quick_entrypoints_arm64.S revision 83b1940e6482b9d8feba5c492507735686650ea5
1/* 2 * Copyright (C) 2014 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "asm_support_arm64.S" 18 19#include "arch/quick_alloc_entrypoints.S" 20 21 22 /* 23 * Macro that sets up the callee save frame to conform with 24 * Runtime::CreateCalleeSaveMethod(kSaveAll) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 27 adrp xIP0, :got:_ZN3art7Runtime9instance_E 28 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 29 30 // Our registers aren't intermixed - just spill in order. 31 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 32 33 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 34 THIS_LOAD_REQUIRES_READ_BARRIER 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ] 36 37 sub sp, sp, #176 38 .cfi_adjust_cfa_offset 176 39 40 // Ugly compile-time check, but we only have the preprocessor. 41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176) 42#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected." 43#endif 44 45 // FP callee-saves 46 stp d8, d9, [sp, #8] 47 stp d10, d11, [sp, #24] 48 stp d12, d13, [sp, #40] 49 stp d14, d15, [sp, #56] 50 51 // Reserved registers 52 stp xSELF, xSUSPEND, [sp, #72] 53 .cfi_rel_offset x18, 72 54 .cfi_rel_offset x19, 80 55 56 // callee-saves 57 stp x20, x21, [sp, #88] 58 .cfi_rel_offset x20, 88 59 .cfi_rel_offset x21, 96 60 61 stp x22, x23, [sp, #104] 62 .cfi_rel_offset x22, 104 63 .cfi_rel_offset x23, 112 64 65 stp x24, x25, [sp, #120] 66 .cfi_rel_offset x24, 120 67 .cfi_rel_offset x25, 128 68 69 stp x26, x27, [sp, #136] 70 .cfi_rel_offset x26, 136 71 .cfi_rel_offset x27, 144 72 73 stp x28, x29, [sp, #152] 74 .cfi_rel_offset x28, 152 75 .cfi_rel_offset x29, 160 76 77 str xLR, [sp, #168] 78 .cfi_rel_offset x30, 168 79 80 // Loads appropriate callee-save-method 81 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 82.endm 83 84 /* 85 * Macro that sets up the callee save frame to conform with 86 * Runtime::CreateCalleeSaveMethod(kRefsOnly). 87 */ 88.macro SETUP_REF_ONLY_CALLEE_SAVE_FRAME 89 adrp xIP0, :got:_ZN3art7Runtime9instance_E 90 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 91 92 // Our registers aren't intermixed - just spill in order. 93 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 94 95 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 96 THIS_LOAD_REQUIRES_READ_BARRIER 97 ldr xIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ] 98 99 sub sp, sp, #96 100 .cfi_adjust_cfa_offset 96 101 102 // Ugly compile-time check, but we only have the preprocessor. 103#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96) 104#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected." 105#endif 106 107 // Callee-saves 108 stp x20, x21, [sp, #8] 109 .cfi_rel_offset x20, 8 110 .cfi_rel_offset x21, 16 111 112 stp x22, x23, [sp, #24] 113 .cfi_rel_offset x22, 24 114 .cfi_rel_offset x23, 32 115 116 stp x24, x25, [sp, #40] 117 .cfi_rel_offset x24, 40 118 .cfi_rel_offset x25, 48 119 120 stp x26, x27, [sp, #56] 121 .cfi_rel_offset x26, 56 122 .cfi_rel_offset x27, 64 123 124 stp x28, x29, [sp, #72] 125 .cfi_rel_offset x28, 72 126 .cfi_rel_offset x29, 80 127 128 // LR 129 str xLR, [sp, #88] 130 .cfi_rel_offset x30, 88 131 132 // Save xSELF to xETR. 133 mov xETR, xSELF 134 135 // Loads appropriate callee-save-method 136 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 137.endm 138 139// TODO: Probably no need to restore registers preserved by aapcs64. 140.macro RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 141 // Restore xSELF. 142 mov xSELF, xETR 143 144 // Callee-saves 145 ldp x20, x21, [sp, #8] 146 .cfi_restore x20 147 .cfi_restore x21 148 149 ldp x22, x23, [sp, #24] 150 .cfi_restore x22 151 .cfi_restore x23 152 153 ldp x24, x25, [sp, #40] 154 .cfi_restore x24 155 .cfi_restore x25 156 157 ldp x26, x27, [sp, #56] 158 .cfi_restore x26 159 .cfi_restore x27 160 161 ldp x28, x29, [sp, #72] 162 .cfi_restore x28 163 .cfi_restore x29 164 165 // LR 166 ldr xLR, [sp, #88] 167 .cfi_restore x30 168 169 add sp, sp, #96 170 .cfi_adjust_cfa_offset -96 171.endm 172 173.macro POP_REF_ONLY_CALLEE_SAVE_FRAME 174 // Restore xSELF as it might be scratched. 175 mov xSELF, xETR 176 // ETR 177 ldr xETR, [sp, #16] 178 .cfi_restore x21 179 180 add sp, sp, #96 181 .cfi_adjust_cfa_offset -96 182.endm 183 184.macro RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 185 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 186 ret 187.endm 188 189 190.macro SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 191 sub sp, sp, #224 192 .cfi_adjust_cfa_offset 224 193 194 // Ugly compile-time check, but we only have the preprocessor. 195#if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 224) 196#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM64) size not as expected." 197#endif 198 199 // FP args 200 stp d0, d1, [sp, #16] 201 stp d2, d3, [sp, #32] 202 stp d4, d5, [sp, #48] 203 stp d6, d7, [sp, #64] 204 205 // args and x20(callee-save) 206 stp x1, x2, [sp, #80] 207 .cfi_rel_offset x1, 80 208 .cfi_rel_offset x2, 88 209 210 stp x3, x4, [sp, #96] 211 .cfi_rel_offset x3, 96 212 .cfi_rel_offset x4, 104 213 214 stp x5, x6, [sp, #112] 215 .cfi_rel_offset x5, 112 216 .cfi_rel_offset x6, 120 217 218 stp x7, x20, [sp, #128] 219 .cfi_rel_offset x7, 128 220 .cfi_rel_offset x20, 136 221 222 // Callee-saves. 223 stp x21, x22, [sp, #144] 224 .cfi_rel_offset x21, 144 225 .cfi_rel_offset x22, 152 226 227 stp x23, x24, [sp, #160] 228 .cfi_rel_offset x23, 160 229 .cfi_rel_offset x24, 168 230 231 stp x25, x26, [sp, #176] 232 .cfi_rel_offset x25, 176 233 .cfi_rel_offset x26, 184 234 235 stp x27, x28, [sp, #192] 236 .cfi_rel_offset x27, 192 237 .cfi_rel_offset x28, 200 238 239 // x29(callee-save) and LR 240 stp x29, xLR, [sp, #208] 241 .cfi_rel_offset x29, 208 242 .cfi_rel_offset x30, 216 243 244 // Save xSELF to xETR. 245 mov xETR, xSELF 246.endm 247 248 /* 249 * Macro that sets up the callee save frame to conform with 250 * Runtime::CreateCalleeSaveMethod(kRefsAndArgs). 251 * 252 * TODO This is probably too conservative - saving FP & LR. 253 */ 254.macro SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME 255 adrp xIP0, :got:_ZN3art7Runtime9instance_E 256 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 257 258 // Our registers aren't intermixed - just spill in order. 259 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 260 261 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 262 THIS_LOAD_REQUIRES_READ_BARRIER 263 ldr xIP0, [xIP0, RUNTIME_REF_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ] 264 265 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 266 267 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 268.endm 269 270// TODO: Probably no need to restore registers preserved by aapcs64. 271.macro RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 272 // Restore xSELF. 273 mov xSELF, xETR 274 275 // FP args 276 ldp d0, d1, [sp, #16] 277 ldp d2, d3, [sp, #32] 278 ldp d4, d5, [sp, #48] 279 ldp d6, d7, [sp, #64] 280 281 // args and x20(callee-save) 282 ldp x1, x2, [sp, #80] 283 .cfi_restore x1 284 .cfi_restore x2 285 286 ldp x3, x4, [sp, #96] 287 .cfi_restore x3 288 .cfi_restore x4 289 290 ldp x5, x6, [sp, #112] 291 .cfi_restore x5 292 .cfi_restore x6 293 294 ldp x7, x20, [sp, #128] 295 .cfi_restore x7 296 .cfi_restore x20 297 298 // Callee-saves. 299 ldp x21, x22, [sp, #144] 300 .cfi_restore x21 301 .cfi_restore x22 302 303 ldp x23, x24, [sp, #160] 304 .cfi_restore x23 305 .cfi_restore x24 306 307 ldp x25, x26, [sp, #176] 308 .cfi_restore x25 309 .cfi_restore x26 310 311 ldp x27, x28, [sp, #192] 312 .cfi_restore x27 313 .cfi_restore x28 314 315 // x29(callee-save) and LR 316 ldp x29, xLR, [sp, #208] 317 .cfi_restore x29 318 .cfi_restore x30 319 320 add sp, sp, #224 321 .cfi_adjust_cfa_offset -224 322.endm 323 324.macro RETURN_IF_RESULT_IS_ZERO 325 cbnz x0, 1f // result non-zero branch over 326 ret // return 3271: 328.endm 329 330.macro RETURN_IF_RESULT_IS_NON_ZERO 331 cbz x0, 1f // result zero branch over 332 ret // return 3331: 334.endm 335 336 /* 337 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 338 * exception is Thread::Current()->exception_ 339 */ 340.macro DELIVER_PENDING_EXCEPTION 341 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 342 mov x0, xSELF 343 mov x1, sp 344 345 // Point of no return. 346 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*, SP) 347 brk 0 // Unreached 348.endm 349 350.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 351 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 352 cbnz \reg, 1f 353 ret 3541: 355 DELIVER_PENDING_EXCEPTION 356.endm 357 358.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 359 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 360.endm 361 362// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 363.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 364 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 365.endm 366 367.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 368 cbnz w0, 1f // result non-zero branch over 369 ret // return 3701: 371 DELIVER_PENDING_EXCEPTION 372.endm 373 374.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 375 .extern \cxx_name 376ENTRY \c_name 377 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 378 mov x0, xSELF // pass Thread::Current 379 mov x1, sp // pass SP 380 b \cxx_name // \cxx_name(Thread*, SP) 381END \c_name 382.endm 383 384.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 385 .extern \cxx_name 386ENTRY \c_name 387 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context. 388 mov x1, xSELF // pass Thread::Current. 389 mov x2, sp // pass SP. 390 b \cxx_name // \cxx_name(arg, Thread*, SP). 391 brk 0 392END \c_name 393.endm 394 395.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 396 .extern \cxx_name 397ENTRY \c_name 398 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 399 mov x2, xSELF // pass Thread::Current 400 mov x3, sp // pass SP 401 b \cxx_name // \cxx_name(arg1, arg2, Thread*, SP) 402 brk 0 403END \c_name 404.endm 405 406 /* 407 * Called by managed code, saves callee saves and then calls artThrowException 408 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 409 */ 410ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 411 412 /* 413 * Called by managed code to create and deliver a NullPointerException. 414 */ 415NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 416 417 /* 418 * Called by managed code to create and deliver an ArithmeticException. 419 */ 420NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 421 422 /* 423 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 424 * index, arg2 holds limit. 425 */ 426TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 427 428 /* 429 * Called by managed code to create and deliver a StackOverflowError. 430 */ 431NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 432 433 /* 434 * Called by managed code to create and deliver a NoSuchMethodError. 435 */ 436ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 437 438 /* 439 * Invoke stack overflow exception from signal handler. 440 * On entry: 441 * xSELF: thread 442 * SP: address of last known frame 443 * IP0: address of next valid SP below protected region in stack 444 * 445 * This is deceptively simple but hides some complexity. It is called in the case of 446 * a stack overflow condition during implicit checks. The signal handler has been 447 * called by the kernel due to a load from the protected stack region. The handler 448 * works out the address of the previous frame and passes this in SP. However there 449 * is a piece of memory somewhere below the current SP that is not accessible (the 450 * memory that caused the signal). The signal handler works out the next 451 * accessible value of SP and passes this in x16/IP0. This code then sets up the SP 452 * to be this new value and calls the code to create and throw the stack overflow 453 * exception. 454 */ 455ENTRY art_quick_throw_stack_overflow_from_signal 456 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 457 mov x0, xSELF // pass Thread::Current 458 mov x1, sp // pass SP 459 mov sp, xIP0 // move SP down to below protected region. 460 b artThrowStackOverflowFromCode // artThrowStackOverflowFromCode(Thread*, SP) 461END art_quick_throw_stack_overflow_from_signal 462 463 /* 464 * All generated callsites for interface invokes and invocation slow paths will load arguments 465 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 466 * the method_idx. This wrapper will save arg1-arg3, load the caller's Method*, align the 467 * stack and call the appropriate C helper. 468 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 469 * 470 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 471 * of the target Method* in x0 and method->code_ in x1. 472 * 473 * If unsuccessful, the helper will return NULL/????. There will be a pending exception in the 474 * thread and we branch to another stub to deliver it. 475 * 476 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 477 * pointing back to the original caller. 478 * 479 * Adapted from ARM32 code. 480 * 481 * Clobbers xIP0. 482 */ 483.macro INVOKE_TRAMPOLINE c_name, cxx_name 484 .extern \cxx_name 485ENTRY \c_name 486 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME // save callee saves in case allocation triggers GC 487 // Helper signature is always 488 // (method_idx, *this_object, *caller_method, *self, sp) 489 490 ldr w2, [sp, #FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE] // pass caller Method* 491 mov x3, xSELF // pass Thread::Current 492 mov x4, sp 493 bl \cxx_name // (method_idx, this, caller, Thread*, SP) 494 mov xIP0, x1 // save Method*->code_ 495 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 496 cbz x0, 1f // did we find the target? if not go to exception delivery 497 br xIP0 // tail call to target 4981: 499 DELIVER_PENDING_EXCEPTION 500END \c_name 501.endm 502 503INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline, artInvokeInterfaceTrampoline 504INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 505 506INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 507INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 508INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 509INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 510 511 512.macro INVOKE_STUB_CREATE_FRAME 513 514SAVE_SIZE=6*8 // x4, x5, xSUSPEND, SP, LR & FP saved. 515SAVE_SIZE_AND_METHOD=SAVE_SIZE+STACK_REFERENCE_SIZE 516 517 518 mov x9, sp // Save stack pointer. 519 .cfi_register sp,x9 520 521 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 522 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 523 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 524 mov sp, x10 // Set new SP. 525 526 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 527 .cfi_def_cfa_register x10 // before this. 528 .cfi_adjust_cfa_offset SAVE_SIZE 529 530 stp x9, xSUSPEND, [x10, #32] // Save old stack pointer and xSUSPEND 531 .cfi_rel_offset sp, 32 532 .cfi_rel_offset x19, 40 533 534 stp x4, x5, [x10, #16] // Save result and shorty addresses. 535 .cfi_rel_offset x4, 16 536 .cfi_rel_offset x5, 24 537 538 stp xFP, xLR, [x10] // Store LR & FP. 539 .cfi_rel_offset x29, 0 540 .cfi_rel_offset x30, 8 541 542 mov xFP, x10 // Use xFP now, as it's callee-saved. 543 .cfi_def_cfa_register x29 544 mov xSELF, x3 // Move thread pointer into SELF register. 545 mov wSUSPEND, #SUSPEND_CHECK_INTERVAL // reset wSUSPEND to suspend check interval 546 547 // Copy arguments into stack frame. 548 // Use simple copy routine for now. 549 // 4 bytes per slot. 550 // X1 - source address 551 // W2 - args length 552 // X9 - destination address. 553 // W10 - temporary 554 add x9, sp, #4 // Destination address is bottom of stack + NULL. 555 556 // Use \@ to differentiate between macro invocations. 557.LcopyParams\@: 558 cmp w2, #0 559 beq .LendCopyParams\@ 560 sub w2, w2, #4 // Need 65536 bytes of range. 561 ldr w10, [x1, x2] 562 str w10, [x9, x2] 563 564 b .LcopyParams\@ 565 566.LendCopyParams\@: 567 568 // Store NULL into StackReference<Method>* at bottom of frame. 569 str wzr, [sp] 570 571#if (STACK_REFERENCE_SIZE != 4) 572#error "STACK_REFERENCE_SIZE(ARM64) size not as expected." 573#endif 574.endm 575 576.macro INVOKE_STUB_CALL_AND_RETURN 577 578 // load method-> METHOD_QUICK_CODE_OFFSET 579 ldr x9, [x0 , #METHOD_QUICK_CODE_OFFSET] 580 // Branch to method. 581 blr x9 582 583 // Restore return value address and shorty address. 584 ldp x4,x5, [xFP, #16] 585 .cfi_restore x4 586 .cfi_restore x5 587 588 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 589 ldrb w10, [x5] 590 591 // Don't set anything for a void type. 592 cmp w10, #'V' 593 beq .Lexit_art_quick_invoke_stub\@ 594 595 cmp w10, #'D' 596 bne .Lreturn_is_float\@ 597 str d0, [x4] 598 b .Lexit_art_quick_invoke_stub\@ 599 600.Lreturn_is_float\@: 601 cmp w10, #'F' 602 bne .Lreturn_is_int\@ 603 str s0, [x4] 604 b .Lexit_art_quick_invoke_stub\@ 605 606 // Just store x0. Doesn't matter if it is 64 or 32 bits. 607.Lreturn_is_int\@: 608 str x0, [x4] 609 610.Lexit_art_quick_invoke_stub\@: 611 ldp x2, xSUSPEND, [xFP, #32] // Restore stack pointer and xSUSPEND. 612 .cfi_restore x19 613 mov sp, x2 614 .cfi_restore sp 615 616 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 617 .cfi_restore x29 618 .cfi_restore x30 619 620 ret 621 622.endm 623 624 625/* 626 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 627 * uint32_t *args, x1 628 * uint32_t argsize, w2 629 * Thread *self, x3 630 * JValue *result, x4 631 * char *shorty); x5 632 * +----------------------+ 633 * | | 634 * | C/C++ frame | 635 * | LR'' | 636 * | FP'' | <- SP' 637 * +----------------------+ 638 * +----------------------+ 639 * | x19 | <- Used as wSUSPEND, won't be restored by managed code. 640 * | SP' | 641 * | X5 | 642 * | X4 | Saved registers 643 * | LR' | 644 * | FP' | <- FP 645 * +----------------------+ 646 * | uint32_t out[n-1] | 647 * | : : | Outs 648 * | uint32_t out[0] | 649 * | StackRef<ArtMethod> | <- SP value=null 650 * +----------------------+ 651 * 652 * Outgoing registers: 653 * x0 - Method* 654 * x1-x7 - integer parameters. 655 * d0-d7 - Floating point parameters. 656 * xSELF = self 657 * wSUSPEND = suspend count 658 * SP = & of ArtMethod* 659 * x1 = "this" pointer. 660 * 661 */ 662ENTRY art_quick_invoke_stub 663 // Spill registers as per AACPS64 calling convention. 664 INVOKE_STUB_CREATE_FRAME 665 666 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 667 // Parse the passed shorty to determine which register to load. 668 // Load addresses for routines that load WXSD registers. 669 adr x11, .LstoreW2 670 adr x12, .LstoreX2 671 adr x13, .LstoreS0 672 adr x14, .LstoreD0 673 674 // Initialize routine offsets to 0 for integers and floats. 675 // x8 for integers, x15 for floating point. 676 mov x8, #0 677 mov x15, #0 678 679 add x10, x5, #1 // Load shorty address, plus one to skip return value. 680 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 681 682 // Loop to fill registers. 683.LfillRegisters: 684 ldrb w17, [x10], #1 // Load next character in signature, and increment. 685 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 686 687 cmp w17, #'F' // is this a float? 688 bne .LisDouble 689 690 cmp x15, # 8*12 // Skip this load if all registers full. 691 beq .Ladvance4 692 693 add x17, x13, x15 // Calculate subroutine to jump to. 694 br x17 695 696.LisDouble: 697 cmp w17, #'D' // is this a double? 698 bne .LisLong 699 700 cmp x15, # 8*12 // Skip this load if all registers full. 701 beq .Ladvance8 702 703 add x17, x14, x15 // Calculate subroutine to jump to. 704 br x17 705 706.LisLong: 707 cmp w17, #'J' // is this a long? 708 bne .LisOther 709 710 cmp x8, # 6*12 // Skip this load if all registers full. 711 beq .Ladvance8 712 713 add x17, x12, x8 // Calculate subroutine to jump to. 714 br x17 715 716.LisOther: // Everything else takes one vReg. 717 cmp x8, # 6*12 // Skip this load if all registers full. 718 beq .Ladvance4 719 720 add x17, x11, x8 // Calculate subroutine to jump to. 721 br x17 722 723.Ladvance4: 724 add x9, x9, #4 725 b .LfillRegisters 726 727.Ladvance8: 728 add x9, x9, #8 729 b .LfillRegisters 730 731// Macro for loading a parameter into a register. 732// counter - the register with offset into these tables 733// size - the size of the register - 4 or 8 bytes. 734// register - the name of the register to be loaded. 735.macro LOADREG counter size register return 736 ldr \register , [x9], #\size 737 add \counter, \counter, 12 738 b \return 739.endm 740 741// Store ints. 742.LstoreW2: 743 LOADREG x8 4 w2 .LfillRegisters 744 LOADREG x8 4 w3 .LfillRegisters 745 LOADREG x8 4 w4 .LfillRegisters 746 LOADREG x8 4 w5 .LfillRegisters 747 LOADREG x8 4 w6 .LfillRegisters 748 LOADREG x8 4 w7 .LfillRegisters 749 750// Store longs. 751.LstoreX2: 752 LOADREG x8 8 x2 .LfillRegisters 753 LOADREG x8 8 x3 .LfillRegisters 754 LOADREG x8 8 x4 .LfillRegisters 755 LOADREG x8 8 x5 .LfillRegisters 756 LOADREG x8 8 x6 .LfillRegisters 757 LOADREG x8 8 x7 .LfillRegisters 758 759// Store singles. 760.LstoreS0: 761 LOADREG x15 4 s0 .LfillRegisters 762 LOADREG x15 4 s1 .LfillRegisters 763 LOADREG x15 4 s2 .LfillRegisters 764 LOADREG x15 4 s3 .LfillRegisters 765 LOADREG x15 4 s4 .LfillRegisters 766 LOADREG x15 4 s5 .LfillRegisters 767 LOADREG x15 4 s6 .LfillRegisters 768 LOADREG x15 4 s7 .LfillRegisters 769 770// Store doubles. 771.LstoreD0: 772 LOADREG x15 8 d0 .LfillRegisters 773 LOADREG x15 8 d1 .LfillRegisters 774 LOADREG x15 8 d2 .LfillRegisters 775 LOADREG x15 8 d3 .LfillRegisters 776 LOADREG x15 8 d4 .LfillRegisters 777 LOADREG x15 8 d5 .LfillRegisters 778 LOADREG x15 8 d6 .LfillRegisters 779 LOADREG x15 8 d7 .LfillRegisters 780 781 782.LcallFunction: 783 784 INVOKE_STUB_CALL_AND_RETURN 785 786END art_quick_invoke_stub 787 788/* extern"C" 789 * void art_quick_invoke_static_stub(ArtMethod *method, x0 790 * uint32_t *args, x1 791 * uint32_t argsize, w2 792 * Thread *self, x3 793 * JValue *result, x4 794 * char *shorty); x5 795 */ 796ENTRY art_quick_invoke_static_stub 797 // Spill registers as per AACPS64 calling convention. 798 INVOKE_STUB_CREATE_FRAME 799 800 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 801 // Parse the passed shorty to determine which register to load. 802 // Load addresses for routines that load WXSD registers. 803 adr x11, .LstoreW1_2 804 adr x12, .LstoreX1_2 805 adr x13, .LstoreS0_2 806 adr x14, .LstoreD0_2 807 808 // Initialize routine offsets to 0 for integers and floats. 809 // x8 for integers, x15 for floating point. 810 mov x8, #0 811 mov x15, #0 812 813 add x10, x5, #1 // Load shorty address, plus one to skip return value. 814 815 // Loop to fill registers. 816.LfillRegisters2: 817 ldrb w17, [x10], #1 // Load next character in signature, and increment. 818 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 819 820 cmp w17, #'F' // is this a float? 821 bne .LisDouble2 822 823 cmp x15, # 8*12 // Skip this load if all registers full. 824 beq .Ladvance4_2 825 826 add x17, x13, x15 // Calculate subroutine to jump to. 827 br x17 828 829.LisDouble2: 830 cmp w17, #'D' // is this a double? 831 bne .LisLong2 832 833 cmp x15, # 8*12 // Skip this load if all registers full. 834 beq .Ladvance8_2 835 836 add x17, x14, x15 // Calculate subroutine to jump to. 837 br x17 838 839.LisLong2: 840 cmp w17, #'J' // is this a long? 841 bne .LisOther2 842 843 cmp x8, # 7*12 // Skip this load if all registers full. 844 beq .Ladvance8_2 845 846 add x17, x12, x8 // Calculate subroutine to jump to. 847 br x17 848 849.LisOther2: // Everything else takes one vReg. 850 cmp x8, # 7*12 // Skip this load if all registers full. 851 beq .Ladvance4_2 852 853 add x17, x11, x8 // Calculate subroutine to jump to. 854 br x17 855 856.Ladvance4_2: 857 add x9, x9, #4 858 b .LfillRegisters2 859 860.Ladvance8_2: 861 add x9, x9, #8 862 b .LfillRegisters2 863 864// Store ints. 865.LstoreW1_2: 866 LOADREG x8 4 w1 .LfillRegisters2 867 LOADREG x8 4 w2 .LfillRegisters2 868 LOADREG x8 4 w3 .LfillRegisters2 869 LOADREG x8 4 w4 .LfillRegisters2 870 LOADREG x8 4 w5 .LfillRegisters2 871 LOADREG x8 4 w6 .LfillRegisters2 872 LOADREG x8 4 w7 .LfillRegisters2 873 874// Store longs. 875.LstoreX1_2: 876 LOADREG x8 8 x1 .LfillRegisters2 877 LOADREG x8 8 x2 .LfillRegisters2 878 LOADREG x8 8 x3 .LfillRegisters2 879 LOADREG x8 8 x4 .LfillRegisters2 880 LOADREG x8 8 x5 .LfillRegisters2 881 LOADREG x8 8 x6 .LfillRegisters2 882 LOADREG x8 8 x7 .LfillRegisters2 883 884// Store singles. 885.LstoreS0_2: 886 LOADREG x15 4 s0 .LfillRegisters2 887 LOADREG x15 4 s1 .LfillRegisters2 888 LOADREG x15 4 s2 .LfillRegisters2 889 LOADREG x15 4 s3 .LfillRegisters2 890 LOADREG x15 4 s4 .LfillRegisters2 891 LOADREG x15 4 s5 .LfillRegisters2 892 LOADREG x15 4 s6 .LfillRegisters2 893 LOADREG x15 4 s7 .LfillRegisters2 894 895// Store doubles. 896.LstoreD0_2: 897 LOADREG x15 8 d0 .LfillRegisters2 898 LOADREG x15 8 d1 .LfillRegisters2 899 LOADREG x15 8 d2 .LfillRegisters2 900 LOADREG x15 8 d3 .LfillRegisters2 901 LOADREG x15 8 d4 .LfillRegisters2 902 LOADREG x15 8 d5 .LfillRegisters2 903 LOADREG x15 8 d6 .LfillRegisters2 904 LOADREG x15 8 d7 .LfillRegisters2 905 906 907.LcallFunction2: 908 909 INVOKE_STUB_CALL_AND_RETURN 910 911END art_quick_invoke_static_stub 912 913 914 915 /* 916 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 917 */ 918 919ENTRY art_quick_do_long_jump 920 // Load FPRs 921 ldp d0, d1, [x1], #16 922 ldp d2, d3, [x1], #16 923 ldp d4, d5, [x1], #16 924 ldp d6, d7, [x1], #16 925 ldp d8, d9, [x1], #16 926 ldp d10, d11, [x1], #16 927 ldp d12, d13, [x1], #16 928 ldp d14, d15, [x1], #16 929 ldp d16, d17, [x1], #16 930 ldp d18, d19, [x1], #16 931 ldp d20, d21, [x1], #16 932 ldp d22, d23, [x1], #16 933 ldp d24, d25, [x1], #16 934 ldp d26, d27, [x1], #16 935 ldp d28, d29, [x1], #16 936 ldp d30, d31, [x1] 937 938 // Load GPRs 939 // TODO: lots of those are smashed, could optimize. 940 add x0, x0, #30*8 941 ldp x30, x1, [x0], #-16 942 ldp x28, x29, [x0], #-16 943 ldp x26, x27, [x0], #-16 944 ldp x24, x25, [x0], #-16 945 ldp x22, x23, [x0], #-16 946 ldp x20, x21, [x0], #-16 947 ldp x18, x19, [x0], #-16 948 ldp x16, x17, [x0], #-16 949 ldp x14, x15, [x0], #-16 950 ldp x12, x13, [x0], #-16 951 ldp x10, x11, [x0], #-16 952 ldp x8, x9, [x0], #-16 953 ldp x6, x7, [x0], #-16 954 ldp x4, x5, [x0], #-16 955 ldp x2, x3, [x0], #-16 956 mov sp, x1 957 958 // TODO: Is it really OK to use LR for the target PC? 959 mov x0, #0 960 mov x1, #0 961 br xLR 962END art_quick_do_long_jump 963 964 /* 965 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 966 * failure. 967 */ 968 .extern artHandleFillArrayDataFromCode 969ENTRY art_quick_handle_fill_data 970 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // Save callee saves in case exception allocation triggers GC. 971 mov x2, xSELF // Pass Thread::Current. 972 mov x3, sp // Pass SP. 973 bl artHandleFillArrayDataFromCode // (Array*, const DexFile::Payload*, Thread*, SP) 974 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 975 RETURN_IF_RESULT_IS_ZERO 976 DELIVER_PENDING_EXCEPTION 977END art_quick_handle_fill_data 978 979 /* 980 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 981 * possibly null object to lock. 982 * 983 * Derived from arm32 code. 984 */ 985 .extern artLockObjectFromCode 986ENTRY art_quick_lock_object 987 cbz w0, .Lslow_lock 988 add x4, x0, #LOCK_WORD_OFFSET // exclusive load/store had no immediate anymore 989.Lretry_lock: 990 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 991 ldxr w1, [x4] 992 cbnz w1, .Lnot_unlocked // already thin locked 993 stxr w3, w2, [x4] 994 cbnz w3, .Lstrex_fail // store failed, retry 995 dmb ishld // full (LoadLoad|LoadStore) memory barrier 996 ret 997.Lstrex_fail: 998 b .Lretry_lock // unlikely forward branch, need to reload and recheck r1/r2 999.Lnot_unlocked: 1000 lsr w3, w1, 30 1001 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1002 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1003 uxth w2, w2 // zero top 16 bits 1004 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1005 // else contention, go to slow path 1006 add w2, w1, #65536 // increment count in lock word placing in w2 for storing 1007 lsr w1, w2, 30 // if either of the top two bits are set, we overflowed. 1008 cbnz w1, .Lslow_lock // if we overflow the count go slow path 1009 str w2, [x0, #LOCK_WORD_OFFSET]// no need for stxr as we hold the lock 1010 ret 1011.Lslow_lock: 1012 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1013 mov x1, xSELF // pass Thread::Current 1014 mov x2, sp // pass SP 1015 bl artLockObjectFromCode // (Object* obj, Thread*, SP) 1016 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1017 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1018END art_quick_lock_object 1019 1020 /* 1021 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1022 * x0 holds the possibly null object to lock. 1023 * 1024 * Derived from arm32 code. 1025 */ 1026 .extern artUnlockObjectFromCode 1027ENTRY art_quick_unlock_object 1028 cbz x0, .Lslow_unlock 1029 ldr w1, [x0, #LOCK_WORD_OFFSET] 1030 lsr w2, w1, 30 1031 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1032 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1033 eor w3, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1034 uxth w3, w3 // zero top 16 bits 1035 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1036 cmp w1, #65536 1037 bpl .Lrecursive_thin_unlock 1038 // transition to unlocked, w3 holds 0 1039 dmb ish // full (LoadStore|StoreStore) memory barrier 1040 str w3, [x0, #LOCK_WORD_OFFSET] 1041 ret 1042.Lrecursive_thin_unlock: 1043 sub w1, w1, #65536 1044 str w1, [x0, #LOCK_WORD_OFFSET] 1045 ret 1046.Lslow_unlock: 1047 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1048 mov x1, xSELF // pass Thread::Current 1049 mov x2, sp // pass SP 1050 bl artUnlockObjectFromCode // (Object* obj, Thread*, SP) 1051 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1052 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1053END art_quick_unlock_object 1054 1055 /* 1056 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1057 * artThrowClassCastException. 1058 */ 1059 .extern artThrowClassCastException 1060ENTRY art_quick_check_cast 1061 // Store arguments and link register 1062 sub sp, sp, #32 // Stack needs to be 16b aligned on calls 1063 .cfi_adjust_cfa_offset 32 1064 stp x0, x1, [sp] 1065 .cfi_rel_offset x0, 0 1066 .cfi_rel_offset x1, 8 1067 stp xSELF, xLR, [sp, #16] 1068 .cfi_rel_offset x18, 16 1069 .cfi_rel_offset x30, 24 1070 1071 // Call runtime code 1072 bl artIsAssignableFromCode 1073 1074 // Check for exception 1075 cbz x0, .Lthrow_class_cast_exception 1076 1077 // Restore and return 1078 ldp x0, x1, [sp] 1079 .cfi_restore x0 1080 .cfi_restore x1 1081 ldp xSELF, xLR, [sp, #16] 1082 .cfi_restore x18 1083 .cfi_restore x30 1084 add sp, sp, #32 1085 .cfi_adjust_cfa_offset -32 1086 ret 1087 1088.Lthrow_class_cast_exception: 1089 // Restore 1090 ldp x0, x1, [sp] 1091 .cfi_restore x0 1092 .cfi_restore x1 1093 ldp xSELF, xLR, [sp, #16] 1094 .cfi_restore x18 1095 .cfi_restore x30 1096 add sp, sp, #32 1097 .cfi_adjust_cfa_offset -32 1098 1099 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 1100 mov x2, xSELF // pass Thread::Current 1101 mov x3, sp // pass SP 1102 b artThrowClassCastException // (Class*, Class*, Thread*, SP) 1103 brk 0 // We should not return here... 1104END art_quick_check_cast 1105 1106 /* 1107 * Entry from managed code for array put operations of objects where the value being stored 1108 * needs to be checked for compatibility. 1109 * x0 = array, x1 = index, x2 = value 1110 * 1111 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1112 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1113 * using index-zero-extension in load/stores. 1114 * 1115 * Temporaries: x3, x4 1116 * TODO: x4 OK? ip seems wrong here. 1117 */ 1118ENTRY art_quick_aput_obj_with_null_and_bound_check 1119 tst x0, x0 1120 bne art_quick_aput_obj_with_bound_check 1121 b art_quick_throw_null_pointer_exception 1122END art_quick_aput_obj_with_null_and_bound_check 1123 1124ENTRY art_quick_aput_obj_with_bound_check 1125 ldr w3, [x0, #ARRAY_LENGTH_OFFSET] 1126 cmp w3, w1 1127 bhi art_quick_aput_obj 1128 mov x0, x1 1129 mov x1, x3 1130 b art_quick_throw_array_bounds 1131END art_quick_aput_obj_with_bound_check 1132 1133ENTRY art_quick_aput_obj 1134 cbz x2, .Ldo_aput_null 1135 ldr w3, [x0, #CLASS_OFFSET] // Heap reference = 32b 1136 // This also zero-extends to x3 1137 ldr w4, [x2, #CLASS_OFFSET] // Heap reference = 32b 1138 // This also zero-extends to x4 1139 ldr w3, [x3, #CLASS_COMPONENT_TYPE_OFFSET] // Heap reference = 32b 1140 // This also zero-extends to x3 1141 cmp w3, w4 // value's type == array's component type - trivial assignability 1142 bne .Lcheck_assignability 1143.Ldo_aput: 1144 add x3, x0, #OBJECT_ARRAY_DATA_OFFSET 1145 // "Compress" = do nothing 1146 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1147 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1148 lsr x0, x0, #7 1149 strb w3, [x3, x0] 1150 ret 1151.Ldo_aput_null: 1152 add x3, x0, #OBJECT_ARRAY_DATA_OFFSET 1153 // "Compress" = do nothing 1154 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1155 ret 1156.Lcheck_assignability: 1157 // Store arguments and link register 1158 sub sp, sp, #48 // Stack needs to be 16b aligned on calls 1159 .cfi_adjust_cfa_offset 48 1160 stp x0, x1, [sp] 1161 .cfi_rel_offset x0, 0 1162 .cfi_rel_offset x1, 8 1163 stp x2, xSELF, [sp, #16] 1164 .cfi_rel_offset x2, 16 1165 .cfi_rel_offset x18, 24 1166 str xLR, [sp, #32] 1167 .cfi_rel_offset x30, 32 1168 1169 // Call runtime code 1170 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1171 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1172 bl artIsAssignableFromCode 1173 1174 // Check for exception 1175 cbz x0, .Lthrow_array_store_exception 1176 1177 // Restore 1178 ldp x0, x1, [sp] 1179 .cfi_restore x0 1180 .cfi_restore x1 1181 ldp x2, xSELF, [sp, #16] 1182 .cfi_restore x2 1183 .cfi_restore x18 1184 ldr xLR, [sp, #32] 1185 .cfi_restore x30 1186 add sp, sp, #48 1187 .cfi_adjust_cfa_offset -48 1188 1189 add x3, x0, #OBJECT_ARRAY_DATA_OFFSET 1190 // "Compress" = do nothing 1191 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1192 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1193 lsr x0, x0, #7 1194 strb w3, [x3, x0] 1195 ret 1196.Lthrow_array_store_exception: 1197 ldp x0, x1, [sp] 1198 .cfi_restore x0 1199 .cfi_restore x1 1200 ldp x2, xSELF, [sp, #16] 1201 .cfi_restore x2 1202 .cfi_restore x18 1203 ldr xLR, [sp, #32] 1204 .cfi_restore x30 1205 add sp, sp, #48 1206 .cfi_adjust_cfa_offset -48 1207 1208 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1209 mov x1, x2 // Pass value. 1210 mov x2, xSELF // Pass Thread::Current. 1211 mov x3, sp // Pass SP. 1212 b artThrowArrayStoreException // (Object*, Object*, Thread*, SP). 1213 brk 0 // Unreached. 1214END art_quick_aput_obj 1215 1216// Macro to facilitate adding new allocation entrypoints. 1217.macro TWO_ARG_DOWNCALL name, entrypoint, return 1218 .extern \entrypoint 1219ENTRY \name 1220 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1221 mov x2, xSELF // pass Thread::Current 1222 mov x3, sp // pass SP 1223 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1224 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1225 \return 1226 DELIVER_PENDING_EXCEPTION 1227END \name 1228.endm 1229 1230// Macro to facilitate adding new array allocation entrypoints. 1231.macro THREE_ARG_DOWNCALL name, entrypoint, return 1232 .extern \entrypoint 1233ENTRY \name 1234 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1235 mov x3, xSELF // pass Thread::Current 1236 mov x4, sp // pass SP 1237 bl \entrypoint 1238 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1239 \return 1240 DELIVER_PENDING_EXCEPTION 1241END \name 1242.endm 1243 1244// Macros taking opportunity of code similarities for downcalls with referrer. 1245.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1246 .extern \entrypoint 1247ENTRY \name 1248 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1249 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1250 mov x2, xSELF // pass Thread::Current 1251 mov x3, sp // pass SP 1252 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1253 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1254 \return 1255END \name 1256.endm 1257 1258.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1259 .extern \entrypoint 1260ENTRY \name 1261 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1262 ldr w2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1263 mov x3, xSELF // pass Thread::Current 1264 mov x4, sp // pass SP 1265 bl \entrypoint 1266 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1267 \return 1268END \name 1269.endm 1270 1271.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1272 .extern \entrypoint 1273ENTRY \name 1274 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1275 ldr w3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1276 mov x4, xSELF // pass Thread::Current 1277 mov x5, sp // pass SP 1278 bl \entrypoint 1279 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1280 \return 1281END \name 1282.endm 1283 1284 /* 1285 * Entry from managed code when uninitialized static storage, this stub will run the class 1286 * initializer and deliver the exception on error. On success the static storage base is 1287 * returned. 1288 */ 1289TWO_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO 1290 1291TWO_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO 1292TWO_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO 1293 1294ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1295ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1296ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1297 1298TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1299TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1300TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1301 1302TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1303TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1304 1305THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1306THREE_ARG_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1307THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1308 1309// This is separated out as the argument order is different. 1310 .extern artSet64StaticFromCode 1311ENTRY art_quick_set64_static 1312 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1313 mov x3, x1 // Store value 1314 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1315 mov x2, x3 // Put value param 1316 mov x3, xSELF // pass Thread::Current 1317 mov x4, sp // pass SP 1318 bl artSet64StaticFromCode 1319 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME 1320 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1321END art_quick_set64_static 1322 1323 /* 1324 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1325 * exception on error. On success the String is returned. x0 holds the referring method, 1326 * w1 holds the string index. The fast path check for hit in strings cache has already been 1327 * performed. 1328 */ 1329TWO_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO 1330 1331// Generate the allocation entrypoints for each allocator. 1332GENERATE_ALL_ALLOC_ENTRYPOINTS 1333 1334 /* 1335 * Called by managed code when the value in wSUSPEND has been decremented to 0. 1336 */ 1337 .extern artTestSuspendFromCode 1338ENTRY art_quick_test_suspend 1339 ldrh w0, [xSELF, #THREAD_FLAGS_OFFSET] // get xSELF->state_and_flags.as_struct.flags 1340 mov wSUSPEND, #SUSPEND_CHECK_INTERVAL // reset wSUSPEND to SUSPEND_CHECK_INTERVAL 1341 cbnz w0, .Lneed_suspend // check flags == 0 1342 ret // return if flags == 0 1343.Lneed_suspend: 1344 mov x0, xSELF 1345 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1346 mov x1, sp 1347 bl artTestSuspendFromCode // (Thread*, SP) 1348 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1349END art_quick_test_suspend 1350 1351ENTRY art_quick_implicit_suspend 1352 mov x0, xSELF 1353 SETUP_REF_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1354 mov x1, sp 1355 bl artTestSuspendFromCode // (Thread*, SP) 1356 RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1357END art_quick_implicit_suspend 1358 1359 /* 1360 * Called by managed code that is attempting to call a method on a proxy class. On entry 1361 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 1362 * method agrees with a ref and args callee save frame. 1363 */ 1364 .extern artQuickProxyInvokeHandler 1365ENTRY art_quick_proxy_invoke_handler 1366 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME 1367 str x0, [sp, #0] // place proxy method at bottom of frame 1368 mov x2, xSELF // pass Thread::Current 1369 mov x3, sp // pass SP 1370 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 1371 // Use xETR as xSELF might be scratched by native function above. 1372 ldr x2, [xETR, THREAD_EXCEPTION_OFFSET] 1373 cbnz x2, .Lexception_in_proxy // success if no exception is pending 1374 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame 1375 fmov d0, x0 // Store result in d0 in case it was float or double 1376 ret // return on success 1377.Lexception_in_proxy: 1378 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 1379 DELIVER_PENDING_EXCEPTION 1380END art_quick_proxy_invoke_handler 1381 1382 /* 1383 * Called to resolve an imt conflict. xIP1 is a hidden argument that holds the target method's 1384 * dex method index. 1385 */ 1386ENTRY art_quick_imt_conflict_trampoline 1387 ldr w0, [sp, #0] // load caller Method* 1388 ldr w0, [x0, #METHOD_DEX_CACHE_METHODS_OFFSET] // load dex_cache_resolved_methods 1389 add x0, x0, #OBJECT_ARRAY_DATA_OFFSET // get starting address of data 1390 ldr w0, [x0, xIP1, lsl 2] // load the target method 1391 b art_quick_invoke_interface_trampoline 1392END art_quick_imt_conflict_trampoline 1393 1394ENTRY art_quick_resolution_trampoline 1395 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME 1396 mov x2, xSELF 1397 mov x3, sp 1398 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 1399 cbz x0, 1f 1400 mov xIP0, x0 // Remember returned code pointer in xIP0. 1401 ldr w0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 1402 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 1403 br xIP0 14041: 1405 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 1406 DELIVER_PENDING_EXCEPTION 1407END art_quick_resolution_trampoline 1408 1409/* 1410 * Generic JNI frame layout: 1411 * 1412 * #-------------------# 1413 * | | 1414 * | caller method... | 1415 * #-------------------# <--- SP on entry 1416 * | Return X30/LR | 1417 * | X29/FP | callee save 1418 * | X28 | callee save 1419 * | X27 | callee save 1420 * | X26 | callee save 1421 * | X25 | callee save 1422 * | X24 | callee save 1423 * | X23 | callee save 1424 * | X22 | callee save 1425 * | X21 | callee save 1426 * | X20 | callee save 1427 * | X7 | arg7 1428 * | X6 | arg6 1429 * | X5 | arg5 1430 * | X4 | arg4 1431 * | X3 | arg3 1432 * | X2 | arg2 1433 * | X1 | arg1 1434 * | D7 | float arg 8 1435 * | D6 | float arg 7 1436 * | D5 | float arg 6 1437 * | D4 | float arg 5 1438 * | D3 | float arg 4 1439 * | D2 | float arg 3 1440 * | D1 | float arg 2 1441 * | D0 | float arg 1 1442 * | Method* | <- X0 1443 * #-------------------# 1444 * | local ref cookie | // 4B 1445 * | handle scope size | // 4B 1446 * #-------------------# 1447 * | JNI Call Stack | 1448 * #-------------------# <--- SP on native call 1449 * | | 1450 * | Stack for Regs | The trampoline assembly will pop these values 1451 * | | into registers for native call 1452 * #-------------------# 1453 * | Native code ptr | 1454 * #-------------------# 1455 * | Free scratch | 1456 * #-------------------# 1457 * | Ptr to (1) | <--- SP 1458 * #-------------------# 1459 */ 1460 /* 1461 * Called to do a generic JNI down-call 1462 */ 1463ENTRY art_quick_generic_jni_trampoline 1464 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 1465 str x0, [sp, #0] // Store native ArtMethod* to bottom of stack. 1466 1467 // Save SP , so we can have static CFI info. 1468 mov x28, sp 1469 .cfi_def_cfa_register x28 1470 1471 // This looks the same, but is different: this will be updated to point to the bottom 1472 // of the frame when the handle scope is inserted. 1473 mov xFP, sp 1474 1475 mov xIP0, #5120 1476 sub sp, sp, xIP0 1477 1478 // prepare for artQuickGenericJniTrampoline call 1479 // (Thread*, SP) 1480 // x0 x1 <= C calling convention 1481 // xSELF xFP <= where they are 1482 1483 mov x0, xSELF // Thread* 1484 mov x1, xFP 1485 bl artQuickGenericJniTrampoline // (Thread*, sp) 1486 1487 // The C call will have registered the complete save-frame on success. 1488 // The result of the call is: 1489 // x0: pointer to native code, 0 on error. 1490 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 1491 1492 // Check for error = 0. 1493 cbz x0, .Lentry_error 1494 1495 // Release part of the alloca. 1496 mov sp, x1 1497 1498 // Save the code pointer 1499 mov xIP0, x0 1500 1501 // Load parameters from frame into registers. 1502 // TODO Check with artQuickGenericJniTrampoline. 1503 // Also, check again APPCS64 - the stack arguments are interleaved. 1504 ldp x0, x1, [sp] 1505 ldp x2, x3, [sp, #16] 1506 ldp x4, x5, [sp, #32] 1507 ldp x6, x7, [sp, #48] 1508 1509 ldp d0, d1, [sp, #64] 1510 ldp d2, d3, [sp, #80] 1511 ldp d4, d5, [sp, #96] 1512 ldp d6, d7, [sp, #112] 1513 1514 add sp, sp, #128 1515 1516 blr xIP0 // native call. 1517 1518 // result sign extension is handled in C code 1519 // prepare for artQuickGenericJniEndTrampoline call 1520 // (Thread*, result, result_f) 1521 // x0 x1 x2 <= C calling convention 1522 mov x1, x0 // Result (from saved) 1523 mov x0, xETR // Thread register, original xSELF might be scratched by native code. 1524 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 1525 1526 bl artQuickGenericJniEndTrampoline 1527 1528 // Tear down the alloca. 1529 mov sp, x28 1530 .cfi_def_cfa_register sp 1531 1532 // Pending exceptions possible. 1533 // Use xETR as xSELF might be scratched by native code 1534 ldr x1, [xETR, THREAD_EXCEPTION_OFFSET] 1535 cbnz x1, .Lexception_in_native 1536 1537 // Tear down the callee-save frame. 1538 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 1539 1540 // store into fpr, for when it's a fpr return... 1541 fmov d0, x0 1542 ret 1543 1544.Lentry_error: 1545 mov sp, x28 1546 .cfi_def_cfa_register sp 1547.Lexception_in_native: 1548 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME 1549 DELIVER_PENDING_EXCEPTION 1550 1551END art_quick_generic_jni_trampoline 1552 1553/* 1554 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 1555 * of a quick call: 1556 * x0 = method being called/to bridge to. 1557 * x1..x7, d0..d7 = arguments to that method. 1558 */ 1559ENTRY art_quick_to_interpreter_bridge 1560 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME // Set up frame and save arguments. 1561 1562 // x0 will contain mirror::ArtMethod* method. 1563 mov x1, xSELF // How to get Thread::Current() ??? 1564 mov x2, sp 1565 1566 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 1567 // mirror::ArtMethod** sp) 1568 bl artQuickToInterpreterBridge 1569 1570 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME // TODO: no need to restore arguments in this case. 1571 1572 fmov d0, x0 1573 1574 RETURN_OR_DELIVER_PENDING_EXCEPTION 1575END art_quick_to_interpreter_bridge 1576 1577 1578// 1579// Instrumentation-related stubs 1580// 1581 .extern artInstrumentationMethodEntryFromCode 1582ENTRY art_quick_instrumentation_entry 1583 SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME 1584 1585 mov x20, x0 // Preserve method reference in a callee-save. 1586 1587 mov x2, xSELF 1588 mov x3, sp 1589 mov x4, xLR 1590 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, SP, LR) 1591 1592 mov xIP0, x0 // x0 = result of call. 1593 mov x0, x20 // Reload method reference. 1594 1595 RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME // Note: will restore xSELF 1596 adr xLR, art_quick_instrumentation_exit 1597 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 1598END art_quick_instrumentation_entry 1599 1600 .extern artInstrumentationMethodExitFromCode 1601ENTRY art_quick_instrumentation_exit 1602 mov xLR, #0 // Clobber LR for later checks. 1603 1604 SETUP_REF_ONLY_CALLEE_SAVE_FRAME 1605 1606 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 1607 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 1608 // easier to have an extra small stack area. 1609 1610 str x0, [sp, #-16]! // Save integer result. 1611 .cfi_adjust_cfa_offset 16 1612 str d0, [sp, #8] // Save floating-point result. 1613 1614 add x1, sp, #16 // Pass SP. 1615 mov x2, x0 // Pass integer result. 1616 fmov x3, d0 // Pass floating-point result. 1617 mov x0, xSELF // Pass Thread. 1618 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 1619 1620 mov xIP0, x0 // Return address from instrumentation call. 1621 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 1622 1623 ldr d0, [sp, #8] // Restore floating-point result. 1624 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 1625 .cfi_adjust_cfa_offset 16 1626 1627 POP_REF_ONLY_CALLEE_SAVE_FRAME 1628 1629 br xIP0 // Tail-call out. 1630END art_quick_instrumentation_exit 1631 1632 /* 1633 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 1634 * will long jump to the upcall with a special exception of -1. 1635 */ 1636 .extern artDeoptimize 1637ENTRY art_quick_deoptimize 1638 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1639 mov x0, xSELF // Pass thread. 1640 mov x1, sp // Pass SP. 1641 bl artDeoptimize // artDeoptimize(Thread*, SP) 1642 brk 0 1643END art_quick_deoptimize 1644 1645 1646 /* 1647 * String's indexOf. 1648 * 1649 * TODO: Not very optimized. 1650 * On entry: 1651 * x0: string object (known non-null) 1652 * w1: char to match (known <= 0xFFFF) 1653 * w2: Starting offset in string data 1654 */ 1655ENTRY art_quick_indexof 1656 ldr w3, [x0, #STRING_COUNT_OFFSET] 1657 ldr w4, [x0, #STRING_OFFSET_OFFSET] 1658 ldr w0, [x0, #STRING_VALUE_OFFSET] // x0 ? 1659 1660 /* Clamp start to [0..count] */ 1661 cmp w2, #0 1662 csel w2, wzr, w2, lt 1663 cmp w2, w3 1664 csel w2, w3, w2, gt 1665 1666 /* Build a pointer to the start of the string data */ 1667 add x0, x0, #STRING_DATA_OFFSET 1668 add x0, x0, x4, lsl #1 1669 1670 /* Save a copy to compute result */ 1671 mov x5, x0 1672 1673 /* Build pointer to start of data to compare and pre-bias */ 1674 add x0, x0, x2, lsl #1 1675 sub x0, x0, #2 1676 1677 /* Compute iteration count */ 1678 sub w2, w3, w2 1679 1680 /* 1681 * At this point we have: 1682 * x0: start of the data to test 1683 * w1: char to compare 1684 * w2: iteration count 1685 * x5: original start of string data 1686 */ 1687 1688 subs w2, w2, #4 1689 b.lt .Lindexof_remainder 1690 1691.Lindexof_loop4: 1692 ldrh w6, [x0, #2]! 1693 ldrh w7, [x0, #2]! 1694 ldrh wIP0, [x0, #2]! 1695 ldrh wIP1, [x0, #2]! 1696 cmp w6, w1 1697 b.eq .Lmatch_0 1698 cmp w7, w1 1699 b.eq .Lmatch_1 1700 cmp wIP0, w1 1701 b.eq .Lmatch_2 1702 cmp wIP1, w1 1703 b.eq .Lmatch_3 1704 subs w2, w2, #4 1705 b.ge .Lindexof_loop4 1706 1707.Lindexof_remainder: 1708 adds w2, w2, #4 1709 b.eq .Lindexof_nomatch 1710 1711.Lindexof_loop1: 1712 ldrh w6, [x0, #2]! 1713 cmp w6, w1 1714 b.eq .Lmatch_3 1715 subs w2, w2, #1 1716 b.ne .Lindexof_loop1 1717 1718.Lindexof_nomatch: 1719 mov x0, #-1 1720 ret 1721 1722.Lmatch_0: 1723 sub x0, x0, #6 1724 sub x0, x0, x5 1725 asr x0, x0, #1 1726 ret 1727.Lmatch_1: 1728 sub x0, x0, #4 1729 sub x0, x0, x5 1730 asr x0, x0, #1 1731 ret 1732.Lmatch_2: 1733 sub x0, x0, #2 1734 sub x0, x0, x5 1735 asr x0, x0, #1 1736 ret 1737.Lmatch_3: 1738 sub x0, x0, x5 1739 asr x0, x0, #1 1740 ret 1741END art_quick_indexof 1742 1743 /* 1744 * String's compareTo. 1745 * 1746 * TODO: Not very optimized. 1747 * 1748 * On entry: 1749 * x0: this object pointer 1750 * x1: comp object pointer 1751 * 1752 */ 1753 .extern __memcmp16 1754ENTRY art_quick_string_compareto 1755 mov x2, x0 // x0 is return, use x2 for first input. 1756 sub x0, x2, x1 // Same string object? 1757 cbnz x0,1f 1758 ret 17591: // Different string objects. 1760 1761 ldr w6, [x2, #STRING_OFFSET_OFFSET] 1762 ldr w5, [x1, #STRING_OFFSET_OFFSET] 1763 ldr w4, [x2, #STRING_COUNT_OFFSET] 1764 ldr w3, [x1, #STRING_COUNT_OFFSET] 1765 ldr w2, [x2, #STRING_VALUE_OFFSET] 1766 ldr w1, [x1, #STRING_VALUE_OFFSET] 1767 1768 /* 1769 * Now: CharArray* Offset Count 1770 * first arg x2 w6 w4 1771 * second arg x1 w5 w3 1772 */ 1773 1774 // x0 := str1.length(w4) - str2.length(w3). ldr zero-extended w3/w4 into x3/x4. 1775 subs x0, x4, x3 1776 // Min(count1, count2) into w3. 1777 csel x3, x3, x4, ge 1778 1779 // Build pointer into string data. 1780 1781 // Add offset in array (substr etc.) (sign extend and << 1). 1782 add x2, x2, w6, sxtw #1 1783 add x1, x1, w5, sxtw #1 1784 1785 // Add offset in CharArray to array. 1786 add x2, x2, #STRING_DATA_OFFSET 1787 add x1, x1, #STRING_DATA_OFFSET 1788 1789 // TODO: Tune this value. 1790 // Check for long string, do memcmp16 for them. 1791 cmp w3, #28 // Constant from arm32. 1792 bgt .Ldo_memcmp16 1793 1794 /* 1795 * Now: 1796 * x2: *first string data 1797 * x1: *second string data 1798 * w3: iteration count 1799 * x0: return value if comparison equal 1800 * x4, x5, x6, x7: free 1801 */ 1802 1803 // Do a simple unrolled loop. 1804.Lloop: 1805 // At least two more elements? 1806 subs w3, w3, #2 1807 b.lt .Lremainder_or_done 1808 1809 ldrh w4, [x2], #2 1810 ldrh w5, [x1], #2 1811 1812 ldrh w6, [x2], #2 1813 ldrh w7, [x1], #2 1814 1815 subs w4, w4, w5 1816 b.ne .Lw4_result 1817 1818 subs w6, w6, w7 1819 b.ne .Lw6_result 1820 1821 b .Lloop 1822 1823.Lremainder_or_done: 1824 adds w3, w3, #1 1825 b.eq .Lremainder 1826 ret 1827 1828.Lremainder: 1829 ldrh w4, [x2], #2 1830 ldrh w5, [x1], #2 1831 subs w4, w4, w5 1832 b.ne .Lw4_result 1833 ret 1834 1835// Result is in w4 1836.Lw4_result: 1837 sxtw x0, w4 1838 ret 1839 1840// Result is in w6 1841.Lw6_result: 1842 sxtw x0, w6 1843 ret 1844 1845.Ldo_memcmp16: 1846 mov x14, x0 // Save x0 and LR. __memcmp16 does not use these temps. 1847 mov x15, xLR // TODO: Codify and check that? 1848 1849 mov x0, x2 1850 uxtw x2, w3 1851 bl __memcmp16 1852 1853 mov xLR, x15 // Restore LR. 1854 1855 cmp x0, #0 // Check the memcmp difference. 1856 csel x0, x0, x14, ne // x0 := x0 != 0 ? x14(prev x0=length diff) : x1. 1857 ret 1858END art_quick_string_compareto 1859 1860// Macro to facilitate adding new entrypoints which call to native function directly. 1861// Currently, xSELF is the only thing we need to take care of between managed code and AAPCS. 1862// But we might introduce more differences. 1863.macro NATIVE_DOWNCALL name, entrypoint 1864 .extern \entrypoint 1865ENTRY \name 1866 stp xSELF, xLR, [sp, #-16]! 1867 bl \entrypoint 1868 ldp xSELF, xLR, [sp], #16 1869 ret 1870END \name 1871.endm 1872 1873NATIVE_DOWNCALL art_quick_fmod fmod 1874NATIVE_DOWNCALL art_quick_fmodf fmodf 1875NATIVE_DOWNCALL art_quick_memcpy memcpy 1876NATIVE_DOWNCALL art_quick_assignable_from_code artIsAssignableFromCode 1877