1 /*
2 * Copyright (c) 1999, 2011, 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/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/methodOop.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 "utilities/top.hpp"
41 #ifdef TARGET_OS_FAMILY_linux
42 # include "thread_linux.inline.hpp"
43 #endif
44 #ifdef TARGET_OS_FAMILY_solaris
45 # include "thread_solaris.inline.hpp"
46 #endif
47 #ifdef TARGET_OS_FAMILY_windows
48 # include "thread_windows.inline.hpp"
49 #endif
50 #ifdef TARGET_OS_FAMILY_bsd
51 # include "thread_bsd.inline.hpp"
52 #endif
53 #ifdef COMPILER2
54 #include "opto/runtime.hpp"
55 #endif
56
57 // Declaration and definition of StubGenerator (no .hpp file).
58 // For a more detailed description of the stub routine structure
59 // see the comment in stubRoutines.hpp
60
61 #define __ _masm->
62 #define a__ ((Assembler*)_masm)->
63
64 #ifdef PRODUCT
65 #define BLOCK_COMMENT(str) /* nothing */
66 #else
67 #define BLOCK_COMMENT(str) __ block_comment(str)
68 #endif
69
70 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
71
72 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
73 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
74
75 // -------------------------------------------------------------------------------------------------------------------------
76 // Stub Code definitions
77
78 static address handle_unsafe_access() {
79 JavaThread* thread = JavaThread::current();
80 address pc = thread->saved_exception_pc();
81 // pc is the instruction which we must emulate
82 // doing a no-op is fine: return garbage from the load
83 // therefore, compute npc
84 address npc = Assembler::locate_next_instruction(pc);
85
86 // request an async exception
87 thread->set_pending_unsafe_access_error();
88
89 // return address of next instruction to execute
90 return npc;
91 }
92
93 class StubGenerator: public StubCodeGenerator {
94 private:
95
96 #ifdef PRODUCT
97 #define inc_counter_np(counter) (0)
98 #else
99 void inc_counter_np_(int& counter) {
100 __ incrementl(ExternalAddress((address)&counter));
101 }
102 #define inc_counter_np(counter) \
103 BLOCK_COMMENT("inc_counter " #counter); \
104 inc_counter_np_(counter);
105 #endif //PRODUCT
106
107 void inc_copy_counter_np(BasicType t) {
108 #ifndef PRODUCT
109 switch (t) {
110 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
111 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
112 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
113 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
114 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
115 }
116 ShouldNotReachHere();
117 #endif //PRODUCT
118 }
119
120 //------------------------------------------------------------------------------------------------------------------------
121 // Call stubs are used to call Java from C
122 //
123 // [ return_from_Java ] <--- rsp
124 // [ argument word n ]
125 // ...
126 // -N [ argument word 1 ]
127 // -7 [ Possible padding for stack alignment ]
128 // -6 [ Possible padding for stack alignment ]
129 // -5 [ Possible padding for stack alignment ]
130 // -4 [ mxcsr save ] <--- rsp_after_call
131 // -3 [ saved rbx, ]
132 // -2 [ saved rsi ]
133 // -1 [ saved rdi ]
134 // 0 [ saved rbp, ] <--- rbp,
135 // 1 [ return address ]
136 // 2 [ ptr. to call wrapper ]
137 // 3 [ result ]
138 // 4 [ result_type ]
139 // 5 [ method ]
140 // 6 [ entry_point ]
141 // 7 [ parameters ]
142 // 8 [ parameter_size ]
143 // 9 [ thread ]
144
145
146 address generate_call_stub(address& return_address) {
147 StubCodeMark mark(this, "StubRoutines", "call_stub");
148 address start = __ pc();
149
150 // stub code parameters / addresses
151 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
152 bool sse_save = false;
153 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
154 const int locals_count_in_bytes (4*wordSize);
155 const Address mxcsr_save (rbp, -4 * wordSize);
156 const Address saved_rbx (rbp, -3 * wordSize);
157 const Address saved_rsi (rbp, -2 * wordSize);
158 const Address saved_rdi (rbp, -1 * wordSize);
159 const Address result (rbp, 3 * wordSize);
160 const Address result_type (rbp, 4 * wordSize);
161 const Address method (rbp, 5 * wordSize);
162 const Address entry_point (rbp, 6 * wordSize);
163 const Address parameters (rbp, 7 * wordSize);
164 const Address parameter_size(rbp, 8 * wordSize);
165 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
166 sse_save = UseSSE > 0;
167
168 // stub code
169 __ enter();
170 __ movptr(rcx, parameter_size); // parameter counter
171 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
172 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
173 __ subptr(rsp, rcx);
174 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
175
176 // save rdi, rsi, & rbx, according to C calling conventions
177 __ movptr(saved_rdi, rdi);
178 __ movptr(saved_rsi, rsi);
179 __ movptr(saved_rbx, rbx);
180 // save and initialize %mxcsr
181 if (sse_save) {
182 Label skip_ldmx;
183 __ stmxcsr(mxcsr_save);
184 __ movl(rax, mxcsr_save);
185 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
186 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
187 __ cmp32(rax, mxcsr_std);
188 __ jcc(Assembler::equal, skip_ldmx);
189 __ ldmxcsr(mxcsr_std);
190 __ bind(skip_ldmx);
191 }
192
193 // make sure the control word is correct.
194 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
195
196 #ifdef ASSERT
197 // make sure we have no pending exceptions
198 { Label L;
199 __ movptr(rcx, thread);
200 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
201 __ jcc(Assembler::equal, L);
202 __ stop("StubRoutines::call_stub: entered with pending exception");
203 __ bind(L);
204 }
205 #endif
206
207 // pass parameters if any
208 BLOCK_COMMENT("pass parameters if any");
209 Label parameters_done;
210 __ movl(rcx, parameter_size); // parameter counter
211 __ testl(rcx, rcx);
212 __ jcc(Assembler::zero, parameters_done);
213
214 // parameter passing loop
215
216 Label loop;
217 // Copy Java parameters in reverse order (receiver last)
218 // Note that the argument order is inverted in the process
219 // source is rdx[rcx: N-1..0]
220 // dest is rsp[rbx: 0..N-1]
221
222 __ movptr(rdx, parameters); // parameter pointer
223 __ xorptr(rbx, rbx);
224
225 __ BIND(loop);
226
227 // get parameter
228 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
229 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
230 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
231 __ increment(rbx);
232 __ decrement(rcx);
233 __ jcc(Assembler::notZero, loop);
234
235 // call Java function
236 __ BIND(parameters_done);
237 __ movptr(rbx, method); // get methodOop
238 __ movptr(rax, entry_point); // get entry_point
239 __ mov(rsi, rsp); // set sender sp
240 BLOCK_COMMENT("call Java function");
241 __ call(rax);
242
243 BLOCK_COMMENT("call_stub_return_address:");
244 return_address = __ pc();
245
246 #ifdef COMPILER2
247 {
248 Label L_skip;
249 if (UseSSE >= 2) {
250 __ verify_FPU(0, "call_stub_return");
251 } else {
252 for (int i = 1; i < 8; i++) {
253 __ ffree(i);
254 }
255
256 // UseSSE <= 1 so double result should be left on TOS
257 __ movl(rsi, result_type);
258 __ cmpl(rsi, T_DOUBLE);
259 __ jcc(Assembler::equal, L_skip);
260 if (UseSSE == 0) {
261 // UseSSE == 0 so float result should be left on TOS
262 __ cmpl(rsi, T_FLOAT);
263 __ jcc(Assembler::equal, L_skip);
264 }
265 __ ffree(0);
266 }
267 __ BIND(L_skip);
268 }
269 #endif // COMPILER2
270
271 // store result depending on type
272 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
273 __ movptr(rdi, result);
274 Label is_long, is_float, is_double, exit;
275 __ movl(rsi, result_type);
276 __ cmpl(rsi, T_LONG);
277 __ jcc(Assembler::equal, is_long);
278 __ cmpl(rsi, T_FLOAT);
279 __ jcc(Assembler::equal, is_float);
280 __ cmpl(rsi, T_DOUBLE);
281 __ jcc(Assembler::equal, is_double);
282
283 // handle T_INT case
284 __ movl(Address(rdi, 0), rax);
285 __ BIND(exit);
286
287 // check that FPU stack is empty
288 __ verify_FPU(0, "generate_call_stub");
289
290 // pop parameters
291 __ lea(rsp, rsp_after_call);
292
293 // restore %mxcsr
294 if (sse_save) {
295 __ ldmxcsr(mxcsr_save);
296 }
297
298 // restore rdi, rsi and rbx,
299 __ movptr(rbx, saved_rbx);
300 __ movptr(rsi, saved_rsi);
301 __ movptr(rdi, saved_rdi);
302 __ addptr(rsp, 4*wordSize);
303
304 // return
305 __ pop(rbp);
306 __ ret(0);
307
308 // handle return types different from T_INT
309 __ BIND(is_long);
310 __ movl(Address(rdi, 0 * wordSize), rax);
311 __ movl(Address(rdi, 1 * wordSize), rdx);
312 __ jmp(exit);
313
314 __ BIND(is_float);
315 // interpreter uses xmm0 for return values
316 if (UseSSE >= 1) {
317 __ movflt(Address(rdi, 0), xmm0);
318 } else {
319 __ fstp_s(Address(rdi, 0));
320 }
321 __ jmp(exit);
322
323 __ BIND(is_double);
324 // interpreter uses xmm0 for return values
325 if (UseSSE >= 2) {
326 __ movdbl(Address(rdi, 0), xmm0);
327 } else {
328 __ fstp_d(Address(rdi, 0));
329 }
330 __ jmp(exit);
331
332 return start;
333 }
334
335
336 //------------------------------------------------------------------------------------------------------------------------
337 // Return point for a Java call if there's an exception thrown in Java code.
338 // The exception is caught and transformed into a pending exception stored in
339 // JavaThread that can be tested from within the VM.
340 //
341 // Note: Usually the parameters are removed by the callee. In case of an exception
342 // crossing an activation frame boundary, that is not the case if the callee
343 // is compiled code => need to setup the rsp.
344 //
345 // rax,: exception oop
346
347 address generate_catch_exception() {
348 StubCodeMark mark(this, "StubRoutines", "catch_exception");
349 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
350 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
351 address start = __ pc();
352
353 // get thread directly
354 __ movptr(rcx, thread);
355 #ifdef ASSERT
356 // verify that threads correspond
357 { Label L;
358 __ get_thread(rbx);
359 __ cmpptr(rbx, rcx);
360 __ jcc(Assembler::equal, L);
361 __ stop("StubRoutines::catch_exception: threads must correspond");
362 __ bind(L);
363 }
364 #endif
365 // set pending exception
366 __ verify_oop(rax);
367 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
368 __ lea(Address(rcx, Thread::exception_file_offset ()),
369 ExternalAddress((address)__FILE__));
370 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
371 // complete return to VM
372 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
373 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
374
375 return start;
376 }
377
378
379 //------------------------------------------------------------------------------------------------------------------------
380 // Continuation point for runtime calls returning with a pending exception.
381 // The pending exception check happened in the runtime or native call stub.
382 // The pending exception in Thread is converted into a Java-level exception.
383 //
384 // Contract with Java-level exception handlers:
385 // rax: exception
386 // rdx: throwing pc
387 //
388 // NOTE: At entry of this stub, exception-pc must be on stack !!
389
390 address generate_forward_exception() {
391 StubCodeMark mark(this, "StubRoutines", "forward exception");
392 address start = __ pc();
393 const Register thread = rcx;
394
395 // other registers used in this stub
396 const Register exception_oop = rax;
397 const Register handler_addr = rbx;
398 const Register exception_pc = rdx;
399
400 // Upon entry, the sp points to the return address returning into Java
401 // (interpreted or compiled) code; i.e., the return address becomes the
402 // throwing pc.
403 //
404 // Arguments pushed before the runtime call are still on the stack but
405 // the exception handler will reset the stack pointer -> ignore them.
406 // A potential result in registers can be ignored as well.
407
408 #ifdef ASSERT
409 // make sure this code is only executed if there is a pending exception
410 { Label L;
411 __ get_thread(thread);
412 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
413 __ jcc(Assembler::notEqual, L);
414 __ stop("StubRoutines::forward exception: no pending exception (1)");
415 __ bind(L);
416 }
417 #endif
418
419 // compute exception handler into rbx,
420 __ get_thread(thread);
421 __ movptr(exception_pc, Address(rsp, 0));
422 BLOCK_COMMENT("call exception_handler_for_return_address");
423 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
424 __ mov(handler_addr, rax);
425
426 // setup rax & rdx, remove return address & clear pending exception
427 __ get_thread(thread);
428 __ pop(exception_pc);
429 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
430 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
431
432 #ifdef ASSERT
433 // make sure exception is set
434 { Label L;
435 __ testptr(exception_oop, exception_oop);
436 __ jcc(Assembler::notEqual, L);
437 __ stop("StubRoutines::forward exception: no pending exception (2)");
438 __ bind(L);
439 }
440 #endif
441
442 // Verify that there is really a valid exception in RAX.
443 __ verify_oop(exception_oop);
444
445 // continue at exception handler (return address removed)
446 // rax: exception
447 // rbx: exception handler
448 // rdx: throwing pc
449 __ jmp(handler_addr);
450
451 return start;
452 }
453
454
455 //----------------------------------------------------------------------------------------------------
456 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
457 //
458 // xchg exists as far back as 8086, lock needed for MP only
459 // Stack layout immediately after call:
460 //
461 // 0 [ret addr ] <--- rsp
462 // 1 [ ex ]
463 // 2 [ dest ]
464 //
465 // Result: *dest <- ex, return (old *dest)
466 //
467 // Note: win32 does not currently use this code
468
469 address generate_atomic_xchg() {
470 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
471 address start = __ pc();
472
473 __ push(rdx);
474 Address exchange(rsp, 2 * wordSize);
475 Address dest_addr(rsp, 3 * wordSize);
476 __ movl(rax, exchange);
477 __ movptr(rdx, dest_addr);
478 __ xchgl(rax, Address(rdx, 0));
479 __ pop(rdx);
480 __ ret(0);
481
482 return start;
483 }
484
485 //----------------------------------------------------------------------------------------------------
486 // Support for void verify_mxcsr()
487 //
488 // This routine is used with -Xcheck:jni to verify that native
489 // JNI code does not return to Java code without restoring the
490 // MXCSR register to our expected state.
491
492
493 address generate_verify_mxcsr() {
494 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
495 address start = __ pc();
496
497 const Address mxcsr_save(rsp, 0);
498
499 if (CheckJNICalls && UseSSE > 0 ) {
500 Label ok_ret;
501 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
502 __ push(rax);
503 __ subptr(rsp, wordSize); // allocate a temp location
504 __ stmxcsr(mxcsr_save);
505 __ movl(rax, mxcsr_save);
506 __ andl(rax, MXCSR_MASK);
507 __ cmp32(rax, mxcsr_std);
508 __ jcc(Assembler::equal, ok_ret);
509
510 __ warn("MXCSR changed by native JNI code.");
511
512 __ ldmxcsr(mxcsr_std);
513
514 __ bind(ok_ret);
515 __ addptr(rsp, wordSize);
516 __ pop(rax);
517 }
518
519 __ ret(0);
520
521 return start;
522 }
523
524
525 //---------------------------------------------------------------------------
526 // Support for void verify_fpu_cntrl_wrd()
527 //
528 // This routine is used with -Xcheck:jni to verify that native
529 // JNI code does not return to Java code without restoring the
530 // FP control word to our expected state.
531
532 address generate_verify_fpu_cntrl_wrd() {
533 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
534 address start = __ pc();
535
536 const Address fpu_cntrl_wrd_save(rsp, 0);
537
538 if (CheckJNICalls) {
539 Label ok_ret;
540 __ push(rax);
541 __ subptr(rsp, wordSize); // allocate a temp location
542 __ fnstcw(fpu_cntrl_wrd_save);
543 __ movl(rax, fpu_cntrl_wrd_save);
544 __ andl(rax, FPU_CNTRL_WRD_MASK);
545 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
546 __ cmp32(rax, fpu_std);
547 __ jcc(Assembler::equal, ok_ret);
548
549 __ warn("Floating point control word changed by native JNI code.");
550
551 __ fldcw(fpu_std);
552
553 __ bind(ok_ret);
554 __ addptr(rsp, wordSize);
555 __ pop(rax);
556 }
557
558 __ ret(0);
559
560 return start;
561 }
562
563 //---------------------------------------------------------------------------
564 // Wrapper for slow-case handling of double-to-integer conversion
565 // d2i or f2i fast case failed either because it is nan or because
566 // of under/overflow.
567 // Input: FPU TOS: float value
568 // Output: rax, (rdx): integer (long) result
569
570 address generate_d2i_wrapper(BasicType t, address fcn) {
571 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
572 address start = __ pc();
573
574 // Capture info about frame layout
575 enum layout { FPUState_off = 0,
576 rbp_off = FPUStateSizeInWords,
577 rdi_off,
578 rsi_off,
579 rcx_off,
580 rbx_off,
581 saved_argument_off,
582 saved_argument_off2, // 2nd half of double
583 framesize
584 };
585
586 assert(FPUStateSizeInWords == 27, "update stack layout");
587
588 // Save outgoing argument to stack across push_FPU_state()
589 __ subptr(rsp, wordSize * 2);
590 __ fstp_d(Address(rsp, 0));
591
592 // Save CPU & FPU state
593 __ push(rbx);
594 __ push(rcx);
595 __ push(rsi);
596 __ push(rdi);
597 __ push(rbp);
598 __ push_FPU_state();
599
600 // push_FPU_state() resets the FP top of stack
601 // Load original double into FP top of stack
602 __ fld_d(Address(rsp, saved_argument_off * wordSize));
603 // Store double into stack as outgoing argument
604 __ subptr(rsp, wordSize*2);
605 __ fst_d(Address(rsp, 0));
606
607 // Prepare FPU for doing math in C-land
608 __ empty_FPU_stack();
609 // Call the C code to massage the double. Result in EAX
610 if (t == T_INT)
611 { BLOCK_COMMENT("SharedRuntime::d2i"); }
612 else if (t == T_LONG)
613 { BLOCK_COMMENT("SharedRuntime::d2l"); }
614 __ call_VM_leaf( fcn, 2 );
615
616 // Restore CPU & FPU state
617 __ pop_FPU_state();
618 __ pop(rbp);
619 __ pop(rdi);
620 __ pop(rsi);
621 __ pop(rcx);
622 __ pop(rbx);
623 __ addptr(rsp, wordSize * 2);
624
625 __ ret(0);
626
627 return start;
628 }
629
630
631 //---------------------------------------------------------------------------
632 // The following routine generates a subroutine to throw an asynchronous
633 // UnknownError when an unsafe access gets a fault that could not be
634 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
635 address generate_handler_for_unsafe_access() {
636 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
637 address start = __ pc();
638
639 __ push(0); // hole for return address-to-be
640 __ pusha(); // push registers
641 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
642 BLOCK_COMMENT("call handle_unsafe_access");
643 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
644 __ movptr(next_pc, rax); // stuff next address
645 __ popa();
646 __ ret(0); // jump to next address
647
648 return start;
649 }
650
651
652 //----------------------------------------------------------------------------------------------------
653 // Non-destructive plausibility checks for oops
654
655 address generate_verify_oop() {
656 StubCodeMark mark(this, "StubRoutines", "verify_oop");
657 address start = __ pc();
658
659 // Incoming arguments on stack after saving rax,:
660 //
661 // [tos ]: saved rdx
662 // [tos + 1]: saved EFLAGS
663 // [tos + 2]: return address
664 // [tos + 3]: char* error message
665 // [tos + 4]: oop object to verify
666 // [tos + 5]: saved rax, - saved by caller and bashed
667
668 Label exit, error;
669 __ pushf();
670 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
671 __ push(rdx); // save rdx
672 // make sure object is 'reasonable'
673 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
674 __ testptr(rax, rax);
675 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
676
677 // Check if the oop is in the right area of memory
678 const int oop_mask = Universe::verify_oop_mask();
679 const int oop_bits = Universe::verify_oop_bits();
680 __ mov(rdx, rax);
681 __ andptr(rdx, oop_mask);
682 __ cmpptr(rdx, oop_bits);
683 __ jcc(Assembler::notZero, error);
684
685 // make sure klass is 'reasonable'
686 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
687 __ testptr(rax, rax);
688 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
689
690 // Check if the klass is in the right area of memory
691 const int klass_mask = Universe::verify_klass_mask();
692 const int klass_bits = Universe::verify_klass_bits();
693 __ mov(rdx, rax);
694 __ andptr(rdx, klass_mask);
695 __ cmpptr(rdx, klass_bits);
696 __ jcc(Assembler::notZero, error);
697
698 // make sure klass' klass is 'reasonable'
699 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
700 __ testptr(rax, rax);
701 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
702
703 __ mov(rdx, rax);
704 __ andptr(rdx, klass_mask);
705 __ cmpptr(rdx, klass_bits);
706 __ jcc(Assembler::notZero, error); // if klass not in right area
707 // of memory it is broken too.
708
709 // return if everything seems ok
710 __ bind(exit);
711 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
712 __ pop(rdx); // restore rdx
713 __ popf(); // restore EFLAGS
714 __ ret(3 * wordSize); // pop arguments
715
716 // handle errors
717 __ bind(error);
718 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
719 __ pop(rdx); // get saved rdx back
720 __ popf(); // get saved EFLAGS off stack -- will be ignored
721 __ pusha(); // push registers (eip = return address & msg are already pushed)
722 BLOCK_COMMENT("call MacroAssembler::debug");
723 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
724 __ popa();
725 __ ret(3 * wordSize); // pop arguments
726 return start;
727 }
728
729 //
730 // Generate pre-barrier for array stores
731 //
732 // Input:
733 // start - starting address
734 // count - element count
735 void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) {
736 assert_different_registers(start, count);
737 BarrierSet* bs = Universe::heap()->barrier_set();
738 switch (bs->kind()) {
739 case BarrierSet::G1SATBCT:
740 case BarrierSet::G1SATBCTLogging:
741 // With G1, don't generate the call if we statically know that the target in uninitialized
742 if (!uninitialized_target) {
743 __ pusha(); // push registers
744 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
745 start, count);
746 __ popa();
747 }
748 break;
749 case BarrierSet::CardTableModRef:
750 case BarrierSet::CardTableExtension:
751 case BarrierSet::ModRef:
752 break;
753 default :
754 ShouldNotReachHere();
755
756 }
757 }
758
759
760 //
761 // Generate a post-barrier for an array store
762 //
763 // start - starting address
764 // count - element count
765 //
766 // The two input registers are overwritten.
767 //
768 void gen_write_ref_array_post_barrier(Register start, Register count) {
769 BarrierSet* bs = Universe::heap()->barrier_set();
770 assert_different_registers(start, count);
771 switch (bs->kind()) {
772 case BarrierSet::G1SATBCT:
773 case BarrierSet::G1SATBCTLogging:
774 {
775 __ pusha(); // push registers
776 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
777 start, count);
778 __ popa();
779 }
780 break;
781
782 case BarrierSet::CardTableModRef:
783 case BarrierSet::CardTableExtension:
784 {
785 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
786 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
787
788 Label L_loop;
789 const Register end = count; // elements count; end == start+count-1
790 assert_different_registers(start, end);
791
792 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
793 __ shrptr(start, CardTableModRefBS::card_shift);
794 __ shrptr(end, CardTableModRefBS::card_shift);
795 __ subptr(end, start); // end --> count
796 __ BIND(L_loop);
797 intptr_t disp = (intptr_t) ct->byte_map_base;
798 Address cardtable(start, count, Address::times_1, disp);
799 __ movb(cardtable, 0);
800 __ decrement(count);
801 __ jcc(Assembler::greaterEqual, L_loop);
802 }
803 break;
804 case BarrierSet::ModRef:
805 break;
806 default :
807 ShouldNotReachHere();
808
809 }
810 }
811
812
813 // Copy 64 bytes chunks
814 //
815 // Inputs:
816 // from - source array address
817 // to_from - destination array address - from
818 // qword_count - 8-bytes element count, negative
819 //
820 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
821 assert( UseSSE >= 2, "supported cpu only" );
822 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
823 // Copy 64-byte chunks
824 __ jmpb(L_copy_64_bytes);
825 __ align(OptoLoopAlignment);
826 __ BIND(L_copy_64_bytes_loop);
827
828 if (UseUnalignedLoadStores) {
829 if (UseAVX >= 2) {
830 __ vmovdqu(xmm0, Address(from, 0));
831 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0);
832 __ vmovdqu(xmm1, Address(from, 32));
833 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1);
834 } else {
835 __ movdqu(xmm0, Address(from, 0));
836 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
837 __ movdqu(xmm1, Address(from, 16));
838 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
839 __ movdqu(xmm2, Address(from, 32));
840 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
841 __ movdqu(xmm3, Address(from, 48));
842 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
843 }
844 } else {
845 __ movq(xmm0, Address(from, 0));
846 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
847 __ movq(xmm1, Address(from, 8));
848 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
849 __ movq(xmm2, Address(from, 16));
850 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
851 __ movq(xmm3, Address(from, 24));
852 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
853 __ movq(xmm4, Address(from, 32));
854 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
855 __ movq(xmm5, Address(from, 40));
856 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
857 __ movq(xmm6, Address(from, 48));
858 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
859 __ movq(xmm7, Address(from, 56));
860 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
861 }
862
863 __ addl(from, 64);
864 __ BIND(L_copy_64_bytes);
865 __ subl(qword_count, 8);
866 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
867
868 if (UseUnalignedLoadStores && (UseAVX >= 2)) {
869 // clean upper bits of YMM registers
870 __ vzeroupper();
871 }
872 __ addl(qword_count, 8);
873 __ jccb(Assembler::zero, L_exit);
874 //
875 // length is too short, just copy qwords
876 //
877 __ BIND(L_copy_8_bytes);
878 __ movq(xmm0, Address(from, 0));
879 __ movq(Address(from, to_from, Address::times_1), xmm0);
880 __ addl(from, 8);
881 __ decrement(qword_count);
882 __ jcc(Assembler::greater, L_copy_8_bytes);
883 __ BIND(L_exit);
884 }
885
886 // Copy 64 bytes chunks
887 //
888 // Inputs:
889 // from - source array address
890 // to_from - destination array address - from
891 // qword_count - 8-bytes element count, negative
892 //
893 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
894 assert( VM_Version::supports_mmx(), "supported cpu only" );
895 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
896 // Copy 64-byte chunks
897 __ jmpb(L_copy_64_bytes);
898 __ align(OptoLoopAlignment);
899 __ BIND(L_copy_64_bytes_loop);
900 __ movq(mmx0, Address(from, 0));
901 __ movq(mmx1, Address(from, 8));
902 __ movq(mmx2, Address(from, 16));
903 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
904 __ movq(mmx3, Address(from, 24));
905 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
906 __ movq(mmx4, Address(from, 32));
907 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
908 __ movq(mmx5, Address(from, 40));
909 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
910 __ movq(mmx6, Address(from, 48));
911 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
912 __ movq(mmx7, Address(from, 56));
913 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
914 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
915 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
916 __ addptr(from, 64);
917 __ BIND(L_copy_64_bytes);
918 __ subl(qword_count, 8);
919 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
920 __ addl(qword_count, 8);
921 __ jccb(Assembler::zero, L_exit);
922 //
923 // length is too short, just copy qwords
924 //
925 __ BIND(L_copy_8_bytes);
926 __ movq(mmx0, Address(from, 0));
927 __ movq(Address(from, to_from, Address::times_1), mmx0);
928 __ addptr(from, 8);
929 __ decrement(qword_count);
930 __ jcc(Assembler::greater, L_copy_8_bytes);
931 __ BIND(L_exit);
932 __ emms();
933 }
934
935 address generate_disjoint_copy(BasicType t, bool aligned,
936 Address::ScaleFactor sf,
937 address* entry, const char *name,
938 bool dest_uninitialized = false) {
939 __ align(CodeEntryAlignment);
940 StubCodeMark mark(this, "StubRoutines", name);
941 address start = __ pc();
942
943 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
944 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
945
946 int shift = Address::times_ptr - sf;
947
948 const Register from = rsi; // source array address
949 const Register to = rdi; // destination array address
950 const Register count = rcx; // elements count
951 const Register to_from = to; // (to - from)
952 const Register saved_to = rdx; // saved destination array address
953
954 __ enter(); // required for proper stackwalking of RuntimeStub frame
955 __ push(rsi);
956 __ push(rdi);
957 __ movptr(from , Address(rsp, 12+ 4));
958 __ movptr(to , Address(rsp, 12+ 8));
959 __ movl(count, Address(rsp, 12+ 12));
960
961 if (entry != NULL) {
962 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
963 BLOCK_COMMENT("Entry:");
964 }
965
966 if (t == T_OBJECT) {
967 __ testl(count, count);
968 __ jcc(Assembler::zero, L_0_count);
969 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
970 __ mov(saved_to, to); // save 'to'
971 }
972
973 __ subptr(to, from); // to --> to_from
974 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
975 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
976 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
977 // align source address at 4 bytes address boundary
978 if (t == T_BYTE) {
979 // One byte misalignment happens only for byte arrays
980 __ testl(from, 1);
981 __ jccb(Assembler::zero, L_skip_align1);
982 __ movb(rax, Address(from, 0));
983 __ movb(Address(from, to_from, Address::times_1, 0), rax);
984 __ increment(from);
985 __ decrement(count);
986 __ BIND(L_skip_align1);
987 }
988 // Two bytes misalignment happens only for byte and short (char) arrays
989 __ testl(from, 2);
990 __ jccb(Assembler::zero, L_skip_align2);
991 __ movw(rax, Address(from, 0));
992 __ movw(Address(from, to_from, Address::times_1, 0), rax);
993 __ addptr(from, 2);
994 __ subl(count, 1<<(shift-1));
995 __ BIND(L_skip_align2);
996 }
997 if (!VM_Version::supports_mmx()) {
998 __ mov(rax, count); // save 'count'
999 __ shrl(count, shift); // bytes count
1000 __ addptr(to_from, from);// restore 'to'
1001 __ rep_mov();
1002 __ subptr(to_from, from);// restore 'to_from'
1003 __ mov(count, rax); // restore 'count'
1004 __ jmpb(L_copy_2_bytes); // all dwords were copied
1005 } else {
1006 if (!UseUnalignedLoadStores) {
1007 // align to 8 bytes, we know we are 4 byte aligned to start
1008 __ testptr(from, 4);
1009 __ jccb(Assembler::zero, L_copy_64_bytes);
1010 __ movl(rax, Address(from, 0));
1011 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1012 __ addptr(from, 4);
1013 __ subl(count, 1<<shift);
1014 }
1015 __ BIND(L_copy_64_bytes);
1016 __ mov(rax, count);
1017 __ shrl(rax, shift+1); // 8 bytes chunk count
1018 //
1019 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
1020 //
1021 if (UseXMMForArrayCopy) {
1022 xmm_copy_forward(from, to_from, rax);
1023 } else {
1024 mmx_copy_forward(from, to_from, rax);
1025 }
1026 }
1027 // copy tailing dword
1028 __ BIND(L_copy_4_bytes);
1029 __ testl(count, 1<<shift);
1030 __ jccb(Assembler::zero, L_copy_2_bytes);
1031 __ movl(rax, Address(from, 0));
1032 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1033 if (t == T_BYTE || t == T_SHORT) {
1034 __ addptr(from, 4);
1035 __ BIND(L_copy_2_bytes);
1036 // copy tailing word
1037 __ testl(count, 1<<(shift-1));
1038 __ jccb(Assembler::zero, L_copy_byte);
1039 __ movw(rax, Address(from, 0));
1040 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1041 if (t == T_BYTE) {
1042 __ addptr(from, 2);
1043 __ BIND(L_copy_byte);
1044 // copy tailing byte
1045 __ testl(count, 1);
1046 __ jccb(Assembler::zero, L_exit);
1047 __ movb(rax, Address(from, 0));
1048 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1049 __ BIND(L_exit);
1050 } else {
1051 __ BIND(L_copy_byte);
1052 }
1053 } else {
1054 __ BIND(L_copy_2_bytes);
1055 }
1056
1057 if (t == T_OBJECT) {
1058 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1059 __ mov(to, saved_to); // restore 'to'
1060 gen_write_ref_array_post_barrier(to, count);
1061 __ BIND(L_0_count);
1062 }
1063 inc_copy_counter_np(t);
1064 __ pop(rdi);
1065 __ pop(rsi);
1066 __ leave(); // required for proper stackwalking of RuntimeStub frame
1067 __ xorptr(rax, rax); // return 0
1068 __ ret(0);
1069 return start;
1070 }
1071
1072
1073 address generate_fill(BasicType t, bool aligned, const char *name) {
1074 __ align(CodeEntryAlignment);
1075 StubCodeMark mark(this, "StubRoutines", name);
1076 address start = __ pc();
1077
1078 BLOCK_COMMENT("Entry:");
1079
1080 const Register to = rdi; // source array address
1081 const Register value = rdx; // value
1082 const Register count = rsi; // elements count
1083
1084 __ enter(); // required for proper stackwalking of RuntimeStub frame
1085 __ push(rsi);
1086 __ push(rdi);
1087 __ movptr(to , Address(rsp, 12+ 4));
1088 __ movl(value, Address(rsp, 12+ 8));
1089 __ movl(count, Address(rsp, 12+ 12));
1090
1091 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1092
1093 __ pop(rdi);
1094 __ pop(rsi);
1095 __ leave(); // required for proper stackwalking of RuntimeStub frame
1096 __ ret(0);
1097 return start;
1098 }
1099
1100 address generate_conjoint_copy(BasicType t, bool aligned,
1101 Address::ScaleFactor sf,
1102 address nooverlap_target,
1103 address* entry, const char *name,
1104 bool dest_uninitialized = false) {
1105 __ align(CodeEntryAlignment);
1106 StubCodeMark mark(this, "StubRoutines", name);
1107 address start = __ pc();
1108
1109 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1110 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1111
1112 int shift = Address::times_ptr - sf;
1113
1114 const Register src = rax; // source array address
1115 const Register dst = rdx; // destination array address
1116 const Register from = rsi; // source array address
1117 const Register to = rdi; // destination array address
1118 const Register count = rcx; // elements count
1119 const Register end = rax; // array end address
1120
1121 __ enter(); // required for proper stackwalking of RuntimeStub frame
1122 __ push(rsi);
1123 __ push(rdi);
1124 __ movptr(src , Address(rsp, 12+ 4)); // from
1125 __ movptr(dst , Address(rsp, 12+ 8)); // to
1126 __ movl2ptr(count, Address(rsp, 12+12)); // count
1127
1128 if (entry != NULL) {
1129 *entry = __ pc(); // Entry point from generic arraycopy stub.
1130 BLOCK_COMMENT("Entry:");
1131 }
1132
1133 // nooverlap_target expects arguments in rsi and rdi.
1134 __ mov(from, src);
1135 __ mov(to , dst);
1136
1137 // arrays overlap test: dispatch to disjoint stub if necessary.
1138 RuntimeAddress nooverlap(nooverlap_target);
1139 __ cmpptr(dst, src);
1140 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1141 __ jump_cc(Assembler::belowEqual, nooverlap);
1142 __ cmpptr(dst, end);
1143 __ jump_cc(Assembler::aboveEqual, nooverlap);
1144
1145 if (t == T_OBJECT) {
1146 __ testl(count, count);
1147 __ jcc(Assembler::zero, L_0_count);
1148 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized);
1149 }
1150
1151 // copy from high to low
1152 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1153 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1154 if (t == T_BYTE || t == T_SHORT) {
1155 // Align the end of destination array at 4 bytes address boundary
1156 __ lea(end, Address(dst, count, sf, 0));
1157 if (t == T_BYTE) {
1158 // One byte misalignment happens only for byte arrays
1159 __ testl(end, 1);
1160 __ jccb(Assembler::zero, L_skip_align1);
1161 __ decrement(count);
1162 __ movb(rdx, Address(from, count, sf, 0));
1163 __ movb(Address(to, count, sf, 0), rdx);
1164 __ BIND(L_skip_align1);
1165 }
1166 // Two bytes misalignment happens only for byte and short (char) arrays
1167 __ testl(end, 2);
1168 __ jccb(Assembler::zero, L_skip_align2);
1169 __ subptr(count, 1<<(shift-1));
1170 __ movw(rdx, Address(from, count, sf, 0));
1171 __ movw(Address(to, count, sf, 0), rdx);
1172 __ BIND(L_skip_align2);
1173 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1174 __ jcc(Assembler::below, L_copy_4_bytes);
1175 }
1176
1177 if (!VM_Version::supports_mmx()) {
1178 __ std();
1179 __ mov(rax, count); // Save 'count'
1180 __ mov(rdx, to); // Save 'to'
1181 __ lea(rsi, Address(from, count, sf, -4));
1182 __ lea(rdi, Address(to , count, sf, -4));
1183 __ shrptr(count, shift); // bytes count
1184 __ rep_mov();
1185 __ cld();
1186 __ mov(count, rax); // restore 'count'
1187 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1188 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1189 __ mov(to, rdx); // restore 'to'
1190 __ jmpb(L_copy_2_bytes); // all dword were copied
1191 } else {
1192 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1193 __ testptr(end, 4);
1194 __ jccb(Assembler::zero, L_copy_8_bytes);
1195 __ subl(count, 1<<shift);
1196 __ movl(rdx, Address(from, count, sf, 0));
1197 __ movl(Address(to, count, sf, 0), rdx);
1198 __ jmpb(L_copy_8_bytes);
1199
1200 __ align(OptoLoopAlignment);
1201 // Move 8 bytes
1202 __ BIND(L_copy_8_bytes_loop);
1203 if (UseXMMForArrayCopy) {
1204 __ movq(xmm0, Address(from, count, sf, 0));
1205 __ movq(Address(to, count, sf, 0), xmm0);
1206 } else {
1207 __ movq(mmx0, Address(from, count, sf, 0));
1208 __ movq(Address(to, count, sf, 0), mmx0);
1209 }
1210 __ BIND(L_copy_8_bytes);
1211 __ subl(count, 2<<shift);
1212 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1213 __ addl(count, 2<<shift);
1214 if (!UseXMMForArrayCopy) {
1215 __ emms();
1216 }
1217 }
1218 __ BIND(L_copy_4_bytes);
1219 // copy prefix qword
1220 __ testl(count, 1<<shift);
1221 __ jccb(Assembler::zero, L_copy_2_bytes);
1222 __ movl(rdx, Address(from, count, sf, -4));
1223 __ movl(Address(to, count, sf, -4), rdx);
1224
1225 if (t == T_BYTE || t == T_SHORT) {
1226 __ subl(count, (1<<shift));
1227 __ BIND(L_copy_2_bytes);
1228 // copy prefix dword
1229 __ testl(count, 1<<(shift-1));
1230 __ jccb(Assembler::zero, L_copy_byte);
1231 __ movw(rdx, Address(from, count, sf, -2));
1232 __ movw(Address(to, count, sf, -2), rdx);
1233 if (t == T_BYTE) {
1234 __ subl(count, 1<<(shift-1));
1235 __ BIND(L_copy_byte);
1236 // copy prefix byte
1237 __ testl(count, 1);
1238 __ jccb(Assembler::zero, L_exit);
1239 __ movb(rdx, Address(from, 0));
1240 __ movb(Address(to, 0), rdx);
1241 __ BIND(L_exit);
1242 } else {
1243 __ BIND(L_copy_byte);
1244 }
1245 } else {
1246 __ BIND(L_copy_2_bytes);
1247 }
1248 if (t == T_OBJECT) {
1249 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1250 gen_write_ref_array_post_barrier(to, count);
1251 __ BIND(L_0_count);
1252 }
1253 inc_copy_counter_np(t);
1254 __ pop(rdi);
1255 __ pop(rsi);
1256 __ leave(); // required for proper stackwalking of RuntimeStub frame
1257 __ xorptr(rax, rax); // return 0
1258 __ ret(0);
1259 return start;
1260 }
1261
1262
1263 address generate_disjoint_long_copy(address* entry, const char *name) {
1264 __ align(CodeEntryAlignment);
1265 StubCodeMark mark(this, "StubRoutines", name);
1266 address start = __ pc();
1267
1268 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1269 const Register from = rax; // source array address
1270 const Register to = rdx; // destination array address
1271 const Register count = rcx; // elements count
1272 const Register to_from = rdx; // (to - from)
1273
1274 __ enter(); // required for proper stackwalking of RuntimeStub frame
1275 __ movptr(from , Address(rsp, 8+0)); // from
1276 __ movptr(to , Address(rsp, 8+4)); // to
1277 __ movl2ptr(count, Address(rsp, 8+8)); // count
1278
1279 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1280 BLOCK_COMMENT("Entry:");
1281
1282 __ subptr(to, from); // to --> to_from
1283 if (VM_Version::supports_mmx()) {
1284 if (UseXMMForArrayCopy) {
1285 xmm_copy_forward(from, to_from, count);
1286 } else {
1287 mmx_copy_forward(from, to_from, count);
1288 }
1289 } else {
1290 __ jmpb(L_copy_8_bytes);
1291 __ align(OptoLoopAlignment);
1292 __ BIND(L_copy_8_bytes_loop);
1293 __ fild_d(Address(from, 0));
1294 __ fistp_d(Address(from, to_from, Address::times_1));
1295 __ addptr(from, 8);
1296 __ BIND(L_copy_8_bytes);
1297 __ decrement(count);
1298 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1299 }
1300 inc_copy_counter_np(T_LONG);
1301 __ leave(); // required for proper stackwalking of RuntimeStub frame
1302 __ xorptr(rax, rax); // return 0
1303 __ ret(0);
1304 return start;
1305 }
1306
1307 address generate_conjoint_long_copy(address nooverlap_target,
1308 address* entry, const char *name) {
1309 __ align(CodeEntryAlignment);
1310 StubCodeMark mark(this, "StubRoutines", name);
1311 address start = __ pc();
1312
1313 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1314 const Register from = rax; // source array address
1315 const Register to = rdx; // destination array address
1316 const Register count = rcx; // elements count
1317 const Register end_from = rax; // source array end address
1318
1319 __ enter(); // required for proper stackwalking of RuntimeStub frame
1320 __ movptr(from , Address(rsp, 8+0)); // from
1321 __ movptr(to , Address(rsp, 8+4)); // to
1322 __ movl2ptr(count, Address(rsp, 8+8)); // count
1323
1324 *entry = __ pc(); // Entry point from generic arraycopy stub.
1325 BLOCK_COMMENT("Entry:");
1326
1327 // arrays overlap test
1328 __ cmpptr(to, from);
1329 RuntimeAddress nooverlap(nooverlap_target);
1330 __ jump_cc(Assembler::belowEqual, nooverlap);
1331 __ lea(end_from, Address(from, count, Address::times_8, 0));
1332 __ cmpptr(to, end_from);
1333 __ movptr(from, Address(rsp, 8)); // from
1334 __ jump_cc(Assembler::aboveEqual, nooverlap);
1335
1336 __ jmpb(L_copy_8_bytes);
1337
1338 __ align(OptoLoopAlignment);
1339 __ BIND(L_copy_8_bytes_loop);
1340 if (VM_Version::supports_mmx()) {
1341 if (UseXMMForArrayCopy) {
1342 __ movq(xmm0, Address(from, count, Address::times_8));
1343 __ movq(Address(to, count, Address::times_8), xmm0);
1344 } else {
1345 __ movq(mmx0, Address(from, count, Address::times_8));
1346 __ movq(Address(to, count, Address::times_8), mmx0);
1347 }
1348 } else {
1349 __ fild_d(Address(from, count, Address::times_8));
1350 __ fistp_d(Address(to, count, Address::times_8));
1351 }
1352 __ BIND(L_copy_8_bytes);
1353 __ decrement(count);
1354 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1355
1356 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1357 __ emms();
1358 }
1359 inc_copy_counter_np(T_LONG);
1360 __ leave(); // required for proper stackwalking of RuntimeStub frame
1361 __ xorptr(rax, rax); // return 0
1362 __ ret(0);
1363 return start;
1364 }
1365
1366
1367 // Helper for generating a dynamic type check.
1368 // The sub_klass must be one of {rbx, rdx, rsi}.
1369 // The temp is killed.
1370 void generate_type_check(Register sub_klass,
1371 Address& super_check_offset_addr,
1372 Address& super_klass_addr,
1373 Register temp,
1374 Label* L_success, Label* L_failure) {
1375 BLOCK_COMMENT("type_check:");
1376
1377 Label L_fallthrough;
1378 #define LOCAL_JCC(assembler_con, label_ptr) \
1379 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1380 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1381
1382 // The following is a strange variation of the fast path which requires
1383 // one less register, because needed values are on the argument stack.
1384 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1385 // L_success, L_failure, NULL);
1386 assert_different_registers(sub_klass, temp);
1387
1388 int sc_offset = in_bytes(Klass::secondary_super_cache_offset());
1389
1390 // if the pointers are equal, we are done (e.g., String[] elements)
1391 __ cmpptr(sub_klass, super_klass_addr);
1392 LOCAL_JCC(Assembler::equal, L_success);
1393
1394 // check the supertype display:
1395 __ movl2ptr(temp, super_check_offset_addr);
1396 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1397 __ movptr(temp, super_check_addr); // load displayed supertype
1398 __ cmpptr(temp, super_klass_addr); // test the super type
1399 LOCAL_JCC(Assembler::equal, L_success);
1400
1401 // if it was a primary super, we can just fail immediately
1402 __ cmpl(super_check_offset_addr, sc_offset);
1403 LOCAL_JCC(Assembler::notEqual, L_failure);
1404
1405 // The repne_scan instruction uses fixed registers, which will get spilled.
1406 // We happen to know this works best when super_klass is in rax.
1407 Register super_klass = temp;
1408 __ movptr(super_klass, super_klass_addr);
1409 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1410 L_success, L_failure);
1411
1412 __ bind(L_fallthrough);
1413
1414 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1415 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1416
1417 #undef LOCAL_JCC
1418 }
1419
1420 //
1421 // Generate checkcasting array copy stub
1422 //
1423 // Input:
1424 // 4(rsp) - source array address
1425 // 8(rsp) - destination array address
1426 // 12(rsp) - element count, can be zero
1427 // 16(rsp) - size_t ckoff (super_check_offset)
1428 // 20(rsp) - oop ckval (super_klass)
1429 //
1430 // Output:
1431 // rax, == 0 - success
1432 // rax, == -1^K - failure, where K is partial transfer count
1433 //
1434 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) {
1435 __ align(CodeEntryAlignment);
1436 StubCodeMark mark(this, "StubRoutines", name);
1437 address start = __ pc();
1438
1439 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1440
1441 // register use:
1442 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1443 // rdi, rsi -- element access (oop, klass)
1444 // rbx, -- temp
1445 const Register from = rax; // source array address
1446 const Register to = rdx; // destination array address
1447 const Register length = rcx; // elements count
1448 const Register elem = rdi; // each oop copied
1449 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1450 const Register temp = rbx; // lone remaining temp
1451
1452 __ enter(); // required for proper stackwalking of RuntimeStub frame
1453
1454 __ push(rsi);
1455 __ push(rdi);
1456 __ push(rbx);
1457
1458 Address from_arg(rsp, 16+ 4); // from
1459 Address to_arg(rsp, 16+ 8); // to
1460 Address length_arg(rsp, 16+12); // elements count
1461 Address ckoff_arg(rsp, 16+16); // super_check_offset
1462 Address ckval_arg(rsp, 16+20); // super_klass
1463
1464 // Load up:
1465 __ movptr(from, from_arg);
1466 __ movptr(to, to_arg);
1467 __ movl2ptr(length, length_arg);
1468
1469 if (entry != NULL) {
1470 *entry = __ pc(); // Entry point from generic arraycopy stub.
1471 BLOCK_COMMENT("Entry:");
1472 }
1473
1474 //---------------------------------------------------------------
1475 // Assembler stub will be used for this call to arraycopy
1476 // if the two arrays are subtypes of Object[] but the
1477 // destination array type is not equal to or a supertype
1478 // of the source type. Each element must be separately
1479 // checked.
1480
1481 // Loop-invariant addresses. They are exclusive end pointers.
1482 Address end_from_addr(from, length, Address::times_ptr, 0);
1483 Address end_to_addr(to, length, Address::times_ptr, 0);
1484
1485 Register end_from = from; // re-use
1486 Register end_to = to; // re-use
1487 Register count = length; // re-use
1488
1489 // Loop-variant addresses. They assume post-incremented count < 0.
1490 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1491 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1492 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1493
1494 // Copy from low to high addresses, indexed from the end of each array.
1495 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized);
1496 __ lea(end_from, end_from_addr);
1497 __ lea(end_to, end_to_addr);
1498 assert(length == count, ""); // else fix next line:
1499 __ negptr(count); // negate and test the length
1500 __ jccb(Assembler::notZero, L_load_element);
1501
1502 // Empty array: Nothing to do.
1503 __ xorptr(rax, rax); // return 0 on (trivial) success
1504 __ jmp(L_done);
1505
1506 // ======== begin loop ========
1507 // (Loop is rotated; its entry is L_load_element.)
1508 // Loop control:
1509 // for (count = -count; count != 0; count++)
1510 // Base pointers src, dst are biased by 8*count,to last element.
1511 __ align(OptoLoopAlignment);
1512
1513 __ BIND(L_store_element);
1514 __ movptr(to_element_addr, elem); // store the oop
1515 __ increment(count); // increment the count toward zero
1516 __ jccb(Assembler::zero, L_do_card_marks);
1517
1518 // ======== loop entry is here ========
1519 __ BIND(L_load_element);
1520 __ movptr(elem, from_element_addr); // load the oop
1521 __ testptr(elem, elem);
1522 __ jccb(Assembler::zero, L_store_element);
1523
1524 // (Could do a trick here: Remember last successful non-null
1525 // element stored and make a quick oop equality check on it.)
1526
1527 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1528 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1529 &L_store_element, NULL);
1530 // (On fall-through, we have failed the element type check.)
1531 // ======== end loop ========
1532
1533 // It was a real error; we must depend on the caller to finish the job.
1534 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1535 // Emit GC store barriers for the oops we have copied (length_arg + count),
1536 // and report their number to the caller.
1537 __ addl(count, length_arg); // transfers = (length - remaining)
1538 __ movl2ptr(rax, count); // save the value
1539 __ notptr(rax); // report (-1^K) to caller
1540 __ movptr(to, to_arg); // reload
1541 assert_different_registers(to, count, rax);
1542 gen_write_ref_array_post_barrier(to, count);
1543 __ jmpb(L_done);
1544
1545 // Come here on success only.
1546 __ BIND(L_do_card_marks);
1547 __ movl2ptr(count, length_arg);
1548 __ movptr(to, to_arg); // reload
1549 gen_write_ref_array_post_barrier(to, count);
1550 __ xorptr(rax, rax); // return 0 on success
1551
1552 // Common exit point (success or failure).
1553 __ BIND(L_done);
1554 __ pop(rbx);
1555 __ pop(rdi);
1556 __ pop(rsi);
1557 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1558 __ leave(); // required for proper stackwalking of RuntimeStub frame
1559 __ ret(0);
1560
1561 return start;
1562 }
1563
1564 //
1565 // Generate 'unsafe' array copy stub
1566 // Though just as safe as the other stubs, it takes an unscaled
1567 // size_t argument instead of an element count.
1568 //
1569 // Input:
1570 // 4(rsp) - source array address
1571 // 8(rsp) - destination array address
1572 // 12(rsp) - byte count, can be zero
1573 //
1574 // Output:
1575 // rax, == 0 - success
1576 // rax, == -1 - need to call System.arraycopy
1577 //
1578 // Examines the alignment of the operands and dispatches
1579 // to a long, int, short, or byte copy loop.
1580 //
1581 address generate_unsafe_copy(const char *name,
1582 address byte_copy_entry,
1583 address short_copy_entry,
1584 address int_copy_entry,
1585 address long_copy_entry) {
1586
1587 Label L_long_aligned, L_int_aligned, L_short_aligned;
1588
1589 __ align(CodeEntryAlignment);
1590 StubCodeMark mark(this, "StubRoutines", name);
1591 address start = __ pc();
1592
1593 const Register from = rax; // source array address
1594 const Register to = rdx; // destination array address
1595 const Register count = rcx; // elements count
1596
1597 __ enter(); // required for proper stackwalking of RuntimeStub frame
1598 __ push(rsi);
1599 __ push(rdi);
1600 Address from_arg(rsp, 12+ 4); // from
1601 Address to_arg(rsp, 12+ 8); // to
1602 Address count_arg(rsp, 12+12); // byte count
1603
1604 // Load up:
1605 __ movptr(from , from_arg);
1606 __ movptr(to , to_arg);
1607 __ movl2ptr(count, count_arg);
1608
1609 // bump this on entry, not on exit:
1610 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1611
1612 const Register bits = rsi;
1613 __ mov(bits, from);
1614 __ orptr(bits, to);
1615 __ orptr(bits, count);
1616
1617 __ testl(bits, BytesPerLong-1);
1618 __ jccb(Assembler::zero, L_long_aligned);
1619
1620 __ testl(bits, BytesPerInt-1);
1621 __ jccb(Assembler::zero, L_int_aligned);
1622
1623 __ testl(bits, BytesPerShort-1);
1624 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1625
1626 __ BIND(L_short_aligned);
1627 __ shrptr(count, LogBytesPerShort); // size => short_count
1628 __ movl(count_arg, count); // update 'count'
1629 __ jump(RuntimeAddress(short_copy_entry));
1630
1631 __ BIND(L_int_aligned);
1632 __ shrptr(count, LogBytesPerInt); // size => int_count
1633 __ movl(count_arg, count); // update 'count'
1634 __ jump(RuntimeAddress(int_copy_entry));
1635
1636 __ BIND(L_long_aligned);
1637 __ shrptr(count, LogBytesPerLong); // size => qword_count
1638 __ movl(count_arg, count); // update 'count'
1639 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1640 __ pop(rsi);
1641 __ jump(RuntimeAddress(long_copy_entry));
1642
1643 return start;
1644 }
1645
1646
1647 // Perform range checks on the proposed arraycopy.
1648 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1649 void arraycopy_range_checks(Register src,
1650 Register src_pos,
1651 Register dst,
1652 Register dst_pos,
1653 Address& length,
1654 Label& L_failed) {
1655 BLOCK_COMMENT("arraycopy_range_checks:");
1656 const Register src_end = src_pos; // source array end position
1657 const Register dst_end = dst_pos; // destination array end position
1658 __ addl(src_end, length); // src_pos + length
1659 __ addl(dst_end, length); // dst_pos + length
1660
1661 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1662 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1663 __ jcc(Assembler::above, L_failed);
1664
1665 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1666 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1667 __ jcc(Assembler::above, L_failed);
1668
1669 BLOCK_COMMENT("arraycopy_range_checks done");
1670 }
1671
1672
1673 //
1674 // Generate generic array copy stubs
1675 //
1676 // Input:
1677 // 4(rsp) - src oop
1678 // 8(rsp) - src_pos
1679 // 12(rsp) - dst oop
1680 // 16(rsp) - dst_pos
1681 // 20(rsp) - element count
1682 //
1683 // Output:
1684 // rax, == 0 - success
1685 // rax, == -1^K - failure, where K is partial transfer count
1686 //
1687 address generate_generic_copy(const char *name,
1688 address entry_jbyte_arraycopy,
1689 address entry_jshort_arraycopy,
1690 address entry_jint_arraycopy,
1691 address entry_oop_arraycopy,
1692 address entry_jlong_arraycopy,
1693 address entry_checkcast_arraycopy) {
1694 Label L_failed, L_failed_0, L_objArray;
1695
1696 { int modulus = CodeEntryAlignment;
1697 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1698 int advance = target - (__ offset() % modulus);
1699 if (advance < 0) advance += modulus;
1700 if (advance > 0) __ nop(advance);
1701 }
1702 StubCodeMark mark(this, "StubRoutines", name);
1703
1704 // Short-hop target to L_failed. Makes for denser prologue code.
1705 __ BIND(L_failed_0);
1706 __ jmp(L_failed);
1707 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1708
1709 __ align(CodeEntryAlignment);
1710 address start = __ pc();
1711
1712 __ enter(); // required for proper stackwalking of RuntimeStub frame
1713 __ push(rsi);
1714 __ push(rdi);
1715
1716 // bump this on entry, not on exit:
1717 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1718
1719 // Input values
1720 Address SRC (rsp, 12+ 4);
1721 Address SRC_POS (rsp, 12+ 8);
1722 Address DST (rsp, 12+12);
1723 Address DST_POS (rsp, 12+16);
1724 Address LENGTH (rsp, 12+20);
1725
1726 //-----------------------------------------------------------------------
1727 // Assembler stub will be used for this call to arraycopy
1728 // if the following conditions are met:
1729 //
1730 // (1) src and dst must not be null.
1731 // (2) src_pos must not be negative.
1732 // (3) dst_pos must not be negative.
1733 // (4) length must not be negative.
1734 // (5) src klass and dst klass should be the same and not NULL.
1735 // (6) src and dst should be arrays.
1736 // (7) src_pos + length must not exceed length of src.
1737 // (8) dst_pos + length must not exceed length of dst.
1738 //
1739
1740 const Register src = rax; // source array oop
1741 const Register src_pos = rsi;
1742 const Register dst = rdx; // destination array oop
1743 const Register dst_pos = rdi;
1744 const Register length = rcx; // transfer count
1745
1746 // if (src == NULL) return -1;
1747 __ movptr(src, SRC); // src oop
1748 __ testptr(src, src);
1749 __ jccb(Assembler::zero, L_failed_0);
1750
1751 // if (src_pos < 0) return -1;
1752 __ movl2ptr(src_pos, SRC_POS); // src_pos
1753 __ testl(src_pos, src_pos);
1754 __ jccb(Assembler::negative, L_failed_0);
1755
1756 // if (dst == NULL) return -1;
1757 __ movptr(dst, DST); // dst oop
1758 __ testptr(dst, dst);
1759 __ jccb(Assembler::zero, L_failed_0);
1760
1761 // if (dst_pos < 0) return -1;
1762 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1763 __ testl(dst_pos, dst_pos);
1764 __ jccb(Assembler::negative, L_failed_0);
1765
1766 // if (length < 0) return -1;
1767 __ movl2ptr(length, LENGTH); // length
1768 __ testl(length, length);
1769 __ jccb(Assembler::negative, L_failed_0);
1770
1771 // if (src->klass() == NULL) return -1;
1772 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1773 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1774 const Register rcx_src_klass = rcx; // array klass
1775 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1776
1777 #ifdef ASSERT
1778 // assert(src->klass() != NULL);
1779 BLOCK_COMMENT("assert klasses not null");
1780 { Label L1, L2;
1781 __ testptr(rcx_src_klass, rcx_src_klass);
1782 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1783 __ bind(L1);
1784 __ stop("broken null klass");
1785 __ bind(L2);
1786 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1787 __ jccb(Assembler::equal, L1); // this would be broken also
1788 BLOCK_COMMENT("assert done");
1789 }
1790 #endif //ASSERT
1791
1792 // Load layout helper (32-bits)
1793 //
1794 // |array_tag| | header_size | element_type | |log2_element_size|
1795 // 32 30 24 16 8 2 0
1796 //
1797 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1798 //
1799
1800 int lh_offset = in_bytes(Klass::layout_helper_offset());
1801 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1802
1803 // Handle objArrays completely differently...
1804 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1805 __ cmpl(src_klass_lh_addr, objArray_lh);
1806 __ jcc(Assembler::equal, L_objArray);
1807
1808 // if (src->klass() != dst->klass()) return -1;
1809 __ cmpptr(rcx_src_klass, dst_klass_addr);
1810 __ jccb(Assembler::notEqual, L_failed_0);
1811
1812 const Register rcx_lh = rcx; // layout helper
1813 assert(rcx_lh == rcx_src_klass, "known alias");
1814 __ movl(rcx_lh, src_klass_lh_addr);
1815
1816 // if (!src->is_Array()) return -1;
1817 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1818 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1819
1820 // At this point, it is known to be a typeArray (array_tag 0x3).
1821 #ifdef ASSERT
1822 { Label L;
1823 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1824 __ jcc(Assembler::greaterEqual, L); // signed cmp
1825 __ stop("must be a primitive array");
1826 __ bind(L);
1827 }
1828 #endif
1829
1830 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1831 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1832
1833 // typeArrayKlass
1834 //
1835 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1836 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1837 //
1838 const Register rsi_offset = rsi; // array offset
1839 const Register src_array = src; // src array offset
1840 const Register dst_array = dst; // dst array offset
1841 const Register rdi_elsize = rdi; // log2 element size
1842
1843 __ mov(rsi_offset, rcx_lh);
1844 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1845 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1846 __ addptr(src_array, rsi_offset); // src array offset
1847 __ addptr(dst_array, rsi_offset); // dst array offset
1848 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1849
1850 // next registers should be set before the jump to corresponding stub
1851 const Register from = src; // source array address
1852 const Register to = dst; // destination array address
1853 const Register count = rcx; // elements count
1854 // some of them should be duplicated on stack
1855 #define FROM Address(rsp, 12+ 4)
1856 #define TO Address(rsp, 12+ 8) // Not used now
1857 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1858
1859 BLOCK_COMMENT("scale indexes to element size");
1860 __ movl2ptr(rsi, SRC_POS); // src_pos
1861 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1862 assert(src_array == from, "");
1863 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1864 __ movl2ptr(rdi, DST_POS); // dst_pos
1865 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1866 assert(dst_array == to, "");
1867 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1868 __ movptr(FROM, from); // src_addr
1869 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1870 __ movl2ptr(count, LENGTH); // elements count
1871
1872 BLOCK_COMMENT("choose copy loop based on element size");
1873 __ cmpl(rdi_elsize, 0);
1874
1875 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1876 __ cmpl(rdi_elsize, LogBytesPerShort);
1877 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1878 __ cmpl(rdi_elsize, LogBytesPerInt);
1879 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1880 #ifdef ASSERT
1881 __ cmpl(rdi_elsize, LogBytesPerLong);
1882 __ jccb(Assembler::notEqual, L_failed);
1883 #endif
1884 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1885 __ pop(rsi);
1886 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1887
1888 __ BIND(L_failed);
1889 __ xorptr(rax, rax);
1890 __ notptr(rax); // return -1
1891 __ pop(rdi);
1892 __ pop(rsi);
1893 __ leave(); // required for proper stackwalking of RuntimeStub frame
1894 __ ret(0);
1895
1896 // objArrayKlass
1897 __ BIND(L_objArray);
1898 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1899
1900 Label L_plain_copy, L_checkcast_copy;
1901 // test array classes for subtyping
1902 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1903 __ jccb(Assembler::notEqual, L_checkcast_copy);
1904
1905 // Identically typed arrays can be copied without element-wise checks.
1906 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1907 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1908
1909 __ BIND(L_plain_copy);
1910 __ movl2ptr(count, LENGTH); // elements count
1911 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1912 __ lea(from, Address(src, src_pos, Address::times_ptr,
1913 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1914 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1915 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1916 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1917 __ movptr(FROM, from); // src_addr
1918 __ movptr(TO, to); // dst_addr
1919 __ movl(COUNT, count); // count
1920 __ jump(RuntimeAddress(entry_oop_arraycopy));
1921
1922 __ BIND(L_checkcast_copy);
1923 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1924 {
1925 // Handy offsets:
1926 int ek_offset = in_bytes(objArrayKlass::element_klass_offset());
1927 int sco_offset = in_bytes(Klass::super_check_offset_offset());
1928
1929 Register rsi_dst_klass = rsi;
1930 Register rdi_temp = rdi;
1931 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1932 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1933 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1934
1935 // Before looking at dst.length, make sure dst is also an objArray.
1936 __ movptr(rsi_dst_klass, dst_klass_addr);
1937 __ cmpl(dst_klass_lh_addr, objArray_lh);
1938 __ jccb(Assembler::notEqual, L_failed);
1939
1940 // It is safe to examine both src.length and dst.length.
1941 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1942 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1943 // (Now src_pos and dst_pos are killed, but not src and dst.)
1944
1945 // We'll need this temp (don't forget to pop it after the type check).
1946 __ push(rbx);
1947 Register rbx_src_klass = rbx;
1948
1949 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1950 __ movptr(rsi_dst_klass, dst_klass_addr);
1951 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1952 Label L_fail_array_check;
1953 generate_type_check(rbx_src_klass,
1954 super_check_offset_addr, dst_klass_addr,
1955 rdi_temp, NULL, &L_fail_array_check);
1956 // (On fall-through, we have passed the array type check.)
1957 __ pop(rbx);
1958 __ jmp(L_plain_copy);
1959
1960 __ BIND(L_fail_array_check);
1961 // Reshuffle arguments so we can call checkcast_arraycopy:
1962
1963 // match initial saves for checkcast_arraycopy
1964 // push(rsi); // already done; see above
1965 // push(rdi); // already done; see above
1966 // push(rbx); // already done; see above
1967
1968 // Marshal outgoing arguments now, freeing registers.
1969 Address from_arg(rsp, 16+ 4); // from
1970 Address to_arg(rsp, 16+ 8); // to
1971 Address length_arg(rsp, 16+12); // elements count
1972 Address ckoff_arg(rsp, 16+16); // super_check_offset
1973 Address ckval_arg(rsp, 16+20); // super_klass
1974
1975 Address SRC_POS_arg(rsp, 16+ 8);
1976 Address DST_POS_arg(rsp, 16+16);
1977 Address LENGTH_arg(rsp, 16+20);
1978 // push rbx, changed the incoming offsets (why not just use rbp,??)
1979 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1980
1981 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1982 __ movl2ptr(length, LENGTH_arg); // reload elements count
1983 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1984 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1985
1986 __ movptr(ckval_arg, rbx); // destination element type
1987 __ movl(rbx, Address(rbx, sco_offset));
1988 __ movl(ckoff_arg, rbx); // corresponding class check offset
1989
1990 __ movl(length_arg, length); // outgoing length argument
1991
1992 __ lea(from, Address(src, src_pos, Address::times_ptr,
1993 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1994 __ movptr(from_arg, from);
1995
1996 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1997 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1998 __ movptr(to_arg, to);
1999 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
2000 }
2001
2002 return start;
2003 }
2004
2005 void generate_arraycopy_stubs() {
2006 address entry;
2007 address entry_jbyte_arraycopy;
2008 address entry_jshort_arraycopy;
2009 address entry_jint_arraycopy;
2010 address entry_oop_arraycopy;
2011 address entry_jlong_arraycopy;
2012 address entry_checkcast_arraycopy;
2013
2014 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
2015 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
2016 "arrayof_jbyte_disjoint_arraycopy");
2017 StubRoutines::_arrayof_jbyte_arraycopy =
2018 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
2019 NULL, "arrayof_jbyte_arraycopy");
2020 StubRoutines::_jbyte_disjoint_arraycopy =
2021 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2022 "jbyte_disjoint_arraycopy");
2023 StubRoutines::_jbyte_arraycopy =
2024 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
2025 &entry_jbyte_arraycopy, "jbyte_arraycopy");
2026
2027 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2028 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2029 "arrayof_jshort_disjoint_arraycopy");
2030 StubRoutines::_arrayof_jshort_arraycopy =
2031 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2032 NULL, "arrayof_jshort_arraycopy");
2033 StubRoutines::_jshort_disjoint_arraycopy =
2034 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2035 "jshort_disjoint_arraycopy");
2036 StubRoutines::_jshort_arraycopy =
2037 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2038 &entry_jshort_arraycopy, "jshort_arraycopy");
2039
2040 // Next arrays are always aligned on 4 bytes at least.
2041 StubRoutines::_jint_disjoint_arraycopy =
2042 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2043 "jint_disjoint_arraycopy");
2044 StubRoutines::_jint_arraycopy =
2045 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2046 &entry_jint_arraycopy, "jint_arraycopy");
2047
2048 StubRoutines::_oop_disjoint_arraycopy =
2049 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2050 "oop_disjoint_arraycopy");
2051 StubRoutines::_oop_arraycopy =
2052 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2053 &entry_oop_arraycopy, "oop_arraycopy");
2054
2055 StubRoutines::_oop_disjoint_arraycopy_uninit =
2056 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2057 "oop_disjoint_arraycopy_uninit",
2058 /*dest_uninitialized*/true);
2059 StubRoutines::_oop_arraycopy_uninit =
2060 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2061 NULL, "oop_arraycopy_uninit",
2062 /*dest_uninitialized*/true);
2063
2064 StubRoutines::_jlong_disjoint_arraycopy =
2065 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2066 StubRoutines::_jlong_arraycopy =
2067 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2068 "jlong_arraycopy");
2069
2070 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2071 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2072 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2073 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2074 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2075 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2076
2077 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy;
2078 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy;
2079 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit;
2080 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy;
2081
2082 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2083 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2084 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit;
2085 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2086
2087 StubRoutines::_checkcast_arraycopy =
2088 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy);
2089 StubRoutines::_checkcast_arraycopy_uninit =
2090 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true);
2091
2092 StubRoutines::_unsafe_arraycopy =
2093 generate_unsafe_copy("unsafe_arraycopy",
2094 entry_jbyte_arraycopy,
2095 entry_jshort_arraycopy,
2096 entry_jint_arraycopy,
2097 entry_jlong_arraycopy);
2098
2099 StubRoutines::_generic_arraycopy =
2100 generate_generic_copy("generic_arraycopy",
2101 entry_jbyte_arraycopy,
2102 entry_jshort_arraycopy,
2103 entry_jint_arraycopy,
2104 entry_oop_arraycopy,
2105 entry_jlong_arraycopy,
2106 entry_checkcast_arraycopy);
2107 }
2108
2109 void generate_math_stubs() {
2110 {
2111 StubCodeMark mark(this, "StubRoutines", "log");
2112 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2113
2114 __ fld_d(Address(rsp, 4));
2115 __ flog();
2116 __ ret(0);
2117 }
2118 {
2119 StubCodeMark mark(this, "StubRoutines", "log10");
2120 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2121
2122 __ fld_d(Address(rsp, 4));
2123 __ flog10();
2124 __ ret(0);
2125 }
2126 {
2127 StubCodeMark mark(this, "StubRoutines", "sin");
2128 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2129
2130 __ fld_d(Address(rsp, 4));
2131 __ trigfunc('s');
2132 __ ret(0);
2133 }
2134 {
2135 StubCodeMark mark(this, "StubRoutines", "cos");
2136 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2137
2138 __ fld_d(Address(rsp, 4));
2139 __ trigfunc('c');
2140 __ ret(0);
2141 }
2142 {
2143 StubCodeMark mark(this, "StubRoutines", "tan");
2144 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2145
2146 __ fld_d(Address(rsp, 4));
2147 __ trigfunc('t');
2148 __ ret(0);
2149 }
2150 {
2151 StubCodeMark mark(this, "StubRoutines", "exp");
2152 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc();
2153
2154 __ fld_d(Address(rsp, 4));
2155 __ exp_with_fallback(0);
2156 __ ret(0);
2157 }
2158 {
2159 StubCodeMark mark(this, "StubRoutines", "pow");
2160 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc();
2161
2162 __ fld_d(Address(rsp, 12));
2163 __ fld_d(Address(rsp, 4));
2164 __ pow_with_fallback(0);
2165 __ ret(0);
2166 }
2167 }
2168
2169 // AES intrinsic stubs
2170 enum {AESBlockSize = 16};
2171
2172 address generate_key_shuffle_mask() {
2173 __ align(16);
2174 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask");
2175 address start = __ pc();
2176 __ emit_data(0x00010203, relocInfo::none, 0 );
2177 __ emit_data(0x04050607, relocInfo::none, 0 );
2178 __ emit_data(0x08090a0b, relocInfo::none, 0 );
2179 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 );
2180 return start;
2181 }
2182
2183 // Utility routine for loading a 128-bit key word in little endian format
2184 // can optionally specify that the shuffle mask is already in an xmmregister
2185 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2186 __ movdqu(xmmdst, Address(key, offset));
2187 if (xmm_shuf_mask != NULL) {
2188 __ pshufb(xmmdst, xmm_shuf_mask);
2189 } else {
2190 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2191 }
2192 }
2193
2194 // aesenc using specified key+offset
2195 // can optionally specify that the shuffle mask is already in an xmmregister
2196 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2197 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2198 __ aesenc(xmmdst, xmmtmp);
2199 }
2200
2201 // aesdec using specified key+offset
2202 // can optionally specify that the shuffle mask is already in an xmmregister
2203 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) {
2204 load_key(xmmtmp, key, offset, xmm_shuf_mask);
2205 __ aesdec(xmmdst, xmmtmp);
2206 }
2207
2208
2209 // Arguments:
2210 //
2211 // Inputs:
2212 // c_rarg0 - source byte array address
2213 // c_rarg1 - destination byte array address
2214 // c_rarg2 - K (key) in little endian int array
2215 //
2216 address generate_aescrypt_encryptBlock() {
2217 assert(UseAES, "need AES instructions and misaligned SSE support");
2218 __ align(CodeEntryAlignment);
2219 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock");
2220 Label L_doLast;
2221 address start = __ pc();
2222
2223 const Register from = rdx; // source array address
2224 const Register to = rdx; // destination array address
2225 const Register key = rcx; // key array address
2226 const Register keylen = rax;
2227 const Address from_param(rbp, 8+0);
2228 const Address to_param (rbp, 8+4);
2229 const Address key_param (rbp, 8+8);
2230
2231 const XMMRegister xmm_result = xmm0;
2232 const XMMRegister xmm_key_shuf_mask = xmm1;
2233 const XMMRegister xmm_temp1 = xmm2;
2234 const XMMRegister xmm_temp2 = xmm3;
2235 const XMMRegister xmm_temp3 = xmm4;
2236 const XMMRegister xmm_temp4 = xmm5;
2237
2238 __ enter(); // required for proper stackwalking of RuntimeStub frame
2239 __ movptr(from, from_param);
2240 __ movptr(key, key_param);
2241
2242 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2243 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2244
2245 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2246 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input
2247 __ movptr(to, to_param);
2248
2249 // For encryption, the java expanded key ordering is just what we need
2250
2251 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask);
2252 __ pxor(xmm_result, xmm_temp1);
2253
2254 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2255 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2256 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2257 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2258
2259 __ aesenc(xmm_result, xmm_temp1);
2260 __ aesenc(xmm_result, xmm_temp2);
2261 __ aesenc(xmm_result, xmm_temp3);
2262 __ aesenc(xmm_result, xmm_temp4);
2263
2264 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2265 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2266 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2267 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2268
2269 __ aesenc(xmm_result, xmm_temp1);
2270 __ aesenc(xmm_result, xmm_temp2);
2271 __ aesenc(xmm_result, xmm_temp3);
2272 __ aesenc(xmm_result, xmm_temp4);
2273
2274 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2275 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2276
2277 __ cmpl(keylen, 44);
2278 __ jccb(Assembler::equal, L_doLast);
2279
2280 __ aesenc(xmm_result, xmm_temp1);
2281 __ aesenc(xmm_result, xmm_temp2);
2282
2283 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2284 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2285
2286 __ cmpl(keylen, 52);
2287 __ jccb(Assembler::equal, L_doLast);
2288
2289 __ aesenc(xmm_result, xmm_temp1);
2290 __ aesenc(xmm_result, xmm_temp2);
2291
2292 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2293 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2294
2295 __ BIND(L_doLast);
2296 __ aesenc(xmm_result, xmm_temp1);
2297 __ aesenclast(xmm_result, xmm_temp2);
2298 __ movdqu(Address(to, 0), xmm_result); // store the result
2299 __ xorptr(rax, rax); // return 0
2300 __ leave(); // required for proper stackwalking of RuntimeStub frame
2301 __ ret(0);
2302
2303 return start;
2304 }
2305
2306
2307 // Arguments:
2308 //
2309 // Inputs:
2310 // c_rarg0 - source byte array address
2311 // c_rarg1 - destination byte array address
2312 // c_rarg2 - K (key) in little endian int array
2313 //
2314 address generate_aescrypt_decryptBlock() {
2315 assert(UseAES, "need AES instructions and misaligned SSE support");
2316 __ align(CodeEntryAlignment);
2317 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock");
2318 Label L_doLast;
2319 address start = __ pc();
2320
2321 const Register from = rdx; // source array address
2322 const Register to = rdx; // destination array address
2323 const Register key = rcx; // key array address
2324 const Register keylen = rax;
2325 const Address from_param(rbp, 8+0);
2326 const Address to_param (rbp, 8+4);
2327 const Address key_param (rbp, 8+8);
2328
2329 const XMMRegister xmm_result = xmm0;
2330 const XMMRegister xmm_key_shuf_mask = xmm1;
2331 const XMMRegister xmm_temp1 = xmm2;
2332 const XMMRegister xmm_temp2 = xmm3;
2333 const XMMRegister xmm_temp3 = xmm4;
2334 const XMMRegister xmm_temp4 = xmm5;
2335
2336 __ enter(); // required for proper stackwalking of RuntimeStub frame
2337 __ movptr(from, from_param);
2338 __ movptr(key, key_param);
2339
2340 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60}
2341 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2342
2343 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2344 __ movdqu(xmm_result, Address(from, 0));
2345 __ movptr(to, to_param);
2346
2347 // for decryption java expanded key ordering is rotated one position from what we want
2348 // so we start from 0x10 here and hit 0x00 last
2349 // we don't know if the key is aligned, hence not using load-execute form
2350 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask);
2351 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask);
2352 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask);
2353 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask);
2354
2355 __ pxor (xmm_result, xmm_temp1);
2356 __ aesdec(xmm_result, xmm_temp2);
2357 __ aesdec(xmm_result, xmm_temp3);
2358 __ aesdec(xmm_result, xmm_temp4);
2359
2360 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask);
2361 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask);
2362 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask);
2363 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask);
2364
2365 __ aesdec(xmm_result, xmm_temp1);
2366 __ aesdec(xmm_result, xmm_temp2);
2367 __ aesdec(xmm_result, xmm_temp3);
2368 __ aesdec(xmm_result, xmm_temp4);
2369
2370 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask);
2371 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask);
2372 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask);
2373
2374 __ cmpl(keylen, 44);
2375 __ jccb(Assembler::equal, L_doLast);
2376
2377 __ aesdec(xmm_result, xmm_temp1);
2378 __ aesdec(xmm_result, xmm_temp2);
2379
2380 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask);
2381 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask);
2382
2383 __ cmpl(keylen, 52);
2384 __ jccb(Assembler::equal, L_doLast);
2385
2386 __ aesdec(xmm_result, xmm_temp1);
2387 __ aesdec(xmm_result, xmm_temp2);
2388
2389 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask);
2390 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask);
2391
2392 __ BIND(L_doLast);
2393 __ aesdec(xmm_result, xmm_temp1);
2394 __ aesdec(xmm_result, xmm_temp2);
2395
2396 // for decryption the aesdeclast operation is always on key+0x00
2397 __ aesdeclast(xmm_result, xmm_temp3);
2398 __ movdqu(Address(to, 0), xmm_result); // store the result
2399 __ xorptr(rax, rax); // return 0
2400 __ leave(); // required for proper stackwalking of RuntimeStub frame
2401 __ ret(0);
2402
2403 return start;
2404 }
2405
2406 void handleSOERegisters(bool saving) {
2407 const int saveFrameSizeInBytes = 4 * wordSize;
2408 const Address saved_rbx (rbp, -3 * wordSize);
2409 const Address saved_rsi (rbp, -2 * wordSize);
2410 const Address saved_rdi (rbp, -1 * wordSize);
2411
2412 if (saving) {
2413 __ subptr(rsp, saveFrameSizeInBytes);
2414 __ movptr(saved_rsi, rsi);
2415 __ movptr(saved_rdi, rdi);
2416 __ movptr(saved_rbx, rbx);
2417 } else {
2418 // restoring
2419 __ movptr(rsi, saved_rsi);
2420 __ movptr(rdi, saved_rdi);
2421 __ movptr(rbx, saved_rbx);
2422 }
2423 }
2424
2425 // Arguments:
2426 //
2427 // Inputs:
2428 // c_rarg0 - source byte array address
2429 // c_rarg1 - destination byte array address
2430 // c_rarg2 - K (key) in little endian int array
2431 // c_rarg3 - r vector byte array address
2432 // c_rarg4 - input length
2433 //
2434 address generate_cipherBlockChaining_encryptAESCrypt() {
2435 assert(UseAES, "need AES instructions and misaligned SSE support");
2436 __ align(CodeEntryAlignment);
2437 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt");
2438 address start = __ pc();
2439
2440 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256;
2441 const Register from = rsi; // source array address
2442 const Register to = rdx; // destination array address
2443 const Register key = rcx; // key array address
2444 const Register rvec = rdi; // r byte array initialized from initvector array address
2445 // and left with the results of the last encryption block
2446 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2447 const Register pos = rax;
2448
2449 // xmm register assignments for the loops below
2450 const XMMRegister xmm_result = xmm0;
2451 const XMMRegister xmm_temp = xmm1;
2452 // first 6 keys preloaded into xmm2-xmm7
2453 const int XMM_REG_NUM_KEY_FIRST = 2;
2454 const int XMM_REG_NUM_KEY_LAST = 7;
2455 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2456
2457 __ enter(); // required for proper stackwalking of RuntimeStub frame
2458 handleSOERegisters(true /*saving*/);
2459
2460 // load registers from incoming parameters
2461 const Address from_param(rbp, 8+0);
2462 const Address to_param (rbp, 8+4);
2463 const Address key_param (rbp, 8+8);
2464 const Address rvec_param (rbp, 8+12);
2465 const Address len_param (rbp, 8+16);
2466 __ movptr(from , from_param);
2467 __ movptr(to , to_param);
2468 __ movptr(key , key_param);
2469 __ movptr(rvec , rvec_param);
2470 __ movptr(len_reg , len_param);
2471
2472 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front
2473 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2474 // load up xmm regs 2 thru 7 with keys 0-5
2475 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2476 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2477 offset += 0x10;
2478 }
2479
2480 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec
2481
2482 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2483 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2484 __ cmpl(rax, 44);
2485 __ jcc(Assembler::notEqual, L_key_192_256);
2486
2487 // 128 bit code follows here
2488 __ movl(pos, 0);
2489 __ align(OptoLoopAlignment);
2490 __ BIND(L_loopTop_128);
2491 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2492 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2493
2494 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2495 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2496 __ aesenc(xmm_result, as_XMMRegister(rnum));
2497 }
2498 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) {
2499 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2500 }
2501 load_key(xmm_temp, key, 0xa0);
2502 __ aesenclast(xmm_result, xmm_temp);
2503
2504 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2505 // no need to store r to memory until we exit
2506 __ addptr(pos, AESBlockSize);
2507 __ subptr(len_reg, AESBlockSize);
2508 __ jcc(Assembler::notEqual, L_loopTop_128);
2509
2510 __ BIND(L_exit);
2511 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object
2512
2513 handleSOERegisters(false /*restoring*/);
2514 __ movl(rax, 0); // return 0 (why?)
2515 __ leave(); // required for proper stackwalking of RuntimeStub frame
2516 __ ret(0);
2517
2518 __ BIND(L_key_192_256);
2519 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2520 __ cmpl(rax, 52);
2521 __ jcc(Assembler::notEqual, L_key_256);
2522
2523 // 192-bit code follows here (could be changed to use more xmm registers)
2524 __ movl(pos, 0);
2525 __ align(OptoLoopAlignment);
2526 __ BIND(L_loopTop_192);
2527 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2528 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2529
2530 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2531 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2532 __ aesenc(xmm_result, as_XMMRegister(rnum));
2533 }
2534 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) {
2535 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2536 }
2537 load_key(xmm_temp, key, 0xc0);
2538 __ aesenclast(xmm_result, xmm_temp);
2539
2540 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2541 // no need to store r to memory until we exit
2542 __ addptr(pos, AESBlockSize);
2543 __ subptr(len_reg, AESBlockSize);
2544 __ jcc(Assembler::notEqual, L_loopTop_192);
2545 __ jmp(L_exit);
2546
2547 __ BIND(L_key_256);
2548 // 256-bit code follows here (could be changed to use more xmm registers)
2549 __ movl(pos, 0);
2550 __ align(OptoLoopAlignment);
2551 __ BIND(L_loopTop_256);
2552 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input
2553 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2554
2555 __ pxor (xmm_result, xmm_key0); // do the aes rounds
2556 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2557 __ aesenc(xmm_result, as_XMMRegister(rnum));
2558 }
2559 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) {
2560 aes_enc_key(xmm_result, xmm_temp, key, key_offset);
2561 }
2562 load_key(xmm_temp, key, 0xe0);
2563 __ aesenclast(xmm_result, xmm_temp);
2564
2565 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2566 // no need to store r to memory until we exit
2567 __ addptr(pos, AESBlockSize);
2568 __ subptr(len_reg, AESBlockSize);
2569 __ jcc(Assembler::notEqual, L_loopTop_256);
2570 __ jmp(L_exit);
2571
2572 return start;
2573 }
2574
2575
2576 // CBC AES Decryption.
2577 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time.
2578 //
2579 // Arguments:
2580 //
2581 // Inputs:
2582 // c_rarg0 - source byte array address
2583 // c_rarg1 - destination byte array address
2584 // c_rarg2 - K (key) in little endian int array
2585 // c_rarg3 - r vector byte array address
2586 // c_rarg4 - input length
2587 //
2588
2589 address generate_cipherBlockChaining_decryptAESCrypt() {
2590 assert(UseAES, "need AES instructions and misaligned SSE support");
2591 __ align(CodeEntryAlignment);
2592 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt");
2593 address start = __ pc();
2594
2595 Label L_exit, L_key_192_256, L_key_256;
2596 Label L_singleBlock_loopTop_128;
2597 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256;
2598 const Register from = rsi; // source array address
2599 const Register to = rdx; // destination array address
2600 const Register key = rcx; // key array address
2601 const Register rvec = rdi; // r byte array initialized from initvector array address
2602 // and left with the results of the last encryption block
2603 const Register len_reg = rbx; // src len (must be multiple of blocksize 16)
2604 const Register pos = rax;
2605
2606 // xmm register assignments for the loops below
2607 const XMMRegister xmm_result = xmm0;
2608 const XMMRegister xmm_temp = xmm1;
2609 // first 6 keys preloaded into xmm2-xmm7
2610 const int XMM_REG_NUM_KEY_FIRST = 2;
2611 const int XMM_REG_NUM_KEY_LAST = 7;
2612 const int FIRST_NON_REG_KEY_offset = 0x70;
2613 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST);
2614
2615 __ enter(); // required for proper stackwalking of RuntimeStub frame
2616 handleSOERegisters(true /*saving*/);
2617
2618 // load registers from incoming parameters
2619 const Address from_param(rbp, 8+0);
2620 const Address to_param (rbp, 8+4);
2621 const Address key_param (rbp, 8+8);
2622 const Address rvec_param (rbp, 8+12);
2623 const Address len_param (rbp, 8+16);
2624 __ movptr(from , from_param);
2625 __ movptr(to , to_param);
2626 __ movptr(key , key_param);
2627 __ movptr(rvec , rvec_param);
2628 __ movptr(len_reg , len_param);
2629
2630 // the java expanded key ordering is rotated one position from what we want
2631 // so we start from 0x10 here and hit 0x00 last
2632 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front
2633 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr()));
2634 // load up xmm regs 2 thru 6 with first 5 keys
2635 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2636 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask);
2637 offset += 0x10;
2638 }
2639
2640 // inside here, use the rvec register to point to previous block cipher
2641 // with which we xor at the end of each newly decrypted block
2642 const Register prev_block_cipher_ptr = rvec;
2643
2644 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256))
2645 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT)));
2646 __ cmpl(rax, 44);
2647 __ jcc(Assembler::notEqual, L_key_192_256);
2648
2649
2650 // 128-bit code follows here, parallelized
2651 __ movl(pos, 0);
2652 __ align(OptoLoopAlignment);
2653 __ BIND(L_singleBlock_loopTop_128);
2654 __ cmpptr(len_reg, 0); // any blocks left??
2655 __ jcc(Assembler::equal, L_exit);
2656 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2657 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2658 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2659 __ aesdec(xmm_result, as_XMMRegister(rnum));
2660 }
2661 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0
2662 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2663 }
2664 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2665 __ aesdeclast(xmm_result, xmm_temp);
2666 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2667 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2668 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2669 // no need to store r to memory until we exit
2670 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2671 __ addptr(pos, AESBlockSize);
2672 __ subptr(len_reg, AESBlockSize);
2673 __ jmp(L_singleBlock_loopTop_128);
2674
2675
2676 __ BIND(L_exit);
2677 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2678 __ movptr(rvec , rvec_param); // restore this since used in loop
2679 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object
2680 handleSOERegisters(false /*restoring*/);
2681 __ movl(rax, 0); // return 0 (why?)
2682 __ leave(); // required for proper stackwalking of RuntimeStub frame
2683 __ ret(0);
2684
2685
2686 __ BIND(L_key_192_256);
2687 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256)
2688 __ cmpl(rax, 52);
2689 __ jcc(Assembler::notEqual, L_key_256);
2690
2691 // 192-bit code follows here (could be optimized to use parallelism)
2692 __ movl(pos, 0);
2693 __ align(OptoLoopAlignment);
2694 __ BIND(L_singleBlock_loopTop_192);
2695 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2696 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2697 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2698 __ aesdec(xmm_result, as_XMMRegister(rnum));
2699 }
2700 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0
2701 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2702 }
2703 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2704 __ aesdeclast(xmm_result, xmm_temp);
2705 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2706 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2707 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2708 // no need to store r to memory until we exit
2709 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2710 __ addptr(pos, AESBlockSize);
2711 __ subptr(len_reg, AESBlockSize);
2712 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192);
2713 __ jmp(L_exit);
2714
2715 __ BIND(L_key_256);
2716 // 256-bit code follows here (could be optimized to use parallelism)
2717 __ movl(pos, 0);
2718 __ align(OptoLoopAlignment);
2719 __ BIND(L_singleBlock_loopTop_256);
2720 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input
2721 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds
2722 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) {
2723 __ aesdec(xmm_result, as_XMMRegister(rnum));
2724 }
2725 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0
2726 aes_dec_key(xmm_result, xmm_temp, key, key_offset);
2727 }
2728 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0
2729 __ aesdeclast(xmm_result, xmm_temp);
2730 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00));
2731 __ pxor (xmm_result, xmm_temp); // xor with the current r vector
2732 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output
2733 // no need to store r to memory until we exit
2734 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr
2735 __ addptr(pos, AESBlockSize);
2736 __ subptr(len_reg, AESBlockSize);
2737 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256);
2738 __ jmp(L_exit);
2739
2740 return start;
2741 }
2742
2743
2744 public:
2745 // Information about frame layout at time of blocking runtime call.
2746 // Note that we only have to preserve callee-saved registers since
2747 // the compilers are responsible for supplying a continuation point
2748 // if they expect all registers to be preserved.
2749 enum layout {
2750 thread_off, // last_java_sp
2751 arg1_off,
2752 arg2_off,
2753 rbp_off, // callee saved register
2754 ret_pc,
2755 framesize
2756 };
2757
2758 private:
2759
2760 #undef __
2761 #define __ masm->
2762
2763 //------------------------------------------------------------------------------------------------------------------------
2764 // Continuation point for throwing of implicit exceptions that are not handled in
2765 // the current activation. Fabricates an exception oop and initiates normal
2766 // exception dispatching in this frame.
2767 //
2768 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2769 // This is no longer true after adapter frames were removed but could possibly
2770 // be brought back in the future if the interpreter code was reworked and it
2771 // was deemed worthwhile. The comment below was left to describe what must
2772 // happen here if callee saves were resurrected. As it stands now this stub
2773 // could actually be a vanilla BufferBlob and have now oopMap at all.
2774 // Since it doesn't make much difference we've chosen to leave it the
2775 // way it was in the callee save days and keep the comment.
2776
2777 // If we need to preserve callee-saved values we need a callee-saved oop map and
2778 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2779 // If the compiler needs all registers to be preserved between the fault
2780 // point and the exception handler then it must assume responsibility for that in
2781 // AbstractCompiler::continuation_for_implicit_null_exception or
2782 // continuation_for_implicit_division_by_zero_exception. All other implicit
2783 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2784 // either at call sites or otherwise assume that stack unwinding will be initiated,
2785 // so caller saved registers were assumed volatile in the compiler.
2786 address generate_throw_exception(const char* name, address runtime_entry,
2787 Register arg1 = noreg, Register arg2 = noreg) {
2788
2789 int insts_size = 256;
2790 int locs_size = 32;
2791
2792 CodeBuffer code(name, insts_size, locs_size);
2793 OopMapSet* oop_maps = new OopMapSet();
2794 MacroAssembler* masm = new MacroAssembler(&code);
2795
2796 address start = __ pc();
2797
2798 // This is an inlined and slightly modified version of call_VM
2799 // which has the ability to fetch the return PC out of
2800 // thread-local storage and also sets up last_Java_sp slightly
2801 // differently than the real call_VM
2802 Register java_thread = rbx;
2803 __ get_thread(java_thread);
2804
2805 __ enter(); // required for proper stackwalking of RuntimeStub frame
2806
2807 // pc and rbp, already pushed
2808 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2809
2810 // Frame is now completed as far as size and linkage.
2811
2812 int frame_complete = __ pc() - start;
2813
2814 // push java thread (becomes first argument of C function)
2815 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2816 if (arg1 != noreg) {
2817 __ movptr(Address(rsp, arg1_off * wordSize), arg1);
2818 }
2819 if (arg2 != noreg) {
2820 assert(arg1 != noreg, "missing reg arg");
2821 __ movptr(Address(rsp, arg2_off * wordSize), arg2);
2822 }
2823
2824 // Set up last_Java_sp and last_Java_fp
2825 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2826
2827 // Call runtime
2828 BLOCK_COMMENT("call runtime_entry");
2829 __ call(RuntimeAddress(runtime_entry));
2830 // Generate oop map
2831 OopMap* map = new OopMap(framesize, 0);
2832 oop_maps->add_gc_map(__ pc() - start, map);
2833
2834 // restore the thread (cannot use the pushed argument since arguments
2835 // may be overwritten by C code generated by an optimizing compiler);
2836 // however can use the register value directly if it is callee saved.
2837 __ get_thread(java_thread);
2838
2839 __ reset_last_Java_frame(java_thread, true, false);
2840
2841 __ leave(); // required for proper stackwalking of RuntimeStub frame
2842
2843 // check for pending exceptions
2844 #ifdef ASSERT
2845 Label L;
2846 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2847 __ jcc(Assembler::notEqual, L);
2848 __ should_not_reach_here();
2849 __ bind(L);
2850 #endif /* ASSERT */
2851 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2852
2853
2854 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2855 return stub->entry_point();
2856 }
2857
2858
2859 void create_control_words() {
2860 // Round to nearest, 53-bit mode, exceptions masked
2861 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2862 // Round to zero, 53-bit mode, exception mased
2863 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2864 // Round to nearest, 24-bit mode, exceptions masked
2865 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2866 // Round to nearest, 64-bit mode, exceptions masked
2867 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2868 // Round to nearest, 64-bit mode, exceptions masked
2869 StubRoutines::_mxcsr_std = 0x1F80;
2870 // Note: the following two constants are 80-bit values
2871 // layout is critical for correct loading by FPU.
2872 // Bias for strict fp multiply/divide
2873 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2874 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2875 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2876 // Un-Bias for strict fp multiply/divide
2877 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2878 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2879 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2880 }
2881
2882 //---------------------------------------------------------------------------
2883 // Initialization
2884
2885 void generate_initial() {
2886 // Generates all stubs and initializes the entry points
2887
2888 //------------------------------------------------------------------------------------------------------------------------
2889 // entry points that exist in all platforms
2890 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2891 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2892 StubRoutines::_forward_exception_entry = generate_forward_exception();
2893
2894 StubRoutines::_call_stub_entry =
2895 generate_call_stub(StubRoutines::_call_stub_return_address);
2896 // is referenced by megamorphic call
2897 StubRoutines::_catch_exception_entry = generate_catch_exception();
2898
2899 // These are currently used by Solaris/Intel
2900 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2901
2902 StubRoutines::_handler_for_unsafe_access_entry =
2903 generate_handler_for_unsafe_access();
2904
2905 // platform dependent
2906 create_control_words();
2907
2908 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2909 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2910 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2911 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2912 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2913 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2914
2915 // Build this early so it's available for the interpreter
2916 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError));
2917 }
2918
2919
2920 void generate_all() {
2921 // Generates all stubs and initializes the entry points
2922
2923 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2924 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2925 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError));
2926 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError));
2927 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call));
2928
2929 //------------------------------------------------------------------------------------------------------------------------
2930 // entry points that are platform specific
2931
2932 // support for verify_oop (must happen after universe_init)
2933 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2934
2935 // arraycopy stubs used by compilers
2936 generate_arraycopy_stubs();
2937
2938 generate_math_stubs();
2939
2940 // don't bother generating these AES intrinsic stubs unless global flag is set
2941 if (UseAESIntrinsics) {
2942 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others
2943
2944 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock();
2945 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock();
2946 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt();
2947 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt();
2948 }
2949 }
2950
2951
2952 public:
2953 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2954 if (all) {
2955 generate_all();
2956 } else {
2957 generate_initial();
2958 }
2959 }
2960 }; // end class declaration
2961
2962
2963 void StubGenerator_generate(CodeBuffer* code, bool all) {
2964 StubGenerator g(code, all);
2965 }