stubGenerator_x86_32.cpp revision 337:9ee9cf798b59
1249259Sdim/* 2249259Sdim * Copyright 1999-2008 Sun Microsystems, Inc. All Rights Reserved. 3249259Sdim * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4249259Sdim * 5249259Sdim * This code is free software; you can redistribute it and/or modify it 6249259Sdim * under the terms of the GNU General Public License version 2 only, as 7249259Sdim * published by the Free Software Foundation. 8249259Sdim * 9249259Sdim * This code is distributed in the hope that it will be useful, but WITHOUT 10249259Sdim * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11249259Sdim * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12249259Sdim * version 2 for more details (a copy is included in the LICENSE file that 13249259Sdim * accompanied this code). 14249259Sdim * 15249259Sdim * You should have received a copy of the GNU General Public License version 16249259Sdim * 2 along with this work; if not, write to the Free Software Foundation, 17249259Sdim * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18249259Sdim * 19249259Sdim * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 20249259Sdim * CA 95054 USA or visit www.sun.com if you need additional information or 21249259Sdim * have any questions. 22249259Sdim * 23249259Sdim */ 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 43const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions 44const int FPU_CNTRL_WRD_MASK = 0xFFFF; 45 46// ------------------------------------------------------------------------------------------------------------------------- 47// Stub Code definitions 48 49static 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 64class StubGenerator: public StubCodeGenerator { 65 private: 66 67#ifdef PRODUCT 68#define inc_counter_np(counter) (0) 69#else 70 void inc_counter_np_(int& counter) { 71 __ incrementl(ExternalAddress((address)&counter)); 72 } 73#define inc_counter_np(counter) \ 74 BLOCK_COMMENT("inc_counter " #counter); \ 75 inc_counter_np_(counter); 76#endif //PRODUCT 77 78 void inc_copy_counter_np(BasicType t) { 79#ifndef PRODUCT 80 switch (t) { 81 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return; 82 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return; 83 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return; 84 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return; 85 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return; 86 } 87 ShouldNotReachHere(); 88#endif //PRODUCT 89 } 90 91 //------------------------------------------------------------------------------------------------------------------------ 92 // Call stubs are used to call Java from C 93 // 94 // [ return_from_Java ] <--- rsp 95 // [ argument word n ] 96 // ... 97 // -N [ argument word 1 ] 98 // -7 [ Possible padding for stack alignment ] 99 // -6 [ Possible padding for stack alignment ] 100 // -5 [ Possible padding for stack alignment ] 101 // -4 [ mxcsr save ] <--- rsp_after_call 102 // -3 [ saved rbx, ] 103 // -2 [ saved rsi ] 104 // -1 [ saved rdi ] 105 // 0 [ saved rbp, ] <--- rbp, 106 // 1 [ return address ] 107 // 2 [ ptr. to call wrapper ] 108 // 3 [ result ] 109 // 4 [ result_type ] 110 // 5 [ method ] 111 // 6 [ entry_point ] 112 // 7 [ parameters ] 113 // 8 [ parameter_size ] 114 // 9 [ thread ] 115 116 117 address generate_call_stub(address& return_address) { 118 StubCodeMark mark(this, "StubRoutines", "call_stub"); 119 address start = __ pc(); 120 121 // stub code parameters / addresses 122 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code"); 123 bool sse_save = false; 124 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()! 125 const int locals_count_in_bytes (4*wordSize); 126 const Address mxcsr_save (rbp, -4 * wordSize); 127 const Address saved_rbx (rbp, -3 * wordSize); 128 const Address saved_rsi (rbp, -2 * wordSize); 129 const Address saved_rdi (rbp, -1 * wordSize); 130 const Address result (rbp, 3 * wordSize); 131 const Address result_type (rbp, 4 * wordSize); 132 const Address method (rbp, 5 * wordSize); 133 const Address entry_point (rbp, 6 * wordSize); 134 const Address parameters (rbp, 7 * wordSize); 135 const Address parameter_size(rbp, 8 * wordSize); 136 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()! 137 sse_save = UseSSE > 0; 138 139 // stub code 140 __ enter(); 141 __ movptr(rcx, parameter_size); // parameter counter 142 __ shlptr(rcx, Interpreter::logStackElementSize()); // convert parameter count to bytes 143 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves 144 __ subptr(rsp, rcx); 145 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack 146 147 // save rdi, rsi, & rbx, according to C calling conventions 148 __ movptr(saved_rdi, rdi); 149 __ movptr(saved_rsi, rsi); 150 __ movptr(saved_rbx, rbx); 151 // save and initialize %mxcsr 152 if (sse_save) { 153 Label skip_ldmx; 154 __ stmxcsr(mxcsr_save); 155 __ movl(rax, mxcsr_save); 156 __ andl(rax, MXCSR_MASK); // Only check control and mask bits 157 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 158 __ cmp32(rax, mxcsr_std); 159 __ jcc(Assembler::equal, skip_ldmx); 160 __ ldmxcsr(mxcsr_std); 161 __ bind(skip_ldmx); 162 } 163 164 // make sure the control word is correct. 165 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 166 167#ifdef ASSERT 168 // make sure we have no pending exceptions 169 { Label L; 170 __ movptr(rcx, thread); 171 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 172 __ jcc(Assembler::equal, L); 173 __ stop("StubRoutines::call_stub: entered with pending exception"); 174 __ bind(L); 175 } 176#endif 177 178 // pass parameters if any 179 BLOCK_COMMENT("pass parameters if any"); 180 Label parameters_done; 181 __ movl(rcx, parameter_size); // parameter counter 182 __ testl(rcx, rcx); 183 __ jcc(Assembler::zero, parameters_done); 184 185 // parameter passing loop 186 187 Label loop; 188 // Copy Java parameters in reverse order (receiver last) 189 // Note that the argument order is inverted in the process 190 // source is rdx[rcx: N-1..0] 191 // dest is rsp[rbx: 0..N-1] 192 193 __ movptr(rdx, parameters); // parameter pointer 194 __ xorptr(rbx, rbx); 195 196 __ BIND(loop); 197 if (TaggedStackInterpreter) { 198 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), 199 -2*wordSize)); // get tag 200 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), 201 Interpreter::expr_tag_offset_in_bytes(0)), rax); // store tag 202 } 203 204 // get parameter 205 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize)); 206 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), 207 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter 208 __ increment(rbx); 209 __ decrement(rcx); 210 __ jcc(Assembler::notZero, loop); 211 212 // call Java function 213 __ BIND(parameters_done); 214 __ movptr(rbx, method); // get methodOop 215 __ movptr(rax, entry_point); // get entry_point 216 __ mov(rsi, rsp); // set sender sp 217 BLOCK_COMMENT("call Java function"); 218 __ call(rax); 219 220 BLOCK_COMMENT("call_stub_return_address:"); 221 return_address = __ pc(); 222 223 Label common_return; 224 225 __ BIND(common_return); 226 227 // store result depending on type 228 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) 229 __ movptr(rdi, result); 230 Label is_long, is_float, is_double, exit; 231 __ movl(rsi, result_type); 232 __ cmpl(rsi, T_LONG); 233 __ jcc(Assembler::equal, is_long); 234 __ cmpl(rsi, T_FLOAT); 235 __ jcc(Assembler::equal, is_float); 236 __ cmpl(rsi, T_DOUBLE); 237 __ jcc(Assembler::equal, is_double); 238 239 // handle T_INT case 240 __ movl(Address(rdi, 0), rax); 241 __ BIND(exit); 242 243 // check that FPU stack is empty 244 __ verify_FPU(0, "generate_call_stub"); 245 246 // pop parameters 247 __ lea(rsp, rsp_after_call); 248 249 // restore %mxcsr 250 if (sse_save) { 251 __ ldmxcsr(mxcsr_save); 252 } 253 254 // restore rdi, rsi and rbx, 255 __ movptr(rbx, saved_rbx); 256 __ movptr(rsi, saved_rsi); 257 __ movptr(rdi, saved_rdi); 258 __ addptr(rsp, 4*wordSize); 259 260 // return 261 __ pop(rbp); 262 __ ret(0); 263 264 // handle return types different from T_INT 265 __ BIND(is_long); 266 __ movl(Address(rdi, 0 * wordSize), rax); 267 __ movl(Address(rdi, 1 * wordSize), rdx); 268 __ jmp(exit); 269 270 __ BIND(is_float); 271 // interpreter uses xmm0 for return values 272 if (UseSSE >= 1) { 273 __ movflt(Address(rdi, 0), xmm0); 274 } else { 275 __ fstp_s(Address(rdi, 0)); 276 } 277 __ jmp(exit); 278 279 __ BIND(is_double); 280 // interpreter uses xmm0 for return values 281 if (UseSSE >= 2) { 282 __ movdbl(Address(rdi, 0), xmm0); 283 } else { 284 __ fstp_d(Address(rdi, 0)); 285 } 286 __ jmp(exit); 287 288 // If we call compiled code directly from the call stub we will 289 // need to adjust the return back to the call stub to a specialized 290 // piece of code that can handle compiled results and cleaning the fpu 291 // stack. compiled code will be set to return here instead of the 292 // return above that handles interpreter returns. 293 294 BLOCK_COMMENT("call_stub_compiled_return:"); 295 StubRoutines::x86::set_call_stub_compiled_return( __ pc()); 296 297#ifdef COMPILER2 298 if (UseSSE >= 2) { 299 __ verify_FPU(0, "call_stub_compiled_return"); 300 } else { 301 for (int i = 1; i < 8; i++) { 302 __ ffree(i); 303 } 304 305 // UseSSE <= 1 so double result should be left on TOS 306 __ movl(rsi, result_type); 307 __ cmpl(rsi, T_DOUBLE); 308 __ jcc(Assembler::equal, common_return); 309 if (UseSSE == 0) { 310 // UseSSE == 0 so float result should be left on TOS 311 __ cmpl(rsi, T_FLOAT); 312 __ jcc(Assembler::equal, common_return); 313 } 314 __ ffree(0); 315 } 316#endif /* COMPILER2 */ 317 __ jmp(common_return); 318 319 return start; 320 } 321 322 323 //------------------------------------------------------------------------------------------------------------------------ 324 // Return point for a Java call if there's an exception thrown in Java code. 325 // The exception is caught and transformed into a pending exception stored in 326 // JavaThread that can be tested from within the VM. 327 // 328 // Note: Usually the parameters are removed by the callee. In case of an exception 329 // crossing an activation frame boundary, that is not the case if the callee 330 // is compiled code => need to setup the rsp. 331 // 332 // rax,: exception oop 333 334 address generate_catch_exception() { 335 StubCodeMark mark(this, "StubRoutines", "catch_exception"); 336 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! 337 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! 338 address start = __ pc(); 339 340 // get thread directly 341 __ movptr(rcx, thread); 342#ifdef ASSERT 343 // verify that threads correspond 344 { Label L; 345 __ get_thread(rbx); 346 __ cmpptr(rbx, rcx); 347 __ jcc(Assembler::equal, L); 348 __ stop("StubRoutines::catch_exception: threads must correspond"); 349 __ bind(L); 350 } 351#endif 352 // set pending exception 353 __ verify_oop(rax); 354 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); 355 __ lea(Address(rcx, Thread::exception_file_offset ()), 356 ExternalAddress((address)__FILE__)); 357 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); 358 // complete return to VM 359 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); 360 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); 361 362 return start; 363 } 364 365 366 //------------------------------------------------------------------------------------------------------------------------ 367 // Continuation point for runtime calls returning with a pending exception. 368 // The pending exception check happened in the runtime or native call stub. 369 // The pending exception in Thread is converted into a Java-level exception. 370 // 371 // Contract with Java-level exception handlers: 372 // rax,: exception 373 // rdx: throwing pc 374 // 375 // NOTE: At entry of this stub, exception-pc must be on stack !! 376 377 address generate_forward_exception() { 378 StubCodeMark mark(this, "StubRoutines", "forward exception"); 379 address start = __ pc(); 380 381 // Upon entry, the sp points to the return address returning into Java 382 // (interpreted or compiled) code; i.e., the return address becomes the 383 // throwing pc. 384 // 385 // Arguments pushed before the runtime call are still on the stack but 386 // the exception handler will reset the stack pointer -> ignore them. 387 // A potential result in registers can be ignored as well. 388 389#ifdef ASSERT 390 // make sure this code is only executed if there is a pending exception 391 { Label L; 392 __ get_thread(rcx); 393 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 394 __ jcc(Assembler::notEqual, L); 395 __ stop("StubRoutines::forward exception: no pending exception (1)"); 396 __ bind(L); 397 } 398#endif 399 400 // compute exception handler into rbx, 401 __ movptr(rax, Address(rsp, 0)); 402 BLOCK_COMMENT("call exception_handler_for_return_address"); 403 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), rax); 404 __ mov(rbx, rax); 405 406 // setup rax, & rdx, remove return address & clear pending exception 407 __ get_thread(rcx); 408 __ pop(rdx); 409 __ movptr(rax, Address(rcx, Thread::pending_exception_offset())); 410 __ movptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 411 412#ifdef ASSERT 413 // make sure exception is set 414 { Label L; 415 __ testptr(rax, rax); 416 __ jcc(Assembler::notEqual, L); 417 __ stop("StubRoutines::forward exception: no pending exception (2)"); 418 __ bind(L); 419 } 420#endif 421 422 // continue at exception handler (return address removed) 423 // rax,: exception 424 // rbx,: exception handler 425 // rdx: throwing pc 426 __ verify_oop(rax); 427 __ jmp(rbx); 428 429 return start; 430 } 431 432 433 //---------------------------------------------------------------------------------------------------- 434 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) 435 // 436 // xchg exists as far back as 8086, lock needed for MP only 437 // Stack layout immediately after call: 438 // 439 // 0 [ret addr ] <--- rsp 440 // 1 [ ex ] 441 // 2 [ dest ] 442 // 443 // Result: *dest <- ex, return (old *dest) 444 // 445 // Note: win32 does not currently use this code 446 447 address generate_atomic_xchg() { 448 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); 449 address start = __ pc(); 450 451 __ push(rdx); 452 Address exchange(rsp, 2 * wordSize); 453 Address dest_addr(rsp, 3 * wordSize); 454 __ movl(rax, exchange); 455 __ movptr(rdx, dest_addr); 456 __ xchgl(rax, Address(rdx, 0)); 457 __ pop(rdx); 458 __ ret(0); 459 460 return start; 461 } 462 463 //---------------------------------------------------------------------------------------------------- 464 // Support for void verify_mxcsr() 465 // 466 // This routine is used with -Xcheck:jni to verify that native 467 // JNI code does not return to Java code without restoring the 468 // MXCSR register to our expected state. 469 470 471 address generate_verify_mxcsr() { 472 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); 473 address start = __ pc(); 474 475 const Address mxcsr_save(rsp, 0); 476 477 if (CheckJNICalls && UseSSE > 0 ) { 478 Label ok_ret; 479 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 480 __ push(rax); 481 __ subptr(rsp, wordSize); // allocate a temp location 482 __ stmxcsr(mxcsr_save); 483 __ movl(rax, mxcsr_save); 484 __ andl(rax, MXCSR_MASK); 485 __ cmp32(rax, mxcsr_std); 486 __ jcc(Assembler::equal, ok_ret); 487 488 __ warn("MXCSR changed by native JNI code."); 489 490 __ ldmxcsr(mxcsr_std); 491 492 __ bind(ok_ret); 493 __ addptr(rsp, wordSize); 494 __ pop(rax); 495 } 496 497 __ ret(0); 498 499 return start; 500 } 501 502 503 //--------------------------------------------------------------------------- 504 // Support for void verify_fpu_cntrl_wrd() 505 // 506 // This routine is used with -Xcheck:jni to verify that native 507 // JNI code does not return to Java code without restoring the 508 // FP control word to our expected state. 509 510 address generate_verify_fpu_cntrl_wrd() { 511 StubCodeMark mark(this, "StubRoutines", "verify_spcw"); 512 address start = __ pc(); 513 514 const Address fpu_cntrl_wrd_save(rsp, 0); 515 516 if (CheckJNICalls) { 517 Label ok_ret; 518 __ push(rax); 519 __ subptr(rsp, wordSize); // allocate a temp location 520 __ fnstcw(fpu_cntrl_wrd_save); 521 __ movl(rax, fpu_cntrl_wrd_save); 522 __ andl(rax, FPU_CNTRL_WRD_MASK); 523 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); 524 __ cmp32(rax, fpu_std); 525 __ jcc(Assembler::equal, ok_ret); 526 527 __ warn("Floating point control word changed by native JNI code."); 528 529 __ fldcw(fpu_std); 530 531 __ bind(ok_ret); 532 __ addptr(rsp, wordSize); 533 __ pop(rax); 534 } 535 536 __ ret(0); 537 538 return start; 539 } 540 541 //--------------------------------------------------------------------------- 542 // Wrapper for slow-case handling of double-to-integer conversion 543 // d2i or f2i fast case failed either because it is nan or because 544 // of under/overflow. 545 // Input: FPU TOS: float value 546 // Output: rax, (rdx): integer (long) result 547 548 address generate_d2i_wrapper(BasicType t, address fcn) { 549 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); 550 address start = __ pc(); 551 552 // Capture info about frame layout 553 enum layout { FPUState_off = 0, 554 rbp_off = FPUStateSizeInWords, 555 rdi_off, 556 rsi_off, 557 rcx_off, 558 rbx_off, 559 saved_argument_off, 560 saved_argument_off2, // 2nd half of double 561 framesize 562 }; 563 564 assert(FPUStateSizeInWords == 27, "update stack layout"); 565 566 // Save outgoing argument to stack across push_FPU_state() 567 __ subptr(rsp, wordSize * 2); 568 __ fstp_d(Address(rsp, 0)); 569 570 // Save CPU & FPU state 571 __ push(rbx); 572 __ push(rcx); 573 __ push(rsi); 574 __ push(rdi); 575 __ push(rbp); 576 __ push_FPU_state(); 577 578 // push_FPU_state() resets the FP top of stack 579 // Load original double into FP top of stack 580 __ fld_d(Address(rsp, saved_argument_off * wordSize)); 581 // Store double into stack as outgoing argument 582 __ subptr(rsp, wordSize*2); 583 __ fst_d(Address(rsp, 0)); 584 585 // Prepare FPU for doing math in C-land 586 __ empty_FPU_stack(); 587 // Call the C code to massage the double. Result in EAX 588 if (t == T_INT) 589 { BLOCK_COMMENT("SharedRuntime::d2i"); } 590 else if (t == T_LONG) 591 { BLOCK_COMMENT("SharedRuntime::d2l"); } 592 __ call_VM_leaf( fcn, 2 ); 593 594 // Restore CPU & FPU state 595 __ pop_FPU_state(); 596 __ pop(rbp); 597 __ pop(rdi); 598 __ pop(rsi); 599 __ pop(rcx); 600 __ pop(rbx); 601 __ addptr(rsp, wordSize * 2); 602 603 __ ret(0); 604 605 return start; 606 } 607 608 609 //--------------------------------------------------------------------------- 610 // The following routine generates a subroutine to throw an asynchronous 611 // UnknownError when an unsafe access gets a fault that could not be 612 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) 613 address generate_handler_for_unsafe_access() { 614 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); 615 address start = __ pc(); 616 617 __ push(0); // hole for return address-to-be 618 __ pusha(); // push registers 619 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); 620 BLOCK_COMMENT("call handle_unsafe_access"); 621 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); 622 __ movptr(next_pc, rax); // stuff next address 623 __ popa(); 624 __ ret(0); // jump to next address 625 626 return start; 627 } 628 629 630 //---------------------------------------------------------------------------------------------------- 631 // Non-destructive plausibility checks for oops 632 633 address generate_verify_oop() { 634 StubCodeMark mark(this, "StubRoutines", "verify_oop"); 635 address start = __ pc(); 636 637 // Incoming arguments on stack after saving rax,: 638 // 639 // [tos ]: saved rdx 640 // [tos + 1]: saved EFLAGS 641 // [tos + 2]: return address 642 // [tos + 3]: char* error message 643 // [tos + 4]: oop object to verify 644 // [tos + 5]: saved rax, - saved by caller and bashed 645 646 Label exit, error; 647 __ pushf(); 648 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); 649 __ push(rdx); // save rdx 650 // make sure object is 'reasonable' 651 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object 652 __ testptr(rax, rax); 653 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok 654 655 // Check if the oop is in the right area of memory 656 const int oop_mask = Universe::verify_oop_mask(); 657 const int oop_bits = Universe::verify_oop_bits(); 658 __ mov(rdx, rax); 659 __ andptr(rdx, oop_mask); 660 __ cmpptr(rdx, oop_bits); 661 __ jcc(Assembler::notZero, error); 662 663 // make sure klass is 'reasonable' 664 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass 665 __ testptr(rax, rax); 666 __ jcc(Assembler::zero, error); // if klass is NULL it is broken 667 668 // Check if the klass is in the right area of memory 669 const int klass_mask = Universe::verify_klass_mask(); 670 const int klass_bits = Universe::verify_klass_bits(); 671 __ mov(rdx, rax); 672 __ andptr(rdx, klass_mask); 673 __ cmpptr(rdx, klass_bits); 674 __ jcc(Assembler::notZero, error); 675 676 // make sure klass' klass is 'reasonable' 677 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass' klass 678 __ testptr(rax, rax); 679 __ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken 680 681 __ mov(rdx, rax); 682 __ andptr(rdx, klass_mask); 683 __ cmpptr(rdx, klass_bits); 684 __ jcc(Assembler::notZero, error); // if klass not in right area 685 // of memory it is broken too. 686 687 // return if everything seems ok 688 __ bind(exit); 689 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 690 __ pop(rdx); // restore rdx 691 __ popf(); // restore EFLAGS 692 __ ret(3 * wordSize); // pop arguments 693 694 // handle errors 695 __ bind(error); 696 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 697 __ pop(rdx); // get saved rdx back 698 __ popf(); // get saved EFLAGS off stack -- will be ignored 699 __ pusha(); // push registers (eip = return address & msg are already pushed) 700 BLOCK_COMMENT("call MacroAssembler::debug"); 701 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); 702 __ popa(); 703 __ ret(3 * wordSize); // pop arguments 704 return start; 705 } 706 707 // 708 // Generate pre-barrier for array stores 709 // 710 // Input: 711 // start - starting address 712 // end - element count 713 void gen_write_ref_array_pre_barrier(Register start, Register count) { 714 assert_different_registers(start, count); 715#if 0 // G1 only 716 BarrierSet* bs = Universe::heap()->barrier_set(); 717 switch (bs->kind()) { 718 case BarrierSet::G1SATBCT: 719 case BarrierSet::G1SATBCTLogging: 720 { 721 __ pusha(); // push registers 722 __ push(count); 723 __ push(start); 724 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre)); 725 __ addl(esp, wordSize * 2); 726 __ popa(); 727 } 728 break; 729 case BarrierSet::CardTableModRef: 730 case BarrierSet::CardTableExtension: 731 case BarrierSet::ModRef: 732 break; 733 default : 734 ShouldNotReachHere(); 735 736 } 737#endif // 0 - G1 only 738 } 739 740 741 // 742 // Generate a post-barrier for an array store 743 // 744 // start - starting address 745 // count - element count 746 // 747 // The two input registers are overwritten. 748 // 749 void gen_write_ref_array_post_barrier(Register start, Register count) { 750 BarrierSet* bs = Universe::heap()->barrier_set(); 751 assert_different_registers(start, count); 752 switch (bs->kind()) { 753#if 0 // G1 only 754 case BarrierSet::G1SATBCT: 755 case BarrierSet::G1SATBCTLogging: 756 { 757 __ pusha(); // push registers 758 __ push(count); 759 __ push(start); 760 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post)); 761 __ addl(esp, wordSize * 2); 762 __ popa(); 763 764 } 765 break; 766#endif // 0 G1 only 767 768 case BarrierSet::CardTableModRef: 769 case BarrierSet::CardTableExtension: 770 { 771 CardTableModRefBS* ct = (CardTableModRefBS*)bs; 772 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 773 774 Label L_loop; 775 const Register end = count; // elements count; end == start+count-1 776 assert_different_registers(start, end); 777 778 __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); 779 __ shrptr(start, CardTableModRefBS::card_shift); 780 __ shrptr(end, CardTableModRefBS::card_shift); 781 __ subptr(end, start); // end --> count 782 __ BIND(L_loop); 783 intptr_t disp = (intptr_t) ct->byte_map_base; 784 Address cardtable(start, count, Address::times_1, disp); 785 __ movb(cardtable, 0); 786 __ decrement(count); 787 __ jcc(Assembler::greaterEqual, L_loop); 788 } 789 break; 790 case BarrierSet::ModRef: 791 break; 792 default : 793 ShouldNotReachHere(); 794 795 } 796 } 797 798 // Copy 64 bytes chunks 799 // 800 // Inputs: 801 // from - source array address 802 // to_from - destination array address - from 803 // qword_count - 8-bytes element count, negative 804 // 805 void mmx_copy_forward(Register from, Register to_from, Register qword_count) { 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(16); 810 __ BIND(L_copy_64_bytes_loop); 811 __ movq(mmx0, Address(from, 0)); 812 __ movq(mmx1, Address(from, 8)); 813 __ movq(mmx2, Address(from, 16)); 814 __ movq(Address(from, to_from, Address::times_1, 0), mmx0); 815 __ movq(mmx3, Address(from, 24)); 816 __ movq(Address(from, to_from, Address::times_1, 8), mmx1); 817 __ movq(mmx4, Address(from, 32)); 818 __ movq(Address(from, to_from, Address::times_1, 16), mmx2); 819 __ movq(mmx5, Address(from, 40)); 820 __ movq(Address(from, to_from, Address::times_1, 24), mmx3); 821 __ movq(mmx6, Address(from, 48)); 822 __ movq(Address(from, to_from, Address::times_1, 32), mmx4); 823 __ movq(mmx7, Address(from, 56)); 824 __ movq(Address(from, to_from, Address::times_1, 40), mmx5); 825 __ movq(Address(from, to_from, Address::times_1, 48), mmx6); 826 __ movq(Address(from, to_from, Address::times_1, 56), mmx7); 827 __ addptr(from, 64); 828 __ BIND(L_copy_64_bytes); 829 __ subl(qword_count, 8); 830 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 831 __ addl(qword_count, 8); 832 __ jccb(Assembler::zero, L_exit); 833 // 834 // length is too short, just copy qwords 835 // 836 __ BIND(L_copy_8_bytes); 837 __ movq(mmx0, Address(from, 0)); 838 __ movq(Address(from, to_from, Address::times_1), mmx0); 839 __ addptr(from, 8); 840 __ decrement(qword_count); 841 __ jcc(Assembler::greater, L_copy_8_bytes); 842 __ BIND(L_exit); 843 __ emms(); 844 } 845 846 address generate_disjoint_copy(BasicType t, bool aligned, 847 Address::ScaleFactor sf, 848 address* entry, const char *name) { 849 __ align(CodeEntryAlignment); 850 StubCodeMark mark(this, "StubRoutines", name); 851 address start = __ pc(); 852 853 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 854 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; 855 856 int shift = Address::times_ptr - sf; 857 858 const Register from = rsi; // source array address 859 const Register to = rdi; // destination array address 860 const Register count = rcx; // elements count 861 const Register to_from = to; // (to - from) 862 const Register saved_to = rdx; // saved destination array address 863 864 __ enter(); // required for proper stackwalking of RuntimeStub frame 865 __ push(rsi); 866 __ push(rdi); 867 __ movptr(from , Address(rsp, 12+ 4)); 868 __ movptr(to , Address(rsp, 12+ 8)); 869 __ movl(count, Address(rsp, 12+ 12)); 870 if (t == T_OBJECT) { 871 __ testl(count, count); 872 __ jcc(Assembler::zero, L_0_count); 873 gen_write_ref_array_pre_barrier(to, count); 874 __ mov(saved_to, to); // save 'to' 875 } 876 877 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 878 BLOCK_COMMENT("Entry:"); 879 880 __ subptr(to, from); // to --> to_from 881 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 882 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 883 if (!aligned && (t == T_BYTE || t == T_SHORT)) { 884 // align source address at 4 bytes address boundary 885 if (t == T_BYTE) { 886 // One byte misalignment happens only for byte arrays 887 __ testl(from, 1); 888 __ jccb(Assembler::zero, L_skip_align1); 889 __ movb(rax, Address(from, 0)); 890 __ movb(Address(from, to_from, Address::times_1, 0), rax); 891 __ increment(from); 892 __ decrement(count); 893 __ BIND(L_skip_align1); 894 } 895 // Two bytes misalignment happens only for byte and short (char) arrays 896 __ testl(from, 2); 897 __ jccb(Assembler::zero, L_skip_align2); 898 __ movw(rax, Address(from, 0)); 899 __ movw(Address(from, to_from, Address::times_1, 0), rax); 900 __ addptr(from, 2); 901 __ subl(count, 1<<(shift-1)); 902 __ BIND(L_skip_align2); 903 } 904 if (!VM_Version::supports_mmx()) { 905 __ mov(rax, count); // save 'count' 906 __ shrl(count, shift); // bytes count 907 __ addptr(to_from, from);// restore 'to' 908 __ rep_mov(); 909 __ subptr(to_from, from);// restore 'to_from' 910 __ mov(count, rax); // restore 'count' 911 __ jmpb(L_copy_2_bytes); // all dwords were copied 912 } else { 913 // align to 8 bytes, we know we are 4 byte aligned to start 914 __ testptr(from, 4); 915 __ jccb(Assembler::zero, L_copy_64_bytes); 916 __ movl(rax, Address(from, 0)); 917 __ movl(Address(from, to_from, Address::times_1, 0), rax); 918 __ addptr(from, 4); 919 __ subl(count, 1<<shift); 920 __ BIND(L_copy_64_bytes); 921 __ mov(rax, count); 922 __ shrl(rax, shift+1); // 8 bytes chunk count 923 // 924 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop 925 // 926 mmx_copy_forward(from, to_from, rax); 927 } 928 // copy tailing dword 929 __ BIND(L_copy_4_bytes); 930 __ testl(count, 1<<shift); 931 __ jccb(Assembler::zero, L_copy_2_bytes); 932 __ movl(rax, Address(from, 0)); 933 __ movl(Address(from, to_from, Address::times_1, 0), rax); 934 if (t == T_BYTE || t == T_SHORT) { 935 __ addptr(from, 4); 936 __ BIND(L_copy_2_bytes); 937 // copy tailing word 938 __ testl(count, 1<<(shift-1)); 939 __ jccb(Assembler::zero, L_copy_byte); 940 __ movw(rax, Address(from, 0)); 941 __ movw(Address(from, to_from, Address::times_1, 0), rax); 942 if (t == T_BYTE) { 943 __ addptr(from, 2); 944 __ BIND(L_copy_byte); 945 // copy tailing byte 946 __ testl(count, 1); 947 __ jccb(Assembler::zero, L_exit); 948 __ movb(rax, Address(from, 0)); 949 __ movb(Address(from, to_from, Address::times_1, 0), rax); 950 __ BIND(L_exit); 951 } else { 952 __ BIND(L_copy_byte); 953 } 954 } else { 955 __ BIND(L_copy_2_bytes); 956 } 957 958 if (t == T_OBJECT) { 959 __ movl(count, Address(rsp, 12+12)); // reread 'count' 960 __ mov(to, saved_to); // restore 'to' 961 gen_write_ref_array_post_barrier(to, count); 962 __ BIND(L_0_count); 963 } 964 inc_copy_counter_np(t); 965 __ pop(rdi); 966 __ pop(rsi); 967 __ leave(); // required for proper stackwalking of RuntimeStub frame 968 __ xorptr(rax, rax); // return 0 969 __ ret(0); 970 return start; 971 } 972 973 974 address generate_conjoint_copy(BasicType t, bool aligned, 975 Address::ScaleFactor sf, 976 address nooverlap_target, 977 address* entry, const char *name) { 978 __ align(CodeEntryAlignment); 979 StubCodeMark mark(this, "StubRoutines", name); 980 address start = __ pc(); 981 982 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 983 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; 984 985 int shift = Address::times_ptr - sf; 986 987 const Register src = rax; // source array address 988 const Register dst = rdx; // destination array address 989 const Register from = rsi; // source array address 990 const Register to = rdi; // destination array address 991 const Register count = rcx; // elements count 992 const Register end = rax; // array end address 993 994 __ enter(); // required for proper stackwalking of RuntimeStub frame 995 __ push(rsi); 996 __ push(rdi); 997 __ movptr(src , Address(rsp, 12+ 4)); // from 998 __ movptr(dst , Address(rsp, 12+ 8)); // to 999 __ movl2ptr(count, Address(rsp, 12+12)); // count 1000 if (t == T_OBJECT) { 1001 gen_write_ref_array_pre_barrier(dst, count); 1002 } 1003 1004 if (entry != NULL) { 1005 *entry = __ pc(); // Entry point from generic arraycopy stub. 1006 BLOCK_COMMENT("Entry:"); 1007 } 1008 1009 if (t == T_OBJECT) { 1010 __ testl(count, count); 1011 __ jcc(Assembler::zero, L_0_count); 1012 } 1013 __ mov(from, src); 1014 __ mov(to , dst); 1015 1016 // arrays overlap test 1017 RuntimeAddress nooverlap(nooverlap_target); 1018 __ cmpptr(dst, src); 1019 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size 1020 __ jump_cc(Assembler::belowEqual, nooverlap); 1021 __ cmpptr(dst, end); 1022 __ jump_cc(Assembler::aboveEqual, nooverlap); 1023 1024 // copy from high to low 1025 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1026 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 1027 if (t == T_BYTE || t == T_SHORT) { 1028 // Align the end of destination array at 4 bytes address boundary 1029 __ lea(end, Address(dst, count, sf, 0)); 1030 if (t == T_BYTE) { 1031 // One byte misalignment happens only for byte arrays 1032 __ testl(end, 1); 1033 __ jccb(Assembler::zero, L_skip_align1); 1034 __ decrement(count); 1035 __ movb(rdx, Address(from, count, sf, 0)); 1036 __ movb(Address(to, count, sf, 0), rdx); 1037 __ BIND(L_skip_align1); 1038 } 1039 // Two bytes misalignment happens only for byte and short (char) arrays 1040 __ testl(end, 2); 1041 __ jccb(Assembler::zero, L_skip_align2); 1042 __ subptr(count, 1<<(shift-1)); 1043 __ movw(rdx, Address(from, count, sf, 0)); 1044 __ movw(Address(to, count, sf, 0), rdx); 1045 __ BIND(L_skip_align2); 1046 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1047 __ jcc(Assembler::below, L_copy_4_bytes); 1048 } 1049 1050 if (!VM_Version::supports_mmx()) { 1051 __ std(); 1052 __ mov(rax, count); // Save 'count' 1053 __ mov(rdx, to); // Save 'to' 1054 __ lea(rsi, Address(from, count, sf, -4)); 1055 __ lea(rdi, Address(to , count, sf, -4)); 1056 __ shrptr(count, shift); // bytes count 1057 __ rep_mov(); 1058 __ cld(); 1059 __ mov(count, rax); // restore 'count' 1060 __ andl(count, (1<<shift)-1); // mask the number of rest elements 1061 __ movptr(from, Address(rsp, 12+4)); // reread 'from' 1062 __ mov(to, rdx); // restore 'to' 1063 __ jmpb(L_copy_2_bytes); // all dword were copied 1064 } else { 1065 // Align to 8 bytes the end of array. It is aligned to 4 bytes already. 1066 __ testptr(end, 4); 1067 __ jccb(Assembler::zero, L_copy_8_bytes); 1068 __ subl(count, 1<<shift); 1069 __ movl(rdx, Address(from, count, sf, 0)); 1070 __ movl(Address(to, count, sf, 0), rdx); 1071 __ jmpb(L_copy_8_bytes); 1072 1073 __ align(16); 1074 // Move 8 bytes 1075 __ BIND(L_copy_8_bytes_loop); 1076 __ movq(mmx0, Address(from, count, sf, 0)); 1077 __ movq(Address(to, count, sf, 0), mmx0); 1078 __ BIND(L_copy_8_bytes); 1079 __ subl(count, 2<<shift); 1080 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1081 __ addl(count, 2<<shift); 1082 __ emms(); 1083 } 1084 __ BIND(L_copy_4_bytes); 1085 // copy prefix qword 1086 __ testl(count, 1<<shift); 1087 __ jccb(Assembler::zero, L_copy_2_bytes); 1088 __ movl(rdx, Address(from, count, sf, -4)); 1089 __ movl(Address(to, count, sf, -4), rdx); 1090 1091 if (t == T_BYTE || t == T_SHORT) { 1092 __ subl(count, (1<<shift)); 1093 __ BIND(L_copy_2_bytes); 1094 // copy prefix dword 1095 __ testl(count, 1<<(shift-1)); 1096 __ jccb(Assembler::zero, L_copy_byte); 1097 __ movw(rdx, Address(from, count, sf, -2)); 1098 __ movw(Address(to, count, sf, -2), rdx); 1099 if (t == T_BYTE) { 1100 __ subl(count, 1<<(shift-1)); 1101 __ BIND(L_copy_byte); 1102 // copy prefix byte 1103 __ testl(count, 1); 1104 __ jccb(Assembler::zero, L_exit); 1105 __ movb(rdx, Address(from, 0)); 1106 __ movb(Address(to, 0), rdx); 1107 __ BIND(L_exit); 1108 } else { 1109 __ BIND(L_copy_byte); 1110 } 1111 } else { 1112 __ BIND(L_copy_2_bytes); 1113 } 1114 if (t == T_OBJECT) { 1115 __ movl2ptr(count, Address(rsp, 12+12)); // reread count 1116 gen_write_ref_array_post_barrier(to, count); 1117 __ BIND(L_0_count); 1118 } 1119 inc_copy_counter_np(t); 1120 __ pop(rdi); 1121 __ pop(rsi); 1122 __ leave(); // required for proper stackwalking of RuntimeStub frame 1123 __ xorptr(rax, rax); // return 0 1124 __ ret(0); 1125 return start; 1126 } 1127 1128 1129 address generate_disjoint_long_copy(address* entry, const char *name) { 1130 __ align(CodeEntryAlignment); 1131 StubCodeMark mark(this, "StubRoutines", name); 1132 address start = __ pc(); 1133 1134 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1135 const Register from = rax; // source array address 1136 const Register to = rdx; // destination array address 1137 const Register count = rcx; // elements count 1138 const Register to_from = rdx; // (to - from) 1139 1140 __ enter(); // required for proper stackwalking of RuntimeStub frame 1141 __ movptr(from , Address(rsp, 8+0)); // from 1142 __ movptr(to , Address(rsp, 8+4)); // to 1143 __ movl2ptr(count, Address(rsp, 8+8)); // count 1144 1145 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 1146 BLOCK_COMMENT("Entry:"); 1147 1148 __ subptr(to, from); // to --> to_from 1149 if (VM_Version::supports_mmx()) { 1150 mmx_copy_forward(from, to_from, count); 1151 } else { 1152 __ jmpb(L_copy_8_bytes); 1153 __ align(16); 1154 __ BIND(L_copy_8_bytes_loop); 1155 __ fild_d(Address(from, 0)); 1156 __ fistp_d(Address(from, to_from, Address::times_1)); 1157 __ addptr(from, 8); 1158 __ BIND(L_copy_8_bytes); 1159 __ decrement(count); 1160 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1161 } 1162 inc_copy_counter_np(T_LONG); 1163 __ leave(); // required for proper stackwalking of RuntimeStub frame 1164 __ xorptr(rax, rax); // return 0 1165 __ ret(0); 1166 return start; 1167 } 1168 1169 address generate_conjoint_long_copy(address nooverlap_target, 1170 address* entry, const char *name) { 1171 __ align(CodeEntryAlignment); 1172 StubCodeMark mark(this, "StubRoutines", name); 1173 address start = __ pc(); 1174 1175 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1176 const Register from = rax; // source array address 1177 const Register to = rdx; // destination array address 1178 const Register count = rcx; // elements count 1179 const Register end_from = rax; // source array end address 1180 1181 __ enter(); // required for proper stackwalking of RuntimeStub frame 1182 __ movptr(from , Address(rsp, 8+0)); // from 1183 __ movptr(to , Address(rsp, 8+4)); // to 1184 __ movl2ptr(count, Address(rsp, 8+8)); // count 1185 1186 *entry = __ pc(); // Entry point from generic arraycopy stub. 1187 BLOCK_COMMENT("Entry:"); 1188 1189 // arrays overlap test 1190 __ cmpptr(to, from); 1191 RuntimeAddress nooverlap(nooverlap_target); 1192 __ jump_cc(Assembler::belowEqual, nooverlap); 1193 __ lea(end_from, Address(from, count, Address::times_8, 0)); 1194 __ cmpptr(to, end_from); 1195 __ movptr(from, Address(rsp, 8)); // from 1196 __ jump_cc(Assembler::aboveEqual, nooverlap); 1197 1198 __ jmpb(L_copy_8_bytes); 1199 1200 __ align(16); 1201 __ BIND(L_copy_8_bytes_loop); 1202 if (VM_Version::supports_mmx()) { 1203 __ movq(mmx0, Address(from, count, Address::times_8)); 1204 __ movq(Address(to, count, Address::times_8), mmx0); 1205 } else { 1206 __ fild_d(Address(from, count, Address::times_8)); 1207 __ fistp_d(Address(to, count, Address::times_8)); 1208 } 1209 __ BIND(L_copy_8_bytes); 1210 __ decrement(count); 1211 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1212 1213 if (VM_Version::supports_mmx()) { 1214 __ emms(); 1215 } 1216 inc_copy_counter_np(T_LONG); 1217 __ leave(); // required for proper stackwalking of RuntimeStub frame 1218 __ xorptr(rax, rax); // return 0 1219 __ ret(0); 1220 return start; 1221 } 1222 1223 1224 // Helper for generating a dynamic type check. 1225 // The sub_klass must be one of {rbx, rdx, rsi}. 1226 // The temp is killed. 1227 void generate_type_check(Register sub_klass, 1228 Address& super_check_offset_addr, 1229 Address& super_klass_addr, 1230 Register temp, 1231 Label* L_success_ptr, Label* L_failure_ptr) { 1232 BLOCK_COMMENT("type_check:"); 1233 1234 Label L_fallthrough; 1235 bool fall_through_on_success = (L_success_ptr == NULL); 1236 if (fall_through_on_success) { 1237 L_success_ptr = &L_fallthrough; 1238 } else { 1239 L_failure_ptr = &L_fallthrough; 1240 } 1241 Label& L_success = *L_success_ptr; 1242 Label& L_failure = *L_failure_ptr; 1243 1244 assert_different_registers(sub_klass, temp); 1245 1246 // a couple of useful fields in sub_klass: 1247 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + 1248 Klass::secondary_supers_offset_in_bytes()); 1249 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + 1250 Klass::secondary_super_cache_offset_in_bytes()); 1251 Address secondary_supers_addr(sub_klass, ss_offset); 1252 Address super_cache_addr( sub_klass, sc_offset); 1253 1254 // if the pointers are equal, we are done (e.g., String[] elements) 1255 __ cmpptr(sub_klass, super_klass_addr); 1256 __ jcc(Assembler::equal, L_success); 1257 1258 // check the supertype display: 1259 __ movl2ptr(temp, super_check_offset_addr); 1260 Address super_check_addr(sub_klass, temp, Address::times_1, 0); 1261 __ movptr(temp, super_check_addr); // load displayed supertype 1262 __ cmpptr(temp, super_klass_addr); // test the super type 1263 __ jcc(Assembler::equal, L_success); 1264 1265 // if it was a primary super, we can just fail immediately 1266 __ cmpl(super_check_offset_addr, sc_offset); 1267 __ jcc(Assembler::notEqual, L_failure); 1268 1269 // Now do a linear scan of the secondary super-klass chain. 1270 // This code is rarely used, so simplicity is a virtue here. 1271 inc_counter_np(SharedRuntime::_partial_subtype_ctr); 1272 { 1273 // The repne_scan instruction uses fixed registers, which we must spill. 1274 // (We need a couple more temps in any case.) 1275 __ push(rax); 1276 __ push(rcx); 1277 __ push(rdi); 1278 assert_different_registers(sub_klass, rax, rcx, rdi); 1279 1280 __ movptr(rdi, secondary_supers_addr); 1281 // Load the array length. 1282 __ movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); 1283 // Skip to start of data. 1284 __ addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); 1285 // Scan rcx words at [edi] for occurance of rax, 1286 // Set NZ/Z based on last compare 1287 __ movptr(rax, super_klass_addr); 1288 __ repne_scan(); 1289 1290 // Unspill the temp. registers: 1291 __ pop(rdi); 1292 __ pop(rcx); 1293 __ pop(rax); 1294 } 1295 __ jcc(Assembler::notEqual, L_failure); 1296 1297 // Success. Cache the super we found and proceed in triumph. 1298 __ movptr(temp, super_klass_addr); // note: rax, is dead 1299 __ movptr(super_cache_addr, temp); 1300 1301 if (!fall_through_on_success) 1302 __ jmp(L_success); 1303 1304 // Fall through on failure! 1305 __ bind(L_fallthrough); 1306 } 1307 1308 // 1309 // Generate checkcasting array copy stub 1310 // 1311 // Input: 1312 // 4(rsp) - source array address 1313 // 8(rsp) - destination array address 1314 // 12(rsp) - element count, can be zero 1315 // 16(rsp) - size_t ckoff (super_check_offset) 1316 // 20(rsp) - oop ckval (super_klass) 1317 // 1318 // Output: 1319 // rax, == 0 - success 1320 // rax, == -1^K - failure, where K is partial transfer count 1321 // 1322 address generate_checkcast_copy(const char *name, address* entry) { 1323 __ align(CodeEntryAlignment); 1324 StubCodeMark mark(this, "StubRoutines", name); 1325 address start = __ pc(); 1326 1327 Label L_load_element, L_store_element, L_do_card_marks, L_done; 1328 1329 // register use: 1330 // rax, rdx, rcx -- loop control (end_from, end_to, count) 1331 // rdi, rsi -- element access (oop, klass) 1332 // rbx, -- temp 1333 const Register from = rax; // source array address 1334 const Register to = rdx; // destination array address 1335 const Register length = rcx; // elements count 1336 const Register elem = rdi; // each oop copied 1337 const Register elem_klass = rsi; // each elem._klass (sub_klass) 1338 const Register temp = rbx; // lone remaining temp 1339 1340 __ enter(); // required for proper stackwalking of RuntimeStub frame 1341 1342 __ push(rsi); 1343 __ push(rdi); 1344 __ push(rbx); 1345 1346 Address from_arg(rsp, 16+ 4); // from 1347 Address to_arg(rsp, 16+ 8); // to 1348 Address length_arg(rsp, 16+12); // elements count 1349 Address ckoff_arg(rsp, 16+16); // super_check_offset 1350 Address ckval_arg(rsp, 16+20); // super_klass 1351 1352 // Load up: 1353 __ movptr(from, from_arg); 1354 __ movptr(to, to_arg); 1355 __ movl2ptr(length, length_arg); 1356 1357 *entry = __ pc(); // Entry point from generic arraycopy stub. 1358 BLOCK_COMMENT("Entry:"); 1359 1360 //--------------------------------------------------------------- 1361 // Assembler stub will be used for this call to arraycopy 1362 // if the two arrays are subtypes of Object[] but the 1363 // destination array type is not equal to or a supertype 1364 // of the source type. Each element must be separately 1365 // checked. 1366 1367 // Loop-invariant addresses. They are exclusive end pointers. 1368 Address end_from_addr(from, length, Address::times_ptr, 0); 1369 Address end_to_addr(to, length, Address::times_ptr, 0); 1370 1371 Register end_from = from; // re-use 1372 Register end_to = to; // re-use 1373 Register count = length; // re-use 1374 1375 // Loop-variant addresses. They assume post-incremented count < 0. 1376 Address from_element_addr(end_from, count, Address::times_ptr, 0); 1377 Address to_element_addr(end_to, count, Address::times_ptr, 0); 1378 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); 1379 1380 // Copy from low to high addresses, indexed from the end of each array. 1381 __ lea(end_from, end_from_addr); 1382 __ lea(end_to, end_to_addr); 1383 gen_write_ref_array_pre_barrier(to, count); 1384 assert(length == count, ""); // else fix next line: 1385 __ negptr(count); // negate and test the length 1386 __ jccb(Assembler::notZero, L_load_element); 1387 1388 // Empty array: Nothing to do. 1389 __ xorptr(rax, rax); // return 0 on (trivial) success 1390 __ jmp(L_done); 1391 1392 // ======== begin loop ======== 1393 // (Loop is rotated; its entry is L_load_element.) 1394 // Loop control: 1395 // for (count = -count; count != 0; count++) 1396 // Base pointers src, dst are biased by 8*count,to last element. 1397 __ align(16); 1398 1399 __ BIND(L_store_element); 1400 __ movptr(to_element_addr, elem); // store the oop 1401 __ increment(count); // increment the count toward zero 1402 __ jccb(Assembler::zero, L_do_card_marks); 1403 1404 // ======== loop entry is here ======== 1405 __ BIND(L_load_element); 1406 __ movptr(elem, from_element_addr); // load the oop 1407 __ testptr(elem, elem); 1408 __ jccb(Assembler::zero, L_store_element); 1409 1410 // (Could do a trick here: Remember last successful non-null 1411 // element stored and make a quick oop equality check on it.) 1412 1413 __ movptr(elem_klass, elem_klass_addr); // query the object klass 1414 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, 1415 &L_store_element, NULL); 1416 // (On fall-through, we have failed the element type check.) 1417 // ======== end loop ======== 1418 1419 // It was a real error; we must depend on the caller to finish the job. 1420 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. 1421 // Emit GC store barriers for the oops we have copied (length_arg + count), 1422 // and report their number to the caller. 1423 __ addl(count, length_arg); // transfers = (length - remaining) 1424 __ movl2ptr(rax, count); // save the value 1425 __ notptr(rax); // report (-1^K) to caller 1426 __ movptr(to, to_arg); // reload 1427 assert_different_registers(to, count, rax); 1428 gen_write_ref_array_post_barrier(to, count); 1429 __ jmpb(L_done); 1430 1431 // Come here on success only. 1432 __ BIND(L_do_card_marks); 1433 __ movl2ptr(count, length_arg); 1434 __ movptr(to, to_arg); // reload 1435 gen_write_ref_array_post_barrier(to, count); 1436 __ xorptr(rax, rax); // return 0 on success 1437 1438 // Common exit point (success or failure). 1439 __ BIND(L_done); 1440 __ pop(rbx); 1441 __ pop(rdi); 1442 __ pop(rsi); 1443 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); 1444 __ leave(); // required for proper stackwalking of RuntimeStub frame 1445 __ ret(0); 1446 1447 return start; 1448 } 1449 1450 // 1451 // Generate 'unsafe' array copy stub 1452 // Though just as safe as the other stubs, it takes an unscaled 1453 // size_t argument instead of an element count. 1454 // 1455 // Input: 1456 // 4(rsp) - source array address 1457 // 8(rsp) - destination array address 1458 // 12(rsp) - byte count, can be zero 1459 // 1460 // Output: 1461 // rax, == 0 - success 1462 // rax, == -1 - need to call System.arraycopy 1463 // 1464 // Examines the alignment of the operands and dispatches 1465 // to a long, int, short, or byte copy loop. 1466 // 1467 address generate_unsafe_copy(const char *name, 1468 address byte_copy_entry, 1469 address short_copy_entry, 1470 address int_copy_entry, 1471 address long_copy_entry) { 1472 1473 Label L_long_aligned, L_int_aligned, L_short_aligned; 1474 1475 __ align(CodeEntryAlignment); 1476 StubCodeMark mark(this, "StubRoutines", name); 1477 address start = __ pc(); 1478 1479 const Register from = rax; // source array address 1480 const Register to = rdx; // destination array address 1481 const Register count = rcx; // elements count 1482 1483 __ enter(); // required for proper stackwalking of RuntimeStub frame 1484 __ push(rsi); 1485 __ push(rdi); 1486 Address from_arg(rsp, 12+ 4); // from 1487 Address to_arg(rsp, 12+ 8); // to 1488 Address count_arg(rsp, 12+12); // byte count 1489 1490 // Load up: 1491 __ movptr(from , from_arg); 1492 __ movptr(to , to_arg); 1493 __ movl2ptr(count, count_arg); 1494 1495 // bump this on entry, not on exit: 1496 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); 1497 1498 const Register bits = rsi; 1499 __ mov(bits, from); 1500 __ orptr(bits, to); 1501 __ orptr(bits, count); 1502 1503 __ testl(bits, BytesPerLong-1); 1504 __ jccb(Assembler::zero, L_long_aligned); 1505 1506 __ testl(bits, BytesPerInt-1); 1507 __ jccb(Assembler::zero, L_int_aligned); 1508 1509 __ testl(bits, BytesPerShort-1); 1510 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); 1511 1512 __ BIND(L_short_aligned); 1513 __ shrptr(count, LogBytesPerShort); // size => short_count 1514 __ movl(count_arg, count); // update 'count' 1515 __ jump(RuntimeAddress(short_copy_entry)); 1516 1517 __ BIND(L_int_aligned); 1518 __ shrptr(count, LogBytesPerInt); // size => int_count 1519 __ movl(count_arg, count); // update 'count' 1520 __ jump(RuntimeAddress(int_copy_entry)); 1521 1522 __ BIND(L_long_aligned); 1523 __ shrptr(count, LogBytesPerLong); // size => qword_count 1524 __ movl(count_arg, count); // update 'count' 1525 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1526 __ pop(rsi); 1527 __ jump(RuntimeAddress(long_copy_entry)); 1528 1529 return start; 1530 } 1531 1532 1533 // Perform range checks on the proposed arraycopy. 1534 // Smashes src_pos and dst_pos. (Uses them up for temps.) 1535 void arraycopy_range_checks(Register src, 1536 Register src_pos, 1537 Register dst, 1538 Register dst_pos, 1539 Address& length, 1540 Label& L_failed) { 1541 BLOCK_COMMENT("arraycopy_range_checks:"); 1542 const Register src_end = src_pos; // source array end position 1543 const Register dst_end = dst_pos; // destination array end position 1544 __ addl(src_end, length); // src_pos + length 1545 __ addl(dst_end, length); // dst_pos + length 1546 1547 // if (src_pos + length > arrayOop(src)->length() ) FAIL; 1548 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); 1549 __ jcc(Assembler::above, L_failed); 1550 1551 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; 1552 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); 1553 __ jcc(Assembler::above, L_failed); 1554 1555 BLOCK_COMMENT("arraycopy_range_checks done"); 1556 } 1557 1558 1559 // 1560 // Generate generic array copy stubs 1561 // 1562 // Input: 1563 // 4(rsp) - src oop 1564 // 8(rsp) - src_pos 1565 // 12(rsp) - dst oop 1566 // 16(rsp) - dst_pos 1567 // 20(rsp) - element count 1568 // 1569 // Output: 1570 // rax, == 0 - success 1571 // rax, == -1^K - failure, where K is partial transfer count 1572 // 1573 address generate_generic_copy(const char *name, 1574 address entry_jbyte_arraycopy, 1575 address entry_jshort_arraycopy, 1576 address entry_jint_arraycopy, 1577 address entry_oop_arraycopy, 1578 address entry_jlong_arraycopy, 1579 address entry_checkcast_arraycopy) { 1580 Label L_failed, L_failed_0, L_objArray; 1581 1582 { int modulus = CodeEntryAlignment; 1583 int target = modulus - 5; // 5 = sizeof jmp(L_failed) 1584 int advance = target - (__ offset() % modulus); 1585 if (advance < 0) advance += modulus; 1586 if (advance > 0) __ nop(advance); 1587 } 1588 StubCodeMark mark(this, "StubRoutines", name); 1589 1590 // Short-hop target to L_failed. Makes for denser prologue code. 1591 __ BIND(L_failed_0); 1592 __ jmp(L_failed); 1593 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); 1594 1595 __ align(CodeEntryAlignment); 1596 address start = __ pc(); 1597 1598 __ enter(); // required for proper stackwalking of RuntimeStub frame 1599 __ push(rsi); 1600 __ push(rdi); 1601 1602 // bump this on entry, not on exit: 1603 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); 1604 1605 // Input values 1606 Address SRC (rsp, 12+ 4); 1607 Address SRC_POS (rsp, 12+ 8); 1608 Address DST (rsp, 12+12); 1609 Address DST_POS (rsp, 12+16); 1610 Address LENGTH (rsp, 12+20); 1611 1612 //----------------------------------------------------------------------- 1613 // Assembler stub will be used for this call to arraycopy 1614 // if the following conditions are met: 1615 // 1616 // (1) src and dst must not be null. 1617 // (2) src_pos must not be negative. 1618 // (3) dst_pos must not be negative. 1619 // (4) length must not be negative. 1620 // (5) src klass and dst klass should be the same and not NULL. 1621 // (6) src and dst should be arrays. 1622 // (7) src_pos + length must not exceed length of src. 1623 // (8) dst_pos + length must not exceed length of dst. 1624 // 1625 1626 const Register src = rax; // source array oop 1627 const Register src_pos = rsi; 1628 const Register dst = rdx; // destination array oop 1629 const Register dst_pos = rdi; 1630 const Register length = rcx; // transfer count 1631 1632 // if (src == NULL) return -1; 1633 __ movptr(src, SRC); // src oop 1634 __ testptr(src, src); 1635 __ jccb(Assembler::zero, L_failed_0); 1636 1637 // if (src_pos < 0) return -1; 1638 __ movl2ptr(src_pos, SRC_POS); // src_pos 1639 __ testl(src_pos, src_pos); 1640 __ jccb(Assembler::negative, L_failed_0); 1641 1642 // if (dst == NULL) return -1; 1643 __ movptr(dst, DST); // dst oop 1644 __ testptr(dst, dst); 1645 __ jccb(Assembler::zero, L_failed_0); 1646 1647 // if (dst_pos < 0) return -1; 1648 __ movl2ptr(dst_pos, DST_POS); // dst_pos 1649 __ testl(dst_pos, dst_pos); 1650 __ jccb(Assembler::negative, L_failed_0); 1651 1652 // if (length < 0) return -1; 1653 __ movl2ptr(length, LENGTH); // length 1654 __ testl(length, length); 1655 __ jccb(Assembler::negative, L_failed_0); 1656 1657 // if (src->klass() == NULL) return -1; 1658 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); 1659 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); 1660 const Register rcx_src_klass = rcx; // array klass 1661 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); 1662 1663#ifdef ASSERT 1664 // assert(src->klass() != NULL); 1665 BLOCK_COMMENT("assert klasses not null"); 1666 { Label L1, L2; 1667 __ testptr(rcx_src_klass, rcx_src_klass); 1668 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL 1669 __ bind(L1); 1670 __ stop("broken null klass"); 1671 __ bind(L2); 1672 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); 1673 __ jccb(Assembler::equal, L1); // this would be broken also 1674 BLOCK_COMMENT("assert done"); 1675 } 1676#endif //ASSERT 1677 1678 // Load layout helper (32-bits) 1679 // 1680 // |array_tag| | header_size | element_type | |log2_element_size| 1681 // 32 30 24 16 8 2 0 1682 // 1683 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 1684 // 1685 1686 int lh_offset = klassOopDesc::header_size() * HeapWordSize + 1687 Klass::layout_helper_offset_in_bytes(); 1688 Address src_klass_lh_addr(rcx_src_klass, lh_offset); 1689 1690 // Handle objArrays completely differently... 1691 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); 1692 __ cmpl(src_klass_lh_addr, objArray_lh); 1693 __ jcc(Assembler::equal, L_objArray); 1694 1695 // if (src->klass() != dst->klass()) return -1; 1696 __ cmpptr(rcx_src_klass, dst_klass_addr); 1697 __ jccb(Assembler::notEqual, L_failed_0); 1698 1699 const Register rcx_lh = rcx; // layout helper 1700 assert(rcx_lh == rcx_src_klass, "known alias"); 1701 __ movl(rcx_lh, src_klass_lh_addr); 1702 1703 // if (!src->is_Array()) return -1; 1704 __ cmpl(rcx_lh, Klass::_lh_neutral_value); 1705 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp 1706 1707 // At this point, it is known to be a typeArray (array_tag 0x3). 1708#ifdef ASSERT 1709 { Label L; 1710 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); 1711 __ jcc(Assembler::greaterEqual, L); // signed cmp 1712 __ stop("must be a primitive array"); 1713 __ bind(L); 1714 } 1715#endif 1716 1717 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); 1718 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1719 1720 // typeArrayKlass 1721 // 1722 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); 1723 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); 1724 // 1725 const Register rsi_offset = rsi; // array offset 1726 const Register src_array = src; // src array offset 1727 const Register dst_array = dst; // dst array offset 1728 const Register rdi_elsize = rdi; // log2 element size 1729 1730 __ mov(rsi_offset, rcx_lh); 1731 __ shrptr(rsi_offset, Klass::_lh_header_size_shift); 1732 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset 1733 __ addptr(src_array, rsi_offset); // src array offset 1734 __ addptr(dst_array, rsi_offset); // dst array offset 1735 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize 1736 1737 // next registers should be set before the jump to corresponding stub 1738 const Register from = src; // source array address 1739 const Register to = dst; // destination array address 1740 const Register count = rcx; // elements count 1741 // some of them should be duplicated on stack 1742#define FROM Address(rsp, 12+ 4) 1743#define TO Address(rsp, 12+ 8) // Not used now 1744#define COUNT Address(rsp, 12+12) // Only for oop arraycopy 1745 1746 BLOCK_COMMENT("scale indexes to element size"); 1747 __ movl2ptr(rsi, SRC_POS); // src_pos 1748 __ shlptr(rsi); // src_pos << rcx (log2 elsize) 1749 assert(src_array == from, ""); 1750 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize 1751 __ movl2ptr(rdi, DST_POS); // dst_pos 1752 __ shlptr(rdi); // dst_pos << rcx (log2 elsize) 1753 assert(dst_array == to, ""); 1754 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize 1755 __ movptr(FROM, from); // src_addr 1756 __ mov(rdi_elsize, rcx_lh); // log2 elsize 1757 __ movl2ptr(count, LENGTH); // elements count 1758 1759 BLOCK_COMMENT("choose copy loop based on element size"); 1760 __ cmpl(rdi_elsize, 0); 1761 1762 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); 1763 __ cmpl(rdi_elsize, LogBytesPerShort); 1764 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); 1765 __ cmpl(rdi_elsize, LogBytesPerInt); 1766 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); 1767#ifdef ASSERT 1768 __ cmpl(rdi_elsize, LogBytesPerLong); 1769 __ jccb(Assembler::notEqual, L_failed); 1770#endif 1771 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1772 __ pop(rsi); 1773 __ jump(RuntimeAddress(entry_jlong_arraycopy)); 1774 1775 __ BIND(L_failed); 1776 __ xorptr(rax, rax); 1777 __ notptr(rax); // return -1 1778 __ pop(rdi); 1779 __ pop(rsi); 1780 __ leave(); // required for proper stackwalking of RuntimeStub frame 1781 __ ret(0); 1782 1783 // objArrayKlass 1784 __ BIND(L_objArray); 1785 // live at this point: rcx_src_klass, src[_pos], dst[_pos] 1786 1787 Label L_plain_copy, L_checkcast_copy; 1788 // test array classes for subtyping 1789 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality 1790 __ jccb(Assembler::notEqual, L_checkcast_copy); 1791 1792 // Identically typed arrays can be copied without element-wise checks. 1793 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); 1794 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1795 1796 __ BIND(L_plain_copy); 1797 __ movl2ptr(count, LENGTH); // elements count 1798 __ movl2ptr(src_pos, SRC_POS); // reload src_pos 1799 __ lea(from, Address(src, src_pos, Address::times_ptr, 1800 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr 1801 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos 1802 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1803 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr 1804 __ movptr(FROM, from); // src_addr 1805 __ movptr(TO, to); // dst_addr 1806 __ movl(COUNT, count); // count 1807 __ jump(RuntimeAddress(entry_oop_arraycopy)); 1808 1809 __ BIND(L_checkcast_copy); 1810 // live at this point: rcx_src_klass, dst[_pos], src[_pos] 1811 { 1812 // Handy offsets: 1813 int ek_offset = (klassOopDesc::header_size() * HeapWordSize + 1814 objArrayKlass::element_klass_offset_in_bytes()); 1815 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + 1816 Klass::super_check_offset_offset_in_bytes()); 1817 1818 Register rsi_dst_klass = rsi; 1819 Register rdi_temp = rdi; 1820 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); 1821 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); 1822 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); 1823 1824 // Before looking at dst.length, make sure dst is also an objArray. 1825 __ movptr(rsi_dst_klass, dst_klass_addr); 1826 __ cmpl(dst_klass_lh_addr, objArray_lh); 1827 __ jccb(Assembler::notEqual, L_failed); 1828 1829 // It is safe to examine both src.length and dst.length. 1830 __ movl2ptr(src_pos, SRC_POS); // reload rsi 1831 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1832 // (Now src_pos and dst_pos are killed, but not src and dst.) 1833 1834 // We'll need this temp (don't forget to pop it after the type check). 1835 __ push(rbx); 1836 Register rbx_src_klass = rbx; 1837 1838 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx 1839 __ movptr(rsi_dst_klass, dst_klass_addr); 1840 Address super_check_offset_addr(rsi_dst_klass, sco_offset); 1841 Label L_fail_array_check; 1842 generate_type_check(rbx_src_klass, 1843 super_check_offset_addr, dst_klass_addr, 1844 rdi_temp, NULL, &L_fail_array_check); 1845 // (On fall-through, we have passed the array type check.) 1846 __ pop(rbx); 1847 __ jmp(L_plain_copy); 1848 1849 __ BIND(L_fail_array_check); 1850 // Reshuffle arguments so we can call checkcast_arraycopy: 1851 1852 // match initial saves for checkcast_arraycopy 1853 // push(rsi); // already done; see above 1854 // push(rdi); // already done; see above 1855 // push(rbx); // already done; see above 1856 1857 // Marshal outgoing arguments now, freeing registers. 1858 Address from_arg(rsp, 16+ 4); // from 1859 Address to_arg(rsp, 16+ 8); // to 1860 Address length_arg(rsp, 16+12); // elements count 1861 Address ckoff_arg(rsp, 16+16); // super_check_offset 1862 Address ckval_arg(rsp, 16+20); // super_klass 1863 1864 Address SRC_POS_arg(rsp, 16+ 8); 1865 Address DST_POS_arg(rsp, 16+16); 1866 Address LENGTH_arg(rsp, 16+20); 1867 // push rbx, changed the incoming offsets (why not just use rbp,??) 1868 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); 1869 1870 __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); 1871 __ movl2ptr(length, LENGTH_arg); // reload elements count 1872 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos 1873 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos 1874 1875 __ movptr(ckval_arg, rbx); // destination element type 1876 __ movl(rbx, Address(rbx, sco_offset)); 1877 __ movl(ckoff_arg, rbx); // corresponding class check offset 1878 1879 __ movl(length_arg, length); // outgoing length argument 1880 1881 __ lea(from, Address(src, src_pos, Address::times_ptr, 1882 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1883 __ movptr(from_arg, from); 1884 1885 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1886 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1887 __ movptr(to_arg, to); 1888 __ jump(RuntimeAddress(entry_checkcast_arraycopy)); 1889 } 1890 1891 return start; 1892 } 1893 1894 void generate_arraycopy_stubs() { 1895 address entry; 1896 address entry_jbyte_arraycopy; 1897 address entry_jshort_arraycopy; 1898 address entry_jint_arraycopy; 1899 address entry_oop_arraycopy; 1900 address entry_jlong_arraycopy; 1901 address entry_checkcast_arraycopy; 1902 1903 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = 1904 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, 1905 "arrayof_jbyte_disjoint_arraycopy"); 1906 StubRoutines::_arrayof_jbyte_arraycopy = 1907 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, 1908 NULL, "arrayof_jbyte_arraycopy"); 1909 StubRoutines::_jbyte_disjoint_arraycopy = 1910 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, 1911 "jbyte_disjoint_arraycopy"); 1912 StubRoutines::_jbyte_arraycopy = 1913 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, 1914 &entry_jbyte_arraycopy, "jbyte_arraycopy"); 1915 1916 StubRoutines::_arrayof_jshort_disjoint_arraycopy = 1917 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, 1918 "arrayof_jshort_disjoint_arraycopy"); 1919 StubRoutines::_arrayof_jshort_arraycopy = 1920 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, 1921 NULL, "arrayof_jshort_arraycopy"); 1922 StubRoutines::_jshort_disjoint_arraycopy = 1923 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, 1924 "jshort_disjoint_arraycopy"); 1925 StubRoutines::_jshort_arraycopy = 1926 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, 1927 &entry_jshort_arraycopy, "jshort_arraycopy"); 1928 1929 // Next arrays are always aligned on 4 bytes at least. 1930 StubRoutines::_jint_disjoint_arraycopy = 1931 generate_disjoint_copy(T_INT, true, Address::times_4, &entry, 1932 "jint_disjoint_arraycopy"); 1933 StubRoutines::_jint_arraycopy = 1934 generate_conjoint_copy(T_INT, true, Address::times_4, entry, 1935 &entry_jint_arraycopy, "jint_arraycopy"); 1936 1937 StubRoutines::_oop_disjoint_arraycopy = 1938 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, 1939 "oop_disjoint_arraycopy"); 1940 StubRoutines::_oop_arraycopy = 1941 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, 1942 &entry_oop_arraycopy, "oop_arraycopy"); 1943 1944 StubRoutines::_jlong_disjoint_arraycopy = 1945 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); 1946 StubRoutines::_jlong_arraycopy = 1947 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, 1948 "jlong_arraycopy"); 1949 1950 StubRoutines::_arrayof_jint_disjoint_arraycopy = 1951 StubRoutines::_jint_disjoint_arraycopy; 1952 StubRoutines::_arrayof_oop_disjoint_arraycopy = 1953 StubRoutines::_oop_disjoint_arraycopy; 1954 StubRoutines::_arrayof_jlong_disjoint_arraycopy = 1955 StubRoutines::_jlong_disjoint_arraycopy; 1956 1957 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; 1958 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; 1959 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; 1960 1961 StubRoutines::_checkcast_arraycopy = 1962 generate_checkcast_copy("checkcast_arraycopy", 1963 &entry_checkcast_arraycopy); 1964 1965 StubRoutines::_unsafe_arraycopy = 1966 generate_unsafe_copy("unsafe_arraycopy", 1967 entry_jbyte_arraycopy, 1968 entry_jshort_arraycopy, 1969 entry_jint_arraycopy, 1970 entry_jlong_arraycopy); 1971 1972 StubRoutines::_generic_arraycopy = 1973 generate_generic_copy("generic_arraycopy", 1974 entry_jbyte_arraycopy, 1975 entry_jshort_arraycopy, 1976 entry_jint_arraycopy, 1977 entry_oop_arraycopy, 1978 entry_jlong_arraycopy, 1979 entry_checkcast_arraycopy); 1980 } 1981 1982 public: 1983 // Information about frame layout at time of blocking runtime call. 1984 // Note that we only have to preserve callee-saved registers since 1985 // the compilers are responsible for supplying a continuation point 1986 // if they expect all registers to be preserved. 1987 enum layout { 1988 thread_off, // last_java_sp 1989 rbp_off, // callee saved register 1990 ret_pc, 1991 framesize 1992 }; 1993 1994 private: 1995 1996#undef __ 1997#define __ masm-> 1998 1999 //------------------------------------------------------------------------------------------------------------------------ 2000 // Continuation point for throwing of implicit exceptions that are not handled in 2001 // the current activation. Fabricates an exception oop and initiates normal 2002 // exception dispatching in this frame. 2003 // 2004 // Previously the compiler (c2) allowed for callee save registers on Java calls. 2005 // This is no longer true after adapter frames were removed but could possibly 2006 // be brought back in the future if the interpreter code was reworked and it 2007 // was deemed worthwhile. The comment below was left to describe what must 2008 // happen here if callee saves were resurrected. As it stands now this stub 2009 // could actually be a vanilla BufferBlob and have now oopMap at all. 2010 // Since it doesn't make much difference we've chosen to leave it the 2011 // way it was in the callee save days and keep the comment. 2012 2013 // If we need to preserve callee-saved values we need a callee-saved oop map and 2014 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. 2015 // If the compiler needs all registers to be preserved between the fault 2016 // point and the exception handler then it must assume responsibility for that in 2017 // AbstractCompiler::continuation_for_implicit_null_exception or 2018 // continuation_for_implicit_division_by_zero_exception. All other implicit 2019 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are 2020 // either at call sites or otherwise assume that stack unwinding will be initiated, 2021 // so caller saved registers were assumed volatile in the compiler. 2022 address generate_throw_exception(const char* name, address runtime_entry, 2023 bool restore_saved_exception_pc) { 2024 2025 int insts_size = 256; 2026 int locs_size = 32; 2027 2028 CodeBuffer code(name, insts_size, locs_size); 2029 OopMapSet* oop_maps = new OopMapSet(); 2030 MacroAssembler* masm = new MacroAssembler(&code); 2031 2032 address start = __ pc(); 2033 2034 // This is an inlined and slightly modified version of call_VM 2035 // which has the ability to fetch the return PC out of 2036 // thread-local storage and also sets up last_Java_sp slightly 2037 // differently than the real call_VM 2038 Register java_thread = rbx; 2039 __ get_thread(java_thread); 2040 if (restore_saved_exception_pc) { 2041 __ movptr(rax, Address(java_thread, in_bytes(JavaThread::saved_exception_pc_offset()))); 2042 __ push(rax); 2043 } 2044 2045 __ enter(); // required for proper stackwalking of RuntimeStub frame 2046 2047 // pc and rbp, already pushed 2048 __ subptr(rsp, (framesize-2) * wordSize); // prolog 2049 2050 // Frame is now completed as far as size and linkage. 2051 2052 int frame_complete = __ pc() - start; 2053 2054 // push java thread (becomes first argument of C function) 2055 __ movptr(Address(rsp, thread_off * wordSize), java_thread); 2056 2057 // Set up last_Java_sp and last_Java_fp 2058 __ set_last_Java_frame(java_thread, rsp, rbp, NULL); 2059 2060 // Call runtime 2061 BLOCK_COMMENT("call runtime_entry"); 2062 __ call(RuntimeAddress(runtime_entry)); 2063 // Generate oop map 2064 OopMap* map = new OopMap(framesize, 0); 2065 oop_maps->add_gc_map(__ pc() - start, map); 2066 2067 // restore the thread (cannot use the pushed argument since arguments 2068 // may be overwritten by C code generated by an optimizing compiler); 2069 // however can use the register value directly if it is callee saved. 2070 __ get_thread(java_thread); 2071 2072 __ reset_last_Java_frame(java_thread, true, false); 2073 2074 __ leave(); // required for proper stackwalking of RuntimeStub frame 2075 2076 // check for pending exceptions 2077#ifdef ASSERT 2078 Label L; 2079 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2080 __ jcc(Assembler::notEqual, L); 2081 __ should_not_reach_here(); 2082 __ bind(L); 2083#endif /* ASSERT */ 2084 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2085 2086 2087 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); 2088 return stub->entry_point(); 2089 } 2090 2091 2092 void create_control_words() { 2093 // Round to nearest, 53-bit mode, exceptions masked 2094 StubRoutines::_fpu_cntrl_wrd_std = 0x027F; 2095 // Round to zero, 53-bit mode, exception mased 2096 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; 2097 // Round to nearest, 24-bit mode, exceptions masked 2098 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; 2099 // Round to nearest, 64-bit mode, exceptions masked 2100 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; 2101 // Round to nearest, 64-bit mode, exceptions masked 2102 StubRoutines::_mxcsr_std = 0x1F80; 2103 // Note: the following two constants are 80-bit values 2104 // layout is critical for correct loading by FPU. 2105 // Bias for strict fp multiply/divide 2106 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 2107 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; 2108 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; 2109 // Un-Bias for strict fp multiply/divide 2110 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 2111 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; 2112 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; 2113 } 2114 2115 //--------------------------------------------------------------------------- 2116 // Initialization 2117 2118 void generate_initial() { 2119 // Generates all stubs and initializes the entry points 2120 2121 //------------------------------------------------------------------------------------------------------------------------ 2122 // entry points that exist in all platforms 2123 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than 2124 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. 2125 StubRoutines::_forward_exception_entry = generate_forward_exception(); 2126 2127 StubRoutines::_call_stub_entry = 2128 generate_call_stub(StubRoutines::_call_stub_return_address); 2129 // is referenced by megamorphic call 2130 StubRoutines::_catch_exception_entry = generate_catch_exception(); 2131 2132 // These are currently used by Solaris/Intel 2133 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); 2134 2135 StubRoutines::_handler_for_unsafe_access_entry = 2136 generate_handler_for_unsafe_access(); 2137 2138 // platform dependent 2139 create_control_words(); 2140 2141 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); 2142 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); 2143 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, 2144 CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); 2145 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, 2146 CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); 2147 } 2148 2149 2150 void generate_all() { 2151 // Generates all stubs and initializes the entry points 2152 2153 // These entry points require SharedInfo::stack0 to be set up in non-core builds 2154 // and need to be relocatable, so they each fabricate a RuntimeStub internally. 2155 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError), false); 2156 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError), false); 2157 StubRoutines::_throw_ArithmeticException_entry = generate_throw_exception("ArithmeticException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_ArithmeticException), true); 2158 StubRoutines::_throw_NullPointerException_entry = generate_throw_exception("NullPointerException throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException), true); 2159 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); 2160 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false); 2161 2162 //------------------------------------------------------------------------------------------------------------------------ 2163 // entry points that are platform specific 2164 2165 // support for verify_oop (must happen after universe_init) 2166 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); 2167 2168 // arraycopy stubs used by compilers 2169 generate_arraycopy_stubs(); 2170 } 2171 2172 2173 public: 2174 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { 2175 if (all) { 2176 generate_all(); 2177 } else { 2178 generate_initial(); 2179 } 2180 } 2181}; // end class declaration 2182 2183 2184void StubGenerator_generate(CodeBuffer* code, bool all) { 2185 StubGenerator g(code, all); 2186} 2187