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
170 // provide initial value for required masks
171 if (UseAVX > 2) {
172 __ movl(rbx, 0xffff);
173 __ kmovdl(k1, rbx);
174 }
175
176 // save and initialize %mxcsr
177 if (sse_save) {
178 Label skip_ldmx;
179 __ stmxcsr(mxcsr_save);
180 __ movl(rax, mxcsr_save);
181 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
182 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
183 __ cmp32(rax, mxcsr_std);
184 __ jcc(Assembler::equal, skip_ldmx);
185 __ ldmxcsr(mxcsr_std);
186 __ bind(skip_ldmx);
187 }
188
189 // make sure the control word is correct.
190 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
191
192 #ifdef ASSERT
193 // make sure we have no pending exceptions
194 { Label L;
195 __ movptr(rcx, thread);
196 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
197 __ jcc(Assembler::equal, L);
198 __ stop("StubRoutines::call_stub: entered with pending exception");
199 __ bind(L);
200 }
201 #endif
202
203 // pass parameters if any
204 BLOCK_COMMENT("pass parameters if any");
205 Label parameters_done;
206 __ movl(rcx, parameter_size); // parameter counter
207 __ testl(rcx, rcx);
208 __ jcc(Assembler::zero, parameters_done);
209
210 // parameter passing loop
211
212 Label loop;
213 // Copy Java parameters in reverse order (receiver last)
214 // Note that the argument order is inverted in the process
215 // source is rdx[rcx: N-1..0]
216 // dest is rsp[rbx: 0..N-1]
217
218 __ movptr(rdx, parameters); // parameter pointer
219 __ xorptr(rbx, rbx);
220
221 __ BIND(loop);
222
223 // get parameter
224 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
225 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
226 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
227 __ increment(rbx);
228 __ decrement(rcx);
229 __ jcc(Assembler::notZero, loop);
230
231 // call Java function
232 __ BIND(parameters_done);
233 __ movptr(rbx, method); // get Method*
234 __ movptr(rax, entry_point); // get entry_point
235 __ mov(rsi, rsp); // set sender sp
236 BLOCK_COMMENT("call Java function");
237 __ call(rax);
238
239 BLOCK_COMMENT("call_stub_return_address:");
240 return_address = __ pc();
241
242 #ifdef COMPILER2
243 {
244 Label L_skip;
245 if (UseSSE >= 2) {
246 __ verify_FPU(0, "call_stub_return");
247 } else {
248 for (int i = 1; i < 8; i++) {
249 __ ffree(i);
250 }
251
252 // UseSSE <= 1 so double result should be left on TOS
253 __ movl(rsi, result_type);
254 __ cmpl(rsi, T_DOUBLE);
255 __ jcc(Assembler::equal, L_skip);
256 if (UseSSE == 0) {
257 // UseSSE == 0 so float result should be left on TOS
258 __ cmpl(rsi, T_FLOAT);
259 __ jcc(Assembler::equal, L_skip);
260 }
261 __ ffree(0);
262 }
263 __ BIND(L_skip);
264 }
265 #endif // COMPILER2
266
267 // store result depending on type
268 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
269 __ movptr(rdi, result);
270 Label is_long, is_float, is_double, exit;
271 __ movl(rsi, result_type);
272 __ cmpl(rsi, T_LONG);
273 __ jcc(Assembler::equal, is_long);
274 __ cmpl(rsi, T_FLOAT);
275 __ jcc(Assembler::equal, is_float);
276 __ cmpl(rsi, T_DOUBLE);
277 __ jcc(Assembler::equal, is_double);
278
279 // handle T_INT case
280 __ movl(Address(rdi, 0), rax);
281 __ BIND(exit);
282
283 // check that FPU stack is empty
284 __ verify_FPU(0, "generate_call_stub");
285
286 // pop parameters
287 __ lea(rsp, rsp_after_call);
288
289 // restore %mxcsr
290 if (sse_save) {
291 __ ldmxcsr(mxcsr_save);
292 }
293
294 // restore rdi, rsi and rbx,
295 __ movptr(rbx, saved_rbx);
296 __ movptr(rsi, saved_rsi);
297 __ movptr(rdi, saved_rdi);
298 __ addptr(rsp, 4*wordSize);
299
300 // return
301 __ pop(rbp);
302 __ ret(0);
303
304 // handle return types different from T_INT
305 __ BIND(is_long);
306 __ movl(Address(rdi, 0 * wordSize), rax);
307 __ movl(Address(rdi, 1 * wordSize), rdx);
308 __ jmp(exit);
309
310 __ BIND(is_float);
311 // interpreter uses xmm0 for return values
312 if (UseSSE >= 1) {
313 __ movflt(Address(rdi, 0), xmm0);
314 } else {
315 __ fstp_s(Address(rdi, 0));
316 }
317 __ jmp(exit);
318
319 __ BIND(is_double);
320 // interpreter uses xmm0 for return values
321 if (UseSSE >= 2) {
322 __ movdbl(Address(rdi, 0), xmm0);
323 } else {
324 __ fstp_d(Address(rdi, 0));
325 }
326 __ jmp(exit);
327
328 return start;
329 }
330
331
332 //------------------------------------------------------------------------------------------------------------------------
333 // Return point for a Java call if there's an exception thrown in Java code.
334 // The exception is caught and transformed into a pending exception stored in
335 // JavaThread that can be tested from within the VM.
336 //
337 // Note: Usually the parameters are removed by the callee. In case of an exception
338 // crossing an activation frame boundary, that is not the case if the callee
339 // is compiled code => need to setup the rsp.
340 //
341 // rax,: exception oop
342
343 address generate_catch_exception() {
344 StubCodeMark mark(this, "StubRoutines", "catch_exception");
345 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
346 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
347 address start = __ pc();
348
349 // get thread directly
350 __ movptr(rcx, thread);
351 #ifdef ASSERT
352 // verify that threads correspond
353 { Label L;
354 __ get_thread(rbx);
355 __ cmpptr(rbx, rcx);
356 __ jcc(Assembler::equal, L);
357 __ stop("StubRoutines::catch_exception: threads must correspond");
358 __ bind(L);
359 }
360 #endif
361 // set pending exception
362 __ verify_oop(rax);
363 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
364 __ lea(Address(rcx, Thread::exception_file_offset ()),
365 ExternalAddress((address)__FILE__));
366 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
367 // complete return to VM
368 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
369 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
370
371 return start;
372 }
373
374
375 //------------------------------------------------------------------------------------------------------------------------
376 // Continuation point for runtime calls returning with a pending exception.
377 // The pending exception check happened in the runtime or native call stub.
378 // The pending exception in Thread is converted into a Java-level exception.
379 //
380 // Contract with Java-level exception handlers:
381 // rax: exception
382 // rdx: throwing pc
383 //
384 // NOTE: At entry of this stub, exception-pc must be on stack !!
385
386 address generate_forward_exception() {
387 StubCodeMark mark(this, "StubRoutines", "forward exception");
388 address start = __ pc();
389 const Register thread = rcx;
390
391 // other registers used in this stub
392 const Register exception_oop = rax;
393 const Register handler_addr = rbx;
394 const Register exception_pc = rdx;
395
396 // Upon entry, the sp points to the return address returning into Java
397 // (interpreted or compiled) code; i.e., the return address becomes the
398 // throwing pc.
399 //
400 // Arguments pushed before the runtime call are still on the stack but
401 // the exception handler will reset the stack pointer -> ignore them.
402 // A potential result in registers can be ignored as well.
403
404 #ifdef ASSERT
405 // make sure this code is only executed if there is a pending exception
406 { Label L;
407 __ get_thread(thread);
408 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
409 __ jcc(Assembler::notEqual, L);
410 __ stop("StubRoutines::forward exception: no pending exception (1)");
411 __ bind(L);
412 }
413 #endif
414
415 // compute exception handler into rbx,
416 __ get_thread(thread);
417 __ movptr(exception_pc, Address(rsp, 0));
418 BLOCK_COMMENT("call exception_handler_for_return_address");
419 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
420 __ mov(handler_addr, rax);
421
422 // setup rax & rdx, remove return address & clear pending exception
423 __ get_thread(thread);
424 __ pop(exception_pc);
425 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
426 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
427
428 #ifdef ASSERT
429 // make sure exception is set
430 { Label L;
431 __ testptr(exception_oop, exception_oop);
432 __ jcc(Assembler::notEqual, L);
433 __ stop("StubRoutines::forward exception: no pending exception (2)");
434 __ bind(L);
435 }
436 #endif
437
438 // Verify that there is really a valid exception in RAX.
439 __ verify_oop(exception_oop);
440
441 // continue at exception handler (return address removed)
442 // rax: exception
443 // rbx: exception handler
444 // rdx: throwing pc
445 __ jmp(handler_addr);
446
447 return start;
448 }
449
450
451 //----------------------------------------------------------------------------------------------------
452 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
453 //
454 // xchg exists as far back as 8086, lock needed for MP only
455 // Stack layout immediately after call:
456 //
457 // 0 [ret addr ] <--- rsp
458 // 1 [ ex ]
459 // 2 [ dest ]
460 //
461 // Result: *dest <- ex, return (old *dest)
462 //
463 // Note: win32 does not currently use this code
464
465 address generate_atomic_xchg() {
466 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
467 address start = __ pc();
468
469 __ push(rdx);
470 Address exchange(rsp, 2 * wordSize);
471 Address dest_addr(rsp, 3 * wordSize);
472 __ movl(rax, exchange);
473 __ movptr(rdx, dest_addr);
474 __ xchgl(rax, Address(rdx, 0));
475 __ pop(rdx);
476 __ ret(0);
477
478 return start;
479 }
480
481 //----------------------------------------------------------------------------------------------------
482 // Support for void verify_mxcsr()
483 //
484 // This routine is used with -Xcheck:jni to verify that native
485 // JNI code does not return to Java code without restoring the
486 // MXCSR register to our expected state.
487
488
489 address generate_verify_mxcsr() {
490 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
491 address start = __ pc();
492
493 const Address mxcsr_save(rsp, 0);
494
495 if (CheckJNICalls && UseSSE > 0 ) {
496 Label ok_ret;
497 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
498 __ push(rax);
499 __ subptr(rsp, wordSize); // allocate a temp location
500 __ stmxcsr(mxcsr_save);
501 __ movl(rax, mxcsr_save);
502 __ andl(rax, MXCSR_MASK);
503 __ cmp32(rax, mxcsr_std);
504 __ jcc(Assembler::equal, ok_ret);
505
506 __ warn("MXCSR changed by native JNI code.");
507
508 __ ldmxcsr(mxcsr_std);
509
510 __ bind(ok_ret);
511 __ addptr(rsp, wordSize);
512 __ pop(rax);
513 }
514
515 __ ret(0);
516
517 return start;
518 }
519
520
521 //---------------------------------------------------------------------------
522 // Support for void verify_fpu_cntrl_wrd()
523 //
524 // This routine is used with -Xcheck:jni to verify that native
525 // JNI code does not return to Java code without restoring the
526 // FP control word to our expected state.
527
528 address generate_verify_fpu_cntrl_wrd() {
529 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
530 address start = __ pc();
531
532 const Address fpu_cntrl_wrd_save(rsp, 0);
533
534 if (CheckJNICalls) {
535 Label ok_ret;
536 __ push(rax);
537 __ subptr(rsp, wordSize); // allocate a temp location
538 __ fnstcw(fpu_cntrl_wrd_save);
539 __ movl(rax, fpu_cntrl_wrd_save);
540 __ andl(rax, FPU_CNTRL_WRD_MASK);
541 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
542 __ cmp32(rax, fpu_std);
543 __ jcc(Assembler::equal, ok_ret);
544
545 __ warn("Floating point control word changed by native JNI code.");
546
547 __ fldcw(fpu_std);
548
549 __ bind(ok_ret);
550 __ addptr(rsp, wordSize);
551 __ pop(rax);
552 }
553
554 __ ret(0);
555
556 return start;
557 }
558
559 //---------------------------------------------------------------------------
560 // Wrapper for slow-case handling of double-to-integer conversion
561 // d2i or f2i fast case failed either because it is nan or because
562 // of under/overflow.
563 // Input: FPU TOS: float value
564 // Output: rax, (rdx): integer (long) result
565
566 address generate_d2i_wrapper(BasicType t, address fcn) {
567 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
568 address start = __ pc();
569
570 // Capture info about frame layout
571 enum layout { FPUState_off = 0,
572 rbp_off = FPUStateSizeInWords,
573 rdi_off,
574 rsi_off,
575 rcx_off,
576 rbx_off,
577 saved_argument_off,
578 saved_argument_off2, // 2nd half of double
579 framesize
580 };
581
582 assert(FPUStateSizeInWords == 27, "update stack layout");
583
584 // Save outgoing argument to stack across push_FPU_state()
585 __ subptr(rsp, wordSize * 2);
586 __ fstp_d(Address(rsp, 0));
587
588 // Save CPU & FPU state
589 __ push(rbx);
590 __ push(rcx);
591 __ push(rsi);
592 __ push(rdi);
593 __ push(rbp);
594 __ push_FPU_state();
595
596 // push_FPU_state() resets the FP top of stack
597 // Load original double into FP top of stack
598 __ fld_d(Address(rsp, saved_argument_off * wordSize));
599 // Store double into stack as outgoing argument
600 __ subptr(rsp, wordSize*2);
601 __ fst_d(Address(rsp, 0));
602
603 // Prepare FPU for doing math in C-land
604 __ empty_FPU_stack();
605 // Call the C code to massage the double. Result in EAX
606 if (t == T_INT)
607 { BLOCK_COMMENT("SharedRuntime::d2i"); }
608 else if (t == T_LONG)
609 { BLOCK_COMMENT("SharedRuntime::d2l"); }
610 __ call_VM_leaf( fcn, 2 );
611
612 // Restore CPU & FPU state
613 __ pop_FPU_state();
614 __ pop(rbp);
615 __ pop(rdi);
616 __ pop(rsi);
617 __ pop(rcx);
618 __ pop(rbx);
619 __ addptr(rsp, wordSize * 2);
620
621 __ ret(0);
622
623 return start;
624 }
625
626
627 //---------------------------------------------------------------------------
628 // The following routine generates a subroutine to throw an asynchronous
629 // UnknownError when an unsafe access gets a fault that could not be
630 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
631 address generate_handler_for_unsafe_access() {
632 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
633 address start = __ pc();
634
635 __ push(0); // hole for return address-to-be
636 __ pusha(); // push registers
637 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
638 BLOCK_COMMENT("call handle_unsafe_access");
639 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
640 __ movptr(next_pc, rax); // stuff next address
641 __ popa();
642 __ ret(0); // jump to next address
643
644 return start;
645 }
646
647
648 //----------------------------------------------------------------------------------------------------
649 // Non-destructive plausibility checks for oops
650
651 address generate_verify_oop() {
652 StubCodeMark mark(this, "StubRoutines", "verify_oop");
653 address start = __ pc();
654
655 // Incoming arguments on stack after saving rax,:
656 //
657 // [tos ]: saved rdx
658 // [tos + 1]: saved EFLAGS
659 // [tos + 2]: return address
660 // [tos + 3]: char* error message
661 // [tos + 4]: oop object to verify
662 // [tos + 5]: saved rax, - saved by caller and bashed
663
664 Label exit, error;
665 __ pushf();
666 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
667 __ push(rdx); // save rdx
668 // make sure object is 'reasonable'
669 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
670 __ testptr(rax, rax);
671 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
672
673 // Check if the oop is in the right area of memory
674 const int oop_mask = Universe::verify_oop_mask();
675 const int oop_bits = Universe::verify_oop_bits();
676 __ mov(rdx, rax);
677 __ andptr(rdx, oop_mask);
678 __ cmpptr(rdx, oop_bits);
679 __ jcc(Assembler::notZero, error);
680
681 // make sure klass is 'reasonable', which is not zero.
682 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
683 __ testptr(rax, rax);
684 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
685
686 // return if everything seems ok
687 __ bind(exit);
688 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
689 __ pop(rdx); // restore rdx
690 __ popf(); // restore EFLAGS
691 __ ret(3 * wordSize); // pop arguments
692
693 // handle errors
694 __ bind(error);
695 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
696 __ pop(rdx); // get saved rdx back
697 __ popf(); // get saved EFLAGS off stack -- will be ignored
698 __ pusha(); // push registers (eip = return address & msg are already pushed)
699 BLOCK_COMMENT("call MacroAssembler::debug");
700 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
701 __ popa();
702 __ ret(3 * wordSize); // pop arguments
703 return start;
704 }
705
706 //
707 // Generate pre-barrier for array stores
708 //
709 // Input:
710 // start - starting address
711 // count - element count
712 void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
713 assert_different_registers(start, count);
714 BarrierSet* bs = Universe::heap()->barrier_set();
715 switch (bs->kind()) {
716 case BarrierSet::G1SATBCTLogging:
717 // With G1, don't generate the call if we statically know that the target in uninitialized
718 if (!uninitialized_target) {
719 __ pusha(); // push registers
720 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
721 start, count);
722 __ popa();
723 }
724 break;
725 case BarrierSet::CardTableModRef:
726 case BarrierSet::CardTableExtension:
727 case BarrierSet::ModRef:
728 break;
729 default :
730 ShouldNotReachHere();
731
732 }
733 }
734
735
736 //
737 // Generate a post-barrier for an array store
738 //
739 // start - starting address
740 // count - element count
741 //
742 // The two input registers are overwritten.
743 //
744 void gen_write_ref_array_post_barrier(Register start, Register count) {
745 BarrierSet* bs = Universe::heap()->barrier_set();
746 assert_different_registers(start, count);
747 switch (bs->kind()) {
748 case BarrierSet::G1SATBCTLogging:
749 {
750 __ pusha(); // push registers
751 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
752 start, count);
753 __ popa();
754 }
755 break;
756
757 case BarrierSet::CardTableModRef:
758 case BarrierSet::CardTableExtension:
759 {
760 CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs);
761 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
762
763 Label L_loop;
764 const Register end = count; // elements count; end == start+count-1
765 assert_different_registers(start, end);
766
767 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
768 __ shrptr(start, CardTableModRefBS::card_shift);
769 __ shrptr(end, CardTableModRefBS::card_shift);
770 __ subptr(end, start); // end --> count
771 __ BIND(L_loop);
772 intptr_t disp = (intptr_t) ct->byte_map_base;
773 Address cardtable(start, count, Address::times_1, disp);
774 __ movb(cardtable, 0);
775 __ decrement(count);
776 __ jcc(Assembler::greaterEqual, L_loop);
777 }
778 break;
779 case BarrierSet::ModRef:
780 break;
781 default :
782 ShouldNotReachHere();
783
784 }
785 }
786
787
788 // Copy 64 bytes chunks
789 //
790 // Inputs:
791 // from - source array address
792 // to_from - destination array address - from
793 // qword_count - 8-bytes element count, negative
794 //
795 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
796 assert( UseSSE >= 2, "supported cpu only" );
797 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
798 // Copy 64-byte chunks
799 __ jmpb(L_copy_64_bytes);
800 __ align(OptoLoopAlignment);
801 __ BIND(L_copy_64_bytes_loop);
802
803 if (UseUnalignedLoadStores) {
804 if (UseAVX > 2) {
805 __ evmovdqu(xmm0, Address(from, 0), Assembler::AVX_512bit);
806 __ evmovdqu(Address(from, to_from, Address::times_1, 0), xmm0, Assembler::AVX_512bit);
807 } else if (UseAVX == 2) {
808 __ vmovdqu(xmm0, Address(from, 0));
809 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
810 __ vmovdqu(xmm1, Address(from, 32));
811 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
812 } else {
813 __ movdqu(xmm0, Address(from, 0));
814 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
815 __ movdqu(xmm1, Address(from, 16));
816 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
817 __ movdqu(xmm2, Address(from, 32));
818 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
819 __ movdqu(xmm3, Address(from, 48));
820 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
821 }
822 } else {
823 __ movq(xmm0, Address(from, 0));
824 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
825 __ movq(xmm1, Address(from, 8));
826 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
827 __ movq(xmm2, Address(from, 16));
828 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
829 __ movq(xmm3, Address(from, 24));
830 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
831 __ movq(xmm4, Address(from, 32));
832 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
833 __ movq(xmm5, Address(from, 40));
834 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
835 __ movq(xmm6, Address(from, 48));
836 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
837 __ movq(xmm7, Address(from, 56));
838 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
839 }
840
841 __ addl(from, 64);
842 __ BIND(L_copy_64_bytes);
843 __ subl(qword_count, 8);
844 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
845
846 if (UseUnalignedLoadStores && (UseAVX == 2)) {
847 // clean upper bits of YMM registers
848 __ vzeroupper();
849 }
850 __ addl(qword_count, 8);
851 __ jccb(Assembler::zero, L_exit);
852 //
853 // length is too short, just copy qwords
854 //
855 __ BIND(L_copy_8_bytes);
856 __ movq(xmm0, Address(from, 0));
857 __ movq(Address(from, to_from, Address::times_1), xmm0);
858 __ addl(from, 8);
859 __ decrement(qword_count);
860 __ jcc(Assembler::greater, L_copy_8_bytes);
861 __ BIND(L_exit);
862 }
863
864 // Copy 64 bytes chunks
865 //
866 // Inputs:
867 // from - source array address
868 // to_from - destination array address - from
869 // qword_count - 8-bytes element count, negative
870 //
871 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
872 assert( VM_Version::supports_mmx(), "supported cpu only" );
873 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
874 // Copy 64-byte chunks
875 __ jmpb(L_copy_64_bytes);
876 __ align(OptoLoopAlignment);
877 __ BIND(L_copy_64_bytes_loop);
878 __ movq(mmx0, Address(from, 0));
879 __ movq(mmx1, Address(from, 8));
880 __ movq(mmx2, Address(from, 16));
881 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
882 __ movq(mmx3, Address(from, 24));
883 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
884 __ movq(mmx4, Address(from, 32));
885 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
886 __ movq(mmx5, Address(from, 40));
887 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
888 __ movq(mmx6, Address(from, 48));
889 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
890 __ movq(mmx7, Address(from, 56));
891 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
892 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
893 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
894 __ addptr(from, 64);
895 __ BIND(L_copy_64_bytes);
896 __ subl(qword_count, 8);
897 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
898 __ addl(qword_count, 8);
899 __ jccb(Assembler::zero, L_exit);
900 //
901 // length is too short, just copy qwords
902 //
903 __ BIND(L_copy_8_bytes);
904 __ movq(mmx0, Address(from, 0));
905 __ movq(Address(from, to_from, Address::times_1), mmx0);
906 __ addptr(from, 8);
907 __ decrement(qword_count);
908 __ jcc(Assembler::greater, L_copy_8_bytes);
909 __ BIND(L_exit);
910 __ emms();
911 }
912
913 address generate_disjoint_copy(BasicType t, bool aligned,
914 Address::ScaleFactor sf,
915 address* entry, const char *name,
916 bool dest_uninitialized = false) {
917 __ align(CodeEntryAlignment);
918 StubCodeMark mark(this, "StubRoutines", name);
919 address start = __ pc();
920
921 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
922 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
923
924 int shift = Address::times_ptr - sf;
925
926 const Register from = rsi; // source array address
927 const Register to = rdi; // destination array address
928 const Register count = rcx; // elements count
929 const Register to_from = to; // (to - from)
930 const Register saved_to = rdx; // saved destination array address
931
932 __ enter(); // required for proper stackwalking of RuntimeStub frame
933 __ push(rsi);
934 __ push(rdi);
935 __ movptr(from , Address(rsp, 12+ 4));
936 __ movptr(to , Address(rsp, 12+ 8));
937 __ movl(count, Address(rsp, 12+ 12));
938
939 if (entry != NULL) {
940 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
941 BLOCK_COMMENT("Entry:");
942 }
943
944 if (t == T_OBJECT) {
945 __ testl(count, count);
946 __ jcc(Assembler::zero, L_0_count);
947 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
948 __ mov(saved_to, to); // save 'to'
949 }
950
951 __ subptr(to, from); // to --> to_from
952 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
953 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
954 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
955 // align source address at 4 bytes address boundary
956 if (t == T_BYTE) {
957 // One byte misalignment happens only for byte arrays
958 __ testl(from, 1);
959 __ jccb(Assembler::zero, L_skip_align1);
960 __ movb(rax, Address(from, 0));
961 __ movb(Address(from, to_from, Address::times_1, 0), rax);
962 __ increment(from);
963 __ decrement(count);
964 __ BIND(L_skip_align1);
965 }
966 // Two bytes misalignment happens only for byte and short (char) arrays
967 __ testl(from, 2);
968 __ jccb(Assembler::zero, L_skip_align2);
969 __ movw(rax, Address(from, 0));
970 __ movw(Address(from, to_from, Address::times_1, 0), rax);
971 __ addptr(from, 2);
972 __ subl(count, 1<<(shift-1));
973 __ BIND(L_skip_align2);
974 }
975 if (!VM_Version::supports_mmx()) {
976 __ mov(rax, count); // save 'count'
977 __ shrl(count, shift); // bytes count
978 __ addptr(to_from, from);// restore 'to'
979 __ rep_mov();
980 __ subptr(to_from, from);// restore 'to_from'
981 __ mov(count, rax); // restore 'count'
982 __ jmpb(L_copy_2_bytes); // all dwords were copied
983 } else {
984 if (!UseUnalignedLoadStores) {
985 // align to 8 bytes, we know we are 4 byte aligned to start
986 __ testptr(from, 4);
987 __ jccb(Assembler::zero, L_copy_64_bytes);
988 __ movl(rax, Address(from, 0));
989 __ movl(Address(from, to_from, Address::times_1, 0), rax);
990 __ addptr(from, 4);
991 __ subl(count, 1<<shift);
992 }
993 __ BIND(L_copy_64_bytes);
994 __ mov(rax, count);
995 __ shrl(rax, shift+1); // 8 bytes chunk count
996 //
997 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
998 //
999 if (UseXMMForArrayCopy) {
1000 xmm_copy_forward(from, to_from, rax);
1001 } else {
1002 mmx_copy_forward(from, to_from, rax);
1003 }
1004 }
1005 // copy tailing dword
1006 __ BIND(L_copy_4_bytes);
1007 __ testl(count, 1<<shift);
1008 __ jccb(Assembler::zero, L_copy_2_bytes);
1009 __ movl(rax, Address(from, 0));
1010 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1011 if (t == T_BYTE || t == T_SHORT) {
1012 __ addptr(from, 4);
1013 __ BIND(L_copy_2_bytes);
1014 // copy tailing word
1015 __ testl(count, 1<<(shift-1));
1016 __ jccb(Assembler::zero, L_copy_byte);
1017 __ movw(rax, Address(from, 0));
1018 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1019 if (t == T_BYTE) {
1020 __ addptr(from, 2);
1021 __ BIND(L_copy_byte);
1022 // copy tailing byte
1023 __ testl(count, 1);
1024 __ jccb(Assembler::zero, L_exit);
1025 __ movb(rax, Address(from, 0));
1026 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1027 __ BIND(L_exit);
1028 } else {
1029 __ BIND(L_copy_byte);
1030 }
1031 } else {
1032 __ BIND(L_copy_2_bytes);
1033 }
1034
1035 if (t == T_OBJECT) {
1036 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1037 __ mov(to, saved_to); // restore 'to'
1038 gen_write_ref_array_post_barrier(to, count);
1039 __ BIND(L_0_count);
1040 }
1041 inc_copy_counter_np(t);
1042 __ pop(rdi);
1043 __ pop(rsi);
1044 __ leave(); // required for proper stackwalking of RuntimeStub frame
1045 __ xorptr(rax, rax); // return 0
1046 __ ret(0);
1047 return start;
1048 }
1049
1050
1051 address generate_fill(BasicType t, bool aligned, const char *name) {
1052 __ align(CodeEntryAlignment);
1053 StubCodeMark mark(this, "StubRoutines", name);
1054 address start = __ pc();
1055
1056 BLOCK_COMMENT("Entry:");
1057
1058 const Register to = rdi; // source array address
1059 const Register value = rdx; // value
1060 const Register count = rsi; // elements count
1061
1062 __ enter(); // required for proper stackwalking of RuntimeStub frame
1063 __ push(rsi);
1064 __ push(rdi);
1065 __ movptr(to , Address(rsp, 12+ 4));
1066 __ movl(value, Address(rsp, 12+ 8));
1067 __ movl(count, Address(rsp, 12+ 12));
1068
1069 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1070
1071 __ pop(rdi);
1072 __ pop(rsi);
1073 __ leave(); // required for proper stackwalking of RuntimeStub frame
1074 __ ret(0);
1075 return start;
1076 }
1077
1078 address generate_conjoint_copy(BasicType t, bool aligned,
1079 Address::ScaleFactor sf,
1080 address nooverlap_target,
1081 address* entry, const char *name,
1082 bool dest_uninitialized = false) {
1083 __ align(CodeEntryAlignment);
1084 StubCodeMark mark(this, "StubRoutines", name);
1085 address start = __ pc();
1086
1087 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1088 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1089
1090 int shift = Address::times_ptr - sf;
1091
1092 const Register src = rax; // source array address
1093 const Register dst = rdx; // destination array address
1094 const Register from = rsi; // source array address
1095 const Register to = rdi; // destination array address
1096 const Register count = rcx; // elements count
1097 const Register end = rax; // array end address
1098
1099 __ enter(); // required for proper stackwalking of RuntimeStub frame
1100 __ push(rsi);
1101 __ push(rdi);
1102 __ movptr(src , Address(rsp, 12+ 4)); // from
1103 __ movptr(dst , Address(rsp, 12+ 8)); // to
1104 __ movl2ptr(count, Address(rsp, 12+12)); // count
1105
1106 if (entry != NULL) {
1107 *entry = __ pc(); // Entry point from generic arraycopy stub.
1108 BLOCK_COMMENT("Entry:");
1109 }
1110
1111 // nooverlap_target expects arguments in rsi and rdi.
1112 __ mov(from, src);
1113 __ mov(to , dst);
1114
1115 // arrays overlap test: dispatch to disjoint stub if necessary.
1116 RuntimeAddress nooverlap(nooverlap_target);
1117 __ cmpptr(dst, src);
1118 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1119 __ jump_cc(Assembler::belowEqual, nooverlap);
1120 __ cmpptr(dst, end);
1121 __ jump_cc(Assembler::aboveEqual, nooverlap);
1122
1123 if (t == T_OBJECT) {
1124 __ testl(count, count);
1125 __ jcc(Assembler::zero, L_0_count);
1126 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1127 }
1128
1129 // copy from high to low
1130 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1131 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1132 if (t == T_BYTE || t == T_SHORT) {
1133 // Align the end of destination array at 4 bytes address boundary
1134 __ lea(end, Address(dst, count, sf, 0));
1135 if (t == T_BYTE) {
1136 // One byte misalignment happens only for byte arrays
1137 __ testl(end, 1);
1138 __ jccb(Assembler::zero, L_skip_align1);
1139 __ decrement(count);
1140 __ movb(rdx, Address(from, count, sf, 0));
1141 __ movb(Address(to, count, sf, 0), rdx);
1142 __ BIND(L_skip_align1);
1143 }
1144 // Two bytes misalignment happens only for byte and short (char) arrays
1145 __ testl(end, 2);
1146 __ jccb(Assembler::zero, L_skip_align2);
1147 __ subptr(count, 1<<(shift-1));
1148 __ movw(rdx, Address(from, count, sf, 0));
1149 __ movw(Address(to, count, sf, 0), rdx);
1150 __ BIND(L_skip_align2);
1151 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1152 __ jcc(Assembler::below, L_copy_4_bytes);
1153 }
1154
1155 if (!VM_Version::supports_mmx()) {
1156 __ std();
1157 __ mov(rax, count); // Save 'count'
1158 __ mov(rdx, to); // Save 'to'
1159 __ lea(rsi, Address(from, count, sf, -4));
1160 __ lea(rdi, Address(to , count, sf, -4));
1161 __ shrptr(count, shift); // bytes count
1162 __ rep_mov();
1163 __ cld();
1164 __ mov(count, rax); // restore 'count'
1165 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1166 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1167 __ mov(to, rdx); // restore 'to'
1168 __ jmpb(L_copy_2_bytes); // all dword were copied
1169 } else {
1170 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1171 __ testptr(end, 4);
1172 __ jccb(Assembler::zero, L_copy_8_bytes);
1173 __ subl(count, 1<<shift);
1174 __ movl(rdx, Address(from, count, sf, 0));
1175 __ movl(Address(to, count, sf, 0), rdx);
1176 __ jmpb(L_copy_8_bytes);
1177
1178 __ align(OptoLoopAlignment);
1179 // Move 8 bytes
1180 __ BIND(L_copy_8_bytes_loop);
1181 if (UseXMMForArrayCopy) {
1182 __ movq(xmm0, Address(from, count, sf, 0));
1183 __ movq(Address(to, count, sf, 0), xmm0);
1184 } else {
1185 __ movq(mmx0, Address(from, count, sf, 0));
1186 __ movq(Address(to, count, sf, 0), mmx0);
1187 }
1188 __ BIND(L_copy_8_bytes);
1189 __ subl(count, 2<<shift);
1190 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1191 __ addl(count, 2<<shift);
1192 if (!UseXMMForArrayCopy) {
1193 __ emms();
1194 }
1195 }
1196 __ BIND(L_copy_4_bytes);
1197 // copy prefix qword
1198 __ testl(count, 1<<shift);
1199 __ jccb(Assembler::zero, L_copy_2_bytes);
1200 __ movl(rdx, Address(from, count, sf, -4));
1201 __ movl(Address(to, count, sf, -4), rdx);
1202
1203 if (t == T_BYTE || t == T_SHORT) {
1204 __ subl(count, (1<<shift));
1205 __ BIND(L_copy_2_bytes);
1206 // copy prefix dword
1207 __ testl(count, 1<<(shift-1));
1208 __ jccb(Assembler::zero, L_copy_byte);
1209 __ movw(rdx, Address(from, count, sf, -2));
1210 __ movw(Address(to, count, sf, -2), rdx);
1211 if (t == T_BYTE) {
1212 __ subl(count, 1<<(shift-1));
1213 __ BIND(L_copy_byte);
1214 // copy prefix byte
1215 __ testl(count, 1);
1216 __ jccb(Assembler::zero, L_exit);
1217 __ movb(rdx, Address(from, 0));
1218 __ movb(Address(to, 0), rdx);
1219 __ BIND(L_exit);
1220 } else {
1221 __ BIND(L_copy_byte);
1222 }
1223 } else {
1224 __ BIND(L_copy_2_bytes);
1225 }
1226 if (t == T_OBJECT) {
1227 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1228 gen_write_ref_array_post_barrier(to, count);
1229 __ BIND(L_0_count);
1230 }
1231 inc_copy_counter_np(t);
1232 __ pop(rdi);
1233 __ pop(rsi);
1234 __ leave(); // required for proper stackwalking of RuntimeStub frame
1235 __ xorptr(rax, rax); // return 0
1236 __ ret(0);
1237 return start;
1238 }
1239
1240
1241 address generate_disjoint_long_copy(address* entry, const char *name) {
1242 __ align(CodeEntryAlignment);
1243 StubCodeMark mark(this, "StubRoutines", name);
1244 address start = __ pc();
1245
1246 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1247 const Register from = rax; // source array address
1248 const Register to = rdx; // destination array address
1249 const Register count = rcx; // elements count
1250 const Register to_from = rdx; // (to - from)
1251
1252 __ enter(); // required for proper stackwalking of RuntimeStub frame
1253 __ movptr(from , Address(rsp, 8+0)); // from
1254 __ movptr(to , Address(rsp, 8+4)); // to
1255 __ movl2ptr(count, Address(rsp, 8+8)); // count
1256
1257 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1258 BLOCK_COMMENT("Entry:");
1259
1260 __ subptr(to, from); // to --> to_from
1261 if (VM_Version::supports_mmx()) {
1262 if (UseXMMForArrayCopy) {
1263 xmm_copy_forward(from, to_from, count);
1264 } else {
1265 mmx_copy_forward(from, to_from, count);
1266 }
1267 } else {
1268 __ jmpb(L_copy_8_bytes);
1269 __ align(OptoLoopAlignment);
1270 __ BIND(L_copy_8_bytes_loop);
1271 __ fild_d(Address(from, 0));
1272 __ fistp_d(Address(from, to_from, Address::times_1));
1273 __ addptr(from, 8);
1274 __ BIND(L_copy_8_bytes);
1275 __ decrement(count);
1276 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1277 }
1278 inc_copy_counter_np(T_LONG);
1279 __ leave(); // required for proper stackwalking of RuntimeStub frame
1280 __ xorptr(rax, rax); // return 0
1281 __ ret(0);
1282 return start;
1283 }
1284
1285 address generate_conjoint_long_copy(address nooverlap_target,
1286 address* entry, const char *name) {
1287 __ align(CodeEntryAlignment);
1288 StubCodeMark mark(this, "StubRoutines", name);
1289 address start = __ pc();
1290
1291 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1292 const Register from = rax; // source array address
1293 const Register to = rdx; // destination array address
1294 const Register count = rcx; // elements count
1295 const Register end_from = rax; // source array end address
1296
1297 __ enter(); // required for proper stackwalking of RuntimeStub frame
1298 __ movptr(from , Address(rsp, 8+0)); // from
1299 __ movptr(to , Address(rsp, 8+4)); // to
1300 __ movl2ptr(count, Address(rsp, 8+8)); // count
1301
1302 *entry = __ pc(); // Entry point from generic arraycopy stub.
1303 BLOCK_COMMENT("Entry:");
1304
1305 // arrays overlap test
1306 __ cmpptr(to, from);
1307 RuntimeAddress nooverlap(nooverlap_target);
1308 __ jump_cc(Assembler::belowEqual, nooverlap);
1309 __ lea(end_from, Address(from, count, Address::times_8, 0));
1310 __ cmpptr(to, end_from);
1311 __ movptr(from, Address(rsp, 8)); // from
1312 __ jump_cc(Assembler::aboveEqual, nooverlap);
1313
1314 __ jmpb(L_copy_8_bytes);
1315
1316 __ align(OptoLoopAlignment);
1317 __ BIND(L_copy_8_bytes_loop);
1318 if (VM_Version::supports_mmx()) {
1319 if (UseXMMForArrayCopy) {
1320 __ movq(xmm0, Address(from, count, Address::times_8));
1321 __ movq(Address(to, count, Address::times_8), xmm0);
1322 } else {
1323 __ movq(mmx0, Address(from, count, Address::times_8));
1324 __ movq(Address(to, count, Address::times_8), mmx0);
1325 }
1326 } else {
1327 __ fild_d(Address(from, count, Address::times_8));
1328 __ fistp_d(Address(to, count, Address::times_8));
1329 }
1330 __ BIND(L_copy_8_bytes);
1331 __ decrement(count);
1332 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1333
1334 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1335 __ emms();
1336 }
1337 inc_copy_counter_np(T_LONG);
1338 __ leave(); // required for proper stackwalking of RuntimeStub frame
1339 __ xorptr(rax, rax); // return 0
1340 __ ret(0);
1341 return start;
1342 }
1343
1344
1345 // Helper for generating a dynamic type check.
1346 // The sub_klass must be one of {rbx, rdx, rsi}.
1347 // The temp is killed.
1348 void generate_type_check(Register sub_klass,
1349 Address& super_check_offset_addr,
1350 Address& super_klass_addr,
1351 Register temp,
1352 Label* L_success, Label* L_failure) {
1353 BLOCK_COMMENT("type_check:");
1354
1355 Label L_fallthrough;
1356 #define LOCAL_JCC(assembler_con, label_ptr) \
1357 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1358 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1359
1360 // The following is a strange variation of the fast path which requires
1361 // one less register, because needed values are on the argument stack.
1362 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1363 // L_success, L_failure, NULL);
1364 assert_different_registers(sub_klass, temp);
1365
1366 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1367
1368 // if the pointers are equal, we are done (e.g., String[] elements)
1369 __ cmpptr(sub_klass, super_klass_addr);
1370 LOCAL_JCC(Assembler::equal, L_success);
1371
1372 // check the supertype display:
1373 __ movl2ptr(temp, super_check_offset_addr);
1374 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1375 __ movptr(temp, super_check_addr); // load displayed supertype
1376 __ cmpptr(temp, super_klass_addr); // test the super type
1377 LOCAL_JCC(Assembler::equal, L_success);
1378
1379 // if it was a primary super, we can just fail immediately
1380 __ cmpl(super_check_offset_addr, sc_offset);
1381 LOCAL_JCC(Assembler::notEqual, L_failure);
1382
1383 // The repne_scan instruction uses fixed registers, which will get spilled.
1384 // We happen to know this works best when super_klass is in rax.
1385 Register super_klass = temp;
1386 __ movptr(super_klass, super_klass_addr);
1387 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1388 L_success, L_failure);
1389
1390 __ bind(L_fallthrough);
1391
1392 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1393 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1394
1395 #undef LOCAL_JCC
1396 }
1397
1398 //
1399 // Generate checkcasting array copy stub
1400 //
1401 // Input:
1402 // 4(rsp) - source array address
1403 // 8(rsp) - destination array address
1404 // 12(rsp) - element count, can be zero
1405 // 16(rsp) - size_t ckoff (super_check_offset)
1406 // 20(rsp) - oop ckval (super_klass)
1407 //
1408 // Output:
1409 // rax, == 0 - success
1410 // rax, == -1^K - failure, where K is partial transfer count
1411 //
1412 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1413 __ align(CodeEntryAlignment);
1414 StubCodeMark mark(this, "StubRoutines", name);
1415 address start = __ pc();
1416
1417 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1418
1419 // register use:
1420 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1421 // rdi, rsi -- element access (oop, klass)
1422 // rbx, -- temp
1423 const Register from = rax; // source array address
1424 const Register to = rdx; // destination array address
1425 const Register length = rcx; // elements count
1426 const Register elem = rdi; // each oop copied
1427 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1428 const Register temp = rbx; // lone remaining temp
1429
1430 __ enter(); // required for proper stackwalking of RuntimeStub frame
1431
1432 __ push(rsi);
1433 __ push(rdi);
1434 __ push(rbx);
1435
1436 Address from_arg(rsp, 16+ 4); // from
1437 Address to_arg(rsp, 16+ 8); // to
1438 Address length_arg(rsp, 16+12); // elements count
1439 Address ckoff_arg(rsp, 16+16); // super_check_offset
1440 Address ckval_arg(rsp, 16+20); // super_klass
1441
1442 // Load up:
1443 __ movptr(from, from_arg);
1444 __ movptr(to, to_arg);
1445 __ movl2ptr(length, length_arg);
1446
1447 if (entry != NULL) {
1448 *entry = __ pc(); // Entry point from generic arraycopy stub.
1449 BLOCK_COMMENT("Entry:");
1450 }
1451
1452 //---------------------------------------------------------------
1453 // Assembler stub will be used for this call to arraycopy
1454 // if the two arrays are subtypes of Object[] but the
1455 // destination array type is not equal to or a supertype
1456 // of the source type. Each element must be separately
1457 // checked.
1458
1459 // Loop-invariant addresses. They are exclusive end pointers.
1460 Address end_from_addr(from, length, Address::times_ptr, 0);
1461 Address end_to_addr(to, length, Address::times_ptr, 0);
1462
1463 Register end_from = from; // re-use
1464 Register end_to = to; // re-use
1465 Register count = length; // re-use
1466
1467 // Loop-variant addresses. They assume post-incremented count < 0.
1468 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1469 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1470 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1471
1472 // Copy from low to high addresses, indexed from the end of each array.
1473 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1474 __ lea(end_from, end_from_addr);
1475 __ lea(end_to, end_to_addr);
1476 assert(length == count, ""); // else fix next line:
1477 __ negptr(count); // negate and test the length
1478 __ jccb(Assembler::notZero, L_load_element);
1479
1480 // Empty array: Nothing to do.
1481 __ xorptr(rax, rax); // return 0 on (trivial) success
1482 __ jmp(L_done);
1483
1484 // ======== begin loop ========
1485 // (Loop is rotated; its entry is L_load_element.)
1486 // Loop control:
1487 // for (count = -count; count != 0; count++)
1488 // Base pointers src, dst are biased by 8*count,to last element.
1489 __ align(OptoLoopAlignment);
1490
1491 __ BIND(L_store_element);
1492 __ movptr(to_element_addr, elem); // store the oop
1493 __ increment(count); // increment the count toward zero
1494 __ jccb(Assembler::zero, L_do_card_marks);
1495
1496 // ======== loop entry is here ========
1497 __ BIND(L_load_element);
1498 __ movptr(elem, from_element_addr); // load the oop
1499 __ testptr(elem, elem);
1500 __ jccb(Assembler::zero, L_store_element);
1501
1502 // (Could do a trick here: Remember last successful non-null
1503 // element stored and make a quick oop equality check on it.)
1504
1505 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1506 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1507 &L_store_element, NULL);
1508 // (On fall-through, we have failed the element type check.)
1509 // ======== end loop ========
1510
1511 // It was a real error; we must depend on the caller to finish the job.
1512 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1513 // Emit GC store barriers for the oops we have copied (length_arg + count),
1514 // and report their number to the caller.
1515 assert_different_registers(to, count, rax);
1516 Label L_post_barrier;
1517 __ addl(count, length_arg); // transfers = (length - remaining)
1518 __ movl2ptr(rax, count); // save the value
1519 __ notptr(rax); // report (-1^K) to caller (does not affect flags)
1520 __ jccb(Assembler::notZero, L_post_barrier);
1521 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier
1522
1523 // Come here on success only.
1524 __ BIND(L_do_card_marks);
1525 __ xorptr(rax, rax); // return 0 on success
1526 __ movl2ptr(count, length_arg);
1527
1528 __ BIND(L_post_barrier);
1529 __ movptr(to, to_arg); // reload
1530 gen_write_ref_array_post_barrier(to, count);
1531
1532 // Common exit point (success or failure).
1533 __ BIND(L_done);
1534 __ pop(rbx);
1535 __ pop(rdi);
1536 __ pop(rsi);
1537 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1538 __ leave(); // required for proper stackwalking of RuntimeStub frame
1539 __ ret(0);
1540
1541 return start;
1542 }
1543
1544 //
1545 // Generate 'unsafe' array copy stub
1546 // Though just as safe as the other stubs, it takes an unscaled
1547 // size_t argument instead of an element count.
1548 //
1549 // Input:
1550 // 4(rsp) - source array address
1551 // 8(rsp) - destination array address
1552 // 12(rsp) - byte count, can be zero
1553 //
1554 // Output:
1555 // rax, == 0 - success
1556 // rax, == -1 - need to call System.arraycopy
1557 //
1558 // Examines the alignment of the operands and dispatches
1559 // to a long, int, short, or byte copy loop.
1560 //
1561 address generate_unsafe_copy(const char *name,
1562 address byte_copy_entry,
1563 address short_copy_entry,
1564 address int_copy_entry,
1565 address long_copy_entry) {
1566
1567 Label L_long_aligned, L_int_aligned, L_short_aligned;
1568
1569 __ align(CodeEntryAlignment);
1570 StubCodeMark mark(this, "StubRoutines", name);
1571 address start = __ pc();
1572
1573 const Register from = rax; // source array address
1574 const Register to = rdx; // destination array address
1575 const Register count = rcx; // elements count
1576
1577 __ enter(); // required for proper stackwalking of RuntimeStub frame
1578 __ push(rsi);
1579 __ push(rdi);
1580 Address from_arg(rsp, 12+ 4); // from
1581 Address to_arg(rsp, 12+ 8); // to
1582 Address count_arg(rsp, 12+12); // byte count
1583
1584 // Load up:
1585 __ movptr(from , from_arg);
1586 __ movptr(to , to_arg);
1587 __ movl2ptr(count, count_arg);
1588
1589 // bump this on entry, not on exit:
1590 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1591
1592 const Register bits = rsi;
1593 __ mov(bits, from);
1594 __ orptr(bits, to);
1595 __ orptr(bits, count);
1596
1597 __ testl(bits, BytesPerLong-1);
1598 __ jccb(Assembler::zero, L_long_aligned);
1599
1600 __ testl(bits, BytesPerInt-1);
1601 __ jccb(Assembler::zero, L_int_aligned);
1602
1603 __ testl(bits, BytesPerShort-1);
1604 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1605
1606 __ BIND(L_short_aligned);
1607 __ shrptr(count, LogBytesPerShort); // size => short_count
1608 __ movl(count_arg, count); // update 'count'
1609 __ jump(RuntimeAddress(short_copy_entry));
1610
1611 __ BIND(L_int_aligned);
1612 __ shrptr(count, LogBytesPerInt); // size => int_count
1613 __ movl(count_arg, count); // update 'count'
1614 __ jump(RuntimeAddress(int_copy_entry));
1615
1616 __ BIND(L_long_aligned);
1617 __ shrptr(count, LogBytesPerLong); // size => qword_count
1618 __ movl(count_arg, count); // update 'count'
1619 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1620 __ pop(rsi);
1621 __ jump(RuntimeAddress(long_copy_entry));
1622
1623 return start;
1624 }
1625
1626
1627 // Perform range checks on the proposed arraycopy.
1628 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1629 void arraycopy_range_checks(Register src,
1630 Register src_pos,
1631 Register dst,
1632 Register dst_pos,
1633 Address& length,
1634 Label& L_failed) {
1635 BLOCK_COMMENT("arraycopy_range_checks:");
1636 const Register src_end = src_pos; // source array end position
1637 const Register dst_end = dst_pos; // destination array end position
1638 __ addl(src_end, length); // src_pos + length
1639 __ addl(dst_end, length); // dst_pos + length
1640
1641 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1642 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1643 __ jcc(Assembler::above, L_failed);
1644
1645 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1646 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1647 __ jcc(Assembler::above, L_failed);
1648
1649 BLOCK_COMMENT("arraycopy_range_checks done");
1650 }
1651
1652
1653 //
1654 // Generate generic array copy stubs
1655 //
1656 // Input:
1657 // 4(rsp) - src oop
1658 // 8(rsp) - src_pos
1659 // 12(rsp) - dst oop
1660 // 16(rsp) - dst_pos
1661 // 20(rsp) - element count
1662 //
1663 // Output:
1664 // rax, == 0 - success
1665 // rax, == -1^K - failure, where K is partial transfer count
1666 //
1667 address generate_generic_copy(const char *name,
1668 address entry_jbyte_arraycopy,
1669 address entry_jshort_arraycopy,
1670 address entry_jint_arraycopy,
1671 address entry_oop_arraycopy,
1672 address entry_jlong_arraycopy,
1673 address entry_checkcast_arraycopy) {
1674 Label L_failed, L_failed_0, L_objArray;
1675
1676 { int modulus = CodeEntryAlignment;
1677 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1678 int advance = target - (__ offset() % modulus);
1679 if (advance < 0) advance += modulus;
1680 if (advance > 0) __ nop(advance);
1681 }
1682 StubCodeMark mark(this, "StubRoutines", name);
1683
1684 // Short-hop target to L_failed. Makes for denser prologue code.
1685 __ BIND(L_failed_0);
1686 __ jmp(L_failed);
1687 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1688
1689 __ align(CodeEntryAlignment);
1690 address start = __ pc();
1691
1692 __ enter(); // required for proper stackwalking of RuntimeStub frame
1693 __ push(rsi);
1694 __ push(rdi);
1695
1696 // bump this on entry, not on exit:
1697 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1698
1699 // Input values
1700 Address SRC (rsp, 12+ 4);
1701 Address SRC_POS (rsp, 12+ 8);
1702 Address DST (rsp, 12+12);
1703 Address DST_POS (rsp, 12+16);
1704 Address LENGTH (rsp, 12+20);
1705
1706 //-----------------------------------------------------------------------
1707 // Assembler stub will be used for this call to arraycopy
1708 // if the following conditions are met:
1709 //
1710 // (1) src and dst must not be null.
1711 // (2) src_pos must not be negative.
1712 // (3) dst_pos must not be negative.
1713 // (4) length must not be negative.
1714 // (5) src klass and dst klass should be the same and not NULL.
1715 // (6) src and dst should be arrays.
1716 // (7) src_pos + length must not exceed length of src.
1717 // (8) dst_pos + length must not exceed length of dst.
1718 //
1719
1720 const Register src = rax; // source array oop
1721 const Register src_pos = rsi;
1722 const Register dst = rdx; // destination array oop
1723 const Register dst_pos = rdi;
1724 const Register length = rcx; // transfer count
1725
1726 // if (src == NULL) return -1;
1727 __ movptr(src, SRC); // src oop
1728 __ testptr(src, src);
1729 __ jccb(Assembler::zero, L_failed_0);
1730
1731 // if (src_pos < 0) return -1;
1732 __ movl2ptr(src_pos, SRC_POS); // src_pos
1733 __ testl(src_pos, src_pos);
1734 __ jccb(Assembler::negative, L_failed_0);
1735
1736 // if (dst == NULL) return -1;
1737 __ movptr(dst, DST); // dst oop
1738 __ testptr(dst, dst);
1739 __ jccb(Assembler::zero, L_failed_0);
1740
1741 // if (dst_pos < 0) return -1;
1742 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1743 __ testl(dst_pos, dst_pos);
1744 __ jccb(Assembler::negative, L_failed_0);
1745
1746 // if (length < 0) return -1;
1747 __ movl2ptr(length, LENGTH); // length
1748 __ testl(length, length);
1749 __ jccb(Assembler::negative, L_failed_0);
1750
1751 // if (src->klass() == NULL) return -1;
1752 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1753 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1754 const Register rcx_src_klass = rcx; // array klass
1755 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1756
1757 #ifdef ASSERT
1758 // assert(src->klass() != NULL);
1759 BLOCK_COMMENT("assert klasses not null");
1760 { Label L1, L2;
1761 __ testptr(rcx_src_klass, rcx_src_klass);
1762 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1763 __ bind(L1);
1764 __ stop("broken null klass");
1765 __ bind(L2);
1766 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1767 __ jccb(Assembler::equal, L1); // this would be broken also
1768 BLOCK_COMMENT("assert done");
1769 }
1770 #endif //ASSERT
1771
1772 // Load layout helper (32-bits)
1773 //
1774 // |array_tag| | header_size | element_type | |log2_element_size|
1775 // 32 30 24 16 8 2 0
1776 //
1777 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1778 //
1779
1780 int lh_offset = in_bytes(Klass::layout_helper_offset());
1781 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1782
1783 // Handle objArrays completely differently...
1784 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1785 __ cmpl(src_klass_lh_addr, objArray_lh);
1786 __ jcc(Assembler::equal, L_objArray);
1787
1788 // if (src->klass() != dst->klass()) return -1;
1789 __ cmpptr(rcx_src_klass, dst_klass_addr);
1790 __ jccb(Assembler::notEqual, L_failed_0);
1791
1792 const Register rcx_lh = rcx; // layout helper
1793 assert(rcx_lh == rcx_src_klass, "known alias");
1794 __ movl(rcx_lh, src_klass_lh_addr);
1795
1796 // if (!src->is_Array()) return -1;
1797 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1798 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1799
1800 // At this point, it is known to be a typeArray (array_tag 0x3).
1801 #ifdef ASSERT
1802 { Label L;
1803 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1804 __ jcc(Assembler::greaterEqual, L); // signed cmp
1805 __ stop("must be a primitive array");
1806 __ bind(L);
1807 }
1808 #endif
1809
1810 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1811 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1812
1813 // TypeArrayKlass
1814 //
1815 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1816 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1817 //
1818 const Register rsi_offset = rsi; // array offset
1819 const Register src_array = src; // src array offset
1820 const Register dst_array = dst; // dst array offset
1821 const Register rdi_elsize = rdi; // log2 element size
1822
1823 __ mov(rsi_offset, rcx_lh);
1824 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1825 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1826 __ addptr(src_array, rsi_offset); // src array offset
1827 __ addptr(dst_array, rsi_offset); // dst array offset
1828 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1829
1830 // next registers should be set before the jump to corresponding stub
1831 const Register from = src; // source array address
1832 const Register to = dst; // destination array address
1833 const Register count = rcx; // elements count
1834 // some of them should be duplicated on stack
1835 #define FROM Address(rsp, 12+ 4)
1836 #define TO Address(rsp, 12+ 8) // Not used now
1837 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1838
1839 BLOCK_COMMENT("scale indexes to element size");
1840 __ movl2ptr(rsi, SRC_POS); // src_pos
1841 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1842 assert(src_array == from, "");
1843 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1844 __ movl2ptr(rdi, DST_POS); // dst_pos
1845 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1846 assert(dst_array == to, "");
1847 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1848 __ movptr(FROM, from); // src_addr
1849 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1850 __ movl2ptr(count, LENGTH); // elements count
1851
1852 BLOCK_COMMENT("choose copy loop based on element size");
1853 __ cmpl(rdi_elsize, 0);
1854
1855 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1856 __ cmpl(rdi_elsize, LogBytesPerShort);
1857 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1858 __ cmpl(rdi_elsize, LogBytesPerInt);
1859 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1860 #ifdef ASSERT
1861 __ cmpl(rdi_elsize, LogBytesPerLong);
1862 __ jccb(Assembler::notEqual, L_failed);
1863 #endif
1864 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1865 __ pop(rsi);
1866 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1867
1868 __ BIND(L_failed);
1869 __ xorptr(rax, rax);
1870 __ notptr(rax); // return -1
1871 __ pop(rdi);
1872 __ pop(rsi);
1873 __ leave(); // required for proper stackwalking of RuntimeStub frame
1874 __ ret(0);
1875
1876 // ObjArrayKlass
1877 __ BIND(L_objArray);
1878 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1879
1880 Label L_plain_copy, L_checkcast_copy;
1881 // test array classes for subtyping
1882 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1883 __ jccb(Assembler::notEqual, L_checkcast_copy);
1884
1885 // Identically typed arrays can be copied without element-wise checks.
1886 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1887 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1888
1889 __ BIND(L_plain_copy);
1890 __ movl2ptr(count, LENGTH); // elements count
1891 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1892 __ lea(from, Address(src, src_pos, Address::times_ptr,
1893 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1894 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1895 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1896 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1897 __ movptr(FROM, from); // src_addr
1898 __ movptr(TO, to); // dst_addr
1899 __ movl(COUNT, count); // count
1900 __ jump(RuntimeAddress(entry_oop_arraycopy));
1901
1902 __ BIND(L_checkcast_copy);
1903 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1904 {
1905 // Handy offsets:
1906 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
1907 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1908
1909 Register rsi_dst_klass = rsi;
1910 Register rdi_temp = rdi;
1911 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1912 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1913 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1914
1915 // Before looking at dst.length, make sure dst is also an objArray.
1916 __ movptr(rsi_dst_klass, dst_klass_addr);
1917 __ cmpl(dst_klass_lh_addr, objArray_lh);
1918 __ jccb(Assembler::notEqual, L_failed);
1919
1920 // It is safe to examine both src.length and dst.length.
1921 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1922 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1923 // (Now src_pos and dst_pos are killed, but not src and dst.)
1924
1925 // We'll need this temp (don't forget to pop it after the type check).
1926 __ push(rbx);
1927 Register rbx_src_klass = rbx;
1928
1929 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1930 __ movptr(rsi_dst_klass, dst_klass_addr);
1931 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1932 Label L_fail_array_check;
1933 generate_type_check(rbx_src_klass,
1934 super_check_offset_addr, dst_klass_addr,
1935 rdi_temp, NULL, &L_fail_array_check);
1936 // (On fall-through, we have passed the array type check.)
1937 __ pop(rbx);
1938 __ jmp(L_plain_copy);
1939
1940 __ BIND(L_fail_array_check);
1941 // Reshuffle arguments so we can call checkcast_arraycopy:
1942
1943 // match initial saves for checkcast_arraycopy
1944 // push(rsi); // already done; see above
1945 // push(rdi); // already done; see above
1946 // push(rbx); // already done; see above
1947
1948 // Marshal outgoing arguments now, freeing registers.
1949 Address from_arg(rsp, 16+ 4); // from
1950 Address to_arg(rsp, 16+ 8); // to
1951 Address length_arg(rsp, 16+12); // elements count
1952 Address ckoff_arg(rsp, 16+16); // super_check_offset
1953 Address ckval_arg(rsp, 16+20); // super_klass
1954
1955 Address SRC_POS_arg(rsp, 16+ 8);
1956 Address DST_POS_arg(rsp, 16+16);
1957 Address LENGTH_arg(rsp, 16+20);
1958 // push rbx, changed the incoming offsets (why not just use rbp,??)
1959 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1960
1961 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1962 __ movl2ptr(length, LENGTH_arg); // reload elements count
1963 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1964 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1965
1966 __ movptr(ckval_arg, rbx); // destination element type
1967 __ movl(rbx, Address(rbx, sco_offset));
1968 __ movl(ckoff_arg, rbx); // corresponding class check offset
1969
1970 __ movl(length_arg, length); // outgoing length argument
1971
1972 __ lea(from, Address(src, src_pos, Address::times_ptr,
1973 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1974 __ movptr(from_arg, from);
1975
1976 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1977 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1978 __ movptr(to_arg, to);
1979 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1980 }
1981
1982 return start;
1983 }
1984
1985 void generate_arraycopy_stubs() {
1986 address entry;
1987 address entry_jbyte_arraycopy;
1988 address entry_jshort_arraycopy;
1989 address entry_jint_arraycopy;
1990 address entry_oop_arraycopy;
1991 address entry_jlong_arraycopy;
1992 address entry_checkcast_arraycopy;
1993
1994 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1995 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1996 "arrayof_jbyte_disjoint_arraycopy");
1997 StubRoutines::_arrayof_jbyte_arraycopy =
1998 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1999 NULL, "arrayof_jbyte_arraycopy");
2000 StubRoutines::_jbyte_disjoint_arraycopy =
2001 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2002 "jbyte_disjoint_arraycopy");
2003 StubRoutines::_jbyte_arraycopy =
2004 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
2005 &entry_jbyte_arraycopy, "jbyte_arraycopy");
2006
2007 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2008 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2009 "arrayof_jshort_disjoint_arraycopy");
2010 StubRoutines::_arrayof_jshort_arraycopy =
2011 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2012 NULL, "arrayof_jshort_arraycopy");
2013 StubRoutines::_jshort_disjoint_arraycopy =
2014 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2015 "jshort_disjoint_arraycopy");
2016 StubRoutines::_jshort_arraycopy =
2017 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2018 &entry_jshort_arraycopy, "jshort_arraycopy");
2019
2020 // Next arrays are always aligned on 4 bytes at least.
2021 StubRoutines::_jint_disjoint_arraycopy =
2022 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2023 "jint_disjoint_arraycopy");
2024 StubRoutines::_jint_arraycopy =
2025 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2026 &entry_jint_arraycopy, "jint_arraycopy");
2027
2028 StubRoutines::_oop_disjoint_arraycopy =
2029 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2030 "oop_disjoint_arraycopy");
2031 StubRoutines::_oop_arraycopy =
2032 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2033 &entry_oop_arraycopy, "oop_arraycopy");
2034
2035 StubRoutines::_oop_disjoint_arraycopy_uninit =
2036 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2037 "oop_disjoint_arraycopy_uninit",
2038 /*dest_uninitialized*/true);
2039 StubRoutines::_oop_arraycopy_uninit =
2040 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2041 NULL, "oop_arraycopy_uninit",
2042 /*dest_uninitialized*/true);
2043
2044 StubRoutines::_jlong_disjoint_arraycopy =
2045 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2046 StubRoutines::_jlong_arraycopy =
2047 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2048 "jlong_arraycopy");
2049
2050 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2051 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2052 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2053 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2054 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2055 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2056
2057 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2058 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2059 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2060 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2061
2062 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2063 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2064 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2065 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2066
2067 StubRoutines::_checkcast_arraycopy =
2068 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2069 StubRoutines::_checkcast_arraycopy_uninit =
2070 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2071
2072 StubRoutines::_unsafe_arraycopy =
2073 generate_unsafe_copy("unsafe_arraycopy",
2074 entry_jbyte_arraycopy,
2075 entry_jshort_arraycopy,
2076 entry_jint_arraycopy,
2077 entry_jlong_arraycopy);
2078
2079 StubRoutines::_generic_arraycopy =
2080 generate_generic_copy("generic_arraycopy",
2081 entry_jbyte_arraycopy,
2082 entry_jshort_arraycopy,
2083 entry_jint_arraycopy,
2084 entry_oop_arraycopy,
2085 entry_jlong_arraycopy,
2086 entry_checkcast_arraycopy);
2087 }
2088
2089 void generate_math_stubs() {
2090 {
2091 StubCodeMark mark(this, "StubRoutines", "log");
2092 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2093
2094 __ fld_d(Address(rsp, 4));
2095 __ flog();
2096 __ ret(0);
2097 }
2098 {
2099 StubCodeMark mark(this, "StubRoutines", "log10");
2100 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2101
2102 __ fld_d(Address(rsp, 4));
2103 __ flog10();
2104 __ ret(0);
2105 }
2106 {
2107 StubCodeMark mark(this, "StubRoutines", "sin");
2108 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2109
2110 __ fld_d(Address(rsp, 4));
2111 __ trigfunc('s');
2112 __ ret(0);
2113 }
2114 {
2115 StubCodeMark mark(this, "StubRoutines", "cos");
2116 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2117
2118 __ fld_d(Address(rsp, 4));
2119 __ trigfunc('c');
2120 __ ret(0);
2121 }
2122 {
2123 StubCodeMark mark(this, "StubRoutines", "tan");
2124 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2125
2126 __ fld_d(Address(rsp, 4));
2127 __ trigfunc('t');
2128 __ ret(0);
2129 }
2130 {
2131 StubCodeMark mark(this, "StubRoutines", "exp");
2132 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2133
2134 __ fld_d(Address(rsp, 4));
2135 __ exp_with_fallback(0);
2136 __ ret(0);
2137 }
2138 {
2139 StubCodeMark mark(this, "StubRoutines", "pow");
2140 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2141
2142 __ fld_d(Address(rsp, 12));
2143 __ fld_d(Address(rsp, 4));
2144 __ pow_with_fallback(0);
2145 __ ret(0);
2146 }
2147 }
2148
2149 // AES intrinsic stubs
2150 enum {AESBlockSize = 16};
2151
2152 address generate_key_shuffle_mask() {
2153 __ align(16);
2154 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2155 address start = __ pc();
2156 __ emit_data(0x00010203, relocInfo::none, 0 );
2157 __ emit_data(0x04050607, relocInfo::none, 0 );
2158 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2159 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2160 return start;
2161 }
2162
2163 // Utility routine for loading a 128-bit key word in little endian format
2164 // can optionally specify that the shuffle mask is already in an xmmregister
2165 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2166 __ movdqu(xmmdst, Address(key, offset));
2167 if (xmm_shuf_mask != NULL) {
2168 __ pshufb(xmmdst, xmm_shuf_mask);
2169 } else {
2170 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2171 }
2172 }
2173
2174 // aesenc using specified key+offset
2175 // can optionally specify that the shuffle mask is already in an xmmregister
2176 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2177 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2178 __ aesenc(xmmdst, xmmtmp);
2179 }
2180
2181 // aesdec using specified key+offset
2182 // can optionally specify that the shuffle mask is already in an xmmregister
2183 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2184 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2185 __ aesdec(xmmdst, xmmtmp);
2186 }
2187
2188
2189 // Arguments:
2190 //
2191 // Inputs:
2192 // c_rarg0 - source byte array address
2193 // c_rarg1 - destination byte array address
2194 // c_rarg2 - K (key) in little endian int array
2195 //
2196 address generate_aescrypt_encryptBlock() {
2197 assert(UseAES, "need AES instructions and misaligned SSE support");
2198 __ align(CodeEntryAlignment);
2199 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2200 Label L_doLast;
2201 address start = __ pc();
2202
2203 const Register from = rdx; // source array address
2204 const Register to = rdx; // destination array address
2205 const Register key = rcx; // key array address
2206 const Register keylen = rax;
2207 const Address from_param(rbp, 8+0);
2208 const Address to_param (rbp, 8+4);
2209 const Address key_param (rbp, 8+8);
2210
2211 const XMMRegister xmm_result = xmm0;
2212 const XMMRegister xmm_key_shuf_mask = xmm1;
2213 const XMMRegister xmm_temp1 = xmm2;
2214 const XMMRegister xmm_temp2 = xmm3;
2215 const XMMRegister xmm_temp3 = xmm4;
2216 const XMMRegister xmm_temp4 = xmm5;
2217
2218 __ enter(); // required for proper stackwalking of RuntimeStub frame
2219 __ movptr(from, from_param);
2220 __ movptr(key, key_param);
2221
2222 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2223 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2224
2225 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2226 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2227 __ movptr(to, to_param);
2228
2229 // For encryption, the java expanded key ordering is just what we need
2230
2231 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2232 __ pxor(xmm_result, xmm_temp1);
2233
2234 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2235 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2236 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2237 load_key(xmm_temp4, key, 0x40, 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, 0x50, xmm_key_shuf_mask);
2245 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2246 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2247 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2248
2249 __ aesenc(xmm_result, xmm_temp1);
2250 __ aesenc(xmm_result, xmm_temp2);
2251 __ aesenc(xmm_result, xmm_temp3);
2252 __ aesenc(xmm_result, xmm_temp4);
2253
2254 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2255 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2256
2257 __ cmpl(keylen, 44);
2258 __ jccb(Assembler::equal, L_doLast);
2259
2260 __ aesenc(xmm_result, xmm_temp1);
2261 __ aesenc(xmm_result, xmm_temp2);
2262
2263 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2264 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2265
2266 __ cmpl(keylen, 52);
2267 __ jccb(Assembler::equal, L_doLast);
2268
2269 __ aesenc(xmm_result, xmm_temp1);
2270 __ aesenc(xmm_result, xmm_temp2);
2271
2272 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2273 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2274
2275 __ BIND(L_doLast);
2276 __ aesenc(xmm_result, xmm_temp1);
2277 __ aesenclast(xmm_result, xmm_temp2);
2278 __ movdqu(Address(to, 0), xmm_result); // store the result
2279 __ xorptr(rax, rax); // return 0
2280 __ leave(); // required for proper stackwalking of RuntimeStub frame
2281 __ ret(0);
2282
2283 return start;
2284 }
2285
2286
2287 // Arguments:
2288 //
2289 // Inputs:
2290 // c_rarg0 - source byte array address
2291 // c_rarg1 - destination byte array address
2292 // c_rarg2 - K (key) in little endian int array
2293 //
2294 address generate_aescrypt_decryptBlock() {
2295 assert(UseAES, "need AES instructions and misaligned SSE support");
2296 __ align(CodeEntryAlignment);
2297 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2298 Label L_doLast;
2299 address start = __ pc();
2300
2301 const Register from = rdx; // source array address
2302 const Register to = rdx; // destination array address
2303 const Register key = rcx; // key array address
2304 const Register keylen = rax;
2305 const Address from_param(rbp, 8+0);
2306 const Address to_param (rbp, 8+4);
2307 const Address key_param (rbp, 8+8);
2308
2309 const XMMRegister xmm_result = xmm0;
2310 const XMMRegister xmm_key_shuf_mask = xmm1;
2311 const XMMRegister xmm_temp1 = xmm2;
2312 const XMMRegister xmm_temp2 = xmm3;
2313 const XMMRegister xmm_temp3 = xmm4;
2314 const XMMRegister xmm_temp4 = xmm5;
2315
2316 __ enter(); // required for proper stackwalking of RuntimeStub frame
2317 __ movptr(from, from_param);
2318 __ movptr(key, key_param);
2319
2320 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2321 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2322
2323 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2324 __ movdqu(xmm_result, Address(from, 0));
2325 __ movptr(to, to_param);
2326
2327 // for decryption java expanded key ordering is rotated one position from what we want
2328 // so we start from 0x10 here and hit 0x00 last
2329 // we don't know if the key is aligned, hence not using load-execute form
2330 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2331 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2332 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2333 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2334
2335 __ pxor (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, 0x50, xmm_key_shuf_mask);
2341 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2342 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2343 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2344
2345 __ aesdec(xmm_result, xmm_temp1);
2346 __ aesdec(xmm_result, xmm_temp2);
2347 __ aesdec(xmm_result, xmm_temp3);
2348 __ aesdec(xmm_result, xmm_temp4);
2349
2350 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2351 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2352 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2353
2354 __ cmpl(keylen, 44);
2355 __ jccb(Assembler::equal, L_doLast);
2356
2357 __ aesdec(xmm_result, xmm_temp1);
2358 __ aesdec(xmm_result, xmm_temp2);
2359
2360 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2361 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2362
2363 __ cmpl(keylen, 52);
2364 __ jccb(Assembler::equal, L_doLast);
2365
2366 __ aesdec(xmm_result, xmm_temp1);
2367 __ aesdec(xmm_result, xmm_temp2);
2368
2369 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2370 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2371
2372 __ BIND(L_doLast);
2373 __ aesdec(xmm_result, xmm_temp1);
2374 __ aesdec(xmm_result, xmm_temp2);
2375
2376 // for decryption the aesdeclast operation is always on key+0x00
2377 __ aesdeclast(xmm_result, xmm_temp3);
2378 __ movdqu(Address(to, 0), xmm_result); // store the result
2379 __ xorptr(rax, rax); // return 0
2380 __ leave(); // required for proper stackwalking of RuntimeStub frame
2381 __ ret(0);
2382
2383 return start;
2384 }
2385
2386 void handleSOERegisters(bool saving) {
2387 const int saveFrameSizeInBytes = 4 * wordSize;
2388 const Address saved_rbx (rbp, -3 * wordSize);
2389 const Address saved_rsi (rbp, -2 * wordSize);
2390 const Address saved_rdi (rbp, -1 * wordSize);
2391
2392 if (saving) {
2393 __ subptr(rsp, saveFrameSizeInBytes);
2394 __ movptr(saved_rsi, rsi);
2395 __ movptr(saved_rdi, rdi);
2396 __ movptr(saved_rbx, rbx);
2397 } else {
2398 // restoring
2399 __ movptr(rsi, saved_rsi);
2400 __ movptr(rdi, saved_rdi);
2401 __ movptr(rbx, saved_rbx);
2402 }
2403 }
2404
2405 // Arguments:
2406 //
2407 // Inputs:
2408 // c_rarg0 - source byte array address
2409 // c_rarg1 - destination byte array address
2410 // c_rarg2 - K (key) in little endian int array
2411 // c_rarg3 - r vector byte array address
2412 // c_rarg4 - input length
2413 //
2414 // Output:
2415 // rax - input length
2416 //
2417 address generate_cipherBlockChaining_encryptAESCrypt() {
2418 assert(UseAES, "need AES instructions and misaligned SSE support");
2419 __ align(CodeEntryAlignment);
2420 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2421 address start = __ pc();
2422
2423 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2424 const Register from = rsi; // source array address
2425 const Register to = rdx; // destination array address
2426 const Register key = rcx; // key array address
2427 const Register rvec = rdi; // r byte array initialized from initvector array address
2428 // and left with the results of the last encryption block
2429 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2430 const Register pos = rax;
2431
2432 // xmm register assignments for the loops below
2433 const XMMRegister xmm_result = xmm0;
2434 const XMMRegister xmm_temp = xmm1;
2435 // first 6 keys preloaded into xmm2-xmm7
2436 const int XMM_REG_NUM_KEY_FIRST = 2;
2437 const int XMM_REG_NUM_KEY_LAST = 7;
2438 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2439
2440 __ enter(); // required for proper stackwalking of RuntimeStub frame
2441 handleSOERegisters(true /*saving*/);
2442
2443 // load registers from incoming parameters
2444 const Address from_param(rbp, 8+0);
2445 const Address to_param (rbp, 8+4);
2446 const Address key_param (rbp, 8+8);
2447 const Address rvec_param (rbp, 8+12);
2448 const Address len_param (rbp, 8+16);
2449 __ movptr(from , from_param);
2450 __ movptr(to , to_param);
2451 __ movptr(key , key_param);
2452 __ movptr(rvec , rvec_param);
2453 __ movptr(len_reg , len_param);
2454
2455 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2456 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2457 // load up xmm regs 2 thru 7 with keys 0-5
2458 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2459 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2460 offset += 0x10;
2461 }
2462
2463 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2464
2465 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2466 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2467 __ cmpl(rax, 44);
2468 __ jcc(Assembler::notEqual, L_key_192_256);
2469
2470 // 128 bit code follows here
2471 __ movl(pos, 0);
2472 __ align(OptoLoopAlignment);
2473 __ BIND(L_loopTop_128);
2474 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2475 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2476
2477 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2478 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2479 __ aesenc(xmm_result, as_XMMRegister(rnum));
2480 }
2481 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2482 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2483 }
2484 load_key(xmm_temp, key, 0xa0);
2485 __ aesenclast(xmm_result, xmm_temp);
2486
2487 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2488 // no need to store r to memory until we exit
2489 __ addptr(pos, AESBlockSize);
2490 __ subptr(len_reg, AESBlockSize);
2491 __ jcc(Assembler::notEqual, L_loopTop_128);
2492
2493 __ BIND(L_exit);
2494 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2495
2496 handleSOERegisters(false /*restoring*/);
2497 __ movptr(rax, len_param); // return length
2498 __ leave(); // required for proper stackwalking of RuntimeStub frame
2499 __ ret(0);
2500
2501 __ BIND(L_key_192_256);
2502 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2503 __ cmpl(rax, 52);
2504 __ jcc(Assembler::notEqual, L_key_256);
2505
2506 // 192-bit code follows here (could be changed to use more xmm registers)
2507 __ movl(pos, 0);
2508 __ align(OptoLoopAlignment);
2509 __ BIND(L_loopTop_192);
2510 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2511 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2512
2513 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2514 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2515 __ aesenc(xmm_result, as_XMMRegister(rnum));
2516 }
2517 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2518 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2519 }
2520 load_key(xmm_temp, key, 0xc0);
2521 __ aesenclast(xmm_result, xmm_temp);
2522
2523 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2524 // no need to store r to memory until we exit
2525 __ addptr(pos, AESBlockSize);
2526 __ subptr(len_reg, AESBlockSize);
2527 __ jcc(Assembler::notEqual, L_loopTop_192);
2528 __ jmp(L_exit);
2529
2530 __ BIND(L_key_256);
2531 // 256-bit code follows here (could be changed to use more xmm registers)
2532 __ movl(pos, 0);
2533 __ align(OptoLoopAlignment);
2534 __ BIND(L_loopTop_256);
2535 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2536 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2537
2538 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2539 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2540 __ aesenc(xmm_result, as_XMMRegister(rnum));
2541 }
2542 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2543 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2544 }
2545 load_key(xmm_temp, key, 0xe0);
2546 __ aesenclast(xmm_result, xmm_temp);
2547
2548 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2549 // no need to store r to memory until we exit
2550 __ addptr(pos, AESBlockSize);
2551 __ subptr(len_reg, AESBlockSize);
2552 __ jcc(Assembler::notEqual, L_loopTop_256);
2553 __ jmp(L_exit);
2554
2555 return start;
2556 }
2557
2558
2559 // CBC AES Decryption.
2560 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2561 //
2562 // Arguments:
2563 //
2564 // Inputs:
2565 // c_rarg0 - source byte array address
2566 // c_rarg1 - destination byte array address
2567 // c_rarg2 - K (key) in little endian int array
2568 // c_rarg3 - r vector byte array address
2569 // c_rarg4 - input length
2570 //
2571 // Output:
2572 // rax - input length
2573 //
2574
2575 address generate_cipherBlockChaining_decryptAESCrypt() {
2576 assert(UseAES, "need AES instructions and misaligned SSE support");
2577 __ align(CodeEntryAlignment);
2578 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2579 address start = __ pc();
2580
2581 Label L_exit, L_key_192_256, L_key_256;
2582 Label L_singleBlock_loopTop_128;
2583 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2584 const Register from = rsi; // source array address
2585 const Register to = rdx; // destination array address
2586 const Register key = rcx; // key array address
2587 const Register rvec = rdi; // r byte array initialized from initvector array address
2588 // and left with the results of the last encryption block
2589 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2590 const Register pos = rax;
2591
2592 // xmm register assignments for the loops below
2593 const XMMRegister xmm_result = xmm0;
2594 const XMMRegister xmm_temp = xmm1;
2595 // first 6 keys preloaded into xmm2-xmm7
2596 const int XMM_REG_NUM_KEY_FIRST = 2;
2597 const int XMM_REG_NUM_KEY_LAST = 7;
2598 const int FIRST_NON_REG_KEY_offset = 0x70;
2599 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2600
2601 __ enter(); // required for proper stackwalking of RuntimeStub frame
2602 handleSOERegisters(true /*saving*/);
2603
2604 // load registers from incoming parameters
2605 const Address from_param(rbp, 8+0);
2606 const Address to_param (rbp, 8+4);
2607 const Address key_param (rbp, 8+8);
2608 const Address rvec_param (rbp, 8+12);
2609 const Address len_param (rbp, 8+16);
2610 __ movptr(from , from_param);
2611 __ movptr(to , to_param);
2612 __ movptr(key , key_param);
2613 __ movptr(rvec , rvec_param);
2614 __ movptr(len_reg , len_param);
2615
2616 // the java expanded key ordering is rotated one position from what we want
2617 // so we start from 0x10 here and hit 0x00 last
2618 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
2619 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2620 // load up xmm regs 2 thru 6 with first 5 keys
2621 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2622 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2623 offset += 0x10;
2624 }
2625
2626 // inside here, use the rvec register to point to previous block cipher
2627 // with which we xor at the end of each newly decrypted block
2628 const Register prev_block_cipher_ptr = rvec;
2629
2630 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2631 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2632 __ cmpl(rax, 44);
2633 __ jcc(Assembler::notEqual, L_key_192_256);
2634
2635
2636 // 128-bit code follows here, parallelized
2637 __ movl(pos, 0);
2638 __ align(OptoLoopAlignment);
2639 __ BIND(L_singleBlock_loopTop_128);
2640 __ cmpptr(len_reg, 0); // any blocks left??
2641 __ jcc(Assembler::equal, L_exit);
2642 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2643 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2644 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2645 __ aesdec(xmm_result, as_XMMRegister(rnum));
2646 }
2647 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
2648 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2649 }
2650 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2651 __ aesdeclast(xmm_result, xmm_temp);
2652 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2653 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2654 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2655 // no need to store r to memory until we exit
2656 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2657 __ addptr(pos, AESBlockSize);
2658 __ subptr(len_reg, AESBlockSize);
2659 __ jmp(L_singleBlock_loopTop_128);
2660
2661
2662 __ BIND(L_exit);
2663 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2664 __ movptr(rvec , rvec_param); // restore this since used in loop
2665 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
2666 handleSOERegisters(false /*restoring*/);
2667 __ movptr(rax, len_param); // return length
2668 __ leave(); // required for proper stackwalking of RuntimeStub frame
2669 __ ret(0);
2670
2671
2672 __ BIND(L_key_192_256);
2673 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2674 __ cmpl(rax, 52);
2675 __ jcc(Assembler::notEqual, L_key_256);
2676
2677 // 192-bit code follows here (could be optimized to use parallelism)
2678 __ movl(pos, 0);
2679 __ align(OptoLoopAlignment);
2680 __ BIND(L_singleBlock_loopTop_192);
2681 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2682 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2683 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2684 __ aesdec(xmm_result, as_XMMRegister(rnum));
2685 }
2686 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
2687 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2688 }
2689 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2690 __ aesdeclast(xmm_result, xmm_temp);
2691 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2692 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2693 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2694 // no need to store r to memory until we exit
2695 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2696 __ addptr(pos, AESBlockSize);
2697 __ subptr(len_reg, AESBlockSize);
2698 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2699 __ jmp(L_exit);
2700
2701 __ BIND(L_key_256);
2702 // 256-bit code follows here (could be optimized to use parallelism)
2703 __ movl(pos, 0);
2704 __ align(OptoLoopAlignment);
2705 __ BIND(L_singleBlock_loopTop_256);
2706 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2707 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2708 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2709 __ aesdec(xmm_result, as_XMMRegister(rnum));
2710 }
2711 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
2712 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2713 }
2714 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2715 __ aesdeclast(xmm_result, xmm_temp);
2716 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2717 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2718 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2719 // no need to store r to memory until we exit
2720 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2721 __ addptr(pos, AESBlockSize);
2722 __ subptr(len_reg, AESBlockSize);
2723 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2724 __ jmp(L_exit);
2725
2726 return start;
2727 }
2728
2729 /**
2730 * Arguments:
2731 *
2732 * Inputs:
2733 * rsp(4) - int crc
2734 * rsp(8) - byte* buf
2735 * rsp(12) - int length
2736 *
2737 * Ouput:
2738 * rax - int crc result
2739 */
2740 address generate_updateBytesCRC32() {
2741 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions");
2742
2743 __ align(CodeEntryAlignment);
2744 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32");
2745
2746 address start = __ pc();
2747
2748 const Register crc = rdx; // crc
2749 const Register buf = rsi; // source java byte array address
2750 const Register len = rcx; // length
2751 const Register table = rdi; // crc_table address (reuse register)
2752 const Register tmp = rbx;
2753 assert_different_registers(crc, buf, len, table, tmp, rax);
2754
2755 BLOCK_COMMENT("Entry:");
2756 __ enter(); // required for proper stackwalking of RuntimeStub frame
2757 __ push(rsi);
2758 __ push(rdi);
2759 __ push(rbx);
2760
2761 Address crc_arg(rbp, 8 + 0);
2762 Address buf_arg(rbp, 8 + 4);
2763 Address len_arg(rbp, 8 + 8);
2764
2765 // Load up:
2766 __ movl(crc, crc_arg);
2767 __ movptr(buf, buf_arg);
2768 __ movl(len, len_arg);
2769
2770 __ kernel_crc32(crc, buf, len, table, tmp);
2771
2772 __ movl(rax, crc);
2773 __ pop(rbx);
2774 __ pop(rdi);
2775 __ pop(rsi);
2776 __ leave(); // required for proper stackwalking of RuntimeStub frame
2777 __ ret(0);
2778
2779 return start;
2780 }
2781
2782 // Safefetch stubs.
2783 void generate_safefetch(const char* name, int size, address* entry,
2784 address* fault_pc, address* continuation_pc) {
2785 // safefetch signatures:
2786 // int SafeFetch32(int* adr, int errValue);
2787 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue);
2788
2789 StubCodeMark mark(this, "StubRoutines", name);
2790
2791 // Entry point, pc or function descriptor.
2792 *entry = __ pc();
2793
2794 __ movl(rax, Address(rsp, 0x8));
2795 __ movl(rcx, Address(rsp, 0x4));
2796 // Load *adr into eax, may fault.
2797 *fault_pc = __ pc();
2798 switch (size) {
2799 case 4:
2800 // int32_t
2801 __ movl(rax, Address(rcx, 0));
2802 break;
2803 case 8:
2804 // int64_t
2805 Unimplemented();
2806 break;
2807 default:
2808 ShouldNotReachHere();
2809 }
2810
2811 // Return errValue or *adr.
2812 *continuation_pc = __ pc();
2813 __ ret(0);
2814 }
2815
2816 public:
2817 // Information about frame layout at time of blocking runtime call.
2818 // Note that we only have to preserve callee-saved registers since
2819 // the compilers are responsible for supplying a continuation point
2820 // if they expect all registers to be preserved.
2821 enum layout {
2822 thread_off, // last_java_sp
2823 arg1_off,
2824 arg2_off,
2825 rbp_off, // callee saved register
2826 ret_pc,
2827 framesize
2828 };
2829
2830 private:
2831
2832 #undef __
2833 #define __ masm->
2834
2835 //------------------------------------------------------------------------------------------------------------------------
2836 // Continuation point for throwing of implicit exceptions that are not handled in
2837 // the current activation. Fabricates an exception oop and initiates normal
2838 // exception dispatching in this frame.
2839 //
2840 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2841 // This is no longer true after adapter frames were removed but could possibly
2842 // be brought back in the future if the interpreter code was reworked and it
2843 // was deemed worthwhile. The comment below was left to describe what must
2844 // happen here if callee saves were resurrected. As it stands now this stub
2845 // could actually be a vanilla BufferBlob and have now oopMap at all.
2846 // Since it doesn't make much difference we've chosen to leave it the
2847 // way it was in the callee save days and keep the comment.
2848
2849 // If we need to preserve callee-saved values we need a callee-saved oop map and
2850 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2851 // If the compiler needs all registers to be preserved between the fault
2852 // point and the exception handler then it must assume responsibility for that in
2853 // AbstractCompiler::continuation_for_implicit_null_exception or
2854 // continuation_for_implicit_division_by_zero_exception. All other implicit
2855 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2856 // either at call sites or otherwise assume that stack unwinding will be initiated,
2857 // so caller saved registers were assumed volatile in the compiler.
2858 address generate_throw_exception(const char* name, address runtime_entry,
2859 Register arg1 = noreg, Register arg2 = noreg) {
2860
2861 int insts_size = 256;
2862 int locs_size = 32;
2863
2864 CodeBuffer code(name, insts_size, locs_size);
2865 OopMapSet* oop_maps = new OopMapSet();
2866 MacroAssembler* masm = new MacroAssembler(&code);
2867
2868 address start = __ pc();
2869
2870 // This is an inlined and slightly modified version of call_VM
2871 // which has the ability to fetch the return PC out of
2872 // thread-local storage and also sets up last_Java_sp slightly
2873 // differently than the real call_VM
2874 Register java_thread = rbx;
2875 __ get_thread(java_thread);
2876
2877 __ enter(); // required for proper stackwalking of RuntimeStub frame
2878
2879 // pc and rbp, already pushed
2880 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2881
2882 // Frame is now completed as far as size and linkage.
2883
2884 int frame_complete = __ pc() - start;
2885
2886 // push java thread (becomes first argument of C function)
2887 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2888 if (arg1 != noreg) {
2889 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
2890 }
2891 if (arg2 != noreg) {
2892 assert(arg1 != noreg, "missing reg arg");
2893 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
2894 }
2895
2896 // Set up last_Java_sp and last_Java_fp
2897 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2898
2899 // Call runtime
2900 BLOCK_COMMENT("call runtime_entry");
2901 __ call(RuntimeAddress(runtime_entry));
2902 // Generate oop map
2903 OopMap* map = new OopMap(framesize, 0);
2904 oop_maps->add_gc_map(__ pc() - start, map);
2905
2906 // restore the thread (cannot use the pushed argument since arguments
2907 // may be overwritten by C code generated by an optimizing compiler);
2908 // however can use the register value directly if it is callee saved.
2909 __ get_thread(java_thread);
2910
2911 __ reset_last_Java_frame(java_thread, true, false);
2912
2913 __ leave(); // required for proper stackwalking of RuntimeStub frame
2914
2915 // check for pending exceptions
2916 #ifdef ASSERT
2917 Label L;
2918 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2919 __ jcc(Assembler::notEqual, L);
2920 __ should_not_reach_here();
2921 __ bind(L);
2922 #endif /* ASSERT */
2923 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2924
2925
2926 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2927 return stub->entry_point();
2928 }
2929
2930
2931 void create_control_words() {
2932 // Round to nearest, 53-bit mode, exceptions masked
2933 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2934 // Round to zero, 53-bit mode, exception mased
2935 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2936 // Round to nearest, 24-bit mode, exceptions masked
2937 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2938 // Round to nearest, 64-bit mode, exceptions masked
2939 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2940 // Round to nearest, 64-bit mode, exceptions masked
2941 StubRoutines::_mxcsr_std = 0x1F80;
2942 // Note: the following two constants are 80-bit values
2943 // layout is critical for correct loading by FPU.
2944 // Bias for strict fp multiply/divide
2945 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2946 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2947 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2948 // Un-Bias for strict fp multiply/divide
2949 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2950 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2951 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2952 }
2953
2954 //---------------------------------------------------------------------------
2955 // Initialization
2956
2957 void generate_initial() {
2958 // Generates all stubs and initializes the entry points
2959
2960 //------------------------------------------------------------------------------------------------------------------------
2961 // entry points that exist in all platforms
2962 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2963 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2964 StubRoutines::_forward_exception_entry = generate_forward_exception();
2965
2966 StubRoutines::_call_stub_entry =
2967 generate_call_stub(StubRoutines::_call_stub_return_address);
2968 // is referenced by megamorphic call
2969 StubRoutines::_catch_exception_entry = generate_catch_exception();
2970
2971 // These are currently used by Solaris/Intel
2972 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2973
2974 StubRoutines::_handler_for_unsafe_access_entry =
2975 generate_handler_for_unsafe_access();
2976
2977 // platform dependent
2978 create_control_words();
2979
2980 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2981 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2982 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2983 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2984 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2985 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2986
2987 // Build this early so it's available for the interpreter
2988 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
2989
2990 if (UseCRC32Intrinsics) {
2991 // set table address before stub generation which use it
2992 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table;
2993 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32();
2994 }
2995 }
2996
2997
2998 void generate_all() {
2999 // Generates all stubs and initializes the entry points
3000
3001 // These entry points require SharedInfo::stack0 to be set up in non-core builds
3002 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
3003 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
3004 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
3005 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
3006
3007 //------------------------------------------------------------------------------------------------------------------------
3008 // entry points that are platform specific
3009
3010 // support for verify_oop (must happen after universe_init)
3011 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
3012
3013 // arraycopy stubs used by compilers
3014 generate_arraycopy_stubs();
3015
3016 generate_math_stubs();
3017
3018 // don't bother generating these AES intrinsic stubs unless global flag is set
3019 if (UseAESIntrinsics) {
3020 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
3021
3022 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
3023 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
3024 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
3025 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
3026 }
3027
3028 // Safefetch stubs.
3029 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry,
3030 &StubRoutines::_safefetch32_fault_pc,
3031 &StubRoutines::_safefetch32_continuation_pc);
3032 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry;
3033 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc;
3034 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc;
3035 }
3036
3037
3038 public:
3039 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
3040 if (all) {
3041 generate_all();
3042 } else {
3043 generate_initial();
3044 }
3045 }
3046 }; // end class declaration
3047
3048
3049 void StubGenerator_generate(CodeBuffer* code, bool all) {
3050 StubGenerator g(code, all);
3051 }