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