1 /*
2 * Copyright 1999-2007 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()), (int32_t)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 // end - 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 __ push(count);
722 __ push(start);
723 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)));
724 __ addptr(rsp, 2*wordSize);
725 __ popa();
726 }
727 break;
728 case BarrierSet::CardTableModRef:
729 case BarrierSet::CardTableExtension:
730 case BarrierSet::ModRef:
731 break;
732 default :
733 ShouldNotReachHere();
734
735 }
736 }
737
738
739 //
740 // Generate a post-barrier for an array store
741 //
742 // start - starting address
743 // count - element count
744 //
745 // The two input registers are overwritten.
746 //
747 void gen_write_ref_array_post_barrier(Register start, Register count) {
748 BarrierSet* bs = Universe::heap()->barrier_set();
749 assert_different_registers(start, count);
750 switch (bs->kind()) {
751 case BarrierSet::G1SATBCT:
752 case BarrierSet::G1SATBCTLogging:
753 {
754 __ pusha(); // push registers
755 __ push(count);
756 __ push(start);
757 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)));
758 __ addptr(rsp, 2*wordSize);
759 __ popa();
760
761 }
762 break;
763
764 case BarrierSet::CardTableModRef:
765 case BarrierSet::CardTableExtension:
766 {
767 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
768 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
769
770 Label L_loop;
771 const Register end = count; // elements count; end == start+count-1
772 assert_different_registers(start, end);
773
774 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
775 __ shrptr(start, CardTableModRefBS::card_shift);
776 __ shrptr(end, CardTableModRefBS::card_shift);
777 __ subptr(end, start); // end --> count
778 __ BIND(L_loop);
779 intptr_t disp = (intptr_t) ct->byte_map_base;
780 Address cardtable(start, count, Address::times_1, disp);
781 __ movb(cardtable, 0);
782 __ decrement(count);
783 __ jcc(Assembler::greaterEqual, L_loop);
784 }
785 break;
786 case BarrierSet::ModRef:
787 break;
788 default :
789 ShouldNotReachHere();
790
791 }
792 }
793
794
795 // Copy 64 bytes chunks
796 //
797 // Inputs:
798 // from - source array address
799 // to_from - destination array address - from
800 // qword_count - 8-bytes element count, negative
801 //
802 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
803 assert( UseSSE >= 2, "supported cpu only" );
804 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
805 // Copy 64-byte chunks
806 __ jmpb(L_copy_64_bytes);
807 __ align(16);
808 __ BIND(L_copy_64_bytes_loop);
809
810 if(UseUnalignedLoadStores) {
811 __ movdqu(xmm0, Address(from, 0));
812 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
813 __ movdqu(xmm1, Address(from, 16));
814 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
815 __ movdqu(xmm2, Address(from, 32));
816 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
817 __ movdqu(xmm3, Address(from, 48));
818 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
819
820 } else {
821 __ movq(xmm0, Address(from, 0));
822 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
823 __ movq(xmm1, Address(from, 8));
824 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
825 __ movq(xmm2, Address(from, 16));
826 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
827 __ movq(xmm3, Address(from, 24));
828 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
829 __ movq(xmm4, Address(from, 32));
830 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
831 __ movq(xmm5, Address(from, 40));
832 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
833 __ movq(xmm6, Address(from, 48));
834 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
835 __ movq(xmm7, Address(from, 56));
836 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
837 }
838
839 __ addl(from, 64);
840 __ BIND(L_copy_64_bytes);
841 __ subl(qword_count, 8);
842 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
843 __ addl(qword_count, 8);
844 __ jccb(Assembler::zero, L_exit);
845 //
846 // length is too short, just copy qwords
847 //
848 __ BIND(L_copy_8_bytes);
849 __ movq(xmm0, Address(from, 0));
850 __ movq(Address(from, to_from, Address::times_1), xmm0);
851 __ addl(from, 8);
852 __ decrement(qword_count);
853 __ jcc(Assembler::greater, L_copy_8_bytes);
854 __ BIND(L_exit);
855 }
856
857 // Copy 64 bytes chunks
858 //
859 // Inputs:
860 // from - source array address
861 // to_from - destination array address - from
862 // qword_count - 8-bytes element count, negative
863 //
864 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
865 assert( VM_Version::supports_mmx(), "supported cpu only" );
866 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
867 // Copy 64-byte chunks
868 __ jmpb(L_copy_64_bytes);
869 __ align(16);
870 __ BIND(L_copy_64_bytes_loop);
871 __ movq(mmx0, Address(from, 0));
872 __ movq(mmx1, Address(from, 8));
873 __ movq(mmx2, Address(from, 16));
874 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
875 __ movq(mmx3, Address(from, 24));
876 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
877 __ movq(mmx4, Address(from, 32));
878 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
879 __ movq(mmx5, Address(from, 40));
880 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
881 __ movq(mmx6, Address(from, 48));
882 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
883 __ movq(mmx7, Address(from, 56));
884 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
885 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
886 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
887 __ addptr(from, 64);
888 __ BIND(L_copy_64_bytes);
889 __ subl(qword_count, 8);
890 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
891 __ addl(qword_count, 8);
892 __ jccb(Assembler::zero, L_exit);
893 //
894 // length is too short, just copy qwords
895 //
896 __ BIND(L_copy_8_bytes);
897 __ movq(mmx0, Address(from, 0));
898 __ movq(Address(from, to_from, Address::times_1), mmx0);
899 __ addptr(from, 8);
900 __ decrement(qword_count);
901 __ jcc(Assembler::greater, L_copy_8_bytes);
902 __ BIND(L_exit);
903 __ emms();
904 }
905
906 address generate_disjoint_copy(BasicType t, bool aligned,
907 Address::ScaleFactor sf,
908 address* entry, const char *name) {
909 __ align(CodeEntryAlignment);
910 StubCodeMark mark(this, "StubRoutines", name);
911 address start = __ pc();
912
913 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
914 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
915
916 int shift = Address::times_ptr - sf;
917
918 const Register from = rsi; // source array address
919 const Register to = rdi; // destination array address
920 const Register count = rcx; // elements count
921 const Register to_from = to; // (to - from)
922 const Register saved_to = rdx; // saved destination array address
923
924 __ enter(); // required for proper stackwalking of RuntimeStub frame
925 __ push(rsi);
926 __ push(rdi);
927 __ movptr(from , Address(rsp, 12+ 4));
928 __ movptr(to , Address(rsp, 12+ 8));
929 __ movl(count, Address(rsp, 12+ 12));
930 if (t == T_OBJECT) {
931 __ testl(count, count);
932 __ jcc(Assembler::zero, L_0_count);
933 gen_write_ref_array_pre_barrier(to, count);
934 __ mov(saved_to, to); // save 'to'
935 }
936
937 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
938 BLOCK_COMMENT("Entry:");
939
940 __ subptr(to, from); // to --> to_from
941 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
942 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
943 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
944 // align source address at 4 bytes address boundary
945 if (t == T_BYTE) {
946 // One byte misalignment happens only for byte arrays
947 __ testl(from, 1);
948 __ jccb(Assembler::zero, L_skip_align1);
949 __ movb(rax, Address(from, 0));
950 __ movb(Address(from, to_from, Address::times_1, 0), rax);
951 __ increment(from);
952 __ decrement(count);
953 __ BIND(L_skip_align1);
954 }
955 // Two bytes misalignment happens only for byte and short (char) arrays
956 __ testl(from, 2);
957 __ jccb(Assembler::zero, L_skip_align2);
958 __ movw(rax, Address(from, 0));
959 __ movw(Address(from, to_from, Address::times_1, 0), rax);
960 __ addptr(from, 2);
961 __ subl(count, 1<<(shift-1));
962 __ BIND(L_skip_align2);
963 }
964 if (!VM_Version::supports_mmx()) {
965 __ mov(rax, count); // save 'count'
966 __ shrl(count, shift); // bytes count
967 __ addptr(to_from, from);// restore 'to'
968 __ rep_mov();
969 __ subptr(to_from, from);// restore 'to_from'
970 __ mov(count, rax); // restore 'count'
971 __ jmpb(L_copy_2_bytes); // all dwords were copied
972 } else {
973 if (!UseUnalignedLoadStores) {
974 // align to 8 bytes, we know we are 4 byte aligned to start
975 __ testptr(from, 4);
976 __ jccb(Assembler::zero, L_copy_64_bytes);
977 __ movl(rax, Address(from, 0));
978 __ movl(Address(from, to_from, Address::times_1, 0), rax);
979 __ addptr(from, 4);
980 __ subl(count, 1<<shift);
981 }
982 __ BIND(L_copy_64_bytes);
983 __ mov(rax, count);
984 __ shrl(rax, shift+1); // 8 bytes chunk count
985 //
986 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
987 //
988 if (UseXMMForArrayCopy) {
989 xmm_copy_forward(from, to_from, rax);
990 } else {
991 mmx_copy_forward(from, to_from, rax);
992 }
993 }
994 // copy tailing dword
995 __ BIND(L_copy_4_bytes);
996 __ testl(count, 1<<shift);
997 __ jccb(Assembler::zero, L_copy_2_bytes);
998 __ movl(rax, Address(from, 0));
999 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1000 if (t == T_BYTE || t == T_SHORT) {
1001 __ addptr(from, 4);
1002 __ BIND(L_copy_2_bytes);
1003 // copy tailing word
1004 __ testl(count, 1<<(shift-1));
1005 __ jccb(Assembler::zero, L_copy_byte);
1006 __ movw(rax, Address(from, 0));
1007 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1008 if (t == T_BYTE) {
1009 __ addptr(from, 2);
1010 __ BIND(L_copy_byte);
1011 // copy tailing byte
1012 __ testl(count, 1);
1013 __ jccb(Assembler::zero, L_exit);
1014 __ movb(rax, Address(from, 0));
1015 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1016 __ BIND(L_exit);
1017 } else {
1018 __ BIND(L_copy_byte);
1019 }
1020 } else {
1021 __ BIND(L_copy_2_bytes);
1022 }
1023
1024 if (t == T_OBJECT) {
1025 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1026 __ mov(to, saved_to); // restore 'to'
1027 gen_write_ref_array_post_barrier(to, count);
1028 __ BIND(L_0_count);
1029 }
1030 inc_copy_counter_np(t);
1031 __ pop(rdi);
1032 __ pop(rsi);
1033 __ leave(); // required for proper stackwalking of RuntimeStub frame
1034 __ xorptr(rax, rax); // return 0
1035 __ ret(0);
1036 return start;
1037 }
1038
1039
1040 address generate_conjoint_copy(BasicType t, bool aligned,
1041 Address::ScaleFactor sf,
1042 address nooverlap_target,
1043 address* entry, const char *name) {
1044 __ align(CodeEntryAlignment);
1045 StubCodeMark mark(this, "StubRoutines", name);
1046 address start = __ pc();
1047
1048 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1049 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1050
1051 int shift = Address::times_ptr - sf;
1052
1053 const Register src = rax; // source array address
1054 const Register dst = rdx; // destination array address
1055 const Register from = rsi; // source array address
1056 const Register to = rdi; // destination array address
1057 const Register count = rcx; // elements count
1058 const Register end = rax; // array end address
1059
1060 __ enter(); // required for proper stackwalking of RuntimeStub frame
1061 __ push(rsi);
1062 __ push(rdi);
1063 __ movptr(src , Address(rsp, 12+ 4)); // from
1064 __ movptr(dst , Address(rsp, 12+ 8)); // to
1065 __ movl2ptr(count, Address(rsp, 12+12)); // count
1066 if (t == T_OBJECT) {
1067 gen_write_ref_array_pre_barrier(dst, count);
1068 }
1069
1070 if (entry != NULL) {
1071 *entry = __ pc(); // Entry point from generic arraycopy stub.
1072 BLOCK_COMMENT("Entry:");
1073 }
1074
1075 if (t == T_OBJECT) {
1076 __ testl(count, count);
1077 __ jcc(Assembler::zero, L_0_count);
1078 }
1079 __ mov(from, src);
1080 __ mov(to , dst);
1081
1082 // arrays overlap test
1083 RuntimeAddress nooverlap(nooverlap_target);
1084 __ cmpptr(dst, src);
1085 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1086 __ jump_cc(Assembler::belowEqual, nooverlap);
1087 __ cmpptr(dst, end);
1088 __ jump_cc(Assembler::aboveEqual, nooverlap);
1089
1090 // copy from high to low
1091 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1092 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1093 if (t == T_BYTE || t == T_SHORT) {
1094 // Align the end of destination array at 4 bytes address boundary
1095 __ lea(end, Address(dst, count, sf, 0));
1096 if (t == T_BYTE) {
1097 // One byte misalignment happens only for byte arrays
1098 __ testl(end, 1);
1099 __ jccb(Assembler::zero, L_skip_align1);
1100 __ decrement(count);
1101 __ movb(rdx, Address(from, count, sf, 0));
1102 __ movb(Address(to, count, sf, 0), rdx);
1103 __ BIND(L_skip_align1);
1104 }
1105 // Two bytes misalignment happens only for byte and short (char) arrays
1106 __ testl(end, 2);
1107 __ jccb(Assembler::zero, L_skip_align2);
1108 __ subptr(count, 1<<(shift-1));
1109 __ movw(rdx, Address(from, count, sf, 0));
1110 __ movw(Address(to, count, sf, 0), rdx);
1111 __ BIND(L_skip_align2);
1112 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1113 __ jcc(Assembler::below, L_copy_4_bytes);
1114 }
1115
1116 if (!VM_Version::supports_mmx()) {
1117 __ std();
1118 __ mov(rax, count); // Save 'count'
1119 __ mov(rdx, to); // Save 'to'
1120 __ lea(rsi, Address(from, count, sf, -4));
1121 __ lea(rdi, Address(to , count, sf, -4));
1122 __ shrptr(count, shift); // bytes count
1123 __ rep_mov();
1124 __ cld();
1125 __ mov(count, rax); // restore 'count'
1126 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1127 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1128 __ mov(to, rdx); // restore 'to'
1129 __ jmpb(L_copy_2_bytes); // all dword were copied
1130 } else {
1131 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1132 __ testptr(end, 4);
1133 __ jccb(Assembler::zero, L_copy_8_bytes);
1134 __ subl(count, 1<<shift);
1135 __ movl(rdx, Address(from, count, sf, 0));
1136 __ movl(Address(to, count, sf, 0), rdx);
1137 __ jmpb(L_copy_8_bytes);
1138
1139 __ align(16);
1140 // Move 8 bytes
1141 __ BIND(L_copy_8_bytes_loop);
1142 if (UseXMMForArrayCopy) {
1143 __ movq(xmm0, Address(from, count, sf, 0));
1144 __ movq(Address(to, count, sf, 0), xmm0);
1145 } else {
1146 __ movq(mmx0, Address(from, count, sf, 0));
1147 __ movq(Address(to, count, sf, 0), mmx0);
1148 }
1149 __ BIND(L_copy_8_bytes);
1150 __ subl(count, 2<<shift);
1151 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1152 __ addl(count, 2<<shift);
1153 if (!UseXMMForArrayCopy) {
1154 __ emms();
1155 }
1156 }
1157 __ BIND(L_copy_4_bytes);
1158 // copy prefix qword
1159 __ testl(count, 1<<shift);
1160 __ jccb(Assembler::zero, L_copy_2_bytes);
1161 __ movl(rdx, Address(from, count, sf, -4));
1162 __ movl(Address(to, count, sf, -4), rdx);
1163
1164 if (t == T_BYTE || t == T_SHORT) {
1165 __ subl(count, (1<<shift));
1166 __ BIND(L_copy_2_bytes);
1167 // copy prefix dword
1168 __ testl(count, 1<<(shift-1));
1169 __ jccb(Assembler::zero, L_copy_byte);
1170 __ movw(rdx, Address(from, count, sf, -2));
1171 __ movw(Address(to, count, sf, -2), rdx);
1172 if (t == T_BYTE) {
1173 __ subl(count, 1<<(shift-1));
1174 __ BIND(L_copy_byte);
1175 // copy prefix byte
1176 __ testl(count, 1);
1177 __ jccb(Assembler::zero, L_exit);
1178 __ movb(rdx, Address(from, 0));
1179 __ movb(Address(to, 0), rdx);
1180 __ BIND(L_exit);
1181 } else {
1182 __ BIND(L_copy_byte);
1183 }
1184 } else {
1185 __ BIND(L_copy_2_bytes);
1186 }
1187 if (t == T_OBJECT) {
1188 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1189 gen_write_ref_array_post_barrier(to, count);
1190 __ BIND(L_0_count);
1191 }
1192 inc_copy_counter_np(t);
1193 __ pop(rdi);
1194 __ pop(rsi);
1195 __ leave(); // required for proper stackwalking of RuntimeStub frame
1196 __ xorptr(rax, rax); // return 0
1197 __ ret(0);
1198 return start;
1199 }
1200
1201
1202 address generate_disjoint_long_copy(address* entry, const char *name) {
1203 __ align(CodeEntryAlignment);
1204 StubCodeMark mark(this, "StubRoutines", name);
1205 address start = __ pc();
1206
1207 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1208 const Register from = rax; // source array address
1209 const Register to = rdx; // destination array address
1210 const Register count = rcx; // elements count
1211 const Register to_from = rdx; // (to - from)
1212
1213 __ enter(); // required for proper stackwalking of RuntimeStub frame
1214 __ movptr(from , Address(rsp, 8+0)); // from
1215 __ movptr(to , Address(rsp, 8+4)); // to
1216 __ movl2ptr(count, Address(rsp, 8+8)); // count
1217
1218 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1219 BLOCK_COMMENT("Entry:");
1220
1221 __ subptr(to, from); // to --> to_from
1222 if (VM_Version::supports_mmx()) {
1223 if (UseXMMForArrayCopy) {
1224 xmm_copy_forward(from, to_from, count);
1225 } else {
1226 mmx_copy_forward(from, to_from, count);
1227 }
1228 } else {
1229 __ jmpb(L_copy_8_bytes);
1230 __ align(16);
1231 __ BIND(L_copy_8_bytes_loop);
1232 __ fild_d(Address(from, 0));
1233 __ fistp_d(Address(from, to_from, Address::times_1));
1234 __ addptr(from, 8);
1235 __ BIND(L_copy_8_bytes);
1236 __ decrement(count);
1237 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1238 }
1239 inc_copy_counter_np(T_LONG);
1240 __ leave(); // required for proper stackwalking of RuntimeStub frame
1241 __ xorptr(rax, rax); // return 0
1242 __ ret(0);
1243 return start;
1244 }
1245
1246 address generate_conjoint_long_copy(address nooverlap_target,
1247 address* entry, const char *name) {
1248 __ align(CodeEntryAlignment);
1249 StubCodeMark mark(this, "StubRoutines", name);
1250 address start = __ pc();
1251
1252 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1253 const Register from = rax; // source array address
1254 const Register to = rdx; // destination array address
1255 const Register count = rcx; // elements count
1256 const Register end_from = rax; // source array end address
1257
1258 __ enter(); // required for proper stackwalking of RuntimeStub frame
1259 __ movptr(from , Address(rsp, 8+0)); // from
1260 __ movptr(to , Address(rsp, 8+4)); // to
1261 __ movl2ptr(count, Address(rsp, 8+8)); // count
1262
1263 *entry = __ pc(); // Entry point from generic arraycopy stub.
1264 BLOCK_COMMENT("Entry:");
1265
1266 // arrays overlap test
1267 __ cmpptr(to, from);
1268 RuntimeAddress nooverlap(nooverlap_target);
1269 __ jump_cc(Assembler::belowEqual, nooverlap);
1270 __ lea(end_from, Address(from, count, Address::times_8, 0));
1271 __ cmpptr(to, end_from);
1272 __ movptr(from, Address(rsp, 8)); // from
1273 __ jump_cc(Assembler::aboveEqual, nooverlap);
1274
1275 __ jmpb(L_copy_8_bytes);
1276
1277 __ align(16);
1278 __ BIND(L_copy_8_bytes_loop);
1279 if (VM_Version::supports_mmx()) {
1280 if (UseXMMForArrayCopy) {
1281 __ movq(xmm0, Address(from, count, Address::times_8));
1282 __ movq(Address(to, count, Address::times_8), xmm0);
1283 } else {
1284 __ movq(mmx0, Address(from, count, Address::times_8));
1285 __ movq(Address(to, count, Address::times_8), mmx0);
1286 }
1287 } else {
1288 __ fild_d(Address(from, count, Address::times_8));
1289 __ fistp_d(Address(to, count, Address::times_8));
1290 }
1291 __ BIND(L_copy_8_bytes);
1292 __ decrement(count);
1293 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1294
1295 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1296 __ emms();
1297 }
1298 inc_copy_counter_np(T_LONG);
1299 __ leave(); // required for proper stackwalking of RuntimeStub frame
1300 __ xorptr(rax, rax); // return 0
1301 __ ret(0);
1302 return start;
1303 }
1304
1305
1306 // Helper for generating a dynamic type check.
1307 // The sub_klass must be one of {rbx, rdx, rsi}.
1308 // The temp is killed.
1309 void generate_type_check(Register sub_klass,
1310 Address& super_check_offset_addr,
1311 Address& super_klass_addr,
1312 Register temp,
1313 Label* L_success_ptr, Label* L_failure_ptr) {
1314 BLOCK_COMMENT("type_check:");
1315
1316 Label L_fallthrough;
1317 bool fall_through_on_success = (L_success_ptr == NULL);
1318 if (fall_through_on_success) {
1319 L_success_ptr = &L_fallthrough;
1320 } else {
1321 L_failure_ptr = &L_fallthrough;
1322 }
1323 Label& L_success = *L_success_ptr;
1324 Label& L_failure = *L_failure_ptr;
1325
1326 assert_different_registers(sub_klass, temp);
1327
1328 // a couple of useful fields in sub_klass:
1329 int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
1330 Klass::secondary_supers_offset_in_bytes());
1331 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1332 Klass::secondary_super_cache_offset_in_bytes());
1333 Address secondary_supers_addr(sub_klass, ss_offset);
1334 Address super_cache_addr( sub_klass, sc_offset);
1335
1336 // if the pointers are equal, we are done (e.g., String[] elements)
1337 __ cmpptr(sub_klass, super_klass_addr);
1338 __ jcc(Assembler::equal, L_success);
1339
1340 // check the supertype display:
1341 __ movl2ptr(temp, super_check_offset_addr);
1342 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1343 __ movptr(temp, super_check_addr); // load displayed supertype
1344 __ cmpptr(temp, super_klass_addr); // test the super type
1345 __ jcc(Assembler::equal, L_success);
1346
1347 // if it was a primary super, we can just fail immediately
1348 __ cmpl(super_check_offset_addr, sc_offset);
1349 __ jcc(Assembler::notEqual, L_failure);
1350
1351 // Now do a linear scan of the secondary super-klass chain.
1352 // This code is rarely used, so simplicity is a virtue here.
1353 inc_counter_np(SharedRuntime::_partial_subtype_ctr);
1354 {
1355 // The repne_scan instruction uses fixed registers, which we must spill.
1356 // (We need a couple more temps in any case.)
1357 __ push(rax);
1358 __ push(rcx);
1359 __ push(rdi);
1360 assert_different_registers(sub_klass, rax, rcx, rdi);
1361
1362 __ movptr(rdi, secondary_supers_addr);
1363 // Load the array length.
1364 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
1365 // Skip to start of data.
1366 __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1367 // Scan rcx words at [edi] for occurance of rax,
1368 // Set NZ/Z based on last compare
1369 __ movptr(rax, super_klass_addr);
1370 __ repne_scan();
1371
1372 // Unspill the temp. registers:
1373 __ pop(rdi);
1374 __ pop(rcx);
1375 __ pop(rax);
1376 }
1377 __ jcc(Assembler::notEqual, L_failure);
1378
1379 // Success. Cache the super we found and proceed in triumph.
1380 __ movptr(temp, super_klass_addr); // note: rax, is dead
1381 __ movptr(super_cache_addr, temp);
1382
1383 if (!fall_through_on_success)
1384 __ jmp(L_success);
1385
1386 // Fall through on failure!
1387 __ bind(L_fallthrough);
1388 }
1389
1390 //
1391 // Generate checkcasting array copy stub
1392 //
1393 // Input:
1394 // 4(rsp) - source array address
1395 // 8(rsp) - destination array address
1396 // 12(rsp) - element count, can be zero
1397 // 16(rsp) - size_t ckoff (super_check_offset)
1398 // 20(rsp) - oop ckval (super_klass)
1399 //
1400 // Output:
1401 // rax, == 0 - success
1402 // rax, == -1^K - failure, where K is partial transfer count
1403 //
1404 address generate_checkcast_copy(const char *name, address* entry) {
1405 __ align(CodeEntryAlignment);
1406 StubCodeMark mark(this, "StubRoutines", name);
1407 address start = __ pc();
1408
1409 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1410
1411 // register use:
1412 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1413 // rdi, rsi -- element access (oop, klass)
1414 // rbx, -- temp
1415 const Register from = rax; // source array address
1416 const Register to = rdx; // destination array address
1417 const Register length = rcx; // elements count
1418 const Register elem = rdi; // each oop copied
1419 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1420 const Register temp = rbx; // lone remaining temp
1421
1422 __ enter(); // required for proper stackwalking of RuntimeStub frame
1423
1424 __ push(rsi);
1425 __ push(rdi);
1426 __ push(rbx);
1427
1428 Address from_arg(rsp, 16+ 4); // from
1429 Address to_arg(rsp, 16+ 8); // to
1430 Address length_arg(rsp, 16+12); // elements count
1431 Address ckoff_arg(rsp, 16+16); // super_check_offset
1432 Address ckval_arg(rsp, 16+20); // super_klass
1433
1434 // Load up:
1435 __ movptr(from, from_arg);
1436 __ movptr(to, to_arg);
1437 __ movl2ptr(length, length_arg);
1438
1439 *entry = __ pc(); // Entry point from generic arraycopy stub.
1440 BLOCK_COMMENT("Entry:");
1441
1442 //---------------------------------------------------------------
1443 // Assembler stub will be used for this call to arraycopy
1444 // if the two arrays are subtypes of Object[] but the
1445 // destination array type is not equal to or a supertype
1446 // of the source type. Each element must be separately
1447 // checked.
1448
1449 // Loop-invariant addresses. They are exclusive end pointers.
1450 Address end_from_addr(from, length, Address::times_ptr, 0);
1451 Address end_to_addr(to, length, Address::times_ptr, 0);
1452
1453 Register end_from = from; // re-use
1454 Register end_to = to; // re-use
1455 Register count = length; // re-use
1456
1457 // Loop-variant addresses. They assume post-incremented count < 0.
1458 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1459 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1460 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1461
1462 // Copy from low to high addresses, indexed from the end of each array.
1463 gen_write_ref_array_pre_barrier(to, count);
1464 __ lea(end_from, end_from_addr);
1465 __ lea(end_to, end_to_addr);
1466 assert(length == count, ""); // else fix next line:
1467 __ negptr(count); // negate and test the length
1468 __ jccb(Assembler::notZero, L_load_element);
1469
1470 // Empty array: Nothing to do.
1471 __ xorptr(rax, rax); // return 0 on (trivial) success
1472 __ jmp(L_done);
1473
1474 // ======== begin loop ========
1475 // (Loop is rotated; its entry is L_load_element.)
1476 // Loop control:
1477 // for (count = -count; count != 0; count++)
1478 // Base pointers src, dst are biased by 8*count,to last element.
1479 __ align(16);
1480
1481 __ BIND(L_store_element);
1482 __ movptr(to_element_addr, elem); // store the oop
1483 __ increment(count); // increment the count toward zero
1484 __ jccb(Assembler::zero, L_do_card_marks);
1485
1486 // ======== loop entry is here ========
1487 __ BIND(L_load_element);
1488 __ movptr(elem, from_element_addr); // load the oop
1489 __ testptr(elem, elem);
1490 __ jccb(Assembler::zero, L_store_element);
1491
1492 // (Could do a trick here: Remember last successful non-null
1493 // element stored and make a quick oop equality check on it.)
1494
1495 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1496 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1497 &L_store_element, NULL);
1498 // (On fall-through, we have failed the element type check.)
1499 // ======== end loop ========
1500
1501 // It was a real error; we must depend on the caller to finish the job.
1502 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1503 // Emit GC store barriers for the oops we have copied (length_arg + count),
1504 // and report their number to the caller.
1505 __ addl(count, length_arg); // transfers = (length - remaining)
1506 __ movl2ptr(rax, count); // save the value
1507 __ notptr(rax); // report (-1^K) to caller
1508 __ movptr(to, to_arg); // reload
1509 assert_different_registers(to, count, rax);
1510 gen_write_ref_array_post_barrier(to, count);
1511 __ jmpb(L_done);
1512
1513 // Come here on success only.
1514 __ BIND(L_do_card_marks);
1515 __ movl2ptr(count, length_arg);
1516 __ movptr(to, to_arg); // reload
1517 gen_write_ref_array_post_barrier(to, count);
1518 __ xorptr(rax, rax); // return 0 on success
1519
1520 // Common exit point (success or failure).
1521 __ BIND(L_done);
1522 __ pop(rbx);
1523 __ pop(rdi);
1524 __ pop(rsi);
1525 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1526 __ leave(); // required for proper stackwalking of RuntimeStub frame
1527 __ ret(0);
1528
1529 return start;
1530 }
1531
1532 //
1533 // Generate 'unsafe' array copy stub
1534 // Though just as safe as the other stubs, it takes an unscaled
1535 // size_t argument instead of an element count.
1536 //
1537 // Input:
1538 // 4(rsp) - source array address
1539 // 8(rsp) - destination array address
1540 // 12(rsp) - byte count, can be zero
1541 //
1542 // Output:
1543 // rax, == 0 - success
1544 // rax, == -1 - need to call System.arraycopy
1545 //
1546 // Examines the alignment of the operands and dispatches
1547 // to a long, int, short, or byte copy loop.
1548 //
1549 address generate_unsafe_copy(const char *name,
1550 address byte_copy_entry,
1551 address short_copy_entry,
1552 address int_copy_entry,
1553 address long_copy_entry) {
1554
1555 Label L_long_aligned, L_int_aligned, L_short_aligned;
1556
1557 __ align(CodeEntryAlignment);
1558 StubCodeMark mark(this, "StubRoutines", name);
1559 address start = __ pc();
1560
1561 const Register from = rax; // source array address
1562 const Register to = rdx; // destination array address
1563 const Register count = rcx; // elements count
1564
1565 __ enter(); // required for proper stackwalking of RuntimeStub frame
1566 __ push(rsi);
1567 __ push(rdi);
1568 Address from_arg(rsp, 12+ 4); // from
1569 Address to_arg(rsp, 12+ 8); // to
1570 Address count_arg(rsp, 12+12); // byte count
1571
1572 // Load up:
1573 __ movptr(from , from_arg);
1574 __ movptr(to , to_arg);
1575 __ movl2ptr(count, count_arg);
1576
1577 // bump this on entry, not on exit:
1578 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1579
1580 const Register bits = rsi;
1581 __ mov(bits, from);
1582 __ orptr(bits, to);
1583 __ orptr(bits, count);
1584
1585 __ testl(bits, BytesPerLong-1);
1586 __ jccb(Assembler::zero, L_long_aligned);
1587
1588 __ testl(bits, BytesPerInt-1);
1589 __ jccb(Assembler::zero, L_int_aligned);
1590
1591 __ testl(bits, BytesPerShort-1);
1592 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1593
1594 __ BIND(L_short_aligned);
1595 __ shrptr(count, LogBytesPerShort); // size => short_count
1596 __ movl(count_arg, count); // update 'count'
1597 __ jump(RuntimeAddress(short_copy_entry));
1598
1599 __ BIND(L_int_aligned);
1600 __ shrptr(count, LogBytesPerInt); // size => int_count
1601 __ movl(count_arg, count); // update 'count'
1602 __ jump(RuntimeAddress(int_copy_entry));
1603
1604 __ BIND(L_long_aligned);
1605 __ shrptr(count, LogBytesPerLong); // size => qword_count
1606 __ movl(count_arg, count); // update 'count'
1607 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1608 __ pop(rsi);
1609 __ jump(RuntimeAddress(long_copy_entry));
1610
1611 return start;
1612 }
1613
1614
1615 // Perform range checks on the proposed arraycopy.
1616 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1617 void arraycopy_range_checks(Register src,
1618 Register src_pos,
1619 Register dst,
1620 Register dst_pos,
1621 Address& length,
1622 Label& L_failed) {
1623 BLOCK_COMMENT("arraycopy_range_checks:");
1624 const Register src_end = src_pos; // source array end position
1625 const Register dst_end = dst_pos; // destination array end position
1626 __ addl(src_end, length); // src_pos + length
1627 __ addl(dst_end, length); // dst_pos + length
1628
1629 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1630 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1631 __ jcc(Assembler::above, L_failed);
1632
1633 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1634 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1635 __ jcc(Assembler::above, L_failed);
1636
1637 BLOCK_COMMENT("arraycopy_range_checks done");
1638 }
1639
1640
1641 //
1642 // Generate generic array copy stubs
1643 //
1644 // Input:
1645 // 4(rsp) - src oop
1646 // 8(rsp) - src_pos
1647 // 12(rsp) - dst oop
1648 // 16(rsp) - dst_pos
1649 // 20(rsp) - element count
1650 //
1651 // Output:
1652 // rax, == 0 - success
1653 // rax, == -1^K - failure, where K is partial transfer count
1654 //
1655 address generate_generic_copy(const char *name,
1656 address entry_jbyte_arraycopy,
1657 address entry_jshort_arraycopy,
1658 address entry_jint_arraycopy,
1659 address entry_oop_arraycopy,
1660 address entry_jlong_arraycopy,
1661 address entry_checkcast_arraycopy) {
1662 Label L_failed, L_failed_0, L_objArray;
1663
1664 { int modulus = CodeEntryAlignment;
1665 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1666 int advance = target - (__ offset() % modulus);
1667 if (advance < 0) advance += modulus;
1668 if (advance > 0) __ nop(advance);
1669 }
1670 StubCodeMark mark(this, "StubRoutines", name);
1671
1672 // Short-hop target to L_failed. Makes for denser prologue code.
1673 __ BIND(L_failed_0);
1674 __ jmp(L_failed);
1675 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1676
1677 __ align(CodeEntryAlignment);
1678 address start = __ pc();
1679
1680 __ enter(); // required for proper stackwalking of RuntimeStub frame
1681 __ push(rsi);
1682 __ push(rdi);
1683
1684 // bump this on entry, not on exit:
1685 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1686
1687 // Input values
1688 Address SRC (rsp, 12+ 4);
1689 Address SRC_POS (rsp, 12+ 8);
1690 Address DST (rsp, 12+12);
1691 Address DST_POS (rsp, 12+16);
1692 Address LENGTH (rsp, 12+20);
1693
1694 //-----------------------------------------------------------------------
1695 // Assembler stub will be used for this call to arraycopy
1696 // if the following conditions are met:
1697 //
1698 // (1) src and dst must not be null.
1699 // (2) src_pos must not be negative.
1700 // (3) dst_pos must not be negative.
1701 // (4) length must not be negative.
1702 // (5) src klass and dst klass should be the same and not NULL.
1703 // (6) src and dst should be arrays.
1704 // (7) src_pos + length must not exceed length of src.
1705 // (8) dst_pos + length must not exceed length of dst.
1706 //
1707
1708 const Register src = rax; // source array oop
1709 const Register src_pos = rsi;
1710 const Register dst = rdx; // destination array oop
1711 const Register dst_pos = rdi;
1712 const Register length = rcx; // transfer count
1713
1714 // if (src == NULL) return -1;
1715 __ movptr(src, SRC); // src oop
1716 __ testptr(src, src);
1717 __ jccb(Assembler::zero, L_failed_0);
1718
1719 // if (src_pos < 0) return -1;
1720 __ movl2ptr(src_pos, SRC_POS); // src_pos
1721 __ testl(src_pos, src_pos);
1722 __ jccb(Assembler::negative, L_failed_0);
1723
1724 // if (dst == NULL) return -1;
1725 __ movptr(dst, DST); // dst oop
1726 __ testptr(dst, dst);
1727 __ jccb(Assembler::zero, L_failed_0);
1728
1729 // if (dst_pos < 0) return -1;
1730 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1731 __ testl(dst_pos, dst_pos);
1732 __ jccb(Assembler::negative, L_failed_0);
1733
1734 // if (length < 0) return -1;
1735 __ movl2ptr(length, LENGTH); // length
1736 __ testl(length, length);
1737 __ jccb(Assembler::negative, L_failed_0);
1738
1739 // if (src->klass() == NULL) return -1;
1740 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1741 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1742 const Register rcx_src_klass = rcx; // array klass
1743 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1744
1745 #ifdef ASSERT
1746 // assert(src->klass() != NULL);
1747 BLOCK_COMMENT("assert klasses not null");
1748 { Label L1, L2;
1749 __ testptr(rcx_src_klass, rcx_src_klass);
1750 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1751 __ bind(L1);
1752 __ stop("broken null klass");
1753 __ bind(L2);
1754 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1755 __ jccb(Assembler::equal, L1); // this would be broken also
1756 BLOCK_COMMENT("assert done");
1757 }
1758 #endif //ASSERT
1759
1760 // Load layout helper (32-bits)
1761 //
1762 // |array_tag| | header_size | element_type | |log2_element_size|
1763 // 32 30 24 16 8 2 0
1764 //
1765 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1766 //
1767
1768 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1769 Klass::layout_helper_offset_in_bytes();
1770 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1771
1772 // Handle objArrays completely differently...
1773 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1774 __ cmpl(src_klass_lh_addr, objArray_lh);
1775 __ jcc(Assembler::equal, L_objArray);
1776
1777 // if (src->klass() != dst->klass()) return -1;
1778 __ cmpptr(rcx_src_klass, dst_klass_addr);
1779 __ jccb(Assembler::notEqual, L_failed_0);
1780
1781 const Register rcx_lh = rcx; // layout helper
1782 assert(rcx_lh == rcx_src_klass, "known alias");
1783 __ movl(rcx_lh, src_klass_lh_addr);
1784
1785 // if (!src->is_Array()) return -1;
1786 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1787 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1788
1789 // At this point, it is known to be a typeArray (array_tag 0x3).
1790 #ifdef ASSERT
1791 { Label L;
1792 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1793 __ jcc(Assembler::greaterEqual, L); // signed cmp
1794 __ stop("must be a primitive array");
1795 __ bind(L);
1796 }
1797 #endif
1798
1799 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1800 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1801
1802 // typeArrayKlass
1803 //
1804 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1805 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1806 //
1807 const Register rsi_offset = rsi; // array offset
1808 const Register src_array = src; // src array offset
1809 const Register dst_array = dst; // dst array offset
1810 const Register rdi_elsize = rdi; // log2 element size
1811
1812 __ mov(rsi_offset, rcx_lh);
1813 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1814 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1815 __ addptr(src_array, rsi_offset); // src array offset
1816 __ addptr(dst_array, rsi_offset); // dst array offset
1817 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1818
1819 // next registers should be set before the jump to corresponding stub
1820 const Register from = src; // source array address
1821 const Register to = dst; // destination array address
1822 const Register count = rcx; // elements count
1823 // some of them should be duplicated on stack
1824 #define FROM Address(rsp, 12+ 4)
1825 #define TO Address(rsp, 12+ 8) // Not used now
1826 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1827
1828 BLOCK_COMMENT("scale indexes to element size");
1829 __ movl2ptr(rsi, SRC_POS); // src_pos
1830 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1831 assert(src_array == from, "");
1832 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1833 __ movl2ptr(rdi, DST_POS); // dst_pos
1834 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1835 assert(dst_array == to, "");
1836 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1837 __ movptr(FROM, from); // src_addr
1838 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1839 __ movl2ptr(count, LENGTH); // elements count
1840
1841 BLOCK_COMMENT("choose copy loop based on element size");
1842 __ cmpl(rdi_elsize, 0);
1843
1844 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1845 __ cmpl(rdi_elsize, LogBytesPerShort);
1846 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1847 __ cmpl(rdi_elsize, LogBytesPerInt);
1848 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1849 #ifdef ASSERT
1850 __ cmpl(rdi_elsize, LogBytesPerLong);
1851 __ jccb(Assembler::notEqual, L_failed);
1852 #endif
1853 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1854 __ pop(rsi);
1855 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1856
1857 __ BIND(L_failed);
1858 __ xorptr(rax, rax);
1859 __ notptr(rax); // return -1
1860 __ pop(rdi);
1861 __ pop(rsi);
1862 __ leave(); // required for proper stackwalking of RuntimeStub frame
1863 __ ret(0);
1864
1865 // objArrayKlass
1866 __ BIND(L_objArray);
1867 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1868
1869 Label L_plain_copy, L_checkcast_copy;
1870 // test array classes for subtyping
1871 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1872 __ jccb(Assembler::notEqual, L_checkcast_copy);
1873
1874 // Identically typed arrays can be copied without element-wise checks.
1875 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1876 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1877
1878 __ BIND(L_plain_copy);
1879 __ movl2ptr(count, LENGTH); // elements count
1880 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1881 __ lea(from, Address(src, src_pos, Address::times_ptr,
1882 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1883 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1884 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1885 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1886 __ movptr(FROM, from); // src_addr
1887 __ movptr(TO, to); // dst_addr
1888 __ movl(COUNT, count); // count
1889 __ jump(RuntimeAddress(entry_oop_arraycopy));
1890
1891 __ BIND(L_checkcast_copy);
1892 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1893 {
1894 // Handy offsets:
1895 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1896 objArrayKlass::element_klass_offset_in_bytes());
1897 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1898 Klass::super_check_offset_offset_in_bytes());
1899
1900 Register rsi_dst_klass = rsi;
1901 Register rdi_temp = rdi;
1902 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1903 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1904 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1905
1906 // Before looking at dst.length, make sure dst is also an objArray.
1907 __ movptr(rsi_dst_klass, dst_klass_addr);
1908 __ cmpl(dst_klass_lh_addr, objArray_lh);
1909 __ jccb(Assembler::notEqual, L_failed);
1910
1911 // It is safe to examine both src.length and dst.length.
1912 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1913 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1914 // (Now src_pos and dst_pos are killed, but not src and dst.)
1915
1916 // We'll need this temp (don't forget to pop it after the type check).
1917 __ push(rbx);
1918 Register rbx_src_klass = rbx;
1919
1920 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1921 __ movptr(rsi_dst_klass, dst_klass_addr);
1922 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1923 Label L_fail_array_check;
1924 generate_type_check(rbx_src_klass,
1925 super_check_offset_addr, dst_klass_addr,
1926 rdi_temp, NULL, &L_fail_array_check);
1927 // (On fall-through, we have passed the array type check.)
1928 __ pop(rbx);
1929 __ jmp(L_plain_copy);
1930
1931 __ BIND(L_fail_array_check);
1932 // Reshuffle arguments so we can call checkcast_arraycopy:
1933
1934 // match initial saves for checkcast_arraycopy
1935 // push(rsi); // already done; see above
1936 // push(rdi); // already done; see above
1937 // push(rbx); // already done; see above
1938
1939 // Marshal outgoing arguments now, freeing registers.
1940 Address from_arg(rsp, 16+ 4); // from
1941 Address to_arg(rsp, 16+ 8); // to
1942 Address length_arg(rsp, 16+12); // elements count
1943 Address ckoff_arg(rsp, 16+16); // super_check_offset
1944 Address ckval_arg(rsp, 16+20); // super_klass
1945
1946 Address SRC_POS_arg(rsp, 16+ 8);
1947 Address DST_POS_arg(rsp, 16+16);
1948 Address LENGTH_arg(rsp, 16+20);
1949 // push rbx, changed the incoming offsets (why not just use rbp,??)
1950 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1951
1952 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1953 __ movl2ptr(length, LENGTH_arg); // reload elements count
1954 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1955 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1956
1957 __ movptr(ckval_arg, rbx); // destination element type
1958 __ movl(rbx, Address(rbx, sco_offset));
1959 __ movl(ckoff_arg, rbx); // corresponding class check offset
1960
1961 __ movl(length_arg, length); // outgoing length argument
1962
1963 __ lea(from, Address(src, src_pos, Address::times_ptr,
1964 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1965 __ movptr(from_arg, from);
1966
1967 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1968 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1969 __ movptr(to_arg, to);
1970 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1971 }
1972
1973 return start;
1974 }
1975
1976 void generate_arraycopy_stubs() {
1977 address entry;
1978 address entry_jbyte_arraycopy;
1979 address entry_jshort_arraycopy;
1980 address entry_jint_arraycopy;
1981 address entry_oop_arraycopy;
1982 address entry_jlong_arraycopy;
1983 address entry_checkcast_arraycopy;
1984
1985 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1986 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1987 "arrayof_jbyte_disjoint_arraycopy");
1988 StubRoutines::_arrayof_jbyte_arraycopy =
1989 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1990 NULL, "arrayof_jbyte_arraycopy");
1991 StubRoutines::_jbyte_disjoint_arraycopy =
1992 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1993 "jbyte_disjoint_arraycopy");
1994 StubRoutines::_jbyte_arraycopy =
1995 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1996 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1997
1998 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1999 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2000 "arrayof_jshort_disjoint_arraycopy");
2001 StubRoutines::_arrayof_jshort_arraycopy =
2002 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2003 NULL, "arrayof_jshort_arraycopy");
2004 StubRoutines::_jshort_disjoint_arraycopy =
2005 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2006 "jshort_disjoint_arraycopy");
2007 StubRoutines::_jshort_arraycopy =
2008 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2009 &entry_jshort_arraycopy, "jshort_arraycopy");
2010
2011 // Next arrays are always aligned on 4 bytes at least.
2012 StubRoutines::_jint_disjoint_arraycopy =
2013 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2014 "jint_disjoint_arraycopy");
2015 StubRoutines::_jint_arraycopy =
2016 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2017 &entry_jint_arraycopy, "jint_arraycopy");
2018
2019 StubRoutines::_oop_disjoint_arraycopy =
2020 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2021 "oop_disjoint_arraycopy");
2022 StubRoutines::_oop_arraycopy =
2023 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2024 &entry_oop_arraycopy, "oop_arraycopy");
2025
2026 StubRoutines::_jlong_disjoint_arraycopy =
2027 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2028 StubRoutines::_jlong_arraycopy =
2029 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2030 "jlong_arraycopy");
2031
2032 StubRoutines::_arrayof_jint_disjoint_arraycopy =
2033 StubRoutines::_jint_disjoint_arraycopy;
2034 StubRoutines::_arrayof_oop_disjoint_arraycopy =
2035 StubRoutines::_oop_disjoint_arraycopy;
2036 StubRoutines::_arrayof_jlong_disjoint_arraycopy =
2037 StubRoutines::_jlong_disjoint_arraycopy;
2038
2039 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2040 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2041 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2042
2043 StubRoutines::_checkcast_arraycopy =
2044 generate_checkcast_copy("checkcast_arraycopy",
2045 &entry_checkcast_arraycopy);
2046
2047 StubRoutines::_unsafe_arraycopy =
2048 generate_unsafe_copy("unsafe_arraycopy",
2049 entry_jbyte_arraycopy,
2050 entry_jshort_arraycopy,
2051 entry_jint_arraycopy,
2052 entry_jlong_arraycopy);
2053
2054 StubRoutines::_generic_arraycopy =
2055 generate_generic_copy("generic_arraycopy",
2056 entry_jbyte_arraycopy,
2057 entry_jshort_arraycopy,
2058 entry_jint_arraycopy,
2059 entry_oop_arraycopy,
2060 entry_jlong_arraycopy,
2061 entry_checkcast_arraycopy);
2062 }
2063
2064 public:
2065 // Information about frame layout at time of blocking runtime call.
2066 // Note that we only have to preserve callee-saved registers since
2067 // the compilers are responsible for supplying a continuation point
2068 // if they expect all registers to be preserved.
2069 enum layout {
2070 thread_off, // last_java_sp
2071 rbp_off, // callee saved register
2072 ret_pc,
2073 framesize
2074 };
2075
2076 private:
2077
2078 #undef __
2079 #define __ masm->
2080
2081 //------------------------------------------------------------------------------------------------------------------------
2082 // Continuation point for throwing of implicit exceptions that are not handled in
2083 // the current activation. Fabricates an exception oop and initiates normal
2084 // exception dispatching in this frame.
2085 //
2086 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2087 // This is no longer true after adapter frames were removed but could possibly
2088 // be brought back in the future if the interpreter code was reworked and it
2089 // was deemed worthwhile. The comment below was left to describe what must
2090 // happen here if callee saves were resurrected. As it stands now this stub
2091 // could actually be a vanilla BufferBlob and have now oopMap at all.
2092 // Since it doesn't make much difference we've chosen to leave it the
2093 // way it was in the callee save days and keep the comment.
2094
2095 // If we need to preserve callee-saved values we need a callee-saved oop map and
2096 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2097 // If the compiler needs all registers to be preserved between the fault
2098 // point and the exception handler then it must assume responsibility for that in
2099 // AbstractCompiler::continuation_for_implicit_null_exception or
2100 // continuation_for_implicit_division_by_zero_exception. All other implicit
2101 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2102 // either at call sites or otherwise assume that stack unwinding will be initiated,
2103 // so caller saved registers were assumed volatile in the compiler.
2104 address generate_throw_exception(const char* name, address runtime_entry,
2105 bool restore_saved_exception_pc) {
2106
2107 int insts_size = 256;
2108 int locs_size = 32;
2109
2110 CodeBuffer code(name, insts_size, locs_size);
2111 OopMapSet* oop_maps = new OopMapSet();
2112 MacroAssembler* masm = new MacroAssembler(&code);
2113
2114 address start = __ pc();
2115
2116 // This is an inlined and slightly modified version of call_VM
2117 // which has the ability to fetch the return PC out of
2118 // thread-local storage and also sets up last_Java_sp slightly
2119 // differently than the real call_VM
2120 Register java_thread = rbx;
2121 __ get_thread(java_thread);
2122 if (restore_saved_exception_pc) {
2123 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2124 __ push(rax);
2125 }
2126
2127 __ enter(); // required for proper stackwalking of RuntimeStub frame
2128
2129 // pc and rbp, already pushed
2130 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2131
2132 // Frame is now completed as far as size and linkage.
2133
2134 int frame_complete = __ pc() - start;
2135
2136 // push java thread (becomes first argument of C function)
2137 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2138
2139 // Set up last_Java_sp and last_Java_fp
2140 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2141
2142 // Call runtime
2143 BLOCK_COMMENT("call runtime_entry");
2144 __ call(RuntimeAddress(runtime_entry));
2145 // Generate oop map
2146 OopMap* map = new OopMap(framesize, 0);
2147 oop_maps->add_gc_map(__ pc() - start, map);
2148
2149 // restore the thread (cannot use the pushed argument since arguments
2150 // may be overwritten by C code generated by an optimizing compiler);
2151 // however can use the register value directly if it is callee saved.
2152 __ get_thread(java_thread);
2153
2154 __ reset_last_Java_frame(java_thread, true, false);
2155
2156 __ leave(); // required for proper stackwalking of RuntimeStub frame
2157
2158 // check for pending exceptions
2159 #ifdef ASSERT
2160 Label L;
2161 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2162 __ jcc(Assembler::notEqual, L);
2163 __ should_not_reach_here();
2164 __ bind(L);
2165 #endif /* ASSERT */
2166 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2167
2168
2169 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2170 return stub->entry_point();
2171 }
2172
2173
2174 void create_control_words() {
2175 // Round to nearest, 53-bit mode, exceptions masked
2176 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2177 // Round to zero, 53-bit mode, exception mased
2178 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2179 // Round to nearest, 24-bit mode, exceptions masked
2180 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2181 // Round to nearest, 64-bit mode, exceptions masked
2182 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2183 // Round to nearest, 64-bit mode, exceptions masked
2184 StubRoutines::_mxcsr_std = 0x1F80;
2185 // Note: the following two constants are 80-bit values
2186 // layout is critical for correct loading by FPU.
2187 // Bias for strict fp multiply/divide
2188 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2189 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2190 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2191 // Un-Bias for strict fp multiply/divide
2192 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2193 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2194 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2195 }
2196
2197 //---------------------------------------------------------------------------
2198 // Initialization
2199
2200 void generate_initial() {
2201 // Generates all stubs and initializes the entry points
2202
2203 //------------------------------------------------------------------------------------------------------------------------
2204 // entry points that exist in all platforms
2205 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2206 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2207 StubRoutines::_forward_exception_entry = generate_forward_exception();
2208
2209 StubRoutines::_call_stub_entry =
2210 generate_call_stub(StubRoutines::_call_stub_return_address);
2211 // is referenced by megamorphic call
2212 StubRoutines::_catch_exception_entry = generate_catch_exception();
2213
2214 // These are currently used by Solaris/Intel
2215 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2216
2217 StubRoutines::_handler_for_unsafe_access_entry =
2218 generate_handler_for_unsafe_access();
2219
2220 // platform dependent
2221 create_control_words();
2222
2223 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2224 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2225 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2226 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2227 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2228 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2229 }
2230
2231
2232 void generate_all() {
2233 // Generates all stubs and initializes the entry points
2234
2235 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2236 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2237 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2238 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
2239 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2240 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2241 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);
2242 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2243
2244 //------------------------------------------------------------------------------------------------------------------------
2245 // entry points that are platform specific
2246
2247 // support for verify_oop (must happen after universe_init)
2248 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2249
2250 // arraycopy stubs used by compilers
2251 generate_arraycopy_stubs();
2252 }
2253
2254
2255 public:
2256 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2257 if (all) {
2258 generate_all();
2259 } else {
2260 generate_initial();
2261 }
2262 }
2263 }; // end class declaration
2264
2265
2266 void StubGenerator_generate(CodeBuffer* code, bool all) {
2267 StubGenerator g(code, all);
2268 }