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
34 #ifdef PRODUCT
35 #define BLOCK_COMMENT(str) /* nothing */
36 #else
37 #define BLOCK_COMMENT(str) __ block_comment(str)
38 #endif
39
40 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
41
42 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
43 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
44
45 // -------------------------------------------------------------------------------------------------------------------------
46 // Stub Code definitions
47
48 static address handle_unsafe_access() {
49 JavaThread* thread = JavaThread::current();
50 address pc = thread->saved_exception_pc();
51 // pc is the instruction which we must emulate
52 // doing a no-op is fine: return garbage from the load
53 // therefore, compute npc
54 address npc = Assembler::locate_next_instruction(pc);
55
56 // request an async exception
57 thread->set_pending_unsafe_access_error();
58
59 // return address of next instruction to execute
60 return npc;
61 }
62
63 class StubGenerator: public StubCodeGenerator {
64 private:
65
66 #ifdef PRODUCT
67 #define inc_counter_np(counter) (0)
68 #else
69 void inc_counter_np_(int& counter) {
70 __ increment(ExternalAddress((address)&counter));
71 }
72 #define inc_counter_np(counter) \
73 BLOCK_COMMENT("inc_counter " #counter); \
74 inc_counter_np_(counter);
75 #endif //PRODUCT
76
77 void inc_copy_counter_np(BasicType t) {
78 #ifndef PRODUCT
79 switch (t) {
80 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
81 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
82 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
83 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
84 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
85 }
86 ShouldNotReachHere();
87 #endif //PRODUCT
88 }
89
90 //------------------------------------------------------------------------------------------------------------------------
91 // Call stubs are used to call Java from C
92 //
93 // [ return_from_Java ] <--- rsp
94 // [ argument word n ]
95 // ...
96 // -N [ argument word 1 ]
97 // -7 [ Possible padding for stack alignment ]
98 // -6 [ Possible padding for stack alignment ]
99 // -5 [ Possible padding for stack alignment ]
100 // -4 [ mxcsr save ] <--- rsp_after_call
101 // -3 [ saved rbx, ]
102 // -2 [ saved rsi ]
103 // -1 [ saved rdi ]
104 // 0 [ saved rbp, ] <--- rbp,
105 // 1 [ return address ]
106 // 2 [ ptr. to call wrapper ]
107 // 3 [ result ]
108 // 4 [ result_type ]
109 // 5 [ method ]
110 // 6 [ entry_point ]
111 // 7 [ parameters ]
112 // 8 [ parameter_size ]
113 // 9 [ thread ]
114
115
116 address generate_call_stub(address& return_address) {
117 StubCodeMark mark(this, "StubRoutines", "call_stub");
118 address start = __ pc();
119
120 // stub code parameters / addresses
121 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
122 bool sse_save = false;
123 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
124 const int locals_count_in_bytes (4*wordSize);
125 const Address mxcsr_save (rbp, -4 * wordSize);
126 const Address saved_rbx (rbp, -3 * wordSize);
127 const Address saved_rsi (rbp, -2 * wordSize);
128 const Address saved_rdi (rbp, -1 * wordSize);
129 const Address result (rbp, 3 * wordSize);
130 const Address result_type (rbp, 4 * wordSize);
131 const Address method (rbp, 5 * wordSize);
132 const Address entry_point (rbp, 6 * wordSize);
133 const Address parameters (rbp, 7 * wordSize);
134 const Address parameter_size(rbp, 8 * wordSize);
135 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
136 sse_save = UseSSE > 0;
137
138 // stub code
139 __ enter();
140 __ movl(rcx, parameter_size); // parameter counter
141 __ shll(rcx, Interpreter::logStackElementSize()); // convert parameter count to bytes
142 __ addl(rcx, locals_count_in_bytes); // reserve space for register saves
143 __ subl(rsp, rcx);
144 __ andl(rsp, -(StackAlignmentInBytes)); // Align stack
145
146 // save rdi, rsi, & rbx, according to C calling conventions
147 __ movl(saved_rdi, rdi);
148 __ movl(saved_rsi, rsi);
149 __ movl(saved_rbx, rbx);
150 // save and initialize %mxcsr
151 if (sse_save) {
152 Label skip_ldmx;
153 __ stmxcsr(mxcsr_save);
154 __ movl(rax, mxcsr_save);
155 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
156 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
157 __ cmp32(rax, mxcsr_std);
158 __ jcc(Assembler::equal, skip_ldmx);
159 __ ldmxcsr(mxcsr_std);
160 __ bind(skip_ldmx);
161 }
162
163 // make sure the control word is correct.
164 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
165
166 #ifdef ASSERT
167 // make sure we have no pending exceptions
168 { Label L;
169 __ movl(rcx, thread);
170 __ cmpl(Address(rcx, Thread::pending_exception_offset()), NULL_WORD);
171 __ jcc(Assembler::equal, L);
172 __ stop("StubRoutines::call_stub: entered with pending exception");
173 __ bind(L);
174 }
175 #endif
176
177 // pass parameters if any
178 BLOCK_COMMENT("pass parameters if any");
179 Label parameters_done;
180 __ movl(rcx, parameter_size); // parameter counter
181 __ testl(rcx, rcx);
182 __ jcc(Assembler::zero, parameters_done);
183
184 // parameter passing loop
185
186 Label loop;
187 // Copy Java parameters in reverse order (receiver last)
188 // Note that the argument order is inverted in the process
189 // source is rdx[rcx: N-1..0]
190 // dest is rsp[rbx: 0..N-1]
191
192 __ movl(rdx, parameters); // parameter pointer
193 __ xorl(rbx, rbx);
194
195 __ BIND(loop);
196 if (TaggedStackInterpreter) {
197 __ movl(rax, Address(rdx, rcx, Interpreter::stackElementScale(),
198 -2*wordSize)); // get tag
199 __ movl(Address(rsp, rbx, Interpreter::stackElementScale(),
200 Interpreter::expr_tag_offset_in_bytes(0)), rax); // store tag
201 }
202
203 // get parameter
204 __ movl(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
205 __ movl(Address(rsp, rbx, Interpreter::stackElementScale(),
206 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
207 __ increment(rbx);
208 __ decrement(rcx);
209 __ jcc(Assembler::notZero, loop);
210
211 // call Java function
212 __ BIND(parameters_done);
213 __ movl(rbx, method); // get methodOop
214 __ movl(rax, entry_point); // get entry_point
215 __ movl(rsi, rsp); // set sender sp
216 BLOCK_COMMENT("call Java function");
217 __ call(rax);
218
219 BLOCK_COMMENT("call_stub_return_address:");
220 return_address = __ pc();
221
222 Label common_return;
223
224 __ BIND(common_return);
225
226 // store result depending on type
227 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
228 __ movl(rdi, result);
229 Label is_long, is_float, is_double, exit;
230 __ movl(rsi, result_type);
231 __ cmpl(rsi, T_LONG);
232 __ jcc(Assembler::equal, is_long);
233 __ cmpl(rsi, T_FLOAT);
234 __ jcc(Assembler::equal, is_float);
235 __ cmpl(rsi, T_DOUBLE);
236 __ jcc(Assembler::equal, is_double);
237
238 // handle T_INT case
239 __ movl(Address(rdi, 0), rax);
240 __ BIND(exit);
241
242 // check that FPU stack is empty
243 __ verify_FPU(0, "generate_call_stub");
244
245 // pop parameters
246 __ leal(rsp, rsp_after_call);
247
248 // restore %mxcsr
249 if (sse_save) {
250 __ ldmxcsr(mxcsr_save);
251 }
252
253 // restore rdi, rsi and rbx,
254 __ movl(rbx, saved_rbx);
255 __ movl(rsi, saved_rsi);
256 __ movl(rdi, saved_rdi);
257 __ addl(rsp, 4*wordSize);
258
259 // return
260 __ popl(rbp);
261 __ ret(0);
262
263 // handle return types different from T_INT
264 __ BIND(is_long);
265 __ movl(Address(rdi, 0 * wordSize), rax);
266 __ movl(Address(rdi, 1 * wordSize), rdx);
267 __ jmp(exit);
268
269 __ BIND(is_float);
270 // interpreter uses xmm0 for return values
271 if (UseSSE >= 1) {
272 __ movflt(Address(rdi, 0), xmm0);
273 } else {
274 __ fstp_s(Address(rdi, 0));
275 }
276 __ jmp(exit);
277
278 __ BIND(is_double);
279 // interpreter uses xmm0 for return values
280 if (UseSSE >= 2) {
281 __ movdbl(Address(rdi, 0), xmm0);
282 } else {
283 __ fstp_d(Address(rdi, 0));
284 }
285 __ jmp(exit);
286
287 // If we call compiled code directly from the call stub we will
288 // need to adjust the return back to the call stub to a specialized
289 // piece of code that can handle compiled results and cleaning the fpu
290 // stack. compiled code will be set to return here instead of the
291 // return above that handles interpreter returns.
292
293 BLOCK_COMMENT("call_stub_compiled_return:");
294 StubRoutines::i486::set_call_stub_compiled_return( __ pc());
295
296 #ifdef COMPILER2
297 if (UseSSE >= 2) {
298 __ verify_FPU(0, "call_stub_compiled_return");
299 } else {
300 for (int i = 1; i < 8; i++) {
301 __ ffree(i);
302 }
303
304 // UseSSE <= 1 so double result should be left on TOS
305 __ movl(rsi, result_type);
306 __ cmpl(rsi, T_DOUBLE);
307 __ jcc(Assembler::equal, common_return);
308 if (UseSSE == 0) {
309 // UseSSE == 0 so float result should be left on TOS
310 __ cmpl(rsi, T_FLOAT);
311 __ jcc(Assembler::equal, common_return);
312 }
313 __ ffree(0);
314 }
315 #endif /* COMPILER2 */
316 __ jmp(common_return);
317
318 return start;
319 }
320
321
322 //------------------------------------------------------------------------------------------------------------------------
323 // Return point for a Java call if there's an exception thrown in Java code.
324 // The exception is caught and transformed into a pending exception stored in
325 // JavaThread that can be tested from within the VM.
326 //
327 // Note: Usually the parameters are removed by the callee. In case of an exception
328 // crossing an activation frame boundary, that is not the case if the callee
329 // is compiled code => need to setup the rsp.
330 //
331 // rax,: exception oop
332
333 address generate_catch_exception() {
334 StubCodeMark mark(this, "StubRoutines", "catch_exception");
335 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
336 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
337 address start = __ pc();
338
339 // get thread directly
340 __ movl(rcx, thread);
341 #ifdef ASSERT
342 // verify that threads correspond
343 { Label L;
344 __ get_thread(rbx);
345 __ cmpl(rbx, rcx);
346 __ jcc(Assembler::equal, L);
347 __ stop("StubRoutines::catch_exception: threads must correspond");
348 __ bind(L);
349 }
350 #endif
351 // set pending exception
352 __ verify_oop(rax);
353 __ movl(Address(rcx, Thread::pending_exception_offset()), rax );
354 __ lea(Address(rcx, Thread::exception_file_offset ()),
355 ExternalAddress((address)__FILE__));
356 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
357 // complete return to VM
358 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
359 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
360
361 return start;
362 }
363
364
365 //------------------------------------------------------------------------------------------------------------------------
366 // Continuation point for runtime calls returning with a pending exception.
367 // The pending exception check happened in the runtime or native call stub.
368 // The pending exception in Thread is converted into a Java-level exception.
369 //
370 // Contract with Java-level exception handlers:
371 // rax,: exception
372 // rdx: throwing pc
373 //
374 // NOTE: At entry of this stub, exception-pc must be on stack !!
375
376 address generate_forward_exception() {
377 StubCodeMark mark(this, "StubRoutines", "forward exception");
378 address start = __ pc();
379
380 // Upon entry, the sp points to the return address returning into Java
381 // (interpreted or compiled) code; i.e., the return address becomes the
382 // throwing pc.
383 //
384 // Arguments pushed before the runtime call are still on the stack but
385 // the exception handler will reset the stack pointer -> ignore them.
386 // A potential result in registers can be ignored as well.
387
388 #ifdef ASSERT
389 // make sure this code is only executed if there is a pending exception
390 { Label L;
391 __ get_thread(rcx);
392 __ cmpl(Address(rcx, Thread::pending_exception_offset()), NULL_WORD);
393 __ jcc(Assembler::notEqual, L);
394 __ stop("StubRoutines::forward exception: no pending exception (1)");
395 __ bind(L);
396 }
397 #endif
398
399 // compute exception handler into rbx,
400 __ movl(rax, Address(rsp, 0));
401 BLOCK_COMMENT("call exception_handler_for_return_address");
402 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rax);
403 __ movl(rbx, rax);
404
405 // setup rax, & rdx, remove return address & clear pending exception
406 __ get_thread(rcx);
407 __ popl(rdx);
408 __ movl(rax, Address(rcx, Thread::pending_exception_offset()));
409 __ movl(Address(rcx, Thread::pending_exception_offset()), NULL_WORD);
410
411 #ifdef ASSERT
412 // make sure exception is set
413 { Label L;
414 __ testl(rax, rax);
415 __ jcc(Assembler::notEqual, L);
416 __ stop("StubRoutines::forward exception: no pending exception (2)");
417 __ bind(L);
418 }
419 #endif
420
421 // continue at exception handler (return address removed)
422 // rax,: exception
423 // rbx,: exception handler
424 // rdx: throwing pc
425 __ verify_oop(rax);
426 __ jmp(rbx);
427
428 return start;
429 }
430
431
432 //----------------------------------------------------------------------------------------------------
433 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
434 //
435 // xchg exists as far back as 8086, lock needed for MP only
436 // Stack layout immediately after call:
437 //
438 // 0 [ret addr ] <--- rsp
439 // 1 [ ex ]
440 // 2 [ dest ]
441 //
442 // Result: *dest <- ex, return (old *dest)
443 //
444 // Note: win32 does not currently use this code
445
446 address generate_atomic_xchg() {
447 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
448 address start = __ pc();
449
450 __ pushl(rdx);
451 Address exchange(rsp, 2 * wordSize);
452 Address dest_addr(rsp, 3 * wordSize);
453 __ movl(rax, exchange);
454 __ movl(rdx, dest_addr);
455 __ xchg(rax, Address(rdx, 0));
456 __ popl(rdx);
457 __ ret(0);
458
459 return start;
460 }
461
462 //----------------------------------------------------------------------------------------------------
463 // Support for void verify_mxcsr()
464 //
465 // This routine is used with -Xcheck:jni to verify that native
466 // JNI code does not return to Java code without restoring the
467 // MXCSR register to our expected state.
468
469
470 address generate_verify_mxcsr() {
471 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
472 address start = __ pc();
473
474 const Address mxcsr_save(rsp, 0);
475
476 if (CheckJNICalls && UseSSE > 0 ) {
477 Label ok_ret;
478 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
479 __ pushl(rax);
480 __ subl(rsp, wordSize); // allocate a temp location
481 __ stmxcsr(mxcsr_save);
482 __ movl(rax, mxcsr_save);
483 __ andl(rax, MXCSR_MASK);
484 __ cmp32(rax, mxcsr_std);
485 __ jcc(Assembler::equal, ok_ret);
486
487 __ warn("MXCSR changed by native JNI code.");
488
489 __ ldmxcsr(mxcsr_std);
490
491 __ bind(ok_ret);
492 __ addl(rsp, wordSize);
493 __ popl(rax);
494 }
495
496 __ ret(0);
497
498 return start;
499 }
500
501
502 //---------------------------------------------------------------------------
503 // Support for void verify_fpu_cntrl_wrd()
504 //
505 // This routine is used with -Xcheck:jni to verify that native
506 // JNI code does not return to Java code without restoring the
507 // FP control word to our expected state.
508
509 address generate_verify_fpu_cntrl_wrd() {
510 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
511 address start = __ pc();
512
513 const Address fpu_cntrl_wrd_save(rsp, 0);
514
515 if (CheckJNICalls) {
516 Label ok_ret;
517 __ pushl(rax);
518 __ subl(rsp, wordSize); // allocate a temp location
519 __ fnstcw(fpu_cntrl_wrd_save);
520 __ movl(rax, fpu_cntrl_wrd_save);
521 __ andl(rax, FPU_CNTRL_WRD_MASK);
522 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
523 __ cmp32(rax, fpu_std);
524 __ jcc(Assembler::equal, ok_ret);
525
526 __ warn("Floating point control word changed by native JNI code.");
527
528 __ fldcw(fpu_std);
529
530 __ bind(ok_ret);
531 __ addl(rsp, wordSize);
532 __ popl(rax);
533 }
534
535 __ ret(0);
536
537 return start;
538 }
539
540 //---------------------------------------------------------------------------
541 // Wrapper for slow-case handling of double-to-integer conversion
542 // d2i or f2i fast case failed either because it is nan or because
543 // of under/overflow.
544 // Input: FPU TOS: float value
545 // Output: rax, (rdx): integer (long) result
546
547 address generate_d2i_wrapper(BasicType t, address fcn) {
548 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
549 address start = __ pc();
550
551 // Capture info about frame layout
552 enum layout { FPUState_off = 0,
553 rbp_off = FPUStateSizeInWords,
554 rdi_off,
555 rsi_off,
556 rcx_off,
557 rbx_off,
558 saved_argument_off,
559 saved_argument_off2, // 2nd half of double
560 framesize
561 };
562
563 assert(FPUStateSizeInWords == 27, "update stack layout");
564
565 // Save outgoing argument to stack across push_FPU_state()
566 __ subl(rsp, wordSize * 2);
567 __ fstp_d(Address(rsp, 0));
568
569 // Save CPU & FPU state
570 __ pushl(rbx);
571 __ pushl(rcx);
572 __ pushl(rsi);
573 __ pushl(rdi);
574 __ pushl(rbp);
575 __ push_FPU_state();
576
577 // push_FPU_state() resets the FP top of stack
578 // Load original double into FP top of stack
579 __ fld_d(Address(rsp, saved_argument_off * wordSize));
580 // Store double into stack as outgoing argument
581 __ subl(rsp, wordSize*2);
582 __ fst_d(Address(rsp, 0));
583
584 // Prepare FPU for doing math in C-land
585 __ empty_FPU_stack();
586 // Call the C code to massage the double. Result in EAX
587 if (t == T_INT)
588 { BLOCK_COMMENT("SharedRuntime::d2i"); }
589 else if (t == T_LONG)
590 { BLOCK_COMMENT("SharedRuntime::d2l"); }
591 __ call_VM_leaf( fcn, 2 );
592
593 // Restore CPU & FPU state
594 __ pop_FPU_state();
595 __ popl(rbp);
596 __ popl(rdi);
597 __ popl(rsi);
598 __ popl(rcx);
599 __ popl(rbx);
600 __ addl(rsp, wordSize * 2);
601
602 __ ret(0);
603
604 return start;
605 }
606
607
608 //---------------------------------------------------------------------------
609 // The following routine generates a subroutine to throw an asynchronous
610 // UnknownError when an unsafe access gets a fault that could not be
611 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
612 address generate_handler_for_unsafe_access() {
613 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
614 address start = __ pc();
615
616 __ pushl(0); // hole for return address-to-be
617 __ pushad(); // push registers
618 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
619 BLOCK_COMMENT("call handle_unsafe_access");
620 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
621 __ movl(next_pc, rax); // stuff next address
622 __ popad();
623 __ ret(0); // jump to next address
624
625 return start;
626 }
627
628
629 //----------------------------------------------------------------------------------------------------
630 // Non-destructive plausibility checks for oops
631
632 address generate_verify_oop() {
633 StubCodeMark mark(this, "StubRoutines", "verify_oop");
634 address start = __ pc();
635
636 // Incoming arguments on stack after saving rax,:
637 //
638 // [tos ]: saved rdx
639 // [tos + 1]: saved EFLAGS
640 // [tos + 2]: return address
641 // [tos + 3]: char* error message
642 // [tos + 4]: oop object to verify
643 // [tos + 5]: saved rax, - saved by caller and bashed
644
645 Label exit, error;
646 __ pushfd();
647 __ increment(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
648 __ pushl(rdx); // save rdx
649 // make sure object is 'reasonable'
650 __ movl(rax, Address(rsp, 4 * wordSize)); // get object
651 __ testl(rax, rax);
652 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
653
654 // Check if the oop is in the right area of memory
655 const int oop_mask = Universe::verify_oop_mask();
656 const int oop_bits = Universe::verify_oop_bits();
657 __ movl(rdx, rax);
658 __ andl(rdx, oop_mask);
659 __ cmpl(rdx, oop_bits);
660 __ jcc(Assembler::notZero, error);
661
662 // make sure klass is 'reasonable'
663 __ movl(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
664 __ testl(rax, rax);
665 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
666
667 // Check if the klass is in the right area of memory
668 const int klass_mask = Universe::verify_klass_mask();
669 const int klass_bits = Universe::verify_klass_bits();
670 __ movl(rdx, rax);
671 __ andl(rdx, klass_mask);
672 __ cmpl(rdx, klass_bits);
673 __ jcc(Assembler::notZero, error);
674
675 // make sure klass' klass is 'reasonable'
676 __ movl(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
677 __ testl(rax, rax);
678 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
679
680 __ movl(rdx, rax);
681 __ andl(rdx, klass_mask);
682 __ cmpl(rdx, klass_bits);
683 __ jcc(Assembler::notZero, error); // if klass not in right area
684 // of memory it is broken too.
685
686 // return if everything seems ok
687 __ bind(exit);
688 __ movl(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
689 __ popl(rdx); // restore rdx
690 __ popfd(); // restore EFLAGS
691 __ ret(3 * wordSize); // pop arguments
692
693 // handle errors
694 __ bind(error);
695 __ movl(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
696 __ popl(rdx); // get saved rdx back
697 __ popfd(); // get saved EFLAGS off stack -- will be ignored
698 __ pushad(); // push registers (eip = return address & msg are already pushed)
699 BLOCK_COMMENT("call MacroAssembler::debug");
700 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug)));
701 __ popad();
702 __ ret(3 * wordSize); // pop arguments
703 return start;
704 }
705
706 //
707 // Generate pre-barrier for array stores
708 //
709 // Input:
710 // start - starting address
711 // end - element count
712 void gen_write_ref_array_pre_barrier(Register start, Register count) {
713 assert_different_registers(start, count);
714 #if 0 // G1 only
715 BarrierSet* bs = Universe::heap()->barrier_set();
716 switch (bs->kind()) {
717 case BarrierSet::G1SATBCT:
718 case BarrierSet::G1SATBCTLogging:
719 {
720 __ pushad(); // push registers
721 __ pushl(count);
722 __ pushl(start);
723 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre));
724 __ addl(esp, wordSize * 2);
725 __ popad();
726 }
727 break;
728 case BarrierSet::CardTableModRef:
729 case BarrierSet::CardTableExtension:
730 case BarrierSet::ModRef:
731 break;
732 default :
733 ShouldNotReachHere();
734
735 }
736 #endif // 0 - G1 only
737 }
738
739
740 //
741 // Generate a post-barrier for an array store
742 //
743 // start - starting address
744 // count - element count
745 //
746 // The two input registers are overwritten.
747 //
748 void gen_write_ref_array_post_barrier(Register start, Register count) {
749 BarrierSet* bs = Universe::heap()->barrier_set();
750 assert_different_registers(start, count);
751 switch (bs->kind()) {
752 #if 0 // G1 only
753 case BarrierSet::G1SATBCT:
754 case BarrierSet::G1SATBCTLogging:
755 {
756 __ pushad(); // push registers
757 __ pushl(count);
758 __ pushl(start);
759 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post));
760 __ addl(esp, wordSize * 2);
761 __ popad();
762
763 }
764 break;
765 #endif // 0 G1 only
766
767 case BarrierSet::CardTableModRef:
768 case BarrierSet::CardTableExtension:
769 {
770 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
771 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
772
773 Label L_loop;
774 const Register end = count; // elements count; end == start+count-1
775 assert_different_registers(start, end);
776
777 __ leal(end, Address(start, count, Address::times_4, -4));
778 __ shrl(start, CardTableModRefBS::card_shift);
779 __ shrl(end, CardTableModRefBS::card_shift);
780 __ subl(end, start); // end --> count
781 __ BIND(L_loop);
782 ExternalAddress base((address)ct->byte_map_base);
783 Address index(start, count, Address::times_1, 0);
784 __ movbyte(ArrayAddress(base, index), 0);
785 __ decrement(count);
786 __ jcc(Assembler::greaterEqual, L_loop);
787 }
788 break;
789 case BarrierSet::ModRef:
790 break;
791 default :
792 ShouldNotReachHere();
793
794 }
795 }
796
797 // Copy 64 bytes chunks
798 //
799 // Inputs:
800 // from - source array address
801 // to_from - destination array address - from
802 // qword_count - 8-bytes element count, negative
803 //
804 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
805 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
806 // Copy 64-byte chunks
807 __ jmpb(L_copy_64_bytes);
808 __ align(16);
809 __ BIND(L_copy_64_bytes_loop);
810 __ movq(mmx0, Address(from, 0));
811 __ movq(mmx1, Address(from, 8));
812 __ movq(mmx2, Address(from, 16));
813 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
814 __ movq(mmx3, Address(from, 24));
815 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
816 __ movq(mmx4, Address(from, 32));
817 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
818 __ movq(mmx5, Address(from, 40));
819 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
820 __ movq(mmx6, Address(from, 48));
821 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
822 __ movq(mmx7, Address(from, 56));
823 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
824 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
825 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
826 __ addl(from, 64);
827 __ BIND(L_copy_64_bytes);
828 __ subl(qword_count, 8);
829 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
830 __ addl(qword_count, 8);
831 __ jccb(Assembler::zero, L_exit);
832 //
833 // length is too short, just copy qwords
834 //
835 __ BIND(L_copy_8_bytes);
836 __ movq(mmx0, Address(from, 0));
837 __ movq(Address(from, to_from, Address::times_1), mmx0);
838 __ addl(from, 8);
839 __ decrement(qword_count);
840 __ jcc(Assembler::greater, L_copy_8_bytes);
841 __ BIND(L_exit);
842 __ emms();
843 }
844
845 address generate_disjoint_copy(BasicType t, bool aligned,
846 Address::ScaleFactor sf,
847 address* entry, const char *name) {
848 __ align(CodeEntryAlignment);
849 StubCodeMark mark(this, "StubRoutines", name);
850 address start = __ pc();
851
852 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
853 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
854
855 int shift = Address::times_4 - sf;
856
857 const Register from = rsi; // source array address
858 const Register to = rdi; // destination array address
859 const Register count = rcx; // elements count
860 const Register to_from = to; // (to - from)
861 const Register saved_to = rdx; // saved destination array address
862
863 __ enter(); // required for proper stackwalking of RuntimeStub frame
864 __ pushl(rsi);
865 __ pushl(rdi);
866 __ movl(from , Address(rsp, 12+ 4));
867 __ movl(to , Address(rsp, 12+ 8));
868 __ movl(count, Address(rsp, 12+ 12));
869 if (t == T_OBJECT) {
870 __ testl(count, count);
871 __ jcc(Assembler::zero, L_0_count);
872 gen_write_ref_array_pre_barrier(to, count);
873 __ movl(saved_to, to); // save 'to'
874 }
875
876 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
877 BLOCK_COMMENT("Entry:");
878
879 __ subl(to, from); // to --> to_from
880 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
881 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
882 if (!aligned && (t == T_BYTE || t == T_SHORT)) {
883 // align source address at 4 bytes address boundary
884 if (t == T_BYTE) {
885 // One byte misalignment happens only for byte arrays
886 __ testl(from, 1);
887 __ jccb(Assembler::zero, L_skip_align1);
888 __ movb(rax, Address(from, 0));
889 __ movb(Address(from, to_from, Address::times_1, 0), rax);
890 __ increment(from);
891 __ decrement(count);
892 __ BIND(L_skip_align1);
893 }
894 // Two bytes misalignment happens only for byte and short (char) arrays
895 __ testl(from, 2);
896 __ jccb(Assembler::zero, L_skip_align2);
897 __ movw(rax, Address(from, 0));
898 __ movw(Address(from, to_from, Address::times_1, 0), rax);
899 __ addl(from, 2);
900 __ subl(count, 1<<(shift-1));
901 __ BIND(L_skip_align2);
902 }
903 if (!VM_Version::supports_mmx()) {
904 __ movl(rax, count); // save 'count'
905 __ shrl(count, shift); // bytes count
906 __ addl(to_from, from); // restore 'to'
907 __ rep_movl();
908 __ subl(to_from, from); // restore 'to_from'
909 __ movl(count, rax); // restore 'count'
910 __ jmpb(L_copy_2_bytes); // all dwords were copied
911 } else {
912 // align to 8 bytes, we know we are 4 byte aligned to start
913 __ testl(from, 4);
914 __ jccb(Assembler::zero, L_copy_64_bytes);
915 __ movl(rax, Address(from, 0));
916 __ movl(Address(from, to_from, Address::times_1, 0), rax);
917 __ addl(from, 4);
918 __ subl(count, 1<<shift);
919 __ BIND(L_copy_64_bytes);
920 __ movl(rax, count);
921 __ shrl(rax, shift+1); // 8 bytes chunk count
922 //
923 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
924 //
925 mmx_copy_forward(from, to_from, rax);
926 }
927 // copy tailing dword
928 __ BIND(L_copy_4_bytes);
929 __ testl(count, 1<<shift);
930 __ jccb(Assembler::zero, L_copy_2_bytes);
931 __ movl(rax, Address(from, 0));
932 __ movl(Address(from, to_from, Address::times_1, 0), rax);
933 if (t == T_BYTE || t == T_SHORT) {
934 __ addl(from, 4);
935 __ BIND(L_copy_2_bytes);
936 // copy tailing word
937 __ testl(count, 1<<(shift-1));
938 __ jccb(Assembler::zero, L_copy_byte);
939 __ movw(rax, Address(from, 0));
940 __ movw(Address(from, to_from, Address::times_1, 0), rax);
941 if (t == T_BYTE) {
942 __ addl(from, 2);
943 __ BIND(L_copy_byte);
944 // copy tailing byte
945 __ testl(count, 1);
946 __ jccb(Assembler::zero, L_exit);
947 __ movb(rax, Address(from, 0));
948 __ movb(Address(from, to_from, Address::times_1, 0), rax);
949 __ BIND(L_exit);
950 } else {
951 __ BIND(L_copy_byte);
952 }
953 } else {
954 __ BIND(L_copy_2_bytes);
955 }
956
957 if (t == T_OBJECT) {
958 __ movl(count, Address(rsp, 12+12)); // reread 'count'
959 __ movl(to, saved_to); // restore 'to'
960 gen_write_ref_array_post_barrier(to, count);
961 __ BIND(L_0_count);
962 }
963 inc_copy_counter_np(t);
964 __ popl(rdi);
965 __ popl(rsi);
966 __ leave(); // required for proper stackwalking of RuntimeStub frame
967 __ xorl(rax, rax); // return 0
968 __ ret(0);
969 return start;
970 }
971
972
973 address generate_conjoint_copy(BasicType t, bool aligned,
974 Address::ScaleFactor sf,
975 address nooverlap_target,
976 address* entry, const char *name) {
977 __ align(CodeEntryAlignment);
978 StubCodeMark mark(this, "StubRoutines", name);
979 address start = __ pc();
980
981 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
982 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
983
984 int shift = Address::times_4 - sf;
985
986 const Register src = rax; // source array address
987 const Register dst = rdx; // destination array address
988 const Register from = rsi; // source array address
989 const Register to = rdi; // destination array address
990 const Register count = rcx; // elements count
991 const Register end = rax; // array end address
992
993 __ enter(); // required for proper stackwalking of RuntimeStub frame
994 __ pushl(rsi);
995 __ pushl(rdi);
996 __ movl(src , Address(rsp, 12+ 4)); // from
997 __ movl(dst , Address(rsp, 12+ 8)); // to
998 __ movl(count, Address(rsp, 12+12)); // count
999 if (t == T_OBJECT) {
1000 gen_write_ref_array_pre_barrier(dst, count);
1001 }
1002
1003 if (entry != NULL) {
1004 *entry = __ pc(); // Entry point from generic arraycopy stub.
1005 BLOCK_COMMENT("Entry:");
1006 }
1007
1008 if (t == T_OBJECT) {
1009 __ testl(count, count);
1010 __ jcc(Assembler::zero, L_0_count);
1011 }
1012 __ movl(from, src);
1013 __ movl(to , dst);
1014
1015 // arrays overlap test
1016 RuntimeAddress nooverlap(nooverlap_target);
1017 __ cmpl(dst, src);
1018 __ leal(end, Address(src, count, sf, 0)); // src + count * elem_size
1019 __ jump_cc(Assembler::belowEqual, nooverlap);
1020 __ cmpl(dst, end);
1021 __ jump_cc(Assembler::aboveEqual, nooverlap);
1022
1023 // copy from high to low
1024 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1025 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1026 if (t == T_BYTE || t == T_SHORT) {
1027 // Align the end of destination array at 4 bytes address boundary
1028 __ leal(end, Address(dst, count, sf, 0));
1029 if (t == T_BYTE) {
1030 // One byte misalignment happens only for byte arrays
1031 __ testl(end, 1);
1032 __ jccb(Assembler::zero, L_skip_align1);
1033 __ decrement(count);
1034 __ movb(rdx, Address(from, count, sf, 0));
1035 __ movb(Address(to, count, sf, 0), rdx);
1036 __ BIND(L_skip_align1);
1037 }
1038 // Two bytes misalignment happens only for byte and short (char) arrays
1039 __ testl(end, 2);
1040 __ jccb(Assembler::zero, L_skip_align2);
1041 __ subl(count, 1<<(shift-1));
1042 __ movw(rdx, Address(from, count, sf, 0));
1043 __ movw(Address(to, count, sf, 0), rdx);
1044 __ BIND(L_skip_align2);
1045 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1046 __ jcc(Assembler::below, L_copy_4_bytes);
1047 }
1048
1049 if (!VM_Version::supports_mmx()) {
1050 __ std();
1051 __ movl(rax, count); // Save 'count'
1052 __ movl(rdx, to); // Save 'to'
1053 __ leal(rsi, Address(from, count, sf, -4));
1054 __ leal(rdi, Address(to , count, sf, -4));
1055 __ shrl(count, shift); // bytes count
1056 __ rep_movl();
1057 __ cld();
1058 __ movl(count, rax); // restore 'count'
1059 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1060 __ movl(from, Address(rsp, 12+4)); // reread 'from'
1061 __ movl(to, rdx); // restore 'to'
1062 __ jmpb(L_copy_2_bytes); // all dword were copied
1063 } else {
1064 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1065 __ testl(end, 4);
1066 __ jccb(Assembler::zero, L_copy_8_bytes);
1067 __ subl(count, 1<<shift);
1068 __ movl(rdx, Address(from, count, sf, 0));
1069 __ movl(Address(to, count, sf, 0), rdx);
1070 __ jmpb(L_copy_8_bytes);
1071
1072 __ align(16);
1073 // Move 8 bytes
1074 __ BIND(L_copy_8_bytes_loop);
1075 __ movq(mmx0, Address(from, count, sf, 0));
1076 __ movq(Address(to, count, sf, 0), mmx0);
1077 __ BIND(L_copy_8_bytes);
1078 __ subl(count, 2<<shift);
1079 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1080 __ addl(count, 2<<shift);
1081 __ emms();
1082 }
1083 __ BIND(L_copy_4_bytes);
1084 // copy prefix qword
1085 __ testl(count, 1<<shift);
1086 __ jccb(Assembler::zero, L_copy_2_bytes);
1087 __ movl(rdx, Address(from, count, sf, -4));
1088 __ movl(Address(to, count, sf, -4), rdx);
1089
1090 if (t == T_BYTE || t == T_SHORT) {
1091 __ subl(count, (1<<shift));
1092 __ BIND(L_copy_2_bytes);
1093 // copy prefix dword
1094 __ testl(count, 1<<(shift-1));
1095 __ jccb(Assembler::zero, L_copy_byte);
1096 __ movw(rdx, Address(from, count, sf, -2));
1097 __ movw(Address(to, count, sf, -2), rdx);
1098 if (t == T_BYTE) {
1099 __ subl(count, 1<<(shift-1));
1100 __ BIND(L_copy_byte);
1101 // copy prefix byte
1102 __ testl(count, 1);
1103 __ jccb(Assembler::zero, L_exit);
1104 __ movb(rdx, Address(from, 0));
1105 __ movb(Address(to, 0), rdx);
1106 __ BIND(L_exit);
1107 } else {
1108 __ BIND(L_copy_byte);
1109 }
1110 } else {
1111 __ BIND(L_copy_2_bytes);
1112 }
1113 if (t == T_OBJECT) {
1114 __ movl(count, Address(rsp, 12+12)); // reread count
1115 gen_write_ref_array_post_barrier(to, count);
1116 __ BIND(L_0_count);
1117 }
1118 inc_copy_counter_np(t);
1119 __ popl(rdi);
1120 __ popl(rsi);
1121 __ leave(); // required for proper stackwalking of RuntimeStub frame
1122 __ xorl(rax, rax); // return 0
1123 __ ret(0);
1124 return start;
1125 }
1126
1127
1128 address generate_disjoint_long_copy(address* entry, const char *name) {
1129 __ align(CodeEntryAlignment);
1130 StubCodeMark mark(this, "StubRoutines", name);
1131 address start = __ pc();
1132
1133 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1134 const Register from = rax; // source array address
1135 const Register to = rdx; // destination array address
1136 const Register count = rcx; // elements count
1137 const Register to_from = rdx; // (to - from)
1138
1139 __ enter(); // required for proper stackwalking of RuntimeStub frame
1140 __ movl(from , Address(rsp, 8+0)); // from
1141 __ movl(to , Address(rsp, 8+4)); // to
1142 __ movl(count, Address(rsp, 8+8)); // count
1143
1144 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1145 BLOCK_COMMENT("Entry:");
1146
1147 __ subl(to, from); // to --> to_from
1148 if (VM_Version::supports_mmx()) {
1149 mmx_copy_forward(from, to_from, count);
1150 } else {
1151 __ jmpb(L_copy_8_bytes);
1152 __ align(16);
1153 __ BIND(L_copy_8_bytes_loop);
1154 __ fild_d(Address(from, 0));
1155 __ fistp_d(Address(from, to_from, Address::times_1));
1156 __ addl(from, 8);
1157 __ BIND(L_copy_8_bytes);
1158 __ decrement(count);
1159 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1160 }
1161 inc_copy_counter_np(T_LONG);
1162 __ leave(); // required for proper stackwalking of RuntimeStub frame
1163 __ xorl(rax, rax); // return 0
1164 __ ret(0);
1165 return start;
1166 }
1167
1168 address generate_conjoint_long_copy(address nooverlap_target,
1169 address* entry, const char *name) {
1170 __ align(CodeEntryAlignment);
1171 StubCodeMark mark(this, "StubRoutines", name);
1172 address start = __ pc();
1173
1174 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1175 const Register from = rax; // source array address
1176 const Register to = rdx; // destination array address
1177 const Register count = rcx; // elements count
1178 const Register end_from = rax; // source array end address
1179
1180 __ enter(); // required for proper stackwalking of RuntimeStub frame
1181 __ movl(from , Address(rsp, 8+0)); // from
1182 __ movl(to , Address(rsp, 8+4)); // to
1183 __ movl(count, Address(rsp, 8+8)); // count
1184
1185 *entry = __ pc(); // Entry point from generic arraycopy stub.
1186 BLOCK_COMMENT("Entry:");
1187
1188 // arrays overlap test
1189 __ cmpl(to, from);
1190 RuntimeAddress nooverlap(nooverlap_target);
1191 __ jump_cc(Assembler::belowEqual, nooverlap);
1192 __ leal(end_from, Address(from, count, Address::times_8, 0));
1193 __ cmpl(to, end_from);
1194 __ movl(from, Address(rsp, 8)); // from
1195 __ jump_cc(Assembler::aboveEqual, nooverlap);
1196
1197 __ jmpb(L_copy_8_bytes);
1198
1199 __ align(16);
1200 __ BIND(L_copy_8_bytes_loop);
1201 if (VM_Version::supports_mmx()) {
1202 __ movq(mmx0, Address(from, count, Address::times_8));
1203 __ movq(Address(to, count, Address::times_8), mmx0);
1204 } else {
1205 __ fild_d(Address(from, count, Address::times_8));
1206 __ fistp_d(Address(to, count, Address::times_8));
1207 }
1208 __ BIND(L_copy_8_bytes);
1209 __ decrement(count);
1210 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1211
1212 if (VM_Version::supports_mmx()) {
1213 __ emms();
1214 }
1215 inc_copy_counter_np(T_LONG);
1216 __ leave(); // required for proper stackwalking of RuntimeStub frame
1217 __ xorl(rax, rax); // return 0
1218 __ ret(0);
1219 return start;
1220 }
1221
1222
1223 // Helper for generating a dynamic type check.
1224 // The sub_klass must be one of {rbx, rdx, rsi}.
1225 // The temp is killed.
1226 void generate_type_check(Register sub_klass,
1227 Address& super_check_offset_addr,
1228 Address& super_klass_addr,
1229 Register temp,
1230 Label* L_success_ptr, Label* L_failure_ptr) {
1231 BLOCK_COMMENT("type_check:");
1232
1233 Label L_fallthrough;
1234 bool fall_through_on_success = (L_success_ptr == NULL);
1235 if (fall_through_on_success) {
1236 L_success_ptr = &L_fallthrough;
1237 } else {
1238 L_failure_ptr = &L_fallthrough;
1239 }
1240 Label& L_success = *L_success_ptr;
1241 Label& L_failure = *L_failure_ptr;
1242
1243 assert_different_registers(sub_klass, temp);
1244
1245 // a couple of useful fields in sub_klass:
1246 int ss_offset = (klassOopDesc::header_size() * HeapWordSize +
1247 Klass::secondary_supers_offset_in_bytes());
1248 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1249 Klass::secondary_super_cache_offset_in_bytes());
1250 Address secondary_supers_addr(sub_klass, ss_offset);
1251 Address super_cache_addr( sub_klass, sc_offset);
1252
1253 // if the pointers are equal, we are done (e.g., String[] elements)
1254 __ cmpl(sub_klass, super_klass_addr);
1255 __ jcc(Assembler::equal, L_success);
1256
1257 // check the supertype display:
1258 __ movl(temp, super_check_offset_addr);
1259 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1260 __ movl(temp, super_check_addr); // load displayed supertype
1261 __ cmpl(temp, super_klass_addr); // test the super type
1262 __ jcc(Assembler::equal, L_success);
1263
1264 // if it was a primary super, we can just fail immediately
1265 __ cmpl(super_check_offset_addr, sc_offset);
1266 __ jcc(Assembler::notEqual, L_failure);
1267
1268 // Now do a linear scan of the secondary super-klass chain.
1269 // This code is rarely used, so simplicity is a virtue here.
1270 inc_counter_np(SharedRuntime::_partial_subtype_ctr);
1271 {
1272 // The repne_scan instruction uses fixed registers, which we must spill.
1273 // (We need a couple more temps in any case.)
1274 __ pushl(rax);
1275 __ pushl(rcx);
1276 __ pushl(rdi);
1277 assert_different_registers(sub_klass, rax, rcx, rdi);
1278
1279 __ movl(rdi, secondary_supers_addr);
1280 // Load the array length.
1281 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
1282 // Skip to start of data.
1283 __ addl(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1284 // Scan rcx words at [edi] for occurance of rax,
1285 // Set NZ/Z based on last compare
1286 __ movl(rax, super_klass_addr);
1287 __ repne_scan();
1288
1289 // Unspill the temp. registers:
1290 __ popl(rdi);
1291 __ popl(rcx);
1292 __ popl(rax);
1293 }
1294 __ jcc(Assembler::notEqual, L_failure);
1295
1296 // Success. Cache the super we found and proceed in triumph.
1297 __ movl(temp, super_klass_addr); // note: rax, is dead
1298 __ movl(super_cache_addr, temp);
1299
1300 if (!fall_through_on_success)
1301 __ jmp(L_success);
1302
1303 // Fall through on failure!
1304 __ bind(L_fallthrough);
1305 }
1306
1307 //
1308 // Generate checkcasting array copy stub
1309 //
1310 // Input:
1311 // 4(rsp) - source array address
1312 // 8(rsp) - destination array address
1313 // 12(rsp) - element count, can be zero
1314 // 16(rsp) - size_t ckoff (super_check_offset)
1315 // 20(rsp) - oop ckval (super_klass)
1316 //
1317 // Output:
1318 // rax, == 0 - success
1319 // rax, == -1^K - failure, where K is partial transfer count
1320 //
1321 address generate_checkcast_copy(const char *name, address* entry) {
1322 __ align(CodeEntryAlignment);
1323 StubCodeMark mark(this, "StubRoutines", name);
1324 address start = __ pc();
1325
1326 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1327
1328 // register use:
1329 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1330 // rdi, rsi -- element access (oop, klass)
1331 // rbx, -- temp
1332 const Register from = rax; // source array address
1333 const Register to = rdx; // destination array address
1334 const Register length = rcx; // elements count
1335 const Register elem = rdi; // each oop copied
1336 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1337 const Register temp = rbx; // lone remaining temp
1338
1339 __ enter(); // required for proper stackwalking of RuntimeStub frame
1340
1341 __ pushl(rsi);
1342 __ pushl(rdi);
1343 __ pushl(rbx);
1344
1345 Address from_arg(rsp, 16+ 4); // from
1346 Address to_arg(rsp, 16+ 8); // to
1347 Address length_arg(rsp, 16+12); // elements count
1348 Address ckoff_arg(rsp, 16+16); // super_check_offset
1349 Address ckval_arg(rsp, 16+20); // super_klass
1350
1351 // Load up:
1352 __ movl(from, from_arg);
1353 __ movl(to, to_arg);
1354 __ movl(length, length_arg);
1355
1356 *entry = __ pc(); // Entry point from generic arraycopy stub.
1357 BLOCK_COMMENT("Entry:");
1358
1359 //---------------------------------------------------------------
1360 // Assembler stub will be used for this call to arraycopy
1361 // if the two arrays are subtypes of Object[] but the
1362 // destination array type is not equal to or a supertype
1363 // of the source type. Each element must be separately
1364 // checked.
1365
1366 // Loop-invariant addresses. They are exclusive end pointers.
1367 Address end_from_addr(from, length, Address::times_4, 0);
1368 Address end_to_addr(to, length, Address::times_4, 0);
1369
1370 Register end_from = from; // re-use
1371 Register end_to = to; // re-use
1372 Register count = length; // re-use
1373
1374 // Loop-variant addresses. They assume post-incremented count < 0.
1375 Address from_element_addr(end_from, count, Address::times_4, 0);
1376 Address to_element_addr(end_to, count, Address::times_4, 0);
1377 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1378
1379 // Copy from low to high addresses, indexed from the end of each array.
1380 __ leal(end_from, end_from_addr);
1381 __ leal(end_to, end_to_addr);
1382 gen_write_ref_array_pre_barrier(to, count);
1383 assert(length == count, ""); // else fix next line:
1384 __ negl(count); // negate and test the length
1385 __ jccb(Assembler::notZero, L_load_element);
1386
1387 // Empty array: Nothing to do.
1388 __ xorl(rax, rax); // return 0 on (trivial) success
1389 __ jmp(L_done);
1390
1391 // ======== begin loop ========
1392 // (Loop is rotated; its entry is L_load_element.)
1393 // Loop control:
1394 // for (count = -count; count != 0; count++)
1395 // Base pointers src, dst are biased by 8*count,to last element.
1396 __ align(16);
1397
1398 __ BIND(L_store_element);
1399 __ movl(to_element_addr, elem); // store the oop
1400 __ increment(count); // increment the count toward zero
1401 __ jccb(Assembler::zero, L_do_card_marks);
1402
1403 // ======== loop entry is here ========
1404 __ BIND(L_load_element);
1405 __ movl(elem, from_element_addr); // load the oop
1406 __ testl(elem, elem);
1407 __ jccb(Assembler::zero, L_store_element);
1408
1409 // (Could do a trick here: Remember last successful non-null
1410 // element stored and make a quick oop equality check on it.)
1411
1412 __ movl(elem_klass, elem_klass_addr); // query the object klass
1413 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1414 &L_store_element, NULL);
1415 // (On fall-through, we have failed the element type check.)
1416 // ======== end loop ========
1417
1418 // It was a real error; we must depend on the caller to finish the job.
1419 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1420 // Emit GC store barriers for the oops we have copied (length_arg + count),
1421 // and report their number to the caller.
1422 __ addl(count, length_arg); // transfers = (length - remaining)
1423 __ movl(rax, count); // save the value
1424 __ notl(rax); // report (-1^K) to caller
1425 __ movl(to, to_arg); // reload
1426 assert_different_registers(to, count, rax);
1427 gen_write_ref_array_post_barrier(to, count);
1428 __ jmpb(L_done);
1429
1430 // Come here on success only.
1431 __ BIND(L_do_card_marks);
1432 __ movl(count, length_arg);
1433 __ movl(to, to_arg); // reload
1434 gen_write_ref_array_post_barrier(to, count);
1435 __ xorl(rax, rax); // return 0 on success
1436
1437 // Common exit point (success or failure).
1438 __ BIND(L_done);
1439 __ popl(rbx);
1440 __ popl(rdi);
1441 __ popl(rsi);
1442 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1443 __ leave(); // required for proper stackwalking of RuntimeStub frame
1444 __ ret(0);
1445
1446 return start;
1447 }
1448
1449 //
1450 // Generate 'unsafe' array copy stub
1451 // Though just as safe as the other stubs, it takes an unscaled
1452 // size_t argument instead of an element count.
1453 //
1454 // Input:
1455 // 4(rsp) - source array address
1456 // 8(rsp) - destination array address
1457 // 12(rsp) - byte count, can be zero
1458 //
1459 // Output:
1460 // rax, == 0 - success
1461 // rax, == -1 - need to call System.arraycopy
1462 //
1463 // Examines the alignment of the operands and dispatches
1464 // to a long, int, short, or byte copy loop.
1465 //
1466 address generate_unsafe_copy(const char *name,
1467 address byte_copy_entry,
1468 address short_copy_entry,
1469 address int_copy_entry,
1470 address long_copy_entry) {
1471
1472 Label L_long_aligned, L_int_aligned, L_short_aligned;
1473
1474 __ align(CodeEntryAlignment);
1475 StubCodeMark mark(this, "StubRoutines", name);
1476 address start = __ pc();
1477
1478 const Register from = rax; // source array address
1479 const Register to = rdx; // destination array address
1480 const Register count = rcx; // elements count
1481
1482 __ enter(); // required for proper stackwalking of RuntimeStub frame
1483 __ pushl(rsi);
1484 __ pushl(rdi);
1485 Address from_arg(rsp, 12+ 4); // from
1486 Address to_arg(rsp, 12+ 8); // to
1487 Address count_arg(rsp, 12+12); // byte count
1488
1489 // Load up:
1490 __ movl(from , from_arg);
1491 __ movl(to , to_arg);
1492 __ movl(count, count_arg);
1493
1494 // bump this on entry, not on exit:
1495 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1496
1497 const Register bits = rsi;
1498 __ movl(bits, from);
1499 __ orl(bits, to);
1500 __ orl(bits, count);
1501
1502 __ testl(bits, BytesPerLong-1);
1503 __ jccb(Assembler::zero, L_long_aligned);
1504
1505 __ testl(bits, BytesPerInt-1);
1506 __ jccb(Assembler::zero, L_int_aligned);
1507
1508 __ testl(bits, BytesPerShort-1);
1509 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1510
1511 __ BIND(L_short_aligned);
1512 __ shrl(count, LogBytesPerShort); // size => short_count
1513 __ movl(count_arg, count); // update 'count'
1514 __ jump(RuntimeAddress(short_copy_entry));
1515
1516 __ BIND(L_int_aligned);
1517 __ shrl(count, LogBytesPerInt); // size => int_count
1518 __ movl(count_arg, count); // update 'count'
1519 __ jump(RuntimeAddress(int_copy_entry));
1520
1521 __ BIND(L_long_aligned);
1522 __ shrl(count, LogBytesPerLong); // size => qword_count
1523 __ movl(count_arg, count); // update 'count'
1524 __ popl(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1525 __ popl(rsi);
1526 __ jump(RuntimeAddress(long_copy_entry));
1527
1528 return start;
1529 }
1530
1531
1532 // Perform range checks on the proposed arraycopy.
1533 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1534 void arraycopy_range_checks(Register src,
1535 Register src_pos,
1536 Register dst,
1537 Register dst_pos,
1538 Address& length,
1539 Label& L_failed) {
1540 BLOCK_COMMENT("arraycopy_range_checks:");
1541 const Register src_end = src_pos; // source array end position
1542 const Register dst_end = dst_pos; // destination array end position
1543 __ addl(src_end, length); // src_pos + length
1544 __ addl(dst_end, length); // dst_pos + length
1545
1546 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1547 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1548 __ jcc(Assembler::above, L_failed);
1549
1550 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1551 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1552 __ jcc(Assembler::above, L_failed);
1553
1554 BLOCK_COMMENT("arraycopy_range_checks done");
1555 }
1556
1557
1558 //
1559 // Generate generic array copy stubs
1560 //
1561 // Input:
1562 // 4(rsp) - src oop
1563 // 8(rsp) - src_pos
1564 // 12(rsp) - dst oop
1565 // 16(rsp) - dst_pos
1566 // 20(rsp) - element count
1567 //
1568 // Output:
1569 // rax, == 0 - success
1570 // rax, == -1^K - failure, where K is partial transfer count
1571 //
1572 address generate_generic_copy(const char *name,
1573 address entry_jbyte_arraycopy,
1574 address entry_jshort_arraycopy,
1575 address entry_jint_arraycopy,
1576 address entry_oop_arraycopy,
1577 address entry_jlong_arraycopy,
1578 address entry_checkcast_arraycopy) {
1579 Label L_failed, L_failed_0, L_objArray;
1580
1581 { int modulus = CodeEntryAlignment;
1582 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1583 int advance = target - (__ offset() % modulus);
1584 if (advance < 0) advance += modulus;
1585 if (advance > 0) __ nop(advance);
1586 }
1587 StubCodeMark mark(this, "StubRoutines", name);
1588
1589 // Short-hop target to L_failed. Makes for denser prologue code.
1590 __ BIND(L_failed_0);
1591 __ jmp(L_failed);
1592 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1593
1594 __ align(CodeEntryAlignment);
1595 address start = __ pc();
1596
1597 __ enter(); // required for proper stackwalking of RuntimeStub frame
1598 __ pushl(rsi);
1599 __ pushl(rdi);
1600
1601 // bump this on entry, not on exit:
1602 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1603
1604 // Input values
1605 Address SRC (rsp, 12+ 4);
1606 Address SRC_POS (rsp, 12+ 8);
1607 Address DST (rsp, 12+12);
1608 Address DST_POS (rsp, 12+16);
1609 Address LENGTH (rsp, 12+20);
1610
1611 //-----------------------------------------------------------------------
1612 // Assembler stub will be used for this call to arraycopy
1613 // if the following conditions are met:
1614 //
1615 // (1) src and dst must not be null.
1616 // (2) src_pos must not be negative.
1617 // (3) dst_pos must not be negative.
1618 // (4) length must not be negative.
1619 // (5) src klass and dst klass should be the same and not NULL.
1620 // (6) src and dst should be arrays.
1621 // (7) src_pos + length must not exceed length of src.
1622 // (8) dst_pos + length must not exceed length of dst.
1623 //
1624
1625 const Register src = rax; // source array oop
1626 const Register src_pos = rsi;
1627 const Register dst = rdx; // destination array oop
1628 const Register dst_pos = rdi;
1629 const Register length = rcx; // transfer count
1630
1631 // if (src == NULL) return -1;
1632 __ movl(src, SRC); // src oop
1633 __ testl(src, src);
1634 __ jccb(Assembler::zero, L_failed_0);
1635
1636 // if (src_pos < 0) return -1;
1637 __ movl(src_pos, SRC_POS); // src_pos
1638 __ testl(src_pos, src_pos);
1639 __ jccb(Assembler::negative, L_failed_0);
1640
1641 // if (dst == NULL) return -1;
1642 __ movl(dst, DST); // dst oop
1643 __ testl(dst, dst);
1644 __ jccb(Assembler::zero, L_failed_0);
1645
1646 // if (dst_pos < 0) return -1;
1647 __ movl(dst_pos, DST_POS); // dst_pos
1648 __ testl(dst_pos, dst_pos);
1649 __ jccb(Assembler::negative, L_failed_0);
1650
1651 // if (length < 0) return -1;
1652 __ movl(length, LENGTH); // length
1653 __ testl(length, length);
1654 __ jccb(Assembler::negative, L_failed_0);
1655
1656 // if (src->klass() == NULL) return -1;
1657 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1658 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1659 const Register rcx_src_klass = rcx; // array klass
1660 __ movl(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1661
1662 #ifdef ASSERT
1663 // assert(src->klass() != NULL);
1664 BLOCK_COMMENT("assert klasses not null");
1665 { Label L1, L2;
1666 __ testl(rcx_src_klass, rcx_src_klass);
1667 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1668 __ bind(L1);
1669 __ stop("broken null klass");
1670 __ bind(L2);
1671 __ cmpl(dst_klass_addr, 0);
1672 __ jccb(Assembler::equal, L1); // this would be broken also
1673 BLOCK_COMMENT("assert done");
1674 }
1675 #endif //ASSERT
1676
1677 // Load layout helper (32-bits)
1678 //
1679 // |array_tag| | header_size | element_type | |log2_element_size|
1680 // 32 30 24 16 8 2 0
1681 //
1682 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1683 //
1684
1685 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1686 Klass::layout_helper_offset_in_bytes();
1687 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1688
1689 // Handle objArrays completely differently...
1690 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1691 __ cmpl(src_klass_lh_addr, objArray_lh);
1692 __ jcc(Assembler::equal, L_objArray);
1693
1694 // if (src->klass() != dst->klass()) return -1;
1695 __ cmpl(rcx_src_klass, dst_klass_addr);
1696 __ jccb(Assembler::notEqual, L_failed_0);
1697
1698 const Register rcx_lh = rcx; // layout helper
1699 assert(rcx_lh == rcx_src_klass, "known alias");
1700 __ movl(rcx_lh, src_klass_lh_addr);
1701
1702 // if (!src->is_Array()) return -1;
1703 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1704 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1705
1706 // At this point, it is known to be a typeArray (array_tag 0x3).
1707 #ifdef ASSERT
1708 { Label L;
1709 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1710 __ jcc(Assembler::greaterEqual, L); // signed cmp
1711 __ stop("must be a primitive array");
1712 __ bind(L);
1713 }
1714 #endif
1715
1716 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1717 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1718
1719 // typeArrayKlass
1720 //
1721 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1722 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1723 //
1724 const Register rsi_offset = rsi; // array offset
1725 const Register src_array = src; // src array offset
1726 const Register dst_array = dst; // dst array offset
1727 const Register rdi_elsize = rdi; // log2 element size
1728
1729 __ movl(rsi_offset, rcx_lh);
1730 __ shrl(rsi_offset, Klass::_lh_header_size_shift);
1731 __ andl(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1732 __ addl(src_array, rsi_offset); // src array offset
1733 __ addl(dst_array, rsi_offset); // dst array offset
1734 __ andl(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1735
1736 // next registers should be set before the jump to corresponding stub
1737 const Register from = src; // source array address
1738 const Register to = dst; // destination array address
1739 const Register count = rcx; // elements count
1740 // some of them should be duplicated on stack
1741 #define FROM Address(rsp, 12+ 4)
1742 #define TO Address(rsp, 12+ 8) // Not used now
1743 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1744
1745 BLOCK_COMMENT("scale indexes to element size");
1746 __ movl(rsi, SRC_POS); // src_pos
1747 __ shll(rsi); // src_pos << rcx (log2 elsize)
1748 assert(src_array == from, "");
1749 __ addl(from, rsi); // from = src_array + SRC_POS << log2 elsize
1750 __ movl(rdi, DST_POS); // dst_pos
1751 __ shll(rdi); // dst_pos << rcx (log2 elsize)
1752 assert(dst_array == to, "");
1753 __ addl(to, rdi); // to = dst_array + DST_POS << log2 elsize
1754 __ movl(FROM, from); // src_addr
1755 __ movl(rdi_elsize, rcx_lh); // log2 elsize
1756 __ movl(count, LENGTH); // elements count
1757
1758 BLOCK_COMMENT("choose copy loop based on element size");
1759 __ cmpl(rdi_elsize, 0);
1760
1761 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1762 __ cmpl(rdi_elsize, LogBytesPerShort);
1763 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1764 __ cmpl(rdi_elsize, LogBytesPerInt);
1765 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1766 #ifdef ASSERT
1767 __ cmpl(rdi_elsize, LogBytesPerLong);
1768 __ jccb(Assembler::notEqual, L_failed);
1769 #endif
1770 __ popl(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1771 __ popl(rsi);
1772 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1773
1774 __ BIND(L_failed);
1775 __ xorl(rax, rax);
1776 __ notl(rax); // return -1
1777 __ popl(rdi);
1778 __ popl(rsi);
1779 __ leave(); // required for proper stackwalking of RuntimeStub frame
1780 __ ret(0);
1781
1782 // objArrayKlass
1783 __ BIND(L_objArray);
1784 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1785
1786 Label L_plain_copy, L_checkcast_copy;
1787 // test array classes for subtyping
1788 __ cmpl(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1789 __ jccb(Assembler::notEqual, L_checkcast_copy);
1790
1791 // Identically typed arrays can be copied without element-wise checks.
1792 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1793 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1794
1795 __ BIND(L_plain_copy);
1796 __ movl(count, LENGTH); // elements count
1797 __ movl(src_pos, SRC_POS); // reload src_pos
1798 __ leal(from, Address(src, src_pos, Address::times_4,
1799 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1800 __ movl(dst_pos, DST_POS); // reload dst_pos
1801 __ leal(to, Address(dst, dst_pos, Address::times_4,
1802 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1803 __ movl(FROM, from); // src_addr
1804 __ movl(TO, to); // dst_addr
1805 __ movl(COUNT, count); // count
1806 __ jump(RuntimeAddress(entry_oop_arraycopy));
1807
1808 __ BIND(L_checkcast_copy);
1809 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1810 {
1811 // Handy offsets:
1812 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1813 objArrayKlass::element_klass_offset_in_bytes());
1814 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1815 Klass::super_check_offset_offset_in_bytes());
1816
1817 Register rsi_dst_klass = rsi;
1818 Register rdi_temp = rdi;
1819 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1820 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1821 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1822
1823 // Before looking at dst.length, make sure dst is also an objArray.
1824 __ movl(rsi_dst_klass, dst_klass_addr);
1825 __ cmpl(dst_klass_lh_addr, objArray_lh);
1826 __ jccb(Assembler::notEqual, L_failed);
1827
1828 // It is safe to examine both src.length and dst.length.
1829 __ movl(src_pos, SRC_POS); // reload rsi
1830 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1831 // (Now src_pos and dst_pos are killed, but not src and dst.)
1832
1833 // We'll need this temp (don't forget to pop it after the type check).
1834 __ pushl(rbx);
1835 Register rbx_src_klass = rbx;
1836
1837 __ movl(rbx_src_klass, rcx_src_klass); // spill away from rcx
1838 __ movl(rsi_dst_klass, dst_klass_addr);
1839 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1840 Label L_fail_array_check;
1841 generate_type_check(rbx_src_klass,
1842 super_check_offset_addr, dst_klass_addr,
1843 rdi_temp, NULL, &L_fail_array_check);
1844 // (On fall-through, we have passed the array type check.)
1845 __ popl(rbx);
1846 __ jmp(L_plain_copy);
1847
1848 __ BIND(L_fail_array_check);
1849 // Reshuffle arguments so we can call checkcast_arraycopy:
1850
1851 // match initial saves for checkcast_arraycopy
1852 // pushl(rsi); // already done; see above
1853 // pushl(rdi); // already done; see above
1854 // pushl(rbx); // already done; see above
1855
1856 // Marshal outgoing arguments now, freeing registers.
1857 Address from_arg(rsp, 16+ 4); // from
1858 Address to_arg(rsp, 16+ 8); // to
1859 Address length_arg(rsp, 16+12); // elements count
1860 Address ckoff_arg(rsp, 16+16); // super_check_offset
1861 Address ckval_arg(rsp, 16+20); // super_klass
1862
1863 Address SRC_POS_arg(rsp, 16+ 8);
1864 Address DST_POS_arg(rsp, 16+16);
1865 Address LENGTH_arg(rsp, 16+20);
1866 // push rbx, changed the incoming offsets (why not just use rbp,??)
1867 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1868
1869 __ movl(rbx, Address(rsi_dst_klass, ek_offset));
1870 __ movl(length, LENGTH_arg); // reload elements count
1871 __ movl(src_pos, SRC_POS_arg); // reload src_pos
1872 __ movl(dst_pos, DST_POS_arg); // reload dst_pos
1873
1874 __ movl(ckval_arg, rbx); // destination element type
1875 __ movl(rbx, Address(rbx, sco_offset));
1876 __ movl(ckoff_arg, rbx); // corresponding class check offset
1877
1878 __ movl(length_arg, length); // outgoing length argument
1879
1880 __ leal(from, Address(src, src_pos, Address::times_4,
1881 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1882 __ movl(from_arg, from);
1883
1884 __ leal(to, Address(dst, dst_pos, Address::times_4,
1885 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1886 __ movl(to_arg, to);
1887 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1888 }
1889
1890 return start;
1891 }
1892
1893 void generate_arraycopy_stubs() {
1894 address entry;
1895 address entry_jbyte_arraycopy;
1896 address entry_jshort_arraycopy;
1897 address entry_jint_arraycopy;
1898 address entry_oop_arraycopy;
1899 address entry_jlong_arraycopy;
1900 address entry_checkcast_arraycopy;
1901
1902 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
1903 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
1904 "arrayof_jbyte_disjoint_arraycopy");
1905 StubRoutines::_arrayof_jbyte_arraycopy =
1906 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
1907 NULL, "arrayof_jbyte_arraycopy");
1908 StubRoutines::_jbyte_disjoint_arraycopy =
1909 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
1910 "jbyte_disjoint_arraycopy");
1911 StubRoutines::_jbyte_arraycopy =
1912 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
1913 &entry_jbyte_arraycopy, "jbyte_arraycopy");
1914
1915 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
1916 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
1917 "arrayof_jshort_disjoint_arraycopy");
1918 StubRoutines::_arrayof_jshort_arraycopy =
1919 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
1920 NULL, "arrayof_jshort_arraycopy");
1921 StubRoutines::_jshort_disjoint_arraycopy =
1922 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
1923 "jshort_disjoint_arraycopy");
1924 StubRoutines::_jshort_arraycopy =
1925 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
1926 &entry_jshort_arraycopy, "jshort_arraycopy");
1927
1928 // Next arrays are always aligned on 4 bytes at least.
1929 StubRoutines::_jint_disjoint_arraycopy =
1930 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
1931 "jint_disjoint_arraycopy");
1932 StubRoutines::_jint_arraycopy =
1933 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
1934 &entry_jint_arraycopy, "jint_arraycopy");
1935
1936 StubRoutines::_oop_disjoint_arraycopy =
1937 generate_disjoint_copy(T_OBJECT, true, Address::times_4, &entry,
1938 "oop_disjoint_arraycopy");
1939 StubRoutines::_oop_arraycopy =
1940 generate_conjoint_copy(T_OBJECT, true, Address::times_4, entry,
1941 &entry_oop_arraycopy, "oop_arraycopy");
1942
1943 StubRoutines::_jlong_disjoint_arraycopy =
1944 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
1945 StubRoutines::_jlong_arraycopy =
1946 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
1947 "jlong_arraycopy");
1948
1949 StubRoutines::_arrayof_jint_disjoint_arraycopy =
1950 StubRoutines::_jint_disjoint_arraycopy;
1951 StubRoutines::_arrayof_oop_disjoint_arraycopy =
1952 StubRoutines::_oop_disjoint_arraycopy;
1953 StubRoutines::_arrayof_jlong_disjoint_arraycopy =
1954 StubRoutines::_jlong_disjoint_arraycopy;
1955
1956 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
1957 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
1958 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
1959
1960 StubRoutines::_checkcast_arraycopy =
1961 generate_checkcast_copy("checkcast_arraycopy",
1962 &entry_checkcast_arraycopy);
1963
1964 StubRoutines::_unsafe_arraycopy =
1965 generate_unsafe_copy("unsafe_arraycopy",
1966 entry_jbyte_arraycopy,
1967 entry_jshort_arraycopy,
1968 entry_jint_arraycopy,
1969 entry_jlong_arraycopy);
1970
1971 StubRoutines::_generic_arraycopy =
1972 generate_generic_copy("generic_arraycopy",
1973 entry_jbyte_arraycopy,
1974 entry_jshort_arraycopy,
1975 entry_jint_arraycopy,
1976 entry_oop_arraycopy,
1977 entry_jlong_arraycopy,
1978 entry_checkcast_arraycopy);
1979 }
1980
1981 public:
1982 // Information about frame layout at time of blocking runtime call.
1983 // Note that we only have to preserve callee-saved registers since
1984 // the compilers are responsible for supplying a continuation point
1985 // if they expect all registers to be preserved.
1986 enum layout {
1987 thread_off, // last_java_sp
1988 rbp_off, // callee saved register
1989 ret_pc,
1990 framesize
1991 };
1992
1993 private:
1994
1995 #undef __
1996 #define __ masm->
1997
1998 //------------------------------------------------------------------------------------------------------------------------
1999 // Continuation point for throwing of implicit exceptions that are not handled in
2000 // the current activation. Fabricates an exception oop and initiates normal
2001 // exception dispatching in this frame.
2002 //
2003 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2004 // This is no longer true after adapter frames were removed but could possibly
2005 // be brought back in the future if the interpreter code was reworked and it
2006 // was deemed worthwhile. The comment below was left to describe what must
2007 // happen here if callee saves were resurrected. As it stands now this stub
2008 // could actually be a vanilla BufferBlob and have now oopMap at all.
2009 // Since it doesn't make much difference we've chosen to leave it the
2010 // way it was in the callee save days and keep the comment.
2011
2012 // If we need to preserve callee-saved values we need a callee-saved oop map and
2013 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2014 // If the compiler needs all registers to be preserved between the fault
2015 // point and the exception handler then it must assume responsibility for that in
2016 // AbstractCompiler::continuation_for_implicit_null_exception or
2017 // continuation_for_implicit_division_by_zero_exception. All other implicit
2018 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2019 // either at call sites or otherwise assume that stack unwinding will be initiated,
2020 // so caller saved registers were assumed volatile in the compiler.
2021 address generate_throw_exception(const char* name, address runtime_entry,
2022 bool restore_saved_exception_pc) {
2023
2024 int insts_size = 256;
2025 int locs_size = 32;
2026
2027 CodeBuffer code(name, insts_size, locs_size);
2028 OopMapSet* oop_maps = new OopMapSet();
2029 MacroAssembler* masm = new MacroAssembler(&code);
2030
2031 address start = __ pc();
2032
2033 // This is an inlined and slightly modified version of call_VM
2034 // which has the ability to fetch the return PC out of
2035 // thread-local storage and also sets up last_Java_sp slightly
2036 // differently than the real call_VM
2037 Register java_thread = rbx;
2038 __ get_thread(java_thread);
2039 if (restore_saved_exception_pc) {
2040 __ movl(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2041 __ pushl(rax);
2042 }
2043
2044 __ enter(); // required for proper stackwalking of RuntimeStub frame
2045
2046 // pc and rbp, already pushed
2047 __ subl(rsp, (framesize-2) * wordSize); // prolog
2048
2049 // Frame is now completed as far as size and linkage.
2050
2051 int frame_complete = __ pc() - start;
2052
2053 // push java thread (becomes first argument of C function)
2054 __ movl(Address(rsp, thread_off * wordSize), java_thread);
2055
2056 // Set up last_Java_sp and last_Java_fp
2057 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2058
2059 // Call runtime
2060 BLOCK_COMMENT("call runtime_entry");
2061 __ call(RuntimeAddress(runtime_entry));
2062 // Generate oop map
2063 OopMap* map = new OopMap(framesize, 0);
2064 oop_maps->add_gc_map(__ pc() - start, map);
2065
2066 // restore the thread (cannot use the pushed argument since arguments
2067 // may be overwritten by C code generated by an optimizing compiler);
2068 // however can use the register value directly if it is callee saved.
2069 __ get_thread(java_thread);
2070
2071 __ reset_last_Java_frame(java_thread, true, false);
2072
2073 __ leave(); // required for proper stackwalking of RuntimeStub frame
2074
2075 // check for pending exceptions
2076 #ifdef ASSERT
2077 Label L;
2078 __ cmpl(Address(java_thread, Thread::pending_exception_offset()), NULL_WORD);
2079 __ jcc(Assembler::notEqual, L);
2080 __ should_not_reach_here();
2081 __ bind(L);
2082 #endif /* ASSERT */
2083 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2084
2085
2086 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2087 return stub->entry_point();
2088 }
2089
2090
2091 void create_control_words() {
2092 // Round to nearest, 53-bit mode, exceptions masked
2093 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2094 // Round to zero, 53-bit mode, exception mased
2095 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2096 // Round to nearest, 24-bit mode, exceptions masked
2097 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2098 // Round to nearest, 64-bit mode, exceptions masked
2099 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2100 // Round to nearest, 64-bit mode, exceptions masked
2101 StubRoutines::_mxcsr_std = 0x1F80;
2102 // Note: the following two constants are 80-bit values
2103 // layout is critical for correct loading by FPU.
2104 // Bias for strict fp multiply/divide
2105 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2106 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2107 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2108 // Un-Bias for strict fp multiply/divide
2109 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2110 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2111 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2112 }
2113
2114 //---------------------------------------------------------------------------
2115 // Initialization
2116
2117 void generate_initial() {
2118 // Generates all stubs and initializes the entry points
2119
2120 //------------------------------------------------------------------------------------------------------------------------
2121 // entry points that exist in all platforms
2122 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2123 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2124 StubRoutines::_forward_exception_entry = generate_forward_exception();
2125
2126 StubRoutines::_call_stub_entry =
2127 generate_call_stub(StubRoutines::_call_stub_return_address);
2128 // is referenced by megamorphic call
2129 StubRoutines::_catch_exception_entry = generate_catch_exception();
2130
2131 // These are currently used by Solaris/Intel
2132 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2133
2134 StubRoutines::_handler_for_unsafe_access_entry =
2135 generate_handler_for_unsafe_access();
2136
2137 // platform dependent
2138 create_control_words();
2139
2140 StubRoutines::i486::_verify_mxcsr_entry = generate_verify_mxcsr();
2141 StubRoutines::i486::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2142 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2143 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2144 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2145 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2146 }
2147
2148
2149 void generate_all() {
2150 // Generates all stubs and initializes the entry points
2151
2152 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2153 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2154 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2155 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
2156 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2157 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2158 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);
2159 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2160
2161 //------------------------------------------------------------------------------------------------------------------------
2162 // entry points that are platform specific
2163
2164 // support for verify_oop (must happen after universe_init)
2165 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2166
2167 // arraycopy stubs used by compilers
2168 generate_arraycopy_stubs();
2169 }
2170
2171
2172 public:
2173 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2174 if (all) {
2175 generate_all();
2176 } else {
2177 generate_initial();
2178 }
2179 }
2180 }; // end class declaration
2181
2182
2183 void StubGenerator_generate(CodeBuffer* code, bool all) {
2184 StubGenerator g(code, all);
2185 }