1 /*
2 * Copyright (c) 1999, 2015, 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::G1SATBCTLogging:
710 // With G1, don't generate the call if we statically know that the target in uninitialized
711 if (!uninitialized_target) {
712 __ pusha(); // push registers
713 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
714 start, count);
715 __ popa();
716 }
717 break;
718 case BarrierSet::CardTableModRef:
719 case BarrierSet::CardTableExtension:
720 case BarrierSet::ModRef:
721 break;
722 default :
723 ShouldNotReachHere();
724
725 }
726 }
727
728
729 //
730 // Generate a post-barrier for an array store
731 //
732 // start - starting address
733 // count - element count
734 //
735 // The two input registers are overwritten.
736 //
737 void gen_write_ref_array_post_barrier(Register start, Register count) {
738 BarrierSet* bs = Universe::heap()->barrier_set();
739 assert_different_registers(start, count);
740 switch (bs->kind()) {
741 case BarrierSet::G1SATBCTLogging:
742 {
743 __ pusha(); // push registers
744 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
745 start, count);
746 __ popa();
747 }
748 break;
749
750 case BarrierSet::CardTableModRef:
751 case BarrierSet::CardTableExtension:
752 {
753 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
754 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
755
756 Label L_loop;
757 const Register end = count; // elements count; end == start+count-1
758 assert_different_registers(start, end);
759
760 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
761 __ shrptr(start, CardTableModRefBS::card_shift);
762 __ shrptr(end, CardTableModRefBS::card_shift);
763 __ subptr(end, start); // end --> count
764 __ BIND(L_loop);
765 intptr_t disp = (intptr_t) ct->byte_map_base;
766 Address cardtable(start, count, Address::times_1, disp);
767 __ movb(cardtable, 0);
768 __ decrement(count);
769 __ jcc(Assembler::greaterEqual, L_loop);
770 }
771 break;
772 case BarrierSet::ModRef:
773 break;
774 default :
775 ShouldNotReachHere();
776
777 }
778 }
779
780
781 // Copy 64 bytes chunks
782 //
783 // Inputs:
784 // from - source array address
785 // to_from - destination array address - from
786 // qword_count - 8-bytes element count, negative
787 //
788 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
789 assert( UseSSE >= 2, "supported cpu only" );
790 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
791 // Copy 64-byte chunks
792 __ jmpb(L_copy_64_bytes);
793 __ align(OptoLoopAlignment);
794 __ BIND(L_copy_64_bytes_loop);
795
796 if (UseUnalignedLoadStores) {
797 if (UseAVX >= 2) {
798 __ vmovdqu(xmm0, Address(from, 0));
799 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
800 __ vmovdqu(xmm1, Address(from, 32));
801 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
802 } else {
803 __ movdqu(xmm0, Address(from, 0));
804 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
805 __ movdqu(xmm1, Address(from, 16));
806 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
807 __ movdqu(xmm2, Address(from, 32));
808 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
809 __ movdqu(xmm3, Address(from, 48));
810 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
811 }
812 } else {
813 __ movq(xmm0, Address(from, 0));
814 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
815 __ movq(xmm1, Address(from, 8));
816 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
817 __ movq(xmm2, Address(from, 16));
818 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
819 __ movq(xmm3, Address(from, 24));
820 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
821 __ movq(xmm4, Address(from, 32));
822 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
823 __ movq(xmm5, Address(from, 40));
824 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
825 __ movq(xmm6, Address(from, 48));
826 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
827 __ movq(xmm7, Address(from, 56));
828 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
829 }
830
831 __ addl(from, 64);
832 __ BIND(L_copy_64_bytes);
833 __ subl(qword_count, 8);
834 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
835
836 if (UseUnalignedLoadStores && (UseAVX >= 2)) {
837 // clean upper bits of YMM registers
838 __ vzeroupper();
839 }
840 __ addl(qword_count, 8);
841 __ jccb(Assembler::zero, L_exit);
842 //
843 // length is too short, just copy qwords
844 //
845 __ BIND(L_copy_8_bytes);
846 __ movq(xmm0, Address(from, 0));
847 __ movq(Address(from, to_from, Address::times_1), xmm0);
848 __ addl(from, 8);
849 __ decrement(qword_count);
850 __ jcc(Assembler::greater, L_copy_8_bytes);
851 __ BIND(L_exit);
852 }
853
854 // Copy 64 bytes chunks
855 //
856 // Inputs:
857 // from - source array address
858 // to_from - destination array address - from
859 // qword_count - 8-bytes element count, negative
860 //
861 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
862 assert( VM_Version::supports_mmx(), "supported cpu only" );
863 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
864 // Copy 64-byte chunks
865 __ jmpb(L_copy_64_bytes);
866 __ align(OptoLoopAlignment);
867 __ BIND(L_copy_64_bytes_loop);
868 __ movq(mmx0, Address(from, 0));
869 __ movq(mmx1, Address(from, 8));
870 __ movq(mmx2, Address(from, 16));
871 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
872 __ movq(mmx3, Address(from, 24));
873 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
874 __ movq(mmx4, Address(from, 32));
875 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
876 __ movq(mmx5, Address(from, 40));
877 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
878 __ movq(mmx6, Address(from, 48));
879 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
880 __ movq(mmx7, Address(from, 56));
881 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
882 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
883 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
884 __ addptr(from, 64);
885 __ BIND(L_copy_64_bytes);
886 __ subl(qword_count, 8);
887 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
888 __ addl(qword_count, 8);
889 __ jccb(Assembler::zero, L_exit);
890 //
891 // length is too short, just copy qwords
892 //
893 __ BIND(L_copy_8_bytes);
894 __ movq(mmx0, Address(from, 0));
895 __ movq(Address(from, to_from, Address::times_1), mmx0);
896 __ addptr(from, 8);
897 __ decrement(qword_count);
898 __ jcc(Assembler::greater, L_copy_8_bytes);
899 __ BIND(L_exit);
900 __ emms();
901 }
902
903 address generate_disjoint_copy(BasicType t, bool aligned,
904 Address::ScaleFactor sf,
905 address* entry, const char *name,
906 bool dest_uninitialized = false) {
907 __ align(CodeEntryAlignment);
908 StubCodeMark mark(this, "StubRoutines", name);
909 address start = __ pc();
910
911 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
912 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
913
914 int shift = Address::times_ptr - sf;
915
916 const Register from = rsi; // source array address
917 const Register to = rdi; // destination array address
918 const Register count = rcx; // elements count
919 const Register to_from = to; // (to - from)
920 const Register saved_to = rdx; // saved destination array address
921
922 __ enter(); // required for proper stackwalking of RuntimeStub frame
923 __ push(rsi);
924 __ push(rdi);
925 __ movptr(from , Address(rsp, 12+ 4));
926 __ movptr(to , Address(rsp, 12+ 8));
927 __ movl(count, Address(rsp, 12+ 12));
928
929 if (entry != NULL) {
930 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
931 BLOCK_COMMENT("Entry:");
932 }
933
934 if (t == T_OBJECT) {
935 __ testl(count, count);
936 __ jcc(Assembler::zero, L_0_count);
937 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
938 __ mov(saved_to, to); // save 'to'
939 }
940
941 __ subptr(to, from); // to --> to_from
942 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
943 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
944 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
945 // align source address at 4 bytes address boundary
946 if (t == T_BYTE) {
947 // One byte misalignment happens only for byte arrays
948 __ testl(from, 1);
949 __ jccb(Assembler::zero, L_skip_align1);
950 __ movb(rax, Address(from, 0));
951 __ movb(Address(from, to_from, Address::times_1, 0), rax);
952 __ increment(from);
953 __ decrement(count);
954 __ BIND(L_skip_align1);
955 }
956 // Two bytes misalignment happens only for byte and short (char) arrays
957 __ testl(from, 2);
958 __ jccb(Assembler::zero, L_skip_align2);
959 __ movw(rax, Address(from, 0));
960 __ movw(Address(from, to_from, Address::times_1, 0), rax);
961 __ addptr(from, 2);
962 __ subl(count, 1<<(shift-1));
963 __ BIND(L_skip_align2);
964 }
965 if (!VM_Version::supports_mmx()) {
966 __ mov(rax, count); // save 'count'
967 __ shrl(count, shift); // bytes count
968 __ addptr(to_from, from);// restore 'to'
969 __ rep_mov();
970 __ subptr(to_from, from);// restore 'to_from'
971 __ mov(count, rax); // restore 'count'
972 __ jmpb(L_copy_2_bytes); // all dwords were copied
973 } else {
974 if (!UseUnalignedLoadStores) {
975 // align to 8 bytes, we know we are 4 byte aligned to start
976 __ testptr(from, 4);
977 __ jccb(Assembler::zero, L_copy_64_bytes);
978 __ movl(rax, Address(from, 0));
979 __ movl(Address(from, to_from, Address::times_1, 0), rax);
980 __ addptr(from, 4);
981 __ subl(count, 1<<shift);
982 }
983 __ BIND(L_copy_64_bytes);
984 __ mov(rax, count);
985 __ shrl(rax, shift+1); // 8 bytes chunk count
986 //
987 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
988 //
989 if (UseXMMForArrayCopy) {
990 xmm_copy_forward(from, to_from, rax);
991 } else {
992 mmx_copy_forward(from, to_from, rax);
993 }
994 }
995 // copy tailing dword
996 __ BIND(L_copy_4_bytes);
997 __ testl(count, 1<<shift);
998 __ jccb(Assembler::zero, L_copy_2_bytes);
999 __ movl(rax, Address(from, 0));
1000 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1001 if (t == T_BYTE || t == T_SHORT) {
1002 __ addptr(from, 4);
1003 __ BIND(L_copy_2_bytes);
1004 // copy tailing word
1005 __ testl(count, 1<<(shift-1));
1006 __ jccb(Assembler::zero, L_copy_byte);
1007 __ movw(rax, Address(from, 0));
1008 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1009 if (t == T_BYTE) {
1010 __ addptr(from, 2);
1011 __ BIND(L_copy_byte);
1012 // copy tailing byte
1013 __ testl(count, 1);
1014 __ jccb(Assembler::zero, L_exit);
1015 __ movb(rax, Address(from, 0));
1016 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1017 __ BIND(L_exit);
1018 } else {
1019 __ BIND(L_copy_byte);
1020 }
1021 } else {
1022 __ BIND(L_copy_2_bytes);
1023 }
1024
1025 if (t == T_OBJECT) {
1026 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1027 __ mov(to, saved_to); // restore 'to'
1028 gen_write_ref_array_post_barrier(to, count);
1029 __ BIND(L_0_count);
1030 }
1031 inc_copy_counter_np(t);
1032 __ pop(rdi);
1033 __ pop(rsi);
1034 __ leave(); // required for proper stackwalking of RuntimeStub frame
1035 __ xorptr(rax, rax); // return 0
1036 __ ret(0);
1037 return start;
1038 }
1039
1040
1041 address generate_fill(BasicType t, bool aligned, const char *name) {
1042 __ align(CodeEntryAlignment);
1043 StubCodeMark mark(this, "StubRoutines", name);
1044 address start = __ pc();
1045
1046 BLOCK_COMMENT("Entry:");
1047
1048 const Register to = rdi; // source array address
1049 const Register value = rdx; // value
1050 const Register count = rsi; // elements count
1051
1052 __ enter(); // required for proper stackwalking of RuntimeStub frame
1053 __ push(rsi);
1054 __ push(rdi);
1055 __ movptr(to , Address(rsp, 12+ 4));
1056 __ movl(value, Address(rsp, 12+ 8));
1057 __ movl(count, Address(rsp, 12+ 12));
1058
1059 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1060
1061 __ pop(rdi);
1062 __ pop(rsi);
1063 __ leave(); // required for proper stackwalking of RuntimeStub frame
1064 __ ret(0);
1065 return start;
1066 }
1067
1068 address generate_conjoint_copy(BasicType t, bool aligned,
1069 Address::ScaleFactor sf,
1070 address nooverlap_target,
1071 address* entry, const char *name,
1072 bool dest_uninitialized = false) {
1073 __ align(CodeEntryAlignment);
1074 StubCodeMark mark(this, "StubRoutines", name);
1075 address start = __ pc();
1076
1077 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1078 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1079
1080 int shift = Address::times_ptr - sf;
1081
1082 const Register src = rax; // source array address
1083 const Register dst = rdx; // destination array address
1084 const Register from = rsi; // source array address
1085 const Register to = rdi; // destination array address
1086 const Register count = rcx; // elements count
1087 const Register end = rax; // array end address
1088
1089 __ enter(); // required for proper stackwalking of RuntimeStub frame
1090 __ push(rsi);
1091 __ push(rdi);
1092 __ movptr(src , Address(rsp, 12+ 4)); // from
1093 __ movptr(dst , Address(rsp, 12+ 8)); // to
1094 __ movl2ptr(count, Address(rsp, 12+12)); // count
1095
1096 if (entry != NULL) {
1097 *entry = __ pc(); // Entry point from generic arraycopy stub.
1098 BLOCK_COMMENT("Entry:");
1099 }
1100
1101 // nooverlap_target expects arguments in rsi and rdi.
1102 __ mov(from, src);
1103 __ mov(to , dst);
1104
1105 // arrays overlap test: dispatch to disjoint stub if necessary.
1106 RuntimeAddress nooverlap(nooverlap_target);
1107 __ cmpptr(dst, src);
1108 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1109 __ jump_cc(Assembler::belowEqual, nooverlap);
1110 __ cmpptr(dst, end);
1111 __ jump_cc(Assembler::aboveEqual, nooverlap);
1112
1113 if (t == T_OBJECT) {
1114 __ testl(count, count);
1115 __ jcc(Assembler::zero, L_0_count);
1116 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1117 }
1118
1119 // copy from high to low
1120 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1121 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1122 if (t == T_BYTE || t == T_SHORT) {
1123 // Align the end of destination array at 4 bytes address boundary
1124 __ lea(end, Address(dst, count, sf, 0));
1125 if (t == T_BYTE) {
1126 // One byte misalignment happens only for byte arrays
1127 __ testl(end, 1);
1128 __ jccb(Assembler::zero, L_skip_align1);
1129 __ decrement(count);
1130 __ movb(rdx, Address(from, count, sf, 0));
1131 __ movb(Address(to, count, sf, 0), rdx);
1132 __ BIND(L_skip_align1);
1133 }
1134 // Two bytes misalignment happens only for byte and short (char) arrays
1135 __ testl(end, 2);
1136 __ jccb(Assembler::zero, L_skip_align2);
1137 __ subptr(count, 1<<(shift-1));
1138 __ movw(rdx, Address(from, count, sf, 0));
1139 __ movw(Address(to, count, sf, 0), rdx);
1140 __ BIND(L_skip_align2);
1141 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1142 __ jcc(Assembler::below, L_copy_4_bytes);
1143 }
1144
1145 if (!VM_Version::supports_mmx()) {
1146 __ std();
1147 __ mov(rax, count); // Save 'count'
1148 __ mov(rdx, to); // Save 'to'
1149 __ lea(rsi, Address(from, count, sf, -4));
1150 __ lea(rdi, Address(to , count, sf, -4));
1151 __ shrptr(count, shift); // bytes count
1152 __ rep_mov();
1153 __ cld();
1154 __ mov(count, rax); // restore 'count'
1155 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1156 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1157 __ mov(to, rdx); // restore 'to'
1158 __ jmpb(L_copy_2_bytes); // all dword were copied
1159 } else {
1160 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1161 __ testptr(end, 4);
1162 __ jccb(Assembler::zero, L_copy_8_bytes);
1163 __ subl(count, 1<<shift);
1164 __ movl(rdx, Address(from, count, sf, 0));
1165 __ movl(Address(to, count, sf, 0), rdx);
1166 __ jmpb(L_copy_8_bytes);
1167
1168 __ align(OptoLoopAlignment);
1169 // Move 8 bytes
1170 __ BIND(L_copy_8_bytes_loop);
1171 if (UseXMMForArrayCopy) {
1172 __ movq(xmm0, Address(from, count, sf, 0));
1173 __ movq(Address(to, count, sf, 0), xmm0);
1174 } else {
1175 __ movq(mmx0, Address(from, count, sf, 0));
1176 __ movq(Address(to, count, sf, 0), mmx0);
1177 }
1178 __ BIND(L_copy_8_bytes);
1179 __ subl(count, 2<<shift);
1180 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1181 __ addl(count, 2<<shift);
1182 if (!UseXMMForArrayCopy) {
1183 __ emms();
1184 }
1185 }
1186 __ BIND(L_copy_4_bytes);
1187 // copy prefix qword
1188 __ testl(count, 1<<shift);
1189 __ jccb(Assembler::zero, L_copy_2_bytes);
1190 __ movl(rdx, Address(from, count, sf, -4));
1191 __ movl(Address(to, count, sf, -4), rdx);
1192
1193 if (t == T_BYTE || t == T_SHORT) {
1194 __ subl(count, (1<<shift));
1195 __ BIND(L_copy_2_bytes);
1196 // copy prefix dword
1197 __ testl(count, 1<<(shift-1));
1198 __ jccb(Assembler::zero, L_copy_byte);
1199 __ movw(rdx, Address(from, count, sf, -2));
1200 __ movw(Address(to, count, sf, -2), rdx);
1201 if (t == T_BYTE) {
1202 __ subl(count, 1<<(shift-1));
1203 __ BIND(L_copy_byte);
1204 // copy prefix byte
1205 __ testl(count, 1);
1206 __ jccb(Assembler::zero, L_exit);
1207 __ movb(rdx, Address(from, 0));
1208 __ movb(Address(to, 0), rdx);
1209 __ BIND(L_exit);
1210 } else {
1211 __ BIND(L_copy_byte);
1212 }
1213 } else {
1214 __ BIND(L_copy_2_bytes);
1215 }
1216 if (t == T_OBJECT) {
1217 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1218 gen_write_ref_array_post_barrier(to, count);
1219 __ BIND(L_0_count);
1220 }
1221 inc_copy_counter_np(t);
1222 __ pop(rdi);
1223 __ pop(rsi);
1224 __ leave(); // required for proper stackwalking of RuntimeStub frame
1225 __ xorptr(rax, rax); // return 0
1226 __ ret(0);
1227 return start;
1228 }
1229
1230
1231 address generate_disjoint_long_copy(address* entry, const char *name) {
1232 __ align(CodeEntryAlignment);
1233 StubCodeMark mark(this, "StubRoutines", name);
1234 address start = __ pc();
1235
1236 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1237 const Register from = rax; // source array address
1238 const Register to = rdx; // destination array address
1239 const Register count = rcx; // elements count
1240 const Register to_from = rdx; // (to - from)
1241
1242 __ enter(); // required for proper stackwalking of RuntimeStub frame
1243 __ movptr(from , Address(rsp, 8+0)); // from
1244 __ movptr(to , Address(rsp, 8+4)); // to
1245 __ movl2ptr(count, Address(rsp, 8+8)); // count
1246
1247 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1248 BLOCK_COMMENT("Entry:");
1249
1250 __ subptr(to, from); // to --> to_from
1251 if (VM_Version::supports_mmx()) {
1252 if (UseXMMForArrayCopy) {
1253 xmm_copy_forward(from, to_from, count);
1254 } else {
1255 mmx_copy_forward(from, to_from, count);
1256 }
1257 } else {
1258 __ jmpb(L_copy_8_bytes);
1259 __ align(OptoLoopAlignment);
1260 __ BIND(L_copy_8_bytes_loop);
1261 __ fild_d(Address(from, 0));
1262 __ fistp_d(Address(from, to_from, Address::times_1));
1263 __ addptr(from, 8);
1264 __ BIND(L_copy_8_bytes);
1265 __ decrement(count);
1266 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1267 }
1268 inc_copy_counter_np(T_LONG);
1269 __ leave(); // required for proper stackwalking of RuntimeStub frame
1270 __ xorptr(rax, rax); // return 0
1271 __ ret(0);
1272 return start;
1273 }
1274
1275 address generate_conjoint_long_copy(address nooverlap_target,
1276 address* entry, const char *name) {
1277 __ align(CodeEntryAlignment);
1278 StubCodeMark mark(this, "StubRoutines", name);
1279 address start = __ pc();
1280
1281 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1282 const Register from = rax; // source array address
1283 const Register to = rdx; // destination array address
1284 const Register count = rcx; // elements count
1285 const Register end_from = rax; // source array end address
1286
1287 __ enter(); // required for proper stackwalking of RuntimeStub frame
1288 __ movptr(from , Address(rsp, 8+0)); // from
1289 __ movptr(to , Address(rsp, 8+4)); // to
1290 __ movl2ptr(count, Address(rsp, 8+8)); // count
1291
1292 *entry = __ pc(); // Entry point from generic arraycopy stub.
1293 BLOCK_COMMENT("Entry:");
1294
1295 // arrays overlap test
1296 __ cmpptr(to, from);
1297 RuntimeAddress nooverlap(nooverlap_target);
1298 __ jump_cc(Assembler::belowEqual, nooverlap);
1299 __ lea(end_from, Address(from, count, Address::times_8, 0));
1300 __ cmpptr(to, end_from);
1301 __ movptr(from, Address(rsp, 8)); // from
1302 __ jump_cc(Assembler::aboveEqual, nooverlap);
1303
1304 __ jmpb(L_copy_8_bytes);
1305
1306 __ align(OptoLoopAlignment);
1307 __ BIND(L_copy_8_bytes_loop);
1308 if (VM_Version::supports_mmx()) {
1309 if (UseXMMForArrayCopy) {
1310 __ movq(xmm0, Address(from, count, Address::times_8));
1311 __ movq(Address(to, count, Address::times_8), xmm0);
1312 } else {
1313 __ movq(mmx0, Address(from, count, Address::times_8));
1314 __ movq(Address(to, count, Address::times_8), mmx0);
1315 }
1316 } else {
1317 __ fild_d(Address(from, count, Address::times_8));
1318 __ fistp_d(Address(to, count, Address::times_8));
1319 }
1320 __ BIND(L_copy_8_bytes);
1321 __ decrement(count);
1322 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1323
1324 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1325 __ emms();
1326 }
1327 inc_copy_counter_np(T_LONG);
1328 __ leave(); // required for proper stackwalking of RuntimeStub frame
1329 __ xorptr(rax, rax); // return 0
1330 __ ret(0);
1331 return start;
1332 }
1333
1334
1335 // Helper for generating a dynamic type check.
1336 // The sub_klass must be one of {rbx, rdx, rsi}.
1337 // The temp is killed.
1338 void generate_type_check(Register sub_klass,
1339 Address& super_check_offset_addr,
1340 Address& super_klass_addr,
1341 Register temp,
1342 Label* L_success, Label* L_failure) {
1343 BLOCK_COMMENT("type_check:");
1344
1345 Label L_fallthrough;
1346 #define LOCAL_JCC(assembler_con, label_ptr) \
1347 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1348 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1349
1350 // The following is a strange variation of the fast path which requires
1351 // one less register, because needed values are on the argument stack.
1352 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1353 // L_success, L_failure, NULL);
1354 assert_different_registers(sub_klass, temp);
1355
1356 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1357
1358 // if the pointers are equal, we are done (e.g., String[] elements)
1359 __ cmpptr(sub_klass, super_klass_addr);
1360 LOCAL_JCC(Assembler::equal, L_success);
1361
1362 // check the supertype display:
1363 __ movl2ptr(temp, super_check_offset_addr);
1364 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1365 __ movptr(temp, super_check_addr); // load displayed supertype
1366 __ cmpptr(temp, super_klass_addr); // test the super type
1367 LOCAL_JCC(Assembler::equal, L_success);
1368
1369 // if it was a primary super, we can just fail immediately
1370 __ cmpl(super_check_offset_addr, sc_offset);
1371 LOCAL_JCC(Assembler::notEqual, L_failure);
1372
1373 // The repne_scan instruction uses fixed registers, which will get spilled.
1374 // We happen to know this works best when super_klass is in rax.
1375 Register super_klass = temp;
1376 __ movptr(super_klass, super_klass_addr);
1377 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1378 L_success, L_failure);
1379
1380 __ bind(L_fallthrough);
1381
1382 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1383 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1384
1385 #undef LOCAL_JCC
1386 }
1387
1388 //
1389 // Generate checkcasting array copy stub
1390 //
1391 // Input:
1392 // 4(rsp) - source array address
1393 // 8(rsp) - destination array address
1394 // 12(rsp) - element count, can be zero
1395 // 16(rsp) - size_t ckoff (super_check_offset)
1396 // 20(rsp) - oop ckval (super_klass)
1397 //
1398 // Output:
1399 // rax, == 0 - success
1400 // rax, == -1^K - failure, where K is partial transfer count
1401 //
1402 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1403 __ align(CodeEntryAlignment);
1404 StubCodeMark mark(this, "StubRoutines", name);
1405 address start = __ pc();
1406
1407 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1408
1409 // register use:
1410 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1411 // rdi, rsi -- element access (oop, klass)
1412 // rbx, -- temp
1413 const Register from = rax; // source array address
1414 const Register to = rdx; // destination array address
1415 const Register length = rcx; // elements count
1416 const Register elem = rdi; // each oop copied
1417 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1418 const Register temp = rbx; // lone remaining temp
1419
1420 __ enter(); // required for proper stackwalking of RuntimeStub frame
1421
1422 __ push(rsi);
1423 __ push(rdi);
1424 __ push(rbx);
1425
1426 Address from_arg(rsp, 16+ 4); // from
1427 Address to_arg(rsp, 16+ 8); // to
1428 Address length_arg(rsp, 16+12); // elements count
1429 Address ckoff_arg(rsp, 16+16); // super_check_offset
1430 Address ckval_arg(rsp, 16+20); // super_klass
1431
1432 // Load up:
1433 __ movptr(from, from_arg);
1434 __ movptr(to, to_arg);
1435 __ movl2ptr(length, length_arg);
1436
1437 if (entry != NULL) {
1438 *entry = __ pc(); // Entry point from generic arraycopy stub.
1439 BLOCK_COMMENT("Entry:");
1440 }
1441
1442 //---------------------------------------------------------------
1443 // Assembler stub will be used for this call to arraycopy
1444 // if the two arrays are subtypes of Object[] but the
1445 // destination array type is not equal to or a supertype
1446 // of the source type. Each element must be separately
1447 // checked.
1448
1449 // Loop-invariant addresses. They are exclusive end pointers.
1450 Address end_from_addr(from, length, Address::times_ptr, 0);
1451 Address end_to_addr(to, length, Address::times_ptr, 0);
1452
1453 Register end_from = from; // re-use
1454 Register end_to = to; // re-use
1455 Register count = length; // re-use
1456
1457 // Loop-variant addresses. They assume post-incremented count < 0.
1458 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1459 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1460 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1461
1462 // Copy from low to high addresses, indexed from the end of each array.
1463 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1464 __ lea(end_from, end_from_addr);
1465 __ lea(end_to, end_to_addr);
1466 assert(length == count, ""); // else fix next line:
1467 __ negptr(count); // negate and test the length
1468 __ jccb(Assembler::notZero, L_load_element);
1469
1470 // Empty array: Nothing to do.
1471 __ xorptr(rax, rax); // return 0 on (trivial) success
1472 __ jmp(L_done);
1473
1474 // ======== begin loop ========
1475 // (Loop is rotated; its entry is L_load_element.)
1476 // Loop control:
1477 // for (count = -count; count != 0; count++)
1478 // Base pointers src, dst are biased by 8*count,to last element.
1479 __ align(OptoLoopAlignment);
1480
1481 __ BIND(L_store_element);
1482 __ movptr(to_element_addr, elem); // store the oop
1483 __ increment(count); // increment the count toward zero
1484 __ jccb(Assembler::zero, L_do_card_marks);
1485
1486 // ======== loop entry is here ========
1487 __ BIND(L_load_element);
1488 __ movptr(elem, from_element_addr); // load the oop
1489 __ testptr(elem, elem);
1490 __ jccb(Assembler::zero, L_store_element);
1491
1492 // (Could do a trick here: Remember last successful non-null
1493 // element stored and make a quick oop equality check on it.)
1494
1495 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1496 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1497 &L_store_element, NULL);
1498 // (On fall-through, we have failed the element type check.)
1499 // ======== end loop ========
1500
1501 // It was a real error; we must depend on the caller to finish the job.
1502 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1503 // Emit GC store barriers for the oops we have copied (length_arg + count),
1504 // and report their number to the caller.
1505 assert_different_registers(to, count, rax);
1506 Label L_post_barrier;
1507 __ addl(count, length_arg); // transfers = (length - remaining)
1508 __ movl2ptr(rax, count); // save the value
1509 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1510 __ jccb(Assembler::notZero, L_post_barrier);
1511 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1512
1513 // Come here on success only.
1514 __ BIND(L_do_card_marks);
1515 __ xorptr(rax, rax); // return 0 on success
1516 __ movl2ptr(count, length_arg);
1517
1518 __ BIND(L_post_barrier);
1519 __ movptr(to, to_arg); // reload
1520 gen_write_ref_array_post_barrier(to, count);
1521
1522 // Common exit point (success or failure).
1523 __ BIND(L_done);
1524 __ pop(rbx);
1525 __ pop(rdi);
1526 __ pop(rsi);
1527 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1528 __ leave(); // required for proper stackwalking of RuntimeStub frame
1529 __ ret(0);
1530
1531 return start;
1532 }
1533
1534 //
1535 // Generate 'unsafe' array copy stub
1536 // Though just as safe as the other stubs, it takes an unscaled
1537 // size_t argument instead of an element count.
1538 //
1539 // Input:
1540 // 4(rsp) - source array address
1541 // 8(rsp) - destination array address
1542 // 12(rsp) - byte count, can be zero
1543 //
1544 // Output:
1545 // rax, == 0 - success
1546 // rax, == -1 - need to call System.arraycopy
1547 //
1548 // Examines the alignment of the operands and dispatches
1549 // to a long, int, short, or byte copy loop.
1550 //
1551 address generate_unsafe_copy(const char *name,
1552 address byte_copy_entry,
1553 address short_copy_entry,
1554 address int_copy_entry,
1555 address long_copy_entry) {
1556
1557 Label L_long_aligned, L_int_aligned, L_short_aligned;
1558
1559 __ align(CodeEntryAlignment);
1560 StubCodeMark mark(this, "StubRoutines", name);
1561 address start = __ pc();
1562
1563 const Register from = rax; // source array address
1564 const Register to = rdx; // destination array address
1565 const Register count = rcx; // elements count
1566
1567 __ enter(); // required for proper stackwalking of RuntimeStub frame
1568 __ push(rsi);
1569 __ push(rdi);
1570 Address from_arg(rsp, 12+ 4); // from
1571 Address to_arg(rsp, 12+ 8); // to
1572 Address count_arg(rsp, 12+12); // byte count
1573
1574 // Load up:
1575 __ movptr(from , from_arg);
1576 __ movptr(to , to_arg);
1577 __ movl2ptr(count, count_arg);
1578
1579 // bump this on entry, not on exit:
1580 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1581
1582 const Register bits = rsi;
1583 __ mov(bits, from);
1584 __ orptr(bits, to);
1585 __ orptr(bits, count);
1586
1587 __ testl(bits, BytesPerLong-1);
1588 __ jccb(Assembler::zero, L_long_aligned);
1589
1590 __ testl(bits, BytesPerInt-1);
1591 __ jccb(Assembler::zero, L_int_aligned);
1592
1593 __ testl(bits, BytesPerShort-1);
1594 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1595
1596 __ BIND(L_short_aligned);
1597 __ shrptr(count, LogBytesPerShort); // size => short_count
1598 __ movl(count_arg, count); // update 'count'
1599 __ jump(RuntimeAddress(short_copy_entry));
1600
1601 __ BIND(L_int_aligned);
1602 __ shrptr(count, LogBytesPerInt); // size => int_count
1603 __ movl(count_arg, count); // update 'count'
1604 __ jump(RuntimeAddress(int_copy_entry));
1605
1606 __ BIND(L_long_aligned);
1607 __ shrptr(count, LogBytesPerLong); // size => qword_count
1608 __ movl(count_arg, count); // update 'count'
1609 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1610 __ pop(rsi);
1611 __ jump(RuntimeAddress(long_copy_entry));
1612
1613 return start;
1614 }
1615
1616
1617 // Perform range checks on the proposed arraycopy.
1618 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1619 void arraycopy_range_checks(Register src,
1620 Register src_pos,
1621 Register dst,
1622 Register dst_pos,
1623 Address& length,
1624 Label& L_failed) {
1625 BLOCK_COMMENT("arraycopy_range_checks:");
1626 const Register src_end = src_pos; // source array end position
1627 const Register dst_end = dst_pos; // destination array end position
1628 __ addl(src_end, length); // src_pos + length
1629 __ addl(dst_end, length); // dst_pos + length
1630
1631 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1632 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1633 __ jcc(Assembler::above, L_failed);
1634
1635 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1636 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1637 __ jcc(Assembler::above, L_failed);
1638
1639 BLOCK_COMMENT("arraycopy_range_checks done");
1640 }
1641
1642
1643 //
1644 // Generate generic array copy stubs
1645 //
1646 // Input:
1647 // 4(rsp) - src oop
1648 // 8(rsp) - src_pos
1649 // 12(rsp) - dst oop
1650 // 16(rsp) - dst_pos
1651 // 20(rsp) - element count
1652 //
1653 // Output:
1654 // rax, == 0 - success
1655 // rax, == -1^K - failure, where K is partial transfer count
1656 //
1657 address generate_generic_copy(const char *name,
1658 address entry_jbyte_arraycopy,
1659 address entry_jshort_arraycopy,
1660 address entry_jint_arraycopy,
1661 address entry_oop_arraycopy,
1662 address entry_jlong_arraycopy,
1663 address entry_checkcast_arraycopy) {
1664 Label L_failed, L_failed_0, L_objArray;
1665
1666 { int modulus = CodeEntryAlignment;
1667 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1668 int advance = target - (__ offset() % modulus);
1669 if (advance < 0) advance += modulus;
1670 if (advance > 0) __ nop(advance);
1671 }
1672 StubCodeMark mark(this, "StubRoutines", name);
1673
1674 // Short-hop target to L_failed. Makes for denser prologue code.
1675 __ BIND(L_failed_0);
1676 __ jmp(L_failed);
1677 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1678
1679 __ align(CodeEntryAlignment);
1680 address start = __ pc();
1681
1682 __ enter(); // required for proper stackwalking of RuntimeStub frame
1683 __ push(rsi);
1684 __ push(rdi);
1685
1686 // bump this on entry, not on exit:
1687 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1688
1689 // Input values
1690 Address SRC (rsp, 12+ 4);
1691 Address SRC_POS (rsp, 12+ 8);
1692 Address DST (rsp, 12+12);
1693 Address DST_POS (rsp, 12+16);
1694 Address LENGTH (rsp, 12+20);
1695
1696 //-----------------------------------------------------------------------
1697 // Assembler stub will be used for this call to arraycopy
1698 // if the following conditions are met:
1699 //
1700 // (1) src and dst must not be null.
1701 // (2) src_pos must not be negative.
1702 // (3) dst_pos must not be negative.
1703 // (4) length must not be negative.
1704 // (5) src klass and dst klass should be the same and not NULL.
1705 // (6) src and dst should be arrays.
1706 // (7) src_pos + length must not exceed length of src.
1707 // (8) dst_pos + length must not exceed length of dst.
1708 //
1709
1710 const Register src = rax; // source array oop
1711 const Register src_pos = rsi;
1712 const Register dst = rdx; // destination array oop
1713 const Register dst_pos = rdi;
1714 const Register length = rcx; // transfer count
1715
1716 // if (src == NULL) return -1;
1717 __ movptr(src, SRC); // src oop
1718 __ testptr(src, src);
1719 __ jccb(Assembler::zero, L_failed_0);
1720
1721 // if (src_pos < 0) return -1;
1722 __ movl2ptr(src_pos, SRC_POS); // src_pos
1723 __ testl(src_pos, src_pos);
1724 __ jccb(Assembler::negative, L_failed_0);
1725
1726 // if (dst == NULL) return -1;
1727 __ movptr(dst, DST); // dst oop
1728 __ testptr(dst, dst);
1729 __ jccb(Assembler::zero, L_failed_0);
1730
1731 // if (dst_pos < 0) return -1;
1732 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1733 __ testl(dst_pos, dst_pos);
1734 __ jccb(Assembler::negative, L_failed_0);
1735
1736 // if (length < 0) return -1;
1737 __ movl2ptr(length, LENGTH); // length
1738 __ testl(length, length);
1739 __ jccb(Assembler::negative, L_failed_0);
1740
1741 // if (src->klass() == NULL) return -1;
1742 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1743 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1744 const Register rcx_src_klass = rcx; // array klass
1745 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1746
1747 #ifdef ASSERT
1748 // assert(src->klass() != NULL);
1749 BLOCK_COMMENT("assert klasses not null");
1750 { Label L1, L2;
1751 __ testptr(rcx_src_klass, rcx_src_klass);
1752 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1753 __ bind(L1);
1754 __ stop("broken null klass");
1755 __ bind(L2);
1756 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1757 __ jccb(Assembler::equal, L1); // this would be broken also
1758 BLOCK_COMMENT("assert done");
1759 }
1760 #endif //ASSERT
1761
1762 // Load layout helper (32-bits)
1763 //
1764 // |array_tag| | header_size | element_type | |log2_element_size|
1765 // 32 30 24 16 8 2 0
1766 //
1767 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1768 //
1769
1770 int lh_offset = in_bytes(Klass::layout_helper_offset());
1771 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1772
1773 // Handle objArrays completely differently...
1774 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1775 __ cmpl(src_klass_lh_addr, objArray_lh);
1776 __ jcc(Assembler::equal, L_objArray);
1777
1778 // if (src->klass() != dst->klass()) return -1;
1779 __ cmpptr(rcx_src_klass, dst_klass_addr);
1780 __ jccb(Assembler::notEqual, L_failed_0);
1781
1782 const Register rcx_lh = rcx; // layout helper
1783 assert(rcx_lh == rcx_src_klass, "known alias");
1784 __ movl(rcx_lh, src_klass_lh_addr);
1785
1786 // if (!src->is_Array()) return -1;
1787 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1788 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1789
1790 // At this point, it is known to be a typeArray (array_tag 0x3).
1791 #ifdef ASSERT
1792 { Label L;
1793 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1794 __ jcc(Assembler::greaterEqual, L); // signed cmp
1795 __ stop("must be a primitive array");
1796 __ bind(L);
1797 }
1798 #endif
1799
1800 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1801 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1802
1803 // TypeArrayKlass
1804 //
1805 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1806 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1807 //
1808 const Register rsi_offset = rsi; // array offset
1809 const Register src_array = src; // src array offset
1810 const Register dst_array = dst; // dst array offset
1811 const Register rdi_elsize = rdi; // log2 element size
1812
1813 __ mov(rsi_offset, rcx_lh);
1814 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1815 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1816 __ addptr(src_array, rsi_offset); // src array offset
1817 __ addptr(dst_array, rsi_offset); // dst array offset
1818 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1819
1820 // next registers should be set before the jump to corresponding stub
1821 const Register from = src; // source array address
1822 const Register to = dst; // destination array address
1823 const Register count = rcx; // elements count
1824 // some of them should be duplicated on stack
1825 #define FROM Address(rsp, 12+ 4)
1826 #define TO Address(rsp, 12+ 8) // Not used now
1827 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1828
1829 BLOCK_COMMENT("scale indexes to element size");
1830 __ movl2ptr(rsi, SRC_POS); // src_pos
1831 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1832 assert(src_array == from, "");
1833 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1834 __ movl2ptr(rdi, DST_POS); // dst_pos
1835 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1836 assert(dst_array == to, "");
1837 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1838 __ movptr(FROM, from); // src_addr
1839 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1840 __ movl2ptr(count, LENGTH); // elements count
1841
1842 BLOCK_COMMENT("choose copy loop based on element size");
1843 __ cmpl(rdi_elsize, 0);
1844
1845 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1846 __ cmpl(rdi_elsize, LogBytesPerShort);
1847 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1848 __ cmpl(rdi_elsize, LogBytesPerInt);
1849 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1850 #ifdef ASSERT
1851 __ cmpl(rdi_elsize, LogBytesPerLong);
1852 __ jccb(Assembler::notEqual, L_failed);
1853 #endif
1854 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1855 __ pop(rsi);
1856 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1857
1858 __ BIND(L_failed);
1859 __ xorptr(rax, rax);
1860 __ notptr(rax); // return -1
1861 __ pop(rdi);
1862 __ pop(rsi);
1863 __ leave(); // required for proper stackwalking of RuntimeStub frame
1864 __ ret(0);
1865
1866 // ObjArrayKlass
1867 __ BIND(L_objArray);
1868 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1869
1870 Label L_plain_copy, L_checkcast_copy;
1871 // test array classes for subtyping
1872 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1873 __ jccb(Assembler::notEqual, L_checkcast_copy);
1874
1875 // Identically typed arrays can be copied without element-wise checks.
1876 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1877 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1878
1879 __ BIND(L_plain_copy);
1880 __ movl2ptr(count, LENGTH); // elements count
1881 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1882 __ lea(from, Address(src, src_pos, Address::times_ptr,
1883 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1884 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1885 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1886 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1887 __ movptr(FROM, from); // src_addr
1888 __ movptr(TO, to); // dst_addr
1889 __ movl(COUNT, count); // count
1890 __ jump(RuntimeAddress(entry_oop_arraycopy));
1891
1892 __ BIND(L_checkcast_copy);
1893 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1894 {
1895 // Handy offsets:
1896 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1897 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1898
1899 Register rsi_dst_klass = rsi;
1900 Register rdi_temp = rdi;
1901 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1902 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1903 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1904
1905 // Before looking at dst.length, make sure dst is also an objArray.
1906 __ movptr(rsi_dst_klass, dst_klass_addr);
1907 __ cmpl(dst_klass_lh_addr, objArray_lh);
1908 __ jccb(Assembler::notEqual, L_failed);
1909
1910 // It is safe to examine both src.length and dst.length.
1911 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1912 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1913 // (Now src_pos and dst_pos are killed, but not src and dst.)
1914
1915 // We'll need this temp (don't forget to pop it after the type check).
1916 __ push(rbx);
1917 Register rbx_src_klass = rbx;
1918
1919 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1920 __ movptr(rsi_dst_klass, dst_klass_addr);
1921 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1922 Label L_fail_array_check;
1923 generate_type_check(rbx_src_klass,
1924 super_check_offset_addr, dst_klass_addr,
1925 rdi_temp, NULL, &L_fail_array_check);
1926 // (On fall-through, we have passed the array type check.)
1927 __ pop(rbx);
1928 __ jmp(L_plain_copy);
1929
1930 __ BIND(L_fail_array_check);
1931 // Reshuffle arguments so we can call checkcast_arraycopy:
1932
1933 // match initial saves for checkcast_arraycopy
1934 // push(rsi); // already done; see above
1935 // push(rdi); // already done; see above
1936 // push(rbx); // already done; see above
1937
1938 // Marshal outgoing arguments now, freeing registers.
1939 Address from_arg(rsp, 16+ 4); // from
1940 Address to_arg(rsp, 16+ 8); // to
1941 Address length_arg(rsp, 16+12); // elements count
1942 Address ckoff_arg(rsp, 16+16); // super_check_offset
1943 Address ckval_arg(rsp, 16+20); // super_klass
1944
1945 Address SRC_POS_arg(rsp, 16+ 8);
1946 Address DST_POS_arg(rsp, 16+16);
1947 Address LENGTH_arg(rsp, 16+20);
1948 // push rbx, changed the incoming offsets (why not just use rbp,??)
1949 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1950
1951 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1952 __ movl2ptr(length, LENGTH_arg); // reload elements count
1953 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1954 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1955
1956 __ movptr(ckval_arg, rbx); // destination element type
1957 __ movl(rbx, Address(rbx, sco_offset));
1958 __ movl(ckoff_arg, rbx); // corresponding class check offset
1959
1960 __ movl(length_arg, length); // outgoing length argument
1961
1962 __ lea(from, Address(src, src_pos, Address::times_ptr,
1963 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1964 __ movptr(from_arg, from);
1965
1966 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1967 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1968 __ movptr(to_arg, to);
1969 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1970 }
1971
1972 return start;
1973 }
1974
1975 void generate_arraycopy_stubs() {
1976 address entry;
1977 address entry_jbyte_arraycopy;
1978 address entry_jshort_arraycopy;
1979 address entry_jint_arraycopy;
1980 address entry_oop_arraycopy;
1981 address entry_jlong_arraycopy;
1982 address entry_checkcast_arraycopy;
1983
1984 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1985 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1986 "arrayof_jbyte_disjoint_arraycopy");
1987 StubRoutines::_arrayof_jbyte_arraycopy =
1988 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1989 NULL, "arrayof_jbyte_arraycopy");
1990 StubRoutines::_jbyte_disjoint_arraycopy =
1991 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1992 "jbyte_disjoint_arraycopy");
1993 StubRoutines::_jbyte_arraycopy =
1994 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1995 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1996
1997 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1998 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1999 "arrayof_jshort_disjoint_arraycopy");
2000 StubRoutines::_arrayof_jshort_arraycopy =
2001 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2002 NULL, "arrayof_jshort_arraycopy");
2003 StubRoutines::_jshort_disjoint_arraycopy =
2004 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2005 "jshort_disjoint_arraycopy");
2006 StubRoutines::_jshort_arraycopy =
2007 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2008 &entry_jshort_arraycopy, "jshort_arraycopy");
2009
2010 // Next arrays are always aligned on 4 bytes at least.
2011 StubRoutines::_jint_disjoint_arraycopy =
2012 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2013 "jint_disjoint_arraycopy");
2014 StubRoutines::_jint_arraycopy =
2015 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2016 &entry_jint_arraycopy, "jint_arraycopy");
2017
2018 StubRoutines::_oop_disjoint_arraycopy =
2019 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2020 "oop_disjoint_arraycopy");
2021 StubRoutines::_oop_arraycopy =
2022 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2023 &entry_oop_arraycopy, "oop_arraycopy");
2024
2025 StubRoutines::_oop_disjoint_arraycopy_uninit =
2026 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2027 "oop_disjoint_arraycopy_uninit",
2028 /*dest_uninitialized*/true);
2029 StubRoutines::_oop_arraycopy_uninit =
2030 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2031 NULL, "oop_arraycopy_uninit",
2032 /*dest_uninitialized*/true);
2033
2034 StubRoutines::_jlong_disjoint_arraycopy =
2035 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2036 StubRoutines::_jlong_arraycopy =
2037 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2038 "jlong_arraycopy");
2039
2040 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2041 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2042 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2043 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2044 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2045 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2046
2047 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2048 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2049 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2050 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2051
2052 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2053 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2054 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2055 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2056
2057 StubRoutines::_checkcast_arraycopy =
2058 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2059 StubRoutines::_checkcast_arraycopy_uninit =
2060 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2061
2062 StubRoutines::_unsafe_arraycopy =
2063 generate_unsafe_copy("unsafe_arraycopy",
2064 entry_jbyte_arraycopy,
2065 entry_jshort_arraycopy,
2066 entry_jint_arraycopy,
2067 entry_jlong_arraycopy);
2068
2069 StubRoutines::_generic_arraycopy =
2070 generate_generic_copy("generic_arraycopy",
2071 entry_jbyte_arraycopy,
2072 entry_jshort_arraycopy,
2073 entry_jint_arraycopy,
2074 entry_oop_arraycopy,
2075 entry_jlong_arraycopy,
2076 entry_checkcast_arraycopy);
2077 }
2078
2079 void generate_math_stubs() {
2080 {
2081 StubCodeMark mark(this, "StubRoutines", "log");
2082 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2083
2084 __ fld_d(Address(rsp, 4));
2085 __ flog();
2086 __ ret(0);
2087 }
2088 {
2089 StubCodeMark mark(this, "StubRoutines", "log10");
2090 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2091
2092 __ fld_d(Address(rsp, 4));
2093 __ flog10();
2094 __ ret(0);
2095 }
2096 {
2097 StubCodeMark mark(this, "StubRoutines", "sin");
2098 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2099
2100 __ fld_d(Address(rsp, 4));
2101 __ trigfunc('s');
2102 __ ret(0);
2103 }
2104 {
2105 StubCodeMark mark(this, "StubRoutines", "cos");
2106 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2107
2108 __ fld_d(Address(rsp, 4));
2109 __ trigfunc('c');
2110 __ ret(0);
2111 }
2112 {
2113 StubCodeMark mark(this, "StubRoutines", "tan");
2114 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2115
2116 __ fld_d(Address(rsp, 4));
2117 __ trigfunc('t');
2118 __ ret(0);
2119 }
2120 {
2121 StubCodeMark mark(this, "StubRoutines", "exp");
2122 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2123
2124 __ fld_d(Address(rsp, 4));
2125 __ exp_with_fallback(0);
2126 __ ret(0);
2127 }
2128 {
2129 StubCodeMark mark(this, "StubRoutines", "pow");
2130 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2131
2132 __ fld_d(Address(rsp, 12));
2133 __ fld_d(Address(rsp, 4));
2134 __ pow_with_fallback(0);
2135 __ ret(0);
2136 }
2137 }
2138
2139 // AES intrinsic stubs
2140 enum {AESBlockSize = 16};
2141
2142 address generate_key_shuffle_mask() {
2143 __ align(16);
2144 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2145 address start = __ pc();
2146 __ emit_data(0x00010203, relocInfo::none, 0 );
2147 __ emit_data(0x04050607, relocInfo::none, 0 );
2148 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2149 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2150 return start;
2151 }
2152
2153 // Utility routine for loading a 128-bit key word in little endian format
2154 // can optionally specify that the shuffle mask is already in an xmmregister
2155 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2156 __ movdqu(xmmdst, Address(key, offset));
2157 if (xmm_shuf_mask != NULL) {
2158 __ pshufb(xmmdst, xmm_shuf_mask);
2159 } else {
2160 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2161 }
2162 }
2163
2164 // aesenc using specified key+offset
2165 // can optionally specify that the shuffle mask is already in an xmmregister
2166 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2167 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2168 __ aesenc(xmmdst, xmmtmp);
2169 }
2170
2171 // aesdec using specified key+offset
2172 // can optionally specify that the shuffle mask is already in an xmmregister
2173 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2174 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2175 __ aesdec(xmmdst, xmmtmp);
2176 }
2177
2178
2179 // Arguments:
2180 //
2181 // Inputs:
2182 // c_rarg0 - source byte array address
2183 // c_rarg1 - destination byte array address
2184 // c_rarg2 - K (key) in little endian int array
2185 //
2186 address generate_aescrypt_encryptBlock() {
2187 assert(UseAES, "need AES instructions and misaligned SSE support");
2188 __ align(CodeEntryAlignment);
2189 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2190 Label L_doLast;
2191 address start = __ pc();
2192
2193 const Register from = rdx; // source array address
2194 const Register to = rdx; // destination array address
2195 const Register key = rcx; // key array address
2196 const Register keylen = rax;
2197 const Address from_param(rbp, 8+0);
2198 const Address to_param (rbp, 8+4);
2199 const Address key_param (rbp, 8+8);
2200
2201 const XMMRegister xmm_result = xmm0;
2202 const XMMRegister xmm_key_shuf_mask = xmm1;
2203 const XMMRegister xmm_temp1 = xmm2;
2204 const XMMRegister xmm_temp2 = xmm3;
2205 const XMMRegister xmm_temp3 = xmm4;
2206 const XMMRegister xmm_temp4 = xmm5;
2207
2208 __ enter(); // required for proper stackwalking of RuntimeStub frame
2209 __ movptr(from, from_param);
2210 __ movptr(key, key_param);
2211
2212 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2213 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2214
2215 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2216 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2217 __ movptr(to, to_param);
2218
2219 // For encryption, the java expanded key ordering is just what we need
2220
2221 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2222 __ pxor(xmm_result, xmm_temp1);
2223
2224 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2225 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2226 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2227 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2228
2229 __ aesenc(xmm_result, xmm_temp1);
2230 __ aesenc(xmm_result, xmm_temp2);
2231 __ aesenc(xmm_result, xmm_temp3);
2232 __ aesenc(xmm_result, xmm_temp4);
2233
2234 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2235 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2236 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2237 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2238
2239 __ aesenc(xmm_result, xmm_temp1);
2240 __ aesenc(xmm_result, xmm_temp2);
2241 __ aesenc(xmm_result, xmm_temp3);
2242 __ aesenc(xmm_result, xmm_temp4);
2243
2244 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2245 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2246
2247 __ cmpl(keylen, 44);
2248 __ jccb(Assembler::equal, L_doLast);
2249
2250 __ aesenc(xmm_result, xmm_temp1);
2251 __ aesenc(xmm_result, xmm_temp2);
2252
2253 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2254 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2255
2256 __ cmpl(keylen, 52);
2257 __ jccb(Assembler::equal, L_doLast);
2258
2259 __ aesenc(xmm_result, xmm_temp1);
2260 __ aesenc(xmm_result, xmm_temp2);
2261
2262 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2263 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2264
2265 __ BIND(L_doLast);
2266 __ aesenc(xmm_result, xmm_temp1);
2267 __ aesenclast(xmm_result, xmm_temp2);
2268 __ movdqu(Address(to, 0), xmm_result); // store the result
2269 __ xorptr(rax, rax); // return 0
2270 __ leave(); // required for proper stackwalking of RuntimeStub frame
2271 __ ret(0);
2272
2273 return start;
2274 }
2275
2276
2277 // Arguments:
2278 //
2279 // Inputs:
2280 // c_rarg0 - source byte array address
2281 // c_rarg1 - destination byte array address
2282 // c_rarg2 - K (key) in little endian int array
2283 //
2284 address generate_aescrypt_decryptBlock() {
2285 assert(UseAES, "need AES instructions and misaligned SSE support");
2286 __ align(CodeEntryAlignment);
2287 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2288 Label L_doLast;
2289 address start = __ pc();
2290
2291 const Register from = rdx; // source array address
2292 const Register to = rdx; // destination array address
2293 const Register key = rcx; // key array address
2294 const Register keylen = rax;
2295 const Address from_param(rbp, 8+0);
2296 const Address to_param (rbp, 8+4);
2297 const Address key_param (rbp, 8+8);
2298
2299 const XMMRegister xmm_result = xmm0;
2300 const XMMRegister xmm_key_shuf_mask = xmm1;
2301 const XMMRegister xmm_temp1 = xmm2;
2302 const XMMRegister xmm_temp2 = xmm3;
2303 const XMMRegister xmm_temp3 = xmm4;
2304 const XMMRegister xmm_temp4 = xmm5;
2305
2306 __ enter(); // required for proper stackwalking of RuntimeStub frame
2307 __ movptr(from, from_param);
2308 __ movptr(key, key_param);
2309
2310 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2311 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2312
2313 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2314 __ movdqu(xmm_result, Address(from, 0));
2315 __ movptr(to, to_param);
2316
2317 // for decryption java expanded key ordering is rotated one position from what we want
2318 // so we start from 0x10 here and hit 0x00 last
2319 // we don't know if the key is aligned, hence not using load-execute form
2320 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2321 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2322 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2323 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2324
2325 __ pxor (xmm_result, xmm_temp1);
2326 __ aesdec(xmm_result, xmm_temp2);
2327 __ aesdec(xmm_result, xmm_temp3);
2328 __ aesdec(xmm_result, xmm_temp4);
2329
2330 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2331 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2332 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2333 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2334
2335 __ aesdec(xmm_result, xmm_temp1);
2336 __ aesdec(xmm_result, xmm_temp2);
2337 __ aesdec(xmm_result, xmm_temp3);
2338 __ aesdec(xmm_result, xmm_temp4);
2339
2340 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2341 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2342 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2343
2344 __ cmpl(keylen, 44);
2345 __ jccb(Assembler::equal, L_doLast);
2346
2347 __ aesdec(xmm_result, xmm_temp1);
2348 __ aesdec(xmm_result, xmm_temp2);
2349
2350 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2351 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2352
2353 __ cmpl(keylen, 52);
2354 __ jccb(Assembler::equal, L_doLast);
2355
2356 __ aesdec(xmm_result, xmm_temp1);
2357 __ aesdec(xmm_result, xmm_temp2);
2358
2359 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2360 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2361
2362 __ BIND(L_doLast);
2363 __ aesdec(xmm_result, xmm_temp1);
2364 __ aesdec(xmm_result, xmm_temp2);
2365
2366 // for decryption the aesdeclast operation is always on key+0x00
2367 __ aesdeclast(xmm_result, xmm_temp3);
2368 __ movdqu(Address(to, 0), xmm_result); // store the result
2369 __ xorptr(rax, rax); // return 0
2370 __ leave(); // required for proper stackwalking of RuntimeStub frame
2371 __ ret(0);
2372
2373 return start;
2374 }
2375
2376 void handleSOERegisters(bool saving) {
2377 const int saveFrameSizeInBytes = 4 * wordSize;
2378 const Address saved_rbx (rbp, -3 * wordSize);
2379 const Address saved_rsi (rbp, -2 * wordSize);
2380 const Address saved_rdi (rbp, -1 * wordSize);
2381
2382 if (saving) {
2383 __ subptr(rsp, saveFrameSizeInBytes);
2384 __ movptr(saved_rsi, rsi);
2385 __ movptr(saved_rdi, rdi);
2386 __ movptr(saved_rbx, rbx);
2387 } else {
2388 // restoring
2389 __ movptr(rsi, saved_rsi);
2390 __ movptr(rdi, saved_rdi);
2391 __ movptr(rbx, saved_rbx);
2392 }
2393 }
2394
2395 // Arguments:
2396 //
2397 // Inputs:
2398 // c_rarg0 - source byte array address
2399 // c_rarg1 - destination byte array address
2400 // c_rarg2 - K (key) in little endian int array
2401 // c_rarg3 - r vector byte array address
2402 // c_rarg4 - input length
2403 //
2404 // Output:
2405 // rax - input length
2406 //
2407 address generate_cipherBlockChaining_encryptAESCrypt() {
2408 assert(UseAES, "need AES instructions and misaligned SSE support");
2409 __ align(CodeEntryAlignment);
2410 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2411 address start = __ pc();
2412
2413 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2414 const Register from = rsi; // source array address
2415 const Register to = rdx; // destination array address
2416 const Register key = rcx; // key array address
2417 const Register rvec = rdi; // r byte array initialized from initvector array address
2418 // and left with the results of the last encryption block
2419 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2420 const Register pos = rax;
2421
2422 // xmm register assignments for the loops below
2423 const XMMRegister xmm_result = xmm0;
2424 const XMMRegister xmm_temp = xmm1;
2425 // first 6 keys preloaded into xmm2-xmm7
2426 const int XMM_REG_NUM_KEY_FIRST = 2;
2427 const int XMM_REG_NUM_KEY_LAST = 7;
2428 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2429
2430 __ enter(); // required for proper stackwalking of RuntimeStub frame
2431 handleSOERegisters(true /*saving*/);
2432
2433 // load registers from incoming parameters
2434 const Address from_param(rbp, 8+0);
2435 const Address to_param (rbp, 8+4);
2436 const Address key_param (rbp, 8+8);
2437 const Address rvec_param (rbp, 8+12);
2438 const Address len_param (rbp, 8+16);
2439 __ movptr(from , from_param);
2440 __ movptr(to , to_param);
2441 __ movptr(key , key_param);
2442 __ movptr(rvec , rvec_param);
2443 __ movptr(len_reg , len_param);
2444
2445 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2446 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2447 // load up xmm regs 2 thru 7 with keys 0-5
2448 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2449 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2450 offset += 0x10;
2451 }
2452
2453 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2454
2455 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2456 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2457 __ cmpl(rax, 44);
2458 __ jcc(Assembler::notEqual, L_key_192_256);
2459
2460 // 128 bit code follows here
2461 __ movl(pos, 0);
2462 __ align(OptoLoopAlignment);
2463 __ BIND(L_loopTop_128);
2464 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2465 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2466
2467 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2468 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2469 __ aesenc(xmm_result, as_XMMRegister(rnum));
2470 }
2471 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2472 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2473 }
2474 load_key(xmm_temp, key, 0xa0);
2475 __ aesenclast(xmm_result, xmm_temp);
2476
2477 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2478 // no need to store r to memory until we exit
2479 __ addptr(pos, AESBlockSize);
2480 __ subptr(len_reg, AESBlockSize);
2481 __ jcc(Assembler::notEqual, L_loopTop_128);
2482
2483 __ BIND(L_exit);
2484 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2485
2486 handleSOERegisters(false /*restoring*/);
2487 __ movptr(rax, len_param); // return length
2488 __ leave(); // required for proper stackwalking of RuntimeStub frame
2489 __ ret(0);
2490
2491 __ BIND(L_key_192_256);
2492 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2493 __ cmpl(rax, 52);
2494 __ jcc(Assembler::notEqual, L_key_256);
2495
2496 // 192-bit code follows here (could be changed to use more xmm registers)
2497 __ movl(pos, 0);
2498 __ align(OptoLoopAlignment);
2499 __ BIND(L_loopTop_192);
2500 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2501 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2502
2503 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2504 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2505 __ aesenc(xmm_result, as_XMMRegister(rnum));
2506 }
2507 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2508 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2509 }
2510 load_key(xmm_temp, key, 0xc0);
2511 __ aesenclast(xmm_result, xmm_temp);
2512
2513 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2514 // no need to store r to memory until we exit
2515 __ addptr(pos, AESBlockSize);
2516 __ subptr(len_reg, AESBlockSize);
2517 __ jcc(Assembler::notEqual, L_loopTop_192);
2518 __ jmp(L_exit);
2519
2520 __ BIND(L_key_256);
2521 // 256-bit code follows here (could be changed to use more xmm registers)
2522 __ movl(pos, 0);
2523 __ align(OptoLoopAlignment);
2524 __ BIND(L_loopTop_256);
2525 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2526 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2527
2528 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2529 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2530 __ aesenc(xmm_result, as_XMMRegister(rnum));
2531 }
2532 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2533 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2534 }
2535 load_key(xmm_temp, key, 0xe0);
2536 __ aesenclast(xmm_result, xmm_temp);
2537
2538 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2539 // no need to store r to memory until we exit
2540 __ addptr(pos, AESBlockSize);
2541 __ subptr(len_reg, AESBlockSize);
2542 __ jcc(Assembler::notEqual, L_loopTop_256);
2543 __ jmp(L_exit);
2544
2545 return start;
2546 }
2547
2548
2549 // CBC AES Decryption.
2550 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2551 //
2552 // Arguments:
2553 //
2554 // Inputs:
2555 // c_rarg0 - source byte array address
2556 // c_rarg1 - destination byte array address
2557 // c_rarg2 - K (key) in little endian int array
2558 // c_rarg3 - r vector byte array address
2559 // c_rarg4 - input length
2560 //
2561 // Output:
2562 // rax - input length
2563 //
2564
2565 address generate_cipherBlockChaining_decryptAESCrypt() {
2566 assert(UseAES, "need AES instructions and misaligned SSE support");
2567 __ align(CodeEntryAlignment);
2568 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2569 address start = __ pc();
2570
2571 Label L_exit, L_key_192_256, L_key_256;
2572 Label L_singleBlock_loopTop_128;
2573 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2574 const Register from = rsi; // source array address
2575 const Register to = rdx; // destination array address
2576 const Register key = rcx; // key array address
2577 const Register rvec = rdi; // r byte array initialized from initvector array address
2578 // and left with the results of the last encryption block
2579 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2580 const Register pos = rax;
2581
2582 // xmm register assignments for the loops below
2583 const XMMRegister xmm_result = xmm0;
2584 const XMMRegister xmm_temp = xmm1;
2585 // first 6 keys preloaded into xmm2-xmm7
2586 const int XMM_REG_NUM_KEY_FIRST = 2;
2587 const int XMM_REG_NUM_KEY_LAST = 7;
2588 const int FIRST_NON_REG_KEY_offset = 0x70;
2589 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2590
2591 __ enter(); // required for proper stackwalking of RuntimeStub frame
2592 handleSOERegisters(true /*saving*/);
2593
2594 // load registers from incoming parameters
2595 const Address from_param(rbp, 8+0);
2596 const Address to_param (rbp, 8+4);
2597 const Address key_param (rbp, 8+8);
2598 const Address rvec_param (rbp, 8+12);
2599 const Address len_param (rbp, 8+16);
2600 __ movptr(from , from_param);
2601 __ movptr(to , to_param);
2602 __ movptr(key , key_param);
2603 __ movptr(rvec , rvec_param);
2604 __ movptr(len_reg , len_param);
2605
2606 // the java expanded key ordering is rotated one position from what we want
2607 // so we start from 0x10 here and hit 0x00 last
2608 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
2609 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2610 // load up xmm regs 2 thru 6 with first 5 keys
2611 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2612 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2613 offset += 0x10;
2614 }
2615
2616 // inside here, use the rvec register to point to previous block cipher
2617 // with which we xor at the end of each newly decrypted block
2618 const Register prev_block_cipher_ptr = rvec;
2619
2620 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2621 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2622 __ cmpl(rax, 44);
2623 __ jcc(Assembler::notEqual, L_key_192_256);
2624
2625
2626 // 128-bit code follows here, parallelized
2627 __ movl(pos, 0);
2628 __ align(OptoLoopAlignment);
2629 __ BIND(L_singleBlock_loopTop_128);
2630 __ cmpptr(len_reg, 0); // any blocks left??
2631 __ jcc(Assembler::equal, L_exit);
2632 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2633 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2634 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2635 __ aesdec(xmm_result, as_XMMRegister(rnum));
2636 }
2637 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
2638 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2639 }
2640 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2641 __ aesdeclast(xmm_result, xmm_temp);
2642 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2643 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2644 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2645 // no need to store r to memory until we exit
2646 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2647 __ addptr(pos, AESBlockSize);
2648 __ subptr(len_reg, AESBlockSize);
2649 __ jmp(L_singleBlock_loopTop_128);
2650
2651
2652 __ BIND(L_exit);
2653 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2654 __ movptr(rvec , rvec_param); // restore this since used in loop
2655 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
2656 handleSOERegisters(false /*restoring*/);
2657 __ movptr(rax, len_param); // return length
2658 __ leave(); // required for proper stackwalking of RuntimeStub frame
2659 __ ret(0);
2660
2661
2662 __ BIND(L_key_192_256);
2663 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2664 __ cmpl(rax, 52);
2665 __ jcc(Assembler::notEqual, L_key_256);
2666
2667 // 192-bit code follows here (could be optimized to use parallelism)
2668 __ movl(pos, 0);
2669 __ align(OptoLoopAlignment);
2670 __ BIND(L_singleBlock_loopTop_192);
2671 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2672 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2673 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2674 __ aesdec(xmm_result, as_XMMRegister(rnum));
2675 }
2676 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
2677 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2678 }
2679 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2680 __ aesdeclast(xmm_result, xmm_temp);
2681 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2682 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2683 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2684 // no need to store r to memory until we exit
2685 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2686 __ addptr(pos, AESBlockSize);
2687 __ subptr(len_reg, AESBlockSize);
2688 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2689 __ jmp(L_exit);
2690
2691 __ BIND(L_key_256);
2692 // 256-bit code follows here (could be optimized to use parallelism)
2693 __ movl(pos, 0);
2694 __ align(OptoLoopAlignment);
2695 __ BIND(L_singleBlock_loopTop_256);
2696 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2697 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2698 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2699 __ aesdec(xmm_result, as_XMMRegister(rnum));
2700 }
2701 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
2702 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2703 }
2704 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2705 __ aesdeclast(xmm_result, xmm_temp);
2706 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2707 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2708 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2709 // no need to store r to memory until we exit
2710 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2711 __ addptr(pos, AESBlockSize);
2712 __ subptr(len_reg, AESBlockSize);
2713 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2714 __ jmp(L_exit);
2715
2716 return start;
2717 }
2718
2719 /**
2720 * Arguments:
2721 *
2722 * Inputs:
2723 * rsp(4) - int crc
2724 * rsp(8) - byte* buf
2725 * rsp(12) - int length
2726 *
2727 * Ouput:
2728 * rax - int crc result
2729 */
2730 address generate_updateBytesCRC32() {
2731 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2732
2733 __ align(CodeEntryAlignment);
2734 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2735
2736 address start = __ pc();
2737
2738 const Register crc = rdx; // crc
2739 const Register buf = rsi; // source java byte array address
2740 const Register len = rcx; // length
2741 const Register table = rdi; // crc_table address (reuse register)
2742 const Register tmp = rbx;
2743 assert_different_registers(crc, buf, len, table, tmp, rax);
2744
2745 BLOCK_COMMENT("Entry:");
2746 __ enter(); // required for proper stackwalking of RuntimeStub frame
2747 __ push(rsi);
2748 __ push(rdi);
2749 __ push(rbx);
2750
2751 Address crc_arg(rbp, 8 + 0);
2752 Address buf_arg(rbp, 8 + 4);
2753 Address len_arg(rbp, 8 + 8);
2754
2755 // Load up:
2756 __ movl(crc, crc_arg);
2757 __ movptr(buf, buf_arg);
2758 __ movl(len, len_arg);
2759
2760 __ kernel_crc32(crc, buf, len, table, tmp);
2761
2762 __ movl(rax, crc);
2763 __ pop(rbx);
2764 __ pop(rdi);
2765 __ pop(rsi);
2766 __ leave(); // required for proper stackwalking of RuntimeStub frame
2767 __ ret(0);
2768
2769 return start;
2770 }
2771
2772 // Safefetch stubs.
2773 void generate_safefetch(const char* name, int size, address* entry,
2774 address* fault_pc, address* continuation_pc) {
2775 // safefetch signatures:
2776 // int SafeFetch32(int* adr, int errValue);
2777 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2778
2779 StubCodeMark mark(this, "StubRoutines", name);
2780
2781 // Entry point, pc or function descriptor.
2782 *entry = __ pc();
2783
2784 __ movl(rax, Address(rsp, 0x8));
2785 __ movl(rcx, Address(rsp, 0x4));
2786 // Load *adr into eax, may fault.
2787 *fault_pc = __ pc();
2788 switch (size) {
2789 case 4:
2790 // int32_t
2791 __ movl(rax, Address(rcx, 0));
2792 break;
2793 case 8:
2794 // int64_t
2795 Unimplemented();
2796 break;
2797 default:
2798 ShouldNotReachHere();
2799 }
2800
2801 // Return errValue or *adr.
2802 *continuation_pc = __ pc();
2803 __ ret(0);
2804 }
2805
2806 public:
2807 // Information about frame layout at time of blocking runtime call.
2808 // Note that we only have to preserve callee-saved registers since
2809 // the compilers are responsible for supplying a continuation point
2810 // if they expect all registers to be preserved.
2811 enum layout {
2812 thread_off, // last_java_sp
2813 arg1_off,
2814 arg2_off,
2815 rbp_off, // callee saved register
2816 ret_pc,
2817 framesize
2818 };
2819
2820 private:
2821
2822 #undef __
2823 #define __ masm->
2824
2825 //------------------------------------------------------------------------------------------------------------------------
2826 // Continuation point for throwing of implicit exceptions that are not handled in
2827 // the current activation. Fabricates an exception oop and initiates normal
2828 // exception dispatching in this frame.
2829 //
2830 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2831 // This is no longer true after adapter frames were removed but could possibly
2832 // be brought back in the future if the interpreter code was reworked and it
2833 // was deemed worthwhile. The comment below was left to describe what must
2834 // happen here if callee saves were resurrected. As it stands now this stub
2835 // could actually be a vanilla BufferBlob and have now oopMap at all.
2836 // Since it doesn't make much difference we've chosen to leave it the
2837 // way it was in the callee save days and keep the comment.
2838
2839 // If we need to preserve callee-saved values we need a callee-saved oop map and
2840 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2841 // If the compiler needs all registers to be preserved between the fault
2842 // point and the exception handler then it must assume responsibility for that in
2843 // AbstractCompiler::continuation_for_implicit_null_exception or
2844 // continuation_for_implicit_division_by_zero_exception. All other implicit
2845 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2846 // either at call sites or otherwise assume that stack unwinding will be initiated,
2847 // so caller saved registers were assumed volatile in the compiler.
2848 address generate_throw_exception(const char* name, address runtime_entry,
2849 Register arg1 = noreg, Register arg2 = noreg) {
2850
2851 int insts_size = 256;
2852 int locs_size = 32;
2853
2854 CodeBuffer code(name, insts_size, locs_size);
2855 OopMapSet* oop_maps = new OopMapSet();
2856 MacroAssembler* masm = new MacroAssembler(&code);
2857
2858 address start = __ pc();
2859
2860 // This is an inlined and slightly modified version of call_VM
2861 // which has the ability to fetch the return PC out of
2862 // thread-local storage and also sets up last_Java_sp slightly
2863 // differently than the real call_VM
2864 Register java_thread = rbx;
2865 __ get_thread(java_thread);
2866
2867 __ enter(); // required for proper stackwalking of RuntimeStub frame
2868
2869 // pc and rbp, already pushed
2870 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2871
2872 // Frame is now completed as far as size and linkage.
2873
2874 int frame_complete = __ pc() - start;
2875
2876 // push java thread (becomes first argument of C function)
2877 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2878 if (arg1 != noreg) {
2879 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
2880 }
2881 if (arg2 != noreg) {
2882 assert(arg1 != noreg, "missing reg arg");
2883 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
2884 }
2885
2886 // Set up last_Java_sp and last_Java_fp
2887 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2888
2889 // Call runtime
2890 BLOCK_COMMENT("call runtime_entry");
2891 __ call(RuntimeAddress(runtime_entry));
2892 // Generate oop map
2893 OopMap* map = new OopMap(framesize, 0);
2894 oop_maps->add_gc_map(__ pc() - start, map);
2895
2896 // restore the thread (cannot use the pushed argument since arguments
2897 // may be overwritten by C code generated by an optimizing compiler);
2898 // however can use the register value directly if it is callee saved.
2899 __ get_thread(java_thread);
2900
2901 __ reset_last_Java_frame(java_thread, true, false);
2902
2903 __ leave(); // required for proper stackwalking of RuntimeStub frame
2904
2905 // check for pending exceptions
2906 #ifdef ASSERT
2907 Label L;
2908 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2909 __ jcc(Assembler::notEqual, L);
2910 __ should_not_reach_here();
2911 __ bind(L);
2912 #endif /* ASSERT */
2913 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2914
2915
2916 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2917 return stub->entry_point();
2918 }
2919
2920
2921 void create_control_words() {
2922 // Round to nearest, 53-bit mode, exceptions masked
2923 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2924 // Round to zero, 53-bit mode, exception mased
2925 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2926 // Round to nearest, 24-bit mode, exceptions masked
2927 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2928 // Round to nearest, 64-bit mode, exceptions masked
2929 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2930 // Round to nearest, 64-bit mode, exceptions masked
2931 StubRoutines::_mxcsr_std = 0x1F80;
2932 // Note: the following two constants are 80-bit values
2933 // layout is critical for correct loading by FPU.
2934 // Bias for strict fp multiply/divide
2935 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2936 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2937 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2938 // Un-Bias for strict fp multiply/divide
2939 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2940 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2941 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2942 }
2943
2944 //---------------------------------------------------------------------------
2945 // Initialization
2946
2947 void generate_initial() {
2948 // Generates all stubs and initializes the entry points
2949
2950 //------------------------------------------------------------------------------------------------------------------------
2951 // entry points that exist in all platforms
2952 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2953 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2954 StubRoutines::_forward_exception_entry = generate_forward_exception();
2955
2956 StubRoutines::_call_stub_entry =
2957 generate_call_stub(StubRoutines::_call_stub_return_address);
2958 // is referenced by megamorphic call
2959 StubRoutines::_catch_exception_entry = generate_catch_exception();
2960
2961 // These are currently used by Solaris/Intel
2962 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2963
2964 StubRoutines::_handler_for_unsafe_access_entry =
2965 generate_handler_for_unsafe_access();
2966
2967 // platform dependent
2968 create_control_words();
2969
2970 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2971 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2972 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2973 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2974 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2975 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2976
2977 // Build this early so it's available for the interpreter
2978 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
2979
2980 if (UseCRC32Intrinsics) {
2981 // set table address before stub generation which use it
2982 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
2983 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
2984 }
2985 }
2986
2987
2988 void generate_all() {
2989 // Generates all stubs and initializes the entry points
2990
2991 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2992 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2993 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
2994 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
2995 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
2996
2997 //------------------------------------------------------------------------------------------------------------------------
2998 // entry points that are platform specific
2999
3000 // support for verify_oop (must happen after universe_init)
3001 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3002
3003 // arraycopy stubs used by compilers
3004 generate_arraycopy_stubs();
3005
3006 generate_math_stubs();
3007
3008 // don't bother generating these AES intrinsic stubs unless global flag is set
3009 if (UseAESIntrinsics) {
3010 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3011
3012 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3013 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3014 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3015 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
3016 }
3017
3018 // Safefetch stubs.
3019 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3020 &StubRoutines::_safefetch32_fault_pc,
3021 &StubRoutines::_safefetch32_continuation_pc);
3022 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3023 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3024 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3025 }
3026
3027
3028 public:
3029 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3030 if (all) {
3031 generate_all();
3032 } else {
3033 generate_initial();
3034 }
3035 }
3036 }; // end class declaration
3037
3038
3039 void StubGenerator_generate(CodeBuffer* code, bool all) {
3040 StubGenerator g(code, all);
3041 }