1 /*
2 * Copyright 1999-2010 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25 #include "incls/_precompiled.incl"
26 #include "incls/_stubGenerator_x86_32.cpp.incl"
27
28 // Declaration and definition of StubGenerator (no .hpp file).
29 // For a more detailed description of the stub routine structure
30 // see the comment in stubRoutines.hpp
31
32 #define __ _masm->
33 #define a__ ((Assembler*)_masm)->
34
35 #ifdef PRODUCT
36 #define BLOCK_COMMENT(str) /* nothing */
37 #else
38 #define BLOCK_COMMENT(str) __ block_comment(str)
39 #endif
40
41 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
42
43 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
44 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
45
46 // -------------------------------------------------------------------------------------------------------------------------
47 // Stub Code definitions
48
49 static address handle_unsafe_access() {
50 JavaThread* thread = JavaThread::current();
51 address pc = thread->saved_exception_pc();
52 // pc is the instruction which we must emulate
53 // doing a no-op is fine: return garbage from the load
54 // therefore, compute npc
55 address npc = Assembler::locate_next_instruction(pc);
56
57 // request an async exception
58 thread->set_pending_unsafe_access_error();
59
60 // return address of next instruction to execute
61 return npc;
62 }
63
64 class StubGenerator: public StubCodeGenerator {
65 private:
66
67 #ifdef PRODUCT
68 #define inc_counter_np(counter) (0)
69 #else
70 void inc_counter_np_(int& counter) {
71 __ incrementl(ExternalAddress((address)&counter));
72 }
73 #define inc_counter_np(counter) \
74 BLOCK_COMMENT("inc_counter " #counter); \
75 inc_counter_np_(counter);
76 #endif //PRODUCT
77
78 void inc_copy_counter_np(BasicType t) {
79 #ifndef PRODUCT
80 switch (t) {
81 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
82 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
83 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
84 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
85 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
86 }
87 ShouldNotReachHere();
88 #endif //PRODUCT
89 }
90
91 //------------------------------------------------------------------------------------------------------------------------
92 // Call stubs are used to call Java from C
93 //
94 // [ return_from_Java ] <--- rsp
95 // [ argument word n ]
96 // ...
97 // -N [ argument word 1 ]
98 // -7 [ Possible padding for stack alignment ]
99 // -6 [ Possible padding for stack alignment ]
100 // -5 [ Possible padding for stack alignment ]
101 // -4 [ mxcsr save ] <--- rsp_after_call
102 // -3 [ saved rbx, ]
103 // -2 [ saved rsi ]
104 // -1 [ saved rdi ]
105 // 0 [ saved rbp, ] <--- rbp,
106 // 1 [ return address ]
107 // 2 [ ptr. to call wrapper ]
108 // 3 [ result ]
109 // 4 [ result_type ]
110 // 5 [ method ]
111 // 6 [ entry_point ]
112 // 7 [ parameters ]
113 // 8 [ parameter_size ]
114 // 9 [ thread ]
115
116
117 address generate_call_stub(address& return_address) {
118 StubCodeMark mark(this, "StubRoutines", "call_stub");
119 address start = __ pc();
120
121 // stub code parameters / addresses
122 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
123 bool sse_save = false;
124 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
125 const int locals_count_in_bytes (4*wordSize);
126 const Address mxcsr_save (rbp, -4 * wordSize);
127 const Address saved_rbx (rbp, -3 * wordSize);
128 const Address saved_rsi (rbp, -2 * wordSize);
129 const Address saved_rdi (rbp, -1 * wordSize);
130 const Address result (rbp, 3 * wordSize);
131 const Address result_type (rbp, 4 * wordSize);
132 const Address method (rbp, 5 * wordSize);
133 const Address entry_point (rbp, 6 * wordSize);
134 const Address parameters (rbp, 7 * wordSize);
135 const Address parameter_size(rbp, 8 * wordSize);
136 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
137 sse_save = UseSSE > 0;
138
139 // stub code
140 __ enter();
141 __ movptr(rcx, parameter_size); // parameter counter
142 __ shlptr(rcx, Interpreter::logStackElementSize()); // convert parameter count to bytes
143 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
144 __ subptr(rsp, rcx);
145 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
146
147 // save rdi, rsi, & rbx, according to C calling conventions
148 __ movptr(saved_rdi, rdi);
149 __ movptr(saved_rsi, rsi);
150 __ movptr(saved_rbx, rbx);
151 // save and initialize %mxcsr
152 if (sse_save) {
153 Label skip_ldmx;
154 __ stmxcsr(mxcsr_save);
155 __ movl(rax, mxcsr_save);
156 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
157 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
158 __ cmp32(rax, mxcsr_std);
159 __ jcc(Assembler::equal, skip_ldmx);
160 __ ldmxcsr(mxcsr_std);
161 __ bind(skip_ldmx);
162 }
163
164 // make sure the control word is correct.
165 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
166
167 #ifdef ASSERT
168 // make sure we have no pending exceptions
169 { Label L;
170 __ movptr(rcx, thread);
171 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
172 __ jcc(Assembler::equal, L);
173 __ stop("StubRoutines::call_stub: entered with pending exception");
174 __ bind(L);
175 }
176 #endif
177
178 // pass parameters if any
179 BLOCK_COMMENT("pass parameters if any");
180 Label parameters_done;
181 __ movl(rcx, parameter_size); // parameter counter
182 __ testl(rcx, rcx);
183 __ jcc(Assembler::zero, parameters_done);
184
185 // parameter passing loop
186
187 Label loop;
188 // Copy Java parameters in reverse order (receiver last)
189 // Note that the argument order is inverted in the process
190 // source is rdx[rcx: N-1..0]
191 // dest is rsp[rbx: 0..N-1]
192
193 __ movptr(rdx, parameters); // parameter pointer
194 __ xorptr(rbx, rbx);
195
196 __ BIND(loop);
197 if (TaggedStackInterpreter) {
198 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(),
199 -2*wordSize)); // get tag
200 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
201 Interpreter::expr_tag_offset_in_bytes(0)), rax); // store tag
202 }
203
204 // get parameter
205 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
206 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
207 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
208 __ increment(rbx);
209 __ decrement(rcx);
210 __ jcc(Assembler::notZero, loop);
211
212 // call Java function
213 __ BIND(parameters_done);
214 __ movptr(rbx, method); // get methodOop
215 __ movptr(rax, entry_point); // get entry_point
216 __ mov(rsi, rsp); // set sender sp
217 BLOCK_COMMENT("call Java function");
218 __ call(rax);
219
220 BLOCK_COMMENT("call_stub_return_address:");
221 return_address = __ pc();
222
223 Label common_return;
224
225 __ BIND(common_return);
226
227 // store result depending on type
228 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
229 __ movptr(rdi, result);
230 Label is_long, is_float, is_double, exit;
231 __ movl(rsi, result_type);
232 __ cmpl(rsi, T_LONG);
233 __ jcc(Assembler::equal, is_long);
234 __ cmpl(rsi, T_FLOAT);
235 __ jcc(Assembler::equal, is_float);
236 __ cmpl(rsi, T_DOUBLE);
237 __ jcc(Assembler::equal, is_double);
238
239 // handle T_INT case
240 __ movl(Address(rdi, 0), rax);
241 __ BIND(exit);
242
243 // check that FPU stack is empty
244 __ verify_FPU(0, "generate_call_stub");
245
246 // pop parameters
247 __ lea(rsp, rsp_after_call);
248
249 // restore %mxcsr
250 if (sse_save) {
251 __ ldmxcsr(mxcsr_save);
252 }
253
254 // restore rdi, rsi and rbx,
255 __ movptr(rbx, saved_rbx);
256 __ movptr(rsi, saved_rsi);
257 __ movptr(rdi, saved_rdi);
258 __ addptr(rsp, 4*wordSize);
259
260 // return
261 __ pop(rbp);
262 __ ret(0);
263
264 // handle return types different from T_INT
265 __ BIND(is_long);
266 __ movl(Address(rdi, 0 * wordSize), rax);
267 __ movl(Address(rdi, 1 * wordSize), rdx);
268 __ jmp(exit);
269
270 __ BIND(is_float);
271 // interpreter uses xmm0 for return values
272 if (UseSSE >= 1) {
273 __ movflt(Address(rdi, 0), xmm0);
274 } else {
275 __ fstp_s(Address(rdi, 0));
276 }
277 __ jmp(exit);
278
279 __ BIND(is_double);
280 // interpreter uses xmm0 for return values
281 if (UseSSE >= 2) {
282 __ movdbl(Address(rdi, 0), xmm0);
283 } else {
284 __ fstp_d(Address(rdi, 0));
285 }
286 __ jmp(exit);
287
288 // If we call compiled code directly from the call stub we will
289 // need to adjust the return back to the call stub to a specialized
290 // piece of code that can handle compiled results and cleaning the fpu
291 // stack. compiled code will be set to return here instead of the
292 // return above that handles interpreter returns.
293
294 BLOCK_COMMENT("call_stub_compiled_return:");
295 StubRoutines::x86::set_call_stub_compiled_return( __ pc());
296
297 #ifdef COMPILER2
298 if (UseSSE >= 2) {
299 __ verify_FPU(0, "call_stub_compiled_return");
300 } else {
301 for (int i = 1; i < 8; i++) {
302 __ ffree(i);
303 }
304
305 // UseSSE <= 1 so double result should be left on TOS
306 __ movl(rsi, result_type);
307 __ cmpl(rsi, T_DOUBLE);
308 __ jcc(Assembler::equal, common_return);
309 if (UseSSE == 0) {
310 // UseSSE == 0 so float result should be left on TOS
311 __ cmpl(rsi, T_FLOAT);
312 __ jcc(Assembler::equal, common_return);
313 }
314 __ ffree(0);
315 }
316 #endif /* COMPILER2 */
317 __ jmp(common_return);
318
319 return start;
320 }
321
322
323 //------------------------------------------------------------------------------------------------------------------------
324 // Return point for a Java call if there's an exception thrown in Java code.
325 // The exception is caught and transformed into a pending exception stored in
326 // JavaThread that can be tested from within the VM.
327 //
328 // Note: Usually the parameters are removed by the callee. In case of an exception
329 // crossing an activation frame boundary, that is not the case if the callee
330 // is compiled code => need to setup the rsp.
331 //
332 // rax,: exception oop
333
334 address generate_catch_exception() {
335 StubCodeMark mark(this, "StubRoutines", "catch_exception");
336 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
337 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
338 address start = __ pc();
339
340 // get thread directly
341 __ movptr(rcx, thread);
342 #ifdef ASSERT
343 // verify that threads correspond
344 { Label L;
345 __ get_thread(rbx);
346 __ cmpptr(rbx, rcx);
347 __ jcc(Assembler::equal, L);
348 __ stop("StubRoutines::catch_exception: threads must correspond");
349 __ bind(L);
350 }
351 #endif
352 // set pending exception
353 __ verify_oop(rax);
354 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
355 __ lea(Address(rcx, Thread::exception_file_offset ()),
356 ExternalAddress((address)__FILE__));
357 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
358 // complete return to VM
359 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
360 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
361
362 return start;
363 }
364
365
366 //------------------------------------------------------------------------------------------------------------------------
367 // Continuation point for runtime calls returning with a pending exception.
368 // The pending exception check happened in the runtime or native call stub.
369 // The pending exception in Thread is converted into a Java-level exception.
370 //
371 // Contract with Java-level exception handlers:
372 // rax,: exception
373 // rdx: throwing pc
374 //
375 // NOTE: At entry of this stub, exception-pc must be on stack !!
376
377 address generate_forward_exception() {
378 StubCodeMark mark(this, "StubRoutines", "forward exception");
379 address start = __ pc();
380
381 // Upon entry, the sp points to the return address returning into Java
382 // (interpreted or compiled) code; i.e., the return address becomes the
383 // throwing pc.
384 //
385 // Arguments pushed before the runtime call are still on the stack but
386 // the exception handler will reset the stack pointer -> ignore them.
387 // A potential result in registers can be ignored as well.
388
389 #ifdef ASSERT
390 // make sure this code is only executed if there is a pending exception
391 { Label L;
392 __ get_thread(rcx);
393 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
394 __ jcc(Assembler::notEqual, L);
395 __ stop("StubRoutines::forward exception: no pending exception (1)");
396 __ bind(L);
397 }
398 #endif
399
400 // compute exception handler into rbx,
401 __ movptr(rax, Address(rsp, 0));
402 BLOCK_COMMENT("call exception_handler_for_return_address");
403 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rax);
404 __ mov(rbx, rax);
405
406 // setup rax, & rdx, remove return address & clear pending exception
407 __ get_thread(rcx);
408 __ pop(rdx);
409 __ movptr(rax, Address(rcx, Thread::pending_exception_offset()));
410 __ movptr(Address(rcx, Thread::pending_exception_offset()), NULL_WORD);
411
412 #ifdef ASSERT
413 // make sure exception is set
414 { Label L;
415 __ testptr(rax, rax);
416 __ jcc(Assembler::notEqual, L);
417 __ stop("StubRoutines::forward exception: no pending exception (2)");
418 __ bind(L);
419 }
420 #endif
421
422 // continue at exception handler (return address removed)
423 // rax,: exception
424 // rbx,: exception handler
425 // rdx: throwing pc
426 __ verify_oop(rax);
427 __ jmp(rbx);
428
429 return start;
430 }
431
432
433 //----------------------------------------------------------------------------------------------------
434 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
435 //
436 // xchg exists as far back as 8086, lock needed for MP only
437 // Stack layout immediately after call:
438 //
439 // 0 [ret addr ] <--- rsp
440 // 1 [ ex ]
441 // 2 [ dest ]
442 //
443 // Result: *dest <- ex, return (old *dest)
444 //
445 // Note: win32 does not currently use this code
446
447 address generate_atomic_xchg() {
448 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
449 address start = __ pc();
450
451 __ push(rdx);
452 Address exchange(rsp, 2 * wordSize);
453 Address dest_addr(rsp, 3 * wordSize);
454 __ movl(rax, exchange);
455 __ movptr(rdx, dest_addr);
456 __ xchgl(rax, Address(rdx, 0));
457 __ pop(rdx);
458 __ ret(0);
459
460 return start;
461 }
462
463 //----------------------------------------------------------------------------------------------------
464 // Support for void verify_mxcsr()
465 //
466 // This routine is used with -Xcheck:jni to verify that native
467 // JNI code does not return to Java code without restoring the
468 // MXCSR register to our expected state.
469
470
471 address generate_verify_mxcsr() {
472 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
473 address start = __ pc();
474
475 const Address mxcsr_save(rsp, 0);
476
477 if (CheckJNICalls && UseSSE > 0 ) {
478 Label ok_ret;
479 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
480 __ push(rax);
481 __ subptr(rsp, wordSize); // allocate a temp location
482 __ stmxcsr(mxcsr_save);
483 __ movl(rax, mxcsr_save);
484 __ andl(rax, MXCSR_MASK);
485 __ cmp32(rax, mxcsr_std);
486 __ jcc(Assembler::equal, ok_ret);
487
488 __ warn("MXCSR changed by native JNI code.");
489
490 __ ldmxcsr(mxcsr_std);
491
492 __ bind(ok_ret);
493 __ addptr(rsp, wordSize);
494 __ pop(rax);
495 }
496
497 __ ret(0);
498
499 return start;
500 }
501
502
503 //---------------------------------------------------------------------------
504 // Support for void verify_fpu_cntrl_wrd()
505 //
506 // This routine is used with -Xcheck:jni to verify that native
507 // JNI code does not return to Java code without restoring the
508 // FP control word to our expected state.
509
510 address generate_verify_fpu_cntrl_wrd() {
511 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
512 address start = __ pc();
513
514 const Address fpu_cntrl_wrd_save(rsp, 0);
515
516 if (CheckJNICalls) {
517 Label ok_ret;
518 __ push(rax);
519 __ subptr(rsp, wordSize); // allocate a temp location
520 __ fnstcw(fpu_cntrl_wrd_save);
521 __ movl(rax, fpu_cntrl_wrd_save);
522 __ andl(rax, FPU_CNTRL_WRD_MASK);
523 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
524 __ cmp32(rax, fpu_std);
525 __ jcc(Assembler::equal, ok_ret);
526
527 __ warn("Floating point control word changed by native JNI code.");
528
529 __ fldcw(fpu_std);
530
531 __ bind(ok_ret);
532 __ addptr(rsp, wordSize);
533 __ pop(rax);
534 }
535
536 __ ret(0);
537
538 return start;
539 }
540
541 //---------------------------------------------------------------------------
542 // Wrapper for slow-case handling of double-to-integer conversion
543 // d2i or f2i fast case failed either because it is nan or because
544 // of under/overflow.
545 // Input: FPU TOS: float value
546 // Output: rax, (rdx): integer (long) result
547
548 address generate_d2i_wrapper(BasicType t, address fcn) {
549 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
550 address start = __ pc();
551
552 // Capture info about frame layout
553 enum layout { FPUState_off = 0,
554 rbp_off = FPUStateSizeInWords,
555 rdi_off,
556 rsi_off,
557 rcx_off,
558 rbx_off,
559 saved_argument_off,
560 saved_argument_off2, // 2nd half of double
561 framesize
562 };
563
564 assert(FPUStateSizeInWords == 27, "update stack layout");
565
566 // Save outgoing argument to stack across push_FPU_state()
567 __ subptr(rsp, wordSize * 2);
568 __ fstp_d(Address(rsp, 0));
569
570 // Save CPU & FPU state
571 __ push(rbx);
572 __ push(rcx);
573 __ push(rsi);
574 __ push(rdi);
575 __ push(rbp);
576 __ push_FPU_state();
577
578 // push_FPU_state() resets the FP top of stack
579 // Load original double into FP top of stack
580 __ fld_d(Address(rsp, saved_argument_off * wordSize));
581 // Store double into stack as outgoing argument
582 __ subptr(rsp, wordSize*2);
583 __ fst_d(Address(rsp, 0));
584
585 // Prepare FPU for doing math in C-land
586 __ empty_FPU_stack();
587 // Call the C code to massage the double. Result in EAX
588 if (t == T_INT)
589 { BLOCK_COMMENT("SharedRuntime::d2i"); }
590 else if (t == T_LONG)
591 { BLOCK_COMMENT("SharedRuntime::d2l"); }
592 __ call_VM_leaf( fcn, 2 );
593
594 // Restore CPU & FPU state
595 __ pop_FPU_state();
596 __ pop(rbp);
597 __ pop(rdi);
598 __ pop(rsi);
599 __ pop(rcx);
600 __ pop(rbx);
601 __ addptr(rsp, wordSize * 2);
602
603 __ ret(0);
604
605 return start;
606 }
607
608
609 //---------------------------------------------------------------------------
610 // The following routine generates a subroutine to throw an asynchronous
611 // UnknownError when an unsafe access gets a fault that could not be
612 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
613 address generate_handler_for_unsafe_access() {
614 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
615 address start = __ pc();
616
617 __ push(0); // hole for return address-to-be
618 __ pusha(); // push registers
619 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
620 BLOCK_COMMENT("call handle_unsafe_access");
621 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
622 __ movptr(next_pc, rax); // stuff next address
623 __ popa();
624 __ ret(0); // jump to next address
625
626 return start;
627 }
628
629
630 //----------------------------------------------------------------------------------------------------
631 // Non-destructive plausibility checks for oops
632
633 address generate_verify_oop() {
634 StubCodeMark mark(this, "StubRoutines", "verify_oop");
635 address start = __ pc();
636
637 // Incoming arguments on stack after saving rax,:
638 //
639 // [tos ]: saved rdx
640 // [tos + 1]: saved EFLAGS
641 // [tos + 2]: return address
642 // [tos + 3]: char* error message
643 // [tos + 4]: oop object to verify
644 // [tos + 5]: saved rax, - saved by caller and bashed
645
646 Label exit, error;
647 __ pushf();
648 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
649 __ push(rdx); // save rdx
650 // make sure object is 'reasonable'
651 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
652 __ testptr(rax, rax);
653 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
654
655 // Check if the oop is in the right area of memory
656 const int oop_mask = Universe::verify_oop_mask();
657 const int oop_bits = Universe::verify_oop_bits();
658 __ mov(rdx, rax);
659 __ andptr(rdx, oop_mask);
660 __ cmpptr(rdx, oop_bits);
661 __ jcc(Assembler::notZero, error);
662
663 // make sure klass is 'reasonable'
664 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
665 __ testptr(rax, rax);
666 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
667
668 // Check if the klass is in the right area of memory
669 const int klass_mask = Universe::verify_klass_mask();
670 const int klass_bits = Universe::verify_klass_bits();
671 __ mov(rdx, rax);
672 __ andptr(rdx, klass_mask);
673 __ cmpptr(rdx, klass_bits);
674 __ jcc(Assembler::notZero, error);
675
676 // make sure klass' klass is 'reasonable'
677 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
678 __ testptr(rax, rax);
679 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
680
681 __ mov(rdx, rax);
682 __ andptr(rdx, klass_mask);
683 __ cmpptr(rdx, klass_bits);
684 __ jcc(Assembler::notZero, error); // if klass not in right area
685 // of memory it is broken too.
686
687 // return if everything seems ok
688 __ bind(exit);
689 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
690 __ pop(rdx); // restore rdx
691 __ popf(); // restore EFLAGS
692 __ ret(3 * wordSize); // pop arguments
693
694 // handle errors
695 __ bind(error);
696 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
697 __ pop(rdx); // get saved rdx back
698 __ popf(); // get saved EFLAGS off stack -- will be ignored
699 __ pusha(); // push registers (eip = return address & msg are already pushed)
700 BLOCK_COMMENT("call MacroAssembler::debug");
701 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
702 __ popa();
703 __ ret(3 * wordSize); // pop arguments
704 return start;
705 }
706
707 //
708 // Generate pre-barrier for array stores
709 //
710 // Input:
711 // start - starting address
712 // count - element count
713 void gen_write_ref_array_pre_barrier(Register start, Register count) {
714 assert_different_registers(start, count);
715 BarrierSet* bs = Universe::heap()->barrier_set();
716 switch (bs->kind()) {
717 case BarrierSet::G1SATBCT:
718 case BarrierSet::G1SATBCTLogging:
719 {
720 __ pusha(); // push registers
721 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
722 start, count);
723 __ popa();
724 }
725 break;
726 case BarrierSet::CardTableModRef:
727 case BarrierSet::CardTableExtension:
728 case BarrierSet::ModRef:
729 break;
730 default :
731 ShouldNotReachHere();
732
733 }
734 }
735
736
737 //
738 // Generate a post-barrier for an array store
739 //
740 // start - starting address
741 // count - element count
742 //
743 // The two input registers are overwritten.
744 //
745 void gen_write_ref_array_post_barrier(Register start, Register count) {
746 BarrierSet* bs = Universe::heap()->barrier_set();
747 assert_different_registers(start, count);
748 switch (bs->kind()) {
749 case BarrierSet::G1SATBCT:
750 case BarrierSet::G1SATBCTLogging:
751 {
752 __ pusha(); // push registers
753 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
754 start, count);
755 __ popa();
756 }
757 break;
758
759 case BarrierSet::CardTableModRef:
760 case BarrierSet::CardTableExtension:
761 {
762 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
763 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
764
765 Label L_loop;
766 const Register end = count; // elements count; end == start+count-1
767 assert_different_registers(start, end);
768
769 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
770 __ shrptr(start, CardTableModRefBS::card_shift);
771 __ shrptr(end, CardTableModRefBS::card_shift);
772 __ subptr(end, start); // end --> count
773 __ BIND(L_loop);
774 intptr_t disp = (intptr_t) ct->byte_map_base;
775 Address cardtable(start, count, Address::times_1, disp);
776 __ movb(cardtable, 0);
777 __ decrement(count);
778 __ jcc(Assembler::greaterEqual, L_loop);
779 }
780 break;
781 case BarrierSet::ModRef:
782 break;
783 default :
784 ShouldNotReachHere();
785
786 }
787 }
788
789
790 // Copy 64 bytes chunks
791 //
792 // Inputs:
793 // from - source array address
794 // to_from - destination array address - from
795 // qword_count - 8-bytes element count, negative
796 //
797 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
798 assert( UseSSE >= 2, "supported cpu only" );
799 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
800 // Copy 64-byte chunks
801 __ jmpb(L_copy_64_bytes);
802 __ align(16);
803 __ BIND(L_copy_64_bytes_loop);
804
805 if(UseUnalignedLoadStores) {
806 __ movdqu(xmm0, Address(from, 0));
807 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
808 __ movdqu(xmm1, Address(from, 16));
809 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
810 __ movdqu(xmm2, Address(from, 32));
811 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
812 __ movdqu(xmm3, Address(from, 48));
813 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
814
815 } else {
816 __ movq(xmm0, Address(from, 0));
817 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
818 __ movq(xmm1, Address(from, 8));
819 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
820 __ movq(xmm2, Address(from, 16));
821 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
822 __ movq(xmm3, Address(from, 24));
823 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
824 __ movq(xmm4, Address(from, 32));
825 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
826 __ movq(xmm5, Address(from, 40));
827 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
828 __ movq(xmm6, Address(from, 48));
829 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
830 __ movq(xmm7, Address(from, 56));
831 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
832 }
833
834 __ addl(from, 64);
835 __ BIND(L_copy_64_bytes);
836 __ subl(qword_count, 8);
837 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
838 __ addl(qword_count, 8);
839 __ jccb(Assembler::zero, L_exit);
840 //
841 // length is too short, just copy qwords
842 //
843 __ BIND(L_copy_8_bytes);
844 __ movq(xmm0, Address(from, 0));
845 __ movq(Address(from, to_from, Address::times_1), xmm0);
846 __ addl(from, 8);
847 __ decrement(qword_count);
848 __ jcc(Assembler::greater, L_copy_8_bytes);
849 __ BIND(L_exit);
850 }
851
852 // Copy 64 bytes chunks
853 //
854 // Inputs:
855 // from - source array address
856 // to_from - destination array address - from
857 // qword_count - 8-bytes element count, negative
858 //
859 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
860 assert( VM_Version::supports_mmx(), "supported cpu only" );
861 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
862 // Copy 64-byte chunks
863 __ jmpb(L_copy_64_bytes);
864 __ align(16);
865 __ BIND(L_copy_64_bytes_loop);
866 __ movq(mmx0, Address(from, 0));
867 __ movq(mmx1, Address(from, 8));
868 __ movq(mmx2, Address(from, 16));
869 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
870 __ movq(mmx3, Address(from, 24));
871 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
872 __ movq(mmx4, Address(from, 32));
873 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
874 __ movq(mmx5, Address(from, 40));
875 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
876 __ movq(mmx6, Address(from, 48));
877 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
878 __ movq(mmx7, Address(from, 56));
879 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
880 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
881 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
882 __ addptr(from, 64);
883 __ BIND(L_copy_64_bytes);
884 __ subl(qword_count, 8);
885 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
886 __ addl(qword_count, 8);
887 __ jccb(Assembler::zero, L_exit);
888 //
889 // length is too short, just copy qwords
890 //
891 __ BIND(L_copy_8_bytes);
892 __ movq(mmx0, Address(from, 0));
893 __ movq(Address(from, to_from, Address::times_1), mmx0);
894 __ addptr(from, 8);
895 __ decrement(qword_count);
896 __ jcc(Assembler::greater, L_copy_8_bytes);
897 __ BIND(L_exit);
898 __ emms();
899 }
900
901 address generate_disjoint_copy(BasicType t, bool aligned,
902 Address::ScaleFactor sf,
903 address* entry, const char *name) {
904 __ align(CodeEntryAlignment);
905 StubCodeMark mark(this, "StubRoutines", name);
906 address start = __ pc();
907
908 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
909 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
910
911 int shift = Address::times_ptr - sf;
912
913 const Register from = rsi; // source array address
914 const Register to = rdi; // destination array address
915 const Register count = rcx; // elements count
916 const Register to_from = to; // (to - from)
917 const Register saved_to = rdx; // saved destination array address
918
919 __ enter(); // required for proper stackwalking of RuntimeStub frame
920 __ push(rsi);
921 __ push(rdi);
922 __ movptr(from , Address(rsp, 12+ 4));
923 __ movptr(to , Address(rsp, 12+ 8));
924 __ movl(count, Address(rsp, 12+ 12));
925 if (t == T_OBJECT) {
926 __ testl(count, count);
927 __ jcc(Assembler::zero, L_0_count);
928 gen_write_ref_array_pre_barrier(to, count);
929 __ mov(saved_to, to); // save 'to'
930 }
931
932 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
933 BLOCK_COMMENT("Entry:");
934
935 __ subptr(to, from); // to --> to_from
936 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
937 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
938 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
939 // align source address at 4 bytes address boundary
940 if (t == T_BYTE) {
941 // One byte misalignment happens only for byte arrays
942 __ testl(from, 1);
943 __ jccb(Assembler::zero, L_skip_align1);
944 __ movb(rax, Address(from, 0));
945 __ movb(Address(from, to_from, Address::times_1, 0), rax);
946 __ increment(from);
947 __ decrement(count);
948 __ BIND(L_skip_align1);
949 }
950 // Two bytes misalignment happens only for byte and short (char) arrays
951 __ testl(from, 2);
952 __ jccb(Assembler::zero, L_skip_align2);
953 __ movw(rax, Address(from, 0));
954 __ movw(Address(from, to_from, Address::times_1, 0), rax);
955 __ addptr(from, 2);
956 __ subl(count, 1<<(shift-1));
957 __ BIND(L_skip_align2);
958 }
959 if (!VM_Version::supports_mmx()) {
960 __ mov(rax, count); // save 'count'
961 __ shrl(count, shift); // bytes count
962 __ addptr(to_from, from);// restore 'to'
963 __ rep_mov();
964 __ subptr(to_from, from);// restore 'to_from'
965 __ mov(count, rax); // restore 'count'
966 __ jmpb(L_copy_2_bytes); // all dwords were copied
967 } else {
968 if (!UseUnalignedLoadStores) {
969 // align to 8 bytes, we know we are 4 byte aligned to start
970 __ testptr(from, 4);
971 __ jccb(Assembler::zero, L_copy_64_bytes);
972 __ movl(rax, Address(from, 0));
973 __ movl(Address(from, to_from, Address::times_1, 0), rax);
974 __ addptr(from, 4);
975 __ subl(count, 1<<shift);
976 }
977 __ BIND(L_copy_64_bytes);
978 __ mov(rax, count);
979 __ shrl(rax, shift+1); // 8 bytes chunk count
980 //
981 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
982 //
983 if (UseXMMForArrayCopy) {
984 xmm_copy_forward(from, to_from, rax);
985 } else {
986 mmx_copy_forward(from, to_from, rax);
987 }
988 }
989 // copy tailing dword
990 __ BIND(L_copy_4_bytes);
991 __ testl(count, 1<<shift);
992 __ jccb(Assembler::zero, L_copy_2_bytes);
993 __ movl(rax, Address(from, 0));
994 __ movl(Address(from, to_from, Address::times_1, 0), rax);
995 if (t == T_BYTE || t == T_SHORT) {
996 __ addptr(from, 4);
997 __ BIND(L_copy_2_bytes);
998 // copy tailing word
999 __ testl(count, 1<<(shift-1));
1000 __ jccb(Assembler::zero, L_copy_byte);
1001 __ movw(rax, Address(from, 0));
1002 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1003 if (t == T_BYTE) {
1004 __ addptr(from, 2);
1005 __ BIND(L_copy_byte);
1006 // copy tailing byte
1007 __ testl(count, 1);
1008 __ jccb(Assembler::zero, L_exit);
1009 __ movb(rax, Address(from, 0));
1010 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1011 __ BIND(L_exit);
1012 } else {
1013 __ BIND(L_copy_byte);
1014 }
1015 } else {
1016 __ BIND(L_copy_2_bytes);
1017 }
1018
1019 if (t == T_OBJECT) {
1020 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1021 __ mov(to, saved_to); // restore 'to'
1022 gen_write_ref_array_post_barrier(to, count);
1023 __ BIND(L_0_count);
1024 }
1025 inc_copy_counter_np(t);
1026 __ pop(rdi);
1027 __ pop(rsi);
1028 __ leave(); // required for proper stackwalking of RuntimeStub frame
1029 __ xorptr(rax, rax); // return 0
1030 __ ret(0);
1031 return start;
1032 }
1033
1034
1035 address generate_conjoint_copy(BasicType t, bool aligned,
1036 Address::ScaleFactor sf,
1037 address nooverlap_target,
1038 address* entry, const char *name) {
1039 __ align(CodeEntryAlignment);
1040 StubCodeMark mark(this, "StubRoutines", name);
1041 address start = __ pc();
1042
1043 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1044 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1045
1046 int shift = Address::times_ptr - sf;
1047
1048 const Register src = rax; // source array address
1049 const Register dst = rdx; // destination array address
1050 const Register from = rsi; // source array address
1051 const Register to = rdi; // destination array address
1052 const Register count = rcx; // elements count
1053 const Register end = rax; // array end address
1054
1055 __ enter(); // required for proper stackwalking of RuntimeStub frame
1056 __ push(rsi);
1057 __ push(rdi);
1058 __ movptr(src , Address(rsp, 12+ 4)); // from
1059 __ movptr(dst , Address(rsp, 12+ 8)); // to
1060 __ movl2ptr(count, Address(rsp, 12+12)); // count
1061 if (t == T_OBJECT) {
1062 gen_write_ref_array_pre_barrier(dst, count);
1063 }
1064
1065 if (entry != NULL) {
1066 *entry = __ pc(); // Entry point from generic arraycopy stub.
1067 BLOCK_COMMENT("Entry:");
1068 }
1069
1070 if (t == T_OBJECT) {
1071 __ testl(count, count);
1072 __ jcc(Assembler::zero, L_0_count);
1073 }
1074 __ mov(from, src);
1075 __ mov(to , dst);
1076
1077 // arrays overlap test
1078 RuntimeAddress nooverlap(nooverlap_target);
1079 __ cmpptr(dst, src);
1080 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1081 __ jump_cc(Assembler::belowEqual, nooverlap);
1082 __ cmpptr(dst, end);
1083 __ jump_cc(Assembler::aboveEqual, nooverlap);
1084
1085 // copy from high to low
1086 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1087 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1088 if (t == T_BYTE || t == T_SHORT) {
1089 // Align the end of destination array at 4 bytes address boundary
1090 __ lea(end, Address(dst, count, sf, 0));
1091 if (t == T_BYTE) {
1092 // One byte misalignment happens only for byte arrays
1093 __ testl(end, 1);
1094 __ jccb(Assembler::zero, L_skip_align1);
1095 __ decrement(count);
1096 __ movb(rdx, Address(from, count, sf, 0));
1097 __ movb(Address(to, count, sf, 0), rdx);
1098 __ BIND(L_skip_align1);
1099 }
1100 // Two bytes misalignment happens only for byte and short (char) arrays
1101 __ testl(end, 2);
1102 __ jccb(Assembler::zero, L_skip_align2);
1103 __ subptr(count, 1<<(shift-1));
1104 __ movw(rdx, Address(from, count, sf, 0));
1105 __ movw(Address(to, count, sf, 0), rdx);
1106 __ BIND(L_skip_align2);
1107 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1108 __ jcc(Assembler::below, L_copy_4_bytes);
1109 }
1110
1111 if (!VM_Version::supports_mmx()) {
1112 __ std();
1113 __ mov(rax, count); // Save 'count'
1114 __ mov(rdx, to); // Save 'to'
1115 __ lea(rsi, Address(from, count, sf, -4));
1116 __ lea(rdi, Address(to , count, sf, -4));
1117 __ shrptr(count, shift); // bytes count
1118 __ rep_mov();
1119 __ cld();
1120 __ mov(count, rax); // restore 'count'
1121 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1122 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1123 __ mov(to, rdx); // restore 'to'
1124 __ jmpb(L_copy_2_bytes); // all dword were copied
1125 } else {
1126 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1127 __ testptr(end, 4);
1128 __ jccb(Assembler::zero, L_copy_8_bytes);
1129 __ subl(count, 1<<shift);
1130 __ movl(rdx, Address(from, count, sf, 0));
1131 __ movl(Address(to, count, sf, 0), rdx);
1132 __ jmpb(L_copy_8_bytes);
1133
1134 __ align(16);
1135 // Move 8 bytes
1136 __ BIND(L_copy_8_bytes_loop);
1137 if (UseXMMForArrayCopy) {
1138 __ movq(xmm0, Address(from, count, sf, 0));
1139 __ movq(Address(to, count, sf, 0), xmm0);
1140 } else {
1141 __ movq(mmx0, Address(from, count, sf, 0));
1142 __ movq(Address(to, count, sf, 0), mmx0);
1143 }
1144 __ BIND(L_copy_8_bytes);
1145 __ subl(count, 2<<shift);
1146 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1147 __ addl(count, 2<<shift);
1148 if (!UseXMMForArrayCopy) {
1149 __ emms();
1150 }
1151 }
1152 __ BIND(L_copy_4_bytes);
1153 // copy prefix qword
1154 __ testl(count, 1<<shift);
1155 __ jccb(Assembler::zero, L_copy_2_bytes);
1156 __ movl(rdx, Address(from, count, sf, -4));
1157 __ movl(Address(to, count, sf, -4), rdx);
1158
1159 if (t == T_BYTE || t == T_SHORT) {
1160 __ subl(count, (1<<shift));
1161 __ BIND(L_copy_2_bytes);
1162 // copy prefix dword
1163 __ testl(count, 1<<(shift-1));
1164 __ jccb(Assembler::zero, L_copy_byte);
1165 __ movw(rdx, Address(from, count, sf, -2));
1166 __ movw(Address(to, count, sf, -2), rdx);
1167 if (t == T_BYTE) {
1168 __ subl(count, 1<<(shift-1));
1169 __ BIND(L_copy_byte);
1170 // copy prefix byte
1171 __ testl(count, 1);
1172 __ jccb(Assembler::zero, L_exit);
1173 __ movb(rdx, Address(from, 0));
1174 __ movb(Address(to, 0), rdx);
1175 __ BIND(L_exit);
1176 } else {
1177 __ BIND(L_copy_byte);
1178 }
1179 } else {
1180 __ BIND(L_copy_2_bytes);
1181 }
1182 if (t == T_OBJECT) {
1183 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1184 gen_write_ref_array_post_barrier(to, count);
1185 __ BIND(L_0_count);
1186 }
1187 inc_copy_counter_np(t);
1188 __ pop(rdi);
1189 __ pop(rsi);
1190 __ leave(); // required for proper stackwalking of RuntimeStub frame
1191 __ xorptr(rax, rax); // return 0
1192 __ ret(0);
1193 return start;
1194 }
1195
1196
1197 address generate_disjoint_long_copy(address* entry, const char *name) {
1198 __ align(CodeEntryAlignment);
1199 StubCodeMark mark(this, "StubRoutines", name);
1200 address start = __ pc();
1201
1202 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1203 const Register from = rax; // source array address
1204 const Register to = rdx; // destination array address
1205 const Register count = rcx; // elements count
1206 const Register to_from = rdx; // (to - from)
1207
1208 __ enter(); // required for proper stackwalking of RuntimeStub frame
1209 __ movptr(from , Address(rsp, 8+0)); // from
1210 __ movptr(to , Address(rsp, 8+4)); // to
1211 __ movl2ptr(count, Address(rsp, 8+8)); // count
1212
1213 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1214 BLOCK_COMMENT("Entry:");
1215
1216 __ subptr(to, from); // to --> to_from
1217 if (VM_Version::supports_mmx()) {
1218 if (UseXMMForArrayCopy) {
1219 xmm_copy_forward(from, to_from, count);
1220 } else {
1221 mmx_copy_forward(from, to_from, count);
1222 }
1223 } else {
1224 __ jmpb(L_copy_8_bytes);
1225 __ align(16);
1226 __ BIND(L_copy_8_bytes_loop);
1227 __ fild_d(Address(from, 0));
1228 __ fistp_d(Address(from, to_from, Address::times_1));
1229 __ addptr(from, 8);
1230 __ BIND(L_copy_8_bytes);
1231 __ decrement(count);
1232 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1233 }
1234 inc_copy_counter_np(T_LONG);
1235 __ leave(); // required for proper stackwalking of RuntimeStub frame
1236 __ xorptr(rax, rax); // return 0
1237 __ ret(0);
1238 return start;
1239 }
1240
1241 address generate_conjoint_long_copy(address nooverlap_target,
1242 address* entry, const char *name) {
1243 __ align(CodeEntryAlignment);
1244 StubCodeMark mark(this, "StubRoutines", name);
1245 address start = __ pc();
1246
1247 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1248 const Register from = rax; // source array address
1249 const Register to = rdx; // destination array address
1250 const Register count = rcx; // elements count
1251 const Register end_from = rax; // source array end address
1252
1253 __ enter(); // required for proper stackwalking of RuntimeStub frame
1254 __ movptr(from , Address(rsp, 8+0)); // from
1255 __ movptr(to , Address(rsp, 8+4)); // to
1256 __ movl2ptr(count, Address(rsp, 8+8)); // count
1257
1258 *entry = __ pc(); // Entry point from generic arraycopy stub.
1259 BLOCK_COMMENT("Entry:");
1260
1261 // arrays overlap test
1262 __ cmpptr(to, from);
1263 RuntimeAddress nooverlap(nooverlap_target);
1264 __ jump_cc(Assembler::belowEqual, nooverlap);
1265 __ lea(end_from, Address(from, count, Address::times_8, 0));
1266 __ cmpptr(to, end_from);
1267 __ movptr(from, Address(rsp, 8)); // from
1268 __ jump_cc(Assembler::aboveEqual, nooverlap);
1269
1270 __ jmpb(L_copy_8_bytes);
1271
1272 __ align(16);
1273 __ BIND(L_copy_8_bytes_loop);
1274 if (VM_Version::supports_mmx()) {
1275 if (UseXMMForArrayCopy) {
1276 __ movq(xmm0, Address(from, count, Address::times_8));
1277 __ movq(Address(to, count, Address::times_8), xmm0);
1278 } else {
1279 __ movq(mmx0, Address(from, count, Address::times_8));
1280 __ movq(Address(to, count, Address::times_8), mmx0);
1281 }
1282 } else {
1283 __ fild_d(Address(from, count, Address::times_8));
1284 __ fistp_d(Address(to, count, Address::times_8));
1285 }
1286 __ BIND(L_copy_8_bytes);
1287 __ decrement(count);
1288 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1289
1290 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1291 __ emms();
1292 }
1293 inc_copy_counter_np(T_LONG);
1294 __ leave(); // required for proper stackwalking of RuntimeStub frame
1295 __ xorptr(rax, rax); // return 0
1296 __ ret(0);
1297 return start;
1298 }
1299
1300
1301 // Helper for generating a dynamic type check.
1302 // The sub_klass must be one of {rbx, rdx, rsi}.
1303 // The temp is killed.
1304 void generate_type_check(Register sub_klass,
1305 Address& super_check_offset_addr,
1306 Address& super_klass_addr,
1307 Register temp,
1308 Label* L_success, Label* L_failure) {
1309 BLOCK_COMMENT("type_check:");
1310
1311 Label L_fallthrough;
1312 #define LOCAL_JCC(assembler_con, label_ptr) \
1313 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1314 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1315
1316 // The following is a strange variation of the fast path which requires
1317 // one less register, because needed values are on the argument stack.
1318 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1319 // L_success, L_failure, NULL);
1320 assert_different_registers(sub_klass, temp);
1321
1322 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1323 Klass::secondary_super_cache_offset_in_bytes());
1324
1325 // if the pointers are equal, we are done (e.g., String[] elements)
1326 __ cmpptr(sub_klass, super_klass_addr);
1327 LOCAL_JCC(Assembler::equal, L_success);
1328
1329 // check the supertype display:
1330 __ movl2ptr(temp, super_check_offset_addr);
1331 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1332 __ movptr(temp, super_check_addr); // load displayed supertype
1333 __ cmpptr(temp, super_klass_addr); // test the super type
1334 LOCAL_JCC(Assembler::equal, L_success);
1335
1336 // if it was a primary super, we can just fail immediately
1337 __ cmpl(super_check_offset_addr, sc_offset);
1338 LOCAL_JCC(Assembler::notEqual, L_failure);
1339
1340 // The repne_scan instruction uses fixed registers, which will get spilled.
1341 // We happen to know this works best when super_klass is in rax.
1342 Register super_klass = temp;
1343 __ movptr(super_klass, super_klass_addr);
1344 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1345 L_success, L_failure);
1346
1347 __ bind(L_fallthrough);
1348
1349 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1350 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1351
1352 #undef LOCAL_JCC
1353 }
1354
1355 //
1356 // Generate checkcasting array copy stub
1357 //
1358 // Input:
1359 // 4(rsp) - source array address
1360 // 8(rsp) - destination array address
1361 // 12(rsp) - element count, can be zero
1362 // 16(rsp) - size_t ckoff (super_check_offset)
1363 // 20(rsp) - oop ckval (super_klass)
1364 //
1365 // Output:
1366 // rax, == 0 - success
1367 // rax, == -1^K - failure, where K is partial transfer count
1368 //
1369 address generate_checkcast_copy(const char *name, address* entry) {
1370 __ align(CodeEntryAlignment);
1371 StubCodeMark mark(this, "StubRoutines", name);
1372 address start = __ pc();
1373
1374 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1375
1376 // register use:
1377 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1378 // rdi, rsi -- element access (oop, klass)
1379 // rbx, -- temp
1380 const Register from = rax; // source array address
1381 const Register to = rdx; // destination array address
1382 const Register length = rcx; // elements count
1383 const Register elem = rdi; // each oop copied
1384 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1385 const Register temp = rbx; // lone remaining temp
1386
1387 __ enter(); // required for proper stackwalking of RuntimeStub frame
1388
1389 __ push(rsi);
1390 __ push(rdi);
1391 __ push(rbx);
1392
1393 Address from_arg(rsp, 16+ 4); // from
1394 Address to_arg(rsp, 16+ 8); // to
1395 Address length_arg(rsp, 16+12); // elements count
1396 Address ckoff_arg(rsp, 16+16); // super_check_offset
1397 Address ckval_arg(rsp, 16+20); // super_klass
1398
1399 // Load up:
1400 __ movptr(from, from_arg);
1401 __ movptr(to, to_arg);
1402 __ movl2ptr(length, length_arg);
1403
1404 *entry = __ pc(); // Entry point from generic arraycopy stub.
1405 BLOCK_COMMENT("Entry:");
1406
1407 //---------------------------------------------------------------
1408 // Assembler stub will be used for this call to arraycopy
1409 // if the two arrays are subtypes of Object[] but the
1410 // destination array type is not equal to or a supertype
1411 // of the source type. Each element must be separately
1412 // checked.
1413
1414 // Loop-invariant addresses. They are exclusive end pointers.
1415 Address end_from_addr(from, length, Address::times_ptr, 0);
1416 Address end_to_addr(to, length, Address::times_ptr, 0);
1417
1418 Register end_from = from; // re-use
1419 Register end_to = to; // re-use
1420 Register count = length; // re-use
1421
1422 // Loop-variant addresses. They assume post-incremented count < 0.
1423 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1424 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1425 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1426
1427 // Copy from low to high addresses, indexed from the end of each array.
1428 gen_write_ref_array_pre_barrier(to, count);
1429 __ lea(end_from, end_from_addr);
1430 __ lea(end_to, end_to_addr);
1431 assert(length == count, ""); // else fix next line:
1432 __ negptr(count); // negate and test the length
1433 __ jccb(Assembler::notZero, L_load_element);
1434
1435 // Empty array: Nothing to do.
1436 __ xorptr(rax, rax); // return 0 on (trivial) success
1437 __ jmp(L_done);
1438
1439 // ======== begin loop ========
1440 // (Loop is rotated; its entry is L_load_element.)
1441 // Loop control:
1442 // for (count = -count; count != 0; count++)
1443 // Base pointers src, dst are biased by 8*count,to last element.
1444 __ align(16);
1445
1446 __ BIND(L_store_element);
1447 __ movptr(to_element_addr, elem); // store the oop
1448 __ increment(count); // increment the count toward zero
1449 __ jccb(Assembler::zero, L_do_card_marks);
1450
1451 // ======== loop entry is here ========
1452 __ BIND(L_load_element);
1453 __ movptr(elem, from_element_addr); // load the oop
1454 __ testptr(elem, elem);
1455 __ jccb(Assembler::zero, L_store_element);
1456
1457 // (Could do a trick here: Remember last successful non-null
1458 // element stored and make a quick oop equality check on it.)
1459
1460 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1461 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1462 &L_store_element, NULL);
1463 // (On fall-through, we have failed the element type check.)
1464 // ======== end loop ========
1465
1466 // It was a real error; we must depend on the caller to finish the job.
1467 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1468 // Emit GC store barriers for the oops we have copied (length_arg + count),
1469 // and report their number to the caller.
1470 __ addl(count, length_arg); // transfers = (length - remaining)
1471 __ movl2ptr(rax, count); // save the value
1472 __ notptr(rax); // report (-1^K) to caller
1473 __ movptr(to, to_arg); // reload
1474 assert_different_registers(to, count, rax);
1475 gen_write_ref_array_post_barrier(to, count);
1476 __ jmpb(L_done);
1477
1478 // Come here on success only.
1479 __ BIND(L_do_card_marks);
1480 __ movl2ptr(count, length_arg);
1481 __ movptr(to, to_arg); // reload
1482 gen_write_ref_array_post_barrier(to, count);
1483 __ xorptr(rax, rax); // return 0 on success
1484
1485 // Common exit point (success or failure).
1486 __ BIND(L_done);
1487 __ pop(rbx);
1488 __ pop(rdi);
1489 __ pop(rsi);
1490 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1491 __ leave(); // required for proper stackwalking of RuntimeStub frame
1492 __ ret(0);
1493
1494 return start;
1495 }
1496
1497 //
1498 // Generate 'unsafe' array copy stub
1499 // Though just as safe as the other stubs, it takes an unscaled
1500 // size_t argument instead of an element count.
1501 //
1502 // Input:
1503 // 4(rsp) - source array address
1504 // 8(rsp) - destination array address
1505 // 12(rsp) - byte count, can be zero
1506 //
1507 // Output:
1508 // rax, == 0 - success
1509 // rax, == -1 - need to call System.arraycopy
1510 //
1511 // Examines the alignment of the operands and dispatches
1512 // to a long, int, short, or byte copy loop.
1513 //
1514 address generate_unsafe_copy(const char *name,
1515 address byte_copy_entry,
1516 address short_copy_entry,
1517 address int_copy_entry,
1518 address long_copy_entry) {
1519
1520 Label L_long_aligned, L_int_aligned, L_short_aligned;
1521
1522 __ align(CodeEntryAlignment);
1523 StubCodeMark mark(this, "StubRoutines", name);
1524 address start = __ pc();
1525
1526 const Register from = rax; // source array address
1527 const Register to = rdx; // destination array address
1528 const Register count = rcx; // elements count
1529
1530 __ enter(); // required for proper stackwalking of RuntimeStub frame
1531 __ push(rsi);
1532 __ push(rdi);
1533 Address from_arg(rsp, 12+ 4); // from
1534 Address to_arg(rsp, 12+ 8); // to
1535 Address count_arg(rsp, 12+12); // byte count
1536
1537 // Load up:
1538 __ movptr(from , from_arg);
1539 __ movptr(to , to_arg);
1540 __ movl2ptr(count, count_arg);
1541
1542 // bump this on entry, not on exit:
1543 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1544
1545 const Register bits = rsi;
1546 __ mov(bits, from);
1547 __ orptr(bits, to);
1548 __ orptr(bits, count);
1549
1550 __ testl(bits, BytesPerLong-1);
1551 __ jccb(Assembler::zero, L_long_aligned);
1552
1553 __ testl(bits, BytesPerInt-1);
1554 __ jccb(Assembler::zero, L_int_aligned);
1555
1556 __ testl(bits, BytesPerShort-1);
1557 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1558
1559 __ BIND(L_short_aligned);
1560 __ shrptr(count, LogBytesPerShort); // size => short_count
1561 __ movl(count_arg, count); // update 'count'
1562 __ jump(RuntimeAddress(short_copy_entry));
1563
1564 __ BIND(L_int_aligned);
1565 __ shrptr(count, LogBytesPerInt); // size => int_count
1566 __ movl(count_arg, count); // update 'count'
1567 __ jump(RuntimeAddress(int_copy_entry));
1568
1569 __ BIND(L_long_aligned);
1570 __ shrptr(count, LogBytesPerLong); // size => qword_count
1571 __ movl(count_arg, count); // update 'count'
1572 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1573 __ pop(rsi);
1574 __ jump(RuntimeAddress(long_copy_entry));
1575
1576 return start;
1577 }
1578
1579
1580 // Perform range checks on the proposed arraycopy.
1581 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1582 void arraycopy_range_checks(Register src,
1583 Register src_pos,
1584 Register dst,
1585 Register dst_pos,
1586 Address& length,
1587 Label& L_failed) {
1588 BLOCK_COMMENT("arraycopy_range_checks:");
1589 const Register src_end = src_pos; // source array end position
1590 const Register dst_end = dst_pos; // destination array end position
1591 __ addl(src_end, length); // src_pos + length
1592 __ addl(dst_end, length); // dst_pos + length
1593
1594 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1595 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1596 __ jcc(Assembler::above, L_failed);
1597
1598 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1599 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1600 __ jcc(Assembler::above, L_failed);
1601
1602 BLOCK_COMMENT("arraycopy_range_checks done");
1603 }
1604
1605
1606 //
1607 // Generate generic array copy stubs
1608 //
1609 // Input:
1610 // 4(rsp) - src oop
1611 // 8(rsp) - src_pos
1612 // 12(rsp) - dst oop
1613 // 16(rsp) - dst_pos
1614 // 20(rsp) - element count
1615 //
1616 // Output:
1617 // rax, == 0 - success
1618 // rax, == -1^K - failure, where K is partial transfer count
1619 //
1620 address generate_generic_copy(const char *name,
1621 address entry_jbyte_arraycopy,
1622 address entry_jshort_arraycopy,
1623 address entry_jint_arraycopy,
1624 address entry_oop_arraycopy,
1625 address entry_jlong_arraycopy,
1626 address entry_checkcast_arraycopy) {
1627 Label L_failed, L_failed_0, L_objArray;
1628
1629 { int modulus = CodeEntryAlignment;
1630 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1631 int advance = target - (__ offset() % modulus);
1632 if (advance < 0) advance += modulus;
1633 if (advance > 0) __ nop(advance);
1634 }
1635 StubCodeMark mark(this, "StubRoutines", name);
1636
1637 // Short-hop target to L_failed. Makes for denser prologue code.
1638 __ BIND(L_failed_0);
1639 __ jmp(L_failed);
1640 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1641
1642 __ align(CodeEntryAlignment);
1643 address start = __ pc();
1644
1645 __ enter(); // required for proper stackwalking of RuntimeStub frame
1646 __ push(rsi);
1647 __ push(rdi);
1648
1649 // bump this on entry, not on exit:
1650 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1651
1652 // Input values
1653 Address SRC (rsp, 12+ 4);
1654 Address SRC_POS (rsp, 12+ 8);
1655 Address DST (rsp, 12+12);
1656 Address DST_POS (rsp, 12+16);
1657 Address LENGTH (rsp, 12+20);
1658
1659 //-----------------------------------------------------------------------
1660 // Assembler stub will be used for this call to arraycopy
1661 // if the following conditions are met:
1662 //
1663 // (1) src and dst must not be null.
1664 // (2) src_pos must not be negative.
1665 // (3) dst_pos must not be negative.
1666 // (4) length must not be negative.
1667 // (5) src klass and dst klass should be the same and not NULL.
1668 // (6) src and dst should be arrays.
1669 // (7) src_pos + length must not exceed length of src.
1670 // (8) dst_pos + length must not exceed length of dst.
1671 //
1672
1673 const Register src = rax; // source array oop
1674 const Register src_pos = rsi;
1675 const Register dst = rdx; // destination array oop
1676 const Register dst_pos = rdi;
1677 const Register length = rcx; // transfer count
1678
1679 // if (src == NULL) return -1;
1680 __ movptr(src, SRC); // src oop
1681 __ testptr(src, src);
1682 __ jccb(Assembler::zero, L_failed_0);
1683
1684 // if (src_pos < 0) return -1;
1685 __ movl2ptr(src_pos, SRC_POS); // src_pos
1686 __ testl(src_pos, src_pos);
1687 __ jccb(Assembler::negative, L_failed_0);
1688
1689 // if (dst == NULL) return -1;
1690 __ movptr(dst, DST); // dst oop
1691 __ testptr(dst, dst);
1692 __ jccb(Assembler::zero, L_failed_0);
1693
1694 // if (dst_pos < 0) return -1;
1695 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1696 __ testl(dst_pos, dst_pos);
1697 __ jccb(Assembler::negative, L_failed_0);
1698
1699 // if (length < 0) return -1;
1700 __ movl2ptr(length, LENGTH); // length
1701 __ testl(length, length);
1702 __ jccb(Assembler::negative, L_failed_0);
1703
1704 // if (src->klass() == NULL) return -1;
1705 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1706 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1707 const Register rcx_src_klass = rcx; // array klass
1708 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1709
1710 #ifdef ASSERT
1711 // assert(src->klass() != NULL);
1712 BLOCK_COMMENT("assert klasses not null");
1713 { Label L1, L2;
1714 __ testptr(rcx_src_klass, rcx_src_klass);
1715 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1716 __ bind(L1);
1717 __ stop("broken null klass");
1718 __ bind(L2);
1719 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1720 __ jccb(Assembler::equal, L1); // this would be broken also
1721 BLOCK_COMMENT("assert done");
1722 }
1723 #endif //ASSERT
1724
1725 // Load layout helper (32-bits)
1726 //
1727 // |array_tag| | header_size | element_type | |log2_element_size|
1728 // 32 30 24 16 8 2 0
1729 //
1730 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1731 //
1732
1733 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1734 Klass::layout_helper_offset_in_bytes();
1735 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1736
1737 // Handle objArrays completely differently...
1738 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1739 __ cmpl(src_klass_lh_addr, objArray_lh);
1740 __ jcc(Assembler::equal, L_objArray);
1741
1742 // if (src->klass() != dst->klass()) return -1;
1743 __ cmpptr(rcx_src_klass, dst_klass_addr);
1744 __ jccb(Assembler::notEqual, L_failed_0);
1745
1746 const Register rcx_lh = rcx; // layout helper
1747 assert(rcx_lh == rcx_src_klass, "known alias");
1748 __ movl(rcx_lh, src_klass_lh_addr);
1749
1750 // if (!src->is_Array()) return -1;
1751 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1752 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1753
1754 // At this point, it is known to be a typeArray (array_tag 0x3).
1755 #ifdef ASSERT
1756 { Label L;
1757 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1758 __ jcc(Assembler::greaterEqual, L); // signed cmp
1759 __ stop("must be a primitive array");
1760 __ bind(L);
1761 }
1762 #endif
1763
1764 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1765 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1766
1767 // typeArrayKlass
1768 //
1769 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1770 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1771 //
1772 const Register rsi_offset = rsi; // array offset
1773 const Register src_array = src; // src array offset
1774 const Register dst_array = dst; // dst array offset
1775 const Register rdi_elsize = rdi; // log2 element size
1776
1777 __ mov(rsi_offset, rcx_lh);
1778 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1779 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1780 __ addptr(src_array, rsi_offset); // src array offset
1781 __ addptr(dst_array, rsi_offset); // dst array offset
1782 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1783
1784 // next registers should be set before the jump to corresponding stub
1785 const Register from = src; // source array address
1786 const Register to = dst; // destination array address
1787 const Register count = rcx; // elements count
1788 // some of them should be duplicated on stack
1789 #define FROM Address(rsp, 12+ 4)
1790 #define TO Address(rsp, 12+ 8) // Not used now
1791 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1792
1793 BLOCK_COMMENT("scale indexes to element size");
1794 __ movl2ptr(rsi, SRC_POS); // src_pos
1795 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1796 assert(src_array == from, "");
1797 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1798 __ movl2ptr(rdi, DST_POS); // dst_pos
1799 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1800 assert(dst_array == to, "");
1801 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1802 __ movptr(FROM, from); // src_addr
1803 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1804 __ movl2ptr(count, LENGTH); // elements count
1805
1806 BLOCK_COMMENT("choose copy loop based on element size");
1807 __ cmpl(rdi_elsize, 0);
1808
1809 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1810 __ cmpl(rdi_elsize, LogBytesPerShort);
1811 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1812 __ cmpl(rdi_elsize, LogBytesPerInt);
1813 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1814 #ifdef ASSERT
1815 __ cmpl(rdi_elsize, LogBytesPerLong);
1816 __ jccb(Assembler::notEqual, L_failed);
1817 #endif
1818 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1819 __ pop(rsi);
1820 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1821
1822 __ BIND(L_failed);
1823 __ xorptr(rax, rax);
1824 __ notptr(rax); // return -1
1825 __ pop(rdi);
1826 __ pop(rsi);
1827 __ leave(); // required for proper stackwalking of RuntimeStub frame
1828 __ ret(0);
1829
1830 // objArrayKlass
1831 __ BIND(L_objArray);
1832 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1833
1834 Label L_plain_copy, L_checkcast_copy;
1835 // test array classes for subtyping
1836 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1837 __ jccb(Assembler::notEqual, L_checkcast_copy);
1838
1839 // Identically typed arrays can be copied without element-wise checks.
1840 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1841 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1842
1843 __ BIND(L_plain_copy);
1844 __ movl2ptr(count, LENGTH); // elements count
1845 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1846 __ lea(from, Address(src, src_pos, Address::times_ptr,
1847 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1848 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1849 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1850 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1851 __ movptr(FROM, from); // src_addr
1852 __ movptr(TO, to); // dst_addr
1853 __ movl(COUNT, count); // count
1854 __ jump(RuntimeAddress(entry_oop_arraycopy));
1855
1856 __ BIND(L_checkcast_copy);
1857 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1858 {
1859 // Handy offsets:
1860 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1861 objArrayKlass::element_klass_offset_in_bytes());
1862 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1863 Klass::super_check_offset_offset_in_bytes());
1864
1865 Register rsi_dst_klass = rsi;
1866 Register rdi_temp = rdi;
1867 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1868 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1869 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1870
1871 // Before looking at dst.length, make sure dst is also an objArray.
1872 __ movptr(rsi_dst_klass, dst_klass_addr);
1873 __ cmpl(dst_klass_lh_addr, objArray_lh);
1874 __ jccb(Assembler::notEqual, L_failed);
1875
1876 // It is safe to examine both src.length and dst.length.
1877 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1878 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1879 // (Now src_pos and dst_pos are killed, but not src and dst.)
1880
1881 // We'll need this temp (don't forget to pop it after the type check).
1882 __ push(rbx);
1883 Register rbx_src_klass = rbx;
1884
1885 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1886 __ movptr(rsi_dst_klass, dst_klass_addr);
1887 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1888 Label L_fail_array_check;
1889 generate_type_check(rbx_src_klass,
1890 super_check_offset_addr, dst_klass_addr,
1891 rdi_temp, NULL, &L_fail_array_check);
1892 // (On fall-through, we have passed the array type check.)
1893 __ pop(rbx);
1894 __ jmp(L_plain_copy);
1895
1896 __ BIND(L_fail_array_check);
1897 // Reshuffle arguments so we can call checkcast_arraycopy:
1898
1899 // match initial saves for checkcast_arraycopy
1900 // push(rsi); // already done; see above
1901 // push(rdi); // already done; see above
1902 // push(rbx); // already done; see above
1903
1904 // Marshal outgoing arguments now, freeing registers.
1905 Address from_arg(rsp, 16+ 4); // from
1906 Address to_arg(rsp, 16+ 8); // to
1907 Address length_arg(rsp, 16+12); // elements count
1908 Address ckoff_arg(rsp, 16+16); // super_check_offset
1909 Address ckval_arg(rsp, 16+20); // super_klass
1910
1911 Address SRC_POS_arg(rsp, 16+ 8);
1912 Address DST_POS_arg(rsp, 16+16);
1913 Address LENGTH_arg(rsp, 16+20);
1914 // push rbx, changed the incoming offsets (why not just use rbp,??)
1915 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1916
1917 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1918 __ movl2ptr(length, LENGTH_arg); // reload elements count
1919 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1920 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1921
1922 __ movptr(ckval_arg, rbx); // destination element type
1923 __ movl(rbx, Address(rbx, sco_offset));
1924 __ movl(ckoff_arg, rbx); // corresponding class check offset
1925
1926 __ movl(length_arg, length); // outgoing length argument
1927
1928 __ lea(from, Address(src, src_pos, Address::times_ptr,
1929 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1930 __ movptr(from_arg, from);
1931
1932 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1933 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1934 __ movptr(to_arg, to);
1935 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1936 }
1937
1938 return start;
1939 }
1940
1941 void generate_arraycopy_stubs() {
1942 address entry;
1943 address entry_jbyte_arraycopy;
1944 address entry_jshort_arraycopy;
1945 address entry_jint_arraycopy;
1946 address entry_oop_arraycopy;
1947 address entry_jlong_arraycopy;
1948 address entry_checkcast_arraycopy;
1949
1950 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1951 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1952 "arrayof_jbyte_disjoint_arraycopy");
1953 StubRoutines::_arrayof_jbyte_arraycopy =
1954 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1955 NULL, "arrayof_jbyte_arraycopy");
1956 StubRoutines::_jbyte_disjoint_arraycopy =
1957 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1958 "jbyte_disjoint_arraycopy");
1959 StubRoutines::_jbyte_arraycopy =
1960 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1961 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1962
1963 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1964 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1965 "arrayof_jshort_disjoint_arraycopy");
1966 StubRoutines::_arrayof_jshort_arraycopy =
1967 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1968 NULL, "arrayof_jshort_arraycopy");
1969 StubRoutines::_jshort_disjoint_arraycopy =
1970 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1971 "jshort_disjoint_arraycopy");
1972 StubRoutines::_jshort_arraycopy =
1973 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1974 &entry_jshort_arraycopy, "jshort_arraycopy");
1975
1976 // Next arrays are always aligned on 4 bytes at least.
1977 StubRoutines::_jint_disjoint_arraycopy =
1978 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1979 "jint_disjoint_arraycopy");
1980 StubRoutines::_jint_arraycopy =
1981 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1982 &entry_jint_arraycopy, "jint_arraycopy");
1983
1984 StubRoutines::_oop_disjoint_arraycopy =
1985 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
1986 "oop_disjoint_arraycopy");
1987 StubRoutines::_oop_arraycopy =
1988 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
1989 &entry_oop_arraycopy, "oop_arraycopy");
1990
1991 StubRoutines::_jlong_disjoint_arraycopy =
1992 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
1993 StubRoutines::_jlong_arraycopy =
1994 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
1995 "jlong_arraycopy");
1996
1997 StubRoutines::_arrayof_jint_disjoint_arraycopy =
1998 StubRoutines::_jint_disjoint_arraycopy;
1999 StubRoutines::_arrayof_oop_disjoint_arraycopy =
2000 StubRoutines::_oop_disjoint_arraycopy;
2001 StubRoutines::_arrayof_jlong_disjoint_arraycopy =
2002 StubRoutines::_jlong_disjoint_arraycopy;
2003
2004 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2005 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2006 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2007
2008 StubRoutines::_checkcast_arraycopy =
2009 generate_checkcast_copy("checkcast_arraycopy",
2010 &entry_checkcast_arraycopy);
2011
2012 StubRoutines::_unsafe_arraycopy =
2013 generate_unsafe_copy("unsafe_arraycopy",
2014 entry_jbyte_arraycopy,
2015 entry_jshort_arraycopy,
2016 entry_jint_arraycopy,
2017 entry_jlong_arraycopy);
2018
2019 StubRoutines::_generic_arraycopy =
2020 generate_generic_copy("generic_arraycopy",
2021 entry_jbyte_arraycopy,
2022 entry_jshort_arraycopy,
2023 entry_jint_arraycopy,
2024 entry_oop_arraycopy,
2025 entry_jlong_arraycopy,
2026 entry_checkcast_arraycopy);
2027 }
2028
2029 void generate_math_stubs() {
2030 {
2031 StubCodeMark mark(this, "StubRoutines", "log");
2032 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2033
2034 __ fld_d(Address(rsp, 4));
2035 __ flog();
2036 __ ret(0);
2037 }
2038 {
2039 StubCodeMark mark(this, "StubRoutines", "log10");
2040 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2041
2042 __ fld_d(Address(rsp, 4));
2043 __ flog10();
2044 __ ret(0);
2045 }
2046 {
2047 StubCodeMark mark(this, "StubRoutines", "sin");
2048 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2049
2050 __ fld_d(Address(rsp, 4));
2051 __ trigfunc('s');
2052 __ ret(0);
2053 }
2054 {
2055 StubCodeMark mark(this, "StubRoutines", "cos");
2056 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2057
2058 __ fld_d(Address(rsp, 4));
2059 __ trigfunc('c');
2060 __ ret(0);
2061 }
2062 {
2063 StubCodeMark mark(this, "StubRoutines", "tan");
2064 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2065
2066 __ fld_d(Address(rsp, 4));
2067 __ trigfunc('t');
2068 __ ret(0);
2069 }
2070
2071 // The intrinsic version of these seem to return the same value as
2072 // the strict version.
2073 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2074 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2075 }
2076
2077 public:
2078 // Information about frame layout at time of blocking runtime call.
2079 // Note that we only have to preserve callee-saved registers since
2080 // the compilers are responsible for supplying a continuation point
2081 // if they expect all registers to be preserved.
2082 enum layout {
2083 thread_off, // last_java_sp
2084 rbp_off, // callee saved register
2085 ret_pc,
2086 framesize
2087 };
2088
2089 private:
2090
2091 #undef __
2092 #define __ masm->
2093
2094 //------------------------------------------------------------------------------------------------------------------------
2095 // Continuation point for throwing of implicit exceptions that are not handled in
2096 // the current activation. Fabricates an exception oop and initiates normal
2097 // exception dispatching in this frame.
2098 //
2099 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2100 // This is no longer true after adapter frames were removed but could possibly
2101 // be brought back in the future if the interpreter code was reworked and it
2102 // was deemed worthwhile. The comment below was left to describe what must
2103 // happen here if callee saves were resurrected. As it stands now this stub
2104 // could actually be a vanilla BufferBlob and have now oopMap at all.
2105 // Since it doesn't make much difference we've chosen to leave it the
2106 // way it was in the callee save days and keep the comment.
2107
2108 // If we need to preserve callee-saved values we need a callee-saved oop map and
2109 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2110 // If the compiler needs all registers to be preserved between the fault
2111 // point and the exception handler then it must assume responsibility for that in
2112 // AbstractCompiler::continuation_for_implicit_null_exception or
2113 // continuation_for_implicit_division_by_zero_exception. All other implicit
2114 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2115 // either at call sites or otherwise assume that stack unwinding will be initiated,
2116 // so caller saved registers were assumed volatile in the compiler.
2117 address generate_throw_exception(const char* name, address runtime_entry,
2118 bool restore_saved_exception_pc) {
2119
2120 int insts_size = 256;
2121 int locs_size = 32;
2122
2123 CodeBuffer code(name, insts_size, locs_size);
2124 OopMapSet* oop_maps = new OopMapSet();
2125 MacroAssembler* masm = new MacroAssembler(&code);
2126
2127 address start = __ pc();
2128
2129 // This is an inlined and slightly modified version of call_VM
2130 // which has the ability to fetch the return PC out of
2131 // thread-local storage and also sets up last_Java_sp slightly
2132 // differently than the real call_VM
2133 Register java_thread = rbx;
2134 __ get_thread(java_thread);
2135 if (restore_saved_exception_pc) {
2136 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2137 __ push(rax);
2138 }
2139
2140 __ enter(); // required for proper stackwalking of RuntimeStub frame
2141
2142 // pc and rbp, already pushed
2143 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2144
2145 // Frame is now completed as far as size and linkage.
2146
2147 int frame_complete = __ pc() - start;
2148
2149 // push java thread (becomes first argument of C function)
2150 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2151
2152 // Set up last_Java_sp and last_Java_fp
2153 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2154
2155 // Call runtime
2156 BLOCK_COMMENT("call runtime_entry");
2157 __ call(RuntimeAddress(runtime_entry));
2158 // Generate oop map
2159 OopMap* map = new OopMap(framesize, 0);
2160 oop_maps->add_gc_map(__ pc() - start, map);
2161
2162 // restore the thread (cannot use the pushed argument since arguments
2163 // may be overwritten by C code generated by an optimizing compiler);
2164 // however can use the register value directly if it is callee saved.
2165 __ get_thread(java_thread);
2166
2167 __ reset_last_Java_frame(java_thread, true, false);
2168
2169 __ leave(); // required for proper stackwalking of RuntimeStub frame
2170
2171 // check for pending exceptions
2172 #ifdef ASSERT
2173 Label L;
2174 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2175 __ jcc(Assembler::notEqual, L);
2176 __ should_not_reach_here();
2177 __ bind(L);
2178 #endif /* ASSERT */
2179 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2180
2181
2182 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2183 return stub->entry_point();
2184 }
2185
2186
2187 void create_control_words() {
2188 // Round to nearest, 53-bit mode, exceptions masked
2189 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2190 // Round to zero, 53-bit mode, exception mased
2191 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2192 // Round to nearest, 24-bit mode, exceptions masked
2193 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2194 // Round to nearest, 64-bit mode, exceptions masked
2195 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2196 // Round to nearest, 64-bit mode, exceptions masked
2197 StubRoutines::_mxcsr_std = 0x1F80;
2198 // Note: the following two constants are 80-bit values
2199 // layout is critical for correct loading by FPU.
2200 // Bias for strict fp multiply/divide
2201 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2202 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2203 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2204 // Un-Bias for strict fp multiply/divide
2205 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2206 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2207 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2208 }
2209
2210 //---------------------------------------------------------------------------
2211 // Initialization
2212
2213 void generate_initial() {
2214 // Generates all stubs and initializes the entry points
2215
2216 //------------------------------------------------------------------------------------------------------------------------
2217 // entry points that exist in all platforms
2218 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2219 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2220 StubRoutines::_forward_exception_entry = generate_forward_exception();
2221
2222 StubRoutines::_call_stub_entry =
2223 generate_call_stub(StubRoutines::_call_stub_return_address);
2224 // is referenced by megamorphic call
2225 StubRoutines::_catch_exception_entry = generate_catch_exception();
2226
2227 // These are currently used by Solaris/Intel
2228 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2229
2230 StubRoutines::_handler_for_unsafe_access_entry =
2231 generate_handler_for_unsafe_access();
2232
2233 // platform dependent
2234 create_control_words();
2235
2236 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2237 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2238 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2239 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2240 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2241 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2242 }
2243
2244
2245 void generate_all() {
2246 // Generates all stubs and initializes the entry points
2247
2248 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2249 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2250 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2251 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
2252 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2253 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2254 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call), false);
2255 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2256
2257 //------------------------------------------------------------------------------------------------------------------------
2258 // entry points that are platform specific
2259
2260 // support for verify_oop (must happen after universe_init)
2261 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2262
2263 // arraycopy stubs used by compilers
2264 generate_arraycopy_stubs();
2265
2266 // generic method handle stubs
2267 if (EnableMethodHandles && SystemDictionary::MethodHandle_klass() != NULL) {
2268 for (MethodHandles::EntryKind ek = MethodHandles::_EK_FIRST;
2269 ek < MethodHandles::_EK_LIMIT;
2270 ek = MethodHandles::EntryKind(1 + (int)ek)) {
2271 StubCodeMark mark(this, "MethodHandle", MethodHandles::entry_name(ek));
2272 MethodHandles::generate_method_handle_stub(_masm, ek);
2273 }
2274 }
2275
2276 generate_math_stubs();
2277 }
2278
2279
2280 public:
2281 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2282 if (all) {
2283 generate_all();
2284 } else {
2285 generate_initial();
2286 }
2287 }
2288 }; // end class declaration
2289
2290
2291 void StubGenerator_generate(CodeBuffer* code, bool all) {
2292 StubGenerator g(code, all);
2293 }