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 case BarrierSet::Epsilon:
723 break;
724 default :
725 ShouldNotReachHere();
726
727 }
728 }
729
730
731 //
732 // Generate a post-barrier for an array store
733 //
734 // start - starting address
735 // count - element count
736 //
737 // The two input registers are overwritten.
738 //
739 void gen_write_ref_array_post_barrier(Register start, Register count) {
740 BarrierSet* bs = Universe::heap()->barrier_set();
741 assert_different_registers(start, count);
742 switch (bs->kind()) {
743 case BarrierSet::G1SATBCT:
744 case BarrierSet::G1SATBCTLogging:
745 {
746 __ pusha(); // push registers
747 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
748 start, count);
749 __ popa();
750 }
751 break;
752
753 case BarrierSet::CardTableModRef:
754 case BarrierSet::CardTableExtension:
755 {
756 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
757 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
758
759 Label L_loop;
760 const Register end = count; // elements count; end == start+count-1
761 assert_different_registers(start, end);
762
763 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
764 __ shrptr(start, CardTableModRefBS::card_shift);
765 __ shrptr(end, CardTableModRefBS::card_shift);
766 __ subptr(end, start); // end --> count
767 __ BIND(L_loop);
768 intptr_t disp = (intptr_t) ct->byte_map_base;
769 Address cardtable(start, count, Address::times_1, disp);
770 __ movb(cardtable, 0);
771 __ decrement(count);
772 __ jcc(Assembler::greaterEqual, L_loop);
773 }
774 break;
775 case BarrierSet::ModRef:
776 case BarrierSet::Epsilon:
777 break;
778 default :
779 ShouldNotReachHere();
780
781 }
782 }
783
784
785 // Copy 64 bytes chunks
786 //
787 // Inputs:
788 // from - source array address
789 // to_from - destination array address - from
790 // qword_count - 8-bytes element count, negative
791 //
792 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
793 assert( UseSSE >= 2, "supported cpu only" );
794 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
795 // Copy 64-byte chunks
796 __ jmpb(L_copy_64_bytes);
797 __ align(OptoLoopAlignment);
798 __ BIND(L_copy_64_bytes_loop);
799
800 if (UseUnalignedLoadStores) {
801 if (UseAVX >= 2) {
802 __ vmovdqu(xmm0, Address(from, 0));
803 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
804 __ vmovdqu(xmm1, Address(from, 32));
805 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
806 } else {
807 __ movdqu(xmm0, Address(from, 0));
808 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
809 __ movdqu(xmm1, Address(from, 16));
810 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
811 __ movdqu(xmm2, Address(from, 32));
812 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
813 __ movdqu(xmm3, Address(from, 48));
814 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
815 }
816 } else {
817 __ movq(xmm0, Address(from, 0));
818 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
819 __ movq(xmm1, Address(from, 8));
820 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
821 __ movq(xmm2, Address(from, 16));
822 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
823 __ movq(xmm3, Address(from, 24));
824 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
825 __ movq(xmm4, Address(from, 32));
826 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
827 __ movq(xmm5, Address(from, 40));
828 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
829 __ movq(xmm6, Address(from, 48));
830 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
831 __ movq(xmm7, Address(from, 56));
832 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
833 }
834
835 __ addl(from, 64);
836 __ BIND(L_copy_64_bytes);
837 __ subl(qword_count, 8);
838 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
839
840 if (UseUnalignedLoadStores && (UseAVX >= 2)) {
841 // clean upper bits of YMM registers
842 __ vpxor(xmm0, xmm0);
843 __ vpxor(xmm1, xmm1);
844 }
845 __ addl(qword_count, 8);
846 __ jccb(Assembler::zero, L_exit);
847 //
848 // length is too short, just copy qwords
849 //
850 __ BIND(L_copy_8_bytes);
851 __ movq(xmm0, Address(from, 0));
852 __ movq(Address(from, to_from, Address::times_1), xmm0);
853 __ addl(from, 8);
854 __ decrement(qword_count);
855 __ jcc(Assembler::greater, L_copy_8_bytes);
856 __ BIND(L_exit);
857 }
858
859 // Copy 64 bytes chunks
860 //
861 // Inputs:
862 // from - source array address
863 // to_from - destination array address - from
864 // qword_count - 8-bytes element count, negative
865 //
866 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
867 assert( VM_Version::supports_mmx(), "supported cpu only" );
868 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
869 // Copy 64-byte chunks
870 __ jmpb(L_copy_64_bytes);
871 __ align(OptoLoopAlignment);
872 __ BIND(L_copy_64_bytes_loop);
873 __ movq(mmx0, Address(from, 0));
874 __ movq(mmx1, Address(from, 8));
875 __ movq(mmx2, Address(from, 16));
876 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
877 __ movq(mmx3, Address(from, 24));
878 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
879 __ movq(mmx4, Address(from, 32));
880 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
881 __ movq(mmx5, Address(from, 40));
882 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
883 __ movq(mmx6, Address(from, 48));
884 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
885 __ movq(mmx7, Address(from, 56));
886 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
887 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
888 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
889 __ addptr(from, 64);
890 __ BIND(L_copy_64_bytes);
891 __ subl(qword_count, 8);
892 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
893 __ addl(qword_count, 8);
894 __ jccb(Assembler::zero, L_exit);
895 //
896 // length is too short, just copy qwords
897 //
898 __ BIND(L_copy_8_bytes);
899 __ movq(mmx0, Address(from, 0));
900 __ movq(Address(from, to_from, Address::times_1), mmx0);
901 __ addptr(from, 8);
902 __ decrement(qword_count);
903 __ jcc(Assembler::greater, L_copy_8_bytes);
904 __ BIND(L_exit);
905 __ emms();
906 }
907
908 address generate_disjoint_copy(BasicType t, bool aligned,
909 Address::ScaleFactor sf,
910 address* entry, const char *name,
911 bool dest_uninitialized = false) {
912 __ align(CodeEntryAlignment);
913 StubCodeMark mark(this, "StubRoutines", name);
914 address start = __ pc();
915
916 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
917 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
918
919 int shift = Address::times_ptr - sf;
920
921 const Register from = rsi; // source array address
922 const Register to = rdi; // destination array address
923 const Register count = rcx; // elements count
924 const Register to_from = to; // (to - from)
925 const Register saved_to = rdx; // saved destination array address
926
927 __ enter(); // required for proper stackwalking of RuntimeStub frame
928 __ push(rsi);
929 __ push(rdi);
930 __ movptr(from , Address(rsp, 12+ 4));
931 __ movptr(to , Address(rsp, 12+ 8));
932 __ movl(count, Address(rsp, 12+ 12));
933
934 if (entry != NULL) {
935 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
936 BLOCK_COMMENT("Entry:");
937 }
938
939 if (t == T_OBJECT) {
940 __ testl(count, count);
941 __ jcc(Assembler::zero, L_0_count);
942 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
943 __ mov(saved_to, to); // save 'to'
944 }
945
946 __ subptr(to, from); // to --> to_from
947 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
948 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
949 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
950 // align source address at 4 bytes address boundary
951 if (t == T_BYTE) {
952 // One byte misalignment happens only for byte arrays
953 __ testl(from, 1);
954 __ jccb(Assembler::zero, L_skip_align1);
955 __ movb(rax, Address(from, 0));
956 __ movb(Address(from, to_from, Address::times_1, 0), rax);
957 __ increment(from);
958 __ decrement(count);
959 __ BIND(L_skip_align1);
960 }
961 // Two bytes misalignment happens only for byte and short (char) arrays
962 __ testl(from, 2);
963 __ jccb(Assembler::zero, L_skip_align2);
964 __ movw(rax, Address(from, 0));
965 __ movw(Address(from, to_from, Address::times_1, 0), rax);
966 __ addptr(from, 2);
967 __ subl(count, 1<<(shift-1));
968 __ BIND(L_skip_align2);
969 }
970 if (!VM_Version::supports_mmx()) {
971 __ mov(rax, count); // save 'count'
972 __ shrl(count, shift); // bytes count
973 __ addptr(to_from, from);// restore 'to'
974 __ rep_mov();
975 __ subptr(to_from, from);// restore 'to_from'
976 __ mov(count, rax); // restore 'count'
977 __ jmpb(L_copy_2_bytes); // all dwords were copied
978 } else {
979 if (!UseUnalignedLoadStores) {
980 // align to 8 bytes, we know we are 4 byte aligned to start
981 __ testptr(from, 4);
982 __ jccb(Assembler::zero, L_copy_64_bytes);
983 __ movl(rax, Address(from, 0));
984 __ movl(Address(from, to_from, Address::times_1, 0), rax);
985 __ addptr(from, 4);
986 __ subl(count, 1<<shift);
987 }
988 __ BIND(L_copy_64_bytes);
989 __ mov(rax, count);
990 __ shrl(rax, shift+1); // 8 bytes chunk count
991 //
992 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
993 //
994 if (UseXMMForArrayCopy) {
995 xmm_copy_forward(from, to_from, rax);
996 } else {
997 mmx_copy_forward(from, to_from, rax);
998 }
999 }
1000 // copy tailing dword
1001 __ BIND(L_copy_4_bytes);
1002 __ testl(count, 1<<shift);
1003 __ jccb(Assembler::zero, L_copy_2_bytes);
1004 __ movl(rax, Address(from, 0));
1005 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1006 if (t == T_BYTE || t == T_SHORT) {
1007 __ addptr(from, 4);
1008 __ BIND(L_copy_2_bytes);
1009 // copy tailing word
1010 __ testl(count, 1<<(shift-1));
1011 __ jccb(Assembler::zero, L_copy_byte);
1012 __ movw(rax, Address(from, 0));
1013 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1014 if (t == T_BYTE) {
1015 __ addptr(from, 2);
1016 __ BIND(L_copy_byte);
1017 // copy tailing byte
1018 __ testl(count, 1);
1019 __ jccb(Assembler::zero, L_exit);
1020 __ movb(rax, Address(from, 0));
1021 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1022 __ BIND(L_exit);
1023 } else {
1024 __ BIND(L_copy_byte);
1025 }
1026 } else {
1027 __ BIND(L_copy_2_bytes);
1028 }
1029
1030 if (t == T_OBJECT) {
1031 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1032 __ mov(to, saved_to); // restore 'to'
1033 gen_write_ref_array_post_barrier(to, count);
1034 __ BIND(L_0_count);
1035 }
1036 inc_copy_counter_np(t);
1037 __ pop(rdi);
1038 __ pop(rsi);
1039 __ leave(); // required for proper stackwalking of RuntimeStub frame
1040 __ xorptr(rax, rax); // return 0
1041 __ ret(0);
1042 return start;
1043 }
1044
1045
1046 address generate_fill(BasicType t, bool aligned, const char *name) {
1047 __ align(CodeEntryAlignment);
1048 StubCodeMark mark(this, "StubRoutines", name);
1049 address start = __ pc();
1050
1051 BLOCK_COMMENT("Entry:");
1052
1053 const Register to = rdi; // source array address
1054 const Register value = rdx; // value
1055 const Register count = rsi; // elements count
1056
1057 __ enter(); // required for proper stackwalking of RuntimeStub frame
1058 __ push(rsi);
1059 __ push(rdi);
1060 __ movptr(to , Address(rsp, 12+ 4));
1061 __ movl(value, Address(rsp, 12+ 8));
1062 __ movl(count, Address(rsp, 12+ 12));
1063
1064 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1065
1066 __ pop(rdi);
1067 __ pop(rsi);
1068 __ leave(); // required for proper stackwalking of RuntimeStub frame
1069 __ ret(0);
1070 return start;
1071 }
1072
1073 address generate_conjoint_copy(BasicType t, bool aligned,
1074 Address::ScaleFactor sf,
1075 address nooverlap_target,
1076 address* entry, const char *name,
1077 bool dest_uninitialized = false) {
1078 __ align(CodeEntryAlignment);
1079 StubCodeMark mark(this, "StubRoutines", name);
1080 address start = __ pc();
1081
1082 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1083 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1084
1085 int shift = Address::times_ptr - sf;
1086
1087 const Register src = rax; // source array address
1088 const Register dst = rdx; // destination array address
1089 const Register from = rsi; // source array address
1090 const Register to = rdi; // destination array address
1091 const Register count = rcx; // elements count
1092 const Register end = rax; // array end address
1093
1094 __ enter(); // required for proper stackwalking of RuntimeStub frame
1095 __ push(rsi);
1096 __ push(rdi);
1097 __ movptr(src , Address(rsp, 12+ 4)); // from
1098 __ movptr(dst , Address(rsp, 12+ 8)); // to
1099 __ movl2ptr(count, Address(rsp, 12+12)); // count
1100
1101 if (entry != NULL) {
1102 *entry = __ pc(); // Entry point from generic arraycopy stub.
1103 BLOCK_COMMENT("Entry:");
1104 }
1105
1106 // nooverlap_target expects arguments in rsi and rdi.
1107 __ mov(from, src);
1108 __ mov(to , dst);
1109
1110 // arrays overlap test: dispatch to disjoint stub if necessary.
1111 RuntimeAddress nooverlap(nooverlap_target);
1112 __ cmpptr(dst, src);
1113 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1114 __ jump_cc(Assembler::belowEqual, nooverlap);
1115 __ cmpptr(dst, end);
1116 __ jump_cc(Assembler::aboveEqual, nooverlap);
1117
1118 if (t == T_OBJECT) {
1119 __ testl(count, count);
1120 __ jcc(Assembler::zero, L_0_count);
1121 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1122 }
1123
1124 // copy from high to low
1125 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1126 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1127 if (t == T_BYTE || t == T_SHORT) {
1128 // Align the end of destination array at 4 bytes address boundary
1129 __ lea(end, Address(dst, count, sf, 0));
1130 if (t == T_BYTE) {
1131 // One byte misalignment happens only for byte arrays
1132 __ testl(end, 1);
1133 __ jccb(Assembler::zero, L_skip_align1);
1134 __ decrement(count);
1135 __ movb(rdx, Address(from, count, sf, 0));
1136 __ movb(Address(to, count, sf, 0), rdx);
1137 __ BIND(L_skip_align1);
1138 }
1139 // Two bytes misalignment happens only for byte and short (char) arrays
1140 __ testl(end, 2);
1141 __ jccb(Assembler::zero, L_skip_align2);
1142 __ subptr(count, 1<<(shift-1));
1143 __ movw(rdx, Address(from, count, sf, 0));
1144 __ movw(Address(to, count, sf, 0), rdx);
1145 __ BIND(L_skip_align2);
1146 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1147 __ jcc(Assembler::below, L_copy_4_bytes);
1148 }
1149
1150 if (!VM_Version::supports_mmx()) {
1151 __ std();
1152 __ mov(rax, count); // Save 'count'
1153 __ mov(rdx, to); // Save 'to'
1154 __ lea(rsi, Address(from, count, sf, -4));
1155 __ lea(rdi, Address(to , count, sf, -4));
1156 __ shrptr(count, shift); // bytes count
1157 __ rep_mov();
1158 __ cld();
1159 __ mov(count, rax); // restore 'count'
1160 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1161 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1162 __ mov(to, rdx); // restore 'to'
1163 __ jmpb(L_copy_2_bytes); // all dword were copied
1164 } else {
1165 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1166 __ testptr(end, 4);
1167 __ jccb(Assembler::zero, L_copy_8_bytes);
1168 __ subl(count, 1<<shift);
1169 __ movl(rdx, Address(from, count, sf, 0));
1170 __ movl(Address(to, count, sf, 0), rdx);
1171 __ jmpb(L_copy_8_bytes);
1172
1173 __ align(OptoLoopAlignment);
1174 // Move 8 bytes
1175 __ BIND(L_copy_8_bytes_loop);
1176 if (UseXMMForArrayCopy) {
1177 __ movq(xmm0, Address(from, count, sf, 0));
1178 __ movq(Address(to, count, sf, 0), xmm0);
1179 } else {
1180 __ movq(mmx0, Address(from, count, sf, 0));
1181 __ movq(Address(to, count, sf, 0), mmx0);
1182 }
1183 __ BIND(L_copy_8_bytes);
1184 __ subl(count, 2<<shift);
1185 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1186 __ addl(count, 2<<shift);
1187 if (!UseXMMForArrayCopy) {
1188 __ emms();
1189 }
1190 }
1191 __ BIND(L_copy_4_bytes);
1192 // copy prefix qword
1193 __ testl(count, 1<<shift);
1194 __ jccb(Assembler::zero, L_copy_2_bytes);
1195 __ movl(rdx, Address(from, count, sf, -4));
1196 __ movl(Address(to, count, sf, -4), rdx);
1197
1198 if (t == T_BYTE || t == T_SHORT) {
1199 __ subl(count, (1<<shift));
1200 __ BIND(L_copy_2_bytes);
1201 // copy prefix dword
1202 __ testl(count, 1<<(shift-1));
1203 __ jccb(Assembler::zero, L_copy_byte);
1204 __ movw(rdx, Address(from, count, sf, -2));
1205 __ movw(Address(to, count, sf, -2), rdx);
1206 if (t == T_BYTE) {
1207 __ subl(count, 1<<(shift-1));
1208 __ BIND(L_copy_byte);
1209 // copy prefix byte
1210 __ testl(count, 1);
1211 __ jccb(Assembler::zero, L_exit);
1212 __ movb(rdx, Address(from, 0));
1213 __ movb(Address(to, 0), rdx);
1214 __ BIND(L_exit);
1215 } else {
1216 __ BIND(L_copy_byte);
1217 }
1218 } else {
1219 __ BIND(L_copy_2_bytes);
1220 }
1221 if (t == T_OBJECT) {
1222 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1223 gen_write_ref_array_post_barrier(to, count);
1224 __ BIND(L_0_count);
1225 }
1226 inc_copy_counter_np(t);
1227 __ pop(rdi);
1228 __ pop(rsi);
1229 __ leave(); // required for proper stackwalking of RuntimeStub frame
1230 __ xorptr(rax, rax); // return 0
1231 __ ret(0);
1232 return start;
1233 }
1234
1235
1236 address generate_disjoint_long_copy(address* entry, const char *name) {
1237 __ align(CodeEntryAlignment);
1238 StubCodeMark mark(this, "StubRoutines", name);
1239 address start = __ pc();
1240
1241 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1242 const Register from = rax; // source array address
1243 const Register to = rdx; // destination array address
1244 const Register count = rcx; // elements count
1245 const Register to_from = rdx; // (to - from)
1246
1247 __ enter(); // required for proper stackwalking of RuntimeStub frame
1248 __ movptr(from , Address(rsp, 8+0)); // from
1249 __ movptr(to , Address(rsp, 8+4)); // to
1250 __ movl2ptr(count, Address(rsp, 8+8)); // count
1251
1252 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1253 BLOCK_COMMENT("Entry:");
1254
1255 __ subptr(to, from); // to --> to_from
1256 if (VM_Version::supports_mmx()) {
1257 if (UseXMMForArrayCopy) {
1258 xmm_copy_forward(from, to_from, count);
1259 } else {
1260 mmx_copy_forward(from, to_from, count);
1261 }
1262 } else {
1263 __ jmpb(L_copy_8_bytes);
1264 __ align(OptoLoopAlignment);
1265 __ BIND(L_copy_8_bytes_loop);
1266 __ fild_d(Address(from, 0));
1267 __ fistp_d(Address(from, to_from, Address::times_1));
1268 __ addptr(from, 8);
1269 __ BIND(L_copy_8_bytes);
1270 __ decrement(count);
1271 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1272 }
1273 inc_copy_counter_np(T_LONG);
1274 __ leave(); // required for proper stackwalking of RuntimeStub frame
1275 __ xorptr(rax, rax); // return 0
1276 __ ret(0);
1277 return start;
1278 }
1279
1280 address generate_conjoint_long_copy(address nooverlap_target,
1281 address* entry, const char *name) {
1282 __ align(CodeEntryAlignment);
1283 StubCodeMark mark(this, "StubRoutines", name);
1284 address start = __ pc();
1285
1286 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1287 const Register from = rax; // source array address
1288 const Register to = rdx; // destination array address
1289 const Register count = rcx; // elements count
1290 const Register end_from = rax; // source array end address
1291
1292 __ enter(); // required for proper stackwalking of RuntimeStub frame
1293 __ movptr(from , Address(rsp, 8+0)); // from
1294 __ movptr(to , Address(rsp, 8+4)); // to
1295 __ movl2ptr(count, Address(rsp, 8+8)); // count
1296
1297 *entry = __ pc(); // Entry point from generic arraycopy stub.
1298 BLOCK_COMMENT("Entry:");
1299
1300 // arrays overlap test
1301 __ cmpptr(to, from);
1302 RuntimeAddress nooverlap(nooverlap_target);
1303 __ jump_cc(Assembler::belowEqual, nooverlap);
1304 __ lea(end_from, Address(from, count, Address::times_8, 0));
1305 __ cmpptr(to, end_from);
1306 __ movptr(from, Address(rsp, 8)); // from
1307 __ jump_cc(Assembler::aboveEqual, nooverlap);
1308
1309 __ jmpb(L_copy_8_bytes);
1310
1311 __ align(OptoLoopAlignment);
1312 __ BIND(L_copy_8_bytes_loop);
1313 if (VM_Version::supports_mmx()) {
1314 if (UseXMMForArrayCopy) {
1315 __ movq(xmm0, Address(from, count, Address::times_8));
1316 __ movq(Address(to, count, Address::times_8), xmm0);
1317 } else {
1318 __ movq(mmx0, Address(from, count, Address::times_8));
1319 __ movq(Address(to, count, Address::times_8), mmx0);
1320 }
1321 } else {
1322 __ fild_d(Address(from, count, Address::times_8));
1323 __ fistp_d(Address(to, count, Address::times_8));
1324 }
1325 __ BIND(L_copy_8_bytes);
1326 __ decrement(count);
1327 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1328
1329 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1330 __ emms();
1331 }
1332 inc_copy_counter_np(T_LONG);
1333 __ leave(); // required for proper stackwalking of RuntimeStub frame
1334 __ xorptr(rax, rax); // return 0
1335 __ ret(0);
1336 return start;
1337 }
1338
1339
1340 // Helper for generating a dynamic type check.
1341 // The sub_klass must be one of {rbx, rdx, rsi}.
1342 // The temp is killed.
1343 void generate_type_check(Register sub_klass,
1344 Address& super_check_offset_addr,
1345 Address& super_klass_addr,
1346 Register temp,
1347 Label* L_success, Label* L_failure) {
1348 BLOCK_COMMENT("type_check:");
1349
1350 Label L_fallthrough;
1351 #define LOCAL_JCC(assembler_con, label_ptr) \
1352 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1353 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1354
1355 // The following is a strange variation of the fast path which requires
1356 // one less register, because needed values are on the argument stack.
1357 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1358 // L_success, L_failure, NULL);
1359 assert_different_registers(sub_klass, temp);
1360
1361 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1362
1363 // if the pointers are equal, we are done (e.g., String[] elements)
1364 __ cmpptr(sub_klass, super_klass_addr);
1365 LOCAL_JCC(Assembler::equal, L_success);
1366
1367 // check the supertype display:
1368 __ movl2ptr(temp, super_check_offset_addr);
1369 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1370 __ movptr(temp, super_check_addr); // load displayed supertype
1371 __ cmpptr(temp, super_klass_addr); // test the super type
1372 LOCAL_JCC(Assembler::equal, L_success);
1373
1374 // if it was a primary super, we can just fail immediately
1375 __ cmpl(super_check_offset_addr, sc_offset);
1376 LOCAL_JCC(Assembler::notEqual, L_failure);
1377
1378 // The repne_scan instruction uses fixed registers, which will get spilled.
1379 // We happen to know this works best when super_klass is in rax.
1380 Register super_klass = temp;
1381 __ movptr(super_klass, super_klass_addr);
1382 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1383 L_success, L_failure);
1384
1385 __ bind(L_fallthrough);
1386
1387 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1388 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1389
1390 #undef LOCAL_JCC
1391 }
1392
1393 //
1394 // Generate checkcasting array copy stub
1395 //
1396 // Input:
1397 // 4(rsp) - source array address
1398 // 8(rsp) - destination array address
1399 // 12(rsp) - element count, can be zero
1400 // 16(rsp) - size_t ckoff (super_check_offset)
1401 // 20(rsp) - oop ckval (super_klass)
1402 //
1403 // Output:
1404 // rax, == 0 - success
1405 // rax, == -1^K - failure, where K is partial transfer count
1406 //
1407 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1408 __ align(CodeEntryAlignment);
1409 StubCodeMark mark(this, "StubRoutines", name);
1410 address start = __ pc();
1411
1412 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1413
1414 // register use:
1415 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1416 // rdi, rsi -- element access (oop, klass)
1417 // rbx, -- temp
1418 const Register from = rax; // source array address
1419 const Register to = rdx; // destination array address
1420 const Register length = rcx; // elements count
1421 const Register elem = rdi; // each oop copied
1422 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1423 const Register temp = rbx; // lone remaining temp
1424
1425 __ enter(); // required for proper stackwalking of RuntimeStub frame
1426
1427 __ push(rsi);
1428 __ push(rdi);
1429 __ push(rbx);
1430
1431 Address from_arg(rsp, 16+ 4); // from
1432 Address to_arg(rsp, 16+ 8); // to
1433 Address length_arg(rsp, 16+12); // elements count
1434 Address ckoff_arg(rsp, 16+16); // super_check_offset
1435 Address ckval_arg(rsp, 16+20); // super_klass
1436
1437 // Load up:
1438 __ movptr(from, from_arg);
1439 __ movptr(to, to_arg);
1440 __ movl2ptr(length, length_arg);
1441
1442 if (entry != NULL) {
1443 *entry = __ pc(); // Entry point from generic arraycopy stub.
1444 BLOCK_COMMENT("Entry:");
1445 }
1446
1447 //---------------------------------------------------------------
1448 // Assembler stub will be used for this call to arraycopy
1449 // if the two arrays are subtypes of Object[] but the
1450 // destination array type is not equal to or a supertype
1451 // of the source type. Each element must be separately
1452 // checked.
1453
1454 // Loop-invariant addresses. They are exclusive end pointers.
1455 Address end_from_addr(from, length, Address::times_ptr, 0);
1456 Address end_to_addr(to, length, Address::times_ptr, 0);
1457
1458 Register end_from = from; // re-use
1459 Register end_to = to; // re-use
1460 Register count = length; // re-use
1461
1462 // Loop-variant addresses. They assume post-incremented count < 0.
1463 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1464 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1465 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1466
1467 // Copy from low to high addresses, indexed from the end of each array.
1468 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1469 __ lea(end_from, end_from_addr);
1470 __ lea(end_to, end_to_addr);
1471 assert(length == count, ""); // else fix next line:
1472 __ negptr(count); // negate and test the length
1473 __ jccb(Assembler::notZero, L_load_element);
1474
1475 // Empty array: Nothing to do.
1476 __ xorptr(rax, rax); // return 0 on (trivial) success
1477 __ jmp(L_done);
1478
1479 // ======== begin loop ========
1480 // (Loop is rotated; its entry is L_load_element.)
1481 // Loop control:
1482 // for (count = -count; count != 0; count++)
1483 // Base pointers src, dst are biased by 8*count,to last element.
1484 __ align(OptoLoopAlignment);
1485
1486 __ BIND(L_store_element);
1487 __ movptr(to_element_addr, elem); // store the oop
1488 __ increment(count); // increment the count toward zero
1489 __ jccb(Assembler::zero, L_do_card_marks);
1490
1491 // ======== loop entry is here ========
1492 __ BIND(L_load_element);
1493 __ movptr(elem, from_element_addr); // load the oop
1494 __ testptr(elem, elem);
1495 __ jccb(Assembler::zero, L_store_element);
1496
1497 // (Could do a trick here: Remember last successful non-null
1498 // element stored and make a quick oop equality check on it.)
1499
1500 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1501 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1502 &L_store_element, NULL);
1503 // (On fall-through, we have failed the element type check.)
1504 // ======== end loop ========
1505
1506 // It was a real error; we must depend on the caller to finish the job.
1507 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1508 // Emit GC store barriers for the oops we have copied (length_arg + count),
1509 // and report their number to the caller.
1510 assert_different_registers(to, count, rax);
1511 Label L_post_barrier;
1512 __ addl(count, length_arg); // transfers = (length - remaining)
1513 __ movl2ptr(rax, count); // save the value
1514 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1515 __ jccb(Assembler::notZero, L_post_barrier);
1516 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1517
1518 // Come here on success only.
1519 __ BIND(L_do_card_marks);
1520 __ xorptr(rax, rax); // return 0 on success
1521 __ movl2ptr(count, length_arg);
1522
1523 __ BIND(L_post_barrier);
1524 __ movptr(to, to_arg); // reload
1525 gen_write_ref_array_post_barrier(to, count);
1526
1527 // Common exit point (success or failure).
1528 __ BIND(L_done);
1529 __ pop(rbx);
1530 __ pop(rdi);
1531 __ pop(rsi);
1532 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1533 __ leave(); // required for proper stackwalking of RuntimeStub frame
1534 __ ret(0);
1535
1536 return start;
1537 }
1538
1539 //
1540 // Generate 'unsafe' array copy stub
1541 // Though just as safe as the other stubs, it takes an unscaled
1542 // size_t argument instead of an element count.
1543 //
1544 // Input:
1545 // 4(rsp) - source array address
1546 // 8(rsp) - destination array address
1547 // 12(rsp) - byte count, can be zero
1548 //
1549 // Output:
1550 // rax, == 0 - success
1551 // rax, == -1 - need to call System.arraycopy
1552 //
1553 // Examines the alignment of the operands and dispatches
1554 // to a long, int, short, or byte copy loop.
1555 //
1556 address generate_unsafe_copy(const char *name,
1557 address byte_copy_entry,
1558 address short_copy_entry,
1559 address int_copy_entry,
1560 address long_copy_entry) {
1561
1562 Label L_long_aligned, L_int_aligned, L_short_aligned;
1563
1564 __ align(CodeEntryAlignment);
1565 StubCodeMark mark(this, "StubRoutines", name);
1566 address start = __ pc();
1567
1568 const Register from = rax; // source array address
1569 const Register to = rdx; // destination array address
1570 const Register count = rcx; // elements count
1571
1572 __ enter(); // required for proper stackwalking of RuntimeStub frame
1573 __ push(rsi);
1574 __ push(rdi);
1575 Address from_arg(rsp, 12+ 4); // from
1576 Address to_arg(rsp, 12+ 8); // to
1577 Address count_arg(rsp, 12+12); // byte count
1578
1579 // Load up:
1580 __ movptr(from , from_arg);
1581 __ movptr(to , to_arg);
1582 __ movl2ptr(count, count_arg);
1583
1584 // bump this on entry, not on exit:
1585 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1586
1587 const Register bits = rsi;
1588 __ mov(bits, from);
1589 __ orptr(bits, to);
1590 __ orptr(bits, count);
1591
1592 __ testl(bits, BytesPerLong-1);
1593 __ jccb(Assembler::zero, L_long_aligned);
1594
1595 __ testl(bits, BytesPerInt-1);
1596 __ jccb(Assembler::zero, L_int_aligned);
1597
1598 __ testl(bits, BytesPerShort-1);
1599 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1600
1601 __ BIND(L_short_aligned);
1602 __ shrptr(count, LogBytesPerShort); // size => short_count
1603 __ movl(count_arg, count); // update 'count'
1604 __ jump(RuntimeAddress(short_copy_entry));
1605
1606 __ BIND(L_int_aligned);
1607 __ shrptr(count, LogBytesPerInt); // size => int_count
1608 __ movl(count_arg, count); // update 'count'
1609 __ jump(RuntimeAddress(int_copy_entry));
1610
1611 __ BIND(L_long_aligned);
1612 __ shrptr(count, LogBytesPerLong); // size => qword_count
1613 __ movl(count_arg, count); // update 'count'
1614 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1615 __ pop(rsi);
1616 __ jump(RuntimeAddress(long_copy_entry));
1617
1618 return start;
1619 }
1620
1621
1622 // Perform range checks on the proposed arraycopy.
1623 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1624 void arraycopy_range_checks(Register src,
1625 Register src_pos,
1626 Register dst,
1627 Register dst_pos,
1628 Address& length,
1629 Label& L_failed) {
1630 BLOCK_COMMENT("arraycopy_range_checks:");
1631 const Register src_end = src_pos; // source array end position
1632 const Register dst_end = dst_pos; // destination array end position
1633 __ addl(src_end, length); // src_pos + length
1634 __ addl(dst_end, length); // dst_pos + length
1635
1636 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1637 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1638 __ jcc(Assembler::above, L_failed);
1639
1640 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1641 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1642 __ jcc(Assembler::above, L_failed);
1643
1644 BLOCK_COMMENT("arraycopy_range_checks done");
1645 }
1646
1647
1648 //
1649 // Generate generic array copy stubs
1650 //
1651 // Input:
1652 // 4(rsp) - src oop
1653 // 8(rsp) - src_pos
1654 // 12(rsp) - dst oop
1655 // 16(rsp) - dst_pos
1656 // 20(rsp) - element count
1657 //
1658 // Output:
1659 // rax, == 0 - success
1660 // rax, == -1^K - failure, where K is partial transfer count
1661 //
1662 address generate_generic_copy(const char *name,
1663 address entry_jbyte_arraycopy,
1664 address entry_jshort_arraycopy,
1665 address entry_jint_arraycopy,
1666 address entry_oop_arraycopy,
1667 address entry_jlong_arraycopy,
1668 address entry_checkcast_arraycopy) {
1669 Label L_failed, L_failed_0, L_objArray;
1670
1671 { int modulus = CodeEntryAlignment;
1672 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1673 int advance = target - (__ offset() % modulus);
1674 if (advance < 0) advance += modulus;
1675 if (advance > 0) __ nop(advance);
1676 }
1677 StubCodeMark mark(this, "StubRoutines", name);
1678
1679 // Short-hop target to L_failed. Makes for denser prologue code.
1680 __ BIND(L_failed_0);
1681 __ jmp(L_failed);
1682 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1683
1684 __ align(CodeEntryAlignment);
1685 address start = __ pc();
1686
1687 __ enter(); // required for proper stackwalking of RuntimeStub frame
1688 __ push(rsi);
1689 __ push(rdi);
1690
1691 // bump this on entry, not on exit:
1692 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1693
1694 // Input values
1695 Address SRC (rsp, 12+ 4);
1696 Address SRC_POS (rsp, 12+ 8);
1697 Address DST (rsp, 12+12);
1698 Address DST_POS (rsp, 12+16);
1699 Address LENGTH (rsp, 12+20);
1700
1701 //-----------------------------------------------------------------------
1702 // Assembler stub will be used for this call to arraycopy
1703 // if the following conditions are met:
1704 //
1705 // (1) src and dst must not be null.
1706 // (2) src_pos must not be negative.
1707 // (3) dst_pos must not be negative.
1708 // (4) length must not be negative.
1709 // (5) src klass and dst klass should be the same and not NULL.
1710 // (6) src and dst should be arrays.
1711 // (7) src_pos + length must not exceed length of src.
1712 // (8) dst_pos + length must not exceed length of dst.
1713 //
1714
1715 const Register src = rax; // source array oop
1716 const Register src_pos = rsi;
1717 const Register dst = rdx; // destination array oop
1718 const Register dst_pos = rdi;
1719 const Register length = rcx; // transfer count
1720
1721 // if (src == NULL) return -1;
1722 __ movptr(src, SRC); // src oop
1723 __ testptr(src, src);
1724 __ jccb(Assembler::zero, L_failed_0);
1725
1726 // if (src_pos < 0) return -1;
1727 __ movl2ptr(src_pos, SRC_POS); // src_pos
1728 __ testl(src_pos, src_pos);
1729 __ jccb(Assembler::negative, L_failed_0);
1730
1731 // if (dst == NULL) return -1;
1732 __ movptr(dst, DST); // dst oop
1733 __ testptr(dst, dst);
1734 __ jccb(Assembler::zero, L_failed_0);
1735
1736 // if (dst_pos < 0) return -1;
1737 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1738 __ testl(dst_pos, dst_pos);
1739 __ jccb(Assembler::negative, L_failed_0);
1740
1741 // if (length < 0) return -1;
1742 __ movl2ptr(length, LENGTH); // length
1743 __ testl(length, length);
1744 __ jccb(Assembler::negative, L_failed_0);
1745
1746 // if (src->klass() == NULL) return -1;
1747 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1748 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1749 const Register rcx_src_klass = rcx; // array klass
1750 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1751
1752 #ifdef ASSERT
1753 // assert(src->klass() != NULL);
1754 BLOCK_COMMENT("assert klasses not null");
1755 { Label L1, L2;
1756 __ testptr(rcx_src_klass, rcx_src_klass);
1757 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1758 __ bind(L1);
1759 __ stop("broken null klass");
1760 __ bind(L2);
1761 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1762 __ jccb(Assembler::equal, L1); // this would be broken also
1763 BLOCK_COMMENT("assert done");
1764 }
1765 #endif //ASSERT
1766
1767 // Load layout helper (32-bits)
1768 //
1769 // |array_tag| | header_size | element_type | |log2_element_size|
1770 // 32 30 24 16 8 2 0
1771 //
1772 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1773 //
1774
1775 int lh_offset = in_bytes(Klass::layout_helper_offset());
1776 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1777
1778 // Handle objArrays completely differently...
1779 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1780 __ cmpl(src_klass_lh_addr, objArray_lh);
1781 __ jcc(Assembler::equal, L_objArray);
1782
1783 // if (src->klass() != dst->klass()) return -1;
1784 __ cmpptr(rcx_src_klass, dst_klass_addr);
1785 __ jccb(Assembler::notEqual, L_failed_0);
1786
1787 const Register rcx_lh = rcx; // layout helper
1788 assert(rcx_lh == rcx_src_klass, "known alias");
1789 __ movl(rcx_lh, src_klass_lh_addr);
1790
1791 // if (!src->is_Array()) return -1;
1792 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1793 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1794
1795 // At this point, it is known to be a typeArray (array_tag 0x3).
1796 #ifdef ASSERT
1797 { Label L;
1798 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1799 __ jcc(Assembler::greaterEqual, L); // signed cmp
1800 __ stop("must be a primitive array");
1801 __ bind(L);
1802 }
1803 #endif
1804
1805 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1806 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1807
1808 // TypeArrayKlass
1809 //
1810 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1811 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1812 //
1813 const Register rsi_offset = rsi; // array offset
1814 const Register src_array = src; // src array offset
1815 const Register dst_array = dst; // dst array offset
1816 const Register rdi_elsize = rdi; // log2 element size
1817
1818 __ mov(rsi_offset, rcx_lh);
1819 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1820 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1821 __ addptr(src_array, rsi_offset); // src array offset
1822 __ addptr(dst_array, rsi_offset); // dst array offset
1823 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1824
1825 // next registers should be set before the jump to corresponding stub
1826 const Register from = src; // source array address
1827 const Register to = dst; // destination array address
1828 const Register count = rcx; // elements count
1829 // some of them should be duplicated on stack
1830 #define FROM Address(rsp, 12+ 4)
1831 #define TO Address(rsp, 12+ 8) // Not used now
1832 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1833
1834 BLOCK_COMMENT("scale indexes to element size");
1835 __ movl2ptr(rsi, SRC_POS); // src_pos
1836 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1837 assert(src_array == from, "");
1838 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1839 __ movl2ptr(rdi, DST_POS); // dst_pos
1840 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1841 assert(dst_array == to, "");
1842 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1843 __ movptr(FROM, from); // src_addr
1844 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1845 __ movl2ptr(count, LENGTH); // elements count
1846
1847 BLOCK_COMMENT("choose copy loop based on element size");
1848 __ cmpl(rdi_elsize, 0);
1849
1850 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1851 __ cmpl(rdi_elsize, LogBytesPerShort);
1852 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1853 __ cmpl(rdi_elsize, LogBytesPerInt);
1854 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1855 #ifdef ASSERT
1856 __ cmpl(rdi_elsize, LogBytesPerLong);
1857 __ jccb(Assembler::notEqual, L_failed);
1858 #endif
1859 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1860 __ pop(rsi);
1861 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1862
1863 __ BIND(L_failed);
1864 __ xorptr(rax, rax);
1865 __ notptr(rax); // return -1
1866 __ pop(rdi);
1867 __ pop(rsi);
1868 __ leave(); // required for proper stackwalking of RuntimeStub frame
1869 __ ret(0);
1870
1871 // ObjArrayKlass
1872 __ BIND(L_objArray);
1873 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1874
1875 Label L_plain_copy, L_checkcast_copy;
1876 // test array classes for subtyping
1877 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1878 __ jccb(Assembler::notEqual, L_checkcast_copy);
1879
1880 // Identically typed arrays can be copied without element-wise checks.
1881 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1882 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1883
1884 __ BIND(L_plain_copy);
1885 __ movl2ptr(count, LENGTH); // elements count
1886 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1887 __ lea(from, Address(src, src_pos, Address::times_ptr,
1888 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1889 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1890 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1891 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1892 __ movptr(FROM, from); // src_addr
1893 __ movptr(TO, to); // dst_addr
1894 __ movl(COUNT, count); // count
1895 __ jump(RuntimeAddress(entry_oop_arraycopy));
1896
1897 __ BIND(L_checkcast_copy);
1898 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1899 {
1900 // Handy offsets:
1901 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1902 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1903
1904 Register rsi_dst_klass = rsi;
1905 Register rdi_temp = rdi;
1906 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1907 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1908 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1909
1910 // Before looking at dst.length, make sure dst is also an objArray.
1911 __ movptr(rsi_dst_klass, dst_klass_addr);
1912 __ cmpl(dst_klass_lh_addr, objArray_lh);
1913 __ jccb(Assembler::notEqual, L_failed);
1914
1915 // It is safe to examine both src.length and dst.length.
1916 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1917 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1918 // (Now src_pos and dst_pos are killed, but not src and dst.)
1919
1920 // We'll need this temp (don't forget to pop it after the type check).
1921 __ push(rbx);
1922 Register rbx_src_klass = rbx;
1923
1924 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1925 __ movptr(rsi_dst_klass, dst_klass_addr);
1926 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1927 Label L_fail_array_check;
1928 generate_type_check(rbx_src_klass,
1929 super_check_offset_addr, dst_klass_addr,
1930 rdi_temp, NULL, &L_fail_array_check);
1931 // (On fall-through, we have passed the array type check.)
1932 __ pop(rbx);
1933 __ jmp(L_plain_copy);
1934
1935 __ BIND(L_fail_array_check);
1936 // Reshuffle arguments so we can call checkcast_arraycopy:
1937
1938 // match initial saves for checkcast_arraycopy
1939 // push(rsi); // already done; see above
1940 // push(rdi); // already done; see above
1941 // push(rbx); // already done; see above
1942
1943 // Marshal outgoing arguments now, freeing registers.
1944 Address from_arg(rsp, 16+ 4); // from
1945 Address to_arg(rsp, 16+ 8); // to
1946 Address length_arg(rsp, 16+12); // elements count
1947 Address ckoff_arg(rsp, 16+16); // super_check_offset
1948 Address ckval_arg(rsp, 16+20); // super_klass
1949
1950 Address SRC_POS_arg(rsp, 16+ 8);
1951 Address DST_POS_arg(rsp, 16+16);
1952 Address LENGTH_arg(rsp, 16+20);
1953 // push rbx, changed the incoming offsets (why not just use rbp,??)
1954 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1955
1956 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1957 __ movl2ptr(length, LENGTH_arg); // reload elements count
1958 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1959 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1960
1961 __ movptr(ckval_arg, rbx); // destination element type
1962 __ movl(rbx, Address(rbx, sco_offset));
1963 __ movl(ckoff_arg, rbx); // corresponding class check offset
1964
1965 __ movl(length_arg, length); // outgoing length argument
1966
1967 __ lea(from, Address(src, src_pos, Address::times_ptr,
1968 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1969 __ movptr(from_arg, from);
1970
1971 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1972 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1973 __ movptr(to_arg, to);
1974 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1975 }
1976
1977 return start;
1978 }
1979
1980 void generate_arraycopy_stubs() {
1981 address entry;
1982 address entry_jbyte_arraycopy;
1983 address entry_jshort_arraycopy;
1984 address entry_jint_arraycopy;
1985 address entry_oop_arraycopy;
1986 address entry_jlong_arraycopy;
1987 address entry_checkcast_arraycopy;
1988
1989 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1990 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1991 "arrayof_jbyte_disjoint_arraycopy");
1992 StubRoutines::_arrayof_jbyte_arraycopy =
1993 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1994 NULL, "arrayof_jbyte_arraycopy");
1995 StubRoutines::_jbyte_disjoint_arraycopy =
1996 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1997 "jbyte_disjoint_arraycopy");
1998 StubRoutines::_jbyte_arraycopy =
1999 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
2000 &entry_jbyte_arraycopy, "jbyte_arraycopy");
2001
2002 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2003 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2004 "arrayof_jshort_disjoint_arraycopy");
2005 StubRoutines::_arrayof_jshort_arraycopy =
2006 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2007 NULL, "arrayof_jshort_arraycopy");
2008 StubRoutines::_jshort_disjoint_arraycopy =
2009 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2010 "jshort_disjoint_arraycopy");
2011 StubRoutines::_jshort_arraycopy =
2012 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2013 &entry_jshort_arraycopy, "jshort_arraycopy");
2014
2015 // Next arrays are always aligned on 4 bytes at least.
2016 StubRoutines::_jint_disjoint_arraycopy =
2017 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2018 "jint_disjoint_arraycopy");
2019 StubRoutines::_jint_arraycopy =
2020 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2021 &entry_jint_arraycopy, "jint_arraycopy");
2022
2023 StubRoutines::_oop_disjoint_arraycopy =
2024 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2025 "oop_disjoint_arraycopy");
2026 StubRoutines::_oop_arraycopy =
2027 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2028 &entry_oop_arraycopy, "oop_arraycopy");
2029
2030 StubRoutines::_oop_disjoint_arraycopy_uninit =
2031 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2032 "oop_disjoint_arraycopy_uninit",
2033 /*dest_uninitialized*/true);
2034 StubRoutines::_oop_arraycopy_uninit =
2035 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2036 NULL, "oop_arraycopy_uninit",
2037 /*dest_uninitialized*/true);
2038
2039 StubRoutines::_jlong_disjoint_arraycopy =
2040 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2041 StubRoutines::_jlong_arraycopy =
2042 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2043 "jlong_arraycopy");
2044
2045 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2046 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2047 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2048 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2049 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2050 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2051
2052 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2053 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2054 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2055 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2056
2057 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2058 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2059 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2060 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2061
2062 StubRoutines::_checkcast_arraycopy =
2063 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2064 StubRoutines::_checkcast_arraycopy_uninit =
2065 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2066
2067 StubRoutines::_unsafe_arraycopy =
2068 generate_unsafe_copy("unsafe_arraycopy",
2069 entry_jbyte_arraycopy,
2070 entry_jshort_arraycopy,
2071 entry_jint_arraycopy,
2072 entry_jlong_arraycopy);
2073
2074 StubRoutines::_generic_arraycopy =
2075 generate_generic_copy("generic_arraycopy",
2076 entry_jbyte_arraycopy,
2077 entry_jshort_arraycopy,
2078 entry_jint_arraycopy,
2079 entry_oop_arraycopy,
2080 entry_jlong_arraycopy,
2081 entry_checkcast_arraycopy);
2082 }
2083
2084 void generate_math_stubs() {
2085 {
2086 StubCodeMark mark(this, "StubRoutines", "log");
2087 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2088
2089 __ fld_d(Address(rsp, 4));
2090 __ flog();
2091 __ ret(0);
2092 }
2093 {
2094 StubCodeMark mark(this, "StubRoutines", "log10");
2095 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2096
2097 __ fld_d(Address(rsp, 4));
2098 __ flog10();
2099 __ ret(0);
2100 }
2101 {
2102 StubCodeMark mark(this, "StubRoutines", "sin");
2103 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2104
2105 __ fld_d(Address(rsp, 4));
2106 __ trigfunc('s');
2107 __ ret(0);
2108 }
2109 {
2110 StubCodeMark mark(this, "StubRoutines", "cos");
2111 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2112
2113 __ fld_d(Address(rsp, 4));
2114 __ trigfunc('c');
2115 __ ret(0);
2116 }
2117 {
2118 StubCodeMark mark(this, "StubRoutines", "tan");
2119 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2120
2121 __ fld_d(Address(rsp, 4));
2122 __ trigfunc('t');
2123 __ ret(0);
2124 }
2125 {
2126 StubCodeMark mark(this, "StubRoutines", "exp");
2127 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2128
2129 __ fld_d(Address(rsp, 4));
2130 __ exp_with_fallback(0);
2131 __ ret(0);
2132 }
2133 {
2134 StubCodeMark mark(this, "StubRoutines", "pow");
2135 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2136
2137 __ fld_d(Address(rsp, 12));
2138 __ fld_d(Address(rsp, 4));
2139 __ pow_with_fallback(0);
2140 __ ret(0);
2141 }
2142 }
2143
2144 // AES intrinsic stubs
2145 enum {AESBlockSize = 16};
2146
2147 address generate_key_shuffle_mask() {
2148 __ align(16);
2149 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2150 address start = __ pc();
2151 __ emit_data(0x00010203, relocInfo::none, 0 );
2152 __ emit_data(0x04050607, relocInfo::none, 0 );
2153 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2154 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2155 return start;
2156 }
2157
2158 // Utility routine for loading a 128-bit key word in little endian format
2159 // can optionally specify that the shuffle mask is already in an xmmregister
2160 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2161 __ movdqu(xmmdst, Address(key, offset));
2162 if (xmm_shuf_mask != NULL) {
2163 __ pshufb(xmmdst, xmm_shuf_mask);
2164 } else {
2165 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2166 }
2167 }
2168
2169 // aesenc using specified key+offset
2170 // can optionally specify that the shuffle mask is already in an xmmregister
2171 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2172 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2173 __ aesenc(xmmdst, xmmtmp);
2174 }
2175
2176 // aesdec using specified key+offset
2177 // can optionally specify that the shuffle mask is already in an xmmregister
2178 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2179 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2180 __ aesdec(xmmdst, xmmtmp);
2181 }
2182
2183
2184 // Arguments:
2185 //
2186 // Inputs:
2187 // c_rarg0 - source byte array address
2188 // c_rarg1 - destination byte array address
2189 // c_rarg2 - K (key) in little endian int array
2190 //
2191 address generate_aescrypt_encryptBlock() {
2192 assert(UseAES, "need AES instructions and misaligned SSE support");
2193 __ align(CodeEntryAlignment);
2194 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2195 Label L_doLast;
2196 address start = __ pc();
2197
2198 const Register from = rdx; // source array address
2199 const Register to = rdx; // destination array address
2200 const Register key = rcx; // key array address
2201 const Register keylen = rax;
2202 const Address from_param(rbp, 8+0);
2203 const Address to_param (rbp, 8+4);
2204 const Address key_param (rbp, 8+8);
2205
2206 const XMMRegister xmm_result = xmm0;
2207 const XMMRegister xmm_key_shuf_mask = xmm1;
2208 const XMMRegister xmm_temp1 = xmm2;
2209 const XMMRegister xmm_temp2 = xmm3;
2210 const XMMRegister xmm_temp3 = xmm4;
2211 const XMMRegister xmm_temp4 = xmm5;
2212
2213 __ enter(); // required for proper stackwalking of RuntimeStub frame
2214 __ movptr(from, from_param);
2215 __ movptr(key, key_param);
2216
2217 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2218 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2219
2220 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2221 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2222 __ movptr(to, to_param);
2223
2224 // For encryption, the java expanded key ordering is just what we need
2225
2226 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2227 __ pxor(xmm_result, xmm_temp1);
2228
2229 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2230 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2231 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2232 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2233
2234 __ aesenc(xmm_result, xmm_temp1);
2235 __ aesenc(xmm_result, xmm_temp2);
2236 __ aesenc(xmm_result, xmm_temp3);
2237 __ aesenc(xmm_result, xmm_temp4);
2238
2239 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2240 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2241 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2242 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2243
2244 __ aesenc(xmm_result, xmm_temp1);
2245 __ aesenc(xmm_result, xmm_temp2);
2246 __ aesenc(xmm_result, xmm_temp3);
2247 __ aesenc(xmm_result, xmm_temp4);
2248
2249 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2250 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2251
2252 __ cmpl(keylen, 44);
2253 __ jccb(Assembler::equal, L_doLast);
2254
2255 __ aesenc(xmm_result, xmm_temp1);
2256 __ aesenc(xmm_result, xmm_temp2);
2257
2258 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2259 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2260
2261 __ cmpl(keylen, 52);
2262 __ jccb(Assembler::equal, L_doLast);
2263
2264 __ aesenc(xmm_result, xmm_temp1);
2265 __ aesenc(xmm_result, xmm_temp2);
2266
2267 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2268 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2269
2270 __ BIND(L_doLast);
2271 __ aesenc(xmm_result, xmm_temp1);
2272 __ aesenclast(xmm_result, xmm_temp2);
2273 __ movdqu(Address(to, 0), xmm_result); // store the result
2274 __ xorptr(rax, rax); // return 0
2275 __ leave(); // required for proper stackwalking of RuntimeStub frame
2276 __ ret(0);
2277
2278 return start;
2279 }
2280
2281
2282 // Arguments:
2283 //
2284 // Inputs:
2285 // c_rarg0 - source byte array address
2286 // c_rarg1 - destination byte array address
2287 // c_rarg2 - K (key) in little endian int array
2288 //
2289 address generate_aescrypt_decryptBlock() {
2290 assert(UseAES, "need AES instructions and misaligned SSE support");
2291 __ align(CodeEntryAlignment);
2292 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2293 Label L_doLast;
2294 address start = __ pc();
2295
2296 const Register from = rdx; // source array address
2297 const Register to = rdx; // destination array address
2298 const Register key = rcx; // key array address
2299 const Register keylen = rax;
2300 const Address from_param(rbp, 8+0);
2301 const Address to_param (rbp, 8+4);
2302 const Address key_param (rbp, 8+8);
2303
2304 const XMMRegister xmm_result = xmm0;
2305 const XMMRegister xmm_key_shuf_mask = xmm1;
2306 const XMMRegister xmm_temp1 = xmm2;
2307 const XMMRegister xmm_temp2 = xmm3;
2308 const XMMRegister xmm_temp3 = xmm4;
2309 const XMMRegister xmm_temp4 = xmm5;
2310
2311 __ enter(); // required for proper stackwalking of RuntimeStub frame
2312 __ movptr(from, from_param);
2313 __ movptr(key, key_param);
2314
2315 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2316 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2317
2318 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2319 __ movdqu(xmm_result, Address(from, 0));
2320 __ movptr(to, to_param);
2321
2322 // for decryption java expanded key ordering is rotated one position from what we want
2323 // so we start from 0x10 here and hit 0x00 last
2324 // we don't know if the key is aligned, hence not using load-execute form
2325 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2326 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2327 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2328 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2329
2330 __ pxor (xmm_result, xmm_temp1);
2331 __ aesdec(xmm_result, xmm_temp2);
2332 __ aesdec(xmm_result, xmm_temp3);
2333 __ aesdec(xmm_result, xmm_temp4);
2334
2335 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2336 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2337 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2338 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2339
2340 __ aesdec(xmm_result, xmm_temp1);
2341 __ aesdec(xmm_result, xmm_temp2);
2342 __ aesdec(xmm_result, xmm_temp3);
2343 __ aesdec(xmm_result, xmm_temp4);
2344
2345 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2346 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2347 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2348
2349 __ cmpl(keylen, 44);
2350 __ jccb(Assembler::equal, L_doLast);
2351
2352 __ aesdec(xmm_result, xmm_temp1);
2353 __ aesdec(xmm_result, xmm_temp2);
2354
2355 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2356 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2357
2358 __ cmpl(keylen, 52);
2359 __ jccb(Assembler::equal, L_doLast);
2360
2361 __ aesdec(xmm_result, xmm_temp1);
2362 __ aesdec(xmm_result, xmm_temp2);
2363
2364 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2365 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2366
2367 __ BIND(L_doLast);
2368 __ aesdec(xmm_result, xmm_temp1);
2369 __ aesdec(xmm_result, xmm_temp2);
2370
2371 // for decryption the aesdeclast operation is always on key+0x00
2372 __ aesdeclast(xmm_result, xmm_temp3);
2373 __ movdqu(Address(to, 0), xmm_result); // store the result
2374 __ xorptr(rax, rax); // return 0
2375 __ leave(); // required for proper stackwalking of RuntimeStub frame
2376 __ ret(0);
2377
2378 return start;
2379 }
2380
2381 void handleSOERegisters(bool saving) {
2382 const int saveFrameSizeInBytes = 4 * wordSize;
2383 const Address saved_rbx (rbp, -3 * wordSize);
2384 const Address saved_rsi (rbp, -2 * wordSize);
2385 const Address saved_rdi (rbp, -1 * wordSize);
2386
2387 if (saving) {
2388 __ subptr(rsp, saveFrameSizeInBytes);
2389 __ movptr(saved_rsi, rsi);
2390 __ movptr(saved_rdi, rdi);
2391 __ movptr(saved_rbx, rbx);
2392 } else {
2393 // restoring
2394 __ movptr(rsi, saved_rsi);
2395 __ movptr(rdi, saved_rdi);
2396 __ movptr(rbx, saved_rbx);
2397 }
2398 }
2399
2400 // Arguments:
2401 //
2402 // Inputs:
2403 // c_rarg0 - source byte array address
2404 // c_rarg1 - destination byte array address
2405 // c_rarg2 - K (key) in little endian int array
2406 // c_rarg3 - r vector byte array address
2407 // c_rarg4 - input length
2408 //
2409 // Output:
2410 // rax - input length
2411 //
2412 address generate_cipherBlockChaining_encryptAESCrypt() {
2413 assert(UseAES, "need AES instructions and misaligned SSE support");
2414 __ align(CodeEntryAlignment);
2415 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2416 address start = __ pc();
2417
2418 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2419 const Register from = rsi; // source array address
2420 const Register to = rdx; // destination array address
2421 const Register key = rcx; // key array address
2422 const Register rvec = rdi; // r byte array initialized from initvector array address
2423 // and left with the results of the last encryption block
2424 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2425 const Register pos = rax;
2426
2427 // xmm register assignments for the loops below
2428 const XMMRegister xmm_result = xmm0;
2429 const XMMRegister xmm_temp = xmm1;
2430 // first 6 keys preloaded into xmm2-xmm7
2431 const int XMM_REG_NUM_KEY_FIRST = 2;
2432 const int XMM_REG_NUM_KEY_LAST = 7;
2433 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2434
2435 __ enter(); // required for proper stackwalking of RuntimeStub frame
2436 handleSOERegisters(true /*saving*/);
2437
2438 // load registers from incoming parameters
2439 const Address from_param(rbp, 8+0);
2440 const Address to_param (rbp, 8+4);
2441 const Address key_param (rbp, 8+8);
2442 const Address rvec_param (rbp, 8+12);
2443 const Address len_param (rbp, 8+16);
2444 __ movptr(from , from_param);
2445 __ movptr(to , to_param);
2446 __ movptr(key , key_param);
2447 __ movptr(rvec , rvec_param);
2448 __ movptr(len_reg , len_param);
2449
2450 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2451 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2452 // load up xmm regs 2 thru 7 with keys 0-5
2453 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2454 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2455 offset += 0x10;
2456 }
2457
2458 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2459
2460 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2461 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2462 __ cmpl(rax, 44);
2463 __ jcc(Assembler::notEqual, L_key_192_256);
2464
2465 // 128 bit code follows here
2466 __ movl(pos, 0);
2467 __ align(OptoLoopAlignment);
2468 __ BIND(L_loopTop_128);
2469 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2470 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2471
2472 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2473 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2474 __ aesenc(xmm_result, as_XMMRegister(rnum));
2475 }
2476 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2477 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2478 }
2479 load_key(xmm_temp, key, 0xa0);
2480 __ aesenclast(xmm_result, xmm_temp);
2481
2482 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2483 // no need to store r to memory until we exit
2484 __ addptr(pos, AESBlockSize);
2485 __ subptr(len_reg, AESBlockSize);
2486 __ jcc(Assembler::notEqual, L_loopTop_128);
2487
2488 __ BIND(L_exit);
2489 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2490
2491 handleSOERegisters(false /*restoring*/);
2492 __ movptr(rax, len_param); // return length
2493 __ leave(); // required for proper stackwalking of RuntimeStub frame
2494 __ ret(0);
2495
2496 __ BIND(L_key_192_256);
2497 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2498 __ cmpl(rax, 52);
2499 __ jcc(Assembler::notEqual, L_key_256);
2500
2501 // 192-bit code follows here (could be changed to use more xmm registers)
2502 __ movl(pos, 0);
2503 __ align(OptoLoopAlignment);
2504 __ BIND(L_loopTop_192);
2505 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2506 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2507
2508 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2509 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2510 __ aesenc(xmm_result, as_XMMRegister(rnum));
2511 }
2512 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2513 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2514 }
2515 load_key(xmm_temp, key, 0xc0);
2516 __ aesenclast(xmm_result, xmm_temp);
2517
2518 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2519 // no need to store r to memory until we exit
2520 __ addptr(pos, AESBlockSize);
2521 __ subptr(len_reg, AESBlockSize);
2522 __ jcc(Assembler::notEqual, L_loopTop_192);
2523 __ jmp(L_exit);
2524
2525 __ BIND(L_key_256);
2526 // 256-bit code follows here (could be changed to use more xmm registers)
2527 __ movl(pos, 0);
2528 __ align(OptoLoopAlignment);
2529 __ BIND(L_loopTop_256);
2530 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2531 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2532
2533 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2534 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2535 __ aesenc(xmm_result, as_XMMRegister(rnum));
2536 }
2537 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2538 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2539 }
2540 load_key(xmm_temp, key, 0xe0);
2541 __ aesenclast(xmm_result, xmm_temp);
2542
2543 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2544 // no need to store r to memory until we exit
2545 __ addptr(pos, AESBlockSize);
2546 __ subptr(len_reg, AESBlockSize);
2547 __ jcc(Assembler::notEqual, L_loopTop_256);
2548 __ jmp(L_exit);
2549
2550 return start;
2551 }
2552
2553
2554 // CBC AES Decryption.
2555 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2556 //
2557 // Arguments:
2558 //
2559 // Inputs:
2560 // c_rarg0 - source byte array address
2561 // c_rarg1 - destination byte array address
2562 // c_rarg2 - K (key) in little endian int array
2563 // c_rarg3 - r vector byte array address
2564 // c_rarg4 - input length
2565 //
2566 // Output:
2567 // rax - input length
2568 //
2569
2570 address generate_cipherBlockChaining_decryptAESCrypt() {
2571 assert(UseAES, "need AES instructions and misaligned SSE support");
2572 __ align(CodeEntryAlignment);
2573 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2574 address start = __ pc();
2575
2576 Label L_exit, L_key_192_256, L_key_256;
2577 Label L_singleBlock_loopTop_128;
2578 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2579 const Register from = rsi; // source array address
2580 const Register to = rdx; // destination array address
2581 const Register key = rcx; // key array address
2582 const Register rvec = rdi; // r byte array initialized from initvector array address
2583 // and left with the results of the last encryption block
2584 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2585 const Register pos = rax;
2586
2587 // xmm register assignments for the loops below
2588 const XMMRegister xmm_result = xmm0;
2589 const XMMRegister xmm_temp = xmm1;
2590 // first 6 keys preloaded into xmm2-xmm7
2591 const int XMM_REG_NUM_KEY_FIRST = 2;
2592 const int XMM_REG_NUM_KEY_LAST = 7;
2593 const int FIRST_NON_REG_KEY_offset = 0x70;
2594 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2595
2596 __ enter(); // required for proper stackwalking of RuntimeStub frame
2597 handleSOERegisters(true /*saving*/);
2598
2599 // load registers from incoming parameters
2600 const Address from_param(rbp, 8+0);
2601 const Address to_param (rbp, 8+4);
2602 const Address key_param (rbp, 8+8);
2603 const Address rvec_param (rbp, 8+12);
2604 const Address len_param (rbp, 8+16);
2605 __ movptr(from , from_param);
2606 __ movptr(to , to_param);
2607 __ movptr(key , key_param);
2608 __ movptr(rvec , rvec_param);
2609 __ movptr(len_reg , len_param);
2610
2611 // the java expanded key ordering is rotated one position from what we want
2612 // so we start from 0x10 here and hit 0x00 last
2613 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
2614 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2615 // load up xmm regs 2 thru 6 with first 5 keys
2616 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2617 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2618 offset += 0x10;
2619 }
2620
2621 // inside here, use the rvec register to point to previous block cipher
2622 // with which we xor at the end of each newly decrypted block
2623 const Register prev_block_cipher_ptr = rvec;
2624
2625 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2626 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2627 __ cmpl(rax, 44);
2628 __ jcc(Assembler::notEqual, L_key_192_256);
2629
2630
2631 // 128-bit code follows here, parallelized
2632 __ movl(pos, 0);
2633 __ align(OptoLoopAlignment);
2634 __ BIND(L_singleBlock_loopTop_128);
2635 __ cmpptr(len_reg, 0); // any blocks left??
2636 __ jcc(Assembler::equal, L_exit);
2637 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2638 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2639 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2640 __ aesdec(xmm_result, as_XMMRegister(rnum));
2641 }
2642 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
2643 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2644 }
2645 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2646 __ aesdeclast(xmm_result, xmm_temp);
2647 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2648 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2649 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2650 // no need to store r to memory until we exit
2651 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2652 __ addptr(pos, AESBlockSize);
2653 __ subptr(len_reg, AESBlockSize);
2654 __ jmp(L_singleBlock_loopTop_128);
2655
2656
2657 __ BIND(L_exit);
2658 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2659 __ movptr(rvec , rvec_param); // restore this since used in loop
2660 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
2661 handleSOERegisters(false /*restoring*/);
2662 __ movptr(rax, len_param); // return length
2663 __ leave(); // required for proper stackwalking of RuntimeStub frame
2664 __ ret(0);
2665
2666
2667 __ BIND(L_key_192_256);
2668 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2669 __ cmpl(rax, 52);
2670 __ jcc(Assembler::notEqual, L_key_256);
2671
2672 // 192-bit code follows here (could be optimized to use parallelism)
2673 __ movl(pos, 0);
2674 __ align(OptoLoopAlignment);
2675 __ BIND(L_singleBlock_loopTop_192);
2676 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2677 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2678 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2679 __ aesdec(xmm_result, as_XMMRegister(rnum));
2680 }
2681 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
2682 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2683 }
2684 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2685 __ aesdeclast(xmm_result, xmm_temp);
2686 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2687 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2688 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2689 // no need to store r to memory until we exit
2690 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2691 __ addptr(pos, AESBlockSize);
2692 __ subptr(len_reg, AESBlockSize);
2693 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2694 __ jmp(L_exit);
2695
2696 __ BIND(L_key_256);
2697 // 256-bit code follows here (could be optimized to use parallelism)
2698 __ movl(pos, 0);
2699 __ align(OptoLoopAlignment);
2700 __ BIND(L_singleBlock_loopTop_256);
2701 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2702 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2703 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2704 __ aesdec(xmm_result, as_XMMRegister(rnum));
2705 }
2706 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
2707 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2708 }
2709 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2710 __ aesdeclast(xmm_result, xmm_temp);
2711 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2712 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2713 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2714 // no need to store r to memory until we exit
2715 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2716 __ addptr(pos, AESBlockSize);
2717 __ subptr(len_reg, AESBlockSize);
2718 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2719 __ jmp(L_exit);
2720
2721 return start;
2722 }
2723
2724 /**
2725 * Arguments:
2726 *
2727 * Inputs:
2728 * rsp(4) - int crc
2729 * rsp(8) - byte* buf
2730 * rsp(12) - int length
2731 *
2732 * Ouput:
2733 * rax - int crc result
2734 */
2735 address generate_updateBytesCRC32() {
2736 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2737
2738 __ align(CodeEntryAlignment);
2739 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2740
2741 address start = __ pc();
2742
2743 const Register crc = rdx; // crc
2744 const Register buf = rsi; // source java byte array address
2745 const Register len = rcx; // length
2746 const Register table = rdi; // crc_table address (reuse register)
2747 const Register tmp = rbx;
2748 assert_different_registers(crc, buf, len, table, tmp, rax);
2749
2750 BLOCK_COMMENT("Entry:");
2751 __ enter(); // required for proper stackwalking of RuntimeStub frame
2752 __ push(rsi);
2753 __ push(rdi);
2754 __ push(rbx);
2755
2756 Address crc_arg(rbp, 8 + 0);
2757 Address buf_arg(rbp, 8 + 4);
2758 Address len_arg(rbp, 8 + 8);
2759
2760 // Load up:
2761 __ movl(crc, crc_arg);
2762 __ movptr(buf, buf_arg);
2763 __ movl(len, len_arg);
2764
2765 __ kernel_crc32(crc, buf, len, table, tmp);
2766
2767 __ movl(rax, crc);
2768 __ pop(rbx);
2769 __ pop(rdi);
2770 __ pop(rsi);
2771 __ leave(); // required for proper stackwalking of RuntimeStub frame
2772 __ ret(0);
2773
2774 return start;
2775 }
2776
2777 // Safefetch stubs.
2778 void generate_safefetch(const char* name, int size, address* entry,
2779 address* fault_pc, address* continuation_pc) {
2780 // safefetch signatures:
2781 // int SafeFetch32(int* adr, int errValue);
2782 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2783
2784 StubCodeMark mark(this, "StubRoutines", name);
2785
2786 // Entry point, pc or function descriptor.
2787 *entry = __ pc();
2788
2789 __ movl(rax, Address(rsp, 0x8));
2790 __ movl(rcx, Address(rsp, 0x4));
2791 // Load *adr into eax, may fault.
2792 *fault_pc = __ pc();
2793 switch (size) {
2794 case 4:
2795 // int32_t
2796 __ movl(rax, Address(rcx, 0));
2797 break;
2798 case 8:
2799 // int64_t
2800 Unimplemented();
2801 break;
2802 default:
2803 ShouldNotReachHere();
2804 }
2805
2806 // Return errValue or *adr.
2807 *continuation_pc = __ pc();
2808 __ ret(0);
2809 }
2810
2811 public:
2812 // Information about frame layout at time of blocking runtime call.
2813 // Note that we only have to preserve callee-saved registers since
2814 // the compilers are responsible for supplying a continuation point
2815 // if they expect all registers to be preserved.
2816 enum layout {
2817 thread_off, // last_java_sp
2818 arg1_off,
2819 arg2_off,
2820 rbp_off, // callee saved register
2821 ret_pc,
2822 framesize
2823 };
2824
2825 private:
2826
2827 #undef __
2828 #define __ masm->
2829
2830 //------------------------------------------------------------------------------------------------------------------------
2831 // Continuation point for throwing of implicit exceptions that are not handled in
2832 // the current activation. Fabricates an exception oop and initiates normal
2833 // exception dispatching in this frame.
2834 //
2835 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2836 // This is no longer true after adapter frames were removed but could possibly
2837 // be brought back in the future if the interpreter code was reworked and it
2838 // was deemed worthwhile. The comment below was left to describe what must
2839 // happen here if callee saves were resurrected. As it stands now this stub
2840 // could actually be a vanilla BufferBlob and have now oopMap at all.
2841 // Since it doesn't make much difference we've chosen to leave it the
2842 // way it was in the callee save days and keep the comment.
2843
2844 // If we need to preserve callee-saved values we need a callee-saved oop map and
2845 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2846 // If the compiler needs all registers to be preserved between the fault
2847 // point and the exception handler then it must assume responsibility for that in
2848 // AbstractCompiler::continuation_for_implicit_null_exception or
2849 // continuation_for_implicit_division_by_zero_exception. All other implicit
2850 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2851 // either at call sites or otherwise assume that stack unwinding will be initiated,
2852 // so caller saved registers were assumed volatile in the compiler.
2853 address generate_throw_exception(const char* name, address runtime_entry,
2854 Register arg1 = noreg, Register arg2 = noreg) {
2855
2856 int insts_size = 256;
2857 int locs_size = 32;
2858
2859 CodeBuffer code(name, insts_size, locs_size);
2860 OopMapSet* oop_maps = new OopMapSet();
2861 MacroAssembler* masm = new MacroAssembler(&code);
2862
2863 address start = __ pc();
2864
2865 // This is an inlined and slightly modified version of call_VM
2866 // which has the ability to fetch the return PC out of
2867 // thread-local storage and also sets up last_Java_sp slightly
2868 // differently than the real call_VM
2869 Register java_thread = rbx;
2870 __ get_thread(java_thread);
2871
2872 __ enter(); // required for proper stackwalking of RuntimeStub frame
2873
2874 // pc and rbp, already pushed
2875 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2876
2877 // Frame is now completed as far as size and linkage.
2878
2879 int frame_complete = __ pc() - start;
2880
2881 // push java thread (becomes first argument of C function)
2882 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2883 if (arg1 != noreg) {
2884 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
2885 }
2886 if (arg2 != noreg) {
2887 assert(arg1 != noreg, "missing reg arg");
2888 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
2889 }
2890
2891 // Set up last_Java_sp and last_Java_fp
2892 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2893
2894 // Call runtime
2895 BLOCK_COMMENT("call runtime_entry");
2896 __ call(RuntimeAddress(runtime_entry));
2897 // Generate oop map
2898 OopMap* map = new OopMap(framesize, 0);
2899 oop_maps->add_gc_map(__ pc() - start, map);
2900
2901 // restore the thread (cannot use the pushed argument since arguments
2902 // may be overwritten by C code generated by an optimizing compiler);
2903 // however can use the register value directly if it is callee saved.
2904 __ get_thread(java_thread);
2905
2906 __ reset_last_Java_frame(java_thread, true);
2907
2908 __ leave(); // required for proper stackwalking of RuntimeStub frame
2909
2910 // check for pending exceptions
2911 #ifdef ASSERT
2912 Label L;
2913 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2914 __ jcc(Assembler::notEqual, L);
2915 __ should_not_reach_here();
2916 __ bind(L);
2917 #endif /* ASSERT */
2918 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2919
2920
2921 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2922 return stub->entry_point();
2923 }
2924
2925
2926 void create_control_words() {
2927 // Round to nearest, 53-bit mode, exceptions masked
2928 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2929 // Round to zero, 53-bit mode, exception mased
2930 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2931 // Round to nearest, 24-bit mode, exceptions masked
2932 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2933 // Round to nearest, 64-bit mode, exceptions masked
2934 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2935 // Round to nearest, 64-bit mode, exceptions masked
2936 StubRoutines::_mxcsr_std = 0x1F80;
2937 // Note: the following two constants are 80-bit values
2938 // layout is critical for correct loading by FPU.
2939 // Bias for strict fp multiply/divide
2940 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2941 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2942 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2943 // Un-Bias for strict fp multiply/divide
2944 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2945 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2946 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2947 }
2948
2949 //---------------------------------------------------------------------------
2950 // Initialization
2951
2952 void generate_initial() {
2953 // Generates all stubs and initializes the entry points
2954
2955 //------------------------------------------------------------------------------------------------------------------------
2956 // entry points that exist in all platforms
2957 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2958 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2959 StubRoutines::_forward_exception_entry = generate_forward_exception();
2960
2961 StubRoutines::_call_stub_entry =
2962 generate_call_stub(StubRoutines::_call_stub_return_address);
2963 // is referenced by megamorphic call
2964 StubRoutines::_catch_exception_entry = generate_catch_exception();
2965
2966 // These are currently used by Solaris/Intel
2967 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2968
2969 StubRoutines::_handler_for_unsafe_access_entry =
2970 generate_handler_for_unsafe_access();
2971
2972 // platform dependent
2973 create_control_words();
2974
2975 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2976 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2977 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2978 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2979 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2980 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2981
2982 // Build this early so it's available for the interpreter
2983 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
2984
2985 if (UseCRC32Intrinsics) {
2986 // set table address before stub generation which use it
2987 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
2988 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
2989 }
2990 }
2991
2992
2993 void generate_all() {
2994 // Generates all stubs and initializes the entry points
2995
2996 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2997 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2998 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
2999 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3000 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3001
3002 //------------------------------------------------------------------------------------------------------------------------
3003 // entry points that are platform specific
3004
3005 // support for verify_oop (must happen after universe_init)
3006 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3007
3008 // arraycopy stubs used by compilers
3009 generate_arraycopy_stubs();
3010
3011 generate_math_stubs();
3012
3013 // don't bother generating these AES intrinsic stubs unless global flag is set
3014 if (UseAESIntrinsics) {
3015 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3016
3017 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3018 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3019 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3020 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
3021 }
3022
3023 // Safefetch stubs.
3024 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3025 &StubRoutines::_safefetch32_fault_pc,
3026 &StubRoutines::_safefetch32_continuation_pc);
3027 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3028 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3029 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3030 }
3031
3032
3033 public:
3034 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3035 if (all) {
3036 generate_all();
3037 } else {
3038 generate_initial();
3039 }
3040 }
3041 }; // end class declaration
3042
3043
3044 void StubGenerator_generate(CodeBuffer* code, bool all) {
3045 StubGenerator g(code, all);
3046 }