get_thread(thread);
401 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
402 __ jcc(Assembler::notEqual, L);
403 __ stop("StubRoutines::forward exception: no pending exception (1)");
404 __ bind(L);
405 }
406 #endif
407
408 // compute exception handler into rbx,
409 __ get_thread(thread);
410 __ movptr(exception_pc, Address(rsp, 0));
411 BLOCK_COMMENT("call exception_handler_for_return_address");
412 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
413 __ mov(handler_addr, rax);
414
415 // setup rax & rdx, remove return address & clear pending exception
416 __ get_thread(thread);
417 __ pop(exception_pc);
418 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
419 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
420
421 #ifdef ASSERT
422 // make sure exception is set
423 { Label L;
424 __ testptr(exception_oop, exception_oop);
425 __ jcc(Assembler::notEqual, L);
426 __ stop("StubRoutines::forward exception: no pending exception (2)");
427 __ bind(L);
428 }
429 #endif
430
431 // Verify that there is really a valid exception in RAX.
432 __ verify_oop(exception_oop);
433
434 // continue at exception handler (return address removed)
435 // rax: exception
436 // rbx: exception handler
437 // rdx: throwing pc
438 __ jmp(handler_addr);
439
440 return start;
441 }
442
443
444 //----------------------------------------------------------------------------------------------------
445 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
446 //
447 // xchg exists as far back as 8086, lock needed for MP only
448 // Stack layout immediately after call:
449 //
450 // 0 [ret addr ] <--- rsp
451 // 1 [ ex ]
452 // 2 [ dest ]
453 //
454 // Result: *dest <- ex, return (old *dest)
455 //
456 // Note: win32 does not currently use this code
457
458 address generate_atomic_xchg() {
459 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
460 address start = __ pc();
461
462 __ push(rdx);
463 Address exchange(rsp, 2 * wordSize);
464 Address dest_addr(rsp, 3 * wordSize);
465 __ movl(rax, exchange);
466 __ movptr(rdx, dest_addr);
467 __ xchgl(rax, Address(rdx, 0));
468 __ pop(rdx);
469 __ ret(0);
470
471 return start;
472 }
473
474 //----------------------------------------------------------------------------------------------------
475 // Support for void verify_mxcsr()
476 //
477 // This routine is used with -Xcheck:jni to verify that native
478 // JNI code does not return to Java code without restoring the
479 // MXCSR register to our expected state.
480
481
482 address generate_verify_mxcsr() {
483 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
484 address start = __ pc();
485
486 const Address mxcsr_save(rsp, 0);
487
488 if (CheckJNICalls && UseSSE > 0 ) {
489 Label ok_ret;
490 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
491 __ push(rax);
492 __ subptr(rsp, wordSize); // allocate a temp location
493 __ stmxcsr(mxcsr_save);
494 __ movl(rax, mxcsr_save);
495 __ andl(rax, MXCSR_MASK);
496 __ cmp32(rax, mxcsr_std);
497 __ jcc(Assembler::equal, ok_ret);
498
499 __ warn("MXCSR changed by native JNI code.");
500
501 __ ldmxcsr(mxcsr_std);
502
503 __ bind(ok_ret);
504 __ addptr(rsp, wordSize);
505 __ pop(rax);
506 }
507
508 __ ret(0);
509
510 return start;
511 }
512
513
514 //---------------------------------------------------------------------------
515 // Support for void verify_fpu_cntrl_wrd()
516 //
517 // This routine is used with -Xcheck:jni to verify that native
518 // JNI code does not return to Java code without restoring the
519 // FP control word to our expected state.
520
521 address generate_verify_fpu_cntrl_wrd() {
522 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
523 address start = __ pc();
524
525 const Address fpu_cntrl_wrd_save(rsp, 0);
526
527 if (CheckJNICalls) {
528 Label ok_ret;
529 __ push(rax);
530 __ subptr(rsp, wordSize); // allocate a temp location
531 __ fnstcw(fpu_cntrl_wrd_save);
532 __ movl(rax, fpu_cntrl_wrd_save);
533 __ andl(rax, FPU_CNTRL_WRD_MASK);
534 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
535 __ cmp32(rax, fpu_std);
536 __ jcc(Assembler::equal, ok_ret);
537
538 __ warn("Floating point control word changed by native JNI code.");
539
540 __ fldcw(fpu_std);
541
542 __ bind(ok_ret);
543 __ addptr(rsp, wordSize);
544 __ pop(rax);
545 }
546
547 __ ret(0);
548
549 return start;
550 }
551
552 //---------------------------------------------------------------------------
553 // Wrapper for slow-case handling of double-to-integer conversion
554 // d2i or f2i fast case failed either because it is nan or because
555 // of under/overflow.
556 // Input: FPU TOS: float value
557 // Output: rax, (rdx): integer (long) result
558
559 address generate_d2i_wrapper(BasicType t, address fcn) {
560 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
561 address start = __ pc();
562
563 // Capture info about frame layout
564 enum layout { FPUState_off = 0,
565 rbp_off = FPUStateSizeInWords,
566 rdi_off,
567 rsi_off,
568 rcx_off,
569 rbx_off,
570 saved_argument_off,
571 saved_argument_off2, // 2nd half of double
572 framesize
573 };
574
575 assert(FPUStateSizeInWords == 27, "update stack layout");
576
577 // Save outgoing argument to stack across push_FPU_state()
578 __ subptr(rsp, wordSize * 2);
579 __ fstp_d(Address(rsp, 0));
580
581 // Save CPU & FPU state
582 __ push(rbx);
583 __ push(rcx);
584 __ push(rsi);
585 __ push(rdi);
586 __ push(rbp);
587 __ push_FPU_state();
588
589 // push_FPU_state() resets the FP top of stack
590 // Load original double into FP top of stack
591 __ fld_d(Address(rsp, saved_argument_off * wordSize));
592 // Store double into stack as outgoing argument
593 __ subptr(rsp, wordSize*2);
594 __ fst_d(Address(rsp, 0));
595
596 // Prepare FPU for doing math in C-land
597 __ empty_FPU_stack();
598 // Call the C code to massage the double. Result in EAX
599 if (t == T_INT)
600 { BLOCK_COMMENT("SharedRuntime::d2i"); }
601 else if (t == T_LONG)
602 { BLOCK_COMMENT("SharedRuntime::d2l"); }
603 __ call_VM_leaf( fcn, 2 );
604
605 // Restore CPU & FPU state
606 __ pop_FPU_state();
607 __ pop(rbp);
608 __ pop(rdi);
609 __ pop(rsi);
610 __ pop(rcx);
611 __ pop(rbx);
612 __ addptr(rsp, wordSize * 2);
613
614 __ ret(0);
615
616 return start;
617 }
618
619
620 //---------------------------------------------------------------------------
621 // The following routine generates a subroutine to throw an asynchronous
622 // UnknownError when an unsafe access gets a fault that could not be
623 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
624 address generate_handler_for_unsafe_access() {
625 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
626 address start = __ pc();
627
628 __ push(0); // hole for return address-to-be
629 __ pusha(); // push registers
630 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
631 BLOCK_COMMENT("call handle_unsafe_access");
632 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
633 __ movptr(next_pc, rax); // stuff next address
634 __ popa();
635 __ ret(0); // jump to next address
636
637 return start;
638 }
639
640
641 //----------------------------------------------------------------------------------------------------
642 // Non-destructive plausibility checks for oops
643
644 address generate_verify_oop() {
645 StubCodeMark mark(this, "StubRoutines", "verify_oop");
646 address start = __ pc();
647
648 // Incoming arguments on stack after saving rax,:
649 //
650 // [tos ]: saved rdx
651 // [tos + 1]: saved EFLAGS
652 // [tos + 2]: return address
653 // [tos + 3]: char* error message
654 // [tos + 4]: oop object to verify
655 // [tos + 5]: saved rax, - saved by caller and bashed
656
657 Label exit, error;
658 __ pushf();
659 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
660 __ push(rdx); // save rdx
661 // make sure object is 'reasonable'
662 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
663 __ testptr(rax, rax);
664 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
665
666 // Check if the oop is in the right area of memory
667 const int oop_mask = Universe::verify_oop_mask();
668 const int oop_bits = Universe::verify_oop_bits();
669 __ mov(rdx, rax);
670 __ andptr(rdx, oop_mask);
671 __ cmpptr(rdx, oop_bits);
672 __ jcc(Assembler::notZero, error);
673
674 // make sure klass is 'reasonable', which is not zero.
675 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
676 __ testptr(rax, rax);
677 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
678
679 // return if everything seems ok
680 __ bind(exit);
681 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
682 __ pop(rdx); // restore rdx
683 __ popf(); // restore EFLAGS
684 __ ret(3 * wordSize); // pop arguments
685
686 // handle errors
687 __ bind(error);
688 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
689 __ pop(rdx); // get saved rdx back
690 __ popf(); // get saved EFLAGS off stack -- will be ignored
691 __ pusha(); // push registers (eip = return address & msg are already pushed)
692 BLOCK_COMMENT("call MacroAssembler::debug");
693 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
694 __ popa();
695 __ ret(3 * wordSize); // pop arguments
696 return start;
697 }
698
699 //
700 // Generate pre-barrier for array stores
701 //
702 // Input:
703 // start - starting address
704 // count - element count
705 void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
706 assert_different_registers(start, count);
707 BarrierSet* bs = Universe::heap()->barrier_set();
708 switch (bs->kind()) {
709 case BarrierSet::G1SATBCT:
710 case BarrierSet::G1SATBCTLogging:
711 // With G1, don't generate the call if we statically know that the target in uninitialized
712 if (!uninitialized_target) {
713 __ pusha(); // push registers
714 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
715 start, count);
716 __ popa();
717 }
718 break;
719 case BarrierSet::CardTableModRef:
720 case BarrierSet::CardTableExtension:
721 case BarrierSet::ModRef:
722 break;
723 default :
724 ShouldNotReachHere();
725
726 }
727 }
728
729
730 //
731 // Generate a post-barrier for an array store
732 //
733 // start - starting address
734 // count - element count
735 //
736 // The two input registers are overwritten.
737 //
738 void gen_write_ref_array_post_barrier(Register start, Register count) {
739 BarrierSet* bs = Universe::heap()->barrier_set();
740 assert_different_registers(start, count);
741 switch (bs->kind()) {
742 case BarrierSet::G1SATBCT:
743 case BarrierSet::G1SATBCTLogging:
744 {
745 __ pusha(); // push registers
746 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
747 start, count);
748 __ popa();
749 }
750 break;
751
752 case BarrierSet::CardTableModRef:
753 case BarrierSet::CardTableExtension:
754 {
755 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
756 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
757
758 Label L_loop;
759 const Register end = count; // elements count; end == start+count-1
760 assert_different_registers(start, end);
761
762 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
763 __ shrptr(start, CardTableModRefBS::card_shift);
764 __ shrptr(end, CardTableModRefBS::card_shift);
765 __ subptr(end, start); // end --> count
766 __ BIND(L_loop);
767 intptr_t disp = (intptr_t) ct->byte_map_base;
768 Address cardtable(start, count, Address::times_1, disp);
769 __ movb(cardtable, 0);
770 __ decrement(count);
771 __ jcc(Assembler::greaterEqual, L_loop);
772 }
773 break;
774 case BarrierSet::ModRef:
775 break;
776 default :
777 ShouldNotReachHere();
778
779 }
780 }
781
782
783 // Copy 64 bytes chunks
784 //
785 // Inputs:
786 // from - source array address
787 // to_from - destination array address - from
788 // qword_count - 8-bytes element count, negative
789 //
790 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
791 assert( UseSSE >= 2, "supported cpu only" );
792 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
793 // Copy 64-byte chunks
794 __ jmpb(L_copy_64_bytes);
795 __ align(OptoLoopAlignment);
796 __ BIND(L_copy_64_bytes_loop);
797
798 if (UseUnalignedLoadStores) {
799 if (UseAVX >= 2) {
800 __ vmovdqu(xmm0, Address(from, 0));
801 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
802 __ vmovdqu(xmm1, Address(from, 32));
803 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
804 } else {
805 __ movdqu(xmm0, Address(from, 0));
806 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
807 __ movdqu(xmm1, Address(from, 16));
808 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
809 __ movdqu(xmm2, Address(from, 32));
810 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
811 __ movdqu(xmm3, Address(from, 48));
812 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
813 }
814 } else {
815 __ movq(xmm0, Address(from, 0));
816 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
817 __ movq(xmm1, Address(from, 8));
818 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
819 __ movq(xmm2, Address(from, 16));
820 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
821 __ movq(xmm3, Address(from, 24));
822 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
823 __ movq(xmm4, Address(from, 32));
824 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
825 __ movq(xmm5, Address(from, 40));
826 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
827 __ movq(xmm6, Address(from, 48));
828 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
829 __ movq(xmm7, Address(from, 56));
830 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
831 }
832
833 __ addl(from, 64);
834 __ BIND(L_copy_64_bytes);
835 __ subl(qword_count, 8);
836 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
837
838 if (UseUnalignedLoadStores && (UseAVX >= 2)) {
839 // clean upper bits of YMM registers
840 __ vzeroupper();
841 }
842 __ addl(qword_count, 8);
843 __ jccb(Assembler::zero, L_exit);
844 //
845 // length is too short, just copy qwords
846 //
847 __ BIND(L_copy_8_bytes);
848 __ movq(xmm0, Address(from, 0));
849 __ movq(Address(from, to_from, Address::times_1), xmm0);
850 __ addl(from, 8);
851 __ decrement(qword_count);
852 __ jcc(Assembler::greater, L_copy_8_bytes);
853 __ BIND(L_exit);
854 }
855
856 // Copy 64 bytes chunks
857 //
858 // Inputs:
859 // from - source array address
860 // to_from - destination array address - from
861 // qword_count - 8-bytes element count, negative
862 //
863 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
864 assert( VM_Version::supports_mmx(), "supported cpu only" );
865 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
866 // Copy 64-byte chunks
867 __ jmpb(L_copy_64_bytes);
868 __ align(OptoLoopAlignment);
869 __ BIND(L_copy_64_bytes_loop);
870 __ movq(mmx0, Address(from, 0));
871 __ movq(mmx1, Address(from, 8));
872 __ movq(mmx2, Address(from, 16));
873 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
874 __ movq(mmx3, Address(from, 24));
875 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
876 __ movq(mmx4, Address(from, 32));
877 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
878 __ movq(mmx5, Address(from, 40));
879 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
880 __ movq(mmx6, Address(from, 48));
881 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
882 __ movq(mmx7, Address(from, 56));
883 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
884 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
885 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
886 __ addptr(from, 64);
887 __ BIND(L_copy_64_bytes);
888 __ subl(qword_count, 8);
889 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
890 __ addl(qword_count, 8);
891 __ jccb(Assembler::zero, L_exit);
892 //
893 // length is too short, just copy qwords
894 //
895 __ BIND(L_copy_8_bytes);
896 __ movq(mmx0, Address(from, 0));
897 __ movq(Address(from, to_from, Address::times_1), mmx0);
898 __ addptr(from, 8);
899 __ decrement(qword_count);
900 __ jcc(Assembler::greater, L_copy_8_bytes);
901 __ BIND(L_exit);
902 __ emms();
903 }
904
905 address generate_disjoint_copy(BasicType t, bool aligned,
906 Address::ScaleFactor sf,
907 address* entry, const char *name,
908 bool dest_uninitialized = false) {
909 __ align(CodeEntryAlignment);
910 StubCodeMark mark(this, "StubRoutines", name);
911 address start = __ pc();
912
913 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
914 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
915
916 int shift = Address::times_ptr - sf;
917
918 const Register from = rsi; // source array address
919 const Register to = rdi; // destination array address
920 const Register count = rcx; // elements count
921 const Register to_from = to; // (to - from)
922 const Register saved_to = rdx; // saved destination array address
923
924 __ enter(); // required for proper stackwalking of RuntimeStub frame
925 __ push(rsi);
926 __ push(rdi);
927 __ movptr(from , Address(rsp, 12+ 4));
928 __ movptr(to , Address(rsp, 12+ 8));
929 __ movl(count, Address(rsp, 12+ 12));
930
931 if (entry != NULL) {
932 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
933 BLOCK_COMMENT("Entry:");
934 }
935
936 if (t == T_OBJECT) {
937 __ testl(count, count);
938 __ jcc(Assembler::zero, L_0_count);
939 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
940 __ mov(saved_to, to); // save 'to'
941 }
942
943 __ subptr(to, from); // to --> to_from
944 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
945 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
946 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
947 // align source address at 4 bytes address boundary
948 if (t == T_BYTE) {
949 // One byte misalignment happens only for byte arrays
950 __ testl(from, 1);
951 __ jccb(Assembler::zero, L_skip_align1);
952 __ movb(rax, Address(from, 0));
953 __ movb(Address(from, to_from, Address::times_1, 0), rax);
954 __ increment(from);
955 __ decrement(count);
956 __ BIND(L_skip_align1);
957 }
958 // Two bytes misalignment happens only for byte and short (char) arrays
959 __ testl(from, 2);
960 __ jccb(Assembler::zero, L_skip_align2);
961 __ movw(rax, Address(from, 0));
962 __ movw(Address(from, to_from, Address::times_1, 0), rax);
963 __ addptr(from, 2);
964 __ subl(count, 1<<(shift-1));
965 __ BIND(L_skip_align2);
966 }
967 if (!VM_Version::supports_mmx()) {
968 __ mov(rax, count); // save 'count'
969 __ shrl(count, shift); // bytes count
970 __ addptr(to_from, from);// restore 'to'
971 __ rep_mov();
972 __ subptr(to_from, from);// restore 'to_from'
973 __ mov(count, rax); // restore 'count'
974 __ jmpb(L_copy_2_bytes); // all dwords were copied
975 } else {
976 if (!UseUnalignedLoadStores) {
977 // align to 8 bytes, we know we are 4 byte aligned to start
978 __ testptr(from, 4);
979 __ jccb(Assembler::zero, L_copy_64_bytes);
980 __ movl(rax, Address(from, 0));
981 __ movl(Address(from, to_from, Address::times_1, 0), rax);
982 __ addptr(from, 4);
983 __ subl(count, 1<<shift);
984 }
985 __ BIND(L_copy_64_bytes);
986 __ mov(rax, count);
987 __ shrl(rax, shift+1); // 8 bytes chunk count
988 //
989 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
990 //
991 if (UseXMMForArrayCopy) {
992 xmm_copy_forward(from, to_from, rax);
993 } else {
994 mmx_copy_forward(from, to_from, rax);
995 }
996 }
997 // copy tailing dword
998 __ BIND(L_copy_4_bytes);
999 __ testl(count, 1<<shift);
1000 __ jccb(Assembler::zero, L_copy_2_bytes);
1001 __ movl(rax, Address(from, 0));
1002 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1003 if (t == T_BYTE || t == T_SHORT) {
1004 __ addptr(from, 4);
1005 __ BIND(L_copy_2_bytes);
1006 // copy tailing word
1007 __ testl(count, 1<<(shift-1));
1008 __ jccb(Assembler::zero, L_copy_byte);
1009 __ movw(rax, Address(from, 0));
1010 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1011 if (t == T_BYTE) {
1012 __ addptr(from, 2);
1013 __ BIND(L_copy_byte);
1014 // copy tailing byte
1015 __ testl(count, 1);
1016 __ jccb(Assembler::zero, L_exit);
1017 __ movb(rax, Address(from, 0));
1018 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1019 __ BIND(L_exit);
1020 } else {
1021 __ BIND(L_copy_byte);
1022 }
1023 } else {
1024 __ BIND(L_copy_2_bytes);
1025 }
1026
1027 if (t == T_OBJECT) {
1028 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1029 __ mov(to, saved_to); // restore 'to'
1030 gen_write_ref_array_post_barrier(to, count);
1031 __ BIND(L_0_count);
1032 }
1033 inc_copy_counter_np(t);
1034 __ pop(rdi);
1035 __ pop(rsi);
1036 __ leave(); // required for proper stackwalking of RuntimeStub frame
1037 __ xorptr(rax, rax); // return 0
1038 __ ret(0);
1039 return start;
1040 }
1041
1042
1043 address generate_fill(BasicType t, bool aligned, const char *name) {
1044 __ align(CodeEntryAlignment);
1045 StubCodeMark mark(this, "StubRoutines", name);
1046 address start = __ pc();
1047
1048 BLOCK_COMMENT("Entry:");
1049
1050 const Register to = rdi; // source array address
1051 const Register value = rdx; // value
1052 const Register count = rsi; // elements count
1053
1054 __ enter(); // required for proper stackwalking of RuntimeStub frame
1055 __ push(rsi);
1056 __ push(rdi);
1057 __ movptr(to , Address(rsp, 12+ 4));
1058 __ movl(value, Address(rsp, 12+ 8));
1059 __ movl(count, Address(rsp, 12+ 12));
1060
1061 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1062
1063 __ pop(rdi);
1064 __ pop(rsi);
1065 __ leave(); // required for proper stackwalking of RuntimeStub frame
1066 __ ret(0);
1067 return start;
1068 }
1069
1070 address generate_conjoint_copy(BasicType t, bool aligned,
1071 Address::ScaleFactor sf,
1072 address nooverlap_target,
1073 address* entry, const char *name,
1074 bool dest_uninitialized = false) {
1075 __ align(CodeEntryAlignment);
1076 StubCodeMark mark(this, "StubRoutines", name);
1077 address start = __ pc();
1078
1079 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1080 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1081
1082 int shift = Address::times_ptr - sf;
1083
1084 const Register src = rax; // source array address
1085 const Register dst = rdx; // destination array address
1086 const Register from = rsi; // source array address
1087 const Register to = rdi; // destination array address
1088 const Register count = rcx; // elements count
1089 const Register end = rax; // array end address
1090
1091 __ enter(); // required for proper stackwalking of RuntimeStub frame
1092 __ push(rsi);
1093 __ push(rdi);
1094 __ movptr(src , Address(rsp, 12+ 4)); // from
1095 __ movptr(dst , Address(rsp, 12+ 8)); // to
1096 __ movl2ptr(count, Address(rsp, 12+12)); // count
1097
1098 if (entry != NULL) {
1099 *entry = __ pc(); // Entry point from generic arraycopy stub.
1100 BLOCK_COMMENT("Entry:");
1101 }
1102
1103 // nooverlap_target expects arguments in rsi and rdi.
1104 __ mov(from, src);
1105 __ mov(to , dst);
1106
1107 // arrays overlap test: dispatch to disjoint stub if necessary.
1108 RuntimeAddress nooverlap(nooverlap_target);
1109 __ cmpptr(dst, src);
1110 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1111 __ jump_cc(Assembler::belowEqual, nooverlap);
1112 __ cmpptr(dst, end);
1113 __ jump_cc(Assembler::aboveEqual, nooverlap);
1114
1115 if (t == T_OBJECT) {
1116 __ testl(count, count);
1117 __ jcc(Assembler::zero, L_0_count);
1118 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1119 }
1120
1121 // copy from high to low
1122 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1123 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1124 if (t == T_BYTE || t == T_SHORT) {
1125 // Align the end of destination array at 4 bytes address boundary
1126 __ lea(end, Address(dst, count, sf, 0));
1127 if (t == T_BYTE) {
1128 // One byte misalignment happens only for byte arrays
1129 __ testl(end, 1);
1130 __ jccb(Assembler::zero, L_skip_align1);
1131 __ decrement(count);
1132 __ movb(rdx, Address(from, count, sf, 0));
1133 __ movb(Address(to, count, sf, 0), rdx);
1134 __ BIND(L_skip_align1);
1135 }
1136 // Two bytes misalignment happens only for byte and short (char) arrays
1137 __ testl(end, 2);
1138 __ jccb(Assembler::zero, L_skip_align2);
1139 __ subptr(count, 1<<(shift-1));
1140 __ movw(rdx, Address(from, count, sf, 0));
1141 __ movw(Address(to, count, sf, 0), rdx);
1142 __ BIND(L_skip_align2);
1143 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1144 __ jcc(Assembler::below, L_copy_4_bytes);
1145 }
1146
1147 if (!VM_Version::supports_mmx()) {
1148 __ std();
1149 __ mov(rax, count); // Save 'count'
1150 __ mov(rdx, to); // Save 'to'
1151 __ lea(rsi, Address(from, count, sf, -4));
1152 __ lea(rdi, Address(to , count, sf, -4));
1153 __ shrptr(count, shift); // bytes count
1154 __ rep_mov();
1155 __ cld();
1156 __ mov(count, rax); // restore 'count'
1157 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1158 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1159 __ mov(to, rdx); // restore 'to'
1160 __ jmpb(L_copy_2_bytes); // all dword were copied
1161 } else {
1162 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1163 __ testptr(end, 4);
1164 __ jccb(Assembler::zero, L_copy_8_bytes);
1165 __ subl(count, 1<<shift);
1166 __ movl(rdx, Address(from, count, sf, 0));
1167 __ movl(Address(to, count, sf, 0), rdx);
1168 __ jmpb(L_copy_8_bytes);
1169
1170 __ align(OptoLoopAlignment);
1171 // Move 8 bytes
1172 __ BIND(L_copy_8_bytes_loop);
1173 if (UseXMMForArrayCopy) {
1174 __ movq(xmm0, Address(from, count, sf, 0));
1175 __ movq(Address(to, count, sf, 0), xmm0);
1176 } else {
1177 __ movq(mmx0, Address(from, count, sf, 0));
1178 __ movq(Address(to, count, sf, 0), mmx0);
1179 }
1180 __ BIND(L_copy_8_bytes);
1181 __ subl(count, 2<<shift);
1182 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1183 __ addl(count, 2<<shift);
1184 if (!UseXMMForArrayCopy) {
1185 __ emms();
1186 }
1187 }
1188 __ BIND(L_copy_4_bytes);
1189 // copy prefix qword
1190 __ testl(count, 1<<shift);
1191 __ jccb(Assembler::zero, L_copy_2_bytes);
1192 __ movl(rdx, Address(from, count, sf, -4));
1193 __ movl(Address(to, count, sf, -4), rdx);
1194
1195 if (t == T_BYTE || t == T_SHORT) {
1196 __ subl(count, (1<<shift));
1197 __ BIND(L_copy_2_bytes);
1198 // copy prefix dword
1199 __ testl(count, 1<<(shift-1));
1200 __ jccb(Assembler::zero, L_copy_byte);
1201 __ movw(rdx, Address(from, count, sf, -2));
1202 __ movw(Address(to, count, sf, -2), rdx);
1203 if (t == T_BYTE) {
1204 __ subl(count, 1<<(shift-1));
1205 __ BIND(L_copy_byte);
1206 // copy prefix byte
1207 __ testl(count, 1);
1208 __ jccb(Assembler::zero, L_exit);
1209 __ movb(rdx, Address(from, 0));
1210 __ movb(Address(to, 0), rdx);
1211 __ BIND(L_exit);
1212 } else {
1213 __ BIND(L_copy_byte);
1214 }
1215 } else {
1216 __ BIND(L_copy_2_bytes);
1217 }
1218 if (t == T_OBJECT) {
1219 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1220 gen_write_ref_array_post_barrier(to, count);
1221 __ BIND(L_0_count);
1222 }
1223 inc_copy_counter_np(t);
1224 __ pop(rdi);
1225 __ pop(rsi);
1226 __ leave(); // required for proper stackwalking of RuntimeStub frame
1227 __ xorptr(rax, rax); // return 0
1228 __ ret(0);
1229 return start;
1230 }
1231
1232
1233 address generate_disjoint_long_copy(address* entry, const char *name) {
1234 __ align(CodeEntryAlignment);
1235 StubCodeMark mark(this, "StubRoutines", name);
1236 address start = __ pc();
1237
1238 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1239 const Register from = rax; // source array address
1240 const Register to = rdx; // destination array address
1241 const Register count = rcx; // elements count
1242 const Register to_from = rdx; // (to - from)
1243
1244 __ enter(); // required for proper stackwalking of RuntimeStub frame
1245 __ movptr(from , Address(rsp, 8+0)); // from
1246 __ movptr(to , Address(rsp, 8+4)); // to
1247 __ movl2ptr(count, Address(rsp, 8+8)); // count
1248
1249 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1250 BLOCK_COMMENT("Entry:");
1251
1252 __ subptr(to, from); // to --> to_from
1253 if (VM_Version::supports_mmx()) {
1254 if (UseXMMForArrayCopy) {
1255 xmm_copy_forward(from, to_from, count);
1256 } else {
1257 mmx_copy_forward(from, to_from, count);
1258 }
1259 } else {
1260 __ jmpb(L_copy_8_bytes);
1261 __ align(OptoLoopAlignment);
1262 __ BIND(L_copy_8_bytes_loop);
1263 __ fild_d(Address(from, 0));
1264 __ fistp_d(Address(from, to_from, Address::times_1));
1265 __ addptr(from, 8);
1266 __ BIND(L_copy_8_bytes);
1267 __ decrement(count);
1268 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1269 }
1270 inc_copy_counter_np(T_LONG);
1271 __ leave(); // required for proper stackwalking of RuntimeStub frame
1272 __ xorptr(rax, rax); // return 0
1273 __ ret(0);
1274 return start;
1275 }
1276
1277 address generate_conjoint_long_copy(address nooverlap_target,
1278 address* entry, const char *name) {
1279 __ align(CodeEntryAlignment);
1280 StubCodeMark mark(this, "StubRoutines", name);
1281 address start = __ pc();
1282
1283 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1284 const Register from = rax; // source array address
1285 const Register to = rdx; // destination array address
1286 const Register count = rcx; // elements count
1287 const Register end_from = rax; // source array end address
1288
1289 __ enter(); // required for proper stackwalking of RuntimeStub frame
1290 __ movptr(from , Address(rsp, 8+0)); // from
1291 __ movptr(to , Address(rsp, 8+4)); // to
1292 __ movl2ptr(count, Address(rsp, 8+8)); // count
1293
1294 *entry = __ pc(); // Entry point from generic arraycopy stub.
1295 BLOCK_COMMENT("Entry:");
1296
1297 // arrays overlap test
1298 __ cmpptr(to, from);
1299 RuntimeAddress nooverlap(nooverlap_target);
1300 __ jump_cc(Assembler::belowEqual, nooverlap);
1301 __ lea(end_from, Address(from, count, Address::times_8, 0));
1302 __ cmpptr(to, end_from);
1303 __ movptr(from, Address(rsp, 8)); // from
1304 __ jump_cc(Assembler::aboveEqual, nooverlap);
1305
1306 __ jmpb(L_copy_8_bytes);
1307
1308 __ align(OptoLoopAlignment);
1309 __ BIND(L_copy_8_bytes_loop);
1310 if (VM_Version::supports_mmx()) {
1311 if (UseXMMForArrayCopy) {
1312 __ movq(xmm0, Address(from, count, Address::times_8));
1313 __ movq(Address(to, count, Address::times_8), xmm0);
1314 } else {
1315 __ movq(mmx0, Address(from, count, Address::times_8));
1316 __ movq(Address(to, count, Address::times_8), mmx0);
1317 }
1318 } else {
1319 __ fild_d(Address(from, count, Address::times_8));
1320 __ fistp_d(Address(to, count, Address::times_8));
1321 }
1322 __ BIND(L_copy_8_bytes);
1323 __ decrement(count);
1324 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1325
1326 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1327 __ emms();
1328 }
1329 inc_copy_counter_np(T_LONG);
1330 __ leave(); // required for proper stackwalking of RuntimeStub frame
1331 __ xorptr(rax, rax); // return 0
1332 __ ret(0);
1333 return start;
1334 }
1335
1336
1337 // Helper for generating a dynamic type check.
1338 // The sub_klass must be one of {rbx, rdx, rsi}.
1339 // The temp is killed.
1340 void generate_type_check(Register sub_klass,
1341 Address& super_check_offset_addr,
1342 Address& super_klass_addr,
1343 Register temp,
1344 Label* L_success, Label* L_failure) {
1345 BLOCK_COMMENT("type_check:");
1346
1347 Label L_fallthrough;
1348 #define LOCAL_JCC(assembler_con, label_ptr) \
1349 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1350 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1351
1352 // The following is a strange variation of the fast path which requires
1353 // one less register, because needed values are on the argument stack.
1354 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1355 // L_success, L_failure, NULL);
1356 assert_different_registers(sub_klass, temp);
1357
1358 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1359
1360 // if the pointers are equal, we are done (e.g., String[] elements)
1361 __ cmpptr(sub_klass, super_klass_addr);
1362 LOCAL_JCC(Assembler::equal, L_success);
1363
1364 // check the supertype display:
1365 __ movl2ptr(temp, super_check_offset_addr);
1366 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1367 __ movptr(temp, super_check_addr); // load displayed supertype
1368 __ cmpptr(temp, super_klass_addr); // test the super type
1369 LOCAL_JCC(Assembler::equal, L_success);
1370
1371 // if it was a primary super, we can just fail immediately
1372 __ cmpl(super_check_offset_addr, sc_offset);
1373 LOCAL_JCC(Assembler::notEqual, L_failure);
1374
1375 // The repne_scan instruction uses fixed registers, which will get spilled.
1376 // We happen to know this works best when super_klass is in rax.
1377 Register super_klass = temp;
1378 __ movptr(super_klass, super_klass_addr);
1379 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1380 L_success, L_failure);
1381
1382 __ bind(L_fallthrough);
1383
1384 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1385 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1386
1387 #undef LOCAL_JCC
1388 }
1389
1390 //
1391 // Generate checkcasting array copy stub
1392 //
1393 // Input:
1394 // 4(rsp) - source array address
1395 // 8(rsp) - destination array address
1396 // 12(rsp) - element count, can be zero
1397 // 16(rsp) - size_t ckoff (super_check_offset)
1398 // 20(rsp) - oop ckval (super_klass)
1399 //
1400 // Output:
1401 // rax, == 0 - success
1402 // rax, == -1^K - failure, where K is partial transfer count
1403 //
1404 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1405 __ align(CodeEntryAlignment);
1406 StubCodeMark mark(this, "StubRoutines", name);
1407 address start = __ pc();
1408
1409 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1410
1411 // register use:
1412 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1413 // rdi, rsi -- element access (oop, klass)
1414 // rbx, -- temp
1415 const Register from = rax; // source array address
1416 const Register to = rdx; // destination array address
1417 const Register length = rcx; // elements count
1418 const Register elem = rdi; // each oop copied
1419 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1420 const Register temp = rbx; // lone remaining temp
1421
1422 __ enter(); // required for proper stackwalking of RuntimeStub frame
1423
1424 __ push(rsi);
1425 __ push(rdi);
1426 __ push(rbx);
1427
1428 Address from_arg(rsp, 16+ 4); // from
1429 Address to_arg(rsp, 16+ 8); // to
1430 Address length_arg(rsp, 16+12); // elements count
1431 Address ckoff_arg(rsp, 16+16); // super_check_offset
1432 Address ckval_arg(rsp, 16+20); // super_klass
1433
1434 // Load up:
1435 __ movptr(from, from_arg);
1436 __ movptr(to, to_arg);
1437 __ movl2ptr(length, length_arg);
1438
1439 if (entry != NULL) {
1440 *entry = __ pc(); // Entry point from generic arraycopy stub.
1441 BLOCK_COMMENT("Entry:");
1442 }
1443
1444 //---------------------------------------------------------------
1445 // Assembler stub will be used for this call to arraycopy
1446 // if the two arrays are subtypes of Object[] but the
1447 // destination array type is not equal to or a supertype
1448 // of the source type. Each element must be separately
1449 // checked.
1450
1451 // Loop-invariant addresses. They are exclusive end pointers.
1452 Address end_from_addr(from, length, Address::times_ptr, 0);
1453 Address end_to_addr(to, length, Address::times_ptr, 0);
1454
1455 Register end_from = from; // re-use
1456 Register end_to = to; // re-use
1457 Register count = length; // re-use
1458
1459 // Loop-variant addresses. They assume post-incremented count < 0.
1460 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1461 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1462 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1463
1464 // Copy from low to high addresses, indexed from the end of each array.
1465 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1466 __ lea(end_from, end_from_addr);
1467 __ lea(end_to, end_to_addr);
1468 assert(length == count, ""); // else fix next line:
1469 __ negptr(count); // negate and test the length
1470 __ jccb(Assembler::notZero, L_load_element);
1471
1472 // Empty array: Nothing to do.
1473 __ xorptr(rax, rax); // return 0 on (trivial) success
1474 __ jmp(L_done);
1475
1476 // ======== begin loop ========
1477 // (Loop is rotated; its entry is L_load_element.)
1478 // Loop control:
1479 // for (count = -count; count != 0; count++)
1480 // Base pointers src, dst are biased by 8*count,to last element.
1481 __ align(OptoLoopAlignment);
1482
1483 __ BIND(L_store_element);
1484 __ movptr(to_element_addr, elem); // store the oop
1485 __ increment(count); // increment the count toward zero
1486 __ jccb(Assembler::zero, L_do_card_marks);
1487
1488 // ======== loop entry is here ========
1489 __ BIND(L_load_element);
1490 __ movptr(elem, from_element_addr); // load the oop
1491 __ testptr(elem, elem);
1492 __ jccb(Assembler::zero, L_store_element);
1493
1494 // (Could do a trick here: Remember last successful non-null
1495 // element stored and make a quick oop equality check on it.)
1496
1497 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1498 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1499 &L_store_element, NULL);
1500 // (On fall-through, we have failed the element type check.)
1501 // ======== end loop ========
1502
1503 // It was a real error; we must depend on the caller to finish the job.
1504 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1505 // Emit GC store barriers for the oops we have copied (length_arg + count),
1506 // and report their number to the caller.
1507 assert_different_registers(to, count, rax);
1508 Label L_post_barrier;
1509 __ addl(count, length_arg); // transfers = (length - remaining)
1510 __ movl2ptr(rax, count); // save the value
1511 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1512 __ jccb(Assembler::notZero, L_post_barrier);
1513 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1514
1515 // Come here on success only.
1516 __ BIND(L_do_card_marks);
1517 __ xorptr(rax, rax); // return 0 on success
1518 __ movl2ptr(count, length_arg);
1519
1520 __ BIND(L_post_barrier);
1521 __ movptr(to, to_arg); // reload
1522 gen_write_ref_array_post_barrier(to, count);
1523
1524 // Common exit point (success or failure).
1525 __ BIND(L_done);
1526 __ pop(rbx);
1527 __ pop(rdi);
1528 __ pop(rsi);
1529 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1530 __ leave(); // required for proper stackwalking of RuntimeStub frame
1531 __ ret(0);
1532
1533 return start;
1534 }
1535
1536 //
1537 // Generate 'unsafe' array copy stub
1538 // Though just as safe as the other stubs, it takes an unscaled
1539 // size_t argument instead of an element count.
1540 //
1541 // Input:
1542 // 4(rsp) - source array address
1543 // 8(rsp) - destination array address
1544 // 12(rsp) - byte count, can be zero
1545 //
1546 // Output:
1547 // rax, == 0 - success
1548 // rax, == -1 - need to call System.arraycopy
1549 //
1550 // Examines the alignment of the operands and dispatches
1551 // to a long, int, short, or byte copy loop.
1552 //
1553 address generate_unsafe_copy(const char *name,
1554 address byte_copy_entry,
1555 address short_copy_entry,
1556 address int_copy_entry,
1557 address long_copy_entry) {
1558
1559 Label L_long_aligned, L_int_aligned, L_short_aligned;
1560
1561 __ align(CodeEntryAlignment);
1562 StubCodeMark mark(this, "StubRoutines", name);
1563 address start = __ pc();
1564
1565 const Register from = rax; // source array address
1566 const Register to = rdx; // destination array address
1567 const Register count = rcx; // elements count
1568
1569 __ enter(); // required for proper stackwalking of RuntimeStub frame
1570 __ push(rsi);
1571 __ push(rdi);
1572 Address from_arg(rsp, 12+ 4); // from
1573 Address to_arg(rsp, 12+ 8); // to
1574 Address count_arg(rsp, 12+12); // byte count
1575
1576 // Load up:
1577 __ movptr(from , from_arg);
1578 __ movptr(to , to_arg);
1579 __ movl2ptr(count, count_arg);
1580
1581 // bump this on entry, not on exit:
1582 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1583
1584 const Register bits = rsi;
1585 __ mov(bits, from);
1586 __ orptr(bits, to);
1587 __ orptr(bits, count);
1588
1589 __ testl(bits, BytesPerLong-1);
1590 __ jccb(Assembler::zero, L_long_aligned);
1591
1592 __ testl(bits, BytesPerInt-1);
1593 __ jccb(Assembler::zero, L_int_aligned);
1594
1595 __ testl(bits, BytesPerShort-1);
1596 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1597
1598 __ BIND(L_short_aligned);
1599 __ shrptr(count, LogBytesPerShort); // size => short_count
1600 __ movl(count_arg, count); // update 'count'
1601 __ jump(RuntimeAddress(short_copy_entry));
1602
1603 __ BIND(L_int_aligned);
1604 __ shrptr(count, LogBytesPerInt); // size => int_count
1605 __ movl(count_arg, count); // update 'count'
1606 __ jump(RuntimeAddress(int_copy_entry));
1607
1608 __ BIND(L_long_aligned);
1609 __ shrptr(count, LogBytesPerLong); // size => qword_count
1610 __ movl(count_arg, count); // update 'count'
1611 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1612 __ pop(rsi);
1613 __ jump(RuntimeAddress(long_copy_entry));
1614
1615 return start;
1616 }
1617
1618
1619 // Perform range checks on the proposed arraycopy.
1620 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1621 void arraycopy_range_checks(Register src,
1622 Register src_pos,
1623 Register dst,
1624 Register dst_pos,
1625 Address& length,
1626 Label& L_failed) {
1627 BLOCK_COMMENT("arraycopy_range_checks:");
1628 const Register src_end = src_pos; // source array end position
1629 const Register dst_end = dst_pos; // destination array end position
1630 __ addl(src_end, length); // src_pos + length
1631 __ addl(dst_end, length); // dst_pos + length
1632
1633 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1634 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1635 __ jcc(Assembler::above, L_failed);
1636
1637 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1638 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1639 __ jcc(Assembler::above, L_failed);
1640
1641 BLOCK_COMMENT("arraycopy_range_checks done");
1642 }
1643
1644
1645 //
1646 // Generate generic array copy stubs
1647 //
1648 // Input:
1649 // 4(rsp) - src oop
1650 // 8(rsp) - src_pos
1651 // 12(rsp) - dst oop
1652 // 16(rsp) - dst_pos
1653 // 20(rsp) - element count
1654 //
1655 // Output:
1656 // rax, == 0 - success
1657 // rax, == -1^K - failure, where K is partial transfer count
1658 //
1659 address generate_generic_copy(const char *name,
1660 address entry_jbyte_arraycopy,
1661 address entry_jshort_arraycopy,
1662 address entry_jint_arraycopy,
1663 address entry_oop_arraycopy,
1664 address entry_jlong_arraycopy,
1665 address entry_checkcast_arraycopy) {
1666 Label L_failed, L_failed_0, L_objArray;
1667
1668 { int modulus = CodeEntryAlignment;
1669 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1670 int advance = target - (__ offset() % modulus);
1671 if (advance < 0) advance += modulus;
1672 if (advance > 0) __ nop(advance);
1673 }
1674 StubCodeMark mark(this, "StubRoutines", name);
1675
1676 // Short-hop target to L_failed. Makes for denser prologue code.
1677 __ BIND(L_failed_0);
1678 __ jmp(L_failed);
1679 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1680
1681 __ align(CodeEntryAlignment);
1682 address start = __ pc();
1683
1684 __ enter(); // required for proper stackwalking of RuntimeStub frame
1685 __ push(rsi);
1686 __ push(rdi);
1687
1688 // bump this on entry, not on exit:
1689 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1690
1691 // Input values
1692 Address SRC (rsp, 12+ 4);
1693 Address SRC_POS (rsp, 12+ 8);
1694 Address DST (rsp, 12+12);
1695 Address DST_POS (rsp, 12+16);
1696 Address LENGTH (rsp, 12+20);
1697
1698 //-----------------------------------------------------------------------
1699 // Assembler stub will be used for this call to arraycopy
1700 // if the following conditions are met:
1701 //
1702 // (1) src and dst must not be null.
1703 // (2) src_pos must not be negative.
1704 // (3) dst_pos must not be negative.
1705 // (4) length must not be negative.
1706 // (5) src klass and dst klass should be the same and not NULL.
1707 // (6) src and dst should be arrays.
1708 // (7) src_pos + length must not exceed length of src.
1709 // (8) dst_pos + length must not exceed length of dst.
1710 //
1711
1712 const Register src = rax; // source array oop
1713 const Register src_pos = rsi;
1714 const Register dst = rdx; // destination array oop
1715 const Register dst_pos = rdi;
1716 const Register length = rcx; // transfer count
1717
1718 // if (src == NULL) return -1;
1719 __ movptr(src, SRC); // src oop
1720 __ testptr(src, src);
1721 __ jccb(Assembler::zero, L_failed_0);
1722
1723 // if (src_pos < 0) return -1;
1724 __ movl2ptr(src_pos, SRC_POS); // src_pos
1725 __ testl(src_pos, src_pos);
1726 __ jccb(Assembler::negative, L_failed_0);
1727
1728 // if (dst == NULL) return -1;
1729 __ movptr(dst, DST); // dst oop
1730 __ testptr(dst, dst);
1731 __ jccb(Assembler::zero, L_failed_0);
1732
1733 // if (dst_pos < 0) return -1;
1734 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1735 __ testl(dst_pos, dst_pos);
1736 __ jccb(Assembler::negative, L_failed_0);
1737
1738 // if (length < 0) return -1;
1739 __ movl2ptr(length, LENGTH); // length
1740 __ testl(length, length);
1741 __ jccb(Assembler::negative, L_failed_0);
1742
1743 // if (src->klass() == NULL) return -1;
1744 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1745 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1746 const Register rcx_src_klass = rcx; // array klass
1747 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1748
1749 #ifdef ASSERT
1750 // assert(src->klass() != NULL);
1751 BLOCK_COMMENT("assert klasses not null");
1752 { Label L1, L2;
1753 __ testptr(rcx_src_klass, rcx_src_klass);
1754 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1755 __ bind(L1);
1756 __ stop("broken null klass");
1757 __ bind(L2);
1758 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1759 __ jccb(Assembler::equal, L1); // this would be broken also
1760 BLOCK_COMMENT("assert done");
1761 }
1762 #endif //ASSERT
1763
1764 // Load layout helper (32-bits)
1765 //
1766 // |array_tag| | header_size | element_type | |log2_element_size|
1767 // 32 30 24 16 8 2 0
1768 //
1769 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1770 //
1771
1772 int lh_offset = in_bytes(Klass::layout_helper_offset());
1773 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1774
1775 // Handle objArrays completely differently...
1776 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1777 __ cmpl(src_klass_lh_addr, objArray_lh);
1778 __ jcc(Assembler::equal, L_objArray);
1779
1780 // if (src->klass() != dst->klass()) return -1;
1781 __ cmpptr(rcx_src_klass, dst_klass_addr);
1782 __ jccb(Assembler::notEqual, L_failed_0);
1783
1784 const Register rcx_lh = rcx; // layout helper
1785 assert(rcx_lh == rcx_src_klass, "known alias");
1786 __ movl(rcx_lh, src_klass_lh_addr);
1787
1788 // if (!src->is_Array()) return -1;
1789 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1790 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1791
1792 // At this point, it is known to be a typeArray (array_tag 0x3).
1793 #ifdef ASSERT
1794 { Label L;
1795 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1796 __ jcc(Assembler::greaterEqual, L); // signed cmp
1797 __ stop("must be a primitive array");
1798 __ bind(L);
1799 }
1800 #endif
1801
1802 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1803 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1804
1805 // TypeArrayKlass
1806 //
1807 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1808 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1809 //
1810 const Register rsi_offset = rsi; // array offset
1811 const Register src_array = src; // src array offset
1812 const Register dst_array = dst; // dst array offset
1813 const Register rdi_elsize = rdi; // log2 element size
1814
1815 __ mov(rsi_offset, rcx_lh);
1816 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1817 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1818 __ addptr(src_array, rsi_offset); // src array offset
1819 __ addptr(dst_array, rsi_offset); // dst array offset
1820 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1821
1822 // next registers should be set before the jump to corresponding stub
1823 const Register from = src; // source array address
1824 const Register to = dst; // destination array address
1825 const Register count = rcx; // elements count
1826 // some of them should be duplicated on stack
1827 #define FROM Address(rsp, 12+ 4)
1828 #define TO Address(rsp, 12+ 8) // Not used now
1829 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1830
1831 BLOCK_COMMENT("scale indexes to element size");
1832 __ movl2ptr(rsi, SRC_POS); // src_pos
1833 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1834 assert(src_array == from, "");
1835 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1836 __ movl2ptr(rdi, DST_POS); // dst_pos
1837 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1838 assert(dst_array == to, "");
1839 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1840 __ movptr(FROM, from); // src_addr
1841 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1842 __ movl2ptr(count, LENGTH); // elements count
1843
1844 BLOCK_COMMENT("choose copy loop based on element size");
1845 __ cmpl(rdi_elsize, 0);
1846
1847 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1848 __ cmpl(rdi_elsize, LogBytesPerShort);
1849 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1850 __ cmpl(rdi_elsize, LogBytesPerInt);
1851 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1852 #ifdef ASSERT
1853 __ cmpl(rdi_elsize, LogBytesPerLong);
1854 __ jccb(Assembler::notEqual, L_failed);
1855 #endif
1856 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1857 __ pop(rsi);
1858 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1859
1860 __ BIND(L_failed);
1861 __ xorptr(rax, rax);
1862 __ notptr(rax); // return -1
1863 __ pop(rdi);
1864 __ pop(rsi);
1865 __ leave(); // required for proper stackwalking of RuntimeStub frame
1866 __ ret(0);
1867
1868 // ObjArrayKlass
1869 __ BIND(L_objArray);
1870 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1871
1872 Label L_plain_copy, L_checkcast_copy;
1873 // test array classes for subtyping
1874 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1875 __ jccb(Assembler::notEqual, L_checkcast_copy);
1876
1877 // Identically typed arrays can be copied without element-wise checks.
1878 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1879 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1880
1881 __ BIND(L_plain_copy);
1882 __ movl2ptr(count, LENGTH); // elements count
1883 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1884 __ lea(from, Address(src, src_pos, Address::times_ptr,
1885 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1886 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1887 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1888 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1889 __ movptr(FROM, from); // src_addr
1890 __ movptr(TO, to); // dst_addr
1891 __ movl(COUNT, count); // count
1892 __ jump(RuntimeAddress(entry_oop_arraycopy));
1893
1894 __ BIND(L_checkcast_copy);
1895 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1896 {
1897 // Handy offsets:
1898 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1899 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1900
1901 Register rsi_dst_klass = rsi;
1902 Register rdi_temp = rdi;
1903 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1904 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1905 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1906
1907 // Before looking at dst.length, make sure dst is also an objArray.
1908 __ movptr(rsi_dst_klass, dst_klass_addr);
1909 __ cmpl(dst_klass_lh_addr, objArray_lh);
1910 __ jccb(Assembler::notEqual, L_failed);
1911
1912 // It is safe to examine both src.length and dst.length.
1913 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1914 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1915 // (Now src_pos and dst_pos are killed, but not src and dst.)
1916
1917 // We'll need this temp (don't forget to pop it after the type check).
1918 __ push(rbx);
1919 Register rbx_src_klass = rbx;
1920
1921 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1922 __ movptr(rsi_dst_klass, dst_klass_addr);
1923 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1924 Label L_fail_array_check;
1925 generate_type_check(rbx_src_klass,
1926 super_check_offset_addr, dst_klass_addr,
1927 rdi_temp, NULL, &L_fail_array_check);
1928 // (On fall-through, we have passed the array type check.)
1929 __ pop(rbx);
1930 __ jmp(L_plain_copy);
1931
1932 __ BIND(L_fail_array_check);
1933 // Reshuffle arguments so we can call checkcast_arraycopy:
1934
1935 // match initial saves for checkcast_arraycopy
1936 // push(rsi); // already done; see above
1937 // push(rdi); // already done; see above
1938 // push(rbx); // already done; see above
1939
1940 // Marshal outgoing arguments now, freeing registers.
1941 Address from_arg(rsp, 16+ 4); // from
1942 Address to_arg(rsp, 16+ 8); // to
1943 Address length_arg(rsp, 16+12); // elements count
1944 Address ckoff_arg(rsp, 16+16); // super_check_offset
1945 Address ckval_arg(rsp, 16+20); // super_klass
1946
1947 Address SRC_POS_arg(rsp, 16+ 8);
1948 Address DST_POS_arg(rsp, 16+16);
1949 Address LENGTH_arg(rsp, 16+20);
1950 // push rbx, changed the incoming offsets (why not just use rbp,??)
1951 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1952
1953 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1954 __ movl2ptr(length, LENGTH_arg); // reload elements count
1955 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1956 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1957
1958 __ movptr(ckval_arg, rbx); // destination element type
1959 __ movl(rbx, Address(rbx, sco_offset));
1960 __ movl(ckoff_arg, rbx); // corresponding class check offset
1961
1962 __ movl(length_arg, length); // outgoing length argument
1963
1964 __ lea(from, Address(src, src_pos, Address::times_ptr,
1965 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1966 __ movptr(from_arg, from);
1967
1968 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1969 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1970 __ movptr(to_arg, to);
1971 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1972 }
1973
1974 return start;
1975 }
1976
1977 void generate_arraycopy_stubs() {
1978 address entry;
1979 address entry_jbyte_arraycopy;
1980 address entry_jshort_arraycopy;
1981 address entry_jint_arraycopy;
1982 address entry_oop_arraycopy;
1983 address entry_jlong_arraycopy;
1984 address entry_checkcast_arraycopy;
1985
1986 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1987 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1988 "arrayof_jbyte_disjoint_arraycopy");
1989 StubRoutines::_arrayof_jbyte_arraycopy =
1990 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1991 NULL, "arrayof_jbyte_arraycopy");
1992 StubRoutines::_jbyte_disjoint_arraycopy =
1993 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1994 "jbyte_disjoint_arraycopy");
1995 StubRoutines::_jbyte_arraycopy =
1996 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1997 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1998
1999 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2000 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2001 "arrayof_jshort_disjoint_arraycopy");
2002 StubRoutines::_arrayof_jshort_arraycopy =
2003 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2004 NULL, "arrayof_jshort_arraycopy");
2005 StubRoutines::_jshort_disjoint_arraycopy =
2006 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2007 "jshort_disjoint_arraycopy");
2008 StubRoutines::_jshort_arraycopy =
2009 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2010 &entry_jshort_arraycopy, "jshort_arraycopy");
2011
2012 // Next arrays are always aligned on 4 bytes at least.
2013 StubRoutines::_jint_disjoint_arraycopy =
2014 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2015 "jint_disjoint_arraycopy");
2016 StubRoutines::_jint_arraycopy =
2017 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2018 &entry_jint_arraycopy, "jint_arraycopy");
2019
2020 StubRoutines::_oop_disjoint_arraycopy =
2021 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2022 "oop_disjoint_arraycopy");
2023 StubRoutines::_oop_arraycopy =
2024 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2025 &entry_oop_arraycopy, "oop_arraycopy");
2026
2027 StubRoutines::_oop_disjoint_arraycopy_uninit =
2028 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2029 "oop_disjoint_arraycopy_uninit",
2030 /*dest_uninitialized*/true);
2031 StubRoutines::_oop_arraycopy_uninit =
2032 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2033 NULL, "oop_arraycopy_uninit",
2034 /*dest_uninitialized*/true);
2035
2036 StubRoutines::_jlong_disjoint_arraycopy =
2037 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2038 StubRoutines::_jlong_arraycopy =
2039 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2040 "jlong_arraycopy");
2041
2042 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2043 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2044 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2045 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2046 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2047 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2048
2049 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2050 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2051 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2052 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2053
2054 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2055 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2056 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2057 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2058
2059 StubRoutines::_checkcast_arraycopy =
2060 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2061 StubRoutines::_checkcast_arraycopy_uninit =
2062 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2063
2064 StubRoutines::_unsafe_arraycopy =
2065 generate_unsafe_copy("unsafe_arraycopy",
2066 entry_jbyte_arraycopy,
2067 entry_jshort_arraycopy,
2068 entry_jint_arraycopy,
2069 entry_jlong_arraycopy);
2070
2071 StubRoutines::_generic_arraycopy =
2072 generate_generic_copy("generic_arraycopy",
2073 entry_jbyte_arraycopy,
2074 entry_jshort_arraycopy,
2075 entry_jint_arraycopy,
2076 entry_oop_arraycopy,
2077 entry_jlong_arraycopy,
2078 entry_checkcast_arraycopy);
2079 }
2080
2081 void generate_math_stubs() {
2082 {
2083 StubCodeMark mark(this, "StubRoutines", "log");
2084 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2085
2086 __ fld_d(Address(rsp, 4));
2087 __ flog();
2088 __ ret(0);
2089 }
2090 {
2091 StubCodeMark mark(this, "StubRoutines", "log10");
2092 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2093
2094 __ fld_d(Address(rsp, 4));
2095 __ flog10();
2096 __ ret(0);
2097 }
2098 {
2099 StubCodeMark mark(this, "StubRoutines", "sin");
2100 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2101
2102 __ fld_d(Address(rsp, 4));
2103 __ trigfunc('s');
2104 __ ret(0);
2105 }
2106 {
2107 StubCodeMark mark(this, "StubRoutines", "cos");
2108 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2109
2110 __ fld_d(Address(rsp, 4));
2111 __ trigfunc('c');
2112 __ ret(0);
2113 }
2114 {
2115 StubCodeMark mark(this, "StubRoutines", "tan");
2116 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2117
2118 __ fld_d(Address(rsp, 4));
2119 __ trigfunc('t');
2120 __ ret(0);
2121 }
2122 {
2123 StubCodeMark mark(this, "StubRoutines", "exp");
2124 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2125
2126 __ fld_d(Address(rsp, 4));
2127 __ exp_with_fallback(0);
2128 __ ret(0);
2129 }
2130 {
2131 StubCodeMark mark(this, "StubRoutines", "pow");
2132 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2133
2134 __ fld_d(Address(rsp, 12));
2135 __ fld_d(Address(rsp, 4));
2136 __ pow_with_fallback(0);
2137 __ ret(0);
2138 }
2139 }
2140
2141 // AES intrinsic stubs
2142 enum {AESBlockSize = 16};
2143
2144 address generate_key_shuffle_mask() {
2145 __ align(16);
2146 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2147 address start = __ pc();
2148 __ emit_data(0x00010203, relocInfo::none, 0 );
2149 __ emit_data(0x04050607, relocInfo::none, 0 );
2150 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2151 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2152 return start;
2153 }
2154
2155 // Utility routine for loading a 128-bit key word in little endian format
2156 // can optionally specify that the shuffle mask is already in an xmmregister
2157 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2158 __ movdqu(xmmdst, Address(key, offset));
2159 if (xmm_shuf_mask != NULL) {
2160 __ pshufb(xmmdst, xmm_shuf_mask);
2161 } else {
2162 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2163 }
2164 }
2165
2166 // aesenc using specified key+offset
2167 // can optionally specify that the shuffle mask is already in an xmmregister
2168 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2169 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2170 __ aesenc(xmmdst, xmmtmp);
2171 }
2172
2173 // aesdec using specified key+offset
2174 // can optionally specify that the shuffle mask is already in an xmmregister
2175 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2176 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2177 __ aesdec(xmmdst, xmmtmp);
2178 }
2179
2180
2181 // Arguments:
2182 //
2183 // Inputs:
2184 // c_rarg0 - source byte array address
2185 // c_rarg1 - destination byte array address
2186 // c_rarg2 - K (key) in little endian int array
2187 //
2188 address generate_aescrypt_encryptBlock() {
2189 assert(UseAES, "need AES instructions and misaligned SSE support");
2190 __ align(CodeEntryAlignment);
2191 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2192 Label L_doLast;
2193 address start = __ pc();
2194
2195 const Register from = rdx; // source array address
2196 const Register to = rdx; // destination array address
2197 const Register key = rcx; // key array address
2198 const Register keylen = rax;
2199 const Address from_param(rbp, 8+0);
2200 const Address to_param (rbp, 8+4);
2201 const Address key_param (rbp, 8+8);
2202
2203 const XMMRegister xmm_result = xmm0;
2204 const XMMRegister xmm_key_shuf_mask = xmm1;
2205 const XMMRegister xmm_temp1 = xmm2;
2206 const XMMRegister xmm_temp2 = xmm3;
2207 const XMMRegister xmm_temp3 = xmm4;
2208 const XMMRegister xmm_temp4 = xmm5;
2209
2210 __ enter(); // required for proper stackwalking of RuntimeStub frame
2211 __ movptr(from, from_param);
2212 __ movptr(key, key_param);
2213
2214 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2215 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2216
2217 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2218 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2219 __ movptr(to, to_param);
2220
2221 // For encryption, the java expanded key ordering is just what we need
2222
2223 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2224 __ pxor(xmm_result, xmm_temp1);
2225
2226 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2227 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2228 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2229 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2230
2231 __ aesenc(xmm_result, xmm_temp1);
2232 __ aesenc(xmm_result, xmm_temp2);
2233 __ aesenc(xmm_result, xmm_temp3);
2234 __ aesenc(xmm_result, xmm_temp4);
2235
2236 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2237 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2238 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2239 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2240
2241 __ aesenc(xmm_result, xmm_temp1);
2242 __ aesenc(xmm_result, xmm_temp2);
2243 __ aesenc(xmm_result, xmm_temp3);
2244 __ aesenc(xmm_result, xmm_temp4);
2245
2246 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2247 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2248
2249 __ cmpl(keylen, 44);
2250 __ jccb(Assembler::equal, L_doLast);
2251
2252 __ aesenc(xmm_result, xmm_temp1);
2253 __ aesenc(xmm_result, xmm_temp2);
2254
2255 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2256 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2257
2258 __ cmpl(keylen, 52);
2259 __ jccb(Assembler::equal, L_doLast);
2260
2261 __ aesenc(xmm_result, xmm_temp1);
2262 __ aesenc(xmm_result, xmm_temp2);
2263
2264 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2265 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2266
2267 __ BIND(L_doLast);
2268 __ aesenc(xmm_result, xmm_temp1);
2269 __ aesenclast(xmm_result, xmm_temp2);
2270 __ movdqu(Address(to, 0), xmm_result); // store the result
2271 __ xorptr(rax, rax); // return 0
2272 __ leave(); // required for proper stackwalking of RuntimeStub frame
2273 __ ret(0);
2274
2275 return start;
2276 }
2277
2278
2279 // Arguments:
2280 //
2281 // Inputs:
2282 // c_rarg0 - source byte array address
2283 // c_rarg1 - destination byte array address
2284 // c_rarg2 - K (key) in little endian int array
2285 //
2286 address generate_aescrypt_decryptBlock() {
2287 assert(UseAES, "need AES instructions and misaligned SSE support");
2288 __ align(CodeEntryAlignment);
2289 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2290 Label L_doLast;
2291 address start = __ pc();
2292
2293 const Register from = rdx; // source array address
2294 const Register to = rdx; // destination array address
2295 const Register key = rcx; // key array address
2296 const Register keylen = rax;
2297 const Address from_param(rbp, 8+0);
2298 const Address to_param (rbp, 8+4);
2299 const Address key_param (rbp, 8+8);
2300
2301 const XMMRegister xmm_result = xmm0;
2302 const XMMRegister xmm_key_shuf_mask = xmm1;
2303 const XMMRegister xmm_temp1 = xmm2;
2304 const XMMRegister xmm_temp2 = xmm3;
2305 const XMMRegister xmm_temp3 = xmm4;
2306 const XMMRegister xmm_temp4 = xmm5;
2307
2308 __ enter(); // required for proper stackwalking of RuntimeStub frame
2309 __ movptr(from, from_param);
2310 __ movptr(key, key_param);
2311
2312 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2313 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2314
2315 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2316 __ movdqu(xmm_result, Address(from, 0));
2317 __ movptr(to, to_param);
2318
2319 // for decryption java expanded key ordering is rotated one position from what we want
2320 // so we start from 0x10 here and hit 0x00 last
2321 // we don't know if the key is aligned, hence not using load-execute form
2322 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2323 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2324 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2325 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2326
2327 __ pxor (xmm_result, xmm_temp1);
2328 __ aesdec(xmm_result, xmm_temp2);
2329 __ aesdec(xmm_result, xmm_temp3);
2330 __ aesdec(xmm_result, xmm_temp4);
2331
2332 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2333 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2334 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2335 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2336
2337 __ aesdec(xmm_result, xmm_temp1);
2338 __ aesdec(xmm_result, xmm_temp2);
2339 __ aesdec(xmm_result, xmm_temp3);
2340 __ aesdec(xmm_result, xmm_temp4);
2341
2342 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2343 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2344 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2345
2346 __ cmpl(keylen, 44);
2347 __ jccb(Assembler::equal, L_doLast);
2348
2349 __ aesdec(xmm_result, xmm_temp1);
2350 __ aesdec(xmm_result, xmm_temp2);
2351
2352 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2353 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2354
2355 __ cmpl(keylen, 52);
2356 __ jccb(Assembler::equal, L_doLast);
2357
2358 __ aesdec(xmm_result, xmm_temp1);
2359 __ aesdec(xmm_result, xmm_temp2);
2360
2361 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2362 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2363
2364 __ BIND(L_doLast);
2365 __ aesdec(xmm_result, xmm_temp1);
2366 __ aesdec(xmm_result, xmm_temp2);
2367
2368 // for decryption the aesdeclast operation is always on key+0x00
2369 __ aesdeclast(xmm_result, xmm_temp3);
2370 __ movdqu(Address(to, 0), xmm_result); // store the result
2371 __ xorptr(rax, rax); // return 0
2372 __ leave(); // required for proper stackwalking of RuntimeStub frame
2373 __ ret(0);
2374
2375 return start;
2376 }
2377
2378 void handleSOERegisters(bool saving) {
2379 const int saveFrameSizeInBytes = 4 * wordSize;
2380 const Address saved_rbx (rbp, -3 * wordSize);
2381 const Address saved_rsi (rbp, -2 * wordSize);
2382 const Address saved_rdi (rbp, -1 * wordSize);
2383
2384 if (saving) {
2385 __ subptr(rsp, saveFrameSizeInBytes);
2386 __ movptr(saved_rsi, rsi);
2387 __ movptr(saved_rdi, rdi);
2388 __ movptr(saved_rbx, rbx);
2389 } else {
2390 // restoring
2391 __ movptr(rsi, saved_rsi);
2392 __ movptr(rdi, saved_rdi);
2393 __ movptr(rbx, saved_rbx);
2394 }
2395 }
2396
2397 // Arguments:
2398 //
2399 // Inputs:
2400 // c_rarg0 - source byte array address
2401 // c_rarg1 - destination byte array address
2402 // c_rarg2 - K (key) in little endian int array
2403 // c_rarg3 - r vector byte array address
2404 // c_rarg4 - input length
2405 //
2406 // Output:
2407 // rax - input length
2408 //
2409 address generate_cipherBlockChaining_encryptAESCrypt() {
2410 assert(UseAES, "need AES instructions and misaligned SSE support");
2411 __ align(CodeEntryAlignment);
2412 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2413 address start = __ pc();
2414
2415 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2416 const Register from = rsi; // source array address
2417 const Register to = rdx; // destination array address
2418 const Register key = rcx; // key array address
2419 const Register rvec = rdi; // r byte array initialized from initvector array address
2420 // and left with the results of the last encryption block
2421 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2422 const Register pos = rax;
2423
2424 // xmm register assignments for the loops below
2425 const XMMRegister xmm_result = xmm0;
2426 const XMMRegister xmm_temp = xmm1;
2427 // first 6 keys preloaded into xmm2-xmm7
2428 const int XMM_REG_NUM_KEY_FIRST = 2;
2429 const int XMM_REG_NUM_KEY_LAST = 7;
2430 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2431
2432 __ enter(); // required for proper stackwalking of RuntimeStub frame
2433 handleSOERegisters(true /*saving*/);
2434
2435 // load registers from incoming parameters
2436 const Address from_param(rbp, 8+0);
2437 const Address to_param (rbp, 8+4);
2438 const Address key_param (rbp, 8+8);
2439 const Address rvec_param (rbp, 8+12);
2440 const Address len_param (rbp, 8+16);
2441 __ movptr(from , from_param);
2442 __ movptr(to , to_param);
2443 __ movptr(key , key_param);
2444 __ movptr(rvec , rvec_param);
2445 __ movptr(len_reg , len_param);
2446
2447 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2448 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2449 // load up xmm regs 2 thru 7 with keys 0-5
2450 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2451 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2452 offset += 0x10;
2453 }
2454
2455 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2456
2457 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2458 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2459 __ cmpl(rax, 44);
2460 __ jcc(Assembler::notEqual, L_key_192_256);
2461
2462 // 128 bit code follows here
2463 __ movl(pos, 0);
2464 __ align(OptoLoopAlignment);
2465 __ BIND(L_loopTop_128);
2466 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2467 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2468
2469 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2470 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2471 __ aesenc(xmm_result, as_XMMRegister(rnum));
2472 }
2473 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2474 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2475 }
2476 load_key(xmm_temp, key, 0xa0);
2477 __ aesenclast(xmm_result, xmm_temp);
2478
2479 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2480 // no need to store r to memory until we exit
2481 __ addptr(pos, AESBlockSize);
2482 __ subptr(len_reg, AESBlockSize);
2483 __ jcc(Assembler::notEqual, L_loopTop_128);
2484
2485 __ BIND(L_exit);
2486 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2487
2488 handleSOERegisters(false /*restoring*/);
2489 __ movptr(rax, len_param); // return length
2490 __ leave(); // required for proper stackwalking of RuntimeStub frame
2491 __ ret(0);
2492
2493 __ BIND(L_key_192_256);
2494 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2495 __ cmpl(rax, 52);
2496 __ jcc(Assembler::notEqual, L_key_256);
2497
2498 // 192-bit code follows here (could be changed to use more xmm registers)
2499 __ movl(pos, 0);
2500 __ align(OptoLoopAlignment);
2501 __ BIND(L_loopTop_192);
2502 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2503 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2504
2505 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2506 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2507 __ aesenc(xmm_result, as_XMMRegister(rnum));
2508 }
2509 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2510 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2511 }
2512 load_key(xmm_temp, key, 0xc0);
2513 __ aesenclast(xmm_result, xmm_temp);
2514
2515 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2516 // no need to store r to memory until we exit
2517 __ addptr(pos, AESBlockSize);
2518 __ subptr(len_reg, AESBlockSize);
2519 __ jcc(Assembler::notEqual, L_loopTop_192);
2520 __ jmp(L_exit);
2521
2522 __ BIND(L_key_256);
2523 // 256-bit code follows here (could be changed to use more xmm registers)
2524 __ movl(pos, 0);
2525 __ align(OptoLoopAlignment);
2526 __ BIND(L_loopTop_256);
2527 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2528 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2529
2530 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2531 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2532 __ aesenc(xmm_result, as_XMMRegister(rnum));
2533 }
2534 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2535 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2536 }
2537 load_key(xmm_temp, key, 0xe0);
2538 __ aesenclast(xmm_result, xmm_temp);
2539
2540 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2541 // no need to store r to memory until we exit
2542 __ addptr(pos, AESBlockSize);
2543 __ subptr(len_reg, AESBlockSize);
2544 __ jcc(Assembler::notEqual, L_loopTop_256);
2545 __ jmp(L_exit);
2546
2547 return start;
2548 }
2549
2550
2551 // CBC AES Decryption.
2552 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2553 //
2554 // Arguments:
2555 //
2556 // Inputs:
2557 // c_rarg0 - source byte array address
2558 // c_rarg1 - destination byte array address
2559 // c_rarg2 - K (key) in little endian int array
2560 // c_rarg3 - r vector byte array address
2561 // c_rarg4 - input length
2562 //
2563 // Output:
2564 // rax - input length
2565 //
2566
2567 address generate_cipherBlockChaining_decryptAESCrypt() {
2568 assert(UseAES, "need AES instructions and misaligned SSE support");
2569 __ align(CodeEntryAlignment);
2570 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2571 address start = __ pc();
2572
2573 Label L_exit, L_key_192_256, L_key_256;
2574 Label L_singleBlock_loopTop_128;
2575 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2576 const Register from = rsi; // source array address
2577 const Register to = rdx; // destination array address
2578 const Register key = rcx; // key array address
2579 const Register rvec = rdi; // r byte array initialized from initvector array address
2580 // and left with the results of the last encryption block
2581 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2582 const Register pos = rax;
2583
2584 // xmm register assignments for the loops below
2585 const XMMRegister xmm_result = xmm0;
2586 const XMMRegister xmm_temp = xmm1;
2587 // first 6 keys preloaded into xmm2-xmm7
2588 const int XMM_REG_NUM_KEY_FIRST = 2;
2589 const int XMM_REG_NUM_KEY_LAST = 7;
2590 const int FIRST_NON_REG_KEY_offset = 0x70;
2591 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2592
2593 __ enter(); // required for proper stackwalking of RuntimeStub frame
2594 handleSOERegisters(true /*saving*/);
2595
2596 // load registers from incoming parameters
2597 const Address from_param(rbp, 8+0);
2598 const Address to_param (rbp, 8+4);
2599 const Address key_param (rbp, 8+8);
2600 const Address rvec_param (rbp, 8+12);
2601 const Address len_param (rbp, 8+16);
2602 __ movptr(from , from_param);
2603 __ movptr(to , to_param);
2604 __ movptr(key , key_param);
2605 __ movptr(rvec , rvec_param);
2606 __ movptr(len_reg , len_param);
2607
2608 // the java expanded key ordering is rotated one position from what we want
2609 // so we start from 0x10 here and hit 0x00 last
2610 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
2611 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2612 // load up xmm regs 2 thru 6 with first 5 keys
2613 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2614 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2615 offset += 0x10;
2616 }
2617
2618 // inside here, use the rvec register to point to previous block cipher
2619 // with which we xor at the end of each newly decrypted block
2620 const Register prev_block_cipher_ptr = rvec;
2621
2622 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2623 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2624 __ cmpl(rax, 44);
2625 __ jcc(Assembler::notEqual, L_key_192_256);
2626
2627
2628 // 128-bit code follows here, parallelized
2629 __ movl(pos, 0);
2630 __ align(OptoLoopAlignment);
2631 __ BIND(L_singleBlock_loopTop_128);
2632 __ cmpptr(len_reg, 0); // any blocks left??
2633 __ jcc(Assembler::equal, L_exit);
2634 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2635 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2636 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2637 __ aesdec(xmm_result, as_XMMRegister(rnum));
2638 }
2639 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
2640 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2641 }
2642 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2643 __ aesdeclast(xmm_result, xmm_temp);
2644 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2645 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2646 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2647 // no need to store r to memory until we exit
2648 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2649 __ addptr(pos, AESBlockSize);
2650 __ subptr(len_reg, AESBlockSize);
2651 __ jmp(L_singleBlock_loopTop_128);
2652
2653
2654 __ BIND(L_exit);
2655 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2656 __ movptr(rvec , rvec_param); // restore this since used in loop
2657 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
2658 handleSOERegisters(false /*restoring*/);
2659 __ movptr(rax, len_param); // return length
2660 __ leave(); // required for proper stackwalking of RuntimeStub frame
2661 __ ret(0);
2662
2663
2664 __ BIND(L_key_192_256);
2665 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2666 __ cmpl(rax, 52);
2667 __ jcc(Assembler::notEqual, L_key_256);
2668
2669 // 192-bit code follows here (could be optimized to use parallelism)
2670 __ movl(pos, 0);
2671 __ align(OptoLoopAlignment);
2672 __ BIND(L_singleBlock_loopTop_192);
2673 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2674 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2675 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2676 __ aesdec(xmm_result, as_XMMRegister(rnum));
2677 }
2678 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
2679 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2680 }
2681 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2682 __ aesdeclast(xmm_result, xmm_temp);
2683 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2684 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2685 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2686 // no need to store r to memory until we exit
2687 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2688 __ addptr(pos, AESBlockSize);
2689 __ subptr(len_reg, AESBlockSize);
2690 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2691 __ jmp(L_exit);
2692
2693 __ BIND(L_key_256);
2694 // 256-bit code follows here (could be optimized to use parallelism)
2695 __ movl(pos, 0);
2696 __ align(OptoLoopAlignment);
2697 __ BIND(L_singleBlock_loopTop_256);
2698 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2699 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2700 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2701 __ aesdec(xmm_result, as_XMMRegister(rnum));
2702 }
2703 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
2704 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2705 }
2706 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2707 __ aesdeclast(xmm_result, xmm_temp);
2708 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2709 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2710 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2711 // no need to store r to memory until we exit
2712 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2713 __ addptr(pos, AESBlockSize);
2714 __ subptr(len_reg, AESBlockSize);
2715 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2716 __ jmp(L_exit);
2717
2718 return start;
2719 }
2720
2721 /**
2722 * Arguments:
2723 *
2724 * Inputs:
2725 * rsp(4) - int crc
2726 * rsp(8) - byte* buf
2727 * rsp(12) - int length
2728 *
2729 * Ouput:
2730 * rax - int crc result
2731 */
2732 address generate_updateBytesCRC32() {
2733 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2734
2735 __ align(CodeEntryAlignment);
2736 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2737
2738 address start = __ pc();
2739
2740 const Register crc = rdx; // crc
2741 const Register buf = rsi; // source java byte array address
2742 const Register len = rcx; // length
2743 const Register table = rdi; // crc_table address (reuse register)
2744 const Register tmp = rbx;
2745 assert_different_registers(crc, buf, len, table, tmp, rax);
2746
2747 BLOCK_COMMENT("Entry:");
2748 __ enter(); // required for proper stackwalking of RuntimeStub frame
2749 __ push(rsi);
2750 __ push(rdi);
2751 __ push(rbx);
2752
2753 Address crc_arg(rbp, 8 + 0);
2754 Address buf_arg(rbp, 8 + 4);
2755 Address len_arg(rbp, 8 + 8);
2756
2757 // Load up:
2758 __ movl(crc, crc_arg);
2759 __ movptr(buf, buf_arg);
2760 __ movl(len, len_arg);
2761
2762 __ kernel_crc32(crc, buf, len, table, tmp);
2763
2764 __ movl(rax, crc);
2765 __ pop(rbx);
2766 __ pop(rdi);
2767 __ pop(rsi);
2768 __ leave(); // required for proper stackwalking of RuntimeStub frame
2769 __ ret(0);
2770
2771 return start;
2772 }
2773
2774 // Safefetch stubs.
2775 void generate_safefetch(const char* name, int size, address* entry,
2776 address* fault_pc, address* continuation_pc) {
2777 // safefetch signatures:
2778 // int SafeFetch32(int* adr, int errValue);
2779 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2780
2781 StubCodeMark mark(this, "StubRoutines", name);
2782
2783 // Entry point, pc or function descriptor.
2784 *entry = __ pc();
2785
2786 __ movl(rax, Address(rsp, 0x8));
2787 __ movl(rcx, Address(rsp, 0x4));
2788 // Load *adr into eax, may fault.
2789 *fault_pc = __ pc();
2790 switch (size) {
2791 case 4:
2792 // int32_t
2793 __ movl(rax, Address(rcx, 0));
2794 break;
2795 case 8:
2796 // int64_t
2797 Unimplemented();
2798 break;
2799 default:
2800 ShouldNotReachHere();
2801 }
2802
2803 // Return errValue or *adr.
2804 *continuation_pc = __ pc();
2805 __ ret(0);
2806 }
2807
2808 public:
2809 // Information about frame layout at time of blocking runtime call.
2810 // Note that we only have to preserve callee-saved registers since
2811 // the compilers are responsible for supplying a continuation point
2812 // if they expect all registers to be preserved.
2813 enum layout {
2814 thread_off, // last_java_sp
2815 arg1_off,
2816 arg2_off,
2817 rbp_off, // callee saved register
2818 ret_pc,
2819 framesize
2820 };
2821
2822 private:
2823
2824 #undef __
2825 #define __ masm->
2826
2827 //------------------------------------------------------------------------------------------------------------------------
2828 // Continuation point for throwing of implicit exceptions that are not handled in
2829 // the current activation. Fabricates an exception oop and initiates normal
2830 // exception dispatching in this frame.
2831 //
2832 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2833 // This is no longer true after adapter frames were removed but could possibly
2834 // be brought back in the future if the interpreter code was reworked and it
2835 // was deemed worthwhile. The comment below was left to describe what must
2836 // happen here if callee saves were resurrected. As it stands now this stub
2837 // could actually be a vanilla BufferBlob and have now oopMap at all.
2838 // Since it doesn't make much difference we've chosen to leave it the
2839 // way it was in the callee save days and keep the comment.
2840
2841 // If we need to preserve callee-saved values we need a callee-saved oop map and
2842 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2843 // If the compiler needs all registers to be preserved between the fault
2844 // point and the exception handler then it must assume responsibility for that in
2845 // AbstractCompiler::continuation_for_implicit_null_exception or
2846 // continuation_for_implicit_division_by_zero_exception. All other implicit
2847 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2848 // either at call sites or otherwise assume that stack unwinding will be initiated,
2849 // so caller saved registers were assumed volatile in the compiler.
2850 address generate_throw_exception(const char* name, address runtime_entry,
2851 Register arg1 = noreg, Register arg2 = noreg) {
2852
2853 int insts_size = 256;
2854 int locs_size = 32;
2855
2856 CodeBuffer code(name, insts_size, locs_size);
2857 OopMapSet* oop_maps = new OopMapSet();
2858 MacroAssembler* masm = new MacroAssembler(&code);
2859
2860 address start = __ pc();
2861
2862 // This is an inlined and slightly modified version of call_VM
2863 // which has the ability to fetch the return PC out of
2864 // thread-local storage and also sets up last_Java_sp slightly
2865 // differently than the real call_VM
2866 Register java_thread = rbx;
2867 __ get_thread(java_thread);
2868
2869 __ enter(); // required for proper stackwalking of RuntimeStub frame
2870
2871 // pc and rbp, already pushed
2872 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2873
2874 // Frame is now completed as far as size and linkage.
2875
2876 int frame_complete = __ pc() - start;
2877
2878 // push java thread (becomes first argument of C function)
2879 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2880 if (arg1 != noreg) {
2881 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
2882 }
2883 if (arg2 != noreg) {
2884 assert(arg1 != noreg, "missing reg arg");
2885 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
2886 }
2887
2888 // Set up last_Java_sp and last_Java_fp
2889 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2890
2891 // Call runtime
2892 BLOCK_COMMENT("call runtime_entry");
2893 __ call(RuntimeAddress(runtime_entry));
2894 // Generate oop map
2895 OopMap* map = new OopMap(framesize, 0);
2896 oop_maps->add_gc_map(__ pc() - start, map);
2897
2898 // restore the thread (cannot use the pushed argument since arguments
2899 // may be overwritten by C code generated by an optimizing compiler);
2900 // however can use the register value directly if it is callee saved.
2901 __ get_thread(java_thread);
2902
2903 __ reset_last_Java_frame(java_thread, true, false);
2904
2905 __ leave(); // required for proper stackwalking of RuntimeStub frame
2906
2907 // check for pending exceptions
2908 #ifdef ASSERT
2909 Label L;
2910 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2911 __ jcc(Assembler::notEqual, L);
2912 __ should_not_reach_here();
2913 __ bind(L);
2914 #endif /* ASSERT */
2915 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2916
2917
2918 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2919 return stub->entry_point();
2920 }
2921
2922
2923 void create_control_words() {
2924 // Round to nearest, 53-bit mode, exceptions masked
2925 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2926 // Round to zero, 53-bit mode, exception mased
2927 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2928 // Round to nearest, 24-bit mode, exceptions masked
2929 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2930 // Round to nearest, 64-bit mode, exceptions masked
2931 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2932 // Round to nearest, 64-bit mode, exceptions masked
2933 StubRoutines::_mxcsr_std = 0x1F80;
2934 // Note: the following two constants are 80-bit values
2935 // layout is critical for correct loading by FPU.
2936 // Bias for strict fp multiply/divide
2937 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2938 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2939 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2940 // Un-Bias for strict fp multiply/divide
2941 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2942 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2943 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2944 }
2945
2946 //---------------------------------------------------------------------------
2947 // Initialization
2948
2949 void generate_initial() {
2950 // Generates all stubs and initializes the entry points
2951
2952 //------------------------------------------------------------------------------------------------------------------------
2953 // entry points that exist in all platforms
2954 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2955 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2956 StubRoutines::_forward_exception_entry = generate_forward_exception();
2957
2958 StubRoutines::_call_stub_entry =
2959 generate_call_stub(StubRoutines::_call_stub_return_address);
2960 // is referenced by megamorphic call
2961 StubRoutines::_catch_exception_entry = generate_catch_exception();
2962
2963 // These are currently used by Solaris/Intel
2964 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2965
2966 StubRoutines::_handler_for_unsafe_access_entry =
2967 generate_handler_for_unsafe_access();
2968
2969 // platform dependent
2970 create_control_words();
2971
2972 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2973 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2974 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2975 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2976 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2977 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2978
2979 // Build this early so it's available for the interpreter
2980 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
2981
2982 if (UseCRC32Intrinsics) {
2983 // set table address before stub generation which use it
2984 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
2985 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
2986 }
2987 }
2988
2989
2990 void generate_all() {
2991 // Generates all stubs and initializes the entry points
2992
2993 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2994 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2995 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
2996 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
2997 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
2998
2999 //------------------------------------------------------------------------------------------------------------------------
3000 // entry points that are platform specific
3001
3002 // support for verify_oop (must happen after universe_init)
3003 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3004
3005 // arraycopy stubs used by compilers
3006 generate_arraycopy_stubs();
3007
3008 generate_math_stubs();
3009
3010 // don't bother generating these AES intrinsic stubs unless global flag is set
3011 if (UseAESIntrinsics) {
3012 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3013
3014 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3015 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3016 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3017 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
3018 }
3019
3020 // Safefetch stubs.
3021 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3022 &StubRoutines::_safefetch32_fault_pc,
3023 &StubRoutines::_safefetch32_continuation_pc);
3024 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3025 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3026 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3027 }
3028
3029
3030 public:
3031 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3032 if (all) {
3033 generate_all();
3034 } else {
3035 generate_initial();
3036 }
3037 }
3038 }; // end class declaration
3039
3040
3041 void StubGenerator_generate(CodeBuffer* code, bool all) {
3042 StubGenerator g(code, all);
3043 }