1 /* 2 * Copyright (c) 1999, 2015, 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/macroAssembler.hpp" 27 #include "asm/macroAssembler.inline.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "nativeInst_x86.hpp" 30 #include "oops/instanceOop.hpp" 31 #include "oops/method.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 "runtime/thread.inline.hpp" 41 #include "utilities/top.hpp" 42 #ifdef COMPILER2 43 #include "opto/runtime.hpp" 44 #endif 45 46 // Declaration and definition of StubGenerator (no .hpp file). 47 // For a more detailed description of the stub routine structure 48 // see the comment in stubRoutines.hpp 49 50 #define __ _masm-> 51 #define a__ ((Assembler*)_masm)-> 52 53 #ifdef PRODUCT 54 #define BLOCK_COMMENT(str) /* nothing */ 55 #else 56 #define BLOCK_COMMENT(str) __ block_comment(str) 57 #endif 58 59 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") 60 61 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions 62 const int FPU_CNTRL_WRD_MASK = 0xFFFF; 63 64 // ------------------------------------------------------------------------------------------------------------------------- 65 // Stub Code definitions 66 67 static address handle_unsafe_access() { 68 JavaThread* thread = JavaThread::current(); 69 address pc = thread->saved_exception_pc(); 70 // pc is the instruction which we must emulate 71 // doing a no-op is fine: return garbage from the load 72 // therefore, compute npc 73 address npc = Assembler::locate_next_instruction(pc); 74 75 // request an async exception 76 thread->set_pending_unsafe_access_error(); 77 78 // return address of next instruction to execute 79 return npc; 80 } 81 82 class StubGenerator: public StubCodeGenerator { 83 private: 84 85 #ifdef PRODUCT 86 #define inc_counter_np(counter) ((void)0) 87 #else 88 void inc_counter_np_(int& counter) { 89 __ incrementl(ExternalAddress((address)&counter)); 90 } 91 #define inc_counter_np(counter) \ 92 BLOCK_COMMENT("inc_counter " #counter); \ 93 inc_counter_np_(counter); 94 #endif //PRODUCT 95 96 void inc_copy_counter_np(BasicType t) { 97 #ifndef PRODUCT 98 switch (t) { 99 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return; 100 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return; 101 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return; 102 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return; 103 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return; 104 } 105 ShouldNotReachHere(); 106 #endif //PRODUCT 107 } 108 109 //------------------------------------------------------------------------------------------------------------------------ 110 // Call stubs are used to call Java from C 111 // 112 // [ return_from_Java ] <--- rsp 113 // [ argument word n ] 114 // ... 115 // -N [ argument word 1 ] 116 // -7 [ Possible padding for stack alignment ] 117 // -6 [ Possible padding for stack alignment ] 118 // -5 [ Possible padding for stack alignment ] 119 // -4 [ mxcsr save ] <--- rsp_after_call 120 // -3 [ saved rbx, ] 121 // -2 [ saved rsi ] 122 // -1 [ saved rdi ] 123 // 0 [ saved rbp, ] <--- rbp, 124 // 1 [ return address ] 125 // 2 [ ptr. to call wrapper ] 126 // 3 [ result ] 127 // 4 [ result_type ] 128 // 5 [ method ] 129 // 6 [ entry_point ] 130 // 7 [ parameters ] 131 // 8 [ parameter_size ] 132 // 9 [ thread ] 133 134 135 address generate_call_stub(address& return_address) { 136 StubCodeMark mark(this, "StubRoutines", "call_stub"); 137 address start = __ pc(); 138 139 // stub code parameters / addresses 140 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code"); 141 bool sse_save = false; 142 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()! 143 const int locals_count_in_bytes (4*wordSize); 144 const Address mxcsr_save (rbp, -4 * wordSize); 145 const Address saved_rbx (rbp, -3 * wordSize); 146 const Address saved_rsi (rbp, -2 * wordSize); 147 const Address saved_rdi (rbp, -1 * wordSize); 148 const Address result (rbp, 3 * wordSize); 149 const Address result_type (rbp, 4 * wordSize); 150 const Address method (rbp, 5 * wordSize); 151 const Address entry_point (rbp, 6 * wordSize); 152 const Address parameters (rbp, 7 * wordSize); 153 const Address parameter_size(rbp, 8 * wordSize); 154 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()! 155 sse_save = UseSSE > 0; 156 157 // stub code 158 __ enter(); 159 __ movptr(rcx, parameter_size); // parameter counter 160 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes 161 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves 162 __ subptr(rsp, rcx); 163 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack 164 165 // save rdi, rsi, & rbx, according to C calling conventions 166 __ movptr(saved_rdi, rdi); 167 __ movptr(saved_rsi, rsi); 168 __ movptr(saved_rbx, rbx); 169 // save and initialize %mxcsr 170 if (sse_save) { 171 Label skip_ldmx; 172 __ stmxcsr(mxcsr_save); 173 __ movl(rax, mxcsr_save); 174 __ andl(rax, MXCSR_MASK); // Only check control and mask bits 175 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 176 __ cmp32(rax, mxcsr_std); 177 __ jcc(Assembler::equal, skip_ldmx); 178 __ ldmxcsr(mxcsr_std); 179 __ bind(skip_ldmx); 180 } 181 182 // make sure the control word is correct. 183 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); 184 185 #ifdef ASSERT 186 // make sure we have no pending exceptions 187 { Label L; 188 __ movptr(rcx, thread); 189 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 190 __ jcc(Assembler::equal, L); 191 __ stop("StubRoutines::call_stub: entered with pending exception"); 192 __ bind(L); 193 } 194 #endif 195 196 // pass parameters if any 197 BLOCK_COMMENT("pass parameters if any"); 198 Label parameters_done; 199 __ movl(rcx, parameter_size); // parameter counter 200 __ testl(rcx, rcx); 201 __ jcc(Assembler::zero, parameters_done); 202 203 // parameter passing loop 204 205 Label loop; 206 // Copy Java parameters in reverse order (receiver last) 207 // Note that the argument order is inverted in the process 208 // source is rdx[rcx: N-1..0] 209 // dest is rsp[rbx: 0..N-1] 210 211 __ movptr(rdx, parameters); // parameter pointer 212 __ xorptr(rbx, rbx); 213 214 __ BIND(loop); 215 216 // get parameter 217 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize)); 218 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), 219 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter 220 __ increment(rbx); 221 __ decrement(rcx); 222 __ jcc(Assembler::notZero, loop); 223 224 // call Java function 225 __ BIND(parameters_done); 226 __ movptr(rbx, method); // get Method* 227 __ movptr(rax, entry_point); // get entry_point 228 __ mov(rsi, rsp); // set sender sp 229 BLOCK_COMMENT("call Java function"); 230 __ call(rax); 231 232 BLOCK_COMMENT("call_stub_return_address:"); 233 return_address = __ pc(); 234 235 #ifdef COMPILER2 236 { 237 Label L_skip; 238 if (UseSSE >= 2) { 239 __ verify_FPU(0, "call_stub_return"); 240 } else { 241 for (int i = 1; i < 8; i++) { 242 __ ffree(i); 243 } 244 245 // UseSSE <= 1 so double result should be left on TOS 246 __ movl(rsi, result_type); 247 __ cmpl(rsi, T_DOUBLE); 248 __ jcc(Assembler::equal, L_skip); 249 if (UseSSE == 0) { 250 // UseSSE == 0 so float result should be left on TOS 251 __ cmpl(rsi, T_FLOAT); 252 __ jcc(Assembler::equal, L_skip); 253 } 254 __ ffree(0); 255 } 256 __ BIND(L_skip); 257 } 258 #endif // COMPILER2 259 260 // store result depending on type 261 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) 262 __ movptr(rdi, result); 263 Label is_long, is_float, is_double, exit; 264 __ movl(rsi, result_type); 265 __ cmpl(rsi, T_LONG); 266 __ jcc(Assembler::equal, is_long); 267 __ cmpl(rsi, T_FLOAT); 268 __ jcc(Assembler::equal, is_float); 269 __ cmpl(rsi, T_DOUBLE); 270 __ jcc(Assembler::equal, is_double); 271 272 // handle T_INT case 273 __ movl(Address(rdi, 0), rax); 274 __ BIND(exit); 275 276 // check that FPU stack is empty 277 __ verify_FPU(0, "generate_call_stub"); 278 279 // pop parameters 280 __ lea(rsp, rsp_after_call); 281 282 // restore %mxcsr 283 if (sse_save) { 284 __ ldmxcsr(mxcsr_save); 285 } 286 287 // restore rdi, rsi and rbx, 288 __ movptr(rbx, saved_rbx); 289 __ movptr(rsi, saved_rsi); 290 __ movptr(rdi, saved_rdi); 291 __ addptr(rsp, 4*wordSize); 292 293 // return 294 __ pop(rbp); 295 __ ret(0); 296 297 // handle return types different from T_INT 298 __ BIND(is_long); 299 __ movl(Address(rdi, 0 * wordSize), rax); 300 __ movl(Address(rdi, 1 * wordSize), rdx); 301 __ jmp(exit); 302 303 __ BIND(is_float); 304 // interpreter uses xmm0 for return values 305 if (UseSSE >= 1) { 306 __ movflt(Address(rdi, 0), xmm0); 307 } else { 308 __ fstp_s(Address(rdi, 0)); 309 } 310 __ jmp(exit); 311 312 __ BIND(is_double); 313 // interpreter uses xmm0 for return values 314 if (UseSSE >= 2) { 315 __ movdbl(Address(rdi, 0), xmm0); 316 } else { 317 __ fstp_d(Address(rdi, 0)); 318 } 319 __ jmp(exit); 320 321 return start; 322 } 323 324 325 //------------------------------------------------------------------------------------------------------------------------ 326 // Return point for a Java call if there's an exception thrown in Java code. 327 // The exception is caught and transformed into a pending exception stored in 328 // JavaThread that can be tested from within the VM. 329 // 330 // Note: Usually the parameters are removed by the callee. In case of an exception 331 // crossing an activation frame boundary, that is not the case if the callee 332 // is compiled code => need to setup the rsp. 333 // 334 // rax,: exception oop 335 336 address generate_catch_exception() { 337 StubCodeMark mark(this, "StubRoutines", "catch_exception"); 338 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! 339 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! 340 address start = __ pc(); 341 342 // get thread directly 343 __ movptr(rcx, thread); 344 #ifdef ASSERT 345 // verify that threads correspond 346 { Label L; 347 __ get_thread(rbx); 348 __ cmpptr(rbx, rcx); 349 __ jcc(Assembler::equal, L); 350 __ stop("StubRoutines::catch_exception: threads must correspond"); 351 __ bind(L); 352 } 353 #endif 354 // set pending exception 355 __ verify_oop(rax); 356 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); 357 __ lea(Address(rcx, Thread::exception_file_offset ()), 358 ExternalAddress((address)__FILE__)); 359 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); 360 // complete return to VM 361 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); 362 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); 363 364 return start; 365 } 366 367 368 //------------------------------------------------------------------------------------------------------------------------ 369 // Continuation point for runtime calls returning with a pending exception. 370 // The pending exception check happened in the runtime or native call stub. 371 // The pending exception in Thread is converted into a Java-level exception. 372 // 373 // Contract with Java-level exception handlers: 374 // rax: exception 375 // rdx: throwing pc 376 // 377 // NOTE: At entry of this stub, exception-pc must be on stack !! 378 379 address generate_forward_exception() { 380 StubCodeMark mark(this, "StubRoutines", "forward exception"); 381 address start = __ pc(); 382 const Register thread = rcx; 383 384 // other registers used in this stub 385 const Register exception_oop = rax; 386 const Register handler_addr = rbx; 387 const Register exception_pc = rdx; 388 389 // Upon entry, the sp points to the return address returning into Java 390 // (interpreted or compiled) code; i.e., the return address becomes the 391 // throwing pc. 392 // 393 // Arguments pushed before the runtime call are still on the stack but 394 // the exception handler will reset the stack pointer -> ignore them. 395 // A potential result in registers can be ignored as well. 396 397 #ifdef ASSERT 398 // make sure this code is only executed if there is a pending exception 399 { Label L; 400 __ get_thread(thread); 401 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 402 __ jcc(Assembler::notEqual, L); 403 __ stop("StubRoutines::forward exception: no pending exception (1)"); 404 __ bind(L); 405 } 406 #endif 407 408 // compute exception handler into rbx, 409 __ get_thread(thread); 410 __ movptr(exception_pc, Address(rsp, 0)); 411 BLOCK_COMMENT("call exception_handler_for_return_address"); 412 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); 413 __ mov(handler_addr, rax); 414 415 // setup rax & rdx, remove return address & clear pending exception 416 __ get_thread(thread); 417 __ pop(exception_pc); 418 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); 419 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); 420 421 #ifdef ASSERT 422 // make sure exception is set 423 { Label L; 424 __ testptr(exception_oop, exception_oop); 425 __ jcc(Assembler::notEqual, L); 426 __ stop("StubRoutines::forward exception: no pending exception (2)"); 427 __ bind(L); 428 } 429 #endif 430 431 // Verify that there is really a valid exception in RAX. 432 __ verify_oop(exception_oop); 433 434 // continue at exception handler (return address removed) 435 // rax: exception 436 // rbx: exception handler 437 // rdx: throwing pc 438 __ jmp(handler_addr); 439 440 return start; 441 } 442 443 444 //---------------------------------------------------------------------------------------------------- 445 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) 446 // 447 // xchg exists as far back as 8086, lock needed for MP only 448 // Stack layout immediately after call: 449 // 450 // 0 [ret addr ] <--- rsp 451 // 1 [ ex ] 452 // 2 [ dest ] 453 // 454 // Result: *dest <- ex, return (old *dest) 455 // 456 // Note: win32 does not currently use this code 457 458 address generate_atomic_xchg() { 459 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); 460 address start = __ pc(); 461 462 __ push(rdx); 463 Address exchange(rsp, 2 * wordSize); 464 Address dest_addr(rsp, 3 * wordSize); 465 __ movl(rax, exchange); 466 __ movptr(rdx, dest_addr); 467 __ xchgl(rax, Address(rdx, 0)); 468 __ pop(rdx); 469 __ ret(0); 470 471 return start; 472 } 473 474 //---------------------------------------------------------------------------------------------------- 475 // Support for void verify_mxcsr() 476 // 477 // This routine is used with -Xcheck:jni to verify that native 478 // JNI code does not return to Java code without restoring the 479 // MXCSR register to our expected state. 480 481 482 address generate_verify_mxcsr() { 483 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); 484 address start = __ pc(); 485 486 const Address mxcsr_save(rsp, 0); 487 488 if (CheckJNICalls && UseSSE > 0 ) { 489 Label ok_ret; 490 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); 491 __ push(rax); 492 __ subptr(rsp, wordSize); // allocate a temp location 493 __ stmxcsr(mxcsr_save); 494 __ movl(rax, mxcsr_save); 495 __ andl(rax, MXCSR_MASK); 496 __ cmp32(rax, mxcsr_std); 497 __ jcc(Assembler::equal, ok_ret); 498 499 __ warn("MXCSR changed by native JNI code."); 500 501 __ ldmxcsr(mxcsr_std); 502 503 __ bind(ok_ret); 504 __ addptr(rsp, wordSize); 505 __ pop(rax); 506 } 507 508 __ ret(0); 509 510 return start; 511 } 512 513 514 //--------------------------------------------------------------------------- 515 // Support for void verify_fpu_cntrl_wrd() 516 // 517 // This routine is used with -Xcheck:jni to verify that native 518 // JNI code does not return to Java code without restoring the 519 // FP control word to our expected state. 520 521 address generate_verify_fpu_cntrl_wrd() { 522 StubCodeMark mark(this, "StubRoutines", "verify_spcw"); 523 address start = __ pc(); 524 525 const Address fpu_cntrl_wrd_save(rsp, 0); 526 527 if (CheckJNICalls) { 528 Label ok_ret; 529 __ push(rax); 530 __ subptr(rsp, wordSize); // allocate a temp location 531 __ fnstcw(fpu_cntrl_wrd_save); 532 __ movl(rax, fpu_cntrl_wrd_save); 533 __ andl(rax, FPU_CNTRL_WRD_MASK); 534 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); 535 __ cmp32(rax, fpu_std); 536 __ jcc(Assembler::equal, ok_ret); 537 538 __ warn("Floating point control word changed by native JNI code."); 539 540 __ fldcw(fpu_std); 541 542 __ bind(ok_ret); 543 __ addptr(rsp, wordSize); 544 __ pop(rax); 545 } 546 547 __ ret(0); 548 549 return start; 550 } 551 552 //--------------------------------------------------------------------------- 553 // Wrapper for slow-case handling of double-to-integer conversion 554 // d2i or f2i fast case failed either because it is nan or because 555 // of under/overflow. 556 // Input: FPU TOS: float value 557 // Output: rax, (rdx): integer (long) result 558 559 address generate_d2i_wrapper(BasicType t, address fcn) { 560 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); 561 address start = __ pc(); 562 563 // Capture info about frame layout 564 enum layout { FPUState_off = 0, 565 rbp_off = FPUStateSizeInWords, 566 rdi_off, 567 rsi_off, 568 rcx_off, 569 rbx_off, 570 saved_argument_off, 571 saved_argument_off2, // 2nd half of double 572 framesize 573 }; 574 575 assert(FPUStateSizeInWords == 27, "update stack layout"); 576 577 // Save outgoing argument to stack across push_FPU_state() 578 __ subptr(rsp, wordSize * 2); 579 __ fstp_d(Address(rsp, 0)); 580 581 // Save CPU & FPU state 582 __ push(rbx); 583 __ push(rcx); 584 __ push(rsi); 585 __ push(rdi); 586 __ push(rbp); 587 __ push_FPU_state(); 588 589 // push_FPU_state() resets the FP top of stack 590 // Load original double into FP top of stack 591 __ fld_d(Address(rsp, saved_argument_off * wordSize)); 592 // Store double into stack as outgoing argument 593 __ subptr(rsp, wordSize*2); 594 __ fst_d(Address(rsp, 0)); 595 596 // Prepare FPU for doing math in C-land 597 __ empty_FPU_stack(); 598 // Call the C code to massage the double. Result in EAX 599 if (t == T_INT) 600 { BLOCK_COMMENT("SharedRuntime::d2i"); } 601 else if (t == T_LONG) 602 { BLOCK_COMMENT("SharedRuntime::d2l"); } 603 __ call_VM_leaf( fcn, 2 ); 604 605 // Restore CPU & FPU state 606 __ pop_FPU_state(); 607 __ pop(rbp); 608 __ pop(rdi); 609 __ pop(rsi); 610 __ pop(rcx); 611 __ pop(rbx); 612 __ addptr(rsp, wordSize * 2); 613 614 __ ret(0); 615 616 return start; 617 } 618 619 620 //--------------------------------------------------------------------------- 621 // The following routine generates a subroutine to throw an asynchronous 622 // UnknownError when an unsafe access gets a fault that could not be 623 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) 624 address generate_handler_for_unsafe_access() { 625 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); 626 address start = __ pc(); 627 628 __ push(0); // hole for return address-to-be 629 __ pusha(); // push registers 630 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); 631 BLOCK_COMMENT("call handle_unsafe_access"); 632 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); 633 __ movptr(next_pc, rax); // stuff next address 634 __ popa(); 635 __ ret(0); // jump to next address 636 637 return start; 638 } 639 640 641 //---------------------------------------------------------------------------------------------------- 642 // Non-destructive plausibility checks for oops 643 644 address generate_verify_oop() { 645 StubCodeMark mark(this, "StubRoutines", "verify_oop"); 646 address start = __ pc(); 647 648 // Incoming arguments on stack after saving rax,: 649 // 650 // [tos ]: saved rdx 651 // [tos + 1]: saved EFLAGS 652 // [tos + 2]: return address 653 // [tos + 3]: char* error message 654 // [tos + 4]: oop object to verify 655 // [tos + 5]: saved rax, - saved by caller and bashed 656 657 Label exit, error; 658 __ pushf(); 659 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); 660 __ push(rdx); // save rdx 661 // make sure object is 'reasonable' 662 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object 663 __ testptr(rax, rax); 664 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok 665 666 // Check if the oop is in the right area of memory 667 const int oop_mask = Universe::verify_oop_mask(); 668 const int oop_bits = Universe::verify_oop_bits(); 669 __ mov(rdx, rax); 670 __ andptr(rdx, oop_mask); 671 __ cmpptr(rdx, oop_bits); 672 __ jcc(Assembler::notZero, error); 673 674 // make sure klass is 'reasonable', which is not zero. 675 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass 676 __ testptr(rax, rax); 677 __ jcc(Assembler::zero, error); // if klass is NULL it is broken 678 679 // return if everything seems ok 680 __ bind(exit); 681 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 682 __ pop(rdx); // restore rdx 683 __ popf(); // restore EFLAGS 684 __ ret(3 * wordSize); // pop arguments 685 686 // handle errors 687 __ bind(error); 688 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back 689 __ pop(rdx); // get saved rdx back 690 __ popf(); // get saved EFLAGS off stack -- will be ignored 691 __ pusha(); // push registers (eip = return address & msg are already pushed) 692 BLOCK_COMMENT("call MacroAssembler::debug"); 693 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); 694 __ popa(); 695 __ ret(3 * wordSize); // pop arguments 696 return start; 697 } 698 699 // 700 // Generate pre-barrier for array stores 701 // 702 // Input: 703 // start - starting address 704 // count - element count 705 void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) { 706 assert_different_registers(start, count); 707 BarrierSet* bs = Universe::heap()->barrier_set(); 708 switch (bs->kind()) { 709 case BarrierSet::G1SATBCTLogging: 710 // With G1, don't generate the call if we statically know that the target in uninitialized 711 if (!uninitialized_target) { 712 __ pusha(); // push registers 713 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), 714 start, count); 715 __ popa(); 716 } 717 break; 718 case BarrierSet::CardTableModRef: 719 case BarrierSet::CardTableExtension: 720 case BarrierSet::ModRef: 721 break; 722 default : 723 ShouldNotReachHere(); 724 725 } 726 } 727 728 729 // 730 // Generate a post-barrier for an array store 731 // 732 // start - starting address 733 // count - element count 734 // 735 // The two input registers are overwritten. 736 // 737 void gen_write_ref_array_post_barrier(Register start, Register count) { 738 BarrierSet* bs = Universe::heap()->barrier_set(); 739 assert_different_registers(start, count); 740 switch (bs->kind()) { 741 case BarrierSet::G1SATBCTLogging: 742 { 743 __ pusha(); // push registers 744 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), 745 start, count); 746 __ popa(); 747 } 748 break; 749 750 case BarrierSet::CardTableModRef: 751 case BarrierSet::CardTableExtension: 752 { 753 CardTableModRefBS* ct = barrier_set_cast<CardTableModRefBS>(bs); 754 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); 755 756 Label L_loop; 757 const Register end = count; // elements count; end == start+count-1 758 assert_different_registers(start, end); 759 760 __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); 761 __ shrptr(start, CardTableModRefBS::card_shift); 762 __ shrptr(end, CardTableModRefBS::card_shift); 763 __ subptr(end, start); // end --> count 764 __ BIND(L_loop); 765 intptr_t disp = (intptr_t) ct->byte_map_base; 766 Address cardtable(start, count, Address::times_1, disp); 767 __ movb(cardtable, 0); 768 __ decrement(count); 769 __ jcc(Assembler::greaterEqual, L_loop); 770 } 771 break; 772 case BarrierSet::ModRef: 773 break; 774 default : 775 ShouldNotReachHere(); 776 777 } 778 } 779 780 781 // Copy 64 bytes chunks 782 // 783 // Inputs: 784 // from - source array address 785 // to_from - destination array address - from 786 // qword_count - 8-bytes element count, negative 787 // 788 void xmm_copy_forward(Register from, Register to_from, Register qword_count) { 789 assert( UseSSE >= 2, "supported cpu only" ); 790 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 791 // Copy 64-byte chunks 792 __ jmpb(L_copy_64_bytes); 793 __ align(OptoLoopAlignment); 794 __ BIND(L_copy_64_bytes_loop); 795 796 if (UseUnalignedLoadStores) { 797 if (UseAVX >= 2) { 798 __ vmovdqu(xmm0, Address(from, 0)); 799 __ vmovdqu(Address(from, to_from, Address::times_1, 0), xmm0); 800 __ vmovdqu(xmm1, Address(from, 32)); 801 __ vmovdqu(Address(from, to_from, Address::times_1, 32), xmm1); 802 } else { 803 __ movdqu(xmm0, Address(from, 0)); 804 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0); 805 __ movdqu(xmm1, Address(from, 16)); 806 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1); 807 __ movdqu(xmm2, Address(from, 32)); 808 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2); 809 __ movdqu(xmm3, Address(from, 48)); 810 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3); 811 } 812 } else { 813 __ movq(xmm0, Address(from, 0)); 814 __ movq(Address(from, to_from, Address::times_1, 0), xmm0); 815 __ movq(xmm1, Address(from, 8)); 816 __ movq(Address(from, to_from, Address::times_1, 8), xmm1); 817 __ movq(xmm2, Address(from, 16)); 818 __ movq(Address(from, to_from, Address::times_1, 16), xmm2); 819 __ movq(xmm3, Address(from, 24)); 820 __ movq(Address(from, to_from, Address::times_1, 24), xmm3); 821 __ movq(xmm4, Address(from, 32)); 822 __ movq(Address(from, to_from, Address::times_1, 32), xmm4); 823 __ movq(xmm5, Address(from, 40)); 824 __ movq(Address(from, to_from, Address::times_1, 40), xmm5); 825 __ movq(xmm6, Address(from, 48)); 826 __ movq(Address(from, to_from, Address::times_1, 48), xmm6); 827 __ movq(xmm7, Address(from, 56)); 828 __ movq(Address(from, to_from, Address::times_1, 56), xmm7); 829 } 830 831 __ addl(from, 64); 832 __ BIND(L_copy_64_bytes); 833 __ subl(qword_count, 8); 834 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 835 836 if (UseUnalignedLoadStores && (UseAVX >= 2)) { 837 // clean upper bits of YMM registers 838 __ vpxor(xmm0, xmm0); 839 __ vpxor(xmm1, xmm1); 840 } 841 __ addl(qword_count, 8); 842 __ jccb(Assembler::zero, L_exit); 843 // 844 // length is too short, just copy qwords 845 // 846 __ BIND(L_copy_8_bytes); 847 __ movq(xmm0, Address(from, 0)); 848 __ movq(Address(from, to_from, Address::times_1), xmm0); 849 __ addl(from, 8); 850 __ decrement(qword_count); 851 __ jcc(Assembler::greater, L_copy_8_bytes); 852 __ BIND(L_exit); 853 } 854 855 // Copy 64 bytes chunks 856 // 857 // Inputs: 858 // from - source array address 859 // to_from - destination array address - from 860 // qword_count - 8-bytes element count, negative 861 // 862 void mmx_copy_forward(Register from, Register to_from, Register qword_count) { 863 assert( VM_Version::supports_mmx(), "supported cpu only" ); 864 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; 865 // Copy 64-byte chunks 866 __ jmpb(L_copy_64_bytes); 867 __ align(OptoLoopAlignment); 868 __ BIND(L_copy_64_bytes_loop); 869 __ movq(mmx0, Address(from, 0)); 870 __ movq(mmx1, Address(from, 8)); 871 __ movq(mmx2, Address(from, 16)); 872 __ movq(Address(from, to_from, Address::times_1, 0), mmx0); 873 __ movq(mmx3, Address(from, 24)); 874 __ movq(Address(from, to_from, Address::times_1, 8), mmx1); 875 __ movq(mmx4, Address(from, 32)); 876 __ movq(Address(from, to_from, Address::times_1, 16), mmx2); 877 __ movq(mmx5, Address(from, 40)); 878 __ movq(Address(from, to_from, Address::times_1, 24), mmx3); 879 __ movq(mmx6, Address(from, 48)); 880 __ movq(Address(from, to_from, Address::times_1, 32), mmx4); 881 __ movq(mmx7, Address(from, 56)); 882 __ movq(Address(from, to_from, Address::times_1, 40), mmx5); 883 __ movq(Address(from, to_from, Address::times_1, 48), mmx6); 884 __ movq(Address(from, to_from, Address::times_1, 56), mmx7); 885 __ addptr(from, 64); 886 __ BIND(L_copy_64_bytes); 887 __ subl(qword_count, 8); 888 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); 889 __ addl(qword_count, 8); 890 __ jccb(Assembler::zero, L_exit); 891 // 892 // length is too short, just copy qwords 893 // 894 __ BIND(L_copy_8_bytes); 895 __ movq(mmx0, Address(from, 0)); 896 __ movq(Address(from, to_from, Address::times_1), mmx0); 897 __ addptr(from, 8); 898 __ decrement(qword_count); 899 __ jcc(Assembler::greater, L_copy_8_bytes); 900 __ BIND(L_exit); 901 __ emms(); 902 } 903 904 address generate_disjoint_copy(BasicType t, bool aligned, 905 Address::ScaleFactor sf, 906 address* entry, const char *name, 907 bool dest_uninitialized = false) { 908 __ align(CodeEntryAlignment); 909 StubCodeMark mark(this, "StubRoutines", name); 910 address start = __ pc(); 911 912 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 913 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; 914 915 int shift = Address::times_ptr - sf; 916 917 const Register from = rsi; // source array address 918 const Register to = rdi; // destination array address 919 const Register count = rcx; // elements count 920 const Register to_from = to; // (to - from) 921 const Register saved_to = rdx; // saved destination array address 922 923 __ enter(); // required for proper stackwalking of RuntimeStub frame 924 __ push(rsi); 925 __ push(rdi); 926 __ movptr(from , Address(rsp, 12+ 4)); 927 __ movptr(to , Address(rsp, 12+ 8)); 928 __ movl(count, Address(rsp, 12+ 12)); 929 930 if (entry != NULL) { 931 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 932 BLOCK_COMMENT("Entry:"); 933 } 934 935 if (t == T_OBJECT) { 936 __ testl(count, count); 937 __ jcc(Assembler::zero, L_0_count); 938 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); 939 __ mov(saved_to, to); // save 'to' 940 } 941 942 __ subptr(to, from); // to --> to_from 943 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 944 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 945 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { 946 // align source address at 4 bytes address boundary 947 if (t == T_BYTE) { 948 // One byte misalignment happens only for byte arrays 949 __ testl(from, 1); 950 __ jccb(Assembler::zero, L_skip_align1); 951 __ movb(rax, Address(from, 0)); 952 __ movb(Address(from, to_from, Address::times_1, 0), rax); 953 __ increment(from); 954 __ decrement(count); 955 __ BIND(L_skip_align1); 956 } 957 // Two bytes misalignment happens only for byte and short (char) arrays 958 __ testl(from, 2); 959 __ jccb(Assembler::zero, L_skip_align2); 960 __ movw(rax, Address(from, 0)); 961 __ movw(Address(from, to_from, Address::times_1, 0), rax); 962 __ addptr(from, 2); 963 __ subl(count, 1<<(shift-1)); 964 __ BIND(L_skip_align2); 965 } 966 if (!VM_Version::supports_mmx()) { 967 __ mov(rax, count); // save 'count' 968 __ shrl(count, shift); // bytes count 969 __ addptr(to_from, from);// restore 'to' 970 __ rep_mov(); 971 __ subptr(to_from, from);// restore 'to_from' 972 __ mov(count, rax); // restore 'count' 973 __ jmpb(L_copy_2_bytes); // all dwords were copied 974 } else { 975 if (!UseUnalignedLoadStores) { 976 // align to 8 bytes, we know we are 4 byte aligned to start 977 __ testptr(from, 4); 978 __ jccb(Assembler::zero, L_copy_64_bytes); 979 __ movl(rax, Address(from, 0)); 980 __ movl(Address(from, to_from, Address::times_1, 0), rax); 981 __ addptr(from, 4); 982 __ subl(count, 1<<shift); 983 } 984 __ BIND(L_copy_64_bytes); 985 __ mov(rax, count); 986 __ shrl(rax, shift+1); // 8 bytes chunk count 987 // 988 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop 989 // 990 if (UseXMMForArrayCopy) { 991 xmm_copy_forward(from, to_from, rax); 992 } else { 993 mmx_copy_forward(from, to_from, rax); 994 } 995 } 996 // copy tailing dword 997 __ BIND(L_copy_4_bytes); 998 __ testl(count, 1<<shift); 999 __ jccb(Assembler::zero, L_copy_2_bytes); 1000 __ movl(rax, Address(from, 0)); 1001 __ movl(Address(from, to_from, Address::times_1, 0), rax); 1002 if (t == T_BYTE || t == T_SHORT) { 1003 __ addptr(from, 4); 1004 __ BIND(L_copy_2_bytes); 1005 // copy tailing word 1006 __ testl(count, 1<<(shift-1)); 1007 __ jccb(Assembler::zero, L_copy_byte); 1008 __ movw(rax, Address(from, 0)); 1009 __ movw(Address(from, to_from, Address::times_1, 0), rax); 1010 if (t == T_BYTE) { 1011 __ addptr(from, 2); 1012 __ BIND(L_copy_byte); 1013 // copy tailing byte 1014 __ testl(count, 1); 1015 __ jccb(Assembler::zero, L_exit); 1016 __ movb(rax, Address(from, 0)); 1017 __ movb(Address(from, to_from, Address::times_1, 0), rax); 1018 __ BIND(L_exit); 1019 } else { 1020 __ BIND(L_copy_byte); 1021 } 1022 } else { 1023 __ BIND(L_copy_2_bytes); 1024 } 1025 1026 if (t == T_OBJECT) { 1027 __ movl(count, Address(rsp, 12+12)); // reread 'count' 1028 __ mov(to, saved_to); // restore 'to' 1029 gen_write_ref_array_post_barrier(to, count); 1030 __ BIND(L_0_count); 1031 } 1032 inc_copy_counter_np(t); 1033 __ pop(rdi); 1034 __ pop(rsi); 1035 __ leave(); // required for proper stackwalking of RuntimeStub frame 1036 __ xorptr(rax, rax); // return 0 1037 __ ret(0); 1038 return start; 1039 } 1040 1041 1042 address generate_fill(BasicType t, bool aligned, const char *name) { 1043 __ align(CodeEntryAlignment); 1044 StubCodeMark mark(this, "StubRoutines", name); 1045 address start = __ pc(); 1046 1047 BLOCK_COMMENT("Entry:"); 1048 1049 const Register to = rdi; // source array address 1050 const Register value = rdx; // value 1051 const Register count = rsi; // elements count 1052 1053 __ enter(); // required for proper stackwalking of RuntimeStub frame 1054 __ push(rsi); 1055 __ push(rdi); 1056 __ movptr(to , Address(rsp, 12+ 4)); 1057 __ movl(value, Address(rsp, 12+ 8)); 1058 __ movl(count, Address(rsp, 12+ 12)); 1059 1060 __ generate_fill(t, aligned, to, value, count, rax, xmm0); 1061 1062 __ pop(rdi); 1063 __ pop(rsi); 1064 __ leave(); // required for proper stackwalking of RuntimeStub frame 1065 __ ret(0); 1066 return start; 1067 } 1068 1069 address generate_conjoint_copy(BasicType t, bool aligned, 1070 Address::ScaleFactor sf, 1071 address nooverlap_target, 1072 address* entry, const char *name, 1073 bool dest_uninitialized = false) { 1074 __ align(CodeEntryAlignment); 1075 StubCodeMark mark(this, "StubRoutines", name); 1076 address start = __ pc(); 1077 1078 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; 1079 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; 1080 1081 int shift = Address::times_ptr - sf; 1082 1083 const Register src = rax; // source array address 1084 const Register dst = rdx; // destination array address 1085 const Register from = rsi; // source array address 1086 const Register to = rdi; // destination array address 1087 const Register count = rcx; // elements count 1088 const Register end = rax; // array end address 1089 1090 __ enter(); // required for proper stackwalking of RuntimeStub frame 1091 __ push(rsi); 1092 __ push(rdi); 1093 __ movptr(src , Address(rsp, 12+ 4)); // from 1094 __ movptr(dst , Address(rsp, 12+ 8)); // to 1095 __ movl2ptr(count, Address(rsp, 12+12)); // count 1096 1097 if (entry != NULL) { 1098 *entry = __ pc(); // Entry point from generic arraycopy stub. 1099 BLOCK_COMMENT("Entry:"); 1100 } 1101 1102 // nooverlap_target expects arguments in rsi and rdi. 1103 __ mov(from, src); 1104 __ mov(to , dst); 1105 1106 // arrays overlap test: dispatch to disjoint stub if necessary. 1107 RuntimeAddress nooverlap(nooverlap_target); 1108 __ cmpptr(dst, src); 1109 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size 1110 __ jump_cc(Assembler::belowEqual, nooverlap); 1111 __ cmpptr(dst, end); 1112 __ jump_cc(Assembler::aboveEqual, nooverlap); 1113 1114 if (t == T_OBJECT) { 1115 __ testl(count, count); 1116 __ jcc(Assembler::zero, L_0_count); 1117 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized); 1118 } 1119 1120 // copy from high to low 1121 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1122 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp 1123 if (t == T_BYTE || t == T_SHORT) { 1124 // Align the end of destination array at 4 bytes address boundary 1125 __ lea(end, Address(dst, count, sf, 0)); 1126 if (t == T_BYTE) { 1127 // One byte misalignment happens only for byte arrays 1128 __ testl(end, 1); 1129 __ jccb(Assembler::zero, L_skip_align1); 1130 __ decrement(count); 1131 __ movb(rdx, Address(from, count, sf, 0)); 1132 __ movb(Address(to, count, sf, 0), rdx); 1133 __ BIND(L_skip_align1); 1134 } 1135 // Two bytes misalignment happens only for byte and short (char) arrays 1136 __ testl(end, 2); 1137 __ jccb(Assembler::zero, L_skip_align2); 1138 __ subptr(count, 1<<(shift-1)); 1139 __ movw(rdx, Address(from, count, sf, 0)); 1140 __ movw(Address(to, count, sf, 0), rdx); 1141 __ BIND(L_skip_align2); 1142 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element 1143 __ jcc(Assembler::below, L_copy_4_bytes); 1144 } 1145 1146 if (!VM_Version::supports_mmx()) { 1147 __ std(); 1148 __ mov(rax, count); // Save 'count' 1149 __ mov(rdx, to); // Save 'to' 1150 __ lea(rsi, Address(from, count, sf, -4)); 1151 __ lea(rdi, Address(to , count, sf, -4)); 1152 __ shrptr(count, shift); // bytes count 1153 __ rep_mov(); 1154 __ cld(); 1155 __ mov(count, rax); // restore 'count' 1156 __ andl(count, (1<<shift)-1); // mask the number of rest elements 1157 __ movptr(from, Address(rsp, 12+4)); // reread 'from' 1158 __ mov(to, rdx); // restore 'to' 1159 __ jmpb(L_copy_2_bytes); // all dword were copied 1160 } else { 1161 // Align to 8 bytes the end of array. It is aligned to 4 bytes already. 1162 __ testptr(end, 4); 1163 __ jccb(Assembler::zero, L_copy_8_bytes); 1164 __ subl(count, 1<<shift); 1165 __ movl(rdx, Address(from, count, sf, 0)); 1166 __ movl(Address(to, count, sf, 0), rdx); 1167 __ jmpb(L_copy_8_bytes); 1168 1169 __ align(OptoLoopAlignment); 1170 // Move 8 bytes 1171 __ BIND(L_copy_8_bytes_loop); 1172 if (UseXMMForArrayCopy) { 1173 __ movq(xmm0, Address(from, count, sf, 0)); 1174 __ movq(Address(to, count, sf, 0), xmm0); 1175 } else { 1176 __ movq(mmx0, Address(from, count, sf, 0)); 1177 __ movq(Address(to, count, sf, 0), mmx0); 1178 } 1179 __ BIND(L_copy_8_bytes); 1180 __ subl(count, 2<<shift); 1181 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1182 __ addl(count, 2<<shift); 1183 if (!UseXMMForArrayCopy) { 1184 __ emms(); 1185 } 1186 } 1187 __ BIND(L_copy_4_bytes); 1188 // copy prefix qword 1189 __ testl(count, 1<<shift); 1190 __ jccb(Assembler::zero, L_copy_2_bytes); 1191 __ movl(rdx, Address(from, count, sf, -4)); 1192 __ movl(Address(to, count, sf, -4), rdx); 1193 1194 if (t == T_BYTE || t == T_SHORT) { 1195 __ subl(count, (1<<shift)); 1196 __ BIND(L_copy_2_bytes); 1197 // copy prefix dword 1198 __ testl(count, 1<<(shift-1)); 1199 __ jccb(Assembler::zero, L_copy_byte); 1200 __ movw(rdx, Address(from, count, sf, -2)); 1201 __ movw(Address(to, count, sf, -2), rdx); 1202 if (t == T_BYTE) { 1203 __ subl(count, 1<<(shift-1)); 1204 __ BIND(L_copy_byte); 1205 // copy prefix byte 1206 __ testl(count, 1); 1207 __ jccb(Assembler::zero, L_exit); 1208 __ movb(rdx, Address(from, 0)); 1209 __ movb(Address(to, 0), rdx); 1210 __ BIND(L_exit); 1211 } else { 1212 __ BIND(L_copy_byte); 1213 } 1214 } else { 1215 __ BIND(L_copy_2_bytes); 1216 } 1217 if (t == T_OBJECT) { 1218 __ movl2ptr(count, Address(rsp, 12+12)); // reread count 1219 gen_write_ref_array_post_barrier(to, count); 1220 __ BIND(L_0_count); 1221 } 1222 inc_copy_counter_np(t); 1223 __ pop(rdi); 1224 __ pop(rsi); 1225 __ leave(); // required for proper stackwalking of RuntimeStub frame 1226 __ xorptr(rax, rax); // return 0 1227 __ ret(0); 1228 return start; 1229 } 1230 1231 1232 address generate_disjoint_long_copy(address* entry, const char *name) { 1233 __ align(CodeEntryAlignment); 1234 StubCodeMark mark(this, "StubRoutines", name); 1235 address start = __ pc(); 1236 1237 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1238 const Register from = rax; // source array address 1239 const Register to = rdx; // destination array address 1240 const Register count = rcx; // elements count 1241 const Register to_from = rdx; // (to - from) 1242 1243 __ enter(); // required for proper stackwalking of RuntimeStub frame 1244 __ movptr(from , Address(rsp, 8+0)); // from 1245 __ movptr(to , Address(rsp, 8+4)); // to 1246 __ movl2ptr(count, Address(rsp, 8+8)); // count 1247 1248 *entry = __ pc(); // Entry point from conjoint arraycopy stub. 1249 BLOCK_COMMENT("Entry:"); 1250 1251 __ subptr(to, from); // to --> to_from 1252 if (VM_Version::supports_mmx()) { 1253 if (UseXMMForArrayCopy) { 1254 xmm_copy_forward(from, to_from, count); 1255 } else { 1256 mmx_copy_forward(from, to_from, count); 1257 } 1258 } else { 1259 __ jmpb(L_copy_8_bytes); 1260 __ align(OptoLoopAlignment); 1261 __ BIND(L_copy_8_bytes_loop); 1262 __ fild_d(Address(from, 0)); 1263 __ fistp_d(Address(from, to_from, Address::times_1)); 1264 __ addptr(from, 8); 1265 __ BIND(L_copy_8_bytes); 1266 __ decrement(count); 1267 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1268 } 1269 inc_copy_counter_np(T_LONG); 1270 __ leave(); // required for proper stackwalking of RuntimeStub frame 1271 __ xorptr(rax, rax); // return 0 1272 __ ret(0); 1273 return start; 1274 } 1275 1276 address generate_conjoint_long_copy(address nooverlap_target, 1277 address* entry, const char *name) { 1278 __ align(CodeEntryAlignment); 1279 StubCodeMark mark(this, "StubRoutines", name); 1280 address start = __ pc(); 1281 1282 Label L_copy_8_bytes, L_copy_8_bytes_loop; 1283 const Register from = rax; // source array address 1284 const Register to = rdx; // destination array address 1285 const Register count = rcx; // elements count 1286 const Register end_from = rax; // source array end address 1287 1288 __ enter(); // required for proper stackwalking of RuntimeStub frame 1289 __ movptr(from , Address(rsp, 8+0)); // from 1290 __ movptr(to , Address(rsp, 8+4)); // to 1291 __ movl2ptr(count, Address(rsp, 8+8)); // count 1292 1293 *entry = __ pc(); // Entry point from generic arraycopy stub. 1294 BLOCK_COMMENT("Entry:"); 1295 1296 // arrays overlap test 1297 __ cmpptr(to, from); 1298 RuntimeAddress nooverlap(nooverlap_target); 1299 __ jump_cc(Assembler::belowEqual, nooverlap); 1300 __ lea(end_from, Address(from, count, Address::times_8, 0)); 1301 __ cmpptr(to, end_from); 1302 __ movptr(from, Address(rsp, 8)); // from 1303 __ jump_cc(Assembler::aboveEqual, nooverlap); 1304 1305 __ jmpb(L_copy_8_bytes); 1306 1307 __ align(OptoLoopAlignment); 1308 __ BIND(L_copy_8_bytes_loop); 1309 if (VM_Version::supports_mmx()) { 1310 if (UseXMMForArrayCopy) { 1311 __ movq(xmm0, Address(from, count, Address::times_8)); 1312 __ movq(Address(to, count, Address::times_8), xmm0); 1313 } else { 1314 __ movq(mmx0, Address(from, count, Address::times_8)); 1315 __ movq(Address(to, count, Address::times_8), mmx0); 1316 } 1317 } else { 1318 __ fild_d(Address(from, count, Address::times_8)); 1319 __ fistp_d(Address(to, count, Address::times_8)); 1320 } 1321 __ BIND(L_copy_8_bytes); 1322 __ decrement(count); 1323 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); 1324 1325 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) { 1326 __ emms(); 1327 } 1328 inc_copy_counter_np(T_LONG); 1329 __ leave(); // required for proper stackwalking of RuntimeStub frame 1330 __ xorptr(rax, rax); // return 0 1331 __ ret(0); 1332 return start; 1333 } 1334 1335 1336 // Helper for generating a dynamic type check. 1337 // The sub_klass must be one of {rbx, rdx, rsi}. 1338 // The temp is killed. 1339 void generate_type_check(Register sub_klass, 1340 Address& super_check_offset_addr, 1341 Address& super_klass_addr, 1342 Register temp, 1343 Label* L_success, Label* L_failure) { 1344 BLOCK_COMMENT("type_check:"); 1345 1346 Label L_fallthrough; 1347 #define LOCAL_JCC(assembler_con, label_ptr) \ 1348 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ 1349 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ 1350 1351 // The following is a strange variation of the fast path which requires 1352 // one less register, because needed values are on the argument stack. 1353 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, 1354 // L_success, L_failure, NULL); 1355 assert_different_registers(sub_klass, temp); 1356 1357 int sc_offset = in_bytes(Klass::secondary_super_cache_offset()); 1358 1359 // if the pointers are equal, we are done (e.g., String[] elements) 1360 __ cmpptr(sub_klass, super_klass_addr); 1361 LOCAL_JCC(Assembler::equal, L_success); 1362 1363 // check the supertype display: 1364 __ movl2ptr(temp, super_check_offset_addr); 1365 Address super_check_addr(sub_klass, temp, Address::times_1, 0); 1366 __ movptr(temp, super_check_addr); // load displayed supertype 1367 __ cmpptr(temp, super_klass_addr); // test the super type 1368 LOCAL_JCC(Assembler::equal, L_success); 1369 1370 // if it was a primary super, we can just fail immediately 1371 __ cmpl(super_check_offset_addr, sc_offset); 1372 LOCAL_JCC(Assembler::notEqual, L_failure); 1373 1374 // The repne_scan instruction uses fixed registers, which will get spilled. 1375 // We happen to know this works best when super_klass is in rax. 1376 Register super_klass = temp; 1377 __ movptr(super_klass, super_klass_addr); 1378 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, 1379 L_success, L_failure); 1380 1381 __ bind(L_fallthrough); 1382 1383 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } 1384 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } 1385 1386 #undef LOCAL_JCC 1387 } 1388 1389 // 1390 // Generate checkcasting array copy stub 1391 // 1392 // Input: 1393 // 4(rsp) - source array address 1394 // 8(rsp) - destination array address 1395 // 12(rsp) - element count, can be zero 1396 // 16(rsp) - size_t ckoff (super_check_offset) 1397 // 20(rsp) - oop ckval (super_klass) 1398 // 1399 // Output: 1400 // rax, == 0 - success 1401 // rax, == -1^K - failure, where K is partial transfer count 1402 // 1403 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) { 1404 __ align(CodeEntryAlignment); 1405 StubCodeMark mark(this, "StubRoutines", name); 1406 address start = __ pc(); 1407 1408 Label L_load_element, L_store_element, L_do_card_marks, L_done; 1409 1410 // register use: 1411 // rax, rdx, rcx -- loop control (end_from, end_to, count) 1412 // rdi, rsi -- element access (oop, klass) 1413 // rbx, -- temp 1414 const Register from = rax; // source array address 1415 const Register to = rdx; // destination array address 1416 const Register length = rcx; // elements count 1417 const Register elem = rdi; // each oop copied 1418 const Register elem_klass = rsi; // each elem._klass (sub_klass) 1419 const Register temp = rbx; // lone remaining temp 1420 1421 __ enter(); // required for proper stackwalking of RuntimeStub frame 1422 1423 __ push(rsi); 1424 __ push(rdi); 1425 __ push(rbx); 1426 1427 Address from_arg(rsp, 16+ 4); // from 1428 Address to_arg(rsp, 16+ 8); // to 1429 Address length_arg(rsp, 16+12); // elements count 1430 Address ckoff_arg(rsp, 16+16); // super_check_offset 1431 Address ckval_arg(rsp, 16+20); // super_klass 1432 1433 // Load up: 1434 __ movptr(from, from_arg); 1435 __ movptr(to, to_arg); 1436 __ movl2ptr(length, length_arg); 1437 1438 if (entry != NULL) { 1439 *entry = __ pc(); // Entry point from generic arraycopy stub. 1440 BLOCK_COMMENT("Entry:"); 1441 } 1442 1443 //--------------------------------------------------------------- 1444 // Assembler stub will be used for this call to arraycopy 1445 // if the two arrays are subtypes of Object[] but the 1446 // destination array type is not equal to or a supertype 1447 // of the source type. Each element must be separately 1448 // checked. 1449 1450 // Loop-invariant addresses. They are exclusive end pointers. 1451 Address end_from_addr(from, length, Address::times_ptr, 0); 1452 Address end_to_addr(to, length, Address::times_ptr, 0); 1453 1454 Register end_from = from; // re-use 1455 Register end_to = to; // re-use 1456 Register count = length; // re-use 1457 1458 // Loop-variant addresses. They assume post-incremented count < 0. 1459 Address from_element_addr(end_from, count, Address::times_ptr, 0); 1460 Address to_element_addr(end_to, count, Address::times_ptr, 0); 1461 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); 1462 1463 // Copy from low to high addresses, indexed from the end of each array. 1464 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); 1465 __ lea(end_from, end_from_addr); 1466 __ lea(end_to, end_to_addr); 1467 assert(length == count, ""); // else fix next line: 1468 __ negptr(count); // negate and test the length 1469 __ jccb(Assembler::notZero, L_load_element); 1470 1471 // Empty array: Nothing to do. 1472 __ xorptr(rax, rax); // return 0 on (trivial) success 1473 __ jmp(L_done); 1474 1475 // ======== begin loop ======== 1476 // (Loop is rotated; its entry is L_load_element.) 1477 // Loop control: 1478 // for (count = -count; count != 0; count++) 1479 // Base pointers src, dst are biased by 8*count,to last element. 1480 __ align(OptoLoopAlignment); 1481 1482 __ BIND(L_store_element); 1483 __ movptr(to_element_addr, elem); // store the oop 1484 __ increment(count); // increment the count toward zero 1485 __ jccb(Assembler::zero, L_do_card_marks); 1486 1487 // ======== loop entry is here ======== 1488 __ BIND(L_load_element); 1489 __ movptr(elem, from_element_addr); // load the oop 1490 __ testptr(elem, elem); 1491 __ jccb(Assembler::zero, L_store_element); 1492 1493 // (Could do a trick here: Remember last successful non-null 1494 // element stored and make a quick oop equality check on it.) 1495 1496 __ movptr(elem_klass, elem_klass_addr); // query the object klass 1497 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, 1498 &L_store_element, NULL); 1499 // (On fall-through, we have failed the element type check.) 1500 // ======== end loop ======== 1501 1502 // It was a real error; we must depend on the caller to finish the job. 1503 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. 1504 // Emit GC store barriers for the oops we have copied (length_arg + count), 1505 // and report their number to the caller. 1506 assert_different_registers(to, count, rax); 1507 Label L_post_barrier; 1508 __ addl(count, length_arg); // transfers = (length - remaining) 1509 __ movl2ptr(rax, count); // save the value 1510 __ notptr(rax); // report (-1^K) to caller (does not affect flags) 1511 __ jccb(Assembler::notZero, L_post_barrier); 1512 __ jmp(L_done); // K == 0, nothing was copied, skip post barrier 1513 1514 // Come here on success only. 1515 __ BIND(L_do_card_marks); 1516 __ xorptr(rax, rax); // return 0 on success 1517 __ movl2ptr(count, length_arg); 1518 1519 __ BIND(L_post_barrier); 1520 __ movptr(to, to_arg); // reload 1521 gen_write_ref_array_post_barrier(to, count); 1522 1523 // Common exit point (success or failure). 1524 __ BIND(L_done); 1525 __ pop(rbx); 1526 __ pop(rdi); 1527 __ pop(rsi); 1528 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); 1529 __ leave(); // required for proper stackwalking of RuntimeStub frame 1530 __ ret(0); 1531 1532 return start; 1533 } 1534 1535 // 1536 // Generate 'unsafe' array copy stub 1537 // Though just as safe as the other stubs, it takes an unscaled 1538 // size_t argument instead of an element count. 1539 // 1540 // Input: 1541 // 4(rsp) - source array address 1542 // 8(rsp) - destination array address 1543 // 12(rsp) - byte count, can be zero 1544 // 1545 // Output: 1546 // rax, == 0 - success 1547 // rax, == -1 - need to call System.arraycopy 1548 // 1549 // Examines the alignment of the operands and dispatches 1550 // to a long, int, short, or byte copy loop. 1551 // 1552 address generate_unsafe_copy(const char *name, 1553 address byte_copy_entry, 1554 address short_copy_entry, 1555 address int_copy_entry, 1556 address long_copy_entry) { 1557 1558 Label L_long_aligned, L_int_aligned, L_short_aligned; 1559 1560 __ align(CodeEntryAlignment); 1561 StubCodeMark mark(this, "StubRoutines", name); 1562 address start = __ pc(); 1563 1564 const Register from = rax; // source array address 1565 const Register to = rdx; // destination array address 1566 const Register count = rcx; // elements count 1567 1568 __ enter(); // required for proper stackwalking of RuntimeStub frame 1569 __ push(rsi); 1570 __ push(rdi); 1571 Address from_arg(rsp, 12+ 4); // from 1572 Address to_arg(rsp, 12+ 8); // to 1573 Address count_arg(rsp, 12+12); // byte count 1574 1575 // Load up: 1576 __ movptr(from , from_arg); 1577 __ movptr(to , to_arg); 1578 __ movl2ptr(count, count_arg); 1579 1580 // bump this on entry, not on exit: 1581 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); 1582 1583 const Register bits = rsi; 1584 __ mov(bits, from); 1585 __ orptr(bits, to); 1586 __ orptr(bits, count); 1587 1588 __ testl(bits, BytesPerLong-1); 1589 __ jccb(Assembler::zero, L_long_aligned); 1590 1591 __ testl(bits, BytesPerInt-1); 1592 __ jccb(Assembler::zero, L_int_aligned); 1593 1594 __ testl(bits, BytesPerShort-1); 1595 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); 1596 1597 __ BIND(L_short_aligned); 1598 __ shrptr(count, LogBytesPerShort); // size => short_count 1599 __ movl(count_arg, count); // update 'count' 1600 __ jump(RuntimeAddress(short_copy_entry)); 1601 1602 __ BIND(L_int_aligned); 1603 __ shrptr(count, LogBytesPerInt); // size => int_count 1604 __ movl(count_arg, count); // update 'count' 1605 __ jump(RuntimeAddress(int_copy_entry)); 1606 1607 __ BIND(L_long_aligned); 1608 __ shrptr(count, LogBytesPerLong); // size => qword_count 1609 __ movl(count_arg, count); // update 'count' 1610 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1611 __ pop(rsi); 1612 __ jump(RuntimeAddress(long_copy_entry)); 1613 1614 return start; 1615 } 1616 1617 1618 // Perform range checks on the proposed arraycopy. 1619 // Smashes src_pos and dst_pos. (Uses them up for temps.) 1620 void arraycopy_range_checks(Register src, 1621 Register src_pos, 1622 Register dst, 1623 Register dst_pos, 1624 Address& length, 1625 Label& L_failed) { 1626 BLOCK_COMMENT("arraycopy_range_checks:"); 1627 const Register src_end = src_pos; // source array end position 1628 const Register dst_end = dst_pos; // destination array end position 1629 __ addl(src_end, length); // src_pos + length 1630 __ addl(dst_end, length); // dst_pos + length 1631 1632 // if (src_pos + length > arrayOop(src)->length() ) FAIL; 1633 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); 1634 __ jcc(Assembler::above, L_failed); 1635 1636 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; 1637 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); 1638 __ jcc(Assembler::above, L_failed); 1639 1640 BLOCK_COMMENT("arraycopy_range_checks done"); 1641 } 1642 1643 1644 // 1645 // Generate generic array copy stubs 1646 // 1647 // Input: 1648 // 4(rsp) - src oop 1649 // 8(rsp) - src_pos 1650 // 12(rsp) - dst oop 1651 // 16(rsp) - dst_pos 1652 // 20(rsp) - element count 1653 // 1654 // Output: 1655 // rax, == 0 - success 1656 // rax, == -1^K - failure, where K is partial transfer count 1657 // 1658 address generate_generic_copy(const char *name, 1659 address entry_jbyte_arraycopy, 1660 address entry_jshort_arraycopy, 1661 address entry_jint_arraycopy, 1662 address entry_oop_arraycopy, 1663 address entry_jlong_arraycopy, 1664 address entry_checkcast_arraycopy) { 1665 Label L_failed, L_failed_0, L_objArray; 1666 1667 { int modulus = CodeEntryAlignment; 1668 int target = modulus - 5; // 5 = sizeof jmp(L_failed) 1669 int advance = target - (__ offset() % modulus); 1670 if (advance < 0) advance += modulus; 1671 if (advance > 0) __ nop(advance); 1672 } 1673 StubCodeMark mark(this, "StubRoutines", name); 1674 1675 // Short-hop target to L_failed. Makes for denser prologue code. 1676 __ BIND(L_failed_0); 1677 __ jmp(L_failed); 1678 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); 1679 1680 __ align(CodeEntryAlignment); 1681 address start = __ pc(); 1682 1683 __ enter(); // required for proper stackwalking of RuntimeStub frame 1684 __ push(rsi); 1685 __ push(rdi); 1686 1687 // bump this on entry, not on exit: 1688 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); 1689 1690 // Input values 1691 Address SRC (rsp, 12+ 4); 1692 Address SRC_POS (rsp, 12+ 8); 1693 Address DST (rsp, 12+12); 1694 Address DST_POS (rsp, 12+16); 1695 Address LENGTH (rsp, 12+20); 1696 1697 //----------------------------------------------------------------------- 1698 // Assembler stub will be used for this call to arraycopy 1699 // if the following conditions are met: 1700 // 1701 // (1) src and dst must not be null. 1702 // (2) src_pos must not be negative. 1703 // (3) dst_pos must not be negative. 1704 // (4) length must not be negative. 1705 // (5) src klass and dst klass should be the same and not NULL. 1706 // (6) src and dst should be arrays. 1707 // (7) src_pos + length must not exceed length of src. 1708 // (8) dst_pos + length must not exceed length of dst. 1709 // 1710 1711 const Register src = rax; // source array oop 1712 const Register src_pos = rsi; 1713 const Register dst = rdx; // destination array oop 1714 const Register dst_pos = rdi; 1715 const Register length = rcx; // transfer count 1716 1717 // if (src == NULL) return -1; 1718 __ movptr(src, SRC); // src oop 1719 __ testptr(src, src); 1720 __ jccb(Assembler::zero, L_failed_0); 1721 1722 // if (src_pos < 0) return -1; 1723 __ movl2ptr(src_pos, SRC_POS); // src_pos 1724 __ testl(src_pos, src_pos); 1725 __ jccb(Assembler::negative, L_failed_0); 1726 1727 // if (dst == NULL) return -1; 1728 __ movptr(dst, DST); // dst oop 1729 __ testptr(dst, dst); 1730 __ jccb(Assembler::zero, L_failed_0); 1731 1732 // if (dst_pos < 0) return -1; 1733 __ movl2ptr(dst_pos, DST_POS); // dst_pos 1734 __ testl(dst_pos, dst_pos); 1735 __ jccb(Assembler::negative, L_failed_0); 1736 1737 // if (length < 0) return -1; 1738 __ movl2ptr(length, LENGTH); // length 1739 __ testl(length, length); 1740 __ jccb(Assembler::negative, L_failed_0); 1741 1742 // if (src->klass() == NULL) return -1; 1743 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); 1744 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); 1745 const Register rcx_src_klass = rcx; // array klass 1746 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); 1747 1748 #ifdef ASSERT 1749 // assert(src->klass() != NULL); 1750 BLOCK_COMMENT("assert klasses not null"); 1751 { Label L1, L2; 1752 __ testptr(rcx_src_klass, rcx_src_klass); 1753 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL 1754 __ bind(L1); 1755 __ stop("broken null klass"); 1756 __ bind(L2); 1757 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); 1758 __ jccb(Assembler::equal, L1); // this would be broken also 1759 BLOCK_COMMENT("assert done"); 1760 } 1761 #endif //ASSERT 1762 1763 // Load layout helper (32-bits) 1764 // 1765 // |array_tag| | header_size | element_type | |log2_element_size| 1766 // 32 30 24 16 8 2 0 1767 // 1768 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 1769 // 1770 1771 int lh_offset = in_bytes(Klass::layout_helper_offset()); 1772 Address src_klass_lh_addr(rcx_src_klass, lh_offset); 1773 1774 // Handle objArrays completely differently... 1775 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); 1776 __ cmpl(src_klass_lh_addr, objArray_lh); 1777 __ jcc(Assembler::equal, L_objArray); 1778 1779 // if (src->klass() != dst->klass()) return -1; 1780 __ cmpptr(rcx_src_klass, dst_klass_addr); 1781 __ jccb(Assembler::notEqual, L_failed_0); 1782 1783 const Register rcx_lh = rcx; // layout helper 1784 assert(rcx_lh == rcx_src_klass, "known alias"); 1785 __ movl(rcx_lh, src_klass_lh_addr); 1786 1787 // if (!src->is_Array()) return -1; 1788 __ cmpl(rcx_lh, Klass::_lh_neutral_value); 1789 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp 1790 1791 // At this point, it is known to be a typeArray (array_tag 0x3). 1792 #ifdef ASSERT 1793 { Label L; 1794 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); 1795 __ jcc(Assembler::greaterEqual, L); // signed cmp 1796 __ stop("must be a primitive array"); 1797 __ bind(L); 1798 } 1799 #endif 1800 1801 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); 1802 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1803 1804 // TypeArrayKlass 1805 // 1806 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); 1807 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); 1808 // 1809 const Register rsi_offset = rsi; // array offset 1810 const Register src_array = src; // src array offset 1811 const Register dst_array = dst; // dst array offset 1812 const Register rdi_elsize = rdi; // log2 element size 1813 1814 __ mov(rsi_offset, rcx_lh); 1815 __ shrptr(rsi_offset, Klass::_lh_header_size_shift); 1816 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset 1817 __ addptr(src_array, rsi_offset); // src array offset 1818 __ addptr(dst_array, rsi_offset); // dst array offset 1819 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize 1820 1821 // next registers should be set before the jump to corresponding stub 1822 const Register from = src; // source array address 1823 const Register to = dst; // destination array address 1824 const Register count = rcx; // elements count 1825 // some of them should be duplicated on stack 1826 #define FROM Address(rsp, 12+ 4) 1827 #define TO Address(rsp, 12+ 8) // Not used now 1828 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy 1829 1830 BLOCK_COMMENT("scale indexes to element size"); 1831 __ movl2ptr(rsi, SRC_POS); // src_pos 1832 __ shlptr(rsi); // src_pos << rcx (log2 elsize) 1833 assert(src_array == from, ""); 1834 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize 1835 __ movl2ptr(rdi, DST_POS); // dst_pos 1836 __ shlptr(rdi); // dst_pos << rcx (log2 elsize) 1837 assert(dst_array == to, ""); 1838 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize 1839 __ movptr(FROM, from); // src_addr 1840 __ mov(rdi_elsize, rcx_lh); // log2 elsize 1841 __ movl2ptr(count, LENGTH); // elements count 1842 1843 BLOCK_COMMENT("choose copy loop based on element size"); 1844 __ cmpl(rdi_elsize, 0); 1845 1846 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); 1847 __ cmpl(rdi_elsize, LogBytesPerShort); 1848 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); 1849 __ cmpl(rdi_elsize, LogBytesPerInt); 1850 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); 1851 #ifdef ASSERT 1852 __ cmpl(rdi_elsize, LogBytesPerLong); 1853 __ jccb(Assembler::notEqual, L_failed); 1854 #endif 1855 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. 1856 __ pop(rsi); 1857 __ jump(RuntimeAddress(entry_jlong_arraycopy)); 1858 1859 __ BIND(L_failed); 1860 __ xorptr(rax, rax); 1861 __ notptr(rax); // return -1 1862 __ pop(rdi); 1863 __ pop(rsi); 1864 __ leave(); // required for proper stackwalking of RuntimeStub frame 1865 __ ret(0); 1866 1867 // ObjArrayKlass 1868 __ BIND(L_objArray); 1869 // live at this point: rcx_src_klass, src[_pos], dst[_pos] 1870 1871 Label L_plain_copy, L_checkcast_copy; 1872 // test array classes for subtyping 1873 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality 1874 __ jccb(Assembler::notEqual, L_checkcast_copy); 1875 1876 // Identically typed arrays can be copied without element-wise checks. 1877 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); 1878 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1879 1880 __ BIND(L_plain_copy); 1881 __ movl2ptr(count, LENGTH); // elements count 1882 __ movl2ptr(src_pos, SRC_POS); // reload src_pos 1883 __ lea(from, Address(src, src_pos, Address::times_ptr, 1884 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr 1885 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos 1886 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1887 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr 1888 __ movptr(FROM, from); // src_addr 1889 __ movptr(TO, to); // dst_addr 1890 __ movl(COUNT, count); // count 1891 __ jump(RuntimeAddress(entry_oop_arraycopy)); 1892 1893 __ BIND(L_checkcast_copy); 1894 // live at this point: rcx_src_klass, dst[_pos], src[_pos] 1895 { 1896 // Handy offsets: 1897 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset()); 1898 int sco_offset = in_bytes(Klass::super_check_offset_offset()); 1899 1900 Register rsi_dst_klass = rsi; 1901 Register rdi_temp = rdi; 1902 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); 1903 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); 1904 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); 1905 1906 // Before looking at dst.length, make sure dst is also an objArray. 1907 __ movptr(rsi_dst_klass, dst_klass_addr); 1908 __ cmpl(dst_klass_lh_addr, objArray_lh); 1909 __ jccb(Assembler::notEqual, L_failed); 1910 1911 // It is safe to examine both src.length and dst.length. 1912 __ movl2ptr(src_pos, SRC_POS); // reload rsi 1913 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); 1914 // (Now src_pos and dst_pos are killed, but not src and dst.) 1915 1916 // We'll need this temp (don't forget to pop it after the type check). 1917 __ push(rbx); 1918 Register rbx_src_klass = rbx; 1919 1920 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx 1921 __ movptr(rsi_dst_klass, dst_klass_addr); 1922 Address super_check_offset_addr(rsi_dst_klass, sco_offset); 1923 Label L_fail_array_check; 1924 generate_type_check(rbx_src_klass, 1925 super_check_offset_addr, dst_klass_addr, 1926 rdi_temp, NULL, &L_fail_array_check); 1927 // (On fall-through, we have passed the array type check.) 1928 __ pop(rbx); 1929 __ jmp(L_plain_copy); 1930 1931 __ BIND(L_fail_array_check); 1932 // Reshuffle arguments so we can call checkcast_arraycopy: 1933 1934 // match initial saves for checkcast_arraycopy 1935 // push(rsi); // already done; see above 1936 // push(rdi); // already done; see above 1937 // push(rbx); // already done; see above 1938 1939 // Marshal outgoing arguments now, freeing registers. 1940 Address from_arg(rsp, 16+ 4); // from 1941 Address to_arg(rsp, 16+ 8); // to 1942 Address length_arg(rsp, 16+12); // elements count 1943 Address ckoff_arg(rsp, 16+16); // super_check_offset 1944 Address ckval_arg(rsp, 16+20); // super_klass 1945 1946 Address SRC_POS_arg(rsp, 16+ 8); 1947 Address DST_POS_arg(rsp, 16+16); 1948 Address LENGTH_arg(rsp, 16+20); 1949 // push rbx, changed the incoming offsets (why not just use rbp,??) 1950 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); 1951 1952 __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); 1953 __ movl2ptr(length, LENGTH_arg); // reload elements count 1954 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos 1955 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos 1956 1957 __ movptr(ckval_arg, rbx); // destination element type 1958 __ movl(rbx, Address(rbx, sco_offset)); 1959 __ movl(ckoff_arg, rbx); // corresponding class check offset 1960 1961 __ movl(length_arg, length); // outgoing length argument 1962 1963 __ lea(from, Address(src, src_pos, Address::times_ptr, 1964 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1965 __ movptr(from_arg, from); 1966 1967 __ lea(to, Address(dst, dst_pos, Address::times_ptr, 1968 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); 1969 __ movptr(to_arg, to); 1970 __ jump(RuntimeAddress(entry_checkcast_arraycopy)); 1971 } 1972 1973 return start; 1974 } 1975 1976 void generate_arraycopy_stubs() { 1977 address entry; 1978 address entry_jbyte_arraycopy; 1979 address entry_jshort_arraycopy; 1980 address entry_jint_arraycopy; 1981 address entry_oop_arraycopy; 1982 address entry_jlong_arraycopy; 1983 address entry_checkcast_arraycopy; 1984 1985 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = 1986 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, 1987 "arrayof_jbyte_disjoint_arraycopy"); 1988 StubRoutines::_arrayof_jbyte_arraycopy = 1989 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, 1990 NULL, "arrayof_jbyte_arraycopy"); 1991 StubRoutines::_jbyte_disjoint_arraycopy = 1992 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, 1993 "jbyte_disjoint_arraycopy"); 1994 StubRoutines::_jbyte_arraycopy = 1995 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, 1996 &entry_jbyte_arraycopy, "jbyte_arraycopy"); 1997 1998 StubRoutines::_arrayof_jshort_disjoint_arraycopy = 1999 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, 2000 "arrayof_jshort_disjoint_arraycopy"); 2001 StubRoutines::_arrayof_jshort_arraycopy = 2002 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, 2003 NULL, "arrayof_jshort_arraycopy"); 2004 StubRoutines::_jshort_disjoint_arraycopy = 2005 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, 2006 "jshort_disjoint_arraycopy"); 2007 StubRoutines::_jshort_arraycopy = 2008 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, 2009 &entry_jshort_arraycopy, "jshort_arraycopy"); 2010 2011 // Next arrays are always aligned on 4 bytes at least. 2012 StubRoutines::_jint_disjoint_arraycopy = 2013 generate_disjoint_copy(T_INT, true, Address::times_4, &entry, 2014 "jint_disjoint_arraycopy"); 2015 StubRoutines::_jint_arraycopy = 2016 generate_conjoint_copy(T_INT, true, Address::times_4, entry, 2017 &entry_jint_arraycopy, "jint_arraycopy"); 2018 2019 StubRoutines::_oop_disjoint_arraycopy = 2020 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, 2021 "oop_disjoint_arraycopy"); 2022 StubRoutines::_oop_arraycopy = 2023 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, 2024 &entry_oop_arraycopy, "oop_arraycopy"); 2025 2026 StubRoutines::_oop_disjoint_arraycopy_uninit = 2027 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, 2028 "oop_disjoint_arraycopy_uninit", 2029 /*dest_uninitialized*/true); 2030 StubRoutines::_oop_arraycopy_uninit = 2031 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, 2032 NULL, "oop_arraycopy_uninit", 2033 /*dest_uninitialized*/true); 2034 2035 StubRoutines::_jlong_disjoint_arraycopy = 2036 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); 2037 StubRoutines::_jlong_arraycopy = 2038 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, 2039 "jlong_arraycopy"); 2040 2041 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); 2042 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); 2043 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); 2044 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); 2045 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); 2046 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); 2047 2048 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; 2049 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; 2050 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; 2051 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; 2052 2053 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; 2054 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; 2055 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; 2056 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; 2057 2058 StubRoutines::_checkcast_arraycopy = 2059 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); 2060 StubRoutines::_checkcast_arraycopy_uninit = 2061 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true); 2062 2063 StubRoutines::_unsafe_arraycopy = 2064 generate_unsafe_copy("unsafe_arraycopy", 2065 entry_jbyte_arraycopy, 2066 entry_jshort_arraycopy, 2067 entry_jint_arraycopy, 2068 entry_jlong_arraycopy); 2069 2070 StubRoutines::_generic_arraycopy = 2071 generate_generic_copy("generic_arraycopy", 2072 entry_jbyte_arraycopy, 2073 entry_jshort_arraycopy, 2074 entry_jint_arraycopy, 2075 entry_oop_arraycopy, 2076 entry_jlong_arraycopy, 2077 entry_checkcast_arraycopy); 2078 } 2079 2080 void generate_math_stubs() { 2081 { 2082 StubCodeMark mark(this, "StubRoutines", "log"); 2083 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); 2084 2085 __ fld_d(Address(rsp, 4)); 2086 __ flog(); 2087 __ ret(0); 2088 } 2089 { 2090 StubCodeMark mark(this, "StubRoutines", "log10"); 2091 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); 2092 2093 __ fld_d(Address(rsp, 4)); 2094 __ flog10(); 2095 __ ret(0); 2096 } 2097 { 2098 StubCodeMark mark(this, "StubRoutines", "sin"); 2099 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); 2100 2101 __ fld_d(Address(rsp, 4)); 2102 __ trigfunc('s'); 2103 __ ret(0); 2104 } 2105 { 2106 StubCodeMark mark(this, "StubRoutines", "cos"); 2107 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); 2108 2109 __ fld_d(Address(rsp, 4)); 2110 __ trigfunc('c'); 2111 __ ret(0); 2112 } 2113 { 2114 StubCodeMark mark(this, "StubRoutines", "tan"); 2115 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); 2116 2117 __ fld_d(Address(rsp, 4)); 2118 __ trigfunc('t'); 2119 __ ret(0); 2120 } 2121 { 2122 StubCodeMark mark(this, "StubRoutines", "exp"); 2123 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc(); 2124 2125 __ fld_d(Address(rsp, 4)); 2126 __ exp_with_fallback(0); 2127 __ ret(0); 2128 } 2129 { 2130 StubCodeMark mark(this, "StubRoutines", "pow"); 2131 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc(); 2132 2133 __ fld_d(Address(rsp, 12)); 2134 __ fld_d(Address(rsp, 4)); 2135 __ pow_with_fallback(0); 2136 __ ret(0); 2137 } 2138 } 2139 2140 // AES intrinsic stubs 2141 enum {AESBlockSize = 16}; 2142 2143 address generate_key_shuffle_mask() { 2144 __ align(16); 2145 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask"); 2146 address start = __ pc(); 2147 __ emit_data(0x00010203, relocInfo::none, 0 ); 2148 __ emit_data(0x04050607, relocInfo::none, 0 ); 2149 __ emit_data(0x08090a0b, relocInfo::none, 0 ); 2150 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 ); 2151 return start; 2152 } 2153 2154 // Utility routine for loading a 128-bit key word in little endian format 2155 // can optionally specify that the shuffle mask is already in an xmmregister 2156 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { 2157 __ movdqu(xmmdst, Address(key, offset)); 2158 if (xmm_shuf_mask != NULL) { 2159 __ pshufb(xmmdst, xmm_shuf_mask); 2160 } else { 2161 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); 2162 } 2163 } 2164 2165 // aesenc using specified key+offset 2166 // can optionally specify that the shuffle mask is already in an xmmregister 2167 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { 2168 load_key(xmmtmp, key, offset, xmm_shuf_mask); 2169 __ aesenc(xmmdst, xmmtmp); 2170 } 2171 2172 // aesdec using specified key+offset 2173 // can optionally specify that the shuffle mask is already in an xmmregister 2174 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { 2175 load_key(xmmtmp, key, offset, xmm_shuf_mask); 2176 __ aesdec(xmmdst, xmmtmp); 2177 } 2178 2179 2180 // Arguments: 2181 // 2182 // Inputs: 2183 // c_rarg0 - source byte array address 2184 // c_rarg1 - destination byte array address 2185 // c_rarg2 - K (key) in little endian int array 2186 // 2187 address generate_aescrypt_encryptBlock() { 2188 assert(UseAES, "need AES instructions and misaligned SSE support"); 2189 __ align(CodeEntryAlignment); 2190 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock"); 2191 Label L_doLast; 2192 address start = __ pc(); 2193 2194 const Register from = rdx; // source array address 2195 const Register to = rdx; // destination array address 2196 const Register key = rcx; // key array address 2197 const Register keylen = rax; 2198 const Address from_param(rbp, 8+0); 2199 const Address to_param (rbp, 8+4); 2200 const Address key_param (rbp, 8+8); 2201 2202 const XMMRegister xmm_result = xmm0; 2203 const XMMRegister xmm_key_shuf_mask = xmm1; 2204 const XMMRegister xmm_temp1 = xmm2; 2205 const XMMRegister xmm_temp2 = xmm3; 2206 const XMMRegister xmm_temp3 = xmm4; 2207 const XMMRegister xmm_temp4 = xmm5; 2208 2209 __ enter(); // required for proper stackwalking of RuntimeStub frame 2210 __ movptr(from, from_param); 2211 __ movptr(key, key_param); 2212 2213 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} 2214 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); 2215 2216 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); 2217 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input 2218 __ movptr(to, to_param); 2219 2220 // For encryption, the java expanded key ordering is just what we need 2221 2222 load_key(xmm_temp1, key, 0x00, xmm_key_shuf_mask); 2223 __ pxor(xmm_result, xmm_temp1); 2224 2225 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); 2226 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); 2227 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); 2228 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); 2229 2230 __ aesenc(xmm_result, xmm_temp1); 2231 __ aesenc(xmm_result, xmm_temp2); 2232 __ aesenc(xmm_result, xmm_temp3); 2233 __ aesenc(xmm_result, xmm_temp4); 2234 2235 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); 2236 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); 2237 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); 2238 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); 2239 2240 __ aesenc(xmm_result, xmm_temp1); 2241 __ aesenc(xmm_result, xmm_temp2); 2242 __ aesenc(xmm_result, xmm_temp3); 2243 __ aesenc(xmm_result, xmm_temp4); 2244 2245 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); 2246 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); 2247 2248 __ cmpl(keylen, 44); 2249 __ jccb(Assembler::equal, L_doLast); 2250 2251 __ aesenc(xmm_result, xmm_temp1); 2252 __ aesenc(xmm_result, xmm_temp2); 2253 2254 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); 2255 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); 2256 2257 __ cmpl(keylen, 52); 2258 __ jccb(Assembler::equal, L_doLast); 2259 2260 __ aesenc(xmm_result, xmm_temp1); 2261 __ aesenc(xmm_result, xmm_temp2); 2262 2263 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); 2264 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); 2265 2266 __ BIND(L_doLast); 2267 __ aesenc(xmm_result, xmm_temp1); 2268 __ aesenclast(xmm_result, xmm_temp2); 2269 __ movdqu(Address(to, 0), xmm_result); // store the result 2270 __ xorptr(rax, rax); // return 0 2271 __ leave(); // required for proper stackwalking of RuntimeStub frame 2272 __ ret(0); 2273 2274 return start; 2275 } 2276 2277 2278 // Arguments: 2279 // 2280 // Inputs: 2281 // c_rarg0 - source byte array address 2282 // c_rarg1 - destination byte array address 2283 // c_rarg2 - K (key) in little endian int array 2284 // 2285 address generate_aescrypt_decryptBlock() { 2286 assert(UseAES, "need AES instructions and misaligned SSE support"); 2287 __ align(CodeEntryAlignment); 2288 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock"); 2289 Label L_doLast; 2290 address start = __ pc(); 2291 2292 const Register from = rdx; // source array address 2293 const Register to = rdx; // destination array address 2294 const Register key = rcx; // key array address 2295 const Register keylen = rax; 2296 const Address from_param(rbp, 8+0); 2297 const Address to_param (rbp, 8+4); 2298 const Address key_param (rbp, 8+8); 2299 2300 const XMMRegister xmm_result = xmm0; 2301 const XMMRegister xmm_key_shuf_mask = xmm1; 2302 const XMMRegister xmm_temp1 = xmm2; 2303 const XMMRegister xmm_temp2 = xmm3; 2304 const XMMRegister xmm_temp3 = xmm4; 2305 const XMMRegister xmm_temp4 = xmm5; 2306 2307 __ enter(); // required for proper stackwalking of RuntimeStub frame 2308 __ movptr(from, from_param); 2309 __ movptr(key, key_param); 2310 2311 // keylen could be only {11, 13, 15} * 4 = {44, 52, 60} 2312 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); 2313 2314 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); 2315 __ movdqu(xmm_result, Address(from, 0)); 2316 __ movptr(to, to_param); 2317 2318 // for decryption java expanded key ordering is rotated one position from what we want 2319 // so we start from 0x10 here and hit 0x00 last 2320 // we don't know if the key is aligned, hence not using load-execute form 2321 load_key(xmm_temp1, key, 0x10, xmm_key_shuf_mask); 2322 load_key(xmm_temp2, key, 0x20, xmm_key_shuf_mask); 2323 load_key(xmm_temp3, key, 0x30, xmm_key_shuf_mask); 2324 load_key(xmm_temp4, key, 0x40, xmm_key_shuf_mask); 2325 2326 __ pxor (xmm_result, xmm_temp1); 2327 __ aesdec(xmm_result, xmm_temp2); 2328 __ aesdec(xmm_result, xmm_temp3); 2329 __ aesdec(xmm_result, xmm_temp4); 2330 2331 load_key(xmm_temp1, key, 0x50, xmm_key_shuf_mask); 2332 load_key(xmm_temp2, key, 0x60, xmm_key_shuf_mask); 2333 load_key(xmm_temp3, key, 0x70, xmm_key_shuf_mask); 2334 load_key(xmm_temp4, key, 0x80, xmm_key_shuf_mask); 2335 2336 __ aesdec(xmm_result, xmm_temp1); 2337 __ aesdec(xmm_result, xmm_temp2); 2338 __ aesdec(xmm_result, xmm_temp3); 2339 __ aesdec(xmm_result, xmm_temp4); 2340 2341 load_key(xmm_temp1, key, 0x90, xmm_key_shuf_mask); 2342 load_key(xmm_temp2, key, 0xa0, xmm_key_shuf_mask); 2343 load_key(xmm_temp3, key, 0x00, xmm_key_shuf_mask); 2344 2345 __ cmpl(keylen, 44); 2346 __ jccb(Assembler::equal, L_doLast); 2347 2348 __ aesdec(xmm_result, xmm_temp1); 2349 __ aesdec(xmm_result, xmm_temp2); 2350 2351 load_key(xmm_temp1, key, 0xb0, xmm_key_shuf_mask); 2352 load_key(xmm_temp2, key, 0xc0, xmm_key_shuf_mask); 2353 2354 __ cmpl(keylen, 52); 2355 __ jccb(Assembler::equal, L_doLast); 2356 2357 __ aesdec(xmm_result, xmm_temp1); 2358 __ aesdec(xmm_result, xmm_temp2); 2359 2360 load_key(xmm_temp1, key, 0xd0, xmm_key_shuf_mask); 2361 load_key(xmm_temp2, key, 0xe0, xmm_key_shuf_mask); 2362 2363 __ BIND(L_doLast); 2364 __ aesdec(xmm_result, xmm_temp1); 2365 __ aesdec(xmm_result, xmm_temp2); 2366 2367 // for decryption the aesdeclast operation is always on key+0x00 2368 __ aesdeclast(xmm_result, xmm_temp3); 2369 __ movdqu(Address(to, 0), xmm_result); // store the result 2370 __ xorptr(rax, rax); // return 0 2371 __ leave(); // required for proper stackwalking of RuntimeStub frame 2372 __ ret(0); 2373 2374 return start; 2375 } 2376 2377 void handleSOERegisters(bool saving) { 2378 const int saveFrameSizeInBytes = 4 * wordSize; 2379 const Address saved_rbx (rbp, -3 * wordSize); 2380 const Address saved_rsi (rbp, -2 * wordSize); 2381 const Address saved_rdi (rbp, -1 * wordSize); 2382 2383 if (saving) { 2384 __ subptr(rsp, saveFrameSizeInBytes); 2385 __ movptr(saved_rsi, rsi); 2386 __ movptr(saved_rdi, rdi); 2387 __ movptr(saved_rbx, rbx); 2388 } else { 2389 // restoring 2390 __ movptr(rsi, saved_rsi); 2391 __ movptr(rdi, saved_rdi); 2392 __ movptr(rbx, saved_rbx); 2393 } 2394 } 2395 2396 // Arguments: 2397 // 2398 // Inputs: 2399 // c_rarg0 - source byte array address 2400 // c_rarg1 - destination byte array address 2401 // c_rarg2 - K (key) in little endian int array 2402 // c_rarg3 - r vector byte array address 2403 // c_rarg4 - input length 2404 // 2405 // Output: 2406 // rax - input length 2407 // 2408 address generate_cipherBlockChaining_encryptAESCrypt() { 2409 assert(UseAES, "need AES instructions and misaligned SSE support"); 2410 __ align(CodeEntryAlignment); 2411 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); 2412 address start = __ pc(); 2413 2414 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256; 2415 const Register from = rsi; // source array address 2416 const Register to = rdx; // destination array address 2417 const Register key = rcx; // key array address 2418 const Register rvec = rdi; // r byte array initialized from initvector array address 2419 // and left with the results of the last encryption block 2420 const Register len_reg = rbx; // src len (must be multiple of blocksize 16) 2421 const Register pos = rax; 2422 2423 // xmm register assignments for the loops below 2424 const XMMRegister xmm_result = xmm0; 2425 const XMMRegister xmm_temp = xmm1; 2426 // first 6 keys preloaded into xmm2-xmm7 2427 const int XMM_REG_NUM_KEY_FIRST = 2; 2428 const int XMM_REG_NUM_KEY_LAST = 7; 2429 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); 2430 2431 __ enter(); // required for proper stackwalking of RuntimeStub frame 2432 handleSOERegisters(true /*saving*/); 2433 2434 // load registers from incoming parameters 2435 const Address from_param(rbp, 8+0); 2436 const Address to_param (rbp, 8+4); 2437 const Address key_param (rbp, 8+8); 2438 const Address rvec_param (rbp, 8+12); 2439 const Address len_param (rbp, 8+16); 2440 __ movptr(from , from_param); 2441 __ movptr(to , to_param); 2442 __ movptr(key , key_param); 2443 __ movptr(rvec , rvec_param); 2444 __ movptr(len_reg , len_param); 2445 2446 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front 2447 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); 2448 // load up xmm regs 2 thru 7 with keys 0-5 2449 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2450 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); 2451 offset += 0x10; 2452 } 2453 2454 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec 2455 2456 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) 2457 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); 2458 __ cmpl(rax, 44); 2459 __ jcc(Assembler::notEqual, L_key_192_256); 2460 2461 // 128 bit code follows here 2462 __ movl(pos, 0); 2463 __ align(OptoLoopAlignment); 2464 __ BIND(L_loopTop_128); 2465 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input 2466 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2467 2468 __ pxor (xmm_result, xmm_key0); // do the aes rounds 2469 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2470 __ aesenc(xmm_result, as_XMMRegister(rnum)); 2471 } 2472 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) { 2473 aes_enc_key(xmm_result, xmm_temp, key, key_offset); 2474 } 2475 load_key(xmm_temp, key, 0xa0); 2476 __ aesenclast(xmm_result, xmm_temp); 2477 2478 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2479 // no need to store r to memory until we exit 2480 __ addptr(pos, AESBlockSize); 2481 __ subptr(len_reg, AESBlockSize); 2482 __ jcc(Assembler::notEqual, L_loopTop_128); 2483 2484 __ BIND(L_exit); 2485 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object 2486 2487 handleSOERegisters(false /*restoring*/); 2488 __ movptr(rax, len_param); // return length 2489 __ leave(); // required for proper stackwalking of RuntimeStub frame 2490 __ ret(0); 2491 2492 __ BIND(L_key_192_256); 2493 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) 2494 __ cmpl(rax, 52); 2495 __ jcc(Assembler::notEqual, L_key_256); 2496 2497 // 192-bit code follows here (could be changed to use more xmm registers) 2498 __ movl(pos, 0); 2499 __ align(OptoLoopAlignment); 2500 __ BIND(L_loopTop_192); 2501 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input 2502 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2503 2504 __ pxor (xmm_result, xmm_key0); // do the aes rounds 2505 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2506 __ aesenc(xmm_result, as_XMMRegister(rnum)); 2507 } 2508 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) { 2509 aes_enc_key(xmm_result, xmm_temp, key, key_offset); 2510 } 2511 load_key(xmm_temp, key, 0xc0); 2512 __ aesenclast(xmm_result, xmm_temp); 2513 2514 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2515 // no need to store r to memory until we exit 2516 __ addptr(pos, AESBlockSize); 2517 __ subptr(len_reg, AESBlockSize); 2518 __ jcc(Assembler::notEqual, L_loopTop_192); 2519 __ jmp(L_exit); 2520 2521 __ BIND(L_key_256); 2522 // 256-bit code follows here (could be changed to use more xmm registers) 2523 __ movl(pos, 0); 2524 __ align(OptoLoopAlignment); 2525 __ BIND(L_loopTop_256); 2526 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input 2527 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2528 2529 __ pxor (xmm_result, xmm_key0); // do the aes rounds 2530 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2531 __ aesenc(xmm_result, as_XMMRegister(rnum)); 2532 } 2533 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) { 2534 aes_enc_key(xmm_result, xmm_temp, key, key_offset); 2535 } 2536 load_key(xmm_temp, key, 0xe0); 2537 __ aesenclast(xmm_result, xmm_temp); 2538 2539 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2540 // no need to store r to memory until we exit 2541 __ addptr(pos, AESBlockSize); 2542 __ subptr(len_reg, AESBlockSize); 2543 __ jcc(Assembler::notEqual, L_loopTop_256); 2544 __ jmp(L_exit); 2545 2546 return start; 2547 } 2548 2549 2550 // CBC AES Decryption. 2551 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time. 2552 // 2553 // Arguments: 2554 // 2555 // Inputs: 2556 // c_rarg0 - source byte array address 2557 // c_rarg1 - destination byte array address 2558 // c_rarg2 - K (key) in little endian int array 2559 // c_rarg3 - r vector byte array address 2560 // c_rarg4 - input length 2561 // 2562 // Output: 2563 // rax - input length 2564 // 2565 2566 address generate_cipherBlockChaining_decryptAESCrypt() { 2567 assert(UseAES, "need AES instructions and misaligned SSE support"); 2568 __ align(CodeEntryAlignment); 2569 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); 2570 address start = __ pc(); 2571 2572 Label L_exit, L_key_192_256, L_key_256; 2573 Label L_singleBlock_loopTop_128; 2574 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256; 2575 const Register from = rsi; // source array address 2576 const Register to = rdx; // destination array address 2577 const Register key = rcx; // key array address 2578 const Register rvec = rdi; // r byte array initialized from initvector array address 2579 // and left with the results of the last encryption block 2580 const Register len_reg = rbx; // src len (must be multiple of blocksize 16) 2581 const Register pos = rax; 2582 2583 // xmm register assignments for the loops below 2584 const XMMRegister xmm_result = xmm0; 2585 const XMMRegister xmm_temp = xmm1; 2586 // first 6 keys preloaded into xmm2-xmm7 2587 const int XMM_REG_NUM_KEY_FIRST = 2; 2588 const int XMM_REG_NUM_KEY_LAST = 7; 2589 const int FIRST_NON_REG_KEY_offset = 0x70; 2590 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); 2591 2592 __ enter(); // required for proper stackwalking of RuntimeStub frame 2593 handleSOERegisters(true /*saving*/); 2594 2595 // load registers from incoming parameters 2596 const Address from_param(rbp, 8+0); 2597 const Address to_param (rbp, 8+4); 2598 const Address key_param (rbp, 8+8); 2599 const Address rvec_param (rbp, 8+12); 2600 const Address len_param (rbp, 8+16); 2601 __ movptr(from , from_param); 2602 __ movptr(to , to_param); 2603 __ movptr(key , key_param); 2604 __ movptr(rvec , rvec_param); 2605 __ movptr(len_reg , len_param); 2606 2607 // the java expanded key ordering is rotated one position from what we want 2608 // so we start from 0x10 here and hit 0x00 last 2609 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front 2610 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); 2611 // load up xmm regs 2 thru 6 with first 5 keys 2612 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2613 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); 2614 offset += 0x10; 2615 } 2616 2617 // inside here, use the rvec register to point to previous block cipher 2618 // with which we xor at the end of each newly decrypted block 2619 const Register prev_block_cipher_ptr = rvec; 2620 2621 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) 2622 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); 2623 __ cmpl(rax, 44); 2624 __ jcc(Assembler::notEqual, L_key_192_256); 2625 2626 2627 // 128-bit code follows here, parallelized 2628 __ movl(pos, 0); 2629 __ align(OptoLoopAlignment); 2630 __ BIND(L_singleBlock_loopTop_128); 2631 __ cmpptr(len_reg, 0); // any blocks left?? 2632 __ jcc(Assembler::equal, L_exit); 2633 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input 2634 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds 2635 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2636 __ aesdec(xmm_result, as_XMMRegister(rnum)); 2637 } 2638 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0 2639 aes_dec_key(xmm_result, xmm_temp, key, key_offset); 2640 } 2641 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 2642 __ aesdeclast(xmm_result, xmm_temp); 2643 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); 2644 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2645 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2646 // no need to store r to memory until we exit 2647 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr 2648 __ addptr(pos, AESBlockSize); 2649 __ subptr(len_reg, AESBlockSize); 2650 __ jmp(L_singleBlock_loopTop_128); 2651 2652 2653 __ BIND(L_exit); 2654 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); 2655 __ movptr(rvec , rvec_param); // restore this since used in loop 2656 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object 2657 handleSOERegisters(false /*restoring*/); 2658 __ movptr(rax, len_param); // return length 2659 __ leave(); // required for proper stackwalking of RuntimeStub frame 2660 __ ret(0); 2661 2662 2663 __ BIND(L_key_192_256); 2664 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) 2665 __ cmpl(rax, 52); 2666 __ jcc(Assembler::notEqual, L_key_256); 2667 2668 // 192-bit code follows here (could be optimized to use parallelism) 2669 __ movl(pos, 0); 2670 __ align(OptoLoopAlignment); 2671 __ BIND(L_singleBlock_loopTop_192); 2672 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input 2673 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds 2674 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2675 __ aesdec(xmm_result, as_XMMRegister(rnum)); 2676 } 2677 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0 2678 aes_dec_key(xmm_result, xmm_temp, key, key_offset); 2679 } 2680 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 2681 __ aesdeclast(xmm_result, xmm_temp); 2682 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); 2683 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2684 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2685 // no need to store r to memory until we exit 2686 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr 2687 __ addptr(pos, AESBlockSize); 2688 __ subptr(len_reg, AESBlockSize); 2689 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192); 2690 __ jmp(L_exit); 2691 2692 __ BIND(L_key_256); 2693 // 256-bit code follows here (could be optimized to use parallelism) 2694 __ movl(pos, 0); 2695 __ align(OptoLoopAlignment); 2696 __ BIND(L_singleBlock_loopTop_256); 2697 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input 2698 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds 2699 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { 2700 __ aesdec(xmm_result, as_XMMRegister(rnum)); 2701 } 2702 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0 2703 aes_dec_key(xmm_result, xmm_temp, key, key_offset); 2704 } 2705 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 2706 __ aesdeclast(xmm_result, xmm_temp); 2707 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); 2708 __ pxor (xmm_result, xmm_temp); // xor with the current r vector 2709 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output 2710 // no need to store r to memory until we exit 2711 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr 2712 __ addptr(pos, AESBlockSize); 2713 __ subptr(len_reg, AESBlockSize); 2714 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256); 2715 __ jmp(L_exit); 2716 2717 return start; 2718 } 2719 2720 /** 2721 * Arguments: 2722 * 2723 * Inputs: 2724 * rsp(4) - int crc 2725 * rsp(8) - byte* buf 2726 * rsp(12) - int length 2727 * 2728 * Ouput: 2729 * rax - int crc result 2730 */ 2731 address generate_updateBytesCRC32() { 2732 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions"); 2733 2734 __ align(CodeEntryAlignment); 2735 StubCodeMark mark(this, "StubRoutines", "updateBytesCRC32"); 2736 2737 address start = __ pc(); 2738 2739 const Register crc = rdx; // crc 2740 const Register buf = rsi; // source java byte array address 2741 const Register len = rcx; // length 2742 const Register table = rdi; // crc_table address (reuse register) 2743 const Register tmp = rbx; 2744 assert_different_registers(crc, buf, len, table, tmp, rax); 2745 2746 BLOCK_COMMENT("Entry:"); 2747 __ enter(); // required for proper stackwalking of RuntimeStub frame 2748 __ push(rsi); 2749 __ push(rdi); 2750 __ push(rbx); 2751 2752 Address crc_arg(rbp, 8 + 0); 2753 Address buf_arg(rbp, 8 + 4); 2754 Address len_arg(rbp, 8 + 8); 2755 2756 // Load up: 2757 __ movl(crc, crc_arg); 2758 __ movptr(buf, buf_arg); 2759 __ movl(len, len_arg); 2760 2761 __ kernel_crc32(crc, buf, len, table, tmp); 2762 2763 __ movl(rax, crc); 2764 __ pop(rbx); 2765 __ pop(rdi); 2766 __ pop(rsi); 2767 __ leave(); // required for proper stackwalking of RuntimeStub frame 2768 __ ret(0); 2769 2770 return start; 2771 } 2772 2773 // Safefetch stubs. 2774 void generate_safefetch(const char* name, int size, address* entry, 2775 address* fault_pc, address* continuation_pc) { 2776 // safefetch signatures: 2777 // int SafeFetch32(int* adr, int errValue); 2778 // intptr_t SafeFetchN (intptr_t* adr, intptr_t errValue); 2779 2780 StubCodeMark mark(this, "StubRoutines", name); 2781 2782 // Entry point, pc or function descriptor. 2783 *entry = __ pc(); 2784 2785 __ movl(rax, Address(rsp, 0x8)); 2786 __ movl(rcx, Address(rsp, 0x4)); 2787 // Load *adr into eax, may fault. 2788 *fault_pc = __ pc(); 2789 switch (size) { 2790 case 4: 2791 // int32_t 2792 __ movl(rax, Address(rcx, 0)); 2793 break; 2794 case 8: 2795 // int64_t 2796 Unimplemented(); 2797 break; 2798 default: 2799 ShouldNotReachHere(); 2800 } 2801 2802 // Return errValue or *adr. 2803 *continuation_pc = __ pc(); 2804 __ ret(0); 2805 } 2806 2807 public: 2808 // Information about frame layout at time of blocking runtime call. 2809 // Note that we only have to preserve callee-saved registers since 2810 // the compilers are responsible for supplying a continuation point 2811 // if they expect all registers to be preserved. 2812 enum layout { 2813 thread_off, // last_java_sp 2814 arg1_off, 2815 arg2_off, 2816 rbp_off, // callee saved register 2817 ret_pc, 2818 framesize 2819 }; 2820 2821 private: 2822 2823 #undef __ 2824 #define __ masm-> 2825 2826 //------------------------------------------------------------------------------------------------------------------------ 2827 // Continuation point for throwing of implicit exceptions that are not handled in 2828 // the current activation. Fabricates an exception oop and initiates normal 2829 // exception dispatching in this frame. 2830 // 2831 // Previously the compiler (c2) allowed for callee save registers on Java calls. 2832 // This is no longer true after adapter frames were removed but could possibly 2833 // be brought back in the future if the interpreter code was reworked and it 2834 // was deemed worthwhile. The comment below was left to describe what must 2835 // happen here if callee saves were resurrected. As it stands now this stub 2836 // could actually be a vanilla BufferBlob and have now oopMap at all. 2837 // Since it doesn't make much difference we've chosen to leave it the 2838 // way it was in the callee save days and keep the comment. 2839 2840 // If we need to preserve callee-saved values we need a callee-saved oop map and 2841 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. 2842 // If the compiler needs all registers to be preserved between the fault 2843 // point and the exception handler then it must assume responsibility for that in 2844 // AbstractCompiler::continuation_for_implicit_null_exception or 2845 // continuation_for_implicit_division_by_zero_exception. All other implicit 2846 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are 2847 // either at call sites or otherwise assume that stack unwinding will be initiated, 2848 // so caller saved registers were assumed volatile in the compiler. 2849 address generate_throw_exception(const char* name, address runtime_entry, 2850 Register arg1 = noreg, Register arg2 = noreg) { 2851 2852 int insts_size = 256; 2853 int locs_size = 32; 2854 2855 CodeBuffer code(name, insts_size, locs_size); 2856 OopMapSet* oop_maps = new OopMapSet(); 2857 MacroAssembler* masm = new MacroAssembler(&code); 2858 2859 address start = __ pc(); 2860 2861 // This is an inlined and slightly modified version of call_VM 2862 // which has the ability to fetch the return PC out of 2863 // thread-local storage and also sets up last_Java_sp slightly 2864 // differently than the real call_VM 2865 Register java_thread = rbx; 2866 __ get_thread(java_thread); 2867 2868 __ enter(); // required for proper stackwalking of RuntimeStub frame 2869 2870 // pc and rbp, already pushed 2871 __ subptr(rsp, (framesize-2) * wordSize); // prolog 2872 2873 // Frame is now completed as far as size and linkage. 2874 2875 int frame_complete = __ pc() - start; 2876 2877 // push java thread (becomes first argument of C function) 2878 __ movptr(Address(rsp, thread_off * wordSize), java_thread); 2879 if (arg1 != noreg) { 2880 __ movptr(Address(rsp, arg1_off * wordSize), arg1); 2881 } 2882 if (arg2 != noreg) { 2883 assert(arg1 != noreg, "missing reg arg"); 2884 __ movptr(Address(rsp, arg2_off * wordSize), arg2); 2885 } 2886 2887 // Set up last_Java_sp and last_Java_fp 2888 __ set_last_Java_frame(java_thread, rsp, rbp, NULL); 2889 2890 // Call runtime 2891 BLOCK_COMMENT("call runtime_entry"); 2892 __ call(RuntimeAddress(runtime_entry)); 2893 // Generate oop map 2894 OopMap* map = new OopMap(framesize, 0); 2895 oop_maps->add_gc_map(__ pc() - start, map); 2896 2897 // restore the thread (cannot use the pushed argument since arguments 2898 // may be overwritten by C code generated by an optimizing compiler); 2899 // however can use the register value directly if it is callee saved. 2900 __ get_thread(java_thread); 2901 2902 __ reset_last_Java_frame(java_thread, true, false); 2903 2904 __ leave(); // required for proper stackwalking of RuntimeStub frame 2905 2906 // check for pending exceptions 2907 #ifdef ASSERT 2908 Label L; 2909 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); 2910 __ jcc(Assembler::notEqual, L); 2911 __ should_not_reach_here(); 2912 __ bind(L); 2913 #endif /* ASSERT */ 2914 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); 2915 2916 2917 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); 2918 return stub->entry_point(); 2919 } 2920 2921 2922 void create_control_words() { 2923 // Round to nearest, 53-bit mode, exceptions masked 2924 StubRoutines::_fpu_cntrl_wrd_std = 0x027F; 2925 // Round to zero, 53-bit mode, exception mased 2926 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; 2927 // Round to nearest, 24-bit mode, exceptions masked 2928 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; 2929 // Round to nearest, 64-bit mode, exceptions masked 2930 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; 2931 // Round to nearest, 64-bit mode, exceptions masked 2932 StubRoutines::_mxcsr_std = 0x1F80; 2933 // Note: the following two constants are 80-bit values 2934 // layout is critical for correct loading by FPU. 2935 // Bias for strict fp multiply/divide 2936 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 2937 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; 2938 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; 2939 // Un-Bias for strict fp multiply/divide 2940 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 2941 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; 2942 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; 2943 } 2944 2945 //--------------------------------------------------------------------------- 2946 // Initialization 2947 2948 void generate_initial() { 2949 // Generates all stubs and initializes the entry points 2950 2951 //------------------------------------------------------------------------------------------------------------------------ 2952 // entry points that exist in all platforms 2953 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than 2954 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. 2955 StubRoutines::_forward_exception_entry = generate_forward_exception(); 2956 2957 StubRoutines::_call_stub_entry = 2958 generate_call_stub(StubRoutines::_call_stub_return_address); 2959 // is referenced by megamorphic call 2960 StubRoutines::_catch_exception_entry = generate_catch_exception(); 2961 2962 // These are currently used by Solaris/Intel 2963 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); 2964 2965 StubRoutines::_handler_for_unsafe_access_entry = 2966 generate_handler_for_unsafe_access(); 2967 2968 // platform dependent 2969 create_control_words(); 2970 2971 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); 2972 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); 2973 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, 2974 CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); 2975 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, 2976 CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); 2977 2978 // Build this early so it's available for the interpreter 2979 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError)); 2980 2981 if (UseCRC32Intrinsics) { 2982 // set table address before stub generation which use it 2983 StubRoutines::_crc_table_adr = (address)StubRoutines::x86::_crc_table; 2984 StubRoutines::_updateBytesCRC32 = generate_updateBytesCRC32(); 2985 } 2986 } 2987 2988 2989 void generate_all() { 2990 // Generates all stubs and initializes the entry points 2991 2992 // These entry points require SharedInfo::stack0 to be set up in non-core builds 2993 // and need to be relocatable, so they each fabricate a RuntimeStub internally. 2994 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError)); 2995 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError)); 2996 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call)); 2997 2998 //------------------------------------------------------------------------------------------------------------------------ 2999 // entry points that are platform specific 3000 3001 // support for verify_oop (must happen after universe_init) 3002 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); 3003 3004 // arraycopy stubs used by compilers 3005 generate_arraycopy_stubs(); 3006 3007 generate_math_stubs(); 3008 3009 // don't bother generating these AES intrinsic stubs unless global flag is set 3010 if (UseAESIntrinsics) { 3011 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others 3012 3013 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); 3014 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); 3015 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); 3016 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt(); 3017 } 3018 3019 // Safefetch stubs. 3020 generate_safefetch("SafeFetch32", sizeof(int), &StubRoutines::_safefetch32_entry, 3021 &StubRoutines::_safefetch32_fault_pc, 3022 &StubRoutines::_safefetch32_continuation_pc); 3023 StubRoutines::_safefetchN_entry = StubRoutines::_safefetch32_entry; 3024 StubRoutines::_safefetchN_fault_pc = StubRoutines::_safefetch32_fault_pc; 3025 StubRoutines::_safefetchN_continuation_pc = StubRoutines::_safefetch32_continuation_pc; 3026 } 3027 3028 3029 public: 3030 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { 3031 if (all) { 3032 generate_all(); 3033 } else { 3034 generate_initial(); 3035 } 3036 } 3037 }; // end class declaration 3038 3039 3040 void StubGenerator_generate(CodeBuffer* code, bool all) { 3041 StubGenerator g(code, all); 3042 }