1 /* 2 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "incls/_precompiled.incl" 26 #include "incls/_stubGenerator_x86_32.cpp.incl" 27 28 // Declaration and definition of StubGenerator (no .hpp file). 29 // For a more detailed description of the stub routine structure 30 // see the comment in stubRoutines.hpp 31 32 #define __ _masm-> 33 #define a__ ((Assembler*)_masm)-> 34 35 #ifdef PRODUCT 36 #define BLOCK_COMMENT(str) /* nothing */ 37 #else 38 #define BLOCK_COMMENT(str) __ block_comment(str) 39 #endif 40 41 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 42 43 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions 44 const int FPU_CNTRL_WRD_MASK = 0xFFFF; 45 46 // ------------------------------------------------------------------------------------------------------------------------- 47 // Stub Code definitions 48 49 static address handle_unsafe_access() { 50 JavaThread* thread = JavaThread::current(); 51 address pc = thread->saved_exception_pc(); 52 // pc is the instruction which we must emulate 53 // doing a no-op is fine: return garbage from the load 54 // therefore, compute npc 55 address npc = Assembler::locate_next_instruction(pc); 56 57 // request an async exception 58 thread->set_pending_unsafe_access_error(); 59 60 // return address of next instruction to execute 61 return npc; 62 } 63 64 class StubGenerator: public StubCodeGenerator { 65 private: 66 67 #ifdef PRODUCT 68 #define inc_counter_np(counter) (0) 69 #else 70 void inc_counter_np_(int& counter) { 71 __ incrementl(ExternalAddress((address)&counter)); 72 } 73 #define inc_counter_np(counter) \ 74 BLOCK_COMMENT("inc_counter " #counter); \ 75 inc_counter_np_(counter); 76 #endif //PRODUCT 77 78 void inc_copy_counter_np(BasicType t) { 79 #ifndef PRODUCT 80 switch (t) { 81 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return; 82 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return; 83 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return; 84 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return; 85 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return; 86 } 87 ShouldNotReachHere(); 88 #endif //PRODUCT 89 } 90 91 //------------------------------------------------------------------------------------------------------------------------ 92 // Call stubs are used to call Java from C 93 // 94 // [ return_from_Java ] <--- rsp 95 // [ argument word n ] 96 // ... 97 // -N [ argument word 1 ] 98 // -7 [ Possible padding for stack alignment ] 99 // -6 [ Possible padding for stack alignment ] 100 // -5 [ Possible padding for stack alignment ] 101 // -4 [ mxcsr save ] <--- rsp_after_call 102 // -3 [ saved rbx, ] 103 // -2 [ saved rsi ] 104 // -1 [ saved rdi ] 105 // 0 [ saved rbp, ] <--- rbp, 106 // 1 [ return address ] 107 // 2 [ ptr. to call wrapper ] 108 // 3 [ result ] 109 // 4 [ result_type ] 110 // 5 [ method ] 111 // 6 [ entry_point ] 112 // 7 [ parameters ] 113 // 8 [ parameter_size ] 114 // 9 [ thread ] 115 116 117 address generate_call_stub(address& return_address) { 118 StubCodeMark mark(this, "StubRoutines", "call_stub"); 119 address start = __ pc(); 120 121 // stub code parameters / addresses 122 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code"); 123 bool sse_save = false; 124 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()! 125 const int locals_count_in_bytes (4*wordSize); 126 const Address mxcsr_save (rbp, -4 * wordSize); 127 const Address saved_rbx (rbp, -3 * wordSize); 128 const Address saved_rsi (rbp, -2 * wordSize); 129 const Address saved_rdi (rbp, -1 * wordSize); 130 const Address result (rbp, 3 * wordSize); 131 const Address result_type (rbp, 4 * wordSize); 132 const Address method (rbp, 5 * wordSize); 133 const Address entry_point (rbp, 6 * wordSize); 134 const Address parameters (rbp, 7 * wordSize); 135 const Address parameter_size(rbp, 8 * wordSize); 136 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()! 137 sse_save = UseSSE > 0; 138 139 // stub code 140 __ enter(); 141 __ movptr(rcx, parameter_size); // parameter counter 142 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes 143 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves 144 __ subptr(rsp, rcx); 145 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack 146 147 // save rdi, rsi, & rbx, according to C calling conventions 148 __ movptr(saved_rdi, rdi); 149 __ movptr(saved_rsi, rsi); 150 __ movptr(saved_rbx, rbx); 151 // save and initialize %mxcsr 152 if (sse_save) { 153 Label skip_ldmx; 154 __ stmxcsr(mxcsr_save); 155 __ movl(rax, mxcsr_save); 156 __ andl(rax, MXCSR_MASK); // Only check control and mask bits 157 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 158 __ cmp32(rax, mxcsr_std); 159 __ jcc(Assembler::equal, skip_ldmx); 160 __ ldmxcsr(mxcsr_std); 161 __ bind(skip_ldmx); 162 } 163 164 // make sure the control word is correct. 165 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 166 167 #ifdef ASSERT 168 // make sure we have no pending exceptions 169 { Label L; 170 __ movptr(rcx, thread); 171 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 172 __ jcc(Assembler::equal, L); 173 __ stop("StubRoutines::call_stub: entered with pending exception"); 174 __ bind(L); 175 } 176 #endif 177 178 // pass parameters if any 179 BLOCK_COMMENT("pass parameters if any"); 180 Label parameters_done; 181 __ movl(rcx, parameter_size); // parameter counter 182 __ testl(rcx, rcx); 183 __ jcc(Assembler::zero, parameters_done); 184 185 // parameter passing loop 186 187 Label loop; 188 // Copy Java parameters in reverse order (receiver last) 189 // Note that the argument order is inverted in the process 190 // source is rdx[rcx: N-1..0] 191 // dest is rsp[rbx: 0..N-1] 192 193 __ movptr(rdx, parameters); // parameter pointer 194 __ xorptr(rbx, rbx); 195 196 __ BIND(loop); 197 198 // get parameter 199 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize)); 200 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), 201 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter 202 __ increment(rbx); 203 __ decrement(rcx); 204 __ jcc(Assembler::notZero, loop); 205 206 // call Java function 207 __ BIND(parameters_done); 208 __ movptr(rbx, method); // get methodOop 209 __ movptr(rax, entry_point); // get entry_point 210 __ mov(rsi, rsp); // set sender sp 211 BLOCK_COMMENT("call Java function"); 212 __ call(rax); 213 214 BLOCK_COMMENT("call_stub_return_address:"); 215 return_address = __ pc(); 216 217 Label common_return; 218 219 __ BIND(common_return); 220 221 // store result depending on type 222 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) 223 __ movptr(rdi, result); 224 Label is_long, is_float, is_double, exit; 225 __ movl(rsi, result_type); 226 __ cmpl(rsi, T_LONG); 227 __ jcc(Assembler::equal, is_long); 228 __ cmpl(rsi, T_FLOAT); 229 __ jcc(Assembler::equal, is_float); 230 __ cmpl(rsi, T_DOUBLE); 231 __ jcc(Assembler::equal, is_double); 232 233 // handle T_INT case 234 __ movl(Address(rdi, 0), rax); 235 __ BIND(exit); 236 237 // check that FPU stack is empty 238 __ verify_FPU(0, "generate_call_stub"); 239 240 // pop parameters 241 __ lea(rsp, rsp_after_call); 242 243 // restore %mxcsr 244 if (sse_save) { 245 __ ldmxcsr(mxcsr_save); 246 } 247 248 // restore rdi, rsi and rbx, 249 __ movptr(rbx, saved_rbx); 250 __ movptr(rsi, saved_rsi); 251 __ movptr(rdi, saved_rdi); 252 __ addptr(rsp, 4*wordSize); 253 254 // return 255 __ pop(rbp); 256 __ ret(0); 257 258 // handle return types different from T_INT 259 __ BIND(is_long); 260 __ movl(Address(rdi, 0 * wordSize), rax); 261 __ movl(Address(rdi, 1 * wordSize), rdx); 262 __ jmp(exit); 263 264 __ BIND(is_float); 265 // interpreter uses xmm0 for return values 266 if (UseSSE >= 1) { 267 __ movflt(Address(rdi, 0), xmm0); 268 } else { 269 __ fstp_s(Address(rdi, 0)); 270 } 271 __ jmp(exit); 272 273 __ BIND(is_double); 274 // interpreter uses xmm0 for return values 275 if (UseSSE >= 2) { 276 __ movdbl(Address(rdi, 0), xmm0); 277 } else { 278 __ fstp_d(Address(rdi, 0)); 279 } 280 __ jmp(exit); 281 282 // If we call compiled code directly from the call stub we will 283 // need to adjust the return back to the call stub to a specialized 284 // piece of code that can handle compiled results and cleaning the fpu 285 // stack. compiled code will be set to return here instead of the 286 // return above that handles interpreter returns. 287 288 BLOCK_COMMENT("call_stub_compiled_return:"); 289 StubRoutines::x86::set_call_stub_compiled_return( __ pc()); 290 291 #ifdef COMPILER2 292 if (UseSSE >= 2) { 293 __ verify_FPU(0, "call_stub_compiled_return"); 294 } else { 295 for (int i = 1; i < 8; i++) { 296 __ ffree(i); 297 } 298 299 // UseSSE <= 1 so double result should be left on TOS 300 __ movl(rsi, result_type); 301 __ cmpl(rsi, T_DOUBLE); 302 __ jcc(Assembler::equal, common_return); 303 if (UseSSE == 0) { 304 // UseSSE == 0 so float result should be left on TOS 305 __ cmpl(rsi, T_FLOAT); 306 __ jcc(Assembler::equal, common_return); 307 } 308 __ ffree(0); 309 } 310 #endif /* COMPILER2 */ 311 __ jmp(common_return); 312 313 return start; 314 } 315 316 317 //------------------------------------------------------------------------------------------------------------------------ 318 // Return point for a Java call if there's an exception thrown in Java code. 319 // The exception is caught and transformed into a pending exception stored in 320 // JavaThread that can be tested from within the VM. 321 // 322 // Note: Usually the parameters are removed by the callee. In case of an exception 323 // crossing an activation frame boundary, that is not the case if the callee 324 // is compiled code => need to setup the rsp. 325 // 326 // rax,: exception oop 327 328 address generate_catch_exception() { 329 StubCodeMark mark(this, "StubRoutines", "catch_exception"); 330 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! 331 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! 332 address start = __ pc(); 333 334 // get thread directly 335 __ movptr(rcx, thread); 336 #ifdef ASSERT 337 // verify that threads correspond 338 { Label L; 339 __ get_thread(rbx); 340 __ cmpptr(rbx, rcx); 341 __ jcc(Assembler::equal, L); 342 __ stop("StubRoutines::catch_exception: threads must correspond"); 343 __ bind(L); 344 } 345 #endif 346 // set pending exception 347 __ verify_oop(rax); 348 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); 349 __ lea(Address(rcx, Thread::exception_file_offset ()), 350 ExternalAddress((address)__FILE__)); 351 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); 352 // complete return to VM 353 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); 354 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); 355 356 return start; 357 } 358 359 360 //------------------------------------------------------------------------------------------------------------------------ 361 // Continuation point for runtime calls returning with a pending exception. 362 // The pending exception check happened in the runtime or native call stub. 363 // The pending exception in Thread is converted into a Java-level exception. 364 // 365 // Contract with Java-level exception handlers: 366 // rax: exception 367 // rdx: throwing pc 368 // 369 // NOTE: At entry of this stub, exception-pc must be on stack !! 370 371 address generate_forward_exception() { 372 StubCodeMark mark(this, "StubRoutines", "forward exception"); 373 address start = __ pc(); 374 const Register thread = rcx; 375 376 // other registers used in this stub 377 const Register exception_oop = rax; 378 const Register handler_addr = rbx; 379 const Register exception_pc = rdx; 380 381 // Upon entry, the sp points to the return address returning into Java 382 // (interpreted or compiled) code; i.e., the return address becomes the 383 // throwing pc. 384 // 385 // Arguments pushed before the runtime call are still on the stack but 386 // the exception handler will reset the stack pointer -> ignore them. 387 // A potential result in registers can be ignored as well. 388 389 #ifdef ASSERT 390 // make sure this code is only executed if there is a pending exception 391 { Label L; 392 __ get_thread(thread); 393 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 394 __ jcc(Assembler::notEqual, L); 395 __ stop("StubRoutines::forward exception: no pending exception (1)"); 396 __ bind(L); 397 } 398 #endif 399 400 // compute exception handler into rbx, 401 __ get_thread(thread); 402 __ movptr(exception_pc, Address(rsp, 0)); 403 BLOCK_COMMENT("call exception_handler_for_return_address"); 404 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); 405 __ mov(handler_addr, rax); 406 407 // setup rax & rdx, remove return address & clear pending exception 408 __ get_thread(thread); 409 __ pop(exception_pc); 410 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); 411 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 412 413 #ifdef ASSERT 414 // make sure exception is set 415 { Label L; 416 __ testptr(exception_oop, exception_oop); 417 __ jcc(Assembler::notEqual, L); 418 __ stop("StubRoutines::forward exception: no pending exception (2)"); 419 __ bind(L); 420 } 421 #endif 422 423 // Verify that there is really a valid exception in RAX. 424 __ verify_oop(exception_oop); 425 426 // Restore SP from BP if the exception PC is a MethodHandle call site. 427 __ cmpl(Address(thread, JavaThread::is_method_handle_return_offset()), 0); 428 __ cmovptr(Assembler::notEqual, rsp, rbp); 429 430 // continue at exception handler (return address removed) 431 // rax: exception 432 // rbx: exception handler 433 // rdx: throwing pc 434 __ jmp(handler_addr); 435 436 return start; 437 } 438 439 440 //---------------------------------------------------------------------------------------------------- 441 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) 442 // 443 // xchg exists as far back as 8086, lock needed for MP only 444 // Stack layout immediately after call: 445 // 446 // 0 [ret addr ] <--- rsp 447 // 1 [ ex ] 448 // 2 [ dest ] 449 // 450 // Result: *dest <- ex, return (old *dest) 451 // 452 // Note: win32 does not currently use this code 453 454 address generate_atomic_xchg() { 455 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); 456 address start = __ pc(); 457 458 __ push(rdx); 459 Address exchange(rsp, 2 * wordSize); 460 Address dest_addr(rsp, 3 * wordSize); 461 __ movl(rax, exchange); 462 __ movptr(rdx, dest_addr); 463 __ xchgl(rax, Address(rdx, 0)); 464 __ pop(rdx); 465 __ ret(0); 466 467 return start; 468 } 469 470 //---------------------------------------------------------------------------------------------------- 471 // Support for void verify_mxcsr() 472 // 473 // This routine is used with -Xcheck:jni to verify that native 474 // JNI code does not return to Java code without restoring the 475 // MXCSR register to our expected state. 476 477 478 address generate_verify_mxcsr() { 479 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); 480 address start = __ pc(); 481 482 const Address mxcsr_save(rsp, 0); 483 484 if (CheckJNICalls && UseSSE > 0 ) { 485 Label ok_ret; 486 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 487 __ push(rax); 488 __ subptr(rsp, wordSize); // allocate a temp location 489 __ stmxcsr(mxcsr_save); 490 __ movl(rax, mxcsr_save); 491 __ andl(rax, MXCSR_MASK); 492 __ cmp32(rax, mxcsr_std); 493 __ jcc(Assembler::equal, ok_ret); 494 495 __ warn("MXCSR changed by native JNI code."); 496 497 __ ldmxcsr(mxcsr_std); 498 499 __ bind(ok_ret); 500 __ addptr(rsp, wordSize); 501 __ pop(rax); 502 } 503 504 __ ret(0); 505 506 return start; 507 } 508 509 510 //--------------------------------------------------------------------------- 511 // Support for void verify_fpu_cntrl_wrd() 512 // 513 // This routine is used with -Xcheck:jni to verify that native 514 // JNI code does not return to Java code without restoring the 515 // FP control word to our expected state. 516 517 address generate_verify_fpu_cntrl_wrd() { 518 StubCodeMark mark(this, "StubRoutines", "verify_spcw"); 519 address start = __ pc(); 520 521 const Address fpu_cntrl_wrd_save(rsp, 0); 522 523 if (CheckJNICalls) { 524 Label ok_ret; 525 __ push(rax); 526 __ subptr(rsp, wordSize); // allocate a temp location 527 __ fnstcw(fpu_cntrl_wrd_save); 528 __ movl(rax, fpu_cntrl_wrd_save); 529 __ andl(rax, FPU_CNTRL_WRD_MASK); 530 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); 531 __ cmp32(rax, fpu_std); 532 __ jcc(Assembler::equal, ok_ret); 533 534 __ warn("Floating point control word changed by native JNI code."); 535 536 __ fldcw(fpu_std); 537 538 __ bind(ok_ret); 539 __ addptr(rsp, wordSize); 540 __ pop(rax); 541 } 542 543 __ ret(0); 544 545 return start; 546 } 547 548 //--------------------------------------------------------------------------- 549 // Wrapper for slow-case handling of double-to-integer conversion 550 // d2i or f2i fast case failed either because it is nan or because 551 // of under/overflow. 552 // Input: FPU TOS: float value 553 // Output: rax, (rdx): integer (long) result 554 555 address generate_d2i_wrapper(BasicType t, address fcn) { 556 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); 557 address start = __ pc(); 558 559 // Capture info about frame layout 560 enum layout { FPUState_off = 0, 561 rbp_off = FPUStateSizeInWords, 562 rdi_off, 563 rsi_off, 564 rcx_off, 565 rbx_off, 566 saved_argument_off, 567 saved_argument_off2, // 2nd half of double 568 framesize 569 }; 570 571 assert(FPUStateSizeInWords == 27, "update stack layout"); 572 573 // Save outgoing argument to stack across push_FPU_state() 574 __ subptr(rsp, wordSize * 2); 575 __ fstp_d(Address(rsp, 0)); 576 577 // Save CPU & FPU state 578 __ push(rbx); 579 __ push(rcx); 580 __ push(rsi); 581 __ push(rdi); 582 __ push(rbp); 583 __ push_FPU_state(); 584 585 // push_FPU_state() resets the FP top of stack 586 // Load original double into FP top of stack 587 __ fld_d(Address(rsp, saved_argument_off * wordSize)); 588 // Store double into stack as outgoing argument 589 __ subptr(rsp, wordSize*2); 590 __ fst_d(Address(rsp, 0)); 591 592 // Prepare FPU for doing math in C-land 593 __ empty_FPU_stack(); 594 // Call the C code to massage the double. Result in EAX 595 if (t == T_INT) 596 { BLOCK_COMMENT("SharedRuntime::d2i"); } 597 else if (t == T_LONG) 598 { BLOCK_COMMENT("SharedRuntime::d2l"); } 599 __ call_VM_leaf( fcn, 2 ); 600 601 // Restore CPU & FPU state 602 __ pop_FPU_state(); 603 __ pop(rbp); 604 __ pop(rdi); 605 __ pop(rsi); 606 __ pop(rcx); 607 __ pop(rbx); 608 __ addptr(rsp, wordSize * 2); 609 610 __ ret(0); 611 612 return start; 613 } 614 615 616 //--------------------------------------------------------------------------- 617 // The following routine generates a subroutine to throw an asynchronous 618 // UnknownError when an unsafe access gets a fault that could not be 619 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) 620 address generate_handler_for_unsafe_access() { 621 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); 622 address start = __ pc(); 623 624 __ push(0); // hole for return address-to-be 625 __ pusha(); // push registers 626 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); 627 BLOCK_COMMENT("call handle_unsafe_access"); 628 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); 629 __ movptr(next_pc, rax); // stuff next address 630 __ popa(); 631 __ ret(0); // jump to next address 632 633 return start; 634 } 635 636 637 //---------------------------------------------------------------------------------------------------- 638 // Non-destructive plausibility checks for oops 639 640 address generate_verify_oop() { 641 StubCodeMark mark(this, "StubRoutines", "verify_oop"); 642 address start = __ pc(); 643 644 // Incoming arguments on stack after saving rax,: 645 // 646 // [tos ]: saved rdx 647 // [tos + 1]: saved EFLAGS 648 // [tos + 2]: return address 649 // [tos + 3]: char* error message 650 // [tos + 4]: oop object to verify 651 // [tos + 5]: saved rax, - saved by caller 652 653 Label exit, error; 654 __ pushf(); 655 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); 656 __ push(rdx); // save rdx 657 // make sure object is 'reasonable' 658 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object 659 __ testptr(rax, rax); 660 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok 661 662 // Check if the oop is in the right area of memory 663 const int oop_mask = Universe::verify_oop_mask(); 664 const int oop_bits = Universe::verify_oop_bits(); 665 __ mov(rdx, rax); 666 __ andptr(rdx, oop_mask); 667 __ cmpptr(rdx, oop_bits); 668 __ jcc(Assembler::notZero, error); 669 670 // make sure klass is 'reasonable' 671 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass 672 __ testptr(rax, rax); 673 __ jcc(Assembler::zero, error); // if klass is NULL it is broken 674 675 // Check if the klass is in the right area of memory 676 const int klass_mask = Universe::verify_klass_mask(); 677 const int klass_bits = Universe::verify_klass_bits(); 678 __ mov(rdx, rax); 679 __ andptr(rdx, klass_mask); 680 __ cmpptr(rdx, klass_bits); 681 __ jcc(Assembler::notZero, error); 682 683 // make sure klass' klass is 'reasonable' 684 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass 685 __ testptr(rax, rax); 686 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken 687 688 __ mov(rdx, rax); 689 __ andptr(rdx, klass_mask); 690 __ cmpptr(rdx, klass_bits); 691 __ jcc(Assembler::notZero, error); // if klass not in right area 692 // of memory it is broken too. 693 694 // return if everything seems ok 695 __ bind(exit); 696 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 697 __ pop(rdx); // restore rdx 698 __ popf(); // restore EFLAGS 699 __ ret(2 * wordSize); // pop arguments 700 701 // handle errors 702 __ bind(error); 703 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 704 __ pop(rdx); // get saved rdx back 705 __ popf(); // get saved EFLAGS off stack -- will be ignored 706 __ pusha(); // push registers (eip = return address & msg are already pushed) 707 BLOCK_COMMENT("call MacroAssembler::debug"); 708 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); 709 __ popa(); 710 __ ret(2 * wordSize); // pop arguments 711 return start; 712 } 713 714 // 715 // Generate pre-barrier for array stores 716 // 717 // Input: 718 // start - starting address 719 // count - element count 720 void gen_write_ref_array_pre_barrier(Register start, Register count) { 721 assert_different_registers(start, count); 722 BarrierSet* bs = Universe::heap()->barrier_set(); 723 switch (bs->kind()) { 724 case BarrierSet::G1SATBCT: 725 case BarrierSet::G1SATBCTLogging: 726 { 727 __ pusha(); // push registers 728 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 729 start, count); 730 __ popa(); 731 } 732 break; 733 case BarrierSet::CardTableModRef: 734 case BarrierSet::CardTableExtension: 735 case BarrierSet::ModRef: 736 break; 737 default : 738 ShouldNotReachHere(); 739 740 } 741 } 742 743 744 // 745 // Generate a post-barrier for an array store 746 // 747 // start - starting address 748 // count - element count 749 // 750 // The two input registers are overwritten. 751 // 752 void gen_write_ref_array_post_barrier(Register start, Register count) { 753 BarrierSet* bs = Universe::heap()->barrier_set(); 754 assert_different_registers(start, count); 755 switch (bs->kind()) { 756 case BarrierSet::G1SATBCT: 757 case BarrierSet::G1SATBCTLogging: 758 { 759 __ pusha(); // push registers 760 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 761 start, count); 762 __ popa(); 763 } 764 break; 765 766 case BarrierSet::CardTableModRef: 767 case BarrierSet::CardTableExtension: 768 { 769 CardTableModRefBS* ct = (CardTableModRefBS*)bs; 770 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 771 772 Label L_loop; 773 const Register end = count; // elements count; end == start+count-1 774 assert_different_registers(start, end); 775 776 __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); 777 __ shrptr(start, CardTableModRefBS::card_shift); 778 __ shrptr(end, CardTableModRefBS::card_shift); 779 __ subptr(end, start); // end --> count 780 __ BIND(L_loop); 781 intptr_t disp = (intptr_t) ct->byte_map_base; 782 Address cardtable(start, count, Address::times_1, disp); 783 __ movb(cardtable, 0); 784 __ decrement(count); 785 __ jcc(Assembler::greaterEqual, L_loop); 786 } 787 break; 788 case BarrierSet::ModRef: 789 break; 790 default : 791 ShouldNotReachHere(); 792 793 } 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 xmm_copy_forward(Register from, Register to_from, Register qword_count) { 805 assert( UseSSE >= 2, "supported cpu only" ); 806 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 807 // Copy 64-byte chunks 808 __ jmpb(L_copy_64_bytes); 809 __ align(OptoLoopAlignment); 810 __ BIND(L_copy_64_bytes_loop); 811 812 if(UseUnalignedLoadStores) { 813 __ movdqu(xmm0, Address(from, 0)); 814 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0); 815 __ movdqu(xmm1, Address(from, 16)); 816 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1); 817 __ movdqu(xmm2, Address(from, 32)); 818 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2); 819 __ movdqu(xmm3, Address(from, 48)); 820 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3); 821 822 } else { 823 __ movq(xmm0, Address(from, 0)); 824 __ movq(Address(from, to_from, Address::times_1, 0), xmm0); 825 __ movq(xmm1, Address(from, 8)); 826 __ movq(Address(from, to_from, Address::times_1, 8), xmm1); 827 __ movq(xmm2, Address(from, 16)); 828 __ movq(Address(from, to_from, Address::times_1, 16), xmm2); 829 __ movq(xmm3, Address(from, 24)); 830 __ movq(Address(from, to_from, Address::times_1, 24), xmm3); 831 __ movq(xmm4, Address(from, 32)); 832 __ movq(Address(from, to_from, Address::times_1, 32), xmm4); 833 __ movq(xmm5, Address(from, 40)); 834 __ movq(Address(from, to_from, Address::times_1, 40), xmm5); 835 __ movq(xmm6, Address(from, 48)); 836 __ movq(Address(from, to_from, Address::times_1, 48), xmm6); 837 __ movq(xmm7, Address(from, 56)); 838 __ movq(Address(from, to_from, Address::times_1, 56), xmm7); 839 } 840 841 __ addl(from, 64); 842 __ BIND(L_copy_64_bytes); 843 __ subl(qword_count, 8); 844 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 845 __ addl(qword_count, 8); 846 __ jccb(Assembler::zero, L_exit); 847 // 848 // length is too short, just copy qwords 849 // 850 __ BIND(L_copy_8_bytes); 851 __ movq(xmm0, Address(from, 0)); 852 __ movq(Address(from, to_from, Address::times_1), xmm0); 853 __ addl(from, 8); 854 __ decrement(qword_count); 855 __ jcc(Assembler::greater, L_copy_8_bytes); 856 __ BIND(L_exit); 857 } 858 859 // Copy 64 bytes chunks 860 // 861 // Inputs: 862 // from - source array address 863 // to_from - destination array address - from 864 // qword_count - 8-bytes element count, negative 865 // 866 void mmx_copy_forward(Register from, Register to_from, Register qword_count) { 867 assert( VM_Version::supports_mmx(), "supported cpu only" ); 868 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 869 // Copy 64-byte chunks 870 __ jmpb(L_copy_64_bytes); 871 __ align(OptoLoopAlignment); 872 __ BIND(L_copy_64_bytes_loop); 873 __ movq(mmx0, Address(from, 0)); 874 __ movq(mmx1, Address(from, 8)); 875 __ movq(mmx2, Address(from, 16)); 876 __ movq(Address(from, to_from, Address::times_1, 0), mmx0); 877 __ movq(mmx3, Address(from, 24)); 878 __ movq(Address(from, to_from, Address::times_1, 8), mmx1); 879 __ movq(mmx4, Address(from, 32)); 880 __ movq(Address(from, to_from, Address::times_1, 16), mmx2); 881 __ movq(mmx5, Address(from, 40)); 882 __ movq(Address(from, to_from, Address::times_1, 24), mmx3); 883 __ movq(mmx6, Address(from, 48)); 884 __ movq(Address(from, to_from, Address::times_1, 32), mmx4); 885 __ movq(mmx7, Address(from, 56)); 886 __ movq(Address(from, to_from, Address::times_1, 40), mmx5); 887 __ movq(Address(from, to_from, Address::times_1, 48), mmx6); 888 __ movq(Address(from, to_from, Address::times_1, 56), mmx7); 889 __ addptr(from, 64); 890 __ BIND(L_copy_64_bytes); 891 __ subl(qword_count, 8); 892 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 893 __ addl(qword_count, 8); 894 __ jccb(Assembler::zero, L_exit); 895 // 896 // length is too short, just copy qwords 897 // 898 __ BIND(L_copy_8_bytes); 899 __ movq(mmx0, Address(from, 0)); 900 __ movq(Address(from, to_from, Address::times_1), mmx0); 901 __ addptr(from, 8); 902 __ decrement(qword_count); 903 __ jcc(Assembler::greater, L_copy_8_bytes); 904 __ BIND(L_exit); 905 __ emms(); 906 } 907 908 address generate_disjoint_copy(BasicType t, bool aligned, 909 Address::ScaleFactor sf, 910 address* entry, const char *name) { 911 __ align(CodeEntryAlignment); 912 StubCodeMark mark(this, "StubRoutines", name); 913 address start = __ pc(); 914 915 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 916 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; 917 918 int shift = Address::times_ptr - sf; 919 920 const Register from = rsi; // source array address 921 const Register to = rdi; // destination array address 922 const Register count = rcx; // elements count 923 const Register to_from = to; // (to - from) 924 const Register saved_to = rdx; // saved destination array address 925 926 __ enter(); // required for proper stackwalking of RuntimeStub frame 927 __ push(rsi); 928 __ push(rdi); 929 __ movptr(from , Address(rsp, 12+ 4)); 930 __ movptr(to , Address(rsp, 12+ 8)); 931 __ movl(count, Address(rsp, 12+ 12)); 932 if (t == T_OBJECT) { 933 __ testl(count, count); 934 __ jcc(Assembler::zero, L_0_count); 935 gen_write_ref_array_pre_barrier(to, count); 936 __ mov(saved_to, to); // save 'to' 937 } 938 939 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 940 BLOCK_COMMENT("Entry:"); 941 942 __ subptr(to, from); // to --> to_from 943 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 944 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 945 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { 946 // align source address at 4 bytes address boundary 947 if (t == T_BYTE) { 948 // One byte misalignment happens only for byte arrays 949 __ testl(from, 1); 950 __ jccb(Assembler::zero, L_skip_align1); 951 __ movb(rax, Address(from, 0)); 952 __ movb(Address(from, to_from, Address::times_1, 0), rax); 953 __ increment(from); 954 __ decrement(count); 955 __ BIND(L_skip_align1); 956 } 957 // Two bytes misalignment happens only for byte and short (char) arrays 958 __ testl(from, 2); 959 __ jccb(Assembler::zero, L_skip_align2); 960 __ movw(rax, Address(from, 0)); 961 __ movw(Address(from, to_from, Address::times_1, 0), rax); 962 __ addptr(from, 2); 963 __ subl(count, 1<<(shift-1)); 964 __ BIND(L_skip_align2); 965 } 966 if (!VM_Version::supports_mmx()) { 967 __ mov(rax, count); // save 'count' 968 __ shrl(count, shift); // bytes count 969 __ addptr(to_from, from);// restore 'to' 970 __ rep_mov(); 971 __ subptr(to_from, from);// restore 'to_from' 972 __ mov(count, rax); // restore 'count' 973 __ jmpb(L_copy_2_bytes); // all dwords were copied 974 } else { 975 if (!UseUnalignedLoadStores) { 976 // align to 8 bytes, we know we are 4 byte aligned to start 977 __ testptr(from, 4); 978 __ jccb(Assembler::zero, L_copy_64_bytes); 979 __ movl(rax, Address(from, 0)); 980 __ movl(Address(from, to_from, Address::times_1, 0), rax); 981 __ addptr(from, 4); 982 __ subl(count, 1<<shift); 983 } 984 __ BIND(L_copy_64_bytes); 985 __ mov(rax, count); 986 __ shrl(rax, shift+1); // 8 bytes chunk count 987 // 988 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop 989 // 990 if (UseXMMForArrayCopy) { 991 xmm_copy_forward(from, to_from, rax); 992 } else { 993 mmx_copy_forward(from, to_from, rax); 994 } 995 } 996 // copy tailing dword 997 __ BIND(L_copy_4_bytes); 998 __ testl(count, 1<<shift); 999 __ jccb(Assembler::zero, L_copy_2_bytes); 1000 __ movl(rax, Address(from, 0)); 1001 __ movl(Address(from, to_from, Address::times_1, 0), rax); 1002 if (t == T_BYTE || t == T_SHORT) { 1003 __ addptr(from, 4); 1004 __ BIND(L_copy_2_bytes); 1005 // copy tailing word 1006 __ testl(count, 1<<(shift-1)); 1007 __ jccb(Assembler::zero, L_copy_byte); 1008 __ movw(rax, Address(from, 0)); 1009 __ movw(Address(from, to_from, Address::times_1, 0), rax); 1010 if (t == T_BYTE) { 1011 __ addptr(from, 2); 1012 __ BIND(L_copy_byte); 1013 // copy tailing byte 1014 __ testl(count, 1); 1015 __ jccb(Assembler::zero, L_exit); 1016 __ movb(rax, Address(from, 0)); 1017 __ movb(Address(from, to_from, Address::times_1, 0), rax); 1018 __ BIND(L_exit); 1019 } else { 1020 __ BIND(L_copy_byte); 1021 } 1022 } else { 1023 __ BIND(L_copy_2_bytes); 1024 } 1025 1026 if (t == T_OBJECT) { 1027 __ movl(count, Address(rsp, 12+12)); // reread 'count' 1028 __ mov(to, saved_to); // restore 'to' 1029 gen_write_ref_array_post_barrier(to, count); 1030 __ BIND(L_0_count); 1031 } 1032 inc_copy_counter_np(t); 1033 __ pop(rdi); 1034 __ pop(rsi); 1035 __ leave(); // required for proper stackwalking of RuntimeStub frame 1036 __ xorptr(rax, rax); // return 0 1037 __ ret(0); 1038 return start; 1039 } 1040 1041 1042 address generate_conjoint_copy(BasicType t, bool aligned, 1043 Address::ScaleFactor sf, 1044 address nooverlap_target, 1045 address* entry, const char *name) { 1046 __ align(CodeEntryAlignment); 1047 StubCodeMark mark(this, "StubRoutines", name); 1048 address start = __ pc(); 1049 1050 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 1051 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; 1052 1053 int shift = Address::times_ptr - sf; 1054 1055 const Register src = rax; // source array address 1056 const Register dst = rdx; // destination array address 1057 const Register from = rsi; // source array address 1058 const Register to = rdi; // destination array address 1059 const Register count = rcx; // elements count 1060 const Register end = rax; // array end address 1061 1062 __ enter(); // required for proper stackwalking of RuntimeStub frame 1063 __ push(rsi); 1064 __ push(rdi); 1065 __ movptr(src , Address(rsp, 12+ 4)); // from 1066 __ movptr(dst , Address(rsp, 12+ 8)); // to 1067 __ movl2ptr(count, Address(rsp, 12+12)); // count 1068 if (t == T_OBJECT) { 1069 gen_write_ref_array_pre_barrier(dst, count); 1070 } 1071 1072 if (entry != NULL) { 1073 *entry = __ pc(); // Entry point from generic arraycopy stub. 1074 BLOCK_COMMENT("Entry:"); 1075 } 1076 1077 if (t == T_OBJECT) { 1078 __ testl(count, count); 1079 __ jcc(Assembler::zero, L_0_count); 1080 } 1081 __ mov(from, src); 1082 __ mov(to , dst); 1083 1084 // arrays overlap test 1085 RuntimeAddress nooverlap(nooverlap_target); 1086 __ cmpptr(dst, src); 1087 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size 1088 __ jump_cc(Assembler::belowEqual, nooverlap); 1089 __ cmpptr(dst, end); 1090 __ jump_cc(Assembler::aboveEqual, nooverlap); 1091 1092 // copy from high to low 1093 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1094 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 1095 if (t == T_BYTE || t == T_SHORT) { 1096 // Align the end of destination array at 4 bytes address boundary 1097 __ lea(end, Address(dst, count, sf, 0)); 1098 if (t == T_BYTE) { 1099 // One byte misalignment happens only for byte arrays 1100 __ testl(end, 1); 1101 __ jccb(Assembler::zero, L_skip_align1); 1102 __ decrement(count); 1103 __ movb(rdx, Address(from, count, sf, 0)); 1104 __ movb(Address(to, count, sf, 0), rdx); 1105 __ BIND(L_skip_align1); 1106 } 1107 // Two bytes misalignment happens only for byte and short (char) arrays 1108 __ testl(end, 2); 1109 __ jccb(Assembler::zero, L_skip_align2); 1110 __ subptr(count, 1<<(shift-1)); 1111 __ movw(rdx, Address(from, count, sf, 0)); 1112 __ movw(Address(to, count, sf, 0), rdx); 1113 __ BIND(L_skip_align2); 1114 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1115 __ jcc(Assembler::below, L_copy_4_bytes); 1116 } 1117 1118 if (!VM_Version::supports_mmx()) { 1119 __ std(); 1120 __ mov(rax, count); // Save 'count' 1121 __ mov(rdx, to); // Save 'to' 1122 __ lea(rsi, Address(from, count, sf, -4)); 1123 __ lea(rdi, Address(to , count, sf, -4)); 1124 __ shrptr(count, shift); // bytes count 1125 __ rep_mov(); 1126 __ cld(); 1127 __ mov(count, rax); // restore 'count' 1128 __ andl(count, (1<<shift)-1); // mask the number of rest elements 1129 __ movptr(from, Address(rsp, 12+4)); // reread 'from' 1130 __ mov(to, rdx); // restore 'to' 1131 __ jmpb(L_copy_2_bytes); // all dword were copied 1132 } else { 1133 // Align to 8 bytes the end of array. It is aligned to 4 bytes already. 1134 __ testptr(end, 4); 1135 __ jccb(Assembler::zero, L_copy_8_bytes); 1136 __ subl(count, 1<<shift); 1137 __ movl(rdx, Address(from, count, sf, 0)); 1138 __ movl(Address(to, count, sf, 0), rdx); 1139 __ jmpb(L_copy_8_bytes); 1140 1141 __ align(OptoLoopAlignment); 1142 // Move 8 bytes 1143 __ BIND(L_copy_8_bytes_loop); 1144 if (UseXMMForArrayCopy) { 1145 __ movq(xmm0, Address(from, count, sf, 0)); 1146 __ movq(Address(to, count, sf, 0), xmm0); 1147 } else { 1148 __ movq(mmx0, Address(from, count, sf, 0)); 1149 __ movq(Address(to, count, sf, 0), mmx0); 1150 } 1151 __ BIND(L_copy_8_bytes); 1152 __ subl(count, 2<<shift); 1153 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1154 __ addl(count, 2<<shift); 1155 if (!UseXMMForArrayCopy) { 1156 __ emms(); 1157 } 1158 } 1159 __ BIND(L_copy_4_bytes); 1160 // copy prefix qword 1161 __ testl(count, 1<<shift); 1162 __ jccb(Assembler::zero, L_copy_2_bytes); 1163 __ movl(rdx, Address(from, count, sf, -4)); 1164 __ movl(Address(to, count, sf, -4), rdx); 1165 1166 if (t == T_BYTE || t == T_SHORT) { 1167 __ subl(count, (1<<shift)); 1168 __ BIND(L_copy_2_bytes); 1169 // copy prefix dword 1170 __ testl(count, 1<<(shift-1)); 1171 __ jccb(Assembler::zero, L_copy_byte); 1172 __ movw(rdx, Address(from, count, sf, -2)); 1173 __ movw(Address(to, count, sf, -2), rdx); 1174 if (t == T_BYTE) { 1175 __ subl(count, 1<<(shift-1)); 1176 __ BIND(L_copy_byte); 1177 // copy prefix byte 1178 __ testl(count, 1); 1179 __ jccb(Assembler::zero, L_exit); 1180 __ movb(rdx, Address(from, 0)); 1181 __ movb(Address(to, 0), rdx); 1182 __ BIND(L_exit); 1183 } else { 1184 __ BIND(L_copy_byte); 1185 } 1186 } else { 1187 __ BIND(L_copy_2_bytes); 1188 } 1189 if (t == T_OBJECT) { 1190 __ movl2ptr(count, Address(rsp, 12+12)); // reread count 1191 gen_write_ref_array_post_barrier(to, count); 1192 __ BIND(L_0_count); 1193 } 1194 inc_copy_counter_np(t); 1195 __ pop(rdi); 1196 __ pop(rsi); 1197 __ leave(); // required for proper stackwalking of RuntimeStub frame 1198 __ xorptr(rax, rax); // return 0 1199 __ ret(0); 1200 return start; 1201 } 1202 1203 1204 address generate_disjoint_long_copy(address* entry, const char *name) { 1205 __ align(CodeEntryAlignment); 1206 StubCodeMark mark(this, "StubRoutines", name); 1207 address start = __ pc(); 1208 1209 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1210 const Register from = rax; // source array address 1211 const Register to = rdx; // destination array address 1212 const Register count = rcx; // elements count 1213 const Register to_from = rdx; // (to - from) 1214 1215 __ enter(); // required for proper stackwalking of RuntimeStub frame 1216 __ movptr(from , Address(rsp, 8+0)); // from 1217 __ movptr(to , Address(rsp, 8+4)); // to 1218 __ movl2ptr(count, Address(rsp, 8+8)); // count 1219 1220 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 1221 BLOCK_COMMENT("Entry:"); 1222 1223 __ subptr(to, from); // to --> to_from 1224 if (VM_Version::supports_mmx()) { 1225 if (UseXMMForArrayCopy) { 1226 xmm_copy_forward(from, to_from, count); 1227 } else { 1228 mmx_copy_forward(from, to_from, count); 1229 } 1230 } else { 1231 __ jmpb(L_copy_8_bytes); 1232 __ align(OptoLoopAlignment); 1233 __ BIND(L_copy_8_bytes_loop); 1234 __ fild_d(Address(from, 0)); 1235 __ fistp_d(Address(from, to_from, Address::times_1)); 1236 __ addptr(from, 8); 1237 __ BIND(L_copy_8_bytes); 1238 __ decrement(count); 1239 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1240 } 1241 inc_copy_counter_np(T_LONG); 1242 __ leave(); // required for proper stackwalking of RuntimeStub frame 1243 __ xorptr(rax, rax); // return 0 1244 __ ret(0); 1245 return start; 1246 } 1247 1248 address generate_conjoint_long_copy(address nooverlap_target, 1249 address* entry, const char *name) { 1250 __ align(CodeEntryAlignment); 1251 StubCodeMark mark(this, "StubRoutines", name); 1252 address start = __ pc(); 1253 1254 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1255 const Register from = rax; // source array address 1256 const Register to = rdx; // destination array address 1257 const Register count = rcx; // elements count 1258 const Register end_from = rax; // source array end address 1259 1260 __ enter(); // required for proper stackwalking of RuntimeStub frame 1261 __ movptr(from , Address(rsp, 8+0)); // from 1262 __ movptr(to , Address(rsp, 8+4)); // to 1263 __ movl2ptr(count, Address(rsp, 8+8)); // count 1264 1265 *entry = __ pc(); // Entry point from generic arraycopy stub. 1266 BLOCK_COMMENT("Entry:"); 1267 1268 // arrays overlap test 1269 __ cmpptr(to, from); 1270 RuntimeAddress nooverlap(nooverlap_target); 1271 __ jump_cc(Assembler::belowEqual, nooverlap); 1272 __ lea(end_from, Address(from, count, Address::times_8, 0)); 1273 __ cmpptr(to, end_from); 1274 __ movptr(from, Address(rsp, 8)); // from 1275 __ jump_cc(Assembler::aboveEqual, nooverlap); 1276 1277 __ jmpb(L_copy_8_bytes); 1278 1279 __ align(OptoLoopAlignment); 1280 __ BIND(L_copy_8_bytes_loop); 1281 if (VM_Version::supports_mmx()) { 1282 if (UseXMMForArrayCopy) { 1283 __ movq(xmm0, Address(from, count, Address::times_8)); 1284 __ movq(Address(to, count, Address::times_8), xmm0); 1285 } else { 1286 __ movq(mmx0, Address(from, count, Address::times_8)); 1287 __ movq(Address(to, count, Address::times_8), mmx0); 1288 } 1289 } else { 1290 __ fild_d(Address(from, count, Address::times_8)); 1291 __ fistp_d(Address(to, count, Address::times_8)); 1292 } 1293 __ BIND(L_copy_8_bytes); 1294 __ decrement(count); 1295 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1296 1297 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) { 1298 __ emms(); 1299 } 1300 inc_copy_counter_np(T_LONG); 1301 __ leave(); // required for proper stackwalking of RuntimeStub frame 1302 __ xorptr(rax, rax); // return 0 1303 __ ret(0); 1304 return start; 1305 } 1306 1307 1308 // Helper for generating a dynamic type check. 1309 // The sub_klass must be one of {rbx, rdx, rsi}. 1310 // The temp is killed. 1311 void generate_type_check(Register sub_klass, 1312 Address& super_check_offset_addr, 1313 Address& super_klass_addr, 1314 Register temp, 1315 Label* L_success, Label* L_failure) { 1316 BLOCK_COMMENT("type_check:"); 1317 1318 Label L_fallthrough; 1319 #define LOCAL_JCC(assembler_con, label_ptr) \ 1320 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ 1321 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ 1322 1323 // The following is a strange variation of the fast path which requires 1324 // one less register, because needed values are on the argument stack. 1325 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, 1326 // L_success, L_failure, NULL); 1327 assert_different_registers(sub_klass, temp); 1328 1329 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 1330 Klass::secondary_super_cache_offset_in_bytes()); 1331 1332 // if the pointers are equal, we are done (e.g., String[] elements) 1333 __ cmpptr(sub_klass, super_klass_addr); 1334 LOCAL_JCC(Assembler::equal, L_success); 1335 1336 // check the supertype display: 1337 __ movl2ptr(temp, super_check_offset_addr); 1338 Address super_check_addr(sub_klass, temp, Address::times_1, 0); 1339 __ movptr(temp, super_check_addr); // load displayed supertype 1340 __ cmpptr(temp, super_klass_addr); // test the super type 1341 LOCAL_JCC(Assembler::equal, L_success); 1342 1343 // if it was a primary super, we can just fail immediately 1344 __ cmpl(super_check_offset_addr, sc_offset); 1345 LOCAL_JCC(Assembler::notEqual, L_failure); 1346 1347 // The repne_scan instruction uses fixed registers, which will get spilled. 1348 // We happen to know this works best when super_klass is in rax. 1349 Register super_klass = temp; 1350 __ movptr(super_klass, super_klass_addr); 1351 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, 1352 L_success, L_failure); 1353 1354 __ bind(L_fallthrough); 1355 1356 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } 1357 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } 1358 1359 #undef LOCAL_JCC 1360 } 1361 1362 // 1363 // Generate checkcasting array copy stub 1364 // 1365 // Input: 1366 // 4(rsp) - source array address 1367 // 8(rsp) - destination array address 1368 // 12(rsp) - element count, can be zero 1369 // 16(rsp) - size_t ckoff (super_check_offset) 1370 // 20(rsp) - oop ckval (super_klass) 1371 // 1372 // Output: 1373 // rax, == 0 - success 1374 // rax, == -1^K - failure, where K is partial transfer count 1375 // 1376 address generate_checkcast_copy(const char *name, address* entry) { 1377 __ align(CodeEntryAlignment); 1378 StubCodeMark mark(this, "StubRoutines", name); 1379 address start = __ pc(); 1380 1381 Label L_load_element, L_store_element, L_do_card_marks, L_done; 1382 1383 // register use: 1384 // rax, rdx, rcx -- loop control (end_from, end_to, count) 1385 // rdi, rsi -- element access (oop, klass) 1386 // rbx, -- temp 1387 const Register from = rax; // source array address 1388 const Register to = rdx; // destination array address 1389 const Register length = rcx; // elements count 1390 const Register elem = rdi; // each oop copied 1391 const Register elem_klass = rsi; // each elem._klass (sub_klass) 1392 const Register temp = rbx; // lone remaining temp 1393 1394 __ enter(); // required for proper stackwalking of RuntimeStub frame 1395 1396 __ push(rsi); 1397 __ push(rdi); 1398 __ push(rbx); 1399 1400 Address from_arg(rsp, 16+ 4); // from 1401 Address to_arg(rsp, 16+ 8); // to 1402 Address length_arg(rsp, 16+12); // elements count 1403 Address ckoff_arg(rsp, 16+16); // super_check_offset 1404 Address ckval_arg(rsp, 16+20); // super_klass 1405 1406 // Load up: 1407 __ movptr(from, from_arg); 1408 __ movptr(to, to_arg); 1409 __ movl2ptr(length, length_arg); 1410 1411 *entry = __ pc(); // Entry point from generic arraycopy stub. 1412 BLOCK_COMMENT("Entry:"); 1413 1414 //--------------------------------------------------------------- 1415 // Assembler stub will be used for this call to arraycopy 1416 // if the two arrays are subtypes of Object[] but the 1417 // destination array type is not equal to or a supertype 1418 // of the source type. Each element must be separately 1419 // checked. 1420 1421 // Loop-invariant addresses. They are exclusive end pointers. 1422 Address end_from_addr(from, length, Address::times_ptr, 0); 1423 Address end_to_addr(to, length, Address::times_ptr, 0); 1424 1425 Register end_from = from; // re-use 1426 Register end_to = to; // re-use 1427 Register count = length; // re-use 1428 1429 // Loop-variant addresses. They assume post-incremented count < 0. 1430 Address from_element_addr(end_from, count, Address::times_ptr, 0); 1431 Address to_element_addr(end_to, count, Address::times_ptr, 0); 1432 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); 1433 1434 // Copy from low to high addresses, indexed from the end of each array. 1435 gen_write_ref_array_pre_barrier(to, count); 1436 __ lea(end_from, end_from_addr); 1437 __ lea(end_to, end_to_addr); 1438 assert(length == count, ""); // else fix next line: 1439 __ negptr(count); // negate and test the length 1440 __ jccb(Assembler::notZero, L_load_element); 1441 1442 // Empty array: Nothing to do. 1443 __ xorptr(rax, rax); // return 0 on (trivial) success 1444 __ jmp(L_done); 1445 1446 // ======== begin loop ======== 1447 // (Loop is rotated; its entry is L_load_element.) 1448 // Loop control: 1449 // for (count = -count; count != 0; count++) 1450 // Base pointers src, dst are biased by 8*count,to last element. 1451 __ align(OptoLoopAlignment); 1452 1453 __ BIND(L_store_element); 1454 __ movptr(to_element_addr, elem); // store the oop 1455 __ increment(count); // increment the count toward zero 1456 __ jccb(Assembler::zero, L_do_card_marks); 1457 1458 // ======== loop entry is here ======== 1459 __ BIND(L_load_element); 1460 __ movptr(elem, from_element_addr); // load the oop 1461 __ testptr(elem, elem); 1462 __ jccb(Assembler::zero, L_store_element); 1463 1464 // (Could do a trick here: Remember last successful non-null 1465 // element stored and make a quick oop equality check on it.) 1466 1467 __ movptr(elem_klass, elem_klass_addr); // query the object klass 1468 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, 1469 &L_store_element, NULL); 1470 // (On fall-through, we have failed the element type check.) 1471 // ======== end loop ======== 1472 1473 // It was a real error; we must depend on the caller to finish the job. 1474 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. 1475 // Emit GC store barriers for the oops we have copied (length_arg + count), 1476 // and report their number to the caller. 1477 __ addl(count, length_arg); // transfers = (length - remaining) 1478 __ movl2ptr(rax, count); // save the value 1479 __ notptr(rax); // report (-1^K) to caller 1480 __ movptr(to, to_arg); // reload 1481 assert_different_registers(to, count, rax); 1482 gen_write_ref_array_post_barrier(to, count); 1483 __ jmpb(L_done); 1484 1485 // Come here on success only. 1486 __ BIND(L_do_card_marks); 1487 __ movl2ptr(count, length_arg); 1488 __ movptr(to, to_arg); // reload 1489 gen_write_ref_array_post_barrier(to, count); 1490 __ xorptr(rax, rax); // return 0 on success 1491 1492 // Common exit point (success or failure). 1493 __ BIND(L_done); 1494 __ pop(rbx); 1495 __ pop(rdi); 1496 __ pop(rsi); 1497 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); 1498 __ leave(); // required for proper stackwalking of RuntimeStub frame 1499 __ ret(0); 1500 1501 return start; 1502 } 1503 1504 // 1505 // Generate 'unsafe' array copy stub 1506 // Though just as safe as the other stubs, it takes an unscaled 1507 // size_t argument instead of an element count. 1508 // 1509 // Input: 1510 // 4(rsp) - source array address 1511 // 8(rsp) - destination array address 1512 // 12(rsp) - byte count, can be zero 1513 // 1514 // Output: 1515 // rax, == 0 - success 1516 // rax, == -1 - need to call System.arraycopy 1517 // 1518 // Examines the alignment of the operands and dispatches 1519 // to a long, int, short, or byte copy loop. 1520 // 1521 address generate_unsafe_copy(const char *name, 1522 address byte_copy_entry, 1523 address short_copy_entry, 1524 address int_copy_entry, 1525 address long_copy_entry) { 1526 1527 Label L_long_aligned, L_int_aligned, L_short_aligned; 1528 1529 __ align(CodeEntryAlignment); 1530 StubCodeMark mark(this, "StubRoutines", name); 1531 address start = __ pc(); 1532 1533 const Register from = rax; // source array address 1534 const Register to = rdx; // destination array address 1535 const Register count = rcx; // elements count 1536 1537 __ enter(); // required for proper stackwalking of RuntimeStub frame 1538 __ push(rsi); 1539 __ push(rdi); 1540 Address from_arg(rsp, 12+ 4); // from 1541 Address to_arg(rsp, 12+ 8); // to 1542 Address count_arg(rsp, 12+12); // byte count 1543 1544 // Load up: 1545 __ movptr(from , from_arg); 1546 __ movptr(to , to_arg); 1547 __ movl2ptr(count, count_arg); 1548 1549 // bump this on entry, not on exit: 1550 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); 1551 1552 const Register bits = rsi; 1553 __ mov(bits, from); 1554 __ orptr(bits, to); 1555 __ orptr(bits, count); 1556 1557 __ testl(bits, BytesPerLong-1); 1558 __ jccb(Assembler::zero, L_long_aligned); 1559 1560 __ testl(bits, BytesPerInt-1); 1561 __ jccb(Assembler::zero, L_int_aligned); 1562 1563 __ testl(bits, BytesPerShort-1); 1564 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); 1565 1566 __ BIND(L_short_aligned); 1567 __ shrptr(count, LogBytesPerShort); // size => short_count 1568 __ movl(count_arg, count); // update 'count' 1569 __ jump(RuntimeAddress(short_copy_entry)); 1570 1571 __ BIND(L_int_aligned); 1572 __ shrptr(count, LogBytesPerInt); // size => int_count 1573 __ movl(count_arg, count); // update 'count' 1574 __ jump(RuntimeAddress(int_copy_entry)); 1575 1576 __ BIND(L_long_aligned); 1577 __ shrptr(count, LogBytesPerLong); // size => qword_count 1578 __ movl(count_arg, count); // update 'count' 1579 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1580 __ pop(rsi); 1581 __ jump(RuntimeAddress(long_copy_entry)); 1582 1583 return start; 1584 } 1585 1586 1587 // Perform range checks on the proposed arraycopy. 1588 // Smashes src_pos and dst_pos. (Uses them up for temps.) 1589 void arraycopy_range_checks(Register src, 1590 Register src_pos, 1591 Register dst, 1592 Register dst_pos, 1593 Address& length, 1594 Label& L_failed) { 1595 BLOCK_COMMENT("arraycopy_range_checks:"); 1596 const Register src_end = src_pos; // source array end position 1597 const Register dst_end = dst_pos; // destination array end position 1598 __ addl(src_end, length); // src_pos + length 1599 __ addl(dst_end, length); // dst_pos + length 1600 1601 // if (src_pos + length > arrayOop(src)->length() ) FAIL; 1602 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); 1603 __ jcc(Assembler::above, L_failed); 1604 1605 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; 1606 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); 1607 __ jcc(Assembler::above, L_failed); 1608 1609 BLOCK_COMMENT("arraycopy_range_checks done"); 1610 } 1611 1612 1613 // 1614 // Generate generic array copy stubs 1615 // 1616 // Input: 1617 // 4(rsp) - src oop 1618 // 8(rsp) - src_pos 1619 // 12(rsp) - dst oop 1620 // 16(rsp) - dst_pos 1621 // 20(rsp) - element count 1622 // 1623 // Output: 1624 // rax, == 0 - success 1625 // rax, == -1^K - failure, where K is partial transfer count 1626 // 1627 address generate_generic_copy(const char *name, 1628 address entry_jbyte_arraycopy, 1629 address entry_jshort_arraycopy, 1630 address entry_jint_arraycopy, 1631 address entry_oop_arraycopy, 1632 address entry_jlong_arraycopy, 1633 address entry_checkcast_arraycopy) { 1634 Label L_failed, L_failed_0, L_objArray; 1635 1636 { int modulus = CodeEntryAlignment; 1637 int target = modulus - 5; // 5 = sizeof jmp(L_failed) 1638 int advance = target - (__ offset() % modulus); 1639 if (advance < 0) advance += modulus; 1640 if (advance > 0) __ nop(advance); 1641 } 1642 StubCodeMark mark(this, "StubRoutines", name); 1643 1644 // Short-hop target to L_failed. Makes for denser prologue code. 1645 __ BIND(L_failed_0); 1646 __ jmp(L_failed); 1647 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); 1648 1649 __ align(CodeEntryAlignment); 1650 address start = __ pc(); 1651 1652 __ enter(); // required for proper stackwalking of RuntimeStub frame 1653 __ push(rsi); 1654 __ push(rdi); 1655 1656 // bump this on entry, not on exit: 1657 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); 1658 1659 // Input values 1660 Address SRC (rsp, 12+ 4); 1661 Address SRC_POS (rsp, 12+ 8); 1662 Address DST (rsp, 12+12); 1663 Address DST_POS (rsp, 12+16); 1664 Address LENGTH (rsp, 12+20); 1665 1666 //----------------------------------------------------------------------- 1667 // Assembler stub will be used for this call to arraycopy 1668 // if the following conditions are met: 1669 // 1670 // (1) src and dst must not be null. 1671 // (2) src_pos must not be negative. 1672 // (3) dst_pos must not be negative. 1673 // (4) length must not be negative. 1674 // (5) src klass and dst klass should be the same and not NULL. 1675 // (6) src and dst should be arrays. 1676 // (7) src_pos + length must not exceed length of src. 1677 // (8) dst_pos + length must not exceed length of dst. 1678 // 1679 1680 const Register src = rax; // source array oop 1681 const Register src_pos = rsi; 1682 const Register dst = rdx; // destination array oop 1683 const Register dst_pos = rdi; 1684 const Register length = rcx; // transfer count 1685 1686 // if (src == NULL) return -1; 1687 __ movptr(src, SRC); // src oop 1688 __ testptr(src, src); 1689 __ jccb(Assembler::zero, L_failed_0); 1690 1691 // if (src_pos < 0) return -1; 1692 __ movl2ptr(src_pos, SRC_POS); // src_pos 1693 __ testl(src_pos, src_pos); 1694 __ jccb(Assembler::negative, L_failed_0); 1695 1696 // if (dst == NULL) return -1; 1697 __ movptr(dst, DST); // dst oop 1698 __ testptr(dst, dst); 1699 __ jccb(Assembler::zero, L_failed_0); 1700 1701 // if (dst_pos < 0) return -1; 1702 __ movl2ptr(dst_pos, DST_POS); // dst_pos 1703 __ testl(dst_pos, dst_pos); 1704 __ jccb(Assembler::negative, L_failed_0); 1705 1706 // if (length < 0) return -1; 1707 __ movl2ptr(length, LENGTH); // length 1708 __ testl(length, length); 1709 __ jccb(Assembler::negative, L_failed_0); 1710 1711 // if (src->klass() == NULL) return -1; 1712 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); 1713 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); 1714 const Register rcx_src_klass = rcx; // array klass 1715 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); 1716 1717 #ifdef ASSERT 1718 // assert(src->klass() != NULL); 1719 BLOCK_COMMENT("assert klasses not null"); 1720 { Label L1, L2; 1721 __ testptr(rcx_src_klass, rcx_src_klass); 1722 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL 1723 __ bind(L1); 1724 __ stop("broken null klass"); 1725 __ bind(L2); 1726 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); 1727 __ jccb(Assembler::equal, L1); // this would be broken also 1728 BLOCK_COMMENT("assert done"); 1729 } 1730 #endif //ASSERT 1731 1732 // Load layout helper (32-bits) 1733 // 1734 // |array_tag| | header_size | element_type | |log2_element_size| 1735 // 32 30 24 16 8 2 0 1736 // 1737 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 1738 // 1739 1740 int lh_offset = klassOopDesc::header_size() * HeapWordSize + 1741 Klass::layout_helper_offset_in_bytes(); 1742 Address src_klass_lh_addr(rcx_src_klass, lh_offset); 1743 1744 // Handle objArrays completely differently... 1745 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); 1746 __ cmpl(src_klass_lh_addr, objArray_lh); 1747 __ jcc(Assembler::equal, L_objArray); 1748 1749 // if (src->klass() != dst->klass()) return -1; 1750 __ cmpptr(rcx_src_klass, dst_klass_addr); 1751 __ jccb(Assembler::notEqual, L_failed_0); 1752 1753 const Register rcx_lh = rcx; // layout helper 1754 assert(rcx_lh == rcx_src_klass, "known alias"); 1755 __ movl(rcx_lh, src_klass_lh_addr); 1756 1757 // if (!src->is_Array()) return -1; 1758 __ cmpl(rcx_lh, Klass::_lh_neutral_value); 1759 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp 1760 1761 // At this point, it is known to be a typeArray (array_tag 0x3). 1762 #ifdef ASSERT 1763 { Label L; 1764 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); 1765 __ jcc(Assembler::greaterEqual, L); // signed cmp 1766 __ stop("must be a primitive array"); 1767 __ bind(L); 1768 } 1769 #endif 1770 1771 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); 1772 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1773 1774 // typeArrayKlass 1775 // 1776 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); 1777 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); 1778 // 1779 const Register rsi_offset = rsi; // array offset 1780 const Register src_array = src; // src array offset 1781 const Register dst_array = dst; // dst array offset 1782 const Register rdi_elsize = rdi; // log2 element size 1783 1784 __ mov(rsi_offset, rcx_lh); 1785 __ shrptr(rsi_offset, Klass::_lh_header_size_shift); 1786 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset 1787 __ addptr(src_array, rsi_offset); // src array offset 1788 __ addptr(dst_array, rsi_offset); // dst array offset 1789 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize 1790 1791 // next registers should be set before the jump to corresponding stub 1792 const Register from = src; // source array address 1793 const Register to = dst; // destination array address 1794 const Register count = rcx; // elements count 1795 // some of them should be duplicated on stack 1796 #define FROM Address(rsp, 12+ 4) 1797 #define TO Address(rsp, 12+ 8) // Not used now 1798 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy 1799 1800 BLOCK_COMMENT("scale indexes to element size"); 1801 __ movl2ptr(rsi, SRC_POS); // src_pos 1802 __ shlptr(rsi); // src_pos << rcx (log2 elsize) 1803 assert(src_array == from, ""); 1804 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize 1805 __ movl2ptr(rdi, DST_POS); // dst_pos 1806 __ shlptr(rdi); // dst_pos << rcx (log2 elsize) 1807 assert(dst_array == to, ""); 1808 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize 1809 __ movptr(FROM, from); // src_addr 1810 __ mov(rdi_elsize, rcx_lh); // log2 elsize 1811 __ movl2ptr(count, LENGTH); // elements count 1812 1813 BLOCK_COMMENT("choose copy loop based on element size"); 1814 __ cmpl(rdi_elsize, 0); 1815 1816 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); 1817 __ cmpl(rdi_elsize, LogBytesPerShort); 1818 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); 1819 __ cmpl(rdi_elsize, LogBytesPerInt); 1820 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); 1821 #ifdef ASSERT 1822 __ cmpl(rdi_elsize, LogBytesPerLong); 1823 __ jccb(Assembler::notEqual, L_failed); 1824 #endif 1825 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1826 __ pop(rsi); 1827 __ jump(RuntimeAddress(entry_jlong_arraycopy)); 1828 1829 __ BIND(L_failed); 1830 __ xorptr(rax, rax); 1831 __ notptr(rax); // return -1 1832 __ pop(rdi); 1833 __ pop(rsi); 1834 __ leave(); // required for proper stackwalking of RuntimeStub frame 1835 __ ret(0); 1836 1837 // objArrayKlass 1838 __ BIND(L_objArray); 1839 // live at this point: rcx_src_klass, src[_pos], dst[_pos] 1840 1841 Label L_plain_copy, L_checkcast_copy; 1842 // test array classes for subtyping 1843 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality 1844 __ jccb(Assembler::notEqual, L_checkcast_copy); 1845 1846 // Identically typed arrays can be copied without element-wise checks. 1847 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); 1848 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1849 1850 __ BIND(L_plain_copy); 1851 __ movl2ptr(count, LENGTH); // elements count 1852 __ movl2ptr(src_pos, SRC_POS); // reload src_pos 1853 __ lea(from, Address(src, src_pos, Address::times_ptr, 1854 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr 1855 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos 1856 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1857 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr 1858 __ movptr(FROM, from); // src_addr 1859 __ movptr(TO, to); // dst_addr 1860 __ movl(COUNT, count); // count 1861 __ jump(RuntimeAddress(entry_oop_arraycopy)); 1862 1863 __ BIND(L_checkcast_copy); 1864 // live at this point: rcx_src_klass, dst[_pos], src[_pos] 1865 { 1866 // Handy offsets: 1867 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + 1868 objArrayKlass::element_klass_offset_in_bytes()); 1869 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 1870 Klass::super_check_offset_offset_in_bytes()); 1871 1872 Register rsi_dst_klass = rsi; 1873 Register rdi_temp = rdi; 1874 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); 1875 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); 1876 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); 1877 1878 // Before looking at dst.length, make sure dst is also an objArray. 1879 __ movptr(rsi_dst_klass, dst_klass_addr); 1880 __ cmpl(dst_klass_lh_addr, objArray_lh); 1881 __ jccb(Assembler::notEqual, L_failed); 1882 1883 // It is safe to examine both src.length and dst.length. 1884 __ movl2ptr(src_pos, SRC_POS); // reload rsi 1885 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1886 // (Now src_pos and dst_pos are killed, but not src and dst.) 1887 1888 // We'll need this temp (don't forget to pop it after the type check). 1889 __ push(rbx); 1890 Register rbx_src_klass = rbx; 1891 1892 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx 1893 __ movptr(rsi_dst_klass, dst_klass_addr); 1894 Address super_check_offset_addr(rsi_dst_klass, sco_offset); 1895 Label L_fail_array_check; 1896 generate_type_check(rbx_src_klass, 1897 super_check_offset_addr, dst_klass_addr, 1898 rdi_temp, NULL, &L_fail_array_check); 1899 // (On fall-through, we have passed the array type check.) 1900 __ pop(rbx); 1901 __ jmp(L_plain_copy); 1902 1903 __ BIND(L_fail_array_check); 1904 // Reshuffle arguments so we can call checkcast_arraycopy: 1905 1906 // match initial saves for checkcast_arraycopy 1907 // push(rsi); // already done; see above 1908 // push(rdi); // already done; see above 1909 // push(rbx); // already done; see above 1910 1911 // Marshal outgoing arguments now, freeing registers. 1912 Address from_arg(rsp, 16+ 4); // from 1913 Address to_arg(rsp, 16+ 8); // to 1914 Address length_arg(rsp, 16+12); // elements count 1915 Address ckoff_arg(rsp, 16+16); // super_check_offset 1916 Address ckval_arg(rsp, 16+20); // super_klass 1917 1918 Address SRC_POS_arg(rsp, 16+ 8); 1919 Address DST_POS_arg(rsp, 16+16); 1920 Address LENGTH_arg(rsp, 16+20); 1921 // push rbx, changed the incoming offsets (why not just use rbp,??) 1922 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); 1923 1924 __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); 1925 __ movl2ptr(length, LENGTH_arg); // reload elements count 1926 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos 1927 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos 1928 1929 __ movptr(ckval_arg, rbx); // destination element type 1930 __ movl(rbx, Address(rbx, sco_offset)); 1931 __ movl(ckoff_arg, rbx); // corresponding class check offset 1932 1933 __ movl(length_arg, length); // outgoing length argument 1934 1935 __ lea(from, Address(src, src_pos, Address::times_ptr, 1936 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1937 __ movptr(from_arg, from); 1938 1939 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1940 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1941 __ movptr(to_arg, to); 1942 __ jump(RuntimeAddress(entry_checkcast_arraycopy)); 1943 } 1944 1945 return start; 1946 } 1947 1948 void generate_arraycopy_stubs() { 1949 address entry; 1950 address entry_jbyte_arraycopy; 1951 address entry_jshort_arraycopy; 1952 address entry_jint_arraycopy; 1953 address entry_oop_arraycopy; 1954 address entry_jlong_arraycopy; 1955 address entry_checkcast_arraycopy; 1956 1957 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = 1958 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, 1959 "arrayof_jbyte_disjoint_arraycopy"); 1960 StubRoutines::_arrayof_jbyte_arraycopy = 1961 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, 1962 NULL, "arrayof_jbyte_arraycopy"); 1963 StubRoutines::_jbyte_disjoint_arraycopy = 1964 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, 1965 "jbyte_disjoint_arraycopy"); 1966 StubRoutines::_jbyte_arraycopy = 1967 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, 1968 &entry_jbyte_arraycopy, "jbyte_arraycopy"); 1969 1970 StubRoutines::_arrayof_jshort_disjoint_arraycopy = 1971 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, 1972 "arrayof_jshort_disjoint_arraycopy"); 1973 StubRoutines::_arrayof_jshort_arraycopy = 1974 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, 1975 NULL, "arrayof_jshort_arraycopy"); 1976 StubRoutines::_jshort_disjoint_arraycopy = 1977 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, 1978 "jshort_disjoint_arraycopy"); 1979 StubRoutines::_jshort_arraycopy = 1980 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, 1981 &entry_jshort_arraycopy, "jshort_arraycopy"); 1982 1983 // Next arrays are always aligned on 4 bytes at least. 1984 StubRoutines::_jint_disjoint_arraycopy = 1985 generate_disjoint_copy(T_INT, true, Address::times_4, &entry, 1986 "jint_disjoint_arraycopy"); 1987 StubRoutines::_jint_arraycopy = 1988 generate_conjoint_copy(T_INT, true, Address::times_4, entry, 1989 &entry_jint_arraycopy, "jint_arraycopy"); 1990 1991 StubRoutines::_oop_disjoint_arraycopy = 1992 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, 1993 "oop_disjoint_arraycopy"); 1994 StubRoutines::_oop_arraycopy = 1995 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, 1996 &entry_oop_arraycopy, "oop_arraycopy"); 1997 1998 StubRoutines::_jlong_disjoint_arraycopy = 1999 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); 2000 StubRoutines::_jlong_arraycopy = 2001 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, 2002 "jlong_arraycopy"); 2003 2004 StubRoutines::_arrayof_jint_disjoint_arraycopy = 2005 StubRoutines::_jint_disjoint_arraycopy; 2006 StubRoutines::_arrayof_oop_disjoint_arraycopy = 2007 StubRoutines::_oop_disjoint_arraycopy; 2008 StubRoutines::_arrayof_jlong_disjoint_arraycopy = 2009 StubRoutines::_jlong_disjoint_arraycopy; 2010 2011 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; 2012 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; 2013 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; 2014 2015 StubRoutines::_checkcast_arraycopy = 2016 generate_checkcast_copy("checkcast_arraycopy", 2017 &entry_checkcast_arraycopy); 2018 2019 StubRoutines::_unsafe_arraycopy = 2020 generate_unsafe_copy("unsafe_arraycopy", 2021 entry_jbyte_arraycopy, 2022 entry_jshort_arraycopy, 2023 entry_jint_arraycopy, 2024 entry_jlong_arraycopy); 2025 2026 StubRoutines::_generic_arraycopy = 2027 generate_generic_copy("generic_arraycopy", 2028 entry_jbyte_arraycopy, 2029 entry_jshort_arraycopy, 2030 entry_jint_arraycopy, 2031 entry_oop_arraycopy, 2032 entry_jlong_arraycopy, 2033 entry_checkcast_arraycopy); 2034 } 2035 2036 void generate_math_stubs() { 2037 { 2038 StubCodeMark mark(this, "StubRoutines", "log"); 2039 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); 2040 2041 __ fld_d(Address(rsp, 4)); 2042 __ flog(); 2043 __ ret(0); 2044 } 2045 { 2046 StubCodeMark mark(this, "StubRoutines", "log10"); 2047 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); 2048 2049 __ fld_d(Address(rsp, 4)); 2050 __ flog10(); 2051 __ ret(0); 2052 } 2053 { 2054 StubCodeMark mark(this, "StubRoutines", "sin"); 2055 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); 2056 2057 __ fld_d(Address(rsp, 4)); 2058 __ trigfunc('s'); 2059 __ ret(0); 2060 } 2061 { 2062 StubCodeMark mark(this, "StubRoutines", "cos"); 2063 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); 2064 2065 __ fld_d(Address(rsp, 4)); 2066 __ trigfunc('c'); 2067 __ ret(0); 2068 } 2069 { 2070 StubCodeMark mark(this, "StubRoutines", "tan"); 2071 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); 2072 2073 __ fld_d(Address(rsp, 4)); 2074 __ trigfunc('t'); 2075 __ ret(0); 2076 } 2077 2078 // The intrinsic version of these seem to return the same value as 2079 // the strict version. 2080 StubRoutines::_intrinsic_exp = SharedRuntime::dexp; 2081 StubRoutines::_intrinsic_pow = SharedRuntime::dpow; 2082 } 2083 2084 public: 2085 // Information about frame layout at time of blocking runtime call. 2086 // Note that we only have to preserve callee-saved registers since 2087 // the compilers are responsible for supplying a continuation point 2088 // if they expect all registers to be preserved. 2089 enum layout { 2090 thread_off, // last_java_sp 2091 rbp_off, // callee saved register 2092 ret_pc, 2093 framesize 2094 }; 2095 2096 private: 2097 2098 #undef __ 2099 #define __ masm-> 2100 2101 //------------------------------------------------------------------------------------------------------------------------ 2102 // Continuation point for throwing of implicit exceptions that are not handled in 2103 // the current activation. Fabricates an exception oop and initiates normal 2104 // exception dispatching in this frame. 2105 // 2106 // Previously the compiler (c2) allowed for callee save registers on Java calls. 2107 // This is no longer true after adapter frames were removed but could possibly 2108 // be brought back in the future if the interpreter code was reworked and it 2109 // was deemed worthwhile. The comment below was left to describe what must 2110 // happen here if callee saves were resurrected. As it stands now this stub 2111 // could actually be a vanilla BufferBlob and have now oopMap at all. 2112 // Since it doesn't make much difference we've chosen to leave it the 2113 // way it was in the callee save days and keep the comment. 2114 2115 // If we need to preserve callee-saved values we need a callee-saved oop map and 2116 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. 2117 // If the compiler needs all registers to be preserved between the fault 2118 // point and the exception handler then it must assume responsibility for that in 2119 // AbstractCompiler::continuation_for_implicit_null_exception or 2120 // continuation_for_implicit_division_by_zero_exception. All other implicit 2121 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are 2122 // either at call sites or otherwise assume that stack unwinding will be initiated, 2123 // so caller saved registers were assumed volatile in the compiler. 2124 address generate_throw_exception(const char* name, address runtime_entry, 2125 bool restore_saved_exception_pc) { 2126 2127 int insts_size = 256; 2128 int locs_size = 32; 2129 2130 CodeBuffer code(name, insts_size, locs_size); 2131 OopMapSet* oop_maps = new OopMapSet(); 2132 MacroAssembler* masm = new MacroAssembler(&code); 2133 2134 address start = __ pc(); 2135 2136 // This is an inlined and slightly modified version of call_VM 2137 // which has the ability to fetch the return PC out of 2138 // thread-local storage and also sets up last_Java_sp slightly 2139 // differently than the real call_VM 2140 Register java_thread = rbx; 2141 __ get_thread(java_thread); 2142 if (restore_saved_exception_pc) { 2143 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset()))); 2144 __ push(rax); 2145 } 2146 2147 __ enter(); // required for proper stackwalking of RuntimeStub frame 2148 2149 // pc and rbp, already pushed 2150 __ subptr(rsp, (framesize-2) * wordSize); // prolog 2151 2152 // Frame is now completed as far as size and linkage. 2153 2154 int frame_complete = __ pc() - start; 2155 2156 // push java thread (becomes first argument of C function) 2157 __ movptr(Address(rsp, thread_off * wordSize), java_thread); 2158 2159 // Set up last_Java_sp and last_Java_fp 2160 __ set_last_Java_frame(java_thread, rsp, rbp, NULL); 2161 2162 // Call runtime 2163 BLOCK_COMMENT("call runtime_entry"); 2164 __ call(RuntimeAddress(runtime_entry)); 2165 // Generate oop map 2166 OopMap* map = new OopMap(framesize, 0); 2167 oop_maps->add_gc_map(__ pc() - start, map); 2168 2169 // restore the thread (cannot use the pushed argument since arguments 2170 // may be overwritten by C code generated by an optimizing compiler); 2171 // however can use the register value directly if it is callee saved. 2172 __ get_thread(java_thread); 2173 2174 __ reset_last_Java_frame(java_thread, true, false); 2175 2176 __ leave(); // required for proper stackwalking of RuntimeStub frame 2177 2178 // check for pending exceptions 2179 #ifdef ASSERT 2180 Label L; 2181 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2182 __ jcc(Assembler::notEqual, L); 2183 __ should_not_reach_here(); 2184 __ bind(L); 2185 #endif /* ASSERT */ 2186 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2187 2188 2189 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); 2190 return stub->entry_point(); 2191 } 2192 2193 2194 void create_control_words() { 2195 // Round to nearest, 53-bit mode, exceptions masked 2196 StubRoutines::_fpu_cntrl_wrd_std = 0x027F; 2197 // Round to zero, 53-bit mode, exception mased 2198 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; 2199 // Round to nearest, 24-bit mode, exceptions masked 2200 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; 2201 // Round to nearest, 64-bit mode, exceptions masked 2202 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; 2203 // Round to nearest, 64-bit mode, exceptions masked 2204 StubRoutines::_mxcsr_std = 0x1F80; 2205 // Note: the following two constants are 80-bit values 2206 // layout is critical for correct loading by FPU. 2207 // Bias for strict fp multiply/divide 2208 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 2209 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; 2210 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; 2211 // Un-Bias for strict fp multiply/divide 2212 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 2213 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; 2214 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; 2215 } 2216 2217 //--------------------------------------------------------------------------- 2218 // Initialization 2219 2220 void generate_initial() { 2221 // Generates all stubs and initializes the entry points 2222 2223 //------------------------------------------------------------------------------------------------------------------------ 2224 // entry points that exist in all platforms 2225 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than 2226 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. 2227 StubRoutines::_forward_exception_entry = generate_forward_exception(); 2228 2229 StubRoutines::_call_stub_entry = 2230 generate_call_stub(StubRoutines::_call_stub_return_address); 2231 // is referenced by megamorphic call 2232 StubRoutines::_catch_exception_entry = generate_catch_exception(); 2233 2234 // These are currently used by Solaris/Intel 2235 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); 2236 2237 StubRoutines::_handler_for_unsafe_access_entry = 2238 generate_handler_for_unsafe_access(); 2239 2240 // platform dependent 2241 create_control_words(); 2242 2243 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); 2244 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); 2245 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, 2246 CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); 2247 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, 2248 CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); 2249 } 2250 2251 2252 void generate_all() { 2253 // Generates all stubs and initializes the entry points 2254 2255 // These entry points require SharedInfo::stack0 to be set up in non-core builds 2256 // and need to be relocatable, so they each fabricate a RuntimeStub internally. 2257 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false); 2258 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false); 2259 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true); 2260 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true); 2261 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); 2262 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false); 2263 2264 //------------------------------------------------------------------------------------------------------------------------ 2265 // entry points that are platform specific 2266 2267 // support for verify_oop (must happen after universe_init) 2268 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); 2269 2270 // arraycopy stubs used by compilers 2271 generate_arraycopy_stubs(); 2272 2273 generate_math_stubs(); 2274 } 2275 2276 2277 public: 2278 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { 2279 if (all) { 2280 generate_all(); 2281 } else { 2282 generate_initial(); 2283 } 2284 } 2285 }; // end class declaration 2286 2287 2288 void StubGenerator_generate(CodeBuffer* code, bool all) { 2289 StubGenerator g(code, all); 2290 }