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 */ 24 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 53 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 57 58 #define __ _masm-> 59 #define a__ ((Assembler*)_masm)-> 60 61 #ifdef PRODUCT 62 #define BLOCK_COMMENT(str) /* nothing */ 63 #else 64 #define BLOCK_COMMENT(str) __ block_comment(str) 65 #endif 66 67 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 68 69 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions 70 const int FPU_CNTRL_WRD_MASK = 0xFFFF; 71 72 // ------------------------------------------------------------------------------------------------------------------------- 73 // Stub Code definitions 74 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); 82 83 // request an async exception 84 thread->set_pending_unsafe_access_error(); 85 86 // return address of next instruction to execute 87 return npc; 88 } 89 90 class StubGenerator: public StubCodeGenerator { 91 private: 92 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 103 104 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 } 116 117 //------------------------------------------------------------------------------------------------------------------------ 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 ] 141 142 143 address generate_call_stub(address& return_address) { 144 StubCodeMark mark(this, "StubRoutines", "call_stub"); 145 address start = __ pc(); 146 147 // 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; 164 165 // 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 172 173 // 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 } 189 190 // make sure the control word is correct. 191 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 192 193 #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 203 204 // 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); 210 211 // parameter passing loop 212 213 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] 218 219 __ movptr(rdx, parameters); // parameter pointer 220 __ xorptr(rbx, rbx); 221 222 __ BIND(loop); 223 224 // 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); 231 232 // 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); 239 240 BLOCK_COMMENT("call_stub_return_address:"); 241 return_address = __ pc(); 242 243 Label common_return; 244 245 __ BIND(common_return); 246 247 // store result depending on type 248 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) 249 __ movptr(rdi, result); 250 Label is_long, is_float, is_double, exit; 251 __ movl(rsi, result_type); 252 __ cmpl(rsi, T_LONG); 253 __ jcc(Assembler::equal, is_long); 254 __ cmpl(rsi, T_FLOAT); 255 __ jcc(Assembler::equal, is_float); 256 __ cmpl(rsi, T_DOUBLE); 257 __ jcc(Assembler::equal, is_double); 258 259 // handle T_INT case 260 __ movl(Address(rdi, 0), rax); 261 __ BIND(exit); 262 263 // check that FPU stack is empty 264 __ verify_FPU(0, "generate_call_stub"); 265 266 // pop parameters 267 __ lea(rsp, rsp_after_call); 268 269 // restore %mxcsr 270 if (sse_save) { 271 __ ldmxcsr(mxcsr_save); 272 } 273 274 // restore rdi, rsi and rbx, 275 __ movptr(rbx, saved_rbx); 276 __ movptr(rsi, saved_rsi); 277 __ movptr(rdi, saved_rdi); 278 __ addptr(rsp, 4*wordSize); 279 280 // return 281 __ pop(rbp); 282 __ ret(0); 283 284 // handle return types different from T_INT 285 __ BIND(is_long); 286 __ movl(Address(rdi, 0 * wordSize), rax); 287 __ movl(Address(rdi, 1 * wordSize), rdx); 288 __ jmp(exit); 289 290 __ BIND(is_float); 291 // interpreter uses xmm0 for return values 292 if (UseSSE >= 1) { 293 __ movflt(Address(rdi, 0), xmm0); 294 } else { 295 __ fstp_s(Address(rdi, 0)); 296 } 297 __ jmp(exit); 298 299 __ BIND(is_double); 300 // interpreter uses xmm0 for return values 301 if (UseSSE >= 2) { 302 __ movdbl(Address(rdi, 0), xmm0); 303 } else { 304 __ fstp_d(Address(rdi, 0)); 305 } 306 __ jmp(exit); 307 308 // If we call compiled code directly from the call stub we will 309 // need to adjust the return back to the call stub to a specialized 310 // piece of code that can handle compiled results and cleaning the fpu 311 // stack. compiled code will be set to return here instead of the 312 // return above that handles interpreter returns. 313 314 BLOCK_COMMENT("call_stub_compiled_return:"); 315 StubRoutines::x86::set_call_stub_compiled_return( __ pc()); 316 317 #ifdef COMPILER2 318 if (UseSSE >= 2) { 319 __ verify_FPU(0, "call_stub_compiled_return"); 320 } else { 321 for (int i = 1; i < 8; i++) { 322 __ ffree(i); 323 } 324 325 // UseSSE <= 1 so double result should be left on TOS 326 __ movl(rsi, result_type); 327 __ cmpl(rsi, T_DOUBLE); 328 __ jcc(Assembler::equal, common_return); 329 if (UseSSE == 0) { 330 // UseSSE == 0 so float result should be left on TOS 331 __ cmpl(rsi, T_FLOAT); 332 __ jcc(Assembler::equal, common_return); 333 } 334 __ ffree(0); 335 } 336 #endif /* COMPILER2 */ 337 __ jmp(common_return); 338 339 return start; 340 } 341 342 343 //------------------------------------------------------------------------------------------------------------------------ 344 // Return point for a Java call if there's an exception thrown in Java code. 345 // The exception is caught and transformed into a pending exception stored in 346 // JavaThread that can be tested from within the VM. 347 // 348 // Note: Usually the parameters are removed by the callee. In case of an exception 349 // crossing an activation frame boundary, that is not the case if the callee 350 // is compiled code => need to setup the rsp. 351 // 352 // rax,: exception oop 353 354 address generate_catch_exception() { 355 StubCodeMark mark(this, "StubRoutines", "catch_exception"); 356 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! 357 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! 358 address start = __ pc(); 359 360 // get thread directly 361 __ movptr(rcx, thread); 362 #ifdef ASSERT 363 // verify that threads correspond 364 { Label L; 365 __ get_thread(rbx); 366 __ cmpptr(rbx, rcx); 367 __ jcc(Assembler::equal, L); 368 __ stop("StubRoutines::catch_exception: threads must correspond"); 369 __ bind(L); 370 } 371 #endif 372 // set pending exception 373 __ verify_oop(rax); 374 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); 375 __ lea(Address(rcx, Thread::exception_file_offset ()), 376 ExternalAddress((address)__FILE__)); 377 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); 378 // complete return to VM 379 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); 380 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); 381 382 return start; 383 } 384 385 386 //------------------------------------------------------------------------------------------------------------------------ 387 // Continuation point for runtime calls returning with a pending exception. 388 // The pending exception check happened in the runtime or native call stub. 389 // The pending exception in Thread is converted into a Java-level exception. 390 // 391 // Contract with Java-level exception handlers: 392 // rax: exception 393 // rdx: throwing pc 394 // 395 // NOTE: At entry of this stub, exception-pc must be on stack !! 396 397 address generate_forward_exception() { 398 StubCodeMark mark(this, "StubRoutines", "forward exception"); 399 address start = __ pc(); 400 const Register thread = rcx; 401 402 // other registers used in this stub 403 const Register exception_oop = rax; 404 const Register handler_addr = rbx; 405 const Register exception_pc = rdx; 406 407 // Upon entry, the sp points to the return address returning into Java 408 // (interpreted or compiled) code; i.e., the return address becomes the 409 // throwing pc. 410 // 411 // Arguments pushed before the runtime call are still on the stack but 412 // the exception handler will reset the stack pointer -> ignore them. 413 // A potential result in registers can be ignored as well. 414 415 #ifdef ASSERT 416 // make sure this code is only executed if there is a pending exception 417 { Label L; 418 __ get_thread(thread); 419 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 420 __ jcc(Assembler::notEqual, L); 421 __ stop("StubRoutines::forward exception: no pending exception (1)"); 422 __ bind(L); 423 } 424 #endif 425 426 // compute exception handler into rbx, 427 __ get_thread(thread); 428 __ movptr(exception_pc, Address(rsp, 0)); 429 BLOCK_COMMENT("call exception_handler_for_return_address"); 430 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); 431 __ mov(handler_addr, rax); 432 433 // setup rax & rdx, remove return address & clear pending exception 434 __ get_thread(thread); 435 __ pop(exception_pc); 436 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); 437 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 438 439 #ifdef ASSERT 440 // make sure exception is set 441 { Label L; 442 __ testptr(exception_oop, exception_oop); 443 __ jcc(Assembler::notEqual, L); 444 __ stop("StubRoutines::forward exception: no pending exception (2)"); 445 __ bind(L); 446 } 447 #endif 448 449 // Verify that there is really a valid exception in RAX. 450 __ verify_oop(exception_oop); 451 452 // Restore SP from BP if the exception PC is a MethodHandle call site. 453 __ cmpl(Address(thread, JavaThread::is_method_handle_return_offset()), 0); 454 __ cmovptr(Assembler::notEqual, rsp, rbp); 455 456 // continue at exception handler (return address removed) 457 // rax: exception 458 // rbx: exception handler 459 // rdx: throwing pc 460 __ jmp(handler_addr); 461 462 return start; 463 } 464 465 466 //---------------------------------------------------------------------------------------------------- 467 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) 468 // 469 // xchg exists as far back as 8086, lock needed for MP only 470 // Stack layout immediately after call: 471 // 472 // 0 [ret addr ] <--- rsp 473 // 1 [ ex ] 474 // 2 [ dest ] 475 // 476 // Result: *dest <- ex, return (old *dest) 477 // 478 // Note: win32 does not currently use this code 479 480 address generate_atomic_xchg() { 481 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); 482 address start = __ pc(); 483 484 __ push(rdx); 485 Address exchange(rsp, 2 * wordSize); 486 Address dest_addr(rsp, 3 * wordSize); 487 __ movl(rax, exchange); 488 __ movptr(rdx, dest_addr); 489 __ xchgl(rax, Address(rdx, 0)); 490 __ pop(rdx); 491 __ ret(0); 492 493 return start; 494 } 495 496 //---------------------------------------------------------------------------------------------------- 497 // Support for void verify_mxcsr() 498 // 499 // This routine is used with -Xcheck:jni to verify that native 500 // JNI code does not return to Java code without restoring the 501 // MXCSR register to our expected state. 502 503 504 address generate_verify_mxcsr() { 505 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); 506 address start = __ pc(); 507 508 const Address mxcsr_save(rsp, 0); 509 510 if (CheckJNICalls && UseSSE > 0 ) { 511 Label ok_ret; 512 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 513 __ push(rax); 514 __ subptr(rsp, wordSize); // allocate a temp location 515 __ stmxcsr(mxcsr_save); 516 __ movl(rax, mxcsr_save); 517 __ andl(rax, MXCSR_MASK); 518 __ cmp32(rax, mxcsr_std); 519 __ jcc(Assembler::equal, ok_ret); 520 521 __ warn("MXCSR changed by native JNI code."); 522 523 __ ldmxcsr(mxcsr_std); 524 525 __ bind(ok_ret); 526 __ addptr(rsp, wordSize); 527 __ pop(rax); 528 } 529 530 __ ret(0); 531 532 return start; 533 } 534 535 536 //--------------------------------------------------------------------------- 537 // Support for void verify_fpu_cntrl_wrd() 538 // 539 // This routine is used with -Xcheck:jni to verify that native 540 // JNI code does not return to Java code without restoring the 541 // FP control word to our expected state. 542 543 address generate_verify_fpu_cntrl_wrd() { 544 StubCodeMark mark(this, "StubRoutines", "verify_spcw"); 545 address start = __ pc(); 546 547 const Address fpu_cntrl_wrd_save(rsp, 0); 548 549 if (CheckJNICalls) { 550 Label ok_ret; 551 __ push(rax); 552 __ subptr(rsp, wordSize); // allocate a temp location 553 __ fnstcw(fpu_cntrl_wrd_save); 554 __ movl(rax, fpu_cntrl_wrd_save); 555 __ andl(rax, FPU_CNTRL_WRD_MASK); 556 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); 557 __ cmp32(rax, fpu_std); 558 __ jcc(Assembler::equal, ok_ret); 559 560 __ warn("Floating point control word changed by native JNI code."); 561 562 __ fldcw(fpu_std); 563 564 __ bind(ok_ret); 565 __ addptr(rsp, wordSize); 566 __ pop(rax); 567 } 568 569 __ ret(0); 570 571 return start; 572 } 573 574 //--------------------------------------------------------------------------- 575 // Wrapper for slow-case handling of double-to-integer conversion 576 // d2i or f2i fast case failed either because it is nan or because 577 // of under/overflow. 578 // Input: FPU TOS: float value 579 // Output: rax, (rdx): integer (long) result 580 581 address generate_d2i_wrapper(BasicType t, address fcn) { 582 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); 583 address start = __ pc(); 584 585 // Capture info about frame layout 586 enum layout { FPUState_off = 0, 587 rbp_off = FPUStateSizeInWords, 588 rdi_off, 589 rsi_off, 590 rcx_off, 591 rbx_off, 592 saved_argument_off, 593 saved_argument_off2, // 2nd half of double 594 framesize 595 }; 596 597 assert(FPUStateSizeInWords == 27, "update stack layout"); 598 599 // Save outgoing argument to stack across push_FPU_state() 600 __ subptr(rsp, wordSize * 2); 601 __ fstp_d(Address(rsp, 0)); 602 603 // Save CPU & FPU state 604 __ push(rbx); 605 __ push(rcx); 606 __ push(rsi); 607 __ push(rdi); 608 __ push(rbp); 609 __ push_FPU_state(); 610 611 // push_FPU_state() resets the FP top of stack 612 // Load original double into FP top of stack 613 __ fld_d(Address(rsp, saved_argument_off * wordSize)); 614 // Store double into stack as outgoing argument 615 __ subptr(rsp, wordSize*2); 616 __ fst_d(Address(rsp, 0)); 617 618 // Prepare FPU for doing math in C-land 619 __ empty_FPU_stack(); 620 // Call the C code to massage the double. Result in EAX 621 if (t == T_INT) 622 { BLOCK_COMMENT("SharedRuntime::d2i"); } 623 else if (t == T_LONG) 624 { BLOCK_COMMENT("SharedRuntime::d2l"); } 625 __ call_VM_leaf( fcn, 2 ); 626 627 // Restore CPU & FPU state 628 __ pop_FPU_state(); 629 __ pop(rbp); 630 __ pop(rdi); 631 __ pop(rsi); 632 __ pop(rcx); 633 __ pop(rbx); 634 __ addptr(rsp, wordSize * 2); 635 636 __ ret(0); 637 638 return start; 639 } 640 641 642 //--------------------------------------------------------------------------- 643 // The following routine generates a subroutine to throw an asynchronous 644 // UnknownError when an unsafe access gets a fault that could not be 645 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) 646 address generate_handler_for_unsafe_access() { 647 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); 648 address start = __ pc(); 649 650 __ push(0); // hole for return address-to-be 651 __ pusha(); // push registers 652 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); 653 BLOCK_COMMENT("call handle_unsafe_access"); 654 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); 655 __ movptr(next_pc, rax); // stuff next address 656 __ popa(); 657 __ ret(0); // jump to next address 658 659 return start; 660 } 661 662 663 //---------------------------------------------------------------------------------------------------- 664 // Non-destructive plausibility checks for oops 665 666 address generate_verify_oop() { 667 StubCodeMark mark(this, "StubRoutines", "verify_oop"); 668 address start = __ pc(); 669 670 // Incoming arguments on stack after saving rax,: 671 // 672 // [tos ]: saved rdx 673 // [tos + 1]: saved EFLAGS 674 // [tos + 2]: return address 675 // [tos + 3]: char* error message 676 // [tos + 4]: oop object to verify 677 // [tos + 5]: saved rax, - saved by caller and bashed 678 679 Label exit, error; 680 __ pushf(); 681 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); 682 __ push(rdx); // save rdx 683 // make sure object is 'reasonable' 684 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object 685 __ testptr(rax, rax); 686 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok 687 688 // Check if the oop is in the right area of memory 689 const int oop_mask = Universe::verify_oop_mask(); 690 const int oop_bits = Universe::verify_oop_bits(); 691 __ mov(rdx, rax); 692 __ andptr(rdx, oop_mask); 693 __ cmpptr(rdx, oop_bits); 694 __ jcc(Assembler::notZero, error); 695 696 // make sure klass is 'reasonable' 697 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass 698 __ testptr(rax, rax); 699 __ jcc(Assembler::zero, error); // if klass is NULL it is broken 700 701 // Check if the klass is in the right area of memory 702 const int klass_mask = Universe::verify_klass_mask(); 703 const int klass_bits = Universe::verify_klass_bits(); 704 __ mov(rdx, rax); 705 __ andptr(rdx, klass_mask); 706 __ cmpptr(rdx, klass_bits); 707 __ jcc(Assembler::notZero, error); 708 709 // make sure klass' klass is 'reasonable' 710 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass 711 __ testptr(rax, rax); 712 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken 713 714 __ mov(rdx, rax); 715 __ andptr(rdx, klass_mask); 716 __ cmpptr(rdx, klass_bits); 717 __ jcc(Assembler::notZero, error); // if klass not in right area 718 // of memory it is broken too. 719 720 // return if everything seems ok 721 __ bind(exit); 722 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 723 __ pop(rdx); // restore rdx 724 __ popf(); // restore EFLAGS 725 __ ret(3 * wordSize); // pop arguments 726 727 // handle errors 728 __ bind(error); 729 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 730 __ pop(rdx); // get saved rdx back 731 __ popf(); // get saved EFLAGS off stack -- will be ignored 732 __ pusha(); // push registers (eip = return address & msg are already pushed) 733 BLOCK_COMMENT("call MacroAssembler::debug"); 734 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); 735 __ popa(); 736 __ ret(3 * wordSize); // pop arguments 737 return start; 738 } 739 740 // 741 // Generate pre-barrier for array stores 742 // 743 // Input: 744 // start - starting address 745 // count - element count 746 void gen_write_ref_array_pre_barrier(Register start, Register count) { 747 assert_different_registers(start, count); 748 BarrierSet* bs = Universe::heap()->barrier_set(); 749 switch (bs->kind()) { 750 case BarrierSet::G1SATBCT: 751 case BarrierSet::G1SATBCTLogging: 752 { 753 __ pusha(); // push registers 754 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 755 start, count); 756 __ popa(); 757 } 758 break; 759 case BarrierSet::CardTableModRef: 760 case BarrierSet::CardTableExtension: 761 case BarrierSet::ModRef: 762 break; 763 default : 764 ShouldNotReachHere(); 765 766 } 767 } 768 769 770 // 771 // Generate a post-barrier for an array store 772 // 773 // start - starting address 774 // count - element count 775 // 776 // The two input registers are overwritten. 777 // 778 void gen_write_ref_array_post_barrier(Register start, Register count) { 779 BarrierSet* bs = Universe::heap()->barrier_set(); 780 assert_different_registers(start, count); 781 switch (bs->kind()) { 782 case BarrierSet::G1SATBCT: 783 case BarrierSet::G1SATBCTLogging: 784 { 785 __ pusha(); // push registers 786 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 787 start, count); 788 __ popa(); 789 } 790 break; 791 792 case BarrierSet::CardTableModRef: 793 case BarrierSet::CardTableExtension: 794 { 795 CardTableModRefBS* ct = (CardTableModRefBS*)bs; 796 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 797 798 Label L_loop; 799 const Register end = count; // elements count; end == start+count-1 800 assert_different_registers(start, end); 801 802 __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); 803 __ shrptr(start, CardTableModRefBS::card_shift); 804 __ shrptr(end, CardTableModRefBS::card_shift); 805 __ subptr(end, start); // end --> count 806 __ BIND(L_loop); 807 intptr_t disp = (intptr_t) ct->byte_map_base; 808 Address cardtable(start, count, Address::times_1, disp); 809 __ movb(cardtable, 0); 810 __ decrement(count); 811 __ jcc(Assembler::greaterEqual, L_loop); 812 } 813 break; 814 case BarrierSet::ModRef: 815 break; 816 default : 817 ShouldNotReachHere(); 818 819 } 820 } 821 822 823 // Copy 64 bytes chunks 824 // 825 // Inputs: 826 // from - source array address 827 // to_from - destination array address - from 828 // qword_count - 8-bytes element count, negative 829 // 830 void xmm_copy_forward(Register from, Register to_from, Register qword_count) { 831 assert( UseSSE >= 2, "supported cpu only" ); 832 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 833 // Copy 64-byte chunks 834 __ jmpb(L_copy_64_bytes); 835 __ align(OptoLoopAlignment); 836 __ BIND(L_copy_64_bytes_loop); 837 838 if(UseUnalignedLoadStores) { 839 __ movdqu(xmm0, Address(from, 0)); 840 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0); 841 __ movdqu(xmm1, Address(from, 16)); 842 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1); 843 __ movdqu(xmm2, Address(from, 32)); 844 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2); 845 __ movdqu(xmm3, Address(from, 48)); 846 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3); 847 848 } else { 849 __ movq(xmm0, Address(from, 0)); 850 __ movq(Address(from, to_from, Address::times_1, 0), xmm0); 851 __ movq(xmm1, Address(from, 8)); 852 __ movq(Address(from, to_from, Address::times_1, 8), xmm1); 853 __ movq(xmm2, Address(from, 16)); 854 __ movq(Address(from, to_from, Address::times_1, 16), xmm2); 855 __ movq(xmm3, Address(from, 24)); 856 __ movq(Address(from, to_from, Address::times_1, 24), xmm3); 857 __ movq(xmm4, Address(from, 32)); 858 __ movq(Address(from, to_from, Address::times_1, 32), xmm4); 859 __ movq(xmm5, Address(from, 40)); 860 __ movq(Address(from, to_from, Address::times_1, 40), xmm5); 861 __ movq(xmm6, Address(from, 48)); 862 __ movq(Address(from, to_from, Address::times_1, 48), xmm6); 863 __ movq(xmm7, Address(from, 56)); 864 __ movq(Address(from, to_from, Address::times_1, 56), xmm7); 865 } 866 867 __ addl(from, 64); 868 __ BIND(L_copy_64_bytes); 869 __ subl(qword_count, 8); 870 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 871 __ addl(qword_count, 8); 872 __ jccb(Assembler::zero, L_exit); 873 // 874 // length is too short, just copy qwords 875 // 876 __ BIND(L_copy_8_bytes); 877 __ movq(xmm0, Address(from, 0)); 878 __ movq(Address(from, to_from, Address::times_1), xmm0); 879 __ addl(from, 8); 880 __ decrement(qword_count); 881 __ jcc(Assembler::greater, L_copy_8_bytes); 882 __ BIND(L_exit); 883 } 884 885 // Copy 64 bytes chunks 886 // 887 // Inputs: 888 // from - source array address 889 // to_from - destination array address - from 890 // qword_count - 8-bytes element count, negative 891 // 892 void mmx_copy_forward(Register from, Register to_from, Register qword_count) { 893 assert( VM_Version::supports_mmx(), "supported cpu only" ); 894 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 895 // Copy 64-byte chunks 896 __ jmpb(L_copy_64_bytes); 897 __ align(OptoLoopAlignment); 898 __ BIND(L_copy_64_bytes_loop); 899 __ movq(mmx0, Address(from, 0)); 900 __ movq(mmx1, Address(from, 8)); 901 __ movq(mmx2, Address(from, 16)); 902 __ movq(Address(from, to_from, Address::times_1, 0), mmx0); 903 __ movq(mmx3, Address(from, 24)); 904 __ movq(Address(from, to_from, Address::times_1, 8), mmx1); 905 __ movq(mmx4, Address(from, 32)); 906 __ movq(Address(from, to_from, Address::times_1, 16), mmx2); 907 __ movq(mmx5, Address(from, 40)); 908 __ movq(Address(from, to_from, Address::times_1, 24), mmx3); 909 __ movq(mmx6, Address(from, 48)); 910 __ movq(Address(from, to_from, Address::times_1, 32), mmx4); 911 __ movq(mmx7, Address(from, 56)); 912 __ movq(Address(from, to_from, Address::times_1, 40), mmx5); 913 __ movq(Address(from, to_from, Address::times_1, 48), mmx6); 914 __ movq(Address(from, to_from, Address::times_1, 56), mmx7); 915 __ addptr(from, 64); 916 __ BIND(L_copy_64_bytes); 917 __ subl(qword_count, 8); 918 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 919 __ addl(qword_count, 8); 920 __ jccb(Assembler::zero, L_exit); 921 // 922 // length is too short, just copy qwords 923 // 924 __ BIND(L_copy_8_bytes); 925 __ movq(mmx0, Address(from, 0)); 926 __ movq(Address(from, to_from, Address::times_1), mmx0); 927 __ addptr(from, 8); 928 __ decrement(qword_count); 929 __ jcc(Assembler::greater, L_copy_8_bytes); 930 __ BIND(L_exit); 931 __ emms(); 932 } 933 934 address generate_disjoint_copy(BasicType t, bool aligned, 935 Address::ScaleFactor sf, 936 address* entry, const char *name) { 937 __ align(CodeEntryAlignment); 938 StubCodeMark mark(this, "StubRoutines", name); 939 address start = __ pc(); 940 941 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 942 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; 943 944 int shift = Address::times_ptr - sf; 945 946 const Register from = rsi; // source array address 947 const Register to = rdi; // destination array address 948 const Register count = rcx; // elements count 949 const Register to_from = to; // (to - from) 950 const Register saved_to = rdx; // saved destination array address 951 952 __ enter(); // required for proper stackwalking of RuntimeStub frame 953 __ push(rsi); 954 __ push(rdi); 955 __ movptr(from , Address(rsp, 12+ 4)); 956 __ movptr(to , Address(rsp, 12+ 8)); 957 __ movl(count, Address(rsp, 12+ 12)); 958 959 if (entry != NULL) { 960 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 961 BLOCK_COMMENT("Entry:"); 962 } 963 964 if (t == T_OBJECT) { 965 __ testl(count, count); 966 __ jcc(Assembler::zero, L_0_count); 967 gen_write_ref_array_pre_barrier(to, count); 968 __ mov(saved_to, to); // save 'to' 969 } 970 971 __ subptr(to, from); // to --> to_from 972 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 973 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 974 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { 975 // align source address at 4 bytes address boundary 976 if (t == T_BYTE) { 977 // One byte misalignment happens only for byte arrays 978 __ testl(from, 1); 979 __ jccb(Assembler::zero, L_skip_align1); 980 __ movb(rax, Address(from, 0)); 981 __ movb(Address(from, to_from, Address::times_1, 0), rax); 982 __ increment(from); 983 __ decrement(count); 984 __ BIND(L_skip_align1); 985 } 986 // Two bytes misalignment happens only for byte and short (char) arrays 987 __ testl(from, 2); 988 __ jccb(Assembler::zero, L_skip_align2); 989 __ movw(rax, Address(from, 0)); 990 __ movw(Address(from, to_from, Address::times_1, 0), rax); 991 __ addptr(from, 2); 992 __ subl(count, 1<<(shift-1)); 993 __ BIND(L_skip_align2); 994 } 995 if (!VM_Version::supports_mmx()) { 996 __ mov(rax, count); // save 'count' 997 __ shrl(count, shift); // bytes count 998 __ addptr(to_from, from);// restore 'to' 999 __ rep_mov(); 1000 __ subptr(to_from, from);// restore 'to_from' 1001 __ mov(count, rax); // restore 'count' 1002 __ jmpb(L_copy_2_bytes); // all dwords were copied 1003 } else { 1004 if (!UseUnalignedLoadStores) { 1005 // align to 8 bytes, we know we are 4 byte aligned to start 1006 __ testptr(from, 4); 1007 __ jccb(Assembler::zero, L_copy_64_bytes); 1008 __ movl(rax, Address(from, 0)); 1009 __ movl(Address(from, to_from, Address::times_1, 0), rax); 1010 __ addptr(from, 4); 1011 __ subl(count, 1<<shift); 1012 } 1013 __ BIND(L_copy_64_bytes); 1014 __ mov(rax, count); 1015 __ shrl(rax, shift+1); // 8 bytes chunk count 1016 // 1017 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop 1018 // 1019 if (UseXMMForArrayCopy) { 1020 xmm_copy_forward(from, to_from, rax); 1021 } else { 1022 mmx_copy_forward(from, to_from, rax); 1023 } 1024 } 1025 // copy tailing dword 1026 __ BIND(L_copy_4_bytes); 1027 __ testl(count, 1<<shift); 1028 __ jccb(Assembler::zero, L_copy_2_bytes); 1029 __ movl(rax, Address(from, 0)); 1030 __ movl(Address(from, to_from, Address::times_1, 0), rax); 1031 if (t == T_BYTE || t == T_SHORT) { 1032 __ addptr(from, 4); 1033 __ BIND(L_copy_2_bytes); 1034 // copy tailing word 1035 __ testl(count, 1<<(shift-1)); 1036 __ jccb(Assembler::zero, L_copy_byte); 1037 __ movw(rax, Address(from, 0)); 1038 __ movw(Address(from, to_from, Address::times_1, 0), rax); 1039 if (t == T_BYTE) { 1040 __ addptr(from, 2); 1041 __ BIND(L_copy_byte); 1042 // copy tailing byte 1043 __ testl(count, 1); 1044 __ jccb(Assembler::zero, L_exit); 1045 __ movb(rax, Address(from, 0)); 1046 __ movb(Address(from, to_from, Address::times_1, 0), rax); 1047 __ BIND(L_exit); 1048 } else { 1049 __ BIND(L_copy_byte); 1050 } 1051 } else { 1052 __ BIND(L_copy_2_bytes); 1053 } 1054 1055 if (t == T_OBJECT) { 1056 __ movl(count, Address(rsp, 12+12)); // reread 'count' 1057 __ mov(to, saved_to); // restore 'to' 1058 gen_write_ref_array_post_barrier(to, count); 1059 __ BIND(L_0_count); 1060 } 1061 inc_copy_counter_np(t); 1062 __ pop(rdi); 1063 __ pop(rsi); 1064 __ leave(); // required for proper stackwalking of RuntimeStub frame 1065 __ xorptr(rax, rax); // return 0 1066 __ ret(0); 1067 return start; 1068 } 1069 1070 1071 address generate_fill(BasicType t, bool aligned, const char *name) { 1072 __ align(CodeEntryAlignment); 1073 StubCodeMark mark(this, "StubRoutines", name); 1074 address start = __ pc(); 1075 1076 BLOCK_COMMENT("Entry:"); 1077 1078 const Register to = rdi; // source array address 1079 const Register value = rdx; // value 1080 const Register count = rsi; // elements count 1081 1082 __ enter(); // required for proper stackwalking of RuntimeStub frame 1083 __ push(rsi); 1084 __ push(rdi); 1085 __ movptr(to , Address(rsp, 12+ 4)); 1086 __ movl(value, Address(rsp, 12+ 8)); 1087 __ movl(count, Address(rsp, 12+ 12)); 1088 1089 __ generate_fill(t, aligned, to, value, count, rax, xmm0); 1090 1091 __ pop(rdi); 1092 __ pop(rsi); 1093 __ leave(); // required for proper stackwalking of RuntimeStub frame 1094 __ ret(0); 1095 return start; 1096 } 1097 1098 address generate_conjoint_copy(BasicType t, bool aligned, 1099 Address::ScaleFactor sf, 1100 address nooverlap_target, 1101 address* entry, const char *name) { 1102 __ align(CodeEntryAlignment); 1103 StubCodeMark mark(this, "StubRoutines", name); 1104 address start = __ pc(); 1105 1106 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 1107 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; 1108 1109 int shift = Address::times_ptr - sf; 1110 1111 const Register src = rax; // source array address 1112 const Register dst = rdx; // destination array address 1113 const Register from = rsi; // source array address 1114 const Register to = rdi; // destination array address 1115 const Register count = rcx; // elements count 1116 const Register end = rax; // array end address 1117 1118 __ enter(); // required for proper stackwalking of RuntimeStub frame 1119 __ push(rsi); 1120 __ push(rdi); 1121 __ movptr(src , Address(rsp, 12+ 4)); // from 1122 __ movptr(dst , Address(rsp, 12+ 8)); // to 1123 __ movl2ptr(count, Address(rsp, 12+12)); // count 1124 1125 if (entry != NULL) { 1126 *entry = __ pc(); // Entry point from generic arraycopy stub. 1127 BLOCK_COMMENT("Entry:"); 1128 } 1129 1130 // nooverlap_target expects arguments in rsi and rdi. 1131 __ mov(from, src); 1132 __ mov(to , dst); 1133 1134 // arrays overlap test: dispatch to disjoint stub if necessary. 1135 RuntimeAddress nooverlap(nooverlap_target); 1136 __ cmpptr(dst, src); 1137 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size 1138 __ jump_cc(Assembler::belowEqual, nooverlap); 1139 __ cmpptr(dst, end); 1140 __ jump_cc(Assembler::aboveEqual, nooverlap); 1141 1142 if (t == T_OBJECT) { 1143 __ testl(count, count); 1144 __ jcc(Assembler::zero, L_0_count); 1145 gen_write_ref_array_pre_barrier(dst, count); 1146 } 1147 1148 // copy from high to low 1149 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1150 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 1151 if (t == T_BYTE || t == T_SHORT) { 1152 // Align the end of destination array at 4 bytes address boundary 1153 __ lea(end, Address(dst, count, sf, 0)); 1154 if (t == T_BYTE) { 1155 // One byte misalignment happens only for byte arrays 1156 __ testl(end, 1); 1157 __ jccb(Assembler::zero, L_skip_align1); 1158 __ decrement(count); 1159 __ movb(rdx, Address(from, count, sf, 0)); 1160 __ movb(Address(to, count, sf, 0), rdx); 1161 __ BIND(L_skip_align1); 1162 } 1163 // Two bytes misalignment happens only for byte and short (char) arrays 1164 __ testl(end, 2); 1165 __ jccb(Assembler::zero, L_skip_align2); 1166 __ subptr(count, 1<<(shift-1)); 1167 __ movw(rdx, Address(from, count, sf, 0)); 1168 __ movw(Address(to, count, sf, 0), rdx); 1169 __ BIND(L_skip_align2); 1170 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1171 __ jcc(Assembler::below, L_copy_4_bytes); 1172 } 1173 1174 if (!VM_Version::supports_mmx()) { 1175 __ std(); 1176 __ mov(rax, count); // Save 'count' 1177 __ mov(rdx, to); // Save 'to' 1178 __ lea(rsi, Address(from, count, sf, -4)); 1179 __ lea(rdi, Address(to , count, sf, -4)); 1180 __ shrptr(count, shift); // bytes count 1181 __ rep_mov(); 1182 __ cld(); 1183 __ mov(count, rax); // restore 'count' 1184 __ andl(count, (1<<shift)-1); // mask the number of rest elements 1185 __ movptr(from, Address(rsp, 12+4)); // reread 'from' 1186 __ mov(to, rdx); // restore 'to' 1187 __ jmpb(L_copy_2_bytes); // all dword were copied 1188 } else { 1189 // Align to 8 bytes the end of array. It is aligned to 4 bytes already. 1190 __ testptr(end, 4); 1191 __ jccb(Assembler::zero, L_copy_8_bytes); 1192 __ subl(count, 1<<shift); 1193 __ movl(rdx, Address(from, count, sf, 0)); 1194 __ movl(Address(to, count, sf, 0), rdx); 1195 __ jmpb(L_copy_8_bytes); 1196 1197 __ align(OptoLoopAlignment); 1198 // Move 8 bytes 1199 __ BIND(L_copy_8_bytes_loop); 1200 if (UseXMMForArrayCopy) { 1201 __ movq(xmm0, Address(from, count, sf, 0)); 1202 __ movq(Address(to, count, sf, 0), xmm0); 1203 } else { 1204 __ movq(mmx0, Address(from, count, sf, 0)); 1205 __ movq(Address(to, count, sf, 0), mmx0); 1206 } 1207 __ BIND(L_copy_8_bytes); 1208 __ subl(count, 2<<shift); 1209 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1210 __ addl(count, 2<<shift); 1211 if (!UseXMMForArrayCopy) { 1212 __ emms(); 1213 } 1214 } 1215 __ BIND(L_copy_4_bytes); 1216 // copy prefix qword 1217 __ testl(count, 1<<shift); 1218 __ jccb(Assembler::zero, L_copy_2_bytes); 1219 __ movl(rdx, Address(from, count, sf, -4)); 1220 __ movl(Address(to, count, sf, -4), rdx); 1221 1222 if (t == T_BYTE || t == T_SHORT) { 1223 __ subl(count, (1<<shift)); 1224 __ BIND(L_copy_2_bytes); 1225 // copy prefix dword 1226 __ testl(count, 1<<(shift-1)); 1227 __ jccb(Assembler::zero, L_copy_byte); 1228 __ movw(rdx, Address(from, count, sf, -2)); 1229 __ movw(Address(to, count, sf, -2), rdx); 1230 if (t == T_BYTE) { 1231 __ subl(count, 1<<(shift-1)); 1232 __ BIND(L_copy_byte); 1233 // copy prefix byte 1234 __ testl(count, 1); 1235 __ jccb(Assembler::zero, L_exit); 1236 __ movb(rdx, Address(from, 0)); 1237 __ movb(Address(to, 0), rdx); 1238 __ BIND(L_exit); 1239 } else { 1240 __ BIND(L_copy_byte); 1241 } 1242 } else { 1243 __ BIND(L_copy_2_bytes); 1244 } 1245 if (t == T_OBJECT) { 1246 __ movl2ptr(count, Address(rsp, 12+12)); // reread count 1247 gen_write_ref_array_post_barrier(to, count); 1248 __ BIND(L_0_count); 1249 } 1250 inc_copy_counter_np(t); 1251 __ pop(rdi); 1252 __ pop(rsi); 1253 __ leave(); // required for proper stackwalking of RuntimeStub frame 1254 __ xorptr(rax, rax); // return 0 1255 __ ret(0); 1256 return start; 1257 } 1258 1259 1260 address generate_disjoint_long_copy(address* entry, const char *name) { 1261 __ align(CodeEntryAlignment); 1262 StubCodeMark mark(this, "StubRoutines", name); 1263 address start = __ pc(); 1264 1265 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1266 const Register from = rax; // source array address 1267 const Register to = rdx; // destination array address 1268 const Register count = rcx; // elements count 1269 const Register to_from = rdx; // (to - from) 1270 1271 __ enter(); // required for proper stackwalking of RuntimeStub frame 1272 __ movptr(from , Address(rsp, 8+0)); // from 1273 __ movptr(to , Address(rsp, 8+4)); // to 1274 __ movl2ptr(count, Address(rsp, 8+8)); // count 1275 1276 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 1277 BLOCK_COMMENT("Entry:"); 1278 1279 __ subptr(to, from); // to --> to_from 1280 if (VM_Version::supports_mmx()) { 1281 if (UseXMMForArrayCopy) { 1282 xmm_copy_forward(from, to_from, count); 1283 } else { 1284 mmx_copy_forward(from, to_from, count); 1285 } 1286 } else { 1287 __ jmpb(L_copy_8_bytes); 1288 __ align(OptoLoopAlignment); 1289 __ BIND(L_copy_8_bytes_loop); 1290 __ fild_d(Address(from, 0)); 1291 __ fistp_d(Address(from, to_from, Address::times_1)); 1292 __ addptr(from, 8); 1293 __ BIND(L_copy_8_bytes); 1294 __ decrement(count); 1295 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1296 } 1297 inc_copy_counter_np(T_LONG); 1298 __ leave(); // required for proper stackwalking of RuntimeStub frame 1299 __ xorptr(rax, rax); // return 0 1300 __ ret(0); 1301 return start; 1302 } 1303 1304 address generate_conjoint_long_copy(address nooverlap_target, 1305 address* entry, const char *name) { 1306 __ align(CodeEntryAlignment); 1307 StubCodeMark mark(this, "StubRoutines", name); 1308 address start = __ pc(); 1309 1310 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1311 const Register from = rax; // source array address 1312 const Register to = rdx; // destination array address 1313 const Register count = rcx; // elements count 1314 const Register end_from = rax; // source array end address 1315 1316 __ enter(); // required for proper stackwalking of RuntimeStub frame 1317 __ movptr(from , Address(rsp, 8+0)); // from 1318 __ movptr(to , Address(rsp, 8+4)); // to 1319 __ movl2ptr(count, Address(rsp, 8+8)); // count 1320 1321 *entry = __ pc(); // Entry point from generic arraycopy stub. 1322 BLOCK_COMMENT("Entry:"); 1323 1324 // arrays overlap test 1325 __ cmpptr(to, from); 1326 RuntimeAddress nooverlap(nooverlap_target); 1327 __ jump_cc(Assembler::belowEqual, nooverlap); 1328 __ lea(end_from, Address(from, count, Address::times_8, 0)); 1329 __ cmpptr(to, end_from); 1330 __ movptr(from, Address(rsp, 8)); // from 1331 __ jump_cc(Assembler::aboveEqual, nooverlap); 1332 1333 __ jmpb(L_copy_8_bytes); 1334 1335 __ align(OptoLoopAlignment); 1336 __ BIND(L_copy_8_bytes_loop); 1337 if (VM_Version::supports_mmx()) { 1338 if (UseXMMForArrayCopy) { 1339 __ movq(xmm0, Address(from, count, Address::times_8)); 1340 __ movq(Address(to, count, Address::times_8), xmm0); 1341 } else { 1342 __ movq(mmx0, Address(from, count, Address::times_8)); 1343 __ movq(Address(to, count, Address::times_8), mmx0); 1344 } 1345 } else { 1346 __ fild_d(Address(from, count, Address::times_8)); 1347 __ fistp_d(Address(to, count, Address::times_8)); 1348 } 1349 __ BIND(L_copy_8_bytes); 1350 __ decrement(count); 1351 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1352 1353 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) { 1354 __ emms(); 1355 } 1356 inc_copy_counter_np(T_LONG); 1357 __ leave(); // required for proper stackwalking of RuntimeStub frame 1358 __ xorptr(rax, rax); // return 0 1359 __ ret(0); 1360 return start; 1361 } 1362 1363 1364 // Helper for generating a dynamic type check. 1365 // The sub_klass must be one of {rbx, rdx, rsi}. 1366 // The temp is killed. 1367 void generate_type_check(Register sub_klass, 1368 Address& super_check_offset_addr, 1369 Address& super_klass_addr, 1370 Register temp, 1371 Label* L_success, Label* L_failure) { 1372 BLOCK_COMMENT("type_check:"); 1373 1374 Label L_fallthrough; 1375 #define LOCAL_JCC(assembler_con, label_ptr) \ 1376 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ 1377 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ 1378 1379 // The following is a strange variation of the fast path which requires 1380 // one less register, because needed values are on the argument stack. 1381 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, 1382 // L_success, L_failure, NULL); 1383 assert_different_registers(sub_klass, temp); 1384 1385 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 1386 Klass::secondary_super_cache_offset_in_bytes()); 1387 1388 // if the pointers are equal, we are done (e.g., String[] elements) 1389 __ cmpptr(sub_klass, super_klass_addr); 1390 LOCAL_JCC(Assembler::equal, L_success); 1391 1392 // check the supertype display: 1393 __ movl2ptr(temp, super_check_offset_addr); 1394 Address super_check_addr(sub_klass, temp, Address::times_1, 0); 1395 __ movptr(temp, super_check_addr); // load displayed supertype 1396 __ cmpptr(temp, super_klass_addr); // test the super type 1397 LOCAL_JCC(Assembler::equal, L_success); 1398 1399 // if it was a primary super, we can just fail immediately 1400 __ cmpl(super_check_offset_addr, sc_offset); 1401 LOCAL_JCC(Assembler::notEqual, L_failure); 1402 1403 // The repne_scan instruction uses fixed registers, which will get spilled. 1404 // We happen to know this works best when super_klass is in rax. 1405 Register super_klass = temp; 1406 __ movptr(super_klass, super_klass_addr); 1407 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, 1408 L_success, L_failure); 1409 1410 __ bind(L_fallthrough); 1411 1412 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } 1413 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } 1414 1415 #undef LOCAL_JCC 1416 } 1417 1418 // 1419 // Generate checkcasting array copy stub 1420 // 1421 // Input: 1422 // 4(rsp) - source array address 1423 // 8(rsp) - destination array address 1424 // 12(rsp) - element count, can be zero 1425 // 16(rsp) - size_t ckoff (super_check_offset) 1426 // 20(rsp) - oop ckval (super_klass) 1427 // 1428 // Output: 1429 // rax, == 0 - success 1430 // rax, == -1^K - failure, where K is partial transfer count 1431 // 1432 address generate_checkcast_copy(const char *name, address* entry) { 1433 __ align(CodeEntryAlignment); 1434 StubCodeMark mark(this, "StubRoutines", name); 1435 address start = __ pc(); 1436 1437 Label L_load_element, L_store_element, L_do_card_marks, L_done; 1438 1439 // register use: 1440 // rax, rdx, rcx -- loop control (end_from, end_to, count) 1441 // rdi, rsi -- element access (oop, klass) 1442 // rbx, -- temp 1443 const Register from = rax; // source array address 1444 const Register to = rdx; // destination array address 1445 const Register length = rcx; // elements count 1446 const Register elem = rdi; // each oop copied 1447 const Register elem_klass = rsi; // each elem._klass (sub_klass) 1448 const Register temp = rbx; // lone remaining temp 1449 1450 __ enter(); // required for proper stackwalking of RuntimeStub frame 1451 1452 __ push(rsi); 1453 __ push(rdi); 1454 __ push(rbx); 1455 1456 Address from_arg(rsp, 16+ 4); // from 1457 Address to_arg(rsp, 16+ 8); // to 1458 Address length_arg(rsp, 16+12); // elements count 1459 Address ckoff_arg(rsp, 16+16); // super_check_offset 1460 Address ckval_arg(rsp, 16+20); // super_klass 1461 1462 // Load up: 1463 __ movptr(from, from_arg); 1464 __ movptr(to, to_arg); 1465 __ movl2ptr(length, length_arg); 1466 1467 if (entry != NULL) { 1468 *entry = __ pc(); // Entry point from generic arraycopy stub. 1469 BLOCK_COMMENT("Entry:"); 1470 } 1471 1472 //--------------------------------------------------------------- 1473 // Assembler stub will be used for this call to arraycopy 1474 // if the two arrays are subtypes of Object[] but the 1475 // destination array type is not equal to or a supertype 1476 // of the source type. Each element must be separately 1477 // checked. 1478 1479 // Loop-invariant addresses. They are exclusive end pointers. 1480 Address end_from_addr(from, length, Address::times_ptr, 0); 1481 Address end_to_addr(to, length, Address::times_ptr, 0); 1482 1483 Register end_from = from; // re-use 1484 Register end_to = to; // re-use 1485 Register count = length; // re-use 1486 1487 // Loop-variant addresses. They assume post-incremented count < 0. 1488 Address from_element_addr(end_from, count, Address::times_ptr, 0); 1489 Address to_element_addr(end_to, count, Address::times_ptr, 0); 1490 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); 1491 1492 // Copy from low to high addresses, indexed from the end of each array. 1493 gen_write_ref_array_pre_barrier(to, count); 1494 __ lea(end_from, end_from_addr); 1495 __ lea(end_to, end_to_addr); 1496 assert(length == count, ""); // else fix next line: 1497 __ negptr(count); // negate and test the length 1498 __ jccb(Assembler::notZero, L_load_element); 1499 1500 // Empty array: Nothing to do. 1501 __ xorptr(rax, rax); // return 0 on (trivial) success 1502 __ jmp(L_done); 1503 1504 // ======== begin loop ======== 1505 // (Loop is rotated; its entry is L_load_element.) 1506 // Loop control: 1507 // for (count = -count; count != 0; count++) 1508 // Base pointers src, dst are biased by 8*count,to last element. 1509 __ align(OptoLoopAlignment); 1510 1511 __ BIND(L_store_element); 1512 __ movptr(to_element_addr, elem); // store the oop 1513 __ increment(count); // increment the count toward zero 1514 __ jccb(Assembler::zero, L_do_card_marks); 1515 1516 // ======== loop entry is here ======== 1517 __ BIND(L_load_element); 1518 __ movptr(elem, from_element_addr); // load the oop 1519 __ testptr(elem, elem); 1520 __ jccb(Assembler::zero, L_store_element); 1521 1522 // (Could do a trick here: Remember last successful non-null 1523 // element stored and make a quick oop equality check on it.) 1524 1525 __ movptr(elem_klass, elem_klass_addr); // query the object klass 1526 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, 1527 &L_store_element, NULL); 1528 // (On fall-through, we have failed the element type check.) 1529 // ======== end loop ======== 1530 1531 // It was a real error; we must depend on the caller to finish the job. 1532 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. 1533 // Emit GC store barriers for the oops we have copied (length_arg + count), 1534 // and report their number to the caller. 1535 __ addl(count, length_arg); // transfers = (length - remaining) 1536 __ movl2ptr(rax, count); // save the value 1537 __ notptr(rax); // report (-1^K) to caller 1538 __ movptr(to, to_arg); // reload 1539 assert_different_registers(to, count, rax); 1540 gen_write_ref_array_post_barrier(to, count); 1541 __ jmpb(L_done); 1542 1543 // Come here on success only. 1544 __ BIND(L_do_card_marks); 1545 __ movl2ptr(count, length_arg); 1546 __ movptr(to, to_arg); // reload 1547 gen_write_ref_array_post_barrier(to, count); 1548 __ xorptr(rax, rax); // return 0 on success 1549 1550 // Common exit point (success or failure). 1551 __ BIND(L_done); 1552 __ pop(rbx); 1553 __ pop(rdi); 1554 __ pop(rsi); 1555 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); 1556 __ leave(); // required for proper stackwalking of RuntimeStub frame 1557 __ ret(0); 1558 1559 return start; 1560 } 1561 1562 // 1563 // Generate 'unsafe' array copy stub 1564 // Though just as safe as the other stubs, it takes an unscaled 1565 // size_t argument instead of an element count. 1566 // 1567 // Input: 1568 // 4(rsp) - source array address 1569 // 8(rsp) - destination array address 1570 // 12(rsp) - byte count, can be zero 1571 // 1572 // Output: 1573 // rax, == 0 - success 1574 // rax, == -1 - need to call System.arraycopy 1575 // 1576 // Examines the alignment of the operands and dispatches 1577 // to a long, int, short, or byte copy loop. 1578 // 1579 address generate_unsafe_copy(const char *name, 1580 address byte_copy_entry, 1581 address short_copy_entry, 1582 address int_copy_entry, 1583 address long_copy_entry) { 1584 1585 Label L_long_aligned, L_int_aligned, L_short_aligned; 1586 1587 __ align(CodeEntryAlignment); 1588 StubCodeMark mark(this, "StubRoutines", name); 1589 address start = __ pc(); 1590 1591 const Register from = rax; // source array address 1592 const Register to = rdx; // destination array address 1593 const Register count = rcx; // elements count 1594 1595 __ enter(); // required for proper stackwalking of RuntimeStub frame 1596 __ push(rsi); 1597 __ push(rdi); 1598 Address from_arg(rsp, 12+ 4); // from 1599 Address to_arg(rsp, 12+ 8); // to 1600 Address count_arg(rsp, 12+12); // byte count 1601 1602 // Load up: 1603 __ movptr(from , from_arg); 1604 __ movptr(to , to_arg); 1605 __ movl2ptr(count, count_arg); 1606 1607 // bump this on entry, not on exit: 1608 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); 1609 1610 const Register bits = rsi; 1611 __ mov(bits, from); 1612 __ orptr(bits, to); 1613 __ orptr(bits, count); 1614 1615 __ testl(bits, BytesPerLong-1); 1616 __ jccb(Assembler::zero, L_long_aligned); 1617 1618 __ testl(bits, BytesPerInt-1); 1619 __ jccb(Assembler::zero, L_int_aligned); 1620 1621 __ testl(bits, BytesPerShort-1); 1622 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); 1623 1624 __ BIND(L_short_aligned); 1625 __ shrptr(count, LogBytesPerShort); // size => short_count 1626 __ movl(count_arg, count); // update 'count' 1627 __ jump(RuntimeAddress(short_copy_entry)); 1628 1629 __ BIND(L_int_aligned); 1630 __ shrptr(count, LogBytesPerInt); // size => int_count 1631 __ movl(count_arg, count); // update 'count' 1632 __ jump(RuntimeAddress(int_copy_entry)); 1633 1634 __ BIND(L_long_aligned); 1635 __ shrptr(count, LogBytesPerLong); // size => qword_count 1636 __ movl(count_arg, count); // update 'count' 1637 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1638 __ pop(rsi); 1639 __ jump(RuntimeAddress(long_copy_entry)); 1640 1641 return start; 1642 } 1643 1644 1645 // Perform range checks on the proposed arraycopy. 1646 // Smashes src_pos and dst_pos. (Uses them up for temps.) 1647 void arraycopy_range_checks(Register src, 1648 Register src_pos, 1649 Register dst, 1650 Register dst_pos, 1651 Address& length, 1652 Label& L_failed) { 1653 BLOCK_COMMENT("arraycopy_range_checks:"); 1654 const Register src_end = src_pos; // source array end position 1655 const Register dst_end = dst_pos; // destination array end position 1656 __ addl(src_end, length); // src_pos + length 1657 __ addl(dst_end, length); // dst_pos + length 1658 1659 // if (src_pos + length > arrayOop(src)->length() ) FAIL; 1660 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); 1661 __ jcc(Assembler::above, L_failed); 1662 1663 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; 1664 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); 1665 __ jcc(Assembler::above, L_failed); 1666 1667 BLOCK_COMMENT("arraycopy_range_checks done"); 1668 } 1669 1670 1671 // 1672 // Generate generic array copy stubs 1673 // 1674 // Input: 1675 // 4(rsp) - src oop 1676 // 8(rsp) - src_pos 1677 // 12(rsp) - dst oop 1678 // 16(rsp) - dst_pos 1679 // 20(rsp) - element count 1680 // 1681 // Output: 1682 // rax, == 0 - success 1683 // rax, == -1^K - failure, where K is partial transfer count 1684 // 1685 address generate_generic_copy(const char *name, 1686 address entry_jbyte_arraycopy, 1687 address entry_jshort_arraycopy, 1688 address entry_jint_arraycopy, 1689 address entry_oop_arraycopy, 1690 address entry_jlong_arraycopy, 1691 address entry_checkcast_arraycopy) { 1692 Label L_failed, L_failed_0, L_objArray; 1693 1694 { int modulus = CodeEntryAlignment; 1695 int target = modulus - 5; // 5 = sizeof jmp(L_failed) 1696 int advance = target - (__ offset() % modulus); 1697 if (advance < 0) advance += modulus; 1698 if (advance > 0) __ nop(advance); 1699 } 1700 StubCodeMark mark(this, "StubRoutines", name); 1701 1702 // Short-hop target to L_failed. Makes for denser prologue code. 1703 __ BIND(L_failed_0); 1704 __ jmp(L_failed); 1705 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); 1706 1707 __ align(CodeEntryAlignment); 1708 address start = __ pc(); 1709 1710 __ enter(); // required for proper stackwalking of RuntimeStub frame 1711 __ push(rsi); 1712 __ push(rdi); 1713 1714 // bump this on entry, not on exit: 1715 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); 1716 1717 // Input values 1718 Address SRC (rsp, 12+ 4); 1719 Address SRC_POS (rsp, 12+ 8); 1720 Address DST (rsp, 12+12); 1721 Address DST_POS (rsp, 12+16); 1722 Address LENGTH (rsp, 12+20); 1723 1724 //----------------------------------------------------------------------- 1725 // Assembler stub will be used for this call to arraycopy 1726 // if the following conditions are met: 1727 // 1728 // (1) src and dst must not be null. 1729 // (2) src_pos must not be negative. 1730 // (3) dst_pos must not be negative. 1731 // (4) length must not be negative. 1732 // (5) src klass and dst klass should be the same and not NULL. 1733 // (6) src and dst should be arrays. 1734 // (7) src_pos + length must not exceed length of src. 1735 // (8) dst_pos + length must not exceed length of dst. 1736 // 1737 1738 const Register src = rax; // source array oop 1739 const Register src_pos = rsi; 1740 const Register dst = rdx; // destination array oop 1741 const Register dst_pos = rdi; 1742 const Register length = rcx; // transfer count 1743 1744 // if (src == NULL) return -1; 1745 __ movptr(src, SRC); // src oop 1746 __ testptr(src, src); 1747 __ jccb(Assembler::zero, L_failed_0); 1748 1749 // if (src_pos < 0) return -1; 1750 __ movl2ptr(src_pos, SRC_POS); // src_pos 1751 __ testl(src_pos, src_pos); 1752 __ jccb(Assembler::negative, L_failed_0); 1753 1754 // if (dst == NULL) return -1; 1755 __ movptr(dst, DST); // dst oop 1756 __ testptr(dst, dst); 1757 __ jccb(Assembler::zero, L_failed_0); 1758 1759 // if (dst_pos < 0) return -1; 1760 __ movl2ptr(dst_pos, DST_POS); // dst_pos 1761 __ testl(dst_pos, dst_pos); 1762 __ jccb(Assembler::negative, L_failed_0); 1763 1764 // if (length < 0) return -1; 1765 __ movl2ptr(length, LENGTH); // length 1766 __ testl(length, length); 1767 __ jccb(Assembler::negative, L_failed_0); 1768 1769 // if (src->klass() == NULL) return -1; 1770 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); 1771 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); 1772 const Register rcx_src_klass = rcx; // array klass 1773 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); 1774 1775 #ifdef ASSERT 1776 // assert(src->klass() != NULL); 1777 BLOCK_COMMENT("assert klasses not null"); 1778 { Label L1, L2; 1779 __ testptr(rcx_src_klass, rcx_src_klass); 1780 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL 1781 __ bind(L1); 1782 __ stop("broken null klass"); 1783 __ bind(L2); 1784 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); 1785 __ jccb(Assembler::equal, L1); // this would be broken also 1786 BLOCK_COMMENT("assert done"); 1787 } 1788 #endif //ASSERT 1789 1790 // Load layout helper (32-bits) 1791 // 1792 // |array_tag| | header_size | element_type | |log2_element_size| 1793 // 32 30 24 16 8 2 0 1794 // 1795 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 1796 // 1797 1798 int lh_offset = klassOopDesc::header_size() * HeapWordSize + 1799 Klass::layout_helper_offset_in_bytes(); 1800 Address src_klass_lh_addr(rcx_src_klass, lh_offset); 1801 1802 // Handle objArrays completely differently... 1803 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); 1804 __ cmpl(src_klass_lh_addr, objArray_lh); 1805 __ jcc(Assembler::equal, L_objArray); 1806 1807 // if (src->klass() != dst->klass()) return -1; 1808 __ cmpptr(rcx_src_klass, dst_klass_addr); 1809 __ jccb(Assembler::notEqual, L_failed_0); 1810 1811 const Register rcx_lh = rcx; // layout helper 1812 assert(rcx_lh == rcx_src_klass, "known alias"); 1813 __ movl(rcx_lh, src_klass_lh_addr); 1814 1815 // if (!src->is_Array()) return -1; 1816 __ cmpl(rcx_lh, Klass::_lh_neutral_value); 1817 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp 1818 1819 // At this point, it is known to be a typeArray (array_tag 0x3). 1820 #ifdef ASSERT 1821 { Label L; 1822 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); 1823 __ jcc(Assembler::greaterEqual, L); // signed cmp 1824 __ stop("must be a primitive array"); 1825 __ bind(L); 1826 } 1827 #endif 1828 1829 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); 1830 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1831 1832 // typeArrayKlass 1833 // 1834 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); 1835 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); 1836 // 1837 const Register rsi_offset = rsi; // array offset 1838 const Register src_array = src; // src array offset 1839 const Register dst_array = dst; // dst array offset 1840 const Register rdi_elsize = rdi; // log2 element size 1841 1842 __ mov(rsi_offset, rcx_lh); 1843 __ shrptr(rsi_offset, Klass::_lh_header_size_shift); 1844 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset 1845 __ addptr(src_array, rsi_offset); // src array offset 1846 __ addptr(dst_array, rsi_offset); // dst array offset 1847 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize 1848 1849 // next registers should be set before the jump to corresponding stub 1850 const Register from = src; // source array address 1851 const Register to = dst; // destination array address 1852 const Register count = rcx; // elements count 1853 // some of them should be duplicated on stack 1854 #define FROM Address(rsp, 12+ 4) 1855 #define TO Address(rsp, 12+ 8) // Not used now 1856 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy 1857 1858 BLOCK_COMMENT("scale indexes to element size"); 1859 __ movl2ptr(rsi, SRC_POS); // src_pos 1860 __ shlptr(rsi); // src_pos << rcx (log2 elsize) 1861 assert(src_array == from, ""); 1862 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize 1863 __ movl2ptr(rdi, DST_POS); // dst_pos 1864 __ shlptr(rdi); // dst_pos << rcx (log2 elsize) 1865 assert(dst_array == to, ""); 1866 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize 1867 __ movptr(FROM, from); // src_addr 1868 __ mov(rdi_elsize, rcx_lh); // log2 elsize 1869 __ movl2ptr(count, LENGTH); // elements count 1870 1871 BLOCK_COMMENT("choose copy loop based on element size"); 1872 __ cmpl(rdi_elsize, 0); 1873 1874 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); 1875 __ cmpl(rdi_elsize, LogBytesPerShort); 1876 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); 1877 __ cmpl(rdi_elsize, LogBytesPerInt); 1878 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); 1879 #ifdef ASSERT 1880 __ cmpl(rdi_elsize, LogBytesPerLong); 1881 __ jccb(Assembler::notEqual, L_failed); 1882 #endif 1883 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1884 __ pop(rsi); 1885 __ jump(RuntimeAddress(entry_jlong_arraycopy)); 1886 1887 __ BIND(L_failed); 1888 __ xorptr(rax, rax); 1889 __ notptr(rax); // return -1 1890 __ pop(rdi); 1891 __ pop(rsi); 1892 __ leave(); // required for proper stackwalking of RuntimeStub frame 1893 __ ret(0); 1894 1895 // objArrayKlass 1896 __ BIND(L_objArray); 1897 // live at this point: rcx_src_klass, src[_pos], dst[_pos] 1898 1899 Label L_plain_copy, L_checkcast_copy; 1900 // test array classes for subtyping 1901 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality 1902 __ jccb(Assembler::notEqual, L_checkcast_copy); 1903 1904 // Identically typed arrays can be copied without element-wise checks. 1905 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); 1906 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1907 1908 __ BIND(L_plain_copy); 1909 __ movl2ptr(count, LENGTH); // elements count 1910 __ movl2ptr(src_pos, SRC_POS); // reload src_pos 1911 __ lea(from, Address(src, src_pos, Address::times_ptr, 1912 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr 1913 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos 1914 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1915 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr 1916 __ movptr(FROM, from); // src_addr 1917 __ movptr(TO, to); // dst_addr 1918 __ movl(COUNT, count); // count 1919 __ jump(RuntimeAddress(entry_oop_arraycopy)); 1920 1921 __ BIND(L_checkcast_copy); 1922 // live at this point: rcx_src_klass, dst[_pos], src[_pos] 1923 { 1924 // Handy offsets: 1925 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + 1926 objArrayKlass::element_klass_offset_in_bytes()); 1927 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 1928 Klass::super_check_offset_offset_in_bytes()); 1929 1930 Register rsi_dst_klass = rsi; 1931 Register rdi_temp = rdi; 1932 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); 1933 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); 1934 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); 1935 1936 // Before looking at dst.length, make sure dst is also an objArray. 1937 __ movptr(rsi_dst_klass, dst_klass_addr); 1938 __ cmpl(dst_klass_lh_addr, objArray_lh); 1939 __ jccb(Assembler::notEqual, L_failed); 1940 1941 // It is safe to examine both src.length and dst.length. 1942 __ movl2ptr(src_pos, SRC_POS); // reload rsi 1943 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1944 // (Now src_pos and dst_pos are killed, but not src and dst.) 1945 1946 // We'll need this temp (don't forget to pop it after the type check). 1947 __ push(rbx); 1948 Register rbx_src_klass = rbx; 1949 1950 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx 1951 __ movptr(rsi_dst_klass, dst_klass_addr); 1952 Address super_check_offset_addr(rsi_dst_klass, sco_offset); 1953 Label L_fail_array_check; 1954 generate_type_check(rbx_src_klass, 1955 super_check_offset_addr, dst_klass_addr, 1956 rdi_temp, NULL, &L_fail_array_check); 1957 // (On fall-through, we have passed the array type check.) 1958 __ pop(rbx); 1959 __ jmp(L_plain_copy); 1960 1961 __ BIND(L_fail_array_check); 1962 // Reshuffle arguments so we can call checkcast_arraycopy: 1963 1964 // match initial saves for checkcast_arraycopy 1965 // push(rsi); // already done; see above 1966 // push(rdi); // already done; see above 1967 // push(rbx); // already done; see above 1968 1969 // Marshal outgoing arguments now, freeing registers. 1970 Address from_arg(rsp, 16+ 4); // from 1971 Address to_arg(rsp, 16+ 8); // to 1972 Address length_arg(rsp, 16+12); // elements count 1973 Address ckoff_arg(rsp, 16+16); // super_check_offset 1974 Address ckval_arg(rsp, 16+20); // super_klass 1975 1976 Address SRC_POS_arg(rsp, 16+ 8); 1977 Address DST_POS_arg(rsp, 16+16); 1978 Address LENGTH_arg(rsp, 16+20); 1979 // push rbx, changed the incoming offsets (why not just use rbp,??) 1980 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); 1981 1982 __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); 1983 __ movl2ptr(length, LENGTH_arg); // reload elements count 1984 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos 1985 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos 1986 1987 __ movptr(ckval_arg, rbx); // destination element type 1988 __ movl(rbx, Address(rbx, sco_offset)); 1989 __ movl(ckoff_arg, rbx); // corresponding class check offset 1990 1991 __ movl(length_arg, length); // outgoing length argument 1992 1993 __ lea(from, Address(src, src_pos, Address::times_ptr, 1994 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1995 __ movptr(from_arg, from); 1996 1997 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1998 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1999 __ movptr(to_arg, to); 2000 __ jump(RuntimeAddress(entry_checkcast_arraycopy)); 2001 } 2002 2003 return start; 2004 } 2005 2006 void generate_arraycopy_stubs() { 2007 address entry; 2008 address entry_jbyte_arraycopy; 2009 address entry_jshort_arraycopy; 2010 address entry_jint_arraycopy; 2011 address entry_oop_arraycopy; 2012 address entry_jlong_arraycopy; 2013 address entry_checkcast_arraycopy; 2014 2015 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = 2016 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, 2017 "arrayof_jbyte_disjoint_arraycopy"); 2018 StubRoutines::_arrayof_jbyte_arraycopy = 2019 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, 2020 NULL, "arrayof_jbyte_arraycopy"); 2021 StubRoutines::_jbyte_disjoint_arraycopy = 2022 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, 2023 "jbyte_disjoint_arraycopy"); 2024 StubRoutines::_jbyte_arraycopy = 2025 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, 2026 &entry_jbyte_arraycopy, "jbyte_arraycopy"); 2027 2028 StubRoutines::_arrayof_jshort_disjoint_arraycopy = 2029 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, 2030 "arrayof_jshort_disjoint_arraycopy"); 2031 StubRoutines::_arrayof_jshort_arraycopy = 2032 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, 2033 NULL, "arrayof_jshort_arraycopy"); 2034 StubRoutines::_jshort_disjoint_arraycopy = 2035 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, 2036 "jshort_disjoint_arraycopy"); 2037 StubRoutines::_jshort_arraycopy = 2038 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, 2039 &entry_jshort_arraycopy, "jshort_arraycopy"); 2040 2041 // Next arrays are always aligned on 4 bytes at least. 2042 StubRoutines::_jint_disjoint_arraycopy = 2043 generate_disjoint_copy(T_INT, true, Address::times_4, &entry, 2044 "jint_disjoint_arraycopy"); 2045 StubRoutines::_jint_arraycopy = 2046 generate_conjoint_copy(T_INT, true, Address::times_4, entry, 2047 &entry_jint_arraycopy, "jint_arraycopy"); 2048 2049 StubRoutines::_oop_disjoint_arraycopy = 2050 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, 2051 "oop_disjoint_arraycopy"); 2052 StubRoutines::_oop_arraycopy = 2053 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, 2054 &entry_oop_arraycopy, "oop_arraycopy"); 2055 2056 StubRoutines::_jlong_disjoint_arraycopy = 2057 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); 2058 StubRoutines::_jlong_arraycopy = 2059 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, 2060 "jlong_arraycopy"); 2061 2062 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); 2063 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); 2064 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); 2065 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); 2066 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); 2067 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); 2068 2069 StubRoutines::_arrayof_jint_disjoint_arraycopy = 2070 StubRoutines::_jint_disjoint_arraycopy; 2071 StubRoutines::_arrayof_oop_disjoint_arraycopy = 2072 StubRoutines::_oop_disjoint_arraycopy; 2073 StubRoutines::_arrayof_jlong_disjoint_arraycopy = 2074 StubRoutines::_jlong_disjoint_arraycopy; 2075 2076 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; 2077 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; 2078 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; 2079 2080 StubRoutines::_checkcast_arraycopy = 2081 generate_checkcast_copy("checkcast_arraycopy", 2082 &entry_checkcast_arraycopy); 2083 2084 StubRoutines::_unsafe_arraycopy = 2085 generate_unsafe_copy("unsafe_arraycopy", 2086 entry_jbyte_arraycopy, 2087 entry_jshort_arraycopy, 2088 entry_jint_arraycopy, 2089 entry_jlong_arraycopy); 2090 2091 StubRoutines::_generic_arraycopy = 2092 generate_generic_copy("generic_arraycopy", 2093 entry_jbyte_arraycopy, 2094 entry_jshort_arraycopy, 2095 entry_jint_arraycopy, 2096 entry_oop_arraycopy, 2097 entry_jlong_arraycopy, 2098 entry_checkcast_arraycopy); 2099 } 2100 2101 void generate_math_stubs() { 2102 { 2103 StubCodeMark mark(this, "StubRoutines", "log"); 2104 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); 2105 2106 __ fld_d(Address(rsp, 4)); 2107 __ flog(); 2108 __ ret(0); 2109 } 2110 { 2111 StubCodeMark mark(this, "StubRoutines", "log10"); 2112 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); 2113 2114 __ fld_d(Address(rsp, 4)); 2115 __ flog10(); 2116 __ ret(0); 2117 } 2118 { 2119 StubCodeMark mark(this, "StubRoutines", "sin"); 2120 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); 2121 2122 __ fld_d(Address(rsp, 4)); 2123 __ trigfunc('s'); 2124 __ ret(0); 2125 } 2126 { 2127 StubCodeMark mark(this, "StubRoutines", "cos"); 2128 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); 2129 2130 __ fld_d(Address(rsp, 4)); 2131 __ trigfunc('c'); 2132 __ ret(0); 2133 } 2134 { 2135 StubCodeMark mark(this, "StubRoutines", "tan"); 2136 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); 2137 2138 __ fld_d(Address(rsp, 4)); 2139 __ trigfunc('t'); 2140 __ ret(0); 2141 } 2142 2143 // The intrinsic version of these seem to return the same value as 2144 // the strict version. 2145 StubRoutines::_intrinsic_exp = SharedRuntime::dexp; 2146 StubRoutines::_intrinsic_pow = SharedRuntime::dpow; 2147 } 2148 2149 public: 2150 // Information about frame layout at time of blocking runtime call. 2151 // Note that we only have to preserve callee-saved registers since 2152 // the compilers are responsible for supplying a continuation point 2153 // if they expect all registers to be preserved. 2154 enum layout { 2155 thread_off, // last_java_sp 2156 rbp_off, // callee saved register 2157 ret_pc, 2158 framesize 2159 }; 2160 2161 private: 2162 2163 #undef __ 2164 #define __ masm-> 2165 2166 //------------------------------------------------------------------------------------------------------------------------ 2167 // Continuation point for throwing of implicit exceptions that are not handled in 2168 // the current activation. Fabricates an exception oop and initiates normal 2169 // exception dispatching in this frame. 2170 // 2171 // Previously the compiler (c2) allowed for callee save registers on Java calls. 2172 // This is no longer true after adapter frames were removed but could possibly 2173 // be brought back in the future if the interpreter code was reworked and it 2174 // was deemed worthwhile. The comment below was left to describe what must 2175 // happen here if callee saves were resurrected. As it stands now this stub 2176 // could actually be a vanilla BufferBlob and have now oopMap at all. 2177 // Since it doesn't make much difference we've chosen to leave it the 2178 // way it was in the callee save days and keep the comment. 2179 2180 // If we need to preserve callee-saved values we need a callee-saved oop map and 2181 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. 2182 // If the compiler needs all registers to be preserved between the fault 2183 // point and the exception handler then it must assume responsibility for that in 2184 // AbstractCompiler::continuation_for_implicit_null_exception or 2185 // continuation_for_implicit_division_by_zero_exception. All other implicit 2186 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are 2187 // either at call sites or otherwise assume that stack unwinding will be initiated, 2188 // so caller saved registers were assumed volatile in the compiler. 2189 address generate_throw_exception(const char* name, address runtime_entry, 2190 bool restore_saved_exception_pc) { 2191 2192 int insts_size = 256; 2193 int locs_size = 32; 2194 2195 CodeBuffer code(name, insts_size, locs_size); 2196 OopMapSet* oop_maps = new OopMapSet(); 2197 MacroAssembler* masm = new MacroAssembler(&code); 2198 2199 address start = __ pc(); 2200 2201 // This is an inlined and slightly modified version of call_VM 2202 // which has the ability to fetch the return PC out of 2203 // thread-local storage and also sets up last_Java_sp slightly 2204 // differently than the real call_VM 2205 Register java_thread = rbx; 2206 __ get_thread(java_thread); 2207 if (restore_saved_exception_pc) { 2208 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset()))); 2209 __ push(rax); 2210 } 2211 2212 __ enter(); // required for proper stackwalking of RuntimeStub frame 2213 2214 // pc and rbp, already pushed 2215 __ subptr(rsp, (framesize-2) * wordSize); // prolog 2216 2217 // Frame is now completed as far as size and linkage. 2218 2219 int frame_complete = __ pc() - start; 2220 2221 // push java thread (becomes first argument of C function) 2222 __ movptr(Address(rsp, thread_off * wordSize), java_thread); 2223 2224 // Set up last_Java_sp and last_Java_fp 2225 __ set_last_Java_frame(java_thread, rsp, rbp, NULL); 2226 2227 // Call runtime 2228 BLOCK_COMMENT("call runtime_entry"); 2229 __ call(RuntimeAddress(runtime_entry)); 2230 // Generate oop map 2231 OopMap* map = new OopMap(framesize, 0); 2232 oop_maps->add_gc_map(__ pc() - start, map); 2233 2234 // restore the thread (cannot use the pushed argument since arguments 2235 // may be overwritten by C code generated by an optimizing compiler); 2236 // however can use the register value directly if it is callee saved. 2237 __ get_thread(java_thread); 2238 2239 __ reset_last_Java_frame(java_thread, true, false); 2240 2241 __ leave(); // required for proper stackwalking of RuntimeStub frame 2242 2243 // check for pending exceptions 2244 #ifdef ASSERT 2245 Label L; 2246 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2247 __ jcc(Assembler::notEqual, L); 2248 __ should_not_reach_here(); 2249 __ bind(L); 2250 #endif /* ASSERT */ 2251 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2252 2253 2254 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); 2255 return stub->entry_point(); 2256 } 2257 2258 2259 void create_control_words() { 2260 // Round to nearest, 53-bit mode, exceptions masked 2261 StubRoutines::_fpu_cntrl_wrd_std = 0x027F; 2262 // Round to zero, 53-bit mode, exception mased 2263 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; 2264 // Round to nearest, 24-bit mode, exceptions masked 2265 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; 2266 // Round to nearest, 64-bit mode, exceptions masked 2267 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; 2268 // Round to nearest, 64-bit mode, exceptions masked 2269 StubRoutines::_mxcsr_std = 0x1F80; 2270 // Note: the following two constants are 80-bit values 2271 // layout is critical for correct loading by FPU. 2272 // Bias for strict fp multiply/divide 2273 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 2274 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; 2275 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; 2276 // Un-Bias for strict fp multiply/divide 2277 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 2278 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; 2279 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; 2280 } 2281 2282 //--------------------------------------------------------------------------- 2283 // Initialization 2284 2285 void generate_initial() { 2286 // Generates all stubs and initializes the entry points 2287 2288 //------------------------------------------------------------------------------------------------------------------------ 2289 // entry points that exist in all platforms 2290 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than 2291 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. 2292 StubRoutines::_forward_exception_entry = generate_forward_exception(); 2293 2294 StubRoutines::_call_stub_entry = 2295 generate_call_stub(StubRoutines::_call_stub_return_address); 2296 // is referenced by megamorphic call 2297 StubRoutines::_catch_exception_entry = generate_catch_exception(); 2298 2299 // These are currently used by Solaris/Intel 2300 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); 2301 2302 StubRoutines::_handler_for_unsafe_access_entry = 2303 generate_handler_for_unsafe_access(); 2304 2305 // platform dependent 2306 create_control_words(); 2307 2308 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); 2309 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); 2310 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, 2311 CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); 2312 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, 2313 CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); 2314 } 2315 2316 2317 void generate_all() { 2318 // Generates all stubs and initializes the entry points 2319 2320 // These entry points require SharedInfo::stack0 to be set up in non-core builds 2321 // and need to be relocatable, so they each fabricate a RuntimeStub internally. 2322 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false); 2323 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false); 2324 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true); 2325 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true); 2326 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); 2327 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false); 2328 2329 //------------------------------------------------------------------------------------------------------------------------ 2330 // entry points that are platform specific 2331 2332 // support for verify_oop (must happen after universe_init) 2333 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); 2334 2335 // arraycopy stubs used by compilers 2336 generate_arraycopy_stubs(); 2337 2338 generate_math_stubs(); 2339 } 2340 2341 2342 public: 2343 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { 2344 if (all) { 2345 generate_all(); 2346 } else { 2347 generate_initial(); 2348 } 2349 } 2350 }; // end class declaration 2351 2352 2353 void StubGenerator_generate(CodeBuffer* code, bool all) { 2354 StubGenerator g(code, all); 2355 }