Mon, 28 Feb 2011 06:07:12 -0800
7012914: JSR 292 MethodHandlesTest C1: frame::verify_return_pc(return_address) failed: must be a return pc
Reviewed-by: never, bdelsart
1 /*
2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "assembler_x86.inline.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "nativeInst_x86.hpp"
30 #include "oops/instanceOop.hpp"
31 #include "oops/methodOop.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/frame.inline.hpp"
36 #include "runtime/handles.inline.hpp"
37 #include "runtime/sharedRuntime.hpp"
38 #include "runtime/stubCodeGenerator.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "utilities/top.hpp"
41 #ifdef TARGET_OS_FAMILY_linux
42 # include "thread_linux.inline.hpp"
43 #endif
44 #ifdef TARGET_OS_FAMILY_solaris
45 # include "thread_solaris.inline.hpp"
46 #endif
47 #ifdef TARGET_OS_FAMILY_windows
48 # include "thread_windows.inline.hpp"
49 #endif
50 #ifdef COMPILER2
51 #include "opto/runtime.hpp"
52 #endif
54 // Declaration and definition of StubGenerator (no .hpp file).
55 // For a more detailed description of the stub routine structure
56 // see the comment in stubRoutines.hpp
58 #define __ _masm->
59 #define a__ ((Assembler*)_masm)->
61 #ifdef PRODUCT
62 #define BLOCK_COMMENT(str) /* nothing */
63 #else
64 #define BLOCK_COMMENT(str) __ block_comment(str)
65 #endif
67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
69 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions
70 const int FPU_CNTRL_WRD_MASK = 0xFFFF;
72 // -------------------------------------------------------------------------------------------------------------------------
73 // Stub Code definitions
75 static address handle_unsafe_access() {
76 JavaThread* thread = JavaThread::current();
77 address pc = thread->saved_exception_pc();
78 // pc is the instruction which we must emulate
79 // doing a no-op is fine: return garbage from the load
80 // therefore, compute npc
81 address npc = Assembler::locate_next_instruction(pc);
83 // request an async exception
84 thread->set_pending_unsafe_access_error();
86 // return address of next instruction to execute
87 return npc;
88 }
90 class StubGenerator: public StubCodeGenerator {
91 private:
93 #ifdef PRODUCT
94 #define inc_counter_np(counter) (0)
95 #else
96 void inc_counter_np_(int& counter) {
97 __ incrementl(ExternalAddress((address)&counter));
98 }
99 #define inc_counter_np(counter) \
100 BLOCK_COMMENT("inc_counter " #counter); \
101 inc_counter_np_(counter);
102 #endif //PRODUCT
104 void inc_copy_counter_np(BasicType t) {
105 #ifndef PRODUCT
106 switch (t) {
107 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return;
108 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return;
109 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return;
110 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return;
111 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return;
112 }
113 ShouldNotReachHere();
114 #endif //PRODUCT
115 }
117 //------------------------------------------------------------------------------------------------------------------------
118 // Call stubs are used to call Java from C
119 //
120 // [ return_from_Java ] <--- rsp
121 // [ argument word n ]
122 // ...
123 // -N [ argument word 1 ]
124 // -7 [ Possible padding for stack alignment ]
125 // -6 [ Possible padding for stack alignment ]
126 // -5 [ Possible padding for stack alignment ]
127 // -4 [ mxcsr save ] <--- rsp_after_call
128 // -3 [ saved rbx, ]
129 // -2 [ saved rsi ]
130 // -1 [ saved rdi ]
131 // 0 [ saved rbp, ] <--- rbp,
132 // 1 [ return address ]
133 // 2 [ ptr. to call wrapper ]
134 // 3 [ result ]
135 // 4 [ result_type ]
136 // 5 [ method ]
137 // 6 [ entry_point ]
138 // 7 [ parameters ]
139 // 8 [ parameter_size ]
140 // 9 [ thread ]
143 address generate_call_stub(address& return_address) {
144 StubCodeMark mark(this, "StubRoutines", "call_stub");
145 address start = __ pc();
147 // stub code parameters / addresses
148 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
149 bool sse_save = false;
150 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()!
151 const int locals_count_in_bytes (4*wordSize);
152 const Address mxcsr_save (rbp, -4 * wordSize);
153 const Address saved_rbx (rbp, -3 * wordSize);
154 const Address saved_rsi (rbp, -2 * wordSize);
155 const Address saved_rdi (rbp, -1 * wordSize);
156 const Address result (rbp, 3 * wordSize);
157 const Address result_type (rbp, 4 * wordSize);
158 const Address method (rbp, 5 * wordSize);
159 const Address entry_point (rbp, 6 * wordSize);
160 const Address parameters (rbp, 7 * wordSize);
161 const Address parameter_size(rbp, 8 * wordSize);
162 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()!
163 sse_save = UseSSE > 0;
165 // stub code
166 __ enter();
167 __ movptr(rcx, parameter_size); // parameter counter
168 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes
169 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves
170 __ subptr(rsp, rcx);
171 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack
173 // save rdi, rsi, & rbx, according to C calling conventions
174 __ movptr(saved_rdi, rdi);
175 __ movptr(saved_rsi, rsi);
176 __ movptr(saved_rbx, rbx);
177 // save and initialize %mxcsr
178 if (sse_save) {
179 Label skip_ldmx;
180 __ stmxcsr(mxcsr_save);
181 __ movl(rax, mxcsr_save);
182 __ andl(rax, MXCSR_MASK); // Only check control and mask bits
183 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
184 __ cmp32(rax, mxcsr_std);
185 __ jcc(Assembler::equal, skip_ldmx);
186 __ ldmxcsr(mxcsr_std);
187 __ bind(skip_ldmx);
188 }
190 // make sure the control word is correct.
191 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
193 #ifdef ASSERT
194 // make sure we have no pending exceptions
195 { Label L;
196 __ movptr(rcx, thread);
197 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
198 __ jcc(Assembler::equal, L);
199 __ stop("StubRoutines::call_stub: entered with pending exception");
200 __ bind(L);
201 }
202 #endif
204 // pass parameters if any
205 BLOCK_COMMENT("pass parameters if any");
206 Label parameters_done;
207 __ movl(rcx, parameter_size); // parameter counter
208 __ testl(rcx, rcx);
209 __ jcc(Assembler::zero, parameters_done);
211 // parameter passing loop
213 Label loop;
214 // Copy Java parameters in reverse order (receiver last)
215 // Note that the argument order is inverted in the process
216 // source is rdx[rcx: N-1..0]
217 // dest is rsp[rbx: 0..N-1]
219 __ movptr(rdx, parameters); // parameter pointer
220 __ xorptr(rbx, rbx);
222 __ BIND(loop);
224 // get parameter
225 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize));
226 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(),
227 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter
228 __ increment(rbx);
229 __ decrement(rcx);
230 __ jcc(Assembler::notZero, loop);
232 // call Java function
233 __ BIND(parameters_done);
234 __ movptr(rbx, method); // get methodOop
235 __ movptr(rax, entry_point); // get entry_point
236 __ mov(rsi, rsp); // set sender sp
237 BLOCK_COMMENT("call Java function");
238 __ call(rax);
240 BLOCK_COMMENT("call_stub_return_address:");
241 return_address = __ pc();
243 #ifdef COMPILER2
244 {
245 Label L_skip;
246 if (UseSSE >= 2) {
247 __ verify_FPU(0, "call_stub_return");
248 } else {
249 for (int i = 1; i < 8; i++) {
250 __ ffree(i);
251 }
253 // UseSSE <= 1 so double result should be left on TOS
254 __ movl(rsi, result_type);
255 __ cmpl(rsi, T_DOUBLE);
256 __ jcc(Assembler::equal, L_skip);
257 if (UseSSE == 0) {
258 // UseSSE == 0 so float result should be left on TOS
259 __ cmpl(rsi, T_FLOAT);
260 __ jcc(Assembler::equal, L_skip);
261 }
262 __ ffree(0);
263 }
264 __ BIND(L_skip);
265 }
266 #endif // COMPILER2
268 // store result depending on type
269 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
270 __ movptr(rdi, result);
271 Label is_long, is_float, is_double, exit;
272 __ movl(rsi, result_type);
273 __ cmpl(rsi, T_LONG);
274 __ jcc(Assembler::equal, is_long);
275 __ cmpl(rsi, T_FLOAT);
276 __ jcc(Assembler::equal, is_float);
277 __ cmpl(rsi, T_DOUBLE);
278 __ jcc(Assembler::equal, is_double);
280 // handle T_INT case
281 __ movl(Address(rdi, 0), rax);
282 __ BIND(exit);
284 // check that FPU stack is empty
285 __ verify_FPU(0, "generate_call_stub");
287 // pop parameters
288 __ lea(rsp, rsp_after_call);
290 // restore %mxcsr
291 if (sse_save) {
292 __ ldmxcsr(mxcsr_save);
293 }
295 // restore rdi, rsi and rbx,
296 __ movptr(rbx, saved_rbx);
297 __ movptr(rsi, saved_rsi);
298 __ movptr(rdi, saved_rdi);
299 __ addptr(rsp, 4*wordSize);
301 // return
302 __ pop(rbp);
303 __ ret(0);
305 // handle return types different from T_INT
306 __ BIND(is_long);
307 __ movl(Address(rdi, 0 * wordSize), rax);
308 __ movl(Address(rdi, 1 * wordSize), rdx);
309 __ jmp(exit);
311 __ BIND(is_float);
312 // interpreter uses xmm0 for return values
313 if (UseSSE >= 1) {
314 __ movflt(Address(rdi, 0), xmm0);
315 } else {
316 __ fstp_s(Address(rdi, 0));
317 }
318 __ jmp(exit);
320 __ BIND(is_double);
321 // interpreter uses xmm0 for return values
322 if (UseSSE >= 2) {
323 __ movdbl(Address(rdi, 0), xmm0);
324 } else {
325 __ fstp_d(Address(rdi, 0));
326 }
327 __ jmp(exit);
329 return start;
330 }
333 //------------------------------------------------------------------------------------------------------------------------
334 // Return point for a Java call if there's an exception thrown in Java code.
335 // The exception is caught and transformed into a pending exception stored in
336 // JavaThread that can be tested from within the VM.
337 //
338 // Note: Usually the parameters are removed by the callee. In case of an exception
339 // crossing an activation frame boundary, that is not the case if the callee
340 // is compiled code => need to setup the rsp.
341 //
342 // rax,: exception oop
344 address generate_catch_exception() {
345 StubCodeMark mark(this, "StubRoutines", "catch_exception");
346 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()!
347 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()!
348 address start = __ pc();
350 // get thread directly
351 __ movptr(rcx, thread);
352 #ifdef ASSERT
353 // verify that threads correspond
354 { Label L;
355 __ get_thread(rbx);
356 __ cmpptr(rbx, rcx);
357 __ jcc(Assembler::equal, L);
358 __ stop("StubRoutines::catch_exception: threads must correspond");
359 __ bind(L);
360 }
361 #endif
362 // set pending exception
363 __ verify_oop(rax);
364 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax );
365 __ lea(Address(rcx, Thread::exception_file_offset ()),
366 ExternalAddress((address)__FILE__));
367 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ );
368 // complete return to VM
369 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
370 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address));
372 return start;
373 }
376 //------------------------------------------------------------------------------------------------------------------------
377 // Continuation point for runtime calls returning with a pending exception.
378 // The pending exception check happened in the runtime or native call stub.
379 // The pending exception in Thread is converted into a Java-level exception.
380 //
381 // Contract with Java-level exception handlers:
382 // rax: exception
383 // rdx: throwing pc
384 //
385 // NOTE: At entry of this stub, exception-pc must be on stack !!
387 address generate_forward_exception() {
388 StubCodeMark mark(this, "StubRoutines", "forward exception");
389 address start = __ pc();
390 const Register thread = rcx;
392 // other registers used in this stub
393 const Register exception_oop = rax;
394 const Register handler_addr = rbx;
395 const Register exception_pc = rdx;
397 // Upon entry, the sp points to the return address returning into Java
398 // (interpreted or compiled) code; i.e., the return address becomes the
399 // throwing pc.
400 //
401 // Arguments pushed before the runtime call are still on the stack but
402 // the exception handler will reset the stack pointer -> ignore them.
403 // A potential result in registers can be ignored as well.
405 #ifdef ASSERT
406 // make sure this code is only executed if there is a pending exception
407 { Label L;
408 __ get_thread(thread);
409 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
410 __ jcc(Assembler::notEqual, L);
411 __ stop("StubRoutines::forward exception: no pending exception (1)");
412 __ bind(L);
413 }
414 #endif
416 // compute exception handler into rbx,
417 __ get_thread(thread);
418 __ movptr(exception_pc, Address(rsp, 0));
419 BLOCK_COMMENT("call exception_handler_for_return_address");
420 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
421 __ mov(handler_addr, rax);
423 // setup rax & rdx, remove return address & clear pending exception
424 __ get_thread(thread);
425 __ pop(exception_pc);
426 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
427 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
429 #ifdef ASSERT
430 // make sure exception is set
431 { Label L;
432 __ testptr(exception_oop, exception_oop);
433 __ jcc(Assembler::notEqual, L);
434 __ stop("StubRoutines::forward exception: no pending exception (2)");
435 __ bind(L);
436 }
437 #endif
439 // Verify that there is really a valid exception in RAX.
440 __ verify_oop(exception_oop);
442 // continue at exception handler (return address removed)
443 // rax: exception
444 // rbx: exception handler
445 // rdx: throwing pc
446 __ jmp(handler_addr);
448 return start;
449 }
452 //----------------------------------------------------------------------------------------------------
453 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest)
454 //
455 // xchg exists as far back as 8086, lock needed for MP only
456 // Stack layout immediately after call:
457 //
458 // 0 [ret addr ] <--- rsp
459 // 1 [ ex ]
460 // 2 [ dest ]
461 //
462 // Result: *dest <- ex, return (old *dest)
463 //
464 // Note: win32 does not currently use this code
466 address generate_atomic_xchg() {
467 StubCodeMark mark(this, "StubRoutines", "atomic_xchg");
468 address start = __ pc();
470 __ push(rdx);
471 Address exchange(rsp, 2 * wordSize);
472 Address dest_addr(rsp, 3 * wordSize);
473 __ movl(rax, exchange);
474 __ movptr(rdx, dest_addr);
475 __ xchgl(rax, Address(rdx, 0));
476 __ pop(rdx);
477 __ ret(0);
479 return start;
480 }
482 //----------------------------------------------------------------------------------------------------
483 // Support for void verify_mxcsr()
484 //
485 // This routine is used with -Xcheck:jni to verify that native
486 // JNI code does not return to Java code without restoring the
487 // MXCSR register to our expected state.
490 address generate_verify_mxcsr() {
491 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr");
492 address start = __ pc();
494 const Address mxcsr_save(rsp, 0);
496 if (CheckJNICalls && UseSSE > 0 ) {
497 Label ok_ret;
498 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std());
499 __ push(rax);
500 __ subptr(rsp, wordSize); // allocate a temp location
501 __ stmxcsr(mxcsr_save);
502 __ movl(rax, mxcsr_save);
503 __ andl(rax, MXCSR_MASK);
504 __ cmp32(rax, mxcsr_std);
505 __ jcc(Assembler::equal, ok_ret);
507 __ warn("MXCSR changed by native JNI code.");
509 __ ldmxcsr(mxcsr_std);
511 __ bind(ok_ret);
512 __ addptr(rsp, wordSize);
513 __ pop(rax);
514 }
516 __ ret(0);
518 return start;
519 }
522 //---------------------------------------------------------------------------
523 // Support for void verify_fpu_cntrl_wrd()
524 //
525 // This routine is used with -Xcheck:jni to verify that native
526 // JNI code does not return to Java code without restoring the
527 // FP control word to our expected state.
529 address generate_verify_fpu_cntrl_wrd() {
530 StubCodeMark mark(this, "StubRoutines", "verify_spcw");
531 address start = __ pc();
533 const Address fpu_cntrl_wrd_save(rsp, 0);
535 if (CheckJNICalls) {
536 Label ok_ret;
537 __ push(rax);
538 __ subptr(rsp, wordSize); // allocate a temp location
539 __ fnstcw(fpu_cntrl_wrd_save);
540 __ movl(rax, fpu_cntrl_wrd_save);
541 __ andl(rax, FPU_CNTRL_WRD_MASK);
542 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std());
543 __ cmp32(rax, fpu_std);
544 __ jcc(Assembler::equal, ok_ret);
546 __ warn("Floating point control word changed by native JNI code.");
548 __ fldcw(fpu_std);
550 __ bind(ok_ret);
551 __ addptr(rsp, wordSize);
552 __ pop(rax);
553 }
555 __ ret(0);
557 return start;
558 }
560 //---------------------------------------------------------------------------
561 // Wrapper for slow-case handling of double-to-integer conversion
562 // d2i or f2i fast case failed either because it is nan or because
563 // of under/overflow.
564 // Input: FPU TOS: float value
565 // Output: rax, (rdx): integer (long) result
567 address generate_d2i_wrapper(BasicType t, address fcn) {
568 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper");
569 address start = __ pc();
571 // Capture info about frame layout
572 enum layout { FPUState_off = 0,
573 rbp_off = FPUStateSizeInWords,
574 rdi_off,
575 rsi_off,
576 rcx_off,
577 rbx_off,
578 saved_argument_off,
579 saved_argument_off2, // 2nd half of double
580 framesize
581 };
583 assert(FPUStateSizeInWords == 27, "update stack layout");
585 // Save outgoing argument to stack across push_FPU_state()
586 __ subptr(rsp, wordSize * 2);
587 __ fstp_d(Address(rsp, 0));
589 // Save CPU & FPU state
590 __ push(rbx);
591 __ push(rcx);
592 __ push(rsi);
593 __ push(rdi);
594 __ push(rbp);
595 __ push_FPU_state();
597 // push_FPU_state() resets the FP top of stack
598 // Load original double into FP top of stack
599 __ fld_d(Address(rsp, saved_argument_off * wordSize));
600 // Store double into stack as outgoing argument
601 __ subptr(rsp, wordSize*2);
602 __ fst_d(Address(rsp, 0));
604 // Prepare FPU for doing math in C-land
605 __ empty_FPU_stack();
606 // Call the C code to massage the double. Result in EAX
607 if (t == T_INT)
608 { BLOCK_COMMENT("SharedRuntime::d2i"); }
609 else if (t == T_LONG)
610 { BLOCK_COMMENT("SharedRuntime::d2l"); }
611 __ call_VM_leaf( fcn, 2 );
613 // Restore CPU & FPU state
614 __ pop_FPU_state();
615 __ pop(rbp);
616 __ pop(rdi);
617 __ pop(rsi);
618 __ pop(rcx);
619 __ pop(rbx);
620 __ addptr(rsp, wordSize * 2);
622 __ ret(0);
624 return start;
625 }
628 //---------------------------------------------------------------------------
629 // The following routine generates a subroutine to throw an asynchronous
630 // UnknownError when an unsafe access gets a fault that could not be
631 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.)
632 address generate_handler_for_unsafe_access() {
633 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access");
634 address start = __ pc();
636 __ push(0); // hole for return address-to-be
637 __ pusha(); // push registers
638 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord);
639 BLOCK_COMMENT("call handle_unsafe_access");
640 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access)));
641 __ movptr(next_pc, rax); // stuff next address
642 __ popa();
643 __ ret(0); // jump to next address
645 return start;
646 }
649 //----------------------------------------------------------------------------------------------------
650 // Non-destructive plausibility checks for oops
652 address generate_verify_oop() {
653 StubCodeMark mark(this, "StubRoutines", "verify_oop");
654 address start = __ pc();
656 // Incoming arguments on stack after saving rax,:
657 //
658 // [tos ]: saved rdx
659 // [tos + 1]: saved EFLAGS
660 // [tos + 2]: return address
661 // [tos + 3]: char* error message
662 // [tos + 4]: oop object to verify
663 // [tos + 5]: saved rax, - saved by caller and bashed
665 Label exit, error;
666 __ pushf();
667 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr()));
668 __ push(rdx); // save rdx
669 // make sure object is 'reasonable'
670 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object
671 __ testptr(rax, rax);
672 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok
674 // Check if the oop is in the right area of memory
675 const int oop_mask = Universe::verify_oop_mask();
676 const int oop_bits = Universe::verify_oop_bits();
677 __ mov(rdx, rax);
678 __ andptr(rdx, oop_mask);
679 __ cmpptr(rdx, oop_bits);
680 __ jcc(Assembler::notZero, error);
682 // make sure klass is 'reasonable'
683 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass
684 __ testptr(rax, rax);
685 __ jcc(Assembler::zero, error); // if klass is NULL it is broken
687 // Check if the klass is in the right area of memory
688 const int klass_mask = Universe::verify_klass_mask();
689 const int klass_bits = Universe::verify_klass_bits();
690 __ mov(rdx, rax);
691 __ andptr(rdx, klass_mask);
692 __ cmpptr(rdx, klass_bits);
693 __ jcc(Assembler::notZero, error);
695 // make sure klass' klass is 'reasonable'
696 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass
697 __ testptr(rax, rax);
698 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
700 __ mov(rdx, rax);
701 __ andptr(rdx, klass_mask);
702 __ cmpptr(rdx, klass_bits);
703 __ jcc(Assembler::notZero, error); // if klass not in right area
704 // of memory it is broken too.
706 // return if everything seems ok
707 __ bind(exit);
708 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
709 __ pop(rdx); // restore rdx
710 __ popf(); // restore EFLAGS
711 __ ret(3 * wordSize); // pop arguments
713 // handle errors
714 __ bind(error);
715 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back
716 __ pop(rdx); // get saved rdx back
717 __ popf(); // get saved EFLAGS off stack -- will be ignored
718 __ pusha(); // push registers (eip = return address & msg are already pushed)
719 BLOCK_COMMENT("call MacroAssembler::debug");
720 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32)));
721 __ popa();
722 __ ret(3 * wordSize); // pop arguments
723 return start;
724 }
726 //
727 // Generate pre-barrier for array stores
728 //
729 // Input:
730 // start - starting address
731 // count - element count
732 void gen_write_ref_array_pre_barrier(Register start, Register count) {
733 assert_different_registers(start, count);
734 BarrierSet* bs = Universe::heap()->barrier_set();
735 switch (bs->kind()) {
736 case BarrierSet::G1SATBCT:
737 case BarrierSet::G1SATBCTLogging:
738 {
739 __ pusha(); // push registers
740 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre),
741 start, count);
742 __ popa();
743 }
744 break;
745 case BarrierSet::CardTableModRef:
746 case BarrierSet::CardTableExtension:
747 case BarrierSet::ModRef:
748 break;
749 default :
750 ShouldNotReachHere();
752 }
753 }
756 //
757 // Generate a post-barrier for an array store
758 //
759 // start - starting address
760 // count - element count
761 //
762 // The two input registers are overwritten.
763 //
764 void gen_write_ref_array_post_barrier(Register start, Register count) {
765 BarrierSet* bs = Universe::heap()->barrier_set();
766 assert_different_registers(start, count);
767 switch (bs->kind()) {
768 case BarrierSet::G1SATBCT:
769 case BarrierSet::G1SATBCTLogging:
770 {
771 __ pusha(); // push registers
772 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post),
773 start, count);
774 __ popa();
775 }
776 break;
778 case BarrierSet::CardTableModRef:
779 case BarrierSet::CardTableExtension:
780 {
781 CardTableModRefBS* ct = (CardTableModRefBS*)bs;
782 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
784 Label L_loop;
785 const Register end = count; // elements count; end == start+count-1
786 assert_different_registers(start, end);
788 __ lea(end, Address(start, count, Address::times_ptr, -wordSize));
789 __ shrptr(start, CardTableModRefBS::card_shift);
790 __ shrptr(end, CardTableModRefBS::card_shift);
791 __ subptr(end, start); // end --> count
792 __ BIND(L_loop);
793 intptr_t disp = (intptr_t) ct->byte_map_base;
794 Address cardtable(start, count, Address::times_1, disp);
795 __ movb(cardtable, 0);
796 __ decrement(count);
797 __ jcc(Assembler::greaterEqual, L_loop);
798 }
799 break;
800 case BarrierSet::ModRef:
801 break;
802 default :
803 ShouldNotReachHere();
805 }
806 }
809 // Copy 64 bytes chunks
810 //
811 // Inputs:
812 // from - source array address
813 // to_from - destination array address - from
814 // qword_count - 8-bytes element count, negative
815 //
816 void xmm_copy_forward(Register from, Register to_from, Register qword_count) {
817 assert( UseSSE >= 2, "supported cpu only" );
818 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
819 // Copy 64-byte chunks
820 __ jmpb(L_copy_64_bytes);
821 __ align(OptoLoopAlignment);
822 __ BIND(L_copy_64_bytes_loop);
824 if(UseUnalignedLoadStores) {
825 __ movdqu(xmm0, Address(from, 0));
826 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0);
827 __ movdqu(xmm1, Address(from, 16));
828 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1);
829 __ movdqu(xmm2, Address(from, 32));
830 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2);
831 __ movdqu(xmm3, Address(from, 48));
832 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3);
834 } else {
835 __ movq(xmm0, Address(from, 0));
836 __ movq(Address(from, to_from, Address::times_1, 0), xmm0);
837 __ movq(xmm1, Address(from, 8));
838 __ movq(Address(from, to_from, Address::times_1, 8), xmm1);
839 __ movq(xmm2, Address(from, 16));
840 __ movq(Address(from, to_from, Address::times_1, 16), xmm2);
841 __ movq(xmm3, Address(from, 24));
842 __ movq(Address(from, to_from, Address::times_1, 24), xmm3);
843 __ movq(xmm4, Address(from, 32));
844 __ movq(Address(from, to_from, Address::times_1, 32), xmm4);
845 __ movq(xmm5, Address(from, 40));
846 __ movq(Address(from, to_from, Address::times_1, 40), xmm5);
847 __ movq(xmm6, Address(from, 48));
848 __ movq(Address(from, to_from, Address::times_1, 48), xmm6);
849 __ movq(xmm7, Address(from, 56));
850 __ movq(Address(from, to_from, Address::times_1, 56), xmm7);
851 }
853 __ addl(from, 64);
854 __ BIND(L_copy_64_bytes);
855 __ subl(qword_count, 8);
856 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
857 __ addl(qword_count, 8);
858 __ jccb(Assembler::zero, L_exit);
859 //
860 // length is too short, just copy qwords
861 //
862 __ BIND(L_copy_8_bytes);
863 __ movq(xmm0, Address(from, 0));
864 __ movq(Address(from, to_from, Address::times_1), xmm0);
865 __ addl(from, 8);
866 __ decrement(qword_count);
867 __ jcc(Assembler::greater, L_copy_8_bytes);
868 __ BIND(L_exit);
869 }
871 // Copy 64 bytes chunks
872 //
873 // Inputs:
874 // from - source array address
875 // to_from - destination array address - from
876 // qword_count - 8-bytes element count, negative
877 //
878 void mmx_copy_forward(Register from, Register to_from, Register qword_count) {
879 assert( VM_Version::supports_mmx(), "supported cpu only" );
880 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit;
881 // Copy 64-byte chunks
882 __ jmpb(L_copy_64_bytes);
883 __ align(OptoLoopAlignment);
884 __ BIND(L_copy_64_bytes_loop);
885 __ movq(mmx0, Address(from, 0));
886 __ movq(mmx1, Address(from, 8));
887 __ movq(mmx2, Address(from, 16));
888 __ movq(Address(from, to_from, Address::times_1, 0), mmx0);
889 __ movq(mmx3, Address(from, 24));
890 __ movq(Address(from, to_from, Address::times_1, 8), mmx1);
891 __ movq(mmx4, Address(from, 32));
892 __ movq(Address(from, to_from, Address::times_1, 16), mmx2);
893 __ movq(mmx5, Address(from, 40));
894 __ movq(Address(from, to_from, Address::times_1, 24), mmx3);
895 __ movq(mmx6, Address(from, 48));
896 __ movq(Address(from, to_from, Address::times_1, 32), mmx4);
897 __ movq(mmx7, Address(from, 56));
898 __ movq(Address(from, to_from, Address::times_1, 40), mmx5);
899 __ movq(Address(from, to_from, Address::times_1, 48), mmx6);
900 __ movq(Address(from, to_from, Address::times_1, 56), mmx7);
901 __ addptr(from, 64);
902 __ BIND(L_copy_64_bytes);
903 __ subl(qword_count, 8);
904 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop);
905 __ addl(qword_count, 8);
906 __ jccb(Assembler::zero, L_exit);
907 //
908 // length is too short, just copy qwords
909 //
910 __ BIND(L_copy_8_bytes);
911 __ movq(mmx0, Address(from, 0));
912 __ movq(Address(from, to_from, Address::times_1), mmx0);
913 __ addptr(from, 8);
914 __ decrement(qword_count);
915 __ jcc(Assembler::greater, L_copy_8_bytes);
916 __ BIND(L_exit);
917 __ emms();
918 }
920 address generate_disjoint_copy(BasicType t, bool aligned,
921 Address::ScaleFactor sf,
922 address* entry, const char *name) {
923 __ align(CodeEntryAlignment);
924 StubCodeMark mark(this, "StubRoutines", name);
925 address start = __ pc();
927 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
928 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes;
930 int shift = Address::times_ptr - sf;
932 const Register from = rsi; // source array address
933 const Register to = rdi; // destination array address
934 const Register count = rcx; // elements count
935 const Register to_from = to; // (to - from)
936 const Register saved_to = rdx; // saved destination array address
938 __ enter(); // required for proper stackwalking of RuntimeStub frame
939 __ push(rsi);
940 __ push(rdi);
941 __ movptr(from , Address(rsp, 12+ 4));
942 __ movptr(to , Address(rsp, 12+ 8));
943 __ movl(count, Address(rsp, 12+ 12));
945 if (entry != NULL) {
946 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
947 BLOCK_COMMENT("Entry:");
948 }
950 if (t == T_OBJECT) {
951 __ testl(count, count);
952 __ jcc(Assembler::zero, L_0_count);
953 gen_write_ref_array_pre_barrier(to, count);
954 __ mov(saved_to, to); // save 'to'
955 }
957 __ subptr(to, from); // to --> to_from
958 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
959 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
960 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) {
961 // align source address at 4 bytes address boundary
962 if (t == T_BYTE) {
963 // One byte misalignment happens only for byte arrays
964 __ testl(from, 1);
965 __ jccb(Assembler::zero, L_skip_align1);
966 __ movb(rax, Address(from, 0));
967 __ movb(Address(from, to_from, Address::times_1, 0), rax);
968 __ increment(from);
969 __ decrement(count);
970 __ BIND(L_skip_align1);
971 }
972 // Two bytes misalignment happens only for byte and short (char) arrays
973 __ testl(from, 2);
974 __ jccb(Assembler::zero, L_skip_align2);
975 __ movw(rax, Address(from, 0));
976 __ movw(Address(from, to_from, Address::times_1, 0), rax);
977 __ addptr(from, 2);
978 __ subl(count, 1<<(shift-1));
979 __ BIND(L_skip_align2);
980 }
981 if (!VM_Version::supports_mmx()) {
982 __ mov(rax, count); // save 'count'
983 __ shrl(count, shift); // bytes count
984 __ addptr(to_from, from);// restore 'to'
985 __ rep_mov();
986 __ subptr(to_from, from);// restore 'to_from'
987 __ mov(count, rax); // restore 'count'
988 __ jmpb(L_copy_2_bytes); // all dwords were copied
989 } else {
990 if (!UseUnalignedLoadStores) {
991 // align to 8 bytes, we know we are 4 byte aligned to start
992 __ testptr(from, 4);
993 __ jccb(Assembler::zero, L_copy_64_bytes);
994 __ movl(rax, Address(from, 0));
995 __ movl(Address(from, to_from, Address::times_1, 0), rax);
996 __ addptr(from, 4);
997 __ subl(count, 1<<shift);
998 }
999 __ BIND(L_copy_64_bytes);
1000 __ mov(rax, count);
1001 __ shrl(rax, shift+1); // 8 bytes chunk count
1002 //
1003 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop
1004 //
1005 if (UseXMMForArrayCopy) {
1006 xmm_copy_forward(from, to_from, rax);
1007 } else {
1008 mmx_copy_forward(from, to_from, rax);
1009 }
1010 }
1011 // copy tailing dword
1012 __ BIND(L_copy_4_bytes);
1013 __ testl(count, 1<<shift);
1014 __ jccb(Assembler::zero, L_copy_2_bytes);
1015 __ movl(rax, Address(from, 0));
1016 __ movl(Address(from, to_from, Address::times_1, 0), rax);
1017 if (t == T_BYTE || t == T_SHORT) {
1018 __ addptr(from, 4);
1019 __ BIND(L_copy_2_bytes);
1020 // copy tailing word
1021 __ testl(count, 1<<(shift-1));
1022 __ jccb(Assembler::zero, L_copy_byte);
1023 __ movw(rax, Address(from, 0));
1024 __ movw(Address(from, to_from, Address::times_1, 0), rax);
1025 if (t == T_BYTE) {
1026 __ addptr(from, 2);
1027 __ BIND(L_copy_byte);
1028 // copy tailing byte
1029 __ testl(count, 1);
1030 __ jccb(Assembler::zero, L_exit);
1031 __ movb(rax, Address(from, 0));
1032 __ movb(Address(from, to_from, Address::times_1, 0), rax);
1033 __ BIND(L_exit);
1034 } else {
1035 __ BIND(L_copy_byte);
1036 }
1037 } else {
1038 __ BIND(L_copy_2_bytes);
1039 }
1041 if (t == T_OBJECT) {
1042 __ movl(count, Address(rsp, 12+12)); // reread 'count'
1043 __ mov(to, saved_to); // restore 'to'
1044 gen_write_ref_array_post_barrier(to, count);
1045 __ BIND(L_0_count);
1046 }
1047 inc_copy_counter_np(t);
1048 __ pop(rdi);
1049 __ pop(rsi);
1050 __ leave(); // required for proper stackwalking of RuntimeStub frame
1051 __ xorptr(rax, rax); // return 0
1052 __ ret(0);
1053 return start;
1054 }
1057 address generate_fill(BasicType t, bool aligned, const char *name) {
1058 __ align(CodeEntryAlignment);
1059 StubCodeMark mark(this, "StubRoutines", name);
1060 address start = __ pc();
1062 BLOCK_COMMENT("Entry:");
1064 const Register to = rdi; // source array address
1065 const Register value = rdx; // value
1066 const Register count = rsi; // elements count
1068 __ enter(); // required for proper stackwalking of RuntimeStub frame
1069 __ push(rsi);
1070 __ push(rdi);
1071 __ movptr(to , Address(rsp, 12+ 4));
1072 __ movl(value, Address(rsp, 12+ 8));
1073 __ movl(count, Address(rsp, 12+ 12));
1075 __ generate_fill(t, aligned, to, value, count, rax, xmm0);
1077 __ pop(rdi);
1078 __ pop(rsi);
1079 __ leave(); // required for proper stackwalking of RuntimeStub frame
1080 __ ret(0);
1081 return start;
1082 }
1084 address generate_conjoint_copy(BasicType t, bool aligned,
1085 Address::ScaleFactor sf,
1086 address nooverlap_target,
1087 address* entry, const char *name) {
1088 __ align(CodeEntryAlignment);
1089 StubCodeMark mark(this, "StubRoutines", name);
1090 address start = __ pc();
1092 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte;
1093 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop;
1095 int shift = Address::times_ptr - sf;
1097 const Register src = rax; // source array address
1098 const Register dst = rdx; // destination array address
1099 const Register from = rsi; // source array address
1100 const Register to = rdi; // destination array address
1101 const Register count = rcx; // elements count
1102 const Register end = rax; // array end address
1104 __ enter(); // required for proper stackwalking of RuntimeStub frame
1105 __ push(rsi);
1106 __ push(rdi);
1107 __ movptr(src , Address(rsp, 12+ 4)); // from
1108 __ movptr(dst , Address(rsp, 12+ 8)); // to
1109 __ movl2ptr(count, Address(rsp, 12+12)); // count
1111 if (entry != NULL) {
1112 *entry = __ pc(); // Entry point from generic arraycopy stub.
1113 BLOCK_COMMENT("Entry:");
1114 }
1116 // nooverlap_target expects arguments in rsi and rdi.
1117 __ mov(from, src);
1118 __ mov(to , dst);
1120 // arrays overlap test: dispatch to disjoint stub if necessary.
1121 RuntimeAddress nooverlap(nooverlap_target);
1122 __ cmpptr(dst, src);
1123 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size
1124 __ jump_cc(Assembler::belowEqual, nooverlap);
1125 __ cmpptr(dst, end);
1126 __ jump_cc(Assembler::aboveEqual, nooverlap);
1128 if (t == T_OBJECT) {
1129 __ testl(count, count);
1130 __ jcc(Assembler::zero, L_0_count);
1131 gen_write_ref_array_pre_barrier(dst, count);
1132 }
1134 // copy from high to low
1135 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1136 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp
1137 if (t == T_BYTE || t == T_SHORT) {
1138 // Align the end of destination array at 4 bytes address boundary
1139 __ lea(end, Address(dst, count, sf, 0));
1140 if (t == T_BYTE) {
1141 // One byte misalignment happens only for byte arrays
1142 __ testl(end, 1);
1143 __ jccb(Assembler::zero, L_skip_align1);
1144 __ decrement(count);
1145 __ movb(rdx, Address(from, count, sf, 0));
1146 __ movb(Address(to, count, sf, 0), rdx);
1147 __ BIND(L_skip_align1);
1148 }
1149 // Two bytes misalignment happens only for byte and short (char) arrays
1150 __ testl(end, 2);
1151 __ jccb(Assembler::zero, L_skip_align2);
1152 __ subptr(count, 1<<(shift-1));
1153 __ movw(rdx, Address(from, count, sf, 0));
1154 __ movw(Address(to, count, sf, 0), rdx);
1155 __ BIND(L_skip_align2);
1156 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element
1157 __ jcc(Assembler::below, L_copy_4_bytes);
1158 }
1160 if (!VM_Version::supports_mmx()) {
1161 __ std();
1162 __ mov(rax, count); // Save 'count'
1163 __ mov(rdx, to); // Save 'to'
1164 __ lea(rsi, Address(from, count, sf, -4));
1165 __ lea(rdi, Address(to , count, sf, -4));
1166 __ shrptr(count, shift); // bytes count
1167 __ rep_mov();
1168 __ cld();
1169 __ mov(count, rax); // restore 'count'
1170 __ andl(count, (1<<shift)-1); // mask the number of rest elements
1171 __ movptr(from, Address(rsp, 12+4)); // reread 'from'
1172 __ mov(to, rdx); // restore 'to'
1173 __ jmpb(L_copy_2_bytes); // all dword were copied
1174 } else {
1175 // Align to 8 bytes the end of array. It is aligned to 4 bytes already.
1176 __ testptr(end, 4);
1177 __ jccb(Assembler::zero, L_copy_8_bytes);
1178 __ subl(count, 1<<shift);
1179 __ movl(rdx, Address(from, count, sf, 0));
1180 __ movl(Address(to, count, sf, 0), rdx);
1181 __ jmpb(L_copy_8_bytes);
1183 __ align(OptoLoopAlignment);
1184 // Move 8 bytes
1185 __ BIND(L_copy_8_bytes_loop);
1186 if (UseXMMForArrayCopy) {
1187 __ movq(xmm0, Address(from, count, sf, 0));
1188 __ movq(Address(to, count, sf, 0), xmm0);
1189 } else {
1190 __ movq(mmx0, Address(from, count, sf, 0));
1191 __ movq(Address(to, count, sf, 0), mmx0);
1192 }
1193 __ BIND(L_copy_8_bytes);
1194 __ subl(count, 2<<shift);
1195 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1196 __ addl(count, 2<<shift);
1197 if (!UseXMMForArrayCopy) {
1198 __ emms();
1199 }
1200 }
1201 __ BIND(L_copy_4_bytes);
1202 // copy prefix qword
1203 __ testl(count, 1<<shift);
1204 __ jccb(Assembler::zero, L_copy_2_bytes);
1205 __ movl(rdx, Address(from, count, sf, -4));
1206 __ movl(Address(to, count, sf, -4), rdx);
1208 if (t == T_BYTE || t == T_SHORT) {
1209 __ subl(count, (1<<shift));
1210 __ BIND(L_copy_2_bytes);
1211 // copy prefix dword
1212 __ testl(count, 1<<(shift-1));
1213 __ jccb(Assembler::zero, L_copy_byte);
1214 __ movw(rdx, Address(from, count, sf, -2));
1215 __ movw(Address(to, count, sf, -2), rdx);
1216 if (t == T_BYTE) {
1217 __ subl(count, 1<<(shift-1));
1218 __ BIND(L_copy_byte);
1219 // copy prefix byte
1220 __ testl(count, 1);
1221 __ jccb(Assembler::zero, L_exit);
1222 __ movb(rdx, Address(from, 0));
1223 __ movb(Address(to, 0), rdx);
1224 __ BIND(L_exit);
1225 } else {
1226 __ BIND(L_copy_byte);
1227 }
1228 } else {
1229 __ BIND(L_copy_2_bytes);
1230 }
1231 if (t == T_OBJECT) {
1232 __ movl2ptr(count, Address(rsp, 12+12)); // reread count
1233 gen_write_ref_array_post_barrier(to, count);
1234 __ BIND(L_0_count);
1235 }
1236 inc_copy_counter_np(t);
1237 __ pop(rdi);
1238 __ pop(rsi);
1239 __ leave(); // required for proper stackwalking of RuntimeStub frame
1240 __ xorptr(rax, rax); // return 0
1241 __ ret(0);
1242 return start;
1243 }
1246 address generate_disjoint_long_copy(address* entry, const char *name) {
1247 __ align(CodeEntryAlignment);
1248 StubCodeMark mark(this, "StubRoutines", name);
1249 address start = __ pc();
1251 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1252 const Register from = rax; // source array address
1253 const Register to = rdx; // destination array address
1254 const Register count = rcx; // elements count
1255 const Register to_from = rdx; // (to - from)
1257 __ enter(); // required for proper stackwalking of RuntimeStub frame
1258 __ movptr(from , Address(rsp, 8+0)); // from
1259 __ movptr(to , Address(rsp, 8+4)); // to
1260 __ movl2ptr(count, Address(rsp, 8+8)); // count
1262 *entry = __ pc(); // Entry point from conjoint arraycopy stub.
1263 BLOCK_COMMENT("Entry:");
1265 __ subptr(to, from); // to --> to_from
1266 if (VM_Version::supports_mmx()) {
1267 if (UseXMMForArrayCopy) {
1268 xmm_copy_forward(from, to_from, count);
1269 } else {
1270 mmx_copy_forward(from, to_from, count);
1271 }
1272 } else {
1273 __ jmpb(L_copy_8_bytes);
1274 __ align(OptoLoopAlignment);
1275 __ BIND(L_copy_8_bytes_loop);
1276 __ fild_d(Address(from, 0));
1277 __ fistp_d(Address(from, to_from, Address::times_1));
1278 __ addptr(from, 8);
1279 __ BIND(L_copy_8_bytes);
1280 __ decrement(count);
1281 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1282 }
1283 inc_copy_counter_np(T_LONG);
1284 __ leave(); // required for proper stackwalking of RuntimeStub frame
1285 __ xorptr(rax, rax); // return 0
1286 __ ret(0);
1287 return start;
1288 }
1290 address generate_conjoint_long_copy(address nooverlap_target,
1291 address* entry, const char *name) {
1292 __ align(CodeEntryAlignment);
1293 StubCodeMark mark(this, "StubRoutines", name);
1294 address start = __ pc();
1296 Label L_copy_8_bytes, L_copy_8_bytes_loop;
1297 const Register from = rax; // source array address
1298 const Register to = rdx; // destination array address
1299 const Register count = rcx; // elements count
1300 const Register end_from = rax; // source array end address
1302 __ enter(); // required for proper stackwalking of RuntimeStub frame
1303 __ movptr(from , Address(rsp, 8+0)); // from
1304 __ movptr(to , Address(rsp, 8+4)); // to
1305 __ movl2ptr(count, Address(rsp, 8+8)); // count
1307 *entry = __ pc(); // Entry point from generic arraycopy stub.
1308 BLOCK_COMMENT("Entry:");
1310 // arrays overlap test
1311 __ cmpptr(to, from);
1312 RuntimeAddress nooverlap(nooverlap_target);
1313 __ jump_cc(Assembler::belowEqual, nooverlap);
1314 __ lea(end_from, Address(from, count, Address::times_8, 0));
1315 __ cmpptr(to, end_from);
1316 __ movptr(from, Address(rsp, 8)); // from
1317 __ jump_cc(Assembler::aboveEqual, nooverlap);
1319 __ jmpb(L_copy_8_bytes);
1321 __ align(OptoLoopAlignment);
1322 __ BIND(L_copy_8_bytes_loop);
1323 if (VM_Version::supports_mmx()) {
1324 if (UseXMMForArrayCopy) {
1325 __ movq(xmm0, Address(from, count, Address::times_8));
1326 __ movq(Address(to, count, Address::times_8), xmm0);
1327 } else {
1328 __ movq(mmx0, Address(from, count, Address::times_8));
1329 __ movq(Address(to, count, Address::times_8), mmx0);
1330 }
1331 } else {
1332 __ fild_d(Address(from, count, Address::times_8));
1333 __ fistp_d(Address(to, count, Address::times_8));
1334 }
1335 __ BIND(L_copy_8_bytes);
1336 __ decrement(count);
1337 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop);
1339 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) {
1340 __ emms();
1341 }
1342 inc_copy_counter_np(T_LONG);
1343 __ leave(); // required for proper stackwalking of RuntimeStub frame
1344 __ xorptr(rax, rax); // return 0
1345 __ ret(0);
1346 return start;
1347 }
1350 // Helper for generating a dynamic type check.
1351 // The sub_klass must be one of {rbx, rdx, rsi}.
1352 // The temp is killed.
1353 void generate_type_check(Register sub_klass,
1354 Address& super_check_offset_addr,
1355 Address& super_klass_addr,
1356 Register temp,
1357 Label* L_success, Label* L_failure) {
1358 BLOCK_COMMENT("type_check:");
1360 Label L_fallthrough;
1361 #define LOCAL_JCC(assembler_con, label_ptr) \
1362 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \
1363 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/
1365 // The following is a strange variation of the fast path which requires
1366 // one less register, because needed values are on the argument stack.
1367 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp,
1368 // L_success, L_failure, NULL);
1369 assert_different_registers(sub_klass, temp);
1371 int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
1372 Klass::secondary_super_cache_offset_in_bytes());
1374 // if the pointers are equal, we are done (e.g., String[] elements)
1375 __ cmpptr(sub_klass, super_klass_addr);
1376 LOCAL_JCC(Assembler::equal, L_success);
1378 // check the supertype display:
1379 __ movl2ptr(temp, super_check_offset_addr);
1380 Address super_check_addr(sub_klass, temp, Address::times_1, 0);
1381 __ movptr(temp, super_check_addr); // load displayed supertype
1382 __ cmpptr(temp, super_klass_addr); // test the super type
1383 LOCAL_JCC(Assembler::equal, L_success);
1385 // if it was a primary super, we can just fail immediately
1386 __ cmpl(super_check_offset_addr, sc_offset);
1387 LOCAL_JCC(Assembler::notEqual, L_failure);
1389 // The repne_scan instruction uses fixed registers, which will get spilled.
1390 // We happen to know this works best when super_klass is in rax.
1391 Register super_klass = temp;
1392 __ movptr(super_klass, super_klass_addr);
1393 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg,
1394 L_success, L_failure);
1396 __ bind(L_fallthrough);
1398 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); }
1399 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); }
1401 #undef LOCAL_JCC
1402 }
1404 //
1405 // Generate checkcasting array copy stub
1406 //
1407 // Input:
1408 // 4(rsp) - source array address
1409 // 8(rsp) - destination array address
1410 // 12(rsp) - element count, can be zero
1411 // 16(rsp) - size_t ckoff (super_check_offset)
1412 // 20(rsp) - oop ckval (super_klass)
1413 //
1414 // Output:
1415 // rax, == 0 - success
1416 // rax, == -1^K - failure, where K is partial transfer count
1417 //
1418 address generate_checkcast_copy(const char *name, address* entry) {
1419 __ align(CodeEntryAlignment);
1420 StubCodeMark mark(this, "StubRoutines", name);
1421 address start = __ pc();
1423 Label L_load_element, L_store_element, L_do_card_marks, L_done;
1425 // register use:
1426 // rax, rdx, rcx -- loop control (end_from, end_to, count)
1427 // rdi, rsi -- element access (oop, klass)
1428 // rbx, -- temp
1429 const Register from = rax; // source array address
1430 const Register to = rdx; // destination array address
1431 const Register length = rcx; // elements count
1432 const Register elem = rdi; // each oop copied
1433 const Register elem_klass = rsi; // each elem._klass (sub_klass)
1434 const Register temp = rbx; // lone remaining temp
1436 __ enter(); // required for proper stackwalking of RuntimeStub frame
1438 __ push(rsi);
1439 __ push(rdi);
1440 __ push(rbx);
1442 Address from_arg(rsp, 16+ 4); // from
1443 Address to_arg(rsp, 16+ 8); // to
1444 Address length_arg(rsp, 16+12); // elements count
1445 Address ckoff_arg(rsp, 16+16); // super_check_offset
1446 Address ckval_arg(rsp, 16+20); // super_klass
1448 // Load up:
1449 __ movptr(from, from_arg);
1450 __ movptr(to, to_arg);
1451 __ movl2ptr(length, length_arg);
1453 if (entry != NULL) {
1454 *entry = __ pc(); // Entry point from generic arraycopy stub.
1455 BLOCK_COMMENT("Entry:");
1456 }
1458 //---------------------------------------------------------------
1459 // Assembler stub will be used for this call to arraycopy
1460 // if the two arrays are subtypes of Object[] but the
1461 // destination array type is not equal to or a supertype
1462 // of the source type. Each element must be separately
1463 // checked.
1465 // Loop-invariant addresses. They are exclusive end pointers.
1466 Address end_from_addr(from, length, Address::times_ptr, 0);
1467 Address end_to_addr(to, length, Address::times_ptr, 0);
1469 Register end_from = from; // re-use
1470 Register end_to = to; // re-use
1471 Register count = length; // re-use
1473 // Loop-variant addresses. They assume post-incremented count < 0.
1474 Address from_element_addr(end_from, count, Address::times_ptr, 0);
1475 Address to_element_addr(end_to, count, Address::times_ptr, 0);
1476 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes());
1478 // Copy from low to high addresses, indexed from the end of each array.
1479 gen_write_ref_array_pre_barrier(to, count);
1480 __ lea(end_from, end_from_addr);
1481 __ lea(end_to, end_to_addr);
1482 assert(length == count, ""); // else fix next line:
1483 __ negptr(count); // negate and test the length
1484 __ jccb(Assembler::notZero, L_load_element);
1486 // Empty array: Nothing to do.
1487 __ xorptr(rax, rax); // return 0 on (trivial) success
1488 __ jmp(L_done);
1490 // ======== begin loop ========
1491 // (Loop is rotated; its entry is L_load_element.)
1492 // Loop control:
1493 // for (count = -count; count != 0; count++)
1494 // Base pointers src, dst are biased by 8*count,to last element.
1495 __ align(OptoLoopAlignment);
1497 __ BIND(L_store_element);
1498 __ movptr(to_element_addr, elem); // store the oop
1499 __ increment(count); // increment the count toward zero
1500 __ jccb(Assembler::zero, L_do_card_marks);
1502 // ======== loop entry is here ========
1503 __ BIND(L_load_element);
1504 __ movptr(elem, from_element_addr); // load the oop
1505 __ testptr(elem, elem);
1506 __ jccb(Assembler::zero, L_store_element);
1508 // (Could do a trick here: Remember last successful non-null
1509 // element stored and make a quick oop equality check on it.)
1511 __ movptr(elem_klass, elem_klass_addr); // query the object klass
1512 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp,
1513 &L_store_element, NULL);
1514 // (On fall-through, we have failed the element type check.)
1515 // ======== end loop ========
1517 // It was a real error; we must depend on the caller to finish the job.
1518 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops.
1519 // Emit GC store barriers for the oops we have copied (length_arg + count),
1520 // and report their number to the caller.
1521 __ addl(count, length_arg); // transfers = (length - remaining)
1522 __ movl2ptr(rax, count); // save the value
1523 __ notptr(rax); // report (-1^K) to caller
1524 __ movptr(to, to_arg); // reload
1525 assert_different_registers(to, count, rax);
1526 gen_write_ref_array_post_barrier(to, count);
1527 __ jmpb(L_done);
1529 // Come here on success only.
1530 __ BIND(L_do_card_marks);
1531 __ movl2ptr(count, length_arg);
1532 __ movptr(to, to_arg); // reload
1533 gen_write_ref_array_post_barrier(to, count);
1534 __ xorptr(rax, rax); // return 0 on success
1536 // Common exit point (success or failure).
1537 __ BIND(L_done);
1538 __ pop(rbx);
1539 __ pop(rdi);
1540 __ pop(rsi);
1541 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr);
1542 __ leave(); // required for proper stackwalking of RuntimeStub frame
1543 __ ret(0);
1545 return start;
1546 }
1548 //
1549 // Generate 'unsafe' array copy stub
1550 // Though just as safe as the other stubs, it takes an unscaled
1551 // size_t argument instead of an element count.
1552 //
1553 // Input:
1554 // 4(rsp) - source array address
1555 // 8(rsp) - destination array address
1556 // 12(rsp) - byte count, can be zero
1557 //
1558 // Output:
1559 // rax, == 0 - success
1560 // rax, == -1 - need to call System.arraycopy
1561 //
1562 // Examines the alignment of the operands and dispatches
1563 // to a long, int, short, or byte copy loop.
1564 //
1565 address generate_unsafe_copy(const char *name,
1566 address byte_copy_entry,
1567 address short_copy_entry,
1568 address int_copy_entry,
1569 address long_copy_entry) {
1571 Label L_long_aligned, L_int_aligned, L_short_aligned;
1573 __ align(CodeEntryAlignment);
1574 StubCodeMark mark(this, "StubRoutines", name);
1575 address start = __ pc();
1577 const Register from = rax; // source array address
1578 const Register to = rdx; // destination array address
1579 const Register count = rcx; // elements count
1581 __ enter(); // required for proper stackwalking of RuntimeStub frame
1582 __ push(rsi);
1583 __ push(rdi);
1584 Address from_arg(rsp, 12+ 4); // from
1585 Address to_arg(rsp, 12+ 8); // to
1586 Address count_arg(rsp, 12+12); // byte count
1588 // Load up:
1589 __ movptr(from , from_arg);
1590 __ movptr(to , to_arg);
1591 __ movl2ptr(count, count_arg);
1593 // bump this on entry, not on exit:
1594 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr);
1596 const Register bits = rsi;
1597 __ mov(bits, from);
1598 __ orptr(bits, to);
1599 __ orptr(bits, count);
1601 __ testl(bits, BytesPerLong-1);
1602 __ jccb(Assembler::zero, L_long_aligned);
1604 __ testl(bits, BytesPerInt-1);
1605 __ jccb(Assembler::zero, L_int_aligned);
1607 __ testl(bits, BytesPerShort-1);
1608 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry));
1610 __ BIND(L_short_aligned);
1611 __ shrptr(count, LogBytesPerShort); // size => short_count
1612 __ movl(count_arg, count); // update 'count'
1613 __ jump(RuntimeAddress(short_copy_entry));
1615 __ BIND(L_int_aligned);
1616 __ shrptr(count, LogBytesPerInt); // size => int_count
1617 __ movl(count_arg, count); // update 'count'
1618 __ jump(RuntimeAddress(int_copy_entry));
1620 __ BIND(L_long_aligned);
1621 __ shrptr(count, LogBytesPerLong); // size => qword_count
1622 __ movl(count_arg, count); // update 'count'
1623 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1624 __ pop(rsi);
1625 __ jump(RuntimeAddress(long_copy_entry));
1627 return start;
1628 }
1631 // Perform range checks on the proposed arraycopy.
1632 // Smashes src_pos and dst_pos. (Uses them up for temps.)
1633 void arraycopy_range_checks(Register src,
1634 Register src_pos,
1635 Register dst,
1636 Register dst_pos,
1637 Address& length,
1638 Label& L_failed) {
1639 BLOCK_COMMENT("arraycopy_range_checks:");
1640 const Register src_end = src_pos; // source array end position
1641 const Register dst_end = dst_pos; // destination array end position
1642 __ addl(src_end, length); // src_pos + length
1643 __ addl(dst_end, length); // dst_pos + length
1645 // if (src_pos + length > arrayOop(src)->length() ) FAIL;
1646 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes()));
1647 __ jcc(Assembler::above, L_failed);
1649 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL;
1650 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes()));
1651 __ jcc(Assembler::above, L_failed);
1653 BLOCK_COMMENT("arraycopy_range_checks done");
1654 }
1657 //
1658 // Generate generic array copy stubs
1659 //
1660 // Input:
1661 // 4(rsp) - src oop
1662 // 8(rsp) - src_pos
1663 // 12(rsp) - dst oop
1664 // 16(rsp) - dst_pos
1665 // 20(rsp) - element count
1666 //
1667 // Output:
1668 // rax, == 0 - success
1669 // rax, == -1^K - failure, where K is partial transfer count
1670 //
1671 address generate_generic_copy(const char *name,
1672 address entry_jbyte_arraycopy,
1673 address entry_jshort_arraycopy,
1674 address entry_jint_arraycopy,
1675 address entry_oop_arraycopy,
1676 address entry_jlong_arraycopy,
1677 address entry_checkcast_arraycopy) {
1678 Label L_failed, L_failed_0, L_objArray;
1680 { int modulus = CodeEntryAlignment;
1681 int target = modulus - 5; // 5 = sizeof jmp(L_failed)
1682 int advance = target - (__ offset() % modulus);
1683 if (advance < 0) advance += modulus;
1684 if (advance > 0) __ nop(advance);
1685 }
1686 StubCodeMark mark(this, "StubRoutines", name);
1688 // Short-hop target to L_failed. Makes for denser prologue code.
1689 __ BIND(L_failed_0);
1690 __ jmp(L_failed);
1691 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed");
1693 __ align(CodeEntryAlignment);
1694 address start = __ pc();
1696 __ enter(); // required for proper stackwalking of RuntimeStub frame
1697 __ push(rsi);
1698 __ push(rdi);
1700 // bump this on entry, not on exit:
1701 inc_counter_np(SharedRuntime::_generic_array_copy_ctr);
1703 // Input values
1704 Address SRC (rsp, 12+ 4);
1705 Address SRC_POS (rsp, 12+ 8);
1706 Address DST (rsp, 12+12);
1707 Address DST_POS (rsp, 12+16);
1708 Address LENGTH (rsp, 12+20);
1710 //-----------------------------------------------------------------------
1711 // Assembler stub will be used for this call to arraycopy
1712 // if the following conditions are met:
1713 //
1714 // (1) src and dst must not be null.
1715 // (2) src_pos must not be negative.
1716 // (3) dst_pos must not be negative.
1717 // (4) length must not be negative.
1718 // (5) src klass and dst klass should be the same and not NULL.
1719 // (6) src and dst should be arrays.
1720 // (7) src_pos + length must not exceed length of src.
1721 // (8) dst_pos + length must not exceed length of dst.
1722 //
1724 const Register src = rax; // source array oop
1725 const Register src_pos = rsi;
1726 const Register dst = rdx; // destination array oop
1727 const Register dst_pos = rdi;
1728 const Register length = rcx; // transfer count
1730 // if (src == NULL) return -1;
1731 __ movptr(src, SRC); // src oop
1732 __ testptr(src, src);
1733 __ jccb(Assembler::zero, L_failed_0);
1735 // if (src_pos < 0) return -1;
1736 __ movl2ptr(src_pos, SRC_POS); // src_pos
1737 __ testl(src_pos, src_pos);
1738 __ jccb(Assembler::negative, L_failed_0);
1740 // if (dst == NULL) return -1;
1741 __ movptr(dst, DST); // dst oop
1742 __ testptr(dst, dst);
1743 __ jccb(Assembler::zero, L_failed_0);
1745 // if (dst_pos < 0) return -1;
1746 __ movl2ptr(dst_pos, DST_POS); // dst_pos
1747 __ testl(dst_pos, dst_pos);
1748 __ jccb(Assembler::negative, L_failed_0);
1750 // if (length < 0) return -1;
1751 __ movl2ptr(length, LENGTH); // length
1752 __ testl(length, length);
1753 __ jccb(Assembler::negative, L_failed_0);
1755 // if (src->klass() == NULL) return -1;
1756 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes());
1757 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes());
1758 const Register rcx_src_klass = rcx; // array klass
1759 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes()));
1761 #ifdef ASSERT
1762 // assert(src->klass() != NULL);
1763 BLOCK_COMMENT("assert klasses not null");
1764 { Label L1, L2;
1765 __ testptr(rcx_src_klass, rcx_src_klass);
1766 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL
1767 __ bind(L1);
1768 __ stop("broken null klass");
1769 __ bind(L2);
1770 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD);
1771 __ jccb(Assembler::equal, L1); // this would be broken also
1772 BLOCK_COMMENT("assert done");
1773 }
1774 #endif //ASSERT
1776 // Load layout helper (32-bits)
1777 //
1778 // |array_tag| | header_size | element_type | |log2_element_size|
1779 // 32 30 24 16 8 2 0
1780 //
1781 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0
1782 //
1784 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
1785 Klass::layout_helper_offset_in_bytes();
1786 Address src_klass_lh_addr(rcx_src_klass, lh_offset);
1788 // Handle objArrays completely differently...
1789 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
1790 __ cmpl(src_klass_lh_addr, objArray_lh);
1791 __ jcc(Assembler::equal, L_objArray);
1793 // if (src->klass() != dst->klass()) return -1;
1794 __ cmpptr(rcx_src_klass, dst_klass_addr);
1795 __ jccb(Assembler::notEqual, L_failed_0);
1797 const Register rcx_lh = rcx; // layout helper
1798 assert(rcx_lh == rcx_src_klass, "known alias");
1799 __ movl(rcx_lh, src_klass_lh_addr);
1801 // if (!src->is_Array()) return -1;
1802 __ cmpl(rcx_lh, Klass::_lh_neutral_value);
1803 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp
1805 // At this point, it is known to be a typeArray (array_tag 0x3).
1806 #ifdef ASSERT
1807 { Label L;
1808 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift));
1809 __ jcc(Assembler::greaterEqual, L); // signed cmp
1810 __ stop("must be a primitive array");
1811 __ bind(L);
1812 }
1813 #endif
1815 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh);
1816 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1818 // typeArrayKlass
1819 //
1820 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize);
1821 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize);
1822 //
1823 const Register rsi_offset = rsi; // array offset
1824 const Register src_array = src; // src array offset
1825 const Register dst_array = dst; // dst array offset
1826 const Register rdi_elsize = rdi; // log2 element size
1828 __ mov(rsi_offset, rcx_lh);
1829 __ shrptr(rsi_offset, Klass::_lh_header_size_shift);
1830 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset
1831 __ addptr(src_array, rsi_offset); // src array offset
1832 __ addptr(dst_array, rsi_offset); // dst array offset
1833 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize
1835 // next registers should be set before the jump to corresponding stub
1836 const Register from = src; // source array address
1837 const Register to = dst; // destination array address
1838 const Register count = rcx; // elements count
1839 // some of them should be duplicated on stack
1840 #define FROM Address(rsp, 12+ 4)
1841 #define TO Address(rsp, 12+ 8) // Not used now
1842 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy
1844 BLOCK_COMMENT("scale indexes to element size");
1845 __ movl2ptr(rsi, SRC_POS); // src_pos
1846 __ shlptr(rsi); // src_pos << rcx (log2 elsize)
1847 assert(src_array == from, "");
1848 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize
1849 __ movl2ptr(rdi, DST_POS); // dst_pos
1850 __ shlptr(rdi); // dst_pos << rcx (log2 elsize)
1851 assert(dst_array == to, "");
1852 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize
1853 __ movptr(FROM, from); // src_addr
1854 __ mov(rdi_elsize, rcx_lh); // log2 elsize
1855 __ movl2ptr(count, LENGTH); // elements count
1857 BLOCK_COMMENT("choose copy loop based on element size");
1858 __ cmpl(rdi_elsize, 0);
1860 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy));
1861 __ cmpl(rdi_elsize, LogBytesPerShort);
1862 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy));
1863 __ cmpl(rdi_elsize, LogBytesPerInt);
1864 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy));
1865 #ifdef ASSERT
1866 __ cmpl(rdi_elsize, LogBytesPerLong);
1867 __ jccb(Assembler::notEqual, L_failed);
1868 #endif
1869 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it.
1870 __ pop(rsi);
1871 __ jump(RuntimeAddress(entry_jlong_arraycopy));
1873 __ BIND(L_failed);
1874 __ xorptr(rax, rax);
1875 __ notptr(rax); // return -1
1876 __ pop(rdi);
1877 __ pop(rsi);
1878 __ leave(); // required for proper stackwalking of RuntimeStub frame
1879 __ ret(0);
1881 // objArrayKlass
1882 __ BIND(L_objArray);
1883 // live at this point: rcx_src_klass, src[_pos], dst[_pos]
1885 Label L_plain_copy, L_checkcast_copy;
1886 // test array classes for subtyping
1887 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality
1888 __ jccb(Assembler::notEqual, L_checkcast_copy);
1890 // Identically typed arrays can be copied without element-wise checks.
1891 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass);
1892 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1894 __ BIND(L_plain_copy);
1895 __ movl2ptr(count, LENGTH); // elements count
1896 __ movl2ptr(src_pos, SRC_POS); // reload src_pos
1897 __ lea(from, Address(src, src_pos, Address::times_ptr,
1898 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr
1899 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos
1900 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1901 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr
1902 __ movptr(FROM, from); // src_addr
1903 __ movptr(TO, to); // dst_addr
1904 __ movl(COUNT, count); // count
1905 __ jump(RuntimeAddress(entry_oop_arraycopy));
1907 __ BIND(L_checkcast_copy);
1908 // live at this point: rcx_src_klass, dst[_pos], src[_pos]
1909 {
1910 // Handy offsets:
1911 int ek_offset = (klassOopDesc::header_size() * HeapWordSize +
1912 objArrayKlass::element_klass_offset_in_bytes());
1913 int sco_offset = (klassOopDesc::header_size() * HeapWordSize +
1914 Klass::super_check_offset_offset_in_bytes());
1916 Register rsi_dst_klass = rsi;
1917 Register rdi_temp = rdi;
1918 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos");
1919 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos");
1920 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset);
1922 // Before looking at dst.length, make sure dst is also an objArray.
1923 __ movptr(rsi_dst_klass, dst_klass_addr);
1924 __ cmpl(dst_klass_lh_addr, objArray_lh);
1925 __ jccb(Assembler::notEqual, L_failed);
1927 // It is safe to examine both src.length and dst.length.
1928 __ movl2ptr(src_pos, SRC_POS); // reload rsi
1929 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed);
1930 // (Now src_pos and dst_pos are killed, but not src and dst.)
1932 // We'll need this temp (don't forget to pop it after the type check).
1933 __ push(rbx);
1934 Register rbx_src_klass = rbx;
1936 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx
1937 __ movptr(rsi_dst_klass, dst_klass_addr);
1938 Address super_check_offset_addr(rsi_dst_klass, sco_offset);
1939 Label L_fail_array_check;
1940 generate_type_check(rbx_src_klass,
1941 super_check_offset_addr, dst_klass_addr,
1942 rdi_temp, NULL, &L_fail_array_check);
1943 // (On fall-through, we have passed the array type check.)
1944 __ pop(rbx);
1945 __ jmp(L_plain_copy);
1947 __ BIND(L_fail_array_check);
1948 // Reshuffle arguments so we can call checkcast_arraycopy:
1950 // match initial saves for checkcast_arraycopy
1951 // push(rsi); // already done; see above
1952 // push(rdi); // already done; see above
1953 // push(rbx); // already done; see above
1955 // Marshal outgoing arguments now, freeing registers.
1956 Address from_arg(rsp, 16+ 4); // from
1957 Address to_arg(rsp, 16+ 8); // to
1958 Address length_arg(rsp, 16+12); // elements count
1959 Address ckoff_arg(rsp, 16+16); // super_check_offset
1960 Address ckval_arg(rsp, 16+20); // super_klass
1962 Address SRC_POS_arg(rsp, 16+ 8);
1963 Address DST_POS_arg(rsp, 16+16);
1964 Address LENGTH_arg(rsp, 16+20);
1965 // push rbx, changed the incoming offsets (why not just use rbp,??)
1966 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, "");
1968 __ movptr(rbx, Address(rsi_dst_klass, ek_offset));
1969 __ movl2ptr(length, LENGTH_arg); // reload elements count
1970 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos
1971 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos
1973 __ movptr(ckval_arg, rbx); // destination element type
1974 __ movl(rbx, Address(rbx, sco_offset));
1975 __ movl(ckoff_arg, rbx); // corresponding class check offset
1977 __ movl(length_arg, length); // outgoing length argument
1979 __ lea(from, Address(src, src_pos, Address::times_ptr,
1980 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1981 __ movptr(from_arg, from);
1983 __ lea(to, Address(dst, dst_pos, Address::times_ptr,
1984 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
1985 __ movptr(to_arg, to);
1986 __ jump(RuntimeAddress(entry_checkcast_arraycopy));
1987 }
1989 return start;
1990 }
1992 void generate_arraycopy_stubs() {
1993 address entry;
1994 address entry_jbyte_arraycopy;
1995 address entry_jshort_arraycopy;
1996 address entry_jint_arraycopy;
1997 address entry_oop_arraycopy;
1998 address entry_jlong_arraycopy;
1999 address entry_checkcast_arraycopy;
2001 StubRoutines::_arrayof_jbyte_disjoint_arraycopy =
2002 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry,
2003 "arrayof_jbyte_disjoint_arraycopy");
2004 StubRoutines::_arrayof_jbyte_arraycopy =
2005 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry,
2006 NULL, "arrayof_jbyte_arraycopy");
2007 StubRoutines::_jbyte_disjoint_arraycopy =
2008 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry,
2009 "jbyte_disjoint_arraycopy");
2010 StubRoutines::_jbyte_arraycopy =
2011 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry,
2012 &entry_jbyte_arraycopy, "jbyte_arraycopy");
2014 StubRoutines::_arrayof_jshort_disjoint_arraycopy =
2015 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry,
2016 "arrayof_jshort_disjoint_arraycopy");
2017 StubRoutines::_arrayof_jshort_arraycopy =
2018 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry,
2019 NULL, "arrayof_jshort_arraycopy");
2020 StubRoutines::_jshort_disjoint_arraycopy =
2021 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry,
2022 "jshort_disjoint_arraycopy");
2023 StubRoutines::_jshort_arraycopy =
2024 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry,
2025 &entry_jshort_arraycopy, "jshort_arraycopy");
2027 // Next arrays are always aligned on 4 bytes at least.
2028 StubRoutines::_jint_disjoint_arraycopy =
2029 generate_disjoint_copy(T_INT, true, Address::times_4, &entry,
2030 "jint_disjoint_arraycopy");
2031 StubRoutines::_jint_arraycopy =
2032 generate_conjoint_copy(T_INT, true, Address::times_4, entry,
2033 &entry_jint_arraycopy, "jint_arraycopy");
2035 StubRoutines::_oop_disjoint_arraycopy =
2036 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry,
2037 "oop_disjoint_arraycopy");
2038 StubRoutines::_oop_arraycopy =
2039 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry,
2040 &entry_oop_arraycopy, "oop_arraycopy");
2042 StubRoutines::_jlong_disjoint_arraycopy =
2043 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy");
2044 StubRoutines::_jlong_arraycopy =
2045 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy,
2046 "jlong_arraycopy");
2048 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill");
2049 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill");
2050 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill");
2051 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill");
2052 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill");
2053 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill");
2055 StubRoutines::_arrayof_jint_disjoint_arraycopy =
2056 StubRoutines::_jint_disjoint_arraycopy;
2057 StubRoutines::_arrayof_oop_disjoint_arraycopy =
2058 StubRoutines::_oop_disjoint_arraycopy;
2059 StubRoutines::_arrayof_jlong_disjoint_arraycopy =
2060 StubRoutines::_jlong_disjoint_arraycopy;
2062 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy;
2063 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy;
2064 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy;
2066 StubRoutines::_checkcast_arraycopy =
2067 generate_checkcast_copy("checkcast_arraycopy",
2068 &entry_checkcast_arraycopy);
2070 StubRoutines::_unsafe_arraycopy =
2071 generate_unsafe_copy("unsafe_arraycopy",
2072 entry_jbyte_arraycopy,
2073 entry_jshort_arraycopy,
2074 entry_jint_arraycopy,
2075 entry_jlong_arraycopy);
2077 StubRoutines::_generic_arraycopy =
2078 generate_generic_copy("generic_arraycopy",
2079 entry_jbyte_arraycopy,
2080 entry_jshort_arraycopy,
2081 entry_jint_arraycopy,
2082 entry_oop_arraycopy,
2083 entry_jlong_arraycopy,
2084 entry_checkcast_arraycopy);
2085 }
2087 void generate_math_stubs() {
2088 {
2089 StubCodeMark mark(this, "StubRoutines", "log");
2090 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc();
2092 __ fld_d(Address(rsp, 4));
2093 __ flog();
2094 __ ret(0);
2095 }
2096 {
2097 StubCodeMark mark(this, "StubRoutines", "log10");
2098 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc();
2100 __ fld_d(Address(rsp, 4));
2101 __ flog10();
2102 __ ret(0);
2103 }
2104 {
2105 StubCodeMark mark(this, "StubRoutines", "sin");
2106 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc();
2108 __ fld_d(Address(rsp, 4));
2109 __ trigfunc('s');
2110 __ ret(0);
2111 }
2112 {
2113 StubCodeMark mark(this, "StubRoutines", "cos");
2114 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc();
2116 __ fld_d(Address(rsp, 4));
2117 __ trigfunc('c');
2118 __ ret(0);
2119 }
2120 {
2121 StubCodeMark mark(this, "StubRoutines", "tan");
2122 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc();
2124 __ fld_d(Address(rsp, 4));
2125 __ trigfunc('t');
2126 __ ret(0);
2127 }
2129 // The intrinsic version of these seem to return the same value as
2130 // the strict version.
2131 StubRoutines::_intrinsic_exp = SharedRuntime::dexp;
2132 StubRoutines::_intrinsic_pow = SharedRuntime::dpow;
2133 }
2135 public:
2136 // Information about frame layout at time of blocking runtime call.
2137 // Note that we only have to preserve callee-saved registers since
2138 // the compilers are responsible for supplying a continuation point
2139 // if they expect all registers to be preserved.
2140 enum layout {
2141 thread_off, // last_java_sp
2142 rbp_off, // callee saved register
2143 ret_pc,
2144 framesize
2145 };
2147 private:
2149 #undef __
2150 #define __ masm->
2152 //------------------------------------------------------------------------------------------------------------------------
2153 // Continuation point for throwing of implicit exceptions that are not handled in
2154 // the current activation. Fabricates an exception oop and initiates normal
2155 // exception dispatching in this frame.
2156 //
2157 // Previously the compiler (c2) allowed for callee save registers on Java calls.
2158 // This is no longer true after adapter frames were removed but could possibly
2159 // be brought back in the future if the interpreter code was reworked and it
2160 // was deemed worthwhile. The comment below was left to describe what must
2161 // happen here if callee saves were resurrected. As it stands now this stub
2162 // could actually be a vanilla BufferBlob and have now oopMap at all.
2163 // Since it doesn't make much difference we've chosen to leave it the
2164 // way it was in the callee save days and keep the comment.
2166 // If we need to preserve callee-saved values we need a callee-saved oop map and
2167 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs.
2168 // If the compiler needs all registers to be preserved between the fault
2169 // point and the exception handler then it must assume responsibility for that in
2170 // AbstractCompiler::continuation_for_implicit_null_exception or
2171 // continuation_for_implicit_division_by_zero_exception. All other implicit
2172 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are
2173 // either at call sites or otherwise assume that stack unwinding will be initiated,
2174 // so caller saved registers were assumed volatile in the compiler.
2175 address generate_throw_exception(const char* name, address runtime_entry,
2176 bool restore_saved_exception_pc) {
2178 int insts_size = 256;
2179 int locs_size = 32;
2181 CodeBuffer code(name, insts_size, locs_size);
2182 OopMapSet* oop_maps = new OopMapSet();
2183 MacroAssembler* masm = new MacroAssembler(&code);
2185 address start = __ pc();
2187 // This is an inlined and slightly modified version of call_VM
2188 // which has the ability to fetch the return PC out of
2189 // thread-local storage and also sets up last_Java_sp slightly
2190 // differently than the real call_VM
2191 Register java_thread = rbx;
2192 __ get_thread(java_thread);
2193 if (restore_saved_exception_pc) {
2194 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset())));
2195 __ push(rax);
2196 }
2198 __ enter(); // required for proper stackwalking of RuntimeStub frame
2200 // pc and rbp, already pushed
2201 __ subptr(rsp, (framesize-2) * wordSize); // prolog
2203 // Frame is now completed as far as size and linkage.
2205 int frame_complete = __ pc() - start;
2207 // push java thread (becomes first argument of C function)
2208 __ movptr(Address(rsp, thread_off * wordSize), java_thread);
2210 // Set up last_Java_sp and last_Java_fp
2211 __ set_last_Java_frame(java_thread, rsp, rbp, NULL);
2213 // Call runtime
2214 BLOCK_COMMENT("call runtime_entry");
2215 __ call(RuntimeAddress(runtime_entry));
2216 // Generate oop map
2217 OopMap* map = new OopMap(framesize, 0);
2218 oop_maps->add_gc_map(__ pc() - start, map);
2220 // restore the thread (cannot use the pushed argument since arguments
2221 // may be overwritten by C code generated by an optimizing compiler);
2222 // however can use the register value directly if it is callee saved.
2223 __ get_thread(java_thread);
2225 __ reset_last_Java_frame(java_thread, true, false);
2227 __ leave(); // required for proper stackwalking of RuntimeStub frame
2229 // check for pending exceptions
2230 #ifdef ASSERT
2231 Label L;
2232 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
2233 __ jcc(Assembler::notEqual, L);
2234 __ should_not_reach_here();
2235 __ bind(L);
2236 #endif /* ASSERT */
2237 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
2240 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false);
2241 return stub->entry_point();
2242 }
2245 void create_control_words() {
2246 // Round to nearest, 53-bit mode, exceptions masked
2247 StubRoutines::_fpu_cntrl_wrd_std = 0x027F;
2248 // Round to zero, 53-bit mode, exception mased
2249 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F;
2250 // Round to nearest, 24-bit mode, exceptions masked
2251 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F;
2252 // Round to nearest, 64-bit mode, exceptions masked
2253 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F;
2254 // Round to nearest, 64-bit mode, exceptions masked
2255 StubRoutines::_mxcsr_std = 0x1F80;
2256 // Note: the following two constants are 80-bit values
2257 // layout is critical for correct loading by FPU.
2258 // Bias for strict fp multiply/divide
2259 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000
2260 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000;
2261 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff;
2262 // Un-Bias for strict fp multiply/divide
2263 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000
2264 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000;
2265 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff;
2266 }
2268 //---------------------------------------------------------------------------
2269 // Initialization
2271 void generate_initial() {
2272 // Generates all stubs and initializes the entry points
2274 //------------------------------------------------------------------------------------------------------------------------
2275 // entry points that exist in all platforms
2276 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than
2277 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp.
2278 StubRoutines::_forward_exception_entry = generate_forward_exception();
2280 StubRoutines::_call_stub_entry =
2281 generate_call_stub(StubRoutines::_call_stub_return_address);
2282 // is referenced by megamorphic call
2283 StubRoutines::_catch_exception_entry = generate_catch_exception();
2285 // These are currently used by Solaris/Intel
2286 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg();
2288 StubRoutines::_handler_for_unsafe_access_entry =
2289 generate_handler_for_unsafe_access();
2291 // platform dependent
2292 create_control_words();
2294 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr();
2295 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd();
2296 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT,
2297 CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
2298 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG,
2299 CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
2300 }
2303 void generate_all() {
2304 // Generates all stubs and initializes the entry points
2306 // These entry points require SharedInfo::stack0 to be set up in non-core builds
2307 // and need to be relocatable, so they each fabricate a RuntimeStub internally.
2308 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false);
2309 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false);
2310 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true);
2311 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true);
2312 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);
2313 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
2315 //------------------------------------------------------------------------------------------------------------------------
2316 // entry points that are platform specific
2318 // support for verify_oop (must happen after universe_init)
2319 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop();
2321 // arraycopy stubs used by compilers
2322 generate_arraycopy_stubs();
2324 generate_math_stubs();
2325 }
2328 public:
2329 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) {
2330 if (all) {
2331 generate_all();
2332 } else {
2333 generate_initial();
2334 }
2335 }
2336 }; // end class declaration
2339 void StubGenerator_generate(CodeBuffer* code, bool all) {
2340 StubGenerator g(code, all);
2341 }